diff options
| author | Kevin Krakauer <krakauer@google.com> | 2020-01-21 13:16:25 -0800 |
|---|---|---|
| committer | Kevin Krakauer <krakauer@google.com> | 2020-01-21 13:16:25 -0800 |
| commit | 62357a0afb5f4128a11dc9a1dfadd2957ec39e2d (patch) | |
| tree | 2f93dc1bb5680434f3bcd69df60d45af89777a94 /pkg | |
| parent | bd292894097ffdf316bc78d81aebd0a2988124f3 (diff) | |
| parent | 2ba6198851dc1e293295d7cadf8c0ae456b68beb (diff) | |
Merge branch 'master' into iptables-write-filter-proto
Diffstat (limited to 'pkg')
141 files changed, 6637 insertions, 4306 deletions
diff --git a/pkg/abi/linux/time.go b/pkg/abi/linux/time.go index 546668bca..5c5a58cd4 100644 --- a/pkg/abi/linux/time.go +++ b/pkg/abi/linux/time.go @@ -234,6 +234,19 @@ type StatxTimestamp struct { _ int32 } +// ToNsec returns the nanosecond representation. +func (sxts StatxTimestamp) ToNsec() int64 { + return int64(sxts.Sec)*1e9 + int64(sxts.Nsec) +} + +// ToNsecCapped returns the safe nanosecond representation. +func (sxts StatxTimestamp) ToNsecCapped() int64 { + if sxts.Sec > maxSecInDuration { + return math.MaxInt64 + } + return sxts.ToNsec() +} + // NsecToStatxTimestamp translates nanoseconds to StatxTimestamp. func NsecToStatxTimestamp(nsec int64) (ts StatxTimestamp) { return StatxTimestamp{ diff --git a/pkg/cpuid/cpuid.go b/pkg/cpuid/cpuid.go index d37047368..cf50ee53f 100644 --- a/pkg/cpuid/cpuid.go +++ b/pkg/cpuid/cpuid.go @@ -657,30 +657,28 @@ func (fs *FeatureSet) FlagsString(cpuinfoOnly bool) string { return strings.Join(s, " ") } -// CPUInfo is to generate a section of one cpu in /proc/cpuinfo. This is a -// minimal /proc/cpuinfo, it is missing some fields like "microcode" that are +// WriteCPUInfoTo is to generate a section of one cpu in /proc/cpuinfo. This is +// a minimal /proc/cpuinfo, it is missing some fields like "microcode" that are // not always printed in Linux. The bogomips field is simply made up. -func (fs FeatureSet) CPUInfo(cpu uint) string { - var b bytes.Buffer - fmt.Fprintf(&b, "processor\t: %d\n", cpu) - fmt.Fprintf(&b, "vendor_id\t: %s\n", fs.VendorID) - fmt.Fprintf(&b, "cpu family\t: %d\n", ((fs.ExtendedFamily<<4)&0xff)|fs.Family) - fmt.Fprintf(&b, "model\t\t: %d\n", ((fs.ExtendedModel<<4)&0xff)|fs.Model) - fmt.Fprintf(&b, "model name\t: %s\n", "unknown") // Unknown for now. - fmt.Fprintf(&b, "stepping\t: %s\n", "unknown") // Unknown for now. - fmt.Fprintf(&b, "cpu MHz\t\t: %.3f\n", cpuFreqMHz) - fmt.Fprintln(&b, "fpu\t\t: yes") - fmt.Fprintln(&b, "fpu_exception\t: yes") - fmt.Fprintf(&b, "cpuid level\t: %d\n", uint32(xSaveInfo)) // Same as ax in vendorID. - fmt.Fprintln(&b, "wp\t\t: yes") - fmt.Fprintf(&b, "flags\t\t: %s\n", fs.FlagsString(true)) - fmt.Fprintf(&b, "bogomips\t: %.02f\n", cpuFreqMHz) // It's bogus anyway. - fmt.Fprintf(&b, "clflush size\t: %d\n", fs.CacheLine) - fmt.Fprintf(&b, "cache_alignment\t: %d\n", fs.CacheLine) - fmt.Fprintf(&b, "address sizes\t: %d bits physical, %d bits virtual\n", 46, 48) - fmt.Fprintln(&b, "power management:") // This is always here, but can be blank. - fmt.Fprintln(&b, "") // The /proc/cpuinfo file ends with an extra newline. - return b.String() +func (fs FeatureSet) WriteCPUInfoTo(cpu uint, b *bytes.Buffer) { + fmt.Fprintf(b, "processor\t: %d\n", cpu) + fmt.Fprintf(b, "vendor_id\t: %s\n", fs.VendorID) + fmt.Fprintf(b, "cpu family\t: %d\n", ((fs.ExtendedFamily<<4)&0xff)|fs.Family) + fmt.Fprintf(b, "model\t\t: %d\n", ((fs.ExtendedModel<<4)&0xff)|fs.Model) + fmt.Fprintf(b, "model name\t: %s\n", "unknown") // Unknown for now. + fmt.Fprintf(b, "stepping\t: %s\n", "unknown") // Unknown for now. + fmt.Fprintf(b, "cpu MHz\t\t: %.3f\n", cpuFreqMHz) + fmt.Fprintln(b, "fpu\t\t: yes") + fmt.Fprintln(b, "fpu_exception\t: yes") + fmt.Fprintf(b, "cpuid level\t: %d\n", uint32(xSaveInfo)) // Same as ax in vendorID. + fmt.Fprintln(b, "wp\t\t: yes") + fmt.Fprintf(b, "flags\t\t: %s\n", fs.FlagsString(true)) + fmt.Fprintf(b, "bogomips\t: %.02f\n", cpuFreqMHz) // It's bogus anyway. + fmt.Fprintf(b, "clflush size\t: %d\n", fs.CacheLine) + fmt.Fprintf(b, "cache_alignment\t: %d\n", fs.CacheLine) + fmt.Fprintf(b, "address sizes\t: %d bits physical, %d bits virtual\n", 46, 48) + fmt.Fprintln(b, "power management:") // This is always here, but can be blank. + fmt.Fprintln(b, "") // The /proc/cpuinfo file ends with an extra newline. } const ( diff --git a/pkg/p9/client_file.go b/pkg/p9/client_file.go index de9357389..0254e4ccc 100644 --- a/pkg/p9/client_file.go +++ b/pkg/p9/client_file.go @@ -165,6 +165,35 @@ func (c *clientFile) SetAttr(valid SetAttrMask, attr SetAttr) error { return c.client.sendRecv(&Tsetattr{FID: c.fid, Valid: valid, SetAttr: attr}, &Rsetattr{}) } +// GetXattr implements File.GetXattr. +func (c *clientFile) GetXattr(name string, size uint64) (string, error) { + if atomic.LoadUint32(&c.closed) != 0 { + return "", syscall.EBADF + } + if !versionSupportsGetSetXattr(c.client.version) { + return "", syscall.EOPNOTSUPP + } + + rgetxattr := Rgetxattr{} + if err := c.client.sendRecv(&Tgetxattr{FID: c.fid, Name: name, Size: size}, &rgetxattr); err != nil { + return "", err + } + + return rgetxattr.Value, nil +} + +// SetXattr implements File.SetXattr. +func (c *clientFile) SetXattr(name, value string, flags uint32) error { + if atomic.LoadUint32(&c.closed) != 0 { + return syscall.EBADF + } + if !versionSupportsGetSetXattr(c.client.version) { + return syscall.EOPNOTSUPP + } + + return c.client.sendRecv(&Tsetxattr{FID: c.fid, Name: name, Value: value, Flags: flags}, &Rsetxattr{}) +} + // Allocate implements File.Allocate. func (c *clientFile) Allocate(mode AllocateMode, offset, length uint64) error { if atomic.LoadUint32(&c.closed) != 0 { diff --git a/pkg/p9/file.go b/pkg/p9/file.go index 96d1f2a8e..4607cfcdf 100644 --- a/pkg/p9/file.go +++ b/pkg/p9/file.go @@ -89,6 +89,22 @@ type File interface { // On the server, SetAttr has a write concurrency guarantee. SetAttr(valid SetAttrMask, attr SetAttr) error + // GetXattr returns extended attributes of this node. + // + // Size indicates the size of the buffer that has been allocated to hold the + // attribute value. If the value is larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + // + // TODO(b/127675828): Determine concurrency guarantees once implemented. + GetXattr(name string, size uint64) (string, error) + + // SetXattr sets extended attributes on this node. + // + // TODO(b/127675828): Determine concurrency guarantees once implemented. + SetXattr(name, value string, flags uint32) error + // Allocate allows the caller to directly manipulate the allocated disk space // for the file. See fallocate(2) for more details. Allocate(mode AllocateMode, offset, length uint64) error diff --git a/pkg/p9/handlers.go b/pkg/p9/handlers.go index b9582c07f..7d6653a07 100644 --- a/pkg/p9/handlers.go +++ b/pkg/p9/handlers.go @@ -913,6 +913,35 @@ func (t *Txattrcreate) handle(cs *connState) message { } // handle implements handler.handle. +func (t *Tgetxattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + val, err := ref.file.GetXattr(t.Name, t.Size) + if err != nil { + return newErr(err) + } + return &Rgetxattr{Value: val} +} + +// handle implements handler.handle. +func (t *Tsetxattr) handle(cs *connState) message { + ref, ok := cs.LookupFID(t.FID) + if !ok { + return newErr(syscall.EBADF) + } + defer ref.DecRef() + + if err := ref.file.SetXattr(t.Name, t.Value, t.Flags); err != nil { + return newErr(err) + } + return &Rsetxattr{} +} + +// handle implements handler.handle. func (t *Treaddir) handle(cs *connState) message { ref, ok := cs.LookupFID(t.Directory) if !ok { diff --git a/pkg/p9/messages.go b/pkg/p9/messages.go index ffdd7e8c6..ceb723d86 100644 --- a/pkg/p9/messages.go +++ b/pkg/p9/messages.go @@ -1611,6 +1611,131 @@ func (r *Rxattrcreate) String() string { return fmt.Sprintf("Rxattrcreate{}") } +// Tgetxattr is a getxattr request. +type Tgetxattr struct { + // FID refers to the file for which to get xattrs. + FID FID + + // Name is the xattr to get. + Name string + + // Size is the buffer size for the xattr to get. + Size uint64 +} + +// Decode implements encoder.Decode. +func (t *Tgetxattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Name = b.ReadString() + t.Size = b.Read64() +} + +// Encode implements encoder.Encode. +func (t *Tgetxattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.WriteString(t.Name) + b.Write64(t.Size) +} + +// Type implements message.Type. +func (*Tgetxattr) Type() MsgType { + return MsgTgetxattr +} + +// String implements fmt.Stringer. +func (t *Tgetxattr) String() string { + return fmt.Sprintf("Tgetxattr{FID: %d, Name: %s, Size: %d}", t.FID, t.Name, t.Size) +} + +// Rgetxattr is a getxattr response. +type Rgetxattr struct { + // Value is the extended attribute value. + Value string +} + +// Decode implements encoder.Decode. +func (r *Rgetxattr) Decode(b *buffer) { + r.Value = b.ReadString() +} + +// Encode implements encoder.Encode. +func (r *Rgetxattr) Encode(b *buffer) { + b.WriteString(r.Value) +} + +// Type implements message.Type. +func (*Rgetxattr) Type() MsgType { + return MsgRgetxattr +} + +// String implements fmt.Stringer. +func (r *Rgetxattr) String() string { + return fmt.Sprintf("Rgetxattr{Value: %s}", r.Value) +} + +// Tsetxattr sets extended attributes. +type Tsetxattr struct { + // FID refers to the file on which to set xattrs. + FID FID + + // Name is the attribute name. + Name string + + // Value is the attribute value. + Value string + + // Linux setxattr(2) flags. + Flags uint32 +} + +// Decode implements encoder.Decode. +func (t *Tsetxattr) Decode(b *buffer) { + t.FID = b.ReadFID() + t.Name = b.ReadString() + t.Value = b.ReadString() + t.Flags = b.Read32() +} + +// Encode implements encoder.Encode. +func (t *Tsetxattr) Encode(b *buffer) { + b.WriteFID(t.FID) + b.WriteString(t.Name) + b.WriteString(t.Value) + b.Write32(t.Flags) +} + +// Type implements message.Type. +func (*Tsetxattr) Type() MsgType { + return MsgTsetxattr +} + +// String implements fmt.Stringer. +func (t *Tsetxattr) String() string { + return fmt.Sprintf("Tsetxattr{FID: %d, Name: %s, Value: %s, Flags: %d}", t.FID, t.Name, t.Value, t.Flags) +} + +// Rsetxattr is a setxattr response. +type Rsetxattr struct { +} + +// Decode implements encoder.Decode. +func (r *Rsetxattr) Decode(b *buffer) { +} + +// Encode implements encoder.Encode. +func (r *Rsetxattr) Encode(b *buffer) { +} + +// Type implements message.Type. +func (*Rsetxattr) Type() MsgType { + return MsgRsetxattr +} + +// String implements fmt.Stringer. +func (r *Rsetxattr) String() string { + return fmt.Sprintf("Rsetxattr{}") +} + // Treaddir is a readdir request. type Treaddir struct { // Directory is the directory FID to read. @@ -2363,6 +2488,10 @@ func init() { msgRegistry.register(MsgRxattrwalk, func() message { return &Rxattrwalk{} }) msgRegistry.register(MsgTxattrcreate, func() message { return &Txattrcreate{} }) msgRegistry.register(MsgRxattrcreate, func() message { return &Rxattrcreate{} }) + msgRegistry.register(MsgTgetxattr, func() message { return &Tgetxattr{} }) + msgRegistry.register(MsgRgetxattr, func() message { return &Rgetxattr{} }) + msgRegistry.register(MsgTsetxattr, func() message { return &Tsetxattr{} }) + msgRegistry.register(MsgRsetxattr, func() message { return &Rsetxattr{} }) msgRegistry.register(MsgTreaddir, func() message { return &Treaddir{} }) msgRegistry.register(MsgRreaddir, func() message { return &Rreaddir{} }) msgRegistry.register(MsgTfsync, func() message { return &Tfsync{} }) diff --git a/pkg/p9/messages_test.go b/pkg/p9/messages_test.go index 6ba6a1654..825c939da 100644 --- a/pkg/p9/messages_test.go +++ b/pkg/p9/messages_test.go @@ -194,6 +194,21 @@ func TestEncodeDecode(t *testing.T) { Flags: 3, }, &Rxattrcreate{}, + &Tgetxattr{ + FID: 1, + Name: "abc", + Size: 2, + }, + &Rgetxattr{ + Value: "xyz", + }, + &Tsetxattr{ + FID: 1, + Name: "abc", + Value: "xyz", + Flags: 2, + }, + &Rsetxattr{}, &Treaddir{ Directory: 1, Offset: 2, diff --git a/pkg/p9/p9.go b/pkg/p9/p9.go index d3090535a..5ab00d625 100644 --- a/pkg/p9/p9.go +++ b/pkg/p9/p9.go @@ -339,6 +339,10 @@ const ( MsgRxattrwalk = 31 MsgTxattrcreate = 32 MsgRxattrcreate = 33 + MsgTgetxattr = 34 + MsgRgetxattr = 35 + MsgTsetxattr = 36 + MsgRsetxattr = 37 MsgTreaddir = 40 MsgRreaddir = 41 MsgTfsync = 50 diff --git a/pkg/p9/version.go b/pkg/p9/version.go index 36a694c58..34a15eb55 100644 --- a/pkg/p9/version.go +++ b/pkg/p9/version.go @@ -26,7 +26,7 @@ const ( // // Clients are expected to start requesting this version number and // to continuously decrement it until a Tversion request succeeds. - highestSupportedVersion uint32 = 9 + highestSupportedVersion uint32 = 10 // lowestSupportedVersion is the lowest supported version X in a // version string of the format 9P2000.L.Google.X. @@ -161,3 +161,9 @@ func versionSupportsFlipcall(v uint32) bool { func VersionSupportsOpenTruncateFlag(v uint32) bool { return v >= 9 } + +// versionSupportsGetSetXattr returns true if version v supports +// the Tgetxattr and Tsetxattr messages. +func versionSupportsGetSetXattr(v uint32) bool { + return v >= 10 +} diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index ae3e364cd..65f22af2b 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -9,17 +9,23 @@ go_library( srcs = [ "aligned.go", "arch.go", + "arch_aarch64.go", "arch_amd64.go", "arch_amd64.s", + "arch_arm64.go", + "arch_state_aarch64.go", "arch_state_x86.go", "arch_x86.go", "auxv.go", + "signal.go", "signal_act.go", "signal_amd64.go", + "signal_arm64.go", "signal_info.go", "signal_stack.go", "stack.go", "syscalls_amd64.go", + "syscalls_arm64.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/arch", visibility = ["//:sandbox"], diff --git a/pkg/sentry/arch/arch_aarch64.go b/pkg/sentry/arch/arch_aarch64.go new file mode 100644 index 000000000..ea4dedbdf --- /dev/null +++ b/pkg/sentry/arch/arch_aarch64.go @@ -0,0 +1,293 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package arch + +import ( + "fmt" + "io" + "syscall" + + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/log" + rpb "gvisor.dev/gvisor/pkg/sentry/arch/registers_go_proto" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" +) + +const ( + // SyscallWidth is the width of insturctions. + SyscallWidth = 4 +) + +// aarch64FPState is aarch64 floating point state. +type aarch64FPState []byte + +// initAarch64FPState (defined in asm files) sets up initial state. +func initAarch64FPState(data *FloatingPointData) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. +} + +func newAarch64FPStateSlice() []byte { + return alignedBytes(4096, 32)[:4096] +} + +// newAarch64FPState returns an initialized floating point state. +// +// The returned state is large enough to store all floating point state +// supported by host, even if the app won't use much of it due to a restricted +// FeatureSet. Since they may still be able to see state not advertised by +// CPUID we must ensure it does not contain any sentry state. +func newAarch64FPState() aarch64FPState { + f := aarch64FPState(newAarch64FPStateSlice()) + initAarch64FPState(f.FloatingPointData()) + return f +} + +// fork creates and returns an identical copy of the aarch64 floating point state. +func (f aarch64FPState) fork() aarch64FPState { + n := aarch64FPState(newAarch64FPStateSlice()) + copy(n, f) + return n +} + +// FloatingPointData returns the raw data pointer. +func (f aarch64FPState) FloatingPointData() *FloatingPointData { + return (*FloatingPointData)(&f[0]) +} + +// NewFloatingPointData returns a new floating point data blob. +// +// This is primarily for use in tests. +func NewFloatingPointData() *FloatingPointData { + return (*FloatingPointData)(&(newAarch64FPState()[0])) +} + +// State contains the common architecture bits for aarch64 (the build tag of this +// file ensures it's only built on aarch64). +type State struct { + // The system registers. + Regs syscall.PtraceRegs `state:".(syscallPtraceRegs)"` + + // Our floating point state. + aarch64FPState `state:"wait"` + + // FeatureSet is a pointer to the currently active feature set. + FeatureSet *cpuid.FeatureSet +} + +// Proto returns a protobuf representation of the system registers in State. +func (s State) Proto() *rpb.Registers { + regs := &rpb.ARM64Registers{ + R0: s.Regs.Regs[0], + R1: s.Regs.Regs[1], + R2: s.Regs.Regs[2], + R3: s.Regs.Regs[3], + R4: s.Regs.Regs[4], + R5: s.Regs.Regs[5], + R6: s.Regs.Regs[6], + R7: s.Regs.Regs[7], + R8: s.Regs.Regs[8], + R9: s.Regs.Regs[9], + R10: s.Regs.Regs[10], + R11: s.Regs.Regs[11], + R12: s.Regs.Regs[12], + R13: s.Regs.Regs[13], + R14: s.Regs.Regs[14], + R15: s.Regs.Regs[15], + R16: s.Regs.Regs[16], + R17: s.Regs.Regs[17], + R18: s.Regs.Regs[18], + R19: s.Regs.Regs[19], + R20: s.Regs.Regs[20], + R21: s.Regs.Regs[21], + R22: s.Regs.Regs[22], + R23: s.Regs.Regs[23], + R24: s.Regs.Regs[24], + R25: s.Regs.Regs[25], + R26: s.Regs.Regs[26], + R27: s.Regs.Regs[27], + R28: s.Regs.Regs[28], + R29: s.Regs.Regs[29], + R30: s.Regs.Regs[30], + Sp: s.Regs.Sp, + Pc: s.Regs.Pc, + Pstate: s.Regs.Pstate, + } + return &rpb.Registers{Arch: &rpb.Registers_Arm64{Arm64: regs}} +} + +// Fork creates and returns an identical copy of the state. +func (s *State) Fork() State { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return State{ + Regs: s.Regs, + FeatureSet: s.FeatureSet, + } +} + +// StateData implements Context.StateData. +func (s *State) StateData() *State { + return s +} + +// CPUIDEmulate emulates a cpuid instruction. +func (s *State) CPUIDEmulate(l log.Logger) { + // TODO(gvisor.dev/issue/1255): cpuid is not supported. +} + +// SingleStep implements Context.SingleStep. +func (s *State) SingleStep() bool { + return false +} + +// SetSingleStep enables single stepping. +func (s *State) SetSingleStep() { + // Set the trap flag. + // TODO(gvisor.dev/issue/1239): ptrace single-step is not supported. +} + +// ClearSingleStep enables single stepping. +func (s *State) ClearSingleStep() { + // Clear the trap flag. + // TODO(gvisor.dev/issue/1239): ptrace single-step is not supported. +} + +// RegisterMap returns a map of all registers. +func (s *State) RegisterMap() (map[string]uintptr, error) { + return map[string]uintptr{ + "R0": uintptr(s.Regs.Regs[0]), + "R1": uintptr(s.Regs.Regs[1]), + "R2": uintptr(s.Regs.Regs[2]), + "R3": uintptr(s.Regs.Regs[3]), + "R4": uintptr(s.Regs.Regs[4]), + "R5": uintptr(s.Regs.Regs[5]), + "R6": uintptr(s.Regs.Regs[6]), + "R7": uintptr(s.Regs.Regs[7]), + "R8": uintptr(s.Regs.Regs[8]), + "R9": uintptr(s.Regs.Regs[9]), + "R10": uintptr(s.Regs.Regs[10]), + "R11": uintptr(s.Regs.Regs[11]), + "R12": uintptr(s.Regs.Regs[12]), + "R13": uintptr(s.Regs.Regs[13]), + "R14": uintptr(s.Regs.Regs[14]), + "R15": uintptr(s.Regs.Regs[15]), + "R16": uintptr(s.Regs.Regs[16]), + "R17": uintptr(s.Regs.Regs[17]), + "R18": uintptr(s.Regs.Regs[18]), + "R19": uintptr(s.Regs.Regs[19]), + "R20": uintptr(s.Regs.Regs[20]), + "R21": uintptr(s.Regs.Regs[21]), + "R22": uintptr(s.Regs.Regs[22]), + "R23": uintptr(s.Regs.Regs[23]), + "R24": uintptr(s.Regs.Regs[24]), + "R25": uintptr(s.Regs.Regs[25]), + "R26": uintptr(s.Regs.Regs[26]), + "R27": uintptr(s.Regs.Regs[27]), + "R28": uintptr(s.Regs.Regs[28]), + "R29": uintptr(s.Regs.Regs[29]), + "R30": uintptr(s.Regs.Regs[30]), + "Sp": uintptr(s.Regs.Sp), + "Pc": uintptr(s.Regs.Pc), + "Pstate": uintptr(s.Regs.Pstate), + }, nil +} + +// PtraceGetRegs implements Context.PtraceGetRegs. +func (s *State) PtraceGetRegs(dst io.Writer) (int, error) { + return dst.Write(binary.Marshal(nil, usermem.ByteOrder, s.ptraceGetRegs())) +} + +func (s *State) ptraceGetRegs() syscall.PtraceRegs { + return s.Regs +} + +var ptraceRegsSize = int(binary.Size(syscall.PtraceRegs{})) + +// PtraceSetRegs implements Context.PtraceSetRegs. +func (s *State) PtraceSetRegs(src io.Reader) (int, error) { + var regs syscall.PtraceRegs + buf := make([]byte, ptraceRegsSize) + if _, err := io.ReadFull(src, buf); err != nil { + return 0, err + } + binary.Unmarshal(buf, usermem.ByteOrder, ®s) + s.Regs = regs + return ptraceRegsSize, nil +} + +// PtraceGetFPRegs implements Context.PtraceGetFPRegs. +func (s *State) PtraceGetFPRegs(dst io.Writer) (int, error) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return 0, nil +} + +// PtraceSetFPRegs implements Context.PtraceSetFPRegs. +func (s *State) PtraceSetFPRegs(src io.Reader) (int, error) { + // TODO(gvisor.dev/issue/1238): floating-point is not supported. + return 0, nil +} + +// Register sets defined in include/uapi/linux/elf.h. +const ( + _NT_PRSTATUS = 1 + _NT_PRFPREG = 2 +) + +// PtraceGetRegSet implements Context.PtraceGetRegSet. +func (s *State) PtraceGetRegSet(regset uintptr, dst io.Writer, maxlen int) (int, error) { + switch regset { + case _NT_PRSTATUS: + if maxlen < ptraceRegsSize { + return 0, syserror.EFAULT + } + return s.PtraceGetRegs(dst) + default: + return 0, syserror.EINVAL + } +} + +// PtraceSetRegSet implements Context.PtraceSetRegSet. +func (s *State) PtraceSetRegSet(regset uintptr, src io.Reader, maxlen int) (int, error) { + switch regset { + case _NT_PRSTATUS: + if maxlen < ptraceRegsSize { + return 0, syserror.EFAULT + } + return s.PtraceSetRegs(src) + default: + return 0, syserror.EINVAL + } +} + +// FullRestore indicates whether a full restore is required. +func (s *State) FullRestore() bool { + return false +} + +// New returns a new architecture context. +func New(arch Arch, fs *cpuid.FeatureSet) Context { + switch arch { + case ARM64: + return &context64{ + State{ + FeatureSet: fs, + }, + } + } + panic(fmt.Sprintf("unknown architecture %v", arch)) +} diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go new file mode 100644 index 000000000..0d5b7d317 --- /dev/null +++ b/pkg/sentry/arch/arch_arm64.go @@ -0,0 +1,266 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "fmt" + "math/rand" + "syscall" + + "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/sentry/limits" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// Host specifies the host architecture. +const Host = ARM64 + +// These constants come directly from Linux. +const ( + // maxAddr64 is the maximum userspace address. It is TASK_SIZE in Linux + // for a 64-bit process. + maxAddr64 usermem.Addr = (1 << 48) + + // maxStackRand64 is the maximum randomization to apply to the stack. + // It is defined by arch/arm64/mm/mmap.c:(STACK_RND_MASK << PAGE_SHIFT) in Linux. + maxStackRand64 = 0x3ffff << 12 // 16 GB + + // maxMmapRand64 is the maximum randomization to apply to the mmap + // layout. It is defined by arch/arm64/mm/mmap.c:arch_mmap_rnd in Linux. + maxMmapRand64 = (1 << 33) * usermem.PageSize + + // minGap64 is the minimum gap to leave at the top of the address space + // for the stack. It is defined by arch/arm64/mm/mmap.c:MIN_GAP in Linux. + minGap64 = (128 << 20) + maxStackRand64 + + // preferredPIELoadAddr is the standard Linux position-independent + // executable base load address. It is ELF_ET_DYN_BASE in Linux. + // + // The Platform {Min,Max}UserAddress() may preclude loading at this + // address. See other preferredFoo comments below. + preferredPIELoadAddr usermem.Addr = maxAddr64 / 6 * 5 +) + +// These constants are selected as heuristics to help make the Platform's +// potentially limited address space conform as closely to Linux as possible. +const ( + preferredTopDownAllocMin usermem.Addr = 0x7e8000000000 + preferredAllocationGap = 128 << 30 // 128 GB + preferredTopDownBaseMin = preferredTopDownAllocMin + preferredAllocationGap + + // minMmapRand64 is the smallest we are willing to make the + // randomization to stay above preferredTopDownBaseMin. + minMmapRand64 = (1 << 18) * usermem.PageSize +) + +// context64 represents an ARM64 context. +type context64 struct { + State +} + +// Arch implements Context.Arch. +func (c *context64) Arch() Arch { + return ARM64 +} + +// Fork returns an exact copy of this context. +func (c *context64) Fork() Context { + return &context64{ + State: c.State.Fork(), + } +} + +// General purpose registers usage on Arm64: +// R0...R7: parameter/result registers. +// R8: indirect result location register. +// R9...R15: temporary rgisters. +// R16: the first intra-procedure-call scratch register. +// R17: the second intra-procedure-call scratch register. +// R18: the platform register. +// R19...R28: callee-saved registers. +// R29: the frame pointer. +// R30: the link register. + +// Return returns the current syscall return value. +func (c *context64) Return() uintptr { + return uintptr(c.Regs.Regs[0]) +} + +// SetReturn sets the syscall return value. +func (c *context64) SetReturn(value uintptr) { + c.Regs.Regs[0] = uint64(value) +} + +// IP returns the current instruction pointer. +func (c *context64) IP() uintptr { + return uintptr(c.Regs.Pc) +} + +// SetIP sets the current instruction pointer. +func (c *context64) SetIP(value uintptr) { + c.Regs.Pc = uint64(value) +} + +// Stack returns the current stack pointer. +func (c *context64) Stack() uintptr { + return uintptr(c.Regs.Sp) +} + +// SetStack sets the current stack pointer. +func (c *context64) SetStack(value uintptr) { + c.Regs.Sp = uint64(value) +} + +// TLS returns the current TLS pointer. +func (c *context64) TLS() uintptr { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // MRS_TPIDR_EL0 + return 0 +} + +// SetTLS sets the current TLS pointer. Returns false if value is invalid. +func (c *context64) SetTLS(value uintptr) bool { + // TODO(gvisor.dev/issue/1238): TLS is not supported. + // MSR_TPIDR_EL0 + return false +} + +// SetRSEQInterruptedIP implements Context.SetRSEQInterruptedIP. +func (c *context64) SetRSEQInterruptedIP(value uintptr) { + c.Regs.Regs[3] = uint64(value) +} + +// Native returns the native type for the given val. +func (c *context64) Native(val uintptr) interface{} { + v := uint64(val) + return &v +} + +// Value returns the generic val for the given native type. +func (c *context64) Value(val interface{}) uintptr { + return uintptr(*val.(*uint64)) +} + +// Width returns the byte width of this architecture. +func (c *context64) Width() uint { + return 8 +} + +// FeatureSet returns the FeatureSet in use. +func (c *context64) FeatureSet() *cpuid.FeatureSet { + return c.State.FeatureSet +} + +// mmapRand returns a random adjustment for randomizing an mmap layout. +func mmapRand(max uint64) usermem.Addr { + return usermem.Addr(rand.Int63n(int64(max))).RoundDown() +} + +// NewMmapLayout implements Context.NewMmapLayout consistently with Linux. +func (c *context64) NewMmapLayout(min, max usermem.Addr, r *limits.LimitSet) (MmapLayout, error) { + min, ok := min.RoundUp() + if !ok { + return MmapLayout{}, syscall.EINVAL + } + if max > maxAddr64 { + max = maxAddr64 + } + max = max.RoundDown() + + if min > max { + return MmapLayout{}, syscall.EINVAL + } + + stackSize := r.Get(limits.Stack) + + // MAX_GAP in Linux. + maxGap := (max / 6) * 5 + gap := usermem.Addr(stackSize.Cur) + if gap < minGap64 { + gap = minGap64 + } + if gap > maxGap { + gap = maxGap + } + defaultDir := MmapTopDown + if stackSize.Cur == limits.Infinity { + defaultDir = MmapBottomUp + } + + topDownMin := max - gap - maxMmapRand64 + maxRand := usermem.Addr(maxMmapRand64) + if topDownMin < preferredTopDownBaseMin { + // Try to keep TopDownBase above preferredTopDownBaseMin by + // shrinking maxRand. + maxAdjust := maxRand - minMmapRand64 + needAdjust := preferredTopDownBaseMin - topDownMin + if needAdjust <= maxAdjust { + maxRand -= needAdjust + } + } + + rnd := mmapRand(uint64(maxRand)) + l := MmapLayout{ + MinAddr: min, + MaxAddr: max, + // TASK_UNMAPPED_BASE in Linux. + BottomUpBase: (max/3 + rnd).RoundDown(), + TopDownBase: (max - gap - rnd).RoundDown(), + DefaultDirection: defaultDir, + // We may have reduced the maximum randomization to keep + // TopDownBase above preferredTopDownBaseMin while maintaining + // our stack gap. Stack allocations must use that max + // randomization to avoiding eating into the gap. + MaxStackRand: uint64(maxRand), + } + + // Final sanity check on the layout. + if !l.Valid() { + panic(fmt.Sprintf("Invalid MmapLayout: %+v", l)) + } + + return l, nil +} + +// PIELoadAddress implements Context.PIELoadAddress. +func (c *context64) PIELoadAddress(l MmapLayout) usermem.Addr { + base := preferredPIELoadAddr + max, ok := base.AddLength(maxMmapRand64) + if !ok { + panic(fmt.Sprintf("preferredPIELoadAddr %#x too large", base)) + } + + if max > l.MaxAddr { + // preferredPIELoadAddr won't fit; fall back to the standard + // Linux behavior of 2/3 of TopDownBase. TSAN won't like this. + // + // Don't bother trying to shrink the randomization for now. + base = l.TopDownBase / 3 * 2 + } + + return base + mmapRand(maxMmapRand64) +} + +// PtracePeekUser implements Context.PtracePeekUser. +func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { + // TODO(gvisor.dev/issue/1239): Full ptrace supporting for Arm64. + return c.Native(0), nil +} + +// PtracePokeUser implements Context.PtracePokeUser. +func (c *context64) PtracePokeUser(addr, data uintptr) error { + // TODO(gvisor.dev/issue/1239): Full ptrace supporting for Arm64. + return nil +} diff --git a/pkg/sentry/socket/rpcinet/device.go b/pkg/sentry/arch/arch_state_aarch64.go index 8cfd5f6e5..0136a85ad 100644 --- a/pkg/sentry/socket/rpcinet/device.go +++ b/pkg/sentry/arch/arch_state_aarch64.go @@ -1,4 +1,4 @@ -// Copyright 2018 The gVisor Authors. +// Copyright 2020 The gVisor Authors. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. @@ -12,8 +12,27 @@ // See the License for the specific language governing permissions and // limitations under the License. -package rpcinet +// +build arm64 -import "gvisor.dev/gvisor/pkg/sentry/device" +package arch -var socketDevice = device.NewAnonDevice() +import ( + "syscall" +) + +type syscallPtraceRegs struct { + Regs [31]uint64 + Sp uint64 + Pc uint64 + Pstate uint64 +} + +// saveRegs is invoked by stateify. +func (s *State) saveRegs() syscallPtraceRegs { + return syscallPtraceRegs(s.Regs) +} + +// loadRegs is invoked by stateify. +func (s *State) loadRegs(r syscallPtraceRegs) { + s.Regs = syscall.PtraceRegs(r) +} diff --git a/pkg/sentry/arch/arch_state_x86.go b/pkg/sentry/arch/arch_state_x86.go index 9061fcc86..84f11b0d1 100644 --- a/pkg/sentry/arch/arch_state_x86.go +++ b/pkg/sentry/arch/arch_state_x86.go @@ -12,6 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. +// +build amd64 i386 + package arch import ( diff --git a/pkg/sentry/arch/registers.proto b/pkg/sentry/arch/registers.proto index 9dc83e241..60c027aab 100644 --- a/pkg/sentry/arch/registers.proto +++ b/pkg/sentry/arch/registers.proto @@ -48,8 +48,45 @@ message AMD64Registers { uint64 gs_base = 27; } +message ARM64Registers { + uint64 r0 = 1; + uint64 r1 = 2; + uint64 r2 = 3; + uint64 r3 = 4; + uint64 r4 = 5; + uint64 r5 = 6; + uint64 r6 = 7; + uint64 r7 = 8; + uint64 r8 = 9; + uint64 r9 = 10; + uint64 r10 = 11; + uint64 r11 = 12; + uint64 r12 = 13; + uint64 r13 = 14; + uint64 r14 = 15; + uint64 r15 = 16; + uint64 r16 = 17; + uint64 r17 = 18; + uint64 r18 = 19; + uint64 r19 = 20; + uint64 r20 = 21; + uint64 r21 = 22; + uint64 r22 = 23; + uint64 r23 = 24; + uint64 r24 = 25; + uint64 r25 = 26; + uint64 r26 = 27; + uint64 r27 = 28; + uint64 r28 = 29; + uint64 r29 = 30; + uint64 r30 = 31; + uint64 sp = 32; + uint64 pc = 33; + uint64 pstate = 34; +} message Registers { oneof arch { AMD64Registers amd64 = 1; + ARM64Registers arm64 = 2; } } diff --git a/pkg/sentry/arch/signal.go b/pkg/sentry/arch/signal.go new file mode 100644 index 000000000..402e46025 --- /dev/null +++ b/pkg/sentry/arch/signal.go @@ -0,0 +1,250 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// SignalAct represents the action that should be taken when a signal is +// delivered, and is equivalent to struct sigaction. +// +// +stateify savable +type SignalAct struct { + Handler uint64 + Flags uint64 + Restorer uint64 // Only used on amd64. + Mask linux.SignalSet +} + +// SerializeFrom implements NativeSignalAct.SerializeFrom. +func (s *SignalAct) SerializeFrom(other *SignalAct) { + *s = *other +} + +// DeserializeTo implements NativeSignalAct.DeserializeTo. +func (s *SignalAct) DeserializeTo(other *SignalAct) { + *other = *s +} + +// SignalStack represents information about a user stack, and is equivalent to +// stack_t. +// +// +stateify savable +type SignalStack struct { + Addr uint64 + Flags uint32 + _ uint32 + Size uint64 +} + +// SerializeFrom implements NativeSignalStack.SerializeFrom. +func (s *SignalStack) SerializeFrom(other *SignalStack) { + *s = *other +} + +// DeserializeTo implements NativeSignalStack.DeserializeTo. +func (s *SignalStack) DeserializeTo(other *SignalStack) { + *other = *s +} + +// SignalInfo represents information about a signal being delivered, and is +// equivalent to struct siginfo in linux kernel(linux/include/uapi/asm-generic/siginfo.h). +// +// +stateify savable +type SignalInfo struct { + Signo int32 // Signal number + Errno int32 // Errno value + Code int32 // Signal code + _ uint32 + + // struct siginfo::_sifields is a union. In SignalInfo, fields in the union + // are accessed through methods. + // + // For reference, here is the definition of _sifields: (_sigfault._trapno, + // which does not exist on x86, omitted for clarity) + // + // union { + // int _pad[SI_PAD_SIZE]; + // + // /* kill() */ + // struct { + // __kernel_pid_t _pid; /* sender's pid */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // } _kill; + // + // /* POSIX.1b timers */ + // struct { + // __kernel_timer_t _tid; /* timer id */ + // int _overrun; /* overrun count */ + // char _pad[sizeof( __ARCH_SI_UID_T) - sizeof(int)]; + // sigval_t _sigval; /* same as below */ + // int _sys_private; /* not to be passed to user */ + // } _timer; + // + // /* POSIX.1b signals */ + // struct { + // __kernel_pid_t _pid; /* sender's pid */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // sigval_t _sigval; + // } _rt; + // + // /* SIGCHLD */ + // struct { + // __kernel_pid_t _pid; /* which child */ + // __ARCH_SI_UID_T _uid; /* sender's uid */ + // int _status; /* exit code */ + // __ARCH_SI_CLOCK_T _utime; + // __ARCH_SI_CLOCK_T _stime; + // } _sigchld; + // + // /* SIGILL, SIGFPE, SIGSEGV, SIGBUS */ + // struct { + // void *_addr; /* faulting insn/memory ref. */ + // short _addr_lsb; /* LSB of the reported address */ + // } _sigfault; + // + // /* SIGPOLL */ + // struct { + // __ARCH_SI_BAND_T _band; /* POLL_IN, POLL_OUT, POLL_MSG */ + // int _fd; + // } _sigpoll; + // + // /* SIGSYS */ + // struct { + // void *_call_addr; /* calling user insn */ + // int _syscall; /* triggering system call number */ + // unsigned int _arch; /* AUDIT_ARCH_* of syscall */ + // } _sigsys; + // } _sifields; + // + // _sifields is padded so that the size of siginfo is SI_MAX_SIZE = 128 + // bytes. + Fields [128 - 16]byte +} + +// FixSignalCodeForUser fixes up si_code. +// +// The si_code we get from Linux may contain the kernel-specific code in the +// top 16 bits if it's positive (e.g., from ptrace). Linux's +// copy_siginfo_to_user does +// err |= __put_user((short)from->si_code, &to->si_code); +// to mask out those bits and we need to do the same. +func (s *SignalInfo) FixSignalCodeForUser() { + if s.Code > 0 { + s.Code &= 0x0000ffff + } +} + +// Pid returns the si_pid field. +func (s *SignalInfo) Pid() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[0:4])) +} + +// SetPid mutates the si_pid field. +func (s *SignalInfo) SetPid(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) +} + +// Uid returns the si_uid field. +func (s *SignalInfo) Uid() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) +} + +// SetUid mutates the si_uid field. +func (s *SignalInfo) SetUid(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) +} + +// Sigval returns the sigval field, which is aliased to both si_int and si_ptr. +func (s *SignalInfo) Sigval() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[8:16]) +} + +// SetSigval mutates the sigval field. +func (s *SignalInfo) SetSigval(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[8:16], val) +} + +// TimerID returns the si_timerid field. +func (s *SignalInfo) TimerID() linux.TimerID { + return linux.TimerID(usermem.ByteOrder.Uint32(s.Fields[0:4])) +} + +// SetTimerID sets the si_timerid field. +func (s *SignalInfo) SetTimerID(val linux.TimerID) { + usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) +} + +// Overrun returns the si_overrun field. +func (s *SignalInfo) Overrun() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) +} + +// SetOverrun sets the si_overrun field. +func (s *SignalInfo) SetOverrun(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) +} + +// Addr returns the si_addr field. +func (s *SignalInfo) Addr() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[0:8]) +} + +// SetAddr sets the si_addr field. +func (s *SignalInfo) SetAddr(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[0:8], val) +} + +// Status returns the si_status field. +func (s *SignalInfo) Status() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) +} + +// SetStatus mutates the si_status field. +func (s *SignalInfo) SetStatus(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) +} + +// CallAddr returns the si_call_addr field. +func (s *SignalInfo) CallAddr() uint64 { + return usermem.ByteOrder.Uint64(s.Fields[0:8]) +} + +// SetCallAddr mutates the si_call_addr field. +func (s *SignalInfo) SetCallAddr(val uint64) { + usermem.ByteOrder.PutUint64(s.Fields[0:8], val) +} + +// Syscall returns the si_syscall field. +func (s *SignalInfo) Syscall() int32 { + return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) +} + +// SetSyscall mutates the si_syscall field. +func (s *SignalInfo) SetSyscall(val int32) { + usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) +} + +// Arch returns the si_arch field. +func (s *SignalInfo) Arch() uint32 { + return usermem.ByteOrder.Uint32(s.Fields[12:16]) +} + +// SetArch mutates the si_arch field. +func (s *SignalInfo) SetArch(val uint32) { + usermem.ByteOrder.PutUint32(s.Fields[12:16], val) +} diff --git a/pkg/sentry/arch/signal_amd64.go b/pkg/sentry/arch/signal_amd64.go index febd6f9b9..1e4f9c3c2 100644 --- a/pkg/sentry/arch/signal_amd64.go +++ b/pkg/sentry/arch/signal_amd64.go @@ -26,236 +26,6 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) -// SignalAct represents the action that should be taken when a signal is -// delivered, and is equivalent to struct sigaction on 64-bit x86. -// -// +stateify savable -type SignalAct struct { - Handler uint64 - Flags uint64 - Restorer uint64 - Mask linux.SignalSet -} - -// SerializeFrom implements NativeSignalAct.SerializeFrom. -func (s *SignalAct) SerializeFrom(other *SignalAct) { - *s = *other -} - -// DeserializeTo implements NativeSignalAct.DeserializeTo. -func (s *SignalAct) DeserializeTo(other *SignalAct) { - *other = *s -} - -// SignalStack represents information about a user stack, and is equivalent to -// stack_t on 64-bit x86. -// -// +stateify savable -type SignalStack struct { - Addr uint64 - Flags uint32 - _ uint32 - Size uint64 -} - -// SerializeFrom implements NativeSignalStack.SerializeFrom. -func (s *SignalStack) SerializeFrom(other *SignalStack) { - *s = *other -} - -// DeserializeTo implements NativeSignalStack.DeserializeTo. -func (s *SignalStack) DeserializeTo(other *SignalStack) { - *other = *s -} - -// SignalInfo represents information about a signal being delivered, and is -// equivalent to struct siginfo on 64-bit x86. -// -// +stateify savable -type SignalInfo struct { - Signo int32 // Signal number - Errno int32 // Errno value - Code int32 // Signal code - _ uint32 - - // struct siginfo::_sifields is a union. In SignalInfo, fields in the union - // are accessed through methods. - // - // For reference, here is the definition of _sifields: (_sigfault._trapno, - // which does not exist on x86, omitted for clarity) - // - // union { - // int _pad[SI_PAD_SIZE]; - // - // /* kill() */ - // struct { - // __kernel_pid_t _pid; /* sender's pid */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // } _kill; - // - // /* POSIX.1b timers */ - // struct { - // __kernel_timer_t _tid; /* timer id */ - // int _overrun; /* overrun count */ - // char _pad[sizeof( __ARCH_SI_UID_T) - sizeof(int)]; - // sigval_t _sigval; /* same as below */ - // int _sys_private; /* not to be passed to user */ - // } _timer; - // - // /* POSIX.1b signals */ - // struct { - // __kernel_pid_t _pid; /* sender's pid */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // sigval_t _sigval; - // } _rt; - // - // /* SIGCHLD */ - // struct { - // __kernel_pid_t _pid; /* which child */ - // __ARCH_SI_UID_T _uid; /* sender's uid */ - // int _status; /* exit code */ - // __ARCH_SI_CLOCK_T _utime; - // __ARCH_SI_CLOCK_T _stime; - // } _sigchld; - // - // /* SIGILL, SIGFPE, SIGSEGV, SIGBUS */ - // struct { - // void *_addr; /* faulting insn/memory ref. */ - // short _addr_lsb; /* LSB of the reported address */ - // } _sigfault; - // - // /* SIGPOLL */ - // struct { - // __ARCH_SI_BAND_T _band; /* POLL_IN, POLL_OUT, POLL_MSG */ - // int _fd; - // } _sigpoll; - // - // /* SIGSYS */ - // struct { - // void *_call_addr; /* calling user insn */ - // int _syscall; /* triggering system call number */ - // unsigned int _arch; /* AUDIT_ARCH_* of syscall */ - // } _sigsys; - // } _sifields; - // - // _sifields is padded so that the size of siginfo is SI_MAX_SIZE = 128 - // bytes. - Fields [128 - 16]byte -} - -// FixSignalCodeForUser fixes up si_code. -// -// The si_code we get from Linux may contain the kernel-specific code in the -// top 16 bits if it's positive (e.g., from ptrace). Linux's -// copy_siginfo_to_user does -// err |= __put_user((short)from->si_code, &to->si_code); -// to mask out those bits and we need to do the same. -func (s *SignalInfo) FixSignalCodeForUser() { - if s.Code > 0 { - s.Code &= 0x0000ffff - } -} - -// Pid returns the si_pid field. -func (s *SignalInfo) Pid() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[0:4])) -} - -// SetPid mutates the si_pid field. -func (s *SignalInfo) SetPid(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) -} - -// Uid returns the si_uid field. -func (s *SignalInfo) Uid() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) -} - -// SetUid mutates the si_uid field. -func (s *SignalInfo) SetUid(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) -} - -// Sigval returns the sigval field, which is aliased to both si_int and si_ptr. -func (s *SignalInfo) Sigval() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[8:16]) -} - -// SetSigval mutates the sigval field. -func (s *SignalInfo) SetSigval(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[8:16], val) -} - -// TimerID returns the si_timerid field. -func (s *SignalInfo) TimerID() linux.TimerID { - return linux.TimerID(usermem.ByteOrder.Uint32(s.Fields[0:4])) -} - -// SetTimerID sets the si_timerid field. -func (s *SignalInfo) SetTimerID(val linux.TimerID) { - usermem.ByteOrder.PutUint32(s.Fields[0:4], uint32(val)) -} - -// Overrun returns the si_overrun field. -func (s *SignalInfo) Overrun() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[4:8])) -} - -// SetOverrun sets the si_overrun field. -func (s *SignalInfo) SetOverrun(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[4:8], uint32(val)) -} - -// Addr returns the si_addr field. -func (s *SignalInfo) Addr() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[0:8]) -} - -// SetAddr sets the si_addr field. -func (s *SignalInfo) SetAddr(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[0:8], val) -} - -// Status returns the si_status field. -func (s *SignalInfo) Status() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) -} - -// SetStatus mutates the si_status field. -func (s *SignalInfo) SetStatus(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) -} - -// CallAddr returns the si_call_addr field. -func (s *SignalInfo) CallAddr() uint64 { - return usermem.ByteOrder.Uint64(s.Fields[0:8]) -} - -// SetCallAddr mutates the si_call_addr field. -func (s *SignalInfo) SetCallAddr(val uint64) { - usermem.ByteOrder.PutUint64(s.Fields[0:8], val) -} - -// Syscall returns the si_syscall field. -func (s *SignalInfo) Syscall() int32 { - return int32(usermem.ByteOrder.Uint32(s.Fields[8:12])) -} - -// SetSyscall mutates the si_syscall field. -func (s *SignalInfo) SetSyscall(val int32) { - usermem.ByteOrder.PutUint32(s.Fields[8:12], uint32(val)) -} - -// Arch returns the si_arch field. -func (s *SignalInfo) Arch() uint32 { - return usermem.ByteOrder.Uint32(s.Fields[12:16]) -} - -// SetArch mutates the si_arch field. -func (s *SignalInfo) SetArch(val uint32) { - usermem.ByteOrder.PutUint32(s.Fields[12:16], val) -} - // SignalContext64 is equivalent to struct sigcontext, the type passed as the // second argument to signal handlers set by signal(2). type SignalContext64 struct { diff --git a/pkg/sentry/arch/signal_arm64.go b/pkg/sentry/arch/signal_arm64.go new file mode 100644 index 000000000..7d0e98935 --- /dev/null +++ b/pkg/sentry/arch/signal_arm64.go @@ -0,0 +1,126 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package arch + +import ( + "encoding/binary" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// SignalContext64 is equivalent to struct sigcontext, the type passed as the +// second argument to signal handlers set by signal(2). +type SignalContext64 struct { + FaultAddr uint64 + Regs [31]uint64 + Sp uint64 + Pc uint64 + Pstate uint64 + _pad [8]byte // __attribute__((__aligned__(16))) + Reserved [4096]uint8 +} + +// UContext64 is equivalent to ucontext on arm64(arch/arm64/include/uapi/asm/ucontext.h). +type UContext64 struct { + Flags uint64 + Link *UContext64 + Stack SignalStack + Sigset linux.SignalSet + // glibc uses a 1024-bit sigset_t + _pad [(1024 - 64) / 8]byte + // sigcontext must be aligned to 16-byte + _pad2 [8]byte + // last for future expansion + MContext SignalContext64 +} + +// NewSignalAct implements Context.NewSignalAct. +func (c *context64) NewSignalAct() NativeSignalAct { + return &SignalAct{} +} + +// NewSignalStack implements Context.NewSignalStack. +func (c *context64) NewSignalStack() NativeSignalStack { + return &SignalStack{} +} + +// SignalSetup implements Context.SignalSetup. +func (c *context64) SignalSetup(st *Stack, act *SignalAct, info *SignalInfo, alt *SignalStack, sigset linux.SignalSet) error { + sp := st.Bottom + + if !(alt.IsEnabled() && sp == alt.Top()) { + sp -= 128 + } + + // Construct the UContext64 now since we need its size. + uc := &UContext64{ + Flags: 0, + Stack: *alt, + MContext: SignalContext64{ + Regs: c.Regs.Regs, + Sp: c.Regs.Sp, + Pc: c.Regs.Pc, + Pstate: c.Regs.Pstate, + }, + Sigset: sigset, + } + + ucSize := binary.Size(uc) + if ucSize < 0 { + panic("can't get size of UContext64") + } + // st.Arch.Width() is for the restorer address. sizeof(siginfo) == 128. + frameSize := int(st.Arch.Width()) + ucSize + 128 + frameBottom := (sp-usermem.Addr(frameSize)) & ^usermem.Addr(15) - 8 + sp = frameBottom + usermem.Addr(frameSize) + st.Bottom = sp + + // Prior to proceeding, figure out if the frame will exhaust the range + // for the signal stack. This is not allowed, and should immediately + // force signal delivery (reverting to the default handler). + if act.IsOnStack() && alt.IsEnabled() && !alt.Contains(frameBottom) { + return syscall.EFAULT + } + + // Adjust the code. + info.FixSignalCodeForUser() + + // Set up the stack frame. + infoAddr, err := st.Push(info) + if err != nil { + return err + } + ucAddr, err := st.Push(uc) + if err != nil { + return err + } + + // Set up registers. + c.Regs.Sp = uint64(st.Bottom) + c.Regs.Pc = act.Handler + c.Regs.Regs[0] = uint64(info.Signo) + c.Regs.Regs[1] = uint64(infoAddr) + c.Regs.Regs[2] = uint64(ucAddr) + + return nil +} + +// SignalRestore implements Context.SignalRestore. +// Only used on intel. +func (c *context64) SignalRestore(st *Stack, rt bool) (linux.SignalSet, SignalStack, error) { + return 0, SignalStack{}, nil +} diff --git a/pkg/sentry/arch/signal_stack.go b/pkg/sentry/arch/signal_stack.go index 5a3228113..d324da705 100644 --- a/pkg/sentry/arch/signal_stack.go +++ b/pkg/sentry/arch/signal_stack.go @@ -12,7 +12,7 @@ // See the License for the specific language governing permissions and // limitations under the License. -// +build i386 amd64 +// +build i386 amd64 arm64 package arch diff --git a/pkg/sentry/arch/syscalls_arm64.go b/pkg/sentry/arch/syscalls_arm64.go new file mode 100644 index 000000000..00d5ef461 --- /dev/null +++ b/pkg/sentry/arch/syscalls_arm64.go @@ -0,0 +1,62 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package arch + +const restartSyscallNr = uintptr(128) + +// SyscallNo returns the syscall number according to the 64-bit convention. +func (c *context64) SyscallNo() uintptr { + return uintptr(c.Regs.Regs[8]) +} + +// SyscallArgs provides syscall arguments according to the 64-bit convention. +// +// Due to the way addresses are mapped for the sentry this binary *must* be +// built in 64-bit mode. So we can just assume the syscall numbers that come +// back match the expected host system call numbers. +// General purpose registers usage on Arm64: +// R0...R7: parameter/result registers. +// R8: indirect result location register. +// R9...R15: temporary registers. +// R16: the first intra-procedure-call scratch register. +// R17: the second intra-procedure-call scratch register. +// R18: the platform register. +// R19...R28: callee-saved registers. +// R29: the frame pointer. +// R30: the link register. +func (c *context64) SyscallArgs() SyscallArguments { + return SyscallArguments{ + SyscallArgument{Value: uintptr(c.Regs.Regs[0])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[1])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[2])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[3])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[4])}, + SyscallArgument{Value: uintptr(c.Regs.Regs[5])}, + } +} + +// RestartSyscall implements Context.RestartSyscall. +func (c *context64) RestartSyscall() { + c.Regs.Pc -= SyscallWidth + c.Regs.Regs[8] = uint64(restartSyscallNr) +} + +// RestartSyscallWithRestartBlock implements Context.RestartSyscallWithRestartBlock. +func (c *context64) RestartSyscallWithRestartBlock() { + c.Regs.Pc -= SyscallWidth + c.Regs.Regs[8] = uint64(restartSyscallNr) +} diff --git a/pkg/sentry/fs/copy_up.go b/pkg/sentry/fs/copy_up.go index 734177e90..e03e3e417 100644 --- a/pkg/sentry/fs/copy_up.go +++ b/pkg/sentry/fs/copy_up.go @@ -18,6 +18,7 @@ import ( "fmt" "io" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -395,12 +396,12 @@ func copyContentsLocked(ctx context.Context, upper *Inode, lower *Inode, size in // Size and permissions are set on upper when the file content is copied // and when the file is created respectively. func copyAttributesLocked(ctx context.Context, upper *Inode, lower *Inode) error { - // Extract attributes fro the lower filesystem. + // Extract attributes from the lower filesystem. lowerAttr, err := lower.UnstableAttr(ctx) if err != nil { return err } - lowerXattr, err := lower.Listxattr() + lowerXattr, err := lower.ListXattr(ctx) if err != nil && err != syserror.EOPNOTSUPP { return err } @@ -421,11 +422,11 @@ func copyAttributesLocked(ctx context.Context, upper *Inode, lower *Inode) error if isXattrOverlay(name) { continue } - value, err := lower.Getxattr(name) + value, err := lower.GetXattr(ctx, name, linux.XATTR_SIZE_MAX) if err != nil { return err } - if err := upper.InodeOperations.Setxattr(upper, name, value); err != nil { + if err := upper.InodeOperations.SetXattr(ctx, upper, name, value, 0 /* flags */); err != nil { return err } } diff --git a/pkg/sentry/fs/file_overlay.go b/pkg/sentry/fs/file_overlay.go index 8a633b1ba..8991207b4 100644 --- a/pkg/sentry/fs/file_overlay.go +++ b/pkg/sentry/fs/file_overlay.go @@ -475,7 +475,7 @@ func readdirEntries(ctx context.Context, o *overlayEntry) (*SortedDentryMap, err // Skip this name if it is a negative entry in the // upper or there exists a whiteout for it. if o.upper != nil { - if overlayHasWhiteout(o.upper, name) { + if overlayHasWhiteout(ctx, o.upper, name) { continue } } diff --git a/pkg/sentry/fs/fsutil/inode.go b/pkg/sentry/fs/fsutil/inode.go index adf5ec69c..df7b74855 100644 --- a/pkg/sentry/fs/fsutil/inode.go +++ b/pkg/sentry/fs/fsutil/inode.go @@ -15,6 +15,7 @@ package fsutil import ( + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -202,7 +203,7 @@ func (i *InodeSimpleAttributes) NotifyModificationAndStatusChange(ctx context.Co } // InodeSimpleExtendedAttributes implements -// fs.InodeOperations.{Get,Set,List}xattr. +// fs.InodeOperations.{Get,Set,List}Xattr. // // +stateify savable type InodeSimpleExtendedAttributes struct { @@ -211,8 +212,8 @@ type InodeSimpleExtendedAttributes struct { xattrs map[string]string } -// Getxattr implements fs.InodeOperations.Getxattr. -func (i *InodeSimpleExtendedAttributes) Getxattr(_ *fs.Inode, name string) (string, error) { +// GetXattr implements fs.InodeOperations.GetXattr. +func (i *InodeSimpleExtendedAttributes) GetXattr(_ context.Context, _ *fs.Inode, name string, _ uint64) (string, error) { i.mu.RLock() value, ok := i.xattrs[name] i.mu.RUnlock() @@ -222,19 +223,31 @@ func (i *InodeSimpleExtendedAttributes) Getxattr(_ *fs.Inode, name string) (stri return value, nil } -// Setxattr implements fs.InodeOperations.Setxattr. -func (i *InodeSimpleExtendedAttributes) Setxattr(_ *fs.Inode, name, value string) error { +// SetXattr implements fs.InodeOperations.SetXattr. +func (i *InodeSimpleExtendedAttributes) SetXattr(_ context.Context, _ *fs.Inode, name, value string, flags uint32) error { i.mu.Lock() + defer i.mu.Unlock() if i.xattrs == nil { + if flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } i.xattrs = make(map[string]string) } + + _, ok := i.xattrs[name] + if ok && flags&linux.XATTR_CREATE != 0 { + return syserror.EEXIST + } + if !ok && flags&linux.XATTR_REPLACE != 0 { + return syserror.ENODATA + } + i.xattrs[name] = value - i.mu.Unlock() return nil } -// Listxattr implements fs.InodeOperations.Listxattr. -func (i *InodeSimpleExtendedAttributes) Listxattr(_ *fs.Inode) (map[string]struct{}, error) { +// ListXattr implements fs.InodeOperations.ListXattr. +func (i *InodeSimpleExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { i.mu.RLock() names := make(map[string]struct{}, len(i.xattrs)) for name := range i.xattrs { @@ -436,18 +449,18 @@ func (InodeNotSymlink) Getlink(context.Context, *fs.Inode) (*fs.Dirent, error) { // extended attributes. type InodeNoExtendedAttributes struct{} -// Getxattr implements fs.InodeOperations.Getxattr. -func (InodeNoExtendedAttributes) Getxattr(*fs.Inode, string) (string, error) { +// GetXattr implements fs.InodeOperations.GetXattr. +func (InodeNoExtendedAttributes) GetXattr(context.Context, *fs.Inode, string, uint64) (string, error) { return "", syserror.EOPNOTSUPP } -// Setxattr implements fs.InodeOperations.Setxattr. -func (InodeNoExtendedAttributes) Setxattr(*fs.Inode, string, string) error { +// SetXattr implements fs.InodeOperations.SetXattr. +func (InodeNoExtendedAttributes) SetXattr(context.Context, *fs.Inode, string, string, uint32) error { return syserror.EOPNOTSUPP } -// Listxattr implements fs.InodeOperations.Listxattr. -func (InodeNoExtendedAttributes) Listxattr(*fs.Inode) (map[string]struct{}, error) { +// ListXattr implements fs.InodeOperations.ListXattr. +func (InodeNoExtendedAttributes) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { return nil, syserror.EOPNOTSUPP } diff --git a/pkg/sentry/fs/gofer/context_file.go b/pkg/sentry/fs/gofer/context_file.go index 44b72582a..2125dafef 100644 --- a/pkg/sentry/fs/gofer/context_file.go +++ b/pkg/sentry/fs/gofer/context_file.go @@ -59,6 +59,20 @@ func (c *contextFile) setAttr(ctx context.Context, valid p9.SetAttrMask, attr p9 return err } +func (c *contextFile) getXattr(ctx context.Context, name string, size uint64) (string, error) { + ctx.UninterruptibleSleepStart(false) + val, err := c.file.GetXattr(name, size) + ctx.UninterruptibleSleepFinish(false) + return val, err +} + +func (c *contextFile) setXattr(ctx context.Context, name, value string, flags uint32) error { + ctx.UninterruptibleSleepStart(false) + err := c.file.SetXattr(name, value, flags) + ctx.UninterruptibleSleepFinish(false) + return err +} + func (c *contextFile) allocate(ctx context.Context, mode p9.AllocateMode, offset, length uint64) error { ctx.UninterruptibleSleepStart(false) err := c.file.Allocate(mode, offset, length) diff --git a/pkg/sentry/fs/gofer/inode.go b/pkg/sentry/fs/gofer/inode.go index 245fe2ef1..98d1a8a48 100644 --- a/pkg/sentry/fs/gofer/inode.go +++ b/pkg/sentry/fs/gofer/inode.go @@ -38,8 +38,7 @@ import ( // // +stateify savable type inodeOperations struct { - fsutil.InodeNotVirtual `state:"nosave"` - fsutil.InodeNoExtendedAttributes `state:"nosave"` + fsutil.InodeNotVirtual `state:"nosave"` // fileState implements fs.CachedFileObject. It exists // to break a circular load dependency between inodeOperations @@ -604,6 +603,21 @@ func (i *inodeOperations) Truncate(ctx context.Context, inode *fs.Inode, length return i.fileState.file.setAttr(ctx, p9.SetAttrMask{Size: true}, p9.SetAttr{Size: uint64(length)}) } +// GetXattr implements fs.InodeOperations.GetXattr. +func (i *inodeOperations) GetXattr(ctx context.Context, inode *fs.Inode, name string, size uint64) (string, error) { + return i.fileState.file.getXattr(ctx, name, size) +} + +// SetXattr implements fs.InodeOperations.SetXattr. +func (i *inodeOperations) SetXattr(ctx context.Context, inode *fs.Inode, name string, value string, flags uint32) error { + return i.fileState.file.setXattr(ctx, name, value, flags) +} + +// ListXattr implements fs.InodeOperations.ListXattr. +func (i *inodeOperations) ListXattr(context.Context, *fs.Inode) (map[string]struct{}, error) { + return nil, syscall.EOPNOTSUPP +} + // Allocate implements fs.InodeOperations.Allocate. func (i *inodeOperations) Allocate(ctx context.Context, inode *fs.Inode, offset, length int64) error { // This can only be called for files anyway. diff --git a/pkg/sentry/fs/inode.go b/pkg/sentry/fs/inode.go index 468043df0..e4cf5a570 100644 --- a/pkg/sentry/fs/inode.go +++ b/pkg/sentry/fs/inode.go @@ -261,28 +261,28 @@ func (i *Inode) UnstableAttr(ctx context.Context) (UnstableAttr, error) { return i.InodeOperations.UnstableAttr(ctx, i) } -// Getxattr calls i.InodeOperations.Getxattr with i as the Inode. -func (i *Inode) Getxattr(name string) (string, error) { +// GetXattr calls i.InodeOperations.GetXattr with i as the Inode. +func (i *Inode) GetXattr(ctx context.Context, name string, size uint64) (string, error) { if i.overlay != nil { - return overlayGetxattr(i.overlay, name) + return overlayGetXattr(ctx, i.overlay, name, size) } - return i.InodeOperations.Getxattr(i, name) + return i.InodeOperations.GetXattr(ctx, i, name, size) } -// Setxattr calls i.InodeOperations.Setxattr with i as the Inode. -func (i *Inode) Setxattr(name, value string) error { +// SetXattr calls i.InodeOperations.SetXattr with i as the Inode. +func (i *Inode) SetXattr(ctx context.Context, d *Dirent, name, value string, flags uint32) error { if i.overlay != nil { - return overlaySetxattr(i.overlay, name, value) + return overlaySetxattr(ctx, i.overlay, d, name, value, flags) } - return i.InodeOperations.Setxattr(i, name, value) + return i.InodeOperations.SetXattr(ctx, i, name, value, flags) } -// Listxattr calls i.InodeOperations.Listxattr with i as the Inode. -func (i *Inode) Listxattr() (map[string]struct{}, error) { +// ListXattr calls i.InodeOperations.ListXattr with i as the Inode. +func (i *Inode) ListXattr(ctx context.Context) (map[string]struct{}, error) { if i.overlay != nil { - return overlayListxattr(i.overlay) + return overlayListXattr(ctx, i.overlay) } - return i.InodeOperations.Listxattr(i) + return i.InodeOperations.ListXattr(ctx, i) } // CheckPermission will check if the caller may access this file in the diff --git a/pkg/sentry/fs/inode_operations.go b/pkg/sentry/fs/inode_operations.go index 5cde9d215..13261cb81 100644 --- a/pkg/sentry/fs/inode_operations.go +++ b/pkg/sentry/fs/inode_operations.go @@ -170,20 +170,27 @@ type InodeOperations interface { // file system events. UnstableAttr(ctx context.Context, inode *Inode) (UnstableAttr, error) - // Getxattr retrieves the value of extended attribute name. Inodes that - // do not support extended attributes return EOPNOTSUPP. Inodes that - // support extended attributes but don't have a value at name return + // GetXattr retrieves the value of extended attribute specified by name. + // Inodes that do not support extended attributes return EOPNOTSUPP. Inodes + // that support extended attributes but don't have a value at name return // ENODATA. - Getxattr(inode *Inode, name string) (string, error) + // + // If this is called through the getxattr(2) syscall, size indicates the + // size of the buffer that the application has allocated to hold the + // attribute value. If the value is larger than size, implementations may + // return ERANGE to indicate that the buffer is too small, but they are also + // free to ignore the hint entirely (i.e. the value returned may be larger + // than size). All size checking is done independently at the syscall layer. + GetXattr(ctx context.Context, inode *Inode, name string, size uint64) (string, error) - // Setxattr sets the value of extended attribute name. Inodes that - // do not support extended attributes return EOPNOTSUPP. - Setxattr(inode *Inode, name, value string) error + // SetXattr sets the value of extended attribute specified by name. Inodes + // that do not support extended attributes return EOPNOTSUPP. + SetXattr(ctx context.Context, inode *Inode, name, value string, flags uint32) error - // Listxattr returns the set of all extended attributes names that + // ListXattr returns the set of all extended attributes names that // have values. Inodes that do not support extended attributes return // EOPNOTSUPP. - Listxattr(inode *Inode) (map[string]struct{}, error) + ListXattr(ctx context.Context, inode *Inode) (map[string]struct{}, error) // Check determines whether an Inode can be accessed with the // requested permission mask using the context (which gives access diff --git a/pkg/sentry/fs/inode_overlay.go b/pkg/sentry/fs/inode_overlay.go index 13d11e001..c477de837 100644 --- a/pkg/sentry/fs/inode_overlay.go +++ b/pkg/sentry/fs/inode_overlay.go @@ -25,13 +25,13 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -func overlayHasWhiteout(parent *Inode, name string) bool { - s, err := parent.Getxattr(XattrOverlayWhiteout(name)) +func overlayHasWhiteout(ctx context.Context, parent *Inode, name string) bool { + s, err := parent.GetXattr(ctx, XattrOverlayWhiteout(name), 1) return err == nil && s == "y" } -func overlayCreateWhiteout(parent *Inode, name string) error { - return parent.InodeOperations.Setxattr(parent, XattrOverlayWhiteout(name), "y") +func overlayCreateWhiteout(ctx context.Context, parent *Inode, name string) error { + return parent.InodeOperations.SetXattr(ctx, parent, XattrOverlayWhiteout(name), "y", 0 /* flags */) } func overlayWriteOut(ctx context.Context, o *overlayEntry) error { @@ -89,7 +89,7 @@ func overlayLookup(ctx context.Context, parent *overlayEntry, inode *Inode, name } // Are we done? - if overlayHasWhiteout(parent.upper, name) { + if overlayHasWhiteout(ctx, parent.upper, name) { if upperInode == nil { parent.copyMu.RUnlock() if negativeUpperChild { @@ -345,7 +345,7 @@ func overlayRemove(ctx context.Context, o *overlayEntry, parent *Dirent, child * } } if child.Inode.overlay.lowerExists { - if err := overlayCreateWhiteout(o.upper, child.name); err != nil { + if err := overlayCreateWhiteout(ctx, o.upper, child.name); err != nil { return err } } @@ -426,7 +426,7 @@ func overlayRename(ctx context.Context, o *overlayEntry, oldParent *Dirent, rena return err } if renamed.Inode.overlay.lowerExists { - if err := overlayCreateWhiteout(oldParent.Inode.overlay.upper, oldName); err != nil { + if err := overlayCreateWhiteout(ctx, oldParent.Inode.overlay.upper, oldName); err != nil { return err } } @@ -528,7 +528,7 @@ func overlayUnstableAttr(ctx context.Context, o *overlayEntry) (UnstableAttr, er return attr, err } -func overlayGetxattr(o *overlayEntry, name string) (string, error) { +func overlayGetXattr(ctx context.Context, o *overlayEntry, name string, size uint64) (string, error) { // Hot path. This is how the overlay checks for whiteout files. // Avoid defers. var ( @@ -544,31 +544,38 @@ func overlayGetxattr(o *overlayEntry, name string) (string, error) { o.copyMu.RLock() if o.upper != nil { - s, err = o.upper.Getxattr(name) + s, err = o.upper.GetXattr(ctx, name, size) } else { - s, err = o.lower.Getxattr(name) + s, err = o.lower.GetXattr(ctx, name, size) } o.copyMu.RUnlock() return s, err } -// TODO(b/146028302): Support setxattr for overlayfs. -func overlaySetxattr(o *overlayEntry, name, value string) error { - return syserror.EOPNOTSUPP +func overlaySetxattr(ctx context.Context, o *overlayEntry, d *Dirent, name, value string, flags uint32) error { + // Don't allow changes to overlay xattrs through a setxattr syscall. + if strings.HasPrefix(XattrOverlayPrefix, name) { + return syserror.EPERM + } + + if err := copyUp(ctx, d); err != nil { + return err + } + return o.upper.SetXattr(ctx, d, name, value, flags) } -func overlayListxattr(o *overlayEntry) (map[string]struct{}, error) { +func overlayListXattr(ctx context.Context, o *overlayEntry) (map[string]struct{}, error) { o.copyMu.RLock() defer o.copyMu.RUnlock() var names map[string]struct{} var err error if o.upper != nil { - names, err = o.upper.Listxattr() + names, err = o.upper.ListXattr(ctx) } else { - names, err = o.lower.Listxattr() + names, err = o.lower.ListXattr(ctx) } for name := range names { - // Same as overlayGetxattr, we shouldn't forward along + // Same as overlayGetXattr, we shouldn't forward along // overlay attributes. if strings.HasPrefix(XattrOverlayPrefix, name) { delete(names, name) diff --git a/pkg/sentry/fs/inode_overlay_test.go b/pkg/sentry/fs/inode_overlay_test.go index 8935aad65..493d98c36 100644 --- a/pkg/sentry/fs/inode_overlay_test.go +++ b/pkg/sentry/fs/inode_overlay_test.go @@ -382,8 +382,8 @@ type dir struct { ReaddirCalled bool } -// Getxattr implements InodeOperations.Getxattr. -func (d *dir) Getxattr(inode *fs.Inode, name string) (string, error) { +// GetXattr implements InodeOperations.GetXattr. +func (d *dir) GetXattr(_ context.Context, _ *fs.Inode, name string, _ uint64) (string, error) { for _, n := range d.negative { if name == fs.XattrOverlayWhiteout(n) { return "y", nil diff --git a/pkg/sentry/fs/lock/lock.go b/pkg/sentry/fs/lock/lock.go index 41b040818..926538d90 100644 --- a/pkg/sentry/fs/lock/lock.go +++ b/pkg/sentry/fs/lock/lock.go @@ -78,6 +78,9 @@ const ( ) // LockEOF is the maximal possible end of a regional file lock. +// +// A BSD-style full file lock can be represented as a regional file lock from +// offset 0 to LockEOF. const LockEOF = math.MaxUint64 // Lock is a regional file lock. It consists of either a single writer diff --git a/pkg/sentry/fs/mounts.go b/pkg/sentry/fs/mounts.go index db3dfd096..a9627a9d1 100644 --- a/pkg/sentry/fs/mounts.go +++ b/pkg/sentry/fs/mounts.go @@ -609,8 +609,11 @@ func (mns *MountNamespace) resolve(ctx context.Context, root, node *Dirent, rema } // Find the node; we resolve relative to the current symlink's parent. + renameMu.RLock() + parent := node.parent + renameMu.RUnlock() *remainingTraversals-- - d, err := mns.FindInode(ctx, root, node.parent, targetPath, remainingTraversals) + d, err := mns.FindInode(ctx, root, parent, targetPath, remainingTraversals) if err != nil { return nil, err } diff --git a/pkg/sentry/fs/proc/BUILD b/pkg/sentry/fs/proc/BUILD index 94d46ab1b..cb37c6c6b 100644 --- a/pkg/sentry/fs/proc/BUILD +++ b/pkg/sentry/fs/proc/BUILD @@ -18,7 +18,6 @@ go_library( "mounts.go", "net.go", "proc.go", - "rpcinet_proc.go", "stat.go", "sys.go", "sys_net.go", @@ -46,7 +45,6 @@ go_library( "//pkg/sentry/limits", "//pkg/sentry/mm", "//pkg/sentry/socket", - "//pkg/sentry/socket/rpcinet", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usage", diff --git a/pkg/sentry/fs/proc/cgroup.go b/pkg/sentry/fs/proc/cgroup.go index 05e31c55d..c4abe319d 100644 --- a/pkg/sentry/fs/proc/cgroup.go +++ b/pkg/sentry/fs/proc/cgroup.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" ) +// LINT.IfChange + func newCGroupInode(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string]string) *fs.Inode { // From man 7 cgroups: "For each cgroup hierarchy of which the process // is a member, there is one entry containing three colon-separated @@ -39,3 +41,5 @@ func newCGroupInode(ctx context.Context, msrc *fs.MountSource, cgroupControllers return newStaticProcInode(ctx, msrc, []byte(data)) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/cpuinfo.go b/pkg/sentry/fs/proc/cpuinfo.go index 3edf36780..df0c4e3a7 100644 --- a/pkg/sentry/fs/proc/cpuinfo.go +++ b/pkg/sentry/fs/proc/cpuinfo.go @@ -15,11 +15,15 @@ package proc import ( + "bytes" + "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { k := kernel.KernelFromContext(ctx) features := k.FeatureSet() @@ -27,9 +31,11 @@ func newCPUInfo(ctx context.Context, msrc *fs.MountSource) *fs.Inode { // Kernel is always initialized with a FeatureSet. panic("cpuinfo read with nil FeatureSet") } - contents := make([]byte, 0, 1024) + var buf bytes.Buffer for i, max := uint(0), k.ApplicationCores(); i < max; i++ { - contents = append(contents, []byte(features.CPUInfo(i))...) + features.WriteCPUInfoTo(i, &buf) } - return newStaticProcInode(ctx, msrc, contents) + return newStaticProcInode(ctx, msrc, buf.Bytes()) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/exec_args.go b/pkg/sentry/fs/proc/exec_args.go index 1d3a2d426..9aaeb780b 100644 --- a/pkg/sentry/fs/proc/exec_args.go +++ b/pkg/sentry/fs/proc/exec_args.go @@ -29,6 +29,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // execArgType enumerates the types of exec arguments that are exposed through // proc. type execArgType int @@ -201,3 +203,5 @@ func (f *execArgFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequen } return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/task.go) diff --git a/pkg/sentry/fs/proc/fds.go b/pkg/sentry/fs/proc/fds.go index bee421d76..2fa3cfa7d 100644 --- a/pkg/sentry/fs/proc/fds.go +++ b/pkg/sentry/fs/proc/fds.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // walkDescriptors finds the descriptor (file-flag pair) for the fd identified // by p, and calls the toInodeOperations callback with that descriptor. This is a helper // method for implementing fs.InodeOperations.Lookup. @@ -277,3 +279,5 @@ func (fdid *fdInfoDir) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs. } return fs.NewFile(ctx, dirent, flags, fops), nil } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/filesystems.go b/pkg/sentry/fs/proc/filesystems.go index e9250c51c..7b3b974ab 100644 --- a/pkg/sentry/fs/proc/filesystems.go +++ b/pkg/sentry/fs/proc/filesystems.go @@ -23,6 +23,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) +// LINT.IfChange + // filesystemsData backs /proc/filesystems. // // +stateify savable @@ -59,3 +61,5 @@ func (*filesystemsData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle // Return the SeqData and advance the generation counter. return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*filesystemsData)(nil)}}, 1 } + +// LINT.ThenChange(../../fsimpl/proc/filesystem.go) diff --git a/pkg/sentry/fs/proc/fs.go b/pkg/sentry/fs/proc/fs.go index f14833805..761d24462 100644 --- a/pkg/sentry/fs/proc/fs.go +++ b/pkg/sentry/fs/proc/fs.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" ) +// LINT.IfChange + // filesystem is a procfs. // // +stateify savable @@ -79,3 +81,5 @@ func (f *filesystem) Mount(ctx context.Context, device string, flags fs.MountSou // never want them cached. return New(ctx, fs.NewNonCachingMountSource(ctx, f, flags), cgroups) } + +// LINT.ThenChange(../../fsimpl/proc/filesystem.go) diff --git a/pkg/sentry/fs/proc/inode.go b/pkg/sentry/fs/proc/inode.go index 0c04f81fa..723f6b661 100644 --- a/pkg/sentry/fs/proc/inode.go +++ b/pkg/sentry/fs/proc/inode.go @@ -26,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) +// LINT.IfChange + // taskOwnedInodeOps wraps an fs.InodeOperations and overrides the UnstableAttr // method to return either the task or root as the owner, depending on the // task's dumpability. @@ -131,3 +133,5 @@ func newProcInode(ctx context.Context, iops fs.InodeOperations, msrc *fs.MountSo } return fs.NewInode(ctx, iops, msrc, sattr) } + +// LINT.ThenChange(../../fsimpl/proc/tasks.go) diff --git a/pkg/sentry/fs/proc/loadavg.go b/pkg/sentry/fs/proc/loadavg.go index 8602b7426..d7d2afcb7 100644 --- a/pkg/sentry/fs/proc/loadavg.go +++ b/pkg/sentry/fs/proc/loadavg.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" ) +// LINT.IfChange + // loadavgData backs /proc/loadavg. // // +stateify savable @@ -53,3 +55,5 @@ func (d *loadavgData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/meminfo.go b/pkg/sentry/fs/proc/meminfo.go index 495f3e3ba..313c6a32b 100644 --- a/pkg/sentry/fs/proc/meminfo.go +++ b/pkg/sentry/fs/proc/meminfo.go @@ -25,6 +25,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/usermem" ) +// LINT.IfChange + // meminfoData backs /proc/meminfo. // // +stateify savable @@ -83,3 +85,5 @@ func (d *meminfoData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) fmt.Fprintf(&buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*meminfoData)(nil)}}, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/mounts.go b/pkg/sentry/fs/proc/mounts.go index e33c4a460..d4efc86e0 100644 --- a/pkg/sentry/fs/proc/mounts.go +++ b/pkg/sentry/fs/proc/mounts.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "sort" + "strings" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -25,6 +26,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // forEachMountSource runs f for the process root mount and each mount that is a // descendant of the root. func forEachMount(t *kernel.Task, fn func(string, *fs.Mount)) { @@ -144,14 +147,35 @@ func (mif *mountInfoFile) ReadSeqFileData(ctx context.Context, handle seqfile.Se // (10) Mount source: filesystem-specific information or "none". fmt.Fprintf(&buf, "none ") - // (11) Superblock options. Only "ro/rw" is supported for now, - // and is the same as the filesystem option. - fmt.Fprintf(&buf, "%s\n", opts) + // (11) Superblock options, and final newline. + fmt.Fprintf(&buf, "%s\n", superBlockOpts(mountPath, mroot.Inode.MountSource)) }) return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*mountInfoFile)(nil)}}, 0 } +func superBlockOpts(mountPath string, msrc *fs.MountSource) string { + // gVisor doesn't (yet) have a concept of super block options, so we + // use the ro/rw bit from the mount flag. + opts := "rw" + if msrc.Flags.ReadOnly { + opts = "ro" + } + + // NOTE(b/147673608): If the mount is a cgroup, we also need to include + // the cgroup name in the options. For now we just read that from the + // path. + // TODO(gvisor.dev/issues/190): Once gVisor has full cgroup support, we + // should get this value from the cgroup itself, and not rely on the + // path. + if msrc.FilesystemType == "cgroup" { + splitPath := strings.Split(mountPath, "/") + cgroupType := splitPath[len(splitPath)-1] + opts += "," + cgroupType + } + return opts +} + // mountsFile is used to implement /proc/[pid]/mounts. // // +stateify savable @@ -195,3 +219,5 @@ func (mf *mountsFile) ReadSeqFileData(ctx context.Context, handle seqfile.SeqHan return []seqfile.SeqData{{Buf: buf.Bytes(), Handle: (*mountsFile)(nil)}}, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/net.go b/pkg/sentry/fs/proc/net.go index 402919924..3f17e98ea 100644 --- a/pkg/sentry/fs/proc/net.go +++ b/pkg/sentry/fs/proc/net.go @@ -38,6 +38,8 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) +// LINT.IfChange + // newNet creates a new proc net entry. func (p *proc) newNetDir(ctx context.Context, k *kernel.Kernel, msrc *fs.MountSource) *fs.Inode { var contents map[string]*fs.Inode @@ -831,3 +833,5 @@ func (n *netUDP) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([]se } return data, 0 } + +// LINT.ThenChange(../../fsimpl/proc/tasks_net.go) diff --git a/pkg/sentry/fs/proc/proc.go b/pkg/sentry/fs/proc/proc.go index 56e92721e..29867dc3a 100644 --- a/pkg/sentry/fs/proc/proc.go +++ b/pkg/sentry/fs/proc/proc.go @@ -27,10 +27,11 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" "gvisor.dev/gvisor/pkg/syserror" ) +// LINT.IfChange + // proc is a root proc node. // // +stateify savable @@ -85,15 +86,9 @@ func New(ctx context.Context, msrc *fs.MountSource, cgroupControllers map[string } // Add more contents that need proc to be initialized. + p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc)) p.AddChild(ctx, "sys", p.newSysDir(ctx, msrc)) - // If we're using rpcinet we will let it manage /proc/net. - if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - p.AddChild(ctx, "net", newRPCInetProcNet(ctx, msrc)) - } else { - p.AddChild(ctx, "net", p.newNetDir(ctx, k, msrc)) - } - return newProcInode(ctx, p, msrc, fs.SpecialDirectory, nil), nil } @@ -249,3 +244,5 @@ func (rpf *rootProcFile) Readdir(ctx context.Context, file *fs.File, ser fs.Dent } return offset, nil } + +// LINT.ThenChange(../../fsimpl/proc/tasks.go) diff --git a/pkg/sentry/fs/proc/rpcinet_proc.go b/pkg/sentry/fs/proc/rpcinet_proc.go deleted file mode 100644 index 01ac97530..000000000 --- a/pkg/sentry/fs/proc/rpcinet_proc.go +++ /dev/null @@ -1,217 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "io" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/waiter" -) - -// rpcInetInode implements fs.InodeOperations. -type rpcInetInode struct { - fsutil.SimpleFileInode - - // filepath is the full path of this rpcInetInode. - filepath string - - k *kernel.Kernel -} - -func newRPCInetInode(ctx context.Context, msrc *fs.MountSource, filepath string, mode linux.FileMode) *fs.Inode { - f := &rpcInetInode{ - SimpleFileInode: *fsutil.NewSimpleFileInode(ctx, fs.RootOwner, fs.FilePermsFromMode(mode), linux.PROC_SUPER_MAGIC), - filepath: filepath, - k: kernel.KernelFromContext(ctx), - } - return newProcInode(ctx, f, msrc, fs.SpecialFile, nil) -} - -// GetFile implements fs.InodeOperations.GetFile. -func (i *rpcInetInode) GetFile(ctx context.Context, dirent *fs.Dirent, flags fs.FileFlags) (*fs.File, error) { - flags.Pread = true - flags.Pwrite = true - fops := &rpcInetFile{ - inode: i, - } - return fs.NewFile(ctx, dirent, flags, fops), nil -} - -// rpcInetFile implements fs.FileOperations as RPCs. -type rpcInetFile struct { - fsutil.FileGenericSeek `state:"nosave"` - fsutil.FileNoIoctl `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileNoopFsync `state:"nosave"` - fsutil.FileNoopRelease `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - waiter.AlwaysReady `state:"nosave"` - - inode *rpcInetInode -} - -// Read implements fs.FileOperations.Read. -// -// This method can panic if an rpcInetInode was created without an rpcinet -// stack. -func (f *rpcInetFile) Read(ctx context.Context, file *fs.File, dst usermem.IOSequence, offset int64) (int64, error) { - if offset < 0 { - return 0, syserror.EINVAL - } - s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) - if !ok { - panic("Network stack is not a rpcinet.") - } - - contents, se := s.RPCReadFile(f.inode.filepath) - if se != nil || offset >= int64(len(contents)) { - return 0, io.EOF - } - - n, err := dst.CopyOut(ctx, contents[offset:]) - return int64(n), err -} - -// Write implements fs.FileOperations.Write. -// -// This method can panic if an rpcInetInode was created without an rpcInet -// stack. -func (f *rpcInetFile) Write(ctx context.Context, file *fs.File, src usermem.IOSequence, offset int64) (int64, error) { - s, ok := f.inode.k.NetworkStack().(*rpcinet.Stack) - if !ok { - panic("Network stack is not a rpcinet.") - } - - if src.NumBytes() == 0 { - return 0, nil - } - - b := make([]byte, src.NumBytes(), src.NumBytes()) - n, err := src.CopyIn(ctx, b) - if err != nil { - return int64(n), err - } - - written, se := s.RPCWriteFile(f.inode.filepath, b) - return int64(written), se.ToError() -} - -// newRPCInetProcNet will build an inode for /proc/net. -func newRPCInetProcNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "arp": newRPCInetInode(ctx, msrc, "/proc/net/arp", 0444), - "dev": newRPCInetInode(ctx, msrc, "/proc/net/dev", 0444), - "if_inet6": newRPCInetInode(ctx, msrc, "/proc/net/if_inet6", 0444), - "ipv6_route": newRPCInetInode(ctx, msrc, "/proc/net/ipv6_route", 0444), - "netlink": newRPCInetInode(ctx, msrc, "/proc/net/netlink", 0444), - "netstat": newRPCInetInode(ctx, msrc, "/proc/net/netstat", 0444), - "packet": newRPCInetInode(ctx, msrc, "/proc/net/packet", 0444), - "protocols": newRPCInetInode(ctx, msrc, "/proc/net/protocols", 0444), - "psched": newRPCInetInode(ctx, msrc, "/proc/net/psched", 0444), - "ptype": newRPCInetInode(ctx, msrc, "/proc/net/ptype", 0444), - "route": newRPCInetInode(ctx, msrc, "/proc/net/route", 0444), - "tcp": newRPCInetInode(ctx, msrc, "/proc/net/tcp", 0444), - "tcp6": newRPCInetInode(ctx, msrc, "/proc/net/tcp6", 0444), - "udp": newRPCInetInode(ctx, msrc, "/proc/net/udp", 0444), - "udp6": newRPCInetInode(ctx, msrc, "/proc/net/udp6", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetProcSysNet will build an inode for /proc/sys/net. -func newRPCInetProcSysNet(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "ipv4": newRPCInetSysNetIPv4Dir(ctx, msrc), - "core": newRPCInetSysNetCore(ctx, msrc), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetSysNetCore builds the /proc/sys/net/core directory. -func newRPCInetSysNetCore(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "default_qdisc": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/default_qdisc", 0444), - "message_burst": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_burst", 0444), - "message_cost": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/message_cost", 0444), - "optmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/optmem_max", 0444), - "rmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_default", 0444), - "rmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/rmem_max", 0444), - "somaxconn": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/somaxconn", 0444), - "wmem_default": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_default", 0444), - "wmem_max": newRPCInetInode(ctx, msrc, "/proc/sys/net/core/wmem_max", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} - -// newRPCInetSysNetIPv4Dir builds the /proc/sys/net/ipv4 directory. -func newRPCInetSysNetIPv4Dir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { - contents := map[string]*fs.Inode{ - "ip_local_port_range": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_port_range", 0444), - "ip_local_reserved_ports": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_local_reserved_ports", 0444), - "ipfrag_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ipfrag_time", 0444), - "ip_nonlocal_bind": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_nonlocal_bind", 0444), - "ip_no_pmtu_disc": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/ip_no_pmtu_disc", 0444), - "tcp_allowed_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_allowed_congestion_control", 0444), - "tcp_available_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_available_congestion_control", 0444), - "tcp_base_mss": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_base_mss", 0444), - "tcp_congestion_control": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_congestion_control", 0644), - "tcp_dsack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_dsack", 0644), - "tcp_early_retrans": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_early_retrans", 0644), - "tcp_fack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fack", 0644), - "tcp_fastopen": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen", 0644), - "tcp_fastopen_key": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fastopen_key", 0444), - "tcp_fin_timeout": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_fin_timeout", 0644), - "tcp_invalid_ratelimit": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_invalid_ratelimit", 0444), - "tcp_keepalive_intvl": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_intvl", 0644), - "tcp_keepalive_probes": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_probes", 0644), - "tcp_keepalive_time": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_keepalive_time", 0644), - "tcp_mem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mem", 0444), - "tcp_mtu_probing": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_mtu_probing", 0644), - "tcp_no_metrics_save": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_no_metrics_save", 0444), - "tcp_probe_interval": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_interval", 0444), - "tcp_probe_threshold": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_probe_threshold", 0444), - "tcp_retries1": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries1", 0644), - "tcp_retries2": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_retries2", 0644), - "tcp_rfc1337": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rfc1337", 0444), - "tcp_rmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_rmem", 0444), - "tcp_sack": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_sack", 0644), - "tcp_slow_start_after_idle": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_slow_start_after_idle", 0644), - "tcp_synack_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_synack_retries", 0644), - "tcp_syn_retries": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_syn_retries", 0644), - "tcp_timestamps": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_timestamps", 0644), - "tcp_wmem": newRPCInetInode(ctx, msrc, "/proc/sys/net/ipv4/tcp_wmem", 0444), - } - - d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) - return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) -} diff --git a/pkg/sentry/fs/proc/stat.go b/pkg/sentry/fs/proc/stat.go index b641effbb..bc5b2bc7b 100644 --- a/pkg/sentry/fs/proc/stat.go +++ b/pkg/sentry/fs/proc/stat.go @@ -24,6 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // statData backs /proc/stat. // // +stateify savable @@ -140,3 +142,5 @@ func (s *statData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) ([] }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/sys.go b/pkg/sentry/fs/proc/sys.go index cd37776c8..2bdcf5f70 100644 --- a/pkg/sentry/fs/proc/sys.go +++ b/pkg/sentry/fs/proc/sys.go @@ -26,11 +26,12 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/fs/ramfs" "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // mmapMinAddrData backs /proc/sys/vm/mmap_min_addr. // // +stateify savable @@ -104,16 +105,10 @@ func (p *proc) newVMDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { func (p *proc) newSysDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode { children := map[string]*fs.Inode{ "kernel": p.newKernelDir(ctx, msrc), + "net": p.newSysNetDir(ctx, msrc), "vm": p.newVMDir(ctx, msrc), } - // If we're using rpcinet we will let it manage /proc/sys/net. - if _, ok := p.k.NetworkStack().(*rpcinet.Stack); ok { - children["net"] = newRPCInetProcSysNet(ctx, msrc) - } else { - children["net"] = p.newSysNetDir(ctx, msrc) - } - d := ramfs.NewDir(ctx, children, fs.RootOwner, fs.FilePermsFromMode(0555)) return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } @@ -160,3 +155,5 @@ func (hf *hostnameFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequ } var _ fs.FileOperations = (*hostnameFile)(nil) + +// LINT.ThenChange(../../fsimpl/proc/tasks_sys.go) diff --git a/pkg/sentry/fs/proc/sys_net.go b/pkg/sentry/fs/proc/sys_net.go index a37e1fa06..b9e8ef35f 100644 --- a/pkg/sentry/fs/proc/sys_net.go +++ b/pkg/sentry/fs/proc/sys_net.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + type tcpMemDir int const ( @@ -364,3 +366,5 @@ func (p *proc) newSysNetDir(ctx context.Context, msrc *fs.MountSource) *fs.Inode d := ramfs.NewDir(ctx, contents, fs.RootOwner, fs.FilePermsFromMode(0555)) return newProcInode(ctx, d, msrc, fs.SpecialDirectory, nil) } + +// LINT.ThenChange(../../fsimpl/proc/tasks_sys.go) diff --git a/pkg/sentry/fs/proc/task.go b/pkg/sentry/fs/proc/task.go index 9bf4b4527..7358d6ef9 100644 --- a/pkg/sentry/fs/proc/task.go +++ b/pkg/sentry/fs/proc/task.go @@ -37,6 +37,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // getTaskMM returns t's MemoryManager. If getTaskMM succeeds, the MemoryManager's // users count is incremented, and must be decremented by the caller when it is // no longer in use. @@ -800,3 +802,5 @@ func (f *auxvecFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc n, err := dst.CopyOut(ctx, buf[offset:]) return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/task.go|../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/uid_gid_map.go b/pkg/sentry/fs/proc/uid_gid_map.go index eea37d15c..3eacc9265 100644 --- a/pkg/sentry/fs/proc/uid_gid_map.go +++ b/pkg/sentry/fs/proc/uid_gid_map.go @@ -30,6 +30,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // idMapInodeOperations implements fs.InodeOperations for // /proc/[pid]/{uid,gid}_map. // @@ -177,3 +179,5 @@ func (imfo *idMapFileOperations) Write(ctx context.Context, file *fs.File, src u // count, even if fewer bytes were used. return int64(srclen), nil } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/proc/uptime.go b/pkg/sentry/fs/proc/uptime.go index 4e903917a..adfe58adb 100644 --- a/pkg/sentry/fs/proc/uptime.go +++ b/pkg/sentry/fs/proc/uptime.go @@ -28,6 +28,8 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +// LINT.IfChange + // uptime is a file containing the system uptime. // // +stateify savable @@ -85,3 +87,5 @@ func (f *uptimeFile) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequenc n, err := dst.CopyOut(ctx, s[offset:]) return int64(n), err } + +// LINT.ThenChange(../../fsimpl/proc/tasks_files.go) diff --git a/pkg/sentry/fs/proc/version.go b/pkg/sentry/fs/proc/version.go index a6d2c3cd3..27fd5b1cb 100644 --- a/pkg/sentry/fs/proc/version.go +++ b/pkg/sentry/fs/proc/version.go @@ -22,6 +22,8 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" ) +// LINT.IfChange + // versionData backs /proc/version. // // +stateify savable @@ -76,3 +78,5 @@ func (v *versionData) ReadSeqFileData(ctx context.Context, h seqfile.SeqHandle) }, }, 0 } + +// LINT.ThenChange(../../fsimpl/proc/task_files.go) diff --git a/pkg/sentry/fs/tmpfs/tmpfs.go b/pkg/sentry/fs/tmpfs/tmpfs.go index 69089c8a8..0f718e236 100644 --- a/pkg/sentry/fs/tmpfs/tmpfs.go +++ b/pkg/sentry/fs/tmpfs/tmpfs.go @@ -148,19 +148,19 @@ func (d *Dir) CreateFifo(ctx context.Context, dir *fs.Inode, name string, perms return d.ramfsDir.CreateFifo(ctx, dir, name, perms) } -// Getxattr implements fs.InodeOperations.Getxattr. -func (d *Dir) Getxattr(i *fs.Inode, name string) (string, error) { - return d.ramfsDir.Getxattr(i, name) +// GetXattr implements fs.InodeOperations.GetXattr. +func (d *Dir) GetXattr(ctx context.Context, i *fs.Inode, name string, size uint64) (string, error) { + return d.ramfsDir.GetXattr(ctx, i, name, size) } -// Setxattr implements fs.InodeOperations.Setxattr. -func (d *Dir) Setxattr(i *fs.Inode, name, value string) error { - return d.ramfsDir.Setxattr(i, name, value) +// SetXattr implements fs.InodeOperations.SetXattr. +func (d *Dir) SetXattr(ctx context.Context, i *fs.Inode, name, value string, flags uint32) error { + return d.ramfsDir.SetXattr(ctx, i, name, value, flags) } -// Listxattr implements fs.InodeOperations.Listxattr. -func (d *Dir) Listxattr(i *fs.Inode) (map[string]struct{}, error) { - return d.ramfsDir.Listxattr(i) +// ListXattr implements fs.InodeOperations.ListXattr. +func (d *Dir) ListXattr(ctx context.Context, i *fs.Inode) (map[string]struct{}, error) { + return d.ramfsDir.ListXattr(ctx, i) } // Lookup implements fs.InodeOperations.Lookup. diff --git a/pkg/sentry/fs/tty/line_discipline.go b/pkg/sentry/fs/tty/line_discipline.go index 894964260..9fe02657e 100644 --- a/pkg/sentry/fs/tty/line_discipline.go +++ b/pkg/sentry/fs/tty/line_discipline.go @@ -140,8 +140,10 @@ func (l *lineDiscipline) setTermios(ctx context.Context, io usermem.IO, args arc // buffer to its read buffer. Anything already in the read buffer is // now readable. if oldCanonEnabled && !l.termios.LEnabled(linux.ICANON) { - l.inQueue.pushWaitBuf(l) + l.inQueue.mu.Lock() + l.inQueue.pushWaitBufLocked(l) l.inQueue.readable = true + l.inQueue.mu.Unlock() l.slaveWaiter.Notify(waiter.EventIn) } diff --git a/pkg/sentry/fs/tty/queue.go b/pkg/sentry/fs/tty/queue.go index 8b5d4699a..21ccc6f32 100644 --- a/pkg/sentry/fs/tty/queue.go +++ b/pkg/sentry/fs/tty/queue.go @@ -197,18 +197,11 @@ func (q *queue) writeBytes(b []byte, l *lineDiscipline) { q.pushWaitBufLocked(l) } -// pushWaitBuf fills the queue's read buffer with data from the wait buffer. +// pushWaitBufLocked fills the queue's read buffer with data from the wait +// buffer. // // Preconditions: // * l.termiosMu must be held for reading. -func (q *queue) pushWaitBuf(l *lineDiscipline) int { - q.mu.Lock() - defer q.mu.Unlock() - return q.pushWaitBufLocked(l) -} - -// Preconditions: -// * l.termiosMu must be held for reading. // * q.mu must be locked. func (q *queue) pushWaitBufLocked(l *lineDiscipline) int { if q.waitBufLen == 0 { diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index 79759e0fc..a4600ad47 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/syserror" ) @@ -40,7 +39,7 @@ func (fs *Filesystem) stepExistingLocked(ctx context.Context, rp *vfs.ResolvingP return nil, syserror.ENOTDIR } // Directory searchable? - if err := d.inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := d.inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } afterSymlink: @@ -182,8 +181,8 @@ func (fs *Filesystem) walkParentDirLocked(ctx context.Context, rp *vfs.Resolving // // Preconditions: Filesystem.mu must be locked for at least reading. parentInode // == parentVFSD.Impl().(*Dentry).Inode. isDir(parentInode) == true. -func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { +func checkCreateLocked(ctx context.Context, rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInode Inode) (string, error) { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return "", err } pc := rp.Component() @@ -206,7 +205,7 @@ func checkCreateLocked(rp *vfs.ResolvingPath, parentVFSD *vfs.Dentry, parentInod // checkDeleteLocked checks that the file represented by vfsd may be deleted. // // Preconditions: Filesystem.mu must be locked for at least reading. -func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { +func checkDeleteLocked(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { parentVFSD := vfsd.Parent() if parentVFSD == nil { return syserror.EBUSY @@ -214,36 +213,12 @@ func checkDeleteLocked(rp *vfs.ResolvingPath, vfsd *vfs.Dentry) error { if parentVFSD.IsDisowned() { return syserror.ENOENT } - if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { + if err := parentVFSD.Impl().(*Dentry).inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite|vfs.MayExec); err != nil { return err } return nil } -// checkRenameLocked checks that a rename operation may be performed on the -// target dentry across the given set of parent directories. The target dentry -// may be nil. -// -// Precondition: isDir(dstInode) == true. -func checkRenameLocked(creds *auth.Credentials, src, dstDir *vfs.Dentry, dstInode Inode) error { - srcDir := src.Parent() - if srcDir == nil { - return syserror.EBUSY - } - if srcDir.IsDisowned() { - return syserror.ENOENT - } - if dstDir.IsDisowned() { - return syserror.ENOENT - } - // Check for creation permissions on dst dir. - if err := dstInode.CheckPermissions(creds, vfs.MayWrite|vfs.MayExec); err != nil { - return err - } - - return nil -} - // Release implements vfs.FilesystemImpl.Release. func (fs *Filesystem) Release() { } @@ -269,7 +244,7 @@ func (fs *Filesystem) GetDentryAt(ctx context.Context, rp *vfs.ResolvingPath, op if !d.isDir() { return nil, syserror.ENOTDIR } - if err := inode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } } @@ -302,7 +277,7 @@ func (fs *Filesystem) LinkAt(ctx context.Context, rp *vfs.ResolvingPath, vd vfs. if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -339,7 +314,7 @@ func (fs *Filesystem) MkdirAt(ctx context.Context, rp *vfs.ResolvingPath, opts v if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -367,7 +342,7 @@ func (fs *Filesystem) MknodAt(ctx context.Context, rp *vfs.ResolvingPath, opts v if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -401,7 +376,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if err != nil { return nil, err } - if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return inode.Open(rp, vfsd, opts.Flags) @@ -420,7 +395,7 @@ func (fs *Filesystem) OpenAt(ctx context.Context, rp *vfs.ResolvingPath, opts vf if mustCreate { return nil, syserror.EEXIST } - if err := inode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := inode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return inode.Open(rp, vfsd, opts.Flags) @@ -432,7 +407,7 @@ afterTrailingSymlink: return nil, err } // Check for search permission in the parent directory. - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayExec); err != nil { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayExec); err != nil { return nil, err } // Reject attempts to open directories with O_CREAT. @@ -450,7 +425,7 @@ afterTrailingSymlink: } if childVFSD == nil { // Already checked for searchability above; now check for writability. - if err := parentInode.CheckPermissions(rp.Credentials(), vfs.MayWrite); err != nil { + if err := parentInode.CheckPermissions(ctx, rp.Credentials(), vfs.MayWrite); err != nil { return nil, err } if err := rp.Mount().CheckBeginWrite(); err != nil { @@ -485,7 +460,7 @@ afterTrailingSymlink: goto afterTrailingSymlink } } - if err := childInode.CheckPermissions(rp.Credentials(), ats); err != nil { + if err := childInode.CheckPermissions(ctx, rp.Credentials(), ats); err != nil { return nil, err } return childInode.Open(rp, childVFSD, opts.Flags) @@ -545,13 +520,13 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa srcVFSD := &src.vfsd // Can we remove the src dentry? - if err := checkDeleteLocked(rp, srcVFSD); err != nil { + if err := checkDeleteLocked(ctx, rp, srcVFSD); err != nil { return err } // Can we create the dst dentry? var dstVFSD *vfs.Dentry - pc, err := checkCreateLocked(rp, dstDirVFSD, dstDirInode) + pc, err := checkCreateLocked(ctx, rp, dstDirVFSD, dstDirInode) switch err { case nil: // Ok, continue with rename as replacement. @@ -607,7 +582,7 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } if !vfsd.Impl().(*Dentry).isDir() { @@ -683,7 +658,7 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ if err != nil { return err } - pc, err := checkCreateLocked(rp, parentVFSD, parentInode) + pc, err := checkCreateLocked(ctx, rp, parentVFSD, parentInode) if err != nil { return err } @@ -712,7 +687,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error return err } defer rp.Mount().EndWrite() - if err := checkDeleteLocked(rp, vfsd); err != nil { + if err := checkDeleteLocked(ctx, rp, vfsd); err != nil { return err } if vfsd.Impl().(*Dentry).isDir() { diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 1d469a0db..1700fffd9 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -262,7 +262,7 @@ func (a *InodeAttrs) SetStat(_ *vfs.Filesystem, opts vfs.SetStatOptions) error { } // CheckPermissions implements Inode.CheckPermissions. -func (a *InodeAttrs) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { +func (a *InodeAttrs) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { mode := a.Mode() return vfs.GenericCheckPermissions( creds, @@ -510,3 +510,47 @@ type InodeSymlink struct { func (InodeSymlink) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { return nil, syserror.ELOOP } + +// StaticDirectory is a standard implementation of a directory with static +// contents. +// +// +stateify savable +type StaticDirectory struct { + InodeNotSymlink + InodeDirectoryNoNewChildren + InodeAttrs + InodeNoDynamicLookup + OrderedChildren +} + +var _ Inode = (*StaticDirectory)(nil) + +// NewStaticDir creates a new static directory and returns its dentry. +func NewStaticDir(creds *auth.Credentials, ino uint64, perm linux.FileMode, children map[string]*Dentry) *Dentry { + inode := &StaticDirectory{} + inode.Init(creds, ino, perm) + + dentry := &Dentry{} + dentry.Init(inode) + + inode.OrderedChildren.Init(OrderedChildrenOptions{}) + links := inode.OrderedChildren.Populate(dentry, children) + inode.IncLinks(links) + + return dentry +} + +// Init initializes StaticDirectory. +func (s *StaticDirectory) Init(creds *auth.Credentials, ino uint64, perm linux.FileMode) { + if perm&^linux.PermissionsMask != 0 { + panic(fmt.Sprintf("Only permission mask must be set: %x", perm&linux.PermissionsMask)) + } + s.InodeAttrs.Init(creds, ino, linux.ModeDirectory|perm) +} + +// Open implements kernfs.Inode. +func (s *StaticDirectory) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &s.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index bb12f39a2..85bcdcc57 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -320,7 +320,7 @@ type inodeMetadata interface { // CheckPermissions checks that creds may access this inode for the // requested access type, per the the rules of // fs/namei.c:generic_permission(). - CheckPermissions(creds *auth.Credentials, atx vfs.AccessTypes) error + CheckPermissions(ctx context.Context, creds *auth.Credentials, atx vfs.AccessTypes) error // Mode returns the (struct stat)::st_mode value for this inode. This is // separated from Stat for performance. diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 068063f4e..f19f12854 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -20,7 +20,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) -type staticSymlink struct { +// StaticSymlink provides an Inode implementation for symlinks that point to +// a immutable target. +type StaticSymlink struct { InodeAttrs InodeNoopRefCount InodeSymlink @@ -28,18 +30,25 @@ type staticSymlink struct { target string } -var _ Inode = (*staticSymlink)(nil) +var _ Inode = (*StaticSymlink)(nil) // NewStaticSymlink creates a new symlink file pointing to 'target'. -func NewStaticSymlink(creds *auth.Credentials, ino uint64, perm linux.FileMode, target string) *Dentry { - inode := &staticSymlink{target: target} - inode.Init(creds, ino, linux.ModeSymlink|perm) +func NewStaticSymlink(creds *auth.Credentials, ino uint64, target string) *Dentry { + inode := &StaticSymlink{} + inode.Init(creds, ino, target) d := &Dentry{} d.Init(inode) return d } -func (s *staticSymlink) Readlink(_ context.Context) (string, error) { +// Init initializes the instance. +func (s *StaticSymlink) Init(creds *auth.Credentials, ino uint64, target string) { + s.target = target + s.InodeAttrs.Init(creds, ino, linux.ModeSymlink|0777) +} + +// Readlink implements Inode. +func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { return s.target, nil } diff --git a/pkg/sentry/fsimpl/proc/BUILD b/pkg/sentry/fsimpl/proc/BUILD index 1f44b3217..f69aa19c4 100644 --- a/pkg/sentry/fsimpl/proc/BUILD +++ b/pkg/sentry/fsimpl/proc/BUILD @@ -7,22 +7,17 @@ go_library( name = "proc", srcs = [ "filesystem.go", - "loadavg.go", - "meminfo.go", - "mounts.go", - "net.go", - "stat.go", - "sys.go", + "subtasks.go", "task.go", "task_files.go", "tasks.go", "tasks_files.go", - "version.go", + "tasks_net.go", + "tasks_sys.go", ], importpath = "gvisor.dev/gvisor/pkg/sentry/fsimpl/proc", deps = [ "//pkg/abi/linux", - "//pkg/binary", "//pkg/log", "//pkg/sentry/context", "//pkg/sentry/fs", @@ -30,8 +25,10 @@ go_library( "//pkg/sentry/inet", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/kernel/time", "//pkg/sentry/limits", "//pkg/sentry/mm", + "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", @@ -39,6 +36,7 @@ go_library( "//pkg/sentry/usermem", "//pkg/sentry/vfs", "//pkg/syserror", + "//pkg/tcpip/header", ], ) @@ -47,7 +45,7 @@ go_test( size = "small", srcs = [ "boot_test.go", - "net_test.go", + "tasks_sys_test.go", "tasks_test.go", ], embed = [":proc"], diff --git a/pkg/sentry/fsimpl/proc/filesystem.go b/pkg/sentry/fsimpl/proc/filesystem.go index d09182c77..f49819187 100644 --- a/pkg/sentry/fsimpl/proc/filesystem.go +++ b/pkg/sentry/fsimpl/proc/filesystem.go @@ -47,7 +47,12 @@ func (ft *procFSType) GetFilesystem(ctx context.Context, vfsObj *vfs.VirtualFile procfs := &kernfs.Filesystem{} procfs.VFSFilesystem().Init(vfsObj, procfs) - _, dentry := newTasksInode(procfs, k, pidns) + var data *InternalData + if opts.InternalData != nil { + data = opts.InternalData.(*InternalData) + } + + _, dentry := newTasksInode(procfs, k, pidns, data.Cgroups) return procfs.VFSFilesystem(), dentry.VFSDentry(), nil } @@ -67,3 +72,20 @@ func newDentry(creds *auth.Credentials, ino uint64, perm linux.FileMode, inode d d.Init(inode) return d } + +type staticFile struct { + kernfs.DynamicBytesFile + vfs.StaticData +} + +var _ dynamicInode = (*staticFile)(nil) + +func newStaticFile(data string) *staticFile { + return &staticFile{StaticData: vfs.StaticData{Data: data}} +} + +// InternalData contains internal data passed in to the procfs mount via +// vfs.GetFilesystemOptions.InternalData. +type InternalData struct { + Cgroups map[string]string +} diff --git a/pkg/sentry/fsimpl/proc/loadavg.go b/pkg/sentry/fsimpl/proc/loadavg.go deleted file mode 100644 index 5351d86e8..000000000 --- a/pkg/sentry/fsimpl/proc/loadavg.go +++ /dev/null @@ -1,42 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" -) - -// loadavgData backs /proc/loadavg. -// -// +stateify savable -type loadavgData struct { - kernfs.DynamicBytesFile -} - -var _ dynamicInode = (*loadavgData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/62345059): Include real data in fields. - // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. - // Column 4-5: currently running processes and the total number of processes. - // Column 6: the last process ID used. - fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/meminfo.go b/pkg/sentry/fsimpl/proc/meminfo.go deleted file mode 100644 index cbdd4f3fc..000000000 --- a/pkg/sentry/fsimpl/proc/meminfo.go +++ /dev/null @@ -1,79 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" -) - -// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. -// -// +stateify savable -type meminfoData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*meminfoData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { - mf := d.k.MemoryFile() - mf.UpdateUsage() - snapshot, totalUsage := usage.MemoryAccounting.Copy() - totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) - anon := snapshot.Anonymous + snapshot.Tmpfs - file := snapshot.PageCache + snapshot.Mapped - // We don't actually have active/inactive LRUs, so just make up numbers. - activeFile := (file / 2) &^ (usermem.PageSize - 1) - inactiveFile := file - activeFile - - fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) - memFree := (totalSize - totalUsage) / 1024 - // We use MemFree as MemAvailable because we don't swap. - // TODO(rahat): When reclaim is implemented the value of MemAvailable - // should change. - fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) - fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) - fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices - fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) - // Emulate a system with no swap, which disables inactivation of anon pages. - fmt.Fprintf(buf, "SwapCache: 0 kB\n") - fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) - fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") - fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) - fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) - fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) - fmt.Fprintf(buf, "SwapTotal: 0 kB\n") - fmt.Fprintf(buf, "SwapFree: 0 kB\n") - fmt.Fprintf(buf, "Dirty: 0 kB\n") - fmt.Fprintf(buf, "Writeback: 0 kB\n") - fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) - fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know - fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) - return nil -} diff --git a/pkg/sentry/fsimpl/proc/mounts.go b/pkg/sentry/fsimpl/proc/mounts.go deleted file mode 100644 index 8683cf677..000000000 --- a/pkg/sentry/fsimpl/proc/mounts.go +++ /dev/null @@ -1,33 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import "gvisor.dev/gvisor/pkg/sentry/kernel" - -// TODO(gvisor.dev/issue/1195): Implement mountInfoFile and mountsFile. - -// mountInfoFile implements vfs.DynamicBytesSource for /proc/[pid]/mountinfo. -// -// +stateify savable -type mountInfoFile struct { - t *kernel.Task -} - -// mountsFile implements vfs.DynamicBytesSource for /proc/[pid]/mounts. -// -// +stateify savable -type mountsFile struct { - t *kernel.Task -} diff --git a/pkg/sentry/fsimpl/proc/net.go b/pkg/sentry/fsimpl/proc/net.go deleted file mode 100644 index fd46eebf8..000000000 --- a/pkg/sentry/fsimpl/proc/net.go +++ /dev/null @@ -1,338 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. -// -// +stateify savable -type ifinet6 struct { - s inet.Stack -} - -var _ vfs.DynamicBytesSource = (*ifinet6)(nil) - -func (n *ifinet6) contents() []string { - var lines []string - nics := n.s.Interfaces() - for id, naddrs := range n.s.InterfaceAddrs() { - nic, ok := nics[id] - if !ok { - // NIC was added after NICNames was called. We'll just - // ignore it. - continue - } - - for _, a := range naddrs { - // IPv6 only. - if a.Family != linux.AF_INET6 { - continue - } - - // Fields: - // IPv6 address displayed in 32 hexadecimal chars without colons - // Netlink device number (interface index) in hexadecimal (use nic id) - // Prefix length in hexadecimal - // Scope value (use 0) - // Interface flags - // Device name - lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) - } - } - return lines -} - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { - for _, l := range n.contents() { - buf.WriteString(l) - } - return nil -} - -// netDev implements vfs.DynamicBytesSource for /proc/net/dev. -// -// +stateify savable -type netDev struct { - s inet.Stack -} - -var _ vfs.DynamicBytesSource = (*netDev)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netDev) Generate(ctx context.Context, buf *bytes.Buffer) error { - interfaces := n.s.Interfaces() - buf.WriteString("Inter-| Receive | Transmit\n") - buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") - - for _, i := range interfaces { - // Implements the same format as - // net/core/net-procfs.c:dev_seq_printf_stats. - var stats inet.StatDev - if err := n.s.Statistics(&stats, i.Name); err != nil { - log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) - continue - } - fmt.Fprintf( - buf, - "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", - i.Name, - // Received - stats[0], // bytes - stats[1], // packets - stats[2], // errors - stats[3], // dropped - stats[4], // fifo - stats[5], // frame - stats[6], // compressed - stats[7], // multicast - // Transmitted - stats[8], // bytes - stats[9], // packets - stats[10], // errors - stats[11], // dropped - stats[12], // fifo - stats[13], // frame - stats[14], // compressed - stats[15], // multicast - ) - } - - return nil -} - -// netUnix implements vfs.DynamicBytesSource for /proc/net/unix. -// -// +stateify savable -type netUnix struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*netUnix)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (n *netUnix) Generate(ctx context.Context, buf *bytes.Buffer) error { - buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") - for _, se := range n.k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) - continue - } - sfile := s.(*fs.File) - if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { - s.DecRef() - // Not a unix socket. - continue - } - sops := sfile.FileOperations.(*unix.SocketOperations) - - addr, err := sops.Endpoint().GetLocalAddress() - if err != nil { - log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) - addr.Addr = "<unknown>" - } - - sockFlags := 0 - if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { - if ce.Listening() { - // For unix domain sockets, linux reports a single flag - // value if the socket is listening, of __SO_ACCEPTCON. - sockFlags = linux.SO_ACCEPTCON - } - } - - // In the socket entry below, the value for the 'Num' field requires - // some consideration. Linux prints the address to the struct - // unix_sock representing a socket in the kernel, but may redact the - // value for unprivileged users depending on the kptr_restrict - // sysctl. - // - // One use for this field is to allow a privileged user to - // introspect into the kernel memory to determine information about - // a socket not available through procfs, such as the socket's peer. - // - // In gvisor, returning a pointer to our internal structures would - // be pointless, as it wouldn't match the memory layout for struct - // unix_sock, making introspection difficult. We could populate a - // struct unix_sock with the appropriate data, but even that - // requires consideration for which kernel version to emulate, as - // the definition of this struct changes over time. - // - // For now, we always redact this pointer. - fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", - (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. - sfile.ReadRefs()-1, // RefCount, don't count our own ref. - 0, // Protocol, always 0 for UDS. - sockFlags, // Flags. - sops.Endpoint().Type(), // Type. - sops.State(), // State. - sfile.InodeID(), // Inode. - ) - - // Path - if len(addr.Addr) != 0 { - if addr.Addr[0] == 0 { - // Abstract path. - fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) - } else { - fmt.Fprintf(buf, " %s", string(addr.Addr)) - } - } - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - return nil -} - -// netTCP implements vfs.DynamicBytesSource for /proc/net/tcp. -// -// +stateify savable -type netTCP struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*netTCP)(nil) - -func (n *netTCP) Generate(ctx context.Context, buf *bytes.Buffer) error { - t := kernel.TaskFromContext(ctx) - buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") - for _, se := range n.k.ListSockets() { - s := se.Sock.Get() - if s == nil { - log.Debugf("Couldn't resolve weakref %+v in socket table, racing with destruction?", se.Sock) - continue - } - sfile := s.(*fs.File) - sops, ok := sfile.FileOperations.(socket.Socket) - if !ok { - panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) - } - if family, stype, _ := sops.Type(); !(family == linux.AF_INET && stype == linux.SOCK_STREAM) { - s.DecRef() - // Not tcp4 sockets. - continue - } - - // Linux's documentation for the fields below can be found at - // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. - // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). - // Note that the header doesn't contain labels for all the fields. - - // Field: sl; entry number. - fmt.Fprintf(buf, "%4d: ", se.ID) - - portBuf := make([]byte, 2) - - // Field: local_adddress. - var localAddr linux.SockAddrInet - if local, _, err := sops.GetSockName(t); err == nil { - localAddr = *local.(*linux.SockAddrInet) - } - binary.LittleEndian.PutUint16(portBuf, localAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(localAddr.Addr[:]), - portBuf) - - // Field: rem_address. - var remoteAddr linux.SockAddrInet - if remote, _, err := sops.GetPeerName(t); err == nil { - remoteAddr = *remote.(*linux.SockAddrInet) - } - binary.LittleEndian.PutUint16(portBuf, remoteAddr.Port) - fmt.Fprintf(buf, "%08X:%04X ", - binary.LittleEndian.Uint32(remoteAddr.Addr[:]), - portBuf) - - // Field: state; socket state. - fmt.Fprintf(buf, "%02X ", sops.State()) - - // Field: tx_queue, rx_queue; number of packets in the transmit and - // receive queue. Unimplemented. - fmt.Fprintf(buf, "%08X:%08X ", 0, 0) - - // Field: tr, tm->when; timer active state and number of jiffies - // until timer expires. Unimplemented. - fmt.Fprintf(buf, "%02X:%08X ", 0, 0) - - // Field: retrnsmt; number of unrecovered RTO timeouts. - // Unimplemented. - fmt.Fprintf(buf, "%08X ", 0) - - // Field: uid. - uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) - if err != nil { - log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) - fmt.Fprintf(buf, "%5d ", 0) - } else { - fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - } - - // Field: timeout; number of unanswered 0-window probes. - // Unimplemented. - fmt.Fprintf(buf, "%8d ", 0) - - // Field: inode. - fmt.Fprintf(buf, "%8d ", sfile.InodeID()) - - // Field: refcount. Don't count the ref we obtain while deferencing - // the weakref to this socket. - fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) - - // Field: Socket struct address. Redacted due to the same reason as - // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. - fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) - - // Field: retransmit timeout. Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: predicted tick of soft clock (delayed ACK control data). - // Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: sending congestion window, Unimplemented. - fmt.Fprintf(buf, "%d ", 0) - - // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. - // Unimplemented, report as large threshold. - fmt.Fprintf(buf, "%d", -1) - - fmt.Fprintf(buf, "\n") - - s.DecRef() - } - - return nil -} diff --git a/pkg/sentry/fsimpl/proc/stat.go b/pkg/sentry/fsimpl/proc/stat.go deleted file mode 100644 index 50894a534..000000000 --- a/pkg/sentry/fsimpl/proc/stat.go +++ /dev/null @@ -1,129 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" -) - -// cpuStats contains the breakdown of CPU time for /proc/stat. -type cpuStats struct { - // user is time spent in userspace tasks with non-positive niceness. - user uint64 - - // nice is time spent in userspace tasks with positive niceness. - nice uint64 - - // system is time spent in non-interrupt kernel context. - system uint64 - - // idle is time spent idle. - idle uint64 - - // ioWait is time spent waiting for IO. - ioWait uint64 - - // irq is time spent in interrupt context. - irq uint64 - - // softirq is time spent in software interrupt context. - softirq uint64 - - // steal is involuntary wait time. - steal uint64 - - // guest is time spent in guests with non-positive niceness. - guest uint64 - - // guestNice is time spent in guests with positive niceness. - guestNice uint64 -} - -// String implements fmt.Stringer. -func (c cpuStats) String() string { - return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) -} - -// statData implements vfs.DynamicBytesSource for /proc/stat. -// -// +stateify savable -type statData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*statData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { - // TODO(b/37226836): We currently export only zero CPU stats. We could - // at least provide some aggregate stats. - var cpu cpuStats - fmt.Fprintf(buf, "cpu %s\n", cpu) - - for c, max := uint(0), s.k.ApplicationCores(); c < max; c++ { - fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) - } - - // The total number of interrupts is dependent on the CPUs and PCI - // devices on the system. See arch_probe_nr_irqs. - // - // Since we don't report real interrupt stats, just choose an arbitrary - // value from a representative VM. - const numInterrupts = 256 - - // The Kernel doesn't handle real interrupts, so report all zeroes. - // TODO(b/37226836): We could count page faults as #PF. - fmt.Fprintf(buf, "intr 0") // total - for i := 0; i < numInterrupts; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - - // Total number of context switches. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "ctxt 0\n") - - // CLOCK_REALTIME timestamp from boot, in seconds. - fmt.Fprintf(buf, "btime %d\n", s.k.Timekeeper().BootTime().Seconds()) - - // Total number of clones. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "processes 0\n") - - // Number of runnable tasks. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_running 0\n") - - // Number of tasks waiting on IO. - // TODO(b/37226836): Count this. - fmt.Fprintf(buf, "procs_blocked 0\n") - - // Number of each softirq handled. - fmt.Fprintf(buf, "softirq 0") // total - for i := 0; i < linux.NumSoftIRQ; i++ { - fmt.Fprintf(buf, " 0") - } - fmt.Fprintf(buf, "\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/subtasks.go b/pkg/sentry/fsimpl/proc/subtasks.go new file mode 100644 index 000000000..91eded415 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/subtasks.go @@ -0,0 +1,128 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "sort" + "strconv" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" +) + +// subtasksInode represents the inode for /proc/[pid]/task/ directory. +// +// +stateify savable +type subtasksInode struct { + kernfs.InodeNotSymlink + kernfs.InodeDirectoryNoNewChildren + kernfs.InodeAttrs + kernfs.OrderedChildren + + task *kernel.Task + pidns *kernel.PIDNamespace + inoGen InoGenerator + cgroupControllers map[string]string +} + +var _ kernfs.Inode = (*subtasksInode)(nil) + +func newSubtasks(task *kernel.Task, pidns *kernel.PIDNamespace, inoGen InoGenerator, cgroupControllers map[string]string) *kernfs.Dentry { + subInode := &subtasksInode{ + task: task, + pidns: pidns, + inoGen: inoGen, + cgroupControllers: cgroupControllers, + } + // Note: credentials are overridden by taskOwnedInode. + subInode.InodeAttrs.Init(task.Credentials(), inoGen.NextIno(), linux.ModeDirectory|0555) + subInode.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + + inode := &taskOwnedInode{Inode: subInode, owner: task} + dentry := &kernfs.Dentry{} + dentry.Init(inode) + + return dentry +} + +// Valid implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) Valid(ctx context.Context) bool { + return true +} + +// Lookup implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + tid, err := strconv.ParseUint(name, 10, 32) + if err != nil { + return nil, syserror.ENOENT + } + + subTask := i.pidns.TaskWithID(kernel.ThreadID(tid)) + if subTask == nil { + return nil, syserror.ENOENT + } + if subTask.ThreadGroup() != i.task.ThreadGroup() { + return nil, syserror.ENOENT + } + + subTaskDentry := newTaskInode(i.inoGen, subTask, i.pidns, false, i.cgroupControllers) + return subTaskDentry.VFSDentry(), nil +} + +// IterDirents implements kernfs.inodeDynamicLookup. +func (i *subtasksInode) IterDirents(ctx context.Context, cb vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + tasks := i.task.ThreadGroup().MemberIDs(i.pidns) + if len(tasks) == 0 { + return offset, syserror.ENOENT + } + + tids := make([]int, 0, len(tasks)) + for _, tid := range tasks { + tids = append(tids, int(tid)) + } + + sort.Ints(tids) + for _, tid := range tids[relOffset:] { + dirent := vfs.Dirent{ + Name: strconv.FormatUint(uint64(tid), 10), + Type: linux.DT_DIR, + Ino: i.inoGen.NextIno(), + NextOff: offset + 1, + } + if !cb.Handle(dirent) { + return offset, nil + } + offset++ + } + return offset, nil +} + +// Open implements kernfs.Inode. +func (i *subtasksInode) Open(rp *vfs.ResolvingPath, vfsd *vfs.Dentry, flags uint32) (*vfs.FileDescription, error) { + fd := &kernfs.GenericDirectoryFD{} + fd.Init(rp.Mount(), vfsd, &i.OrderedChildren, flags) + return fd.VFSFileDescription(), nil +} + +// Stat implements kernfs.Inode. +func (i *subtasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { + stat := i.InodeAttrs.Stat(vsfs) + stat.Nlink += uint32(i.task.ThreadGroup().Count()) + return stat +} diff --git a/pkg/sentry/fsimpl/proc/sys.go b/pkg/sentry/fsimpl/proc/sys.go deleted file mode 100644 index b88256e12..000000000 --- a/pkg/sentry/fsimpl/proc/sys.go +++ /dev/null @@ -1,51 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/kernel" - "gvisor.dev/gvisor/pkg/sentry/vfs" -) - -// mmapMinAddrData implements vfs.DynamicBytesSource for -// /proc/sys/vm/mmap_min_addr. -// -// +stateify savable -type mmapMinAddrData struct { - k *kernel.Kernel -} - -var _ vfs.DynamicBytesSource = (*mmapMinAddrData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) - return nil -} - -// +stateify savable -type overcommitMemory struct{} - -var _ vfs.DynamicBytesSource = (*overcommitMemory)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (d *overcommitMemory) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "0\n") - return nil -} diff --git a/pkg/sentry/fsimpl/proc/task.go b/pkg/sentry/fsimpl/proc/task.go index 11a64c777..a0580f20d 100644 --- a/pkg/sentry/fsimpl/proc/task.go +++ b/pkg/sentry/fsimpl/proc/task.go @@ -15,6 +15,9 @@ package proc import ( + "bytes" + "fmt" + "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" @@ -40,33 +43,37 @@ type taskInode struct { var _ kernfs.Inode = (*taskInode)(nil) -func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool) *kernfs.Dentry { +func newTaskInode(inoGen InoGenerator, task *kernel.Task, pidns *kernel.PIDNamespace, isThreadGroup bool, cgroupControllers map[string]string) *kernfs.Dentry { contents := map[string]*kernfs.Dentry{ - //"auxv": newAuxvec(t, msrc), - //"cmdline": newExecArgInode(t, msrc, cmdlineExecArg), - //"comm": newComm(t, msrc), - //"environ": newExecArgInode(t, msrc, environExecArg), + "auxv": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &auxvData{task: task}), + "cmdline": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: cmdlineDataArg}), + "comm": newComm(task, inoGen.NextIno(), 0444), + "environ": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &cmdlineData{task: task, arg: environDataArg}), //"exe": newExe(t, msrc), //"fd": newFdDir(t, msrc), //"fdinfo": newFdInfoDir(t, msrc), - //"gid_map": newGIDMap(t, msrc), - "io": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, newIO(task, isThreadGroup)), - "maps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &mapsData{task: task}), + "gid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: true}), + "io": newTaskOwnedFile(task, inoGen.NextIno(), 0400, newIO(task, isThreadGroup)), + "maps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &mapsData{task: task}), //"mountinfo": seqfile.NewSeqFileInode(t, &mountInfoFile{t: t}, msrc), //"mounts": seqfile.NewSeqFileInode(t, &mountsFile{t: t}, msrc), - //"ns": newNamespaceDir(t, msrc), - "smaps": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &smapsData{task: task}), - "stat": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &taskStatData{t: task, pidns: pidns, tgstats: isThreadGroup}), - "statm": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statmData{t: task}), - "status": newTaskOwnedFile(task, inoGen.NextIno(), defaultPermission, &statusData{t: task, pidns: pidns}), - //"uid_map": newUIDMap(t, msrc), + "ns": newTaskOwnedDir(task, inoGen.NextIno(), 0511, map[string]*kernfs.Dentry{ + "net": newNamespaceSymlink(task, inoGen.NextIno(), "net"), + "pid": newNamespaceSymlink(task, inoGen.NextIno(), "pid"), + "user": newNamespaceSymlink(task, inoGen.NextIno(), "user"), + }), + "smaps": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &smapsData{task: task}), + "stat": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &taskStatData{task: task, pidns: pidns, tgstats: isThreadGroup}), + "statm": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statmData{task: task}), + "status": newTaskOwnedFile(task, inoGen.NextIno(), 0444, &statusData{task: task, pidns: pidns}), + "uid_map": newTaskOwnedFile(task, inoGen.NextIno(), 0644, &idMapData{task: task, gids: false}), } if isThreadGroup { - //contents["task"] = p.newSubtasks(t, msrc) + contents["task"] = newSubtasks(task, pidns, inoGen, cgroupControllers) + } + if len(cgroupControllers) > 0 { + contents["cgroup"] = newTaskOwnedFile(task, inoGen.NextIno(), 0444, newCgroupData(cgroupControllers)) } - //if len(p.cgroupControllers) > 0 { - // contents["cgroup"] = newCGroupInode(t, msrc, p.cgroupControllers) - //} taskInode := &taskInode{task: task} // Note: credentials are overridden by taskOwnedInode. @@ -127,6 +134,23 @@ func newTaskOwnedFile(task *kernel.Task, ino uint64, perm linux.FileMode, inode return d } +func newTaskOwnedDir(task *kernel.Task, ino uint64, perm linux.FileMode, children map[string]*kernfs.Dentry) *kernfs.Dentry { + dir := &kernfs.StaticDirectory{} + + // Note: credentials are overridden by taskOwnedInode. + dir.Init(task.Credentials(), ino, perm) + + inode := &taskOwnedInode{Inode: dir, owner: task} + d := &kernfs.Dentry{} + d.Init(inode) + + dir.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + links := dir.OrderedChildren.Populate(d, children) + dir.IncLinks(links) + + return d +} + // Stat implements kernfs.Inode. func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { stat := i.Inode.Stat(fs) @@ -137,7 +161,7 @@ func (i *taskOwnedInode) Stat(fs *vfs.Filesystem) linux.Statx { } // CheckPermissions implements kernfs.Inode. -func (i *taskOwnedInode) CheckPermissions(creds *auth.Credentials, ats vfs.AccessTypes) error { +func (i *taskOwnedInode) CheckPermissions(_ context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { mode := i.Mode() uid, gid := i.getOwner(mode) return vfs.GenericCheckPermissions( @@ -188,3 +212,38 @@ func newIO(t *kernel.Task, isThreadGroup bool) *ioData { } return &ioData{ioUsage: t} } + +func newNamespaceSymlink(task *kernel.Task, ino uint64, ns string) *kernfs.Dentry { + // Namespace symlinks should contain the namespace name and the inode number + // for the namespace instance, so for example user:[123456]. We currently fake + // the inode number by sticking the symlink inode in its place. + target := fmt.Sprintf("%s:[%d]", ns, ino) + + inode := &kernfs.StaticSymlink{} + // Note: credentials are overridden by taskOwnedInode. + inode.Init(task.Credentials(), ino, target) + + taskInode := &taskOwnedInode{Inode: inode, owner: task} + d := &kernfs.Dentry{} + d.Init(taskInode) + return d +} + +// newCgroupData creates inode that shows cgroup information. +// From man 7 cgroups: "For each cgroup hierarchy of which the process is a +// member, there is one entry containing three colon-separated fields: +// hierarchy-ID:controller-list:cgroup-path" +func newCgroupData(controllers map[string]string) dynamicInode { + buf := bytes.Buffer{} + + // The hierarchy ids must be positive integers (for cgroup v1), but the + // exact number does not matter, so long as they are unique. We can + // just use a counter, but since linux sorts this file in descending + // order, we must count down to preserve this behavior. + i := len(controllers) + for name, dir := range controllers { + fmt.Fprintf(&buf, "%d:%s:%s\n", i, name, dir) + i-- + } + return newStaticFile(buf.String()) +} diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index 93f0e1aa8..7bc352ae9 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -17,15 +17,20 @@ package proc import ( "bytes" "fmt" + "io" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/mm" + "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" ) // mm gets the kernel task's MemoryManager. No additional reference is taken on @@ -41,6 +46,256 @@ func getMM(task *kernel.Task) *mm.MemoryManager { return tmm } +// getMMIncRef returns t's MemoryManager. If getMMIncRef succeeds, the +// MemoryManager's users count is incremented, and must be decremented by the +// caller when it is no longer in use. +func getMMIncRef(task *kernel.Task) (*mm.MemoryManager, error) { + if task.ExitState() == kernel.TaskExitDead { + return nil, syserror.ESRCH + } + var m *mm.MemoryManager + task.WithMuLocked(func(t *kernel.Task) { + m = t.MemoryManager() + }) + if m == nil || !m.IncUsers() { + return nil, io.EOF + } + return m, nil +} + +type bufferWriter struct { + buf *bytes.Buffer +} + +// WriteFromBlocks writes up to srcs.NumBytes() bytes from srcs and returns +// the number of bytes written. It may return a partial write without an +// error (i.e. (n, nil) where 0 < n < srcs.NumBytes()). It should not +// return a full write with an error (i.e. srcs.NumBytes(), err) where err +// != nil). +func (w *bufferWriter) WriteFromBlocks(srcs safemem.BlockSeq) (uint64, error) { + written := srcs.NumBytes() + for !srcs.IsEmpty() { + w.buf.Write(srcs.Head().ToSlice()) + srcs = srcs.Tail() + } + return written, nil +} + +// auxvData implements vfs.DynamicBytesSource for /proc/[pid]/auxv. +// +// +stateify savable +type auxvData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*auxvData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *auxvData) Generate(ctx context.Context, buf *bytes.Buffer) error { + m, err := getMMIncRef(d.task) + if err != nil { + return err + } + defer m.DecUsers(ctx) + + // Space for buffer with AT_NULL (0) terminator at the end. + auxv := m.Auxv() + buf.Grow((len(auxv) + 1) * 16) + for _, e := range auxv { + var tmp [8]byte + usermem.ByteOrder.PutUint64(tmp[:], e.Key) + buf.Write(tmp[:]) + + usermem.ByteOrder.PutUint64(tmp[:], uint64(e.Value)) + buf.Write(tmp[:]) + } + return nil +} + +// execArgType enumerates the types of exec arguments that are exposed through +// proc. +type execArgType int + +const ( + cmdlineDataArg execArgType = iota + environDataArg +) + +// cmdlineData implements vfs.DynamicBytesSource for /proc/[pid]/cmdline. +// +// +stateify savable +type cmdlineData struct { + kernfs.DynamicBytesFile + + task *kernel.Task + + // arg is the type of exec argument this file contains. + arg execArgType +} + +var _ dynamicInode = (*cmdlineData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *cmdlineData) Generate(ctx context.Context, buf *bytes.Buffer) error { + m, err := getMMIncRef(d.task) + if err != nil { + return err + } + defer m.DecUsers(ctx) + + // Figure out the bounds of the exec arg we are trying to read. + var ar usermem.AddrRange + switch d.arg { + case cmdlineDataArg: + ar = usermem.AddrRange{ + Start: m.ArgvStart(), + End: m.ArgvEnd(), + } + case environDataArg: + ar = usermem.AddrRange{ + Start: m.EnvvStart(), + End: m.EnvvEnd(), + } + default: + panic(fmt.Sprintf("unknown exec arg type %v", d.arg)) + } + if ar.Start == 0 || ar.End == 0 { + // Don't attempt to read before the start/end are set up. + return io.EOF + } + + // N.B. Technically this should be usermem.IOOpts.IgnorePermissions = true + // until Linux 4.9 (272ddc8b3735 "proc: don't use FOLL_FORCE for reading + // cmdline and environment"). + writer := &bufferWriter{buf: buf} + if n, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(ar), writer, usermem.IOOpts{}); n == 0 || err != nil { + // Nothing to copy or something went wrong. + return err + } + + // On Linux, if the NULL byte at the end of the argument vector has been + // overwritten, it continues reading the environment vector as part of + // the argument vector. + if d.arg == cmdlineDataArg && buf.Bytes()[buf.Len()-1] != 0 { + if end := bytes.IndexByte(buf.Bytes(), 0); end != -1 { + // If we found a NULL character somewhere else in argv, truncate the + // return up to the NULL terminator (including it). + buf.Truncate(end) + return nil + } + + // There is no NULL terminator in the string, return into envp. + arEnvv := usermem.AddrRange{ + Start: m.EnvvStart(), + End: m.EnvvEnd(), + } + + // Upstream limits the returned amount to one page of slop. + // https://elixir.bootlin.com/linux/v4.20/source/fs/proc/base.c#L208 + // we'll return one page total between argv and envp because of the + // above page restrictions. + if buf.Len() >= usermem.PageSize { + // Returned at least one page already, nothing else to add. + return nil + } + remaining := usermem.PageSize - buf.Len() + if int(arEnvv.Length()) > remaining { + end, ok := arEnvv.Start.AddLength(uint64(remaining)) + if !ok { + return syserror.EFAULT + } + arEnvv.End = end + } + if _, err := m.CopyInTo(ctx, usermem.AddrRangeSeqOf(arEnvv), writer, usermem.IOOpts{}); err != nil { + return err + } + + // Linux will return envp up to and including the first NULL character, + // so find it. + if end := bytes.IndexByte(buf.Bytes()[ar.Length():], 0); end != -1 { + buf.Truncate(end) + } + } + + return nil +} + +// +stateify savable +type commInode struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +func newComm(task *kernel.Task, ino uint64, perm linux.FileMode) *kernfs.Dentry { + inode := &commInode{task: task} + inode.DynamicBytesFile.Init(task.Credentials(), ino, &commData{task: task}, perm) + + d := &kernfs.Dentry{} + d.Init(inode) + return d +} + +func (i *commInode) CheckPermissions(ctx context.Context, creds *auth.Credentials, ats vfs.AccessTypes) error { + // This file can always be read or written by members of the same thread + // group. See fs/proc/base.c:proc_tid_comm_permission. + // + // N.B. This check is currently a no-op as we don't yet support writing and + // this file is world-readable anyways. + t := kernel.TaskFromContext(ctx) + if t != nil && t.ThreadGroup() == i.task.ThreadGroup() && !ats.MayExec() { + return nil + } + + return i.DynamicBytesFile.CheckPermissions(ctx, creds, ats) +} + +// commData implements vfs.DynamicBytesSource for /proc/[pid]/comm. +// +// +stateify savable +type commData struct { + kernfs.DynamicBytesFile + + task *kernel.Task +} + +var _ dynamicInode = (*commData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *commData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(d.task.Name()) + buf.WriteString("\n") + return nil +} + +// idMapData implements vfs.DynamicBytesSource for /proc/[pid]/{gid_map|uid_map}. +// +// +stateify savable +type idMapData struct { + kernfs.DynamicBytesFile + + task *kernel.Task + gids bool +} + +var _ dynamicInode = (*idMapData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *idMapData) Generate(ctx context.Context, buf *bytes.Buffer) error { + var entries []auth.IDMapEntry + if d.gids { + entries = d.task.UserNamespace().GIDMap() + } else { + entries = d.task.UserNamespace().UIDMap() + } + for _, e := range entries { + fmt.Fprintf(buf, "%10d %10d %10d\n", e.FirstID, e.FirstParentID, e.Length) + } + return nil +} + // mapsData implements vfs.DynamicBytesSource for /proc/[pid]/maps. // // +stateify savable @@ -83,7 +338,7 @@ func (d *smapsData) Generate(ctx context.Context, buf *bytes.Buffer) error { type taskStatData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task // If tgstats is true, accumulate fault stats (not implemented) and CPU // time across all tasks in t's thread group. @@ -98,40 +353,40 @@ var _ dynamicInode = (*taskStatData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.t)) - fmt.Fprintf(buf, "(%s) ", s.t.Name()) - fmt.Fprintf(buf, "%c ", s.t.StateStatus()[0]) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfTask(s.task)) + fmt.Fprintf(buf, "(%s) ", s.task.Name()) + fmt.Fprintf(buf, "%c ", s.task.StateStatus()[0]) ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { + if parent := s.task.Parent(); parent != nil { ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) } fmt.Fprintf(buf, "%d ", ppid) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.t.ThreadGroup().ProcessGroup())) - fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.t.ThreadGroup().Session())) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfProcessGroup(s.task.ThreadGroup().ProcessGroup())) + fmt.Fprintf(buf, "%d ", s.pidns.IDOfSession(s.task.ThreadGroup().Session())) fmt.Fprintf(buf, "0 0 " /* tty_nr tpgid */) fmt.Fprintf(buf, "0 " /* flags */) fmt.Fprintf(buf, "0 0 0 0 " /* minflt cminflt majflt cmajflt */) var cputime usage.CPUStats if s.tgstats { - cputime = s.t.ThreadGroup().CPUStats() + cputime = s.task.ThreadGroup().CPUStats() } else { - cputime = s.t.CPUStats() + cputime = s.task.CPUStats() } fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - cputime = s.t.ThreadGroup().JoinedChildCPUStats() + cputime = s.task.ThreadGroup().JoinedChildCPUStats() fmt.Fprintf(buf, "%d %d ", linux.ClockTFromDuration(cputime.UserTime), linux.ClockTFromDuration(cputime.SysTime)) - fmt.Fprintf(buf, "%d %d ", s.t.Priority(), s.t.Niceness()) - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Count()) + fmt.Fprintf(buf, "%d %d ", s.task.Priority(), s.task.Niceness()) + fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Count()) // itrealvalue. Since kernel 2.6.17, this field is no longer // maintained, and is hard coded as 0. fmt.Fprintf(buf, "0 ") // Start time is relative to boot time, expressed in clock ticks. - fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.t.StartTime().Sub(s.t.Kernel().Timekeeper().BootTime()))) + fmt.Fprintf(buf, "%d ", linux.ClockTFromDuration(s.task.StartTime().Sub(s.task.Kernel().Timekeeper().BootTime()))) var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if mm := t.MemoryManager(); mm != nil { vss = mm.VirtualMemorySize() rss = mm.ResidentSetSize() @@ -140,14 +395,14 @@ func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "%d %d ", vss, rss/usermem.PageSize) // rsslim. - fmt.Fprintf(buf, "%d ", s.t.ThreadGroup().Limits().Get(limits.Rss).Cur) + fmt.Fprintf(buf, "%d ", s.task.ThreadGroup().Limits().Get(limits.Rss).Cur) fmt.Fprintf(buf, "0 0 0 0 0 " /* startcode endcode startstack kstkesp kstkeip */) fmt.Fprintf(buf, "0 0 0 0 0 " /* signal blocked sigignore sigcatch wchan */) fmt.Fprintf(buf, "0 0 " /* nswap cnswap */) terminationSignal := linux.Signal(0) - if s.t == s.t.ThreadGroup().Leader() { - terminationSignal = s.t.ThreadGroup().TerminationSignal() + if s.task == s.task.ThreadGroup().Leader() { + terminationSignal = s.task.ThreadGroup().TerminationSignal() } fmt.Fprintf(buf, "%d ", terminationSignal) fmt.Fprintf(buf, "0 0 0 " /* processor rt_priority policy */) @@ -164,7 +419,7 @@ func (s *taskStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { type statmData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task } var _ dynamicInode = (*statmData)(nil) @@ -172,7 +427,7 @@ var _ dynamicInode = (*statmData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { var vss, rss uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if mm := t.MemoryManager(); mm != nil { vss = mm.VirtualMemorySize() rss = mm.ResidentSetSize() @@ -189,7 +444,7 @@ func (s *statmData) Generate(ctx context.Context, buf *bytes.Buffer) error { type statusData struct { kernfs.DynamicBytesFile - t *kernel.Task + task *kernel.Task pidns *kernel.PIDNamespace } @@ -197,23 +452,23 @@ var _ dynamicInode = (*statusData)(nil) // Generate implements vfs.DynamicBytesSource.Generate. func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { - fmt.Fprintf(buf, "Name:\t%s\n", s.t.Name()) - fmt.Fprintf(buf, "State:\t%s\n", s.t.StateStatus()) - fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.t.ThreadGroup())) - fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.t)) + fmt.Fprintf(buf, "Name:\t%s\n", s.task.Name()) + fmt.Fprintf(buf, "State:\t%s\n", s.task.StateStatus()) + fmt.Fprintf(buf, "Tgid:\t%d\n", s.pidns.IDOfThreadGroup(s.task.ThreadGroup())) + fmt.Fprintf(buf, "Pid:\t%d\n", s.pidns.IDOfTask(s.task)) ppid := kernel.ThreadID(0) - if parent := s.t.Parent(); parent != nil { + if parent := s.task.Parent(); parent != nil { ppid = s.pidns.IDOfThreadGroup(parent.ThreadGroup()) } fmt.Fprintf(buf, "PPid:\t%d\n", ppid) tpid := kernel.ThreadID(0) - if tracer := s.t.Tracer(); tracer != nil { + if tracer := s.task.Tracer(); tracer != nil { tpid = s.pidns.IDOfTask(tracer) } fmt.Fprintf(buf, "TracerPid:\t%d\n", tpid) var fds int var vss, rss, data uint64 - s.t.WithMuLocked(func(t *kernel.Task) { + s.task.WithMuLocked(func(t *kernel.Task) { if fdTable := t.FDTable(); fdTable != nil { fds = fdTable.Size() } @@ -227,13 +482,13 @@ func (s *statusData) Generate(ctx context.Context, buf *bytes.Buffer) error { fmt.Fprintf(buf, "VmSize:\t%d kB\n", vss>>10) fmt.Fprintf(buf, "VmRSS:\t%d kB\n", rss>>10) fmt.Fprintf(buf, "VmData:\t%d kB\n", data>>10) - fmt.Fprintf(buf, "Threads:\t%d\n", s.t.ThreadGroup().Count()) - creds := s.t.Credentials() + fmt.Fprintf(buf, "Threads:\t%d\n", s.task.ThreadGroup().Count()) + creds := s.task.Credentials() fmt.Fprintf(buf, "CapInh:\t%016x\n", creds.InheritableCaps) fmt.Fprintf(buf, "CapPrm:\t%016x\n", creds.PermittedCaps) fmt.Fprintf(buf, "CapEff:\t%016x\n", creds.EffectiveCaps) fmt.Fprintf(buf, "CapBnd:\t%016x\n", creds.BoundingCaps) - fmt.Fprintf(buf, "Seccomp:\t%d\n", s.t.SeccompMode()) + fmt.Fprintf(buf, "Seccomp:\t%d\n", s.task.SeccompMode()) // We unconditionally report a single NUMA node. See // pkg/sentry/syscalls/linux/sys_mempolicy.go. fmt.Fprintf(buf, "Mems_allowed:\t1\n") diff --git a/pkg/sentry/fsimpl/proc/tasks.go b/pkg/sentry/fsimpl/proc/tasks.go index d8f92d52f..51f634716 100644 --- a/pkg/sentry/fsimpl/proc/tasks.go +++ b/pkg/sentry/fsimpl/proc/tasks.go @@ -15,6 +15,7 @@ package proc import ( + "bytes" "sort" "strconv" @@ -28,9 +29,8 @@ import ( ) const ( - defaultPermission = 0444 - selfName = "self" - threadSelfName = "thread-self" + selfName = "self" + threadSelfName = "thread-self" ) // InoGenerator generates unique inode numbers for a given filesystem. @@ -54,22 +54,28 @@ type tasksInode struct { // Linux. So handle them outside of OrderedChildren. selfSymlink *vfs.Dentry threadSelfSymlink *vfs.Dentry + + // cgroupControllers is a map of controller name to directory in the + // cgroup hierarchy. These controllers are immutable and will be listed + // in /proc/pid/cgroup if not nil. + cgroupControllers map[string]string } var _ kernfs.Inode = (*tasksInode)(nil) -func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace) (*tasksInode, *kernfs.Dentry) { +func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNamespace, cgroupControllers map[string]string) (*tasksInode, *kernfs.Dentry) { root := auth.NewRootCredentials(pidns.UserNamespace()) contents := map[string]*kernfs.Dentry{ - //"cpuinfo": newCPUInfo(ctx, msrc), - //"filesystems": seqfile.NewSeqFileInode(ctx, &filesystemsData{}, msrc), - "loadavg": newDentry(root, inoGen.NextIno(), defaultPermission, &loadavgData{}), - "meminfo": newDentry(root, inoGen.NextIno(), defaultPermission, &meminfoData{k: k}), - "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), defaultPermission, "self/mounts"), - "stat": newDentry(root, inoGen.NextIno(), defaultPermission, &statData{k: k}), - //"uptime": newUptime(ctx, msrc), - //"version": newVersionData(root, inoGen.NextIno(), k), - "version": newDentry(root, inoGen.NextIno(), defaultPermission, &versionData{k: k}), + "cpuinfo": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(cpuInfoData(k))), + //"filesystems": newDentry(root, inoGen.NextIno(), 0444, &filesystemsData{}), + "loadavg": newDentry(root, inoGen.NextIno(), 0444, &loadavgData{}), + "sys": newSysDir(root, inoGen), + "meminfo": newDentry(root, inoGen.NextIno(), 0444, &meminfoData{}), + "mounts": kernfs.NewStaticSymlink(root, inoGen.NextIno(), "self/mounts"), + "net": newNetDir(root, inoGen, k), + "stat": newDentry(root, inoGen.NextIno(), 0444, &statData{}), + "uptime": newDentry(root, inoGen.NextIno(), 0444, &uptimeData{}), + "version": newDentry(root, inoGen.NextIno(), 0444, &versionData{}), } inode := &tasksInode{ @@ -77,6 +83,7 @@ func newTasksInode(inoGen InoGenerator, k *kernel.Kernel, pidns *kernel.PIDNames inoGen: inoGen, selfSymlink: newSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), threadSelfSymlink: newThreadSelfSymlink(root, inoGen.NextIno(), 0444, pidns).VFSDentry(), + cgroupControllers: cgroupControllers, } inode.InodeAttrs.Init(root, inoGen.NextIno(), linux.ModeDirectory|0555) @@ -110,7 +117,7 @@ func (i *tasksInode) Lookup(ctx context.Context, name string) (*vfs.Dentry, erro return nil, syserror.ENOENT } - taskDentry := newTaskInode(i.inoGen, task, i.pidns, true) + taskDentry := newTaskInode(i.inoGen, task, i.pidns, true, i.cgroupControllers) return taskDentry.VFSDentry(), nil } @@ -216,3 +223,20 @@ func (i *tasksInode) Stat(vsfs *vfs.Filesystem) linux.Statx { return stat } + +func cpuInfoData(k *kernel.Kernel) string { + features := k.FeatureSet() + if features == nil { + // Kernel is always initialized with a FeatureSet. + panic("cpuinfo read with nil FeatureSet") + } + var buf bytes.Buffer + for i, max := uint(0), k.ApplicationCores(); i < max; i++ { + features.WriteCPUInfoTo(i, &buf) + } + return buf.String() +} + +func shmData(v uint64) dynamicInode { + return newStaticFile(strconv.FormatUint(v, 10)) +} diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 91f30a798..ad3760e39 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -15,6 +15,7 @@ package proc import ( + "bytes" "fmt" "strconv" @@ -23,6 +24,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/time" + "gvisor.dev/gvisor/pkg/sentry/usage" + "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" ) @@ -90,3 +94,244 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { } return fmt.Sprintf("%d/task/%d", tgid, tid), nil } + +// cpuStats contains the breakdown of CPU time for /proc/stat. +type cpuStats struct { + // user is time spent in userspace tasks with non-positive niceness. + user uint64 + + // nice is time spent in userspace tasks with positive niceness. + nice uint64 + + // system is time spent in non-interrupt kernel context. + system uint64 + + // idle is time spent idle. + idle uint64 + + // ioWait is time spent waiting for IO. + ioWait uint64 + + // irq is time spent in interrupt context. + irq uint64 + + // softirq is time spent in software interrupt context. + softirq uint64 + + // steal is involuntary wait time. + steal uint64 + + // guest is time spent in guests with non-positive niceness. + guest uint64 + + // guestNice is time spent in guests with positive niceness. + guestNice uint64 +} + +// String implements fmt.Stringer. +func (c cpuStats) String() string { + return fmt.Sprintf("%d %d %d %d %d %d %d %d %d %d", c.user, c.nice, c.system, c.idle, c.ioWait, c.irq, c.softirq, c.steal, c.guest, c.guestNice) +} + +// statData implements vfs.DynamicBytesSource for /proc/stat. +// +// +stateify savable +type statData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*statData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (s *statData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // TODO(b/37226836): We currently export only zero CPU stats. We could + // at least provide some aggregate stats. + var cpu cpuStats + fmt.Fprintf(buf, "cpu %s\n", cpu) + + for c, max := uint(0), s.k.ApplicationCores(); c < max; c++ { + fmt.Fprintf(buf, "cpu%d %s\n", c, cpu) + } + + // The total number of interrupts is dependent on the CPUs and PCI + // devices on the system. See arch_probe_nr_irqs. + // + // Since we don't report real interrupt stats, just choose an arbitrary + // value from a representative VM. + const numInterrupts = 256 + + // The Kernel doesn't handle real interrupts, so report all zeroes. + // TODO(b/37226836): We could count page faults as #PF. + fmt.Fprintf(buf, "intr 0") // total + for i := 0; i < numInterrupts; i++ { + fmt.Fprintf(buf, " 0") + } + fmt.Fprintf(buf, "\n") + + // Total number of context switches. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "ctxt 0\n") + + // CLOCK_REALTIME timestamp from boot, in seconds. + fmt.Fprintf(buf, "btime %d\n", s.k.Timekeeper().BootTime().Seconds()) + + // Total number of clones. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "processes 0\n") + + // Number of runnable tasks. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "procs_running 0\n") + + // Number of tasks waiting on IO. + // TODO(b/37226836): Count this. + fmt.Fprintf(buf, "procs_blocked 0\n") + + // Number of each softirq handled. + fmt.Fprintf(buf, "softirq 0") // total + for i := 0; i < linux.NumSoftIRQ; i++ { + fmt.Fprintf(buf, " 0") + } + fmt.Fprintf(buf, "\n") + return nil +} + +// loadavgData backs /proc/loadavg. +// +// +stateify savable +type loadavgData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*loadavgData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *loadavgData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // TODO(b/62345059): Include real data in fields. + // Column 1-3: CPU and IO utilization of the last 1, 5, and 10 minute periods. + // Column 4-5: currently running processes and the total number of processes. + // Column 6: the last process ID used. + fmt.Fprintf(buf, "%.2f %.2f %.2f %d/%d %d\n", 0.00, 0.00, 0.00, 0, 0, 0) + return nil +} + +// meminfoData implements vfs.DynamicBytesSource for /proc/meminfo. +// +// +stateify savable +type meminfoData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*meminfoData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *meminfoData) Generate(ctx context.Context, buf *bytes.Buffer) error { + mf := d.k.MemoryFile() + mf.UpdateUsage() + snapshot, totalUsage := usage.MemoryAccounting.Copy() + totalSize := usage.TotalMemory(mf.TotalSize(), totalUsage) + anon := snapshot.Anonymous + snapshot.Tmpfs + file := snapshot.PageCache + snapshot.Mapped + // We don't actually have active/inactive LRUs, so just make up numbers. + activeFile := (file / 2) &^ (usermem.PageSize - 1) + inactiveFile := file - activeFile + + fmt.Fprintf(buf, "MemTotal: %8d kB\n", totalSize/1024) + memFree := (totalSize - totalUsage) / 1024 + // We use MemFree as MemAvailable because we don't swap. + // TODO(rahat): When reclaim is implemented the value of MemAvailable + // should change. + fmt.Fprintf(buf, "MemFree: %8d kB\n", memFree) + fmt.Fprintf(buf, "MemAvailable: %8d kB\n", memFree) + fmt.Fprintf(buf, "Buffers: 0 kB\n") // memory usage by block devices + fmt.Fprintf(buf, "Cached: %8d kB\n", (file+snapshot.Tmpfs)/1024) + // Emulate a system with no swap, which disables inactivation of anon pages. + fmt.Fprintf(buf, "SwapCache: 0 kB\n") + fmt.Fprintf(buf, "Active: %8d kB\n", (anon+activeFile)/1024) + fmt.Fprintf(buf, "Inactive: %8d kB\n", inactiveFile/1024) + fmt.Fprintf(buf, "Active(anon): %8d kB\n", anon/1024) + fmt.Fprintf(buf, "Inactive(anon): 0 kB\n") + fmt.Fprintf(buf, "Active(file): %8d kB\n", activeFile/1024) + fmt.Fprintf(buf, "Inactive(file): %8d kB\n", inactiveFile/1024) + fmt.Fprintf(buf, "Unevictable: 0 kB\n") // TODO(b/31823263) + fmt.Fprintf(buf, "Mlocked: 0 kB\n") // TODO(b/31823263) + fmt.Fprintf(buf, "SwapTotal: 0 kB\n") + fmt.Fprintf(buf, "SwapFree: 0 kB\n") + fmt.Fprintf(buf, "Dirty: 0 kB\n") + fmt.Fprintf(buf, "Writeback: 0 kB\n") + fmt.Fprintf(buf, "AnonPages: %8d kB\n", anon/1024) + fmt.Fprintf(buf, "Mapped: %8d kB\n", file/1024) // doesn't count mapped tmpfs, which we don't know + fmt.Fprintf(buf, "Shmem: %8d kB\n", snapshot.Tmpfs/1024) + return nil +} + +// uptimeData implements vfs.DynamicBytesSource for /proc/uptime. +// +// +stateify savable +type uptimeData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*uptimeData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (*uptimeData) Generate(ctx context.Context, buf *bytes.Buffer) error { + k := kernel.KernelFromContext(ctx) + now := time.NowFromContext(ctx) + + // Pretend that we've spent zero time sleeping (second number). + fmt.Fprintf(buf, "%.2f 0.00\n", now.Sub(k.Timekeeper().BootTime()).Seconds()) + return nil +} + +// versionData implements vfs.DynamicBytesSource for /proc/version. +// +// +stateify savable +type versionData struct { + kernfs.DynamicBytesFile + + // k is the owning Kernel. + k *kernel.Kernel +} + +var _ dynamicInode = (*versionData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { + init := v.k.GlobalInit() + if init == nil { + // Attempted to read before the init Task is created. This can + // only occur during startup, which should never need to read + // this file. + panic("Attempted to read version before initial Task is available") + } + + // /proc/version takes the form: + // + // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) + // (COMPILER_VERSION) VERSION" + // + // where: + // - SYSNAME, RELEASE, and VERSION are the same as returned by + // sys_utsname + // - COMPILE_USER is the user that build the kernel + // - COMPILE_HOST is the hostname of the machine on which the kernel + // was built + // - COMPILER_VERSION is the version reported by the building compiler + // + // Since we don't really want to expose build information to + // applications, those fields are omitted. + // + // FIXME(mpratt): Using Version from the init task SyscallTable + // disregards the different version a task may have (e.g., in a uts + // namespace). + ver := init.Leader().SyscallTable().Version + fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_net.go b/pkg/sentry/fsimpl/proc/tasks_net.go new file mode 100644 index 000000000..3dbf3ba41 --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_net.go @@ -0,0 +1,784 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + "io" + "reflect" + "time" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/inet" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/unix" + "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" + "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip/header" +) + +func newNetDir(root *auth.Credentials, inoGen InoGenerator, k *kernel.Kernel) *kernfs.Dentry { + var contents map[string]*kernfs.Dentry + if stack := k.NetworkStack(); stack != nil { + const ( + arp = "IP address HW type Flags HW address Mask Device" + netlink = "sk Eth Pid Groups Rmem Wmem Dump Locks Drops Inode" + packet = "sk RefCnt Type Proto Iface R Rmem User Inode" + protocols = "protocol size sockets memory press maxhdr slab module cl co di ac io in de sh ss gs se re sp bi br ha uh gp em" + ptype = "Type Device Function" + upd6 = " sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode" + ) + psched := fmt.Sprintf("%08x %08x %08x %08x\n", uint64(time.Microsecond/time.Nanosecond), 64, 1000000, uint64(time.Second/time.Nanosecond)) + + contents = map[string]*kernfs.Dentry{ + "dev": newDentry(root, inoGen.NextIno(), 0444, &netDevData{stack: stack}), + "snmp": newDentry(root, inoGen.NextIno(), 0444, &netSnmpData{stack: stack}), + + // The following files are simple stubs until they are implemented in + // netstack, if the file contains a header the stub is just the header + // otherwise it is an empty file. + "arp": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(arp)), + "netlink": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(netlink)), + "netstat": newDentry(root, inoGen.NextIno(), 0444, &netStatData{}), + "packet": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(packet)), + "protocols": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(protocols)), + + // Linux sets psched values to: nsec per usec, psched tick in ns, 1000000, + // high res timer ticks per sec (ClockGetres returns 1ns resolution). + "psched": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(psched)), + "ptype": newDentry(root, inoGen.NextIno(), 0444, newStaticFile(ptype)), + "route": newDentry(root, inoGen.NextIno(), 0444, &netRouteData{stack: stack}), + "tcp": newDentry(root, inoGen.NextIno(), 0444, &netTCPData{kernel: k}), + "udp": newDentry(root, inoGen.NextIno(), 0444, &netUDPData{kernel: k}), + "unix": newDentry(root, inoGen.NextIno(), 0444, &netUnixData{kernel: k}), + } + + if stack.SupportsIPv6() { + contents["if_inet6"] = newDentry(root, inoGen.NextIno(), 0444, &ifinet6{stack: stack}) + contents["ipv6_route"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")) + contents["tcp6"] = newDentry(root, inoGen.NextIno(), 0444, &netTCP6Data{kernel: k}) + contents["udp6"] = newDentry(root, inoGen.NextIno(), 0444, newStaticFile(upd6)) + } + } + + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, contents) +} + +// ifinet6 implements vfs.DynamicBytesSource for /proc/net/if_inet6. +// +// +stateify savable +type ifinet6 struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*ifinet6)(nil) + +func (n *ifinet6) contents() []string { + var lines []string + nics := n.stack.Interfaces() + for id, naddrs := range n.stack.InterfaceAddrs() { + nic, ok := nics[id] + if !ok { + // NIC was added after NICNames was called. We'll just ignore it. + continue + } + + for _, a := range naddrs { + // IPv6 only. + if a.Family != linux.AF_INET6 { + continue + } + + // Fields: + // IPv6 address displayed in 32 hexadecimal chars without colons + // Netlink device number (interface index) in hexadecimal (use nic id) + // Prefix length in hexadecimal + // Scope value (use 0) + // Interface flags + // Device name + lines = append(lines, fmt.Sprintf("%032x %02x %02x %02x %02x %8s\n", a.Addr, id, a.PrefixLen, 0, a.Flags, nic.Name)) + } + } + return lines +} + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *ifinet6) Generate(ctx context.Context, buf *bytes.Buffer) error { + for _, l := range n.contents() { + buf.WriteString(l) + } + return nil +} + +// netDevData implements vfs.DynamicBytesSource for /proc/net/dev. +// +// +stateify savable +type netDevData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netDevData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netDevData) Generate(ctx context.Context, buf *bytes.Buffer) error { + interfaces := n.stack.Interfaces() + buf.WriteString("Inter-| Receive | Transmit\n") + buf.WriteString(" face |bytes packets errs drop fifo frame compressed multicast|bytes packets errs drop fifo colls carrier compressed\n") + + for _, i := range interfaces { + // Implements the same format as + // net/core/net-procfs.c:dev_seq_printf_stats. + var stats inet.StatDev + if err := n.stack.Statistics(&stats, i.Name); err != nil { + log.Warningf("Failed to retrieve interface statistics for %v: %v", i.Name, err) + continue + } + fmt.Fprintf( + buf, + "%6s: %7d %7d %4d %4d %4d %5d %10d %9d %8d %7d %4d %4d %4d %5d %7d %10d\n", + i.Name, + // Received + stats[0], // bytes + stats[1], // packets + stats[2], // errors + stats[3], // dropped + stats[4], // fifo + stats[5], // frame + stats[6], // compressed + stats[7], // multicast + // Transmitted + stats[8], // bytes + stats[9], // packets + stats[10], // errors + stats[11], // dropped + stats[12], // fifo + stats[13], // frame + stats[14], // compressed + stats[15], // multicast + ) + } + + return nil +} + +// netUnixData implements vfs.DynamicBytesSource for /proc/net/unix. +// +// +stateify savable +type netUnixData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netUnixData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (n *netUnixData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString("Num RefCount Protocol Flags Type St Inode Path\n") + for _, se := range n.kernel.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref %v in socket table, racing with destruction?", se.Sock) + continue + } + sfile := s.(*fs.File) + if family, _, _ := sfile.FileOperations.(socket.Socket).Type(); family != linux.AF_UNIX { + s.DecRef() + // Not a unix socket. + continue + } + sops := sfile.FileOperations.(*unix.SocketOperations) + + addr, err := sops.Endpoint().GetLocalAddress() + if err != nil { + log.Warningf("Failed to retrieve socket name from %+v: %v", sfile, err) + addr.Addr = "<unknown>" + } + + sockFlags := 0 + if ce, ok := sops.Endpoint().(transport.ConnectingEndpoint); ok { + if ce.Listening() { + // For unix domain sockets, linux reports a single flag + // value if the socket is listening, of __SO_ACCEPTCON. + sockFlags = linux.SO_ACCEPTCON + } + } + + // In the socket entry below, the value for the 'Num' field requires + // some consideration. Linux prints the address to the struct + // unix_sock representing a socket in the kernel, but may redact the + // value for unprivileged users depending on the kptr_restrict + // sysctl. + // + // One use for this field is to allow a privileged user to + // introspect into the kernel memory to determine information about + // a socket not available through procfs, such as the socket's peer. + // + // In gvisor, returning a pointer to our internal structures would + // be pointless, as it wouldn't match the memory layout for struct + // unix_sock, making introspection difficult. We could populate a + // struct unix_sock with the appropriate data, but even that + // requires consideration for which kernel version to emulate, as + // the definition of this struct changes over time. + // + // For now, we always redact this pointer. + fmt.Fprintf(buf, "%#016p: %08X %08X %08X %04X %02X %5d", + (*unix.SocketOperations)(nil), // Num, pointer to kernel socket struct. + sfile.ReadRefs()-1, // RefCount, don't count our own ref. + 0, // Protocol, always 0 for UDS. + sockFlags, // Flags. + sops.Endpoint().Type(), // Type. + sops.State(), // State. + sfile.InodeID(), // Inode. + ) + + // Path + if len(addr.Addr) != 0 { + if addr.Addr[0] == 0 { + // Abstract path. + fmt.Fprintf(buf, " @%s", string(addr.Addr[1:])) + } else { + fmt.Fprintf(buf, " %s", string(addr.Addr)) + } + } + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +func networkToHost16(n uint16) uint16 { + // n is in network byte order, so is big-endian. The most-significant byte + // should be stored in the lower address. + // + // We manually inline binary.BigEndian.Uint16() because Go does not support + // non-primitive consts, so binary.BigEndian is a (mutable) var, so calls to + // binary.BigEndian.Uint16() require a read of binary.BigEndian and an + // interface method call, defeating inlining. + buf := [2]byte{byte(n >> 8 & 0xff), byte(n & 0xff)} + return usermem.ByteOrder.Uint16(buf[:]) +} + +func writeInetAddr(w io.Writer, family int, i linux.SockAddr) { + switch family { + case linux.AF_INET: + var a linux.SockAddrInet + if i != nil { + a = *i.(*linux.SockAddrInet) + } + + // linux.SockAddrInet.Port is stored in the network byte order and is + // printed like a number in host byte order. Note that all numbers in host + // byte order are printed with the most-significant byte first when + // formatted with %X. See get_tcp4_sock() and udp4_format_sock() in Linux. + port := networkToHost16(a.Port) + + // linux.SockAddrInet.Addr is stored as a byte slice in big-endian order + // (i.e. most-significant byte in index 0). Linux represents this as a + // __be32 which is a typedef for an unsigned int, and is printed with + // %X. This means that for a little-endian machine, Linux prints the + // least-significant byte of the address first. To emulate this, we first + // invert the byte order for the address using usermem.ByteOrder.Uint32, + // which makes it have the equivalent encoding to a __be32 on a little + // endian machine. Note that this operation is a no-op on a big endian + // machine. Then similar to Linux, we format it with %X, which will print + // the most-significant byte of the __be32 address first, which is now + // actually the least-significant byte of the original address in + // linux.SockAddrInet.Addr on little endian machines, due to the conversion. + addr := usermem.ByteOrder.Uint32(a.Addr[:]) + + fmt.Fprintf(w, "%08X:%04X ", addr, port) + case linux.AF_INET6: + var a linux.SockAddrInet6 + if i != nil { + a = *i.(*linux.SockAddrInet6) + } + + port := networkToHost16(a.Port) + addr0 := usermem.ByteOrder.Uint32(a.Addr[0:4]) + addr1 := usermem.ByteOrder.Uint32(a.Addr[4:8]) + addr2 := usermem.ByteOrder.Uint32(a.Addr[8:12]) + addr3 := usermem.ByteOrder.Uint32(a.Addr[12:16]) + fmt.Fprintf(w, "%08X%08X%08X%08X:%04X ", addr0, addr1, addr2, addr3, port) + } +} + +func commonGenerateTCP(ctx context.Context, buf *bytes.Buffer, k *kernel.Kernel, family int) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + for _, se := range k.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if fa, stype, _ := sops.Type(); !(family == fa && stype == linux.SOCK_STREAM) { + s.DecRef() + // Not tcp4 sockets. + continue + } + + // Linux's documentation for the fields below can be found at + // https://www.kernel.org/doc/Documentation/networking/proc_net_tcp.txt. + // For Linux's implementation, see net/ipv4/tcp_ipv4.c:get_tcp4_sock(). + // Note that the header doesn't contain labels for all the fields. + + // Field: sl; entry number. + fmt.Fprintf(buf, "%4d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddr + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = local + } + } + writeInetAddr(buf, family, localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddr + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = remote + } + } + writeInetAddr(buf, family, remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when; timer active state and number of jiffies + // until timer expires. Unimplemented. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt; number of unrecovered RTO timeouts. + // Unimplemented. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout; number of unanswered 0-window probes. + // Unimplemented. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: refcount. Don't count the ref we obtain while deferencing + // the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: retransmit timeout. Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: predicted tick of soft clock (delayed ACK control data). + // Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: (ack.quick<<1)|ack.pingpong, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: sending congestion window, Unimplemented. + fmt.Fprintf(buf, "%d ", 0) + + // Field: Slow start size threshold, -1 if threshold >= 0xFFFF. + // Unimplemented, report as large threshold. + fmt.Fprintf(buf, "%d", -1) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + + return nil +} + +// netTCPData implements vfs.DynamicBytesSource for /proc/net/tcp. +// +// +stateify savable +type netTCPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCPData)(nil) + +func (d *netTCPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address rem_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode \n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET) +} + +// netTCP6Data implements vfs.DynamicBytesSource for /proc/net/tcp6. +// +// +stateify savable +type netTCP6Data struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netTCP6Data)(nil) + +func (d *netTCP6Data) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString(" sl local_address remote_address st tx_queue rx_queue tr tm->when retrnsmt uid timeout inode\n") + return commonGenerateTCP(ctx, buf, d.kernel, linux.AF_INET6) +} + +// netUDPData implements vfs.DynamicBytesSource for /proc/net/udp. +// +// +stateify savable +type netUDPData struct { + kernfs.DynamicBytesFile + + kernel *kernel.Kernel +} + +var _ dynamicInode = (*netUDPData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *netUDPData) Generate(ctx context.Context, buf *bytes.Buffer) error { + // t may be nil here if our caller is not part of a task goroutine. This can + // happen for example if we're here for "sentryctl cat". When t is nil, + // degrade gracefully and retrieve what we can. + t := kernel.TaskFromContext(ctx) + + for _, se := range d.kernel.ListSockets() { + s := se.Sock.Get() + if s == nil { + log.Debugf("Couldn't resolve weakref with ID %v in socket table, racing with destruction?", se.ID) + continue + } + sfile := s.(*fs.File) + sops, ok := sfile.FileOperations.(socket.Socket) + if !ok { + panic(fmt.Sprintf("Found non-socket file in socket table: %+v", sfile)) + } + if family, stype, _ := sops.Type(); family != linux.AF_INET || stype != linux.SOCK_DGRAM { + s.DecRef() + // Not udp4 socket. + continue + } + + // For Linux's implementation, see net/ipv4/udp.c:udp4_format_sock(). + + // Field: sl; entry number. + fmt.Fprintf(buf, "%5d: ", se.ID) + + // Field: local_adddress. + var localAddr linux.SockAddrInet + if t != nil { + if local, _, err := sops.GetSockName(t); err == nil { + localAddr = *local.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &localAddr) + + // Field: rem_address. + var remoteAddr linux.SockAddrInet + if t != nil { + if remote, _, err := sops.GetPeerName(t); err == nil { + remoteAddr = *remote.(*linux.SockAddrInet) + } + } + writeInetAddr(buf, linux.AF_INET, &remoteAddr) + + // Field: state; socket state. + fmt.Fprintf(buf, "%02X ", sops.State()) + + // Field: tx_queue, rx_queue; number of packets in the transmit and + // receive queue. Unimplemented. + fmt.Fprintf(buf, "%08X:%08X ", 0, 0) + + // Field: tr, tm->when. Always 0 for UDP. + fmt.Fprintf(buf, "%02X:%08X ", 0, 0) + + // Field: retrnsmt. Always 0 for UDP. + fmt.Fprintf(buf, "%08X ", 0) + + // Field: uid. + uattr, err := sfile.Dirent.Inode.UnstableAttr(ctx) + if err != nil { + log.Warningf("Failed to retrieve unstable attr for socket file: %v", err) + fmt.Fprintf(buf, "%5d ", 0) + } else { + creds := auth.CredentialsFromContext(ctx) + fmt.Fprintf(buf, "%5d ", uint32(uattr.Owner.UID.In(creds.UserNamespace).OrOverflow())) + } + + // Field: timeout. Always 0 for UDP. + fmt.Fprintf(buf, "%8d ", 0) + + // Field: inode. + fmt.Fprintf(buf, "%8d ", sfile.InodeID()) + + // Field: ref; reference count on the socket inode. Don't count the ref + // we obtain while deferencing the weakref to this socket. + fmt.Fprintf(buf, "%d ", sfile.ReadRefs()-1) + + // Field: Socket struct address. Redacted due to the same reason as + // the 'Num' field in /proc/net/unix, see netUnix.ReadSeqFileData. + fmt.Fprintf(buf, "%#016p ", (*socket.Socket)(nil)) + + // Field: drops; number of dropped packets. Unimplemented. + fmt.Fprintf(buf, "%d", 0) + + fmt.Fprintf(buf, "\n") + + s.DecRef() + } + return nil +} + +// netSnmpData implements vfs.DynamicBytesSource for /proc/net/snmp. +// +// +stateify savable +type netSnmpData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netSnmpData)(nil) + +type snmpLine struct { + prefix string + header string +} + +var snmp = []snmpLine{ + { + prefix: "Ip", + header: "Forwarding DefaultTTL InReceives InHdrErrors InAddrErrors ForwDatagrams InUnknownProtos InDiscards InDelivers OutRequests OutDiscards OutNoRoutes ReasmTimeout ReasmReqds ReasmOKs ReasmFails FragOKs FragFails FragCreates", + }, + { + prefix: "Icmp", + header: "InMsgs InErrors InCsumErrors InDestUnreachs InTimeExcds InParmProbs InSrcQuenchs InRedirects InEchos InEchoReps InTimestamps InTimestampReps InAddrMasks InAddrMaskReps OutMsgs OutErrors OutDestUnreachs OutTimeExcds OutParmProbs OutSrcQuenchs OutRedirects OutEchos OutEchoReps OutTimestamps OutTimestampReps OutAddrMasks OutAddrMaskReps", + }, + { + prefix: "IcmpMsg", + }, + { + prefix: "Tcp", + header: "RtoAlgorithm RtoMin RtoMax MaxConn ActiveOpens PassiveOpens AttemptFails EstabResets CurrEstab InSegs OutSegs RetransSegs InErrs OutRsts InCsumErrors", + }, + { + prefix: "Udp", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, + { + prefix: "UdpLite", + header: "InDatagrams NoPorts InErrors OutDatagrams RcvbufErrors SndbufErrors InCsumErrors IgnoredMulti", + }, +} + +func toSlice(a interface{}) []uint64 { + v := reflect.Indirect(reflect.ValueOf(a)) + return v.Slice(0, v.Len()).Interface().([]uint64) +} + +func sprintSlice(s []uint64) string { + if len(s) == 0 { + return "" + } + r := fmt.Sprint(s) + return r[1 : len(r)-1] // Remove "[]" introduced by fmt of slice. +} + +// Generate implements vfs.DynamicBytesSource. +func (d *netSnmpData) Generate(ctx context.Context, buf *bytes.Buffer) error { + types := []interface{}{ + &inet.StatSNMPIP{}, + &inet.StatSNMPICMP{}, + nil, // TODO(gvisor.dev/issue/628): Support IcmpMsg stats. + &inet.StatSNMPTCP{}, + &inet.StatSNMPUDP{}, + &inet.StatSNMPUDPLite{}, + } + for i, stat := range types { + line := snmp[i] + if stat == nil { + fmt.Fprintf(buf, "%s:\n", line.prefix) + fmt.Fprintf(buf, "%s:\n", line.prefix) + continue + } + if err := d.stack.Statistics(stat, line.prefix); err != nil { + if err == syserror.EOPNOTSUPP { + log.Infof("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } else { + log.Warningf("Failed to retrieve %s of /proc/net/snmp: %v", line.prefix, err) + } + } + + fmt.Fprintf(buf, "%s: %s\n", line.prefix, line.header) + + if line.prefix == "Tcp" { + tcp := stat.(*inet.StatSNMPTCP) + // "Tcp" needs special processing because MaxConn is signed. RFC 2012. + fmt.Sprintf("%s: %s %d %s\n", line.prefix, sprintSlice(tcp[:3]), int64(tcp[3]), sprintSlice(tcp[4:])) + } else { + fmt.Sprintf("%s: %s\n", line.prefix, sprintSlice(toSlice(stat))) + } + } + return nil +} + +// netRouteData implements vfs.DynamicBytesSource for /proc/net/route. +// +// +stateify savable +type netRouteData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netRouteData)(nil) + +// Generate implements vfs.DynamicBytesSource. +// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. +func (d *netRouteData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%-127s\n", "Iface\tDestination\tGateway\tFlags\tRefCnt\tUse\tMetric\tMask\tMTU\tWindow\tIRTT") + + interfaces := d.stack.Interfaces() + for _, rt := range d.stack.RouteTable() { + // /proc/net/route only includes ipv4 routes. + if rt.Family != linux.AF_INET { + continue + } + + // /proc/net/route does not include broadcast or multicast routes. + if rt.Type == linux.RTN_BROADCAST || rt.Type == linux.RTN_MULTICAST { + continue + } + + iface, ok := interfaces[rt.OutputInterface] + if !ok || iface.Name == "lo" { + continue + } + + var ( + gw uint32 + prefix uint32 + flags = linux.RTF_UP + ) + if len(rt.GatewayAddr) == header.IPv4AddressSize { + flags |= linux.RTF_GATEWAY + gw = usermem.ByteOrder.Uint32(rt.GatewayAddr) + } + if len(rt.DstAddr) == header.IPv4AddressSize { + prefix = usermem.ByteOrder.Uint32(rt.DstAddr) + } + l := fmt.Sprintf( + "%s\t%08X\t%08X\t%04X\t%d\t%d\t%d\t%08X\t%d\t%d\t%d", + iface.Name, + prefix, + gw, + flags, + 0, // RefCnt. + 0, // Use. + 0, // Metric. + (uint32(1)<<rt.DstLen)-1, + 0, // MTU. + 0, // Window. + 0, // RTT. + ) + fmt.Fprintf(buf, "%-127s\n", l) + } + return nil +} + +// netStatData implements vfs.DynamicBytesSource for /proc/net/netstat. +// +// +stateify savable +type netStatData struct { + kernfs.DynamicBytesFile + + stack inet.Stack +} + +var _ dynamicInode = (*netStatData)(nil) + +// Generate implements vfs.DynamicBytesSource. +// See Linux's net/ipv4/fib_trie.c:fib_route_seq_show. +func (d *netStatData) Generate(ctx context.Context, buf *bytes.Buffer) error { + buf.WriteString("TcpExt: SyncookiesSent SyncookiesRecv SyncookiesFailed " + + "EmbryonicRsts PruneCalled RcvPruned OfoPruned OutOfWindowIcmps " + + "LockDroppedIcmps ArpFilter TW TWRecycled TWKilled PAWSPassive " + + "PAWSActive PAWSEstab DelayedACKs DelayedACKLocked DelayedACKLost " + + "ListenOverflows ListenDrops TCPPrequeued TCPDirectCopyFromBacklog " + + "TCPDirectCopyFromPrequeue TCPPrequeueDropped TCPHPHits TCPHPHitsToUser " + + "TCPPureAcks TCPHPAcks TCPRenoRecovery TCPSackRecovery TCPSACKReneging " + + "TCPFACKReorder TCPSACKReorder TCPRenoReorder TCPTSReorder TCPFullUndo " + + "TCPPartialUndo TCPDSACKUndo TCPLossUndo TCPLostRetransmit " + + "TCPRenoFailures TCPSackFailures TCPLossFailures TCPFastRetrans " + + "TCPForwardRetrans TCPSlowStartRetrans TCPTimeouts TCPLossProbes " + + "TCPLossProbeRecovery TCPRenoRecoveryFail TCPSackRecoveryFail " + + "TCPSchedulerFailed TCPRcvCollapsed TCPDSACKOldSent TCPDSACKOfoSent " + + "TCPDSACKRecv TCPDSACKOfoRecv TCPAbortOnData TCPAbortOnClose " + + "TCPAbortOnMemory TCPAbortOnTimeout TCPAbortOnLinger TCPAbortFailed " + + "TCPMemoryPressures TCPSACKDiscard TCPDSACKIgnoredOld " + + "TCPDSACKIgnoredNoUndo TCPSpuriousRTOs TCPMD5NotFound TCPMD5Unexpected " + + "TCPMD5Failure TCPSackShifted TCPSackMerged TCPSackShiftFallback " + + "TCPBacklogDrop TCPMinTTLDrop TCPDeferAcceptDrop IPReversePathFilter " + + "TCPTimeWaitOverflow TCPReqQFullDoCookies TCPReqQFullDrop TCPRetransFail " + + "TCPRcvCoalesce TCPOFOQueue TCPOFODrop TCPOFOMerge TCPChallengeACK " + + "TCPSYNChallenge TCPFastOpenActive TCPFastOpenActiveFail " + + "TCPFastOpenPassive TCPFastOpenPassiveFail TCPFastOpenListenOverflow " + + "TCPFastOpenCookieReqd TCPSpuriousRtxHostQueues BusyPollRxPackets " + + "TCPAutoCorking TCPFromZeroWindowAdv TCPToZeroWindowAdv " + + "TCPWantZeroWindowAdv TCPSynRetrans TCPOrigDataSent TCPHystartTrainDetect " + + "TCPHystartTrainCwnd TCPHystartDelayDetect TCPHystartDelayCwnd " + + "TCPACKSkippedSynRecv TCPACKSkippedPAWS TCPACKSkippedSeq " + + "TCPACKSkippedFinWait2 TCPACKSkippedTimeWait TCPACKSkippedChallenge " + + "TCPWinProbe TCPKeepAlive TCPMTUPFail TCPMTUPSuccess") + return nil +} diff --git a/pkg/sentry/fsimpl/proc/tasks_sys.go b/pkg/sentry/fsimpl/proc/tasks_sys.go new file mode 100644 index 000000000..aabf2bf0c --- /dev/null +++ b/pkg/sentry/fsimpl/proc/tasks_sys.go @@ -0,0 +1,143 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package proc + +import ( + "bytes" + "fmt" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" +) + +// newSysDir returns the dentry corresponding to /proc/sys directory. +func newSysDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "kernel": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "hostname": newDentry(root, inoGen.NextIno(), 0444, &hostnameData{}), + "shmall": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMALL)), + "shmmax": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMAX)), + "shmmni": newDentry(root, inoGen.NextIno(), 0444, shmData(linux.SHMMNI)), + }), + "vm": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "mmap_min_addr": newDentry(root, inoGen.NextIno(), 0444, &mmapMinAddrData{}), + "overcommit_memory": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0\n")), + }), + "net": newSysNetDir(root, inoGen), + }) +} + +// newSysNetDir returns the dentry corresponding to /proc/sys/net directory. +func newSysNetDir(root *auth.Credentials, inoGen InoGenerator) *kernfs.Dentry { + return kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "net": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "ipv4": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + // Add tcp_sack. + // TODO(gvisor.dev/issue/1195): tcp_sack allows write(2) + // "tcp_sack": newTCPSackInode(ctx, msrc, s), + + // The following files are simple stubs until they are implemented in + // netstack, most of these files are configuration related. We use the + // value closest to the actual netstack behavior or any empty file, all + // of these files will have mode 0444 (read-only for all users). + "ip_local_port_range": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("16000 65535")), + "ip_local_reserved_ports": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "ipfrag_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("30")), + "ip_nonlocal_bind": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "ip_no_pmtu_disc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + + // tcp_allowed_congestion_control tell the user what they are able to + // do as an unprivledged process so we leave it empty. + "tcp_allowed_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "tcp_available_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + "tcp_congestion_control": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("reno")), + + // Many of the following stub files are features netstack doesn't + // support. The unsupported features return "0" to indicate they are + // disabled. + "tcp_base_mss": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1280")), + "tcp_dsack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_early_retrans": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fack": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_fastopen_key": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("")), + "tcp_invalid_ratelimit": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_intvl": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_probes": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_keepalive_time": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("7200")), + "tcp_mtu_probing": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_no_metrics_save": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_probe_interval": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_probe_threshold": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "tcp_retries1": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), + "tcp_retries2": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("15")), + "tcp_rfc1337": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_slow_start_after_idle": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + "tcp_synack_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), + "tcp_syn_retries": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("3")), + "tcp_timestamps": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("1")), + }), + "core": kernfs.NewStaticDir(root, inoGen.NextIno(), 0555, map[string]*kernfs.Dentry{ + "default_qdisc": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("pfifo_fast")), + "message_burst": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("10")), + "message_cost": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("5")), + "optmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("0")), + "rmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "rmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "somaxconn": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("128")), + "wmem_default": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + "wmem_max": newDentry(root, inoGen.NextIno(), 0444, newStaticFile("212992")), + }), + }), + }) +} + +// mmapMinAddrData implements vfs.DynamicBytesSource for +// /proc/sys/vm/mmap_min_addr. +// +// +stateify savable +type mmapMinAddrData struct { + kernfs.DynamicBytesFile + + k *kernel.Kernel +} + +var _ dynamicInode = (*mmapMinAddrData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (d *mmapMinAddrData) Generate(ctx context.Context, buf *bytes.Buffer) error { + fmt.Fprintf(buf, "%d\n", d.k.Platform.MinUserAddress()) + return nil +} + +// hostnameData implements vfs.DynamicBytesSource for /proc/sys/kernel/hostname. +// +// +stateify savable +type hostnameData struct { + kernfs.DynamicBytesFile +} + +var _ dynamicInode = (*hostnameData)(nil) + +// Generate implements vfs.DynamicBytesSource.Generate. +func (*hostnameData) Generate(ctx context.Context, buf *bytes.Buffer) error { + utsns := kernel.UTSNamespaceFromContext(ctx) + buf.WriteString(utsns.HostName()) + buf.WriteString("\n") + return nil +} diff --git a/pkg/sentry/fsimpl/proc/net_test.go b/pkg/sentry/fsimpl/proc/tasks_sys_test.go index 20a77a8ca..0a1d3f34b 100644 --- a/pkg/sentry/fsimpl/proc/net_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_sys_test.go @@ -31,7 +31,7 @@ func newIPv6TestStack() *inet.TestStack { } func TestIfinet6NoAddresses(t *testing.T) { - n := &ifinet6{s: newIPv6TestStack()} + n := &ifinet6{stack: newIPv6TestStack()} var buf bytes.Buffer n.Generate(contexttest.Context(t), &buf) if buf.Len() > 0 { @@ -62,7 +62,7 @@ func TestIfinet6(t *testing.T) { "101112131415161718191a1b1c1d1e1f 02 80 00 00 eth1\n": {}, } - n := &ifinet6{s: s} + n := &ifinet6{stack: s} contents := n.contents() if len(contents) != len(want) { t.Errorf("Got len(n.contents()) = %d, want = %d", len(contents), len(want)) diff --git a/pkg/sentry/fsimpl/proc/tasks_test.go b/pkg/sentry/fsimpl/proc/tasks_test.go index ca8c87ec2..002d2f73b 100644 --- a/pkg/sentry/fsimpl/proc/tasks_test.go +++ b/pkg/sentry/fsimpl/proc/tasks_test.go @@ -69,12 +69,16 @@ func checkDots(dirs []vfs.Dirent) ([]vfs.Dirent, error) { func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ + "cpuinfo": {Type: linux.DT_REG}, "loadavg": {Type: linux.DT_REG}, "meminfo": {Type: linux.DT_REG}, "mounts": {Type: linux.DT_LNK}, + "net": {Type: linux.DT_DIR}, "self": selfLink, "stat": {Type: linux.DT_REG}, + "sys": {Type: linux.DT_DIR}, "thread-self": threadSelfLink, + "uptime": {Type: linux.DT_REG}, "version": {Type: linux.DT_REG}, } return checkFiles(gots, wants) @@ -82,12 +86,21 @@ func checkTasksStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { func checkTaskStaticFiles(gots []vfs.Dirent) ([]vfs.Dirent, error) { wants := map[string]vfs.Dirent{ - "io": {Type: linux.DT_REG}, - "maps": {Type: linux.DT_REG}, - "smaps": {Type: linux.DT_REG}, - "stat": {Type: linux.DT_REG}, - "statm": {Type: linux.DT_REG}, - "status": {Type: linux.DT_REG}, + "auxv": {Type: linux.DT_REG}, + "cgroup": {Type: linux.DT_REG}, + "cmdline": {Type: linux.DT_REG}, + "comm": {Type: linux.DT_REG}, + "environ": {Type: linux.DT_REG}, + "gid_map": {Type: linux.DT_REG}, + "io": {Type: linux.DT_REG}, + "maps": {Type: linux.DT_REG}, + "ns": {Type: linux.DT_DIR}, + "smaps": {Type: linux.DT_REG}, + "stat": {Type: linux.DT_REG}, + "statm": {Type: linux.DT_REG}, + "status": {Type: linux.DT_REG}, + "task": {Type: linux.DT_DIR}, + "uid_map": {Type: linux.DT_REG}, } return checkFiles(gots, wants) } @@ -133,7 +146,15 @@ func setup() (context.Context, *vfs.VirtualFilesystem, vfs.VirtualDentry, error) vfsObj.MustRegisterFilesystemType("procfs", &procFSType{}, &vfs.RegisterFilesystemTypeOptions{ AllowUserMount: true, }) - mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &vfs.GetFilesystemOptions{}) + fsOpts := vfs.GetFilesystemOptions{ + InternalData: &InternalData{ + Cgroups: map[string]string{ + "cpuset": "/foo/cpuset", + "memory": "/foo/memory", + }, + }, + } + mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "procfs", &fsOpts) if err != nil { return nil, nil, vfs.VirtualDentry{}, fmt.Errorf("NewMountNamespace(): %v", err) } diff --git a/pkg/sentry/fsimpl/proc/version.go b/pkg/sentry/fsimpl/proc/version.go deleted file mode 100644 index 367f2396b..000000000 --- a/pkg/sentry/fsimpl/proc/version.go +++ /dev/null @@ -1,70 +0,0 @@ -// Copyright 2019 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package proc - -import ( - "bytes" - "fmt" - - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" - "gvisor.dev/gvisor/pkg/sentry/kernel" -) - -// versionData implements vfs.DynamicBytesSource for /proc/version. -// -// +stateify savable -type versionData struct { - kernfs.DynamicBytesFile - - // k is the owning Kernel. - k *kernel.Kernel -} - -var _ dynamicInode = (*versionData)(nil) - -// Generate implements vfs.DynamicBytesSource.Generate. -func (v *versionData) Generate(ctx context.Context, buf *bytes.Buffer) error { - init := v.k.GlobalInit() - if init == nil { - // Attempted to read before the init Task is created. This can - // only occur during startup, which should never need to read - // this file. - panic("Attempted to read version before initial Task is available") - } - - // /proc/version takes the form: - // - // "SYSNAME version RELEASE (COMPILE_USER@COMPILE_HOST) - // (COMPILER_VERSION) VERSION" - // - // where: - // - SYSNAME, RELEASE, and VERSION are the same as returned by - // sys_utsname - // - COMPILE_USER is the user that build the kernel - // - COMPILE_HOST is the hostname of the machine on which the kernel - // was built - // - COMPILER_VERSION is the version reported by the building compiler - // - // Since we don't really want to expose build information to - // applications, those fields are omitted. - // - // FIXME(mpratt): Using Version from the init task SyscallTable - // disregards the different version a task may have (e.g., in a uts - // namespace). - ver := init.Leader().SyscallTable().Version - fmt.Fprintf(buf, "%s version %s %s\n", ver.Sysname, ver.Release, ver.Version) - return nil -} diff --git a/pkg/sentry/fsimpl/tmpfs/BUILD b/pkg/sentry/fsimpl/tmpfs/BUILD index 82f5c2f41..7601c7c04 100644 --- a/pkg/sentry/fsimpl/tmpfs/BUILD +++ b/pkg/sentry/fsimpl/tmpfs/BUILD @@ -40,6 +40,7 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/pipe", + "//pkg/sentry/kernel/time", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", @@ -77,6 +78,7 @@ go_test( srcs = [ "pipe_test.go", "regular_file_test.go", + "stat_test.go", ], embed = [":tmpfs"], deps = [ diff --git a/pkg/sentry/fsimpl/tmpfs/filesystem.go b/pkg/sentry/fsimpl/tmpfs/filesystem.go index 26979729e..4cd7e9aea 100644 --- a/pkg/sentry/fsimpl/tmpfs/filesystem.go +++ b/pkg/sentry/fsimpl/tmpfs/filesystem.go @@ -56,7 +56,8 @@ afterSymlink: } next := nextVFSD.Impl().(*dentry) if symlink, ok := next.inode.impl.(*symlink); ok && rp.ShouldFollowSymlink() { - // TODO: symlink traversals update access time + // TODO(gvisor.dev/issues/1197): Symlink traversals updates + // access time. if err := rp.HandleSymlink(symlink.target); err != nil { return nil, err } @@ -501,7 +502,8 @@ func (fs *filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa oldParent.inode.decLinksLocked() newParent.inode.incLinksLocked() } - // TODO: update timestamps and parent directory sizes + // TODO(gvisor.dev/issues/1197): Update timestamps and parent directory + // sizes. vfsObj.CommitRenameReplaceDentry(renamedVFSD, &newParent.vfsd, newName, replacedVFSD) return nil } @@ -555,15 +557,11 @@ func (fs *filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error func (fs *filesystem) SetStatAt(ctx context.Context, rp *vfs.ResolvingPath, opts vfs.SetStatOptions) error { fs.mu.RLock() defer fs.mu.RUnlock() - _, err := resolveLocked(rp) + d, err := resolveLocked(rp) if err != nil { return err } - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM + return d.inode.setStat(opts.Stat) } // StatAt implements vfs.FilesystemImpl.StatAt. @@ -587,7 +585,7 @@ func (fs *filesystem) StatFSAt(ctx context.Context, rp *vfs.ResolvingPath) (linu if err != nil { return linux.Statfs{}, err } - // TODO: actually implement statfs + // TODO(gvisor.dev/issues/1197): Actually implement statfs. return linux.Statfs{}, syserror.ENOSYS } diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file.go b/pkg/sentry/fsimpl/tmpfs/regular_file.go index f200e767d..5fa70cc6d 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file.go @@ -63,6 +63,41 @@ func (fs *filesystem) newRegularFile(creds *auth.Credentials, mode linux.FileMod return &file.inode } +// truncate grows or shrinks the file to the given size. It returns true if the +// file size was updated. +func (rf *regularFile) truncate(size uint64) (bool, error) { + rf.mu.Lock() + defer rf.mu.Unlock() + + if size == rf.size { + // Nothing to do. + return false, nil + } + + if size > rf.size { + // Growing the file. + if rf.seals&linux.F_SEAL_GROW != 0 { + // Seal does not allow growth. + return false, syserror.EPERM + } + rf.size = size + return true, nil + } + + // Shrinking the file + if rf.seals&linux.F_SEAL_SHRINK != 0 { + // Seal does not allow shrink. + return false, syserror.EPERM + } + + // TODO(gvisor.dev/issues/1197): Invalidate mappings once we have + // mappings. + + rf.data.Truncate(size, rf.memFile) + rf.size = size + return true, nil +} + type regularFileFD struct { fileDescription diff --git a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go index 3731c5b6f..034a29fdb 100644 --- a/pkg/sentry/fsimpl/tmpfs/regular_file_test.go +++ b/pkg/sentry/fsimpl/tmpfs/regular_file_test.go @@ -18,6 +18,7 @@ import ( "bytes" "fmt" "io" + "sync/atomic" "testing" "gvisor.dev/gvisor/pkg/abi/linux" @@ -29,10 +30,12 @@ import ( "gvisor.dev/gvisor/pkg/sentry/vfs" ) -// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If -// the returned err is not nil, then cleanup should be called when the FD is no -// longer needed. -func newFileFD(ctx context.Context, filename string) (*vfs.FileDescription, func(), error) { +// nextFileID is used to generate unique file names. +var nextFileID int64 + +// newTmpfsRoot creates a new tmpfs mount, and returns the root. If the error +// is not nil, then cleanup should be called when the root is no longer needed. +func newTmpfsRoot(ctx context.Context) (*vfs.VirtualFilesystem, vfs.VirtualDentry, func(), error) { creds := auth.CredentialsFromContext(ctx) vfsObj := vfs.New() @@ -41,36 +44,124 @@ func newFileFD(ctx context.Context, filename string) (*vfs.FileDescription, func }) mntns, err := vfsObj.NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{}) if err != nil { - return nil, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) + return nil, vfs.VirtualDentry{}, nil, fmt.Errorf("failed to create tmpfs root mount: %v", err) } root := mntns.Root() + return vfsObj, root, func() { + root.DecRef() + mntns.DecRef(vfsObj) + }, nil +} + +// newFileFD creates a new file in a new tmpfs mount, and returns the FD. If +// the returned err is not nil, then cleanup should be called when the FD is no +// longer needed. +func newFileFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + filename := fmt.Sprintf("tmpfs-test-file-%d", atomic.AddInt64(&nextFileID, 1)) // Create the file that will be write/read. fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ - Root: root, - Start: root, - Path: fspath.Parse(filename), - FollowFinalSymlink: true, + Root: root, + Start: root, + Path: fspath.Parse(filename), }, &vfs.OpenOptions{ Flags: linux.O_RDWR | linux.O_CREAT | linux.O_EXCL, - Mode: 0644, + Mode: linux.ModeRegular | mode, }) if err != nil { - root.DecRef() - mntns.DecRef(vfsObj) + cleanup() return nil, nil, fmt.Errorf("failed to create file %q: %v", filename, err) } - return fd, func() { - root.DecRef() - mntns.DecRef(vfsObj) - }, nil + return fd, cleanup, nil +} + +// newDirFD is like newFileFD, but for directories. +func newDirFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + dirname := fmt.Sprintf("tmpfs-test-dir-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the dir. + if err := vfsObj.MkdirAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.MkdirOptions{ + Mode: linux.ModeDirectory | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create directory %q: %v", dirname, err) + } + + // Open the dir and return it. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(dirname), + }, &vfs.OpenOptions{ + Flags: linux.O_RDONLY | linux.O_DIRECTORY, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open directory %q: %v", dirname, err) + } + + return fd, cleanup, nil +} + +// newPipeFD is like newFileFD, but for pipes. +func newPipeFD(ctx context.Context, mode linux.FileMode) (*vfs.FileDescription, func(), error) { + creds := auth.CredentialsFromContext(ctx) + vfsObj, root, cleanup, err := newTmpfsRoot(ctx) + if err != nil { + return nil, nil, err + } + + pipename := fmt.Sprintf("tmpfs-test-pipe-%d", atomic.AddInt64(&nextFileID, 1)) + + // Create the pipe. + if err := vfsObj.MknodAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(pipename), + }, &vfs.MknodOptions{ + Mode: linux.ModeNamedPipe | mode, + }); err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to create pipe %q: %v", pipename, err) + } + + // Open the pipe and return it. + fd, err := vfsObj.OpenAt(ctx, creds, &vfs.PathOperation{ + Root: root, + Start: root, + Path: fspath.Parse(pipename), + }, &vfs.OpenOptions{ + Flags: linux.O_RDWR, + }) + if err != nil { + cleanup() + return nil, nil, fmt.Errorf("failed to open pipe %q: %v", pipename, err) + } + + return fd, cleanup, nil } // Test that we can write some data to a file and read it back.` func TestSimpleWriteRead(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "simpleReadWrite") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } @@ -116,7 +207,7 @@ func TestSimpleWriteRead(t *testing.T) { func TestPWrite(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "PRead") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } @@ -171,7 +262,7 @@ func TestPWrite(t *testing.T) { func TestPRead(t *testing.T) { ctx := contexttest.Context(t) - fd, cleanup, err := newFileFD(ctx, "PRead") + fd, cleanup, err := newFileFD(ctx, 0644) if err != nil { t.Fatal(err) } @@ -222,3 +313,124 @@ func TestPRead(t *testing.T) { } } } + +func TestTruncate(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + // Fill the file with some data. + data := bytes.Repeat([]byte("gVisor is awsome"), 100) + written, err := fd.Write(ctx, usermem.BytesIOSequence(data), vfs.WriteOptions{}) + if err != nil { + t.Fatalf("fd.Write failed: %v", err) + } + + // Size should be same as written. + sizeStatOpts := vfs.StatOptions{Mask: linux.STATX_SIZE} + stat, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := int64(stat.Size), written; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + + // Truncate down. + newSize := uint64(10) + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + // Size should be updated. + statAfterTruncateDown, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := statAfterTruncateDown.Size, newSize; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + // We should only read newSize worth of data. + buf := make([]byte, 1000) + if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { + t.Fatalf("fd.PRead failed: %v", err) + } else if uint64(n) != newSize { + t.Errorf("fd.PRead got size %d, want %d", n, newSize) + } + // Mtime and Ctime should be bumped. + if got := statAfterTruncateDown.Mtime.ToNsec(); got <= stat.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want > %v", got, stat.Mtime) + } + if got := statAfterTruncateDown.Ctime.ToNsec(); got <= stat.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) + } + + // Truncate up. + newSize = 100 + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + // Size should be updated. + statAfterTruncateUp, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + if got, want := statAfterTruncateUp.Size, newSize; got != want { + t.Errorf("fd.Stat got size %d, want %d", got, want) + } + // We should read newSize worth of data. + buf = make([]byte, 1000) + if n, err := fd.PRead(ctx, usermem.BytesIOSequence(buf), 0, vfs.ReadOptions{}); err != nil && err != io.EOF { + t.Fatalf("fd.PRead failed: %v", err) + } else if uint64(n) != newSize { + t.Errorf("fd.PRead got size %d, want %d", n, newSize) + } + // Bytes should be null after 10, since we previously truncated to 10. + for i := uint64(10); i < newSize; i++ { + if buf[i] != 0 { + t.Errorf("fd.PRead got byte %d=%x, want 0", i, buf[i]) + break + } + } + // Mtime and Ctime should be bumped. + if got := statAfterTruncateUp.Mtime.ToNsec(); got <= statAfterTruncateDown.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want > %v", got, statAfterTruncateDown.Mtime) + } + if got := statAfterTruncateUp.Ctime.ToNsec(); got <= statAfterTruncateDown.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want > %v", got, stat.Ctime) + } + + // Truncate to the current size. + newSize = statAfterTruncateUp.Size + if err := fd.SetStat(ctx, vfs.SetStatOptions{ + Stat: linux.Statx{ + Mask: linux.STATX_SIZE, + Size: newSize, + }, + }); err != nil { + t.Errorf("fd.Truncate failed: %v", err) + } + statAfterTruncateNoop, err := fd.Stat(ctx, sizeStatOpts) + if err != nil { + t.Fatalf("fd.Stat failed: %v", err) + } + // Mtime and Ctime should not be bumped, since operation is a noop. + if got := statAfterTruncateNoop.Mtime.ToNsec(); got != statAfterTruncateUp.Mtime.ToNsec() { + t.Errorf("fd.Stat got Mtime %v, want %v", got, statAfterTruncateUp.Mtime) + } + if got := statAfterTruncateNoop.Ctime.ToNsec(); got != statAfterTruncateUp.Ctime.ToNsec() { + t.Errorf("fd.Stat got Ctime %v, want %v", got, statAfterTruncateUp.Ctime) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/stat_test.go b/pkg/sentry/fsimpl/tmpfs/stat_test.go new file mode 100644 index 000000000..ebe035dee --- /dev/null +++ b/pkg/sentry/fsimpl/tmpfs/stat_test.go @@ -0,0 +1,232 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tmpfs + +import ( + "fmt" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/contexttest" + "gvisor.dev/gvisor/pkg/sentry/vfs" +) + +func TestStatAfterCreate(t *testing.T) { + ctx := contexttest.Context(t) + mode := linux.FileMode(0644) + + // Run with different file types. + // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + for _, typ := range []string{"file", "dir", "pipe"} { + t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { + var ( + fd *vfs.FileDescription + cleanup func() + err error + ) + switch typ { + case "file": + fd, cleanup, err = newFileFD(ctx, mode) + case "dir": + fd, cleanup, err = newDirFD(ctx, mode) + case "pipe": + fd, cleanup, err = newPipeFD(ctx, mode) + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + if err != nil { + t.Fatal(err) + } + defer cleanup() + + got, err := fd.Stat(ctx, vfs.StatOptions{}) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Atime, Ctime, Mtime should all be current time (non-zero). + atime, ctime, mtime := got.Atime.ToNsec(), got.Ctime.ToNsec(), got.Mtime.ToNsec() + if atime != ctime || ctime != mtime { + t.Errorf("got atime=%d ctime=%d mtime=%d, wanted equal values", atime, ctime, mtime) + } + if atime == 0 { + t.Errorf("got atime=%d, want non-zero", atime) + } + + // Btime should be 0, as it is not set by tmpfs. + if btime := got.Btime.ToNsec(); btime != 0 { + t.Errorf("got btime %d, want 0", got.Btime.ToNsec()) + } + + // Size should be 0. + if got.Size != 0 { + t.Errorf("got size %d, want 0", got.Size) + } + + // Nlink should be 1 for files, 2 for dirs. + wantNlink := uint32(1) + if typ == "dir" { + wantNlink = 2 + } + if got.Nlink != wantNlink { + t.Errorf("got nlink %d, want %d", got.Nlink, wantNlink) + } + + // UID and GID are set from context creds. + creds := auth.CredentialsFromContext(ctx) + if got.UID != uint32(creds.EffectiveKUID) { + t.Errorf("got uid %d, want %d", got.UID, uint32(creds.EffectiveKUID)) + } + if got.GID != uint32(creds.EffectiveKGID) { + t.Errorf("got gid %d, want %d", got.GID, uint32(creds.EffectiveKGID)) + } + + // Mode. + wantMode := uint16(mode) + switch typ { + case "file": + wantMode |= linux.S_IFREG + case "dir": + wantMode |= linux.S_IFDIR + case "pipe": + wantMode |= linux.S_IFIFO + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + + if got.Mode != wantMode { + t.Errorf("got mode %x, want %x", got.Mode, wantMode) + } + + // Ino. + if got.Ino == 0 { + t.Errorf("got ino %d, want not 0", got.Ino) + } + }) + } +} + +func TestSetStatAtime(t *testing.T) { + ctx := contexttest.Context(t) + fd, cleanup, err := newFileFD(ctx, 0644) + if err != nil { + t.Fatal(err) + } + defer cleanup() + + allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} + + // Get initial stat. + initialStat, err := fd.Stat(ctx, allStatOptions) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Set atime, but without the mask. + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ + Mask: 0, + Atime: linux.NsecToStatxTimestamp(100), + }}); err != nil { + t.Errorf("SetStat atime without mask failed: %v") + } + // Atime should be unchanged. + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != initialStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) + } + + // Set atime, this time included in the mask. + setStat := linux.Statx{ + Mask: linux.STATX_ATIME, + Atime: linux.NsecToStatxTimestamp(100), + } + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { + t.Errorf("SetStat atime with mask failed: %v") + } + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != setStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) + } +} + +func TestSetStat(t *testing.T) { + ctx := contexttest.Context(t) + mode := linux.FileMode(0644) + + // Run with different file types. + // TODO(gvisor.dev/issues/1197): Also test symlinks and sockets. + for _, typ := range []string{"file", "dir", "pipe"} { + t.Run(fmt.Sprintf("type=%q", typ), func(t *testing.T) { + var ( + fd *vfs.FileDescription + cleanup func() + err error + ) + switch typ { + case "file": + fd, cleanup, err = newFileFD(ctx, mode) + case "dir": + fd, cleanup, err = newDirFD(ctx, mode) + case "pipe": + fd, cleanup, err = newPipeFD(ctx, mode) + default: + panic(fmt.Sprintf("unknown typ %q", typ)) + } + if err != nil { + t.Fatal(err) + } + defer cleanup() + + allStatOptions := vfs.StatOptions{Mask: linux.STATX_ALL} + + // Get initial stat. + initialStat, err := fd.Stat(ctx, allStatOptions) + if err != nil { + t.Fatalf("Stat failed: %v", err) + } + + // Set atime, but without the mask. + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: linux.Statx{ + Mask: 0, + Atime: linux.NsecToStatxTimestamp(100), + }}); err != nil { + t.Errorf("SetStat atime without mask failed: %v") + } + // Atime should be unchanged. + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != initialStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, initialStat.Atime) + } + + // Set atime, this time included in the mask. + setStat := linux.Statx{ + Mask: linux.STATX_ATIME, + Atime: linux.NsecToStatxTimestamp(100), + } + if err := fd.SetStat(ctx, vfs.SetStatOptions{Stat: setStat}); err != nil { + t.Errorf("SetStat atime with mask failed: %v") + } + if gotStat, err := fd.Stat(ctx, allStatOptions); err != nil { + t.Errorf("Stat got error: %v", err) + } else if gotStat.Atime != setStat.Atime { + t.Errorf("Stat got atime %d, want %d", gotStat.Atime, setStat.Atime) + } + }) + } +} diff --git a/pkg/sentry/fsimpl/tmpfs/tmpfs.go b/pkg/sentry/fsimpl/tmpfs/tmpfs.go index 701826f90..1d4889c89 100644 --- a/pkg/sentry/fsimpl/tmpfs/tmpfs.go +++ b/pkg/sentry/fsimpl/tmpfs/tmpfs.go @@ -31,6 +31,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" @@ -47,6 +48,9 @@ type filesystem struct { // memFile is used to allocate pages to for regular files. memFile *pgalloc.MemoryFile + // clock is a realtime clock used to set timestamps in file operations. + clock time.Clock + // mu serializes changes to the Dentry tree. mu sync.RWMutex @@ -59,8 +63,10 @@ func (fstype FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt if memFileProvider == nil { panic("MemoryFileProviderFromContext returned nil") } + clock := time.RealtimeClockFromContext(ctx) fs := filesystem{ memFile: memFileProvider.MemoryFile(), + clock: clock, } fs.vfsfs.Init(vfsObj, &fs) root := fs.newDentry(fs.newDirectory(creds, 01777)) @@ -116,6 +122,9 @@ func (d *dentry) DecRef() { // inode represents a filesystem object. type inode struct { + // clock is a realtime clock used to set timestamps in file operations. + clock time.Clock + // refs is a reference count. refs is accessed using atomic memory // operations. // @@ -126,26 +135,37 @@ type inode struct { // filesystem.RmdirAt() drops the reference. refs int64 - // Inode metadata; protected by mu and accessed using atomic memory - // operations unless otherwise specified. - mu sync.RWMutex + // Inode metadata. Writing multiple fields atomically requires holding + // mu, othewise atomic operations can be used. + mu sync.Mutex mode uint32 // excluding file type bits, which are based on impl nlink uint32 // protected by filesystem.mu instead of inode.mu uid uint32 // auth.KUID, but stored as raw uint32 for sync/atomic gid uint32 // auth.KGID, but ... ino uint64 // immutable + // Linux's tmpfs has no concept of btime. + atime int64 // nanoseconds + ctime int64 // nanoseconds + mtime int64 // nanoseconds + impl interface{} // immutable } const maxLinks = math.MaxUint32 func (i *inode) init(impl interface{}, fs *filesystem, creds *auth.Credentials, mode linux.FileMode) { + i.clock = fs.clock i.refs = 1 i.mode = uint32(mode) i.uid = uint32(creds.EffectiveKUID) i.gid = uint32(creds.EffectiveKGID) i.ino = atomic.AddUint64(&fs.nextInoMinusOne, 1) + // Tmpfs creation sets atime, ctime, and mtime to current time. + now := i.clock.Now().Nanoseconds() + i.atime = now + i.ctime = now + i.mtime = now // i.nlink initialized by caller i.impl = impl } @@ -213,15 +233,24 @@ func (i *inode) checkPermissions(creds *auth.Credentials, ats vfs.AccessTypes, i // Go won't inline this function, and returning linux.Statx (which is quite // big) means spending a lot of time in runtime.duffcopy(), so instead it's an // output parameter. +// +// Note that Linux does not guarantee to return consistent data (in the case of +// a concurrent modification), so we do not require holding inode.mu. func (i *inode) statTo(stat *linux.Statx) { - stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | linux.STATX_UID | linux.STATX_GID | linux.STATX_INO + stat.Mask = linux.STATX_TYPE | linux.STATX_MODE | linux.STATX_NLINK | + linux.STATX_UID | linux.STATX_GID | linux.STATX_INO | linux.STATX_ATIME | + linux.STATX_BTIME | linux.STATX_CTIME | linux.STATX_MTIME stat.Blksize = 1 // usermem.PageSize in tmpfs stat.Nlink = atomic.LoadUint32(&i.nlink) stat.UID = atomic.LoadUint32(&i.uid) stat.GID = atomic.LoadUint32(&i.gid) stat.Mode = uint16(atomic.LoadUint32(&i.mode)) stat.Ino = i.ino - // TODO: device number + // Linux's tmpfs has no concept of btime, so zero-value is returned. + stat.Atime = linux.NsecToStatxTimestamp(i.atime) + stat.Ctime = linux.NsecToStatxTimestamp(i.ctime) + stat.Mtime = linux.NsecToStatxTimestamp(i.mtime) + // TODO(gvisor.dev/issues/1197): Device number. switch impl := i.impl.(type) { case *regularFile: stat.Mode |= linux.S_IFREG @@ -245,6 +274,75 @@ func (i *inode) statTo(stat *linux.Statx) { } } +func (i *inode) setStat(stat linux.Statx) error { + if stat.Mask == 0 { + return nil + } + i.mu.Lock() + var ( + needsMtimeBump bool + needsCtimeBump bool + ) + mask := stat.Mask + if mask&linux.STATX_MODE != 0 { + atomic.StoreUint32(&i.mode, uint32(stat.Mode)) + needsCtimeBump = true + } + if mask&linux.STATX_UID != 0 { + atomic.StoreUint32(&i.uid, stat.UID) + needsCtimeBump = true + } + if mask&linux.STATX_GID != 0 { + atomic.StoreUint32(&i.gid, stat.GID) + needsCtimeBump = true + } + if mask&linux.STATX_SIZE != 0 { + switch impl := i.impl.(type) { + case *regularFile: + updated, err := impl.truncate(stat.Size) + if err != nil { + return err + } + if updated { + needsMtimeBump = true + needsCtimeBump = true + } + case *directory: + return syserror.EISDIR + case *symlink: + return syserror.EINVAL + case *namedPipe: + // Nothing. + default: + panic(fmt.Sprintf("unknown inode type: %T", i.impl)) + } + } + if mask&linux.STATX_ATIME != 0 { + atomic.StoreInt64(&i.atime, stat.Atime.ToNsecCapped()) + needsCtimeBump = true + } + if mask&linux.STATX_MTIME != 0 { + atomic.StoreInt64(&i.mtime, stat.Mtime.ToNsecCapped()) + needsCtimeBump = true + // Ignore the mtime bump, since we just set it ourselves. + needsMtimeBump = false + } + if mask&linux.STATX_CTIME != 0 { + atomic.StoreInt64(&i.ctime, stat.Ctime.ToNsecCapped()) + // Ignore the ctime bump, since we just set it ourselves. + needsCtimeBump = false + } + now := i.clock.Now().Nanoseconds() + if needsMtimeBump { + atomic.StoreInt64(&i.mtime, now) + } + if needsCtimeBump { + atomic.StoreInt64(&i.ctime, now) + } + i.mu.Unlock() + return nil +} + // allocatedBlocksForSize returns the number of 512B blocks needed to // accommodate the given size in bytes, as appropriate for struct // stat::st_blocks and struct statx::stx_blocks. (Note that this 512B block @@ -291,9 +389,5 @@ func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linu // SetStat implements vfs.FileDescriptionImpl.SetStat. func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { - if opts.Stat.Mask == 0 { - return nil - } - // TODO: implement inode.setStat - return syserror.EPERM + return fd.inode().setStat(opts.Stat) } diff --git a/pkg/sentry/platform/kvm/machine_amd64.go b/pkg/sentry/platform/kvm/machine_amd64.go index b99fe425e..873e39dc7 100644 --- a/pkg/sentry/platform/kvm/machine_amd64.go +++ b/pkg/sentry/platform/kvm/machine_amd64.go @@ -90,7 +90,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go index 7ae47f291..3b1f20219 100644 --- a/pkg/sentry/platform/kvm/machine_arm64.go +++ b/pkg/sentry/platform/kvm/machine_arm64.go @@ -97,7 +97,9 @@ func (m *machine) dropPageTables(pt *pagetables.PageTables) { // Clear from all PCIDs. for _, c := range m.vCPUs { - c.PCIDs.Drop(pt) + if c.PCIDs != nil { + c.PCIDs.Drop(pt) + } } } diff --git a/pkg/sentry/platform/ring0/entry_arm64.s b/pkg/sentry/platform/ring0/entry_arm64.s index 64e9c0845..da07815ff 100644 --- a/pkg/sentry/platform/ring0/entry_arm64.s +++ b/pkg/sentry/platform/ring0/entry_arm64.s @@ -474,6 +474,16 @@ TEXT ·El1_sync(SB),NOSPLIT,$0 B el1_invalid el1_da: + WORD $0xd538d092 //MRS TPIDR_EL1, R18 + WORD $0xd538601a //MRS FAR_EL1, R26 + + MOVD R26, CPU_FAULT_ADDR(RSV_REG) + + MOVD $0, CPU_ERROR_TYPE(RSV_REG) + + MOVD $PageFault, R3 + MOVD R3, CPU_VECTOR_CODE(RSV_REG) + B ·Halt(SB) el1_ia: diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4301b697c..1684dfc24 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -327,7 +327,7 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { } // PackTOS packs an IP_TOS socket control message. -func PackTOS(t *kernel.Task, tos int8, buf []byte) []byte { +func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { return putCmsgStruct( buf, linux.SOL_IP, diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index f7b069eed..2662fbc0f 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -138,13 +138,14 @@ var Metrics = tcpip.Stats{ }, }, IP: tcpip.IPStats{ - PacketsReceived: mustCreateMetric("/netstack/ip/packets_received", "Total number of IP packets received from the link layer in nic.DeliverNetworkPacket."), - InvalidAddressesReceived: mustCreateMetric("/netstack/ip/invalid_addresses_received", "Total number of IP packets received with an unknown or invalid destination address."), - PacketsDelivered: mustCreateMetric("/netstack/ip/packets_delivered", "Total number of incoming IP packets that are successfully delivered to the transport layer via HandlePacket."), - PacketsSent: mustCreateMetric("/netstack/ip/packets_sent", "Total number of IP packets sent via WritePacket."), - OutgoingPacketErrors: mustCreateMetric("/netstack/ip/outgoing_packet_errors", "Total number of IP packets which failed to write to a link-layer endpoint."), - MalformedPacketsReceived: mustCreateMetric("/netstack/ip/malformed_packets_received", "Total number of IP packets which failed IP header validation checks."), - MalformedFragmentsReceived: mustCreateMetric("/netstack/ip/malformed_fragments_received", "Total number of IP fragments which failed IP fragment validation checks."), + PacketsReceived: mustCreateMetric("/netstack/ip/packets_received", "Total number of IP packets received from the link layer in nic.DeliverNetworkPacket."), + InvalidDestinationAddressesReceived: mustCreateMetric("/netstack/ip/invalid_addresses_received", "Total number of IP packets received with an unknown or invalid destination address."), + InvalidSourceAddressesReceived: mustCreateMetric("/netstack/ip/invalid_source_addresses_received", "Total number of IP packets received with an unknown or invalid source address."), + PacketsDelivered: mustCreateMetric("/netstack/ip/packets_delivered", "Total number of incoming IP packets that are successfully delivered to the transport layer via HandlePacket."), + PacketsSent: mustCreateMetric("/netstack/ip/packets_sent", "Total number of IP packets sent via WritePacket."), + OutgoingPacketErrors: mustCreateMetric("/netstack/ip/outgoing_packet_errors", "Total number of IP packets which failed to write to a link-layer endpoint."), + MalformedPacketsReceived: mustCreateMetric("/netstack/ip/malformed_packets_received", "Total number of IP packets which failed IP header validation checks."), + MalformedFragmentsReceived: mustCreateMetric("/netstack/ip/malformed_fragments_received", "Total number of IP fragments which failed IP fragment validation checks."), }, TCP: tcpip.TCPStats{ ActiveConnectionOpenings: mustCreateMetric("/netstack/tcp/active_connection_openings", "Number of connections opened successfully via Connect."), @@ -324,22 +325,15 @@ func bytesToIPAddress(addr []byte) tcpip.Address { // converts it to the FullAddress format. It supports AF_UNIX, AF_INET, // AF_INET6, and AF_PACKET addresses. // -// strict indicates whether addresses with the AF_UNSPEC family are accepted of not. -// // AddressAndFamily returns an address and its family. -func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, uint16, *syserr.Error) { +func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { // Make sure we have at least 2 bytes for the address family. if len(addr) < 2 { return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument } - family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (strict || family != linux.AF_UNSPEC) { - return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported - } - // Get the rest of the fields based on the address family. - switch family { + switch family := usermem.ByteOrder.Uint16(addr); family { case linux.AF_UNIX: path := addr[2:] if len(path) > linux.UnixPathMax { @@ -638,10 +632,40 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { return r } +func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error { + if family == uint16(s.family) { + return nil + } + if !exact && family == linux.AF_INET && s.family == linux.AF_INET6 { + v, err := s.Endpoint.GetSockOptBool(tcpip.V6OnlyOption) + if err != nil { + return syserr.TranslateNetstackError(err) + } + if !v { + return nil + } + } + return syserr.ErrInvalidArgument +} + +// mapFamily maps the AF_INET ANY address to the IPv4-mapped IPv6 ANY if the +// receiver's family is AF_INET6. +// +// This is a hack to work around the fact that both IPv4 and IPv6 ANY are +// represented by the empty string. +// +// TODO(gvisor.dev/issues/1556): remove this function. +func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { + if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET { + addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" + } + return addr +} + // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - addr, family, err := AddressAndFamily(s.family, sockaddr, false /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } @@ -653,6 +677,12 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo } return syserr.TranslateNetstackError(err) } + + if err := s.checkFamily(family, false /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) + // Always return right away in the non-blocking case. if !blocking { return syserr.TranslateNetstackError(s.Endpoint.Connect(addr)) @@ -681,10 +711,14 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, _, err := AddressAndFamily(s.family, sockaddr, true /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } + if err := s.checkFamily(family, true /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) // Issue the bind request to the endpoint. return syserr.TranslateNetstackError(s.Endpoint.Bind(addr)) @@ -1243,11 +1277,11 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf if err != nil { return nil, syserr.TranslateNetstackError(err) } - var o uint32 + var o int32 if v { o = 1 } - return int32(o), nil + return o, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1352,6 +1386,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return int32(v), nil + case linux.IP_RECVTOS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIP(t, name) } @@ -1870,6 +1919,13 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + case linux.IP_RECVTOS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1890,7 +1946,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, linux.IP_RECVORIGDSTADDR, - linux.IP_RECVTOS, linux.IP_RECVTTL, linux.IP_RETOPTS, linux.IP_TRANSPARENT, @@ -2088,8 +2143,8 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) case linux.AF_INET6: var out linux.SockAddrInet6 - if len(addr.Addr) == 4 { - // Copy address is v4-mapped format. + if len(addr.Addr) == header.IPv4AddressSize { + // Copy address in v4-mapped format. copy(out.Addr[12:], addr.Addr) out.Addr[10] = 0xff out.Addr[11] = 0xff @@ -2310,7 +2365,14 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe } func (s *SocketOperations) controlMessages() socket.ControlMessages { - return socket.ControlMessages{IP: tcpip.ControlMessages{HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, Timestamp: s.readCM.Timestamp}} + return socket.ControlMessages{ + IP: tcpip.ControlMessages{ + HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, + Timestamp: s.readCM.Timestamp, + HasTOS: s.readCM.HasTOS, + TOS: s.readCM.TOS, + }, + } } // updateTimestamp sets the timestamp for SIOCGSTAMP. It should be called after @@ -2403,10 +2465,14 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] var addr *tcpip.FullAddress if len(to) > 0 { - addrBuf, _, err := AddressAndFamily(s.family, to, true /* strict */) + addrBuf, family, err := AddressAndFamily(to) if err != nil { return 0, err } + if err := s.checkFamily(family, false /* exact */); err != nil { + return 0, err + } + addrBuf = s.mapFamily(addrBuf, family) addr = &addrBuf } diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index a0db2d4fd..31ea66eca 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -148,25 +148,25 @@ func (s *Stack) Statistics(stat interface{}, arg string) error { case *inet.StatSNMPIP: ip := Metrics.IP *stats = inet.StatSNMPIP{ - 0, // TODO(gvisor.dev/issue/969): Support Ip/Forwarding. - 0, // TODO(gvisor.dev/issue/969): Support Ip/DefaultTTL. - ip.PacketsReceived.Value(), // InReceives. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InHdrErrors. - ip.InvalidAddressesReceived.Value(), // InAddrErrors. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ForwDatagrams. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InUnknownProtos. - 0, // TODO(gvisor.dev/issue/969): Support Ip/InDiscards. - ip.PacketsDelivered.Value(), // InDelivers. - ip.PacketsSent.Value(), // OutRequests. - ip.OutgoingPacketErrors.Value(), // OutDiscards. - 0, // TODO(gvisor.dev/issue/969): Support Ip/OutNoRoutes. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmTimeout. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmReqds. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragOKs. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragFails. - 0, // TODO(gvisor.dev/issue/969): Support Ip/FragCreates. + 0, // TODO(gvisor.dev/issue/969): Support Ip/Forwarding. + 0, // TODO(gvisor.dev/issue/969): Support Ip/DefaultTTL. + ip.PacketsReceived.Value(), // InReceives. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InHdrErrors. + ip.InvalidDestinationAddressesReceived.Value(), // InAddrErrors. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ForwDatagrams. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InUnknownProtos. + 0, // TODO(gvisor.dev/issue/969): Support Ip/InDiscards. + ip.PacketsDelivered.Value(), // InDelivers. + ip.PacketsSent.Value(), // OutRequests. + ip.OutgoingPacketErrors.Value(), // OutDiscards. + 0, // TODO(gvisor.dev/issue/969): Support Ip/OutNoRoutes. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmTimeout. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmReqds. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmOKs. + 0, // TODO(gvisor.dev/issue/969): Support Ip/ReasmFails. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragOKs. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragFails. + 0, // TODO(gvisor.dev/issue/969): Support Ip/FragCreates. } case *inet.StatSNMPICMP: in := Metrics.ICMP.V4PacketsReceived.ICMPv4PacketStats diff --git a/pkg/sentry/socket/rpcinet/BUILD b/pkg/sentry/socket/rpcinet/BUILD deleted file mode 100644 index 4668b87d1..000000000 --- a/pkg/sentry/socket/rpcinet/BUILD +++ /dev/null @@ -1,69 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") -load("@io_bazel_rules_go//proto:def.bzl", "go_proto_library") -load("@rules_cc//cc:defs.bzl", "cc_proto_library") - -package(licenses = ["notice"]) - -go_library( - name = "rpcinet", - srcs = [ - "device.go", - "rpcinet.go", - "socket.go", - "stack.go", - "stack_unsafe.go", - ], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet", - visibility = ["//pkg/sentry:internal"], - deps = [ - ":syscall_rpc_go_proto", - "//pkg/abi/linux", - "//pkg/binary", - "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/device", - "//pkg/sentry/fs", - "//pkg/sentry/fs/fsutil", - "//pkg/sentry/inet", - "//pkg/sentry/kernel", - "//pkg/sentry/kernel/time", - "//pkg/sentry/socket", - "//pkg/sentry/socket/hostinet", - "//pkg/sentry/socket/rpcinet/conn", - "//pkg/sentry/socket/rpcinet/notifier", - "//pkg/sentry/unimpl", - "//pkg/sentry/usermem", - "//pkg/syserr", - "//pkg/syserror", - "//pkg/tcpip", - "//pkg/tcpip/buffer", - "//pkg/tcpip/stack", - "//pkg/unet", - "//pkg/waiter", - ], -) - -proto_library( - name = "syscall_rpc_proto", - srcs = ["syscall_rpc.proto"], - visibility = [ - "//visibility:public", - ], -) - -cc_proto_library( - name = "syscall_rpc_cc_proto", - visibility = [ - "//visibility:public", - ], - deps = [":syscall_rpc_proto"], -) - -go_proto_library( - name = "syscall_rpc_go_proto", - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto", - proto = ":syscall_rpc_proto", - visibility = [ - "//visibility:public", - ], -) diff --git a/pkg/sentry/socket/rpcinet/conn/BUILD b/pkg/sentry/socket/rpcinet/conn/BUILD deleted file mode 100644 index b2677c659..000000000 --- a/pkg/sentry/socket/rpcinet/conn/BUILD +++ /dev/null @@ -1,18 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "conn", - srcs = ["conn.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn", - visibility = ["//pkg/sentry:internal"], - deps = [ - "//pkg/binary", - "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", - "//pkg/sync", - "//pkg/syserr", - "//pkg/unet", - "@com_github_golang_protobuf//proto:go_default_library", - ], -) diff --git a/pkg/sentry/socket/rpcinet/conn/conn.go b/pkg/sentry/socket/rpcinet/conn/conn.go deleted file mode 100644 index 02f39c767..000000000 --- a/pkg/sentry/socket/rpcinet/conn/conn.go +++ /dev/null @@ -1,187 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package conn is an RPC connection to a syscall RPC server. -package conn - -import ( - "fmt" - "sync/atomic" - "syscall" - - "github.com/golang/protobuf/proto" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/unet" - - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" -) - -type request struct { - response []byte - ready chan struct{} - ignoreResult bool -} - -// RPCConnection represents a single RPC connection to a syscall gofer. -type RPCConnection struct { - // reqID is the ID of the last request and must be accessed atomically. - reqID uint64 - - sendMu sync.Mutex - socket *unet.Socket - - reqMu sync.Mutex - requests map[uint64]request -} - -// NewRPCConnection initializes a RPC connection to a socket gofer. -func NewRPCConnection(s *unet.Socket) *RPCConnection { - conn := &RPCConnection{socket: s, requests: map[uint64]request{}} - go func() { // S/R-FIXME(b/77962828) - var nums [16]byte - for { - for n := 0; n < len(nums); { - nn, err := conn.socket.Read(nums[n:]) - if err != nil { - panic(fmt.Sprint("error reading length from socket rpc gofer: ", err)) - } - n += nn - } - - b := make([]byte, binary.LittleEndian.Uint64(nums[:8])) - id := binary.LittleEndian.Uint64(nums[8:]) - - for n := 0; n < len(b); { - nn, err := conn.socket.Read(b[n:]) - if err != nil { - panic(fmt.Sprint("error reading request from socket rpc gofer: ", err)) - } - n += nn - } - - conn.reqMu.Lock() - r := conn.requests[id] - if r.ignoreResult { - delete(conn.requests, id) - } else { - r.response = b - conn.requests[id] = r - } - conn.reqMu.Unlock() - close(r.ready) - } - }() - return conn -} - -// NewRequest makes a request to the RPC gofer and returns the request ID and a -// channel which will be closed once the request completes. -func (c *RPCConnection) NewRequest(req pb.SyscallRequest, ignoreResult bool) (uint64, chan struct{}) { - b, err := proto.Marshal(&req) - if err != nil { - panic(fmt.Sprint("invalid proto: ", err)) - } - - id := atomic.AddUint64(&c.reqID, 1) - ch := make(chan struct{}) - - c.reqMu.Lock() - c.requests[id] = request{ready: ch, ignoreResult: ignoreResult} - c.reqMu.Unlock() - - c.sendMu.Lock() - defer c.sendMu.Unlock() - - var nums [16]byte - binary.LittleEndian.PutUint64(nums[:8], uint64(len(b))) - binary.LittleEndian.PutUint64(nums[8:], id) - for n := 0; n < len(nums); { - nn, err := c.socket.Write(nums[n:]) - if err != nil { - panic(fmt.Sprint("error writing length and ID to socket gofer: ", err)) - } - n += nn - } - - for n := 0; n < len(b); { - nn, err := c.socket.Write(b[n:]) - if err != nil { - panic(fmt.Sprint("error writing request to socket gofer: ", err)) - } - n += nn - } - - return id, ch -} - -// RPCReadFile will execute the ReadFile helper RPC method which avoids the -// common pattern of open(2), read(2), close(2) by doing all three operations -// as a single RPC. It will read the entire file or return EFBIG if the file -// was too large. -func (c *RPCConnection) RPCReadFile(path string) ([]byte, *syserr.Error) { - req := &pb.SyscallRequest_ReadFile{&pb.ReadFileRequest{ - Path: path, - }} - - id, ch := c.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-ch - - res := c.Request(id).Result.(*pb.SyscallResponse_ReadFile).ReadFile.Result - if e, ok := res.(*pb.ReadFileResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.ReadFileResponse_Data).Data, nil -} - -// RPCWriteFile will execute the WriteFile helper RPC method which avoids the -// common pattern of open(2), write(2), write(2), close(2) by doing all -// operations as a single RPC. -func (c *RPCConnection) RPCWriteFile(path string, data []byte) (int64, *syserr.Error) { - req := &pb.SyscallRequest_WriteFile{&pb.WriteFileRequest{ - Path: path, - Content: data, - }} - - id, ch := c.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-ch - - res := c.Request(id).Result.(*pb.SyscallResponse_WriteFile).WriteFile - if e := res.ErrorNumber; e != 0 { - return int64(res.Written), syserr.FromHost(syscall.Errno(e)) - } - - return int64(res.Written), nil -} - -// Request retrieves the request corresponding to the given request ID. -// -// The channel returned by NewRequest must have been closed before Request can -// be called. This will happen automatically, do not manually close the -// channel. -func (c *RPCConnection) Request(id uint64) pb.SyscallResponse { - c.reqMu.Lock() - r := c.requests[id] - delete(c.requests, id) - c.reqMu.Unlock() - - var resp pb.SyscallResponse - if err := proto.Unmarshal(r.response, &resp); err != nil { - panic(fmt.Sprint("invalid proto: ", err)) - } - - return resp -} diff --git a/pkg/sentry/socket/rpcinet/notifier/BUILD b/pkg/sentry/socket/rpcinet/notifier/BUILD deleted file mode 100644 index a5954f22b..000000000 --- a/pkg/sentry/socket/rpcinet/notifier/BUILD +++ /dev/null @@ -1,17 +0,0 @@ -load("//tools/go_stateify:defs.bzl", "go_library") - -package(licenses = ["notice"]) - -go_library( - name = "notifier", - srcs = ["notifier.go"], - importpath = "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier", - visibility = ["//:sandbox"], - deps = [ - "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", - "//pkg/sentry/socket/rpcinet/conn", - "//pkg/sync", - "//pkg/waiter", - "@org_golang_x_sys//unix:go_default_library", - ], -) diff --git a/pkg/sentry/socket/rpcinet/notifier/notifier.go b/pkg/sentry/socket/rpcinet/notifier/notifier.go deleted file mode 100644 index 82b75d6dd..000000000 --- a/pkg/sentry/socket/rpcinet/notifier/notifier.go +++ /dev/null @@ -1,231 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package notifier implements an FD notifier implementation over RPC. -package notifier - -import ( - "fmt" - "syscall" - - "golang.org/x/sys/unix" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sync" - "gvisor.dev/gvisor/pkg/waiter" -) - -type fdInfo struct { - queue *waiter.Queue - waiting bool -} - -// Notifier holds all the state necessary to issue notifications when IO events -// occur in the observed FDs. -type Notifier struct { - // rpcConn is the connection that is used for sending RPCs. - rpcConn *conn.RPCConnection - - // epFD is the epoll file descriptor used to register for io - // notifications. - epFD uint32 - - // mu protects fdMap. - mu sync.Mutex - - // fdMap maps file descriptors to their notification queues and waiting - // status. - fdMap map[uint32]*fdInfo -} - -// NewRPCNotifier creates a new notifier object. -func NewRPCNotifier(cn *conn.RPCConnection) (*Notifier, error) { - id, c := cn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCreate1{&pb.EpollCreate1Request{}}}, false /* ignoreResult */) - <-c - - res := cn.Request(id).Result.(*pb.SyscallResponse_EpollCreate1).EpollCreate1.Result - if e, ok := res.(*pb.EpollCreate1Response_ErrorNumber); ok { - return nil, syscall.Errno(e.ErrorNumber) - } - - w := &Notifier{ - rpcConn: cn, - epFD: res.(*pb.EpollCreate1Response_Fd).Fd, - fdMap: make(map[uint32]*fdInfo), - } - - go w.waitAndNotify() // S/R-FIXME(b/77962828) - - return w, nil -} - -// waitFD waits on mask for fd. The fdMap mutex must be hold. -func (n *Notifier) waitFD(fd uint32, fi *fdInfo, mask waiter.EventMask) error { - if !fi.waiting && mask == 0 { - return nil - } - - e := pb.EpollEvent{ - Events: mask.ToLinux() | unix.EPOLLET, - Fd: fd, - } - - switch { - case !fi.waiting && mask != 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_ADD, Fd: fd, Event: &e}}}, false /* ignoreResult */) - <-c - - e := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollCtl).EpollCtl.ErrorNumber - if e != 0 { - return syscall.Errno(e) - } - - fi.waiting = true - case fi.waiting && mask == 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_DEL, Fd: fd}}}, false /* ignoreResult */) - <-c - n.rpcConn.Request(id) - - fi.waiting = false - case fi.waiting && mask != 0: - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollCtl{&pb.EpollCtlRequest{Epfd: n.epFD, Op: syscall.EPOLL_CTL_MOD, Fd: fd, Event: &e}}}, false /* ignoreResult */) - <-c - - e := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollCtl).EpollCtl.ErrorNumber - if e != 0 { - return syscall.Errno(e) - } - } - - return nil -} - -// addFD adds an FD to the list of FDs observed by n. -func (n *Notifier) addFD(fd uint32, queue *waiter.Queue) { - n.mu.Lock() - defer n.mu.Unlock() - - // Panic if we're already notifying on this FD. - if _, ok := n.fdMap[fd]; ok { - panic(fmt.Sprintf("File descriptor %d added twice", fd)) - } - - // We have nothing to wait for at the moment. Just add it to the map. - n.fdMap[fd] = &fdInfo{queue: queue} -} - -// updateFD updates the set of events the FD needs to be notified on. -func (n *Notifier) updateFD(fd uint32) error { - n.mu.Lock() - defer n.mu.Unlock() - - if fi, ok := n.fdMap[fd]; ok { - return n.waitFD(fd, fi, fi.queue.Events()) - } - - return nil -} - -// RemoveFD removes an FD from the list of FDs observed by n. -func (n *Notifier) removeFD(fd uint32) { - n.mu.Lock() - defer n.mu.Unlock() - - // Remove from map, then from epoll object. - n.waitFD(fd, n.fdMap[fd], 0) - delete(n.fdMap, fd) -} - -// hasFD returns true if the FD is in the list of observed FDs. -func (n *Notifier) hasFD(fd uint32) bool { - n.mu.Lock() - defer n.mu.Unlock() - - _, ok := n.fdMap[fd] - return ok -} - -// waitAndNotify loops waiting for io event notifications from the epoll -// object. Once notifications arrive, they are dispatched to the -// registered queue. -func (n *Notifier) waitAndNotify() error { - for { - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_EpollWait{&pb.EpollWaitRequest{Fd: n.epFD, NumEvents: 100, Msec: -1}}}, false /* ignoreResult */) - <-c - - res := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_EpollWait).EpollWait.Result - if e, ok := res.(*pb.EpollWaitResponse_ErrorNumber); ok { - err := syscall.Errno(e.ErrorNumber) - // NOTE(magi): I don't think epoll_wait can return EAGAIN but I'm being - // conseratively careful here since exiting the notification thread - // would be really bad. - if err == syscall.EINTR || err == syscall.EAGAIN { - continue - } - return err - } - - n.mu.Lock() - for _, e := range res.(*pb.EpollWaitResponse_Events).Events.Events { - if fi, ok := n.fdMap[e.Fd]; ok { - fi.queue.Notify(waiter.EventMaskFromLinux(e.Events)) - } - } - n.mu.Unlock() - } -} - -// AddFD adds an FD to the list of observed FDs. -func (n *Notifier) AddFD(fd uint32, queue *waiter.Queue) error { - n.addFD(fd, queue) - return nil -} - -// UpdateFD updates the set of events the FD needs to be notified on. -func (n *Notifier) UpdateFD(fd uint32) error { - return n.updateFD(fd) -} - -// RemoveFD removes an FD from the list of observed FDs. -func (n *Notifier) RemoveFD(fd uint32) { - n.removeFD(fd) -} - -// HasFD returns true if the FD is in the list of observed FDs. -// -// This should only be used by tests to assert that FDs are correctly -// registered. -func (n *Notifier) HasFD(fd uint32) bool { - return n.hasFD(fd) -} - -// NonBlockingPoll polls the given fd in non-blocking fashion. It is used just -// to query the FD's current state; this method will block on the RPC response -// although the syscall is non-blocking. -func (n *Notifier) NonBlockingPoll(fd uint32, mask waiter.EventMask) waiter.EventMask { - for { - id, c := n.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Poll{&pb.PollRequest{Fd: fd, Events: mask.ToLinux()}}}, false /* ignoreResult */) - <-c - - res := n.rpcConn.Request(id).Result.(*pb.SyscallResponse_Poll).Poll.Result - if e, ok := res.(*pb.PollResponse_ErrorNumber); ok { - if syscall.Errno(e.ErrorNumber) == syscall.EINTR { - continue - } - return mask - } - - return waiter.EventMaskFromLinux(res.(*pb.PollResponse_Events).Events) - } -} diff --git a/pkg/sentry/socket/rpcinet/rpcinet.go b/pkg/sentry/socket/rpcinet/rpcinet.go deleted file mode 100644 index 5d4fd4dac..000000000 --- a/pkg/sentry/socket/rpcinet/rpcinet.go +++ /dev/null @@ -1,16 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -// Package rpcinet implements sockets using an RPC for each syscall. -package rpcinet diff --git a/pkg/sentry/socket/rpcinet/socket.go b/pkg/sentry/socket/rpcinet/socket.go deleted file mode 100644 index ddb76d9d4..000000000 --- a/pkg/sentry/socket/rpcinet/socket.go +++ /dev/null @@ -1,909 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "sync/atomic" - "syscall" - "time" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" - "gvisor.dev/gvisor/pkg/sentry/kernel" - ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/socket" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sentry/unimpl" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/tcpip" - "gvisor.dev/gvisor/pkg/tcpip/buffer" - "gvisor.dev/gvisor/pkg/waiter" -) - -// socketOperations implements fs.FileOperations and socket.Socket for a socket -// implemented using a host socket. -type socketOperations struct { - fsutil.FilePipeSeek `state:"nosave"` - fsutil.FileNotDirReaddir `state:"nosave"` - fsutil.FileNoFsync `state:"nosave"` - fsutil.FileNoMMap `state:"nosave"` - fsutil.FileNoSplice `state:"nosave"` - fsutil.FileNoopFlush `state:"nosave"` - fsutil.FileUseInodeUnstableAttr `state:"nosave"` - socket.SendReceiveTimeout - - family int // Read-only. - stype linux.SockType // Read-only. - protocol int // Read-only. - - fd uint32 // must be O_NONBLOCK - wq *waiter.Queue - rpcConn *conn.RPCConnection - notifier *notifier.Notifier - - // shState is the state of the connection with respect to shutdown. Because - // we're mixing non-blocking semantics on the other side we have to adapt for - // some strange differences between blocking and non-blocking sockets. - shState int32 -} - -// Verify that we actually implement socket.Socket. -var _ = socket.Socket(&socketOperations{}) - -// New creates a new RPC socket. -func newSocketFile(ctx context.Context, stack *Stack, family int, skType linux.SockType, protocol int) (*fs.File, *syserr.Error) { - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Socket{&pb.SocketRequest{Family: int64(family), Type: int64(skType | syscall.SOCK_NONBLOCK), Protocol: int64(protocol)}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Socket).Socket.Result - if e, ok := res.(*pb.SocketResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - fd := res.(*pb.SocketResponse_Fd).Fd - - var wq waiter.Queue - stack.notifier.AddFD(fd, &wq) - - dirent := socket.NewDirent(ctx, socketDevice) - defer dirent.DecRef() - return fs.NewFile(ctx, dirent, fs.FileFlags{Read: true, Write: true}, &socketOperations{ - family: family, - stype: skType, - protocol: protocol, - wq: &wq, - fd: fd, - rpcConn: stack.rpcConn, - notifier: stack.notifier, - }), nil -} - -func isBlockingErrno(err error) bool { - return err == syscall.EAGAIN || err == syscall.EWOULDBLOCK -} - -func translateIOSyscallError(err error) error { - if isBlockingErrno(err) { - return syserror.ErrWouldBlock - } - return err -} - -// setShutdownFlags will set the shutdown flag so we can handle blocking reads -// after a read shutdown. -func (s *socketOperations) setShutdownFlags(how int) { - var f tcpip.ShutdownFlags - switch how { - case linux.SHUT_RD: - f = tcpip.ShutdownRead - case linux.SHUT_WR: - f = tcpip.ShutdownWrite - case linux.SHUT_RDWR: - f = tcpip.ShutdownWrite | tcpip.ShutdownRead - } - - // Atomically update the flags. - for { - old := atomic.LoadInt32(&s.shState) - if atomic.CompareAndSwapInt32(&s.shState, old, old|int32(f)) { - break - } - } -} - -func (s *socketOperations) resetShutdownFlags() { - atomic.StoreInt32(&s.shState, 0) -} - -func (s *socketOperations) isShutRdSet() bool { - return atomic.LoadInt32(&s.shState)&int32(tcpip.ShutdownRead) != 0 -} - -func (s *socketOperations) isShutWrSet() bool { - return atomic.LoadInt32(&s.shState)&int32(tcpip.ShutdownWrite) != 0 -} - -// Release implements fs.FileOperations.Release. -func (s *socketOperations) Release() { - s.notifier.RemoveFD(s.fd) - - // We always need to close the FD. - _, _ = s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Close{&pb.CloseRequest{Fd: s.fd}}}, true /* ignoreResult */) -} - -// Readiness implements waiter.Waitable.Readiness. -func (s *socketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { - return s.notifier.NonBlockingPoll(s.fd, mask) -} - -// EventRegister implements waiter.Waitable.EventRegister. -func (s *socketOperations) EventRegister(e *waiter.Entry, mask waiter.EventMask) { - s.wq.EventRegister(e, mask) - s.notifier.UpdateFD(s.fd) -} - -// EventUnregister implements waiter.Waitable.EventUnregister. -func (s *socketOperations) EventUnregister(e *waiter.Entry) { - s.wq.EventUnregister(e) - s.notifier.UpdateFD(s.fd) -} - -func rpcRead(t *kernel.Task, req *pb.SyscallRequest_Read) (*pb.ReadResponse_Data, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Read).Read.Result - if e, ok := res.(*pb.ReadResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.ReadResponse_Data), nil -} - -// Read implements fs.FileOperations.Read. -func (s *socketOperations) Read(ctx context.Context, _ *fs.File, dst usermem.IOSequence, _ int64) (int64, error) { - req := &pb.SyscallRequest_Read{&pb.ReadRequest{ - Fd: s.fd, - Length: uint32(dst.NumBytes()), - }} - - res, se := rpcRead(ctx.(*kernel.Task), req) - if se == nil { - n, e := dst.CopyOut(ctx, res.Data) - return int64(n), e - } - - return 0, se.ToError() -} - -func rpcWrite(t *kernel.Task, req *pb.SyscallRequest_Write) (uint32, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Write).Write.Result - if e, ok := res.(*pb.WriteResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.WriteResponse_Length).Length, nil -} - -// Write implements fs.FileOperations.Write. -func (s *socketOperations) Write(ctx context.Context, _ *fs.File, src usermem.IOSequence, _ int64) (int64, error) { - t := ctx.(*kernel.Task) - v := buffer.NewView(int(src.NumBytes())) - - // Copy all the data into the buffer. - if _, err := src.CopyIn(t, v); err != nil { - return 0, err - } - - n, err := rpcWrite(t, &pb.SyscallRequest_Write{&pb.WriteRequest{Fd: s.fd, Data: v}}) - if n > 0 && n < uint32(src.NumBytes()) { - // The FileOperations.Write interface expects us to return ErrWouldBlock in - // the event of a partial write. - return int64(n), syserror.ErrWouldBlock - } - return int64(n), err.ToError() -} - -func rpcConnect(t *kernel.Task, fd uint32, sockaddr []byte) *syserr.Error { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Connect{&pb.ConnectRequest{Fd: uint32(fd), Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Connect).Connect.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Connect implements socket.Socket.Connect. -func (s *socketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - if !blocking { - e := rpcConnect(t, s.fd, sockaddr) - if e == nil { - // Reset the shutdown state on new connects. - s.resetShutdownFlags() - } - return e - } - - // Register for notification when the endpoint becomes writable, then - // initiate the connection. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventOut|waiter.EventIn|waiter.EventHUp) - defer s.EventUnregister(&e) - for { - if err := rpcConnect(t, s.fd, sockaddr); err == nil || err != syserr.ErrInProgress && err != syserr.ErrAlreadyInProgress { - if err == nil { - // Reset the shutdown state on new connects. - s.resetShutdownFlags() - } - return err - } - - // It's pending, so we have to wait for a notification, and fetch the - // result once the wait completes. - if err := t.Block(ch); err != nil { - return syserr.FromError(err) - } - } -} - -func rpcAccept(t *kernel.Task, fd uint32, peer bool) (*pb.AcceptResponse_ResultPayload, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Accept{&pb.AcceptRequest{Fd: fd, Peer: peer, Flags: syscall.SOCK_NONBLOCK}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Accept).Accept.Result - if e, ok := res.(*pb.AcceptResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - return res.(*pb.AcceptResponse_Payload).Payload, nil -} - -// Accept implements socket.Socket.Accept. -func (s *socketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { - payload, se := rpcAccept(t, s.fd, peerRequested) - - // Check if we need to block. - if blocking && se == syserr.ErrTryAgain { - // Register for notifications. - e, ch := waiter.NewChannelEntry(nil) - // FIXME(b/119878986): This waiter.EventHUp is a partial - // measure, need to figure out how to translate linux events to - // internal events. - s.EventRegister(&e, waiter.EventIn|waiter.EventHUp) - defer s.EventUnregister(&e) - - // Try to accept the connection again; if it fails, then wait until we - // get a notification. - for { - if payload, se = rpcAccept(t, s.fd, peerRequested); se != syserr.ErrTryAgain { - break - } - - if err := t.Block(ch); err != nil { - return 0, nil, 0, syserr.FromError(err) - } - } - } - - // Handle any error from accept. - if se != nil { - return 0, nil, 0, se - } - - var wq waiter.Queue - s.notifier.AddFD(payload.Fd, &wq) - - dirent := socket.NewDirent(t, socketDevice) - defer dirent.DecRef() - fileFlags := fs.FileFlags{ - Read: true, - Write: true, - NonSeekable: true, - NonBlocking: flags&linux.SOCK_NONBLOCK != 0, - } - file := fs.NewFile(t, dirent, fileFlags, &socketOperations{ - family: s.family, - stype: s.stype, - protocol: s.protocol, - wq: &wq, - fd: payload.Fd, - rpcConn: s.rpcConn, - notifier: s.notifier, - }) - defer file.DecRef() - - fd, err := t.NewFDFrom(0, file, kernel.FDFlags{ - CloseOnExec: flags&linux.SOCK_CLOEXEC != 0, - }) - if err != nil { - return 0, nil, 0, syserr.FromError(err) - } - t.Kernel().RecordSocket(file) - - if peerRequested { - return fd, socket.UnmarshalSockAddr(s.family, payload.Address.Address), payload.Address.Length, nil - } - - return fd, nil, 0, nil -} - -// Bind implements socket.Socket.Bind. -func (s *socketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Bind{&pb.BindRequest{Fd: s.fd, Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Bind).Bind.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Listen implements socket.Socket.Listen. -func (s *socketOperations) Listen(t *kernel.Task, backlog int) *syserr.Error { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Listen{&pb.ListenRequest{Fd: s.fd, Backlog: int64(backlog)}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Listen).Listen.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// Shutdown implements socket.Socket.Shutdown. -func (s *socketOperations) Shutdown(t *kernel.Task, how int) *syserr.Error { - // We save the shutdown state because of strange differences on linux - // related to recvs on blocking vs. non-blocking sockets after a SHUT_RD. - // We need to emulate that behavior on the blocking side. - // TODO(b/120096741): There is a possible race that can exist with loopback, - // where data could possibly be lost. - s.setShutdownFlags(how) - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Shutdown{&pb.ShutdownRequest{Fd: s.fd, How: int64(how)}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Shutdown).Shutdown.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - - return nil -} - -// GetSockOpt implements socket.Socket.GetSockOpt. -func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPtr usermem.Addr, outLen int) (interface{}, *syserr.Error) { - // SO_RCVTIMEO and SO_SNDTIMEO are special because blocking is performed - // within the sentry. - if level == linux.SOL_SOCKET && name == linux.SO_RCVTIMEO { - if outLen < linux.SizeOfTimeval { - return nil, syserr.ErrInvalidArgument - } - - return linux.NsecToTimeval(s.RecvTimeout()), nil - } - if level == linux.SOL_SOCKET && name == linux.SO_SNDTIMEO { - if outLen < linux.SizeOfTimeval { - return nil, syserr.ErrInvalidArgument - } - - return linux.NsecToTimeval(s.SendTimeout()), nil - } - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetSockOpt{&pb.GetSockOptRequest{Fd: s.fd, Level: int64(level), Name: int64(name), Length: uint32(outLen)}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetSockOpt).GetSockOpt.Result - if e, ok := res.(*pb.GetSockOptResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.GetSockOptResponse_Opt).Opt, nil -} - -// SetSockOpt implements socket.Socket.SetSockOpt. -func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt []byte) *syserr.Error { - // Because blocking actually happens within the sentry we need to inspect - // this socket option to determine if it's a SO_RCVTIMEO or SO_SNDTIMEO, - // and if so, we will save it and use it as the deadline for recv(2) - // or send(2) related syscalls. - if level == linux.SOL_SOCKET && name == linux.SO_RCVTIMEO { - if len(opt) < linux.SizeOfTimeval { - return syserr.ErrInvalidArgument - } - - var v linux.Timeval - binary.Unmarshal(opt[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - if v.Usec < 0 || v.Usec >= int64(time.Second/time.Microsecond) { - return syserr.ErrDomain - } - s.SetRecvTimeout(v.ToNsecCapped()) - return nil - } - if level == linux.SOL_SOCKET && name == linux.SO_SNDTIMEO { - if len(opt) < linux.SizeOfTimeval { - return syserr.ErrInvalidArgument - } - - var v linux.Timeval - binary.Unmarshal(opt[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - if v.Usec < 0 || v.Usec >= int64(time.Second/time.Microsecond) { - return syserr.ErrDomain - } - s.SetSendTimeout(v.ToNsecCapped()) - return nil - } - - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_SetSockOpt{&pb.SetSockOptRequest{Fd: s.fd, Level: int64(level), Name: int64(name), Opt: opt}}}, false /* ignoreResult */) - <-c - - if e := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_SetSockOpt).SetSockOpt.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -// GetPeerName implements socket.Socket.GetPeerName. -func (s *socketOperations) GetPeerName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetPeerName{&pb.GetPeerNameRequest{Fd: s.fd}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetPeerName).GetPeerName.Result - if e, ok := res.(*pb.GetPeerNameResponse_ErrorNumber); ok { - return nil, 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - addr := res.(*pb.GetPeerNameResponse_Address).Address - return socket.UnmarshalSockAddr(s.family, addr.Address), addr.Length, nil -} - -// GetSockName implements socket.Socket.GetSockName. -func (s *socketOperations) GetSockName(t *kernel.Task) (linux.SockAddr, uint32, *syserr.Error) { - stack := t.NetworkContext().(*Stack) - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_GetSockName{&pb.GetSockNameRequest{Fd: s.fd}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_GetSockName).GetSockName.Result - if e, ok := res.(*pb.GetSockNameResponse_ErrorNumber); ok { - return nil, 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - addr := res.(*pb.GetSockNameResponse_Address).Address - return socket.UnmarshalSockAddr(s.family, addr.Address), addr.Length, nil -} - -func rpcIoctl(t *kernel.Task, fd, cmd uint32, arg []byte) ([]byte, error) { - stack := t.NetworkContext().(*Stack) - - id, c := stack.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Ioctl{&pb.IOCtlRequest{Fd: fd, Cmd: cmd, Arg: arg}}}, false /* ignoreResult */) - <-c - - res := stack.rpcConn.Request(id).Result.(*pb.SyscallResponse_Ioctl).Ioctl.Result - if e, ok := res.(*pb.IOCtlResponse_ErrorNumber); ok { - return nil, syscall.Errno(e.ErrorNumber) - } - - return res.(*pb.IOCtlResponse_Value).Value, nil -} - -// ifconfIoctlFromStack populates a struct ifconf for the SIOCGIFCONF ioctl. -func ifconfIoctlFromStack(ctx context.Context, io usermem.IO, ifc *linux.IFConf) error { - // If Ptr is NULL, return the necessary buffer size via Len. - // Otherwise, write up to Len bytes starting at Ptr containing ifreq - // structs. - t := ctx.(*kernel.Task) - s := t.NetworkContext().(*Stack) - if s == nil { - return syserr.ErrNoDevice.ToError() - } - - if ifc.Ptr == 0 { - ifc.Len = int32(len(s.Interfaces())) * int32(linux.SizeOfIFReq) - return nil - } - - max := ifc.Len - ifc.Len = 0 - for key, ifaceAddrs := range s.InterfaceAddrs() { - iface := s.Interfaces()[key] - for _, ifaceAddr := range ifaceAddrs { - // Don't write past the end of the buffer. - if ifc.Len+int32(linux.SizeOfIFReq) > max { - break - } - if ifaceAddr.Family != linux.AF_INET { - continue - } - - // Populate ifr.ifr_addr. - ifr := linux.IFReq{} - ifr.SetName(iface.Name) - usermem.ByteOrder.PutUint16(ifr.Data[0:2], uint16(ifaceAddr.Family)) - usermem.ByteOrder.PutUint16(ifr.Data[2:4], 0) - copy(ifr.Data[4:8], ifaceAddr.Addr[:4]) - - // Copy the ifr to userspace. - dst := uintptr(ifc.Ptr) + uintptr(ifc.Len) - ifc.Len += int32(linux.SizeOfIFReq) - if _, err := usermem.CopyObjectOut(ctx, io, usermem.Addr(dst), ifr, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return err - } - } - } - return nil -} - -// Ioctl implements fs.FileOperations.Ioctl. -func (s *socketOperations) Ioctl(ctx context.Context, _ *fs.File, io usermem.IO, args arch.SyscallArguments) (uintptr, error) { - t := ctx.(*kernel.Task) - - cmd := uint32(args[1].Int()) - arg := args[2].Pointer() - - var buf []byte - switch cmd { - // The following ioctls take 4 byte argument parameters. - case syscall.TIOCINQ, - syscall.TIOCOUTQ: - buf = make([]byte, 4) - // The following ioctls have args which are sizeof(struct ifreq). - case syscall.SIOCGIFADDR, - syscall.SIOCGIFBRDADDR, - syscall.SIOCGIFDSTADDR, - syscall.SIOCGIFFLAGS, - syscall.SIOCGIFHWADDR, - syscall.SIOCGIFINDEX, - syscall.SIOCGIFMAP, - syscall.SIOCGIFMETRIC, - syscall.SIOCGIFMTU, - syscall.SIOCGIFNAME, - syscall.SIOCGIFNETMASK, - syscall.SIOCGIFTXQLEN: - buf = make([]byte, linux.SizeOfIFReq) - case syscall.SIOCGIFCONF: - // SIOCGIFCONF has slightly different behavior than the others, in that it - // will need to populate the array of ifreqs. - var ifc linux.IFConf - if _, err := usermem.CopyObjectIn(ctx, io, args[2].Pointer(), &ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }); err != nil { - return 0, err - } - - if err := ifconfIoctlFromStack(ctx, io, &ifc); err != nil { - return 0, err - } - _, err := usermem.CopyObjectOut(ctx, io, args[2].Pointer(), ifc, usermem.IOOpts{ - AddressSpaceActive: true, - }) - - return 0, err - - case linux.SIOCGIFMEM, linux.SIOCGIFPFLAGS, linux.SIOCGMIIPHY, linux.SIOCGMIIREG: - unimpl.EmitUnimplementedEvent(ctx) - - default: - return 0, syserror.ENOTTY - } - - _, err := io.CopyIn(ctx, arg, buf, usermem.IOOpts{ - AddressSpaceActive: true, - }) - - if err != nil { - return 0, err - } - - v, err := rpcIoctl(t, s.fd, cmd, buf) - if err != nil { - return 0, err - } - - if len(v) != len(buf) { - return 0, syserror.EINVAL - } - - _, err = io.CopyOut(ctx, arg, v, usermem.IOOpts{ - AddressSpaceActive: true, - }) - return 0, err -} - -func rpcRecvMsg(t *kernel.Task, req *pb.SyscallRequest_Recvmsg) (*pb.RecvmsgResponse_ResultPayload, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Recvmsg).Recvmsg.Result - if e, ok := res.(*pb.RecvmsgResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.RecvmsgResponse_Payload).Payload, nil -} - -// Because we only support SO_TIMESTAMP we will search control messages for -// that value and set it if so, all other control messages will be ignored. -func (s *socketOperations) extractControlMessages(payload *pb.RecvmsgResponse_ResultPayload) socket.ControlMessages { - c := socket.ControlMessages{} - if len(payload.GetCmsgData()) > 0 { - // Parse the control messages looking for SO_TIMESTAMP. - msgs, e := syscall.ParseSocketControlMessage(payload.GetCmsgData()) - if e != nil { - return socket.ControlMessages{} - } - for _, m := range msgs { - if m.Header.Level != linux.SOL_SOCKET || m.Header.Type != linux.SO_TIMESTAMP { - continue - } - - // Let's parse the time stamp and set it. - if len(m.Data) < linux.SizeOfTimeval { - // Give up on locating the SO_TIMESTAMP option. - return socket.ControlMessages{} - } - - var v linux.Timeval - binary.Unmarshal(m.Data[:linux.SizeOfTimeval], usermem.ByteOrder, &v) - c.IP.HasTimestamp = true - c.IP.Timestamp = v.ToNsecCapped() - break - } - } - return c -} - -// RecvMsg implements socket.Socket.RecvMsg. -func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { - req := &pb.SyscallRequest_Recvmsg{&pb.RecvmsgRequest{ - Fd: s.fd, - Length: uint32(dst.NumBytes()), - Sender: senderRequested, - Trunc: flags&linux.MSG_TRUNC != 0, - Peek: flags&linux.MSG_PEEK != 0, - CmsgLength: uint32(controlDataLen), - }} - - res, err := rpcRecvMsg(t, req) - if err == nil { - var e error - var n int - if len(res.Data) > 0 { - n, e = dst.CopyOut(t, res.Data) - if e == nil && n != len(res.Data) { - panic("CopyOut failed to copy full buffer") - } - } - c := s.extractControlMessages(res) - return int(res.Length), 0, socket.UnmarshalSockAddr(s.family, res.Address.GetAddress()), res.Address.GetLength(), c, syserr.FromError(e) - } - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain || flags&linux.MSG_DONTWAIT != 0 { - return 0, 0, nil, 0, socket.ControlMessages{}, err - } - - // We'll have to block. Register for notifications and keep trying to - // send all the data. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventIn) - defer s.EventUnregister(&e) - - for { - res, err := rpcRecvMsg(t, req) - if err == nil { - var e error - var n int - if len(res.Data) > 0 { - n, e = dst.CopyOut(t, res.Data) - if e == nil && n != len(res.Data) { - panic("CopyOut failed to copy full buffer") - } - } - c := s.extractControlMessages(res) - return int(res.Length), 0, socket.UnmarshalSockAddr(s.family, res.Address.GetAddress()), res.Address.GetLength(), c, syserr.FromError(e) - } - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain { - return 0, 0, nil, 0, socket.ControlMessages{}, err - } - - if s.isShutRdSet() { - // Blocking would have caused us to block indefinitely so we return 0, - // this is the same behavior as Linux. - return 0, 0, nil, 0, socket.ControlMessages{}, nil - } - - if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { - if err == syserror.ETIMEDOUT { - return 0, 0, nil, 0, socket.ControlMessages{}, syserr.ErrTryAgain - } - return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) - } - } -} - -func rpcSendMsg(t *kernel.Task, req *pb.SyscallRequest_Sendmsg) (uint32, *syserr.Error) { - s := t.NetworkContext().(*Stack) - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Sendmsg).Sendmsg.Result - if e, ok := res.(*pb.SendmsgResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.SendmsgResponse_Length).Length, nil -} - -// SendMsg implements socket.Socket.SendMsg. -func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to []byte, flags int, haveDeadline bool, deadline ktime.Time, controlMessages socket.ControlMessages) (int, *syserr.Error) { - // Whitelist flags. - if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { - return 0, syserr.ErrInvalidArgument - } - - // Reject Unix control messages. - if !controlMessages.Unix.Empty() { - return 0, syserr.ErrInvalidArgument - } - - v := buffer.NewView(int(src.NumBytes())) - - // Copy all the data into the buffer. - if _, err := src.CopyIn(t, v); err != nil { - return 0, syserr.FromError(err) - } - - // TODO(bgeffon): this needs to change to map directly to a SendMsg syscall - // in the RPC. - totalWritten := 0 - n, err := rpcSendMsg(t, &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: uint32(s.fd), - Data: v, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }}) - - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain || flags&linux.MSG_DONTWAIT != 0 { - return int(n), err - } - - if n > 0 { - totalWritten += int(n) - v.TrimFront(int(n)) - } - - // We'll have to block. Register for notification and keep trying to - // send all the data. - e, ch := waiter.NewChannelEntry(nil) - s.EventRegister(&e, waiter.EventOut) - defer s.EventUnregister(&e) - - for { - n, err := rpcSendMsg(t, &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: uint32(s.fd), - Data: v, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }}) - - if n > 0 { - totalWritten += int(n) - v.TrimFront(int(n)) - - if err == nil && totalWritten < int(src.NumBytes()) { - continue - } - } - - if err != syserr.ErrWouldBlock && err != syserr.ErrTryAgain { - // We eat the error in this situation. - return int(totalWritten), nil - } - - if err := t.BlockWithDeadline(ch, haveDeadline, deadline); err != nil { - if err == syserror.ETIMEDOUT { - return int(totalWritten), syserr.ErrTryAgain - } - return int(totalWritten), syserr.FromError(err) - } - } -} - -// State implements socket.Socket.State. -func (s *socketOperations) State() uint32 { - // TODO(b/127845868): Define a new rpc to query the socket state. - return 0 -} - -// Type implements socket.Socket.Type. -func (s *socketOperations) Type() (family int, skType linux.SockType, protocol int) { - return s.family, s.stype, s.protocol -} - -type socketProvider struct { - family int -} - -// Socket implements socket.Provider.Socket. -func (p *socketProvider) Socket(t *kernel.Task, stypeflags linux.SockType, protocol int) (*fs.File, *syserr.Error) { - // Check that we are using the RPC network stack. - stack := t.NetworkContext() - if stack == nil { - return nil, nil - } - - s, ok := stack.(*Stack) - if !ok { - return nil, nil - } - - // Only accept TCP and UDP. - // - // Try to restrict the flags we will accept to minimize backwards - // incompatibility with netstack. - stype := stypeflags & linux.SOCK_TYPE_MASK - switch stype { - case syscall.SOCK_STREAM: - switch protocol { - case 0, syscall.IPPROTO_TCP: - // ok - default: - return nil, nil - } - case syscall.SOCK_DGRAM: - switch protocol { - case 0, syscall.IPPROTO_UDP: - // ok - default: - return nil, nil - } - default: - return nil, nil - } - - return newSocketFile(t, s, p.family, stype, protocol) -} - -// Pair implements socket.Provider.Pair. -func (p *socketProvider) Pair(t *kernel.Task, stype linux.SockType, protocol int) (*fs.File, *fs.File, *syserr.Error) { - // Not supported by AF_INET/AF_INET6. - return nil, nil, nil -} - -func init() { - for _, family := range []int{syscall.AF_INET, syscall.AF_INET6} { - socket.RegisterProvider(family, &socketProvider{family}) - } -} diff --git a/pkg/sentry/socket/rpcinet/stack.go b/pkg/sentry/socket/rpcinet/stack.go deleted file mode 100644 index f7878a760..000000000 --- a/pkg/sentry/socket/rpcinet/stack.go +++ /dev/null @@ -1,177 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "fmt" - "syscall" - - "gvisor.dev/gvisor/pkg/sentry/inet" - "gvisor.dev/gvisor/pkg/sentry/socket/hostinet" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" - "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/notifier" - "gvisor.dev/gvisor/pkg/syserr" - "gvisor.dev/gvisor/pkg/tcpip/stack" - "gvisor.dev/gvisor/pkg/unet" -) - -// Stack implements inet.Stack for RPC backed sockets. -type Stack struct { - interfaces map[int32]inet.Interface - interfaceAddrs map[int32][]inet.InterfaceAddr - routes []inet.Route - rpcConn *conn.RPCConnection - notifier *notifier.Notifier -} - -// NewStack returns a Stack containing the current state of the host network -// stack. -func NewStack(fd int32) (*Stack, error) { - sock, err := unet.NewSocket(int(fd)) - if err != nil { - return nil, err - } - - stack := &Stack{ - interfaces: make(map[int32]inet.Interface), - interfaceAddrs: make(map[int32][]inet.InterfaceAddr), - rpcConn: conn.NewRPCConnection(sock), - } - - var e error - stack.notifier, e = notifier.NewRPCNotifier(stack.rpcConn) - if e != nil { - return nil, e - } - - links, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETLINK) - if err != nil { - return nil, fmt.Errorf("RTM_GETLINK failed: %v", err) - } - - addrs, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETADDR) - if err != nil { - return nil, fmt.Errorf("RTM_GETADDR failed: %v", err) - } - - e = hostinet.ExtractHostInterfaces(links, addrs, stack.interfaces, stack.interfaceAddrs) - if e != nil { - return nil, e - } - - routes, err := stack.DoNetlinkRouteRequest(syscall.RTM_GETROUTE) - if err != nil { - return nil, fmt.Errorf("RTM_GETROUTE failed: %v", err) - } - - stack.routes, e = hostinet.ExtractHostRoutes(routes) - if e != nil { - return nil, e - } - - return stack, nil -} - -// RPCReadFile will execute the ReadFile helper RPC method which avoids the -// common pattern of open(2), read(2), close(2) by doing all three operations -// as a single RPC. It will read the entire file or return EFBIG if the file -// was too large. -func (s *Stack) RPCReadFile(path string) ([]byte, *syserr.Error) { - return s.rpcConn.RPCReadFile(path) -} - -// RPCWriteFile will execute the WriteFile helper RPC method which avoids the -// common pattern of open(2), write(2), write(2), close(2) by doing all -// operations as a single RPC. -func (s *Stack) RPCWriteFile(path string, data []byte) (int64, *syserr.Error) { - return s.rpcConn.RPCWriteFile(path, data) -} - -// Interfaces implements inet.Stack.Interfaces. -func (s *Stack) Interfaces() map[int32]inet.Interface { - interfaces := make(map[int32]inet.Interface) - for k, v := range s.interfaces { - interfaces[k] = v - } - return interfaces -} - -// InterfaceAddrs implements inet.Stack.InterfaceAddrs. -func (s *Stack) InterfaceAddrs() map[int32][]inet.InterfaceAddr { - addrs := make(map[int32][]inet.InterfaceAddr) - for k, v := range s.interfaceAddrs { - addrs[k] = append([]inet.InterfaceAddr(nil), v...) - } - return addrs -} - -// SupportsIPv6 implements inet.Stack.SupportsIPv6. -func (s *Stack) SupportsIPv6() bool { - panic("rpcinet handles procfs directly this method should not be called") -} - -// TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. -func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - panic("rpcinet handles procfs directly this method should not be called") -} - -// SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. -func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - panic("rpcinet handles procfs directly this method should not be called") - -} - -// TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. -func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - panic("rpcinet handles procfs directly this method should not be called") - -} - -// SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. -func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - panic("rpcinet handles procfs directly this method should not be called") -} - -// TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. -func (s *Stack) TCPSACKEnabled() (bool, error) { - panic("rpcinet handles procfs directly this method should not be called") -} - -// SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. -func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - panic("rpcinet handles procfs directly this method should not be called") -} - -// Statistics implements inet.Stack.Statistics. -func (s *Stack) Statistics(stat interface{}, arg string) error { - return syserr.ErrEndpointOperation.ToError() -} - -// RouteTable implements inet.Stack.RouteTable. -func (s *Stack) RouteTable() []inet.Route { - return append([]inet.Route(nil), s.routes...) -} - -// Resume implements inet.Stack.Resume. -func (s *Stack) Resume() {} - -// RegisteredEndpoints implements inet.Stack.RegisteredEndpoints. -func (s *Stack) RegisteredEndpoints() []stack.TransportEndpoint { return nil } - -// CleanupEndpoints implements inet.Stack.CleanupEndpoints. -func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } - -// RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. -func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} diff --git a/pkg/sentry/socket/rpcinet/stack_unsafe.go b/pkg/sentry/socket/rpcinet/stack_unsafe.go deleted file mode 100644 index a94bdad83..000000000 --- a/pkg/sentry/socket/rpcinet/stack_unsafe.go +++ /dev/null @@ -1,193 +0,0 @@ -// Copyright 2018 The gVisor Authors. -// -// Licensed under the Apache License, Version 2.0 (the "License"); -// you may not use this file except in compliance with the License. -// You may obtain a copy of the License at -// -// http://www.apache.org/licenses/LICENSE-2.0 -// -// Unless required by applicable law or agreed to in writing, software -// distributed under the License is distributed on an "AS IS" BASIS, -// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -// See the License for the specific language governing permissions and -// limitations under the License. - -package rpcinet - -import ( - "syscall" - "unsafe" - - "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" - pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/pkg/syserr" -) - -// NewNetlinkRouteRequest builds a netlink message for getting the RIB, -// the routing information base. -func newNetlinkRouteRequest(proto, seq, family int) []byte { - rr := &syscall.NetlinkRouteRequest{} - rr.Header.Len = uint32(syscall.NLMSG_HDRLEN + syscall.SizeofRtGenmsg) - rr.Header.Type = uint16(proto) - rr.Header.Flags = syscall.NLM_F_DUMP | syscall.NLM_F_REQUEST - rr.Header.Seq = uint32(seq) - rr.Data.Family = uint8(family) - return netlinkRRtoWireFormat(rr) -} - -func netlinkRRtoWireFormat(rr *syscall.NetlinkRouteRequest) []byte { - b := make([]byte, rr.Header.Len) - *(*uint32)(unsafe.Pointer(&b[0:4][0])) = rr.Header.Len - *(*uint16)(unsafe.Pointer(&b[4:6][0])) = rr.Header.Type - *(*uint16)(unsafe.Pointer(&b[6:8][0])) = rr.Header.Flags - *(*uint32)(unsafe.Pointer(&b[8:12][0])) = rr.Header.Seq - *(*uint32)(unsafe.Pointer(&b[12:16][0])) = rr.Header.Pid - b[16] = byte(rr.Data.Family) - return b -} - -func (s *Stack) getNetlinkFd() (uint32, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Socket{&pb.SocketRequest{Family: int64(syscall.AF_NETLINK), Type: int64(syscall.SOCK_RAW | syscall.SOCK_NONBLOCK), Protocol: int64(syscall.NETLINK_ROUTE)}}}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Socket).Socket.Result - if e, ok := res.(*pb.SocketResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - return res.(*pb.SocketResponse_Fd).Fd, nil -} - -func (s *Stack) bindNetlinkFd(fd uint32, sockaddr []byte) *syserr.Error { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Bind{&pb.BindRequest{Fd: fd, Address: sockaddr}}}, false /* ignoreResult */) - <-c - - if e := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Bind).Bind.ErrorNumber; e != 0 { - return syserr.FromHost(syscall.Errno(e)) - } - return nil -} - -func (s *Stack) closeNetlinkFd(fd uint32) { - _, _ = s.rpcConn.NewRequest(pb.SyscallRequest{Args: &pb.SyscallRequest_Close{&pb.CloseRequest{Fd: fd}}}, true /* ignoreResult */) -} - -func (s *Stack) rpcSendMsg(req *pb.SyscallRequest_Sendmsg) (uint32, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Sendmsg).Sendmsg.Result - if e, ok := res.(*pb.SendmsgResponse_ErrorNumber); ok { - return 0, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.SendmsgResponse_Length).Length, nil -} - -func (s *Stack) sendMsg(fd uint32, buf []byte, to []byte, flags int) (int, *syserr.Error) { - // Whitelist flags. - if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_EOR|syscall.MSG_FASTOPEN|syscall.MSG_MORE|syscall.MSG_NOSIGNAL) != 0 { - return 0, syserr.ErrInvalidArgument - } - - req := &pb.SyscallRequest_Sendmsg{&pb.SendmsgRequest{ - Fd: fd, - Data: buf, - Address: to, - More: flags&linux.MSG_MORE != 0, - EndOfRecord: flags&linux.MSG_EOR != 0, - }} - - n, err := s.rpcSendMsg(req) - return int(n), err -} - -func (s *Stack) rpcRecvMsg(req *pb.SyscallRequest_Recvmsg) (*pb.RecvmsgResponse_ResultPayload, *syserr.Error) { - id, c := s.rpcConn.NewRequest(pb.SyscallRequest{Args: req}, false /* ignoreResult */) - <-c - - res := s.rpcConn.Request(id).Result.(*pb.SyscallResponse_Recvmsg).Recvmsg.Result - if e, ok := res.(*pb.RecvmsgResponse_ErrorNumber); ok { - return nil, syserr.FromHost(syscall.Errno(e.ErrorNumber)) - } - - return res.(*pb.RecvmsgResponse_Payload).Payload, nil -} - -func (s *Stack) recvMsg(fd, l, flags uint32) ([]byte, *syserr.Error) { - req := &pb.SyscallRequest_Recvmsg{&pb.RecvmsgRequest{ - Fd: fd, - Length: l, - Sender: false, - Trunc: flags&linux.MSG_TRUNC != 0, - Peek: flags&linux.MSG_PEEK != 0, - }} - - res, err := s.rpcRecvMsg(req) - if err != nil { - return nil, err - } - return res.Data, nil -} - -func (s *Stack) netlinkRequest(proto, family int) ([]byte, error) { - fd, err := s.getNetlinkFd() - if err != nil { - return nil, err.ToError() - } - defer s.closeNetlinkFd(fd) - - lsa := syscall.SockaddrNetlink{Family: syscall.AF_NETLINK} - b := binary.Marshal(nil, usermem.ByteOrder, &lsa) - if err := s.bindNetlinkFd(fd, b); err != nil { - return nil, err.ToError() - } - - wb := newNetlinkRouteRequest(proto, 1, family) - _, err = s.sendMsg(fd, wb, b, 0) - if err != nil { - return nil, err.ToError() - } - - var tab []byte -done: - for { - rb, err := s.recvMsg(fd, uint32(syscall.Getpagesize()), 0) - nr := len(rb) - if err != nil { - return nil, err.ToError() - } - - if nr < syscall.NLMSG_HDRLEN { - return nil, syserr.ErrInvalidArgument.ToError() - } - - tab = append(tab, rb...) - msgs, e := syscall.ParseNetlinkMessage(rb) - if e != nil { - return nil, e - } - - for _, m := range msgs { - if m.Header.Type == syscall.NLMSG_DONE { - break done - } - if m.Header.Type == syscall.NLMSG_ERROR { - return nil, syserr.ErrInvalidArgument.ToError() - } - } - } - - return tab, nil -} - -// DoNetlinkRouteRequest returns routing information base, also known as RIB, -// which consists of network facility information, states and parameters. -func (s *Stack) DoNetlinkRouteRequest(req int) ([]syscall.NetlinkMessage, error) { - data, err := s.netlinkRequest(req, syscall.AF_UNSPEC) - if err != nil { - return nil, err - } - return syscall.ParseNetlinkMessage(data) -} diff --git a/pkg/sentry/socket/rpcinet/syscall_rpc.proto b/pkg/sentry/socket/rpcinet/syscall_rpc.proto deleted file mode 100644 index b677e9eb3..000000000 --- a/pkg/sentry/socket/rpcinet/syscall_rpc.proto +++ /dev/null @@ -1,352 +0,0 @@ -syntax = "proto3"; - -// package syscall_rpc is a set of networking related system calls that can be -// forwarded to a socket gofer. -// -package syscall_rpc; - -message SendmsgRequest { - uint32 fd = 1; - bytes data = 2 [ctype = CORD]; - bytes address = 3; - bool more = 4; - bool end_of_record = 5; -} - -message SendmsgResponse { - oneof result { - uint32 error_number = 1; - uint32 length = 2; - } -} - -message IOCtlRequest { - uint32 fd = 1; - uint32 cmd = 2; - bytes arg = 3; -} - -message IOCtlResponse { - oneof result { - uint32 error_number = 1; - bytes value = 2; - } -} - -message RecvmsgRequest { - uint32 fd = 1; - uint32 length = 2; - bool sender = 3; - bool peek = 4; - bool trunc = 5; - uint32 cmsg_length = 6; -} - -message OpenRequest { - bytes path = 1; - uint32 flags = 2; - uint32 mode = 3; -} - -message OpenResponse { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message ReadRequest { - uint32 fd = 1; - uint32 length = 2; -} - -message ReadResponse { - oneof result { - uint32 error_number = 1; - bytes data = 2 [ctype = CORD]; - } -} - -message ReadFileRequest { - string path = 1; -} - -message ReadFileResponse { - oneof result { - uint32 error_number = 1; - bytes data = 2 [ctype = CORD]; - } -} - -message WriteRequest { - uint32 fd = 1; - bytes data = 2 [ctype = CORD]; -} - -message WriteResponse { - oneof result { - uint32 error_number = 1; - uint32 length = 2; - } -} - -message WriteFileRequest { - string path = 1; - bytes content = 2; -} - -message WriteFileResponse { - uint32 error_number = 1; - uint32 written = 2; -} - -message AddressResponse { - bytes address = 1; - uint32 length = 2; -} - -message RecvmsgResponse { - message ResultPayload { - bytes data = 1 [ctype = CORD]; - AddressResponse address = 2; - uint32 length = 3; - bytes cmsg_data = 4; - } - oneof result { - uint32 error_number = 1; - ResultPayload payload = 2; - } -} - -message BindRequest { - uint32 fd = 1; - bytes address = 2; -} - -message BindResponse { - uint32 error_number = 1; -} - -message AcceptRequest { - uint32 fd = 1; - bool peer = 2; - int64 flags = 3; -} - -message AcceptResponse { - message ResultPayload { - uint32 fd = 1; - AddressResponse address = 2; - } - oneof result { - uint32 error_number = 1; - ResultPayload payload = 2; - } -} - -message ConnectRequest { - uint32 fd = 1; - bytes address = 2; -} - -message ConnectResponse { - uint32 error_number = 1; -} - -message ListenRequest { - uint32 fd = 1; - int64 backlog = 2; -} - -message ListenResponse { - uint32 error_number = 1; -} - -message ShutdownRequest { - uint32 fd = 1; - int64 how = 2; -} - -message ShutdownResponse { - uint32 error_number = 1; -} - -message CloseRequest { - uint32 fd = 1; -} - -message CloseResponse { - uint32 error_number = 1; -} - -message GetSockOptRequest { - uint32 fd = 1; - int64 level = 2; - int64 name = 3; - uint32 length = 4; -} - -message GetSockOptResponse { - oneof result { - uint32 error_number = 1; - bytes opt = 2; - } -} - -message SetSockOptRequest { - uint32 fd = 1; - int64 level = 2; - int64 name = 3; - bytes opt = 4; -} - -message SetSockOptResponse { - uint32 error_number = 1; -} - -message GetSockNameRequest { - uint32 fd = 1; -} - -message GetSockNameResponse { - oneof result { - uint32 error_number = 1; - AddressResponse address = 2; - } -} - -message GetPeerNameRequest { - uint32 fd = 1; -} - -message GetPeerNameResponse { - oneof result { - uint32 error_number = 1; - AddressResponse address = 2; - } -} - -message SocketRequest { - int64 family = 1; - int64 type = 2; - int64 protocol = 3; -} - -message SocketResponse { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message EpollWaitRequest { - uint32 fd = 1; - uint32 num_events = 2; - sint64 msec = 3; -} - -message EpollEvent { - uint32 fd = 1; - uint32 events = 2; -} - -message EpollEvents { - repeated EpollEvent events = 1; -} - -message EpollWaitResponse { - oneof result { - uint32 error_number = 1; - EpollEvents events = 2; - } -} - -message EpollCtlRequest { - uint32 epfd = 1; - int64 op = 2; - uint32 fd = 3; - EpollEvent event = 4; -} - -message EpollCtlResponse { - uint32 error_number = 1; -} - -message EpollCreate1Request { - int64 flag = 1; -} - -message EpollCreate1Response { - oneof result { - uint32 error_number = 1; - uint32 fd = 2; - } -} - -message PollRequest { - uint32 fd = 1; - uint32 events = 2; -} - -message PollResponse { - oneof result { - uint32 error_number = 1; - uint32 events = 2; - } -} - -message SyscallRequest { - oneof args { - SocketRequest socket = 1; - SendmsgRequest sendmsg = 2; - RecvmsgRequest recvmsg = 3; - BindRequest bind = 4; - AcceptRequest accept = 5; - ConnectRequest connect = 6; - ListenRequest listen = 7; - ShutdownRequest shutdown = 8; - CloseRequest close = 9; - GetSockOptRequest get_sock_opt = 10; - SetSockOptRequest set_sock_opt = 11; - GetSockNameRequest get_sock_name = 12; - GetPeerNameRequest get_peer_name = 13; - EpollWaitRequest epoll_wait = 14; - EpollCtlRequest epoll_ctl = 15; - EpollCreate1Request epoll_create1 = 16; - PollRequest poll = 17; - ReadRequest read = 18; - WriteRequest write = 19; - OpenRequest open = 20; - IOCtlRequest ioctl = 21; - WriteFileRequest write_file = 22; - ReadFileRequest read_file = 23; - } -} - -message SyscallResponse { - oneof result { - SocketResponse socket = 1; - SendmsgResponse sendmsg = 2; - RecvmsgResponse recvmsg = 3; - BindResponse bind = 4; - AcceptResponse accept = 5; - ConnectResponse connect = 6; - ListenResponse listen = 7; - ShutdownResponse shutdown = 8; - CloseResponse close = 9; - GetSockOptResponse get_sock_opt = 10; - SetSockOptResponse set_sock_opt = 11; - GetSockNameResponse get_sock_name = 12; - GetPeerNameResponse get_peer_name = 13; - EpollWaitResponse epoll_wait = 14; - EpollCtlResponse epoll_ctl = 15; - EpollCreate1Response epoll_create1 = 16; - PollResponse poll = 17; - ReadResponse read = 18; - WriteResponse write = 19; - OpenResponse open = 20; - IOCtlResponse ioctl = 21; - WriteFileResponse write_file = 22; - ReadFileResponse read_file = 23; - } -} diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 91effe89a..7f49ba864 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -116,13 +116,16 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) + addr, family, err := netstack.AddressAndFamily(sockaddr) if err != nil { if err == syserr.ErrAddressFamilyNotSupported { err = syserr.ErrInvalidArgument } return "", err } + if family != linux.AF_UNIX { + return "", syserr.ErrInvalidArgument + } // The address is trimmed by GetAddress. p := string(addr.Addr) diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index 51f2efb39..b6d7177f4 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -341,7 +341,7 @@ func sockAddr(t *kernel.Task, addr usermem.Addr, length uint32) string { switch family { case linux.AF_INET, linux.AF_INET6, linux.AF_UNIX: - fa, _, err := netstack.AddressAndFamily(int(family), b, true /* strict */) + fa, _, err := netstack.AddressAndFamily(b) if err != nil { return fmt.Sprintf("%#x {Family: %s, error extracting address: %v}", addr, familyStr, err) } diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index aa05e208a..917f74e07 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -13,6 +13,8 @@ go_library( "sigset.go", "sys_aio.go", "sys_capability.go", + "sys_clone_amd64.go", + "sys_clone_arm64.go", "sys_epoll.go", "sys_eventfd.go", "sys_file.go", @@ -40,6 +42,8 @@ go_library( "sys_socket.go", "sys_splice.go", "sys_stat.go", + "sys_stat_amd64.go", + "sys_stat_arm64.go", "sys_sync.go", "sys_sysinfo.go", "sys_syslog.go", diff --git a/pkg/sentry/syscalls/linux/linux64_amd64.go b/pkg/sentry/syscalls/linux/linux64_amd64.go index 479c5f6ff..c76771a54 100644 --- a/pkg/sentry/syscalls/linux/linux64_amd64.go +++ b/pkg/sentry/syscalls/linux/linux64_amd64.go @@ -228,12 +228,12 @@ var AMD64 = &kernel.SyscallTable{ 185: syscalls.Error("security", syserror.ENOSYS, "Not implemented in Linux.", nil), 186: syscalls.Supported("gettid", Gettid), 187: syscalls.Supported("readahead", Readahead), - 188: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), - 189: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 190: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 191: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), - 192: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 193: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 188: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 189: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 190: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 191: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 192: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 193: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), 194: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 195: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 196: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), diff --git a/pkg/sentry/syscalls/linux/linux64_arm64.go b/pkg/sentry/syscalls/linux/linux64_arm64.go index d3f61f5e8..d3587fda6 100644 --- a/pkg/sentry/syscalls/linux/linux64_arm64.go +++ b/pkg/sentry/syscalls/linux/linux64_arm64.go @@ -41,12 +41,12 @@ var ARM64 = &kernel.SyscallTable{ 2: syscalls.PartiallySupported("io_submit", IoSubmit, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 3: syscalls.PartiallySupported("io_cancel", IoCancel, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), 4: syscalls.PartiallySupported("io_getevents", IoGetevents, "Generally supported with exceptions. User ring optimizations are not implemented.", []string{"gvisor.dev/issue/204"}), - 5: syscalls.PartiallySupported("setxattr", Setxattr, "Only supported for tmpfs.", nil), - 6: syscalls.Error("lsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 7: syscalls.Error("fsetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 8: syscalls.PartiallySupported("getxattr", Getxattr, "Only supported for tmpfs.", nil), - 9: syscalls.ErrorWithEvent("lgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), - 10: syscalls.ErrorWithEvent("fgetxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), + 5: syscalls.PartiallySupported("setxattr", SetXattr, "Only supported for tmpfs.", nil), + 6: syscalls.PartiallySupported("lsetxattr", LSetXattr, "Only supported for tmpfs.", nil), + 7: syscalls.PartiallySupported("fsetxattr", FSetXattr, "Only supported for tmpfs.", nil), + 8: syscalls.PartiallySupported("getxattr", GetXattr, "Only supported for tmpfs.", nil), + 9: syscalls.PartiallySupported("lgetxattr", LGetXattr, "Only supported for tmpfs.", nil), + 10: syscalls.PartiallySupported("fgetxattr", FGetXattr, "Only supported for tmpfs.", nil), 11: syscalls.ErrorWithEvent("listxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 12: syscalls.ErrorWithEvent("llistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), 13: syscalls.ErrorWithEvent("flistxattr", syserror.ENOTSUP, "Requires filesystem support.", nil), @@ -61,6 +61,7 @@ var ARM64 = &kernel.SyscallTable{ 22: syscalls.Supported("epoll_pwait", EpollPwait), 23: syscalls.Supported("dup", Dup), 24: syscalls.Supported("dup3", Dup3), + 25: syscalls.PartiallySupported("fcntl", Fcntl, "Not all options are supported.", nil), 26: syscalls.Supported("inotify_init1", InotifyInit1), 27: syscalls.PartiallySupported("inotify_add_watch", InotifyAddWatch, "inotify events are only available inside the sandbox.", nil), 28: syscalls.PartiallySupported("inotify_rm_watch", InotifyRmWatch, "inotify events are only available inside the sandbox.", nil), @@ -78,7 +79,9 @@ var ARM64 = &kernel.SyscallTable{ 40: syscalls.PartiallySupported("mount", Mount, "Not all options or file systems are supported.", nil), 41: syscalls.Error("pivot_root", syserror.EPERM, "", nil), 42: syscalls.Error("nfsservctl", syserror.ENOSYS, "Removed after Linux 3.1.", nil), + 43: syscalls.PartiallySupported("statfs", Statfs, "Depends on the backing file system implementation.", nil), 44: syscalls.PartiallySupported("fstatfs", Fstatfs, "Depends on the backing file system implementation.", nil), + 45: syscalls.Supported("truncate", Truncate), 46: syscalls.Supported("ftruncate", Ftruncate), 47: syscalls.PartiallySupported("fallocate", Fallocate, "Not all options are supported.", nil), 48: syscalls.Supported("faccessat", Faccessat), @@ -112,6 +115,7 @@ var ARM64 = &kernel.SyscallTable{ 76: syscalls.PartiallySupported("splice", Splice, "Stub implementation.", []string{"gvisor.dev/issue/138"}), // TODO(b/29354098) 77: syscalls.Supported("tee", Tee), 78: syscalls.Supported("readlinkat", Readlinkat), + 79: syscalls.Supported("fstatat", Fstatat), 80: syscalls.Supported("fstat", Fstat), 81: syscalls.PartiallySupported("sync", Sync, "Full data flush is not guaranteed at this time.", nil), 82: syscalls.PartiallySupported("fsync", Fsync, "Full data flush is not guaranteed at this time.", nil), @@ -254,6 +258,8 @@ var ARM64 = &kernel.SyscallTable{ 219: syscalls.Error("keyctl", syserror.EACCES, "Not available to user.", nil), 220: syscalls.PartiallySupported("clone", Clone, "Mount namespace (CLONE_NEWNS) not supported. Options CLONE_PARENT, CLONE_SYSVSEM not supported.", nil), 221: syscalls.Supported("execve", Execve), + 222: syscalls.PartiallySupported("mmap", Mmap, "Generally supported with exceptions. Options MAP_FIXED_NOREPLACE, MAP_SHARED_VALIDATE, MAP_SYNC MAP_GROWSDOWN, MAP_HUGETLB are not supported.", nil), + 223: syscalls.PartiallySupported("fadvise64", Fadvise64, "Not all options are supported.", nil), 224: syscalls.CapError("swapon", linux.CAP_SYS_ADMIN, "", nil), 225: syscalls.CapError("swapoff", linux.CAP_SYS_ADMIN, "", nil), 226: syscalls.Supported("mprotect", Mprotect), @@ -299,6 +305,8 @@ var ARM64 = &kernel.SyscallTable{ 282: syscalls.ErrorWithEvent("userfaultfd", syserror.ENOSYS, "", []string{"gvisor.dev/issue/266"}), // TODO(b/118906345) 283: syscalls.ErrorWithEvent("membarrier", syserror.ENOSYS, "", []string{"gvisor.dev/issue/267"}), // TODO(gvisor.dev/issue/267) 284: syscalls.PartiallySupported("mlock2", Mlock2, "Stub implementation. The sandbox lacks appropriate permissions.", nil), + + // Syscalls after 284 are "backports" from versions of Linux after 4.4. 285: syscalls.ErrorWithEvent("copy_file_range", syserror.ENOSYS, "", nil), 286: syscalls.Supported("preadv2", Preadv2), 287: syscalls.PartiallySupported("pwritev2", Pwritev2, "Flag RWF_HIPRI is not supported.", nil), diff --git a/pkg/sentry/syscalls/linux/sys_clone_amd64.go b/pkg/sentry/syscalls/linux/sys_clone_amd64.go new file mode 100644 index 000000000..dd43cf18d --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_clone_amd64.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// Clone implements linux syscall clone(2). +// sys_clone has so many flavors. We implement the default one in linux 3.11 +// x86_64: +// sys_clone(clone_flags, newsp, parent_tidptr, child_tidptr, tls_val) +func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := int(args[0].Int()) + stack := args[1].Pointer() + parentTID := args[2].Pointer() + childTID := args[3].Pointer() + tls := args[4].Pointer() + return clone(t, flags, stack, parentTID, childTID, tls) +} diff --git a/pkg/sentry/syscalls/linux/sys_clone_arm64.go b/pkg/sentry/syscalls/linux/sys_clone_arm64.go new file mode 100644 index 000000000..cf68a8949 --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_clone_arm64.go @@ -0,0 +1,35 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// +build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/sentry/arch" + "gvisor.dev/gvisor/pkg/sentry/kernel" +) + +// Clone implements linux syscall clone(2). +// sys_clone has so many flavors, and we implement the default one in linux 3.11 +// arm64(kernel/fork.c with CONFIG_CLONE_BACKWARDS defined in the config file): +// sys_clone(clone_flags, newsp, parent_tidptr, tls_val, child_tidptr) +func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + flags := int(args[0].Int()) + stack := args[1].Pointer() + parentTID := args[2].Pointer() + tls := args[3].Pointer() + childTID := args[4].Pointer() + return clone(t, flags, stack, parentTID, childTID, tls) +} diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index 5556bc276..69b17b799 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -16,7 +16,6 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -125,56 +124,6 @@ func fstat(t *kernel.Task, f *fs.File, statAddr usermem.Addr) error { return copyOutStat(t, statAddr, f.Dirent.Inode.StableAttr, uattr) } -// copyOutStat copies the attributes (sattr, uattr) to the struct stat at -// address dst in t's address space. It encodes the stat struct to bytes -// manually, as stat() is a very common syscall for many applications, and -// t.CopyObjectOut has noticeable performance impact due to its many slice -// allocations and use of reflection. -func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { - b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] - - // Dev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) - // Ino (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) - // Nlink (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links) - // Mode (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) - // UID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) - // GID (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) - // Padding (uint32) - b = binary.AppendUint32(b, usermem.ByteOrder, 0) - // Rdev (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) - // Size (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) - // Blksize (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.BlockSize)) - // Blocks (uint64) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) - - // ATime - atime := uattr.AccessTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) - - // MTime - mtime := uattr.ModificationTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) - - // CTime - ctime := uattr.StatusChangeTime.Timespec() - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) - b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) - - _, err := t.CopyOutBytes(dst, b) - return err -} - // Statx implements linux syscall statx(2). func Statx(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { fd := args[0].Int() diff --git a/pkg/sentry/syscalls/linux/sys_stat_amd64.go b/pkg/sentry/syscalls/linux/sys_stat_amd64.go new file mode 100644 index 000000000..58afb4a9a --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_amd64.go @@ -0,0 +1,75 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build amd64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// copyOutStat copies the attributes (sattr, uattr) to the struct stat at +// address dst in t's address space. It encodes the stat struct to bytes +// manually, as stat() is a very common syscall for many applications, and +// t.CopyObjectOut has noticeable performance impact due to its many slice +// allocations and use of reflection. +func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { + b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] + + // Dev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) + // Ino (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) + // Nlink (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uattr.Links) + // Mode (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) + // UID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + // GID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) + // Padding (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, 0) + // Rdev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) + // Size (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) + // Blksize (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.BlockSize)) + // Blocks (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) + + // ATime + atime := uattr.AccessTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) + + // MTime + mtime := uattr.ModificationTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) + + // CTime + ctime := uattr.StatusChangeTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) + + _, err := t.CopyOutBytes(dst, b) + return err +} diff --git a/pkg/sentry/syscalls/linux/sys_stat_arm64.go b/pkg/sentry/syscalls/linux/sys_stat_arm64.go new file mode 100644 index 000000000..3e1251e0b --- /dev/null +++ b/pkg/sentry/syscalls/linux/sys_stat_arm64.go @@ -0,0 +1,77 @@ +// Copyright 2020 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//+build arm64 + +package linux + +import ( + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/usermem" +) + +// copyOutStat copies the attributes (sattr, uattr) to the struct stat at +// address dst in t's address space. It encodes the stat struct to bytes +// manually, as stat() is a very common syscall for many applications, and +// t.CopyObjectOut has noticeable performance impact due to its many slice +// allocations and use of reflection. +func copyOutStat(t *kernel.Task, dst usermem.Addr, sattr fs.StableAttr, uattr fs.UnstableAttr) error { + b := t.CopyScratchBuffer(int(linux.SizeOfStat))[:0] + + // Dev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.DeviceID)) + // Ino (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(sattr.InodeID)) + // Mode (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, sattr.Type.LinuxType()|uint32(uattr.Perms.LinuxMode())) + // Nlink (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Links)) + // UID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.UID.In(t.UserNamespace()).OrOverflow())) + // GID (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(uattr.Owner.GID.In(t.UserNamespace()).OrOverflow())) + // Rdev (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(linux.MakeDeviceID(sattr.DeviceFileMajor, sattr.DeviceFileMinor))) + // Padding (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, 0) + // Size (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Size)) + // Blksize (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, uint32(sattr.BlockSize)) + // Padding (uint32) + b = binary.AppendUint32(b, usermem.ByteOrder, 0) + // Blocks (uint64) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(uattr.Usage/512)) + + // ATime + atime := uattr.AccessTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(atime.Nsec)) + + // MTime + mtime := uattr.ModificationTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(mtime.Nsec)) + + // CTime + ctime := uattr.StatusChangeTime.Timespec() + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Sec)) + b = binary.AppendUint64(b, usermem.ByteOrder, uint64(ctime.Nsec)) + + _, err := t.CopyOutBytes(dst, b) + return err +} diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 4115116ff..b47c3b5c4 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -220,19 +220,6 @@ func clone(t *kernel.Task, flags int, stack usermem.Addr, parentTID usermem.Addr return uintptr(ntid), ctrl, err } -// Clone implements linux syscall clone(2). -// sys_clone has so many flavors. We implement the default one in linux 3.11 -// x86_64: -// sys_clone(clone_flags, newsp, parent_tidptr, child_tidptr, tls_val) -func Clone(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - flags := int(args[0].Int()) - stack := args[1].Pointer() - parentTID := args[2].Pointer() - childTID := args[3].Pointer() - tls := args[4].Pointer() - return clone(t, flags, stack, parentTID, childTID, tls) -} - // Fork implements Linux syscall fork(2). func Fork(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { // "A call to fork() is equivalent to a call to clone(2) specifying flags diff --git a/pkg/sentry/syscalls/linux/sys_xattr.go b/pkg/sentry/syscalls/linux/sys_xattr.go index 97d9a65ea..e35c077d6 100644 --- a/pkg/sentry/syscalls/linux/sys_xattr.go +++ b/pkg/sentry/syscalls/linux/sys_xattr.go @@ -25,12 +25,46 @@ import ( "gvisor.dev/gvisor/pkg/syserror" ) -// Getxattr implements linux syscall getxattr(2). -func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// GetXattr implements linux syscall getxattr(2). +func GetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getXattrFromPath(t, args, true) +} + +// LGetXattr implements linux syscall lgetxattr(2). +func LGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return getXattrFromPath(t, args, false) +} + +// FGetXattr implements linux syscall fgetxattr(2). +func FGetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := uint64(args[3].SizeT()) + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + n, value, err := getXattr(t, f.Dirent, nameAddr, size) + if err != nil { + return 0, nil, err + } + + if _, err := t.CopyOutBytes(valueAddr, []byte(value)); err != nil { + return 0, nil, err + } + return uintptr(n), nil, nil +} + +func getXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() - size := args[3].SizeT() + size := uint64(args[3].SizeT()) path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) if err != nil { @@ -38,21 +72,15 @@ func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } valueLen := 0 - err = fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - value, err := getxattr(t, d, dirPath, nameAddr) - if err != nil { - return err + err = fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR } - valueLen = len(value) - if size == 0 { - return nil - } - if size > linux.XATTR_SIZE_MAX { - size = linux.XATTR_SIZE_MAX - } - if valueLen > int(size) { - return syserror.ERANGE + n, value, err := getXattr(t, d, nameAddr, size) + valueLen = n + if err != nil { + return err } _, err = t.CopyOutBytes(valueAddr, []byte(value)) @@ -64,34 +92,78 @@ func Getxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return uintptr(valueLen), nil, nil } -// getxattr implements getxattr from the given *fs.Dirent. -func getxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr usermem.Addr) (string, error) { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return "", syserror.ENOTDIR - } - +// getXattr implements getxattr(2) from the given *fs.Dirent. +func getXattr(t *kernel.Task, d *fs.Dirent, nameAddr usermem.Addr, size uint64) (int, string, error) { if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Read: true}); err != nil { - return "", err + return 0, "", err } name, err := copyInXattrName(t, nameAddr) if err != nil { - return "", err + return 0, "", err } if !strings.HasPrefix(name, linux.XATTR_USER_PREFIX) { - return "", syserror.EOPNOTSUPP + return 0, "", syserror.EOPNOTSUPP + } + + // If getxattr(2) is called with size 0, the size of the value will be + // returned successfully even if it is nonzero. In that case, we need to + // retrieve the entire attribute value so we can return the correct size. + requestedSize := size + if size == 0 || size > linux.XATTR_SIZE_MAX { + requestedSize = linux.XATTR_SIZE_MAX + } + + value, err := d.Inode.GetXattr(t, name, requestedSize) + if err != nil { + return 0, "", err } + n := len(value) + if uint64(n) > requestedSize { + return 0, "", syserror.ERANGE + } + + // Don't copy out the attribute value if size is 0. + if size == 0 { + return n, "", nil + } + return n, value, nil +} + +// SetXattr implements linux syscall setxattr(2). +func SetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return setXattrFromPath(t, args, true) +} - return d.Inode.Getxattr(name) +// LSetXattr implements linux syscall lsetxattr(2). +func LSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + return setXattrFromPath(t, args, false) } -// Setxattr implements linux syscall setxattr(2). -func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { +// FSetXattr implements linux syscall fsetxattr(2). +func FSetXattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + fd := args[0].Int() + nameAddr := args[1].Pointer() + valueAddr := args[2].Pointer() + size := uint64(args[3].SizeT()) + flags := args[4].Uint() + + // TODO(b/113957122): Return EBADF if the fd was opened with O_PATH. + f := t.GetFile(fd) + if f == nil { + return 0, nil, syserror.EBADF + } + defer f.DecRef() + + return 0, nil, setXattr(t, f.Dirent, nameAddr, valueAddr, uint64(size), flags) +} + +func setXattrFromPath(t *kernel.Task, args arch.SyscallArguments, resolveSymlink bool) (uintptr, *kernel.SyscallControl, error) { pathAddr := args[0].Pointer() nameAddr := args[1].Pointer() valueAddr := args[2].Pointer() - size := args[3].SizeT() + size := uint64(args[3].SizeT()) flags := args[4].Uint() path, dirPath, err := copyInPath(t, pathAddr, false /* allowEmpty */) @@ -99,19 +171,19 @@ func Setxattr(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc return 0, nil, err } - if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { - return 0, nil, syserror.EINVAL - } + return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, resolveSymlink, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { + if dirPath && !fs.IsDir(d.Inode.StableAttr) { + return syserror.ENOTDIR + } - return 0, nil, fileOpOn(t, linux.AT_FDCWD, path, true /* resolve */, func(root *fs.Dirent, d *fs.Dirent, _ uint) error { - return setxattr(t, d, dirPath, nameAddr, valueAddr, size, flags) + return setXattr(t, d, nameAddr, valueAddr, uint64(size), flags) }) } -// setxattr implements setxattr from the given *fs.Dirent. -func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr usermem.Addr, size uint, flags uint32) error { - if dirPath && !fs.IsDir(d.Inode.StableAttr) { - return syserror.ENOTDIR +// setXattr implements setxattr(2) from the given *fs.Dirent. +func setXattr(t *kernel.Task, d *fs.Dirent, nameAddr, valueAddr usermem.Addr, size uint64, flags uint32) error { + if flags&^(linux.XATTR_CREATE|linux.XATTR_REPLACE) != 0 { + return syserror.EINVAL } if err := checkXattrPermissions(t, d.Inode, fs.PermMask{Write: true}); err != nil { @@ -127,7 +199,7 @@ func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr us return syserror.E2BIG } buf := make([]byte, size) - if _, err = t.CopyInBytes(valueAddr, buf); err != nil { + if _, err := t.CopyInBytes(valueAddr, buf); err != nil { return err } value := string(buf) @@ -136,7 +208,7 @@ func setxattr(t *kernel.Task, d *fs.Dirent, dirPath bool, nameAddr, valueAddr us return syserror.EOPNOTSUPP } - return d.Inode.Setxattr(name, value) + return d.Inode.SetXattr(t, d, name, value, flags) } func copyInXattrName(t *kernel.Task, nameAddr usermem.Addr) (string, error) { diff --git a/pkg/sentry/vfs/permissions.go b/pkg/sentry/vfs/permissions.go index f1edb0680..d279d05ca 100644 --- a/pkg/sentry/vfs/permissions.go +++ b/pkg/sentry/vfs/permissions.go @@ -30,6 +30,26 @@ const ( MayExec = 1 ) +// OnlyRead returns true if access _only_ allows read. +func (a AccessTypes) OnlyRead() bool { + return a == MayRead +} + +// MayRead returns true if access allows read. +func (a AccessTypes) MayRead() bool { + return a&MayRead != 0 +} + +// MayWrite returns true if access allows write. +func (a AccessTypes) MayWrite() bool { + return a&MayWrite != 0 +} + +// MayExec returns true if access allows exec. +func (a AccessTypes) MayExec() bool { + return a&MayExec != 0 +} + // GenericCheckPermissions checks that creds has the given access rights on a // file with the given permissions, UID, and GID, subject to the rules of // fs/namei.c:generic_permission(). isDir is true if the file is a directory. @@ -53,7 +73,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo } // CAP_DAC_READ_SEARCH allows the caller to read and search arbitrary // directories, and read arbitrary non-directory files. - if (isDir && (ats&MayWrite == 0)) || ats == MayRead { + if (isDir && !ats.MayWrite()) || ats.OnlyRead() { if creds.HasCapability(linux.CAP_DAC_READ_SEARCH) { return nil } @@ -61,7 +81,7 @@ func GenericCheckPermissions(creds *auth.Credentials, ats AccessTypes, isDir boo // CAP_DAC_OVERRIDE allows arbitrary access to directories, read/write // access to non-directory files, and execute access to non-directory files // for which at least one execute bit is set. - if isDir || (ats&MayExec == 0) || (mode&0111 != 0) { + if isDir || !ats.MayExec() || (mode&0111 != 0) { if creds.HasCapability(linux.CAP_DAC_OVERRIDE) { return nil } diff --git a/pkg/sleep/sleep_test.go b/pkg/sleep/sleep_test.go index 130806c86..af47e2ba1 100644 --- a/pkg/sleep/sleep_test.go +++ b/pkg/sleep/sleep_test.go @@ -376,6 +376,37 @@ func TestRace(t *testing.T) { } } +// TestRaceInOrder tests that multiple wakers can continuously send wake requests to +// the sleeper and that the wakers are retrieved in the order asserted. +func TestRaceInOrder(t *testing.T) { + const wakers = 100 + const wakeRequests = 10000 + + w := make([]Waker, wakers) + s := Sleeper{} + + // Associate each waker and start goroutines that will assert them. + for i := range w { + s.AddWaker(&w[i], i) + } + go func() { + n := 0 + for n < wakeRequests { + wk := w[n%len(w)] + wk.Assert() + n++ + } + }() + + // Wait for all wake up notifications from all wakers. + for i := 0; i < wakeRequests; i++ { + v, _ := s.Fetch(true) + if got, want := v, i%wakers; got != want { + t.Fatalf("got %d want %d", got, want) + } + } +} + // BenchmarkSleeperMultiSelect measures how long it takes to fetch a wake up // from 4 wakers when at least one is already asserted. func BenchmarkSleeperMultiSelect(b *testing.B) { diff --git a/pkg/tcpip/checker/checker.go b/pkg/tcpip/checker/checker.go index 2f15bf1f1..885d773b0 100644 --- a/pkg/tcpip/checker/checker.go +++ b/pkg/tcpip/checker/checker.go @@ -33,6 +33,9 @@ type NetworkChecker func(*testing.T, []header.Network) // TransportChecker is a function to check a property of a transport packet. type TransportChecker func(*testing.T, header.Transport) +// ControlMessagesChecker is a function to check a property of ancillary data. +type ControlMessagesChecker func(*testing.T, tcpip.ControlMessages) + // IPv4 checks the validity and properties of the given IPv4 packet. It is // expected to be used in conjunction with other network checkers for specific // properties. For example, to check the source and destination address, one @@ -158,6 +161,19 @@ func FragmentFlags(flags uint8) NetworkChecker { } } +// ReceiveTOS creates a checker that checks the TOS field in ControlMessages. +func ReceiveTOS(want uint8) ControlMessagesChecker { + return func(t *testing.T, cm tcpip.ControlMessages) { + t.Helper() + if !cm.HasTOS { + t.Fatalf("got cm.HasTOS = %t, want cm.TOS = %d", cm.HasTOS, want) + } + if got := cm.TOS; got != want { + t.Fatalf("got cm.TOS = %d, want %d", got, want) + } + } +} + // TOS creates a checker that checks the TOS field. func TOS(tos uint8, label uint32) NetworkChecker { return func(t *testing.T, h []header.Network) { @@ -754,3 +770,9 @@ func NDPNSTargetAddress(want tcpip.Address) TransportChecker { } } } + +// NDPRS creates a checker that checks that the packet contains a valid NDP +// Router Solicitation message (as per the raw wire format). +func NDPRS() NetworkChecker { + return NDP(header.ICMPv6RouterSolicit, header.NDPRSMinimumSize) +} diff --git a/pkg/tcpip/header/BUILD b/pkg/tcpip/header/BUILD index f2061c778..cd747d100 100644 --- a/pkg/tcpip/header/BUILD +++ b/pkg/tcpip/header/BUILD @@ -20,6 +20,7 @@ go_library( "ndp_neighbor_solicit.go", "ndp_options.go", "ndp_router_advert.go", + "ndp_router_solicit.go", "tcp.go", "udp.go", ], diff --git a/pkg/tcpip/header/ipv6.go b/pkg/tcpip/header/ipv6.go index 135a60b12..70e6ce095 100644 --- a/pkg/tcpip/header/ipv6.go +++ b/pkg/tcpip/header/ipv6.go @@ -84,6 +84,13 @@ const ( // The address is ff02::1. IPv6AllNodesMulticastAddress tcpip.Address = "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" + // IPv6AllRoutersMulticastAddress is a link-local multicast group that + // all IPv6 routers MUST join, as per RFC 4291, section 2.8. Packets + // destined to this address will reach all routers on a link. + // + // The address is ff02::2. + IPv6AllRoutersMulticastAddress tcpip.Address = "\xff\x02\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" + // IPv6MinimumMTU is the minimum MTU required by IPv6, per RFC 2460, // section 5. IPv6MinimumMTU = 1280 @@ -333,6 +340,17 @@ func IsV6LinkLocalAddress(addr tcpip.Address) bool { return addr[0] == 0xfe && (addr[1]&0xc0) == 0x80 } +// IsV6UniqueLocalAddress determines if the provided address is an IPv6 +// unique-local address (within the prefix FC00::/7). +func IsV6UniqueLocalAddress(addr tcpip.Address) bool { + if len(addr) != IPv6AddressSize { + return false + } + // According to RFC 4193 section 3.1, a unique local address has the prefix + // FC00::/7. + return (addr[0] & 0xfe) == 0xfc +} + // AppendOpaqueInterfaceIdentifier appends a 64 bit opaque interface identifier // (IID) to buf as outlined by RFC 7217 and returns the extended buffer. // @@ -371,3 +389,35 @@ func LinkLocalAddrWithOpaqueIID(nicName string, dadCounter uint8, secretKey []by return tcpip.Address(AppendOpaqueInterfaceIdentifier(lladdrb[:IIDOffsetInIPv6Address], IPv6LinkLocalPrefix.Subnet(), nicName, dadCounter, secretKey)) } + +// IPv6AddressScope is the scope of an IPv6 address. +type IPv6AddressScope int + +const ( + // LinkLocalScope indicates a link-local address. + LinkLocalScope IPv6AddressScope = iota + + // UniqueLocalScope indicates a unique-local address. + UniqueLocalScope + + // GlobalScope indicates a global address. + GlobalScope +) + +// ScopeForIPv6Address returns the scope for an IPv6 address. +func ScopeForIPv6Address(addr tcpip.Address) (IPv6AddressScope, *tcpip.Error) { + if len(addr) != IPv6AddressSize { + return GlobalScope, tcpip.ErrBadAddress + } + + switch { + case IsV6LinkLocalAddress(addr): + return LinkLocalScope, nil + + case IsV6UniqueLocalAddress(addr): + return UniqueLocalScope, nil + + default: + return GlobalScope, nil + } +} diff --git a/pkg/tcpip/header/ipv6_test.go b/pkg/tcpip/header/ipv6_test.go index 1994003ed..29f54bc57 100644 --- a/pkg/tcpip/header/ipv6_test.go +++ b/pkg/tcpip/header/ipv6_test.go @@ -25,7 +25,13 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/header" ) -const linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") +const ( + linkAddr = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + linkLocalAddr = tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + uniqueLocalAddr1 = tcpip.Address("\xfc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + uniqueLocalAddr2 = tcpip.Address("\xfd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + globalAddr = tcpip.Address("\xa0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") +) func TestEthernetAdddressToModifiedEUI64(t *testing.T) { expectedIID := [header.IIDSize]byte{0, 2, 3, 255, 254, 4, 5, 6} @@ -206,3 +212,91 @@ func TestLinkLocalAddrWithOpaqueIID(t *testing.T) { }) } } + +func TestIsV6UniqueLocalAddress(t *testing.T) { + tests := []struct { + name string + addr tcpip.Address + expected bool + }{ + { + name: "Valid Unique 1", + addr: uniqueLocalAddr1, + expected: true, + }, + { + name: "Valid Unique 2", + addr: uniqueLocalAddr1, + expected: true, + }, + { + name: "Link Local", + addr: linkLocalAddr, + expected: false, + }, + { + name: "Global", + addr: globalAddr, + expected: false, + }, + { + name: "IPv4", + addr: "\x01\x02\x03\x04", + expected: false, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + if got := header.IsV6UniqueLocalAddress(test.addr); got != test.expected { + t.Errorf("got header.IsV6UniqueLocalAddress(%s) = %t, want = %t", test.addr, got, test.expected) + } + }) + } +} + +func TestScopeForIPv6Address(t *testing.T) { + tests := []struct { + name string + addr tcpip.Address + scope header.IPv6AddressScope + err *tcpip.Error + }{ + { + name: "Unique Local", + addr: uniqueLocalAddr1, + scope: header.UniqueLocalScope, + err: nil, + }, + { + name: "Link Local", + addr: linkLocalAddr, + scope: header.LinkLocalScope, + err: nil, + }, + { + name: "Global", + addr: globalAddr, + scope: header.GlobalScope, + err: nil, + }, + { + name: "IPv4", + addr: "\x01\x02\x03\x04", + scope: header.GlobalScope, + err: tcpip.ErrBadAddress, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + got, err := header.ScopeForIPv6Address(test.addr) + if err != test.err { + t.Errorf("got header.IsV6UniqueLocalAddress(%s) = (_, %v), want = (_, %v)", test.addr, err, test.err) + } + if got != test.scope { + t.Errorf("got header.IsV6UniqueLocalAddress(%s) = (%d, _), want = (%d, _)", test.addr, got, test.scope) + } + }) + } +} diff --git a/pkg/tcpip/header/ndp_router_solicit.go b/pkg/tcpip/header/ndp_router_solicit.go new file mode 100644 index 000000000..9e67ba95d --- /dev/null +++ b/pkg/tcpip/header/ndp_router_solicit.go @@ -0,0 +1,36 @@ +// Copyright 2019 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package header + +// NDPRouterSolicit is an NDP Router Solicitation message. It will only contain +// the body of an ICMPv6 packet. +// +// See RFC 4861 section 4.1 for more details. +type NDPRouterSolicit []byte + +const ( + // NDPRSMinimumSize is the minimum size of a valid NDP Router + // Solicitation message (body of an ICMPv6 packet). + NDPRSMinimumSize = 4 + + // ndpRSOptionsOffset is the start of the NDP options in an + // NDPRouterSolicit. + ndpRSOptionsOffset = 4 +) + +// Options returns an NDPOptions of the the options body. +func (b NDPRouterSolicit) Options() NDPOptions { + return NDPOptions(b[ndpRSOptionsOffset:]) +} diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 5c813d8a0..fc06b5b87 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -154,6 +154,8 @@ func (it *IPTables) Check(hook Hook, pkt tcpip.PacketBuffer) bool { return false case Stolen, Queue, Repeat, None, Jump, Return, Continue: panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) + default: + panic(fmt.Sprintf("Unknown verdict %v.", verdict)) } } @@ -175,6 +177,8 @@ func (it *IPTables) checkTable(hook Hook, pkt tcpip.PacketBuffer, tablename stri continue case Stolen, Queue, Repeat, None, Jump, Return: panic(fmt.Sprintf("Unimplemented verdict %v.", verdict)) + default: + panic(fmt.Sprintf("Unknown verdict %v.", verdict)) } } diff --git a/pkg/tcpip/stack/BUILD b/pkg/tcpip/stack/BUILD index 705e984c1..783351a69 100644 --- a/pkg/tcpip/stack/BUILD +++ b/pkg/tcpip/stack/BUILD @@ -51,7 +51,7 @@ go_library( go_test( name = "stack_x_test", - size = "small", + size = "medium", srcs = [ "ndp_test.go", "stack_test.go", diff --git a/pkg/tcpip/stack/ndp.go b/pkg/tcpip/stack/ndp.go index a9dd322db..7d4b41dfa 100644 --- a/pkg/tcpip/stack/ndp.go +++ b/pkg/tcpip/stack/ndp.go @@ -17,6 +17,7 @@ package stack import ( "fmt" "log" + "math/rand" "time" "gvisor.dev/gvisor/pkg/tcpip" @@ -38,24 +39,36 @@ const ( // Default = 1s (from RFC 4861 section 10). defaultRetransmitTimer = time.Second + // defaultMaxRtrSolicitations is the default number of Router + // Solicitation messages to send when a NIC becomes enabled. + // + // Default = 3 (from RFC 4861 section 10). + defaultMaxRtrSolicitations = 3 + + // defaultRtrSolicitationInterval is the default amount of time between + // sending Router Solicitation messages. + // + // Default = 4s (from 4861 section 10). + defaultRtrSolicitationInterval = 4 * time.Second + + // defaultMaxRtrSolicitationDelay is the default maximum amount of time + // to wait before sending the first Router Solicitation message. + // + // Default = 1s (from 4861 section 10). + defaultMaxRtrSolicitationDelay = time.Second + // defaultHandleRAs is the default configuration for whether or not to // handle incoming Router Advertisements as a host. - // - // Default = true. defaultHandleRAs = true // defaultDiscoverDefaultRouters is the default configuration for // whether or not to discover default routers from incoming Router // Advertisements, as a host. - // - // Default = true. defaultDiscoverDefaultRouters = true // defaultDiscoverOnLinkPrefixes is the default configuration for // whether or not to discover on-link prefixes from incoming Router // Advertisements' Prefix Information option, as a host. - // - // Default = true. defaultDiscoverOnLinkPrefixes = true // defaultAutoGenGlobalAddresses is the default configuration for @@ -74,26 +87,31 @@ const ( // value of 0 means unspecified, so the smallest valid value is 1. // Note, the unit of the RetransmitTimer field in the Router // Advertisement is milliseconds. - // - // Min = 1ms. minimumRetransmitTimer = time.Millisecond + // minimumRtrSolicitationInterval is the minimum amount of time to wait + // between sending Router Solicitation messages. This limit is imposed + // to make sure that Router Solicitation messages are not sent all at + // once, defeating the purpose of sending the initial few messages. + minimumRtrSolicitationInterval = 500 * time.Millisecond + + // minimumMaxRtrSolicitationDelay is the minimum amount of time to wait + // before sending the first Router Solicitation message. It is 0 because + // we cannot have a negative delay. + minimumMaxRtrSolicitationDelay = 0 + // MaxDiscoveredDefaultRouters is the maximum number of discovered // default routers. The stack should stop discovering new routers after // discovering MaxDiscoveredDefaultRouters routers. // // This value MUST be at minimum 2 as per RFC 4861 section 6.3.4, and // SHOULD be more. - // - // Max = 10. MaxDiscoveredDefaultRouters = 10 // MaxDiscoveredOnLinkPrefixes is the maximum number of discovered // on-link prefixes. The stack should stop discovering new on-link // prefixes after discovering MaxDiscoveredOnLinkPrefixes on-link // prefixes. - // - // Max = 10. MaxDiscoveredOnLinkPrefixes = 10 // validPrefixLenForAutoGen is the expected prefix length that an @@ -245,9 +263,24 @@ type NDPConfigurations struct { // The amount of time to wait between sending Neighbor solicitation // messages. // - // Must be greater than 0.5s. + // Must be greater than or equal to 1ms. RetransmitTimer time.Duration + // The number of Router Solicitation messages to send when the NIC + // becomes enabled. + MaxRtrSolicitations uint8 + + // The amount of time between transmitting Router Solicitation messages. + // + // Must be greater than or equal to 0.5s. + RtrSolicitationInterval time.Duration + + // The maximum amount of time before transmitting the first Router + // Solicitation message. + // + // Must be greater than or equal to 0s. + MaxRtrSolicitationDelay time.Duration + // HandleRAs determines whether or not Router Advertisements will be // processed. HandleRAs bool @@ -278,12 +311,15 @@ type NDPConfigurations struct { // default values. func DefaultNDPConfigurations() NDPConfigurations { return NDPConfigurations{ - DupAddrDetectTransmits: defaultDupAddrDetectTransmits, - RetransmitTimer: defaultRetransmitTimer, - HandleRAs: defaultHandleRAs, - DiscoverDefaultRouters: defaultDiscoverDefaultRouters, - DiscoverOnLinkPrefixes: defaultDiscoverOnLinkPrefixes, - AutoGenGlobalAddresses: defaultAutoGenGlobalAddresses, + DupAddrDetectTransmits: defaultDupAddrDetectTransmits, + RetransmitTimer: defaultRetransmitTimer, + MaxRtrSolicitations: defaultMaxRtrSolicitations, + RtrSolicitationInterval: defaultRtrSolicitationInterval, + MaxRtrSolicitationDelay: defaultMaxRtrSolicitationDelay, + HandleRAs: defaultHandleRAs, + DiscoverDefaultRouters: defaultDiscoverDefaultRouters, + DiscoverOnLinkPrefixes: defaultDiscoverOnLinkPrefixes, + AutoGenGlobalAddresses: defaultAutoGenGlobalAddresses, } } @@ -292,10 +328,24 @@ func DefaultNDPConfigurations() NDPConfigurations { // // If RetransmitTimer is less than minimumRetransmitTimer, then a value of // defaultRetransmitTimer will be used. +// +// If RtrSolicitationInterval is less than minimumRtrSolicitationInterval, then +// a value of defaultRtrSolicitationInterval will be used. +// +// If MaxRtrSolicitationDelay is less than minimumMaxRtrSolicitationDelay, then +// a value of defaultMaxRtrSolicitationDelay will be used. func (c *NDPConfigurations) validate() { if c.RetransmitTimer < minimumRetransmitTimer { c.RetransmitTimer = defaultRetransmitTimer } + + if c.RtrSolicitationInterval < minimumRtrSolicitationInterval { + c.RtrSolicitationInterval = defaultRtrSolicitationInterval + } + + if c.MaxRtrSolicitationDelay < minimumMaxRtrSolicitationDelay { + c.MaxRtrSolicitationDelay = defaultMaxRtrSolicitationDelay + } } // ndpState is the per-interface NDP state. @@ -316,6 +366,10 @@ type ndpState struct { // Information option. onLinkPrefixes map[tcpip.Subnet]onLinkPrefixState + // The timer used to send the next router solicitation message. + // If routers are being solicited, rtrSolicitTimer MUST NOT be nil. + rtrSolicitTimer *time.Timer + // The addresses generated by SLAAC. autoGenAddresses map[tcpip.Address]autoGenAddressState @@ -378,13 +432,12 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, ref *ref // Should not attempt to perform DAD on an address that is currently in // the DAD process. if _, ok := ndp.dad[addr]; ok { - // Should never happen because we should only ever call this - // function for newly created addresses. If we attemped to - // "add" an address that already existed, we would returned an - // error since we attempted to add a duplicate address, or its - // reference count would have been increased without doing the - // work that would have been done for an address that was brand - // new. See NIC.addPermanentAddressLocked. + // Should never happen because we should only ever call this function for + // newly created addresses. If we attemped to "add" an address that already + // existed, we would get an error since we attempted to add a duplicate + // address, or its reference count would have been increased without doing + // the work that would have been done for an address that was brand new. + // See NIC.addAddressLocked. panic(fmt.Sprintf("ndpdad: already performing DAD for addr %s on NIC(%d)", addr, ndp.nic.ID())) } @@ -501,10 +554,12 @@ func (ndp *ndpState) doDuplicateAddressDetection(addr tcpip.Address, remaining u // address. panic(fmt.Sprintf("ndpdad: NIC(%d) is not in the solicited-node multicast group (%s) but it has addr %s", ndp.nic.ID(), snmc, addr)) } + snmcRef.incRef() // Use the unspecified address as the source address when performing // DAD. r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, snmc, ndp.nic.linkEP.LinkAddress(), snmcRef, false, false) + defer r.Release() hdr := buffer.NewPrependable(int(r.MaxHeaderLength()) + header.ICMPv6NeighborSolicitMinimumSize) pkt := header.ICMPv6(hdr.Prepend(header.ICMPv6NeighborSolicitMinimumSize)) @@ -1108,7 +1163,7 @@ func (ndp *ndpState) cleanupAutoGenAddrResourcesAndNotify(addr tcpip.Address) bo // // The NIC that ndp belongs to MUST be locked. func (ndp *ndpState) cleanupHostOnlyState() { - for addr, _ := range ndp.autoGenAddresses { + for addr := range ndp.autoGenAddresses { ndp.invalidateAutoGenAddress(addr) } @@ -1116,7 +1171,7 @@ func (ndp *ndpState) cleanupHostOnlyState() { log.Fatalf("ndp: still have auto-generated addresses after cleaning up, found = %d", got) } - for prefix, _ := range ndp.onLinkPrefixes { + for prefix := range ndp.onLinkPrefixes { ndp.invalidateOnLinkPrefix(prefix) } @@ -1124,7 +1179,7 @@ func (ndp *ndpState) cleanupHostOnlyState() { log.Fatalf("ndp: still have discovered on-link prefixes after cleaning up, found = %d", got) } - for router, _ := range ndp.defaultRouters { + for router := range ndp.defaultRouters { ndp.invalidateDefaultRouter(router) } @@ -1132,3 +1187,84 @@ func (ndp *ndpState) cleanupHostOnlyState() { log.Fatalf("ndp: still have discovered default routers after cleaning up, found = %d", got) } } + +// startSolicitingRouters starts soliciting routers, as per RFC 4861 section +// 6.3.7. If routers are already being solicited, this function does nothing. +// +// The NIC ndp belongs to MUST be locked. +func (ndp *ndpState) startSolicitingRouters() { + if ndp.rtrSolicitTimer != nil { + // We are already soliciting routers. + return + } + + remaining := ndp.configs.MaxRtrSolicitations + if remaining == 0 { + return + } + + // Calculate the random delay before sending our first RS, as per RFC + // 4861 section 6.3.7. + var delay time.Duration + if ndp.configs.MaxRtrSolicitationDelay > 0 { + delay = time.Duration(rand.Int63n(int64(ndp.configs.MaxRtrSolicitationDelay))) + } + + ndp.rtrSolicitTimer = time.AfterFunc(delay, func() { + // Send an RS message with the unspecified source address. + ref := ndp.nic.getRefOrCreateTemp(header.IPv6ProtocolNumber, header.IPv6Any, NeverPrimaryEndpoint, true) + r := makeRoute(header.IPv6ProtocolNumber, header.IPv6Any, header.IPv6AllRoutersMulticastAddress, ndp.nic.linkEP.LinkAddress(), ref, false, false) + defer r.Release() + + payloadSize := header.ICMPv6HeaderSize + header.NDPRSMinimumSize + hdr := buffer.NewPrependable(header.IPv6MinimumSize + payloadSize) + pkt := header.ICMPv6(hdr.Prepend(payloadSize)) + pkt.SetType(header.ICMPv6RouterSolicit) + pkt.SetChecksum(header.ICMPv6Checksum(pkt, r.LocalAddress, r.RemoteAddress, buffer.VectorisedView{})) + + sent := r.Stats().ICMP.V6PacketsSent + if err := r.WritePacket(nil, + NetworkHeaderParams{ + Protocol: header.ICMPv6ProtocolNumber, + TTL: header.NDPHopLimit, + TOS: DefaultTOS, + }, tcpip.PacketBuffer{Header: hdr}, + ); err != nil { + sent.Dropped.Increment() + log.Printf("startSolicitingRouters: error writing NDP router solicit message on NIC(%d); err = %s", ndp.nic.ID(), err) + // Don't send any more messages if we had an error. + remaining = 0 + } else { + sent.RouterSolicit.Increment() + remaining-- + } + + ndp.nic.mu.Lock() + defer ndp.nic.mu.Unlock() + if remaining == 0 { + ndp.rtrSolicitTimer = nil + } else if ndp.rtrSolicitTimer != nil { + // Note, we need to explicitly check to make sure that + // the timer field is not nil because if it was nil but + // we still reached this point, then we know the NIC + // was requested to stop soliciting routers so we don't + // need to send the next Router Solicitation message. + ndp.rtrSolicitTimer.Reset(ndp.configs.RtrSolicitationInterval) + } + }) + +} + +// stopSolicitingRouters stops soliciting routers. If routers are not currently +// being solicited, this function does nothing. +// +// The NIC ndp belongs to MUST be locked. +func (ndp *ndpState) stopSolicitingRouters() { + if ndp.rtrSolicitTimer == nil { + // Nothing to do. + return + } + + ndp.rtrSolicitTimer.Stop() + ndp.rtrSolicitTimer = nil +} diff --git a/pkg/tcpip/stack/ndp_test.go b/pkg/tcpip/stack/ndp_test.go index f9bc18c55..1a52e0e68 100644 --- a/pkg/tcpip/stack/ndp_test.go +++ b/pkg/tcpip/stack/ndp_test.go @@ -35,12 +35,12 @@ import ( ) const ( - addr1 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01" - addr2 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02" - addr3 = "\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03" - linkAddr1 = "\x02\x02\x03\x04\x05\x06" - linkAddr2 = "\x02\x02\x03\x04\x05\x07" - linkAddr3 = "\x02\x02\x03\x04\x05\x08" + addr1 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + addr2 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + addr3 = tcpip.Address("\x0a\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x03") + linkAddr1 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x06") + linkAddr2 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x07") + linkAddr3 = tcpip.LinkAddress("\x02\x02\x03\x04\x05\x08") defaultTimeout = 100 * time.Millisecond ) @@ -1732,9 +1732,11 @@ func stackAndNdpDispatcherWithDefaultRoute(t *testing.T, nicID tcpip.NICID) (*nd return ndpDisp, e, s } -// addrForNewConnection returns the local address used when creating a new -// connection. -func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address { +// addrForNewConnectionTo returns the local address used when creating a new +// connection to addr. +func addrForNewConnectionTo(t *testing.T, s *stack.Stack, addr tcpip.FullAddress) tcpip.Address { + t.Helper() + wq := waiter.Queue{} we, ch := waiter.NewChannelEntry(nil) wq.EventRegister(&we, waiter.EventIn) @@ -1748,8 +1750,8 @@ func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address { if err := ep.SetSockOptBool(tcpip.V6OnlyOption, true); err != nil { t.Fatalf("SetSockOpt(tcpip.V6OnlyOption, true): %s", err) } - if err := ep.Connect(dstAddr); err != nil { - t.Fatalf("ep.Connect(%+v): %s", dstAddr, err) + if err := ep.Connect(addr); err != nil { + t.Fatalf("ep.Connect(%+v): %s", addr, err) } got, err := ep.GetLocalAddress() if err != nil { @@ -1758,9 +1760,19 @@ func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address { return got.Addr } +// addrForNewConnection returns the local address used when creating a new +// connection. +func addrForNewConnection(t *testing.T, s *stack.Stack) tcpip.Address { + t.Helper() + + return addrForNewConnectionTo(t, s, dstAddr) +} + // addrForNewConnectionWithAddr returns the local address used when creating a // new connection with a specific local address. func addrForNewConnectionWithAddr(t *testing.T, s *stack.Stack, addr tcpip.FullAddress) tcpip.Address { + t.Helper() + wq := waiter.Queue{} we, ch := waiter.NewChannelEntry(nil) wq.EventRegister(&we, waiter.EventIn) @@ -2433,6 +2445,119 @@ func TestAutoGenAddrRemoval(t *testing.T) { } } +// TestAutoGenAddrAfterRemoval tests adding a SLAAC address that was previously +// assigned to the NIC but is in the permanentExpired state. +func TestAutoGenAddrAfterRemoval(t *testing.T) { + t.Parallel() + + const nicID = 1 + + prefix1, _, addr1 := prefixSubnetAddr(0, linkAddr1) + prefix2, _, addr2 := prefixSubnetAddr(1, linkAddr1) + ndpDisp, e, s := stackAndNdpDispatcherWithDefaultRoute(t, nicID) + + expectAutoGenAddrEvent := func(addr tcpip.AddressWithPrefix, eventType ndpAutoGenAddrEventType) { + t.Helper() + + select { + case e := <-ndpDisp.autoGenAddrC: + if diff := checkAutoGenAddrEvent(e, addr, eventType); diff != "" { + t.Errorf("auto-gen addr event mismatch (-want +got):\n%s", diff) + } + default: + t.Fatal("expected addr auto gen event") + } + } + + expectPrimaryAddr := func(addr tcpip.AddressWithPrefix) { + t.Helper() + + if got, err := s.GetMainNICAddress(nicID, header.IPv6ProtocolNumber); err != nil { + t.Fatalf("s.GetMainNICAddress(%d, %d): %s", nicID, header.IPv6ProtocolNumber, err) + } else if got != addr { + t.Errorf("got s.GetMainNICAddress(%d, %d) = %s, want = %s", nicID, header.IPv6ProtocolNumber, got, addr) + } + + if got := addrForNewConnection(t, s); got != addr.Address { + t.Errorf("got addrForNewConnection = %s, want = %s", got, addr.Address) + } + } + + // Receive a PI to auto-generate addr1 with a large valid and preferred + // lifetime. + const largeLifetimeSeconds = 999 + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix1, true, true, largeLifetimeSeconds, largeLifetimeSeconds)) + expectAutoGenAddrEvent(addr1, newAddr) + expectPrimaryAddr(addr1) + + // Add addr2 as a static address. + protoAddr2 := tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: addr2, + } + if err := s.AddProtocolAddressWithOptions(nicID, protoAddr2, stack.FirstPrimaryEndpoint); err != nil { + t.Fatalf("AddProtocolAddressWithOptions(%d, %+v, %d, %s) = %s", nicID, protoAddr2, stack.FirstPrimaryEndpoint, err) + } + // addr2 should be more preferred now since it is at the front of the primary + // list. + expectPrimaryAddr(addr2) + + // Get a route using addr2 to increment its reference count then remove it + // to leave it in the permanentExpired state. + r, err := s.FindRoute(nicID, addr2.Address, addr3, header.IPv6ProtocolNumber, false) + if err != nil { + t.Fatalf("FindRoute(%d, %s, %s, %d, false): %s", nicID, addr2.Address, addr3, header.IPv6ProtocolNumber, err) + } + defer r.Release() + if err := s.RemoveAddress(nicID, addr2.Address); err != nil { + t.Fatalf("s.RemoveAddress(%d, %s): %s", nicID, addr2.Address, err) + } + // addr1 should be preferred again since addr2 is in the expired state. + expectPrimaryAddr(addr1) + + // Receive a PI to auto-generate addr2 as valid and preferred. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, largeLifetimeSeconds)) + expectAutoGenAddrEvent(addr2, newAddr) + // addr2 should be more preferred now that it is closer to the front of the + // primary list and not deprecated. + expectPrimaryAddr(addr2) + + // Removing the address should result in an invalidation event immediately. + // It should still be in the permanentExpired state because r is still held. + // + // We remove addr2 here to make sure addr2 was marked as a SLAAC address + // (it was previously marked as a static address). + if err := s.RemoveAddress(1, addr2.Address); err != nil { + t.Fatalf("RemoveAddress(_, %s) = %s", addr2.Address, err) + } + expectAutoGenAddrEvent(addr2, invalidatedAddr) + // addr1 should be more preferred since addr2 is in the expired state. + expectPrimaryAddr(addr1) + + // Receive a PI to auto-generate addr2 as valid and deprecated. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, 0)) + expectAutoGenAddrEvent(addr2, newAddr) + // addr1 should still be more preferred since addr2 is deprecated, even though + // it is closer to the front of the primary list. + expectPrimaryAddr(addr1) + + // Receive a PI to refresh addr2's preferred lifetime. + e.InjectInbound(header.IPv6ProtocolNumber, raBufWithPI(llAddr3, 0, prefix2, true, true, largeLifetimeSeconds, largeLifetimeSeconds)) + select { + case <-ndpDisp.autoGenAddrC: + t.Fatal("unexpectedly got an auto gen addr event") + default: + } + // addr2 should be more preferred now that it is not deprecated. + expectPrimaryAddr(addr2) + + if err := s.RemoveAddress(1, addr2.Address); err != nil { + t.Fatalf("RemoveAddress(_, %s) = %s", addr2.Address, err) + } + expectAutoGenAddrEvent(addr2, invalidatedAddr) + expectPrimaryAddr(addr1) +} + // TestAutoGenAddrStaticConflict tests that if SLAAC generates an address that // is already assigned to the NIC, the static address remains. func TestAutoGenAddrStaticConflict(t *testing.T) { @@ -3086,3 +3211,227 @@ func TestDHCPv6ConfigurationFromNDPDA(t *testing.T) { e.InjectInbound(header.IPv6ProtocolNumber, raBufWithDHCPv6(llAddr2, false, true)) expectNoDHCPv6Event() } + +// TestRouterSolicitation tests the initial Router Solicitations that are sent +// when a NIC newly becomes enabled. +func TestRouterSolicitation(t *testing.T) { + t.Parallel() + + tests := []struct { + name string + maxRtrSolicit uint8 + rtrSolicitInt time.Duration + effectiveRtrSolicitInt time.Duration + maxRtrSolicitDelay time.Duration + effectiveMaxRtrSolicitDelay time.Duration + }{ + { + name: "Single RS with delay", + maxRtrSolicit: 1, + rtrSolicitInt: time.Second, + effectiveRtrSolicitInt: time.Second, + maxRtrSolicitDelay: time.Second, + effectiveMaxRtrSolicitDelay: time.Second, + }, + { + name: "Two RS with delay", + maxRtrSolicit: 2, + rtrSolicitInt: time.Second, + effectiveRtrSolicitInt: time.Second, + maxRtrSolicitDelay: 500 * time.Millisecond, + effectiveMaxRtrSolicitDelay: 500 * time.Millisecond, + }, + { + name: "Single RS without delay", + maxRtrSolicit: 1, + rtrSolicitInt: time.Second, + effectiveRtrSolicitInt: time.Second, + maxRtrSolicitDelay: 0, + effectiveMaxRtrSolicitDelay: 0, + }, + { + name: "Two RS without delay and invalid zero interval", + maxRtrSolicit: 2, + rtrSolicitInt: 0, + effectiveRtrSolicitInt: 4 * time.Second, + maxRtrSolicitDelay: 0, + effectiveMaxRtrSolicitDelay: 0, + }, + { + name: "Three RS without delay", + maxRtrSolicit: 3, + rtrSolicitInt: 500 * time.Millisecond, + effectiveRtrSolicitInt: 500 * time.Millisecond, + maxRtrSolicitDelay: 0, + effectiveMaxRtrSolicitDelay: 0, + }, + { + name: "Two RS with invalid negative delay", + maxRtrSolicit: 2, + rtrSolicitInt: time.Second, + effectiveRtrSolicitInt: time.Second, + maxRtrSolicitDelay: -3 * time.Second, + effectiveMaxRtrSolicitDelay: time.Second, + }, + } + + // This Run will not return until the parallel tests finish. + // + // We need this because we need to do some teardown work after the + // parallel tests complete. + // + // See https://godoc.org/testing#hdr-Subtests_and_Sub_benchmarks for + // more details. + t.Run("group", func(t *testing.T) { + for _, test := range tests { + test := test + + t.Run(test.name, func(t *testing.T) { + t.Parallel() + e := channel.New(int(test.maxRtrSolicit), 1280, linkAddr1) + waitForPkt := func(timeout time.Duration) { + t.Helper() + select { + case p := <-e.C: + if p.Proto != header.IPv6ProtocolNumber { + t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) + } + checker.IPv6(t, + p.Pkt.Header.View(), + checker.SrcAddr(header.IPv6Any), + checker.DstAddr(header.IPv6AllRoutersMulticastAddress), + checker.TTL(header.NDPHopLimit), + checker.NDPRS(), + ) + + case <-time.After(timeout): + t.Fatal("timed out waiting for packet") + } + } + waitForNothing := func(timeout time.Duration) { + t.Helper() + select { + case <-e.C: + t.Fatal("unexpectedly got a packet") + case <-time.After(timeout): + } + } + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + MaxRtrSolicitations: test.maxRtrSolicit, + RtrSolicitationInterval: test.rtrSolicitInt, + MaxRtrSolicitationDelay: test.maxRtrSolicitDelay, + }, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Make sure each RS got sent at the right + // times. + remaining := test.maxRtrSolicit + if remaining > 0 { + waitForPkt(test.effectiveMaxRtrSolicitDelay + defaultTimeout) + remaining-- + } + for ; remaining > 0; remaining-- { + waitForNothing(test.effectiveRtrSolicitInt - defaultTimeout) + waitForPkt(2 * defaultTimeout) + } + + // Make sure no more RS. + if test.effectiveRtrSolicitInt > test.effectiveMaxRtrSolicitDelay { + waitForNothing(test.effectiveRtrSolicitInt + defaultTimeout) + } else { + waitForNothing(test.effectiveMaxRtrSolicitDelay + defaultTimeout) + } + + // Make sure the counter got properly + // incremented. + if got, want := s.Stats().ICMP.V6PacketsSent.RouterSolicit.Value(), uint64(test.maxRtrSolicit); got != want { + t.Fatalf("got sent RouterSolicit = %d, want = %d", got, want) + } + }) + } + }) +} + +// TestStopStartSolicitingRouters tests that when forwarding is enabled or +// disabled, router solicitations are stopped or started, respecitively. +func TestStopStartSolicitingRouters(t *testing.T) { + t.Parallel() + + const interval = 500 * time.Millisecond + const delay = time.Second + const maxRtrSolicitations = 3 + e := channel.New(maxRtrSolicitations, 1280, linkAddr1) + waitForPkt := func(timeout time.Duration) { + t.Helper() + select { + case p := <-e.C: + if p.Proto != header.IPv6ProtocolNumber { + t.Fatalf("got Proto = %d, want = %d", p.Proto, header.IPv6ProtocolNumber) + } + checker.IPv6(t, p.Pkt.Header.View(), + checker.SrcAddr(header.IPv6Any), + checker.DstAddr(header.IPv6AllRoutersMulticastAddress), + checker.TTL(header.NDPHopLimit), + checker.NDPRS()) + + case <-time.After(timeout): + t.Fatal("timed out waiting for packet") + } + } + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + NDPConfigs: stack.NDPConfigurations{ + MaxRtrSolicitations: maxRtrSolicitations, + RtrSolicitationInterval: interval, + MaxRtrSolicitationDelay: delay, + }, + }) + if err := s.CreateNIC(1, e); err != nil { + t.Fatalf("CreateNIC(1) = %s", err) + } + + // Enable forwarding which should stop router solicitations. + s.SetForwarding(true) + select { + case <-e.C: + // A single RS may have been sent before forwarding was enabled. + select { + case <-e.C: + t.Fatal("Should not have sent more than one RS message") + case <-time.After(interval + defaultTimeout): + } + case <-time.After(delay + defaultTimeout): + } + + // Enabling forwarding again should do nothing. + s.SetForwarding(true) + select { + case <-e.C: + t.Fatal("unexpectedly got a packet after becoming a router") + case <-time.After(delay + defaultTimeout): + } + + // Disable forwarding which should start router solicitations. + s.SetForwarding(false) + waitForPkt(delay + defaultTimeout) + waitForPkt(interval + defaultTimeout) + waitForPkt(interval + defaultTimeout) + select { + case <-e.C: + t.Fatal("unexpectedly got an extra packet after sending out the expected RSs") + case <-time.After(interval + defaultTimeout): + } + + // Disabling forwarding again should do nothing. + s.SetForwarding(false) + select { + case <-e.C: + t.Fatal("unexpectedly got a packet after becoming a router") + case <-time.After(delay + defaultTimeout): + } +} diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index d50693634..de88c0bfa 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -15,6 +15,8 @@ package stack import ( + "log" + "sort" "strings" "sync/atomic" @@ -175,49 +177,72 @@ func (n *NIC) enable() *tcpip.Error { } // Do not auto-generate an IPv6 link-local address for loopback devices. - if !n.stack.autoGenIPv6LinkLocal || n.isLoopback() { - return nil - } + if n.stack.autoGenIPv6LinkLocal && !n.isLoopback() { + var addr tcpip.Address + if oIID := n.stack.opaqueIIDOpts; oIID.NICNameFromID != nil { + addr = header.LinkLocalAddrWithOpaqueIID(oIID.NICNameFromID(n.ID(), n.name), 0, oIID.SecretKey) + } else { + l2addr := n.linkEP.LinkAddress() - var addr tcpip.Address - if oIID := n.stack.opaqueIIDOpts; oIID.NICNameFromID != nil { - addr = header.LinkLocalAddrWithOpaqueIID(oIID.NICNameFromID(n.ID(), n.name), 0, oIID.SecretKey) - } else { - l2addr := n.linkEP.LinkAddress() + // Only attempt to generate the link-local address if we have a valid MAC + // address. + // + // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by + // LinkEndpoint.LinkAddress) before reaching this point. + if !header.IsValidUnicastEthernetAddress(l2addr) { + return nil + } - // Only attempt to generate the link-local address if we have a valid MAC - // address. - // - // TODO(b/141011931): Validate a LinkEndpoint's link address (provided by - // LinkEndpoint.LinkAddress) before reaching this point. - if !header.IsValidUnicastEthernetAddress(l2addr) { - return nil + addr = header.LinkLocalAddr(l2addr) } - addr = header.LinkLocalAddr(l2addr) + if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ + Protocol: header.IPv6ProtocolNumber, + AddressWithPrefix: tcpip.AddressWithPrefix{ + Address: addr, + PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen, + }, + }, CanBePrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { + return err + } } - _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ - Protocol: header.IPv6ProtocolNumber, - AddressWithPrefix: tcpip.AddressWithPrefix{ - Address: addr, - PrefixLen: header.IPv6LinkLocalPrefix.PrefixLen, - }, - }, CanBePrimaryEndpoint) + // If we are operating as a router, then do not solicit routers since we + // won't process the RAs anyways. + // + // Routers do not process Router Advertisements (RA) the same way a host + // does. That is, routers do not learn from RAs (e.g. on-link prefixes + // and default routers). Therefore, soliciting RAs from other routers on + // a link is unnecessary for routers. + if !n.stack.forwarding { + n.ndp.startSolicitingRouters() + } - return err + return nil } // becomeIPv6Router transitions n into an IPv6 router. // // When transitioning into an IPv6 router, host-only state (NDP discovered // routers, discovered on-link prefixes, and auto-generated addresses) will -// be cleaned up/invalidated. +// be cleaned up/invalidated and NDP router solicitations will be stopped. func (n *NIC) becomeIPv6Router() { n.mu.Lock() defer n.mu.Unlock() n.ndp.cleanupHostOnlyState() + n.ndp.stopSolicitingRouters() +} + +// becomeIPv6Host transitions n into an IPv6 host. +// +// When transitioning into an IPv6 host, NDP router solicitations will be +// started. +func (n *NIC) becomeIPv6Host() { + n.mu.Lock() + defer n.mu.Unlock() + + n.ndp.startSolicitingRouters() } // attachLinkEndpoint attaches the NIC to the endpoint, which will enable it @@ -251,13 +276,17 @@ func (n *NIC) setSpoofing(enable bool) { n.mu.Unlock() } -// primaryEndpoint returns the primary endpoint of n for the given network -// protocol. -// // primaryEndpoint will return the first non-deprecated endpoint if such an -// endpoint exists. If no non-deprecated endpoint exists, the first deprecated -// endpoint will be returned. -func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedNetworkEndpoint { +// endpoint exists for the given protocol and remoteAddr. If no non-deprecated +// endpoint exists, the first deprecated endpoint will be returned. +// +// If an IPv6 primary endpoint is requested, Source Address Selection (as +// defined by RFC 6724 section 5) will be performed. +func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber, remoteAddr tcpip.Address) *referencedNetworkEndpoint { + if protocol == header.IPv6ProtocolNumber && remoteAddr != "" { + return n.primaryIPv6Endpoint(remoteAddr) + } + n.mu.RLock() defer n.mu.RUnlock() @@ -296,6 +325,103 @@ func (n *NIC) primaryEndpoint(protocol tcpip.NetworkProtocolNumber) *referencedN return deprecatedEndpoint } +// ipv6AddrCandidate is an IPv6 candidate for Source Address Selection (RFC +// 6724 section 5). +type ipv6AddrCandidate struct { + ref *referencedNetworkEndpoint + scope header.IPv6AddressScope +} + +// primaryIPv6Endpoint returns an IPv6 endpoint following Source Address +// Selection (RFC 6724 section 5). +// +// Note, only rules 1-3 are followed. +// +// remoteAddr must be a valid IPv6 address. +func (n *NIC) primaryIPv6Endpoint(remoteAddr tcpip.Address) *referencedNetworkEndpoint { + n.mu.RLock() + defer n.mu.RUnlock() + + primaryAddrs := n.primary[header.IPv6ProtocolNumber] + + if len(primaryAddrs) == 0 { + return nil + } + + // Create a candidate set of available addresses we can potentially use as a + // source address. + cs := make([]ipv6AddrCandidate, 0, len(primaryAddrs)) + for _, r := range primaryAddrs { + // If r is not valid for outgoing connections, it is not a valid endpoint. + if !r.isValidForOutgoing() { + continue + } + + addr := r.ep.ID().LocalAddress + scope, err := header.ScopeForIPv6Address(addr) + if err != nil { + // Should never happen as we got r from the primary IPv6 endpoint list and + // ScopeForIPv6Address only returns an error if addr is not an IPv6 + // address. + log.Fatalf("header.ScopeForIPv6Address(%s): %s", addr, err) + } + + cs = append(cs, ipv6AddrCandidate{ + ref: r, + scope: scope, + }) + } + + remoteScope, err := header.ScopeForIPv6Address(remoteAddr) + if err != nil { + // primaryIPv6Endpoint should never be called with an invalid IPv6 address. + log.Fatalf("header.ScopeForIPv6Address(%s): %s", remoteAddr, err) + } + + // Sort the addresses as per RFC 6724 section 5 rules 1-3. + // + // TODO(b/146021396): Implement rules 4-8 of RFC 6724 section 5. + sort.Slice(cs, func(i, j int) bool { + sa := cs[i] + sb := cs[j] + + // Prefer same address as per RFC 6724 section 5 rule 1. + if sa.ref.ep.ID().LocalAddress == remoteAddr { + return true + } + if sb.ref.ep.ID().LocalAddress == remoteAddr { + return false + } + + // Prefer appropriate scope as per RFC 6724 section 5 rule 2. + if sa.scope < sb.scope { + return sa.scope >= remoteScope + } else if sb.scope < sa.scope { + return sb.scope < remoteScope + } + + // Avoid deprecated addresses as per RFC 6724 section 5 rule 3. + if saDep, sbDep := sa.ref.deprecated, sb.ref.deprecated; saDep != sbDep { + // If sa is not deprecated, it is preferred over sb. + return sbDep + } + + // sa and sb are equal, return the endpoint that is closest to the front of + // the primary endpoint list. + return i < j + }) + + // Return the most preferred address that can have its reference count + // incremented. + for _, c := range cs { + if r := c.ref; r.tryIncRef() { + return r + } + } + + return nil +} + // hasPermanentAddrLocked returns true if n has a permanent (including currently // tentative) address, addr. func (n *NIC) hasPermanentAddrLocked(addr tcpip.Address) bool { @@ -407,18 +533,34 @@ func (n *NIC) getRefOrCreateTemp(protocol tcpip.NetworkProtocolNumber, address t return ref } -func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) (*referencedNetworkEndpoint, *tcpip.Error) { - id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address} +// addAddressLocked adds a new protocolAddress to n. +// +// If n already has the address in a non-permanent state, and the kind given is +// permanent, that address will be promoted in place and its properties set to +// the properties provided. Otherwise, it returns tcpip.ErrDuplicateAddress. +func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) { + // TODO(b/141022673): Validate IP addresses before adding them. + + // Sanity check. + id := NetworkEndpointID{LocalAddress: protocolAddress.AddressWithPrefix.Address} if ref, ok := n.endpoints[id]; ok { + // Endpoint already exists. + if kind != permanent { + return nil, tcpip.ErrDuplicateAddress + } switch ref.getKind() { case permanentTentative, permanent: // The NIC already have a permanent endpoint with that address. return nil, tcpip.ErrDuplicateAddress case permanentExpired, temporary: - // Promote the endpoint to become permanent and respect - // the new peb. + // Promote the endpoint to become permanent and respect the new peb, + // configType and deprecated status. if ref.tryIncRef() { + // TODO(b/147748385): Perform Duplicate Address Detection when promoting + // an IPv6 endpoint to permanent. ref.setKind(permanent) + ref.deprecated = deprecated + ref.configType = configType refs := n.primary[ref.protocol] for i, r := range refs { @@ -450,19 +592,6 @@ func (n *NIC) addPermanentAddressLocked(protocolAddress tcpip.ProtocolAddress, p } } - return n.addAddressLocked(protocolAddress, peb, permanent, static, false) -} - -func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior, kind networkEndpointKind, configType networkEndpointConfigType, deprecated bool) (*referencedNetworkEndpoint, *tcpip.Error) { - // TODO(b/141022673): Validate IP address before adding them. - - // Sanity check. - id := NetworkEndpointID{protocolAddress.AddressWithPrefix.Address} - if _, ok := n.endpoints[id]; ok { - // Endpoint already exists. - return nil, tcpip.ErrDuplicateAddress - } - netProto, ok := n.stack.networkProtocols[protocolAddress.Protocol] if !ok { return nil, tcpip.ErrUnknownProtocol @@ -527,7 +656,7 @@ func (n *NIC) addAddressLocked(protocolAddress tcpip.ProtocolAddress, peb Primar func (n *NIC) AddAddress(protocolAddress tcpip.ProtocolAddress, peb PrimaryEndpointBehavior) *tcpip.Error { // Add the endpoint. n.mu.Lock() - _, err := n.addPermanentAddressLocked(protocolAddress, peb) + _, err := n.addAddressLocked(protocolAddress, peb, permanent, static, false /* deprecated */) n.mu.Unlock() return err @@ -660,7 +789,7 @@ func (n *NIC) RemoveAddressRange(subnet tcpip.Subnet) { n.mu.Unlock() } -// Subnets returns the Subnets associated with this NIC. +// AddressRanges returns the Subnets associated with this NIC. func (n *NIC) AddressRanges() []tcpip.Subnet { n.mu.RLock() defer n.mu.RUnlock() @@ -803,13 +932,13 @@ func (n *NIC) joinGroupLocked(protocol tcpip.NetworkProtocolNumber, addr tcpip.A if !ok { return tcpip.ErrUnknownProtocol } - if _, err := n.addPermanentAddressLocked(tcpip.ProtocolAddress{ + if _, err := n.addAddressLocked(tcpip.ProtocolAddress{ Protocol: protocol, AddressWithPrefix: tcpip.AddressWithPrefix{ Address: addr, PrefixLen: netProto.DefaultPrefixLen(), }, - }, NeverPrimaryEndpoint); err != nil { + }, NeverPrimaryEndpoint, permanent, static, false /* deprecated */); err != nil { return err } } @@ -855,7 +984,7 @@ func handlePacket(protocol tcpip.NetworkProtocolNumber, dst, src tcpip.Address, // DeliverNetworkPacket finds the appropriate network protocol endpoint and // hands the packet over for further processing. This function is called when -// the NIC receives a packet from the physical interface. +// the NIC receives a packet from the link endpoint. // Note that the ownership of the slice backing vv is retained by the caller. // This rule applies only to the slice itself, not to the items of the slice; // the ownership of the items is not retained by the caller. @@ -900,6 +1029,14 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link src, dst := netProto.ParseAddresses(pkt.Data.First()) + if n.stack.handleLocal && !n.isLoopback() && n.getRef(protocol, src) != nil { + // The source address is one of our own, so we never should have gotten a + // packet like this unless handleLocal is false. Loopback also calls this + // function even though the packets didn't come from the physical interface + // so don't drop those. + n.stack.stats.IP.InvalidSourceAddressesReceived.Increment() + return + } if ref := n.getRef(protocol, dst); ref != nil { handlePacket(protocol, dst, src, linkEP.LinkAddress(), remote, ref, pkt) return @@ -912,7 +1049,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link if n.stack.Forwarding() { r, err := n.stack.FindRoute(0, "", dst, protocol, false /* multicastLoop */) if err != nil { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() return } defer r.Release() @@ -950,7 +1087,7 @@ func (n *NIC) DeliverNetworkPacket(linkEP LinkEndpoint, remote, local tcpip.Link // If a packet socket handled the packet, don't treat it as invalid. if len(packetEPs) == 0 { - n.stack.stats.IP.InvalidAddressesReceived.Increment() + n.stack.stats.IP.InvalidDestinationAddressesReceived.Increment() } } @@ -1187,7 +1324,8 @@ type referencedNetworkEndpoint struct { kind networkEndpointKind // configType is the method that was used to configure this endpoint. - // This must never change after the endpoint is added to a NIC. + // This must never change except during endpoint creation and promotion to + // permanent. configType networkEndpointConfigType // deprecated indicates whether or not the endpoint should be considered diff --git a/pkg/tcpip/stack/stack.go b/pkg/tcpip/stack/stack.go index a47ceba54..7057b110e 100644 --- a/pkg/tcpip/stack/stack.go +++ b/pkg/tcpip/stack/stack.go @@ -423,7 +423,11 @@ type Stack struct { // handleLocal allows non-loopback interfaces to loop packets. handleLocal bool - // tables are the iptables packet filtering and manipulation rules. + // tablesMu protects iptables. + tablesMu sync.RWMutex + + // tables are the iptables packet filtering and manipulation rules. The are + // protected by tablesMu.` tables iptables.IPTables // resumableEndpoints is a list of endpoints that need to be resumed if the @@ -547,6 +551,49 @@ type TransportEndpointInfo struct { RegisterNICID tcpip.NICID } +// AddrNetProto unwraps the specified address if it is a V4-mapped V6 address +// and returns the network protocol number to be used to communicate with the +// specified address. It returns an error if the passed address is incompatible +// with the receiver. +func (e *TransportEndpointInfo) AddrNetProto(addr tcpip.FullAddress, v6only bool) (tcpip.FullAddress, tcpip.NetworkProtocolNumber, *tcpip.Error) { + netProto := e.NetProto + switch len(addr.Addr) { + case header.IPv4AddressSize: + netProto = header.IPv4ProtocolNumber + case header.IPv6AddressSize: + if header.IsV4MappedAddress(addr.Addr) { + netProto = header.IPv4ProtocolNumber + addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] + if addr.Addr == header.IPv4Any { + addr.Addr = "" + } + } + } + + switch len(e.ID.LocalAddress) { + case header.IPv4AddressSize: + if len(addr.Addr) == header.IPv6AddressSize { + return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState + } + case header.IPv6AddressSize: + if len(addr.Addr) == header.IPv4AddressSize { + return tcpip.FullAddress{}, 0, tcpip.ErrNetworkUnreachable + } + } + + switch { + case netProto == e.NetProto: + case netProto == header.IPv4ProtocolNumber && e.NetProto == header.IPv6ProtocolNumber: + if v6only { + return tcpip.FullAddress{}, 0, tcpip.ErrNoRoute + } + default: + return tcpip.FullAddress{}, 0, tcpip.ErrInvalidEndpointState + } + + return addr, netProto, nil +} + // IsEndpointInfo is an empty method to implement the tcpip.EndpointInfo // marker interface. func (*TransportEndpointInfo) IsEndpointInfo() {} @@ -707,7 +754,9 @@ func (s *Stack) Stats() tcpip.Stats { // SetForwarding enables or disables the packet forwarding between NICs. // // When forwarding becomes enabled, any host-only state on all NICs will be -// cleaned up. +// cleaned up and if IPv6 is enabled, NDP Router Solicitations will be started. +// When forwarding becomes disabled and if IPv6 is enabled, NDP Router +// Solicitations will be stopped. func (s *Stack) SetForwarding(enable bool) { // TODO(igudger, bgeffon): Expose via /proc/sys/net/ipv4/ip_forward. s.mu.Lock() @@ -729,6 +778,10 @@ func (s *Stack) SetForwarding(enable bool) { for _, nic := range s.nics { nic.becomeIPv6Router() } + } else { + for _, nic := range s.nics { + nic.becomeIPv6Host() + } } } @@ -869,7 +922,7 @@ func (s *Stack) CheckNIC(id tcpip.NICID) bool { return false } -// NICSubnets returns a map of NICIDs to their associated subnets. +// NICAddressRanges returns a map of NICIDs to their associated subnets. func (s *Stack) NICAddressRanges() map[tcpip.NICID][]tcpip.Subnet { s.mu.RLock() defer s.mu.RUnlock() @@ -1063,9 +1116,9 @@ func (s *Stack) GetMainNICAddress(id tcpip.NICID, protocol tcpip.NetworkProtocol return nic.primaryAddress(protocol), nil } -func (s *Stack) getRefEP(nic *NIC, localAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { +func (s *Stack) getRefEP(nic *NIC, localAddr, remoteAddr tcpip.Address, netProto tcpip.NetworkProtocolNumber) (ref *referencedNetworkEndpoint) { if len(localAddr) == 0 { - return nic.primaryEndpoint(netProto) + return nic.primaryEndpoint(netProto, remoteAddr) } return nic.findEndpoint(netProto, localAddr, CanBePrimaryEndpoint) } @@ -1081,7 +1134,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n needRoute := !(isBroadcast || isMulticast || header.IsV6LinkLocalAddress(remoteAddr)) if id != 0 && !needRoute { if nic, ok := s.nics[id]; ok { - if ref := s.getRefEP(nic, localAddr, netProto); ref != nil { + if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { return makeRoute(netProto, ref.ep.ID().LocalAddress, remoteAddr, nic.linkEP.LinkAddress(), ref, s.handleLocal && !nic.isLoopback(), multicastLoop && !nic.isLoopback()), nil } } @@ -1091,7 +1144,7 @@ func (s *Stack) FindRoute(id tcpip.NICID, localAddr, remoteAddr tcpip.Address, n continue } if nic, ok := s.nics[route.NIC]; ok { - if ref := s.getRefEP(nic, localAddr, netProto); ref != nil { + if ref := s.getRefEP(nic, localAddr, remoteAddr, netProto); ref != nil { if len(remoteAddr) == 0 { // If no remote address was provided, then the route // provided will refer to the link local address. @@ -1545,12 +1598,17 @@ func (s *Stack) LeaveGroup(protocol tcpip.NetworkProtocolNumber, nicID tcpip.NIC // IPTables returns the stack's iptables. func (s *Stack) IPTables() iptables.IPTables { - return s.tables + s.tablesMu.RLock() + t := s.tables + s.tablesMu.RUnlock() + return t } // SetIPTables sets the stack's iptables. func (s *Stack) SetIPTables(ipt iptables.IPTables) { + s.tablesMu.Lock() s.tables = ipt + s.tablesMu.Unlock() } // ICMPLimit returns the maximum number of ICMP messages that can be sent diff --git a/pkg/tcpip/stack/stack_test.go b/pkg/tcpip/stack/stack_test.go index cf41e02eb..dad288642 100644 --- a/pkg/tcpip/stack/stack_test.go +++ b/pkg/tcpip/stack/stack_test.go @@ -35,6 +35,7 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/link/loopback" "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" + "gvisor.dev/gvisor/pkg/tcpip/transport/udp" ) const ( @@ -2411,3 +2412,154 @@ func TestNewPEBOnPromotionToPermanent(t *testing.T) { } } } + +func TestIPv6SourceAddressSelectionScopeAndSameAddress(t *testing.T) { + const ( + linkLocalAddr1 = tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + linkLocalAddr2 = tcpip.Address("\xfe\x80\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + uniqueLocalAddr1 = tcpip.Address("\xfc\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + uniqueLocalAddr2 = tcpip.Address("\xfd\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + globalAddr1 = tcpip.Address("\xa0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01") + globalAddr2 = tcpip.Address("\xa0\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02") + nicID = 1 + ) + + // Rule 3 is not tested here, and is instead tested by NDP's AutoGenAddr test. + tests := []struct { + name string + nicAddrs []tcpip.Address + connectAddr tcpip.Address + expectedLocalAddr tcpip.Address + }{ + // Test Rule 1 of RFC 6724 section 5. + { + name: "Same Global most preferred (last address)", + nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1}, + connectAddr: globalAddr1, + expectedLocalAddr: globalAddr1, + }, + { + name: "Same Global most preferred (first address)", + nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1}, + connectAddr: globalAddr1, + expectedLocalAddr: globalAddr1, + }, + { + name: "Same Link Local most preferred (last address)", + nicAddrs: []tcpip.Address{globalAddr1, uniqueLocalAddr1, linkLocalAddr1}, + connectAddr: linkLocalAddr1, + expectedLocalAddr: linkLocalAddr1, + }, + { + name: "Same Link Local most preferred (first address)", + nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1}, + connectAddr: linkLocalAddr1, + expectedLocalAddr: linkLocalAddr1, + }, + { + name: "Same Unique Local most preferred (last address)", + nicAddrs: []tcpip.Address{uniqueLocalAddr1, globalAddr1, linkLocalAddr1}, + connectAddr: uniqueLocalAddr1, + expectedLocalAddr: uniqueLocalAddr1, + }, + { + name: "Same Unique Local most preferred (first address)", + nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1}, + connectAddr: uniqueLocalAddr1, + expectedLocalAddr: uniqueLocalAddr1, + }, + + // Test Rule 2 of RFC 6724 section 5. + { + name: "Global most preferred (last address)", + nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1}, + connectAddr: globalAddr2, + expectedLocalAddr: globalAddr1, + }, + { + name: "Global most preferred (first address)", + nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1}, + connectAddr: globalAddr2, + expectedLocalAddr: globalAddr1, + }, + { + name: "Link Local most preferred (last address)", + nicAddrs: []tcpip.Address{globalAddr1, uniqueLocalAddr1, linkLocalAddr1}, + connectAddr: linkLocalAddr2, + expectedLocalAddr: linkLocalAddr1, + }, + { + name: "Link Local most preferred (first address)", + nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, globalAddr1}, + connectAddr: linkLocalAddr2, + expectedLocalAddr: linkLocalAddr1, + }, + { + name: "Unique Local most preferred (last address)", + nicAddrs: []tcpip.Address{uniqueLocalAddr1, globalAddr1, linkLocalAddr1}, + connectAddr: uniqueLocalAddr2, + expectedLocalAddr: uniqueLocalAddr1, + }, + { + name: "Unique Local most preferred (first address)", + nicAddrs: []tcpip.Address{globalAddr1, linkLocalAddr1, uniqueLocalAddr1}, + connectAddr: uniqueLocalAddr2, + expectedLocalAddr: uniqueLocalAddr1, + }, + + // Test returning the endpoint that is closest to the front when + // candidate addresses are "equal" from the perspective of RFC 6724 + // section 5. + { + name: "Unique Local for Global", + nicAddrs: []tcpip.Address{linkLocalAddr1, uniqueLocalAddr1, uniqueLocalAddr2}, + connectAddr: globalAddr2, + expectedLocalAddr: uniqueLocalAddr1, + }, + { + name: "Link Local for Global", + nicAddrs: []tcpip.Address{linkLocalAddr1, linkLocalAddr2}, + connectAddr: globalAddr2, + expectedLocalAddr: linkLocalAddr1, + }, + { + name: "Link Local for Unique Local", + nicAddrs: []tcpip.Address{linkLocalAddr1, linkLocalAddr2}, + connectAddr: uniqueLocalAddr2, + expectedLocalAddr: linkLocalAddr1, + }, + } + + for _, test := range tests { + t.Run(test.name, func(t *testing.T) { + e := channel.New(0, 1280, linkAddr1) + s := stack.New(stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + }) + if err := s.CreateNIC(nicID, e); err != nil { + t.Fatalf("CreateNIC(%d, _) = %s", nicID, err) + } + s.SetRouteTable([]tcpip.Route{{ + Destination: header.IPv6EmptySubnet, + Gateway: llAddr3, + NIC: nicID, + }}) + s.AddLinkAddress(nicID, llAddr3, linkAddr3) + + for _, a := range test.nicAddrs { + if err := s.AddAddress(nicID, ipv6.ProtocolNumber, a); err != nil { + t.Errorf("s.AddAddress(%d, %d, %s): %s", nicID, ipv6.ProtocolNumber, a, err) + } + } + + if t.Failed() { + t.FailNow() + } + + if got := addrForNewConnectionTo(t, s, tcpip.FullAddress{Addr: test.connectAddr, NIC: nicID, Port: 1234}); got != test.expectedLocalAddr { + t.Errorf("got local address = %s, want = %s", got, test.expectedLocalAddr) + } + }) + } +} diff --git a/pkg/tcpip/stack/transport_demuxer.go b/pkg/tcpip/stack/transport_demuxer.go index f384a91de..d686e6eb8 100644 --- a/pkg/tcpip/stack/transport_demuxer.go +++ b/pkg/tcpip/stack/transport_demuxer.go @@ -104,7 +104,14 @@ func (epsByNic *endpointsByNic) handlePacket(r *Route, id TransportEndpointID, p return } // multiPortEndpoints are guaranteed to have at least one element. - selectEndpoint(id, mpep, epsByNic.seed).HandlePacket(r, id, pkt) + transEP := selectEndpoint(id, mpep, epsByNic.seed) + if queuedProtocol, mustQueue := mpep.demux.queuedProtocols[protocolIDs{mpep.netProto, mpep.transProto}]; mustQueue { + queuedProtocol.QueuePacket(r, transEP, id, pkt) + epsByNic.mu.RUnlock() + return + } + + transEP.HandlePacket(r, id, pkt) epsByNic.mu.RUnlock() // Don't use defer for performance reasons. } @@ -130,7 +137,7 @@ func (epsByNic *endpointsByNic) handleControlPacket(n *NIC, id TransportEndpoint // registerEndpoint returns true if it succeeds. It fails and returns // false if ep already has an element with the same key. -func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { +func (epsByNic *endpointsByNic) registerEndpoint(d *transportDemuxer, netProto tcpip.NetworkProtocolNumber, transProto tcpip.TransportProtocolNumber, t TransportEndpoint, reusePort bool, bindToDevice tcpip.NICID) *tcpip.Error { epsByNic.mu.Lock() defer epsByNic.mu.Unlock() @@ -140,7 +147,7 @@ func (epsByNic *endpointsByNic) registerEndpoint(t TransportEndpoint, reusePort } // This is a new binding. - multiPortEp := &multiPortEndpoint{} + multiPortEp := &multiPortEndpoint{demux: d, netProto: netProto, transProto: transProto} multiPortEp.endpointsMap = make(map[TransportEndpoint]int) multiPortEp.reuse = reusePort epsByNic.endpoints[bindToDevice] = multiPortEp @@ -168,18 +175,34 @@ func (epsByNic *endpointsByNic) unregisterEndpoint(bindToDevice tcpip.NICID, t T // newTransportDemuxer. type transportDemuxer struct { // protocol is immutable. - protocol map[protocolIDs]*transportEndpoints + protocol map[protocolIDs]*transportEndpoints + queuedProtocols map[protocolIDs]queuedTransportProtocol +} + +// queuedTransportProtocol if supported by a protocol implementation will cause +// the dispatcher to delivery packets to the QueuePacket method instead of +// calling HandlePacket directly on the endpoint. +type queuedTransportProtocol interface { + QueuePacket(r *Route, ep TransportEndpoint, id TransportEndpointID, pkt tcpip.PacketBuffer) } func newTransportDemuxer(stack *Stack) *transportDemuxer { - d := &transportDemuxer{protocol: make(map[protocolIDs]*transportEndpoints)} + d := &transportDemuxer{ + protocol: make(map[protocolIDs]*transportEndpoints), + queuedProtocols: make(map[protocolIDs]queuedTransportProtocol), + } // Add each network and transport pair to the demuxer. for netProto := range stack.networkProtocols { for proto := range stack.transportProtocols { - d.protocol[protocolIDs{netProto, proto}] = &transportEndpoints{ + protoIDs := protocolIDs{netProto, proto} + d.protocol[protoIDs] = &transportEndpoints{ endpoints: make(map[TransportEndpointID]*endpointsByNic), } + qTransProto, isQueued := (stack.transportProtocols[proto].proto).(queuedTransportProtocol) + if isQueued { + d.queuedProtocols[protoIDs] = qTransProto + } } } @@ -209,7 +232,11 @@ func (d *transportDemuxer) registerEndpoint(netProtos []tcpip.NetworkProtocolNum // // +stateify savable type multiPortEndpoint struct { - mu sync.RWMutex `state:"nosave"` + mu sync.RWMutex `state:"nosave"` + demux *transportDemuxer + netProto tcpip.NetworkProtocolNumber + transProto tcpip.TransportProtocolNumber + endpointsArr []TransportEndpoint endpointsMap map[TransportEndpoint]int // reuse indicates if more than one endpoint is allowed. @@ -258,13 +285,22 @@ func selectEndpoint(id TransportEndpointID, mpep *multiPortEndpoint, seed uint32 func (ep *multiPortEndpoint) handlePacketAll(r *Route, id TransportEndpointID, pkt tcpip.PacketBuffer) { ep.mu.RLock() + queuedProtocol, mustQueue := ep.demux.queuedProtocols[protocolIDs{ep.netProto, ep.transProto}] for i, endpoint := range ep.endpointsArr { // HandlePacket takes ownership of pkt, so each endpoint needs // its own copy except for the final one. if i == len(ep.endpointsArr)-1 { + if mustQueue { + queuedProtocol.QueuePacket(r, endpoint, id, pkt) + break + } endpoint.HandlePacket(r, id, pkt) break } + if mustQueue { + queuedProtocol.QueuePacket(r, endpoint, id, pkt.Clone()) + continue + } endpoint.HandlePacket(r, id, pkt.Clone()) } ep.mu.RUnlock() // Don't use defer for performance reasons. @@ -357,7 +393,7 @@ func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocol if epsByNic, ok := eps.endpoints[id]; ok { // There was already a binding. - return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) + return epsByNic.registerEndpoint(d, netProto, protocol, ep, reusePort, bindToDevice) } // This is a new binding. @@ -367,7 +403,7 @@ func (d *transportDemuxer) singleRegisterEndpoint(netProto tcpip.NetworkProtocol } eps.endpoints[id] = epsByNic - return epsByNic.registerEndpoint(ep, reusePort, bindToDevice) + return epsByNic.registerEndpoint(d, netProto, protocol, ep, reusePort, bindToDevice) } // unregisterEndpoint unregisters the endpoint with the given id such that it diff --git a/pkg/tcpip/tcpip.go b/pkg/tcpip/tcpip.go index a0a3d1e53..3fc823a36 100644 --- a/pkg/tcpip/tcpip.go +++ b/pkg/tcpip/tcpip.go @@ -321,7 +321,7 @@ type ControlMessages struct { HasTOS bool // TOS is the IPv4 type of service of the associated packet. - TOS int8 + TOS uint8 // HasTClass indicates whether Tclass is valid/set. HasTClass bool @@ -496,9 +496,13 @@ type WriteOptions struct { type SockOptBool int const ( + // ReceiveTOSOption is used by SetSockOpt/GetSockOpt to specify if the TOS + // ancillary message is passed with incoming packets. + ReceiveTOSOption SockOptBool = iota + // V6OnlyOption is used by {G,S}etSockOptBool to specify whether an IPv6 // socket is to be restricted to sending and receiving IPv6 packets only. - V6OnlyOption SockOptBool = iota + V6OnlyOption ) // SockOptInt represents socket options which values have the int type. @@ -895,9 +899,13 @@ type IPStats struct { // link layer in nic.DeliverNetworkPacket. PacketsReceived *StatCounter - // InvalidAddressesReceived is the total number of IP packets received - // with an unknown or invalid destination address. - InvalidAddressesReceived *StatCounter + // InvalidDestinationAddressesReceived is the total number of IP packets + // received with an unknown or invalid destination address. + InvalidDestinationAddressesReceived *StatCounter + + // InvalidSourceAddressesReceived is the total number of IP packets received + // with a source address that should never have been received on the wire. + InvalidSourceAddressesReceived *StatCounter // PacketsDelivered is the total number of incoming IP packets that // are successfully delivered to the transport layer via HandlePacket. diff --git a/pkg/tcpip/transport/icmp/endpoint.go b/pkg/tcpip/transport/icmp/endpoint.go index 330786f4c..42afb3f5b 100644 --- a/pkg/tcpip/transport/icmp/endpoint.go +++ b/pkg/tcpip/transport/icmp/endpoint.go @@ -288,7 +288,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c toCopy := *to to = &toCopy - netProto, err := e.checkV4Mapped(to, true) + netProto, err := e.checkV4Mapped(to) if err != nil { return 0, nil, err } @@ -475,18 +475,12 @@ func send6(r *stack.Route, ident uint16, data buffer.View, ttl uint8) *tcpip.Err }) } -func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if header.IsV4MappedAddress(addr.Addr) { - return 0, tcpip.ErrNoRoute - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); !allowMismatch && l != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState +func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, false /* v6only */) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } @@ -518,7 +512,7 @@ func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } @@ -631,7 +625,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } diff --git a/pkg/tcpip/transport/tcp/BUILD b/pkg/tcpip/transport/tcp/BUILD index 353bd06f4..0e3ab05ad 100644 --- a/pkg/tcpip/transport/tcp/BUILD +++ b/pkg/tcpip/transport/tcp/BUILD @@ -16,6 +16,18 @@ go_template_instance( }, ) +go_template_instance( + name = "tcp_endpoint_list", + out = "tcp_endpoint_list.go", + package = "tcp", + prefix = "endpoint", + template = "//pkg/ilist:generic_list", + types = { + "Element": "*endpoint", + "Linker": "*endpoint", + }, +) + go_library( name = "tcp", srcs = [ @@ -23,6 +35,7 @@ go_library( "connect.go", "cubic.go", "cubic_state.go", + "dispatcher.go", "endpoint.go", "endpoint_state.go", "forwarder.go", @@ -38,6 +51,7 @@ go_library( "segment_state.go", "snd.go", "snd_state.go", + "tcp_endpoint_list.go", "tcp_segment_list.go", "timer.go", ], @@ -45,7 +59,6 @@ go_library( imports = ["gvisor.dev/gvisor/pkg/tcpip/buffer"], visibility = ["//visibility:public"], deps = [ - "//pkg/log", "//pkg/rand", "//pkg/sleep", "//pkg/sync", diff --git a/pkg/tcpip/transport/tcp/accept.go b/pkg/tcpip/transport/tcp/accept.go index 1ea996936..1a2e3efa9 100644 --- a/pkg/tcpip/transport/tcp/accept.go +++ b/pkg/tcpip/transport/tcp/accept.go @@ -285,7 +285,7 @@ func (l *listenContext) createEndpointAndPerformHandshake(s *segment, opts *head // listenEP is nil when listenContext is used by tcp.Forwarder. if l.listenEP != nil { l.listenEP.mu.Lock() - if l.listenEP.state != StateListen { + if l.listenEP.EndpointState() != StateListen { l.listenEP.mu.Unlock() return nil, tcpip.ErrConnectionAborted } @@ -344,11 +344,12 @@ func (l *listenContext) closeAllPendingEndpoints() { // instead. func (e *endpoint) deliverAccepted(n *endpoint) { e.mu.Lock() - state := e.state + state := e.EndpointState() e.pendingAccepted.Add(1) defer e.pendingAccepted.Done() acceptedChan := e.acceptedChan e.mu.Unlock() + if state == StateListen { acceptedChan <- n e.waiterQueue.Notify(waiter.EventIn) @@ -562,8 +563,8 @@ func (e *endpoint) handleListenSegment(ctx *listenContext, s *segment) { // We do not use transitionToStateEstablishedLocked here as there is // no handshake state available when doing a SYN cookie based accept. n.stack.Stats().TCP.CurrentEstablished.Increment() - n.state = StateEstablished n.isConnectNotified = true + n.setEndpointState(StateEstablished) // Do the delivery in a separate goroutine so // that we don't block the listen loop in case @@ -596,7 +597,7 @@ func (e *endpoint) protocolListenLoop(rcvWnd seqnum.Size) *tcpip.Error { // handleSynSegment() from attempting to queue new connections // to the endpoint. e.mu.Lock() - e.state = StateClose + e.setEndpointState(StateClose) // close any endpoints in SYN-RCVD state. ctx.closeAllPendingEndpoints() diff --git a/pkg/tcpip/transport/tcp/connect.go b/pkg/tcpip/transport/tcp/connect.go index 613ec1775..a2f384384 100644 --- a/pkg/tcpip/transport/tcp/connect.go +++ b/pkg/tcpip/transport/tcp/connect.go @@ -190,7 +190,7 @@ func (h *handshake) resetToSynRcvd(iss seqnum.Value, irs seqnum.Value, opts *hea h.mss = opts.MSS h.sndWndScale = opts.WS h.ep.mu.Lock() - h.ep.state = StateSynRecv + h.ep.setEndpointState(StateSynRecv) h.ep.mu.Unlock() } @@ -274,14 +274,14 @@ func (h *handshake) synSentState(s *segment) *tcpip.Error { // SYN-RCVD state. h.state = handshakeSynRcvd h.ep.mu.Lock() - h.ep.state = StateSynRecv ttl := h.ep.ttl + h.ep.setEndpointState(StateSynRecv) h.ep.mu.Unlock() synOpts := header.TCPSynOptions{ WS: int(h.effectiveRcvWndScale()), TS: rcvSynOpts.TS, TSVal: h.ep.timestamp(), - TSEcr: h.ep.recentTS, + TSEcr: h.ep.recentTimestamp(), // We only send SACKPermitted if the other side indicated it // permits SACK. This is not explicitly defined in the RFC but @@ -341,7 +341,7 @@ func (h *handshake) synRcvdState(s *segment) *tcpip.Error { WS: h.rcvWndScale, TS: h.ep.sendTSOk, TSVal: h.ep.timestamp(), - TSEcr: h.ep.recentTS, + TSEcr: h.ep.recentTimestamp(), SACKPermitted: h.ep.sackPermitted, MSS: h.ep.amss, } @@ -501,7 +501,7 @@ func (h *handshake) execute() *tcpip.Error { WS: h.rcvWndScale, TS: true, TSVal: h.ep.timestamp(), - TSEcr: h.ep.recentTS, + TSEcr: h.ep.recentTimestamp(), SACKPermitted: bool(sackEnabled), MSS: h.ep.amss, } @@ -792,7 +792,7 @@ func (e *endpoint) makeOptions(sackBlocks []header.SACKBlock) []byte { // Ref: https://tools.ietf.org/html/rfc7323#section-5.4. offset += header.EncodeNOP(options[offset:]) offset += header.EncodeNOP(options[offset:]) - offset += header.EncodeTSOption(e.timestamp(), uint32(e.recentTS), options[offset:]) + offset += header.EncodeTSOption(e.timestamp(), e.recentTimestamp(), options[offset:]) } if e.sackPermitted && len(sackBlocks) > 0 { offset += header.EncodeNOP(options[offset:]) @@ -811,7 +811,7 @@ func (e *endpoint) makeOptions(sackBlocks []header.SACKBlock) []byte { // sendRaw sends a TCP segment to the endpoint's peer. func (e *endpoint) sendRaw(data buffer.VectorisedView, flags byte, seq, ack seqnum.Value, rcvWnd seqnum.Size) *tcpip.Error { var sackBlocks []header.SACKBlock - if e.state == StateEstablished && e.rcv.pendingBufSize > 0 && (flags&header.TCPFlagAck != 0) { + if e.EndpointState() == StateEstablished && e.rcv.pendingBufSize > 0 && (flags&header.TCPFlagAck != 0) { sackBlocks = e.sack.Blocks[:e.sack.NumBlocks] } options := e.makeOptions(sackBlocks) @@ -848,6 +848,9 @@ func (e *endpoint) handleWrite() *tcpip.Error { } func (e *endpoint) handleClose() *tcpip.Error { + if !e.EndpointState().connected() { + return nil + } // Drain the send queue. e.handleWrite() @@ -864,11 +867,7 @@ func (e *endpoint) handleClose() *tcpip.Error { func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { // Only send a reset if the connection is being aborted for a reason // other than receiving a reset. - if e.state == StateEstablished || e.state == StateCloseWait { - e.stack.Stats().TCP.EstablishedResets.Increment() - e.stack.Stats().TCP.CurrentEstablished.Decrement() - } - e.state = StateError + e.setEndpointState(StateError) e.HardError = err if err != tcpip.ErrConnectionReset && err != tcpip.ErrTimeout { // The exact sequence number to be used for the RST is the same as the @@ -888,9 +887,12 @@ func (e *endpoint) resetConnectionLocked(err *tcpip.Error) { } // completeWorkerLocked is called by the worker goroutine when it's about to -// exit. It marks the worker as completed and performs cleanup work if requested -// by Close(). +// exit. func (e *endpoint) completeWorkerLocked() { + // Worker is terminating(either due to moving to + // CLOSED or ERROR state, ensure we release all + // registrations port reservations even if the socket + // itself is not yet closed by the application. e.workerRunning = false if e.workerCleanup { e.cleanupLocked() @@ -917,8 +919,7 @@ func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { e.rcvAutoParams.prevCopied = int(h.rcvWnd) e.rcvListMu.Unlock() } - h.ep.stack.Stats().TCP.CurrentEstablished.Increment() - e.state = StateEstablished + e.setEndpointState(StateEstablished) } // transitionToStateCloseLocked ensures that the endpoint is @@ -927,11 +928,12 @@ func (e *endpoint) transitionToStateEstablishedLocked(h *handshake) { // delivered to this endpoint from the demuxer when the endpoint // is transitioned to StateClose. func (e *endpoint) transitionToStateCloseLocked() { - if e.state == StateClose { + if e.EndpointState() == StateClose { return } + // Mark the endpoint as fully closed for reads/writes. e.cleanupLocked() - e.state = StateClose + e.setEndpointState(StateClose) e.stack.Stats().TCP.EstablishedClosed.Increment() } @@ -946,7 +948,9 @@ func (e *endpoint) tryDeliverSegmentFromClosedEndpoint(s *segment) { s.decRef() return } - ep.(*endpoint).enqueueSegment(s) + if ep.(*endpoint).enqueueSegment(s) { + ep.(*endpoint).newSegmentWaker.Assert() + } } func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { @@ -955,9 +959,8 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { // except SYN-SENT, all reset (RST) segments are // validated by checking their SEQ-fields." So // we only process it if it's acceptable. - s.decRef() e.mu.Lock() - switch e.state { + switch e.EndpointState() { // In case of a RST in CLOSE-WAIT linux moves // the socket to closed state with an error set // to indicate EPIPE. @@ -981,103 +984,53 @@ func (e *endpoint) handleReset(s *segment) (ok bool, err *tcpip.Error) { e.transitionToStateCloseLocked() e.HardError = tcpip.ErrAborted e.mu.Unlock() + e.notifyProtocolGoroutine(notifyTickleWorker) return false, nil default: e.mu.Unlock() + // RFC 793, page 37 states that "in all states + // except SYN-SENT, all reset (RST) segments are + // validated by checking their SEQ-fields." So + // we only process it if it's acceptable. + + // Notify protocol goroutine. This is required when + // handleSegment is invoked from the processor goroutine + // rather than the worker goroutine. + e.notifyProtocolGoroutine(notifyResetByPeer) return false, tcpip.ErrConnectionReset } } return true, nil } -// handleSegments pulls segments from the queue and processes them. It returns -// no error if the protocol loop should continue, an error otherwise. -func (e *endpoint) handleSegments() *tcpip.Error { +// handleSegments processes all inbound segments. +func (e *endpoint) handleSegments(fastPath bool) *tcpip.Error { checkRequeue := true for i := 0; i < maxSegmentsPerWake; i++ { + if e.EndpointState() == StateClose || e.EndpointState() == StateError { + return nil + } s := e.segmentQueue.dequeue() if s == nil { checkRequeue = false break } - // Invoke the tcp probe if installed. - if e.probe != nil { - e.probe(e.completeState()) + cont, err := e.handleSegment(s) + if err != nil { + s.decRef() + return err } - - if s.flagIsSet(header.TCPFlagRst) { - if ok, err := e.handleReset(s); !ok { - return err - } - } else if s.flagIsSet(header.TCPFlagSyn) { - // See: https://tools.ietf.org/html/rfc5961#section-4.1 - // 1) If the SYN bit is set, irrespective of the sequence number, TCP - // MUST send an ACK (also referred to as challenge ACK) to the remote - // peer: - // - // <SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK> - // - // After sending the acknowledgment, TCP MUST drop the unacceptable - // segment and stop processing further. - // - // By sending an ACK, the remote peer is challenged to confirm the loss - // of the previous connection and the request to start a new connection. - // A legitimate peer, after restart, would not have a TCB in the - // synchronized state. Thus, when the ACK arrives, the peer should send - // a RST segment back with the sequence number derived from the ACK - // field that caused the RST. - - // This RST will confirm that the remote peer has indeed closed the - // previous connection. Upon receipt of a valid RST, the local TCP - // endpoint MUST terminate its connection. The local TCP endpoint - // should then rely on SYN retransmission from the remote end to - // re-establish the connection. - - e.snd.sendAck() - } else if s.flagIsSet(header.TCPFlagAck) { - // Patch the window size in the segment according to the - // send window scale. - s.window <<= e.snd.sndWndScale - - // RFC 793, page 41 states that "once in the ESTABLISHED - // state all segments must carry current acknowledgment - // information." - drop, err := e.rcv.handleRcvdSegment(s) - if err != nil { - s.decRef() - return err - } - if drop { - s.decRef() - continue - } - - // Now check if the received segment has caused us to transition - // to a CLOSED state, if yes then terminate processing and do - // not invoke the sender. - e.mu.RLock() - state := e.state - e.mu.RUnlock() - if state == StateClose { - // When we get into StateClose while processing from the queue, - // return immediately and let the protocolMainloop handle it. - // - // We can reach StateClose only while processing a previous segment - // or a notification from the protocolMainLoop (caller goroutine). - // This means that with this return, the segment dequeue below can - // never occur on a closed endpoint. - s.decRef() - return nil - } - e.snd.handleRcvdSegment(s) + if !cont { + s.decRef() + return nil } - s.decRef() } - // If the queue is not empty, make sure we'll wake up in the next - // iteration. - if checkRequeue && !e.segmentQueue.empty() { + // When fastPath is true we don't want to wake up the worker + // goroutine. If the endpoint has more segments to process the + // dispatcher will call handleSegments again anyway. + if !fastPath && checkRequeue && !e.segmentQueue.empty() { e.newSegmentWaker.Assert() } @@ -1086,11 +1039,88 @@ func (e *endpoint) handleSegments() *tcpip.Error { e.snd.sendAck() } - e.resetKeepaliveTimer(true) + e.resetKeepaliveTimer(true /* receivedData */) return nil } +// handleSegment handles a given segment and notifies the worker goroutine if +// if the connection should be terminated. +func (e *endpoint) handleSegment(s *segment) (cont bool, err *tcpip.Error) { + // Invoke the tcp probe if installed. + if e.probe != nil { + e.probe(e.completeState()) + } + + if s.flagIsSet(header.TCPFlagRst) { + if ok, err := e.handleReset(s); !ok { + return false, err + } + } else if s.flagIsSet(header.TCPFlagSyn) { + // See: https://tools.ietf.org/html/rfc5961#section-4.1 + // 1) If the SYN bit is set, irrespective of the sequence number, TCP + // MUST send an ACK (also referred to as challenge ACK) to the remote + // peer: + // + // <SEQ=SND.NXT><ACK=RCV.NXT><CTL=ACK> + // + // After sending the acknowledgment, TCP MUST drop the unacceptable + // segment and stop processing further. + // + // By sending an ACK, the remote peer is challenged to confirm the loss + // of the previous connection and the request to start a new connection. + // A legitimate peer, after restart, would not have a TCB in the + // synchronized state. Thus, when the ACK arrives, the peer should send + // a RST segment back with the sequence number derived from the ACK + // field that caused the RST. + + // This RST will confirm that the remote peer has indeed closed the + // previous connection. Upon receipt of a valid RST, the local TCP + // endpoint MUST terminate its connection. The local TCP endpoint + // should then rely on SYN retransmission from the remote end to + // re-establish the connection. + + e.snd.sendAck() + } else if s.flagIsSet(header.TCPFlagAck) { + // Patch the window size in the segment according to the + // send window scale. + s.window <<= e.snd.sndWndScale + + // RFC 793, page 41 states that "once in the ESTABLISHED + // state all segments must carry current acknowledgment + // information." + drop, err := e.rcv.handleRcvdSegment(s) + if err != nil { + return false, err + } + if drop { + return true, nil + } + + // Now check if the received segment has caused us to transition + // to a CLOSED state, if yes then terminate processing and do + // not invoke the sender. + e.mu.RLock() + state := e.state + e.mu.RUnlock() + if state == StateClose { + // When we get into StateClose while processing from the queue, + // return immediately and let the protocolMainloop handle it. + // + // We can reach StateClose only while processing a previous segment + // or a notification from the protocolMainLoop (caller goroutine). + // This means that with this return, the segment dequeue below can + // never occur on a closed endpoint. + s.decRef() + return false, nil + } + + e.snd.handleRcvdSegment(s) + } + + return true, nil +} + // keepaliveTimerExpired is called when the keepaliveTimer fires. We send TCP // keepalive packets periodically when the connection is idle. If we don't hear // from the other side after a number of tries, we terminate the connection. @@ -1160,7 +1190,7 @@ func (e *endpoint) disableKeepaliveTimer() { // protocolMainLoop is the main loop of the TCP protocol. It runs in its own // goroutine and is responsible for sending segments and handling received // segments. -func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { +func (e *endpoint) protocolMainLoop(handshake bool, wakerInitDone chan<- struct{}) *tcpip.Error { var closeTimer *time.Timer var closeWaker sleep.Waker @@ -1182,6 +1212,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } e.mu.Unlock() + e.workMu.Unlock() // When the protocol loop exits we should wake up our waiters. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) } @@ -1193,7 +1224,7 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { initialRcvWnd := e.initialReceiveWindow() h := newHandshake(e, seqnum.Size(initialRcvWnd)) e.mu.Lock() - h.ep.state = StateSynSent + h.ep.setEndpointState(StateSynSent) e.mu.Unlock() if err := h.execute(); err != nil { @@ -1202,12 +1233,11 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.lastErrorMu.Unlock() e.mu.Lock() - e.state = StateError + e.setEndpointState(StateError) e.HardError = err // Lock released below. epilogue() - return err } } @@ -1215,7 +1245,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { e.keepalive.timer.init(&e.keepalive.waker) defer e.keepalive.timer.cleanup() - // Tell waiters that the endpoint is connected and writable. e.mu.Lock() drained := e.drainDone != nil e.mu.Unlock() @@ -1224,8 +1253,6 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { <-e.undrain } - e.waiterQueue.Notify(waiter.EventOut) - // Set up the functions that will be called when the main protocol loop // wakes up. funcs := []struct { @@ -1241,17 +1268,14 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { f: e.handleClose, }, { - w: &e.newSegmentWaker, - f: e.handleSegments, - }, - { w: &closeWaker, f: func() *tcpip.Error { // This means the socket is being closed due - // to the TCP_FIN_WAIT2 timeout was hit. Just + // to the TCP-FIN-WAIT2 timeout was hit. Just // mark the socket as closed. e.mu.Lock() e.transitionToStateCloseLocked() + e.workerCleanup = true e.mu.Unlock() return nil }, @@ -1267,6 +1291,12 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { }, }, { + w: &e.newSegmentWaker, + f: func() *tcpip.Error { + return e.handleSegments(false /* fastPath */) + }, + }, + { w: &e.keepalive.waker, f: e.keepaliveTimerExpired, }, @@ -1293,14 +1323,16 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { } if n¬ifyReset != 0 { - e.mu.Lock() - e.resetConnectionLocked(tcpip.ErrConnectionAborted) - e.mu.Unlock() + return tcpip.ErrConnectionAborted + } + + if n¬ifyResetByPeer != 0 { + return tcpip.ErrConnectionReset } if n¬ifyClose != 0 && closeTimer == nil { e.mu.Lock() - if e.state == StateFinWait2 && e.closed { + if e.EndpointState() == StateFinWait2 && e.closed { // The socket has been closed and we are in FIN_WAIT2 // so start the FIN_WAIT2 timer. closeTimer = time.AfterFunc(e.tcpLingerTimeout, func() { @@ -1320,11 +1352,11 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { if n¬ifyDrain != 0 { for !e.segmentQueue.empty() { - if err := e.handleSegments(); err != nil { + if err := e.handleSegments(false /* fastPath */); err != nil { return err } } - if e.state != StateClose && e.state != StateError { + if e.EndpointState() != StateClose && e.EndpointState() != StateError { // Only block the worker if the endpoint // is not in closed state or error state. close(e.drainDone) @@ -1349,14 +1381,21 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { s.AddWaker(funcs[i].w, i) } + // Notify the caller that the waker initialization is complete and the + // endpoint is ready. + if wakerInitDone != nil { + close(wakerInitDone) + } + + // Tell waiters that the endpoint is connected and writable. + e.waiterQueue.Notify(waiter.EventOut) + // The following assertions and notifications are needed for restored // endpoints. Fresh newly created endpoints have empty states and should // not invoke any. - e.segmentQueue.mu.Lock() - if !e.segmentQueue.list.Empty() { + if !e.segmentQueue.empty() { e.newSegmentWaker.Assert() } - e.segmentQueue.mu.Unlock() e.rcvListMu.Lock() if !e.rcvList.Empty() { @@ -1371,28 +1410,53 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { // Main loop. Handle segments until both send and receive ends of the // connection have completed. + cleanupOnError := func(err *tcpip.Error) { + e.mu.Lock() + e.workerCleanup = true + if err != nil { + e.resetConnectionLocked(err) + } + // Lock released below. + epilogue() + } - for e.state != StateTimeWait && e.state != StateClose && e.state != StateError { +loop: + for e.EndpointState() != StateTimeWait && e.EndpointState() != StateClose && e.EndpointState() != StateError { e.mu.Unlock() e.workMu.Unlock() v, _ := s.Fetch(true) e.workMu.Lock() - if err := funcs[v].f(); err != nil { - e.mu.Lock() - // Ensure we release all endpoint registration and route - // references as the connection is now in an error - // state. - e.workerCleanup = true - e.resetConnectionLocked(err) - // Lock released below. - epilogue() + // We need to double check here because the notification maybe + // stale by the time we got around to processing it. + // + // NOTE: since we now hold the workMu the processors cannot + // change the state of the endpoint so it's safe to proceed + // after this check. + switch e.EndpointState() { + case StateError: + // If the endpoint has already transitioned to an ERROR + // state just pass nil here as any reset that may need + // to be sent etc should already have been done and we + // just want to terminate the loop and cleanup the + // endpoint. + cleanupOnError(nil) return nil + case StateTimeWait: + fallthrough + case StateClose: + e.mu.Lock() + break loop + default: + if err := funcs[v].f(); err != nil { + cleanupOnError(err) + return nil + } + e.mu.Lock() } - e.mu.Lock() } - state := e.state + state := e.EndpointState() e.mu.Unlock() var reuseTW func() if state == StateTimeWait { @@ -1405,13 +1469,15 @@ func (e *endpoint) protocolMainLoop(handshake bool) *tcpip.Error { s.Done() // Wake up any waiters before we enter TIME_WAIT. e.waiterQueue.Notify(waiter.EventHUp | waiter.EventErr | waiter.EventIn | waiter.EventOut) + e.mu.Lock() + e.workerCleanup = true + e.mu.Unlock() reuseTW = e.doTimeWait() } // Mark endpoint as closed. e.mu.Lock() - if e.state != StateError { - e.stack.Stats().TCP.CurrentEstablished.Decrement() + if e.EndpointState() != StateError { e.transitionToStateCloseLocked() } @@ -1468,7 +1534,11 @@ func (e *endpoint) handleTimeWaitSegments() (extendTimeWait bool, reuseTW func() tcpEP := listenEP.(*endpoint) if EndpointState(tcpEP.State()) == StateListen { reuseTW = func() { - tcpEP.enqueueSegment(s) + if !tcpEP.enqueueSegment(s) { + s.decRef() + return + } + tcpEP.newSegmentWaker.Assert() } // We explicitly do not decRef // the segment as it's still diff --git a/pkg/tcpip/transport/tcp/dispatcher.go b/pkg/tcpip/transport/tcp/dispatcher.go new file mode 100644 index 000000000..e18012ac0 --- /dev/null +++ b/pkg/tcpip/transport/tcp/dispatcher.go @@ -0,0 +1,224 @@ +// Copyright 2018 The gVisor Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package tcp + +import ( + "gvisor.dev/gvisor/pkg/rand" + "gvisor.dev/gvisor/pkg/sleep" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/hash/jenkins" + "gvisor.dev/gvisor/pkg/tcpip/header" + "gvisor.dev/gvisor/pkg/tcpip/stack" +) + +// epQueue is a queue of endpoints. +type epQueue struct { + mu sync.Mutex + list endpointList +} + +// enqueue adds e to the queue if the endpoint is not already on the queue. +func (q *epQueue) enqueue(e *endpoint) { + q.mu.Lock() + if e.pendingProcessing { + q.mu.Unlock() + return + } + q.list.PushBack(e) + e.pendingProcessing = true + q.mu.Unlock() +} + +// dequeue removes and returns the first element from the queue if available, +// returns nil otherwise. +func (q *epQueue) dequeue() *endpoint { + q.mu.Lock() + if e := q.list.Front(); e != nil { + q.list.Remove(e) + e.pendingProcessing = false + q.mu.Unlock() + return e + } + q.mu.Unlock() + return nil +} + +// empty returns true if the queue is empty, false otherwise. +func (q *epQueue) empty() bool { + q.mu.Lock() + v := q.list.Empty() + q.mu.Unlock() + return v +} + +// processor is responsible for processing packets queued to a tcp endpoint. +type processor struct { + epQ epQueue + newEndpointWaker sleep.Waker + id int +} + +func newProcessor(id int) *processor { + p := &processor{ + id: id, + } + go p.handleSegments() + return p +} + +func (p *processor) queueEndpoint(ep *endpoint) { + // Queue an endpoint for processing by the processor goroutine. + p.epQ.enqueue(ep) + p.newEndpointWaker.Assert() +} + +func (p *processor) handleSegments() { + const newEndpointWaker = 1 + s := sleep.Sleeper{} + s.AddWaker(&p.newEndpointWaker, newEndpointWaker) + defer s.Done() + for { + s.Fetch(true) + for ep := p.epQ.dequeue(); ep != nil; ep = p.epQ.dequeue() { + if ep.segmentQueue.empty() { + continue + } + + // If socket has transitioned out of connected state + // then just let the worker handle the packet. + // + // NOTE: We read this outside of e.mu lock which means + // that by the time we get to handleSegments the + // endpoint may not be in ESTABLISHED. But this should + // be fine as all normal shutdown states are handled by + // handleSegments and if the endpoint moves to a + // CLOSED/ERROR state then handleSegments is a noop. + if ep.EndpointState() != StateEstablished { + ep.newSegmentWaker.Assert() + continue + } + + if !ep.workMu.TryLock() { + ep.newSegmentWaker.Assert() + continue + } + // If the endpoint is in a connected state then we do + // direct delivery to ensure low latency and avoid + // scheduler interactions. + if err := ep.handleSegments(true /* fastPath */); err != nil || ep.EndpointState() == StateClose { + // Send any active resets if required. + if err != nil { + ep.mu.Lock() + ep.resetConnectionLocked(err) + ep.mu.Unlock() + } + ep.notifyProtocolGoroutine(notifyTickleWorker) + ep.workMu.Unlock() + continue + } + + if !ep.segmentQueue.empty() { + p.epQ.enqueue(ep) + } + + ep.workMu.Unlock() + } + } +} + +// dispatcher manages a pool of TCP endpoint processors which are responsible +// for the processing of inbound segments. This fixed pool of processor +// goroutines do full tcp processing. The processor is selected based on the +// hash of the endpoint id to ensure that delivery for the same endpoint happens +// in-order. +type dispatcher struct { + processors []*processor + seed uint32 +} + +func newDispatcher(nProcessors int) *dispatcher { + processors := []*processor{} + for i := 0; i < nProcessors; i++ { + processors = append(processors, newProcessor(i)) + } + return &dispatcher{ + processors: processors, + seed: generateRandUint32(), + } +} + +func (d *dispatcher) queuePacket(r *stack.Route, stackEP stack.TransportEndpoint, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { + ep := stackEP.(*endpoint) + s := newSegment(r, id, pkt) + if !s.parse() { + ep.stack.Stats().MalformedRcvdPackets.Increment() + ep.stack.Stats().TCP.InvalidSegmentsReceived.Increment() + ep.stats.ReceiveErrors.MalformedPacketsReceived.Increment() + s.decRef() + return + } + + if !s.csumValid { + ep.stack.Stats().MalformedRcvdPackets.Increment() + ep.stack.Stats().TCP.ChecksumErrors.Increment() + ep.stats.ReceiveErrors.ChecksumErrors.Increment() + s.decRef() + return + } + + ep.stack.Stats().TCP.ValidSegmentsReceived.Increment() + ep.stats.SegmentsReceived.Increment() + if (s.flags & header.TCPFlagRst) != 0 { + ep.stack.Stats().TCP.ResetsReceived.Increment() + } + + if !ep.enqueueSegment(s) { + s.decRef() + return + } + + // For sockets not in established state let the worker goroutine + // handle the packets. + if ep.EndpointState() != StateEstablished { + ep.newSegmentWaker.Assert() + return + } + + d.selectProcessor(id).queueEndpoint(ep) +} + +func generateRandUint32() uint32 { + b := make([]byte, 4) + if _, err := rand.Read(b); err != nil { + panic(err) + } + return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 +} + +func (d *dispatcher) selectProcessor(id stack.TransportEndpointID) *processor { + payload := []byte{ + byte(id.LocalPort), + byte(id.LocalPort >> 8), + byte(id.RemotePort), + byte(id.RemotePort >> 8)} + + h := jenkins.Sum32(d.seed) + h.Write(payload) + h.Write([]byte(id.LocalAddress)) + h.Write([]byte(id.RemoteAddress)) + + return d.processors[h.Sum32()%uint32(len(d.processors))] +} diff --git a/pkg/tcpip/transport/tcp/endpoint.go b/pkg/tcpip/transport/tcp/endpoint.go index cca511fb9..4797f11d1 100644 --- a/pkg/tcpip/transport/tcp/endpoint.go +++ b/pkg/tcpip/transport/tcp/endpoint.go @@ -120,6 +120,7 @@ const ( notifyMTUChanged notifyDrain notifyReset + notifyResetByPeer notifyKeepaliveChanged notifyMSSChanged // notifyTickleWorker is used to tickle the protocol main loop during a @@ -127,6 +128,7 @@ const ( // ensures the loop terminates if the final state of the endpoint is // say TIME_WAIT. notifyTickleWorker + notifyError ) // SACKInfo holds TCP SACK related information for a given endpoint. @@ -283,6 +285,18 @@ func (*EndpointInfo) IsEndpointInfo() {} type endpoint struct { EndpointInfo + // endpointEntry is used to queue endpoints for processing to the + // a given tcp processor goroutine. + // + // Precondition: epQueue.mu must be held to read/write this field.. + endpointEntry `state:"nosave"` + + // pendingProcessing is true if this endpoint is queued for processing + // to a TCP processor. + // + // Precondition: epQueue.mu must be held to read/write this field.. + pendingProcessing bool `state:"nosave"` + // workMu is used to arbitrate which goroutine may perform protocol // work. Only the main protocol goroutine is expected to call Lock() on // it, but other goroutines (e.g., send) may call TryLock() to eagerly @@ -324,6 +338,7 @@ type endpoint struct { // The following fields are protected by the mutex. mu sync.RWMutex `state:"nosave"` + // state must be read/set using the EndpointState()/setEndpointState() methods. state EndpointState `state:".(EndpointState)"` // origEndpointState is only used during a restore phase to save the @@ -359,7 +374,7 @@ type endpoint struct { workerRunning bool // workerCleanup specifies if the worker goroutine must perform cleanup - // before exitting. This can only be set to true when workerRunning is + // before exiting. This can only be set to true when workerRunning is // also true, and they're both protected by the mutex. workerCleanup bool @@ -371,6 +386,8 @@ type endpoint struct { // recentTS is the timestamp that should be sent in the TSEcr field of // the timestamp for future segments sent by the endpoint. This field is // updated if required when a new segment is received by this endpoint. + // + // recentTS must be read/written atomically. recentTS uint32 // tsOffset is a randomized offset added to the value of the @@ -567,6 +584,47 @@ func (e *endpoint) ResumeWork() { e.workMu.Unlock() } +// setEndpointState updates the state of the endpoint to state atomically. This +// method is unexported as the only place we should update the state is in this +// package but we allow the state to be read freely without holding e.mu. +// +// Precondition: e.mu must be held to call this method. +func (e *endpoint) setEndpointState(state EndpointState) { + oldstate := EndpointState(atomic.LoadUint32((*uint32)(&e.state))) + switch state { + case StateEstablished: + e.stack.Stats().TCP.CurrentEstablished.Increment() + case StateError: + fallthrough + case StateClose: + if oldstate == StateCloseWait || oldstate == StateEstablished { + e.stack.Stats().TCP.EstablishedResets.Increment() + } + fallthrough + default: + if oldstate == StateEstablished { + e.stack.Stats().TCP.CurrentEstablished.Decrement() + } + } + atomic.StoreUint32((*uint32)(&e.state), uint32(state)) +} + +// EndpointState returns the current state of the endpoint. +func (e *endpoint) EndpointState() EndpointState { + return EndpointState(atomic.LoadUint32((*uint32)(&e.state))) +} + +// setRecentTimestamp atomically sets the recentTS field to the +// provided value. +func (e *endpoint) setRecentTimestamp(recentTS uint32) { + atomic.StoreUint32(&e.recentTS, recentTS) +} + +// recentTimestamp atomically reads and returns the value of the recentTS field. +func (e *endpoint) recentTimestamp() uint32 { + return atomic.LoadUint32(&e.recentTS) +} + // keepalive is a synchronization wrapper used to appease stateify. See the // comment in endpoint, where it is used. // @@ -656,7 +714,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { e.mu.RLock() defer e.mu.RUnlock() - switch e.state { + switch e.EndpointState() { case StateInitial, StateBound, StateConnecting, StateSynSent, StateSynRecv: // Ready for nothing. @@ -672,7 +730,7 @@ func (e *endpoint) Readiness(mask waiter.EventMask) waiter.EventMask { } } } - if e.state.connected() { + if e.EndpointState().connected() { // Determine if the endpoint is writable if requested. if (mask & waiter.EventOut) != 0 { e.sndBufMu.Lock() @@ -733,14 +791,20 @@ func (e *endpoint) Close() { // Issue a shutdown so that the peer knows we won't send any more data // if we're connected, or stop accepting if we're listening. e.Shutdown(tcpip.ShutdownWrite | tcpip.ShutdownRead) + e.closeNoShutdown() +} +// closeNoShutdown closes the endpoint without doing a full shutdown. This is +// used when a connection needs to be aborted with a RST and we want to skip +// a full 4 way TCP shutdown. +func (e *endpoint) closeNoShutdown() { e.mu.Lock() // For listening sockets, we always release ports inline so that they // are immediately available for reuse after Close() is called. If also // registered, we unregister as well otherwise the next user would fail // in Listen() when trying to register. - if e.state == StateListen && e.isPortReserved { + if e.EndpointState() == StateListen && e.isPortReserved { if e.isRegistered { e.stack.StartTransportEndpointCleanup(e.boundNICID, e.effectiveNetProtos, ProtocolNumber, e.ID, e, e.boundBindToDevice) e.isRegistered = false @@ -780,6 +844,8 @@ func (e *endpoint) closePendingAcceptableConnectionsLocked() { defer close(done) for n := range e.acceptedChan { n.notifyProtocolGoroutine(notifyReset) + // close all connections that have completed but + // not accepted by the application. n.Close() } }() @@ -797,11 +863,13 @@ func (e *endpoint) closePendingAcceptableConnectionsLocked() { // after Close() is called and the worker goroutine (if any) is done with its // work. func (e *endpoint) cleanupLocked() { + // Close all endpoints that might have been accepted by TCP but not by // the client. if e.acceptedChan != nil { e.closePendingAcceptableConnectionsLocked() } + e.workerCleanup = false if e.isRegistered { @@ -920,7 +988,7 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, // reads to proceed before returning a ECONNRESET. e.rcvListMu.Lock() bufUsed := e.rcvBufUsed - if s := e.state; !s.connected() && s != StateClose && bufUsed == 0 { + if s := e.EndpointState(); !s.connected() && s != StateClose && bufUsed == 0 { e.rcvListMu.Unlock() he := e.HardError e.mu.RUnlock() @@ -944,7 +1012,7 @@ func (e *endpoint) Read(*tcpip.FullAddress) (buffer.View, tcpip.ControlMessages, func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) { if e.rcvBufUsed == 0 { - if e.rcvClosed || !e.state.connected() { + if e.rcvClosed || !e.EndpointState().connected() { return buffer.View{}, tcpip.ErrClosedForReceive } return buffer.View{}, tcpip.ErrWouldBlock @@ -980,8 +1048,8 @@ func (e *endpoint) readLocked() (buffer.View, *tcpip.Error) { // Caller must hold e.mu and e.sndBufMu func (e *endpoint) isEndpointWritableLocked() (int, *tcpip.Error) { // The endpoint cannot be written to if it's not connected. - if !e.state.connected() { - switch e.state { + if !e.EndpointState().connected() { + switch e.EndpointState() { case StateError: return 0, e.HardError default: @@ -1039,42 +1107,86 @@ func (e *endpoint) Write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, perr } - if !opts.Atomic { // See above. - e.mu.RLock() - e.sndBufMu.Lock() + if opts.Atomic { + // Add data to the send queue. + s := newSegmentFromView(&e.route, e.ID, v) + e.sndBufUsed += len(v) + e.sndBufInQueue += seqnum.Size(len(v)) + e.sndQueue.PushBack(s) + e.sndBufMu.Unlock() + // Release the endpoint lock to prevent deadlocks due to lock + // order inversion when acquiring workMu. + e.mu.RUnlock() + } - // Because we released the lock before copying, check state again - // to make sure the endpoint is still in a valid state for a write. - avail, err = e.isEndpointWritableLocked() - if err != nil { + if e.workMu.TryLock() { + // Since we released locks in between it's possible that the + // endpoint transitioned to a CLOSED/ERROR states so make + // sure endpoint is still writable before trying to write. + if !opts.Atomic { // See above. + e.mu.RLock() + e.sndBufMu.Lock() + + // Because we released the lock before copying, check state again + // to make sure the endpoint is still in a valid state for a write. + avail, err = e.isEndpointWritableLocked() + if err != nil { + e.sndBufMu.Unlock() + e.mu.RUnlock() + e.stats.WriteErrors.WriteClosed.Increment() + return 0, nil, err + } + + // Discard any excess data copied in due to avail being reduced due + // to a simultaneous write call to the socket. + if avail < len(v) { + v = v[:avail] + } + // Add data to the send queue. + s := newSegmentFromView(&e.route, e.ID, v) + e.sndBufUsed += len(v) + e.sndBufInQueue += seqnum.Size(len(v)) + e.sndQueue.PushBack(s) e.sndBufMu.Unlock() + // Release the endpoint lock to prevent deadlocks due to lock + // order inversion when acquiring workMu. e.mu.RUnlock() - e.stats.WriteErrors.WriteClosed.Increment() - return 0, nil, err - } - // Discard any excess data copied in due to avail being reduced due - // to a simultaneous write call to the socket. - if avail < len(v) { - v = v[:avail] } - } - - // Add data to the send queue. - s := newSegmentFromView(&e.route, e.ID, v) - e.sndBufUsed += len(v) - e.sndBufInQueue += seqnum.Size(len(v)) - e.sndQueue.PushBack(s) - e.sndBufMu.Unlock() - // Release the endpoint lock to prevent deadlocks due to lock - // order inversion when acquiring workMu. - e.mu.RUnlock() - - if e.workMu.TryLock() { // Do the work inline. e.handleWrite() e.workMu.Unlock() } else { + if !opts.Atomic { // See above. + e.mu.RLock() + e.sndBufMu.Lock() + + // Because we released the lock before copying, check state again + // to make sure the endpoint is still in a valid state for a write. + avail, err = e.isEndpointWritableLocked() + if err != nil { + e.sndBufMu.Unlock() + e.mu.RUnlock() + e.stats.WriteErrors.WriteClosed.Increment() + return 0, nil, err + } + + // Discard any excess data copied in due to avail being reduced due + // to a simultaneous write call to the socket. + if avail < len(v) { + v = v[:avail] + } + // Add data to the send queue. + s := newSegmentFromView(&e.route, e.ID, v) + e.sndBufUsed += len(v) + e.sndBufInQueue += seqnum.Size(len(v)) + e.sndQueue.PushBack(s) + e.sndBufMu.Unlock() + // Release the endpoint lock to prevent deadlocks due to lock + // order inversion when acquiring workMu. + e.mu.RUnlock() + + } // Let the protocol goroutine do the work. e.sndWaker.Assert() } @@ -1091,7 +1203,7 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro // The endpoint can be read if it's connected, or if it's already closed // but has some pending unread data. - if s := e.state; !s.connected() && s != StateClose { + if s := e.EndpointState(); !s.connected() && s != StateClose { if s == StateError { return 0, tcpip.ControlMessages{}, e.HardError } @@ -1103,7 +1215,7 @@ func (e *endpoint) Peek(vec [][]byte) (int64, tcpip.ControlMessages, *tcpip.Erro defer e.rcvListMu.Unlock() if e.rcvBufUsed == 0 { - if e.rcvClosed || !e.state.connected() { + if e.rcvClosed || !e.EndpointState().connected() { e.stats.ReadErrors.ReadClosed.Increment() return 0, tcpip.ControlMessages{}, tcpip.ErrClosedForReceive } @@ -1187,7 +1299,7 @@ func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { defer e.mu.Unlock() // We only allow this to be set when we're in the initial state. - if e.state != StateInitial { + if e.EndpointState() != StateInitial { return tcpip.ErrInvalidEndpointState } @@ -1402,14 +1514,14 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // Acquire the work mutex as we may need to // reinitialize the congestion control state. e.mu.Lock() - state := e.state + state := e.EndpointState() e.cc = v e.mu.Unlock() switch state { case StateEstablished: e.workMu.Lock() e.mu.Lock() - if e.state == state { + if e.EndpointState() == state { e.snd.cc = e.snd.initCongestionControl(e.cc) } e.mu.Unlock() @@ -1472,7 +1584,7 @@ func (e *endpoint) readyReceiveSize() (int, *tcpip.Error) { defer e.mu.RUnlock() // The endpoint cannot be in listen state. - if e.state == StateListen { + if e.EndpointState() == StateListen { return 0, tcpip.ErrInvalidEndpointState } @@ -1691,26 +1803,11 @@ func (e *endpoint) GetSockOpt(opt interface{}) *tcpip.Error { } func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if header.IsV4MappedAddress(addr.Addr) { - // Fail if using a v4 mapped address on a v6only endpoint. - if e.v6only { - return 0, tcpip.ErrNoRoute - } - - netProto = header.IPv4ProtocolNumber - addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] - if addr.Addr == header.IPv4Any { - addr.Addr = "" - } - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); l != 0 && len(addr.Addr) != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } @@ -1746,7 +1843,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc return err } - if e.state.connected() { + if e.EndpointState().connected() { // The endpoint is already connected. If caller hasn't been // notified yet, return success. if !e.isConnectNotified { @@ -1758,7 +1855,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } nicID := addr.NIC - switch e.state { + switch e.EndpointState() { case StateBound: // If we're already bound to a NIC but the caller is requesting // that we use a different one now, we cannot proceed. @@ -1865,7 +1962,7 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } e.isRegistered = true - e.state = StateConnecting + e.setEndpointState(StateConnecting) e.route = r.Clone() e.boundNICID = nicID e.effectiveNetProtos = netProtos @@ -1886,14 +1983,13 @@ func (e *endpoint) connect(addr tcpip.FullAddress, handshake bool, run bool) *tc } e.segmentQueue.mu.Unlock() e.snd.updateMaxPayloadSize(int(e.route.MTU()), 0) - e.state = StateEstablished - e.stack.Stats().TCP.CurrentEstablished.Increment() + e.setEndpointState(StateEstablished) } if run { e.workerRunning = true e.stack.Stats().TCP.ActiveConnectionOpenings.Increment() - go e.protocolMainLoop(handshake) // S/R-SAFE: will be drained before save. + go e.protocolMainLoop(handshake, nil) // S/R-SAFE: will be drained before save. } return tcpip.ErrConnectStarted @@ -1911,7 +2007,7 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { e.shutdownFlags |= flags finQueued := false switch { - case e.state.connected(): + case e.EndpointState().connected(): // Close for read. if (e.shutdownFlags & tcpip.ShutdownRead) != 0 { // Mark read side as closed. @@ -1923,8 +2019,18 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { // If we're fully closed and we have unread data we need to abort // the connection with a RST. if (e.shutdownFlags&tcpip.ShutdownWrite) != 0 && rcvBufUsed > 0 { - e.notifyProtocolGoroutine(notifyReset) e.mu.Unlock() + // Try to send an active reset immediately if the + // work mutex is available. + if e.workMu.TryLock() { + e.mu.Lock() + e.resetConnectionLocked(tcpip.ErrConnectionAborted) + e.notifyProtocolGoroutine(notifyTickleWorker) + e.mu.Unlock() + e.workMu.Unlock() + } else { + e.notifyProtocolGoroutine(notifyReset) + } return nil } } @@ -1946,11 +2052,10 @@ func (e *endpoint) Shutdown(flags tcpip.ShutdownFlags) *tcpip.Error { finQueued = true // Mark endpoint as closed. e.sndClosed = true - e.sndBufMu.Unlock() } - case e.state == StateListen: + case e.EndpointState() == StateListen: // Tell protocolListenLoop to stop. if flags&tcpip.ShutdownRead != 0 { e.notifyProtocolGoroutine(notifyClose) @@ -1991,7 +2096,7 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { // When the endpoint shuts down, it sets workerCleanup to true, and from // that point onward, acceptedChan is the responsibility of the cleanup() // method (and should not be touched anywhere else, including here). - if e.state == StateListen && !e.workerCleanup { + if e.EndpointState() == StateListen && !e.workerCleanup { // Adjust the size of the channel iff we can fix existing // pending connections into the new one. if len(e.acceptedChan) > backlog { @@ -2009,7 +2114,7 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { return nil } - if e.state == StateInitial { + if e.EndpointState() == StateInitial { // The listen is called on an unbound socket, the socket is // automatically bound to a random free port with the local // address set to INADDR_ANY. @@ -2019,7 +2124,7 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { } // Endpoint must be bound before it can transition to listen mode. - if e.state != StateBound { + if e.EndpointState() != StateBound { e.stats.ReadErrors.InvalidEndpointState.Increment() return tcpip.ErrInvalidEndpointState } @@ -2030,24 +2135,27 @@ func (e *endpoint) listen(backlog int) *tcpip.Error { } e.isRegistered = true - e.state = StateListen + e.setEndpointState(StateListen) + if e.acceptedChan == nil { e.acceptedChan = make(chan *endpoint, backlog) } e.workerRunning = true - go e.protocolListenLoop( // S/R-SAFE: drained on save. seqnum.Size(e.receiveBufferAvailable())) - return nil } // startAcceptedLoop sets up required state and starts a goroutine with the // main loop for accepted connections. func (e *endpoint) startAcceptedLoop(waiterQueue *waiter.Queue) { + e.mu.Lock() e.waiterQueue = waiterQueue e.workerRunning = true - go e.protocolMainLoop(false) // S/R-SAFE: drained on save. + e.mu.Unlock() + wakerInitDone := make(chan struct{}) + go e.protocolMainLoop(false, wakerInitDone) // S/R-SAFE: drained on save. + <-wakerInitDone } // Accept returns a new endpoint if a peer has established a connection @@ -2057,7 +2165,7 @@ func (e *endpoint) Accept() (tcpip.Endpoint, *waiter.Queue, *tcpip.Error) { defer e.mu.RUnlock() // Endpoint must be in listen state before it can accept connections. - if e.state != StateListen { + if e.EndpointState() != StateListen { return nil, nil, tcpip.ErrInvalidEndpointState } @@ -2084,7 +2192,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { // Don't allow binding once endpoint is not in the initial state // anymore. This is because once the endpoint goes into a connected or // listen state, it is already bound. - if e.state != StateInitial { + if e.EndpointState() != StateInitial { return tcpip.ErrAlreadyBound } @@ -2146,7 +2254,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) (err *tcpip.Error) { } // Mark endpoint as bound. - e.state = StateBound + e.setEndpointState(StateBound) return nil } @@ -2168,7 +2276,7 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { e.mu.RLock() defer e.mu.RUnlock() - if !e.state.connected() { + if !e.EndpointState().connected() { return tcpip.FullAddress{}, tcpip.ErrNotConnected } @@ -2179,45 +2287,22 @@ func (e *endpoint) GetRemoteAddress() (tcpip.FullAddress, *tcpip.Error) { }, nil } -// HandlePacket is called by the stack when new packets arrive to this transport -// endpoint. func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { - s := newSegment(r, id, pkt) - if !s.parse() { - e.stack.Stats().MalformedRcvdPackets.Increment() - e.stack.Stats().TCP.InvalidSegmentsReceived.Increment() - e.stats.ReceiveErrors.MalformedPacketsReceived.Increment() - s.decRef() - return - } - - if !s.csumValid { - e.stack.Stats().MalformedRcvdPackets.Increment() - e.stack.Stats().TCP.ChecksumErrors.Increment() - e.stats.ReceiveErrors.ChecksumErrors.Increment() - s.decRef() - return - } - - e.stack.Stats().TCP.ValidSegmentsReceived.Increment() - e.stats.SegmentsReceived.Increment() - if (s.flags & header.TCPFlagRst) != 0 { - e.stack.Stats().TCP.ResetsReceived.Increment() - } - - e.enqueueSegment(s) + // TCP HandlePacket is not required anymore as inbound packets first + // land at the Dispatcher which then can either delivery using the + // worker go routine or directly do the invoke the tcp processing inline + // based on the state of the endpoint. } -func (e *endpoint) enqueueSegment(s *segment) { +func (e *endpoint) enqueueSegment(s *segment) bool { // Send packet to worker goroutine. - if e.segmentQueue.enqueue(s) { - e.newSegmentWaker.Assert() - } else { + if !e.segmentQueue.enqueue(s) { // The queue is full, so we drop the segment. e.stack.Stats().DroppedPackets.Increment() e.stats.ReceiveErrors.SegmentQueueDropped.Increment() - s.decRef() + return false } + return true } // HandleControlPacket implements stack.TransportEndpoint.HandleControlPacket. @@ -2334,8 +2419,8 @@ func (e *endpoint) rcvWndScaleForHandshake() int { // updateRecentTimestamp updates the recent timestamp using the algorithm // described in https://tools.ietf.org/html/rfc7323#section-4.3 func (e *endpoint) updateRecentTimestamp(tsVal uint32, maxSentAck seqnum.Value, segSeq seqnum.Value) { - if e.sendTSOk && seqnum.Value(e.recentTS).LessThan(seqnum.Value(tsVal)) && segSeq.LessThanEq(maxSentAck) { - e.recentTS = tsVal + if e.sendTSOk && seqnum.Value(e.recentTimestamp()).LessThan(seqnum.Value(tsVal)) && segSeq.LessThanEq(maxSentAck) { + e.setRecentTimestamp(tsVal) } } @@ -2345,7 +2430,7 @@ func (e *endpoint) updateRecentTimestamp(tsVal uint32, maxSentAck seqnum.Value, func (e *endpoint) maybeEnableTimestamp(synOpts *header.TCPSynOptions) { if synOpts.TS { e.sendTSOk = true - e.recentTS = synOpts.TSVal + e.setRecentTimestamp(synOpts.TSVal) } } @@ -2434,7 +2519,7 @@ func (e *endpoint) completeState() stack.TCPEndpointState { // Endpoint TCP Option state. s.SendTSOk = e.sendTSOk - s.RecentTS = e.recentTS + s.RecentTS = e.recentTimestamp() s.TSOffset = e.tsOffset s.SACKPermitted = e.sackPermitted s.SACK.Blocks = make([]header.SACKBlock, e.sack.NumBlocks) @@ -2541,9 +2626,7 @@ func (e *endpoint) initGSO() { // State implements tcpip.Endpoint.State. It exports the endpoint's protocol // state for diagnostics. func (e *endpoint) State() uint32 { - e.mu.Lock() - defer e.mu.Unlock() - return uint32(e.state) + return uint32(e.EndpointState()) } // Info returns a copy of the endpoint info. diff --git a/pkg/tcpip/transport/tcp/endpoint_state.go b/pkg/tcpip/transport/tcp/endpoint_state.go index 4b8d867bc..4a46f0ec5 100644 --- a/pkg/tcpip/transport/tcp/endpoint_state.go +++ b/pkg/tcpip/transport/tcp/endpoint_state.go @@ -16,6 +16,7 @@ package tcp import ( "fmt" + "sync/atomic" "time" "gvisor.dev/gvisor/pkg/sync" @@ -48,7 +49,7 @@ func (e *endpoint) beforeSave() { e.mu.Lock() defer e.mu.Unlock() - switch e.state { + switch e.EndpointState() { case StateInitial, StateBound: // TODO(b/138137272): this enumeration duplicates // EndpointState.connected. remove it. @@ -70,31 +71,30 @@ func (e *endpoint) beforeSave() { fallthrough case StateListen, StateConnecting: e.drainSegmentLocked() - if e.state != StateClose && e.state != StateError { + if e.EndpointState() != StateClose && e.EndpointState() != StateError { if !e.workerRunning { panic("endpoint has no worker running in listen, connecting, or connected state") } break } - fallthrough case StateError, StateClose: - for (e.state == StateError || e.state == StateClose) && e.workerRunning { + for e.workerRunning { e.mu.Unlock() time.Sleep(100 * time.Millisecond) e.mu.Lock() } if e.workerRunning { - panic("endpoint still has worker running in closed or error state") + panic(fmt.Sprintf("endpoint: %+v still has worker running in closed or error state", e.ID)) } default: - panic(fmt.Sprintf("endpoint in unknown state %v", e.state)) + panic(fmt.Sprintf("endpoint in unknown state %v", e.EndpointState())) } if e.waiterQueue != nil && !e.waiterQueue.IsEmpty() { panic("endpoint still has waiters upon save") } - if e.state != StateClose && !((e.state == StateBound || e.state == StateListen) == e.isPortReserved) { + if e.EndpointState() != StateClose && !((e.EndpointState() == StateBound || e.EndpointState() == StateListen) == e.isPortReserved) { panic("endpoints which are not in the closed state must have a reserved port IFF they are in bound or listen state") } } @@ -135,7 +135,7 @@ func (e *endpoint) loadAcceptedChan(acceptedEndpoints []*endpoint) { // saveState is invoked by stateify. func (e *endpoint) saveState() EndpointState { - return e.state + return e.EndpointState() } // Endpoint loading must be done in the following ordering by their state, to @@ -151,7 +151,8 @@ var connectingLoading sync.WaitGroup func (e *endpoint) loadState(state EndpointState) { // This is to ensure that the loading wait groups include all applicable // endpoints before any asynchronous calls to the Wait() methods. - if state.connected() { + // For restore purposes we treat TimeWait like a connected endpoint. + if state.connected() || state == StateTimeWait { connectedLoading.Add(1) } switch state { @@ -160,13 +161,14 @@ func (e *endpoint) loadState(state EndpointState) { case StateConnecting, StateSynSent, StateSynRecv: connectingLoading.Add(1) } - e.state = state + // Directly update the state here rather than using e.setEndpointState + // as the endpoint is still being loaded and the stack reference to increment + // metrics is not yet initialized. + atomic.StoreUint32((*uint32)(&e.state), uint32(state)) } // afterLoad is invoked by stateify. func (e *endpoint) afterLoad() { - // Freeze segment queue before registering to prevent any segments - // from being delivered while it is being restored. e.origEndpointState = e.state // Restore the endpoint to InitialState as it will be moved to // its origEndpointState during Resume. @@ -180,7 +182,6 @@ func (e *endpoint) Resume(s *stack.Stack) { e.segmentQueue.setLimit(MaxUnprocessedSegments) e.workMu.Init() state := e.origEndpointState - switch state { case StateInitial, StateBound, StateListen, StateConnecting, StateEstablished: var ss SendBufferSizeOption @@ -276,7 +277,7 @@ func (e *endpoint) Resume(s *stack.Stack) { listenLoading.Wait() connectingLoading.Wait() bind() - e.state = StateClose + e.setEndpointState(StateClose) tcpip.AsyncLoading.Done() }() } @@ -288,6 +289,7 @@ func (e *endpoint) Resume(s *stack.Stack) { e.stack.CompleteTransportEndpointCleanup(e) tcpip.DeleteDanglingEndpoint(e) } + } // saveLastError is invoked by stateify. diff --git a/pkg/tcpip/transport/tcp/protocol.go b/pkg/tcpip/transport/tcp/protocol.go index 9a8f64aa6..958c06fa7 100644 --- a/pkg/tcpip/transport/tcp/protocol.go +++ b/pkg/tcpip/transport/tcp/protocol.go @@ -21,6 +21,7 @@ package tcp import ( + "runtime" "strings" "time" @@ -104,6 +105,7 @@ type protocol struct { moderateReceiveBuffer bool tcpLingerTimeout time.Duration tcpTimeWaitTimeout time.Duration + dispatcher *dispatcher } // Number returns the tcp protocol number. @@ -134,6 +136,14 @@ func (*protocol) ParsePorts(v buffer.View) (src, dst uint16, err *tcpip.Error) { return h.SourcePort(), h.DestinationPort(), nil } +// QueuePacket queues packets targeted at an endpoint after hashing the packet +// to a specific processing queue. Each queue is serviced by its own processor +// goroutine which is responsible for dequeuing and doing full TCP dispatch of +// the packet. +func (p *protocol) QueuePacket(r *stack.Route, ep stack.TransportEndpoint, id stack.TransportEndpointID, pkt tcpip.PacketBuffer) { + p.dispatcher.queuePacket(r, ep, id, pkt) +} + // HandleUnknownDestinationPacket handles packets targeted at this protocol but // that don't match any existing endpoint. // @@ -330,5 +340,6 @@ func NewProtocol() stack.TransportProtocol { availableCongestionControl: []string{ccReno, ccCubic}, tcpLingerTimeout: DefaultTCPLingerTimeout, tcpTimeWaitTimeout: DefaultTCPTimeWaitTimeout, + dispatcher: newDispatcher(runtime.GOMAXPROCS(0)), } } diff --git a/pkg/tcpip/transport/tcp/rcv.go b/pkg/tcpip/transport/tcp/rcv.go index 05c8488f8..958f03ac1 100644 --- a/pkg/tcpip/transport/tcp/rcv.go +++ b/pkg/tcpip/transport/tcp/rcv.go @@ -169,19 +169,19 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // We just received a FIN, our next state depends on whether we sent a // FIN already or not. r.ep.mu.Lock() - switch r.ep.state { + switch r.ep.EndpointState() { case StateEstablished: - r.ep.state = StateCloseWait + r.ep.setEndpointState(StateCloseWait) case StateFinWait1: if s.flagIsSet(header.TCPFlagAck) { // FIN-ACK, transition to TIME-WAIT. - r.ep.state = StateTimeWait + r.ep.setEndpointState(StateTimeWait) } else { // Simultaneous close, expecting a final ACK. - r.ep.state = StateClosing + r.ep.setEndpointState(StateClosing) } case StateFinWait2: - r.ep.state = StateTimeWait + r.ep.setEndpointState(StateTimeWait) } r.ep.mu.Unlock() @@ -205,16 +205,16 @@ func (r *receiver) consumeSegment(s *segment, segSeq seqnum.Value, segLen seqnum // shutdown states. if s.flagIsSet(header.TCPFlagAck) && s.ackNumber == r.ep.snd.sndNxt { r.ep.mu.Lock() - switch r.ep.state { + switch r.ep.EndpointState() { case StateFinWait1: - r.ep.state = StateFinWait2 + r.ep.setEndpointState(StateFinWait2) // Notify protocol goroutine that we have received an // ACK to our FIN so that it can start the FIN_WAIT2 // timer to abort connection if the other side does // not close within 2MSL. r.ep.notifyProtocolGoroutine(notifyClose) case StateClosing: - r.ep.state = StateTimeWait + r.ep.setEndpointState(StateTimeWait) case StateLastAck: r.ep.transitionToStateCloseLocked() } @@ -267,7 +267,6 @@ func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, clo switch state { case StateCloseWait, StateClosing, StateLastAck: if !s.sequenceNumber.LessThanEq(r.rcvNxt) { - s.decRef() // Just drop the segment as we have // already received a FIN and this // segment is after the sequence number @@ -284,7 +283,6 @@ func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, clo // trigger a RST. endDataSeq := s.sequenceNumber.Add(seqnum.Size(s.data.Size())) if rcvClosed && r.rcvNxt.LessThan(endDataSeq) { - s.decRef() return true, tcpip.ErrConnectionAborted } if state == StateFinWait1 { @@ -314,7 +312,6 @@ func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, clo // the last actual data octet in a segment in // which it occurs. if closed && (!s.flagIsSet(header.TCPFlagFin) || s.sequenceNumber.Add(s.logicalLen()) != r.rcvNxt+1) { - s.decRef() return true, tcpip.ErrConnectionAborted } } @@ -336,7 +333,7 @@ func (r *receiver) handleRcvdSegmentClosing(s *segment, state EndpointState, clo // r as they arrive. It is called by the protocol main loop. func (r *receiver) handleRcvdSegment(s *segment) (drop bool, err *tcpip.Error) { r.ep.mu.RLock() - state := r.ep.state + state := r.ep.EndpointState() closed := r.ep.closed r.ep.mu.RUnlock() diff --git a/pkg/tcpip/transport/tcp/snd.go b/pkg/tcpip/transport/tcp/snd.go index fdff7ed81..b74b61e7d 100644 --- a/pkg/tcpip/transport/tcp/snd.go +++ b/pkg/tcpip/transport/tcp/snd.go @@ -705,17 +705,15 @@ func (s *sender) maybeSendSegment(seg *segment, limit int, end seqnum.Value) (se } seg.flags = header.TCPFlagAck | header.TCPFlagFin segEnd = seg.sequenceNumber.Add(1) - // Transition to FIN-WAIT1 state since we're initiating an active close. - s.ep.mu.Lock() - switch s.ep.state { + // Update the state to reflect that we have now + // queued a FIN. + switch s.ep.EndpointState() { case StateCloseWait: - // We've already received a FIN and are now sending our own. The - // sender is now awaiting a final ACK for this FIN. - s.ep.state = StateLastAck + s.ep.setEndpointState(StateLastAck) default: - s.ep.state = StateFinWait1 + s.ep.setEndpointState(StateFinWait1) } - s.ep.mu.Unlock() + } else { // We're sending a non-FIN segment. if seg.flags&header.TCPFlagFin != 0 { diff --git a/pkg/tcpip/transport/tcp/tcp_test.go b/pkg/tcpip/transport/tcp/tcp_test.go index 6edfa8dce..a9dfbe857 100644 --- a/pkg/tcpip/transport/tcp/tcp_test.go +++ b/pkg/tcpip/transport/tcp/tcp_test.go @@ -293,7 +293,6 @@ func TestTCPResetSentForACKWhenNotUsingSynCookies(t *testing.T) { checker.SeqNum(uint32(c.IRS+1)), checker.AckNum(uint32(iss)+1), checker.TCPFlags(header.TCPFlagFin|header.TCPFlagAck))) - finHeaders := &context.Headers{ SrcPort: context.TestPort, DstPort: context.StackPort, @@ -459,6 +458,9 @@ func TestConnectResetAfterClose(t *testing.T) { checker.IPv4(t, b, checker.TCP( checker.DstPort(context.TestPort), + // RST is always generated with sndNxt which if the FIN + // has been sent will be 1 higher than the sequence number + // of the FIN itself. checker.SeqNum(uint32(c.IRS)+2), checker.AckNum(0), checker.TCPFlags(header.TCPFlagRst), @@ -1500,6 +1502,9 @@ func TestRstOnCloseWithUnreadDataFinConvertRst(t *testing.T) { checker.TCP( checker.DstPort(context.TestPort), checker.TCPFlags(header.TCPFlagAck|header.TCPFlagRst), + // RST is always generated with sndNxt which if the FIN + // has been sent will be 1 higher than the sequence + // number of the FIN itself. checker.SeqNum(uint32(c.IRS)+2), )) // The RST puts the endpoint into an error state. @@ -5441,6 +5446,7 @@ func TestReceiveBufferAutoTuningApplicationLimited(t *testing.T) { rawEP.SendPacketWithTS(b[start:start+mss], tsVal) packetsSent++ } + // Resume the worker so that it only sees the packets once all of them // are waiting to be read. worker.ResumeWork() @@ -5508,7 +5514,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { stk := c.Stack() // Set lower limits for auto-tuning tests. This is required because the // test stops the worker which can cause packets to be dropped because - // the segment queue holding unprocessed packets is limited to 500. + // the segment queue holding unprocessed packets is limited to 300. const receiveBufferSize = 80 << 10 // 80KB. const maxReceiveBufferSize = receiveBufferSize * 10 if err := stk.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.ReceiveBufferSizeOption{1, receiveBufferSize, maxReceiveBufferSize}); err != nil { @@ -5563,6 +5569,7 @@ func TestReceiveBufferAutoTuning(t *testing.T) { totalSent += mss packetsSent++ } + // Resume it so that it only sees the packets once all of them // are waiting to be read. worker.ResumeWork() diff --git a/pkg/tcpip/transport/udp/endpoint.go b/pkg/tcpip/transport/udp/endpoint.go index a4ff29a7d..c9cbed8f4 100644 --- a/pkg/tcpip/transport/udp/endpoint.go +++ b/pkg/tcpip/transport/udp/endpoint.go @@ -31,6 +31,7 @@ type udpPacket struct { senderAddress tcpip.FullAddress data buffer.VectorisedView `state:".(buffer.VectorisedView)"` timestamp int64 + tos uint8 } // EndpointState represents the state of a UDP endpoint. @@ -113,6 +114,10 @@ type endpoint struct { // applied while sending packets. Defaults to 0 as on Linux. sendTOS uint8 + // receiveTOS determines if the incoming IPv4 TOS header field is passed + // as ancillary data to ControlMessages on Read. + receiveTOS bool + // shutdownFlags represent the current shutdown state of the endpoint. shutdownFlags tcpip.ShutdownFlags @@ -243,7 +248,18 @@ func (e *endpoint) Read(addr *tcpip.FullAddress) (buffer.View, tcpip.ControlMess *addr = p.senderAddress } - return p.data.ToView(), tcpip.ControlMessages{HasTimestamp: true, Timestamp: p.timestamp}, nil + cm := tcpip.ControlMessages{ + HasTimestamp: true, + Timestamp: p.timestamp, + } + e.mu.RLock() + receiveTOS := e.receiveTOS + e.mu.RUnlock() + if receiveTOS { + cm.HasTOS = true + cm.TOS = p.tos + } + return p.data.ToView(), cm, nil } // prepareForWrite prepares the endpoint for sending data. In particular, it @@ -402,7 +418,7 @@ func (e *endpoint) write(p tcpip.Payloader, opts tcpip.WriteOptions) (int64, <-c return 0, nil, tcpip.ErrBroadcastDisabled } - netProto, err := e.checkV4Mapped(to, false) + netProto, err := e.checkV4Mapped(to) if err != nil { return 0, nil, err } @@ -458,6 +474,12 @@ func (e *endpoint) Peek([][]byte) (int64, tcpip.ControlMessages, *tcpip.Error) { // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool. func (e *endpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { switch opt { + case tcpip.ReceiveTOSOption: + e.mu.Lock() + e.receiveTOS = v + e.mu.Unlock() + return nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { @@ -501,7 +523,7 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { defer e.mu.Unlock() fa := tcpip.FullAddress{Addr: v.InterfaceAddr} - netProto, err := e.checkV4Mapped(&fa, false) + netProto, err := e.checkV4Mapped(&fa) if err != nil { return err } @@ -664,15 +686,21 @@ func (e *endpoint) SetSockOpt(opt interface{}) *tcpip.Error { // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool. func (e *endpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { switch opt { + case tcpip.ReceiveTOSOption: + e.mu.RLock() + v := e.receiveTOS + e.mu.RUnlock() + return v, nil + case tcpip.V6OnlyOption: // We only recognize this option on v6 endpoints. if e.NetProto != header.IPv6ProtocolNumber { return false, tcpip.ErrUnknownProtocolOption } - e.mu.Lock() + e.mu.RLock() v := e.v6only - e.mu.Unlock() + e.mu.RUnlock() return v, nil } @@ -839,35 +867,12 @@ func sendUDP(r *stack.Route, data buffer.VectorisedView, localPort, remotePort u return nil } -func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress, allowMismatch bool) (tcpip.NetworkProtocolNumber, *tcpip.Error) { - netProto := e.NetProto - if len(addr.Addr) == 0 { - return netProto, nil - } - if header.IsV4MappedAddress(addr.Addr) { - // Fail if using a v4 mapped address on a v6only endpoint. - if e.v6only { - return 0, tcpip.ErrNoRoute - } - - netProto = header.IPv4ProtocolNumber - addr.Addr = addr.Addr[header.IPv6AddressSize-header.IPv4AddressSize:] - if addr.Addr == header.IPv4Any { - addr.Addr = "" - } - - // Fail if we are bound to an IPv6 address. - if !allowMismatch && len(e.ID.LocalAddress) == 16 { - return 0, tcpip.ErrNetworkUnreachable - } - } - - // Fail if we're bound to an address length different from the one we're - // checking. - if l := len(e.ID.LocalAddress); l != 0 && l != len(addr.Addr) { - return 0, tcpip.ErrInvalidEndpointState +func (e *endpoint) checkV4Mapped(addr *tcpip.FullAddress) (tcpip.NetworkProtocolNumber, *tcpip.Error) { + unwrapped, netProto, err := e.TransportEndpointInfo.AddrNetProto(*addr, e.v6only) + if err != nil { + return 0, err } - + *addr = unwrapped return netProto, nil } @@ -916,7 +921,7 @@ func (e *endpoint) Disconnect() *tcpip.Error { // Connect connects the endpoint to its peer. Specifying a NIC is optional. func (e *endpoint) Connect(addr tcpip.FullAddress) *tcpip.Error { - netProto, err := e.checkV4Mapped(&addr, false) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } @@ -1074,7 +1079,7 @@ func (e *endpoint) bindLocked(addr tcpip.FullAddress) *tcpip.Error { return tcpip.ErrInvalidEndpointState } - netProto, err := e.checkV4Mapped(&addr, true) + netProto, err := e.checkV4Mapped(&addr) if err != nil { return err } @@ -1238,6 +1243,12 @@ func (e *endpoint) HandlePacket(r *stack.Route, id stack.TransportEndpointID, pk e.rcvList.PushBack(packet) e.rcvBufSize += pkt.Data.Size() + // Save any useful information from the network header to the packet. + switch r.NetProto { + case header.IPv4ProtocolNumber: + packet.tos, _ = header.IPv4(pkt.NetworkHeader).TOS() + } + packet.timestamp = e.stack.NowNanoseconds() e.rcvMu.Unlock() diff --git a/pkg/tcpip/transport/udp/udp_test.go b/pkg/tcpip/transport/udp/udp_test.go index d33507156..c6927cfe3 100644 --- a/pkg/tcpip/transport/udp/udp_test.go +++ b/pkg/tcpip/transport/udp/udp_test.go @@ -56,6 +56,7 @@ const ( multicastAddr = "\xe8\x2b\xd3\xea" multicastV6Addr = "\xff\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00" broadcastAddr = header.IPv4Broadcast + testTOS = 0x80 // defaultMTU is the MTU, in bytes, used throughout the tests, except // where another value is explicitly used. It is chosen to match the MTU @@ -273,11 +274,16 @@ type testContext struct { func newDualTestContext(t *testing.T, mtu uint32) *testContext { t.Helper() - - s := stack.New(stack.Options{ + return newDualTestContextWithOptions(t, mtu, stack.Options{ NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, }) +} + +func newDualTestContextWithOptions(t *testing.T, mtu uint32, options stack.Options) *testContext { + t.Helper() + + s := stack.New(options) ep := channel.New(256, mtu, "") wep := stack.LinkEndpoint(ep) @@ -453,6 +459,7 @@ func (c *testContext) injectV4Packet(payload []byte, h *header4Tuple, valid bool ip := header.IPv4(buf) ip.Encode(&header.IPv4Fields{ IHL: header.IPv4MinimumSize, + TOS: testTOS, TotalLength: uint16(len(buf)), TTL: 65, Protocol: uint8(udp.ProtocolNumber), @@ -552,8 +559,8 @@ func TestBindToDeviceOption(t *testing.T) { // testReadInternal sends a packet of the given test flow into the stack by // injecting it into the link endpoint. It then attempts to read it from the // UDP endpoint and depending on if this was expected to succeed verifies its -// correctness. -func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool) { +// correctness including any additional checker functions provided. +func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expectReadError bool, checkers ...checker.ControlMessagesChecker) { c.t.Helper() payload := newPayload() @@ -568,12 +575,12 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe epstats := c.ep.Stats().(*tcpip.TransportEndpointStats).Clone() var addr tcpip.FullAddress - v, _, err := c.ep.Read(&addr) + v, cm, err := c.ep.Read(&addr) if err == tcpip.ErrWouldBlock { // Wait for data to become available. select { case <-ch: - v, _, err = c.ep.Read(&addr) + v, cm, err = c.ep.Read(&addr) case <-time.After(300 * time.Millisecond): if packetShouldBeDropped { @@ -606,15 +613,21 @@ func testReadInternal(c *testContext, flow testFlow, packetShouldBeDropped, expe if !bytes.Equal(payload, v) { c.t.Fatalf("bad payload: got %x, want %x", v, payload) } + + // Run any checkers against the ControlMessages. + for _, f := range checkers { + f(c.t, cm) + } + c.checkEndpointReadStats(1, epstats, err) } // testRead sends a packet of the given test flow into the stack by injecting it // into the link endpoint. It then reads it from the UDP endpoint and verifies -// its correctness. -func testRead(c *testContext, flow testFlow) { +// its correctness including any additional checker functions provided. +func testRead(c *testContext, flow testFlow, checkers ...checker.ControlMessagesChecker) { c.t.Helper() - testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */) + testReadInternal(c, flow, false /* packetShouldBeDropped */, false /* expectReadError */, checkers...) } // testFailingRead sends a packet of the given test flow into the stack by @@ -755,6 +768,49 @@ func TestV6ReadOnV6(t *testing.T) { testRead(c, unicastV6) } +// TestV4ReadSelfSource checks that packets coming from a local IP address are +// correctly dropped when handleLocal is true and not otherwise. +func TestV4ReadSelfSource(t *testing.T) { + for _, tt := range []struct { + name string + handleLocal bool + wantErr *tcpip.Error + wantInvalidSource uint64 + }{ + {"HandleLocal", false, nil, 0}, + {"NoHandleLocal", true, tcpip.ErrWouldBlock, 1}, + } { + t.Run(tt.name, func(t *testing.T) { + c := newDualTestContextWithOptions(t, defaultMTU, stack.Options{ + NetworkProtocols: []stack.NetworkProtocol{ipv4.NewProtocol(), ipv6.NewProtocol()}, + TransportProtocols: []stack.TransportProtocol{udp.NewProtocol()}, + HandleLocal: tt.handleLocal, + }) + defer c.cleanup() + + c.createEndpointForFlow(unicastV4) + + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + t.Fatalf("Bind failed: %s", err) + } + + payload := newPayload() + h := unicastV4.header4Tuple(incoming) + h.srcAddr = h.dstAddr + + c.injectV4Packet(payload, &h, true /* valid */) + + if got := c.s.Stats().IP.InvalidSourceAddressesReceived.Value(); got != tt.wantInvalidSource { + t.Errorf("c.s.Stats().IP.InvalidSourceAddressesReceived got %d, want %d", got, tt.wantInvalidSource) + } + + if _, _, err := c.ep.Read(nil); err != tt.wantErr { + t.Errorf("c.ep.Read() got error %v, want %v", err, tt.wantErr) + } + }) + } +} + func TestV4ReadOnV4(t *testing.T) { c := newDualTestContext(t, defaultMTU) defer c.cleanup() @@ -1288,7 +1344,7 @@ func TestTOSV4(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv4TOSOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1323,7 +1379,7 @@ func TestTOSV6(t *testing.T) { c.createEndpointForFlow(flow) - const tos = 0xC0 + const tos = testTOS var v tcpip.IPv6TrafficClassOption if err := c.ep.GetSockOpt(&v); err != nil { c.t.Errorf("GetSockopt failed: %s", err) @@ -1350,6 +1406,47 @@ func TestTOSV6(t *testing.T) { } } +func TestReceiveTOSV4(t *testing.T) { + for _, flow := range []testFlow{unicastV4, broadcast} { + t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { + c := newDualTestContext(t, defaultMTU) + defer c.cleanup() + + c.createEndpointForFlow(flow) + + // Verify that setting and reading the option works. + v, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) + } + // Test for expected default value. + if v != false { + c.t.Errorf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", v, false) + } + + want := true + if err := c.ep.SetSockOptBool(tcpip.ReceiveTOSOption, want); err != nil { + c.t.Fatalf("SetSockOptBool(tcpip.ReceiveTOSOption, %t) failed: %s", want, err) + } + + got, err := c.ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + c.t.Fatal("GetSockOptBool(tcpip.ReceiveTOSOption) failed:", err) + } + if got != want { + c.t.Fatalf("got GetSockOptBool(tcpip.ReceiveTOSOption) = %t, want = %t", got, want) + } + + // Verify that the correct received TOS is handed through as + // ancillary data to the ControlMessages struct. + if err := c.ep.Bind(tcpip.FullAddress{Port: stackPort}); err != nil { + c.t.Fatal("Bind failed:", err) + } + testRead(c, flow, checker.ReceiveTOS(testTOS)) + }) + } +} + func TestMulticastInterfaceOption(t *testing.T) { for _, flow := range []testFlow{multicastV4, multicastV4in6, multicastV6, multicastV6Only} { t.Run(fmt.Sprintf("flow:%s", flow), func(t *testing.T) { diff --git a/pkg/waiter/waiter.go b/pkg/waiter/waiter.go index f708e95fa..707eb085b 100644 --- a/pkg/waiter/waiter.go +++ b/pkg/waiter/waiter.go @@ -62,7 +62,7 @@ import ( ) // EventMask represents io events as used in the poll() syscall. -type EventMask uint16 +type EventMask uint64 // Events that waiters can wait on. The meaning is the same as those in the // poll() syscall. |