diff options
Diffstat (limited to 'pkg')
73 files changed, 461 insertions, 406 deletions
diff --git a/pkg/abi/linux/aio.go b/pkg/abi/linux/aio.go index 86ee3f8b5..5fc099892 100644 --- a/pkg/abi/linux/aio.go +++ b/pkg/abi/linux/aio.go @@ -42,6 +42,8 @@ const ( // // The priority field is currently ignored in the implementation below. Also // note that the IOCB_FLAG_RESFD feature is not supported. +// +// +marshal type IOCallback struct { Data uint64 Key uint32 @@ -64,6 +66,7 @@ type IOCallback struct { // IOEvent describes an I/O result. // +// +marshal // +stateify savable type IOEvent struct { Data uint64 diff --git a/pkg/abi/linux/bpf.go b/pkg/abi/linux/bpf.go index aa3d3ce70..9422fcf69 100644 --- a/pkg/abi/linux/bpf.go +++ b/pkg/abi/linux/bpf.go @@ -16,6 +16,7 @@ package linux // BPFInstruction is a raw BPF virtual machine instruction. // +// +marshal slice:BPFInstructionSlice // +stateify savable type BPFInstruction struct { // OpCode is the operation to execute. diff --git a/pkg/abi/linux/capability.go b/pkg/abi/linux/capability.go index 965f74663..afd16cc27 100644 --- a/pkg/abi/linux/capability.go +++ b/pkg/abi/linux/capability.go @@ -177,12 +177,16 @@ const ( ) // CapUserHeader is equivalent to Linux's cap_user_header_t. +// +// +marshal type CapUserHeader struct { Version uint32 Pid int32 } // CapUserData is equivalent to Linux's cap_user_data_t. +// +// +marshal slice:CapUserDataSlice type CapUserData struct { Effective uint32 Permitted uint32 diff --git a/pkg/abi/linux/fcntl.go b/pkg/abi/linux/fcntl.go index 9242e80a5..cc3571fad 100644 --- a/pkg/abi/linux/fcntl.go +++ b/pkg/abi/linux/fcntl.go @@ -45,6 +45,8 @@ const ( ) // Flock is the lock structure for F_SETLK. +// +// +marshal type Flock struct { Type int16 Whence int16 @@ -63,6 +65,8 @@ const ( ) // FOwnerEx is the owner structure for F_SETOWN_EX and F_GETOWN_EX. +// +// +marshal type FOwnerEx struct { Type int32 PID int32 diff --git a/pkg/abi/linux/ipc.go b/pkg/abi/linux/ipc.go index 22acd2d43..c6e65df62 100644 --- a/pkg/abi/linux/ipc.go +++ b/pkg/abi/linux/ipc.go @@ -37,6 +37,8 @@ const IPC_PRIVATE = 0 // features like 32-bit UIDs. // IPCPerm is equivalent to struct ipc64_perm. +// +// +marshal type IPCPerm struct { Key uint32 UID uint32 diff --git a/pkg/abi/linux/linux.go b/pkg/abi/linux/linux.go index 281acdbde..3b4abece1 100644 --- a/pkg/abi/linux/linux.go +++ b/pkg/abi/linux/linux.go @@ -12,7 +12,8 @@ // See the License for the specific language governing permissions and // limitations under the License. -// Package linux contains the constants and types needed to interface with a Linux kernel. +// Package linux contains the constants and types needed to interface with a +// Linux kernel. package linux // NumSoftIRQ is the number of software IRQs, exposed via /proc/stat. @@ -21,6 +22,8 @@ package linux const NumSoftIRQ = 10 // Sysinfo is the structure provided by sysinfo on linux versions > 2.3.48. +// +// +marshal type Sysinfo struct { Uptime int64 Loads [3]uint64 @@ -34,6 +37,6 @@ type Sysinfo struct { _ [6]byte // Pad Procs to 64bits. TotalHigh uint64 FreeHigh uint64 - Unit uint32 - /* The _f field in the glibc version of Sysinfo has size 0 on AMD64 */ + Unit uint32 `marshal:"unaligned"` // Struct ends mid-64-bit-word. + // The _f field in the glibc version of Sysinfo has size 0 on AMD64. } diff --git a/pkg/abi/linux/netlink.go b/pkg/abi/linux/netlink.go index 0ba086c76..b41f94a69 100644 --- a/pkg/abi/linux/netlink.go +++ b/pkg/abi/linux/netlink.go @@ -40,6 +40,8 @@ const ( ) // SockAddrNetlink is struct sockaddr_nl, from uapi/linux/netlink.h. +// +// +marshal type SockAddrNetlink struct { Family uint16 _ uint16 diff --git a/pkg/abi/linux/poll.go b/pkg/abi/linux/poll.go index c04d26e4c..3443a5768 100644 --- a/pkg/abi/linux/poll.go +++ b/pkg/abi/linux/poll.go @@ -15,6 +15,8 @@ package linux // PollFD is struct pollfd, used by poll(2)/ppoll(2), from uapi/asm-generic/poll.h. +// +// +marshal slice:PollFDSlice type PollFD struct { FD int32 Events int16 diff --git a/pkg/abi/linux/rusage.go b/pkg/abi/linux/rusage.go index d8302dc85..e29d0ac7e 100644 --- a/pkg/abi/linux/rusage.go +++ b/pkg/abi/linux/rusage.go @@ -26,6 +26,8 @@ const ( ) // Rusage represents the Linux struct rusage. +// +// +marshal type Rusage struct { UTime Timeval STime Timeval diff --git a/pkg/abi/linux/sem.go b/pkg/abi/linux/sem.go index de422c519..487a626cc 100644 --- a/pkg/abi/linux/sem.go +++ b/pkg/abi/linux/sem.go @@ -35,6 +35,8 @@ const ( const SEM_UNDO = 0x1000 // SemidDS is equivalent to struct semid64_ds. +// +// +marshal type SemidDS struct { SemPerm IPCPerm SemOTime TimeT @@ -45,6 +47,8 @@ type SemidDS struct { } // Sembuf is equivalent to struct sembuf. +// +// +marshal slice:SembufSlice type Sembuf struct { SemNum uint16 SemOp int16 diff --git a/pkg/abi/linux/shm.go b/pkg/abi/linux/shm.go index e45aadb10..274b1e847 100644 --- a/pkg/abi/linux/shm.go +++ b/pkg/abi/linux/shm.go @@ -51,6 +51,8 @@ const ( // ShmidDS is equivalent to struct shmid64_ds. Source: // include/uapi/asm-generic/shmbuf.h +// +// +marshal type ShmidDS struct { ShmPerm IPCPerm ShmSegsz uint64 @@ -66,6 +68,8 @@ type ShmidDS struct { } // ShmParams is equivalent to struct shminfo. Source: include/uapi/linux/shm.h +// +// +marshal type ShmParams struct { ShmMax uint64 ShmMin uint64 @@ -75,6 +79,8 @@ type ShmParams struct { } // ShmInfo is equivalent to struct shm_info. Source: include/uapi/linux/shm.h +// +// +marshal type ShmInfo struct { UsedIDs int32 // Number of currently existing segments. _ [4]byte diff --git a/pkg/abi/linux/signal.go b/pkg/abi/linux/signal.go index 1c330e763..6ca57ffbb 100644 --- a/pkg/abi/linux/signal.go +++ b/pkg/abi/linux/signal.go @@ -214,6 +214,8 @@ const ( ) // Sigevent represents struct sigevent. +// +// +marshal type Sigevent struct { Value uint64 // union sigval {int, void*} Signo int32 diff --git a/pkg/abi/linux/socket.go b/pkg/abi/linux/socket.go index e37c8727d..d156d41e4 100644 --- a/pkg/abi/linux/socket.go +++ b/pkg/abi/linux/socket.go @@ -14,7 +14,10 @@ package linux -import "gvisor.dev/gvisor/pkg/binary" +import ( + "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal" +) // Address families, from linux/socket.h. const ( @@ -265,6 +268,8 @@ type InetMulticastRequestWithNIC struct { type Inet6Addr [16]byte // SockAddrInet6 is struct sockaddr_in6, from uapi/linux/in6.h. +// +// +marshal type SockAddrInet6 struct { Family uint16 Port uint16 @@ -274,6 +279,8 @@ type SockAddrInet6 struct { } // SockAddrLink is a struct sockaddr_ll, from uapi/linux/if_packet.h. +// +// +marshal type SockAddrLink struct { Family uint16 Protocol uint16 @@ -290,6 +297,8 @@ type SockAddrLink struct { const UnixPathMax = 108 // SockAddrUnix is struct sockaddr_un, from uapi/linux/un.h. +// +// +marshal type SockAddrUnix struct { Family uint16 Path [UnixPathMax]int8 @@ -299,6 +308,8 @@ type SockAddrUnix struct { // equivalent to struct sockaddr. SockAddr ensures that a well-defined set of // types can be used as socket addresses. type SockAddr interface { + marshal.Marshallable + // implementsSockAddr exists purely to allow a type to indicate that they // implement this interface. This method is a no-op and shouldn't be called. implementsSockAddr() diff --git a/pkg/abi/linux/time.go b/pkg/abi/linux/time.go index e6860ed49..206f5af7e 100644 --- a/pkg/abi/linux/time.go +++ b/pkg/abi/linux/time.go @@ -93,6 +93,8 @@ const ( const maxSecInDuration = math.MaxInt64 / int64(time.Second) // TimeT represents time_t in <time.h>. It represents time in seconds. +// +// +marshal type TimeT int64 // NsecToTimeT translates nanoseconds to TimeT (seconds). @@ -102,7 +104,7 @@ func NsecToTimeT(nsec int64) TimeT { // Timespec represents struct timespec in <time.h>. // -// +marshal +// +marshal slice:TimespecSlice type Timespec struct { Sec int64 Nsec int64 @@ -158,7 +160,7 @@ const SizeOfTimeval = 16 // Timeval represents struct timeval in <time.h>. // -// +marshal +// +marshal slice:TimevalSlice type Timeval struct { Sec int64 Usec int64 @@ -196,6 +198,8 @@ func DurationToTimeval(dur time.Duration) Timeval { } // Itimerspec represents struct itimerspec in <time.h>. +// +// +marshal type Itimerspec struct { Interval Timespec Value Timespec @@ -206,12 +210,16 @@ type Itimerspec struct { // struct timeval it_interval; /* next value */ // struct timeval it_value; /* current value */ // }; +// +// +marshal type ItimerVal struct { Interval Timeval Value Timeval } // ClockT represents type clock_t. +// +// +marshal type ClockT int64 // ClockTFromDuration converts time.Duration to clock_t. @@ -220,6 +228,8 @@ func ClockTFromDuration(d time.Duration) ClockT { } // Tms represents struct tms, used by times(2). +// +// +marshal type Tms struct { UTime ClockT STime ClockT @@ -229,6 +239,8 @@ type Tms struct { // TimerID represents type timer_t, which identifies a POSIX per-process // interval timer. +// +// +marshal type TimerID int32 // StatxTimestamp represents struct statx_timestamp. diff --git a/pkg/abi/linux/utsname.go b/pkg/abi/linux/utsname.go index 60f220a67..cb7c95437 100644 --- a/pkg/abi/linux/utsname.go +++ b/pkg/abi/linux/utsname.go @@ -26,6 +26,8 @@ const ( ) // UtsName represents struct utsname, the struct returned by uname(2). +// +// +marshal type UtsName struct { Sysname [UTSLen + 1]byte Nodename [UTSLen + 1]byte diff --git a/pkg/sentry/arch/BUILD b/pkg/sentry/arch/BUILD index dc8f0135d..99e2b3389 100644 --- a/pkg/sentry/arch/BUILD +++ b/pkg/sentry/arch/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/cpuid", "//pkg/log", "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/sentry/limits", "//pkg/sync", "//pkg/syserror", diff --git a/pkg/sentry/arch/arch.go b/pkg/sentry/arch/arch.go index a903d031c..d75d665ae 100644 --- a/pkg/sentry/arch/arch.go +++ b/pkg/sentry/arch/arch.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/cpuid" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/usermem" ) @@ -72,12 +73,12 @@ type Context interface { // with return values of varying sizes (for example ARCH_GETFS). This // is a simple utility function to convert to the native size in these // cases, and then we can CopyOut. - Native(val uintptr) interface{} + Native(val uintptr) marshal.Marshallable // Value converts a native type back to a generic value. // Once a value has been converted to native via the above call -- it // can be converted back here. - Value(val interface{}) uintptr + Value(val marshal.Marshallable) uintptr // Width returns the number of bytes for a native value. Width() uint @@ -205,7 +206,7 @@ type Context interface { // equivalent of arch_ptrace(): // PtracePeekUser implements ptrace(PTRACE_PEEKUSR). - PtracePeekUser(addr uintptr) (interface{}, error) + PtracePeekUser(addr uintptr) (marshal.Marshallable, error) // PtracePokeUser implements ptrace(PTRACE_POKEUSR). PtracePokeUser(addr, data uintptr) error diff --git a/pkg/sentry/arch/arch_amd64.go b/pkg/sentry/arch/arch_amd64.go index 1c3e3c14c..c7d3a206d 100644 --- a/pkg/sentry/arch/arch_amd64.go +++ b/pkg/sentry/arch/arch_amd64.go @@ -23,6 +23,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/usermem" ) @@ -179,14 +181,14 @@ func (c *context64) SetOldRSeqInterruptedIP(value uintptr) { } // Native returns the native type for the given val. -func (c *context64) Native(val uintptr) interface{} { - v := uint64(val) +func (c *context64) Native(val uintptr) marshal.Marshallable { + v := primitive.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)) +func (c *context64) Value(val marshal.Marshallable) uintptr { + return uintptr(*val.(*primitive.Uint64)) } // Width returns the byte width of this architecture. @@ -293,7 +295,7 @@ func (c *context64) PIELoadAddress(l MmapLayout) usermem.Addr { const userStructSize = 928 // PtracePeekUser implements Context.PtracePeekUser. -func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { +func (c *context64) PtracePeekUser(addr uintptr) (marshal.Marshallable, error) { if addr&7 != 0 || addr >= userStructSize { return nil, syscall.EIO } diff --git a/pkg/sentry/arch/arch_arm64.go b/pkg/sentry/arch/arch_arm64.go index 550741d8c..680d23a9f 100644 --- a/pkg/sentry/arch/arch_arm64.go +++ b/pkg/sentry/arch/arch_arm64.go @@ -22,6 +22,8 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/cpuid" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/usermem" ) @@ -163,14 +165,14 @@ func (c *context64) SetOldRSeqInterruptedIP(value uintptr) { } // Native returns the native type for the given val. -func (c *context64) Native(val uintptr) interface{} { - v := uint64(val) +func (c *context64) Native(val uintptr) marshal.Marshallable { + v := primitive.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)) +func (c *context64) Value(val marshal.Marshallable) uintptr { + return uintptr(*val.(*primitive.Uint64)) } // Width returns the byte width of this architecture. @@ -274,7 +276,7 @@ func (c *context64) PIELoadAddress(l MmapLayout) usermem.Addr { } // PtracePeekUser implements Context.PtracePeekUser. -func (c *context64) PtracePeekUser(addr uintptr) (interface{}, error) { +func (c *context64) PtracePeekUser(addr uintptr) (marshal.Marshallable, error) { // TODO(gvisor.dev/issue/1239): Full ptrace supporting for Arm64. return c.Native(0), nil } diff --git a/pkg/sentry/fsimpl/verity/BUILD b/pkg/sentry/fsimpl/verity/BUILD index 814eb2085..bc8e38431 100644 --- a/pkg/sentry/fsimpl/verity/BUILD +++ b/pkg/sentry/fsimpl/verity/BUILD @@ -13,6 +13,7 @@ go_library( "//pkg/abi/linux", "//pkg/context", "//pkg/fspath", + "//pkg/marshal/primitive", "//pkg/merkletree", "//pkg/sentry/arch", "//pkg/sentry/fs/lock", diff --git a/pkg/sentry/fsimpl/verity/verity.go b/pkg/sentry/fsimpl/verity/verity.go index e352ce197..249cc1341 100644 --- a/pkg/sentry/fsimpl/verity/verity.go +++ b/pkg/sentry/fsimpl/verity/verity.go @@ -28,6 +28,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/merkletree" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -598,7 +599,10 @@ func (fd *fileDescription) getFlags(ctx context.Context, uio usermem.IO, args ar if len(fd.d.rootHash) != 0 { f |= linux.FS_VERITY_FL } - _, err := kernel.TaskFromContext(ctx).CopyOut(args[2].Pointer(), f) + + t := kernel.TaskFromContext(ctx) + addr := args[2].Pointer() + _, err := primitive.CopyInt32Out(t, addr, f) return 0, err } diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD index 66cf4f207..a43c549f1 100644 --- a/pkg/sentry/kernel/BUILD +++ b/pkg/sentry/kernel/BUILD @@ -197,6 +197,7 @@ go_library( "gvisor.dev/gvisor/pkg/sentry/device", "gvisor.dev/gvisor/pkg/tcpip", ], + marshal = True, visibility = ["//:sandbox"], deps = [ ":uncaught_signal_go_proto", @@ -213,6 +214,7 @@ go_library( "//pkg/fspath", "//pkg/log", "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/metric", "//pkg/refs", "//pkg/refs_vfs2", diff --git a/pkg/sentry/kernel/auth/BUILD b/pkg/sentry/kernel/auth/BUILD index 2bc49483a..869e49ebc 100644 --- a/pkg/sentry/kernel/auth/BUILD +++ b/pkg/sentry/kernel/auth/BUILD @@ -57,6 +57,7 @@ go_library( "id_map_set.go", "user_namespace.go", ], + marshal = True, visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/kernel/auth/id.go b/pkg/sentry/kernel/auth/id.go index 0a58ba17c..4c32ee703 100644 --- a/pkg/sentry/kernel/auth/id.go +++ b/pkg/sentry/kernel/auth/id.go @@ -19,9 +19,13 @@ import ( ) // UID is a user ID in an unspecified user namespace. +// +// +marshal type UID uint32 // GID is a group ID in an unspecified user namespace. +// +// +marshal slice:GIDSlice type GID uint32 // In the root user namespace, user/group IDs have a 1-to-1 relationship with diff --git a/pkg/sentry/kernel/ptrace.go b/pkg/sentry/kernel/ptrace.go index 50df179c3..1145faf13 100644 --- a/pkg/sentry/kernel/ptrace.go +++ b/pkg/sentry/kernel/ptrace.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/mm" "gvisor.dev/gvisor/pkg/syserror" @@ -999,18 +1000,15 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { // at the address specified by the data parameter, and the return value // is the error flag." - ptrace(2) word := t.Arch().Native(0) - if _, err := usermem.CopyObjectIn(t, target.MemoryManager(), addr, word, usermem.IOOpts{ - IgnorePermissions: true, - }); err != nil { + if _, err := word.CopyIn(target.AsCopyContext(usermem.IOOpts{IgnorePermissions: true}), addr); err != nil { return err } - _, err := t.CopyOut(data, word) + _, err := word.CopyOut(t, data) return err case linux.PTRACE_POKETEXT, linux.PTRACE_POKEDATA: - _, err := usermem.CopyObjectOut(t, target.MemoryManager(), addr, t.Arch().Native(uintptr(data)), usermem.IOOpts{ - IgnorePermissions: true, - }) + word := t.Arch().Native(uintptr(data)) + _, err := word.CopyOut(target.AsCopyContext(usermem.IOOpts{IgnorePermissions: true}), addr) return err case linux.PTRACE_GETREGSET: @@ -1078,12 +1076,12 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if target.ptraceSiginfo == nil { return syserror.EINVAL } - _, err := t.CopyOut(data, target.ptraceSiginfo) + _, err := target.ptraceSiginfo.CopyOut(t, data) return err case linux.PTRACE_SETSIGINFO: var info arch.SignalInfo - if _, err := t.CopyIn(data, &info); err != nil { + if _, err := info.CopyIn(t, data); err != nil { return err } t.tg.pidns.owner.mu.RLock() @@ -1098,7 +1096,8 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { if addr != linux.SignalSetSize { return syserror.EINVAL } - _, err := t.CopyOut(data, target.SignalMask()) + mask := target.SignalMask() + _, err := mask.CopyOut(t, data) return err case linux.PTRACE_SETSIGMASK: @@ -1106,7 +1105,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { return syserror.EINVAL } var mask linux.SignalSet - if _, err := t.CopyIn(data, &mask); err != nil { + if _, err := mask.CopyIn(t, data); err != nil { return err } // The target's task goroutine is stopped, so this is safe: @@ -1121,7 +1120,7 @@ func (t *Task) Ptrace(req int64, pid ThreadID, addr, data usermem.Addr) error { case linux.PTRACE_GETEVENTMSG: t.tg.pidns.owner.mu.RLock() defer t.tg.pidns.owner.mu.RUnlock() - _, err := t.CopyOut(usermem.Addr(data), target.ptraceEventMsg) + _, err := primitive.CopyUint64Out(t, usermem.Addr(data), target.ptraceEventMsg) return err // PEEKSIGINFO is unimplemented but seems to have no users anywhere. diff --git a/pkg/sentry/kernel/ptrace_amd64.go b/pkg/sentry/kernel/ptrace_amd64.go index cef1276ec..609ad3941 100644 --- a/pkg/sentry/kernel/ptrace_amd64.go +++ b/pkg/sentry/kernel/ptrace_amd64.go @@ -30,7 +30,7 @@ func (t *Task) ptraceArch(target *Task, req int64, addr, data usermem.Addr) erro if err != nil { return err } - _, err = t.CopyOut(data, n) + _, err = n.CopyOut(t, data) return err case linux.PTRACE_POKEUSR: // aka PTRACE_POKEUSER diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go index 9d7a9128f..fce1064a7 100644 --- a/pkg/sentry/kernel/task_clone.go +++ b/pkg/sentry/kernel/task_clone.go @@ -341,12 +341,12 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) { nt.SetClearTID(opts.ChildTID) } if opts.ChildSetTID { - // Can't use Task.CopyOut, which assumes AddressSpaceActive. - usermem.CopyObjectOut(t, nt.MemoryManager(), opts.ChildTID, nt.ThreadID(), usermem.IOOpts{}) + ctid := nt.ThreadID() + ctid.CopyOut(nt.AsCopyContext(usermem.IOOpts{AddressSpaceActive: false}), opts.ChildTID) } ntid := t.tg.pidns.IDOfTask(nt) if opts.ParentSetTID { - t.CopyOut(opts.ParentTID, ntid) + ntid.CopyOut(t, opts.ParentTID) } kind := ptraceCloneKindClone diff --git a/pkg/sentry/kernel/task_exit.go b/pkg/sentry/kernel/task_exit.go index b76f7f503..b400a8b41 100644 --- a/pkg/sentry/kernel/task_exit.go +++ b/pkg/sentry/kernel/task_exit.go @@ -248,7 +248,8 @@ func (*runExitMain) execute(t *Task) taskRunState { signaled := t.tg.exiting && t.tg.exitStatus.Signaled() t.tg.signalHandlers.mu.Unlock() if !signaled { - if _, err := t.CopyOut(t.cleartid, ThreadID(0)); err == nil { + zero := ThreadID(0) + if _, err := zero.CopyOut(t, t.cleartid); err == nil { t.Futex().Wake(t, t.cleartid, false, ^uint32(0), 1) } // If the CopyOut fails, there's nothing we can do. diff --git a/pkg/sentry/kernel/task_futex.go b/pkg/sentry/kernel/task_futex.go index 4b535c949..c80391475 100644 --- a/pkg/sentry/kernel/task_futex.go +++ b/pkg/sentry/kernel/task_futex.go @@ -16,6 +16,7 @@ package kernel import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/kernel/futex" "gvisor.dev/gvisor/pkg/usermem" ) @@ -87,7 +88,7 @@ func (t *Task) exitRobustList() { return } - next := rl.List + next := primitive.Uint64(rl.List) done := 0 var pendingLockAddr usermem.Addr if rl.ListOpPending != 0 { @@ -99,12 +100,12 @@ func (t *Task) exitRobustList() { // We traverse to the next element of the list before we // actually wake anything. This prevents the race where waking // this futex causes a modification of the list. - thisLockAddr := usermem.Addr(next + rl.FutexOffset) + thisLockAddr := usermem.Addr(uint64(next) + rl.FutexOffset) // Try to decode the next element in the list before waking the // current futex. But don't check the error until after we've // woken the current futex. Linux does it in this order too - _, nextErr := t.CopyIn(usermem.Addr(next), &next) + _, nextErr := next.CopyIn(t, usermem.Addr(next)) // Wakeup the current futex if it's not pending. if thisLockAddr != pendingLockAddr { diff --git a/pkg/sentry/kernel/task_run.go b/pkg/sentry/kernel/task_run.go index aa3a573c0..8dc3fec90 100644 --- a/pkg/sentry/kernel/task_run.go +++ b/pkg/sentry/kernel/task_run.go @@ -141,7 +141,7 @@ func (*runApp) handleCPUIDInstruction(t *Task) error { region := trace.StartRegion(t.traceContext, cpuidRegion) expected := arch.CPUIDInstruction[:] found := make([]byte, len(expected)) - _, err := t.CopyIn(usermem.Addr(t.Arch().IP()), &found) + _, err := t.CopyInBytes(usermem.Addr(t.Arch().IP()), found) if err == nil && bytes.Equal(expected, found) { // Skip the cpuid instruction. t.Arch().CPUIDEmulate(t) diff --git a/pkg/sentry/kernel/task_syscall.go b/pkg/sentry/kernel/task_syscall.go index 2dbf86547..0141459e7 100644 --- a/pkg/sentry/kernel/task_syscall.go +++ b/pkg/sentry/kernel/task_syscall.go @@ -22,6 +22,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bits" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/memmap" @@ -287,7 +288,7 @@ func (t *Task) doVsyscall(addr usermem.Addr, sysno uintptr) taskRunState { // Grab the caller up front, to make sure there's a sensible stack. caller := t.Arch().Native(uintptr(0)) - if _, err := t.CopyIn(usermem.Addr(t.Arch().Stack()), caller); err != nil { + if _, err := caller.CopyIn(t, usermem.Addr(t.Arch().Stack())); err != nil { t.Debugf("vsyscall %d: error reading return address from stack: %v", sysno, err) t.forceSignal(linux.SIGSEGV, false /* unconditional */) t.SendSignal(SignalInfoPriv(linux.SIGSEGV)) @@ -323,7 +324,7 @@ func (t *Task) doVsyscall(addr usermem.Addr, sysno uintptr) taskRunState { type runVsyscallAfterPtraceEventSeccomp struct { addr usermem.Addr sysno uintptr - caller interface{} + caller marshal.Marshallable } func (r *runVsyscallAfterPtraceEventSeccomp) execute(t *Task) taskRunState { @@ -346,7 +347,7 @@ func (r *runVsyscallAfterPtraceEventSeccomp) execute(t *Task) taskRunState { return t.doVsyscallInvoke(sysno, t.Arch().SyscallArgs(), r.caller) } -func (t *Task) doVsyscallInvoke(sysno uintptr, args arch.SyscallArguments, caller interface{}) taskRunState { +func (t *Task) doVsyscallInvoke(sysno uintptr, args arch.SyscallArguments, caller marshal.Marshallable) taskRunState { rval, ctrl, err := t.executeSyscall(sysno, args) if ctrl != nil { t.Debugf("vsyscall %d, caller %x: syscall control: %v", sysno, t.Arch().Value(caller), ctrl) diff --git a/pkg/sentry/kernel/task_usermem.go b/pkg/sentry/kernel/task_usermem.go index 0cb86e390..14d765af1 100644 --- a/pkg/sentry/kernel/task_usermem.go +++ b/pkg/sentry/kernel/task_usermem.go @@ -43,17 +43,6 @@ func (t *Task) Deactivate() { } } -// CopyIn copies a fixed-size value or slice of fixed-size values in from the -// task's memory. The copy will fail with syscall.EFAULT if it traverses user -// memory that is unmapped or not readable by the user. -// -// This Task's AddressSpace must be active. -func (t *Task) CopyIn(addr usermem.Addr, dst interface{}) (int, error) { - return usermem.CopyObjectIn(t, t.MemoryManager(), addr, dst, usermem.IOOpts{ - AddressSpaceActive: true, - }) -} - // CopyInBytes is a fast version of CopyIn if the caller can serialize the // data without reflection and pass in a byte slice. // @@ -64,17 +53,6 @@ func (t *Task) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) { }) } -// CopyOut copies a fixed-size value or slice of fixed-size values out to the -// task's memory. The copy will fail with syscall.EFAULT if it traverses user -// memory that is unmapped or not writeable by the user. -// -// This Task's AddressSpace must be active. -func (t *Task) CopyOut(addr usermem.Addr, src interface{}) (int, error) { - return usermem.CopyObjectOut(t, t.MemoryManager(), addr, src, usermem.IOOpts{ - AddressSpaceActive: true, - }) -} - // CopyOutBytes is a fast version of CopyOut if the caller can serialize the // data without reflection and pass in a byte slice. // @@ -114,7 +92,7 @@ func (t *Task) CopyInVector(addr usermem.Addr, maxElemSize, maxTotalSize int) ([ var v []string for { argAddr := t.Arch().Native(0) - if _, err := t.CopyIn(addr, argAddr); err != nil { + if _, err := argAddr.CopyIn(t, addr); err != nil { return v, err } if t.Arch().Value(argAddr) == 0 { @@ -302,29 +280,29 @@ func (t *Task) IovecsIOSequence(addr usermem.Addr, iovcnt int, opts usermem.IOOp }, nil } -// CopyContextWithOpts wraps a task to allow copying memory to and from the +// CopyContext wraps a task to allow copying memory to and from the // task memory with user specified usermem.IOOpts. -type CopyContextWithOpts struct { +type CopyContext struct { *Task opts usermem.IOOpts } -// AsCopyContextWithOpts wraps the task and returns it as CopyContextWithOpts. -func (t *Task) AsCopyContextWithOpts(opts usermem.IOOpts) *CopyContextWithOpts { - return &CopyContextWithOpts{t, opts} +// AsCopyContext wraps the task and returns it as CopyContext. +func (t *Task) AsCopyContext(opts usermem.IOOpts) *CopyContext { + return &CopyContext{t, opts} } // CopyInString copies a string in from the task's memory. -func (t *CopyContextWithOpts) CopyInString(addr usermem.Addr, maxLen int) (string, error) { +func (t *CopyContext) CopyInString(addr usermem.Addr, maxLen int) (string, error) { return usermem.CopyStringIn(t, t.MemoryManager(), addr, maxLen, t.opts) } // CopyInBytes copies task memory into dst from an IO context. -func (t *CopyContextWithOpts) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) { +func (t *CopyContext) CopyInBytes(addr usermem.Addr, dst []byte) (int, error) { return t.MemoryManager().CopyIn(t, addr, dst, t.opts) } // CopyOutBytes copies src into task memoryfrom an IO context. -func (t *CopyContextWithOpts) CopyOutBytes(addr usermem.Addr, src []byte) (int, error) { +func (t *CopyContext) CopyOutBytes(addr usermem.Addr, src []byte) (int, error) { return t.MemoryManager().CopyOut(t, addr, src, t.opts) } diff --git a/pkg/sentry/kernel/threads.go b/pkg/sentry/kernel/threads.go index 872e1a82d..5ae5906e8 100644 --- a/pkg/sentry/kernel/threads.go +++ b/pkg/sentry/kernel/threads.go @@ -36,6 +36,8 @@ import ( const TasksLimit = (1 << 16) // ThreadID is a generic thread identifier. +// +// +marshal type ThreadID int32 // String returns a decimal representation of the ThreadID. diff --git a/pkg/sentry/strace/BUILD b/pkg/sentry/strace/BUILD index 88d5db9fc..a920180d3 100644 --- a/pkg/sentry/strace/BUILD +++ b/pkg/sentry/strace/BUILD @@ -28,6 +28,7 @@ go_library( "//pkg/binary", "//pkg/bits", "//pkg/eventchannel", + "//pkg/marshal/primitive", "//pkg/seccomp", "//pkg/sentry/arch", "//pkg/sentry/kernel", diff --git a/pkg/sentry/strace/epoll.go b/pkg/sentry/strace/epoll.go index 5d51a7792..ae3b998c8 100644 --- a/pkg/sentry/strace/epoll.go +++ b/pkg/sentry/strace/epoll.go @@ -26,7 +26,7 @@ import ( func epollEvent(t *kernel.Task, eventAddr usermem.Addr) string { var e linux.EpollEvent - if _, err := t.CopyIn(eventAddr, &e); err != nil { + if _, err := e.CopyIn(t, eventAddr); err != nil { return fmt.Sprintf("%#x {error reading event: %v}", eventAddr, err) } var sb strings.Builder @@ -41,7 +41,7 @@ func epollEvents(t *kernel.Task, eventsAddr usermem.Addr, numEvents, maxBytes ui addr := eventsAddr for i := uint64(0); i < numEvents; i++ { var e linux.EpollEvent - if _, err := t.CopyIn(addr, &e); err != nil { + if _, err := e.CopyIn(t, addr); err != nil { fmt.Fprintf(&sb, "{error reading event at %#x: %v}", addr, err) continue } diff --git a/pkg/sentry/strace/socket.go b/pkg/sentry/strace/socket.go index 08e97e6c4..cc5f70cd4 100644 --- a/pkg/sentry/strace/socket.go +++ b/pkg/sentry/strace/socket.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/socket/netlink" "gvisor.dev/gvisor/pkg/sentry/socket/netstack" @@ -166,7 +167,7 @@ func cmsghdr(t *kernel.Task, addr usermem.Addr, length uint64, maxBytes uint64) } buf := make([]byte, length) - if _, err := t.CopyIn(addr, &buf); err != nil { + if _, err := t.CopyInBytes(addr, buf); err != nil { return fmt.Sprintf("%#x (error decoding control: %v)", addr, err) } @@ -302,7 +303,7 @@ func cmsghdr(t *kernel.Task, addr usermem.Addr, length uint64, maxBytes uint64) func msghdr(t *kernel.Task, addr usermem.Addr, printContent bool, maxBytes uint64) string { var msg slinux.MessageHeader64 - if err := slinux.CopyInMessageHeader64(t, addr, &msg); err != nil { + if _, err := msg.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding msghdr: %v)", addr, err) } s := fmt.Sprintf( @@ -380,9 +381,9 @@ func postSockAddr(t *kernel.Task, addr usermem.Addr, lengthPtr usermem.Addr) str func copySockLen(t *kernel.Task, addr usermem.Addr) (uint32, error) { // socklen_t is 32-bits. - var l uint32 - _, err := t.CopyIn(addr, &l) - return l, err + var l primitive.Uint32 + _, err := l.CopyIn(t, addr) + return uint32(l), err } func sockLenPointer(t *kernel.Task, addr usermem.Addr) string { @@ -436,22 +437,22 @@ func getSockOptVal(t *kernel.Task, level, optname uint64, optVal usermem.Addr, o func sockOptVal(t *kernel.Task, level, optname uint64, optVal usermem.Addr, optLen uint64, maximumBlobSize uint) string { switch optLen { case 1: - var v uint8 - _, err := t.CopyIn(optVal, &v) + var v primitive.Uint8 + _, err := v.CopyIn(t, optVal) if err != nil { return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) } return fmt.Sprintf("%#x {value=%v}", optVal, v) case 2: - var v uint16 - _, err := t.CopyIn(optVal, &v) + var v primitive.Uint16 + _, err := v.CopyIn(t, optVal) if err != nil { return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) } return fmt.Sprintf("%#x {value=%v}", optVal, v) case 4: - var v uint32 - _, err := t.CopyIn(optVal, &v) + var v primitive.Uint32 + _, err := v.CopyIn(t, optVal) if err != nil { return fmt.Sprintf("%#x {error reading optval: %v}", optVal, err) } diff --git a/pkg/sentry/strace/strace.go b/pkg/sentry/strace/strace.go index 87b239730..52281ccc2 100644 --- a/pkg/sentry/strace/strace.go +++ b/pkg/sentry/strace/strace.go @@ -21,13 +21,13 @@ import ( "fmt" "strconv" "strings" - "syscall" "time" "gvisor.dev/gvisor/pkg/abi" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/bits" "gvisor.dev/gvisor/pkg/eventchannel" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/seccomp" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -91,7 +91,7 @@ func iovecs(t *kernel.Task, addr usermem.Addr, iovcnt int, printContent bool, ma } b := make([]byte, size) - amt, err := t.CopyIn(ar.Start, b) + amt, err := t.CopyInBytes(ar.Start, b) if err != nil { iovs[i] = fmt.Sprintf("{base=%#x, len=%d, %q..., error decoding string: %v}", ar.Start, ar.Length(), b[:amt], err) continue @@ -118,7 +118,7 @@ func dump(t *kernel.Task, addr usermem.Addr, size uint, maximumBlobSize uint) st } b := make([]byte, size) - amt, err := t.CopyIn(addr, b) + amt, err := t.CopyInBytes(addr, b) if err != nil { return fmt.Sprintf("%#x (error decoding string: %s)", addr, err) } @@ -199,7 +199,7 @@ func fdVFS2(t *kernel.Task, fd int32) string { func fdpair(t *kernel.Task, addr usermem.Addr) string { var fds [2]int32 - _, err := t.CopyIn(addr, &fds) + _, err := primitive.CopyInt32SliceIn(t, addr, fds[:]) if err != nil { return fmt.Sprintf("%#x (error decoding fds: %s)", addr, err) } @@ -209,7 +209,7 @@ func fdpair(t *kernel.Task, addr usermem.Addr) string { func uname(t *kernel.Task, addr usermem.Addr) string { var u linux.UtsName - if _, err := t.CopyIn(addr, &u); err != nil { + if _, err := u.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding utsname: %s)", addr, err) } @@ -222,7 +222,7 @@ func utimensTimespec(t *kernel.Task, addr usermem.Addr) string { } var tim linux.Timespec - if _, err := t.CopyIn(addr, &tim); err != nil { + if _, err := tim.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding timespec: %s)", addr, err) } @@ -244,7 +244,7 @@ func timespec(t *kernel.Task, addr usermem.Addr) string { } var tim linux.Timespec - if _, err := t.CopyIn(addr, &tim); err != nil { + if _, err := tim.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding timespec: %s)", addr, err) } return fmt.Sprintf("%#x {sec=%v nsec=%v}", addr, tim.Sec, tim.Nsec) @@ -256,7 +256,7 @@ func timeval(t *kernel.Task, addr usermem.Addr) string { } var tim linux.Timeval - if _, err := t.CopyIn(addr, &tim); err != nil { + if _, err := tim.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding timeval: %s)", addr, err) } @@ -268,8 +268,8 @@ func utimbuf(t *kernel.Task, addr usermem.Addr) string { return "null" } - var utim syscall.Utimbuf - if _, err := t.CopyIn(addr, &utim); err != nil { + var utim linux.Utime + if _, err := utim.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding utimbuf: %s)", addr, err) } @@ -282,7 +282,7 @@ func stat(t *kernel.Task, addr usermem.Addr) string { } var stat linux.Stat - if _, err := t.CopyIn(addr, &stat); err != nil { + if _, err := stat.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding stat: %s)", addr, err) } return fmt.Sprintf("%#x {dev=%d, ino=%d, mode=%s, nlink=%d, uid=%d, gid=%d, rdev=%d, size=%d, blksize=%d, blocks=%d, atime=%s, mtime=%s, ctime=%s}", addr, stat.Dev, stat.Ino, linux.FileMode(stat.Mode), stat.Nlink, stat.UID, stat.GID, stat.Rdev, stat.Size, stat.Blksize, stat.Blocks, time.Unix(stat.ATime.Sec, stat.ATime.Nsec), time.Unix(stat.MTime.Sec, stat.MTime.Nsec), time.Unix(stat.CTime.Sec, stat.CTime.Nsec)) @@ -330,7 +330,7 @@ func rusage(t *kernel.Task, addr usermem.Addr) string { } var ru linux.Rusage - if _, err := t.CopyIn(addr, &ru); err != nil { + if _, err := ru.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding rusage: %s)", addr, err) } return fmt.Sprintf("%#x %+v", addr, ru) @@ -342,7 +342,7 @@ func capHeader(t *kernel.Task, addr usermem.Addr) string { } var hdr linux.CapUserHeader - if _, err := t.CopyIn(addr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, addr); err != nil { return fmt.Sprintf("%#x (error decoding header: %s)", addr, err) } @@ -367,7 +367,7 @@ func capData(t *kernel.Task, hdrAddr, dataAddr usermem.Addr) string { } var hdr linux.CapUserHeader - if _, err := t.CopyIn(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, hdrAddr); err != nil { return fmt.Sprintf("%#x (error decoding header: %v)", dataAddr, err) } @@ -376,7 +376,7 @@ func capData(t *kernel.Task, hdrAddr, dataAddr usermem.Addr) string { switch hdr.Version { case linux.LINUX_CAPABILITY_VERSION_1: var data linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := data.CopyIn(t, dataAddr); err != nil { return fmt.Sprintf("%#x (error decoding data: %v)", dataAddr, err) } p = uint64(data.Permitted) @@ -384,7 +384,7 @@ func capData(t *kernel.Task, hdrAddr, dataAddr usermem.Addr) string { e = uint64(data.Effective) case linux.LINUX_CAPABILITY_VERSION_2, linux.LINUX_CAPABILITY_VERSION_3: var data [2]linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := linux.CopyCapUserDataSliceIn(t, dataAddr, data[:]); err != nil { return fmt.Sprintf("%#x (error decoding data: %v)", dataAddr, err) } p = uint64(data[0].Permitted) | (uint64(data[1].Permitted) << 32) diff --git a/pkg/sentry/syscalls/linux/BUILD b/pkg/sentry/syscalls/linux/BUILD index 86f224c86..75752b2e6 100644 --- a/pkg/sentry/syscalls/linux/BUILD +++ b/pkg/sentry/syscalls/linux/BUILD @@ -56,6 +56,7 @@ go_library( "sys_xattr.go", "timespec.go", ], + marshal = True, visibility = ["//:sandbox"], deps = [ "//pkg/abi", diff --git a/pkg/sentry/syscalls/linux/sys_aio.go b/pkg/sentry/syscalls/linux/sys_aio.go index e9d64dec5..0bf313a13 100644 --- a/pkg/sentry/syscalls/linux/sys_aio.go +++ b/pkg/sentry/syscalls/linux/sys_aio.go @@ -17,6 +17,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -36,7 +37,7 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca // // The context pointer _must_ be zero initially. var idIn uint64 - if _, err := t.CopyIn(idAddr, &idIn); err != nil { + if _, err := primitive.CopyUint64In(t, idAddr, &idIn); err != nil { return 0, nil, err } if idIn != 0 { @@ -49,7 +50,7 @@ func IoSetup(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca } // Copy out the new ID. - if _, err := t.CopyOut(idAddr, &id); err != nil { + if _, err := primitive.CopyUint64Out(t, idAddr, id); err != nil { t.MemoryManager().DestroyAIOContext(t, id) return 0, nil, err } @@ -142,7 +143,7 @@ func IoGetevents(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S ev := v.(*linux.IOEvent) // Copy out the result. - if _, err := t.CopyOut(eventsAddr, ev); err != nil { + if _, err := ev.CopyOut(t, eventsAddr); err != nil { if count > 0 { return uintptr(count), nil, nil } @@ -338,21 +339,27 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } for i := int32(0); i < nrEvents; i++ { - // Copy in the address. - cbAddrNative := t.Arch().Native(0) - if _, err := t.CopyIn(addr, cbAddrNative); err != nil { - if i > 0 { - // Some successful. - return uintptr(i), nil, nil + // Copy in the callback address. + var cbAddr usermem.Addr + switch t.Arch().Width() { + case 8: + var cbAddrP primitive.Uint64 + if _, err := cbAddrP.CopyIn(t, addr); err != nil { + if i > 0 { + // Some successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err } - // Nothing done. - return 0, nil, err + cbAddr = usermem.Addr(cbAddrP) + default: + return 0, nil, syserror.ENOSYS } // Copy in this callback. var cb linux.IOCallback - cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) - if _, err := t.CopyIn(cbAddr, &cb); err != nil { + if _, err := cb.CopyIn(t, cbAddr); err != nil { if i > 0 { // Some have been successful. diff --git a/pkg/sentry/syscalls/linux/sys_capability.go b/pkg/sentry/syscalls/linux/sys_capability.go index adf5ea5f2..d3b85e11b 100644 --- a/pkg/sentry/syscalls/linux/sys_capability.go +++ b/pkg/sentry/syscalls/linux/sys_capability.go @@ -45,7 +45,7 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal dataAddr := args[1].Pointer() var hdr linux.CapUserHeader - if _, err := t.CopyIn(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, hdrAddr); err != nil { return 0, nil, err } // hdr.Pid doesn't need to be valid if this capget() is a "version probe" @@ -65,7 +65,7 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal Permitted: uint32(p), Inheritable: uint32(i), } - _, err = t.CopyOut(dataAddr, &data) + _, err = data.CopyOut(t, dataAddr) return 0, nil, err case linux.LINUX_CAPABILITY_VERSION_2, linux.LINUX_CAPABILITY_VERSION_3: @@ -88,12 +88,12 @@ func Capget(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal Inheritable: uint32(i >> 32), }, } - _, err = t.CopyOut(dataAddr, &data) + _, err = linux.CopyCapUserDataSliceOut(t, dataAddr, data[:]) return 0, nil, err default: hdr.Version = linux.HighestCapabilityVersion - if _, err := t.CopyOut(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyOut(t, hdrAddr); err != nil { return 0, nil, err } if dataAddr != 0 { @@ -109,7 +109,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal dataAddr := args[1].Pointer() var hdr linux.CapUserHeader - if _, err := t.CopyIn(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyIn(t, hdrAddr); err != nil { return 0, nil, err } switch hdr.Version { @@ -118,7 +118,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, syserror.EPERM } var data linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := data.CopyIn(t, dataAddr); err != nil { return 0, nil, err } p := auth.CapabilitySet(data.Permitted) & auth.AllCapabilities @@ -131,7 +131,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return 0, nil, syserror.EPERM } var data [2]linux.CapUserData - if _, err := t.CopyIn(dataAddr, &data); err != nil { + if _, err := linux.CopyCapUserDataSliceIn(t, dataAddr, data[:]); err != nil { return 0, nil, err } p := (auth.CapabilitySet(data[0].Permitted) | (auth.CapabilitySet(data[1].Permitted) << 32)) & auth.AllCapabilities @@ -141,7 +141,7 @@ func Capset(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal default: hdr.Version = linux.HighestCapabilityVersion - if _, err := t.CopyOut(hdrAddr, &hdr); err != nil { + if _, err := hdr.CopyOut(t, hdrAddr); err != nil { return 0, nil, err } return 0, nil, syserror.EINVAL diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go index 07c77e442..98331eb3c 100644 --- a/pkg/sentry/syscalls/linux/sys_file.go +++ b/pkg/sentry/syscalls/linux/sys_file.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/lock" @@ -613,7 +614,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIONBIO: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.Flags() @@ -627,7 +628,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOASYNC: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.Flags() @@ -641,15 +642,14 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOSETOWN, linux.SIOCSPGRP: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } fSetOwn(t, file, set) return 0, nil, nil case linux.FIOGETOWN, linux.SIOCGPGRP: - who := fGetOwn(t, file) - _, err := t.CopyOut(args[2].Pointer(), &who) + _, err := primitive.CopyInt32Out(t, args[2].Pointer(), fGetOwn(t, file)) return 0, nil, err default: @@ -694,7 +694,7 @@ func Getcwd(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } // Top it off with a terminator. - _, err = t.CopyOut(addr+usermem.Addr(bytes), []byte("\x00")) + _, err = t.CopyOutBytes(addr+usermem.Addr(bytes), []byte("\x00")) return uintptr(bytes + 1), nil, err } @@ -962,7 +962,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Copy in the lock request. flockAddr := args[2].Pointer() var flock linux.Flock - if _, err := t.CopyIn(flockAddr, &flock); err != nil { + if _, err := flock.CopyIn(t, flockAddr); err != nil { return 0, nil, err } @@ -1052,12 +1052,12 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.F_GETOWN_EX: addr := args[2].Pointer() owner := fGetOwnEx(t, file) - _, err := t.CopyOut(addr, &owner) + _, err := owner.CopyOut(t, addr) return 0, nil, err case linux.F_SETOWN_EX: addr := args[2].Pointer() var owner linux.FOwnerEx - _, err := t.CopyIn(addr, &owner) + _, err := owner.CopyIn(t, addr) if err != nil { return 0, nil, err } @@ -1922,7 +1922,7 @@ func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times linux.Utime - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := times.CopyIn(t, timesAddr); err != nil { return 0, nil, err } ts = fs.TimeSpec{ @@ -1942,7 +1942,7 @@ func Utimes(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } ts = fs.TimeSpec{ @@ -1970,7 +1970,7 @@ func Utimensat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timespec - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimespecSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } if !timespecIsValid(times[0]) || !timespecIsValid(times[1]) { @@ -2004,7 +2004,7 @@ func Futimesat(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ts := defaultSetToSystemTimeSpec() if timesAddr != 0 { var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return 0, nil, err } if times[0].Usec >= 1e6 || times[0].Usec < 0 || diff --git a/pkg/sentry/syscalls/linux/sys_futex.go b/pkg/sentry/syscalls/linux/sys_futex.go index 12b2fa690..f39ce0639 100644 --- a/pkg/sentry/syscalls/linux/sys_futex.go +++ b/pkg/sentry/syscalls/linux/sys_futex.go @@ -306,8 +306,8 @@ func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel // Despite the syscall using the name 'pid' for this variable, it is // very much a tid. tid := args[0].Int() - head := args[1].Pointer() - size := args[2].Pointer() + headAddr := args[1].Pointer() + sizeAddr := args[2].Pointer() if tid < 0 { return 0, nil, syserror.EINVAL @@ -321,12 +321,16 @@ func GetRobustList(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel } // Copy out head pointer. - if _, err := t.CopyOut(head, uint64(ot.GetRobustList())); err != nil { + head := t.Arch().Native(uintptr(ot.GetRobustList())) + if _, err := head.CopyOut(t, headAddr); err != nil { return 0, nil, err } - // Copy out size, which is a constant. - if _, err := t.CopyOut(size, uint64(linux.SizeOfRobustListHead)); err != nil { + // Copy out size, which is a constant. Note that while size isn't + // an address, it is defined as the arch-dependent size_t, so it + // needs to be converted to a native-sized int. + size := t.Arch().Native(uintptr(linux.SizeOfRobustListHead)) + if _, err := size.CopyOut(t, sizeAddr); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_getdents.go b/pkg/sentry/syscalls/linux/sys_getdents.go index 59004cefe..b25f7d881 100644 --- a/pkg/sentry/syscalls/linux/sys_getdents.go +++ b/pkg/sentry/syscalls/linux/sys_getdents.go @@ -19,7 +19,6 @@ import ( "io" "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" @@ -93,19 +92,23 @@ func getdents(t *kernel.Task, fd int32, addr usermem.Addr, size int, f func(*dir } } -// oldDirentHdr is a fixed sized header matching the fixed size -// fields found in the old linux dirent struct. +// oldDirentHdr is a fixed sized header matching the fixed size fields found in +// the old linux dirent struct. +// +// +marshal type oldDirentHdr struct { Ino uint64 Off uint64 - Reclen uint16 + Reclen uint16 `marshal:"unaligned"` // Struct ends mid-word. } -// direntHdr is a fixed sized header matching the fixed size -// fields found in the new linux dirent struct. +// direntHdr is a fixed sized header matching the fixed size fields found in the +// new linux dirent struct. +// +// +marshal type direntHdr struct { OldHdr oldDirentHdr - Typ uint8 + Typ uint8 `marshal:"unaligned"` // Struct ends mid-word. } // dirent contains the data pointed to by a new linux dirent struct. @@ -134,20 +137,20 @@ func newDirent(width uint, name string, attr fs.DentAttr, offset uint64) *dirent // the old linux dirent format. func smallestDirent(a arch.Context) uint { d := dirent{} - return uint(binary.Size(d.Hdr.OldHdr)) + a.Width() + 1 + return uint(d.Hdr.OldHdr.SizeBytes()) + a.Width() + 1 } // smallestDirent64 returns the size of the smallest possible dirent using // the new linux dirent format. func smallestDirent64(a arch.Context) uint { d := dirent{} - return uint(binary.Size(d.Hdr)) + a.Width() + return uint(d.Hdr.SizeBytes()) + a.Width() } // padRec pads the name field until the rec length is a multiple of the width, // which must be a power of 2. It returns the padded rec length. func (d *dirent) padRec(width int) uint16 { - a := int(binary.Size(d.Hdr)) + len(d.Name) + a := d.Hdr.SizeBytes() + len(d.Name) r := (a + width) &^ (width - 1) padding := r - a d.Name = append(d.Name, make([]byte, padding)...) @@ -157,7 +160,7 @@ func (d *dirent) padRec(width int) uint16 { // Serialize64 serializes a Dirent struct to a byte slice, keeping the new // linux dirent format. Returns the number of bytes serialized or an error. func (d *dirent) Serialize64(w io.Writer) (int, error) { - n1, err := w.Write(binary.Marshal(nil, usermem.ByteOrder, d.Hdr)) + n1, err := d.Hdr.WriteTo(w) if err != nil { return 0, err } @@ -165,14 +168,14 @@ func (d *dirent) Serialize64(w io.Writer) (int, error) { if err != nil { return 0, err } - return n1 + n2, nil + return int(n1) + n2, nil } // Serialize serializes a Dirent struct to a byte slice, using the old linux // dirent format. // Returns the number of bytes serialized or an error. func (d *dirent) Serialize(w io.Writer) (int, error) { - n1, err := w.Write(binary.Marshal(nil, usermem.ByteOrder, d.Hdr.OldHdr)) + n1, err := d.Hdr.OldHdr.WriteTo(w) if err != nil { return 0, err } @@ -184,7 +187,7 @@ func (d *dirent) Serialize(w io.Writer) (int, error) { if err != nil { return 0, err } - return n1 + n2 + n3, nil + return int(n1) + n2 + n3, nil } // direntSerializer implements fs.InodeOperationsInfoSerializer, serializing dirents to an diff --git a/pkg/sentry/syscalls/linux/sys_identity.go b/pkg/sentry/syscalls/linux/sys_identity.go index 715ac45e6..a29d307e5 100644 --- a/pkg/sentry/syscalls/linux/sys_identity.go +++ b/pkg/sentry/syscalls/linux/sys_identity.go @@ -49,13 +49,13 @@ func Getresuid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys ruid := c.RealKUID.In(c.UserNamespace).OrOverflow() euid := c.EffectiveKUID.In(c.UserNamespace).OrOverflow() suid := c.SavedKUID.In(c.UserNamespace).OrOverflow() - if _, err := t.CopyOut(ruidAddr, ruid); err != nil { + if _, err := ruid.CopyOut(t, ruidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(euidAddr, euid); err != nil { + if _, err := euid.CopyOut(t, euidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(suidAddr, suid); err != nil { + if _, err := suid.CopyOut(t, suidAddr); err != nil { return 0, nil, err } return 0, nil, nil @@ -84,13 +84,13 @@ func Getresgid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys rgid := c.RealKGID.In(c.UserNamespace).OrOverflow() egid := c.EffectiveKGID.In(c.UserNamespace).OrOverflow() sgid := c.SavedKGID.In(c.UserNamespace).OrOverflow() - if _, err := t.CopyOut(rgidAddr, rgid); err != nil { + if _, err := rgid.CopyOut(t, rgidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(egidAddr, egid); err != nil { + if _, err := egid.CopyOut(t, egidAddr); err != nil { return 0, nil, err } - if _, err := t.CopyOut(sgidAddr, sgid); err != nil { + if _, err := sgid.CopyOut(t, sgidAddr); err != nil { return 0, nil, err } return 0, nil, nil @@ -157,7 +157,7 @@ func Getgroups(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys for i, kgid := range kgids { gids[i] = kgid.In(t.UserNamespace()).OrOverflow() } - if _, err := t.CopyOut(args[1].Pointer(), gids); err != nil { + if _, err := auth.CopyGIDSliceOut(t, args[1].Pointer(), gids); err != nil { return 0, nil, err } return uintptr(len(gids)), nil, nil @@ -173,7 +173,7 @@ func Setgroups(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, t.SetExtraGIDs(nil) } gids := make([]auth.GID, size) - if _, err := t.CopyIn(args[1].Pointer(), &gids); err != nil { + if _, err := auth.CopyGIDSliceIn(t, args[1].Pointer(), gids); err != nil { return 0, nil, err } return 0, nil, t.SetExtraGIDs(gids) diff --git a/pkg/sentry/syscalls/linux/sys_mmap.go b/pkg/sentry/syscalls/linux/sys_mmap.go index d0109baa4..8ab062bca 100644 --- a/pkg/sentry/syscalls/linux/sys_mmap.go +++ b/pkg/sentry/syscalls/linux/sys_mmap.go @@ -239,7 +239,7 @@ func Mincore(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca return 0, nil, syserror.ENOMEM } resident := bytes.Repeat([]byte{1}, int(mapped/usermem.PageSize)) - _, err := t.CopyOut(vec, resident) + _, err := t.CopyOutBytes(vec, resident) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_pipe.go b/pkg/sentry/syscalls/linux/sys_pipe.go index c55beb39b..849a47476 100644 --- a/pkg/sentry/syscalls/linux/sys_pipe.go +++ b/pkg/sentry/syscalls/linux/sys_pipe.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -46,7 +47,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags uint) (uintptr, error) { return 0, err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(t, fd); file != nil { file.DecRef(t) diff --git a/pkg/sentry/syscalls/linux/sys_poll.go b/pkg/sentry/syscalls/linux/sys_poll.go index 789e2ed5b..254f4c9f9 100644 --- a/pkg/sentry/syscalls/linux/sys_poll.go +++ b/pkg/sentry/syscalls/linux/sys_poll.go @@ -162,7 +162,7 @@ func CopyInPollFDs(t *kernel.Task, addr usermem.Addr, nfds uint) ([]linux.PollFD pfd := make([]linux.PollFD, nfds) if nfds > 0 { - if _, err := t.CopyIn(addr, &pfd); err != nil { + if _, err := linux.CopyPollFDSliceIn(t, addr, pfd); err != nil { return nil, err } } @@ -189,7 +189,7 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) // The poll entries are copied out regardless of whether // any are set or not. This aligns with the Linux behavior. if nfds > 0 && err == nil { - if _, err := t.CopyOut(addr, pfd); err != nil { + if _, err := linux.CopyPollFDSliceOut(t, addr, pfd); err != nil { return remainingTimeout, 0, err } } @@ -202,7 +202,7 @@ func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialBy set := make([]byte, nBytes) if addr != 0 { - if _, err := t.CopyIn(addr, &set); err != nil { + if _, err := t.CopyInBytes(addr, set); err != nil { return nil, err } // If we only use part of the last byte, mask out the extraneous bits. @@ -329,19 +329,19 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Copy updated vectors back. if readFDs != 0 { - if _, err := t.CopyOut(readFDs, r); err != nil { + if _, err := t.CopyOutBytes(readFDs, r); err != nil { return 0, err } } if writeFDs != 0 { - if _, err := t.CopyOut(writeFDs, w); err != nil { + if _, err := t.CopyOutBytes(writeFDs, w); err != nil { return 0, err } } if exceptFDs != 0 { - if _, err := t.CopyOut(exceptFDs, e); err != nil { + if _, err := t.CopyOutBytes(exceptFDs, e); err != nil { return 0, err } } diff --git a/pkg/sentry/syscalls/linux/sys_prctl.go b/pkg/sentry/syscalls/linux/sys_prctl.go index 64a725296..a892d2c62 100644 --- a/pkg/sentry/syscalls/linux/sys_prctl.go +++ b/pkg/sentry/syscalls/linux/sys_prctl.go @@ -18,6 +18,7 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -43,7 +44,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall return 0, nil, nil case linux.PR_GET_PDEATHSIG: - _, err := t.CopyOut(args[1].Pointer(), int32(t.ParentDeathSignal())) + _, err := primitive.CopyInt32Out(t, args[1].Pointer(), int32(t.ParentDeathSignal())) return 0, nil, err case linux.PR_GET_DUMPABLE: @@ -110,7 +111,7 @@ func Prctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall buf[len] = 0 len++ } - _, err := t.CopyOut(addr, buf[:len]) + _, err := t.CopyOutBytes(addr, buf[:len]) if err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_rlimit.go b/pkg/sentry/syscalls/linux/sys_rlimit.go index d5d5b6959..309c183a3 100644 --- a/pkg/sentry/syscalls/linux/sys_rlimit.go +++ b/pkg/sentry/syscalls/linux/sys_rlimit.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/limits" @@ -26,17 +27,13 @@ import ( // rlimit describes an implementation of 'struct rlimit', which may vary from // system-to-system. type rlimit interface { + marshal.Marshallable + // toLimit converts an rlimit to a limits.Limit. toLimit() *limits.Limit // fromLimit converts a limits.Limit to an rlimit. fromLimit(lim limits.Limit) - - // copyIn copies an rlimit from the untrusted app to the kernel. - copyIn(t *kernel.Task, addr usermem.Addr) error - - // copyOut copies an rlimit from the kernel to the untrusted app. - copyOut(t *kernel.Task, addr usermem.Addr) error } // newRlimit returns the appropriate rlimit type for 'struct rlimit' on this system. @@ -50,6 +47,7 @@ func newRlimit(t *kernel.Task) (rlimit, error) { } } +// +marshal type rlimit64 struct { Cur uint64 Max uint64 @@ -70,12 +68,12 @@ func (r *rlimit64) fromLimit(lim limits.Limit) { } func (r *rlimit64) copyIn(t *kernel.Task, addr usermem.Addr) error { - _, err := t.CopyIn(addr, r) + _, err := r.CopyIn(t, addr) return err } func (r *rlimit64) copyOut(t *kernel.Task, addr usermem.Addr) error { - _, err := t.CopyOut(addr, *r) + _, err := r.CopyOut(t, addr) return err } @@ -140,7 +138,8 @@ func Getrlimit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys return 0, nil, err } rlim.fromLimit(lim) - return 0, nil, rlim.copyOut(t, addr) + _, err = rlim.CopyOut(t, addr) + return 0, nil, err } // Setrlimit implements linux syscall setrlimit(2). @@ -155,7 +154,7 @@ func Setrlimit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys if err != nil { return 0, nil, err } - if err := rlim.copyIn(t, addr); err != nil { + if _, err := rlim.CopyIn(t, addr); err != nil { return 0, nil, syserror.EFAULT } _, err = prlimit64(t, resource, rlim.toLimit()) diff --git a/pkg/sentry/syscalls/linux/sys_rusage.go b/pkg/sentry/syscalls/linux/sys_rusage.go index 1674c7445..ac5c98a54 100644 --- a/pkg/sentry/syscalls/linux/sys_rusage.go +++ b/pkg/sentry/syscalls/linux/sys_rusage.go @@ -80,7 +80,7 @@ func Getrusage(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys } ru := getrusage(t, which) - _, err := t.CopyOut(addr, &ru) + _, err := ru.CopyOut(t, addr) return 0, nil, err } @@ -104,7 +104,7 @@ func Times(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall CUTime: linux.ClockTFromDuration(cs2.UserTime), CSTime: linux.ClockTFromDuration(cs2.SysTime), } - if _, err := t.CopyOut(addr, &r); err != nil { + if _, err := r.CopyOut(t, addr); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_sched.go b/pkg/sentry/syscalls/linux/sys_sched.go index 99f6993f5..cd6f4dd94 100644 --- a/pkg/sentry/syscalls/linux/sys_sched.go +++ b/pkg/sentry/syscalls/linux/sys_sched.go @@ -27,6 +27,8 @@ const ( ) // SchedParam replicates struct sched_param in sched.h. +// +// +marshal type SchedParam struct { schedPriority int64 } @@ -45,7 +47,7 @@ func SchedGetparam(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel return 0, nil, syserror.ESRCH } r := SchedParam{schedPriority: onlyPriority} - if _, err := t.CopyOut(param, r); err != nil { + if _, err := r.CopyOut(t, param); err != nil { return 0, nil, err } @@ -79,7 +81,7 @@ func SchedSetscheduler(t *kernel.Task, args arch.SyscallArguments) (uintptr, *ke return 0, nil, syserror.ESRCH } var r SchedParam - if _, err := t.CopyIn(param, &r); err != nil { + if _, err := r.CopyIn(t, param); err != nil { return 0, nil, syserror.EINVAL } if r.schedPriority != onlyPriority { diff --git a/pkg/sentry/syscalls/linux/sys_seccomp.go b/pkg/sentry/syscalls/linux/sys_seccomp.go index 5b7a66f4d..4fdb4463c 100644 --- a/pkg/sentry/syscalls/linux/sys_seccomp.go +++ b/pkg/sentry/syscalls/linux/sys_seccomp.go @@ -24,6 +24,8 @@ import ( ) // userSockFprog is equivalent to Linux's struct sock_fprog on amd64. +// +// +marshal type userSockFprog struct { // Len is the length of the filter in BPF instructions. Len uint16 @@ -33,7 +35,7 @@ type userSockFprog struct { // Filter is a user pointer to the struct sock_filter array that makes up // the filter program. Filter is a uint64 rather than a usermem.Addr // because usermem.Addr is actually uintptr, which is not a fixed-size - // type, and encoding/binary.Read objects to this. + // type. Filter uint64 } @@ -54,11 +56,11 @@ func seccomp(t *kernel.Task, mode, flags uint64, addr usermem.Addr) error { } var fprog userSockFprog - if _, err := t.CopyIn(addr, &fprog); err != nil { + if _, err := fprog.CopyIn(t, addr); err != nil { return err } filter := make([]linux.BPFInstruction, int(fprog.Len)) - if _, err := t.CopyIn(usermem.Addr(fprog.Filter), &filter); err != nil { + if _, err := linux.CopyBPFInstructionSliceIn(t, usermem.Addr(fprog.Filter), filter); err != nil { return err } compiledFilter, err := bpf.Compile(filter) diff --git a/pkg/sentry/syscalls/linux/sys_sem.go b/pkg/sentry/syscalls/linux/sys_sem.go index 5f54f2456..47dadb800 100644 --- a/pkg/sentry/syscalls/linux/sys_sem.go +++ b/pkg/sentry/syscalls/linux/sys_sem.go @@ -18,6 +18,7 @@ import ( "math" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -66,7 +67,7 @@ func Semop(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall } ops := make([]linux.Sembuf, nsops) - if _, err := t.CopyIn(sembufAddr, ops); err != nil { + if _, err := linux.CopySembufSliceIn(t, sembufAddr, ops); err != nil { return 0, nil, err } @@ -116,8 +117,8 @@ func Semctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal case linux.IPC_SET: arg := args[3].Pointer() - s := linux.SemidDS{} - if _, err := t.CopyIn(arg, &s); err != nil { + var s linux.SemidDS + if _, err := s.CopyIn(t, arg); err != nil { return 0, nil, err } @@ -188,7 +189,7 @@ func setValAll(t *kernel.Task, id int32, array usermem.Addr) error { return syserror.EINVAL } vals := make([]uint16, set.Size()) - if _, err := t.CopyIn(array, vals); err != nil { + if _, err := primitive.CopyUint16SliceIn(t, array, vals); err != nil { return err } creds := auth.CredentialsFromContext(t) @@ -217,7 +218,7 @@ func getValAll(t *kernel.Task, id int32, array usermem.Addr) error { if err != nil { return err } - _, err = t.CopyOut(array, vals) + _, err = primitive.CopyUint16SliceOut(t, array, vals) return err } diff --git a/pkg/sentry/syscalls/linux/sys_shm.go b/pkg/sentry/syscalls/linux/sys_shm.go index f0ae8fa8e..584064143 100644 --- a/pkg/sentry/syscalls/linux/sys_shm.go +++ b/pkg/sentry/syscalls/linux/sys_shm.go @@ -112,18 +112,18 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal stat, err := segment.IPCStat(t) if err == nil { - _, err = t.CopyOut(buf, stat) + _, err = stat.CopyOut(t, buf) } return 0, nil, err case linux.IPC_INFO: params := r.IPCInfo() - _, err := t.CopyOut(buf, params) + _, err := params.CopyOut(t, buf) return 0, nil, err case linux.SHM_INFO: info := r.ShmInfo() - _, err := t.CopyOut(buf, info) + _, err := info.CopyOut(t, buf) return 0, nil, err } @@ -137,11 +137,10 @@ func Shmctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal switch cmd { case linux.IPC_SET: var ds linux.ShmidDS - _, err = t.CopyIn(buf, &ds) - if err != nil { + if _, err = ds.CopyIn(t, buf); err != nil { return 0, nil, err } - err = segment.Set(t, &ds) + err := segment.Set(t, &ds) return 0, nil, err case linux.IPC_RMID: diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 161d14ded..9feaca0da 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -67,10 +67,10 @@ const flagsOffset = 48 const sizeOfInt32 = 4 // messageHeader64Len is the length of a MessageHeader64 struct. -var messageHeader64Len = uint64(binary.Size(MessageHeader64{})) +var messageHeader64Len = uint64((*MessageHeader64)(nil).SizeBytes()) // multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct. -var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{})) +var multipleMessageHeader64Len = uint64((*multipleMessageHeader64)(nil).SizeBytes()) // baseRecvFlags are the flags that are accepted across recvmsg(2), // recvmmsg(2), and recvfrom(2). @@ -78,6 +78,8 @@ const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | // MessageHeader64 is the 64-bit representation of the msghdr struct used in // the recvmsg and sendmsg syscalls. +// +// +marshal type MessageHeader64 struct { // Name is the optional pointer to a network address buffer. Name uint64 @@ -106,30 +108,14 @@ type MessageHeader64 struct { // multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in // the recvmmsg and sendmmsg syscalls. +// +// +marshal type multipleMessageHeader64 struct { msgHdr MessageHeader64 msgLen uint32 _ int32 } -// CopyInMessageHeader64 copies a message header from user to kernel memory. -func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error { - b := t.CopyScratchBuffer(52) - if _, err := t.CopyInBytes(addr, b); err != nil { - return err - } - - msg.Name = usermem.ByteOrder.Uint64(b[0:]) - msg.NameLen = usermem.ByteOrder.Uint32(b[8:]) - msg.Iov = usermem.ByteOrder.Uint64(b[16:]) - msg.IovLen = usermem.ByteOrder.Uint64(b[24:]) - msg.Control = usermem.ByteOrder.Uint64(b[32:]) - msg.ControlLen = usermem.ByteOrder.Uint64(b[40:]) - msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:])) - - return nil -} - // CaptureAddress allocates memory for and copies a socket address structure // from the untrusted address space range. func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { @@ -148,10 +134,10 @@ func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, // writeAddress writes a sockaddr structure and its length to an output buffer // in the unstrusted address space range. If the address is bigger than the // buffer, it is truncated. -func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { +func writeAddress(t *kernel.Task, addr linux.SockAddr, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { // Get the buffer length. var bufLen uint32 - if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil { + if _, err := primitive.CopyUint32In(t, addrLenPtr, &bufLen); err != nil { return err } @@ -160,7 +146,7 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Write the length unconditionally. - if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil { + if _, err := primitive.CopyUint32Out(t, addrLenPtr, addrLen); err != nil { return err } @@ -173,7 +159,8 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Copy as much of the address as will fit in the buffer. - encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr) + encodedAddr := t.CopyScratchBuffer(addr.SizeBytes()) + addr.MarshalUnsafe(encodedAddr) if bufLen > uint32(len(encodedAddr)) { bufLen = uint32(len(encodedAddr)) } @@ -247,7 +234,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Copy the file descriptors out. - if _, err := t.CopyOut(socks, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, socks, fds); err != nil { for _, fd := range fds { if file, _ := t.FDTable().Remove(t, fd); file != nil { file.DecRef(t) @@ -456,8 +443,8 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Read the length. Reject negative values. - optLen := int32(0) - if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + var optLen int32 + if _, err := primitive.CopyInt32In(t, optLenAddr, &optLen); err != nil { return 0, nil, err } if optLen < 0 { @@ -471,7 +458,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } vLen := int32(binary.Size(v)) - if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + if _, err := primitive.CopyInt32Out(t, optLenAddr, vLen); err != nil { return 0, nil, err } @@ -733,7 +720,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -748,7 +735,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) { // Capture the message header and io vectors. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -780,7 +767,7 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i if int(msg.Flags) != mflags { // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } } @@ -817,17 +804,17 @@ func recvSingleMsg(t *kernel.Task, s socket.Socket, msgPtr usermem.Addr, flags i } // Copy the control data to the caller. - if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { + if _, err := primitive.CopyUint64Out(t, msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { return 0, err } if len(controlData) > 0 { - if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil { + if _, err := t.CopyOutBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } @@ -996,7 +983,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -1011,7 +998,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr usermem.Addr, flags int32) (uintptr, error) { // Capture the message header. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -1022,7 +1009,7 @@ func sendSingleMsg(t *kernel.Task, s socket.Socket, file *fs.File, msgPtr userme return 0, syserror.ENOBUFS } controlData = make([]byte, msg.ControlLen) - if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { + if _, err := t.CopyInBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } diff --git a/pkg/sentry/syscalls/linux/sys_splice.go b/pkg/sentry/syscalls/linux/sys_splice.go index c69941feb..46616c961 100644 --- a/pkg/sentry/syscalls/linux/sys_splice.go +++ b/pkg/sentry/syscalls/linux/sys_splice.go @@ -16,6 +16,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -141,7 +142,7 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc // Copy in the offset. var offset int64 - if _, err := t.CopyIn(offsetAddr, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, offsetAddr, &offset); err != nil { return 0, nil, err } @@ -149,11 +150,11 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc n, err = doSplice(t, outFile, inFile, fs.SpliceOpts{ Length: count, SrcOffset: true, - SrcStart: offset, + SrcStart: int64(offset), }, outFile.Flags().NonBlocking) // Copy out the new offset. - if _, err := t.CopyOut(offsetAddr, n+offset); err != nil { + if _, err := primitive.CopyInt64Out(t, offsetAddr, offset+n); err != nil { return 0, nil, err } } else { @@ -228,7 +229,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } var offset int64 - if _, err := t.CopyIn(outOffset, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, outOffset, &offset); err != nil { return 0, nil, err } @@ -246,7 +247,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal } var offset int64 - if _, err := t.CopyIn(inOffset, &offset); err != nil { + if _, err := primitive.CopyInt64In(t, inOffset, &offset); err != nil { return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_stat.go b/pkg/sentry/syscalls/linux/sys_stat.go index a5826f2dd..cda29a8b5 100644 --- a/pkg/sentry/syscalls/linux/sys_stat.go +++ b/pkg/sentry/syscalls/linux/sys_stat.go @@ -221,7 +221,7 @@ func statx(t *kernel.Task, sattr fs.StableAttr, uattr fs.UnstableAttr, statxAddr DevMajor: uint32(devMajor), DevMinor: devMinor, } - _, err := t.CopyOut(statxAddr, &s) + _, err := s.CopyOut(t, statxAddr) return err } @@ -283,7 +283,7 @@ func statfsImpl(t *kernel.Task, d *fs.Dirent, addr usermem.Addr) error { FragmentSize: d.Inode.StableAttr.BlockSize, // Leave other fields 0 like simple_statfs does. } - _, err = t.CopyOut(addr, &statfs) + _, err = statfs.CopyOut(t, addr) return err } diff --git a/pkg/sentry/syscalls/linux/sys_sysinfo.go b/pkg/sentry/syscalls/linux/sys_sysinfo.go index 297de052a..674d341b6 100644 --- a/pkg/sentry/syscalls/linux/sys_sysinfo.go +++ b/pkg/sentry/syscalls/linux/sys_sysinfo.go @@ -43,6 +43,6 @@ func Sysinfo(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca FreeRAM: memFree, Unit: 1, } - _, err := t.CopyOut(addr, si) + _, err := si.CopyOut(t, addr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_thread.go b/pkg/sentry/syscalls/linux/sys_thread.go index 101096038..39ca9ea97 100644 --- a/pkg/sentry/syscalls/linux/sys_thread.go +++ b/pkg/sentry/syscalls/linux/sys_thread.go @@ -19,6 +19,7 @@ import ( "syscall" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fsbridge" @@ -311,13 +312,13 @@ func wait4(t *kernel.Task, pid int, statusAddr usermem.Addr, options int, rusage return 0, err } if statusAddr != 0 { - if _, err := t.CopyOut(statusAddr, wr.Status); err != nil { + if _, err := primitive.CopyUint32Out(t, statusAddr, wr.Status); err != nil { return 0, err } } if rusageAddr != 0 { ru := getrusage(wr.Task, linux.RUSAGE_BOTH) - if _, err := t.CopyOut(rusageAddr, &ru); err != nil { + if _, err := ru.CopyOut(t, rusageAddr); err != nil { return 0, err } } @@ -395,14 +396,14 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // as well. if infop != 0 { var si arch.SignalInfo - _, err = t.CopyOut(infop, &si) + _, err = si.CopyOut(t, infop) } } return 0, nil, err } if rusageAddr != 0 { ru := getrusage(wr.Task, linux.RUSAGE_BOTH) - if _, err := t.CopyOut(rusageAddr, &ru); err != nil { + if _, err := ru.CopyOut(t, rusageAddr); err != nil { return 0, nil, err } } @@ -441,7 +442,7 @@ func Waitid(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal default: t.Warningf("waitid got incomprehensible wait status %d", s) } - _, err = t.CopyOut(infop, &si) + _, err = si.CopyOut(t, infop) return 0, nil, err } @@ -558,9 +559,7 @@ func Getcpu(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // third argument to this system call is nowadays unused. if cpu != 0 { - buf := t.CopyScratchBuffer(4) - usermem.ByteOrder.PutUint32(buf, uint32(t.CPU())) - if _, err := t.CopyOutBytes(cpu, buf); err != nil { + if _, err := primitive.CopyInt32Out(t, cpu, t.CPU()); err != nil { return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/sys_time.go b/pkg/sentry/syscalls/linux/sys_time.go index a2a24a027..c5054d2f1 100644 --- a/pkg/sentry/syscalls/linux/sys_time.go +++ b/pkg/sentry/syscalls/linux/sys_time.go @@ -19,6 +19,7 @@ import ( "time" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" @@ -168,7 +169,7 @@ func Time(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallC return uintptr(r), nil, nil } - if _, err := t.CopyOut(addr, r); err != nil { + if _, err := r.CopyOut(t, addr); err != nil { return 0, nil, err } return uintptr(r), nil, nil @@ -334,8 +335,8 @@ func Gettimeofday(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. // Ask the time package for the timezone. _, offset := time.Now().Zone() // This int32 array mimics linux's struct timezone. - timezone := [2]int32{-int32(offset) / 60, 0} - _, err := t.CopyOut(tz, timezone) + timezone := []int32{-int32(offset) / 60, 0} + _, err := primitive.CopyInt32SliceOut(t, tz, timezone) return 0, nil, err } return 0, nil, nil diff --git a/pkg/sentry/syscalls/linux/sys_timer.go b/pkg/sentry/syscalls/linux/sys_timer.go index a4c400f87..45eef4feb 100644 --- a/pkg/sentry/syscalls/linux/sys_timer.go +++ b/pkg/sentry/syscalls/linux/sys_timer.go @@ -21,81 +21,63 @@ import ( "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" - "gvisor.dev/gvisor/pkg/usermem" ) const nsecPerSec = int64(time.Second) -// copyItimerValIn copies an ItimerVal from the untrusted app range to the -// kernel. The ItimerVal may be either 32 or 64 bits. -// A NULL address is allowed because because Linux allows -// setitimer(which, NULL, &old_value) which disables the timer. -// There is a KERN_WARN message saying this misfeature will be removed. -// However, that hasn't happened as of 3.19, so we continue to support it. -func copyItimerValIn(t *kernel.Task, addr usermem.Addr) (linux.ItimerVal, error) { - if addr == usermem.Addr(0) { - return linux.ItimerVal{}, nil - } - - switch t.Arch().Width() { - case 8: - // Native size, just copy directly. - var itv linux.ItimerVal - if _, err := t.CopyIn(addr, &itv); err != nil { - return linux.ItimerVal{}, err - } - - return itv, nil - default: - return linux.ItimerVal{}, syserror.ENOSYS - } -} - -// copyItimerValOut copies an ItimerVal to the untrusted app range. -// The ItimerVal may be either 32 or 64 bits. -// A NULL address is allowed, in which case no copy takes place -func copyItimerValOut(t *kernel.Task, addr usermem.Addr, itv *linux.ItimerVal) error { - if addr == usermem.Addr(0) { - return nil - } - - switch t.Arch().Width() { - case 8: - // Native size, just copy directly. - _, err := t.CopyOut(addr, itv) - return err - default: - return syserror.ENOSYS - } -} - // Getitimer implements linux syscall getitimer(2). func Getitimer(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { + if t.Arch().Width() != 8 { + // Definition of linux.ItimerVal assumes 64-bit architecture. + return 0, nil, syserror.ENOSYS + } + timerID := args[0].Int() - val := args[1].Pointer() + addr := args[1].Pointer() olditv, err := t.Getitimer(timerID) if err != nil { return 0, nil, err } - return 0, nil, copyItimerValOut(t, val, &olditv) + // A NULL address is allowed, in which case no copy out takes place. + if addr == 0 { + return 0, nil, nil + } + _, err = olditv.CopyOut(t, addr) + return 0, nil, err } // Setitimer implements linux syscall setitimer(2). func Setitimer(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { - timerID := args[0].Int() - newVal := args[1].Pointer() - oldVal := args[2].Pointer() + if t.Arch().Width() != 8 { + // Definition of linux.ItimerVal assumes 64-bit architecture. + return 0, nil, syserror.ENOSYS + } - newitv, err := copyItimerValIn(t, newVal) - if err != nil { - return 0, nil, err + timerID := args[0].Int() + newAddr := args[1].Pointer() + oldAddr := args[2].Pointer() + + var newitv linux.ItimerVal + // A NULL address is allowed because because Linux allows + // setitimer(which, NULL, &old_value) which disables the timer. There is a + // KERN_WARN message saying this misfeature will be removed. However, that + // hasn't happened as of 3.19, so we continue to support it. + if newAddr != 0 { + if _, err := newitv.CopyIn(t, newAddr); err != nil { + return 0, nil, err + } } olditv, err := t.Setitimer(timerID, newitv) if err != nil { return 0, nil, err } - return 0, nil, copyItimerValOut(t, oldVal, &olditv) + // A NULL address is allowed, in which case no copy out takes place. + if oldAddr == 0 { + return 0, nil, nil + } + _, err = olditv.CopyOut(t, oldAddr) + return 0, nil, err } // Alarm implements linux syscall alarm(2). @@ -131,7 +113,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S var sev *linux.Sigevent if sevp != 0 { sev = &linux.Sigevent{} - if _, err = t.CopyIn(sevp, sev); err != nil { + if _, err = sev.CopyIn(t, sevp); err != nil { return 0, nil, err } } @@ -141,7 +123,7 @@ func TimerCreate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.S return 0, nil, err } - if _, err := t.CopyOut(timerIDp, &id); err != nil { + if _, err := id.CopyOut(t, timerIDp); err != nil { t.IntervalTimerDelete(id) return 0, nil, err } @@ -157,7 +139,7 @@ func TimerSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. oldValAddr := args[3].Pointer() var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } oldVal, err := t.IntervalTimerSettime(timerID, newVal, flags&linux.TIMER_ABSTIME != 0) @@ -165,9 +147,8 @@ func TimerSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. return 0, nil, err } if oldValAddr != 0 { - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { - return 0, nil, err - } + _, err = oldVal.CopyOut(t, oldValAddr) + return 0, nil, err } return 0, nil, nil } @@ -181,7 +162,7 @@ func TimerGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel. if err != nil { return 0, nil, err } - _, err = t.CopyOut(curValAddr, &curVal) + _, err = curVal.CopyOut(t, curValAddr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_timerfd.go b/pkg/sentry/syscalls/linux/sys_timerfd.go index 34b03e4ee..cadd9d348 100644 --- a/pkg/sentry/syscalls/linux/sys_timerfd.go +++ b/pkg/sentry/syscalls/linux/sys_timerfd.go @@ -81,7 +81,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tf.Clock()) @@ -91,7 +91,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, oldS := tf.SetTime(newS) if oldValAddr != 0 { oldVal := ktime.ItimerspecFromSetting(tm, oldS) - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + if _, err := oldVal.CopyOut(t, oldValAddr); err != nil { return 0, nil, err } } @@ -116,6 +116,6 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, s := tf.GetTime() curVal := ktime.ItimerspecFromSetting(tm, s) - _, err := t.CopyOut(curValAddr, &curVal) + _, err := curVal.CopyOut(t, curValAddr) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/sys_tls_amd64.go b/pkg/sentry/syscalls/linux/sys_tls_amd64.go index b3eb96a1c..6ddd30d5c 100644 --- a/pkg/sentry/syscalls/linux/sys_tls_amd64.go +++ b/pkg/sentry/syscalls/linux/sys_tls_amd64.go @@ -18,6 +18,7 @@ package linux import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" @@ -30,17 +31,19 @@ func ArchPrctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys case linux.ARCH_GET_FS: addr := args[1].Pointer() fsbase := t.Arch().TLS() - _, err := t.CopyOut(addr, uint64(fsbase)) - if err != nil { - return 0, nil, err + switch t.Arch().Width() { + case 8: + if _, err := primitive.CopyUint64Out(t, addr, uint64(fsbase)); err != nil { + return 0, nil, err + } + default: + return 0, nil, syserror.ENOSYS } - case linux.ARCH_SET_FS: fsbase := args[1].Uint64() if !t.Arch().SetTLS(uintptr(fsbase)) { return 0, nil, syserror.EPERM } - case linux.ARCH_GET_GS, linux.ARCH_SET_GS: t.Kernel().EmitUnimplementedEvent(t) fallthrough diff --git a/pkg/sentry/syscalls/linux/sys_utsname.go b/pkg/sentry/syscalls/linux/sys_utsname.go index e9d702e8e..66c5974f5 100644 --- a/pkg/sentry/syscalls/linux/sys_utsname.go +++ b/pkg/sentry/syscalls/linux/sys_utsname.go @@ -46,7 +46,7 @@ func Uname(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall // Copy out the result. va := args[0].Pointer() - _, err := t.CopyOut(va, u) + _, err := u.CopyOut(t, va) return 0, nil, err } diff --git a/pkg/sentry/syscalls/linux/vfs2/aio.go b/pkg/sentry/syscalls/linux/vfs2/aio.go index 42559bf69..6d0a38330 100644 --- a/pkg/sentry/syscalls/linux/vfs2/aio.go +++ b/pkg/sentry/syscalls/linux/vfs2/aio.go @@ -17,6 +17,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsimpl/eventfd" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -38,21 +39,27 @@ func IoSubmit(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc } for i := int32(0); i < nrEvents; i++ { - // Copy in the address. - cbAddrNative := t.Arch().Native(0) - if _, err := t.CopyIn(addr, cbAddrNative); err != nil { - if i > 0 { - // Some successful. - return uintptr(i), nil, nil + // Copy in the callback address. + var cbAddr usermem.Addr + switch t.Arch().Width() { + case 8: + var cbAddrP primitive.Uint64 + if _, err := cbAddrP.CopyIn(t, addr); err != nil { + if i > 0 { + // Some successful. + return uintptr(i), nil, nil + } + // Nothing done. + return 0, nil, err } - // Nothing done. - return 0, nil, err + cbAddr = usermem.Addr(cbAddrP) + default: + return 0, nil, syserror.ENOSYS } // Copy in this callback. var cb linux.IOCallback - cbAddr := usermem.Addr(t.Arch().Value(cbAddrNative)) - if _, err := t.CopyIn(cbAddr, &cb); err != nil { + if _, err := cb.CopyIn(t, cbAddr); err != nil { if i > 0 { // Some have been successful. return uintptr(i), nil, nil diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go index fdd8f88c5..d8b8d9783 100644 --- a/pkg/sentry/syscalls/linux/vfs2/fd.go +++ b/pkg/sentry/syscalls/linux/vfs2/fd.go @@ -181,11 +181,11 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall if !hasOwner { return 0, nil, nil } - _, err := t.CopyOut(args[2].Pointer(), &owner) + _, err := owner.CopyOut(t, args[2].Pointer()) return 0, nil, err case linux.F_SETOWN_EX: var owner linux.FOwnerEx - _, err := t.CopyIn(args[2].Pointer(), &owner) + _, err := owner.CopyIn(t, args[2].Pointer()) if err != nil { return 0, nil, err } @@ -286,7 +286,7 @@ func posixLock(t *kernel.Task, args arch.SyscallArguments, file *vfs.FileDescrip // Copy in the lock request. flockAddr := args[2].Pointer() var flock linux.Flock - if _, err := t.CopyIn(flockAddr, &flock); err != nil { + if _, err := flock.CopyIn(t, flockAddr); err != nil { return err } diff --git a/pkg/sentry/syscalls/linux/vfs2/ioctl.go b/pkg/sentry/syscalls/linux/vfs2/ioctl.go index baa8a49af..2806c3f6f 100644 --- a/pkg/sentry/syscalls/linux/vfs2/ioctl.go +++ b/pkg/sentry/syscalls/linux/vfs2/ioctl.go @@ -16,6 +16,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserror" @@ -47,7 +48,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIONBIO: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.StatusFlags() @@ -60,7 +61,7 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall case linux.FIOASYNC: var set int32 - if _, err := t.CopyIn(args[2].Pointer(), &set); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &set); err != nil { return 0, nil, err } flags := file.StatusFlags() @@ -82,12 +83,12 @@ func Ioctl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall who = owner.PID } } - _, err := t.CopyOut(args[2].Pointer(), &who) + _, err := primitive.CopyInt32Out(t, args[2].Pointer(), who) return 0, nil, err case linux.FIOSETOWN, linux.SIOCSPGRP: var who int32 - if _, err := t.CopyIn(args[2].Pointer(), &who); err != nil { + if _, err := primitive.CopyInt32In(t, args[2].Pointer(), &who); err != nil { return 0, nil, err } ownerType := int32(linux.F_OWNER_PID) diff --git a/pkg/sentry/syscalls/linux/vfs2/pipe.go b/pkg/sentry/syscalls/linux/vfs2/pipe.go index 3aa6d939d..ee38fdca0 100644 --- a/pkg/sentry/syscalls/linux/vfs2/pipe.go +++ b/pkg/sentry/syscalls/linux/vfs2/pipe.go @@ -16,6 +16,7 @@ package vfs2 import ( "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fsimpl/pipefs" "gvisor.dev/gvisor/pkg/sentry/kernel" @@ -51,7 +52,7 @@ func pipe2(t *kernel.Task, addr usermem.Addr, flags int32) error { if err != nil { return err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(t, fd); file != nil { file.DecRef(t) diff --git a/pkg/sentry/syscalls/linux/vfs2/poll.go b/pkg/sentry/syscalls/linux/vfs2/poll.go index 79ad64039..c22e4ce54 100644 --- a/pkg/sentry/syscalls/linux/vfs2/poll.go +++ b/pkg/sentry/syscalls/linux/vfs2/poll.go @@ -165,7 +165,7 @@ func copyInPollFDs(t *kernel.Task, addr usermem.Addr, nfds uint) ([]linux.PollFD pfd := make([]linux.PollFD, nfds) if nfds > 0 { - if _, err := t.CopyIn(addr, &pfd); err != nil { + if _, err := linux.CopyPollFDSliceIn(t, addr, pfd); err != nil { return nil, err } } @@ -192,7 +192,7 @@ func doPoll(t *kernel.Task, addr usermem.Addr, nfds uint, timeout time.Duration) // The poll entries are copied out regardless of whether // any are set or not. This aligns with the Linux behavior. if nfds > 0 && err == nil { - if _, err := t.CopyOut(addr, pfd); err != nil { + if _, err := linux.CopyPollFDSliceOut(t, addr, pfd); err != nil { return remainingTimeout, 0, err } } @@ -205,7 +205,7 @@ func CopyInFDSet(t *kernel.Task, addr usermem.Addr, nBytes, nBitsInLastPartialBy set := make([]byte, nBytes) if addr != 0 { - if _, err := t.CopyIn(addr, &set); err != nil { + if _, err := t.CopyInBytes(addr, set); err != nil { return nil, err } // If we only use part of the last byte, mask out the extraneous bits. @@ -332,19 +332,19 @@ func doSelect(t *kernel.Task, nfds int, readFDs, writeFDs, exceptFDs usermem.Add // Copy updated vectors back. if readFDs != 0 { - if _, err := t.CopyOut(readFDs, r); err != nil { + if _, err := t.CopyOutBytes(readFDs, r); err != nil { return 0, err } } if writeFDs != 0 { - if _, err := t.CopyOut(writeFDs, w); err != nil { + if _, err := t.CopyOutBytes(writeFDs, w); err != nil { return 0, err } } if exceptFDs != 0 { - if _, err := t.CopyOut(exceptFDs, e); err != nil { + if _, err := t.CopyOutBytes(exceptFDs, e); err != nil { return 0, err } } @@ -497,6 +497,12 @@ func Select(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal return n, nil, err } +// +marshal +type sigSetWithSize struct { + sigsetAddr uint64 + sizeofSigset uint64 +} + // Pselect implements linux syscall pselect(2). func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) { nfds := int(args[0].Int()) // select(2) uses an int. @@ -538,12 +544,6 @@ func Pselect(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysca return n, nil, err } -// +marshal -type sigSetWithSize struct { - sigsetAddr uint64 - sizeofSigset uint64 -} - // copyTimespecInToDuration copies a Timespec from the untrusted app range, // validates it and converts it to a Duration. // diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go index 5e6eb13ba..1ee37e5a8 100644 --- a/pkg/sentry/syscalls/linux/vfs2/setstat.go +++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go @@ -346,7 +346,7 @@ func populateSetStatOptionsForUtimes(t *kernel.Task, timesAddr usermem.Addr, opt return nil } var times [2]linux.Timeval - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimevalSliceIn(t, timesAddr, times[:]); err != nil { return err } if times[0].Usec < 0 || times[0].Usec > 999999 || times[1].Usec < 0 || times[1].Usec > 999999 { @@ -410,7 +410,7 @@ func populateSetStatOptionsForUtimens(t *kernel.Task, timesAddr usermem.Addr, op return nil } var times [2]linux.Timespec - if _, err := t.CopyIn(timesAddr, ×); err != nil { + if _, err := linux.CopyTimespecSliceIn(t, timesAddr, times[:]); err != nil { return err } if times[0].Nsec != linux.UTIME_OMIT { diff --git a/pkg/sentry/syscalls/linux/vfs2/socket.go b/pkg/sentry/syscalls/linux/vfs2/socket.go index c899e3a72..bfae6b7e9 100644 --- a/pkg/sentry/syscalls/linux/vfs2/socket.go +++ b/pkg/sentry/syscalls/linux/vfs2/socket.go @@ -66,10 +66,10 @@ const flagsOffset = 48 const sizeOfInt32 = 4 // messageHeader64Len is the length of a MessageHeader64 struct. -var messageHeader64Len = uint64(binary.Size(MessageHeader64{})) +var messageHeader64Len = uint64((*MessageHeader64)(nil).SizeBytes()) // multipleMessageHeader64Len is the length of a multipeMessageHeader64 struct. -var multipleMessageHeader64Len = uint64(binary.Size(multipleMessageHeader64{})) +var multipleMessageHeader64Len = uint64((*multipleMessageHeader64)(nil).SizeBytes()) // baseRecvFlags are the flags that are accepted across recvmsg(2), // recvmmsg(2), and recvfrom(2). @@ -77,6 +77,8 @@ const baseRecvFlags = linux.MSG_OOB | linux.MSG_DONTROUTE | linux.MSG_DONTWAIT | // MessageHeader64 is the 64-bit representation of the msghdr struct used in // the recvmsg and sendmsg syscalls. +// +// +marshal type MessageHeader64 struct { // Name is the optional pointer to a network address buffer. Name uint64 @@ -105,30 +107,14 @@ type MessageHeader64 struct { // multipleMessageHeader64 is the 64-bit representation of the mmsghdr struct used in // the recvmmsg and sendmmsg syscalls. +// +// +marshal type multipleMessageHeader64 struct { msgHdr MessageHeader64 msgLen uint32 _ int32 } -// CopyInMessageHeader64 copies a message header from user to kernel memory. -func CopyInMessageHeader64(t *kernel.Task, addr usermem.Addr, msg *MessageHeader64) error { - b := t.CopyScratchBuffer(52) - if _, err := t.CopyInBytes(addr, b); err != nil { - return err - } - - msg.Name = usermem.ByteOrder.Uint64(b[0:]) - msg.NameLen = usermem.ByteOrder.Uint32(b[8:]) - msg.Iov = usermem.ByteOrder.Uint64(b[16:]) - msg.IovLen = usermem.ByteOrder.Uint64(b[24:]) - msg.Control = usermem.ByteOrder.Uint64(b[32:]) - msg.ControlLen = usermem.ByteOrder.Uint64(b[40:]) - msg.Flags = int32(usermem.ByteOrder.Uint32(b[48:])) - - return nil -} - // CaptureAddress allocates memory for and copies a socket address structure // from the untrusted address space range. func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, error) { @@ -147,10 +133,10 @@ func CaptureAddress(t *kernel.Task, addr usermem.Addr, addrlen uint32) ([]byte, // writeAddress writes a sockaddr structure and its length to an output buffer // in the unstrusted address space range. If the address is bigger than the // buffer, it is truncated. -func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { +func writeAddress(t *kernel.Task, addr linux.SockAddr, addrLen uint32, addrPtr usermem.Addr, addrLenPtr usermem.Addr) error { // Get the buffer length. var bufLen uint32 - if _, err := t.CopyIn(addrLenPtr, &bufLen); err != nil { + if _, err := primitive.CopyUint32In(t, addrLenPtr, &bufLen); err != nil { return err } @@ -159,7 +145,7 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Write the length unconditionally. - if _, err := t.CopyOut(addrLenPtr, addrLen); err != nil { + if _, err := primitive.CopyUint32Out(t, addrLenPtr, addrLen); err != nil { return err } @@ -172,7 +158,8 @@ func writeAddress(t *kernel.Task, addr interface{}, addrLen uint32, addrPtr user } // Copy as much of the address as will fit in the buffer. - encodedAddr := binary.Marshal(nil, usermem.ByteOrder, addr) + encodedAddr := t.CopyScratchBuffer(addr.SizeBytes()) + addr.MarshalUnsafe(encodedAddr) if bufLen > uint32(len(encodedAddr)) { bufLen = uint32(len(encodedAddr)) } @@ -250,7 +237,7 @@ func SocketPair(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy return 0, nil, err } - if _, err := t.CopyOut(addr, fds); err != nil { + if _, err := primitive.CopyInt32SliceOut(t, addr, fds); err != nil { for _, fd := range fds { if _, file := t.FDTable().Remove(t, fd); file != nil { file.DecRef(t) @@ -459,8 +446,8 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } // Read the length. Reject negative values. - optLen := int32(0) - if _, err := t.CopyIn(optLenAddr, &optLen); err != nil { + var optLen int32 + if _, err := primitive.CopyInt32In(t, optLenAddr, &optLen); err != nil { return 0, nil, err } if optLen < 0 { @@ -474,7 +461,7 @@ func GetSockOpt(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sy } vLen := int32(binary.Size(v)) - if _, err := t.CopyOut(optLenAddr, vLen); err != nil { + if _, err := primitive.CopyInt32Out(t, optLenAddr, vLen); err != nil { return 0, nil, err } @@ -736,7 +723,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -751,7 +738,7 @@ func RecvMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, flags int32, haveDeadline bool, deadline ktime.Time) (uintptr, error) { // Capture the message header and io vectors. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -783,7 +770,7 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla if int(msg.Flags) != mflags { // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } } @@ -820,17 +807,17 @@ func recvSingleMsg(t *kernel.Task, s socket.SocketVFS2, msgPtr usermem.Addr, fla } // Copy the control data to the caller. - if _, err := t.CopyOut(msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { + if _, err := primitive.CopyUint64Out(t, msgPtr+controlLenOffset, uint64(len(controlData))); err != nil { return 0, err } if len(controlData) > 0 { - if _, err := t.CopyOut(usermem.Addr(msg.Control), controlData); err != nil { + if _, err := t.CopyOutBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } // Copy out the flags to the caller. - if _, err := t.CopyOut(msgPtr+flagsOffset, int32(mflags)); err != nil { + if _, err := primitive.CopyInt32Out(t, msgPtr+flagsOffset, int32(mflags)); err != nil { return 0, err } @@ -999,7 +986,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if !ok { return 0, nil, syserror.EFAULT } - if _, err = t.CopyOut(lp, uint32(n)); err != nil { + if _, err = primitive.CopyUint32Out(t, lp, uint32(n)); err != nil { break } count++ @@ -1014,7 +1001,7 @@ func SendMMsg(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescription, msgPtr usermem.Addr, flags int32) (uintptr, error) { // Capture the message header. var msg MessageHeader64 - if err := CopyInMessageHeader64(t, msgPtr, &msg); err != nil { + if _, err := msg.CopyIn(t, msgPtr); err != nil { return 0, err } @@ -1025,7 +1012,7 @@ func sendSingleMsg(t *kernel.Task, s socket.SocketVFS2, file *vfs.FileDescriptio return 0, syserror.ENOBUFS } controlData = make([]byte, msg.ControlLen) - if _, err := t.CopyIn(usermem.Addr(msg.Control), &controlData); err != nil { + if _, err := t.CopyInBytes(usermem.Addr(msg.Control), controlData); err != nil { return 0, err } } diff --git a/pkg/sentry/syscalls/linux/vfs2/splice.go b/pkg/sentry/syscalls/linux/vfs2/splice.go index 5543cfac2..f55d74cd2 100644 --- a/pkg/sentry/syscalls/linux/vfs2/splice.go +++ b/pkg/sentry/syscalls/linux/vfs2/splice.go @@ -19,6 +19,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/pipe" @@ -89,7 +90,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if inFile.Options().DenyPRead { return 0, nil, syserror.EINVAL } - if _, err := t.CopyIn(inOffsetPtr, &inOffset); err != nil { + if _, err := primitive.CopyInt64In(t, inOffsetPtr, &inOffset); err != nil { return 0, nil, err } if inOffset < 0 { @@ -104,7 +105,7 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal if outFile.Options().DenyPWrite { return 0, nil, syserror.EINVAL } - if _, err := t.CopyIn(outOffsetPtr, &outOffset); err != nil { + if _, err := primitive.CopyInt64In(t, outOffsetPtr, &outOffset); err != nil { return 0, nil, err } if outOffset < 0 { @@ -160,12 +161,12 @@ func Splice(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscal // Copy updated offsets out. if inOffsetPtr != 0 { - if _, err := t.CopyOut(inOffsetPtr, &inOffset); err != nil { + if _, err := primitive.CopyInt64Out(t, inOffsetPtr, inOffset); err != nil { return 0, nil, err } } if outOffsetPtr != 0 { - if _, err := t.CopyOut(outOffsetPtr, &outOffset); err != nil { + if _, err := primitive.CopyInt64Out(t, outOffsetPtr, outOffset); err != nil { return 0, nil, err } } @@ -303,9 +304,12 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if inFile.Options().DenyPRead { return 0, nil, syserror.ESPIPE } - if _, err := t.CopyIn(offsetAddr, &offset); err != nil { + var offsetP primitive.Int64 + if _, err := offsetP.CopyIn(t, offsetAddr); err != nil { return 0, nil, err } + offset = int64(offsetP) + if offset < 0 { return 0, nil, syserror.EINVAL } @@ -422,7 +426,8 @@ func Sendfile(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sysc if offsetAddr != 0 { // Copy out the new offset. - if _, err := t.CopyOut(offsetAddr, offset); err != nil { + offsetP := primitive.Uint64(offset) + if _, err := offsetP.CopyOut(t, offsetAddr); err != nil { return 0, nil, err } } diff --git a/pkg/sentry/syscalls/linux/vfs2/timerfd.go b/pkg/sentry/syscalls/linux/vfs2/timerfd.go index 7a26890ef..250870c03 100644 --- a/pkg/sentry/syscalls/linux/vfs2/timerfd.go +++ b/pkg/sentry/syscalls/linux/vfs2/timerfd.go @@ -87,7 +87,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne } var newVal linux.Itimerspec - if _, err := t.CopyIn(newValAddr, &newVal); err != nil { + if _, err := newVal.CopyIn(t, newValAddr); err != nil { return 0, nil, err } newS, err := ktime.SettingFromItimerspec(newVal, flags&linux.TFD_TIMER_ABSTIME != 0, tfd.Clock()) @@ -97,7 +97,7 @@ func TimerfdSettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, oldS := tfd.SetTime(newS) if oldValAddr != 0 { oldVal := ktime.ItimerspecFromSetting(tm, oldS) - if _, err := t.CopyOut(oldValAddr, &oldVal); err != nil { + if _, err := oldVal.CopyOut(t, oldValAddr); err != nil { return 0, nil, err } } @@ -122,6 +122,6 @@ func TimerfdGettime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kerne tm, s := tfd.GetTime() curVal := ktime.ItimerspecFromSetting(tm, s) - _, err := t.CopyOut(curValAddr, &curVal) + _, err := curVal.CopyOut(t, curValAddr) return 0, nil, err } |