diff options
Diffstat (limited to 'pkg/sentry/socket')
27 files changed, 764 insertions, 325 deletions
diff --git a/pkg/sentry/socket/BUILD b/pkg/sentry/socket/BUILD index c0fd3425b..a3f775d15 100644 --- a/pkg/sentry/socket/BUILD +++ b/pkg/sentry/socket/BUILD @@ -10,6 +10,7 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/context", + "//pkg/marshal", "//pkg/sentry/device", "//pkg/sentry/fs", "//pkg/sentry/fs/fsutil", @@ -20,6 +21,5 @@ go_library( "//pkg/syserr", "//pkg/tcpip", "//pkg/usermem", - "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index e76e498de..b6ebe29d6 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -21,6 +21,8 @@ go_library( "//pkg/context", "//pkg/fdnotifier", "//pkg/log", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/safemem", "//pkg/sentry/arch", "//pkg/sentry/device", @@ -37,11 +39,12 @@ go_library( "//pkg/sentry/vfs", "//pkg/syserr", "//pkg/syserror", + "//pkg/tcpip", + "//pkg/tcpip/network/ipv4", + "//pkg/tcpip/network/ipv6", "//pkg/tcpip/stack", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", - "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/hostinet/socket.go b/pkg/sentry/socket/hostinet/socket.go index 242e6bf76..7d3c4a01c 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -24,6 +24,8 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -36,8 +38,6 @@ import ( "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const ( diff --git a/pkg/sentry/socket/hostinet/socket_vfs2.go b/pkg/sentry/socket/hostinet/socket_vfs2.go index 8a1d52ebf..163af329b 100644 --- a/pkg/sentry/socket/hostinet/socket_vfs2.go +++ b/pkg/sentry/socket/hostinet/socket_vfs2.go @@ -52,6 +52,7 @@ var _ = socket.SocketVFS2(&socketVFS2{}) func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol int, fd int, flags uint32) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) s := &socketVFS2{ socketOpsCommon: socketOpsCommon{ @@ -77,6 +78,13 @@ func newVFS2Socket(t *kernel.Task, family int, stype linux.SockType, protocol in return vfsfd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *socketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *socketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) @@ -97,11 +105,6 @@ func (s *socketVFS2) Ioctl(ctx context.Context, uio usermem.IO, args arch.Syscal return ioctl(ctx, s.fd, uio, args) } -// Allocate implements vfs.FileDescriptionImpl.Allocate. -func (s *socketVFS2) Allocate(ctx context.Context, mode, offset, length uint64) error { - return syserror.ENODEV -} - // PRead implements vfs.FileDescriptionImpl.PRead. func (s *socketVFS2) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { return 0, syserror.ESPIPE diff --git a/pkg/sentry/socket/hostinet/stack.go b/pkg/sentry/socket/hostinet/stack.go index fda3dcb35..faa61160e 100644 --- a/pkg/sentry/socket/hostinet/stack.go +++ b/pkg/sentry/socket/hostinet/stack.go @@ -30,6 +30,9 @@ import ( "gvisor.dev/gvisor/pkg/sentry/inet" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/tcpip" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv4" + "gvisor.dev/gvisor/pkg/tcpip/network/ipv6" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -59,6 +62,8 @@ type Stack struct { tcpSACKEnabled bool netDevFile *os.File netSNMPFile *os.File + ipv4Forwarding bool + ipv6Forwarding bool } // NewStack returns an empty Stack containing no configuration. @@ -118,6 +123,13 @@ func (s *Stack) Configure() error { s.netSNMPFile = f } + s.ipv6Forwarding = false + if ipForwarding, err := ioutil.ReadFile("/proc/sys/net/ipv6/conf/all/forwarding"); err == nil { + s.ipv6Forwarding = strings.TrimSpace(string(ipForwarding)) != "0" + } else { + log.Warningf("Failed to read if ipv6 forwarding is enabled, setting to false") + } + return nil } @@ -468,3 +480,21 @@ func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { return nil } // RestoreCleanupEndpoints implements inet.Stack.RestoreCleanupEndpoints. func (s *Stack) RestoreCleanupEndpoints([]stack.TransportEndpoint) {} + +// Forwarding implements inet.Stack.Forwarding. +func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { + switch protocol { + case ipv4.ProtocolNumber: + return s.ipv4Forwarding + case ipv6.ProtocolNumber: + return s.ipv6Forwarding + default: + log.Warningf("Forwarding(%v) failed: unsupported protocol", protocol) + return false + } +} + +// SetForwarding implements inet.Stack.SetForwarding. +func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error { + return syserror.EACCES +} diff --git a/pkg/sentry/socket/netfilter/extensions.go b/pkg/sentry/socket/netfilter/extensions.go index 0336a32d8..549787955 100644 --- a/pkg/sentry/socket/netfilter/extensions.go +++ b/pkg/sentry/socket/netfilter/extensions.go @@ -19,6 +19,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -37,7 +39,7 @@ type matchMaker interface { // name is the matcher name as stored in the xt_entry_match struct. name() string - // marshal converts from an stack.Matcher to an ABI struct. + // marshal converts from a stack.Matcher to an ABI struct. marshal(matcher stack.Matcher) []byte // unmarshal converts from the ABI matcher struct to an @@ -93,3 +95,71 @@ func unmarshalMatcher(match linux.XTEntryMatch, filter stack.IPHeaderFilter, buf } return matchMaker.unmarshal(buf, filter) } + +// targetMaker knows how to (un)marshal a target. Once registered, +// marshalTarget and unmarshalTarget can be used. +type targetMaker interface { + // id uniquely identifies the target. + id() stack.TargetID + + // marshal converts from a stack.Target to an ABI struct. + marshal(target stack.Target) []byte + + // unmarshal converts from the ABI matcher struct to a stack.Target. + unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) +} + +// targetMakers maps the TargetID of supported targets to the targetMaker that +// marshals and unmarshals it. It is immutable after package initialization. +var targetMakers = map[stack.TargetID]targetMaker{} + +func targetRevision(name string, netProto tcpip.NetworkProtocolNumber, rev uint8) (uint8, bool) { + tid := stack.TargetID{ + Name: name, + NetworkProtocol: netProto, + Revision: rev, + } + if _, ok := targetMakers[tid]; !ok { + return 0, false + } + + // Return the highest supported revision unless rev is higher. + for _, other := range targetMakers { + otherID := other.id() + if name == otherID.Name && netProto == otherID.NetworkProtocol && otherID.Revision > rev { + rev = uint8(otherID.Revision) + } + } + return rev, true +} + +// registerTargetMaker should be called by target extensions to register them +// with the netfilter package. +func registerTargetMaker(tm targetMaker) { + if _, ok := targetMakers[tm.id()]; ok { + panic(fmt.Sprintf("multiple targets registered with name %q.", tm.id())) + } + targetMakers[tm.id()] = tm +} + +func marshalTarget(target stack.Target) []byte { + targetMaker, ok := targetMakers[target.ID()] + if !ok { + panic(fmt.Sprintf("unknown target of type %T with id %+v.", target, target.ID())) + } + return targetMaker.marshal(target) +} + +func unmarshalTarget(target linux.XTEntryTarget, filter stack.IPHeaderFilter, buf []byte) (stack.Target, *syserr.Error) { + tid := stack.TargetID{ + Name: target.Name.String(), + NetworkProtocol: filter.NetworkProtocol(), + Revision: target.Revision, + } + targetMaker, ok := targetMakers[tid] + if !ok { + nflog("unsupported target with name %q", target.Name.String()) + return nil, syserr.ErrInvalidArgument + } + return targetMaker.unmarshal(buf, filter) +} diff --git a/pkg/sentry/socket/netfilter/ipv4.go b/pkg/sentry/socket/netfilter/ipv4.go index e4c55a100..b560fae0d 100644 --- a/pkg/sentry/socket/netfilter/ipv4.go +++ b/pkg/sentry/socket/netfilter/ipv4.go @@ -181,18 +181,23 @@ func modifyEntries4(stk *stack.Stack, optVal []byte, replace *linux.IPTReplace, nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return nil, syserr.ErrInvalidArgument } - target, err := parseTarget(filter, optVal[:targetSize]) - if err != nil { - nflog("failed to parse target: %v", err) - return nil, syserr.ErrInvalidArgument - } - optVal = optVal[targetSize:] - table.Rules = append(table.Rules, stack.Rule{ + rule := stack.Rule{ Filter: filter, - Target: target, Matchers: matchers, - }) + } + + { + target, err := parseTarget(filter, optVal[:targetSize], false /* ipv6 */) + if err != nil { + nflog("failed to parse target: %v", err) + return nil, err + } + rule.Target = target + } + optVal = optVal[targetSize:] + + table.Rules = append(table.Rules, rule) offsets[offset] = int(entryIdx) offset += uint32(entry.NextOffset) diff --git a/pkg/sentry/socket/netfilter/ipv6.go b/pkg/sentry/socket/netfilter/ipv6.go index 3b2c1becd..4253f7bf4 100644 --- a/pkg/sentry/socket/netfilter/ipv6.go +++ b/pkg/sentry/socket/netfilter/ipv6.go @@ -184,18 +184,23 @@ func modifyEntries6(stk *stack.Stack, optVal []byte, replace *linux.IPTReplace, nflog("entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) return nil, syserr.ErrInvalidArgument } - target, err := parseTarget(filter, optVal[:targetSize]) - if err != nil { - nflog("failed to parse target: %v", err) - return nil, syserr.ErrInvalidArgument - } - optVal = optVal[targetSize:] - table.Rules = append(table.Rules, stack.Rule{ + rule := stack.Rule{ Filter: filter, - Target: target, Matchers: matchers, - }) + } + + { + target, err := parseTarget(filter, optVal[:targetSize], true /* ipv6 */) + if err != nil { + nflog("failed to parse target: %v", err) + return nil, err + } + rule.Target = target + } + optVal = optVal[targetSize:] + + table.Rules = append(table.Rules, rule) offsets[offset] = int(entryIdx) offset += uint32(entry.NextOffset) diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 3e1735079..904a12e38 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/syserr" + "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) @@ -195,7 +196,7 @@ func SetEntries(stk *stack.Stack, optVal []byte, ipv6 bool) *syserr.Error { // Check the user chains. for ruleIdx, rule := range table.Rules { - if _, ok := rule.Target.(stack.UserChainTarget); !ok { + if _, ok := rule.Target.(*stack.UserChainTarget); !ok { continue } @@ -216,7 +217,7 @@ func SetEntries(stk *stack.Stack, optVal []byte, ipv6 bool) *syserr.Error { // Set each jump to point to the appropriate rule. Right now they hold byte // offsets. for ruleIdx, rule := range table.Rules { - jump, ok := rule.Target.(JumpTarget) + jump, ok := rule.Target.(*JumpTarget) if !ok { continue } @@ -307,7 +308,7 @@ func validUnderflow(rule stack.Rule, ipv6 bool) bool { return false } switch rule.Target.(type) { - case stack.AcceptTarget, stack.DropTarget: + case *stack.AcceptTarget, *stack.DropTarget: return true default: return false @@ -318,7 +319,7 @@ func isUnconditionalAccept(rule stack.Rule, ipv6 bool) bool { if !validUnderflow(rule, ipv6) { return false } - _, ok := rule.Target.(stack.AcceptTarget) + _, ok := rule.Target.(*stack.AcceptTarget) return ok } @@ -337,3 +338,20 @@ func hookFromLinux(hook int) stack.Hook { } panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) } + +// TargetRevision returns a linux.XTGetRevision for a given target. It sets +// Revision to the highest supported value, unless the provided revision number +// is larger. +func TargetRevision(t *kernel.Task, revPtr usermem.Addr, netProto tcpip.NetworkProtocolNumber) (linux.XTGetRevision, *syserr.Error) { + // Read in the target name and version. + var rev linux.XTGetRevision + if _, err := rev.CopyIn(t, revPtr); err != nil { + return linux.XTGetRevision{}, syserr.FromError(err) + } + maxSupported, ok := targetRevision(rev.Name.String(), netProto, rev.Revision) + if !ok { + return linux.XTGetRevision{}, syserr.ErrProtocolNotSupported + } + rev.Revision = maxSupported + return rev, nil +} diff --git a/pkg/sentry/socket/netfilter/targets.go b/pkg/sentry/socket/netfilter/targets.go index 87e41abd8..0e14447fe 100644 --- a/pkg/sentry/socket/netfilter/targets.go +++ b/pkg/sentry/socket/netfilter/targets.go @@ -15,255 +15,357 @@ package netfilter import ( - "errors" "fmt" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/header" "gvisor.dev/gvisor/pkg/tcpip/stack" "gvisor.dev/gvisor/pkg/usermem" ) -// errorTargetName is used to mark targets as error targets. Error targets -// shouldn't be reached - an error has occurred if we fall through to one. -const errorTargetName = "ERROR" +func init() { + // Standard targets include ACCEPT, DROP, RETURN, and JUMP. + registerTargetMaker(&standardTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&standardTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) + + // Both user chains and actual errors are represented in iptables by + // error targets. + registerTargetMaker(&errorTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&errorTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) + + registerTargetMaker(&redirectTargetMaker{ + NetworkProtocol: header.IPv4ProtocolNumber, + }) + registerTargetMaker(&nfNATTargetMaker{ + NetworkProtocol: header.IPv6ProtocolNumber, + }) +} -// redirectTargetName is used to mark targets as redirect targets. Redirect -// targets should be reached for only NAT and Mangle tables. These targets will -// change the destination port/destination IP for packets. -const redirectTargetName = "REDIRECT" +type standardTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} -func marshalTarget(target stack.Target) []byte { +func (sm *standardTargetMaker) id() stack.TargetID { + // Standard targets have the empty string as a name and no revisions. + return stack.TargetID{ + NetworkProtocol: sm.NetworkProtocol, + } +} +func (*standardTargetMaker) marshal(target stack.Target) []byte { + // Translate verdicts the same way as the iptables tool. + var verdict int32 switch tg := target.(type) { - case stack.AcceptTarget: - return marshalStandardTarget(stack.RuleAccept) - case stack.DropTarget: - return marshalStandardTarget(stack.RuleDrop) - case stack.ErrorTarget: - return marshalErrorTarget(errorTargetName) - case stack.UserChainTarget: - return marshalErrorTarget(tg.Name) - case stack.ReturnTarget: - return marshalStandardTarget(stack.RuleReturn) - case stack.RedirectTarget: - return marshalRedirectTarget(tg) - case JumpTarget: - return marshalJumpTarget(tg) + case *stack.AcceptTarget: + verdict = -linux.NF_ACCEPT - 1 + case *stack.DropTarget: + verdict = -linux.NF_DROP - 1 + case *stack.ReturnTarget: + verdict = linux.NF_RETURN + case *JumpTarget: + verdict = int32(tg.Offset) default: panic(fmt.Errorf("unknown target of type %T", target)) } -} - -func marshalStandardTarget(verdict stack.RuleVerdict) []byte { - nflog("convert to binary: marshalling standard target") // The target's name will be the empty string. - target := linux.XTStandardTarget{ + xt := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTStandardTarget, }, - Verdict: translateFromStandardVerdict(verdict), + Verdict: verdict, } ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + return binary.Marshal(ret, usermem.ByteOrder, xt) +} + +func (*standardTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) != linux.SizeOfXTStandardTarget { + nflog("buf has wrong size for standard target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + var standardTarget linux.XTStandardTarget + buf = buf[:linux.SizeOfXTStandardTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) + + if standardTarget.Verdict < 0 { + // A Verdict < 0 indicates a non-jump verdict. + return translateToStandardTarget(standardTarget.Verdict, filter.NetworkProtocol()) + } + // A verdict >= 0 indicates a jump. + return &JumpTarget{ + Offset: uint32(standardTarget.Verdict), + NetworkProtocol: filter.NetworkProtocol(), + }, nil +} + +type errorTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (em *errorTargetMaker) id() stack.TargetID { + // Error targets have no revision. + return stack.TargetID{ + Name: stack.ErrorTargetName, + NetworkProtocol: em.NetworkProtocol, + } } -func marshalErrorTarget(errorName string) []byte { +func (*errorTargetMaker) marshal(target stack.Target) []byte { + var errorName string + switch tg := target.(type) { + case *stack.ErrorTarget: + errorName = stack.ErrorTargetName + case *stack.UserChainTarget: + errorName = tg.Name + default: + panic(fmt.Sprintf("errorMakerTarget cannot marshal unknown type %T", target)) + } + // This is an error target named error - target := linux.XTErrorTarget{ + xt := linux.XTErrorTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTErrorTarget, }, } - copy(target.Name[:], errorName) - copy(target.Target.Name[:], errorTargetName) + copy(xt.Name[:], errorName) + copy(xt.Target.Name[:], stack.ErrorTargetName) ret := make([]byte, 0, linux.SizeOfXTErrorTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + return binary.Marshal(ret, usermem.ByteOrder, xt) } -func marshalRedirectTarget(rt stack.RedirectTarget) []byte { +func (*errorTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) != linux.SizeOfXTErrorTarget { + nflog("buf has insufficient size for error target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + var errorTarget linux.XTErrorTarget + buf = buf[:linux.SizeOfXTErrorTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) + + // Error targets are used in 2 cases: + // * An actual error case. These rules have an error + // named stack.ErrorTargetName. The last entry of the table + // is usually an error case to catch any packets that + // somehow fall through every rule. + // * To mark the start of a user defined chain. These + // rules have an error with the name of the chain. + switch name := errorTarget.Name.String(); name { + case stack.ErrorTargetName: + return &stack.ErrorTarget{NetworkProtocol: filter.NetworkProtocol()}, nil + default: + // User defined chain. + return &stack.UserChainTarget{ + Name: name, + NetworkProtocol: filter.NetworkProtocol(), + }, nil + } +} + +type redirectTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (rm *redirectTargetMaker) id() stack.TargetID { + return stack.TargetID{ + Name: stack.RedirectTargetName, + NetworkProtocol: rm.NetworkProtocol, + } +} + +func (*redirectTargetMaker) marshal(target stack.Target) []byte { + rt := target.(*stack.RedirectTarget) // This is a redirect target named redirect - target := linux.XTRedirectTarget{ + xt := linux.XTRedirectTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTRedirectTarget, }, } - copy(target.Target.Name[:], redirectTargetName) + copy(xt.Target.Name[:], stack.RedirectTargetName) ret := make([]byte, 0, linux.SizeOfXTRedirectTarget) - target.NfRange.RangeSize = 1 - if rt.RangeProtoSpecified { - target.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + xt.NfRange.RangeSize = 1 + xt.NfRange.RangeIPV4.Flags |= linux.NF_NAT_RANGE_PROTO_SPECIFIED + xt.NfRange.RangeIPV4.MinPort = htons(rt.Port) + xt.NfRange.RangeIPV4.MaxPort = xt.NfRange.RangeIPV4.MinPort + return binary.Marshal(ret, usermem.ByteOrder, xt) +} + +func (*redirectTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if len(buf) < linux.SizeOfXTRedirectTarget { + nflog("redirectTargetMaker: buf has insufficient size for redirect target %d", len(buf)) + return nil, syserr.ErrInvalidArgument + } + + if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { + nflog("redirectTargetMaker: bad proto %d", p) + return nil, syserr.ErrInvalidArgument + } + + var redirectTarget linux.XTRedirectTarget + buf = buf[:linux.SizeOfXTRedirectTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) + + // Copy linux.XTRedirectTarget to stack.RedirectTarget. + target := stack.RedirectTarget{NetworkProtocol: filter.NetworkProtocol()} + + // RangeSize should be 1. + nfRange := redirectTarget.NfRange + if nfRange.RangeSize != 1 { + nflog("redirectTargetMaker: bad rangesize %d", nfRange.RangeSize) + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): Check if the flags are valid. + // Also check if we need to map ports or IP. + // For now, redirect target only supports destination port change. + // Port range and IP range are not supported yet. + if nfRange.RangeIPV4.Flags != linux.NF_NAT_RANGE_PROTO_SPECIFIED { + nflog("redirectTargetMaker: invalid range flags %d", nfRange.RangeIPV4.Flags) + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): Port range is not supported yet. + if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { + nflog("redirectTargetMaker: MinPort != MaxPort (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) + return nil, syserr.ErrInvalidArgument } - // Convert port from little endian to big endian. - port := make([]byte, 2) - binary.LittleEndian.PutUint16(port, rt.MinPort) - target.NfRange.RangeIPV4.MinPort = binary.BigEndian.Uint16(port) - binary.LittleEndian.PutUint16(port, rt.MaxPort) - target.NfRange.RangeIPV4.MaxPort = binary.BigEndian.Uint16(port) - return binary.Marshal(ret, usermem.ByteOrder, target) + if nfRange.RangeIPV4.MinIP != nfRange.RangeIPV4.MaxIP { + nflog("redirectTargetMaker: MinIP != MaxIP (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) + return nil, syserr.ErrInvalidArgument + } + + target.Addr = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) + target.Port = ntohs(nfRange.RangeIPV4.MinPort) + + return &target, nil } -func marshalJumpTarget(jt JumpTarget) []byte { - nflog("convert to binary: marshalling jump target") +type nfNATTarget struct { + Target linux.XTEntryTarget + Range linux.NFNATRange +} - // The target's name will be the empty string. - target := linux.XTStandardTarget{ +const nfNATMarhsalledSize = linux.SizeOfXTEntryTarget + linux.SizeOfNFNATRange + +type nfNATTargetMaker struct { + NetworkProtocol tcpip.NetworkProtocolNumber +} + +func (rm *nfNATTargetMaker) id() stack.TargetID { + return stack.TargetID{ + Name: stack.RedirectTargetName, + NetworkProtocol: rm.NetworkProtocol, + } +} + +func (*nfNATTargetMaker) marshal(target stack.Target) []byte { + rt := target.(*stack.RedirectTarget) + nt := nfNATTarget{ Target: linux.XTEntryTarget{ - TargetSize: linux.SizeOfXTStandardTarget, + TargetSize: nfNATMarhsalledSize, + }, + Range: linux.NFNATRange{ + Flags: linux.NF_NAT_RANGE_PROTO_SPECIFIED, }, - // Verdict is overloaded by the ABI. When positive, it holds - // the jump offset from the start of the table. - Verdict: int32(jt.Offset), } + copy(nt.Target.Name[:], stack.RedirectTargetName) + copy(nt.Range.MinAddr[:], rt.Addr) + copy(nt.Range.MaxAddr[:], rt.Addr) - ret := make([]byte, 0, linux.SizeOfXTStandardTarget) - return binary.Marshal(ret, usermem.ByteOrder, target) + nt.Range.MinProto = htons(rt.Port) + nt.Range.MaxProto = nt.Range.MinProto + + ret := make([]byte, 0, nfNATMarhsalledSize) + return binary.Marshal(ret, usermem.ByteOrder, nt) } -// translateFromStandardVerdict translates verdicts the same way as the iptables -// tool. -func translateFromStandardVerdict(verdict stack.RuleVerdict) int32 { - switch verdict { - case stack.RuleAccept: - return -linux.NF_ACCEPT - 1 - case stack.RuleDrop: - return -linux.NF_DROP - 1 - case stack.RuleReturn: - return linux.NF_RETURN - default: - // TODO(gvisor.dev/issue/170): Support Jump. - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) +func (*nfNATTargetMaker) unmarshal(buf []byte, filter stack.IPHeaderFilter) (stack.Target, *syserr.Error) { + if size := nfNATMarhsalledSize; len(buf) < size { + nflog("nfNATTargetMaker: buf has insufficient size (%d) for nfNAT target (%d)", len(buf), size) + return nil, syserr.ErrInvalidArgument } + + if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { + nflog("nfNATTargetMaker: bad proto %d", p) + return nil, syserr.ErrInvalidArgument + } + + var natRange linux.NFNATRange + buf = buf[linux.SizeOfXTEntryTarget:nfNATMarhsalledSize] + binary.Unmarshal(buf, usermem.ByteOrder, &natRange) + + // We don't support port or address ranges. + if natRange.MinAddr != natRange.MaxAddr { + nflog("nfNATTargetMaker: MinAddr and MaxAddr are different") + return nil, syserr.ErrInvalidArgument + } + if natRange.MinProto != natRange.MaxProto { + nflog("nfNATTargetMaker: MinProto and MaxProto are different") + return nil, syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/3549): Check for other flags. + // For now, redirect target only supports destination change. + if natRange.Flags != linux.NF_NAT_RANGE_PROTO_SPECIFIED { + nflog("nfNATTargetMaker: invalid range flags %d", natRange.Flags) + return nil, syserr.ErrInvalidArgument + } + + target := stack.RedirectTarget{ + NetworkProtocol: filter.NetworkProtocol(), + Addr: tcpip.Address(natRange.MinAddr[:]), + Port: ntohs(natRange.MinProto), + } + + return &target, nil } // translateToStandardTarget translates from the value in a // linux.XTStandardTarget to an stack.Verdict. -func translateToStandardTarget(val int32) (stack.Target, error) { +func translateToStandardTarget(val int32, netProto tcpip.NetworkProtocolNumber) (stack.Target, *syserr.Error) { // TODO(gvisor.dev/issue/170): Support other verdicts. switch val { case -linux.NF_ACCEPT - 1: - return stack.AcceptTarget{}, nil + return &stack.AcceptTarget{NetworkProtocol: netProto}, nil case -linux.NF_DROP - 1: - return stack.DropTarget{}, nil + return &stack.DropTarget{NetworkProtocol: netProto}, nil case -linux.NF_QUEUE - 1: - return nil, errors.New("unsupported iptables verdict QUEUE") + nflog("unsupported iptables verdict QUEUE") + return nil, syserr.ErrInvalidArgument case linux.NF_RETURN: - return stack.ReturnTarget{}, nil + return &stack.ReturnTarget{NetworkProtocol: netProto}, nil default: - return nil, fmt.Errorf("unknown iptables verdict %d", val) + nflog("unknown iptables verdict %d", val) + return nil, syserr.ErrInvalidArgument } } // parseTarget parses a target from optVal. optVal should contain only the // target. -func parseTarget(filter stack.IPHeaderFilter, optVal []byte) (stack.Target, error) { +func parseTarget(filter stack.IPHeaderFilter, optVal []byte, ipv6 bool) (stack.Target, *syserr.Error) { nflog("set entries: parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { - return nil, fmt.Errorf("optVal has insufficient size for entry target %d", len(optVal)) + nflog("optVal has insufficient size for entry target %d", len(optVal)) + return nil, syserr.ErrInvalidArgument } var target linux.XTEntryTarget buf := optVal[:linux.SizeOfXTEntryTarget] binary.Unmarshal(buf, usermem.ByteOrder, &target) - switch target.Name.String() { - case "": - // Standard target. - if len(optVal) != linux.SizeOfXTStandardTarget { - return nil, fmt.Errorf("optVal has wrong size for standard target %d", len(optVal)) - } - var standardTarget linux.XTStandardTarget - buf = optVal[:linux.SizeOfXTStandardTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) - - if standardTarget.Verdict < 0 { - // A Verdict < 0 indicates a non-jump verdict. - return translateToStandardTarget(standardTarget.Verdict) - } - // A verdict >= 0 indicates a jump. - return JumpTarget{Offset: uint32(standardTarget.Verdict)}, nil - - case errorTargetName: - // Error target. - if len(optVal) != linux.SizeOfXTErrorTarget { - return nil, fmt.Errorf("optVal has insufficient size for error target %d", len(optVal)) - } - var errorTarget linux.XTErrorTarget - buf = optVal[:linux.SizeOfXTErrorTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &errorTarget) - - // Error targets are used in 2 cases: - // * An actual error case. These rules have an error - // named errorTargetName. The last entry of the table - // is usually an error case to catch any packets that - // somehow fall through every rule. - // * To mark the start of a user defined chain. These - // rules have an error with the name of the chain. - switch name := errorTarget.Name.String(); name { - case errorTargetName: - nflog("set entries: error target") - return stack.ErrorTarget{}, nil - default: - // User defined chain. - nflog("set entries: user-defined target %q", name) - return stack.UserChainTarget{Name: name}, nil - } - - case redirectTargetName: - // Redirect target. - if len(optVal) < linux.SizeOfXTRedirectTarget { - return nil, fmt.Errorf("netfilter.SetEntries: optVal has insufficient size for redirect target %d", len(optVal)) - } - - if p := filter.Protocol; p != header.TCPProtocolNumber && p != header.UDPProtocolNumber { - return nil, fmt.Errorf("netfilter.SetEntries: bad proto %d", p) - } - - var redirectTarget linux.XTRedirectTarget - buf = optVal[:linux.SizeOfXTRedirectTarget] - binary.Unmarshal(buf, usermem.ByteOrder, &redirectTarget) - - // Copy linux.XTRedirectTarget to stack.RedirectTarget. - var target stack.RedirectTarget - nfRange := redirectTarget.NfRange - - // RangeSize should be 1. - if nfRange.RangeSize != 1 { - return nil, fmt.Errorf("netfilter.SetEntries: bad rangesize %d", nfRange.RangeSize) - } - - // TODO(gvisor.dev/issue/170): Check if the flags are valid. - // Also check if we need to map ports or IP. - // For now, redirect target only supports destination port change. - // Port range and IP range are not supported yet. - if nfRange.RangeIPV4.Flags&linux.NF_NAT_RANGE_PROTO_SPECIFIED == 0 { - return nil, fmt.Errorf("netfilter.SetEntries: invalid range flags %d", nfRange.RangeIPV4.Flags) - } - target.RangeProtoSpecified = true - - target.MinIP = tcpip.Address(nfRange.RangeIPV4.MinIP[:]) - target.MaxIP = tcpip.Address(nfRange.RangeIPV4.MaxIP[:]) - - // TODO(gvisor.dev/issue/170): Port range is not supported yet. - if nfRange.RangeIPV4.MinPort != nfRange.RangeIPV4.MaxPort { - return nil, fmt.Errorf("netfilter.SetEntries: minport != maxport (%d, %d)", nfRange.RangeIPV4.MinPort, nfRange.RangeIPV4.MaxPort) - } - - // Convert port from big endian to little endian. - port := make([]byte, 2) - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MinPort) - target.MinPort = binary.LittleEndian.Uint16(port) - - binary.BigEndian.PutUint16(port, nfRange.RangeIPV4.MaxPort) - target.MaxPort = binary.LittleEndian.Uint16(port) - return target, nil - } - // Unknown target. - return nil, fmt.Errorf("unknown target %q doesn't exist or isn't supported yet", target.Name.String()) + return unmarshalTarget(target, filter, optVal) } // JumpTarget implements stack.Target. @@ -274,9 +376,31 @@ type JumpTarget struct { // RuleNum is the rule to jump to. RuleNum int + + // NetworkProtocol is the network protocol the target is used with. + NetworkProtocol tcpip.NetworkProtocolNumber +} + +// ID implements Target.ID. +func (jt *JumpTarget) ID() stack.TargetID { + return stack.TargetID{ + NetworkProtocol: jt.NetworkProtocol, + } } // Action implements stack.Target.Action. -func (jt JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { +func (jt *JumpTarget) Action(*stack.PacketBuffer, *stack.ConnTrack, stack.Hook, *stack.GSO, *stack.Route, tcpip.Address) (stack.RuleVerdict, int) { return stack.RuleJump, jt.RuleNum } + +func ntohs(port uint16) uint16 { + buf := make([]byte, 2) + binary.BigEndian.PutUint16(buf, port) + return usermem.ByteOrder.Uint16(buf) +} + +func htons(port uint16) uint16 { + buf := make([]byte, 2) + usermem.ByteOrder.PutUint16(buf, port) + return binary.BigEndian.Uint16(buf) +} diff --git a/pkg/sentry/socket/netfilter/tcp_matcher.go b/pkg/sentry/socket/netfilter/tcp_matcher.go index 0bfd6c1f4..844acfede 100644 --- a/pkg/sentry/socket/netfilter/tcp_matcher.go +++ b/pkg/sentry/socket/netfilter/tcp_matcher.go @@ -97,17 +97,33 @@ func (*TCPMatcher) Name() string { // Match implements Matcher.Match. func (tm *TCPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader().View()) + // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved + // into the stack.Check codepath as matchers are added. + switch pkt.NetworkProtocolNumber { + case header.IPv4ProtocolNumber: + netHeader := header.IPv4(pkt.NetworkHeader().View()) + if netHeader.TransportProtocol() != header.TCPProtocolNumber { + return false, false + } - if netHeader.TransportProtocol() != header.TCPProtocolNumber { - return false, false - } + // We don't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + if frag == 1 { + return false, true + } + return false, false + } - // We dont't match fragments. - if frag := netHeader.FragmentOffset(); frag != 0 { - if frag == 1 { - return false, true + case header.IPv6ProtocolNumber: + // As in Linux, we do not perform an IPv6 fragment check. See + // xt_action_param.fragoff in + // include/linux/netfilter/x_tables.h. + if header.IPv6(pkt.NetworkHeader().View()).TransportProtocol() != header.TCPProtocolNumber { + return false, false } + + default: + // We don't know the network protocol. return false, false } diff --git a/pkg/sentry/socket/netfilter/udp_matcher.go b/pkg/sentry/socket/netfilter/udp_matcher.go index 7ed05461d..63201201c 100644 --- a/pkg/sentry/socket/netfilter/udp_matcher.go +++ b/pkg/sentry/socket/netfilter/udp_matcher.go @@ -94,19 +94,33 @@ func (*UDPMatcher) Name() string { // Match implements Matcher.Match. func (um *UDPMatcher) Match(hook stack.Hook, pkt *stack.PacketBuffer, interfaceName string) (bool, bool) { - netHeader := header.IPv4(pkt.NetworkHeader().View()) - // TODO(gvisor.dev/issue/170): Proto checks should ultimately be moved // into the stack.Check codepath as matchers are added. - if netHeader.TransportProtocol() != header.UDPProtocolNumber { - return false, false - } + switch pkt.NetworkProtocolNumber { + case header.IPv4ProtocolNumber: + netHeader := header.IPv4(pkt.NetworkHeader().View()) + if netHeader.TransportProtocol() != header.UDPProtocolNumber { + return false, false + } - // We dont't match fragments. - if frag := netHeader.FragmentOffset(); frag != 0 { - if frag == 1 { - return false, true + // We don't match fragments. + if frag := netHeader.FragmentOffset(); frag != 0 { + if frag == 1 { + return false, true + } + return false, false } + + case header.IPv6ProtocolNumber: + // As in Linux, we do not perform an IPv6 fragment check. See + // xt_action_param.fragoff in + // include/linux/netfilter/x_tables.h. + if header.IPv6(pkt.NetworkHeader().View()).TransportProtocol() != header.UDPProtocolNumber { + return false, false + } + + default: + // We don't know the network protocol. return false, false } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 0546801bf..1f926aa91 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -16,6 +16,8 @@ go_library( "//pkg/abi/linux", "//pkg/binary", "//pkg/context", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/sentry/arch", "//pkg/sentry/device", "//pkg/sentry/fs", @@ -36,8 +38,6 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", - "//tools/go_marshal/primitive", ], ) diff --git a/pkg/sentry/socket/netlink/provider_vfs2.go b/pkg/sentry/socket/netlink/provider_vfs2.go index bb205be0d..e8930f031 100644 --- a/pkg/sentry/socket/netlink/provider_vfs2.go +++ b/pkg/sentry/socket/netlink/provider_vfs2.go @@ -52,6 +52,7 @@ func (*socketProviderVFS2) Socket(t *kernel.Task, stype linux.SockType, protocol vfsfd := &s.vfsfd mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) if err := vfsfd.Init(s, linux.O_RDWR, mnt, d, &vfs.FileDescriptionOptions{ DenyPRead: true, DenyPWrite: true, diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 68a9b9a96..5ddcd4be5 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -21,6 +21,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" @@ -38,8 +40,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) const sizeOfInt32 int = 4 diff --git a/pkg/sentry/socket/netlink/socket_vfs2.go b/pkg/sentry/socket/netlink/socket_vfs2.go index a38d25da9..c83b23242 100644 --- a/pkg/sentry/socket/netlink/socket_vfs2.go +++ b/pkg/sentry/socket/netlink/socket_vfs2.go @@ -82,6 +82,13 @@ func NewVFS2(t *kernel.Task, skType linux.SockType, protocol Protocol) (*SocketV return fd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index 1fb777a6c..fae3b6783 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -22,6 +22,8 @@ go_library( "//pkg/binary", "//pkg/context", "//pkg/log", + "//pkg/marshal", + "//pkg/marshal/primitive", "//pkg/metric", "//pkg/safemem", "//pkg/sentry/arch", @@ -51,8 +53,6 @@ go_library( "//pkg/tcpip/transport/udp", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", - "//tools/go_marshal/primitive", "@org_golang_x_sys//unix:go_default_library", ], ) diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 0bf21f7d8..87e30d742 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -40,6 +40,8 @@ import ( "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/metric" "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" @@ -62,8 +64,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip/transport/udp" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) func mustCreateMetric(name, description string) *tcpip.StatCounter { @@ -158,6 +158,9 @@ var Metrics = tcpip.Stats{ OutgoingPacketErrors: mustCreateMetric("/netstack/ip/outgoing_packet_errors", "Total number of IP packets which failed to write to a link-layer endpoint."), MalformedPacketsReceived: mustCreateMetric("/netstack/ip/malformed_packets_received", "Total number of IP packets which failed IP header validation checks."), MalformedFragmentsReceived: mustCreateMetric("/netstack/ip/malformed_fragments_received", "Total number of IP fragments which failed IP fragment validation checks."), + IPTablesPreroutingDropped: mustCreateMetric("/netstack/ip/iptables/prerouting_dropped", "Total number of IP packets dropped in the Prerouting chain."), + IPTablesInputDropped: mustCreateMetric("/netstack/ip/iptables/input_dropped", "Total number of IP packets dropped in the Input chain."), + IPTablesOutputDropped: mustCreateMetric("/netstack/ip/iptables/output_dropped", "Total number of IP packets dropped in the Output chain."), }, TCP: tcpip.TCPStats{ ActiveConnectionOpenings: mustCreateMetric("/netstack/tcp/active_connection_openings", "Number of connections opened successfully via Connect."), @@ -195,7 +198,6 @@ var Metrics = tcpip.Stats{ PacketsSent: mustCreateMetric("/netstack/udp/packets_sent", "Number of UDP datagrams sent."), PacketSendErrors: mustCreateMetric("/netstack/udp/packet_send_errors", "Number of UDP datagrams failed to be sent."), ChecksumErrors: mustCreateMetric("/netstack/udp/checksum_errors", "Number of UDP datagrams dropped due to bad checksums."), - InvalidSourceAddress: mustCreateMetric("/netstack/udp/invalid_source", "Number of UDP datagrams dropped due to invalid source address."), }, } @@ -857,7 +859,7 @@ func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { // blockingAccept implements a blocking version of accept(2), that is, if no // connections are ready to be accept, it will block until one becomes ready. -func (s *socketOpsCommon) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *waiter.Queue, *syserr.Error) { +func (s *socketOpsCommon) blockingAccept(t *kernel.Task, peerAddr *tcpip.FullAddress) (tcpip.Endpoint, *waiter.Queue, *syserr.Error) { // Register for notifications. e, ch := waiter.NewChannelEntry(nil) s.EventRegister(&e, waiter.EventIn) @@ -866,7 +868,7 @@ func (s *socketOpsCommon) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *waite // Try to accept the connection again; if it fails, then wait until we // get a notification. for { - if ep, wq, err := s.Endpoint.Accept(); err != tcpip.ErrWouldBlock { + if ep, wq, err := s.Endpoint.Accept(peerAddr); err != tcpip.ErrWouldBlock { return ep, wq, syserr.TranslateNetstackError(err) } @@ -879,15 +881,18 @@ func (s *socketOpsCommon) blockingAccept(t *kernel.Task) (tcpip.Endpoint, *waite // Accept implements the linux syscall accept(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { - // Issue the accept request to get the new endpoint. - ep, wq, terr := s.Endpoint.Accept() + var peerAddr *tcpip.FullAddress + if peerRequested { + peerAddr = &tcpip.FullAddress{} + } + ep, wq, terr := s.Endpoint.Accept(peerAddr) if terr != nil { if terr != tcpip.ErrWouldBlock || !blocking { return 0, nil, 0, syserr.TranslateNetstackError(terr) } var err *syserr.Error - ep, wq, err = s.blockingAccept(t) + ep, wq, err = s.blockingAccept(t, peerAddr) if err != nil { return 0, nil, 0, err } @@ -907,13 +912,8 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, var addr linux.SockAddr var addrLen uint32 - if peerRequested { - // Get address of the peer and write it to peer slice. - var err *syserr.Error - addr, addrLen, err = ns.FileOperations.(*SocketOperations).GetPeerName(t) - if err != nil { - return 0, nil, 0, err - } + if peerAddr != nil { + addr, addrLen = ConvertAddress(s.family, *peerAddr) } fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ @@ -1187,7 +1187,7 @@ func getSockOptSocket(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, fam var v tcpip.LingerOption var linger linux.Linger if err := ep.GetSockOpt(&v); err != nil { - return &linger, nil + return nil, syserr.TranslateNetstackError(err) } if v.Enabled { @@ -1512,8 +1512,17 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name return &vP, nil case linux.IP6T_ORIGINAL_DST: - // TODO(gvisor.dev/issue/170): ip6tables. - return nil, syserr.ErrInvalidArgument + if outLen < int(binary.Size(linux.SockAddrInet6{})) { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.OriginalDestinationOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + a, _ := ConvertAddress(linux.AF_INET6, tcpip.FullAddress(v)) + return a.(*linux.SockAddrInet6), nil case linux.IP6T_SO_GET_INFO: if outLen < linux.SizeOfIPTGetinfo { @@ -1555,6 +1564,26 @@ func getSockOptIPv6(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name } return &entries, nil + case linux.IP6T_SO_GET_REVISION_TARGET: + if outLen < linux.SizeOfXTGetRevision { + return nil, syserr.ErrInvalidArgument + } + + // Only valid for raw IPv6 sockets. + if family, skType, _ := s.Type(); family != linux.AF_INET6 || skType != linux.SOCK_RAW { + return nil, syserr.ErrProtocolNotAvailable + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + ret, err := netfilter.TargetRevision(t, outPtr, header.IPv6ProtocolNumber) + if err != nil { + return nil, err + } + return &ret, nil + default: emitUnimplementedEventIPv6(t, name) } @@ -1718,6 +1747,26 @@ func getSockOptIP(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, name in } return &entries, nil + case linux.IPT_SO_GET_REVISION_TARGET: + if outLen < linux.SizeOfXTGetRevision { + return nil, syserr.ErrInvalidArgument + } + + // Only valid for raw IPv4 sockets. + if family, skType, _ := s.Type(); family != linux.AF_INET || skType != linux.SOCK_RAW { + return nil, syserr.ErrProtocolNotAvailable + } + + stack := inet.StackFromContext(t) + if stack == nil { + return nil, syserr.ErrNoDevice + } + ret, err := netfilter.TargetRevision(t, outPtr, header.IPv4ProtocolNumber) + if err != nil { + return nil, err + } + return &ret, nil + default: emitUnimplementedEventIP(t, name) } @@ -1770,10 +1819,16 @@ func SetSockOpt(t *kernel.Task, s socket.SocketOps, ep commonEndpoint, level int case linux.SOL_IP: return setSockOptIP(t, s, ep, name, optVal) + case linux.SOL_PACKET: + // gVisor doesn't support any SOL_PACKET options just return not + // supported. Returning nil here will result in tcpdump thinking AF_PACKET + // features are supported and proceed to use them and break. + t.Kernel().EmitUnimplementedEvent(t) + return syserr.ErrProtocolNotAvailable + case linux.SOL_UDP, linux.SOL_ICMPV6, - linux.SOL_RAW, - linux.SOL_PACKET: + linux.SOL_RAW: t.Kernel().EmitUnimplementedEvent(t) } diff --git a/pkg/sentry/socket/netstack/netstack_vfs2.go b/pkg/sentry/socket/netstack/netstack_vfs2.go index 1f7d17f5f..4c6791fff 100644 --- a/pkg/sentry/socket/netstack/netstack_vfs2.go +++ b/pkg/sentry/socket/netstack/netstack_vfs2.go @@ -18,6 +18,8 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/amutex" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" "gvisor.dev/gvisor/pkg/sentry/arch" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" @@ -29,8 +31,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" - "gvisor.dev/gvisor/tools/go_marshal/primitive" ) // SocketVFS2 encapsulates all the state needed to represent a network stack @@ -79,6 +79,13 @@ func NewVFS2(t *kernel.Task, family int, skType linux.SockType, protocol int, qu return vfsfd, nil } +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.socketOpsCommon.Release(ctx) +} + // Readiness implements waiter.Waitable.Readiness. func (s *SocketVFS2) Readiness(mask waiter.EventMask) waiter.EventMask { return s.socketOpsCommon.Readiness(mask) @@ -151,14 +158,18 @@ func (s *SocketVFS2) Write(ctx context.Context, src usermem.IOSequence, opts vfs // tcpip.Endpoint. func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { // Issue the accept request to get the new endpoint. - ep, wq, terr := s.Endpoint.Accept() + var peerAddr *tcpip.FullAddress + if peerRequested { + peerAddr = &tcpip.FullAddress{} + } + ep, wq, terr := s.Endpoint.Accept(peerAddr) if terr != nil { if terr != tcpip.ErrWouldBlock || !blocking { return 0, nil, 0, syserr.TranslateNetstackError(terr) } var err *syserr.Error - ep, wq, err = s.blockingAccept(t) + ep, wq, err = s.blockingAccept(t, peerAddr) if err != nil { return 0, nil, 0, err } @@ -176,13 +187,9 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block var addr linux.SockAddr var addrLen uint32 - if peerRequested { + if peerAddr != nil { // Get address of the peer and write it to peer slice. - var err *syserr.Error - addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t) - if err != nil { - return 0, nil, 0, err - } + addr, addrLen = ConvertAddress(s.family, *peerAddr) } fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ diff --git a/pkg/sentry/socket/netstack/stack.go b/pkg/sentry/socket/netstack/stack.go index f0fe18684..1028d2a6e 100644 --- a/pkg/sentry/socket/netstack/stack.go +++ b/pkg/sentry/socket/netstack/stack.go @@ -155,7 +155,7 @@ func (s *Stack) AddInterfaceAddr(idx int32, addr inet.InterfaceAddr) error { // TCPReceiveBufferSize implements inet.Stack.TCPReceiveBufferSize. func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { - var rs tcp.ReceiveBufferSizeOption + var rs tcpip.TCPReceiveBufferSizeRangeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &rs) return inet.TCPBufferSize{ Min: rs.Min, @@ -166,17 +166,17 @@ func (s *Stack) TCPReceiveBufferSize() (inet.TCPBufferSize, error) { // SetTCPReceiveBufferSize implements inet.Stack.SetTCPReceiveBufferSize. func (s *Stack) SetTCPReceiveBufferSize(size inet.TCPBufferSize) error { - rs := tcp.ReceiveBufferSizeOption{ + rs := tcpip.TCPReceiveBufferSizeRangeOption{ Min: size.Min, Default: size.Default, Max: size.Max, } - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, rs)).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, &rs)).ToError() } // TCPSendBufferSize implements inet.Stack.TCPSendBufferSize. func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { - var ss tcp.SendBufferSizeOption + var ss tcpip.TCPSendBufferSizeRangeOption err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &ss) return inet.TCPBufferSize{ Min: ss.Min, @@ -187,29 +187,30 @@ func (s *Stack) TCPSendBufferSize() (inet.TCPBufferSize, error) { // SetTCPSendBufferSize implements inet.Stack.SetTCPSendBufferSize. func (s *Stack) SetTCPSendBufferSize(size inet.TCPBufferSize) error { - ss := tcp.SendBufferSizeOption{ + ss := tcpip.TCPSendBufferSizeRangeOption{ Min: size.Min, Default: size.Default, Max: size.Max, } - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, ss)).ToError() + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, &ss)).ToError() } // TCPSACKEnabled implements inet.Stack.TCPSACKEnabled. func (s *Stack) TCPSACKEnabled() (bool, error) { - var sack tcp.SACKEnabled + var sack tcpip.TCPSACKEnabled err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &sack) return bool(sack), syserr.TranslateNetstackError(err).ToError() } // SetTCPSACKEnabled implements inet.Stack.SetTCPSACKEnabled. func (s *Stack) SetTCPSACKEnabled(enabled bool) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.SACKEnabled(enabled))).ToError() + opt := tcpip.TCPSACKEnabled(enabled) + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, &opt)).ToError() } // TCPRecovery implements inet.Stack.TCPRecovery. func (s *Stack) TCPRecovery() (inet.TCPLossRecovery, error) { - var recovery tcp.Recovery + var recovery tcpip.TCPRecovery if err := s.Stack.TransportProtocolOption(tcp.ProtocolNumber, &recovery); err != nil { return 0, syserr.TranslateNetstackError(err).ToError() } @@ -218,7 +219,8 @@ func (s *Stack) TCPRecovery() (inet.TCPLossRecovery, error) { // SetTCPRecovery implements inet.Stack.SetTCPRecovery. func (s *Stack) SetTCPRecovery(recovery inet.TCPLossRecovery) error { - return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, tcp.Recovery(recovery))).ToError() + opt := tcpip.TCPRecovery(recovery) + return syserr.TranslateNetstackError(s.Stack.SetTransportProtocolOption(tcp.ProtocolNumber, &opt)).ToError() } // Statistics implements inet.Stack.Statistics. @@ -410,3 +412,24 @@ func (s *Stack) CleanupEndpoints() []stack.TransportEndpoint { func (s *Stack) RestoreCleanupEndpoints(es []stack.TransportEndpoint) { s.Stack.RestoreCleanupEndpoints(es) } + +// Forwarding implements inet.Stack.Forwarding. +func (s *Stack) Forwarding(protocol tcpip.NetworkProtocolNumber) bool { + switch protocol { + case ipv4.ProtocolNumber, ipv6.ProtocolNumber: + return s.Stack.Forwarding(protocol) + default: + panic(fmt.Sprintf("Forwarding(%v) failed: unsupported protocol", protocol)) + } +} + +// SetForwarding implements inet.Stack.SetForwarding. +func (s *Stack) SetForwarding(protocol tcpip.NetworkProtocolNumber, enable bool) error { + switch protocol { + case ipv4.ProtocolNumber, ipv6.ProtocolNumber: + s.Stack.SetForwarding(protocol, enable) + default: + panic(fmt.Sprintf("SetForwarding(%v) failed: unsupported protocol", protocol)) + } + return nil +} diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 04b259d27..fd31479e5 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -25,6 +25,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/device" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" @@ -35,7 +36,6 @@ import ( "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" - "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // ControlMessages represents the union of unix control messages and tcpip diff --git a/pkg/sentry/socket/unix/BUILD b/pkg/sentry/socket/unix/BUILD index cb953e4dc..cc7408698 100644 --- a/pkg/sentry/socket/unix/BUILD +++ b/pkg/sentry/socket/unix/BUILD @@ -7,10 +7,21 @@ go_template_instance( name = "socket_refs", out = "socket_refs.go", package = "unix", - prefix = "socketOpsCommon", + prefix = "socketOperations", template = "//pkg/refs_vfs2:refs_template", types = { - "T": "socketOpsCommon", + "T": "SocketOperations", + }, +) + +go_template_instance( + name = "socket_vfs2_refs", + out = "socket_vfs2_refs.go", + package = "unix", + prefix = "socketVFS2", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "SocketVFS2", }, ) @@ -20,6 +31,7 @@ go_library( "device.go", "io.go", "socket_refs.go", + "socket_vfs2_refs.go", "unix.go", "unix_vfs2.go", ], @@ -29,6 +41,7 @@ go_library( "//pkg/context", "//pkg/fspath", "//pkg/log", + "//pkg/marshal", "//pkg/refs", "//pkg/safemem", "//pkg/sentry/arch", @@ -49,6 +62,5 @@ go_library( "//pkg/tcpip", "//pkg/usermem", "//pkg/waiter", - "//tools/go_marshal/marshal", ], ) diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index e3a75b519..aa4f3c04d 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -391,7 +391,7 @@ func (e *connectionedEndpoint) Listen(backlog int) *syserr.Error { } // Accept accepts a new connection. -func (e *connectionedEndpoint) Accept() (Endpoint, *syserr.Error) { +func (e *connectionedEndpoint) Accept(peerAddr *tcpip.FullAddress) (Endpoint, *syserr.Error) { e.Lock() defer e.Unlock() @@ -401,6 +401,18 @@ func (e *connectionedEndpoint) Accept() (Endpoint, *syserr.Error) { select { case ne := <-e.acceptedChan: + if peerAddr != nil { + ne.Lock() + c := ne.connected + ne.Unlock() + if c != nil { + addr, err := c.GetLocalAddress() + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + *peerAddr = addr + } + } return ne, nil default: diff --git a/pkg/sentry/socket/unix/transport/connectionless.go b/pkg/sentry/socket/unix/transport/connectionless.go index 4751b2fd8..f8aacca13 100644 --- a/pkg/sentry/socket/unix/transport/connectionless.go +++ b/pkg/sentry/socket/unix/transport/connectionless.go @@ -144,12 +144,12 @@ func (e *connectionlessEndpoint) Connect(ctx context.Context, server BoundEndpoi } // Listen starts listening on the connection. -func (e *connectionlessEndpoint) Listen(int) *syserr.Error { +func (*connectionlessEndpoint) Listen(int) *syserr.Error { return syserr.ErrNotSupported } // Accept accepts a new connection. -func (e *connectionlessEndpoint) Accept() (Endpoint, *syserr.Error) { +func (*connectionlessEndpoint) Accept(*tcpip.FullAddress) (Endpoint, *syserr.Error) { return nil, syserr.ErrNotSupported } diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index cc9d650fb..d6fc03520 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -151,7 +151,10 @@ type Endpoint interface { // block if no new connections are available. // // The returned Queue is the wait queue for the newly created endpoint. - Accept() (Endpoint, *syserr.Error) + // + // peerAddr if not nil will be populated with the address of the connected + // peer on a successful accept. + Accept(peerAddr *tcpip.FullAddress) (Endpoint, *syserr.Error) // Bind binds the endpoint to a specific local address and port. // Specifying a NIC is optional. @@ -743,6 +746,9 @@ type baseEndpoint struct { // path is not empty if the endpoint has been bound, // or may be used if the endpoint is connected. path string + + // linger is used for SO_LINGER socket option. + linger tcpip.LingerOption } // EventRegister implements waiter.Waitable.EventRegister. @@ -838,8 +844,14 @@ func (e *baseEndpoint) SendMsg(ctx context.Context, data [][]byte, c ControlMess return n, err } -// SetSockOpt sets a socket option. Currently not supported. -func (e *baseEndpoint) SetSockOpt(tcpip.SettableSocketOption) *tcpip.Error { +// SetSockOpt sets a socket option. +func (e *baseEndpoint) SetSockOpt(opt tcpip.SettableSocketOption) *tcpip.Error { + switch v := opt.(type) { + case *tcpip.LingerOption: + e.Lock() + e.linger = *v + e.Unlock() + } return nil } @@ -942,8 +954,11 @@ func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { // GetSockOpt implements tcpip.Endpoint.GetSockOpt. func (e *baseEndpoint) GetSockOpt(opt tcpip.GettableSocketOption) *tcpip.Error { - switch opt.(type) { + switch o := opt.(type) { case *tcpip.LingerOption: + e.Lock() + *o = e.linger + e.Unlock() return nil default: diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 0a7a26495..f80011ce4 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -24,6 +24,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/arch" "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/fs/fsutil" @@ -39,7 +40,6 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketOperations is a Unix socket. It is similar to a netstack socket, @@ -55,6 +55,7 @@ type SocketOperations struct { fsutil.FileNoopFlush `state:"nosave"` fsutil.FileUseInodeUnstableAttr `state:"nosave"` + socketOperationsRefs socketOpsCommon } @@ -84,11 +85,27 @@ func NewWithDirent(ctx context.Context, d *fs.Dirent, ep transport.Endpoint, sty return fs.NewFile(ctx, d, flags, &s) } +// DecRef implements RefCounter.DecRef. +func (s *SocketOperations) DecRef(ctx context.Context) { + s.socketOperationsRefs.DecRef(func() { + s.ep.Close(ctx) + if s.abstractNamespace != nil { + s.abstractNamespace.Remove(s.abstractName, s) + } + }) +} + +// Release implemements fs.FileOperations.Release. +func (s *SocketOperations) Release(ctx context.Context) { + // Release only decrements a reference on s because s may be referenced in + // the abstract socket namespace. + s.DecRef(ctx) +} + // socketOpsCommon contains the socket operations common to VFS1 and VFS2. // // +stateify savable type socketOpsCommon struct { - socketOpsCommonRefs socket.SendReceiveTimeout ep transport.Endpoint @@ -101,23 +118,6 @@ type socketOpsCommon struct { abstractNamespace *kernel.AbstractSocketNamespace } -// DecRef implements RefCounter.DecRef. -func (s *socketOpsCommon) DecRef(ctx context.Context) { - s.socketOpsCommonRefs.DecRef(func() { - s.ep.Close(ctx) - if s.abstractNamespace != nil { - s.abstractNamespace.Remove(s.abstractName, s) - } - }) -} - -// Release implemements fs.FileOperations.Release. -func (s *socketOpsCommon) Release(ctx context.Context) { - // Release only decrements a reference on s because s may be referenced in - // the abstract socket namespace. - s.DecRef(ctx) -} - func (s *socketOpsCommon) isPacket() bool { switch s.stype { case linux.SOCK_DGRAM, linux.SOCK_SEQPACKET: @@ -205,7 +205,7 @@ func (s *socketOpsCommon) Listen(t *kernel.Task, backlog int) *syserr.Error { // blockingAccept implements a blocking version of accept(2), that is, if no // connections are ready to be accept, it will block until one becomes ready. -func (s *SocketOperations) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr.Error) { +func (s *SocketOperations) blockingAccept(t *kernel.Task, peerAddr *tcpip.FullAddress) (transport.Endpoint, *syserr.Error) { // Register for notifications. e, ch := waiter.NewChannelEntry(nil) s.EventRegister(&e, waiter.EventIn) @@ -214,7 +214,7 @@ func (s *SocketOperations) blockingAccept(t *kernel.Task) (transport.Endpoint, * // Try to accept the connection; if it fails, then wait until we get a // notification. for { - if ep, err := s.ep.Accept(); err != syserr.ErrWouldBlock { + if ep, err := s.ep.Accept(peerAddr); err != syserr.ErrWouldBlock { return ep, err } @@ -227,15 +227,18 @@ func (s *SocketOperations) blockingAccept(t *kernel.Task) (transport.Endpoint, * // Accept implements the linux syscall accept(2) for sockets backed by // a transport.Endpoint. func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { - // Issue the accept request to get the new endpoint. - ep, err := s.ep.Accept() + var peerAddr *tcpip.FullAddress + if peerRequested { + peerAddr = &tcpip.FullAddress{} + } + ep, err := s.ep.Accept(peerAddr) if err != nil { if err != syserr.ErrWouldBlock || !blocking { return 0, nil, 0, err } var err *syserr.Error - ep, err = s.blockingAccept(t) + ep, err = s.blockingAccept(t, peerAddr) if err != nil { return 0, nil, 0, err } @@ -252,13 +255,8 @@ func (s *SocketOperations) Accept(t *kernel.Task, peerRequested bool, flags int, var addr linux.SockAddr var addrLen uint32 - if peerRequested { - // Get address of the peer. - var err *syserr.Error - addr, addrLen, err = ns.FileOperations.(*SocketOperations).GetPeerName(t) - if err != nil { - return 0, nil, 0, err - } + if peerAddr != nil { + addr, addrLen = netstack.ConvertAddress(linux.AF_UNIX, *peerAddr) } fd, e := t.NewFDFrom(0, ns, kernel.FDFlags{ diff --git a/pkg/sentry/socket/unix/unix_vfs2.go b/pkg/sentry/socket/unix/unix_vfs2.go index 65a285b8f..3345124cc 100644 --- a/pkg/sentry/socket/unix/unix_vfs2.go +++ b/pkg/sentry/socket/unix/unix_vfs2.go @@ -18,6 +18,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fspath" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/arch" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/sockfs" @@ -32,17 +33,19 @@ import ( "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/usermem" "gvisor.dev/gvisor/pkg/waiter" - "gvisor.dev/gvisor/tools/go_marshal/marshal" ) // SocketVFS2 implements socket.SocketVFS2 (and by extension, // vfs.FileDescriptionImpl) for Unix sockets. +// +// +stateify savable type SocketVFS2 struct { vfsfd vfs.FileDescription vfs.FileDescriptionDefaultImpl vfs.DentryMetadataFileDescriptionImpl vfs.LockFD + socketVFS2Refs socketOpsCommon } @@ -53,6 +56,7 @@ var _ = socket.SocketVFS2(&SocketVFS2{}) func NewSockfsFile(t *kernel.Task, ep transport.Endpoint, stype linux.SockType) (*vfs.FileDescription, *syserr.Error) { mnt := t.Kernel().SocketMount() d := sockfs.NewDentry(t.Credentials(), mnt) + defer d.DecRef(t) fd, err := NewFileDescription(ep, stype, linux.O_RDWR, mnt, d, &vfs.FileLocks{}) if err != nil { @@ -88,6 +92,25 @@ func NewFileDescription(ep transport.Endpoint, stype linux.SockType, flags uint3 return vfsfd, nil } +// DecRef implements RefCounter.DecRef. +func (s *SocketVFS2) DecRef(ctx context.Context) { + s.socketVFS2Refs.DecRef(func() { + t := kernel.TaskFromContext(ctx) + t.Kernel().DeleteSocketVFS2(&s.vfsfd) + s.ep.Close(ctx) + if s.abstractNamespace != nil { + s.abstractNamespace.Remove(s.abstractName, s) + } + }) +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (s *SocketVFS2) Release(ctx context.Context) { + // Release only decrements a reference on s because s may be referenced in + // the abstract socket namespace. + s.DecRef(ctx) +} + // GetSockOpt implements the linux syscall getsockopt(2) for sockets backed by // a transport.Endpoint. func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem.Addr, outLen int) (marshal.Marshallable, *syserr.Error) { @@ -96,7 +119,7 @@ func (s *SocketVFS2) GetSockOpt(t *kernel.Task, level, name int, outPtr usermem. // blockingAccept implements a blocking version of accept(2), that is, if no // connections are ready to be accept, it will block until one becomes ready. -func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr.Error) { +func (s *SocketVFS2) blockingAccept(t *kernel.Task, peerAddr *tcpip.FullAddress) (transport.Endpoint, *syserr.Error) { // Register for notifications. e, ch := waiter.NewChannelEntry(nil) s.socketOpsCommon.EventRegister(&e, waiter.EventIn) @@ -105,7 +128,7 @@ func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr // Try to accept the connection; if it fails, then wait until we get a // notification. for { - if ep, err := s.ep.Accept(); err != syserr.ErrWouldBlock { + if ep, err := s.ep.Accept(peerAddr); err != syserr.ErrWouldBlock { return ep, err } @@ -118,15 +141,18 @@ func (s *SocketVFS2) blockingAccept(t *kernel.Task) (transport.Endpoint, *syserr // Accept implements the linux syscall accept(2) for sockets backed by // a transport.Endpoint. func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, blocking bool) (int32, linux.SockAddr, uint32, *syserr.Error) { - // Issue the accept request to get the new endpoint. - ep, err := s.ep.Accept() + var peerAddr *tcpip.FullAddress + if peerRequested { + peerAddr = &tcpip.FullAddress{} + } + ep, err := s.ep.Accept(peerAddr) if err != nil { if err != syserr.ErrWouldBlock || !blocking { return 0, nil, 0, err } var err *syserr.Error - ep, err = s.blockingAccept(t) + ep, err = s.blockingAccept(t, peerAddr) if err != nil { return 0, nil, 0, err } @@ -144,13 +170,8 @@ func (s *SocketVFS2) Accept(t *kernel.Task, peerRequested bool, flags int, block var addr linux.SockAddr var addrLen uint32 - if peerRequested { - // Get address of the peer. - var err *syserr.Error - addr, addrLen, err = ns.Impl().(*SocketVFS2).GetPeerName(t) - if err != nil { - return 0, nil, 0, err - } + if peerAddr != nil { + addr, addrLen = netstack.ConvertAddress(linux.AF_UNIX, *peerAddr) } fd, e := t.NewFDFromVFS2(0, ns, kernel.FDFlags{ |