diff options
Diffstat (limited to 'pkg/sentry/socket')
24 files changed, 830 insertions, 232 deletions
diff --git a/pkg/sentry/socket/control/BUILD b/pkg/sentry/socket/control/BUILD index 4a6e83a8b..357517ed4 100644 --- a/pkg/sentry/socket/control/BUILD +++ b/pkg/sentry/socket/control/BUILD @@ -17,6 +17,7 @@ go_library( "//pkg/sentry/fs", "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", + "//pkg/sentry/socket", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", "//pkg/syserror", diff --git a/pkg/sentry/socket/control/control.go b/pkg/sentry/socket/control/control.go index 4e95101b7..1684dfc24 100644 --- a/pkg/sentry/socket/control/control.go +++ b/pkg/sentry/socket/control/control.go @@ -23,6 +23,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/fs" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" @@ -194,15 +195,15 @@ func putCmsg(buf []byte, flags int, msgType uint32, align uint, data []int32) ([ // the available space, we must align down. // // align must be >= 4 and each data int32 is 4 bytes. The length of the - // header is already aligned, so if we align to the with of the data there + // header is already aligned, so if we align to the width of the data there // are two cases: // 1. The aligned length is less than the length of the header. The // unaligned length was also less than the length of the header, so we // can't write anything. // 2. The aligned length is greater than or equal to the length of the - // header. We can write the header plus zero or more datas. We can't write - // a partial int32, so the length of the message will be - // min(aligned length, header + datas). + // header. We can write the header plus zero or more bytes of data. We can't + // write a partial int32, so the length of the message will be + // min(aligned length, header + data). if space < linux.SizeOfControlMessageHeader { flags |= linux.MSG_CTRUNC return buf, flags @@ -239,12 +240,12 @@ func putCmsgStruct(buf []byte, msgLevel, msgType uint32, align uint, data interf buf = binary.Marshal(buf, usermem.ByteOrder, data) - // Check if we went over. + // If the control message data brought us over capacity, omit it. if cap(buf) != cap(ob) { return hdrBuf } - // Fix up length. + // Update control message length to include data. putUint64(ob, uint64(len(buf)-len(ob))) return alignSlice(buf, align) @@ -320,35 +321,109 @@ func PackInq(t *kernel.Task, inq int32, buf []byte) []byte { buf, linux.SOL_TCP, linux.TCP_INQ, - 4, + t.Arch().Width(), inq, ) } +// PackTOS packs an IP_TOS socket control message. +func PackTOS(t *kernel.Task, tos uint8, buf []byte) []byte { + return putCmsgStruct( + buf, + linux.SOL_IP, + linux.IP_TOS, + t.Arch().Width(), + tos, + ) +} + +// PackTClass packs an IPV6_TCLASS socket control message. +func PackTClass(t *kernel.Task, tClass int32, buf []byte) []byte { + return putCmsgStruct( + buf, + linux.SOL_IPV6, + linux.IPV6_TCLASS, + t.Arch().Width(), + tClass, + ) +} + +// PackControlMessages packs control messages into the given buffer. +// +// We skip control messages specific to Unix domain sockets. +// +// Note that some control messages may be truncated if they do not fit under +// the capacity of buf. +func PackControlMessages(t *kernel.Task, cmsgs socket.ControlMessages, buf []byte) []byte { + if cmsgs.IP.HasTimestamp { + buf = PackTimestamp(t, cmsgs.IP.Timestamp, buf) + } + + if cmsgs.IP.HasInq { + // In Linux, TCP_CM_INQ is added after SO_TIMESTAMP. + buf = PackInq(t, cmsgs.IP.Inq, buf) + } + + if cmsgs.IP.HasTOS { + buf = PackTOS(t, cmsgs.IP.TOS, buf) + } + + if cmsgs.IP.HasTClass { + buf = PackTClass(t, cmsgs.IP.TClass, buf) + } + + return buf +} + +// cmsgSpace is equivalent to CMSG_SPACE in Linux. +func cmsgSpace(t *kernel.Task, dataLen int) int { + return linux.SizeOfControlMessageHeader + AlignUp(dataLen, t.Arch().Width()) +} + +// CmsgsSpace returns the number of bytes needed to fit the control messages +// represented in cmsgs. +func CmsgsSpace(t *kernel.Task, cmsgs socket.ControlMessages) int { + space := 0 + + if cmsgs.IP.HasTimestamp { + space += cmsgSpace(t, linux.SizeOfTimeval) + } + + if cmsgs.IP.HasInq { + space += cmsgSpace(t, linux.SizeOfControlMessageInq) + } + + if cmsgs.IP.HasTOS { + space += cmsgSpace(t, linux.SizeOfControlMessageTOS) + } + + if cmsgs.IP.HasTClass { + space += cmsgSpace(t, linux.SizeOfControlMessageTClass) + } + + return space +} + // Parse parses a raw socket control message into portable objects. -func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport.ControlMessages, error) { +func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (socket.ControlMessages, error) { var ( - fds linux.ControlMessageRights - haveCreds bool - creds linux.ControlMessageCredentials + cmsgs socket.ControlMessages + fds linux.ControlMessageRights ) for i := 0; i < len(buf); { if i+linux.SizeOfControlMessageHeader > len(buf) { - return transport.ControlMessages{}, syserror.EINVAL + return cmsgs, syserror.EINVAL } var h linux.ControlMessageHeader binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageHeader], usermem.ByteOrder, &h) if h.Length < uint64(linux.SizeOfControlMessageHeader) { - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } if h.Length > uint64(len(buf)-i) { - return transport.ControlMessages{}, syserror.EINVAL - } - if h.Level != linux.SOL_SOCKET { - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } i += linux.SizeOfControlMessageHeader @@ -358,59 +433,85 @@ func Parse(t *kernel.Task, socketOrEndpoint interface{}, buf []byte) (transport. // sizeof(long) in CMSG_ALIGN. width := t.Arch().Width() - switch h.Type { - case linux.SCM_RIGHTS: - rightsSize := AlignDown(length, linux.SizeOfControlMessageRight) - numRights := rightsSize / linux.SizeOfControlMessageRight - - if len(fds)+numRights > linux.SCM_MAX_FD { - return transport.ControlMessages{}, syserror.EINVAL + switch h.Level { + case linux.SOL_SOCKET: + switch h.Type { + case linux.SCM_RIGHTS: + rightsSize := AlignDown(length, linux.SizeOfControlMessageRight) + numRights := rightsSize / linux.SizeOfControlMessageRight + + if len(fds)+numRights > linux.SCM_MAX_FD { + return socket.ControlMessages{}, syserror.EINVAL + } + + for j := i; j < i+rightsSize; j += linux.SizeOfControlMessageRight { + fds = append(fds, int32(usermem.ByteOrder.Uint32(buf[j:j+linux.SizeOfControlMessageRight]))) + } + + i += AlignUp(length, width) + + case linux.SCM_CREDENTIALS: + if length < linux.SizeOfControlMessageCredentials { + return socket.ControlMessages{}, syserror.EINVAL + } + + var creds linux.ControlMessageCredentials + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageCredentials], usermem.ByteOrder, &creds) + scmCreds, err := NewSCMCredentials(t, creds) + if err != nil { + return socket.ControlMessages{}, err + } + cmsgs.Unix.Credentials = scmCreds + i += AlignUp(length, width) + + default: + // Unknown message type. + return socket.ControlMessages{}, syserror.EINVAL } - - for j := i; j < i+rightsSize; j += linux.SizeOfControlMessageRight { - fds = append(fds, int32(usermem.ByteOrder.Uint32(buf[j:j+linux.SizeOfControlMessageRight]))) + case linux.SOL_IP: + switch h.Type { + case linux.IP_TOS: + if length < linux.SizeOfControlMessageTOS { + return socket.ControlMessages{}, syserror.EINVAL + } + cmsgs.IP.HasTOS = true + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTOS], usermem.ByteOrder, &cmsgs.IP.TOS) + i += AlignUp(length, width) + + default: + return socket.ControlMessages{}, syserror.EINVAL } - - i += AlignUp(length, width) - - case linux.SCM_CREDENTIALS: - if length < linux.SizeOfControlMessageCredentials { - return transport.ControlMessages{}, syserror.EINVAL + case linux.SOL_IPV6: + switch h.Type { + case linux.IPV6_TCLASS: + if length < linux.SizeOfControlMessageTClass { + return socket.ControlMessages{}, syserror.EINVAL + } + cmsgs.IP.HasTClass = true + binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageTClass], usermem.ByteOrder, &cmsgs.IP.TClass) + i += AlignUp(length, width) + + default: + return socket.ControlMessages{}, syserror.EINVAL } - - binary.Unmarshal(buf[i:i+linux.SizeOfControlMessageCredentials], usermem.ByteOrder, &creds) - haveCreds = true - i += AlignUp(length, width) - default: - // Unknown message type. - return transport.ControlMessages{}, syserror.EINVAL + return socket.ControlMessages{}, syserror.EINVAL } } - var credentials SCMCredentials - if haveCreds { - var err error - if credentials, err = NewSCMCredentials(t, creds); err != nil { - return transport.ControlMessages{}, err - } - } else { - credentials = makeCreds(t, socketOrEndpoint) + if cmsgs.Unix.Credentials == nil { + cmsgs.Unix.Credentials = makeCreds(t, socketOrEndpoint) } - var rights SCMRights if len(fds) > 0 { - var err error - if rights, err = NewSCMRights(t, fds); err != nil { - return transport.ControlMessages{}, err + rights, err := NewSCMRights(t, fds) + if err != nil { + return socket.ControlMessages{}, err } + cmsgs.Unix.Rights = rights } - if credentials == nil && rights == nil { - return transport.ControlMessages{}, nil - } - - return transport.ControlMessages{Credentials: credentials, Rights: rights}, nil + return cmsgs, nil } func makeCreds(t *kernel.Task, socketOrEndpoint interface{}) SCMCredentials { diff --git a/pkg/sentry/socket/hostinet/BUILD b/pkg/sentry/socket/hostinet/BUILD index 8b66a719d..4c44c7c0f 100644 --- a/pkg/sentry/socket/hostinet/BUILD +++ b/pkg/sentry/socket/hostinet/BUILD @@ -29,10 +29,12 @@ go_library( "//pkg/sentry/kernel/time", "//pkg/sentry/safemem", "//pkg/sentry/socket", + "//pkg/sentry/socket/control", "//pkg/sentry/usermem", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip/stack", "//pkg/waiter", + "@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 92beb1bcf..c957b0f1d 100644 --- a/pkg/sentry/socket/hostinet/socket.go +++ b/pkg/sentry/socket/hostinet/socket.go @@ -18,6 +18,7 @@ import ( "fmt" "syscall" + "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" "gvisor.dev/gvisor/pkg/fdnotifier" @@ -29,6 +30,7 @@ import ( ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" + "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" @@ -41,6 +43,10 @@ const ( // sizeofSockaddr is the size in bytes of the largest sockaddr type // supported by this package. sizeofSockaddr = syscall.SizeofSockaddrInet6 // sizeof(sockaddr_in6) > sizeof(sockaddr_in) + + // maxControlLen is the maximum size of a control message buffer used in a + // recvmsg or sendmsg syscall. + maxControlLen = 1024 ) // socketOperations implements fs.FileOperations and socket.Socket for a socket @@ -281,26 +287,32 @@ func (s *socketOperations) GetSockOpt(t *kernel.Task, level int, name int, outPt // Whitelist options and constrain option length. var optlen int switch level { - case syscall.SOL_IPV6: + case linux.SOL_IP: + switch name { + case linux.IP_TOS, linux.IP_RECVTOS: + optlen = sizeofInt32 + } + case linux.SOL_IPV6: switch name { - case syscall.IPV6_V6ONLY: + case linux.IPV6_TCLASS, linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } - case syscall.SOL_SOCKET: + case linux.SOL_SOCKET: switch name { - case syscall.SO_ERROR, syscall.SO_KEEPALIVE, syscall.SO_SNDBUF, syscall.SO_RCVBUF, syscall.SO_REUSEADDR: + case linux.SO_ERROR, linux.SO_KEEPALIVE, linux.SO_SNDBUF, linux.SO_RCVBUF, linux.SO_REUSEADDR: optlen = sizeofInt32 - case syscall.SO_LINGER: + case linux.SO_LINGER: optlen = syscall.SizeofLinger } - case syscall.SOL_TCP: + case linux.SOL_TCP: switch name { - case syscall.TCP_NODELAY: + case linux.TCP_NODELAY: optlen = sizeofInt32 - case syscall.TCP_INFO: + case linux.TCP_INFO: optlen = int(linux.SizeOfTCPInfo) } } + if optlen == 0 { return nil, syserr.ErrProtocolNotAvailable // ENOPROTOOPT } @@ -320,19 +332,24 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ // Whitelist options and constrain option length. var optlen int switch level { - case syscall.SOL_IPV6: + case linux.SOL_IP: + switch name { + case linux.IP_TOS, linux.IP_RECVTOS: + optlen = sizeofInt32 + } + case linux.SOL_IPV6: switch name { - case syscall.IPV6_V6ONLY: + case linux.IPV6_TCLASS, linux.IPV6_RECVTCLASS, linux.IPV6_V6ONLY: optlen = sizeofInt32 } - case syscall.SOL_SOCKET: + case linux.SOL_SOCKET: switch name { - case syscall.SO_SNDBUF, syscall.SO_RCVBUF, syscall.SO_REUSEADDR: + case linux.SO_SNDBUF, linux.SO_RCVBUF, linux.SO_REUSEADDR: optlen = sizeofInt32 } - case syscall.SOL_TCP: + case linux.SOL_TCP: switch name { - case syscall.TCP_NODELAY: + case linux.TCP_NODELAY: optlen = sizeofInt32 } } @@ -354,11 +371,11 @@ func (s *socketOperations) SetSockOpt(t *kernel.Task, level int, name int, opt [ } // RecvMsg implements socket.Socket.RecvMsg. -func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlDataLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { +func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, haveDeadline bool, deadline ktime.Time, senderRequested bool, controlLen uint64) (int, int, linux.SockAddr, uint32, socket.ControlMessages, *syserr.Error) { // Whitelist flags. // // FIXME(jamieliu): We can't support MSG_ERRQUEUE because it uses ancillary - // messages that netstack/tcpip/transport/unix doesn't understand. Kill the + // messages that gvisor/pkg/tcpip/transport/unix doesn't understand. Kill the // Socket interface's dependence on netstack. if flags&^(syscall.MSG_DONTWAIT|syscall.MSG_PEEK|syscall.MSG_TRUNC) != 0 { return 0, 0, nil, 0, socket.ControlMessages{}, syserr.ErrInvalidArgument @@ -370,6 +387,7 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags senderAddrBuf = make([]byte, sizeofSockaddr) } + var controlBuf []byte var msgFlags int recvmsgToBlocks := safemem.ReaderFunc(func(dsts safemem.BlockSeq) (uint64, error) { @@ -384,11 +402,6 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags // We always do a non-blocking recv*(). sysflags := flags | syscall.MSG_DONTWAIT - if dsts.NumBlocks() == 1 { - // Skip allocating []syscall.Iovec. - return recvfrom(s.fd, dsts.Head().ToSlice(), sysflags, &senderAddrBuf) - } - iovs := iovecsFromBlockSeq(dsts) msg := syscall.Msghdr{ Iov: &iovs[0], @@ -398,12 +411,21 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags msg.Name = &senderAddrBuf[0] msg.Namelen = uint32(len(senderAddrBuf)) } + if controlLen > 0 { + if controlLen > maxControlLen { + controlLen = maxControlLen + } + controlBuf = make([]byte, controlLen) + msg.Control = &controlBuf[0] + msg.Controllen = controlLen + } n, err := recvmsg(s.fd, &msg, sysflags) if err != nil { return 0, err } senderAddrBuf = senderAddrBuf[:msg.Namelen] msgFlags = int(msg.Flags) + controlLen = uint64(msg.Controllen) return n, nil }) @@ -429,14 +451,38 @@ func (s *socketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags n, err = dst.CopyOutFrom(t, recvmsgToBlocks) } } - - // We don't allow control messages. - msgFlags &^= linux.MSG_CTRUNC + if err != nil { + return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) + } if senderRequested { senderAddr = socket.UnmarshalSockAddr(s.family, senderAddrBuf) } - return int(n), msgFlags, senderAddr, uint32(len(senderAddrBuf)), socket.ControlMessages{}, syserr.FromError(err) + + unixControlMessages, err := unix.ParseSocketControlMessage(controlBuf[:controlLen]) + if err != nil { + return 0, 0, nil, 0, socket.ControlMessages{}, syserr.FromError(err) + } + + controlMessages := socket.ControlMessages{} + for _, unixCmsg := range unixControlMessages { + switch unixCmsg.Header.Level { + case syscall.SOL_IP: + switch unixCmsg.Header.Type { + case syscall.IP_TOS: + controlMessages.IP.HasTOS = true + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTOS], usermem.ByteOrder, &controlMessages.IP.TOS) + } + case syscall.SOL_IPV6: + switch unixCmsg.Header.Type { + case syscall.IPV6_TCLASS: + controlMessages.IP.HasTClass = true + binary.Unmarshal(unixCmsg.Data[:linux.SizeOfControlMessageTClass], usermem.ByteOrder, &controlMessages.IP.TClass) + } + } + } + + return int(n), msgFlags, senderAddr, uint32(len(senderAddrBuf)), controlMessages, nil } // SendMsg implements socket.Socket.SendMsg. @@ -446,6 +492,14 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] return 0, syserr.ErrInvalidArgument } + space := uint64(control.CmsgsSpace(t, controlMessages)) + if space > maxControlLen { + space = maxControlLen + } + controlBuf := make([]byte, 0, space) + // PackControlMessages will append up to space bytes to controlBuf. + controlBuf = control.PackControlMessages(t, controlMessages, controlBuf) + sendmsgFromBlocks := safemem.WriterFunc(func(srcs safemem.BlockSeq) (uint64, error) { // Refuse to do anything if any part of src.Addrs was unusable. if uint64(src.NumBytes()) != srcs.NumBytes() { @@ -458,7 +512,7 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] // We always do a non-blocking send*(). sysflags := flags | syscall.MSG_DONTWAIT - if srcs.NumBlocks() == 1 { + if srcs.NumBlocks() == 1 && len(controlBuf) == 0 { // Skip allocating []syscall.Iovec. src := srcs.Head() n, _, errno := syscall.Syscall6(syscall.SYS_SENDTO, uintptr(s.fd), src.Addr(), uintptr(src.Len()), uintptr(sysflags), uintptr(firstBytePtr(to)), uintptr(len(to))) @@ -477,6 +531,10 @@ func (s *socketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] msg.Name = &to[0] msg.Namelen = uint32(len(to)) } + if len(controlBuf) != 0 { + msg.Control = &controlBuf[0] + msg.Controllen = uint64(len(controlBuf)) + } return sendmsg(s.fd, &msg, sysflags) }) diff --git a/pkg/sentry/socket/netfilter/BUILD b/pkg/sentry/socket/netfilter/BUILD index 5eb06bbf4..b70047d81 100644 --- a/pkg/sentry/socket/netfilter/BUILD +++ b/pkg/sentry/socket/netfilter/BUILD @@ -14,6 +14,7 @@ go_library( deps = [ "//pkg/abi/linux", "//pkg/binary", + "//pkg/log", "//pkg/sentry/kernel", "//pkg/sentry/usermem", "//pkg/syserr", diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index 9f87c32f1..a9cfc1749 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -21,6 +21,7 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/log" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserr" @@ -35,6 +36,7 @@ const errorTargetName = "ERROR" // metadata is opaque to netstack. It holds data that we need to translate // between Linux's and netstack's iptables representations. +// TODO(gvisor.dev/issue/170): This might be removable. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -51,7 +53,7 @@ func GetInfo(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr) (linux.IPTG } // Find the appropriate table. - table, err := findTable(ep, info.TableName()) + table, err := findTable(ep, info.Name) if err != nil { return linux.IPTGetinfo{}, err } @@ -82,30 +84,31 @@ func GetEntries(t *kernel.Task, ep tcpip.Endpoint, outPtr usermem.Addr, outLen i } // Find the appropriate table. - table, err := findTable(ep, userEntries.TableName()) + table, err := findTable(ep, userEntries.Name) if err != nil { return linux.KernelIPTGetEntries{}, err } // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, _, err := convertNetstackToBinary(userEntries.TableName(), table) + entries, _, err := convertNetstackToBinary(userEntries.Name.String(), table) if err != nil { return linux.KernelIPTGetEntries{}, err } if binary.Size(entries) > uintptr(outLen) { + log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument } return entries, nil } -func findTable(ep tcpip.Endpoint, tableName string) (iptables.Table, *syserr.Error) { +func findTable(ep tcpip.Endpoint, tablename linux.TableName) (iptables.Table, *syserr.Error) { ipt, err := ep.IPTables() if err != nil { return iptables.Table{}, syserr.FromError(err) } - table, ok := ipt.Tables[tableName] + table, ok := ipt.Tables[tablename.String()] if !ok { return iptables.Table{}, syserr.ErrInvalidArgument } @@ -135,110 +138,68 @@ func FillDefaultIPTables(stack *stack.Stack) { // format expected by the iptables tool. Linux stores each table as a binary // blob that can only be traversed by parsing a bit, reading some offsets, // jumping to those offsets, parsing again, etc. -func convertNetstackToBinary(name string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) { +func convertNetstackToBinary(tablename string, table iptables.Table) (linux.KernelIPTGetEntries, metadata, *syserr.Error) { // Return values. var entries linux.KernelIPTGetEntries var meta metadata // The table name has to fit in the struct. - if linux.XT_TABLE_MAXNAMELEN < len(name) { + if linux.XT_TABLE_MAXNAMELEN < len(tablename) { + log.Warningf("Table name %q too long.", tablename) return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument } - copy(entries.Name[:], name) - - // Deal with the built in chains first (INPUT, OUTPUT, etc.). Each of - // these chains ends with an unconditional policy entry. - for hook := iptables.Prerouting; hook < iptables.NumHooks; hook++ { - chain, ok := table.BuiltinChains[hook] - if !ok { - // This table doesn't support this hook. - continue - } - - // Sanity check. - if len(chain.Rules) < 1 { - return linux.KernelIPTGetEntries{}, metadata{}, syserr.ErrInvalidArgument - } + copy(entries.Name[:], tablename) - for ruleIdx, rule := range chain.Rules { - // If this is the first rule of a builtin chain, set - // the metadata hook entry point. - if ruleIdx == 0 { + for ruleIdx, rule := range table.Rules { + // Is this a chain entry point? + for hook, hookRuleIdx := range table.BuiltinChains { + if hookRuleIdx == ruleIdx { meta.HookEntry[hook] = entries.Size } - - // Each rule corresponds to an entry. - entry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ - NextOffset: linux.SizeOfIPTEntry, - TargetOffset: linux.SizeOfIPTEntry, - }, + } + // Is this a chain underflow point? + for underflow, underflowRuleIdx := range table.Underflows { + if underflowRuleIdx == ruleIdx { + meta.Underflow[underflow] = entries.Size } + } - for _, matcher := range rule.Matchers { - // Serialize the matcher and add it to the - // entry. - serialized := marshalMatcher(matcher) - entry.Elems = append(entry.Elems, serialized...) - entry.NextOffset += uint16(len(serialized)) - entry.TargetOffset += uint16(len(serialized)) - } + // Each rule corresponds to an entry. + entry := linux.KernelIPTEntry{ + IPTEntry: linux.IPTEntry{ + NextOffset: linux.SizeOfIPTEntry, + TargetOffset: linux.SizeOfIPTEntry, + }, + } - // Serialize and append the target. - serialized := marshalTarget(rule.Target) + for _, matcher := range rule.Matchers { + // Serialize the matcher and add it to the + // entry. + serialized := marshalMatcher(matcher) entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) - - // The underflow rule is the last rule in the chain, - // and is an unconditional rule (i.e. it matches any - // packet). This is enforced when saving iptables. - if ruleIdx == len(chain.Rules)-1 { - meta.Underflow[hook] = entries.Size - } - - entries.Size += uint32(entry.NextOffset) - entries.Entrytable = append(entries.Entrytable, entry) - meta.NumEntries++ + entry.TargetOffset += uint16(len(serialized)) } - } - - // TODO(gvisor.dev/issue/170): Deal with the user chains here. Each of - // these starts with an error node holding the chain's name and ends - // with an unconditional return. + // Serialize and append the target. + serialized := marshalTarget(rule.Target) + entry.Elems = append(entry.Elems, serialized...) + entry.NextOffset += uint16(len(serialized)) - // Lastly, each table ends with an unconditional error target rule as - // its final entry. - errorEntry := linux.KernelIPTEntry{ - IPTEntry: linux.IPTEntry{ - NextOffset: linux.SizeOfIPTEntry, - TargetOffset: linux.SizeOfIPTEntry, - }, + entries.Size += uint32(entry.NextOffset) + entries.Entrytable = append(entries.Entrytable, entry) + meta.NumEntries++ } - var errorTarget linux.XTErrorTarget - errorTarget.Target.TargetSize = linux.SizeOfXTErrorTarget - copy(errorTarget.ErrorName[:], errorTargetName) - copy(errorTarget.Target.Name[:], errorTargetName) - - // Serialize and add it to the list of entries. - errorTargetBuf := make([]byte, 0, linux.SizeOfXTErrorTarget) - serializedErrorTarget := binary.Marshal(errorTargetBuf, usermem.ByteOrder, errorTarget) - errorEntry.Elems = append(errorEntry.Elems, serializedErrorTarget...) - errorEntry.NextOffset += uint16(len(serializedErrorTarget)) - - entries.Size += uint32(errorEntry.NextOffset) - entries.Entrytable = append(entries.Entrytable, errorEntry) - meta.NumEntries++ - meta.Size = entries.Size + meta.Size = entries.Size return entries, meta, nil } func marshalMatcher(matcher iptables.Matcher) []byte { switch matcher.(type) { default: - // TODO(gvisor.dev/issue/170): We don't support any matchers yet, so - // any call to marshalMatcher will panic. + // TODO(gvisor.dev/issue/170): We don't support any matchers + // yet, so any call to marshalMatcher will panic. panic(fmt.Errorf("unknown matcher of type %T", matcher)) } } @@ -246,28 +207,46 @@ func marshalMatcher(matcher iptables.Matcher) []byte { func marshalTarget(target iptables.Target) []byte { switch target.(type) { case iptables.UnconditionalAcceptTarget: - return marshalUnconditionalAcceptTarget() + return marshalStandardTarget(iptables.Accept) + case iptables.UnconditionalDropTarget: + return marshalStandardTarget(iptables.Drop) + case iptables.ErrorTarget: + return marshalErrorTarget() default: panic(fmt.Errorf("unknown target of type %T", target)) } } -func marshalUnconditionalAcceptTarget() []byte { +func marshalStandardTarget(verdict iptables.Verdict) []byte { // The target's name will be the empty string. target := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ TargetSize: linux.SizeOfXTStandardTarget, }, - Verdict: translateStandardVerdict(iptables.Accept), + Verdict: translateFromStandardVerdict(verdict), } ret := make([]byte, 0, linux.SizeOfXTStandardTarget) return binary.Marshal(ret, usermem.ByteOrder, target) } -// translateStandardVerdict translates verdicts the same way as the iptables +func marshalErrorTarget() []byte { + // This is an error target named error + target := linux.XTErrorTarget{ + Target: linux.XTEntryTarget{ + TargetSize: linux.SizeOfXTErrorTarget, + }, + } + copy(target.Name[:], errorTargetName) + copy(target.Target.Name[:], errorTargetName) + + ret := make([]byte, 0, linux.SizeOfXTErrorTarget) + return binary.Marshal(ret, usermem.ByteOrder, target) +} + +// translateFromStandardVerdict translates verdicts the same way as the iptables // tool. -func translateStandardVerdict(verdict iptables.Verdict) int32 { +func translateFromStandardVerdict(verdict iptables.Verdict) int32 { switch verdict { case iptables.Accept: return -linux.NF_ACCEPT - 1 @@ -280,7 +259,258 @@ func translateStandardVerdict(verdict iptables.Verdict) int32 { case iptables.Jump: // TODO(gvisor.dev/issue/170): Support Jump. panic("Jump isn't supported yet") + } + panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) +} + +// translateToStandardVerdict translates from the value in a +// linux.XTStandardTarget to an iptables.Verdict. +func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { + // TODO(gvisor.dev/issue/170): Support other verdicts. + switch val { + case -linux.NF_ACCEPT - 1: + return iptables.Accept, nil + case -linux.NF_DROP - 1: + return iptables.Drop, nil + case -linux.NF_QUEUE - 1: + log.Warningf("Unsupported iptables verdict QUEUE.") + case linux.NF_RETURN: + log.Warningf("Unsupported iptables verdict RETURN.") + default: + log.Warningf("Unknown iptables verdict %d.", val) + } + return iptables.Invalid, syserr.ErrInvalidArgument +} + +// SetEntries sets iptables rules for a single table. See +// net/ipv4/netfilter/ip_tables.c:translate_table for reference. +func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { + printReplace(optVal) + + // Get the basic rules data (struct ipt_replace). + if len(optVal) < linux.SizeOfIPTReplace { + log.Warningf("netfilter.SetEntries: optVal has insufficient size for replace %d", len(optVal)) + return syserr.ErrInvalidArgument + } + var replace linux.IPTReplace + replaceBuf := optVal[:linux.SizeOfIPTReplace] + optVal = optVal[linux.SizeOfIPTReplace:] + binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) + + // TODO(gvisor.dev/issue/170): Support other tables. + var table iptables.Table + switch replace.Name.String() { + case iptables.TablenameFilter: + table = iptables.EmptyFilterTable() default: - panic(fmt.Sprintf("unknown standard verdict: %d", verdict)) + log.Warningf("We don't yet support writing to the %q table (gvisor.dev/issue/170)", replace.Name.String()) + return syserr.ErrInvalidArgument + } + + // Convert input into a list of rules and their offsets. + var offset uint32 + var offsets []uint32 + for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + // Get the struct ipt_entry. + if len(optVal) < linux.SizeOfIPTEntry { + log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal)) + return syserr.ErrInvalidArgument + } + var entry linux.IPTEntry + buf := optVal[:linux.SizeOfIPTEntry] + optVal = optVal[linux.SizeOfIPTEntry:] + binary.Unmarshal(buf, usermem.ByteOrder, &entry) + if entry.TargetOffset != linux.SizeOfIPTEntry { + // TODO(gvisor.dev/issue/170): Support matchers. + return syserr.ErrInvalidArgument + } + + // TODO(gvisor.dev/issue/170): We should support IPTIP + // filtering. We reject any nonzero IPTIP values for now. + emptyIPTIP := linux.IPTIP{} + if entry.IP != emptyIPTIP { + log.Warningf("netfilter: non-empty struct iptip found") + return syserr.ErrInvalidArgument + } + + // Get the target of the rule. + target, consumed, err := parseTarget(optVal) + if err != nil { + return err + } + optVal = optVal[consumed:] + + table.Rules = append(table.Rules, iptables.Rule{Target: target}) + offsets = append(offsets, offset) + offset += linux.SizeOfIPTEntry + consumed + } + + // Go through the list of supported hooks for this table and, for each + // one, set the rule it corresponds to. + for hook, _ := range replace.HookEntry { + if table.ValidHooks()&uint32(hook) != 0 { + hk := hookFromLinux(hook) + for ruleIdx, offset := range offsets { + if offset == replace.HookEntry[hook] { + table.BuiltinChains[hk] = ruleIdx + } + if offset == replace.Underflow[hook] { + table.Underflows[hk] = ruleIdx + } + } + if ruleIdx := table.BuiltinChains[hk]; ruleIdx == iptables.HookUnset { + log.Warningf("Hook %v is unset.", hk) + return syserr.ErrInvalidArgument + } + if ruleIdx := table.Underflows[hk]; ruleIdx == iptables.HookUnset { + log.Warningf("Underflow %v is unset.", hk) + return syserr.ErrInvalidArgument + } + } + } + + ipt := stack.IPTables() + table.SetMetadata(metadata{ + HookEntry: replace.HookEntry, + Underflow: replace.Underflow, + NumEntries: replace.NumEntries, + Size: replace.Size, + }) + ipt.Tables[replace.Name.String()] = table + stack.SetIPTables(ipt) + + return nil +} + +// parseTarget parses a target from the start of optVal and returns the target +// along with the number of bytes it occupies in optVal. +func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { + if len(optVal) < linux.SizeOfXTEntryTarget { + log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) + return nil, 0, 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 { + log.Warningf("netfilter.SetEntries: optVal has insufficient size for standard target %d", len(optVal)) + return nil, 0, syserr.ErrInvalidArgument + } + var standardTarget linux.XTStandardTarget + buf = optVal[:linux.SizeOfXTStandardTarget] + binary.Unmarshal(buf, usermem.ByteOrder, &standardTarget) + + verdict, err := translateToStandardVerdict(standardTarget.Verdict) + if err != nil { + return nil, 0, err + } + switch verdict { + case iptables.Accept: + return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil + case iptables.Drop: + // TODO(gvisor.dev/issue/170): Return an + // iptables.UnconditionalDropTarget to support DROP. + log.Infof("netfilter DROP is not supported yet.") + return nil, 0, syserr.ErrInvalidArgument + default: + panic(fmt.Sprintf("Unknown verdict: %v", verdict)) + } + + case errorTargetName: + // Error target. + if len(optVal) < linux.SizeOfXTErrorTarget { + log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) + return nil, 0, syserr.ErrInvalidArgument + } + 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 errorTarget.Name.String() { + case errorTargetName: + return iptables.ErrorTarget{}, linux.SizeOfXTErrorTarget, nil + default: + log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) + return nil, 0, syserr.ErrInvalidArgument + } + } + + // Unknown target. + log.Infof("Unknown target %q doesn't exist or isn't supported yet.", target.Name.String()) + return nil, 0, syserr.ErrInvalidArgument +} + +func hookFromLinux(hook int) iptables.Hook { + switch hook { + case linux.NF_INET_PRE_ROUTING: + return iptables.Prerouting + case linux.NF_INET_LOCAL_IN: + return iptables.Input + case linux.NF_INET_FORWARD: + return iptables.Forward + case linux.NF_INET_LOCAL_OUT: + return iptables.Output + case linux.NF_INET_POST_ROUTING: + return iptables.Postrouting + } + panic(fmt.Sprintf("Unknown hook %d does not correspond to a builtin chain", hook)) +} + +// printReplace prints information about the struct ipt_replace in optVal. It +// is only for debugging. +func printReplace(optVal []byte) { + // Basic replace info. + var replace linux.IPTReplace + replaceBuf := optVal[:linux.SizeOfIPTReplace] + optVal = optVal[linux.SizeOfIPTReplace:] + binary.Unmarshal(replaceBuf, usermem.ByteOrder, &replace) + log.Infof("Replacing table %q: %+v", replace.Name.String(), replace) + + // Read in the list of entries at the end of replace. + var totalOffset uint16 + for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + var entry linux.IPTEntry + entryBuf := optVal[:linux.SizeOfIPTEntry] + binary.Unmarshal(entryBuf, usermem.ByteOrder, &entry) + log.Infof("Entry %d (total offset %d): %+v", entryIdx, totalOffset, entry) + + totalOffset += entry.NextOffset + if entry.TargetOffset == linux.SizeOfIPTEntry { + log.Infof("Entry has no matches.") + } else { + log.Infof("Entry has matches.") + } + + var target linux.XTEntryTarget + targetBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTEntryTarget] + binary.Unmarshal(targetBuf, usermem.ByteOrder, &target) + log.Infof("Target named %q: %+v", target.Name.String(), target) + + switch target.Name.String() { + case "": + var standardTarget linux.XTStandardTarget + stBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTStandardTarget] + binary.Unmarshal(stBuf, usermem.ByteOrder, &standardTarget) + log.Infof("Standard target with verdict %q (%d).", linux.VerdictStrings[standardTarget.Verdict], standardTarget.Verdict) + case errorTargetName: + var errorTarget linux.XTErrorTarget + etBuf := optVal[entry.TargetOffset : entry.TargetOffset+linux.SizeOfXTErrorTarget] + binary.Unmarshal(etBuf, usermem.ByteOrder, &errorTarget) + log.Infof("Error target with name %q.", errorTarget.Name.String()) + default: + log.Infof("Unknown target type.") + } + + optVal = optVal[entry.NextOffset:] } } diff --git a/pkg/sentry/socket/netlink/BUILD b/pkg/sentry/socket/netlink/BUILD index 79589e3c8..103933144 100644 --- a/pkg/sentry/socket/netlink/BUILD +++ b/pkg/sentry/socket/netlink/BUILD @@ -22,12 +22,12 @@ go_library( "//pkg/sentry/kernel", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/time", - "//pkg/sentry/safemem", "//pkg/sentry/socket", "//pkg/sentry/socket/netlink/port", "//pkg/sentry/socket/unix", "//pkg/sentry/socket/unix/transport", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/netlink/port/BUILD b/pkg/sentry/socket/netlink/port/BUILD index 463544c1a..2d9f4ba9b 100644 --- a/pkg/sentry/socket/netlink/port/BUILD +++ b/pkg/sentry/socket/netlink/port/BUILD @@ -8,6 +8,7 @@ go_library( srcs = ["port.go"], importpath = "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port", visibility = ["//pkg/sentry:internal"], + deps = ["//pkg/sync"], ) go_test( diff --git a/pkg/sentry/socket/netlink/port/port.go b/pkg/sentry/socket/netlink/port/port.go index e9d3275b1..2cd3afc22 100644 --- a/pkg/sentry/socket/netlink/port/port.go +++ b/pkg/sentry/socket/netlink/port/port.go @@ -24,7 +24,8 @@ import ( "fmt" "math" "math/rand" - "sync" + + "gvisor.dev/gvisor/pkg/sync" ) // maxPorts is a sanity limit on the maximum number of ports to allocate per diff --git a/pkg/sentry/socket/netlink/socket.go b/pkg/sentry/socket/netlink/socket.go index 4a1b87a9a..cea56f4ed 100644 --- a/pkg/sentry/socket/netlink/socket.go +++ b/pkg/sentry/socket/netlink/socket.go @@ -17,7 +17,6 @@ package netlink import ( "math" - "sync" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/binary" @@ -29,12 +28,12 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/socket" "gvisor.dev/gvisor/pkg/sentry/socket/netlink/port" "gvisor.dev/gvisor/pkg/sentry/socket/unix" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -500,29 +499,29 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have trunc := flags&linux.MSG_TRUNC != 0 r := unix.EndpointReader{ + Ctx: t, Endpoint: s.ep, Peek: flags&linux.MSG_PEEK != 0, } + doRead := func() (int64, error) { + return dst.CopyOutFrom(t, &r) + } + // If MSG_TRUNC is set with a zero byte destination then we still need // to read the message and discard it, or in the case where MSG_PEEK is // set, leave it be. In both cases the full message length must be - // returned. However, the memory manager for the destination will not read - // the endpoint if the destination is zero length. - // - // In order for the endpoint to be read when the destination size is zero, - // we must cause a read of the endpoint by using a separate fake zero - // length block sequence and calling the EndpointReader directly. + // returned. if trunc && dst.Addrs.NumBytes() == 0 { - // Perform a read to a zero byte block sequence. We can ignore the - // original destination since it was zero bytes. The length returned by - // ReadToBlocks is ignored and we return the full message length to comply - // with MSG_TRUNC. - _, err := r.ReadToBlocks(safemem.BlockSeqOf(safemem.BlockFromSafeSlice(make([]byte, 0)))) - return int(r.MsgSize), linux.MSG_TRUNC, from, fromLen, socket.ControlMessages{}, syserr.FromError(err) + doRead = func() (int64, error) { + err := r.Truncate() + // Always return zero for bytes read since the destination size is + // zero. + return 0, err + } } - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 { + if n, err := doRead(); err != syserror.ErrWouldBlock || flags&linux.MSG_DONTWAIT != 0 { var mflags int if n < int64(r.MsgSize) { mflags |= linux.MSG_TRUNC @@ -540,7 +539,7 @@ func (s *Socket) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags int, have defer s.EventUnregister(&e) for { - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock { + if n, err := doRead(); err != syserror.ErrWouldBlock { var mflags int if n < int64(r.MsgSize) { mflags |= linux.MSG_TRUNC diff --git a/pkg/sentry/socket/netstack/BUILD b/pkg/sentry/socket/netstack/BUILD index e414d8055..f78784569 100644 --- a/pkg/sentry/socket/netstack/BUILD +++ b/pkg/sentry/socket/netstack/BUILD @@ -34,6 +34,7 @@ go_library( "//pkg/sentry/socket/netfilter", "//pkg/sentry/unimpl", "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserr", "//pkg/syserror", "//pkg/tcpip", diff --git a/pkg/sentry/socket/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index d92399efd..d2f7e987d 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -29,7 +29,6 @@ import ( "io" "math" "reflect" - "sync" "syscall" "time" @@ -49,6 +48,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/socket/netfilter" "gvisor.dev/gvisor/pkg/sentry/unimpl" "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/syserror" "gvisor.dev/gvisor/pkg/tcpip" @@ -151,6 +151,8 @@ var Metrics = tcpip.Stats{ PassiveConnectionOpenings: mustCreateMetric("/netstack/tcp/passive_connection_openings", "Number of connections opened successfully via Listen."), CurrentEstablished: mustCreateMetric("/netstack/tcp/current_established", "Number of connections in either ESTABLISHED or CLOSE-WAIT state now."), EstablishedResets: mustCreateMetric("/netstack/tcp/established_resets", "Number of times TCP connections have made a direct transition to the CLOSED state from either the ESTABLISHED state or the CLOSE-WAIT state"), + EstablishedClosed: mustCreateMetric("/netstack/tcp/established_closed", "number of times established TCP connections made a transition to CLOSED state."), + EstablishedTimedout: mustCreateMetric("/netstack/tcp/established_timedout", "Number of times an established connection was reset because of keep-alive time out."), ListenOverflowSynDrop: mustCreateMetric("/netstack/tcp/listen_overflow_syn_drop", "Number of times the listen queue overflowed and a SYN was dropped."), ListenOverflowAckDrop: mustCreateMetric("/netstack/tcp/listen_overflow_ack_drop", "Number of times the listen queue overflowed and the final ACK in the handshake was dropped."), ListenOverflowSynCookieSent: mustCreateMetric("/netstack/tcp/listen_overflow_syn_cookie_sent", "Number of times a SYN cookie was sent."), @@ -220,17 +222,25 @@ type commonEndpoint interface { // transport.Endpoint.SetSockOpt. SetSockOpt(interface{}) *tcpip.Error + // SetSockOptBool implements tcpip.Endpoint.SetSockOptBool and + // transport.Endpoint.SetSockOptBool. + SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error + // SetSockOptInt implements tcpip.Endpoint.SetSockOptInt and // transport.Endpoint.SetSockOptInt. - SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error + SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error // GetSockOpt implements tcpip.Endpoint.GetSockOpt and // transport.Endpoint.GetSockOpt. GetSockOpt(interface{}) *tcpip.Error + // GetSockOptBool implements tcpip.Endpoint.GetSockOptBool and + // transport.Endpoint.GetSockOpt. + GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) + // GetSockOptInt implements tcpip.Endpoint.GetSockOptInt and // transport.Endpoint.GetSockOpt. - GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) + GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) } // SocketOperations encapsulates all the state needed to represent a network stack @@ -314,22 +324,15 @@ func bytesToIPAddress(addr []byte) tcpip.Address { // converts it to the FullAddress format. It supports AF_UNIX, AF_INET, // AF_INET6, and AF_PACKET addresses. // -// strict indicates whether addresses with the AF_UNSPEC family are accepted of not. -// // AddressAndFamily returns an address and its family. -func AddressAndFamily(sfamily int, addr []byte, strict bool) (tcpip.FullAddress, uint16, *syserr.Error) { +func AddressAndFamily(addr []byte) (tcpip.FullAddress, uint16, *syserr.Error) { // Make sure we have at least 2 bytes for the address family. if len(addr) < 2 { return tcpip.FullAddress{}, 0, syserr.ErrInvalidArgument } - family := usermem.ByteOrder.Uint16(addr) - if family != uint16(sfamily) && (!strict && family != linux.AF_UNSPEC) { - return tcpip.FullAddress{}, family, syserr.ErrAddressFamilyNotSupported - } - // Get the rest of the fields based on the address family. - switch family { + switch family := usermem.ByteOrder.Uint16(addr); family { case linux.AF_UNIX: path := addr[2:] if len(path) > linux.UnixPathMax { @@ -628,10 +631,40 @@ func (s *SocketOperations) Readiness(mask waiter.EventMask) waiter.EventMask { return r } +func (s *SocketOperations) checkFamily(family uint16, exact bool) *syserr.Error { + if family == uint16(s.family) { + return nil + } + if !exact && family == linux.AF_INET && s.family == linux.AF_INET6 { + v, err := s.Endpoint.GetSockOptBool(tcpip.V6OnlyOption) + if err != nil { + return syserr.TranslateNetstackError(err) + } + if !v { + return nil + } + } + return syserr.ErrInvalidArgument +} + +// mapFamily maps the AF_INET ANY address to the IPv4-mapped IPv6 ANY if the +// receiver's family is AF_INET6. +// +// This is a hack to work around the fact that both IPv4 and IPv6 ANY are +// represented by the empty string. +// +// TODO(gvisor.dev/issues/1556): remove this function. +func (s *SocketOperations) mapFamily(addr tcpip.FullAddress, family uint16) tcpip.FullAddress { + if len(addr.Addr) == 0 && s.family == linux.AF_INET6 && family == linux.AF_INET { + addr.Addr = "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\xff\xff\x00\x00\x00\x00" + } + return addr +} + // Connect implements the linux syscall connect(2) for sockets backed by // tpcip.Endpoint. func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking bool) *syserr.Error { - addr, family, err := AddressAndFamily(s.family, sockaddr, false /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } @@ -643,6 +676,12 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo } return syserr.TranslateNetstackError(err) } + + if err := s.checkFamily(family, false /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) + // Always return right away in the non-blocking case. if !blocking { return syserr.TranslateNetstackError(s.Endpoint.Connect(addr)) @@ -671,10 +710,14 @@ func (s *SocketOperations) Connect(t *kernel.Task, sockaddr []byte, blocking boo // Bind implements the linux syscall bind(2) for sockets backed by // tcpip.Endpoint. func (s *SocketOperations) Bind(t *kernel.Task, sockaddr []byte) *syserr.Error { - addr, _, err := AddressAndFamily(s.family, sockaddr, true /* strict */) + addr, family, err := AddressAndFamily(sockaddr) if err != nil { return err } + if err := s.checkFamily(family, true /* exact */); err != nil { + return err + } + addr = s.mapFamily(addr, family) // Issue the bind request to the endpoint. return syserr.TranslateNetstackError(s.Endpoint.Bind(addr)) @@ -975,13 +1018,23 @@ func getSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, family if err := ep.GetSockOpt(&v); err != nil { return nil, syserr.TranslateNetstackError(err) } - if len(v) == 0 { + if v == 0 { return []byte{}, nil } if outLen < linux.IFNAMSIZ { return nil, syserr.ErrInvalidArgument } - return append([]byte(v), 0), nil + s := t.NetworkContext() + if s == nil { + return nil, syserr.ErrNoDevice + } + nic, ok := s.Interfaces()[int32(v)] + if !ok { + // The NICID no longer indicates a valid interface, probably because that + // interface was removed. + return nil, syserr.ErrUnknownDevice + } + return append([]byte(nic.Name), 0), nil case linux.SO_BROADCAST: if outLen < sizeOfInt32 { @@ -1125,6 +1178,18 @@ func getSockOptTCP(t *kernel.Task, ep commonEndpoint, name, outLen int) (interfa return int32(time.Duration(v) / time.Second), nil + case linux.TCP_USER_TIMEOUT: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + var v tcpip.TCPUserTimeoutOption + if err := ep.GetSockOpt(&v); err != nil { + return nil, syserr.TranslateNetstackError(err) + } + + return int32(time.Duration(v) / time.Millisecond), nil + case linux.TCP_INFO: var v tcpip.TCPInfoOption if err := ep.GetSockOpt(&v); err != nil { @@ -1199,12 +1264,15 @@ func getSockOptIPv6(t *kernel.Task, ep commonEndpoint, name, outLen int) (interf return nil, syserr.ErrInvalidArgument } - var v tcpip.V6OnlyOption - if err := ep.GetSockOpt(&v); err != nil { + v, err := ep.GetSockOptBool(tcpip.V6OnlyOption) + if err != nil { return nil, syserr.TranslateNetstackError(err) } - - return int32(v), nil + var o int32 + if v { + o = 1 + } + return o, nil case linux.IPV6_PATHMTU: t.Kernel().EmitUnimplementedEvent(t) @@ -1309,6 +1377,21 @@ func getSockOptIP(t *kernel.Task, ep commonEndpoint, name, outLen int, family in } return int32(v), nil + case linux.IP_RECVTOS: + if outLen < sizeOfInt32 { + return nil, syserr.ErrInvalidArgument + } + + v, err := ep.GetSockOptBool(tcpip.ReceiveTOSOption) + if err != nil { + return nil, syserr.TranslateNetstackError(err) + } + var o int32 + if v { + o = 1 + } + return o, nil + default: emitUnimplementedEventIP(t, name) } @@ -1342,6 +1425,26 @@ func (s *SocketOperations) SetSockOpt(t *kernel.Task, level int, name int, optVa return nil } + if s.skType == linux.SOCK_RAW && level == linux.IPPROTO_IP { + switch name { + case linux.IPT_SO_SET_REPLACE: + if len(optVal) < linux.SizeOfIPTReplace { + return syserr.ErrInvalidArgument + } + + stack := inet.StackFromContext(t) + if stack == nil { + return syserr.ErrNoDevice + } + // Stack must be a netstack stack. + return netfilter.SetEntries(stack.(*Stack).Stack, optVal) + + case linux.IPT_SO_SET_ADD_COUNTERS: + // TODO(gvisor.dev/issue/170): Counter support. + return nil + } + } + return SetSockOpt(t, s, s.Endpoint, level, name, optVal) } @@ -1413,7 +1516,20 @@ func setSockOptSocket(t *kernel.Task, s socket.Socket, ep commonEndpoint, name i if n == -1 { n = len(optVal) } - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(optVal[:n]))) + name := string(optVal[:n]) + if name == "" { + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(0))) + } + s := t.NetworkContext() + if s == nil { + return syserr.ErrNoDevice + } + for nicID, nic := range s.Interfaces() { + if nic.Name == name { + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.BindToDeviceOption(nicID))) + } + } + return syserr.ErrUnknownDevice case linux.SO_BROADCAST: if len(optVal) < sizeOfInt32 { @@ -1561,6 +1677,17 @@ func setSockOptTCP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) * } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.KeepaliveIntervalOption(time.Second * time.Duration(v)))) + case linux.TCP_USER_TIMEOUT: + if len(optVal) < sizeOfInt32 { + return syserr.ErrInvalidArgument + } + + v := int32(usermem.ByteOrder.Uint32(optVal)) + if v < 0 { + return syserr.ErrInvalidArgument + } + return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.TCPUserTimeoutOption(time.Millisecond * time.Duration(v)))) + case linux.TCP_CONGESTION: v := tcpip.CongestionControlOption(optVal) if err := ep.SetSockOpt(v); err != nil { @@ -1596,7 +1723,7 @@ func setSockOptIPv6(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) } v := usermem.ByteOrder.Uint32(optVal) - return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.V6OnlyOption(v))) + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.V6OnlyOption, v != 0)) case linux.IPV6_ADD_MEMBERSHIP, linux.IPV6_DROP_MEMBERSHIP, @@ -1783,6 +1910,13 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s } return syserr.TranslateNetstackError(ep.SetSockOpt(tcpip.IPv4TOSOption(v))) + case linux.IP_RECVTOS: + v, err := parseIntOrChar(optVal) + if err != nil { + return err + } + return syserr.TranslateNetstackError(ep.SetSockOptBool(tcpip.ReceiveTOSOption, v != 0)) + case linux.IP_ADD_SOURCE_MEMBERSHIP, linux.IP_BIND_ADDRESS_NO_PORT, linux.IP_BLOCK_SOURCE, @@ -1803,7 +1937,6 @@ func setSockOptIP(t *kernel.Task, ep commonEndpoint, name int, optVal []byte) *s linux.IP_RECVFRAGSIZE, linux.IP_RECVOPTS, linux.IP_RECVORIGDSTADDR, - linux.IP_RECVTOS, linux.IP_RECVTTL, linux.IP_RETOPTS, linux.IP_TRANSPARENT, @@ -2001,8 +2134,8 @@ func ConvertAddress(family int, addr tcpip.FullAddress) (linux.SockAddr, uint32) case linux.AF_INET6: var out linux.SockAddrInet6 - if len(addr.Addr) == 4 { - // Copy address is v4-mapped format. + if len(addr.Addr) == header.IPv4AddressSize { + // Copy address in v4-mapped format. copy(out.Addr[12:], addr.Addr) out.Addr[10] = 0xff out.Addr[11] = 0xff @@ -2223,7 +2356,14 @@ func (s *SocketOperations) nonBlockingRead(ctx context.Context, dst usermem.IOSe } func (s *SocketOperations) controlMessages() socket.ControlMessages { - return socket.ControlMessages{IP: tcpip.ControlMessages{HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, Timestamp: s.readCM.Timestamp}} + return socket.ControlMessages{ + IP: tcpip.ControlMessages{ + HasTimestamp: s.readCM.HasTimestamp && s.sockOptTimestamp, + Timestamp: s.readCM.Timestamp, + HasTOS: s.readCM.HasTOS, + TOS: s.readCM.TOS, + }, + } } // updateTimestamp sets the timestamp for SIOCGSTAMP. It should be called after @@ -2316,10 +2456,14 @@ func (s *SocketOperations) SendMsg(t *kernel.Task, src usermem.IOSequence, to [] var addr *tcpip.FullAddress if len(to) > 0 { - addrBuf, _, err := AddressAndFamily(s.family, to, true /* strict */) + addrBuf, family, err := AddressAndFamily(to) if err != nil { return 0, err } + if err := s.checkFamily(family, false /* exact */); err != nil { + return 0, err + } + addrBuf = s.mapFamily(addrBuf, family) addr = &addrBuf } diff --git a/pkg/sentry/socket/rpcinet/conn/BUILD b/pkg/sentry/socket/rpcinet/conn/BUILD index 23eadcb1b..b2677c659 100644 --- a/pkg/sentry/socket/rpcinet/conn/BUILD +++ b/pkg/sentry/socket/rpcinet/conn/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/binary", "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", + "//pkg/sync", "//pkg/syserr", "//pkg/unet", "@com_github_golang_protobuf//proto:go_default_library", diff --git a/pkg/sentry/socket/rpcinet/conn/conn.go b/pkg/sentry/socket/rpcinet/conn/conn.go index 356adad99..02f39c767 100644 --- a/pkg/sentry/socket/rpcinet/conn/conn.go +++ b/pkg/sentry/socket/rpcinet/conn/conn.go @@ -17,12 +17,12 @@ package conn import ( "fmt" - "sync" "sync/atomic" "syscall" "github.com/golang/protobuf/proto" "gvisor.dev/gvisor/pkg/binary" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/unet" diff --git a/pkg/sentry/socket/rpcinet/notifier/BUILD b/pkg/sentry/socket/rpcinet/notifier/BUILD index a3585e10d..a5954f22b 100644 --- a/pkg/sentry/socket/rpcinet/notifier/BUILD +++ b/pkg/sentry/socket/rpcinet/notifier/BUILD @@ -10,6 +10,7 @@ go_library( deps = [ "//pkg/sentry/socket/rpcinet:syscall_rpc_go_proto", "//pkg/sentry/socket/rpcinet/conn", + "//pkg/sync", "//pkg/waiter", "@org_golang_x_sys//unix:go_default_library", ], diff --git a/pkg/sentry/socket/rpcinet/notifier/notifier.go b/pkg/sentry/socket/rpcinet/notifier/notifier.go index 7efe4301f..82b75d6dd 100644 --- a/pkg/sentry/socket/rpcinet/notifier/notifier.go +++ b/pkg/sentry/socket/rpcinet/notifier/notifier.go @@ -17,12 +17,12 @@ package notifier import ( "fmt" - "sync" "syscall" "golang.org/x/sys/unix" "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/conn" pb "gvisor.dev/gvisor/pkg/sentry/socket/rpcinet/syscall_rpc_go_proto" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/rpcinet/syscall_rpc.proto b/pkg/sentry/socket/rpcinet/syscall_rpc.proto index 9586f5923..b677e9eb3 100644 --- a/pkg/sentry/socket/rpcinet/syscall_rpc.proto +++ b/pkg/sentry/socket/rpcinet/syscall_rpc.proto @@ -3,7 +3,6 @@ syntax = "proto3"; // package syscall_rpc is a set of networking related system calls that can be // forwarded to a socket gofer. // -// TODO(b/77963526): Document individual RPCs. package syscall_rpc; message SendmsgRequest { diff --git a/pkg/sentry/socket/socket.go b/pkg/sentry/socket/socket.go index 8c250c325..2389a9cdb 100644 --- a/pkg/sentry/socket/socket.go +++ b/pkg/sentry/socket/socket.go @@ -43,6 +43,11 @@ type ControlMessages struct { IP tcpip.ControlMessages } +// Release releases Unix domain socket credentials and rights. +func (c *ControlMessages) Release() { + c.Unix.Release() +} + // Socket is the interface containing socket syscalls used by the syscall layer // to redirect them to the appropriate implementation. type Socket interface { diff --git a/pkg/sentry/socket/unix/io.go b/pkg/sentry/socket/unix/io.go index 2ec1a662d..2447f24ef 100644 --- a/pkg/sentry/socket/unix/io.go +++ b/pkg/sentry/socket/unix/io.go @@ -83,6 +83,19 @@ type EndpointReader struct { ControlTrunc bool } +// Truncate calls RecvMsg on the endpoint without writing to a destination. +func (r *EndpointReader) Truncate() error { + // Ignore bytes read since it will always be zero. + _, ms, c, ct, err := r.Endpoint.RecvMsg(r.Ctx, [][]byte{}, r.Creds, r.NumRights, r.Peek, r.From) + r.Control = c + r.ControlTrunc = ct + r.MsgSize = ms + if err != nil { + return err.ToError() + } + return nil +} + // ReadToBlocks implements safemem.Reader.ReadToBlocks. func (r *EndpointReader) ReadToBlocks(dsts safemem.BlockSeq) (uint64, error) { return safemem.FromVecReaderFunc{func(bufs [][]byte) (int64, error) { diff --git a/pkg/sentry/socket/unix/transport/BUILD b/pkg/sentry/socket/unix/transport/BUILD index 788ad70d2..d7ba95dff 100644 --- a/pkg/sentry/socket/unix/transport/BUILD +++ b/pkg/sentry/socket/unix/transport/BUILD @@ -32,6 +32,7 @@ go_library( "//pkg/ilist", "//pkg/refs", "//pkg/sentry/context", + "//pkg/sync", "//pkg/syserr", "//pkg/tcpip", "//pkg/tcpip/buffer", diff --git a/pkg/sentry/socket/unix/transport/connectioned.go b/pkg/sentry/socket/unix/transport/connectioned.go index dea11e253..9e6fbc111 100644 --- a/pkg/sentry/socket/unix/transport/connectioned.go +++ b/pkg/sentry/socket/unix/transport/connectioned.go @@ -15,10 +15,9 @@ package transport import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/waiter" diff --git a/pkg/sentry/socket/unix/transport/queue.go b/pkg/sentry/socket/unix/transport/queue.go index e27b1c714..5dcd3d95e 100644 --- a/pkg/sentry/socket/unix/transport/queue.go +++ b/pkg/sentry/socket/unix/transport/queue.go @@ -15,9 +15,8 @@ package transport import ( - "sync" - "gvisor.dev/gvisor/pkg/refs" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/waiter" ) diff --git a/pkg/sentry/socket/unix/transport/unix.go b/pkg/sentry/socket/unix/transport/unix.go index 529a7a7a9..fcc0da332 100644 --- a/pkg/sentry/socket/unix/transport/unix.go +++ b/pkg/sentry/socket/unix/transport/unix.go @@ -16,11 +16,11 @@ package transport import ( - "sync" "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserr" "gvisor.dev/gvisor/pkg/tcpip" "gvisor.dev/gvisor/pkg/tcpip/buffer" @@ -175,17 +175,25 @@ type Endpoint interface { // types. SetSockOpt(opt interface{}) *tcpip.Error + // SetSockOptBool sets a socket option for simple cases when a value has + // the int type. + SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error + // SetSockOptInt sets a socket option for simple cases when a value has // the int type. - SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error + SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error // GetSockOpt gets a socket option. opt should be a pointer to one of the // tcpip.*Option types. GetSockOpt(opt interface{}) *tcpip.Error + // GetSockOptBool gets a socket option for simple cases when a return + // value has the int type. + GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) + // GetSockOptInt gets a socket option for simple cases when a return // value has the int type. - GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) + GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) // State returns the current state of the socket, as represented by Linux in // procfs. @@ -851,11 +859,19 @@ func (e *baseEndpoint) SetSockOpt(opt interface{}) *tcpip.Error { return nil } -func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOpt, v int) *tcpip.Error { +func (e *baseEndpoint) SetSockOptBool(opt tcpip.SockOptBool, v bool) *tcpip.Error { return nil } -func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOpt) (int, *tcpip.Error) { +func (e *baseEndpoint) SetSockOptInt(opt tcpip.SockOptInt, v int) *tcpip.Error { + return nil +} + +func (e *baseEndpoint) GetSockOptBool(opt tcpip.SockOptBool) (bool, *tcpip.Error) { + return false, tcpip.ErrUnknownProtocolOption +} + +func (e *baseEndpoint) GetSockOptInt(opt tcpip.SockOptInt) (int, *tcpip.Error) { switch opt { case tcpip.ReceiveQueueSizeOption: v := 0 diff --git a/pkg/sentry/socket/unix/unix.go b/pkg/sentry/socket/unix/unix.go index 1aaae8487..7f49ba864 100644 --- a/pkg/sentry/socket/unix/unix.go +++ b/pkg/sentry/socket/unix/unix.go @@ -116,10 +116,16 @@ func (s *SocketOperations) Endpoint() transport.Endpoint { // extractPath extracts and validates the address. func extractPath(sockaddr []byte) (string, *syserr.Error) { - addr, _, err := netstack.AddressAndFamily(linux.AF_UNIX, sockaddr, true /* strict */) + addr, family, err := netstack.AddressAndFamily(sockaddr) if err != nil { + if err == syserr.ErrAddressFamilyNotSupported { + err = syserr.ErrInvalidArgument + } return "", err } + if family != linux.AF_UNIX { + return "", syserr.ErrInvalidArgument + } // The address is trimmed by GetAddress. p := string(addr.Addr) @@ -541,8 +547,27 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags if senderRequested { r.From = &tcpip.FullAddress{} } + + doRead := func() (int64, error) { + return dst.CopyOutFrom(t, &r) + } + + // If MSG_TRUNC is set with a zero byte destination then we still need + // to read the message and discard it, or in the case where MSG_PEEK is + // set, leave it be. In both cases the full message length must be + // returned. + if trunc && dst.Addrs.NumBytes() == 0 { + doRead = func() (int64, error) { + err := r.Truncate() + // Always return zero for bytes read since the destination size is + // zero. + return 0, err + } + + } + var total int64 - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock || dontWait { + if n, err := doRead(); err != syserror.ErrWouldBlock || dontWait { var from linux.SockAddr var fromLen uint32 if r.From != nil && len([]byte(r.From.Addr)) != 0 { @@ -577,7 +602,7 @@ func (s *SocketOperations) RecvMsg(t *kernel.Task, dst usermem.IOSequence, flags defer s.EventUnregister(&e) for { - if n, err := dst.CopyOutFrom(t, &r); err != syserror.ErrWouldBlock { + if n, err := doRead(); err != syserror.ErrWouldBlock { var from linux.SockAddr var fromLen uint32 if r.From != nil { |