diff options
| -rw-r--r-- | pkg/abi/linux/netfilter.go | 89 | ||||
| -rw-r--r-- | pkg/sentry/socket/netfilter/BUILD | 1 | ||||
| -rw-r--r-- | pkg/sentry/socket/netfilter/netfilter.go | 408 | ||||
| -rw-r--r-- | pkg/sentry/socket/netstack/netstack.go | 20 | ||||
| -rw-r--r-- | pkg/sentry/syscalls/linux/sys_socket.go | 2 | ||||
| -rw-r--r-- | pkg/tcpip/iptables/BUILD | 5 | ||||
| -rw-r--r-- | pkg/tcpip/iptables/iptables.go | 120 | ||||
| -rw-r--r-- | pkg/tcpip/iptables/targets.go | 16 | ||||
| -rw-r--r-- | pkg/tcpip/iptables/types.go | 49 |
9 files changed, 515 insertions, 195 deletions
diff --git a/pkg/abi/linux/netfilter.go b/pkg/abi/linux/netfilter.go index 269ba5567..33fcc6c95 100644 --- a/pkg/abi/linux/netfilter.go +++ b/pkg/abi/linux/netfilter.go @@ -42,6 +42,15 @@ const ( NF_RETURN = -NF_REPEAT - 1 ) +// VerdictStrings maps int verdicts to the strings they represent. It is used +// for debugging. +var VerdictStrings = map[int32]string{ + -NF_DROP - 1: "DROP", + -NF_ACCEPT - 1: "ACCEPT", + -NF_QUEUE - 1: "QUEUE", + NF_RETURN: "RETURN", +} + // Socket options. These correspond to values in // include/uapi/linux/netfilter_ipv4/ip_tables.h. const ( @@ -179,7 +188,7 @@ const SizeOfXTCounters = 16 // the user data. type XTEntryMatch struct { MatchSize uint16 - Name [XT_EXTENSION_MAXNAMELEN]byte + Name ExtensionName Revision uint8 // Data is omitted here because it would cause XTEntryMatch to be an // extra byte larger (see http://www.catb.org/esr/structure-packing/). @@ -199,7 +208,7 @@ const SizeOfXTEntryMatch = 32 // the user data. type XTEntryTarget struct { TargetSize uint16 - Name [XT_EXTENSION_MAXNAMELEN]byte + Name ExtensionName Revision uint8 // Data is omitted here because it would cause XTEntryTarget to be an // extra byte larger (see http://www.catb.org/esr/structure-packing/). @@ -226,9 +235,9 @@ const SizeOfXTStandardTarget = 40 // ErrorName. It corresponds to struct xt_error_target in // include/uapi/linux/netfilter/x_tables.h. type XTErrorTarget struct { - Target XTEntryTarget - ErrorName [XT_FUNCTION_MAXNAMELEN]byte - _ [2]byte + Target XTEntryTarget + Name ErrorName + _ [2]byte } // SizeOfXTErrorTarget is the size of an XTErrorTarget. @@ -237,7 +246,7 @@ const SizeOfXTErrorTarget = 64 // IPTGetinfo is the argument for the IPT_SO_GET_INFO sockopt. It corresponds // to struct ipt_getinfo in include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTGetinfo struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName ValidHooks uint32 HookEntry [NF_INET_NUMHOOKS]uint32 Underflow [NF_INET_NUMHOOKS]uint32 @@ -248,16 +257,11 @@ type IPTGetinfo struct { // SizeOfIPTGetinfo is the size of an IPTGetinfo. const SizeOfIPTGetinfo = 84 -// TableName returns the table name. -func (info *IPTGetinfo) TableName() string { - return tableName(info.Name[:]) -} - // IPTGetEntries is the argument for the IPT_SO_GET_ENTRIES sockopt. It // corresponds to struct ipt_get_entries in // include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTGetEntries struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName Size uint32 _ [4]byte // Entrytable is omitted here because it would cause IPTGetEntries to @@ -266,34 +270,22 @@ type IPTGetEntries struct { // Entrytable [0]IPTEntry } -// TableName returns the entries' table name. -func (entries *IPTGetEntries) TableName() string { - return tableName(entries.Name[:]) -} - // SizeOfIPTGetEntries is the size of an IPTGetEntries. const SizeOfIPTGetEntries = 40 -// KernelIPTGetEntries is identical to IPTEntry, but includes the Elems field. -// This struct marshaled via the binary package to write an KernelIPTGetEntries -// to userspace. +// KernelIPTGetEntries is identical to IPTGetEntries, but includes the +// Entrytable field. This struct marshaled via the binary package to write an +// KernelIPTGetEntries to userspace. type KernelIPTGetEntries struct { - Name [XT_TABLE_MAXNAMELEN]byte - Size uint32 - _ [4]byte + IPTGetEntries Entrytable []KernelIPTEntry } -// TableName returns the entries' table name. -func (entries *KernelIPTGetEntries) TableName() string { - return tableName(entries.Name[:]) -} - // IPTReplace is the argument for the IPT_SO_SET_REPLACE sockopt. It // corresponds to struct ipt_replace in // include/uapi/linux/netfilter_ipv4/ip_tables.h. type IPTReplace struct { - Name [XT_TABLE_MAXNAMELEN]byte + Name TableName ValidHooks uint32 NumEntries uint32 Size uint32 @@ -306,14 +298,45 @@ type IPTReplace struct { // Entries [0]IPTEntry } +// KernelIPTReplace is identical to IPTReplace, but includes the Entries field. +type KernelIPTReplace struct { + IPTReplace + Entries [0]IPTEntry +} + // SizeOfIPTReplace is the size of an IPTReplace. const SizeOfIPTReplace = 96 -func tableName(name []byte) string { - for i, c := range name { +// ExtensionName holds the name of a netfilter extension. +type ExtensionName [XT_EXTENSION_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (en ExtensionName) String() string { + return goString(en[:]) +} + +// TableName holds the name of a netfilter table. +type TableName [XT_TABLE_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (tn TableName) String() string { + return goString(tn[:]) +} + +// ErrorName holds the name of a netfilter error. These can also hold +// user-defined chains. +type ErrorName [XT_FUNCTION_MAXNAMELEN]byte + +// String implements fmt.Stringer. +func (en ErrorName) String() string { + return goString(en[:]) +} + +func goString(cstring []byte) string { + for i, c := range cstring { if c == 0 { - return string(name[:i]) + return string(cstring[:i]) } } - return string(name) + return string(cstring) } 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/netstack/netstack.go b/pkg/sentry/socket/netstack/netstack.go index 0affb8071..099319327 100644 --- a/pkg/sentry/socket/netstack/netstack.go +++ b/pkg/sentry/socket/netstack/netstack.go @@ -1377,6 +1377,26 @@ 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) } diff --git a/pkg/sentry/syscalls/linux/sys_socket.go b/pkg/sentry/syscalls/linux/sys_socket.go index 4b5aafcc0..cda517a81 100644 --- a/pkg/sentry/syscalls/linux/sys_socket.go +++ b/pkg/sentry/syscalls/linux/sys_socket.go @@ -41,7 +41,7 @@ const maxListenBacklog = 1024 const maxAddrLen = 200 // maxOptLen is the maximum sockopt parameter length we're willing to accept. -const maxOptLen = 1024 +const maxOptLen = 1024 * 8 // maxControlLen is the maximum length of the msghdr.msg_control buffer we're // willing to accept. Note that this limit is smaller than Linux, which allows diff --git a/pkg/tcpip/iptables/BUILD b/pkg/tcpip/iptables/BUILD index cc5f531e2..64769c333 100644 --- a/pkg/tcpip/iptables/BUILD +++ b/pkg/tcpip/iptables/BUILD @@ -11,5 +11,8 @@ go_library( ], importpath = "gvisor.dev/gvisor/pkg/tcpip/iptables", visibility = ["//visibility:public"], - deps = ["//pkg/tcpip/buffer"], + deps = [ + "//pkg/log", + "//pkg/tcpip/buffer", + ], ) diff --git a/pkg/tcpip/iptables/iptables.go b/pkg/tcpip/iptables/iptables.go index 68c68d4aa..647970133 100644 --- a/pkg/tcpip/iptables/iptables.go +++ b/pkg/tcpip/iptables/iptables.go @@ -16,66 +16,114 @@ // tool. package iptables +// Table names. const ( - tablenameNat = "nat" - tablenameMangle = "mangle" + TablenameNat = "nat" + TablenameMangle = "mangle" + TablenameFilter = "filter" ) // Chain names as defined by net/ipv4/netfilter/ip_tables.c. const ( - chainNamePrerouting = "PREROUTING" - chainNameInput = "INPUT" - chainNameForward = "FORWARD" - chainNameOutput = "OUTPUT" - chainNamePostrouting = "POSTROUTING" + ChainNamePrerouting = "PREROUTING" + ChainNameInput = "INPUT" + ChainNameForward = "FORWARD" + ChainNameOutput = "OUTPUT" + ChainNamePostrouting = "POSTROUTING" ) +// HookUnset indicates that there is no hook set for an entrypoint or +// underflow. +const HookUnset = -1 + // DefaultTables returns a default set of tables. Each chain is set to accept // all packets. func DefaultTables() IPTables { + // TODO(gvisor.dev/issue/170): We may be able to swap out some strings for + // iotas. return IPTables{ Tables: map[string]Table{ - tablenameNat: Table{ - BuiltinChains: map[Hook]Chain{ - Prerouting: unconditionalAcceptChain(chainNamePrerouting), - Input: unconditionalAcceptChain(chainNameInput), - Output: unconditionalAcceptChain(chainNameOutput), - Postrouting: unconditionalAcceptChain(chainNamePostrouting), + TablenameNat: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, }, - DefaultTargets: map[Hook]Target{ - Prerouting: UnconditionalAcceptTarget{}, - Input: UnconditionalAcceptTarget{}, - Output: UnconditionalAcceptTarget{}, - Postrouting: UnconditionalAcceptTarget{}, + Underflows: map[Hook]int{ + Prerouting: 0, + Input: 1, + Output: 2, + Postrouting: 3, }, - UserChains: map[string]Chain{}, + UserChains: map[string]int{}, }, - tablenameMangle: Table{ - BuiltinChains: map[Hook]Chain{ - Prerouting: unconditionalAcceptChain(chainNamePrerouting), - Output: unconditionalAcceptChain(chainNameOutput), + TablenameMangle: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Prerouting: 0, + Output: 1, }, - DefaultTargets: map[Hook]Target{ - Prerouting: UnconditionalAcceptTarget{}, - Output: UnconditionalAcceptTarget{}, + Underflows: map[Hook]int{ + Prerouting: 0, + Output: 1, }, - UserChains: map[string]Chain{}, + UserChains: map[string]int{}, + }, + TablenameFilter: Table{ + Rules: []Rule{ + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: UnconditionalAcceptTarget{}}, + Rule{Target: ErrorTarget{}}, + }, + BuiltinChains: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + Underflows: map[Hook]int{ + Input: 0, + Forward: 1, + Output: 2, + }, + UserChains: map[string]int{}, }, }, Priorities: map[Hook][]string{ - Prerouting: []string{tablenameMangle, tablenameNat}, - Output: []string{tablenameMangle, tablenameNat}, + Input: []string{TablenameNat, TablenameFilter}, + Prerouting: []string{TablenameMangle, TablenameNat}, + Output: []string{TablenameMangle, TablenameNat, TablenameFilter}, }, } } -func unconditionalAcceptChain(name string) Chain { - return Chain{ - Name: name, - Rules: []Rule{ - Rule{ - Target: UnconditionalAcceptTarget{}, - }, +// EmptyFilterTable returns a Table with no rules and the filter table chains +// mapped to HookUnset. +func EmptyFilterTable() Table { + return Table{ + Rules: []Rule{}, + BuiltinChains: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, + }, + Underflows: map[Hook]int{ + Input: HookUnset, + Forward: HookUnset, + Output: HookUnset, }, + UserChains: map[string]int{}, } } diff --git a/pkg/tcpip/iptables/targets.go b/pkg/tcpip/iptables/targets.go index 19a7f77e3..b94a4c941 100644 --- a/pkg/tcpip/iptables/targets.go +++ b/pkg/tcpip/iptables/targets.go @@ -16,7 +16,10 @@ package iptables -import "gvisor.dev/gvisor/pkg/tcpip/buffer" +import ( + "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/tcpip/buffer" +) // UnconditionalAcceptTarget accepts all packets. type UnconditionalAcceptTarget struct{} @@ -33,3 +36,14 @@ type UnconditionalDropTarget struct{} func (UnconditionalDropTarget) Action(packet buffer.VectorisedView) (Verdict, string) { return Drop, "" } + +// ErrorTarget logs an error and drops the packet. It represents a target that +// should be unreachable. +type ErrorTarget struct{} + +// Action implements Target.Action. +func (ErrorTarget) Action(packet buffer.VectorisedView) (Verdict, string) { + log.Warningf("ErrorTarget triggered.") + return Drop, "" + +} diff --git a/pkg/tcpip/iptables/types.go b/pkg/tcpip/iptables/types.go index 42a79ef9f..540f8c0b4 100644 --- a/pkg/tcpip/iptables/types.go +++ b/pkg/tcpip/iptables/types.go @@ -61,9 +61,12 @@ const ( type Verdict int const ( + // Invalid indicates an unkonwn or erroneous verdict. + Invalid Verdict = iota + // Accept indicates the packet should continue traversing netstack as // normal. - Accept Verdict = iota + Accept // Drop inicates the packet should be dropped, stopping traversing // netstack. @@ -104,29 +107,22 @@ type IPTables struct { Priorities map[Hook][]string } -// A Table defines a set of chains and hooks into the network stack. The -// currently supported tables are: -// * nat -// * mangle +// A Table defines a set of chains and hooks into the network stack. It is +// really just a list of rules with some metadata for entrypoints and such. type Table struct { - // BuiltinChains holds the un-deletable chains built into netstack. If - // a hook isn't present in the map, this table doesn't utilize that - // hook. - BuiltinChains map[Hook]Chain + // Rules holds the rules that make up the table. + Rules []Rule - // DefaultTargets holds a target for each hook that will be executed if - // chain traversal doesn't yield a verdict. - DefaultTargets map[Hook]Target + // BuiltinChains maps builtin chains to their entrypoint rule in Rules. + BuiltinChains map[Hook]int + + // Underflows maps builtin chains to their underflow rule in Rules + // (i.e. the rule to execute if the chain returns without a verdict). + Underflows map[Hook]int // UserChains holds user-defined chains for the keyed by name. Users // can give their chains arbitrary names. - UserChains map[string]Chain - - // Chains maps names to chains for both builtin and user-defined chains. - // Its entries point to Chains already either in BuiltinChains or - // UserChains, and its purpose is to make looking up tables by name - // fast. - Chains map[string]*Chain + UserChains map[string]int // Metadata holds information about the Table that is useful to users // of IPTables, but not to the netstack IPTables code itself. @@ -152,21 +148,6 @@ func (table *Table) SetMetadata(metadata interface{}) { table.metadata = metadata } -// A Chain defines a list of rules for packet processing. When a packet -// traverses a chain, it is checked against each rule until either a rule -// returns a verdict or the chain ends. -// -// By convention, builtin chains end with a rule that matches everything and -// returns either Accept or Drop. User-defined chains end with Return. These -// aren't strictly necessary here, but the iptables tool writes tables this way. -type Chain struct { - // Name is the chain name. - Name string - - // Rules is the list of rules to traverse. - Rules []Rule -} - // A Rule is a packet processing rule. It consists of two pieces. First it // contains zero or more matchers, each of which is a specification of which // packets this rule applies to. If there are no matchers in the rule, it |