diff options
Diffstat (limited to 'pkg/sentry')
-rw-r--r-- | pkg/sentry/socket/netfilter/netfilter.go | 220 |
1 files changed, 193 insertions, 27 deletions
diff --git a/pkg/sentry/socket/netfilter/netfilter.go b/pkg/sentry/socket/netfilter/netfilter.go index c65c36081..6c88a50a6 100644 --- a/pkg/sentry/socket/netfilter/netfilter.go +++ b/pkg/sentry/socket/netfilter/netfilter.go @@ -36,7 +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. +// TODO(gvisor.dev/issue/170): Use metadata to check correctness. type metadata struct { HookEntry [linux.NF_INET_NUMHOOKS]uint32 Underflow [linux.NF_INET_NUMHOOKS]uint32 @@ -44,6 +44,14 @@ type metadata struct { Size uint32 } +const enableDebugLog = true + +func nflog(format string, args ...interface{}) { + if enableDebugLog { + log.Infof("netfilter: "+format, args...) + } +} + // GetInfo returns information about iptables. func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPTGetinfo, *syserr.Error) { // Read in the struct and table name. @@ -72,6 +80,8 @@ func GetInfo(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr) (linux.IPT info.NumEntries = metadata.NumEntries info.Size = metadata.Size + nflog("GetInfo returning info: %+v", info) + return info, nil } @@ -80,21 +90,26 @@ func GetEntries(t *kernel.Task, stack *stack.Stack, outPtr usermem.Addr, outLen // Read in the struct and table name. var userEntries linux.IPTGetEntries if _, err := t.CopyIn(outPtr, &userEntries); err != nil { + log.Warningf("netfilter: couldn't copy in entries %q", userEntries.Name) return linux.KernelIPTGetEntries{}, syserr.FromError(err) } // Find the appropriate table. table, err := findTable(stack, userEntries.Name) if err != nil { + log.Warningf("netfilter: couldn't find table %q", userEntries.Name) return linux.KernelIPTGetEntries{}, err } // Convert netstack's iptables rules to something that the iptables // tool can understand. - entries, _, err := convertNetstackToBinary(userEntries.Name.String(), table) + entries, meta, err := convertNetstackToBinary(userEntries.Name.String(), table) if err != nil { return linux.KernelIPTGetEntries{}, err } + if meta != table.Metadata().(metadata) { + panic(fmt.Sprintf("Table %q metadata changed between writing and reading. Was saved as %+v, but is now %+v", userEntries.Name.String(), table.Metadata().(metadata), meta)) + } if binary.Size(entries) > uintptr(outLen) { log.Warningf("Insufficient GetEntries output size: %d", uintptr(outLen)) return linux.KernelIPTGetEntries{}, syserr.ErrInvalidArgument @@ -148,15 +163,19 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern copy(entries.Name[:], tablename) for ruleIdx, rule := range table.Rules { + nflog("Current offset: %d", entries.Size) + // Is this a chain entry point? for hook, hookRuleIdx := range table.BuiltinChains { if hookRuleIdx == ruleIdx { + nflog("Found hook %d at offset %d", hook, entries.Size) meta.HookEntry[hook] = entries.Size } } // Is this a chain underflow point? for underflow, underflowRuleIdx := range table.Underflows { if underflowRuleIdx == ruleIdx { + nflog("Found underflow %d at offset %d", underflow, entries.Size) meta.Underflow[underflow] = entries.Size } } @@ -176,6 +195,10 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Serialize the matcher and add it to the // entry. serialized := marshalMatcher(matcher) + nflog("matcher serialized as: %v", serialized) + if len(serialized)%8 != 0 { + panic(fmt.Sprintf("matcher %T is not 64-bit aligned", matcher)) + } entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) entry.TargetOffset += uint16(len(serialized)) @@ -183,20 +206,29 @@ func convertNetstackToBinary(tablename string, table iptables.Table) (linux.Kern // Serialize and append the target. serialized := marshalTarget(rule.Target) + if len(serialized)%8 != 0 { + panic(fmt.Sprintf("target %T is not 64-bit aligned", rule.Target)) + } entry.Elems = append(entry.Elems, serialized...) entry.NextOffset += uint16(len(serialized)) + nflog("Adding entry: %+v", entry) + entries.Size += uint32(entry.NextOffset) entries.Entrytable = append(entries.Entrytable, entry) meta.NumEntries++ } + nflog("Finished with an marshalled size of %d", meta.Size) meta.Size = entries.Size return entries, meta, nil } +// TODO: SOMEHOW THIS IS NOT GETTING APPENDED! func marshalMatcher(matcher iptables.Matcher) []byte { - switch matcher.(type) { + switch m := matcher.(type) { + case *iptables.UDPMatcher: + return marshalUDPMatcher(m) default: // TODO(gvisor.dev/issue/170): We don't support any matchers // yet, so any call to marshalMatcher will panic. @@ -204,6 +236,39 @@ func marshalMatcher(matcher iptables.Matcher) []byte { } } +func marshalUDPMatcher(matcher *iptables.UDPMatcher) []byte { + nflog("Marshalling UDP matcher: %+v", matcher) + + linuxMatcher := linux.KernelXTEntryMatch{ + XTEntryMatch: linux.XTEntryMatch{ + MatchSize: linux.SizeOfXTEntryMatch + linux.SizeOfXTUDP + 6, + // Name: "udp", + }, + Data: make([]byte, 0, linux.SizeOfXTUDP), + } + copy(linuxMatcher.Name[:], "udp") + + xtudp := linux.XTUDP{ + SourcePortStart: matcher.Data.SourcePortStart, + SourcePortEnd: matcher.Data.SourcePortEnd, + DestinationPortStart: matcher.Data.DestinationPortStart, + DestinationPortEnd: matcher.Data.DestinationPortEnd, + InverseFlags: matcher.Data.InverseFlags, + } + nflog("marshalUDPMatcher: xtudp: %+v", xtudp) + linuxMatcher.Data = binary.Marshal(linuxMatcher.Data, usermem.ByteOrder, xtudp) + nflog("marshalUDPMatcher: linuxMatcher: %+v", linuxMatcher) + + // We have to pad this struct size to a multiple of 8 bytes, so we make + // this a little longer than it needs to be. + buf := make([]byte, 0, linux.SizeOfXTEntryMatch+linux.SizeOfXTUDP+6) + buf = binary.Marshal(buf, usermem.ByteOrder, linuxMatcher) + buf = append(buf, []byte{0, 0, 0, 0, 0, 0}...) + nflog("Marshalled into matcher of size %d", len(buf)) + nflog("marshalUDPMatcher: buf is: %v", buf) + return buf[:] +} + func marshalTarget(target iptables.Target) []byte { switch target.(type) { case iptables.UnconditionalAcceptTarget: @@ -218,6 +283,9 @@ func marshalTarget(target iptables.Target) []byte { } func marshalStandardTarget(verdict iptables.Verdict) []byte { + nflog("Marshalling standard target with size %d", linux.SizeOfXTStandardTarget) + + // TODO: Must be aligned. // The target's name will be the empty string. target := linux.XTStandardTarget{ Target: linux.XTEntryTarget{ @@ -285,7 +353,7 @@ func translateToStandardVerdict(val int32) (iptables.Verdict, *syserr.Error) { // 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) + // printReplace(optVal) // Get the basic rules data (struct ipt_replace). if len(optVal) < linux.SizeOfIPTReplace { @@ -307,10 +375,14 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return syserr.ErrInvalidArgument } + nflog("Setting entries in table %q", replace.Name.String()) + // Convert input into a list of rules and their offsets. var offset uint32 var offsets []uint32 for entryIdx := uint32(0); entryIdx < replace.NumEntries; entryIdx++ { + nflog("Processing entry at offset %d", offset) + // Get the struct ipt_entry. if len(optVal) < linux.SizeOfIPTEntry { log.Warningf("netfilter: optVal has insufficient size for entry %d", len(optVal)) @@ -318,10 +390,12 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { } 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. + initialOptValLen := len(optVal) + optVal = optVal[linux.SizeOfIPTEntry:] + + if entry.TargetOffset < linux.SizeOfIPTEntry { + log.Warningf("netfilter: entry has too-small target offset %d", entry.TargetOffset) return syserr.ErrInvalidArgument } @@ -332,19 +406,41 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { return err } + // TODO: Matchers (and maybe targets) can specify that they only work for certiain protocols, hooks, tables. + // Get matchers. + matchersSize := entry.TargetOffset - linux.SizeOfIPTEntry + if len(optVal) < int(matchersSize) { + log.Warningf("netfilter: entry doesn't have enough room for its matchers (only %d bytes remain)", len(optVal)) + } + matchers, err := parseMatchers(filter, optVal[:matchersSize]) + if err != nil { + log.Warningf("netfilter: failed to parse matchers: %v", err) + return err + } + optVal = optVal[matchersSize:] + // Get the target of the rule. - target, consumed, err := parseTarget(optVal) + targetSize := entry.NextOffset - entry.TargetOffset + if len(optVal) < int(targetSize) { + log.Warningf("netfilter: entry doesn't have enough room for its target (only %d bytes remain)", len(optVal)) + } + target, err := parseTarget(optVal[:targetSize]) if err != nil { return err } - optVal = optVal[consumed:] + optVal = optVal[targetSize:] table.Rules = append(table.Rules, iptables.Rule{ - Filter: filter, - Target: target, + Filter: filter, + Target: target, + Matchers: matchers, }) offsets = append(offsets, offset) - offset += linux.SizeOfIPTEntry + consumed + offset += uint32(entry.NextOffset) + + if initialOptValLen-len(optVal) != int(entry.NextOffset) { + log.Warningf("netfilter: entry NextOffset is %d, but entry took up %d bytes", entry.NextOffset, initialOptValLen-len(optVal)) + } } // Go through the list of supported hooks for this table and, for each @@ -401,12 +497,82 @@ func SetEntries(stack *stack.Stack, optVal []byte) *syserr.Error { 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) { +// parseMatchers parses 0 or more matchers from optVal. optVal should contain +// only the matchers. +func parseMatchers(filter iptables.IPHeaderFilter, optVal []byte) ([]iptables.Matcher, *syserr.Error) { + nflog("Parsing matchers of size %d", len(optVal)) + var matchers []iptables.Matcher + for len(optVal) > 0 { + nflog("parseMatchers: optVal has len %d", len(optVal)) + // Get the XTEntryMatch. + if len(optVal) < linux.SizeOfXTEntryMatch { + log.Warningf("netfilter: optVal has insufficient size for entry match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + var match linux.XTEntryMatch + buf := optVal[:linux.SizeOfXTEntryMatch] + binary.Unmarshal(buf, usermem.ByteOrder, &match) + nflog("parseMatchers: parsed entry match %q: %+v", match.Name.String(), match) + + // Check some invariants. + if match.MatchSize < linux.SizeOfXTEntryMatch { + log.Warningf("netfilter: match size is too small, must be at least %d", linux.SizeOfXTEntryMatch) + return nil, syserr.ErrInvalidArgument + } + if len(optVal) < int(match.MatchSize) { + log.Warningf("netfilter: optVal has insufficient size for match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + + buf = optVal[linux.SizeOfXTEntryMatch:match.MatchSize] + var matcher iptables.Matcher + var err error + switch match.Name.String() { + case "udp": + if len(buf) < linux.SizeOfXTUDP { + log.Warningf("netfilter: optVal has insufficient size for UDP match: %d", len(optVal)) + return nil, syserr.ErrInvalidArgument + } + var matchData linux.XTUDP + // For alignment reasons, the match's total size may exceed what's + // strictly necessary to hold matchData. + binary.Unmarshal(buf[:linux.SizeOfXTUDP], usermem.ByteOrder, &matchData) + log.Infof("parseMatchers: parsed XTUDP: %+v", matchData) + matcher, err = iptables.NewUDPMatcher(filter, iptables.UDPMatcherData{ + SourcePortStart: matchData.SourcePortStart, + SourcePortEnd: matchData.SourcePortEnd, + DestinationPortStart: matchData.DestinationPortStart, + DestinationPortEnd: matchData.DestinationPortEnd, + InverseFlags: matchData.InverseFlags, + }) + if err != nil { + log.Warningf("netfilter: failed to create UDP matcher: %v", err) + return nil, syserr.ErrInvalidArgument + } + + default: + log.Warningf("netfilter: unsupported matcher with name %q", match.Name.String()) + return nil, syserr.ErrInvalidArgument + } + + matchers = append(matchers, matcher) + + // TODO: Support revision. + // TODO: Support proto -- matchers usually specify which proto(s) they work with. + optVal = optVal[match.MatchSize:] + } + + // TODO: Check that optVal is exhausted. + return matchers, nil +} + +// parseTarget parses a target from optVal. optVal should contain only the +// target. +func parseTarget(optVal []byte) (iptables.Target, *syserr.Error) { + nflog("Parsing target of size %d", len(optVal)) if len(optVal) < linux.SizeOfXTEntryTarget { log.Warningf("netfilter: optVal has insufficient size for entry target %d", len(optVal)) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } var target linux.XTEntryTarget buf := optVal[:linux.SizeOfXTEntryTarget] @@ -414,9 +580,9 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { 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 + if len(optVal) != linux.SizeOfXTStandardTarget { + log.Warningf("netfilter.SetEntries: optVal has wrong size for standard target %d", len(optVal)) + return nil, syserr.ErrInvalidArgument } var standardTarget linux.XTStandardTarget buf = optVal[:linux.SizeOfXTStandardTarget] @@ -424,22 +590,22 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { verdict, err := translateToStandardVerdict(standardTarget.Verdict) if err != nil { - return nil, 0, err + return nil, err } switch verdict { case iptables.Accept: - return iptables.UnconditionalAcceptTarget{}, linux.SizeOfXTStandardTarget, nil + return iptables.UnconditionalAcceptTarget{}, nil case iptables.Drop: - return iptables.UnconditionalDropTarget{}, linux.SizeOfXTStandardTarget, nil + return iptables.UnconditionalDropTarget{}, nil default: panic(fmt.Sprintf("Unknown verdict: %v", verdict)) } case errorTargetName: // Error target. - if len(optVal) < linux.SizeOfXTErrorTarget { + if len(optVal) != linux.SizeOfXTErrorTarget { log.Infof("netfilter.SetEntries: optVal has insufficient size for error target %d", len(optVal)) - return nil, 0, syserr.ErrInvalidArgument + return nil, syserr.ErrInvalidArgument } var errorTarget linux.XTErrorTarget buf = optVal[:linux.SizeOfXTErrorTarget] @@ -454,16 +620,16 @@ func parseTarget(optVal []byte) (iptables.Target, uint32, *syserr.Error) { // rules have an error with the name of the chain. switch errorTarget.Name.String() { case errorTargetName: - return iptables.ErrorTarget{}, linux.SizeOfXTErrorTarget, nil + return iptables.ErrorTarget{}, nil default: log.Infof("Unknown error target %q doesn't exist or isn't supported yet.", errorTarget.Name.String()) - return nil, 0, syserr.ErrInvalidArgument + return nil, 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 + return nil, syserr.ErrInvalidArgument } func filterFromIPTIP(iptip linux.IPTIP) (iptables.IPHeaderFilter, *syserr.Error) { |