diff options
Diffstat (limited to 'pkg')
-rw-r--r-- | pkg/tcpip/network/ip/generic_multicast_protocol.go | 130 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/igmp.go | 47 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv4/ipv4.go | 21 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ipv6.go | 85 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/mld.go | 83 | ||||
-rw-r--r-- | pkg/tcpip/network/ipv6/ndp.go | 15 | ||||
-rw-r--r-- | pkg/tcpip/stack/addressable_endpoint_state.go | 8 | ||||
-rw-r--r-- | pkg/tcpip/stack/nic.go | 6 |
8 files changed, 343 insertions, 52 deletions
diff --git a/pkg/tcpip/network/ip/generic_multicast_protocol.go b/pkg/tcpip/network/ip/generic_multicast_protocol.go index c8c17ab15..f85c5ff9d 100644 --- a/pkg/tcpip/network/ip/generic_multicast_protocol.go +++ b/pkg/tcpip/network/ip/generic_multicast_protocol.go @@ -30,6 +30,23 @@ type hostState int // The states below are generic across IGMPv2 (RFC 2236 section 6) and MLDv1 // (RFC 2710 section 5). Even though the states are generic across both IGMPv2 // and MLDv1, IGMPv2 terminology will be used. +// +// ______________receive query______________ +// | | +// | _____send or receive report_____ | +// | | | | +// V | V | +// +-------+ +-----------+ +------------+ +-------------------+ +--------+ | +// | Non-M | | Pending-M | | Delaying-M | | Queued Delaying-M | | Idle-M | - +// +-------+ +-----------+ +------------+ +-------------------+ +--------+ +// | ^ | ^ | ^ | ^ +// | | | | | | | | +// ---------- ------- ---------- ------------- +// initialize new send inital fail to send send or receive +// group membership report delayed report report +// +// Not shown in the diagram above, but any state may transition into the non +// member state when a group is left. const ( // nonMember is the "'Non-Member' state, when the host does not belong to the // group on the interface. This is the initial state for all memberships on @@ -41,6 +58,15 @@ const ( // but without advertising the membership to the network. nonMember hostState = iota + // pendingMember is a newly joined member that is waiting to successfully send + // the initial set of reports. + // + // This is not an RFC defined state; it is an implementation specific state to + // track that the initial report needs to be sent. + // + // MAY NOT transition to the idle member state from this state. + pendingMember + // delayingMember is the "'Delaying Member' state, when the host belongs to // the group on the interface and has a report delay timer running for that // membership." @@ -48,6 +74,16 @@ const ( // 'Delaying Listener' is the MLDv1 term used to describe this state. delayingMember + // queuedDelayingMember is a delayingMember that failed to send a report after + // its delayed report timer fired. Hosts in this state are waiting to attempt + // retransmission of the delayed report. + // + // This is not an RFC defined state; it is an implementation specific state to + // track that the delayed report needs to be sent. + // + // May transition to idle member if a report is received for a group. + queuedDelayingMember + // idleMember is the "Idle Member" state, when the host belongs to the group // on the interface and does not have a report delay timer running for that // membership. @@ -56,6 +92,17 @@ const ( idleMember ) +func (s hostState) isDelayingMember() bool { + switch s { + case nonMember, pendingMember, idleMember: + return false + case delayingMember, queuedDelayingMember: + return true + default: + panic(fmt.Sprintf("unrecognized host state = %d", s)) + } +} + // multicastGroupState holds the Generic Multicast Protocol state for a // multicast group. type multicastGroupState struct { @@ -124,7 +171,10 @@ type GenericMulticastProtocolOptions struct { // can be represented by GenericMulticastProtocolState. type MulticastGroupProtocol interface { // SendReport sends a multicast report for the specified group address. - SendReport(groupAddress tcpip.Address) *tcpip.Error + // + // Returns false if the caller should queue the report to be sent later. Note, + // returning false does not mean that the receiver hit an error. + SendReport(groupAddress tcpip.Address) (sent bool, err *tcpip.Error) // SendLeave sends a multicast leave for the specified group address. SendLeave(groupAddress tcpip.Address) *tcpip.Error @@ -166,6 +216,9 @@ type GenericMulticastProtocolState struct { // // The GenericMulticastProtocolState will only grab the lock when timers/jobs // fire. +// +// Note: the methods on opts.Protocol will always be called while protocolMU is +// held. func (g *GenericMulticastProtocolState) Init(protocolMU *sync.RWMutex, opts GenericMulticastProtocolOptions) { if g.memberships != nil { panic("attempted to initialize generic membership protocol state twice") @@ -212,6 +265,29 @@ func (g *GenericMulticastProtocolState) InitializeGroupsLocked() { } } +// SendQueuedReportsLocked attempts to send reports for groups that failed to +// send reports during their last attempt. +// +// Precondition: g.protocolMU must be locked. +func (g *GenericMulticastProtocolState) SendQueuedReportsLocked() { + for groupAddress, info := range g.memberships { + switch info.state { + case nonMember, delayingMember, idleMember: + case pendingMember: + // pendingMembers failed to send their initial unsolicited report so try + // to send the report and queue the extra unsolicited reports. + g.maybeSendInitialReportLocked(groupAddress, &info) + case queuedDelayingMember: + // queuedDelayingMembers failed to send their delayed reports so try to + // send the report and transition them to the idle state. + g.maybeSendDelayedReportLocked(groupAddress, &info) + default: + panic(fmt.Sprintf("unrecognized host state = %d", info.state)) + } + g.memberships[groupAddress] = info + } +} + // JoinGroupLocked handles joining a new group. // // If dontInitialize is true, the group will be not be initialized and will be @@ -239,8 +315,7 @@ func (g *GenericMulticastProtocolState) JoinGroupLocked(groupAddress tcpip.Addre panic(fmt.Sprintf("expected to find group state for group = %s", groupAddress)) } - info.lastToSendReport = g.opts.Protocol.SendReport(groupAddress) == nil - info.state = idleMember + g.maybeSendDelayedReportLocked(groupAddress, &info) g.memberships[groupAddress] = info }), } @@ -347,7 +422,7 @@ func (g *GenericMulticastProtocolState) HandleReportLocked(groupAddress tcpip.Ad // multicast address while it has a timer running for that same address // on that interface, it stops its timer and does not send a Report for // that address, thus suppressing duplicate reports on the link. - if info, ok := g.memberships[groupAddress]; ok && info.state == delayingMember { + if info, ok := g.memberships[groupAddress]; ok && info.state.isDelayingMember() { info.delayedReportJob.Cancel() info.lastToSendReport = false info.state = idleMember @@ -360,10 +435,10 @@ func (g *GenericMulticastProtocolState) HandleReportLocked(groupAddress tcpip.Ad // Precondition: g.protocolMU must be locked. func (g *GenericMulticastProtocolState) initializeNewMemberLocked(groupAddress tcpip.Address, info *multicastGroupState) { if info.state != nonMember { - panic(fmt.Sprintf("state for group %s is not non-member; state = %d", groupAddress, info.state)) + panic(fmt.Sprintf("host must be in non-member state to be initialized; group = %s, state = %d", groupAddress, info.state)) } - info.state = idleMember + info.lastToSendReport = false if groupAddress == g.opts.AllNodesAddress { // As per RFC 2236 section 6 page 10 (for IGMPv2), @@ -379,9 +454,25 @@ func (g *GenericMulticastProtocolState) initializeNewMemberLocked(groupAddress t // case. The node starts in Idle Listener state for that address on // every interface, never transitions to another state, and never sends // a Report or Done for that address. + info.state = idleMember return } + info.state = pendingMember + g.maybeSendInitialReportLocked(groupAddress, info) +} + +// maybeSendInitialReportLocked attempts to start transmission of the initial +// set of reports after newly joining a group. +// +// Host must be in pending member state. +// +// Precondition: g.protocolMU must be locked. +func (g *GenericMulticastProtocolState) maybeSendInitialReportLocked(groupAddress tcpip.Address, info *multicastGroupState) { + if info.state != pendingMember { + panic(fmt.Sprintf("host must be in pending member state to send initial reports; group = %s, state = %d", groupAddress, info.state)) + } + // As per RFC 2236 section 3 page 5 (for IGMPv2), // // When a host joins a multicast group, it should immediately transmit an @@ -399,8 +490,30 @@ func (g *GenericMulticastProtocolState) initializeNewMemberLocked(groupAddress t // // TODO(gvisor.dev/issue/4901): Support a configurable number of initial // unsolicited reports. - info.lastToSendReport = g.opts.Protocol.SendReport(groupAddress) == nil - g.setDelayTimerForAddressRLocked(groupAddress, info, g.opts.MaxUnsolicitedReportDelay) + sent, err := g.opts.Protocol.SendReport(groupAddress) + if err == nil && sent { + info.lastToSendReport = true + g.setDelayTimerForAddressRLocked(groupAddress, info, g.opts.MaxUnsolicitedReportDelay) + } +} + +// maybeSendDelayedReportLocked attempts to send the delayed report. +// +// Host must be in pending, delaying or queued delaying member state. +// +// Precondition: g.protocolMU must be locked. +func (g *GenericMulticastProtocolState) maybeSendDelayedReportLocked(groupAddress tcpip.Address, info *multicastGroupState) { + if !info.state.isDelayingMember() { + panic(fmt.Sprintf("host must be in delaying or queued delaying member state to send delayed reports; group = %s, state = %d", groupAddress, info.state)) + } + + sent, err := g.opts.Protocol.SendReport(groupAddress) + if err == nil && sent { + info.lastToSendReport = true + info.state = idleMember + } else { + info.state = queuedDelayingMember + } } // maybeSendLeave attempts to send a leave message. @@ -531,6 +644,7 @@ func (g *GenericMulticastProtocolState) setDelayTimerForAddressRLocked(groupAddr // TODO: Reset the timer if time remaining is greater than maxResponseTime. return } + info.state = delayingMember info.delayedReportJob.Cancel() info.delayedReportJob.Schedule(g.calculateDelayTimerDuration(maxResponseTime)) diff --git a/pkg/tcpip/network/ipv4/igmp.go b/pkg/tcpip/network/ipv4/igmp.go index a3a7176a0..fb7a9e68e 100644 --- a/pkg/tcpip/network/ipv4/igmp.go +++ b/pkg/tcpip/network/ipv4/igmp.go @@ -96,7 +96,9 @@ type igmpState struct { } // SendReport implements ip.MulticastGroupProtocol. -func (igmp *igmpState) SendReport(groupAddress tcpip.Address) *tcpip.Error { +// +// Precondition: igmp.ep.mu must be read locked. +func (igmp *igmpState) SendReport(groupAddress tcpip.Address) (bool, *tcpip.Error) { igmpType := header.IGMPv2MembershipReport if igmp.v1Present() { igmpType = header.IGMPv1MembershipReport @@ -105,6 +107,8 @@ func (igmp *igmpState) SendReport(groupAddress tcpip.Address) *tcpip.Error { } // SendLeave implements ip.MulticastGroupProtocol. +// +// Precondition: igmp.ep.mu must be read locked. func (igmp *igmpState) SendLeave(groupAddress tcpip.Address) *tcpip.Error { // As per RFC 2236 Section 6, Page 8: "If the interface state says the // Querier is running IGMPv1, this action SHOULD be skipped. If the flag @@ -113,7 +117,8 @@ func (igmp *igmpState) SendLeave(groupAddress tcpip.Address) *tcpip.Error { if igmp.v1Present() { return nil } - return igmp.writePacket(header.IPv4AllRoutersGroup, groupAddress, header.IGMPLeaveGroup) + _, err := igmp.writePacket(header.IPv4AllRoutersGroup, groupAddress, header.IGMPLeaveGroup) + return err } // init sets up an igmpState struct, and is required to be called before using @@ -235,9 +240,10 @@ func (igmp *igmpState) handleMembershipReport(groupAddress tcpip.Address) { igmp.genericMulticastProtocol.HandleReportLocked(groupAddress) } -// writePacket assembles and sends an IGMP packet with the provided fields, -// incrementing the provided stat counter on success. -func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip.Address, igmpType header.IGMPType) *tcpip.Error { +// writePacket assembles and sends an IGMP packet. +// +// Precondition: igmp.ep.mu must be read locked. +func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip.Address, igmpType header.IGMPType) (bool, *tcpip.Error) { igmpData := header.IGMP(buffer.NewView(header.IGMPReportMinimumSize)) igmpData.SetType(igmpType) igmpData.SetGroupAddress(groupAddress) @@ -248,9 +254,13 @@ func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip Data: buffer.View(igmpData).ToVectorisedView(), }) - // TODO(gvisor.dev/issue/4888): We should not use the unspecified address, - // rather we should select an appropriate local address. - localAddr := header.IPv4Any + addressEndpoint := igmp.ep.acquireOutgoingPrimaryAddressRLocked(destAddress, false /* allowExpired */) + if addressEndpoint == nil { + return false, nil + } + localAddr := addressEndpoint.AddressWithPrefix().Address + addressEndpoint.DecRef() + addressEndpoint = nil igmp.ep.addIPHeader(localAddr, destAddress, pkt, stack.NetworkHeaderParams{ Protocol: header.IGMPProtocolNumber, TTL: header.IGMPTTL, @@ -259,22 +269,22 @@ func (igmp *igmpState) writePacket(destAddress tcpip.Address, groupAddress tcpip &header.IPv4SerializableRouterAlertOption{}, }) - sent := igmp.ep.protocol.stack.Stats().IGMP.PacketsSent + sentStats := igmp.ep.protocol.stack.Stats().IGMP.PacketsSent if err := igmp.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv4Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil { - sent.Dropped.Increment() - return err + sentStats.Dropped.Increment() + return false, err } switch igmpType { case header.IGMPv1MembershipReport: - sent.V1MembershipReport.Increment() + sentStats.V1MembershipReport.Increment() case header.IGMPv2MembershipReport: - sent.V2MembershipReport.Increment() + sentStats.V2MembershipReport.Increment() case header.IGMPLeaveGroup: - sent.LeaveGroup.Increment() + sentStats.LeaveGroup.Increment() default: panic(fmt.Sprintf("unrecognized igmp type = %d", igmpType)) } - return nil + return true, nil } // joinGroup handles adding a new group to the membership map, setting up the @@ -325,3 +335,10 @@ func (igmp *igmpState) softLeaveAll() { func (igmp *igmpState) initializeAll() { igmp.genericMulticastProtocol.InitializeGroupsLocked() } + +// sendQueuedReports attempts to send any reports that are queued for sending. +// +// Precondition: igmp.ep.mu must be locked. +func (igmp *igmpState) sendQueuedReports() { + igmp.genericMulticastProtocol.SendQueuedReportsLocked() +} diff --git a/pkg/tcpip/network/ipv4/ipv4.go b/pkg/tcpip/network/ipv4/ipv4.go index c63ecca4a..e9ff70d04 100644 --- a/pkg/tcpip/network/ipv4/ipv4.go +++ b/pkg/tcpip/network/ipv4/ipv4.go @@ -172,7 +172,7 @@ func (e *endpoint) Disable() { } func (e *endpoint) disableLocked() { - if !e.setEnabled(false) { + if !e.isEnabled() { return } @@ -189,6 +189,10 @@ func (e *endpoint) disableLocked() { if err := e.mu.addressableEndpointState.RemovePermanentAddress(ipv4BroadcastAddr.Address); err != nil && err != tcpip.ErrBadLocalAddress { panic(fmt.Sprintf("unexpected error when removing address = %s: %s", ipv4BroadcastAddr.Address, err)) } + + if !e.setEnabled(false) { + panic("should have only done work to disable the endpoint if it was enabled") + } } // DefaultTTL is the default time-to-live value for this endpoint. @@ -780,7 +784,12 @@ func (e *endpoint) Close() { func (e *endpoint) AddAndAcquirePermanentAddress(addr tcpip.AddressWithPrefix, peb stack.PrimaryEndpointBehavior, configType stack.AddressConfigType, deprecated bool) (stack.AddressEndpoint, *tcpip.Error) { e.mu.Lock() defer e.mu.Unlock() - return e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) + + ep, err := e.mu.addressableEndpointState.AddAndAcquirePermanentAddress(addr, peb, configType, deprecated) + if err == nil { + e.mu.igmp.sendQueuedReports() + } + return ep, err } // RemovePermanentAddress implements stack.AddressableEndpoint. @@ -815,6 +824,14 @@ func (e *endpoint) AcquireAssignedAddress(localAddr tcpip.Address, allowTemp boo func (e *endpoint) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { e.mu.RLock() defer e.mu.RUnlock() + return e.acquireOutgoingPrimaryAddressRLocked(remoteAddr, allowExpired) +} + +// acquireOutgoingPrimaryAddressRLocked is like AcquireOutgoingPrimaryAddress +// but with locking requirements +// +// Precondition: igmp.ep.mu must be read locked. +func (e *endpoint) acquireOutgoingPrimaryAddressRLocked(remoteAddr tcpip.Address, allowExpired bool) stack.AddressEndpoint { return e.mu.addressableEndpointState.AcquireOutgoingPrimaryAddress(remoteAddr, allowExpired) } diff --git a/pkg/tcpip/network/ipv6/ipv6.go b/pkg/tcpip/network/ipv6/ipv6.go index 7288e309c..e506e99e9 100644 --- a/pkg/tcpip/network/ipv6/ipv6.go +++ b/pkg/tcpip/network/ipv6/ipv6.go @@ -121,6 +121,45 @@ type OpaqueInterfaceIdentifierOptions struct { SecretKey []byte } +// onAddressAssignedLocked handles an address being assigned. +// +// Precondition: e.mu must be exclusively locked. +func (e *endpoint) onAddressAssignedLocked(addr tcpip.Address) { + // As per RFC 2710 section 3, + // + // All MLD messages described in this document are sent with a link-local + // IPv6 Source Address, ... + // + // If we just completed DAD for a link-local address, then attempt to send any + // queued MLD reports. Note, we may have sent reports already for some of the + // groups before we had a valid link-local address to use as the source for + // the MLD messages, but that was only so that MLD snooping switches are aware + // of our membership to groups - routers would not have handled those reports. + // + // As per RFC 3590 section 4, + // + // MLD Report and Done messages are sent with a link-local address as + // the IPv6 source address, if a valid address is available on the + // interface. If a valid link-local address is not available (e.g., one + // has not been configured), the message is sent with the unspecified + // address (::) as the IPv6 source address. + // + // Once a valid link-local address is available, a node SHOULD generate + // new MLD Report messages for all multicast addresses joined on the + // interface. + // + // Routers receiving an MLD Report or Done message with the unspecified + // address as the IPv6 source address MUST silently discard the packet + // without taking any action on the packets contents. + // + // Snooping switches MUST manage multicast forwarding state based on MLD + // Report and Done messages sent with the unspecified address as the + // IPv6 source address. + if header.IsV6LinkLocalAddress(addr) { + e.mu.mld.sendQueuedReports() + } +} + // InvalidateDefaultRouter implements stack.NDPEndpoint. func (e *endpoint) InvalidateDefaultRouter(rtr tcpip.Address) { e.mu.Lock() @@ -333,7 +372,7 @@ func (e *endpoint) Disable() { } func (e *endpoint) disableLocked() { - if !e.setEnabled(false) { + if !e.Enabled() { return } @@ -349,6 +388,10 @@ func (e *endpoint) disableLocked() { // Leave groups from the perspective of MLD so that routers know that // we are no longer interested in the group. e.mu.mld.softLeaveAll() + + if !e.setEnabled(false) { + panic("should have only done work to disable the endpoint if it was enabled") + } } // stopDADForPermanentAddressesLocked stops DAD for all permaneent addresses. @@ -1176,13 +1219,6 @@ func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPre return addressEndpoint, nil } - snmc := header.SolicitedNodeAddr(addr.Address) - if err := e.joinGroupLocked(snmc); err != nil { - // joinGroupLocked only returns an error if the group address is not a valid - // IPv6 multicast address. - panic(fmt.Sprintf("e.joinGroupLocked(%s): %s", snmc, err)) - } - addressEndpoint.SetKind(stack.PermanentTentative) if e.Enabled() { @@ -1191,6 +1227,13 @@ func (e *endpoint) addAndAcquirePermanentAddressLocked(addr tcpip.AddressWithPre } } + snmc := header.SolicitedNodeAddr(addr.Address) + if err := e.joinGroupLocked(snmc); err != nil { + // joinGroupLocked only returns an error if the group address is not a valid + // IPv6 multicast address. + panic(fmt.Sprintf("e.joinGroupLocked(%s): %s", snmc, err)) + } + return addressEndpoint, nil } @@ -1292,6 +1335,26 @@ func (e *endpoint) AcquireOutgoingPrimaryAddress(remoteAddr tcpip.Address, allow return e.acquireOutgoingPrimaryAddressRLocked(remoteAddr, allowExpired) } +// getLinkLocalAddressRLocked returns a link-local address from the primary list +// of addresses, if one is available. +// +// See stack.PrimaryEndpointBehavior for more details about the primary list. +// +// Precondition: e.mu must be read locked. +func (e *endpoint) getLinkLocalAddressRLocked() tcpip.Address { + var linkLocalAddr tcpip.Address + e.mu.addressableEndpointState.ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) bool { + if addressEndpoint.IsAssigned(false /* allowExpired */) { + if addr := addressEndpoint.AddressWithPrefix().Address; header.IsV6LinkLocalAddress(addr) { + linkLocalAddr = addr + return false + } + } + return true + }) + return linkLocalAddr +} + // acquireOutgoingPrimaryAddressRLocked is like AcquireOutgoingPrimaryAddress // but with locking requirements. // @@ -1311,10 +1374,10 @@ func (e *endpoint) acquireOutgoingPrimaryAddressRLocked(remoteAddr tcpip.Address // Create a candidate set of available addresses we can potentially use as a // source address. var cs []addrCandidate - e.mu.addressableEndpointState.ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) { + e.mu.addressableEndpointState.ForEachPrimaryEndpoint(func(addressEndpoint stack.AddressEndpoint) bool { // If r is not valid for outgoing connections, it is not a valid endpoint. if !addressEndpoint.IsAssigned(allowExpired) { - return + return true } addr := addressEndpoint.AddressWithPrefix().Address @@ -1330,6 +1393,8 @@ func (e *endpoint) acquireOutgoingPrimaryAddressRLocked(remoteAddr tcpip.Address addressEndpoint: addressEndpoint, scope: scope, }) + + return true }) remoteScope, err := header.ScopeForIPv6Address(remoteAddr) diff --git a/pkg/tcpip/network/ipv6/mld.go b/pkg/tcpip/network/ipv6/mld.go index b67eafdba..48644d9c8 100644 --- a/pkg/tcpip/network/ipv6/mld.go +++ b/pkg/tcpip/network/ipv6/mld.go @@ -59,13 +59,18 @@ type mldState struct { } // SendReport implements ip.MulticastGroupProtocol. -func (mld *mldState) SendReport(groupAddress tcpip.Address) *tcpip.Error { +// +// Precondition: mld.ep.mu must be read locked. +func (mld *mldState) SendReport(groupAddress tcpip.Address) (bool, *tcpip.Error) { return mld.writePacket(groupAddress, groupAddress, header.ICMPv6MulticastListenerReport) } // SendLeave implements ip.MulticastGroupProtocol. +// +// Precondition: mld.ep.mu must be read locked. func (mld *mldState) SendLeave(groupAddress tcpip.Address) *tcpip.Error { - return mld.writePacket(header.IPv6AllRoutersMulticastAddress, groupAddress, header.ICMPv6MulticastListenerDone) + _, err := mld.writePacket(header.IPv6AllRoutersMulticastAddress, groupAddress, header.ICMPv6MulticastListenerDone) + return err } // init sets up an mldState struct, and is required to be called before using @@ -147,7 +152,17 @@ func (mld *mldState) initializeAll() { mld.genericMulticastProtocol.InitializeGroupsLocked() } -func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldType header.ICMPv6Type) *tcpip.Error { +// sendQueuedReports attempts to send any reports that are queued for sending. +// +// Precondition: mld.ep.mu must be locked. +func (mld *mldState) sendQueuedReports() { + mld.genericMulticastProtocol.SendQueuedReportsLocked() +} + +// writePacket assembles and sends an MLD packet. +// +// Precondition: mld.ep.mu must be read locked. +func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldType header.ICMPv6Type) (bool, *tcpip.Error) { sentStats := mld.ep.protocol.stack.Stats().ICMP.V6.PacketsSent var mldStat *tcpip.StatCounter switch mldType { @@ -162,9 +177,61 @@ func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldTyp icmp := header.ICMPv6(buffer.NewView(header.ICMPv6HeaderSize + header.MLDMinimumSize)) icmp.SetType(mldType) header.MLD(icmp.MessageBody()).SetMulticastAddress(groupAddress) - // TODO(gvisor.dev/issue/4888): We should not use the unspecified address, - // rather we should select an appropriate local address. - localAddress := header.IPv6Any + // As per RFC 2710 section 3, + // + // All MLD messages described in this document are sent with a link-local + // IPv6 Source Address, an IPv6 Hop Limit of 1, and an IPv6 Router Alert + // option in a Hop-by-Hop Options header. + // + // However, this would cause problems with Duplicate Address Detection with + // the first address as MLD snooping switches may not send multicast traffic + // that DAD depends on to the node performing DAD without the MLD report, as + // documented in RFC 4816: + // + // Note that when a node joins a multicast address, it typically sends a + // Multicast Listener Discovery (MLD) report message [RFC2710] [RFC3810] + // for the multicast address. In the case of Duplicate Address + // Detection, the MLD report message is required in order to inform MLD- + // snooping switches, rather than routers, to forward multicast packets. + // In the above description, the delay for joining the multicast address + // thus means delaying transmission of the corresponding MLD report + // message. Since the MLD specifications do not request a random delay + // to avoid race conditions, just delaying Neighbor Solicitation would + // cause congestion by the MLD report messages. The congestion would + // then prevent the MLD-snooping switches from working correctly and, as + // a result, prevent Duplicate Address Detection from working. The + // requirement to include the delay for the MLD report in this case + // avoids this scenario. [RFC3590] also talks about some interaction + // issues between Duplicate Address Detection and MLD, and specifies + // which source address should be used for the MLD report in this case. + // + // As per RFC 3590 section 4, we should still send out MLD reports with an + // unspecified source address if we do not have an assigned link-local + // address to use as the source address to ensure DAD works as expected on + // networks with MLD snooping switches: + // + // MLD Report and Done messages are sent with a link-local address as + // the IPv6 source address, if a valid address is available on the + // interface. If a valid link-local address is not available (e.g., one + // has not been configured), the message is sent with the unspecified + // address (::) as the IPv6 source address. + // + // Once a valid link-local address is available, a node SHOULD generate + // new MLD Report messages for all multicast addresses joined on the + // interface. + // + // Routers receiving an MLD Report or Done message with the unspecified + // address as the IPv6 source address MUST silently discard the packet + // without taking any action on the packets contents. + // + // Snooping switches MUST manage multicast forwarding state based on MLD + // Report and Done messages sent with the unspecified address as the + // IPv6 source address. + localAddress := mld.ep.getLinkLocalAddressRLocked() + if len(localAddress) == 0 { + localAddress = header.IPv6Any + } + icmp.SetChecksum(header.ICMPv6Checksum(icmp, localAddress, destAddress, buffer.VectorisedView{})) pkt := stack.NewPacketBuffer(stack.PacketBufferOptions{ @@ -180,8 +247,8 @@ func (mld *mldState) writePacket(destAddress, groupAddress tcpip.Address, mldTyp // Membership Reports. if err := mld.ep.nic.WritePacketToRemote(header.EthernetAddressFromMulticastIPv6Address(destAddress), nil /* gso */, ProtocolNumber, pkt); err != nil { sentStats.Dropped.Increment() - return err + return false, err } mldStat.Increment() - return nil + return localAddress != header.IPv6Any, nil } diff --git a/pkg/tcpip/network/ipv6/ndp.go b/pkg/tcpip/network/ipv6/ndp.go index 2f5e2e82c..3b892aeda 100644 --- a/pkg/tcpip/network/ipv6/ndp.go +++ b/pkg/tcpip/network/ipv6/ndp.go @@ -647,6 +647,7 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressE ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, true, nil) } + ndp.ep.onAddressAssignedLocked(addr) return nil } @@ -690,12 +691,14 @@ func (ndp *ndpState) startDuplicateAddressDetection(addr tcpip.Address, addressE ndpDisp.OnDuplicateAddressDetectionStatus(ndp.ep.nic.ID(), addr, dadDone, err) } - // If DAD resolved for a stable SLAAC address, attempt generation of a - // temporary SLAAC address. - if dadDone && addressEndpoint.ConfigType() == stack.AddressConfigSlaac { - // Reset the generation attempts counter as we are starting the generation - // of a new address for the SLAAC prefix. - ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) + if dadDone { + if addressEndpoint.ConfigType() == stack.AddressConfigSlaac { + // Reset the generation attempts counter as we are starting the + // generation of a new address for the SLAAC prefix. + ndp.regenerateTempSLAACAddr(addressEndpoint.AddressWithPrefix().Subnet(), true /* resetGenAttempts */) + } + + ndp.ep.onAddressAssignedLocked(addr) } }), } diff --git a/pkg/tcpip/stack/addressable_endpoint_state.go b/pkg/tcpip/stack/addressable_endpoint_state.go index 6e4f5fa46..cd423bf71 100644 --- a/pkg/tcpip/stack/addressable_endpoint_state.go +++ b/pkg/tcpip/stack/addressable_endpoint_state.go @@ -82,12 +82,16 @@ func (a *AddressableEndpointState) ForEachEndpoint(f func(AddressEndpoint) bool) } // ForEachPrimaryEndpoint calls f for each primary address. -func (a *AddressableEndpointState) ForEachPrimaryEndpoint(f func(AddressEndpoint)) { +// +// Once f returns false, f will no longer be called. +func (a *AddressableEndpointState) ForEachPrimaryEndpoint(f func(AddressEndpoint) bool) { a.mu.RLock() defer a.mu.RUnlock() for _, ep := range a.mu.primary { - f(ep) + if !f(ep) { + return + } } } diff --git a/pkg/tcpip/stack/nic.go b/pkg/tcpip/stack/nic.go index 5887aa1ed..a6237dd5f 100644 --- a/pkg/tcpip/stack/nic.go +++ b/pkg/tcpip/stack/nic.go @@ -172,7 +172,7 @@ func (n *NIC) disable() { // // n MUST be locked. func (n *NIC) disableLocked() { - if !n.setEnabled(false) { + if !n.Enabled() { return } @@ -184,6 +184,10 @@ func (n *NIC) disableLocked() { for _, ep := range n.networkEndpoints { ep.Disable() } + + if !n.setEnabled(false) { + panic("should have only done work to disable the NIC if it was enabled") + } } // enable enables n. |