Index: head/sys/netinet/in_pcb.h =================================================================== --- head/sys/netinet/in_pcb.h (revision 339038) +++ head/sys/netinet/in_pcb.h (revision 339039) @@ -1,894 +1,896 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1990, 1993 * The Regents of the University of California. * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)in_pcb.h 8.1 (Berkeley) 6/10/93 * $FreeBSD$ */ #ifndef _NETINET_IN_PCB_H_ #define _NETINET_IN_PCB_H_ #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #include #include #include #endif #include #define in6pcb inpcb /* for KAME src sync over BSD*'s */ #define in6p_sp inp_sp /* for KAME src sync over BSD*'s */ /* * struct inpcb is the common protocol control block structure used in most * IP transport protocols. * * Pointers to local and foreign host table entries, local and foreign socket * numbers, and pointers up (to a socket structure) and down (to a * protocol-specific control block) are stored here. */ CK_LIST_HEAD(inpcbhead, inpcb); CK_LIST_HEAD(inpcbporthead, inpcbport); CK_LIST_HEAD(inpcblbgrouphead, inpcblbgroup); typedef uint64_t inp_gen_t; /* * PCB with AF_INET6 null bind'ed laddr can receive AF_INET input packet. * So, AF_INET6 null laddr is also used as AF_INET null laddr, by utilizing * the following structure. */ struct in_addr_4in6 { u_int32_t ia46_pad32[3]; struct in_addr ia46_addr4; }; union in_dependaddr { struct in_addr_4in6 id46_addr; struct in6_addr id6_addr; }; /* * NOTE: ipv6 addrs should be 64-bit aligned, per RFC 2553. in_conninfo has * some extra padding to accomplish this. * NOTE 2: tcp_syncache.c uses first 5 32-bit words, which identify fport, * lport, faddr to generate hash, so these fields shouldn't be moved. */ struct in_endpoints { u_int16_t ie_fport; /* foreign port */ u_int16_t ie_lport; /* local port */ /* protocol dependent part, local and foreign addr */ union in_dependaddr ie_dependfaddr; /* foreign host table entry */ union in_dependaddr ie_dependladdr; /* local host table entry */ #define ie_faddr ie_dependfaddr.id46_addr.ia46_addr4 #define ie_laddr ie_dependladdr.id46_addr.ia46_addr4 #define ie6_faddr ie_dependfaddr.id6_addr #define ie6_laddr ie_dependladdr.id6_addr u_int32_t ie6_zoneid; /* scope zone id */ }; /* * XXX The defines for inc_* are hacks and should be changed to direct * references. */ struct in_conninfo { u_int8_t inc_flags; u_int8_t inc_len; u_int16_t inc_fibnum; /* XXX was pad, 16 bits is plenty */ /* protocol dependent part */ struct in_endpoints inc_ie; }; /* * Flags for inc_flags. */ #define INC_ISIPV6 0x01 #define INC_IPV6MINMTU 0x02 #define inc_fport inc_ie.ie_fport #define inc_lport inc_ie.ie_lport #define inc_faddr inc_ie.ie_faddr #define inc_laddr inc_ie.ie_laddr #define inc6_faddr inc_ie.ie6_faddr #define inc6_laddr inc_ie.ie6_laddr #define inc6_zoneid inc_ie.ie6_zoneid #if defined(_KERNEL) || defined(_WANT_INPCB) /* * struct inpcb captures the network layer state for TCP, UDP, and raw IPv4 and * IPv6 sockets. In the case of TCP and UDP, further per-connection state is * hung off of inp_ppcb most of the time. Almost all fields of struct inpcb * are static after creation or protected by a per-inpcb rwlock, inp_lock. A * few fields are protected by multiple locks as indicated in the locking notes * below. For these fields, all of the listed locks must be write-locked for * any modifications. However, these fields can be safely read while any one of * the listed locks are read-locked. This model can permit greater concurrency * for read operations. For example, connections can be looked up while only * holding a read lock on the global pcblist lock. This is important for * performance when attempting to find the connection for a packet given its IP * and port tuple. * * One noteworthy exception is that the global pcbinfo lock follows a different * set of rules in relation to the inp_list field. Rather than being * write-locked for modifications and read-locked for list iterations, it must * be read-locked during modifications and write-locked during list iterations. * This ensures that the relatively rare global list iterations safely walk a * stable snapshot of connections while allowing more common list modifications * to safely grab the pcblist lock just while adding or removing a connection * from the global list. * * Key: * (b) - Protected by the hpts lock. * (c) - Constant after initialization * (e) - Protected by the net_epoch_prempt epoch * (g) - Protected by the pcbgroup lock * (i) - Protected by the inpcb lock * (p) - Protected by the pcbinfo lock for the inpcb * (l) - Protected by the pcblist lock for the inpcb * (h) - Protected by the pcbhash lock for the inpcb * (s) - Protected by another subsystem's locks * (x) - Undefined locking * * Notes on the tcp_hpts: * * First Hpts lock order is * 1) INP_WLOCK() * 2) HPTS_LOCK() i.e. hpts->pmtx * * To insert a TCB on the hpts you *must* be holding the INP_WLOCK(). * You may check the inp->inp_in_hpts flag without the hpts lock. * The hpts is the only one that will clear this flag holding * only the hpts lock. This means that in your tcp_output() * routine when you test for the inp_in_hpts flag to be 1 * it may be transitioning to 0 (by the hpts). * That's ok since that will just mean an extra call to tcp_output * that most likely will find the call you executed * (when the mis-match occured) will have put the TCB back * on the hpts and it will return. If your * call did not add the inp back to the hpts then you will either * over-send or the cwnd will block you from sending more. * * Note you should also be holding the INP_WLOCK() when you * call the remove from the hpts as well. Though usually * you are either doing this from a timer, where you need and have * the INP_WLOCK() or from destroying your TCB where again * you should already have the INP_WLOCK(). * * The inp_hpts_cpu, inp_hpts_cpu_set, inp_input_cpu and * inp_input_cpu_set fields are controlled completely by * the hpts. Do not ever set these. The inp_hpts_cpu_set * and inp_input_cpu_set fields indicate if the hpts has * setup the respective cpu field. It is advised if this * field is 0, to enqueue the packet with the appropriate * hpts_immediate() call. If the _set field is 1, then * you may compare the inp_*_cpu field to the curcpu and * may want to again insert onto the hpts if these fields * are not equal (i.e. you are not on the expected CPU). * * A note on inp_hpts_calls and inp_input_calls, these * flags are set when the hpts calls either the output * or do_segment routines respectively. If the routine * being called wants to use this, then it needs to * clear the flag before returning. The hpts will not * clear the flag. The flags can be used to tell if * the hpts is the function calling the respective * routine. * * A few other notes: * * When a read lock is held, stability of the field is guaranteed; to write * to a field, a write lock must generally be held. * * netinet/netinet6-layer code should not assume that the inp_socket pointer * is safe to dereference without inp_lock being held, even for protocols * other than TCP (where the inpcb persists during TIMEWAIT even after the * socket has been freed), or there may be close(2)-related races. * * The inp_vflag field is overloaded, and would otherwise ideally be (c). * * TODO: Currently only the TCP stack is leveraging the global pcbinfo lock * read-lock usage during modification, this model can be applied to other * protocols (especially SCTP). */ struct icmp6_filter; struct inpcbpolicy; struct m_snd_tag; struct inpcb { /* Cache line #1 (amd64) */ CK_LIST_ENTRY(inpcb) inp_hash; /* [w](h/i) [r](e/i) hash list */ CK_LIST_ENTRY(inpcb) inp_pcbgrouphash; /* (g/i) hash list */ struct rwlock inp_lock; /* Cache line #2 (amd64) */ #define inp_start_zero inp_hpts #define inp_zero_size (sizeof(struct inpcb) - \ offsetof(struct inpcb, inp_start_zero)) TAILQ_ENTRY(inpcb) inp_hpts; /* pacing out queue next lock(b) */ uint32_t inp_hpts_request; /* Current hpts request, zero if * fits in the pacing window (i&b). */ /* * Note the next fields are protected by a * different lock (hpts-lock). This means that * they must correspond in size to the smallest * protectable bit field (uint8_t on x86, and * other platfomrs potentially uint32_t?). Also * since CPU switches can occur at different times the two * fields can *not* be collapsed into a signal bit field. */ #if defined(__amd64__) || defined(__i386__) volatile uint8_t inp_in_hpts; /* on output hpts (lock b) */ volatile uint8_t inp_in_input; /* on input hpts (lock b) */ #else volatile uint32_t inp_in_hpts; /* on output hpts (lock b) */ volatile uint32_t inp_in_input; /* on input hpts (lock b) */ #endif volatile uint16_t inp_hpts_cpu; /* Lock (i) */ u_int inp_refcount; /* (i) refcount */ int inp_flags; /* (i) generic IP/datagram flags */ int inp_flags2; /* (i) generic IP/datagram flags #2*/ volatile uint16_t inp_input_cpu; /* Lock (i) */ volatile uint8_t inp_hpts_cpu_set :1, /* on output hpts (i) */ inp_input_cpu_set : 1, /* on input hpts (i) */ inp_hpts_calls :1, /* (i) from output hpts */ inp_input_calls :1, /* (i) from input hpts */ inp_spare_bits2 : 4; uint8_t inp_spare_byte; /* Compiler hole */ void *inp_ppcb; /* (i) pointer to per-protocol pcb */ struct socket *inp_socket; /* (i) back pointer to socket */ uint32_t inp_hptsslot; /* Hpts wheel slot this tcb is Lock(i&b) */ uint32_t inp_hpts_drop_reas; /* reason we are dropping the PCB (lock i&b) */ TAILQ_ENTRY(inpcb) inp_input; /* pacing in queue next lock(b) */ struct inpcbinfo *inp_pcbinfo; /* (c) PCB list info */ struct inpcbgroup *inp_pcbgroup; /* (g/i) PCB group list */ CK_LIST_ENTRY(inpcb) inp_pcbgroup_wild; /* (g/i/h) group wildcard entry */ struct ucred *inp_cred; /* (c) cache of socket cred */ u_int32_t inp_flow; /* (i) IPv6 flow information */ u_char inp_vflag; /* (i) IP version flag (v4/v6) */ u_char inp_ip_ttl; /* (i) time to live proto */ u_char inp_ip_p; /* (c) protocol proto */ u_char inp_ip_minttl; /* (i) minimum TTL or drop */ uint32_t inp_flowid; /* (x) flow id / queue id */ struct m_snd_tag *inp_snd_tag; /* (i) send tag for outgoing mbufs */ uint32_t inp_flowtype; /* (x) M_HASHTYPE value */ uint32_t inp_rss_listen_bucket; /* (x) overridden RSS listen bucket */ /* Local and foreign ports, local and foreign addr. */ struct in_conninfo inp_inc; /* (i) list for PCB's local port */ /* MAC and IPSEC policy information. */ struct label *inp_label; /* (i) MAC label */ struct inpcbpolicy *inp_sp; /* (s) for IPSEC */ /* Protocol-dependent part; options. */ struct { u_char inp_ip_tos; /* (i) type of service proto */ struct mbuf *inp_options; /* (i) IP options */ struct ip_moptions *inp_moptions; /* (i) mcast options */ }; struct { /* (i) IP options */ struct mbuf *in6p_options; /* (i) IP6 options for outgoing packets */ struct ip6_pktopts *in6p_outputopts; /* (i) IP multicast options */ struct ip6_moptions *in6p_moptions; /* (i) ICMPv6 code type filter */ struct icmp6_filter *in6p_icmp6filt; /* (i) IPV6_CHECKSUM setsockopt */ int in6p_cksum; short in6p_hops; }; CK_LIST_ENTRY(inpcb) inp_portlist; /* (i/h) */ struct inpcbport *inp_phd; /* (i/h) head of this list */ inp_gen_t inp_gencnt; /* (c) generation count */ void *spare_ptr; /* Spare pointer. */ rt_gen_t inp_rt_cookie; /* generation for route entry */ union { /* cached L3 information */ struct route inp_route; struct route_in6 inp_route6; }; CK_LIST_ENTRY(inpcb) inp_list; /* (p/l) list for all PCBs for proto */ /* (e[r]) for list iteration */ /* (p[w]/l) for addition/removal */ struct epoch_context inp_epoch_ctx; }; #endif /* _KERNEL */ #define inp_fport inp_inc.inc_fport #define inp_lport inp_inc.inc_lport #define inp_faddr inp_inc.inc_faddr #define inp_laddr inp_inc.inc_laddr #define in6p_faddr inp_inc.inc6_faddr #define in6p_laddr inp_inc.inc6_laddr #define in6p_zoneid inp_inc.inc6_zoneid #define in6p_flowinfo inp_flow #define inp_vnet inp_pcbinfo->ipi_vnet /* * The range of the generation count, as used in this implementation, is 9e19. * We would have to create 300 billion connections per second for this number * to roll over in a year. This seems sufficiently unlikely that we simply * don't concern ourselves with that possibility. */ /* * Interface exported to userland by various protocols which use inpcbs. Hack * alert -- only define if struct xsocket is in scope. * Fields prefixed with "xi_" are unique to this structure, and the rest * match fields in the struct inpcb, to ease coding and porting. * * Legend: * (s) - used by userland utilities in src * (p) - used by utilities in ports * (3) - is known to be used by third party software not in ports * (n) - no known usage */ #ifdef _SYS_SOCKETVAR_H_ struct xinpcb { ksize_t xi_len; /* length of this structure */ struct xsocket xi_socket; /* (s,p) */ struct in_conninfo inp_inc; /* (s,p) */ uint64_t inp_gencnt; /* (s,p) */ kvaddr_t inp_ppcb; /* (s) netstat(1) */ int64_t inp_spare64[4]; uint32_t inp_flow; /* (s) */ uint32_t inp_flowid; /* (s) */ uint32_t inp_flowtype; /* (s) */ int32_t inp_flags; /* (s,p) */ int32_t inp_flags2; /* (s) */ int32_t inp_rss_listen_bucket; /* (n) */ int32_t in6p_cksum; /* (n) */ int32_t inp_spare32[4]; uint16_t in6p_hops; /* (n) */ uint8_t inp_ip_tos; /* (n) */ int8_t pad8; uint8_t inp_vflag; /* (s,p) */ uint8_t inp_ip_ttl; /* (n) */ uint8_t inp_ip_p; /* (n) */ uint8_t inp_ip_minttl; /* (n) */ int8_t inp_spare8[4]; } __aligned(8); struct xinpgen { ksize_t xig_len; /* length of this structure */ u_int xig_count; /* number of PCBs at this time */ uint32_t _xig_spare32; inp_gen_t xig_gen; /* generation count at this time */ so_gen_t xig_sogen; /* socket generation count this time */ uint64_t _xig_spare64[4]; } __aligned(8); #ifdef _KERNEL void in_pcbtoxinpcb(const struct inpcb *, struct xinpcb *); #endif #endif /* _SYS_SOCKETVAR_H_ */ struct inpcbport { struct epoch_context phd_epoch_ctx; CK_LIST_ENTRY(inpcbport) phd_hash; struct inpcbhead phd_pcblist; u_short phd_port; }; struct in_pcblist { int il_count; struct epoch_context il_epoch_ctx; struct inpcbinfo *il_pcbinfo; struct inpcb *il_inp_list[0]; }; /*- * Global data structure for each high-level protocol (UDP, TCP, ...) in both * IPv4 and IPv6. Holds inpcb lists and information for managing them. * * Each pcbinfo is protected by three locks: ipi_lock, ipi_hash_lock and * ipi_list_lock: * - ipi_lock covering the global pcb list stability during loop iteration, * - ipi_hash_lock covering the hashed lookup tables, * - ipi_list_lock covering mutable global fields (such as the global * pcb list) * * The lock order is: * * ipi_lock (before) * inpcb locks (before) * ipi_list locks (before) * {ipi_hash_lock, pcbgroup locks} * * Locking key: * * (c) Constant or nearly constant after initialisation * (e) - Protected by the net_epoch_prempt epoch * (g) Locked by ipi_lock * (l) Locked by ipi_list_lock * (h) Read using either net_epoch_preempt or inpcb lock; write requires both ipi_hash_lock and inpcb lock * (p) Protected by one or more pcbgroup locks * (x) Synchronisation properties poorly defined */ struct inpcbinfo { /* * Global lock protecting inpcb list modification */ struct mtx ipi_lock; /* * Global list of inpcbs on the protocol. */ struct inpcbhead *ipi_listhead; /* [r](e) [w](g/l) */ u_int ipi_count; /* (l) */ /* * Generation count -- incremented each time a connection is allocated * or freed. */ u_quad_t ipi_gencnt; /* (l) */ /* * Fields associated with port lookup and allocation. */ u_short ipi_lastport; /* (x) */ u_short ipi_lastlow; /* (x) */ u_short ipi_lasthi; /* (x) */ /* * UMA zone from which inpcbs are allocated for this protocol. */ struct uma_zone *ipi_zone; /* (c) */ /* * Connection groups associated with this protocol. These fields are * constant, but pcbgroup structures themselves are protected by * per-pcbgroup locks. */ struct inpcbgroup *ipi_pcbgroups; /* (c) */ u_int ipi_npcbgroups; /* (c) */ u_int ipi_hashfields; /* (c) */ /* * Global lock protecting modification non-pcbgroup hash lookup tables. */ struct mtx ipi_hash_lock; /* * Global hash of inpcbs, hashed by local and foreign addresses and * port numbers. */ struct inpcbhead *ipi_hashbase; /* (h) */ u_long ipi_hashmask; /* (h) */ /* * Global hash of inpcbs, hashed by only local port number. */ struct inpcbporthead *ipi_porthashbase; /* (h) */ u_long ipi_porthashmask; /* (h) */ /* * List of wildcard inpcbs for use with pcbgroups. In the past, was * per-pcbgroup but is now global. All pcbgroup locks must be held * to modify the list, so any is sufficient to read it. */ struct inpcbhead *ipi_wildbase; /* (p) */ u_long ipi_wildmask; /* (p) */ /* * Load balance groups used for the SO_REUSEPORT_LB option, * hashed by local port. */ struct inpcblbgrouphead *ipi_lbgrouphashbase; /* (h) */ u_long ipi_lbgrouphashmask; /* (h) */ /* * Pointer to network stack instance */ struct vnet *ipi_vnet; /* (c) */ /* * general use 2 */ void *ipi_pspare[2]; /* * Global lock protecting global inpcb list, inpcb count, etc. */ struct rwlock ipi_list_lock; }; #ifdef _KERNEL /* * Connection groups hold sets of connections that have similar CPU/thread * affinity. Each connection belongs to exactly one connection group. */ struct inpcbgroup { /* * Per-connection group hash of inpcbs, hashed by local and foreign * addresses and port numbers. */ struct inpcbhead *ipg_hashbase; /* (c) */ u_long ipg_hashmask; /* (c) */ /* * Notional affinity of this pcbgroup. */ u_int ipg_cpu; /* (p) */ /* * Per-connection group lock, not to be confused with ipi_lock. * Protects the hash table hung off the group, but also the global * wildcard list in inpcbinfo. */ struct mtx ipg_lock; } __aligned(CACHE_LINE_SIZE); /* * Load balance groups used for the SO_REUSEPORT_LB socket option. Each group * (or unique address:port combination) can be re-used at most * INPCBLBGROUP_SIZMAX (256) times. The inpcbs are stored in il_inp which * is dynamically resized as processes bind/unbind to that specific group. */ struct inpcblbgroup { CK_LIST_ENTRY(inpcblbgroup) il_list; struct epoch_context il_epoch_ctx; uint16_t il_lport; /* (c) */ u_char il_vflag; /* (c) */ u_char il_pad; uint32_t il_pad2; union in_dependaddr il_dependladdr; /* (c) */ #define il_laddr il_dependladdr.id46_addr.ia46_addr4 #define il6_laddr il_dependladdr.id6_addr uint32_t il_inpsiz; /* max count in il_inp[] (h) */ uint32_t il_inpcnt; /* cur count in il_inp[] (h) */ struct inpcb *il_inp[]; /* (h) */ }; #define INP_LOCK_INIT(inp, d, t) \ rw_init_flags(&(inp)->inp_lock, (t), RW_RECURSE | RW_DUPOK) #define INP_LOCK_DESTROY(inp) rw_destroy(&(inp)->inp_lock) #define INP_RLOCK(inp) rw_rlock(&(inp)->inp_lock) #define INP_WLOCK(inp) rw_wlock(&(inp)->inp_lock) #define INP_TRY_RLOCK(inp) rw_try_rlock(&(inp)->inp_lock) #define INP_TRY_WLOCK(inp) rw_try_wlock(&(inp)->inp_lock) #define INP_RUNLOCK(inp) rw_runlock(&(inp)->inp_lock) #define INP_WUNLOCK(inp) rw_wunlock(&(inp)->inp_lock) #define INP_TRY_UPGRADE(inp) rw_try_upgrade(&(inp)->inp_lock) #define INP_DOWNGRADE(inp) rw_downgrade(&(inp)->inp_lock) #define INP_WLOCKED(inp) rw_wowned(&(inp)->inp_lock) #define INP_LOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_LOCKED) #define INP_RLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_RLOCKED) #define INP_WLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_WLOCKED) #define INP_UNLOCK_ASSERT(inp) rw_assert(&(inp)->inp_lock, RA_UNLOCKED) /* * These locking functions are for inpcb consumers outside of sys/netinet, * more specifically, they were added for the benefit of TOE drivers. The * macros are reserved for use by the stack. */ void inp_wlock(struct inpcb *); void inp_wunlock(struct inpcb *); void inp_rlock(struct inpcb *); void inp_runlock(struct inpcb *); #ifdef INVARIANT_SUPPORT void inp_lock_assert(struct inpcb *); void inp_unlock_assert(struct inpcb *); #else #define inp_lock_assert(inp) do {} while (0) #define inp_unlock_assert(inp) do {} while (0) #endif void inp_apply_all(void (*func)(struct inpcb *, void *), void *arg); int inp_ip_tos_get(const struct inpcb *inp); void inp_ip_tos_set(struct inpcb *inp, int val); struct socket * inp_inpcbtosocket(struct inpcb *inp); struct tcpcb * inp_inpcbtotcpcb(struct inpcb *inp); void inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp, uint32_t *faddr, uint16_t *fp); int inp_so_options(const struct inpcb *inp); #endif /* _KERNEL */ #define INP_INFO_LOCK_INIT(ipi, d) \ mtx_init(&(ipi)->ipi_lock, (d), NULL, MTX_DEF| MTX_RECURSE) #define INP_INFO_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_lock) #define INP_INFO_RLOCK_ET(ipi, et) NET_EPOCH_ENTER_ET((et)) #define INP_INFO_WLOCK(ipi) mtx_lock(&(ipi)->ipi_lock) #define INP_INFO_TRY_WLOCK(ipi) mtx_trylock(&(ipi)->ipi_lock) #define INP_INFO_WLOCKED(ipi) mtx_owned(&(ipi)->ipi_lock) #define INP_INFO_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT_ET((et)) #define INP_INFO_RUNLOCK_TP(ipi, tp) NET_EPOCH_EXIT_ET(*(tp)->t_inpcb->inp_et) #define INP_INFO_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_lock) #define INP_INFO_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_lock)) #define INP_INFO_RLOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt)) #define INP_INFO_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_lock, MA_OWNED) +#define INP_INFO_WUNLOCK_ASSERT(ipi) \ + mtx_assert(&(ipi)->ipi_lock, MA_NOTOWNED) #define INP_INFO_UNLOCK_ASSERT(ipi) MPASS(!in_epoch(net_epoch_preempt) && !mtx_owned(&(ipi)->ipi_lock)) #define INP_LIST_LOCK_INIT(ipi, d) \ rw_init_flags(&(ipi)->ipi_list_lock, (d), 0) #define INP_LIST_LOCK_DESTROY(ipi) rw_destroy(&(ipi)->ipi_list_lock) #define INP_LIST_RLOCK(ipi) rw_rlock(&(ipi)->ipi_list_lock) #define INP_LIST_WLOCK(ipi) rw_wlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_RLOCK(ipi) rw_try_rlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_WLOCK(ipi) rw_try_wlock(&(ipi)->ipi_list_lock) #define INP_LIST_TRY_UPGRADE(ipi) rw_try_upgrade(&(ipi)->ipi_list_lock) #define INP_LIST_RUNLOCK(ipi) rw_runlock(&(ipi)->ipi_list_lock) #define INP_LIST_WUNLOCK(ipi) rw_wunlock(&(ipi)->ipi_list_lock) #define INP_LIST_LOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_LOCKED) #define INP_LIST_RLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_RLOCKED) #define INP_LIST_WLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_WLOCKED) #define INP_LIST_UNLOCK_ASSERT(ipi) \ rw_assert(&(ipi)->ipi_list_lock, RA_UNLOCKED) #define INP_HASH_LOCK_INIT(ipi, d) mtx_init(&(ipi)->ipi_hash_lock, (d), NULL, MTX_DEF) #define INP_HASH_LOCK_DESTROY(ipi) mtx_destroy(&(ipi)->ipi_hash_lock) #define INP_HASH_RLOCK(ipi) struct epoch_tracker inp_hash_et; epoch_enter_preempt(net_epoch_preempt, &inp_hash_et) #define INP_HASH_RLOCK_ET(ipi, et) epoch_enter_preempt(net_epoch_preempt, &(et)) #define INP_HASH_WLOCK(ipi) mtx_lock(&(ipi)->ipi_hash_lock) #define INP_HASH_RUNLOCK(ipi) NET_EPOCH_EXIT_ET(inp_hash_et) #define INP_HASH_RUNLOCK_ET(ipi, et) NET_EPOCH_EXIT_ET((et)) #define INP_HASH_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_hash_lock) #define INP_HASH_LOCK_ASSERT(ipi) MPASS(in_epoch(net_epoch_preempt) || mtx_owned(&(ipi)->ipi_hash_lock)) #define INP_HASH_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_hash_lock, MA_OWNED); #define INP_GROUP_LOCK_INIT(ipg, d) mtx_init(&(ipg)->ipg_lock, (d), NULL, \ MTX_DEF | MTX_DUPOK) #define INP_GROUP_LOCK_DESTROY(ipg) mtx_destroy(&(ipg)->ipg_lock) #define INP_GROUP_LOCK(ipg) mtx_lock(&(ipg)->ipg_lock) #define INP_GROUP_LOCK_ASSERT(ipg) mtx_assert(&(ipg)->ipg_lock, MA_OWNED) #define INP_GROUP_UNLOCK(ipg) mtx_unlock(&(ipg)->ipg_lock) #define INP_PCBHASH(faddr, lport, fport, mask) \ (((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) & (mask)) #define INP_PCBPORTHASH(lport, mask) \ (ntohs((lport)) & (mask)) #define INP_PCBLBGROUP_PORTHASH(lport, mask) \ (ntohs((lport)) & (mask)) #define INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) \ ((faddr) ^ ((faddr) >> 16) ^ ntohs((lport) ^ (fport))) #define INP6_PCBHASHKEY(faddr) ((faddr)->s6_addr32[3]) /* * Flags for inp_vflags -- historically version flags only */ #define INP_IPV4 0x1 #define INP_IPV6 0x2 #define INP_IPV6PROTO 0x4 /* opened under IPv6 protocol */ /* * Flags for inp_flags. */ #define INP_RECVOPTS 0x00000001 /* receive incoming IP options */ #define INP_RECVRETOPTS 0x00000002 /* receive IP options for reply */ #define INP_RECVDSTADDR 0x00000004 /* receive IP dst address */ #define INP_HDRINCL 0x00000008 /* user supplies entire IP header */ #define INP_HIGHPORT 0x00000010 /* user wants "high" port binding */ #define INP_LOWPORT 0x00000020 /* user wants "low" port binding */ #define INP_ANONPORT 0x00000040 /* port chosen for user */ #define INP_RECVIF 0x00000080 /* receive incoming interface */ #define INP_MTUDISC 0x00000100 /* user can do MTU discovery */ /* 0x000200 unused: was INP_FAITH */ #define INP_RECVTTL 0x00000400 /* receive incoming IP TTL */ #define INP_DONTFRAG 0x00000800 /* don't fragment packet */ #define INP_BINDANY 0x00001000 /* allow bind to any address */ #define INP_INHASHLIST 0x00002000 /* in_pcbinshash() has been called */ #define INP_RECVTOS 0x00004000 /* receive incoming IP TOS */ #define IN6P_IPV6_V6ONLY 0x00008000 /* restrict AF_INET6 socket for v6 */ #define IN6P_PKTINFO 0x00010000 /* receive IP6 dst and I/F */ #define IN6P_HOPLIMIT 0x00020000 /* receive hoplimit */ #define IN6P_HOPOPTS 0x00040000 /* receive hop-by-hop options */ #define IN6P_DSTOPTS 0x00080000 /* receive dst options after rthdr */ #define IN6P_RTHDR 0x00100000 /* receive routing header */ #define IN6P_RTHDRDSTOPTS 0x00200000 /* receive dstoptions before rthdr */ #define IN6P_TCLASS 0x00400000 /* receive traffic class value */ #define IN6P_AUTOFLOWLABEL 0x00800000 /* attach flowlabel automatically */ #define INP_TIMEWAIT 0x01000000 /* in TIMEWAIT, ppcb is tcptw */ #define INP_ONESBCAST 0x02000000 /* send all-ones broadcast */ #define INP_DROPPED 0x04000000 /* protocol drop flag */ #define INP_SOCKREF 0x08000000 /* strong socket reference */ #define INP_RESERVED_0 0x10000000 /* reserved field */ #define INP_RESERVED_1 0x20000000 /* reserved field */ #define IN6P_RFC2292 0x40000000 /* used RFC2292 API on the socket */ #define IN6P_MTU 0x80000000 /* receive path MTU */ #define INP_CONTROLOPTS (INP_RECVOPTS|INP_RECVRETOPTS|INP_RECVDSTADDR|\ INP_RECVIF|INP_RECVTTL|INP_RECVTOS|\ IN6P_PKTINFO|IN6P_HOPLIMIT|IN6P_HOPOPTS|\ IN6P_DSTOPTS|IN6P_RTHDR|IN6P_RTHDRDSTOPTS|\ IN6P_TCLASS|IN6P_AUTOFLOWLABEL|IN6P_RFC2292|\ IN6P_MTU) /* * Flags for inp_flags2. */ #define INP_2UNUSED1 0x00000001 #define INP_2UNUSED2 0x00000002 #define INP_PCBGROUPWILD 0x00000004 /* in pcbgroup wildcard list */ #define INP_REUSEPORT 0x00000008 /* SO_REUSEPORT option is set */ #define INP_FREED 0x00000010 /* inp itself is not valid */ #define INP_REUSEADDR 0x00000020 /* SO_REUSEADDR option is set */ #define INP_BINDMULTI 0x00000040 /* IP_BINDMULTI option is set */ #define INP_RSS_BUCKET_SET 0x00000080 /* IP_RSS_LISTEN_BUCKET is set */ #define INP_RECVFLOWID 0x00000100 /* populate recv datagram with flow info */ #define INP_RECVRSSBUCKETID 0x00000200 /* populate recv datagram with bucket id */ #define INP_RATE_LIMIT_CHANGED 0x00000400 /* rate limit needs attention */ #define INP_ORIGDSTADDR 0x00000800 /* receive IP dst address/port */ #define INP_CANNOT_DO_ECN 0x00001000 /* The stack does not do ECN */ #define INP_REUSEPORT_LB 0x00002000 /* SO_REUSEPORT_LB option is set */ /* * Flags passed to in_pcblookup*() functions. */ #define INPLOOKUP_WILDCARD 0x00000001 /* Allow wildcard sockets. */ #define INPLOOKUP_RLOCKPCB 0x00000002 /* Return inpcb read-locked. */ #define INPLOOKUP_WLOCKPCB 0x00000004 /* Return inpcb write-locked. */ #define INPLOOKUP_MASK (INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB | \ INPLOOKUP_WLOCKPCB) #define sotoinpcb(so) ((struct inpcb *)(so)->so_pcb) #define sotoin6pcb(so) sotoinpcb(so) /* for KAME src sync over BSD*'s */ #define INP_SOCKAF(so) so->so_proto->pr_domain->dom_family #define INP_CHECK_SOCKAF(so, af) (INP_SOCKAF(so) == af) /* * Constants for pcbinfo.ipi_hashfields. */ #define IPI_HASHFIELDS_NONE 0 #define IPI_HASHFIELDS_2TUPLE 1 #define IPI_HASHFIELDS_4TUPLE 2 #ifdef _KERNEL VNET_DECLARE(int, ipport_reservedhigh); VNET_DECLARE(int, ipport_reservedlow); VNET_DECLARE(int, ipport_lowfirstauto); VNET_DECLARE(int, ipport_lowlastauto); VNET_DECLARE(int, ipport_firstauto); VNET_DECLARE(int, ipport_lastauto); VNET_DECLARE(int, ipport_hifirstauto); VNET_DECLARE(int, ipport_hilastauto); VNET_DECLARE(int, ipport_randomized); VNET_DECLARE(int, ipport_randomcps); VNET_DECLARE(int, ipport_randomtime); VNET_DECLARE(int, ipport_stoprandom); VNET_DECLARE(int, ipport_tcpallocs); #define V_ipport_reservedhigh VNET(ipport_reservedhigh) #define V_ipport_reservedlow VNET(ipport_reservedlow) #define V_ipport_lowfirstauto VNET(ipport_lowfirstauto) #define V_ipport_lowlastauto VNET(ipport_lowlastauto) #define V_ipport_firstauto VNET(ipport_firstauto) #define V_ipport_lastauto VNET(ipport_lastauto) #define V_ipport_hifirstauto VNET(ipport_hifirstauto) #define V_ipport_hilastauto VNET(ipport_hilastauto) #define V_ipport_randomized VNET(ipport_randomized) #define V_ipport_randomcps VNET(ipport_randomcps) #define V_ipport_randomtime VNET(ipport_randomtime) #define V_ipport_stoprandom VNET(ipport_stoprandom) #define V_ipport_tcpallocs VNET(ipport_tcpallocs) void in_pcbinfo_destroy(struct inpcbinfo *); void in_pcbinfo_init(struct inpcbinfo *, const char *, struct inpcbhead *, int, int, char *, uma_init, u_int); int in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi); struct inpcbgroup * in_pcbgroup_byhash(struct inpcbinfo *, u_int, uint32_t); struct inpcbgroup * in_pcbgroup_byinpcb(struct inpcb *); struct inpcbgroup * in_pcbgroup_bytuple(struct inpcbinfo *, struct in_addr, u_short, struct in_addr, u_short); void in_pcbgroup_destroy(struct inpcbinfo *); int in_pcbgroup_enabled(struct inpcbinfo *); void in_pcbgroup_init(struct inpcbinfo *, u_int, int); void in_pcbgroup_remove(struct inpcb *); void in_pcbgroup_update(struct inpcb *); void in_pcbgroup_update_mbuf(struct inpcb *, struct mbuf *); void in_pcbpurgeif0(struct inpcbinfo *, struct ifnet *); int in_pcballoc(struct socket *, struct inpcbinfo *); int in_pcbbind(struct inpcb *, struct sockaddr *, struct ucred *); int in_pcb_lport(struct inpcb *, struct in_addr *, u_short *, struct ucred *, int); int in_pcbbind_setup(struct inpcb *, struct sockaddr *, in_addr_t *, u_short *, struct ucred *); int in_pcbconnect(struct inpcb *, struct sockaddr *, struct ucred *); int in_pcbconnect_mbuf(struct inpcb *, struct sockaddr *, struct ucred *, struct mbuf *); int in_pcbconnect_setup(struct inpcb *, struct sockaddr *, in_addr_t *, u_short *, in_addr_t *, u_short *, struct inpcb **, struct ucred *); void in_pcbdetach(struct inpcb *); void in_pcbdisconnect(struct inpcb *); void in_pcbdrop(struct inpcb *); void in_pcbfree(struct inpcb *); int in_pcbinshash(struct inpcb *); int in_pcbinshash_nopcbgroup(struct inpcb *); int in_pcbladdr(struct inpcb *, struct in_addr *, struct in_addr *, struct ucred *); struct inpcb * in_pcblookup_local(struct inpcbinfo *, struct in_addr, u_short, int, struct ucred *); struct inpcb * in_pcblookup(struct inpcbinfo *, struct in_addr, u_int, struct in_addr, u_int, int, struct ifnet *); struct inpcb * in_pcblookup_mbuf(struct inpcbinfo *, struct in_addr, u_int, struct in_addr, u_int, int, struct ifnet *, struct mbuf *); void in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr, int, struct inpcb *(*)(struct inpcb *, int)); void in_pcbref(struct inpcb *); void in_pcbrehash(struct inpcb *); void in_pcbrehash_mbuf(struct inpcb *, struct mbuf *); int in_pcbrele(struct inpcb *); int in_pcbrele_rlocked(struct inpcb *); int in_pcbrele_wlocked(struct inpcb *); void in_pcblist_rele_rlocked(epoch_context_t ctx); void in_losing(struct inpcb *); void in_pcbsetsolabel(struct socket *so); int in_getpeeraddr(struct socket *so, struct sockaddr **nam); int in_getsockaddr(struct socket *so, struct sockaddr **nam); struct sockaddr * in_sockaddr(in_port_t port, struct in_addr *addr); void in_pcbsosetlabel(struct socket *so); #ifdef RATELIMIT int in_pcbattach_txrtlmt(struct inpcb *, struct ifnet *, uint32_t, uint32_t, uint32_t); void in_pcbdetach_txrtlmt(struct inpcb *); int in_pcbmodify_txrtlmt(struct inpcb *, uint32_t); int in_pcbquery_txrtlmt(struct inpcb *, uint32_t *); int in_pcbquery_txrlevel(struct inpcb *, uint32_t *); void in_pcboutput_txrtlmt(struct inpcb *, struct ifnet *, struct mbuf *); void in_pcboutput_eagain(struct inpcb *); #endif #endif /* _KERNEL */ #endif /* !_NETINET_IN_PCB_H_ */ Index: head/sys/netinet/siftr.c =================================================================== --- head/sys/netinet/siftr.c (revision 339038) +++ head/sys/netinet/siftr.c (revision 339039) @@ -1,1568 +1,1568 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2007-2009 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010, The FreeBSD Foundation * All rights reserved. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by Lawrence Stewart under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /****************************************************** * Statistical Information For TCP Research (SIFTR) * * A FreeBSD kernel module that adds very basic intrumentation to the * TCP stack, allowing internal stats to be recorded to a log file * for experimental, debugging and performance analysis purposes. * * SIFTR was first released in 2007 by James Healy and Lawrence Stewart whilst * working on the NewTCP research project at Swinburne University of * Technology's Centre for Advanced Internet Architectures, Melbourne, * Australia, which was made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * More details are available at: * http://caia.swin.edu.au/urp/newtcp/ * * Work on SIFTR v1.2.x was sponsored by the FreeBSD Foundation as part of * the "Enhancing the FreeBSD TCP Implementation" project 2008-2009. * More details are available at: * http://www.freebsdfoundation.org/ * http://caia.swin.edu.au/freebsd/etcp09/ * * Lawrence Stewart is the current maintainer, and all contact regarding * SIFTR should be directed to him via email: lastewart@swin.edu.au * * Initial release date: June 2007 * Most recent update: September 2010 ******************************************************/ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SIFTR_IPV6 #include #include #endif /* SIFTR_IPV6 */ #include /* * Three digit version number refers to X.Y.Z where: * X is the major version number * Y is bumped to mark backwards incompatible changes * Z is bumped to mark backwards compatible changes */ #define V_MAJOR 1 #define V_BACKBREAK 2 #define V_BACKCOMPAT 4 #define MODVERSION __CONCAT(V_MAJOR, __CONCAT(V_BACKBREAK, V_BACKCOMPAT)) #define MODVERSION_STR __XSTRING(V_MAJOR) "." __XSTRING(V_BACKBREAK) "." \ __XSTRING(V_BACKCOMPAT) #define HOOK 0 #define UNHOOK 1 #define SIFTR_EXPECTED_MAX_TCP_FLOWS 65536 #define SYS_NAME "FreeBSD" #define PACKET_TAG_SIFTR 100 #define PACKET_COOKIE_SIFTR 21749576 #define SIFTR_LOG_FILE_MODE 0644 #define SIFTR_DISABLE 0 #define SIFTR_ENABLE 1 /* * Hard upper limit on the length of log messages. Bump this up if you add new * data fields such that the line length could exceed the below value. */ #define MAX_LOG_MSG_LEN 200 /* XXX: Make this a sysctl tunable. */ #define SIFTR_ALQ_BUFLEN (1000*MAX_LOG_MSG_LEN) /* * 1 byte for IP version * IPv4: src/dst IP (4+4) + src/dst port (2+2) = 12 bytes * IPv6: src/dst IP (16+16) + src/dst port (2+2) = 36 bytes */ #ifdef SIFTR_IPV6 #define FLOW_KEY_LEN 37 #else #define FLOW_KEY_LEN 13 #endif #ifdef SIFTR_IPV6 #define SIFTR_IPMODE 6 #else #define SIFTR_IPMODE 4 #endif /* useful macros */ #define CAST_PTR_INT(X) (*((int*)(X))) #define UPPER_SHORT(X) (((X) & 0xFFFF0000) >> 16) #define LOWER_SHORT(X) ((X) & 0x0000FFFF) #define FIRST_OCTET(X) (((X) & 0xFF000000) >> 24) #define SECOND_OCTET(X) (((X) & 0x00FF0000) >> 16) #define THIRD_OCTET(X) (((X) & 0x0000FF00) >> 8) #define FOURTH_OCTET(X) ((X) & 0x000000FF) static MALLOC_DEFINE(M_SIFTR, "siftr", "dynamic memory used by SIFTR"); static MALLOC_DEFINE(M_SIFTR_PKTNODE, "siftr_pktnode", "SIFTR pkt_node struct"); static MALLOC_DEFINE(M_SIFTR_HASHNODE, "siftr_hashnode", "SIFTR flow_hash_node struct"); /* Used as links in the pkt manager queue. */ struct pkt_node { /* Timestamp of pkt as noted in the pfil hook. */ struct timeval tval; /* Direction pkt is travelling; either PFIL_IN or PFIL_OUT. */ uint8_t direction; /* IP version pkt_node relates to; either INP_IPV4 or INP_IPV6. */ uint8_t ipver; /* Hash of the pkt which triggered the log message. */ uint32_t hash; /* Local/foreign IP address. */ #ifdef SIFTR_IPV6 uint32_t ip_laddr[4]; uint32_t ip_faddr[4]; #else uint8_t ip_laddr[4]; uint8_t ip_faddr[4]; #endif /* Local TCP port. */ uint16_t tcp_localport; /* Foreign TCP port. */ uint16_t tcp_foreignport; /* Congestion Window (bytes). */ u_long snd_cwnd; /* Sending Window (bytes). */ u_long snd_wnd; /* Receive Window (bytes). */ u_long rcv_wnd; /* Unused (was: Bandwidth Controlled Window (bytes)). */ u_long snd_bwnd; /* Slow Start Threshold (bytes). */ u_long snd_ssthresh; /* Current state of the TCP FSM. */ int conn_state; /* Max Segment Size (bytes). */ u_int max_seg_size; /* * Smoothed RTT stored as found in the TCP control block * in units of (TCP_RTT_SCALE*hz). */ int smoothed_rtt; /* Is SACK enabled? */ u_char sack_enabled; /* Window scaling for snd window. */ u_char snd_scale; /* Window scaling for recv window. */ u_char rcv_scale; /* TCP control block flags. */ u_int flags; /* Retransmit timeout length. */ int rxt_length; /* Size of the TCP send buffer in bytes. */ u_int snd_buf_hiwater; /* Current num bytes in the send socket buffer. */ u_int snd_buf_cc; /* Size of the TCP receive buffer in bytes. */ u_int rcv_buf_hiwater; /* Current num bytes in the receive socket buffer. */ u_int rcv_buf_cc; /* Number of bytes inflight that we are waiting on ACKs for. */ u_int sent_inflight_bytes; /* Number of segments currently in the reassembly queue. */ int t_segqlen; /* Flowid for the connection. */ u_int flowid; /* Flow type for the connection. */ u_int flowtype; /* Link to next pkt_node in the list. */ STAILQ_ENTRY(pkt_node) nodes; }; struct flow_hash_node { uint16_t counter; uint8_t key[FLOW_KEY_LEN]; LIST_ENTRY(flow_hash_node) nodes; }; struct siftr_stats { /* # TCP pkts seen by the SIFTR PFIL hooks, including any skipped. */ uint64_t n_in; uint64_t n_out; /* # pkts skipped due to failed malloc calls. */ uint32_t nskip_in_malloc; uint32_t nskip_out_malloc; /* # pkts skipped due to failed mtx acquisition. */ uint32_t nskip_in_mtx; uint32_t nskip_out_mtx; /* # pkts skipped due to failed inpcb lookups. */ uint32_t nskip_in_inpcb; uint32_t nskip_out_inpcb; /* # pkts skipped due to failed tcpcb lookups. */ uint32_t nskip_in_tcpcb; uint32_t nskip_out_tcpcb; /* # pkts skipped due to stack reinjection. */ uint32_t nskip_in_dejavu; uint32_t nskip_out_dejavu; }; DPCPU_DEFINE_STATIC(struct siftr_stats, ss); static volatile unsigned int siftr_exit_pkt_manager_thread = 0; static unsigned int siftr_enabled = 0; static unsigned int siftr_pkts_per_log = 1; static unsigned int siftr_generate_hashes = 0; /* static unsigned int siftr_binary_log = 0; */ static char siftr_logfile[PATH_MAX] = "/var/log/siftr.log"; static char siftr_logfile_shadow[PATH_MAX] = "/var/log/siftr.log"; static u_long siftr_hashmask; STAILQ_HEAD(pkthead, pkt_node) pkt_queue = STAILQ_HEAD_INITIALIZER(pkt_queue); LIST_HEAD(listhead, flow_hash_node) *counter_hash; static int wait_for_pkt; static struct alq *siftr_alq = NULL; static struct mtx siftr_pkt_queue_mtx; static struct mtx siftr_pkt_mgr_mtx; static struct thread *siftr_pkt_manager_thr = NULL; /* * pfil.h defines PFIL_IN as 1 and PFIL_OUT as 2, * which we use as an index into this array. */ static char direction[3] = {'\0', 'i','o'}; /* Required function prototypes. */ static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS); static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS); /* Declare the net.inet.siftr sysctl tree and populate it. */ SYSCTL_DECL(_net_inet_siftr); SYSCTL_NODE(_net_inet, OID_AUTO, siftr, CTLFLAG_RW, NULL, "siftr related settings"); SYSCTL_PROC(_net_inet_siftr, OID_AUTO, enabled, CTLTYPE_UINT|CTLFLAG_RW, &siftr_enabled, 0, &siftr_sysctl_enabled_handler, "IU", "switch siftr module operations on/off"); SYSCTL_PROC(_net_inet_siftr, OID_AUTO, logfile, CTLTYPE_STRING|CTLFLAG_RW, &siftr_logfile_shadow, sizeof(siftr_logfile_shadow), &siftr_sysctl_logfile_name_handler, "A", "file to save siftr log messages to"); SYSCTL_UINT(_net_inet_siftr, OID_AUTO, ppl, CTLFLAG_RW, &siftr_pkts_per_log, 1, "number of packets between generating a log message"); SYSCTL_UINT(_net_inet_siftr, OID_AUTO, genhashes, CTLFLAG_RW, &siftr_generate_hashes, 0, "enable packet hash generation"); /* XXX: TODO SYSCTL_UINT(_net_inet_siftr, OID_AUTO, binary, CTLFLAG_RW, &siftr_binary_log, 0, "write log files in binary instead of ascii"); */ /* Begin functions. */ static void siftr_process_pkt(struct pkt_node * pkt_node) { struct flow_hash_node *hash_node; struct listhead *counter_list; struct siftr_stats *ss; struct ale *log_buf; uint8_t key[FLOW_KEY_LEN]; uint8_t found_match, key_offset; hash_node = NULL; ss = DPCPU_PTR(ss); found_match = 0; key_offset = 1; /* * Create the key that will be used to create a hash index * into our hash table. Our key consists of: * ipversion, localip, localport, foreignip, foreignport */ key[0] = pkt_node->ipver; memcpy(key + key_offset, &pkt_node->ip_laddr, sizeof(pkt_node->ip_laddr)); key_offset += sizeof(pkt_node->ip_laddr); memcpy(key + key_offset, &pkt_node->tcp_localport, sizeof(pkt_node->tcp_localport)); key_offset += sizeof(pkt_node->tcp_localport); memcpy(key + key_offset, &pkt_node->ip_faddr, sizeof(pkt_node->ip_faddr)); key_offset += sizeof(pkt_node->ip_faddr); memcpy(key + key_offset, &pkt_node->tcp_foreignport, sizeof(pkt_node->tcp_foreignport)); counter_list = counter_hash + (hash32_buf(key, sizeof(key), 0) & siftr_hashmask); /* * If the list is not empty i.e. the hash index has * been used by another flow previously. */ if (LIST_FIRST(counter_list) != NULL) { /* * Loop through the hash nodes in the list. * There should normally only be 1 hash node in the list, * except if there have been collisions at the hash index * computed by hash32_buf(). */ LIST_FOREACH(hash_node, counter_list, nodes) { /* * Check if the key for the pkt we are currently * processing is the same as the key stored in the * hash node we are currently processing. * If they are the same, then we've found the * hash node that stores the counter for the flow * the pkt belongs to. */ if (memcmp(hash_node->key, key, sizeof(key)) == 0) { found_match = 1; break; } } } /* If this flow hash hasn't been seen before or we have a collision. */ if (hash_node == NULL || !found_match) { /* Create a new hash node to store the flow's counter. */ hash_node = malloc(sizeof(struct flow_hash_node), M_SIFTR_HASHNODE, M_WAITOK); if (hash_node != NULL) { /* Initialise our new hash node list entry. */ hash_node->counter = 0; memcpy(hash_node->key, key, sizeof(key)); LIST_INSERT_HEAD(counter_list, hash_node, nodes); } else { /* Malloc failed. */ if (pkt_node->direction == PFIL_IN) ss->nskip_in_malloc++; else ss->nskip_out_malloc++; return; } } else if (siftr_pkts_per_log > 1) { /* * Taking the remainder of the counter divided * by the current value of siftr_pkts_per_log * and storing that in counter provides a neat * way to modulate the frequency of log * messages being written to the log file. */ hash_node->counter = (hash_node->counter + 1) % siftr_pkts_per_log; /* * If we have not seen enough packets since the last time * we wrote a log message for this connection, return. */ if (hash_node->counter > 0) return; } log_buf = alq_getn(siftr_alq, MAX_LOG_MSG_LEN, ALQ_WAITOK); if (log_buf == NULL) return; /* Should only happen if the ALQ is shutting down. */ #ifdef SIFTR_IPV6 pkt_node->ip_laddr[3] = ntohl(pkt_node->ip_laddr[3]); pkt_node->ip_faddr[3] = ntohl(pkt_node->ip_faddr[3]); if (pkt_node->ipver == INP_IPV6) { /* IPv6 packet */ pkt_node->ip_laddr[0] = ntohl(pkt_node->ip_laddr[0]); pkt_node->ip_laddr[1] = ntohl(pkt_node->ip_laddr[1]); pkt_node->ip_laddr[2] = ntohl(pkt_node->ip_laddr[2]); pkt_node->ip_faddr[0] = ntohl(pkt_node->ip_faddr[0]); pkt_node->ip_faddr[1] = ntohl(pkt_node->ip_faddr[1]); pkt_node->ip_faddr[2] = ntohl(pkt_node->ip_faddr[2]); /* Construct an IPv6 log message. */ log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN, "%c,0x%08x,%zd.%06ld,%x:%x:%x:%x:%x:%x:%x:%x,%u,%x:%x:%x:" "%x:%x:%x:%x:%x,%u,%ld,%ld,%ld,%ld,%ld,%u,%u,%u,%u,%u,%u," "%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n", direction[pkt_node->direction], pkt_node->hash, pkt_node->tval.tv_sec, pkt_node->tval.tv_usec, UPPER_SHORT(pkt_node->ip_laddr[0]), LOWER_SHORT(pkt_node->ip_laddr[0]), UPPER_SHORT(pkt_node->ip_laddr[1]), LOWER_SHORT(pkt_node->ip_laddr[1]), UPPER_SHORT(pkt_node->ip_laddr[2]), LOWER_SHORT(pkt_node->ip_laddr[2]), UPPER_SHORT(pkt_node->ip_laddr[3]), LOWER_SHORT(pkt_node->ip_laddr[3]), ntohs(pkt_node->tcp_localport), UPPER_SHORT(pkt_node->ip_faddr[0]), LOWER_SHORT(pkt_node->ip_faddr[0]), UPPER_SHORT(pkt_node->ip_faddr[1]), LOWER_SHORT(pkt_node->ip_faddr[1]), UPPER_SHORT(pkt_node->ip_faddr[2]), LOWER_SHORT(pkt_node->ip_faddr[2]), UPPER_SHORT(pkt_node->ip_faddr[3]), LOWER_SHORT(pkt_node->ip_faddr[3]), ntohs(pkt_node->tcp_foreignport), pkt_node->snd_ssthresh, pkt_node->snd_cwnd, pkt_node->snd_bwnd, pkt_node->snd_wnd, pkt_node->rcv_wnd, pkt_node->snd_scale, pkt_node->rcv_scale, pkt_node->conn_state, pkt_node->max_seg_size, pkt_node->smoothed_rtt, pkt_node->sack_enabled, pkt_node->flags, pkt_node->rxt_length, pkt_node->snd_buf_hiwater, pkt_node->snd_buf_cc, pkt_node->rcv_buf_hiwater, pkt_node->rcv_buf_cc, pkt_node->sent_inflight_bytes, pkt_node->t_segqlen, pkt_node->flowid, pkt_node->flowtype); } else { /* IPv4 packet */ pkt_node->ip_laddr[0] = FIRST_OCTET(pkt_node->ip_laddr[3]); pkt_node->ip_laddr[1] = SECOND_OCTET(pkt_node->ip_laddr[3]); pkt_node->ip_laddr[2] = THIRD_OCTET(pkt_node->ip_laddr[3]); pkt_node->ip_laddr[3] = FOURTH_OCTET(pkt_node->ip_laddr[3]); pkt_node->ip_faddr[0] = FIRST_OCTET(pkt_node->ip_faddr[3]); pkt_node->ip_faddr[1] = SECOND_OCTET(pkt_node->ip_faddr[3]); pkt_node->ip_faddr[2] = THIRD_OCTET(pkt_node->ip_faddr[3]); pkt_node->ip_faddr[3] = FOURTH_OCTET(pkt_node->ip_faddr[3]); #endif /* SIFTR_IPV6 */ /* Construct an IPv4 log message. */ log_buf->ae_bytesused = snprintf(log_buf->ae_data, MAX_LOG_MSG_LEN, "%c,0x%08x,%jd.%06ld,%u.%u.%u.%u,%u,%u.%u.%u.%u,%u,%ld,%ld," "%ld,%ld,%ld,%u,%u,%u,%u,%u,%u,%u,%d,%u,%u,%u,%u,%u,%u,%u,%u\n", direction[pkt_node->direction], pkt_node->hash, (intmax_t)pkt_node->tval.tv_sec, pkt_node->tval.tv_usec, pkt_node->ip_laddr[0], pkt_node->ip_laddr[1], pkt_node->ip_laddr[2], pkt_node->ip_laddr[3], ntohs(pkt_node->tcp_localport), pkt_node->ip_faddr[0], pkt_node->ip_faddr[1], pkt_node->ip_faddr[2], pkt_node->ip_faddr[3], ntohs(pkt_node->tcp_foreignport), pkt_node->snd_ssthresh, pkt_node->snd_cwnd, pkt_node->snd_bwnd, pkt_node->snd_wnd, pkt_node->rcv_wnd, pkt_node->snd_scale, pkt_node->rcv_scale, pkt_node->conn_state, pkt_node->max_seg_size, pkt_node->smoothed_rtt, pkt_node->sack_enabled, pkt_node->flags, pkt_node->rxt_length, pkt_node->snd_buf_hiwater, pkt_node->snd_buf_cc, pkt_node->rcv_buf_hiwater, pkt_node->rcv_buf_cc, pkt_node->sent_inflight_bytes, pkt_node->t_segqlen, pkt_node->flowid, pkt_node->flowtype); #ifdef SIFTR_IPV6 } #endif alq_post_flags(siftr_alq, log_buf, 0); } static void siftr_pkt_manager_thread(void *arg) { STAILQ_HEAD(pkthead, pkt_node) tmp_pkt_queue = STAILQ_HEAD_INITIALIZER(tmp_pkt_queue); struct pkt_node *pkt_node, *pkt_node_temp; uint8_t draining; draining = 2; mtx_lock(&siftr_pkt_mgr_mtx); /* draining == 0 when queue has been flushed and it's safe to exit. */ while (draining) { /* * Sleep until we are signalled to wake because thread has * been told to exit or until 1 tick has passed. */ mtx_sleep(&wait_for_pkt, &siftr_pkt_mgr_mtx, PWAIT, "pktwait", 1); /* Gain exclusive access to the pkt_node queue. */ mtx_lock(&siftr_pkt_queue_mtx); /* * Move pkt_queue to tmp_pkt_queue, which leaves * pkt_queue empty and ready to receive more pkt_nodes. */ STAILQ_CONCAT(&tmp_pkt_queue, &pkt_queue); /* * We've finished making changes to the list. Unlock it * so the pfil hooks can continue queuing pkt_nodes. */ mtx_unlock(&siftr_pkt_queue_mtx); /* * We can't hold a mutex whilst calling siftr_process_pkt * because ALQ might sleep waiting for buffer space. */ mtx_unlock(&siftr_pkt_mgr_mtx); /* Flush all pkt_nodes to the log file. */ STAILQ_FOREACH_SAFE(pkt_node, &tmp_pkt_queue, nodes, pkt_node_temp) { siftr_process_pkt(pkt_node); STAILQ_REMOVE_HEAD(&tmp_pkt_queue, nodes); free(pkt_node, M_SIFTR_PKTNODE); } KASSERT(STAILQ_EMPTY(&tmp_pkt_queue), ("SIFTR tmp_pkt_queue not empty after flush")); mtx_lock(&siftr_pkt_mgr_mtx); /* * If siftr_exit_pkt_manager_thread gets set during the window * where we are draining the tmp_pkt_queue above, there might * still be pkts in pkt_queue that need to be drained. * Allow one further iteration to occur after * siftr_exit_pkt_manager_thread has been set to ensure * pkt_queue is completely empty before we kill the thread. * * siftr_exit_pkt_manager_thread is set only after the pfil * hooks have been removed, so only 1 extra iteration * is needed to drain the queue. */ if (siftr_exit_pkt_manager_thread) draining--; } mtx_unlock(&siftr_pkt_mgr_mtx); /* Calls wakeup on this thread's struct thread ptr. */ kthread_exit(); } static uint32_t hash_pkt(struct mbuf *m, uint32_t offset) { uint32_t hash; hash = 0; while (m != NULL && offset > m->m_len) { /* * The IP packet payload does not start in this mbuf, so * need to figure out which mbuf it starts in and what offset * into the mbuf's data region the payload starts at. */ offset -= m->m_len; m = m->m_next; } while (m != NULL) { /* Ensure there is data in the mbuf */ if ((m->m_len - offset) > 0) hash = hash32_buf(m->m_data + offset, m->m_len - offset, hash); m = m->m_next; offset = 0; } return (hash); } /* * Check if a given mbuf has the SIFTR mbuf tag. If it does, log the fact that * it's a reinjected packet and return. If it doesn't, tag the mbuf and return. * Return value >0 means the caller should skip processing this mbuf. */ static inline int siftr_chkreinject(struct mbuf *m, int dir, struct siftr_stats *ss) { if (m_tag_locate(m, PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, NULL) != NULL) { if (dir == PFIL_IN) ss->nskip_in_dejavu++; else ss->nskip_out_dejavu++; return (1); } else { struct m_tag *tag = m_tag_alloc(PACKET_COOKIE_SIFTR, PACKET_TAG_SIFTR, 0, M_NOWAIT); if (tag == NULL) { if (dir == PFIL_IN) ss->nskip_in_malloc++; else ss->nskip_out_malloc++; return (1); } m_tag_prepend(m, tag); } return (0); } /* * Look up an inpcb for a packet. Return the inpcb pointer if found, or NULL * otherwise. */ static inline struct inpcb * siftr_findinpcb(int ipver, struct ip *ip, struct mbuf *m, uint16_t sport, uint16_t dport, int dir, struct siftr_stats *ss) { struct inpcb *inp; /* We need the tcbinfo lock. */ - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); if (dir == PFIL_IN) inp = (ipver == INP_IPV4 ? in_pcblookup(&V_tcbinfo, ip->ip_src, sport, ip->ip_dst, dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) : #ifdef SIFTR_IPV6 in6_pcblookup(&V_tcbinfo, &((struct ip6_hdr *)ip)->ip6_src, sport, &((struct ip6_hdr *)ip)->ip6_dst, dport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) #else NULL #endif ); else inp = (ipver == INP_IPV4 ? in_pcblookup(&V_tcbinfo, ip->ip_dst, dport, ip->ip_src, sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) : #ifdef SIFTR_IPV6 in6_pcblookup(&V_tcbinfo, &((struct ip6_hdr *)ip)->ip6_dst, dport, &((struct ip6_hdr *)ip)->ip6_src, sport, INPLOOKUP_RLOCKPCB, m->m_pkthdr.rcvif) #else NULL #endif ); /* If we can't find the inpcb, bail. */ if (inp == NULL) { if (dir == PFIL_IN) ss->nskip_in_inpcb++; else ss->nskip_out_inpcb++; } return (inp); } static inline void siftr_siftdata(struct pkt_node *pn, struct inpcb *inp, struct tcpcb *tp, int ipver, int dir, int inp_locally_locked) { #ifdef SIFTR_IPV6 if (ipver == INP_IPV4) { pn->ip_laddr[3] = inp->inp_laddr.s_addr; pn->ip_faddr[3] = inp->inp_faddr.s_addr; #else *((uint32_t *)pn->ip_laddr) = inp->inp_laddr.s_addr; *((uint32_t *)pn->ip_faddr) = inp->inp_faddr.s_addr; #endif #ifdef SIFTR_IPV6 } else { pn->ip_laddr[0] = inp->in6p_laddr.s6_addr32[0]; pn->ip_laddr[1] = inp->in6p_laddr.s6_addr32[1]; pn->ip_laddr[2] = inp->in6p_laddr.s6_addr32[2]; pn->ip_laddr[3] = inp->in6p_laddr.s6_addr32[3]; pn->ip_faddr[0] = inp->in6p_faddr.s6_addr32[0]; pn->ip_faddr[1] = inp->in6p_faddr.s6_addr32[1]; pn->ip_faddr[2] = inp->in6p_faddr.s6_addr32[2]; pn->ip_faddr[3] = inp->in6p_faddr.s6_addr32[3]; } #endif pn->tcp_localport = inp->inp_lport; pn->tcp_foreignport = inp->inp_fport; pn->snd_cwnd = tp->snd_cwnd; pn->snd_wnd = tp->snd_wnd; pn->rcv_wnd = tp->rcv_wnd; pn->snd_bwnd = 0; /* Unused, kept for compat. */ pn->snd_ssthresh = tp->snd_ssthresh; pn->snd_scale = tp->snd_scale; pn->rcv_scale = tp->rcv_scale; pn->conn_state = tp->t_state; pn->max_seg_size = tp->t_maxseg; pn->smoothed_rtt = tp->t_srtt; pn->sack_enabled = (tp->t_flags & TF_SACK_PERMIT) != 0; pn->flags = tp->t_flags; pn->rxt_length = tp->t_rxtcur; pn->snd_buf_hiwater = inp->inp_socket->so_snd.sb_hiwat; pn->snd_buf_cc = sbused(&inp->inp_socket->so_snd); pn->rcv_buf_hiwater = inp->inp_socket->so_rcv.sb_hiwat; pn->rcv_buf_cc = sbused(&inp->inp_socket->so_rcv); pn->sent_inflight_bytes = tp->snd_max - tp->snd_una; pn->t_segqlen = tp->t_segqlen; pn->flowid = inp->inp_flowid; pn->flowtype = inp->inp_flowtype; /* We've finished accessing the tcb so release the lock. */ if (inp_locally_locked) INP_RUNLOCK(inp); pn->ipver = ipver; pn->direction = dir; /* * Significantly more accurate than using getmicrotime(), but slower! * Gives true microsecond resolution at the expense of a hit to * maximum pps throughput processing when SIFTR is loaded and enabled. */ microtime(&pn->tval); TCP_PROBE1(siftr, &pn); } /* * pfil hook that is called for each IPv4 packet making its way through the * stack in either direction. * The pfil subsystem holds a non-sleepable mutex somewhere when * calling our hook function, so we can't sleep at all. * It's very important to use the M_NOWAIT flag with all function calls * that support it so that they won't sleep, otherwise you get a panic. */ static int siftr_chkpkt(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, struct inpcb *inp) { struct pkt_node *pn; struct ip *ip; struct tcphdr *th; struct tcpcb *tp; struct siftr_stats *ss; unsigned int ip_hl; int inp_locally_locked; inp_locally_locked = 0; ss = DPCPU_PTR(ss); /* * m_pullup is not required here because ip_{input|output} * already do the heavy lifting for us. */ ip = mtod(*m, struct ip *); /* Only continue processing if the packet is TCP. */ if (ip->ip_p != IPPROTO_TCP) goto ret; /* * If a kernel subsystem reinjects packets into the stack, our pfil * hook will be called multiple times for the same packet. * Make sure we only process unique packets. */ if (siftr_chkreinject(*m, dir, ss)) goto ret; if (dir == PFIL_IN) ss->n_in++; else ss->n_out++; /* * Create a tcphdr struct starting at the correct offset * in the IP packet. ip->ip_hl gives the ip header length * in 4-byte words, so multiply it to get the size in bytes. */ ip_hl = (ip->ip_hl << 2); th = (struct tcphdr *)((caddr_t)ip + ip_hl); /* * If the pfil hooks don't provide a pointer to the * inpcb, we need to find it ourselves and lock it. */ if (!inp) { /* Find the corresponding inpcb for this pkt. */ inp = siftr_findinpcb(INP_IPV4, ip, *m, th->th_sport, th->th_dport, dir, ss); if (inp == NULL) goto ret; else inp_locally_locked = 1; } INP_LOCK_ASSERT(inp); /* Find the TCP control block that corresponds with this packet */ tp = intotcpcb(inp); /* * If we can't find the TCP control block (happens occasionaly for a * packet sent during the shutdown phase of a TCP connection), * or we're in the timewait state, bail */ if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) { if (dir == PFIL_IN) ss->nskip_in_tcpcb++; else ss->nskip_out_tcpcb++; goto inp_unlock; } pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); if (pn == NULL) { if (dir == PFIL_IN) ss->nskip_in_malloc++; else ss->nskip_out_malloc++; goto inp_unlock; } siftr_siftdata(pn, inp, tp, INP_IPV4, dir, inp_locally_locked); if (siftr_generate_hashes) { if ((*m)->m_pkthdr.csum_flags & CSUM_TCP) { /* * For outbound packets, the TCP checksum isn't * calculated yet. This is a problem for our packet * hashing as the receiver will calc a different hash * to ours if we don't include the correct TCP checksum * in the bytes being hashed. To work around this * problem, we manually calc the TCP checksum here in * software. We unset the CSUM_TCP flag so the lower * layers don't recalc it. */ (*m)->m_pkthdr.csum_flags &= ~CSUM_TCP; /* * Calculate the TCP checksum in software and assign * to correct TCP header field, which will follow the * packet mbuf down the stack. The trick here is that * tcp_output() sets th->th_sum to the checksum of the * pseudo header for us already. Because of the nature * of the checksumming algorithm, we can sum over the * entire IP payload (i.e. TCP header and data), which * will include the already calculated pseduo header * checksum, thus giving us the complete TCP checksum. * * To put it in simple terms, if checksum(1,2,3,4)=10, * then checksum(1,2,3,4,5) == checksum(10,5). * This property is what allows us to "cheat" and * checksum only the IP payload which has the TCP * th_sum field populated with the pseudo header's * checksum, and not need to futz around checksumming * pseudo header bytes and TCP header/data in one hit. * Refer to RFC 1071 for more info. * * NB: in_cksum_skip(struct mbuf *m, int len, int skip) * in_cksum_skip 2nd argument is NOT the number of * bytes to read from the mbuf at "skip" bytes offset * from the start of the mbuf (very counter intuitive!). * The number of bytes to read is calculated internally * by the function as len-skip i.e. to sum over the IP * payload (TCP header + data) bytes, it is INCORRECT * to call the function like this: * in_cksum_skip(at, ip->ip_len - offset, offset) * Rather, it should be called like this: * in_cksum_skip(at, ip->ip_len, offset) * which means read "ip->ip_len - offset" bytes from * the mbuf cluster "at" at offset "offset" bytes from * the beginning of the "at" mbuf's data pointer. */ th->th_sum = in_cksum_skip(*m, ntohs(ip->ip_len), ip_hl); } /* * XXX: Having to calculate the checksum in software and then * hash over all bytes is really inefficient. Would be nice to * find a way to create the hash and checksum in the same pass * over the bytes. */ pn->hash = hash_pkt(*m, ip_hl); } mtx_lock(&siftr_pkt_queue_mtx); STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); mtx_unlock(&siftr_pkt_queue_mtx); goto ret; inp_unlock: if (inp_locally_locked) INP_RUNLOCK(inp); ret: /* Returning 0 ensures pfil will not discard the pkt */ return (0); } #ifdef SIFTR_IPV6 static int siftr_chkpkt6(void *arg, struct mbuf **m, struct ifnet *ifp, int dir, struct inpcb *inp) { struct pkt_node *pn; struct ip6_hdr *ip6; struct tcphdr *th; struct tcpcb *tp; struct siftr_stats *ss; unsigned int ip6_hl; int inp_locally_locked; inp_locally_locked = 0; ss = DPCPU_PTR(ss); /* * m_pullup is not required here because ip6_{input|output} * already do the heavy lifting for us. */ ip6 = mtod(*m, struct ip6_hdr *); /* * Only continue processing if the packet is TCP * XXX: We should follow the next header fields * as shown on Pg 6 RFC 2460, but right now we'll * only check pkts that have no extension headers. */ if (ip6->ip6_nxt != IPPROTO_TCP) goto ret6; /* * If a kernel subsystem reinjects packets into the stack, our pfil * hook will be called multiple times for the same packet. * Make sure we only process unique packets. */ if (siftr_chkreinject(*m, dir, ss)) goto ret6; if (dir == PFIL_IN) ss->n_in++; else ss->n_out++; ip6_hl = sizeof(struct ip6_hdr); /* * Create a tcphdr struct starting at the correct offset * in the ipv6 packet. ip->ip_hl gives the ip header length * in 4-byte words, so multiply it to get the size in bytes. */ th = (struct tcphdr *)((caddr_t)ip6 + ip6_hl); /* * For inbound packets, the pfil hooks don't provide a pointer to the * inpcb, so we need to find it ourselves and lock it. */ if (!inp) { /* Find the corresponding inpcb for this pkt. */ inp = siftr_findinpcb(INP_IPV6, (struct ip *)ip6, *m, th->th_sport, th->th_dport, dir, ss); if (inp == NULL) goto ret6; else inp_locally_locked = 1; } /* Find the TCP control block that corresponds with this packet. */ tp = intotcpcb(inp); /* * If we can't find the TCP control block (happens occasionaly for a * packet sent during the shutdown phase of a TCP connection), * or we're in the timewait state, bail. */ if (tp == NULL || inp->inp_flags & INP_TIMEWAIT) { if (dir == PFIL_IN) ss->nskip_in_tcpcb++; else ss->nskip_out_tcpcb++; goto inp_unlock6; } pn = malloc(sizeof(struct pkt_node), M_SIFTR_PKTNODE, M_NOWAIT|M_ZERO); if (pn == NULL) { if (dir == PFIL_IN) ss->nskip_in_malloc++; else ss->nskip_out_malloc++; goto inp_unlock6; } siftr_siftdata(pn, inp, tp, INP_IPV6, dir, inp_locally_locked); /* XXX: Figure out how to generate hashes for IPv6 packets. */ mtx_lock(&siftr_pkt_queue_mtx); STAILQ_INSERT_TAIL(&pkt_queue, pn, nodes); mtx_unlock(&siftr_pkt_queue_mtx); goto ret6; inp_unlock6: if (inp_locally_locked) INP_RUNLOCK(inp); ret6: /* Returning 0 ensures pfil will not discard the pkt. */ return (0); } #endif /* #ifdef SIFTR_IPV6 */ static int siftr_pfil(int action) { struct pfil_head *pfh_inet; #ifdef SIFTR_IPV6 struct pfil_head *pfh_inet6; #endif VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); pfh_inet = pfil_head_get(PFIL_TYPE_AF, AF_INET); #ifdef SIFTR_IPV6 pfh_inet6 = pfil_head_get(PFIL_TYPE_AF, AF_INET6); #endif if (action == HOOK) { pfil_add_hook(siftr_chkpkt, NULL, PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); #ifdef SIFTR_IPV6 pfil_add_hook(siftr_chkpkt6, NULL, PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); #endif } else if (action == UNHOOK) { pfil_remove_hook(siftr_chkpkt, NULL, PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet); #ifdef SIFTR_IPV6 pfil_remove_hook(siftr_chkpkt6, NULL, PFIL_IN | PFIL_OUT | PFIL_WAITOK, pfh_inet6); #endif } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } static int siftr_sysctl_logfile_name_handler(SYSCTL_HANDLER_ARGS) { struct alq *new_alq; int error; error = sysctl_handle_string(oidp, arg1, arg2, req); /* Check for error or same filename */ if (error != 0 || req->newptr == NULL || strncmp(siftr_logfile, arg1, arg2) == 0) goto done; /* Filname changed */ error = alq_open(&new_alq, arg1, curthread->td_ucred, SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); if (error != 0) goto done; /* * If disabled, siftr_alq == NULL so we simply close * the alq as we've proved it can be opened. * If enabled, close the existing alq and switch the old * for the new. */ if (siftr_alq == NULL) { alq_close(new_alq); } else { alq_close(siftr_alq); siftr_alq = new_alq; } /* Update filename upon success */ strlcpy(siftr_logfile, arg1, arg2); done: return (error); } static int siftr_manage_ops(uint8_t action) { struct siftr_stats totalss; struct timeval tval; struct flow_hash_node *counter, *tmp_counter; struct sbuf *s; int i, key_index, error; uint32_t bytes_to_write, total_skipped_pkts; uint16_t lport, fport; uint8_t *key, ipver __unused; #ifdef SIFTR_IPV6 uint32_t laddr[4]; uint32_t faddr[4]; #else uint8_t laddr[4]; uint8_t faddr[4]; #endif error = 0; total_skipped_pkts = 0; /* Init an autosizing sbuf that initially holds 200 chars. */ if ((s = sbuf_new(NULL, NULL, 200, SBUF_AUTOEXTEND)) == NULL) return (-1); if (action == SIFTR_ENABLE) { /* * Create our alq * XXX: We should abort if alq_open fails! */ alq_open(&siftr_alq, siftr_logfile, curthread->td_ucred, SIFTR_LOG_FILE_MODE, SIFTR_ALQ_BUFLEN, 0); STAILQ_INIT(&pkt_queue); DPCPU_ZERO(ss); siftr_exit_pkt_manager_thread = 0; kthread_add(&siftr_pkt_manager_thread, NULL, NULL, &siftr_pkt_manager_thr, RFNOWAIT, 0, "siftr_pkt_manager_thr"); siftr_pfil(HOOK); microtime(&tval); sbuf_printf(s, "enable_time_secs=%jd\tenable_time_usecs=%06ld\t" "siftrver=%s\thz=%u\ttcp_rtt_scale=%u\tsysname=%s\t" "sysver=%u\tipmode=%u\n", (intmax_t)tval.tv_sec, tval.tv_usec, MODVERSION_STR, hz, TCP_RTT_SCALE, SYS_NAME, __FreeBSD_version, SIFTR_IPMODE); sbuf_finish(s); alq_writen(siftr_alq, sbuf_data(s), sbuf_len(s), ALQ_WAITOK); } else if (action == SIFTR_DISABLE && siftr_pkt_manager_thr != NULL) { /* * Remove the pfil hook functions. All threads currently in * the hook functions are allowed to exit before siftr_pfil() * returns. */ siftr_pfil(UNHOOK); /* This will block until the pkt manager thread unlocks it. */ mtx_lock(&siftr_pkt_mgr_mtx); /* Tell the pkt manager thread that it should exit now. */ siftr_exit_pkt_manager_thread = 1; /* * Wake the pkt_manager thread so it realises that * siftr_exit_pkt_manager_thread == 1 and exits gracefully. * The wakeup won't be delivered until we unlock * siftr_pkt_mgr_mtx so this isn't racy. */ wakeup(&wait_for_pkt); /* Wait for the pkt_manager thread to exit. */ mtx_sleep(siftr_pkt_manager_thr, &siftr_pkt_mgr_mtx, PWAIT, "thrwait", 0); siftr_pkt_manager_thr = NULL; mtx_unlock(&siftr_pkt_mgr_mtx); totalss.n_in = DPCPU_VARSUM(ss, n_in); totalss.n_out = DPCPU_VARSUM(ss, n_out); totalss.nskip_in_malloc = DPCPU_VARSUM(ss, nskip_in_malloc); totalss.nskip_out_malloc = DPCPU_VARSUM(ss, nskip_out_malloc); totalss.nskip_in_mtx = DPCPU_VARSUM(ss, nskip_in_mtx); totalss.nskip_out_mtx = DPCPU_VARSUM(ss, nskip_out_mtx); totalss.nskip_in_tcpcb = DPCPU_VARSUM(ss, nskip_in_tcpcb); totalss.nskip_out_tcpcb = DPCPU_VARSUM(ss, nskip_out_tcpcb); totalss.nskip_in_inpcb = DPCPU_VARSUM(ss, nskip_in_inpcb); totalss.nskip_out_inpcb = DPCPU_VARSUM(ss, nskip_out_inpcb); total_skipped_pkts = totalss.nskip_in_malloc + totalss.nskip_out_malloc + totalss.nskip_in_mtx + totalss.nskip_out_mtx + totalss.nskip_in_tcpcb + totalss.nskip_out_tcpcb + totalss.nskip_in_inpcb + totalss.nskip_out_inpcb; microtime(&tval); sbuf_printf(s, "disable_time_secs=%jd\tdisable_time_usecs=%06ld\t" "num_inbound_tcp_pkts=%ju\tnum_outbound_tcp_pkts=%ju\t" "total_tcp_pkts=%ju\tnum_inbound_skipped_pkts_malloc=%u\t" "num_outbound_skipped_pkts_malloc=%u\t" "num_inbound_skipped_pkts_mtx=%u\t" "num_outbound_skipped_pkts_mtx=%u\t" "num_inbound_skipped_pkts_tcpcb=%u\t" "num_outbound_skipped_pkts_tcpcb=%u\t" "num_inbound_skipped_pkts_inpcb=%u\t" "num_outbound_skipped_pkts_inpcb=%u\t" "total_skipped_tcp_pkts=%u\tflow_list=", (intmax_t)tval.tv_sec, tval.tv_usec, (uintmax_t)totalss.n_in, (uintmax_t)totalss.n_out, (uintmax_t)(totalss.n_in + totalss.n_out), totalss.nskip_in_malloc, totalss.nskip_out_malloc, totalss.nskip_in_mtx, totalss.nskip_out_mtx, totalss.nskip_in_tcpcb, totalss.nskip_out_tcpcb, totalss.nskip_in_inpcb, totalss.nskip_out_inpcb, total_skipped_pkts); /* * Iterate over the flow hash, printing a summary of each * flow seen and freeing any malloc'd memory. * The hash consists of an array of LISTs (man 3 queue). */ for (i = 0; i <= siftr_hashmask; i++) { LIST_FOREACH_SAFE(counter, counter_hash + i, nodes, tmp_counter) { key = counter->key; key_index = 1; ipver = key[0]; memcpy(laddr, key + key_index, sizeof(laddr)); key_index += sizeof(laddr); memcpy(&lport, key + key_index, sizeof(lport)); key_index += sizeof(lport); memcpy(faddr, key + key_index, sizeof(faddr)); key_index += sizeof(faddr); memcpy(&fport, key + key_index, sizeof(fport)); #ifdef SIFTR_IPV6 laddr[3] = ntohl(laddr[3]); faddr[3] = ntohl(faddr[3]); if (ipver == INP_IPV6) { laddr[0] = ntohl(laddr[0]); laddr[1] = ntohl(laddr[1]); laddr[2] = ntohl(laddr[2]); faddr[0] = ntohl(faddr[0]); faddr[1] = ntohl(faddr[1]); faddr[2] = ntohl(faddr[2]); sbuf_printf(s, "%x:%x:%x:%x:%x:%x:%x:%x;%u-" "%x:%x:%x:%x:%x:%x:%x:%x;%u,", UPPER_SHORT(laddr[0]), LOWER_SHORT(laddr[0]), UPPER_SHORT(laddr[1]), LOWER_SHORT(laddr[1]), UPPER_SHORT(laddr[2]), LOWER_SHORT(laddr[2]), UPPER_SHORT(laddr[3]), LOWER_SHORT(laddr[3]), ntohs(lport), UPPER_SHORT(faddr[0]), LOWER_SHORT(faddr[0]), UPPER_SHORT(faddr[1]), LOWER_SHORT(faddr[1]), UPPER_SHORT(faddr[2]), LOWER_SHORT(faddr[2]), UPPER_SHORT(faddr[3]), LOWER_SHORT(faddr[3]), ntohs(fport)); } else { laddr[0] = FIRST_OCTET(laddr[3]); laddr[1] = SECOND_OCTET(laddr[3]); laddr[2] = THIRD_OCTET(laddr[3]); laddr[3] = FOURTH_OCTET(laddr[3]); faddr[0] = FIRST_OCTET(faddr[3]); faddr[1] = SECOND_OCTET(faddr[3]); faddr[2] = THIRD_OCTET(faddr[3]); faddr[3] = FOURTH_OCTET(faddr[3]); #endif sbuf_printf(s, "%u.%u.%u.%u;%u-%u.%u.%u.%u;%u,", laddr[0], laddr[1], laddr[2], laddr[3], ntohs(lport), faddr[0], faddr[1], faddr[2], faddr[3], ntohs(fport)); #ifdef SIFTR_IPV6 } #endif free(counter, M_SIFTR_HASHNODE); } LIST_INIT(counter_hash + i); } sbuf_printf(s, "\n"); sbuf_finish(s); i = 0; do { bytes_to_write = min(SIFTR_ALQ_BUFLEN, sbuf_len(s)-i); alq_writen(siftr_alq, sbuf_data(s)+i, bytes_to_write, ALQ_WAITOK); i += bytes_to_write; } while (i < sbuf_len(s)); alq_close(siftr_alq); siftr_alq = NULL; } sbuf_delete(s); /* * XXX: Should be using ret to check if any functions fail * and set error appropriately */ return (error); } static int siftr_sysctl_enabled_handler(SYSCTL_HANDLER_ARGS) { if (req->newptr == NULL) goto skip; /* If the value passed in isn't 0 or 1, return an error. */ if (CAST_PTR_INT(req->newptr) != 0 && CAST_PTR_INT(req->newptr) != 1) return (1); /* If we are changing state (0 to 1 or 1 to 0). */ if (CAST_PTR_INT(req->newptr) != siftr_enabled ) if (siftr_manage_ops(CAST_PTR_INT(req->newptr))) { siftr_manage_ops(SIFTR_DISABLE); return (1); } skip: return (sysctl_handle_int(oidp, arg1, arg2, req)); } static void siftr_shutdown_handler(void *arg) { siftr_manage_ops(SIFTR_DISABLE); } /* * Module is being unloaded or machine is shutting down. Take care of cleanup. */ static int deinit_siftr(void) { /* Cleanup. */ siftr_manage_ops(SIFTR_DISABLE); hashdestroy(counter_hash, M_SIFTR, siftr_hashmask); mtx_destroy(&siftr_pkt_queue_mtx); mtx_destroy(&siftr_pkt_mgr_mtx); return (0); } /* * Module has just been loaded into the kernel. */ static int init_siftr(void) { EVENTHANDLER_REGISTER(shutdown_pre_sync, siftr_shutdown_handler, NULL, SHUTDOWN_PRI_FIRST); /* Initialise our flow counter hash table. */ counter_hash = hashinit(SIFTR_EXPECTED_MAX_TCP_FLOWS, M_SIFTR, &siftr_hashmask); mtx_init(&siftr_pkt_queue_mtx, "siftr_pkt_queue_mtx", NULL, MTX_DEF); mtx_init(&siftr_pkt_mgr_mtx, "siftr_pkt_mgr_mtx", NULL, MTX_DEF); /* Print message to the user's current terminal. */ uprintf("\nStatistical Information For TCP Research (SIFTR) %s\n" " http://caia.swin.edu.au/urp/newtcp\n\n", MODVERSION_STR); return (0); } /* * This is the function that is called to load and unload the module. * When the module is loaded, this function is called once with * "what" == MOD_LOAD * When the module is unloaded, this function is called twice with * "what" = MOD_QUIESCE first, followed by "what" = MOD_UNLOAD second * When the system is shut down e.g. CTRL-ALT-DEL or using the shutdown command, * this function is called once with "what" = MOD_SHUTDOWN * When the system is shut down, the handler isn't called until the very end * of the shutdown sequence i.e. after the disks have been synced. */ static int siftr_load_handler(module_t mod, int what, void *arg) { int ret; switch (what) { case MOD_LOAD: ret = init_siftr(); break; case MOD_QUIESCE: case MOD_SHUTDOWN: ret = deinit_siftr(); break; case MOD_UNLOAD: ret = 0; break; default: ret = EINVAL; break; } return (ret); } static moduledata_t siftr_mod = { .name = "siftr", .evhand = siftr_load_handler, }; /* * Param 1: name of the kernel module * Param 2: moduledata_t struct containing info about the kernel module * and the execution entry point for the module * Param 3: From sysinit_sub_id enumeration in /usr/include/sys/kernel.h * Defines the module initialisation order * Param 4: From sysinit_elem_order enumeration in /usr/include/sys/kernel.h * Defines the initialisation order of this kld relative to others * within the same subsystem as defined by param 3 */ DECLARE_MODULE(siftr, siftr_mod, SI_SUB_LAST, SI_ORDER_ANY); MODULE_DEPEND(siftr, alq, 1, 1, 1); MODULE_VERSION(siftr, MODVERSION); Index: head/sys/netinet/tcp_hpts.c =================================================================== --- head/sys/netinet/tcp_hpts.c (revision 339038) +++ head/sys/netinet/tcp_hpts.c (revision 339039) @@ -1,1902 +1,1902 @@ /*- * Copyright (c) 2016-2018 Netflix Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_tcpdebug.h" /** * Some notes about usage. * * The tcp_hpts system is designed to provide a high precision timer * system for tcp. Its main purpose is to provide a mechanism for * pacing packets out onto the wire. It can be used in two ways * by a given TCP stack (and those two methods can be used simultaneously). * * First, and probably the main thing its used by Rack and BBR for, it can * be used to call tcp_output() of a transport stack at some time in the future. * The normal way this is done is that tcp_output() of the stack schedules * itself to be called again by calling tcp_hpts_insert(tcpcb, slot). The * slot is the time from now that the stack wants to be called but it * must be converted to tcp_hpts's notion of slot. This is done with * one of the macros HPTS_MS_TO_SLOTS or HPTS_USEC_TO_SLOTS. So a typical * call from the tcp_output() routine might look like: * * tcp_hpts_insert(tp, HPTS_USEC_TO_SLOTS(550)); * * The above would schedule tcp_ouput() to be called in 550 useconds. * Note that if using this mechanism the stack will want to add near * its top a check to prevent unwanted calls (from user land or the * arrival of incoming ack's). So it would add something like: * * if (inp->inp_in_hpts) * return; * * to prevent output processing until the time alotted has gone by. * Of course this is a bare bones example and the stack will probably * have more consideration then just the above. * * Now the tcp_hpts system will call tcp_output in one of two forms, * it will first check to see if the stack as defined a * tfb_tcp_output_wtime() function, if so that is the routine it * will call, if that function is not defined then it will call the * tfb_tcp_output() function. The only difference between these * two calls is that the former passes the time in to the function * so the function does not have to access the time (which tcp_hpts * already has). What these functions do is of course totally up * to the individual tcp stack. * * Now the second function (actually two functions I guess :D) * the tcp_hpts system provides is the ability to either abort * a connection (later) or process input on a connection. * Why would you want to do this? To keep processor locality. * * So in order to use the input redirection function the * stack changes its tcp_do_segment() routine to instead * of process the data call the function: * * tcp_queue_pkt_to_input() * * You will note that the arguments to this function look * a lot like tcp_do_segments's arguments. This function * will assure that the tcp_hpts system will * call the functions tfb_tcp_hpts_do_segment() from the * correct CPU. Note that multiple calls can get pushed * into the tcp_hpts system this will be indicated by * the next to last argument to tfb_tcp_hpts_do_segment() * (nxt_pkt). If nxt_pkt is a 1 then another packet is * coming. If nxt_pkt is a 0 then this is the last call * that the tcp_hpts system has available for the tcp stack. * * The other point of the input system is to be able to safely * drop a tcp connection without worrying about the recursive * locking that may be occuring on the INP_WLOCK. So if * a stack wants to drop a connection it calls: * * tcp_set_inp_to_drop(tp, ETIMEDOUT) * * To schedule the tcp_hpts system to call * * tcp_drop(tp, drop_reason) * * at a future point. This is quite handy to prevent locking * issues when dropping connections. * */ #include #include #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #include #include #include #include #ifdef tcpdebug #include #endif /* tcpdebug */ #ifdef tcp_offload #include #endif #include "opt_rss.h" MALLOC_DEFINE(M_TCPHPTS, "tcp_hpts", "TCP hpts"); #ifdef RSS #include #include static int tcp_bind_threads = 1; #else static int tcp_bind_threads = 0; #endif TUNABLE_INT("net.inet.tcp.bind_hptss", &tcp_bind_threads); static uint32_t tcp_hpts_logging_size = DEFAULT_HPTS_LOG; TUNABLE_INT("net.inet.tcp.hpts_logging_sz", &tcp_hpts_logging_size); static struct tcp_hptsi tcp_pace; static void tcp_wakehpts(struct tcp_hpts_entry *p); static void tcp_wakeinput(struct tcp_hpts_entry *p); static void tcp_input_data(struct tcp_hpts_entry *hpts, struct timeval *tv); static void tcp_hptsi(struct tcp_hpts_entry *hpts, struct timeval *ctick); static void tcp_hpts_thread(void *ctx); static void tcp_init_hptsi(void *st); int32_t tcp_min_hptsi_time = DEFAULT_MIN_SLEEP; static int32_t tcp_hpts_callout_skip_swi = 0; SYSCTL_NODE(_net_inet_tcp, OID_AUTO, hpts, CTLFLAG_RW, 0, "TCP Hpts controls"); #define timersub(tvp, uvp, vvp) \ do { \ (vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec; \ (vvp)->tv_usec = (tvp)->tv_usec - (uvp)->tv_usec; \ if ((vvp)->tv_usec < 0) { \ (vvp)->tv_sec--; \ (vvp)->tv_usec += 1000000; \ } \ } while (0) static int32_t logging_on = 0; static int32_t hpts_sleep_max = (NUM_OF_HPTSI_SLOTS - 2); static int32_t tcp_hpts_precision = 120; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, precision, CTLFLAG_RW, &tcp_hpts_precision, 120, "Value for PRE() precision of callout"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, logging, CTLFLAG_RW, &logging_on, 0, "Turn on logging if compiled in"); counter_u64_t hpts_loops; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, loops, CTLFLAG_RD, &hpts_loops, "Number of times hpts had to loop to catch up"); counter_u64_t back_tosleep; SYSCTL_COUNTER_U64(_net_inet_tcp_hpts, OID_AUTO, no_tcbsfound, CTLFLAG_RD, &back_tosleep, "Number of times hpts found no tcbs"); static int32_t in_newts_every_tcb = 0; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, in_tsperpcb, CTLFLAG_RW, &in_newts_every_tcb, 0, "Do we have a new cts every tcb we process for input"); static int32_t in_ts_percision = 0; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, in_tspercision, CTLFLAG_RW, &in_ts_percision, 0, "Do we use percise timestamp for clients on input"); static int32_t out_newts_every_tcb = 0; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, out_tsperpcb, CTLFLAG_RW, &out_newts_every_tcb, 0, "Do we have a new cts every tcb we process for output"); static int32_t out_ts_percision = 0; SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, out_tspercision, CTLFLAG_RW, &out_ts_percision, 0, "Do we use a percise timestamp for every output cts"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, maxsleep, CTLFLAG_RW, &hpts_sleep_max, 0, "The maximum time the hpts will sleep <1 - 254>"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, minsleep, CTLFLAG_RW, &tcp_min_hptsi_time, 0, "The minimum time the hpts must sleep before processing more slots"); SYSCTL_INT(_net_inet_tcp_hpts, OID_AUTO, skip_swi, CTLFLAG_RW, &tcp_hpts_callout_skip_swi, 0, "Do we have the callout call directly to the hpts?"); static void __tcp_hpts_log_it(struct tcp_hpts_entry *hpts, struct inpcb *inp, int event, uint32_t slot, uint32_t ticknow, int32_t line) { struct hpts_log *pl; HPTS_MTX_ASSERT(hpts); if (hpts->p_log == NULL) return; pl = &hpts->p_log[hpts->p_log_at]; hpts->p_log_at++; if (hpts->p_log_at >= hpts->p_logsize) { hpts->p_log_at = 0; hpts->p_log_wrapped = 1; } pl->inp = inp; if (inp) { pl->t_paceslot = inp->inp_hptsslot; pl->t_hptsreq = inp->inp_hpts_request; pl->p_onhpts = inp->inp_in_hpts; pl->p_oninput = inp->inp_in_input; } else { pl->t_paceslot = 0; pl->t_hptsreq = 0; pl->p_onhpts = 0; pl->p_oninput = 0; } pl->is_notempty = 1; pl->event = event; pl->line = line; pl->cts = tcp_get_usecs(NULL); pl->p_curtick = hpts->p_curtick; pl->p_prevtick = hpts->p_prevtick; pl->p_on_queue_cnt = hpts->p_on_queue_cnt; pl->ticknow = ticknow; pl->slot_req = slot; pl->p_nxt_slot = hpts->p_nxt_slot; pl->p_cur_slot = hpts->p_cur_slot; pl->p_hpts_sleep_time = hpts->p_hpts_sleep_time; pl->p_flags = (hpts->p_cpu & 0x7f); pl->p_flags <<= 7; pl->p_flags |= (hpts->p_num & 0x7f); pl->p_flags <<= 2; if (hpts->p_hpts_active) { pl->p_flags |= HPTS_HPTS_ACTIVE; } } #define tcp_hpts_log_it(a, b, c, d, e) __tcp_hpts_log_it(a, b, c, d, e, __LINE__) static void hpts_timeout_swi(void *arg) { struct tcp_hpts_entry *hpts; hpts = (struct tcp_hpts_entry *)arg; swi_sched(hpts->ie_cookie, 0); } static void hpts_timeout_dir(void *arg) { tcp_hpts_thread(arg); } static inline void hpts_sane_pace_remove(struct tcp_hpts_entry *hpts, struct inpcb *inp, struct hptsh *head, int clear) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_hpts_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_hpts == 0) { /* We are not on the hpts? */ panic("%s: hpts:%p inp:%p not on the hpts?", __FUNCTION__, hpts, inp); } if (TAILQ_EMPTY(head) && (hpts->p_on_queue_cnt != 0)) { /* We should not be empty with a queue count */ panic("%s hpts:%p hpts bucket empty but cnt:%d", __FUNCTION__, hpts, hpts->p_on_queue_cnt); } #endif TAILQ_REMOVE(head, inp, inp_hpts); hpts->p_on_queue_cnt--; if (hpts->p_on_queue_cnt < 0) { /* Count should not go negative .. */ #ifdef INVARIANTS panic("Hpts goes negative inp:%p hpts:%p", inp, hpts); #endif hpts->p_on_queue_cnt = 0; } if (clear) { inp->inp_hpts_request = 0; inp->inp_in_hpts = 0; } } static inline void hpts_sane_pace_insert(struct tcp_hpts_entry *hpts, struct inpcb *inp, struct hptsh *head, int line, int noref) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_hpts_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if ((noref == 0) && (inp->inp_in_hpts == 1)) { /* We are already on the hpts? */ panic("%s: hpts:%p inp:%p already on the hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_INSERT_TAIL(head, inp, inp_hpts); inp->inp_in_hpts = 1; hpts->p_on_queue_cnt++; if (noref == 0) { in_pcbref(inp); } } static inline void hpts_sane_input_remove(struct tcp_hpts_entry *hpts, struct inpcb *inp, int clear) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_input_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_input == 0) { /* We are not on the input hpts? */ panic("%s: hpts:%p inp:%p not on the input hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_REMOVE(&hpts->p_input, inp, inp_input); hpts->p_on_inqueue_cnt--; if (hpts->p_on_inqueue_cnt < 0) { #ifdef INVARIANTS panic("Hpts in goes negative inp:%p hpts:%p", inp, hpts); #endif hpts->p_on_inqueue_cnt = 0; } #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { /* We should not be empty with a queue count */ panic("%s hpts:%p in_hpts input empty but cnt:%d", __FUNCTION__, hpts, hpts->p_on_inqueue_cnt); } #endif if (clear) inp->inp_in_input = 0; } static inline void hpts_sane_input_insert(struct tcp_hpts_entry *hpts, struct inpcb *inp, int line) { #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx) == 0) { /* We don't own the mutex? */ panic("%s: hpts:%p inp:%p no hpts mutex", __FUNCTION__, hpts, inp); } if (hpts->p_cpu != inp->inp_input_cpu) { /* It is not the right cpu/mutex? */ panic("%s: hpts:%p inp:%p incorrect CPU", __FUNCTION__, hpts, inp); } if (inp->inp_in_input == 1) { /* We are already on the input hpts? */ panic("%s: hpts:%p inp:%p already on the input hpts?", __FUNCTION__, hpts, inp); } #endif TAILQ_INSERT_TAIL(&hpts->p_input, inp, inp_input); inp->inp_in_input = 1; hpts->p_on_inqueue_cnt++; in_pcbref(inp); } static int sysctl_tcp_hpts_log(SYSCTL_HANDLER_ARGS) { struct tcp_hpts_entry *hpts; size_t sz; int32_t logging_was, i; int32_t error = 0; /* * HACK: Turn off logging so no locks are required this really needs * a memory barrier :) */ logging_was = logging_on; logging_on = 0; if (!req->oldptr) { /* How much? */ sz = 0; for (i = 0; i < tcp_pace.rp_num_hptss; i++) { hpts = tcp_pace.rp_ent[i]; if (hpts->p_log == NULL) continue; sz += (sizeof(struct hpts_log) * hpts->p_logsize); } error = SYSCTL_OUT(req, 0, sz); } else { for (i = 0; i < tcp_pace.rp_num_hptss; i++) { hpts = tcp_pace.rp_ent[i]; if (hpts->p_log == NULL) continue; if (hpts->p_log_wrapped) sz = (sizeof(struct hpts_log) * hpts->p_logsize); else sz = (sizeof(struct hpts_log) * hpts->p_log_at); error = SYSCTL_OUT(req, hpts->p_log, sz); } } logging_on = logging_was; return error; } SYSCTL_PROC(_net_inet_tcp_hpts, OID_AUTO, log, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_tcp_hpts_log, "A", "tcp hptsi log"); static void tcp_wakehpts(struct tcp_hpts_entry *hpts) { HPTS_MTX_ASSERT(hpts); swi_sched(hpts->ie_cookie, 0); if (hpts->p_hpts_active == 2) { /* Rare sleeping on a ENOBUF */ wakeup_one(hpts); } } static void tcp_wakeinput(struct tcp_hpts_entry *hpts) { HPTS_MTX_ASSERT(hpts); swi_sched(hpts->ie_cookie, 0); if (hpts->p_hpts_active == 2) { /* Rare sleeping on a ENOBUF */ wakeup_one(hpts); } } struct tcp_hpts_entry * tcp_cur_hpts(struct inpcb *inp) { int32_t hpts_num; struct tcp_hpts_entry *hpts; hpts_num = inp->inp_hpts_cpu; hpts = tcp_pace.rp_ent[hpts_num]; return (hpts); } struct tcp_hpts_entry * tcp_hpts_lock(struct inpcb *inp) { struct tcp_hpts_entry *hpts; int32_t hpts_num; again: hpts_num = inp->inp_hpts_cpu; hpts = tcp_pace.rp_ent[hpts_num]; #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); if (hpts_num != inp->inp_hpts_cpu) { mtx_unlock(&hpts->p_mtx); goto again; } return (hpts); } struct tcp_hpts_entry * tcp_input_lock(struct inpcb *inp) { struct tcp_hpts_entry *hpts; int32_t hpts_num; again: hpts_num = inp->inp_input_cpu; hpts = tcp_pace.rp_ent[hpts_num]; #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); if (hpts_num != inp->inp_input_cpu) { mtx_unlock(&hpts->p_mtx); goto again; } return (hpts); } static void tcp_remove_hpts_ref(struct inpcb *inp, struct tcp_hpts_entry *hpts, int line) { int32_t add_freed; if (inp->inp_flags2 & INP_FREED) { /* * Need to play a special trick so that in_pcbrele_wlocked * does not return 1 when it really should have returned 0. */ add_freed = 1; inp->inp_flags2 &= ~INP_FREED; } else { add_freed = 0; } #ifndef INP_REF_DEBUG if (in_pcbrele_wlocked(inp)) { /* * This should not happen. We have the inpcb referred to by * the main socket (why we are called) and the hpts. It * should always return 0. */ panic("inpcb:%p release ret 1", inp); } #else if (__in_pcbrele_wlocked(inp, line)) { /* * This should not happen. We have the inpcb referred to by * the main socket (why we are called) and the hpts. It * should always return 0. */ panic("inpcb:%p release ret 1", inp); } #endif if (add_freed) { inp->inp_flags2 |= INP_FREED; } } static void tcp_hpts_remove_locked_output(struct tcp_hpts_entry *hpts, struct inpcb *inp, int32_t flags, int32_t line) { if (inp->inp_in_hpts) { hpts_sane_pace_remove(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], 1); tcp_remove_hpts_ref(inp, hpts, line); } } static void tcp_hpts_remove_locked_input(struct tcp_hpts_entry *hpts, struct inpcb *inp, int32_t flags, int32_t line) { HPTS_MTX_ASSERT(hpts); if (inp->inp_in_input) { hpts_sane_input_remove(hpts, inp, 1); tcp_remove_hpts_ref(inp, hpts, line); } } /* * Called normally with the INP_LOCKED but it * does not matter, the hpts lock is the key * but the lock order allows us to hold the * INP lock and then get the hpts lock. * * Valid values in the flags are * HPTS_REMOVE_OUTPUT - remove from the output of the hpts. * HPTS_REMOVE_INPUT - remove from the input of the hpts. * Note that you can or both values together and get two * actions. */ void __tcp_hpts_remove(struct inpcb *inp, int32_t flags, int32_t line) { struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); if (flags & HPTS_REMOVE_OUTPUT) { hpts = tcp_hpts_lock(inp); tcp_hpts_remove_locked_output(hpts, inp, flags, line); mtx_unlock(&hpts->p_mtx); } if (flags & HPTS_REMOVE_INPUT) { hpts = tcp_input_lock(inp); tcp_hpts_remove_locked_input(hpts, inp, flags, line); mtx_unlock(&hpts->p_mtx); } } static inline int hpts_tick(struct tcp_hpts_entry *hpts, int32_t plus) { return ((hpts->p_prevtick + plus) % NUM_OF_HPTSI_SLOTS); } static int tcp_queue_to_hpts_immediate_locked(struct inpcb *inp, struct tcp_hpts_entry *hpts, int32_t line, int32_t noref) { int32_t need_wake = 0; uint32_t ticknow = 0; HPTS_MTX_ASSERT(hpts); if (inp->inp_in_hpts == 0) { /* Ok we need to set it on the hpts in the current slot */ if (hpts->p_hpts_active == 0) { /* A sleeping hpts we want in next slot to run */ if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_INSERT_SLEEPER, 0, hpts_tick(hpts, 1)); } inp->inp_hptsslot = hpts_tick(hpts, 1); inp->inp_hpts_request = 0; if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_SLEEP_BEFORE, 1, ticknow); } need_wake = 1; } else if ((void *)inp == hpts->p_inp) { /* * We can't allow you to go into the same slot we * are in. We must put you out. */ inp->inp_hptsslot = hpts->p_nxt_slot; } else inp->inp_hptsslot = hpts->p_cur_slot; hpts_sane_pace_insert(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], line, noref); inp->inp_hpts_request = 0; if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_IMMEDIATE, 0, 0); } if (need_wake) { /* * Activate the hpts if it is sleeping and its * timeout is not 1. */ if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_WAKEUP_HPTS, 0, ticknow); } hpts->p_direct_wake = 1; tcp_wakehpts(hpts); } } return (need_wake); } int __tcp_queue_to_hpts_immediate(struct inpcb *inp, int32_t line) { int32_t ret; struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); ret = tcp_queue_to_hpts_immediate_locked(inp, hpts, line, 0); mtx_unlock(&hpts->p_mtx); return (ret); } static void tcp_hpts_insert_locked(struct tcp_hpts_entry *hpts, struct inpcb *inp, uint32_t slot, uint32_t cts, int32_t line, struct hpts_diag *diag, int32_t noref) { int32_t need_new_to = 0; int32_t need_wakeup = 0; uint32_t largest_slot; uint32_t ticknow = 0; uint32_t slot_calc; HPTS_MTX_ASSERT(hpts); if (diag) { memset(diag, 0, sizeof(struct hpts_diag)); diag->p_hpts_active = hpts->p_hpts_active; diag->p_nxt_slot = hpts->p_nxt_slot; diag->p_cur_slot = hpts->p_cur_slot; diag->slot_req = slot; } if ((inp->inp_in_hpts == 0) || noref) { inp->inp_hpts_request = slot; if (slot == 0) { /* Immediate */ tcp_queue_to_hpts_immediate_locked(inp, hpts, line, noref); return; } if (hpts->p_hpts_active) { /* * Its slot - 1 since nxt_slot is the next tick that * will go off since the hpts is awake */ if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_INSERT_NORMAL, slot, 0); } /* * We want to make sure that we don't place a inp in * the range of p_cur_slot <-> p_nxt_slot. If we * take from p_nxt_slot to the end, plus p_cur_slot * and then take away 2, we will know how many is * the max slots we can use. */ if (hpts->p_nxt_slot > hpts->p_cur_slot) { /* * Non-wrap case nxt_slot <-> cur_slot we * don't want to land in. So the diff gives * us what is taken away from the number of * slots. */ largest_slot = NUM_OF_HPTSI_SLOTS - (hpts->p_nxt_slot - hpts->p_cur_slot); } else if (hpts->p_nxt_slot == hpts->p_cur_slot) { largest_slot = NUM_OF_HPTSI_SLOTS - 2; } else { /* * Wrap case so the diff gives us the number * of slots that we can land in. */ largest_slot = hpts->p_cur_slot - hpts->p_nxt_slot; } /* * We take away two so we never have a problem (20 * usec's) out of 1024000 usecs */ largest_slot -= 2; if (inp->inp_hpts_request > largest_slot) { /* * Restrict max jump of slots and remember * leftover */ slot = largest_slot; inp->inp_hpts_request -= largest_slot; } else { /* This one will run when we hit it */ inp->inp_hpts_request = 0; } if (hpts->p_nxt_slot == hpts->p_cur_slot) slot_calc = (hpts->p_nxt_slot + slot) % NUM_OF_HPTSI_SLOTS; else slot_calc = (hpts->p_nxt_slot + slot - 1) % NUM_OF_HPTSI_SLOTS; if (slot_calc == hpts->p_cur_slot) { #ifdef INVARIANTS /* TSNH */ panic("Hpts:%p impossible slot calculation slot_calc:%u slot:%u largest:%u\n", hpts, slot_calc, slot, largest_slot); #endif if (slot_calc) slot_calc--; else slot_calc = NUM_OF_HPTSI_SLOTS - 1; } inp->inp_hptsslot = slot_calc; if (diag) { diag->inp_hptsslot = inp->inp_hptsslot; } } else { /* * The hpts is sleeping, we need to figure out where * it will wake up at and if we need to reschedule * its time-out. */ uint32_t have_slept, yet_to_sleep; uint32_t slot_now; struct timeval tv; ticknow = tcp_gethptstick(&tv); slot_now = ticknow % NUM_OF_HPTSI_SLOTS; /* * The user wants to be inserted at (slot_now + * slot) % NUM_OF_HPTSI_SLOTS, so lets set that up. */ largest_slot = NUM_OF_HPTSI_SLOTS - 2; if (inp->inp_hpts_request > largest_slot) { /* Adjust the residual in inp_hpts_request */ slot = largest_slot; inp->inp_hpts_request -= largest_slot; } else { /* No residual it all fits */ inp->inp_hpts_request = 0; } inp->inp_hptsslot = (slot_now + slot) % NUM_OF_HPTSI_SLOTS; if (diag) { diag->slot_now = slot_now; diag->inp_hptsslot = inp->inp_hptsslot; diag->p_on_min_sleep = hpts->p_on_min_sleep; } if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_INSERT_SLEEPER, slot, ticknow); } /* Now do we need to restart the hpts's timer? */ if (TSTMP_GT(ticknow, hpts->p_curtick)) have_slept = ticknow - hpts->p_curtick; else have_slept = 0; if (have_slept < hpts->p_hpts_sleep_time) { /* This should be what happens */ yet_to_sleep = hpts->p_hpts_sleep_time - have_slept; } else { /* We are over-due */ yet_to_sleep = 0; need_wakeup = 1; } if (diag) { diag->have_slept = have_slept; diag->yet_to_sleep = yet_to_sleep; diag->hpts_sleep_time = hpts->p_hpts_sleep_time; } if ((hpts->p_on_min_sleep == 0) && (yet_to_sleep > slot)) { /* * We need to reschedule the hptss time-out. */ hpts->p_hpts_sleep_time = slot; need_new_to = slot * HPTS_TICKS_PER_USEC; } } hpts_sane_pace_insert(hpts, inp, &hpts->p_hptss[inp->inp_hptsslot], line, noref); if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_INSERTED, slot, ticknow); } /* * Now how far is the hpts sleeping to? if active is 1, its * up and ticking we do nothing, otherwise we may need to * reschedule its callout if need_new_to is set from above. */ if (need_wakeup) { if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_RESCHEDULE, 1, 0); } hpts->p_direct_wake = 1; tcp_wakehpts(hpts); if (diag) { diag->need_new_to = 0; diag->co_ret = 0xffff0000; } } else if (need_new_to) { int32_t co_ret; struct timeval tv; sbintime_t sb; tv.tv_sec = 0; tv.tv_usec = 0; while (need_new_to > HPTS_USEC_IN_SEC) { tv.tv_sec++; need_new_to -= HPTS_USEC_IN_SEC; } tv.tv_usec = need_new_to; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { co_ret = callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { co_ret = callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } if (diag) { diag->need_new_to = need_new_to; diag->co_ret = co_ret; } } } else { #ifdef INVARIANTS panic("Hpts:%p tp:%p already on hpts and add?", hpts, inp); #endif } } uint32_t tcp_hpts_insert_diag(struct inpcb *inp, uint32_t slot, int32_t line, struct hpts_diag *diag){ struct tcp_hpts_entry *hpts; uint32_t slot_on, cts; struct timeval tv; /* * We now return the next-slot the hpts will be on, beyond its * current run (if up) or where it was when it stopped if it is * sleeping. */ INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); if (in_ts_percision) microuptime(&tv); else getmicrouptime(&tv); cts = tcp_tv_to_usectick(&tv); tcp_hpts_insert_locked(hpts, inp, slot, cts, line, diag, 0); slot_on = hpts->p_nxt_slot; mtx_unlock(&hpts->p_mtx); return (slot_on); } uint32_t __tcp_hpts_insert(struct inpcb *inp, uint32_t slot, int32_t line){ return (tcp_hpts_insert_diag(inp, slot, line, NULL)); } int __tcp_queue_to_input_locked(struct inpcb *inp, struct tcp_hpts_entry *hpts, int32_t line) { int32_t retval = 0; HPTS_MTX_ASSERT(hpts); if (inp->inp_in_input == 0) { /* Ok we need to set it on the hpts in the current slot */ hpts_sane_input_insert(hpts, inp, line); retval = 1; if (hpts->p_hpts_active == 0) { /* * Activate the hpts if it is sleeping. */ if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_WAKEUP_INPUT, 0, 0); } retval = 2; hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } } else if (hpts->p_hpts_active == 0) { retval = 4; hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } return (retval); } void tcp_queue_pkt_to_input(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, int32_t tlen, int32_t drop_hdrlen, uint8_t iptos) { /* Setup packet for input first */ INP_WLOCK_ASSERT(tp->t_inpcb); m->m_pkthdr.pace_thoff = (uint16_t) ((caddr_t)th - mtod(m, caddr_t)); m->m_pkthdr.pace_tlen = (uint16_t) tlen; m->m_pkthdr.pace_drphdrlen = drop_hdrlen; m->m_pkthdr.pace_tos = iptos; m->m_pkthdr.pace_lock = (curthread->td_epochnest != 0); if (tp->t_in_pkt == NULL) { tp->t_in_pkt = m; tp->t_tail_pkt = m; } else { tp->t_tail_pkt->m_nextpkt = m; tp->t_tail_pkt = m; } } int32_t __tcp_queue_to_input(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, int32_t tlen, int32_t drop_hdrlen, uint8_t iptos, int32_t line){ struct tcp_hpts_entry *hpts; int32_t ret; tcp_queue_pkt_to_input(tp, m, th, tlen, drop_hdrlen, iptos); hpts = tcp_input_lock(tp->t_inpcb); ret = __tcp_queue_to_input_locked(tp->t_inpcb, hpts, line); mtx_unlock(&hpts->p_mtx); return (ret); } void __tcp_set_inp_to_drop(struct inpcb *inp, uint16_t reason, int32_t line) { struct tcp_hpts_entry *hpts; struct tcpcb *tp; tp = intotcpcb(inp); hpts = tcp_input_lock(tp->t_inpcb); if (inp->inp_in_input == 0) { /* Ok we need to set it on the hpts in the current slot */ hpts_sane_input_insert(hpts, inp, line); if (hpts->p_hpts_active == 0) { /* * Activate the hpts if it is sleeping. */ hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } } else if (hpts->p_hpts_active == 0) { hpts->p_direct_wake = 1; tcp_wakeinput(hpts); } inp->inp_hpts_drop_reas = reason; mtx_unlock(&hpts->p_mtx); } static uint16_t hpts_random_cpu(struct inpcb *inp){ /* * No flow type set distribute the load randomly. */ uint16_t cpuid; uint32_t ran; /* * If one has been set use it i.e. we want both in and out on the * same hpts. */ if (inp->inp_input_cpu_set) { return (inp->inp_input_cpu); } else if (inp->inp_hpts_cpu_set) { return (inp->inp_hpts_cpu); } /* Nothing set use a random number */ ran = arc4random(); cpuid = (ran & 0xffff) % mp_ncpus; return (cpuid); } static uint16_t hpts_cpuid(struct inpcb *inp){ u_int cpuid; /* * If one has been set use it i.e. we want both in and out on the * same hpts. */ if (inp->inp_input_cpu_set) { return (inp->inp_input_cpu); } else if (inp->inp_hpts_cpu_set) { return (inp->inp_hpts_cpu); } /* If one is set the other must be the same */ #ifdef RSS cpuid = rss_hash2cpuid(inp->inp_flowid, inp->inp_flowtype); if (cpuid == NETISR_CPUID_NONE) return (hpts_random_cpu(inp)); else return (cpuid); #else /* * We don't have a flowid -> cpuid mapping, so cheat and just map * unknown cpuids to curcpu. Not the best, but apparently better * than defaulting to swi 0. */ if (inp->inp_flowtype != M_HASHTYPE_NONE) { cpuid = inp->inp_flowid % mp_ncpus; return (cpuid); } cpuid = hpts_random_cpu(inp); return (cpuid); #endif } /* * Do NOT try to optimize the processing of inp's * by first pulling off all the inp's into a temporary * list (e.g. TAILQ_CONCAT). If you do that the subtle * interactions of switching CPU's will kill because of * problems in the linked list manipulation. Basically * you would switch cpu's with the hpts mutex locked * but then while you were processing one of the inp's * some other one that you switch will get a new * packet on the different CPU. It will insert it * on the new hptss input list. Creating a temporary * link in the inp will not fix it either, since * the other hpts will be doing the same thing and * you will both end up using the temporary link. * * You will die in an ASSERT for tailq corruption if you * run INVARIANTS or you will die horribly without * INVARIANTS in some unknown way with a corrupt linked * list. */ static void tcp_input_data(struct tcp_hpts_entry *hpts, struct timeval *tv) { struct mbuf *m, *n; struct tcpcb *tp; struct inpcb *inp; uint16_t drop_reason; int16_t set_cpu; uint32_t did_prefetch = 0; int32_t ti_locked = TI_UNLOCKED; struct epoch_tracker et; HPTS_MTX_ASSERT(hpts); while ((inp = TAILQ_FIRST(&hpts->p_input)) != NULL) { HPTS_MTX_ASSERT(hpts); hpts_sane_input_remove(hpts, inp, 0); if (inp->inp_input_cpu_set == 0) { set_cpu = 1; } else { set_cpu = 0; } hpts->p_inp = inp; drop_reason = inp->inp_hpts_drop_reas; inp->inp_in_input = 0; mtx_unlock(&hpts->p_mtx); CURVNET_SET(inp->inp_vnet); if (drop_reason) { INP_INFO_RLOCK_ET(&V_tcbinfo, et); ti_locked = TI_RLOCKED; } else { ti_locked = TI_UNLOCKED; } INP_WLOCK(inp); if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) || (inp->inp_flags2 & INP_FREED)) { out: hpts->p_inp = NULL; if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); } if (in_pcbrele_wlocked(inp) == 0) { INP_WUNLOCK(inp); } ti_locked = TI_UNLOCKED; CURVNET_RESTORE(); mtx_lock(&hpts->p_mtx); continue; } tp = intotcpcb(inp); if ((tp == NULL) || (tp->t_inpcb == NULL)) { goto out; } if (drop_reason) { /* This tcb is being destroyed for drop_reason */ m = tp->t_in_pkt; if (m) n = m->m_nextpkt; else n = NULL; tp->t_in_pkt = NULL; while (m) { m_freem(m); m = n; if (m) n = m->m_nextpkt; } tp = tcp_drop(tp, drop_reason); INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); if (tp == NULL) { INP_WLOCK(inp); } if (in_pcbrele_wlocked(inp) == 0) INP_WUNLOCK(inp); CURVNET_RESTORE(); mtx_lock(&hpts->p_mtx); continue; } if (set_cpu) { /* * Setup so the next time we will move to the right * CPU. This should be a rare event. It will * sometimes happens when we are the client side * (usually not the server). Somehow tcp_output() * gets called before the tcp_do_segment() sets the * intial state. This means the r_cpu and r_hpts_cpu * is 0. We get on the hpts, and then tcp_input() * gets called setting up the r_cpu to the correct * value. The hpts goes off and sees the mis-match. * We simply correct it here and the CPU will switch * to the new hpts nextime the tcb gets added to the * the hpts (not this time) :-) */ tcp_set_hpts(inp); } m = tp->t_in_pkt; n = NULL; if (m != NULL && (m->m_pkthdr.pace_lock == TI_RLOCKED || tp->t_state != TCPS_ESTABLISHED)) { ti_locked = TI_RLOCKED; INP_INFO_RLOCK_ET(&V_tcbinfo, et); m = tp->t_in_pkt; } if (in_newts_every_tcb) { if (in_ts_percision) microuptime(tv); else getmicrouptime(tv); } if (tp->t_fb_ptr != NULL) { kern_prefetch(tp->t_fb_ptr, &did_prefetch); did_prefetch = 1; } /* Any input work to do, if so do it first */ if ((m != NULL) && (m == tp->t_in_pkt)) { struct tcphdr *th; int32_t tlen, drop_hdrlen, nxt_pkt; uint8_t iptos; n = m->m_nextpkt; tp->t_in_pkt = tp->t_tail_pkt = NULL; while (m) { th = (struct tcphdr *)(mtod(m, caddr_t)+m->m_pkthdr.pace_thoff); tlen = m->m_pkthdr.pace_tlen; drop_hdrlen = m->m_pkthdr.pace_drphdrlen; iptos = m->m_pkthdr.pace_tos; m->m_nextpkt = NULL; if (n) nxt_pkt = 1; else nxt_pkt = 0; inp->inp_input_calls = 1; if (tp->t_fb->tfb_tcp_hpts_do_segment) { /* Use the hpts specific do_segment */ (*tp->t_fb->tfb_tcp_hpts_do_segment) (m, th, inp->inp_socket, tp, drop_hdrlen, tlen, iptos, nxt_pkt, tv); } else { /* Use the default do_segment */ (*tp->t_fb->tfb_tcp_do_segment) (m, th, inp->inp_socket, tp, drop_hdrlen, tlen, iptos); } if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); /* * Do segment returns unlocked we need the * lock again but we also need some kasserts * here. */ - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); INP_UNLOCK_ASSERT(inp); m = n; if (m) n = m->m_nextpkt; if (m != NULL && m->m_pkthdr.pace_lock == TI_RLOCKED) { INP_INFO_RLOCK_ET(&V_tcbinfo, et); ti_locked = TI_RLOCKED; } else ti_locked = TI_UNLOCKED; INP_WLOCK(inp); /* * Since we have an opening here we must * re-check if the tcb went away while we * were getting the lock(s). */ if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) || (inp->inp_flags2 & INP_FREED)) { while (m) { m_freem(m); m = n; if (m) n = m->m_nextpkt; } goto out; } /* * Now that we hold the INP lock, check if * we need to upgrade our lock. */ if (ti_locked == TI_UNLOCKED && (tp->t_state != TCPS_ESTABLISHED)) { ti_locked = TI_RLOCKED; INP_INFO_RLOCK_ET(&V_tcbinfo, et); } } /** end while(m) */ } /** end if ((m != NULL) && (m == tp->t_in_pkt)) */ if (in_pcbrele_wlocked(inp) == 0) INP_WUNLOCK(inp); if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); INP_UNLOCK_ASSERT(inp); ti_locked = TI_UNLOCKED; mtx_lock(&hpts->p_mtx); hpts->p_inp = NULL; CURVNET_RESTORE(); } } static int tcp_hpts_est_run(struct tcp_hpts_entry *hpts) { int32_t ticks_to_run; if (hpts->p_prevtick && (SEQ_GT(hpts->p_curtick, hpts->p_prevtick))) { ticks_to_run = hpts->p_curtick - hpts->p_prevtick; if (ticks_to_run >= (NUM_OF_HPTSI_SLOTS - 1)) { ticks_to_run = NUM_OF_HPTSI_SLOTS - 2; } } else { if (hpts->p_prevtick == hpts->p_curtick) { /* This happens when we get woken up right away */ return (-1); } ticks_to_run = 1; } /* Set in where we will be when we catch up */ hpts->p_nxt_slot = (hpts->p_cur_slot + ticks_to_run) % NUM_OF_HPTSI_SLOTS; if (hpts->p_nxt_slot == hpts->p_cur_slot) { panic("Impossible math -- hpts:%p p_nxt_slot:%d p_cur_slot:%d ticks_to_run:%d", hpts, hpts->p_nxt_slot, hpts->p_cur_slot, ticks_to_run); } return (ticks_to_run); } static void tcp_hptsi(struct tcp_hpts_entry *hpts, struct timeval *ctick) { struct tcpcb *tp; struct inpcb *inp = NULL, *ninp; struct timeval tv; int32_t ticks_to_run, i, error, tick_now, interum_tick; int32_t paced_cnt = 0; int32_t did_prefetch = 0; int32_t prefetch_ninp = 0; int32_t prefetch_tp = 0; uint32_t cts; int16_t set_cpu; HPTS_MTX_ASSERT(hpts); hpts->p_curtick = tcp_tv_to_hptstick(ctick); cts = tcp_tv_to_usectick(ctick); memcpy(&tv, ctick, sizeof(struct timeval)); hpts->p_cur_slot = hpts_tick(hpts, 1); /* Figure out if we had missed ticks */ again: HPTS_MTX_ASSERT(hpts); ticks_to_run = tcp_hpts_est_run(hpts); if (!TAILQ_EMPTY(&hpts->p_input)) { tcp_input_data(hpts, &tv); } #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { panic("tp:%p in_hpts input empty but cnt:%d", hpts, hpts->p_on_inqueue_cnt); } #endif HPTS_MTX_ASSERT(hpts); /* Reset the ticks to run and time if we need too */ interum_tick = tcp_gethptstick(&tv); if (interum_tick != hpts->p_curtick) { /* Save off the new time we execute to */ *ctick = tv; hpts->p_curtick = interum_tick; cts = tcp_tv_to_usectick(&tv); hpts->p_cur_slot = hpts_tick(hpts, 1); ticks_to_run = tcp_hpts_est_run(hpts); } if (ticks_to_run == -1) { goto no_run; } if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_SETTORUN, ticks_to_run, 0); } if (hpts->p_on_queue_cnt == 0) { goto no_one; } HPTS_MTX_ASSERT(hpts); for (i = 0; i < ticks_to_run; i++) { /* * Calculate our delay, if there are no extra ticks there * was not any */ hpts->p_delayed_by = (ticks_to_run - (i + 1)) * HPTS_TICKS_PER_USEC; HPTS_MTX_ASSERT(hpts); while ((inp = TAILQ_FIRST(&hpts->p_hptss[hpts->p_cur_slot])) != NULL) { /* For debugging */ if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_HPTSI, ticks_to_run, i); } hpts->p_inp = inp; paced_cnt++; if (hpts->p_cur_slot != inp->inp_hptsslot) { panic("Hpts:%p inp:%p slot mis-aligned %u vs %u", hpts, inp, hpts->p_cur_slot, inp->inp_hptsslot); } /* Now pull it */ if (inp->inp_hpts_cpu_set == 0) { set_cpu = 1; } else { set_cpu = 0; } hpts_sane_pace_remove(hpts, inp, &hpts->p_hptss[hpts->p_cur_slot], 0); if ((ninp = TAILQ_FIRST(&hpts->p_hptss[hpts->p_cur_slot])) != NULL) { /* We prefetch the next inp if possible */ kern_prefetch(ninp, &prefetch_ninp); prefetch_ninp = 1; } if (inp->inp_hpts_request) { /* * This guy is deferred out further in time * then our wheel had on it. Push him back * on the wheel. */ int32_t remaining_slots; remaining_slots = ticks_to_run - (i + 1); if (inp->inp_hpts_request > remaining_slots) { /* * Keep INVARIANTS happy by clearing * the flag */ tcp_hpts_insert_locked(hpts, inp, inp->inp_hpts_request, cts, __LINE__, NULL, 1); hpts->p_inp = NULL; continue; } inp->inp_hpts_request = 0; } /* * We clear the hpts flag here after dealing with * remaining slots. This way anyone looking with the * TCB lock will see its on the hpts until just * before we unlock. */ inp->inp_in_hpts = 0; mtx_unlock(&hpts->p_mtx); INP_WLOCK(inp); if (in_pcbrele_wlocked(inp)) { mtx_lock(&hpts->p_mtx); if (logging_on) tcp_hpts_log_it(hpts, hpts->p_inp, HPTSLOG_INP_DONE, 0, 1); hpts->p_inp = NULL; continue; } if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { out_now: #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif INP_WUNLOCK(inp); mtx_lock(&hpts->p_mtx); if (logging_on) tcp_hpts_log_it(hpts, hpts->p_inp, HPTSLOG_INP_DONE, 0, 3); hpts->p_inp = NULL; continue; } tp = intotcpcb(inp); if ((tp == NULL) || (tp->t_inpcb == NULL)) { goto out_now; } if (set_cpu) { /* * Setup so the next time we will move to * the right CPU. This should be a rare * event. It will sometimes happens when we * are the client side (usually not the * server). Somehow tcp_output() gets called * before the tcp_do_segment() sets the * intial state. This means the r_cpu and * r_hpts_cpu is 0. We get on the hpts, and * then tcp_input() gets called setting up * the r_cpu to the correct value. The hpts * goes off and sees the mis-match. We * simply correct it here and the CPU will * switch to the new hpts nextime the tcb * gets added to the the hpts (not this one) * :-) */ tcp_set_hpts(inp); } if (out_newts_every_tcb) { struct timeval sv; if (out_ts_percision) microuptime(&sv); else getmicrouptime(&sv); cts = tcp_tv_to_usectick(&sv); } CURVNET_SET(inp->inp_vnet); /* * There is a hole here, we get the refcnt on the * inp so it will still be preserved but to make * sure we can get the INP we need to hold the p_mtx * above while we pull out the tp/inp, as long as * fini gets the lock first we are assured of having * a sane INP we can lock and test. */ #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx before tcp-output:%d", hpts, __LINE__); } #endif if (tp->t_fb_ptr != NULL) { kern_prefetch(tp->t_fb_ptr, &did_prefetch); did_prefetch = 1; } inp->inp_hpts_calls = 1; if (tp->t_fb->tfb_tcp_output_wtime != NULL) { error = (*tp->t_fb->tfb_tcp_output_wtime) (tp, &tv); } else { error = tp->t_fb->tfb_tcp_output(tp); } if (ninp && ninp->inp_ppcb) { /* * If we have a nxt inp, see if we can * prefetch its ppcb. Note this may seem * "risky" since we have no locks (other * than the previous inp) and there no * assurance that ninp was not pulled while * we were processing inp and freed. If this * occured it could mean that either: * * a) Its NULL (which is fine we won't go * here) b) Its valid (which is cool we * will prefetch it) c) The inp got * freed back to the slab which was * reallocated. Then the piece of memory was * re-used and something else (not an * address) is in inp_ppcb. If that occurs * we don't crash, but take a TLB shootdown * performance hit (same as if it was NULL * and we tried to pre-fetch it). * * Considering that the likelyhood of is * quite rare we will take a risk on doing * this. If performance drops after testing * we can always take this out. NB: the * kern_prefetch on amd64 actually has * protection against a bad address now via * the DMAP_() tests. This will prevent the * TLB hit, and instead if occurs just * cause us to load cache with a useless * address (to us). */ kern_prefetch(ninp->inp_ppcb, &prefetch_tp); prefetch_tp = 1; } INP_WUNLOCK(inp); INP_UNLOCK_ASSERT(inp); CURVNET_RESTORE(); #ifdef INVARIANTS if (mtx_owned(&hpts->p_mtx)) { panic("Hpts:%p owns mtx prior-to lock line:%d", hpts, __LINE__); } #endif mtx_lock(&hpts->p_mtx); if (logging_on) tcp_hpts_log_it(hpts, hpts->p_inp, HPTSLOG_INP_DONE, 0, 4); hpts->p_inp = NULL; } HPTS_MTX_ASSERT(hpts); hpts->p_inp = NULL; hpts->p_cur_slot++; if (hpts->p_cur_slot >= NUM_OF_HPTSI_SLOTS) { hpts->p_cur_slot = 0; } } no_one: HPTS_MTX_ASSERT(hpts); hpts->p_prevtick = hpts->p_curtick; hpts->p_delayed_by = 0; /* * Check to see if we took an excess amount of time and need to run * more ticks (if we did not hit eno-bufs). */ /* Re-run any input that may be there */ (void)tcp_gethptstick(&tv); if (!TAILQ_EMPTY(&hpts->p_input)) { tcp_input_data(hpts, &tv); } #ifdef INVARIANTS if (TAILQ_EMPTY(&hpts->p_input) && (hpts->p_on_inqueue_cnt != 0)) { panic("tp:%p in_hpts input empty but cnt:%d", hpts, hpts->p_on_inqueue_cnt); } #endif tick_now = tcp_gethptstick(&tv); if (SEQ_GT(tick_now, hpts->p_prevtick)) { struct timeval res; /* Did we really spend a full tick or more in here? */ timersub(&tv, ctick, &res); if (res.tv_sec || (res.tv_usec >= HPTS_TICKS_PER_USEC)) { counter_u64_add(hpts_loops, 1); if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_TOLONG, (uint32_t) res.tv_usec, tick_now); } *ctick = res; hpts->p_curtick = tick_now; goto again; } } no_run: { uint32_t t = 0, i, fnd = 0; if (hpts->p_on_queue_cnt) { /* * Find next slot that is occupied and use that to * be the sleep time. */ for (i = 1, t = hpts->p_nxt_slot; i < NUM_OF_HPTSI_SLOTS; i++) { if (TAILQ_EMPTY(&hpts->p_hptss[t]) == 0) { fnd = 1; break; } t = (t + 1) % NUM_OF_HPTSI_SLOTS; } if (fnd) { hpts->p_hpts_sleep_time = i; } else { counter_u64_add(back_tosleep, 1); #ifdef INVARIANTS panic("Hpts:%p cnt:%d but non found", hpts, hpts->p_on_queue_cnt); #endif hpts->p_on_queue_cnt = 0; goto non_found; } t++; } else { /* No one on the wheel sleep for all but 2 slots */ non_found: if (hpts_sleep_max == 0) hpts_sleep_max = 1; hpts->p_hpts_sleep_time = min((NUM_OF_HPTSI_SLOTS - 2), hpts_sleep_max); t = 0; } if (logging_on) { tcp_hpts_log_it(hpts, inp, HPTSLOG_SLEEPSET, t, (hpts->p_hpts_sleep_time * HPTS_TICKS_PER_USEC)); } } } void __tcp_set_hpts(struct inpcb *inp, int32_t line) { struct tcp_hpts_entry *hpts; INP_WLOCK_ASSERT(inp); hpts = tcp_hpts_lock(inp); if ((inp->inp_in_hpts == 0) && (inp->inp_hpts_cpu_set == 0)) { inp->inp_hpts_cpu = hpts_cpuid(inp); inp->inp_hpts_cpu_set = 1; } mtx_unlock(&hpts->p_mtx); hpts = tcp_input_lock(inp); if ((inp->inp_input_cpu_set == 0) && (inp->inp_in_input == 0)) { inp->inp_input_cpu = hpts_cpuid(inp); inp->inp_input_cpu_set = 1; } mtx_unlock(&hpts->p_mtx); } uint16_t tcp_hpts_delayedby(struct inpcb *inp){ return (tcp_pace.rp_ent[inp->inp_hpts_cpu]->p_delayed_by); } static void tcp_hpts_thread(void *ctx) { struct tcp_hpts_entry *hpts; struct timeval tv; sbintime_t sb; hpts = (struct tcp_hpts_entry *)ctx; mtx_lock(&hpts->p_mtx); if (hpts->p_direct_wake) { /* Signaled by input */ if (logging_on) tcp_hpts_log_it(hpts, NULL, HPTSLOG_AWAKE, 1, 1); callout_stop(&hpts->co); } else { /* Timed out */ if (callout_pending(&hpts->co) || !callout_active(&hpts->co)) { if (logging_on) tcp_hpts_log_it(hpts, NULL, HPTSLOG_AWAKE, 2, 2); mtx_unlock(&hpts->p_mtx); return; } callout_deactivate(&hpts->co); if (logging_on) tcp_hpts_log_it(hpts, NULL, HPTSLOG_AWAKE, 3, 3); } hpts->p_hpts_active = 1; (void)tcp_gethptstick(&tv); tcp_hptsi(hpts, &tv); HPTS_MTX_ASSERT(hpts); tv.tv_sec = 0; tv.tv_usec = hpts->p_hpts_sleep_time * HPTS_TICKS_PER_USEC; if (tcp_min_hptsi_time && (tv.tv_usec < tcp_min_hptsi_time)) { tv.tv_usec = tcp_min_hptsi_time; hpts->p_on_min_sleep = 1; } else { /* Clear the min sleep flag */ hpts->p_on_min_sleep = 0; } hpts->p_hpts_active = 0; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } hpts->p_direct_wake = 0; mtx_unlock(&hpts->p_mtx); } #undef timersub static void tcp_init_hptsi(void *st) { int32_t i, j, error, bound = 0, created = 0; size_t sz, asz; struct timeval tv; sbintime_t sb; struct tcp_hpts_entry *hpts; char unit[16]; uint32_t ncpus = mp_ncpus ? mp_ncpus : MAXCPU; tcp_pace.rp_proc = NULL; tcp_pace.rp_num_hptss = ncpus; hpts_loops = counter_u64_alloc(M_WAITOK); back_tosleep = counter_u64_alloc(M_WAITOK); sz = (tcp_pace.rp_num_hptss * sizeof(struct tcp_hpts_entry *)); tcp_pace.rp_ent = malloc(sz, M_TCPHPTS, M_WAITOK | M_ZERO); asz = sizeof(struct hptsh) * NUM_OF_HPTSI_SLOTS; for (i = 0; i < tcp_pace.rp_num_hptss; i++) { tcp_pace.rp_ent[i] = malloc(sizeof(struct tcp_hpts_entry), M_TCPHPTS, M_WAITOK | M_ZERO); tcp_pace.rp_ent[i]->p_hptss = malloc(asz, M_TCPHPTS, M_WAITOK); hpts = tcp_pace.rp_ent[i]; /* * Init all the hpts structures that are not specifically * zero'd by the allocations. Also lets attach them to the * appropriate sysctl block as well. */ mtx_init(&hpts->p_mtx, "tcp_hpts_lck", "hpts", MTX_DEF | MTX_DUPOK); TAILQ_INIT(&hpts->p_input); for (j = 0; j < NUM_OF_HPTSI_SLOTS; j++) { TAILQ_INIT(&hpts->p_hptss[j]); } sysctl_ctx_init(&hpts->hpts_ctx); sprintf(unit, "%d", i); hpts->hpts_root = SYSCTL_ADD_NODE(&hpts->hpts_ctx, SYSCTL_STATIC_CHILDREN(_net_inet_tcp_hpts), OID_AUTO, unit, CTLFLAG_RW, 0, ""); SYSCTL_ADD_INT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "in_qcnt", CTLFLAG_RD, &hpts->p_on_inqueue_cnt, 0, "Count TCB's awaiting input processing"); SYSCTL_ADD_INT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "out_qcnt", CTLFLAG_RD, &hpts->p_on_queue_cnt, 0, "Count TCB's awaiting output processing"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "active", CTLFLAG_RD, &hpts->p_hpts_active, 0, "Is the hpts active"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "curslot", CTLFLAG_RD, &hpts->p_cur_slot, 0, "What the current slot is if active"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "curtick", CTLFLAG_RD, &hpts->p_curtick, 0, "What the current tick on if active"); SYSCTL_ADD_UINT(&hpts->hpts_ctx, SYSCTL_CHILDREN(hpts->hpts_root), OID_AUTO, "logsize", CTLFLAG_RD, &hpts->p_logsize, 0, "Hpts logging buffer size"); hpts->p_hpts_sleep_time = NUM_OF_HPTSI_SLOTS - 2; hpts->p_num = i; hpts->p_prevtick = hpts->p_curtick = tcp_gethptstick(&tv); hpts->p_prevtick -= 1; hpts->p_prevtick %= NUM_OF_HPTSI_SLOTS; hpts->p_cpu = 0xffff; hpts->p_nxt_slot = 1; hpts->p_logsize = tcp_hpts_logging_size; if (hpts->p_logsize) { sz = (sizeof(struct hpts_log) * hpts->p_logsize); hpts->p_log = malloc(sz, M_TCPHPTS, M_WAITOK | M_ZERO); } callout_init(&hpts->co, 1); } /* * Now lets start ithreads to handle the hptss. */ CPU_FOREACH(i) { hpts = tcp_pace.rp_ent[i]; hpts->p_cpu = i; error = swi_add(&hpts->ie, "hpts", tcp_hpts_thread, (void *)hpts, SWI_NET, INTR_MPSAFE, &hpts->ie_cookie); if (error) { panic("Can't add hpts:%p i:%d err:%d", hpts, i, error); } created++; if (tcp_bind_threads) { if (intr_event_bind(hpts->ie, i) == 0) bound++; } tv.tv_sec = 0; tv.tv_usec = hpts->p_hpts_sleep_time * HPTS_TICKS_PER_USEC; sb = tvtosbt(tv); if (tcp_hpts_callout_skip_swi == 0) { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_swi, hpts, hpts->p_cpu, (C_DIRECT_EXEC | C_PREL(tcp_hpts_precision))); } else { callout_reset_sbt_on(&hpts->co, sb, 0, hpts_timeout_dir, hpts, hpts->p_cpu, C_PREL(tcp_hpts_precision)); } } printf("TCP Hpts created %d swi interrupt thread and bound %d\n", created, bound); return; } SYSINIT(tcphptsi, SI_SUB_KTHREAD_IDLE, SI_ORDER_ANY, tcp_init_hptsi, NULL); MODULE_VERSION(tcphpts, 1); Index: head/sys/netinet/tcp_input.c =================================================================== --- head/sys/netinet/tcp_input.c (revision 339038) +++ head/sys/netinet/tcp_input.c (revision 339039) @@ -1,3820 +1,3820 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, * James Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include #include #ifdef TCP_HHOOK #include #endif #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #include #ifdef TCPDEBUG #include #endif /* TCPDEBUG */ #ifdef TCP_OFFLOAD #include #endif #include #include #include const int tcprexmtthresh = 3; int tcp_log_in_vain = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); VNET_DEFINE(int, blackhole) = 0; #define V_blackhole VNET(blackhole) SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole), 0, "Do not send RST on segments to closed ports"); VNET_DEFINE(int, tcp_delack_enabled) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_delack_enabled), 0, "Delay ACK to try and piggyback it onto a data packet"); VNET_DEFINE(int, drop_synfin) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(drop_synfin), 0, "Drop TCP packets with SYN+FIN set"); VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc6675_pipe), 0, "Use calculated pipe/in-flight bytes per RFC 6675"); VNET_DEFINE(int, tcp_do_rfc3042) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3042), 0, "Enable RFC 3042 (Limited Transmit)"); VNET_DEFINE(int, tcp_do_rfc3390) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3390), 0, "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); VNET_DEFINE(int, tcp_initcwnd_segments) = 10; SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, "Slow-start flight size (initial congestion window) in number of segments"); VNET_DEFINE(int, tcp_do_rfc3465) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3465), 0, "Enable RFC 3465 (Appropriate Byte Counting)"); VNET_DEFINE(int, tcp_abc_l_var) = 2; SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_abc_l_var), 2, "Cap the max cwnd increment during slow-start to this number of segments"); static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); VNET_DEFINE(int, tcp_do_ecn) = 2; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_ecn), 0, "TCP ECN support"); VNET_DEFINE(int, tcp_ecn_maxretries) = 1; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ecn_maxretries), 0, "Max retries before giving up on ECN"); VNET_DEFINE(int, tcp_insecure_syn) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_syn), 0, "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); VNET_DEFINE(int, tcp_insecure_rst) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_rst), 0, "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); VNET_DEFINE(int, tcp_recvspace) = 1024*64; #define V_tcp_recvspace VNET(tcp_recvspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autorcvbuf), 0, "Enable automatic receive buffer sizing"); VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_inc), 0, "Incrementor step size of automatic receive buffer"); VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_max), 0, "Max size of automatic receive buffer"); VNET_DEFINE(struct inpcbhead, tcb); #define tcb6 tcb /* for KAME src sync over BSD*'s */ VNET_DEFINE(struct inpcbinfo, tcbinfo); /* * TCP statistics are stored in an array of counter(9)s, which size matches * size of struct tcpstat. TCP running connection count is a regular array. */ VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, "TCP connection counts by TCP state"); static void tcp_vnet_init(const void *unused) { COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); } VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, tcp_vnet_init, NULL); #ifdef VIMAGE static void tcp_vnet_uninit(const void *unused) { COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); VNET_PCPUSTAT_FREE(tcpstat); } VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, tcp_vnet_uninit, NULL); #endif /* VIMAGE */ /* * Kernel module interface for updating tcpstat. The argument is an index * into tcpstat treated as an array. */ void kmod_tcpstat_inc(int statnum) { counter_u64_add(VNET(tcpstat)[statnum], 1); } #ifdef TCP_HHOOK /* * Wrapper for the TCP established input helper hook. */ void hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, tp->osd); } } #endif /* * CC wrapper hook functions */ void cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, uint16_t type) { INP_WLOCK_ASSERT(tp->t_inpcb); tp->ccv->nsegs = nsegs; tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); if (tp->snd_cwnd <= tp->snd_wnd) tp->ccv->flags |= CCF_CWND_LIMITED; else tp->ccv->flags &= ~CCF_CWND_LIMITED; if (type == CC_ACK) { if (tp->snd_cwnd > tp->snd_ssthresh) { tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); if (tp->t_bytes_acked >= tp->snd_cwnd) { tp->t_bytes_acked -= tp->snd_cwnd; tp->ccv->flags |= CCF_ABC_SENTAWND; } } else { tp->ccv->flags &= ~CCF_ABC_SENTAWND; tp->t_bytes_acked = 0; } } if (CC_ALGO(tp)->ack_received != NULL) { /* XXXLAS: Find a way to live without this */ tp->ccv->curack = th->th_ack; CC_ALGO(tp)->ack_received(tp->ccv, type); } } void cc_conn_init(struct tcpcb *tp) { struct hc_metrics_lite metrics; struct inpcb *inp = tp->t_inpcb; u_int maxseg; int rtt; INP_WLOCK_ASSERT(tp->t_inpcb); tcp_hc_get(&inp->inp_inc, &metrics); maxseg = tcp_maxseg(tp); if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { tp->t_srtt = rtt; tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; TCPSTAT_INC(tcps_usedrtt); if (metrics.rmx_rttvar) { tp->t_rttvar = metrics.rmx_rttvar; TCPSTAT_INC(tcps_usedrttvar); } else { /* default variation is +- 1 rtt */ tp->t_rttvar = tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, tp->t_rttmin, TCPTV_REXMTMAX); } if (metrics.rmx_ssthresh) { /* * There's some sort of gateway or interface * buffer limit on the path. Use this to set * the slow start threshold, but set the * threshold to no less than 2*mss. */ tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); TCPSTAT_INC(tcps_usedssthresh); } /* * Set the initial slow-start flight size. * * RFC5681 Section 3.1 specifies the default conservative values. * RFC3390 specifies slightly more aggressive values. * RFC6928 increases it to ten segments. * Support for user specified value for initial flight size. * * If a SYN or SYN/ACK was lost and retransmitted, we have to * reduce the initial CWND to one segment as congestion is likely * requiring us to be cautious. */ if (tp->snd_cwnd == 1) tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ else if (V_tcp_initcwnd_segments) tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); else if (V_tcp_do_rfc3390) tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); else { /* Per RFC5681 Section 3.1 */ if (maxseg > 2190) tp->snd_cwnd = 2 * maxseg; else if (maxseg > 1095) tp->snd_cwnd = 3 * maxseg; else tp->snd_cwnd = 4 * maxseg; } if (CC_ALGO(tp)->conn_init != NULL) CC_ALGO(tp)->conn_init(tp->ccv); } void inline cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) { u_int maxseg; INP_WLOCK_ASSERT(tp->t_inpcb); switch(type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(tp->t_flags)) { tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_ECN_PERMIT) tp->t_flags |= TF_ECN_SND_CWR; } break; case CC_ECN: if (!IN_CONGRECOVERY(tp->t_flags)) { TCPSTAT_INC(tcps_ecn_rcwnd); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_ECN_PERMIT) tp->t_flags |= TF_ECN_SND_CWR; } break; case CC_RTO: maxseg = tcp_maxseg(tp); tp->t_dupacks = 0; tp->t_bytes_acked = 0; EXIT_RECOVERY(tp->t_flags); tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / maxseg) * maxseg; tp->snd_cwnd = maxseg; break; case CC_RTO_ERR: TCPSTAT_INC(tcps_sndrexmitbad); /* RTO was unnecessary, so reset everything. */ tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; tp->snd_recover = tp->snd_recover_prev; if (tp->t_flags & TF_WASFRECOVERY) ENTER_FASTRECOVERY(tp->t_flags); if (tp->t_flags & TF_WASCRECOVERY) ENTER_CONGRECOVERY(tp->t_flags); tp->snd_nxt = tp->snd_max; tp->t_flags &= ~TF_PREVVALID; tp->t_badrxtwin = 0; break; } if (CC_ALGO(tp)->cong_signal != NULL) { if (th != NULL) tp->ccv->curack = th->th_ack; CC_ALGO(tp)->cong_signal(tp->ccv, type); } } void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) { INP_WLOCK_ASSERT(tp->t_inpcb); /* XXXLAS: KASSERT that we're in recovery? */ if (CC_ALGO(tp)->post_recovery != NULL) { tp->ccv->curack = th->th_ack; CC_ALGO(tp)->post_recovery(tp->ccv); } /* XXXLAS: EXIT_RECOVERY ? */ tp->t_bytes_acked = 0; } /* * Indicate whether this ack should be delayed. We can delay the ack if * following conditions are met: * - There is no delayed ack timer in progress. * - Our last ack wasn't a 0-sized window. We never want to delay * the ack that opens up a 0-sized window. * - LRO wasn't used for this segment. We make sure by checking that the * segment size is not larger than the MSS. */ #define DELAY_ACK(tp, tlen) \ ((!tcp_timer_active(tp, TT_DELACK) && \ (tp->t_flags & TF_RXWIN0SENT) == 0) && \ (tlen <= tp->t_maxseg) && \ (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) static void inline cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) { INP_WLOCK_ASSERT(tp->t_inpcb); if (CC_ALGO(tp)->ecnpkt_handler != NULL) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->ccv->flags |= CCF_IPHDR_CE; break; case IPTOS_ECN_ECT0: tp->ccv->flags &= ~CCF_IPHDR_CE; break; case IPTOS_ECN_ECT1: tp->ccv->flags &= ~CCF_IPHDR_CE; break; } if (th->th_flags & TH_CWR) tp->ccv->flags |= CCF_TCPHDR_CWR; else tp->ccv->flags &= ~CCF_TCPHDR_CWR; if (tp->t_flags & TF_DELACK) tp->ccv->flags |= CCF_DELACK; else tp->ccv->flags &= ~CCF_DELACK; CC_ALGO(tp)->ecnpkt_handler(tp->ccv); if (tp->ccv->flags & CCF_ACKNOW) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } } /* * TCP input handling is split into multiple parts: * tcp6_input is a thin wrapper around tcp_input for the extended * ip6_protox[] call format in ip6_input * tcp_input handles primary segment validation, inpcb lookup and * SYN processing on listen sockets * tcp_do_segment processes the ACK and text of the segment for * establishing, established and closing connections */ #ifdef INET6 int tcp6_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct in6_ifaddr *ia6; struct ip6_hdr *ip6; IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); /* * draft-itojun-ipv6-tcp-to-anycast * better place to put this in? */ ip6 = mtod(m, struct ip6_hdr *); ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { struct ip6_hdr *ip6; ifa_free(&ia6->ia_ifa); ip6 = mtod(m, struct ip6_hdr *); icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); return (IPPROTO_DONE); } if (ia6) ifa_free(&ia6->ia_ifa); return (tcp_input(mp, offp, proto)); } #endif /* INET6 */ int tcp_input(struct mbuf **mp, int *offp, int proto) { struct mbuf *m = *mp; struct tcphdr *th = NULL; struct ip *ip = NULL; struct inpcb *inp = NULL; struct tcpcb *tp = NULL; struct socket *so = NULL; u_char *optp = NULL; int off0; int optlen = 0; #ifdef INET int len; #endif int tlen = 0, off; int drop_hdrlen; int thflags; int rstreason = 0; /* For badport_bandlim accounting purposes */ uint8_t iptos; struct m_tag *fwd_tag = NULL; struct epoch_tracker et; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6; #else const void *ip6 = NULL; #endif /* INET6 */ struct tcpopt to; /* options in this segment */ char *s = NULL; /* address and port logging */ int ti_locked; #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif #ifdef INET6 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; #endif off0 = *offp; m = *mp; *mp = NULL; to.to_flags = 0; TCPSTAT_INC(tcps_rcvtotal); #ifdef INET6 if (isipv6) { /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); if (m == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in6_cksum_pseudo(ip6, tlen, IPPROTO_TCP, m->m_pkthdr.csum_data); th->th_sum ^= 0xffff; } else th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } /* * Be proactive about unspecified IPv6 address in source. * As we use all-zero to indicate unbounded/unconnected pcb, * unspecified IPv6 address can be used to confuse us. * * Note that packets with unspecified IPv6 destination is * already dropped in ip6_input. */ if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { /* XXX stat */ goto drop; } iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { /* * Get IP and TCP header together in first mbuf. * Note: IP leaves IP header in first mbuf. */ if (off0 > sizeof (struct ip)) { ip_stripoptions(m); off0 = sizeof(struct ip); } if (m->m_len < sizeof (struct tcpiphdr)) { if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); tlen = ntohs(ip->ip_len) - off0; iptos = ip->ip_tos; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + tlen + IPPROTO_TCP)); th->th_sum ^= 0xffff; } else { struct ipovly *ipov = (struct ipovly *)ip; /* * Checksum extended TCP header and data. */ len = off0 + tlen; bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = htons(tlen); th->th_sum = in_cksum(m, len); /* Reset length for SDT probes. */ ip->ip_len = htons(len); /* Reset TOS bits */ ip->ip_tos = iptos; /* Re-initialization for later version check */ ip->ip_v = IPVERSION; ip->ip_hl = off0 >> 2; } if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } } #endif /* INET */ /* * Check that TCP offset makes sense, * pull out TCP options and adjust length. XXX */ off = th->th_off << 2; if (off < sizeof (struct tcphdr) || off > tlen) { TCPSTAT_INC(tcps_rcvbadoff); goto drop; } tlen -= off; /* tlen is used instead of ti->ti_len */ if (off > sizeof (struct tcphdr)) { #ifdef INET6 if (isipv6) { IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (m->m_len < sizeof(struct ip) + off) { if ((m = m_pullup(m, sizeof (struct ip) + off)) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); } } #endif optlen = off - sizeof (struct tcphdr); optp = (u_char *)(th + 1); } thflags = th->th_flags; /* * Convert TCP protocol specific fields to host format. */ tcp_fields_to_host(th); /* * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. */ drop_hdrlen = off0 + off; /* * Locate pcb for segment; if we're likely to add or remove a * connection then first acquire pcbinfo lock. There are three cases * where we might discover later we need a write lock despite the * flags: ACKs moving a connection out of the syncache, ACKs for a * connection in TIMEWAIT and SYNs not targeting a listening socket. */ if ((thflags & (TH_FIN | TH_RST)) != 0) { INP_INFO_RLOCK_ET(&V_tcbinfo, et); ti_locked = TI_RLOCKED; } else ti_locked = TI_UNLOCKED; /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ( #ifdef INET6 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) #ifdef INET || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) #endif #endif #if defined(INET) && !defined(INET6) (m->m_flags & M_IP_NEXTHOP) #endif ) fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); findpcb: #ifdef INVARIANTS if (ti_locked == TI_RLOCKED) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } else { - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); } #endif #ifdef INET6 if (isipv6 && fwd_tag != NULL) { struct sockaddr_in6 *next_hop6; next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &next_hop6->sin6_addr, next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); } } else if (isipv6) { inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET if (fwd_tag != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag+1); /* * Transparently forwarded. Pretend to be the destination. * already got one like this? */ inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(&V_tcbinfo, ip->ip_src, th->th_sport, next_hop->sin_addr, next_hop->sin_port ? ntohs(next_hop->sin_port) : th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); } } else inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); #endif /* INET */ /* * If the INPCB does not exist then all data in the incoming * segment is discarded and an appropriate RST is sent back. * XXX MRT Send RST using which routing table? */ if (inp == NULL) { /* * Log communication attempts to ports that are not * in use. */ if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || tcp_log_in_vain == 2) { if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) log(LOG_INFO, "%s; %s: Connection attempt " "to closed port\n", s, __func__); } /* * When blackholing do not respond with a RST but * completely ignore the segment and drop it. */ if ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole == 2) goto dropunlock; rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } INP_WLOCK_ASSERT(inp); /* * While waiting for inp lock during the lookup, another thread * can have dropped the inpcb, in which case we need to loop back * and try to find a new inpcb to deliver to. */ if (inp->inp_flags & INP_DROPPED) { INP_WUNLOCK(inp); inp = NULL; goto findpcb; } if ((inp->inp_flowtype == M_HASHTYPE_NONE) && (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && ((inp->inp_socket == NULL) || (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); } #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6) && IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { goto dropunlock; } #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4) && IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { goto dropunlock; } #endif /* INET */ #endif /* IPSEC */ /* * Check the minimum TTL for socket. */ if (inp->inp_ip_minttl != 0) { #ifdef INET6 if (isipv6) { if (inp->inp_ip_minttl > ip6->ip6_hlim) goto dropunlock; } else #endif if (inp->inp_ip_minttl > ip->ip_ttl) goto dropunlock; } /* * A previous connection in TIMEWAIT state is supposed to catch stray * or duplicate segments arriving late. If this segment was a * legitimate new connection attempt, the old INPCB gets removed and * we can try again to find a listening socket. * * At this point, due to earlier optimism, we may hold only an inpcb * lock, and not the inpcbinfo write lock. If so, we need to try to * acquire it, or if that fails, acquire a reference on the inpcb, * drop all locks, acquire a global write lock, and then re-acquire * the inpcb lock. We may at that point discover that another thread * has tried to free the inpcb, in which case we need to loop back * and try to find a new inpcb to deliver to. * * XXXRW: It may be time to rethink timewait locking. */ if (inp->inp_flags & INP_TIMEWAIT) { if (ti_locked == TI_UNLOCKED) { INP_INFO_RLOCK_ET(&V_tcbinfo, et); ti_locked = TI_RLOCKED; } INP_INFO_RLOCK_ASSERT(&V_tcbinfo); if (thflags & TH_SYN) tcp_dooptions(&to, optp, optlen, TO_SYN); /* * NB: tcp_twcheck unlocks the INP and frees the mbuf. */ if (tcp_twcheck(inp, &to, th, m, tlen)) goto findpcb; INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); return (IPPROTO_DONE); } /* * The TCPCB may no longer exist if the connection is winding * down or it is in the CLOSED state. Either way we drop the * segment and send an appropriate response. */ tp = intotcpcb(inp); if (tp == NULL || tp->t_state == TCPS_CLOSED) { rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) { tcp_offload_input(tp, m); m = NULL; /* consumed by the TOE driver */ goto dropunlock; } #endif /* * We've identified a valid inpcb, but it could be that we need an * inpcbinfo write lock but don't hold it. In this case, attempt to * acquire using the same strategy as the TIMEWAIT case above. If we * relock, we have to jump back to 'relocked' as the connection might * now be in TIMEWAIT. */ #ifdef INVARIANTS if ((thflags & (TH_FIN | TH_RST)) != 0) INP_INFO_RLOCK_ASSERT(&V_tcbinfo); #endif if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && !IS_FASTOPEN(tp->t_flags)))) { if (ti_locked == TI_UNLOCKED) { INP_INFO_RLOCK_ET(&V_tcbinfo, et); ti_locked = TI_RLOCKED; } INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } #ifdef MAC INP_WLOCK_ASSERT(inp); if (mac_inpcb_check_deliver(inp, m)) goto dropunlock; #endif so = inp->inp_socket; KASSERT(so != NULL, ("%s: so == NULL", __func__)); #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) { ostate = tp->t_state; #ifdef INET6 if (isipv6) { bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); } else #endif bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); tcp_savetcp = *th; } #endif /* TCPDEBUG */ /* * When the socket is accepting connections (the INPCB is in LISTEN * state) we look into the SYN cache if this is a new connection * attempt or the completion of a previous one. */ KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), ("%s: so accepting but tp %p not listening", __func__, tp)); if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { struct in_conninfo inc; bzero(&inc, sizeof(inc)); #ifdef INET6 if (isipv6) { inc.inc_flags |= INC_ISIPV6; if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) inc.inc_flags |= INC_IPV6MINMTU; inc.inc6_faddr = ip6->ip6_src; inc.inc6_laddr = ip6->ip6_dst; } else #endif { inc.inc_faddr = ip->ip_src; inc.inc_laddr = ip->ip_dst; } inc.inc_fport = th->th_sport; inc.inc_lport = th->th_dport; inc.inc_fibnum = so->so_fibnum; /* * Check for an existing connection attempt in syncache if * the flag is only ACK. A successful lookup creates a new * socket appended to the listen queue in SYN_RECEIVED state. */ if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* * Parse the TCP options here because * syncookies need access to the reflected * timestamp. */ tcp_dooptions(&to, optp, optlen, 0); /* * NB: syncache_expand() doesn't unlock * inp and tcpinfo locks. */ rstreason = syncache_expand(&inc, &to, th, &so, m); if (rstreason < 0) { /* * A failing TCP MD5 signature comparison * must result in the segment being dropped * and must not produce any response back * to the sender. */ goto dropunlock; } else if (rstreason == 0) { /* * No syncache entry or ACK was not * for our SYN/ACK. Send a RST. * NB: syncache did its own logging * of the failure cause. */ rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } tfo_socket_result: if (so == NULL) { /* * We completed the 3-way handshake * but could not allocate a socket * either due to memory shortage, * listen queue length limits or * global socket limits. Send RST * or wait and have the remote end * retransmit the ACK for another * try. */ if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Socket allocation failed due to " "limits or memory shortage, %s\n", s, __func__, V_tcp_sc_rst_sock_fail ? "sending RST" : "try again"); if (V_tcp_sc_rst_sock_fail) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } else goto dropunlock; } /* * Socket is created in state SYN_RECEIVED. * Unlock the listen socket, lock the newly * created socket and update the tp variable. */ INP_WUNLOCK(inp); /* listen socket */ inp = sotoinpcb(so); /* * New connection inpcb is already locked by * syncache_expand(). */ INP_WLOCK_ASSERT(inp); tp = intotcpcb(inp); KASSERT(tp->t_state == TCPS_SYN_RECEIVED, ("%s: ", __func__)); /* * Process the segment and the data it * contains. tcp_do_segment() consumes * the mbuf chain and unlocks the inpcb. */ TCP_PROBE5(receive, NULL, tp, m, tp, th); tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); return (IPPROTO_DONE); } /* * Segment flag validation for new connection attempts: * * Our (SYN|ACK) response was rejected. * Check with syncache and remove entry to prevent * retransmits. * * NB: syncache_chkrst does its own logging of failure * causes. */ if (thflags & TH_RST) { syncache_chkrst(&inc, th); goto dropunlock; } /* * We can't do anything without SYN. */ if ((thflags & TH_SYN) == 0) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN is missing, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * (SYN|ACK) is bogus on a listen socket. */ if (thflags & TH_ACK) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|ACK invalid, segment rejected\n", s, __func__); syncache_badack(&inc); /* XXX: Not needed! */ TCPSTAT_INC(tcps_badsyn); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } /* * If the drop_synfin option is enabled, drop all * segments with both the SYN and FIN bits set. * This prevents e.g. nmap from identifying the * TCP/IP stack. * XXX: Poor reasoning. nmap has other methods * and is constantly refining its stack detection * strategies. * XXX: This is a violation of the TCP specification * and was used by RFC1644. */ if ((thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * Segment's flags are (SYN) or (SYN|FIN). * * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored * as they do not affect the state of the TCP FSM. * The data pointed to by TH_URG and th_urp is ignored. */ KASSERT((thflags & (TH_RST|TH_ACK)) == 0, ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); KASSERT(thflags & (TH_SYN), ("%s: Listen socket: TH_SYN not set", __func__)); #ifdef INET6 /* * If deprecated address is forbidden, * we do not accept SYN to deprecated interface * address to prevent any new inbound connection from * getting established. * When we do not accept SYN, we send a TCP RST, * with deprecated source address (instead of dropping * it). We compromise it as it is much better for peer * to send a RST, and RST will be the final packet * for the exchange. * * If we do not forbid deprecated addresses, we accept * the SYN packet. RFC2462 does not suggest dropping * SYN in this case. * If we decipher RFC2462 5.5.4, it says like this: * 1. use of deprecated addr with existing * communication is okay - "SHOULD continue to be * used" * 2. use of it with new communication: * (2a) "SHOULD NOT be used if alternate address * with sufficient scope is available" * (2b) nothing mentioned otherwise. * Here we fall into (2b) case as we have no choice in * our source address selection - we must obey the peer. * * The wording in RFC2462 is confusing, and there are * multiple description text for deprecated address * handling - worse, they are not exactly the same. * I believe 5.5.4 is the best one, so we follow 5.5.4. */ if (isipv6 && !V_ip6_use_deprecated) { struct in6_ifaddr *ia6; ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); if (ia6 != NULL && (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { ifa_free(&ia6->ia_ifa); if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to deprecated " "IPv6 address rejected\n", s, __func__); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } if (ia6) ifa_free(&ia6->ia_ifa); } #endif /* INET6 */ /* * Basic sanity checks on incoming SYN requests: * Don't respond if the destination is a link layer * broadcast according to RFC1122 4.2.3.10, p. 104. * If it is from this socket it must be forged. * Don't respond if the source or destination is a * global or subnet broad- or multicast address. * Note that it is quite possible to receive unicast * link-layer packets with a broadcast IP address. Use * in_broadcast() to find them. */ if (m->m_flags & (M_BCAST|M_MCAST)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from broad- or multicast " "link layer address ignored\n", s, __func__); goto dropunlock; } #ifdef INET6 if (isipv6) { if (th->th_dport == th->th_sport && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to/from self " "ignored\n", s, __func__); goto dropunlock; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to multicast " "address ignored\n", s, __func__); goto dropunlock; } } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (th->th_dport == th->th_sport && ip->ip_dst.s_addr == ip->ip_src.s_addr) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to self " "ignored\n", s, __func__); goto dropunlock; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to broad- " "or multicast address ignored\n", s, __func__); goto dropunlock; } } #endif /* * SYN appears to be valid. Create compressed TCP state * for syncache. */ #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); tcp_dooptions(&to, optp, optlen, TO_SYN); if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) goto tfo_socket_result; /* * Entry added to syncache and mbuf consumed. * Only the listen socket is unlocked by syncache_add(). */ if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); ti_locked = TI_UNLOCKED; } - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); return (IPPROTO_DONE); } else if (tp->t_state == TCPS_LISTEN) { /* * When a listen socket is torn down the SO_ACCEPTCONN * flag is removed first while connections are drained * from the accept queue in a unlock/lock cycle of the * ACCEPT_LOCK, opening a race condition allowing a SYN * attempt go through unhandled. */ goto dropunlock; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) { tcp_dooptions(&to, optp, optlen, thflags); if ((to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); goto dropunlock; } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to.to_signature) != 0) goto dropunlock; } #endif TCP_PROBE5(receive, NULL, tp, m, tp, th); /* * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later * state. tcp_do_segment() always consumes the mbuf chain, unlocks * the inpcb, and unlocks pcbinfo. */ tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); if (ti_locked == TI_RLOCKED) INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); return (IPPROTO_DONE); dropwithreset: TCP_PROBE5(receive, NULL, tp, m, tp, th); if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); ti_locked = TI_UNLOCKED; } #ifdef INVARIANTS else { KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " "ti_locked: %d", __func__, ti_locked)); - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); } #endif if (inp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); m = NULL; /* mbuf chain got consumed. */ goto drop; dropunlock: if (m != NULL) TCP_PROBE5(receive, NULL, tp, m, tp, th); if (ti_locked == TI_RLOCKED) { INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); ti_locked = TI_UNLOCKED; } #ifdef INVARIANTS else { KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " "ti_locked: %d", __func__, ti_locked)); - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); } #endif if (inp != NULL) INP_WUNLOCK(inp); drop: - INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); + INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); if (s != NULL) free(s, M_TCPLOG); if (m != NULL) m_freem(m); return (IPPROTO_DONE); } /* * Automatic sizing of receive socket buffer. Often the send * buffer size is not optimally adjusted to the actual network * conditions at hand (delay bandwidth product). Setting the * buffer size too small limits throughput on links with high * bandwidth and high delay (eg. trans-continental/oceanic links). * * On the receive side the socket buffer memory is only rarely * used to any significant extent. This allows us to be much * more aggressive in scaling the receive socket buffer. For * the case that the buffer space is actually used to a large * extent and we run out of kernel memory we can simply drop * the new segments; TCP on the sender will just retransmit it * later. Setting the buffer size too big may only consume too * much kernel memory if the application doesn't read() from * the socket or packet loss or reordering makes use of the * reassembly queue. * * The criteria to step up the receive buffer one notch are: * 1. Application has not set receive buffer size with * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. * 2. the number of bytes received during the time it takes * one timestamp to be reflected back to us (the RTT); * 3. received bytes per RTT is within seven eighth of the * current socket buffer size; * 4. receive buffer size has not hit maximal automatic size; * * This algorithm does one step per RTT at most and only if * we receive a bulk stream w/o packet losses or reorderings. * Shrinking the buffer during idle times is not necessary as * it doesn't consume any memory when idle. * * TODO: Only step up if the application is actually serving * the buffer to better manage the socket buffer resources. */ int tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int tlen) { int newsize = 0; if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && tp->t_srtt != 0 && tp->rfbuf_ts != 0 && TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > (tp->t_srtt >> TCP_RTT_SHIFT)) { if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { newsize = min(so->so_rcv.sb_hiwat + V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); } TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); /* Start over with next RTT. */ tp->rfbuf_ts = 0; tp->rfbuf_cnt = 0; } else { tp->rfbuf_cnt += tlen; /* add up */ } return (newsize); } void tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) { int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; int rstreason, todrop, win; uint32_t tiwin; uint16_t nsegs; char *s; struct in_conninfo *inc; struct mbuf *mfree; struct tcpopt to; int tfo_syn; #ifdef TCPDEBUG /* * The size of tcp_saveipgen must be the size of the max ip header, * now IPv6. */ u_char tcp_saveipgen[IP6_HDR_LEN]; struct tcphdr tcp_savetcp; short ostate = 0; #endif thflags = th->th_flags; inc = &tp->t_inpcb->inp_inc; tp->sackhint.last_sack_ack = 0; sack_changed = 0; nsegs = max(1, m->m_pkthdr.lro_nsegs); /* * If this is either a state-changing packet or current state isn't * established, we require a write lock on tcbinfo. Otherwise, we * allow the tcbinfo to be in either alocked or unlocked, as the * caller may have unnecessarily acquired a write lock due to a race. */ if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || tp->t_state != TCPS_ESTABLISHED) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); } INP_WLOCK_ASSERT(tp->t_inpcb); KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", __func__)); #ifdef TCPPCAP /* Save segment, if requested. */ tcp_pcap_add(th, m, &(tp->t_inpkts)); #endif TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, tlen, NULL, true); if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); free(s, M_TCPLOG); } goto drop; } /* * If a segment with the ACK-bit set arrives in the SYN-SENT state * check SEQ.ACK first. */ if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } /* * Segment received on connection. * Reset idle time and keep-alive timer. * XXX: This should be done after segment * validation to ignore broken/spoofed segs. */ tp->t_rcvtime = ticks; /* * Scale up the window into a 32-bit value. * For the SYN_SENT state the scale is zero. */ tiwin = th->th_win << tp->snd_scale; /* * TCP ECN processing. */ if (tp->t_flags & TF_ECN_PERMIT) { if (thflags & TH_CWR) tp->t_flags &= ~TF_ECN_SND_ECE; switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->t_flags |= TF_ECN_SND_ECE; TCPSTAT_INC(tcps_ecn_ce); break; case IPTOS_ECN_ECT0: TCPSTAT_INC(tcps_ecn_ect0); break; case IPTOS_ECN_ECT1: TCPSTAT_INC(tcps_ecn_ect1); break; } /* Process a packet differently from RFC3168. */ cc_ecnpkt_handler(tp, th, iptos); /* Congestion experienced. */ if (thflags & TH_ECE) { cc_cong_signal(tp, th, CC_ECN); } } /* * Parse options on any incoming segment. */ tcp_dooptions(&to, (u_char *)(th + 1), (th->th_off << 2) - sizeof(struct tcphdr), (thflags & TH_SYN) ? TO_SYN : 0); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if ((tp->t_flags & TF_SIGNATURE) != 0 && (to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_sigopt); /* XXX: should drop? */ } #endif /* * If echoed timestamp is later than the current time, * fall back to non RFC1323 RTT calculation. Normalize * timestamp if syncookies were used when this connection * was established. */ if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { to.to_tsecr -= tp->ts_offset; if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) to.to_tsecr = 0; else if (tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Process options only when we get SYN/ACK back. The SYN case * for incoming connections is handled in tcp_syncache. * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. * XXX this is traditional behavior, may need to be cleaned up. */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) { tp->t_flags |= TF_RCVD_SCALE; tp->snd_scale = to.to_wscale; } /* * Initial send window. It will be updated with * the next incoming segment to the scaled value. */ tp->snd_wnd = th->th_win; if (to.to_flags & TOF_TS) { tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to.to_tsval; tp->ts_recent_age = tcp_ts_getticks(); } if (to.to_flags & TOF_MSS) tcp_mss(tp, to.to_mss); if ((tp->t_flags & TF_SACK_PERMIT) && (to.to_flags & TOF_SACKPERM) == 0) tp->t_flags &= ~TF_SACK_PERMIT; if (IS_FASTOPEN(tp->t_flags)) { if (to.to_flags & TOF_FASTOPEN) { uint16_t mss; if (to.to_flags & TOF_MSS) mss = to.to_mss; else if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) mss = TCP6_MSS; else mss = TCP_MSS; tcp_fastopen_update_cache(tp, mss, to.to_tfo_len, to.to_tfo_cookie); } else tcp_fastopen_disable_path(tp); } } /* * If timestamps were negotiated during SYN/ACK they should * appear on every segment during this session and vice versa. */ if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "no action\n", s, __func__); free(s, M_TCPLOG); } } if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "no action\n", s, __func__); free(s, M_TCPLOG); } } /* * Header prediction: check for the two common cases * of a uni-directional data xfer. If the packet has * no control flags, is in-sequence, the window didn't * change and we're not retransmitting, it's a * candidate. If the length is zero and the ack moved * forward, we're the sender side of the xfer. Just * free the data acked & wake any higher level process * that was blocked waiting for space. If the length * is non-zero and the ack didn't move, we're the * receiver side. If we're getting packets in-order * (the reassembly queue is empty), add the data to * the socket buffer and note that we need a delayed ack. * Make sure that the hidden state-flags are also off. * Since we check for TCPS_ESTABLISHED first, it can only * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && th->th_seq == tp->rcv_nxt && (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && tp->snd_nxt == tp->snd_max && tiwin && tiwin == tp->snd_wnd && ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && SEGQ_EMPTY(tp) && ((to.to_flags & TOF_TS) == 0 || TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { /* * If last ACK falls within this segment's sequence numbers, * record the timestamp. * NOTE that the test is modified according to the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } if (tlen == 0) { if (SEQ_GT(th->th_ack, tp->snd_una) && SEQ_LEQ(th->th_ack, tp->snd_max) && !IN_RECOVERY(tp->t_flags) && (to.to_flags & TOF_SACK) == 0 && TAILQ_EMPTY(&tp->snd_holes)) { /* * This is a pure ack for outstanding data. */ TCPSTAT_INC(tcps_predack); /* * "bad retransmit" recovery without timestamps. */ if ((to.to_flags & TOF_TS) == 0 && tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && (int)(ticks - tp->t_badrxtwin) < 0) { cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Recalculate the transmit timer / rtt. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } acked = BYTES_THIS_ACK(tp, th); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); sbdrop(&so->so_snd, acked); if (SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* * Let the congestion control algorithm update * congestion control related information. This * typically means increasing the congestion * window. */ cc_ack_received(tp, th, nsegs, CC_ACK); tp->snd_una = th->th_ack; /* * Pull snd_wl2 up to prevent seq wrap relative * to th_ack. */ tp->snd_wl2 = th->th_ack; tp->t_dupacks = 0; m_freem(m); /* * If all outstanding data are acked, stop * retransmit timer, otherwise restart timer * using current (possibly backed-off) value. * If process is waiting for space, * wakeup/selwakeup/signal. If data * are ready to send, let tcp_output * decide between more output or persist. */ #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); if (tp->snd_una == tp->snd_max) tcp_timer_activate(tp, TT_REXMT, 0); else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); sowwakeup(so); if (sbavail(&so->so_snd)) (void) tp->t_fb->tfb_tcp_output(tp); goto check_delack; } } else if (th->th_ack == tp->snd_una && tlen <= sbspace(&so->so_rcv)) { int newsize = 0; /* automatic sockbuf scaling */ /* * This is a pure, in-sequence data packet with * nothing on the reassembly queue and we have enough * buffer space to take it. */ /* Clean receiver SACK report if present */ if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) tcp_clean_sackreport(tp); TCPSTAT_INC(tcps_preddat); tp->rcv_nxt += tlen; /* * Pull snd_wl1 up to prevent seq wrap relative to * th_seq. */ tp->snd_wl1 = th->th_seq; /* * Pull rcv_up up to prevent seq wrap relative to * rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; TCPSTAT_ADD(tcps_rcvpack, nsegs); TCPSTAT_ADD(tcps_rcvbyte, tlen); #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); newsize = tcp_autorcvbuf(m, th, so, tp, tlen); /* Add data to socket buffer. */ SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { m_freem(m); } else { /* * Set new socket buffer size. * Give up when limit is reached. */ if (newsize) if (!sbreserve_locked(&so->so_rcv, newsize, so, NULL)) so->so_rcv.sb_flags &= ~SB_AUTOSIZE; m_adj(m, drop_hdrlen); /* delayed header drop */ sbappendstream_locked(&so->so_rcv, m, 0); } /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); if (DELAY_ACK(tp, tlen)) { tp->t_flags |= TF_DELACK; } else { tp->t_flags |= TF_ACKNOW; tp->t_fb->tfb_tcp_output(tp); } goto check_delack; } } /* * Calculate amount of space in receive window, * and then do TCP input processing. * Receive window is amount of space in rcv queue, * but not less than advertised window. */ win = sbspace(&so->so_rcv); if (win < 0) win = 0; tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); switch (tp->t_state) { /* * If the state is SYN_RECEIVED: * if seg contains an ACK, but not for our SYN/ACK, send a RST. */ case TCPS_SYN_RECEIVED: if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } if (IS_FASTOPEN(tp->t_flags)) { /* * When a TFO connection is in SYN_RECEIVED, the * only valid packets are the initial SYN, a * retransmit/copy of the initial SYN (possibly with * a subset of the original data), a valid ACK, a * FIN, or a RST. */ if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } else if (thflags & TH_SYN) { /* non-initial SYN is ignored */ if ((tcp_timer_active(tp, TT_DELACK) || tcp_timer_active(tp, TT_REXMT))) goto drop; } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { goto drop; } } break; /* * If the state is SYN_SENT: * if seg contains a RST with valid ACK (SEQ.ACK has already * been verified), then drop the connection. * if seg contains a RST without an ACK, drop the seg. * if seg does not contain SYN, then drop the seg. * Otherwise this is an acceptable SYN segment * initialize tp->rcv_nxt and tp->irs * if seg contains ack then advance tp->snd_una * if seg contains an ECE and ECN support is enabled, the stream * is ECN capable. * if SYN has been acked change to ESTABLISHED else SYN_RCVD state * arrange for segment to be acked (eventually) * continue processing rest of data/controls, beginning with URG */ case TCPS_SYN_SENT: if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { TCP_PROBE5(connect__refused, NULL, tp, m, tp, th); tp = tcp_drop(tp, ECONNREFUSED); } if (thflags & TH_RST) goto drop; if (!(thflags & TH_SYN)) goto drop; tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { int tfo_partial_ack = 0; TCPSTAT_INC(tcps_connects); soisconnected(so); #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif /* Do window scaling on this connection? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); tp->snd_una++; /* SYN is acked */ /* * If not all the data that was sent in the TFO SYN * has been acked, resend the remainder right away. */ if (IS_FASTOPEN(tp->t_flags) && (tp->snd_una != tp->snd_max)) { tp->snd_nxt = th->th_ack; tfo_partial_ack = 1; } /* * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); else tp->t_flags |= TF_ACKNOW; if ((thflags & TH_ECE) && V_tcp_do_ecn) { tp->t_flags |= TF_ECN_PERMIT; TCPSTAT_INC(tcps_ecn_shs); } /* * Received in SYN_SENT[*] state. * Transitions: * SYN_SENT --> ESTABLISHED * SYN_SENT* --> FIN_WAIT_1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(connect__established, NULL, tp, m, tp, th); cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } } else { /* * Received initial SYN in SYN-SENT[*] state => * simultaneous open. * If it succeeds, connection is * half-synchronized. * Otherwise, do 3-way handshake: * SYN-SENT -> SYN-RECEIVED * SYN-SENT* -> SYN-RECEIVED* */ tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); tcp_timer_activate(tp, TT_REXMT, 0); tcp_state_change(tp, TCPS_SYN_RECEIVED); } INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); /* * Advance th->th_seq to correspond to first data byte. * If data, trim to stay within window, * dropping FIN if necessary. */ th->th_seq++; if (tlen > tp->rcv_wnd) { todrop = tlen - tp->rcv_wnd; m_adj(m, -todrop); tlen = tp->rcv_wnd; thflags &= ~TH_FIN; TCPSTAT_INC(tcps_rcvpackafterwin); TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); } tp->snd_wl1 = th->th_seq - 1; tp->rcv_up = th->th_seq; /* * Client side of transaction: already sent SYN and data. * If the remote host used T/TCP to validate the SYN, * our data will be ACK'd; if so, enter normal data segment * processing in the middle of step 5, ack processing. * Otherwise, goto step 6. */ if (thflags & TH_ACK) goto process_ACK; goto step6; /* * If the state is LAST_ACK or CLOSING or TIME_WAIT: * do normal processing. * * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. */ case TCPS_LAST_ACK: case TCPS_CLOSING: break; /* continue normal processing */ } /* * States other than LISTEN or SYN_SENT. * First check the RST flag and sequence number since reset segments * are exempt from the timestamp and connection count tests. This * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix * below which allowed reset segments in half the sequence space * to fall though and be processed (which gives forged reset * segments with a random sequence number a 50 percent chance of * killing a connection). * Then check timestamp, if present. * Then check the connection count, if present. * Then check that at least some bytes of segment are within * receive window. If segment begins before rcv_nxt, * drop leading data (and SYN); if nothing left, just ack. */ if (thflags & TH_RST) { /* * RFC5961 Section 3.2 * * - RST drops connection only if SEG.SEQ == RCV.NXT. * - If RST is in window, we send challenge ACK. * * Note: to take into account delayed ACKs, we should * test against last_ack_sent instead of rcv_nxt. * Note 2: we handle special case of closed window, not * covered by the RFC. */ if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); KASSERT(tp->t_state != TCPS_SYN_SENT, ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", __func__, th, tp)); if (V_tcp_insecure_rst || tp->last_ack_sent == th->th_seq) { TCPSTAT_INC(tcps_drops); /* Drop the connection. */ switch (tp->t_state) { case TCPS_SYN_RECEIVED: so->so_error = ECONNREFUSED; goto close; case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: so->so_error = ECONNRESET; close: /* FALLTHROUGH */ default: tp = tcp_close(tp); } } else { TCPSTAT_INC(tcps_badrst); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } } goto drop; } /* * RFC5961 Section 4.2 * Send challenge ACK for any SYN in synchronized state. */ if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && tp->t_state != TCPS_SYN_RECEIVED) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); TCPSTAT_INC(tcps_badsyn); if (V_tcp_insecure_syn && SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { tp = tcp_drop(tp, ECONNRESET); rstreason = BANDLIM_UNLIMITED; } else { /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } goto drop; } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { /* Check to see if ts_recent is over 24 days old. */ if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { /* * Invalidate ts_recent. If this segment updates * ts_recent, the age will be reset later and ts_recent * will get a valid value. If it does not, setting * ts_recent to zero will at least satisfy the * requirement that zero be placed in the timestamp * echo reply when ts_recent isn't valid. The * age isn't reset until we get a valid ts_recent * because we don't want out-of-order segments to be * dropped when ts_recent is old. */ tp->ts_recent = 0; } else { TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, tlen); TCPSTAT_INC(tcps_pawsdrop); if (tlen) goto dropafterack; goto drop; } } /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive * window. Check the sequence number versus IRS since we know * the sequence numbers haven't wrapped. This is a partial fix * for the "LAND" DoS attack. */ if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } todrop = tp->rcv_nxt - th->th_seq; if (todrop > 0) { if (thflags & TH_SYN) { thflags &= ~TH_SYN; th->th_seq++; if (th->th_urp > 1) th->th_urp--; else thflags &= ~TH_URG; todrop--; } /* * Following if statement from Stevens, vol. 2, p. 960. */ if (todrop > tlen || (todrop == tlen && (thflags & TH_FIN) == 0)) { /* * Any valid FIN must be to the left of the window. * At this point the FIN must be a duplicate or out * of sequence; drop it. */ thflags &= ~TH_FIN; /* * Send an ACK to resynchronize and drop any data. * But keep on processing for RST or ACK. */ tp->t_flags |= TF_ACKNOW; todrop = tlen; TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, todrop); } else { TCPSTAT_INC(tcps_rcvpartduppack); TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); } drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; if (th->th_urp > todrop) th->th_urp -= todrop; else { thflags &= ~TH_URG; th->th_urp = 0; } } /* * If new data are received on a connection after the * user processes are gone, then RST the other end. */ if ((so->so_state & SS_NOFDREF) && tp->t_state > TCPS_CLOSE_WAIT && tlen) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " "after socket was closed, " "sending RST and removing tcpcb\n", s, __func__, tcpstates[tp->t_state], tlen); free(s, M_TCPLOG); } tp = tcp_close(tp); TCPSTAT_INC(tcps_rcvafterclose); rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } /* * If segment ends after window, drop trailing data * (and PUSH and FIN); if nothing left, just ACK. */ todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); if (todrop > 0) { TCPSTAT_INC(tcps_rcvpackafterwin); if (todrop >= tlen) { TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); /* * If window is closed can only take segments at * window edge, and have to drop data and PUSH from * incoming segments. Continue processing, but * remember to ack. Otherwise, drop segment * and ack. */ if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { tp->t_flags |= TF_ACKNOW; TCPSTAT_INC(tcps_rcvwinprobe); } else goto dropafterack; } else TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); m_adj(m, -todrop); tlen -= todrop; thflags &= ~(TH_PUSH|TH_FIN); } /* * If last ACK falls within this segment's sequence numbers, * record its timestamp. * NOTE: * 1) That the test incorporates suggestions from the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). * 2) That updating only on newer timestamps interferes with * our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. * 3) That we modify the segment boundary check to be * Last.ACK.Sent <= SEG.SEQ + SEG.Len * instead of RFC1323's * Last.ACK.Sent < SEG.SEQ + SEG.Len, * This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated * Vol. 2 p.869. In such cases, we can still calculate the * RTT correctly when RCV.NXT == Last.ACK.Sent. */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN|TH_FIN)) != 0))) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN * flag is on (half-synchronized state), then queue data for * later processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_state == TCPS_SYN_RECEIVED || (tp->t_flags & TF_NEEDSYN)) { if (tp->t_state == TCPS_SYN_RECEIVED && IS_FASTOPEN(tp->t_flags)) { tp->snd_wnd = tiwin; cc_conn_init(tp); } goto step6; } else if (tp->t_flags & TF_ACKNOW) goto dropafterack; else goto drop; } /* * Ack processing. */ switch (tp->t_state) { /* * In SYN_RECEIVED state, the ack ACKs our SYN, so enter * ESTABLISHED state and continue processing. * The ACK was checked above. */ case TCPS_SYN_RECEIVED: TCPSTAT_INC(tcps_connects); soisconnected(so); /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; tp->snd_wnd = tiwin; } /* * Make transitions: * SYN-RECEIVED -> ESTABLISHED * SYN-RECEIVED* -> FIN-WAIT-1 */ tp->t_starttime = ticks; if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; /* * Account for the ACK of our SYN prior to * regular ACK processing below. */ tp->snd_una++; } if (tp->t_flags & TF_NEEDFIN) { tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(accept__established, NULL, tp, m, tp, th); /* * TFO connections call cc_conn_init() during SYN * processing. Calling it again here for such * connections is not harmless as it would undo the * snd_cwnd reduction that occurs when a TFO SYN|ACK * is retransmitted. */ if (!IS_FASTOPEN(tp->t_flags)) cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } /* * If segment contains data or ACK, will call tcp_reass() * later; if not, do so now to pass queued data to user. */ if (tlen == 0 && (thflags & TH_FIN) == 0) (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, (struct mbuf *)0); tp->snd_wl1 = th->th_seq - 1; /* FALLTHROUGH */ /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range * ACKs. If the ack is in the range * tp->snd_una < th->th_ack <= tp->snd_max * then advance tp->snd_una to th->th_ack and drop * data from the retransmission queue. If this ACK reflects * more up to date window information we update our window information. */ case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: if (SEQ_GT(th->th_ack, tp->snd_max)) { TCPSTAT_INC(tcps_rcvacktoomuch); goto dropafterack; } if ((tp->t_flags & TF_SACK_PERMIT) && ((to.to_flags & TOF_SACK) || !TAILQ_EMPTY(&tp->snd_holes))) sack_changed = tcp_sack_doack(tp, &to, th->th_ack); else /* * Reset the value so that previous (valid) value * from the last ack with SACK doesn't get used. */ tp->sackhint.sacked_bytes = 0; #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif if (SEQ_LEQ(th->th_ack, tp->snd_una)) { u_int maxseg; maxseg = tcp_maxseg(tp); if (tlen == 0 && (tiwin == tp->snd_wnd || (tp->t_flags & TF_SACK_PERMIT))) { /* * If this is the first time we've seen a * FIN from the remote, this is not a * duplicate and it needs to be processed * normally. This happens during a * simultaneous close. */ if ((thflags & TH_FIN) && (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { tp->t_dupacks = 0; break; } TCPSTAT_INC(tcps_rcvdupack); /* * If we have outstanding data (other than * a window probe), this is a completely * duplicate ack (ie, window info didn't * change and FIN isn't set), * the ack is the biggest we've * seen and we've seen exactly our rexmt * threshold of them, assume a packet * has been dropped and retransmit it. * Kludge snd_nxt & the congestion * window so we send only this one * packet. * * We know we're losing at the current * window size so do congestion avoidance * (set ssthresh to half the current window * and pull our congestion window back to * the new ssthresh). * * Dup acks mean that packets have left the * network (they're now cached at the receiver) * so bump cwnd by the amount in the receiver * to keep a constant cwnd packets in the * network. * * When using TCP ECN, notify the peer that * we reduced the cwnd. */ /* * Following 2 kinds of acks should not affect * dupack counting: * 1) Old acks * 2) Acks with SACK but without any new SACK * information in them. These could result from * any anomaly in the network like a switch * duplicating packets or a possible DoS attack. */ if (th->th_ack != tp->snd_una || ((tp->t_flags & TF_SACK_PERMIT) && !sack_changed)) break; else if (!tcp_timer_active(tp, TT_REXMT)) tp->t_dupacks = 0; else if (++tp->t_dupacks > tcprexmtthresh || IN_FASTRECOVERY(tp->t_flags)) { cc_ack_received(tp, th, nsegs, CC_DUPACK); if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags)) { int awnd; /* * Compute the amount of data in flight first. * We can inject new data into the pipe iff * we have less than 1/2 the original window's * worth of data in flight. */ if (V_tcp_do_rfc6675_pipe) awnd = tcp_compute_pipe(tp); else awnd = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; if (awnd < tp->snd_ssthresh) { tp->snd_cwnd += maxseg; if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; } } else tp->snd_cwnd += maxseg; (void) tp->t_fb->tfb_tcp_output(tp); goto drop; } else if (tp->t_dupacks == tcprexmtthresh) { tcp_seq onxt = tp->snd_nxt; /* * If we're doing sack, check to * see if we're already in sack * recovery. If we're not doing sack, * check to see if we're in newreno * recovery. */ if (tp->t_flags & TF_SACK_PERMIT) { if (IN_FASTRECOVERY(tp->t_flags)) { tp->t_dupacks = 0; break; } } else { if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->t_dupacks = 0; break; } } /* Congestion signal before ack. */ cc_cong_signal(tp, th, CC_NDUPACK); cc_ack_received(tp, th, nsegs, CC_DUPACK); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; if (tp->t_flags & TF_SACK_PERMIT) { TCPSTAT_INC( tcps_sack_recovery_episode); tp->sack_newdata = tp->snd_nxt; tp->snd_cwnd = maxseg; (void) tp->t_fb->tfb_tcp_output(tp); goto drop; } tp->snd_nxt = th->th_ack; tp->snd_cwnd = maxseg; (void) tp->t_fb->tfb_tcp_output(tp); KASSERT(tp->snd_limited <= 2, ("%s: tp->snd_limited too big", __func__)); tp->snd_cwnd = tp->snd_ssthresh + maxseg * (tp->t_dupacks - tp->snd_limited); if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; goto drop; } else if (V_tcp_do_rfc3042) { /* * Process first and second duplicate * ACKs. Each indicates a segment * leaving the network, creating room * for more. Make sure we can send a * packet on reception of each duplicate * ACK by increasing snd_cwnd by one * segment. Restore the original * snd_cwnd after packet transmission. */ cc_ack_received(tp, th, nsegs, CC_DUPACK); uint32_t oldcwnd = tp->snd_cwnd; tcp_seq oldsndmax = tp->snd_max; u_int sent; int avail; KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2, ("%s: dupacks not 1 or 2", __func__)); if (tp->t_dupacks == 1) tp->snd_limited = 0; tp->snd_cwnd = (tp->snd_nxt - tp->snd_una) + (tp->t_dupacks - tp->snd_limited) * maxseg; /* * Only call tcp_output when there * is new data available to be sent. * Otherwise we would send pure ACKs. */ SOCKBUF_LOCK(&so->so_snd); avail = sbavail(&so->so_snd) - (tp->snd_nxt - tp->snd_una); SOCKBUF_UNLOCK(&so->so_snd); if (avail > 0) (void) tp->t_fb->tfb_tcp_output(tp); sent = tp->snd_max - oldsndmax; if (sent > maxseg) { KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) || (sent == maxseg + 1 && tp->t_flags & TF_SENTFIN), ("%s: sent too much", __func__)); tp->snd_limited = 2; } else if (sent > 0) ++tp->snd_limited; tp->snd_cwnd = oldcwnd; goto drop; } } break; } else { /* * This ack is advancing the left edge, reset the * counter. */ tp->t_dupacks = 0; /* * If this ack also has new SACK info, increment the * counter as per rfc6675. */ if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) tp->t_dupacks++; } KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("%s: th_ack <= snd_una", __func__)); /* * If the congestion window was inflated to account * for the other side's cached packets, retract it. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (tp->t_flags & TF_SACK_PERMIT) tcp_sack_partialack(tp, th); else tcp_newreno_partial_ack(tp, th); } else cc_post_recovery(tp, th); } /* * If we reach this point, ACK is not a duplicate, * i.e., it ACKs something we sent. */ if (tp->t_flags & TF_NEEDSYN) { /* * T/TCP: Connection was half-synchronized, and our * SYN has been ACK'd (so connection is now fully * synchronized). Go to non-starred state, * increment snd_una for ACK of SYN, and check if * we can do window scaling. */ tp->t_flags &= ~TF_NEEDSYN; tp->snd_una++; /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; /* Send window already scaled. */ } } process_ACK: INP_WLOCK_ASSERT(tp->t_inpcb); acked = BYTES_THIS_ACK(tp, th); KASSERT(acked >= 0, ("%s: acked unexepectedly negative " "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, tp->snd_una, th->th_ack, tp, m)); TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); /* * If we just performed our first retransmit, and the ACK * arrives within our recovery window, then it was a mistake * to do the retransmit in the first place. Recover our * original cwnd and ssthresh, and proceed to transmit where * we left off. */ if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) cc_cong_signal(tp, th, CC_RTO_ERR); /* * If we have a timestamp reply, update smoothed * round trip time. If no timestamp is present but * transmit timer is running and timed sequence * number was acked, update smoothed round trip time. * Since we now have an rtt measurement, cancel the * timer backoff (cf., Phil Karn's retransmit alg.). * Recompute the initial retransmit timer. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } /* * If all outstanding data is acked, stop retransmit * timer and remember to restart (more output or persist). * If there is more data to be acked, restart retransmit * timer, using current (possibly backed-off) value. */ if (th->th_ack == tp->snd_max) { tcp_timer_activate(tp, TT_REXMT, 0); needoutput = 1; } else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); /* * If no data (only SYN) was ACK'd, * skip rest of ACK processing. */ if (acked == 0) goto step6; /* * Let the congestion control algorithm update congestion * control related information. This typically means increasing * the congestion window. */ cc_ack_received(tp, th, nsegs, CC_ACK); SOCKBUF_LOCK(&so->so_snd); if (acked > sbavail(&so->so_snd)) { if (tp->snd_wnd >= sbavail(&so->so_snd)) tp->snd_wnd -= sbavail(&so->so_snd); else tp->snd_wnd = 0; mfree = sbcut_locked(&so->so_snd, (int)sbavail(&so->so_snd)); ourfinisacked = 1; } else { mfree = sbcut_locked(&so->so_snd, acked); if (tp->snd_wnd >= (uint32_t) acked) tp->snd_wnd -= acked; else tp->snd_wnd = 0; ourfinisacked = 0; } /* NB: sowwakeup_locked() does an implicit unlock. */ sowwakeup_locked(so); m_freem(mfree); /* Detect una wraparound. */ if (!IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* XXXLAS: Can this be moved up into cc_post_recovery? */ if (IN_RECOVERY(tp->t_flags) && SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_RECOVERY(tp->t_flags); } tp->snd_una = th->th_ack; if (tp->t_flags & TF_SACK_PERMIT) { if (SEQ_GT(tp->snd_una, tp->snd_recover)) tp->snd_recover = tp->snd_una; } if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; switch (tp->t_state) { /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now acknowledged * then enter FIN_WAIT_2. */ case TCPS_FIN_WAIT_1: if (ourfinisacked) { /* * If we can't receive any more * data, then closing user can proceed. * Starting the timer is contrary to the * specification, but if we don't get a FIN * we'll hang forever. * * XXXjl: * we should release the tp also, and use a * compressed state. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } break; /* * In CLOSING STATE in addition to the processing for * the ESTABLISHED state if the ACK acknowledges our FIN * then enter the TIME-WAIT state, otherwise ignore * the segment. */ case TCPS_CLOSING: if (ourfinisacked) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); tcp_twstart(tp); m_freem(m); return; } break; /* * In LAST_ACK, we may still be waiting for data to drain * and/or to be acked, as well as for the ack of our FIN. * If our FIN is now acknowledged, delete the TCB, * enter the closed state and return. */ case TCPS_LAST_ACK: if (ourfinisacked) { INP_INFO_RLOCK_ASSERT(&V_tcbinfo); tp = tcp_close(tp); goto drop; } break; } } step6: INP_WLOCK_ASSERT(tp->t_inpcb); /* * Update window information. * Don't look at window if no ACK: TAC's send garbage on first SYN. */ if ((thflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) TCPSTAT_INC(tcps_rcvwinupd); tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; needoutput = 1; } /* * Process segments with URG. */ if ((thflags & TH_URG) && th->th_urp && TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* * This is a kludge, but if we receive and accept * random urgent pointers, we'll crash in * soreceive. It's hard to imagine someone * actually wanting to send this much urgent data. */ SOCKBUF_LOCK(&so->so_rcv); if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { th->th_urp = 0; /* XXX */ thflags &= ~TH_URG; /* XXX */ SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, * then mark the data stream. This should not happen * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since * a FIN has been received from the remote side. * In these states we ignore the URG. * * According to RFC961 (Assigned Protocols), * the urgent pointer points to the last octet * of urgent data. We continue, however, * to consider it to indicate the first octet * of data past the urgent section as the original * spec states (in one of two places). */ if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = sbavail(&so->so_rcv) + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) so->so_rcv.sb_state |= SBS_RCVATMARK; sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); } SOCKBUF_UNLOCK(&so->so_rcv); /* * Remove out of band data so doesn't get presented to user. * This can happen independent of advancing the URG pointer, * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ if (th->th_urp <= (uint32_t)tlen && !(so->so_options & SO_OOBINLINE)) { /* hdr drop is delayed */ tcp_pulloutofband(so, th, m, drop_hdrlen); } } else { /* * If no out of band data is expected, * pull receive urgent pointer along * with the receive window. */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; } dodata: /* XXX */ INP_WLOCK_ASSERT(tp->t_inpcb); /* * Process the segment text, merging it into the TCP sequencing queue, * and arranging for acknowledgment of receipt if necessary. * This process logically involves adjusting tp->rcv_wnd as data * is presented to the user (this happens in tcp_usrreq.c, * case PRU_RCVD). If a FIN has already been received on this * connection then we just ignore the text. */ tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && IS_FASTOPEN(tp->t_flags)); if ((tlen || (thflags & TH_FIN) || tfo_syn) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { tcp_seq save_start = th->th_seq; m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly queue * with control block tp. Set thflags to whether reassembly now * includes a segment with FIN. This handles the common case * inline (segment is the next to be received on an established * connection, and the queue is empty), avoiding linkage into * and removal from the queue and repetition of various * conversions. * Set DELACK for segments received in order, but ack * immediately when segments are out of order (so * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && SEGQ_EMPTY(tp) && (TCPS_HAVEESTABLISHED(tp->t_state) || tfo_syn)) { if (DELAY_ACK(tp, tlen) || tfo_syn) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt += tlen; thflags = th->th_flags & TH_FIN; TCPSTAT_INC(tcps_rcvpack); TCPSTAT_ADD(tcps_rcvbyte, tlen); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) m_freem(m); else sbappendstream_locked(&so->so_rcv, m, 0); /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); } else { /* * XXX: Due to the header drop above "th" is * theoretically invalid by now. Fortunately * m_adj() doesn't actually frees any mbufs * when trimming from the head. */ thflags = tcp_reass(tp, th, &save_start, &tlen, m); tp->t_flags |= TF_ACKNOW; } if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) tcp_update_sack_list(tp, save_start, save_start + tlen); #if 0 /* * Note the amount of data that peer has sent into * our window, in order to estimate the sender's * buffer size. * XXX: Unused. */ if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); else len = so->so_rcv.sb_hiwat; #endif } else { m_freem(m); thflags &= ~TH_FIN; } /* * If FIN is received ACK the FIN and let the user know * that the connection is closing. */ if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { socantrcvmore(so); /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, * so it may be piggybacked when SYN is sent. * Otherwise, since we received a FIN then no * more input can be expected, send ACK now. */ if (tp->t_flags & TF_NEEDSYN) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt++; } switch (tp->t_state) { /* * In SYN_RECEIVED and ESTABLISHED STATES * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tcp_state_change(tp, TCPS_CLOSE_WAIT); break; /* * If still in FIN_WAIT_1 STATE FIN has not been acked so * enter the CLOSING state. */ case TCPS_FIN_WAIT_1: tcp_state_change(tp, TCPS_CLOSING); break; /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the other * standard timers. */ case TCPS_FIN_WAIT_2: INP_INFO_RLOCK_ASSERT(&V_tcbinfo); tcp_twstart(tp); return; } } #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); /* * Return any desired output. */ if (needoutput || (tp->t_flags & TF_ACKNOW)) (void) tp->t_fb->tfb_tcp_output(tp); check_delack: INP_WLOCK_ASSERT(tp->t_inpcb); if (tp->t_flags & TF_DELACK) { tp->t_flags &= ~TF_DELACK; tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } INP_WUNLOCK(tp->t_inpcb); return; dropafterack: /* * Generate an ACK dropping incoming segment if it occupies * sequence space, where the ACK reflects our state. * * We can now skip the test for the RST flag since all * paths to this code happen after packets containing * RST have been dropped. * * In the SYN-RECEIVED state, don't send an ACK unless the * segment we received passes the SYN-RECEIVED ACK test. * If it fails send a RST. This breaks the loop in the * "LAND" DoS attack, and also prevents an ACK storm * between two listening ports that have been sent forged * SYN segments, each with the source address of the other. */ if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && (SEQ_GT(tp->snd_una, th->th_ack) || SEQ_GT(th->th_ack, tp->snd_max)) ) { rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } #ifdef TCPDEBUG if (so->so_options & SO_DEBUG) tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); tp->t_flags |= TF_ACKNOW; (void) tp->t_fb->tfb_tcp_output(tp); INP_WUNLOCK(tp->t_inpcb); m_freem(m); return; dropwithreset: if (tp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(tp->t_inpcb); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); return; drop: /* * Drop space held by incoming segment and return. */ #ifdef TCPDEBUG if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, &tcp_savetcp, 0); #endif TCP_PROBE3(debug__input, tp, th, m); if (tp != NULL) INP_WUNLOCK(tp->t_inpcb); m_freem(m); } /* * Issue RST and make ACK acceptable to originator of segment. * The mbuf must still include the original packet header. * tp may be NULL. */ void tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen, int rstreason) { #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; #endif if (tp != NULL) { INP_WLOCK_ASSERT(tp->t_inpcb); } /* Don't bother if destination was broadcast/multicast. */ if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) goto drop; #ifdef INET6 if (mtod(m, struct ip *)->ip_v == 6) { ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) goto drop; /* IPv6 anycast check is done at tcp6_input() */ } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip = mtod(m, struct ip *); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) goto drop; } #endif /* Perform bandwidth limiting. */ if (badport_bandlim(rstreason) < 0) goto drop; /* tcp_respond consumes the mbuf chain. */ if (th->th_flags & TH_ACK) { tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, TH_RST); } else { if (th->th_flags & TH_SYN) tlen++; if (th->th_flags & TH_FIN) tlen++; tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); } return; drop: m_freem(m); } /* * Parse TCP options and place in tcpopt. */ void tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) { int opt, optlen; to->to_flags = 0; for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { if (cnt < 2) break; optlen = cp[1]; if (optlen < 2 || optlen > cnt) break; } switch (opt) { case TCPOPT_MAXSEG: if (optlen != TCPOLEN_MAXSEG) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_MSS; bcopy((char *)cp + 2, (char *)&to->to_mss, sizeof(to->to_mss)); to->to_mss = ntohs(to->to_mss); break; case TCPOPT_WINDOW: if (optlen != TCPOLEN_WINDOW) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_SCALE; to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); break; case TCPOPT_TIMESTAMP: if (optlen != TCPOLEN_TIMESTAMP) continue; to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); to->to_tsval = ntohl(to->to_tsval); bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); to->to_tsecr = ntohl(to->to_tsecr); break; case TCPOPT_SIGNATURE: /* * In order to reply to a host which has set the * TCP_SIGNATURE option in its initial SYN, we have * to record the fact that the option was observed * here for the syncache code to perform the correct * response. */ if (optlen != TCPOLEN_SIGNATURE) continue; to->to_flags |= TOF_SIGNATURE; to->to_signature = cp + 2; break; case TCPOPT_SACK_PERMITTED: if (optlen != TCPOLEN_SACK_PERMITTED) continue; if (!(flags & TO_SYN)) continue; if (!V_tcp_do_sack) continue; to->to_flags |= TOF_SACKPERM; break; case TCPOPT_SACK: if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) continue; if (flags & TO_SYN) continue; to->to_flags |= TOF_SACK; to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; to->to_sacks = cp + 2; TCPSTAT_INC(tcps_sack_rcv_blocks); break; case TCPOPT_FAST_OPEN: /* * Cookie length validation is performed by the * server side cookie checking code or the client * side cookie cache update code. */ if (!(flags & TO_SYN)) continue; if (!V_tcp_fastopen_client_enable && !V_tcp_fastopen_server_enable) continue; to->to_flags |= TOF_FASTOPEN; to->to_tfo_len = optlen - 2; to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; break; default: continue; } } } /* * Pull out of band byte out of a segment so * it doesn't appear in the user's data queue. * It is still reflected in the segment length for * sequencing purposes. */ void tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) { int cnt = off + th->th_urp - 1; while (cnt >= 0) { if (m->m_len > cnt) { char *cp = mtod(m, caddr_t) + cnt; struct tcpcb *tp = sototcpcb(so); INP_WLOCK_ASSERT(tp->t_inpcb); tp->t_iobc = *cp; tp->t_oobflags |= TCPOOB_HAVEDATA; bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); m->m_len--; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len--; return; } cnt -= m->m_len; m = m->m_next; if (m == NULL) break; } panic("tcp_pulloutofband"); } /* * Collect new round-trip time estimate * and update averages and current timeout. */ void tcp_xmit_timer(struct tcpcb *tp, int rtt) { int delta; INP_WLOCK_ASSERT(tp->t_inpcb); TCPSTAT_INC(tcps_rttupdated); tp->t_rttupdated++; if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { /* * srtt is stored as fixed point with 5 bits after the * binary point (i.e., scaled by 8). The following magic * is equivalent to the smoothing algorithm in rfc793 with * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed * point). Adjust rtt to origin 0. */ delta = ((rtt - 1) << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); if ((tp->t_srtt += delta) <= 0) tp->t_srtt = 1; /* * We accumulate a smoothed rtt variance (actually, a * smoothed mean difference), then set the retransmit * timer to smoothed rtt + 4 times the smoothed variance. * rttvar is stored as fixed point with 4 bits after the * binary point (scaled by 16). The following is * equivalent to rfc793 smoothing with an alpha of .75 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces * rfc793's wired-in beta. */ if (delta < 0) delta = -delta; delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); if ((tp->t_rttvar += delta) <= 0) tp->t_rttvar = 1; if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } else { /* * No rtt measurement yet - use the unsmoothed rtt. * Set the variance to half the rtt (so our first * retransmit happens at 3*rtt). */ tp->t_srtt = rtt << TCP_RTT_SHIFT; tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); tp->t_rttbest = tp->t_srtt + tp->t_rttvar; } tp->t_rtttime = 0; tp->t_rxtshift = 0; /* * the retransmit should happen at rtt + 4 * rttvar. * Because of the way we do the smoothing, srtt and rttvar * will each average +1/2 tick of bias. When we compute * the retransmit timer, we want 1/2 tick of rounding and * 1 extra tick because of +-1/2 tick uncertainty in the * firing of the timer. The bias will give us exactly the * 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); /* * We received an ack for a packet that wasn't retransmitted; * it is probably safe to discard any error indications we've * received recently. This isn't quite right, but close enough * for now (a route might have failed after we sent a segment, * and the return path might not be symmetrical). */ tp->t_softerror = 0; } /* * Determine a reasonable value for maxseg size. * If the route is known, check route for mtu. * If none, use an mss that can be handled on the outgoing interface * without forcing IP to fragment. If no route is found, route has no mtu, * or the destination isn't local, use a default, hopefully conservative * size (usually 512 or the default IP max size, but no more than the mtu * of the interface), as we can't discover anything about intervening * gateways or networks. We also initialize the congestion/slow start * window to be a single segment if the destination isn't local. * While looking at the routing entry, we also initialize other path-dependent * parameters from pre-set or cached values in the routing entry. * * NOTE that resulting t_maxseg doesn't include space for TCP options or * IP options, e.g. IPSEC data, since length of this data may vary, and * thus it is calculated for every segment separately in tcp_output(). * * NOTE that this routine is only called when we process an incoming * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS * settings are handled in tcp_mssopt(). */ void tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) { int mss = 0; uint32_t maxmtu = 0; struct inpcb *inp = tp->t_inpcb; struct hc_metrics_lite metrics; #ifdef INET6 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; size_t min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : sizeof (struct tcpiphdr); #else const size_t min_protoh = sizeof(struct tcpiphdr); #endif INP_WLOCK_ASSERT(tp->t_inpcb); if (mtuoffer != -1) { KASSERT(offer == -1, ("%s: conflict", __func__)); offer = mtuoffer - min_protoh; } /* Initialize. */ #ifdef INET6 if (isipv6) { maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_v6mssdflt; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { maxmtu = tcp_maxmtu(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_mssdflt; } #endif /* * No route to sender, stay with default mss and return. */ if (maxmtu == 0) { /* * In case we return early we need to initialize metrics * to a defined state as tcp_hc_get() would do for us * if there was no cache hit. */ if (metricptr != NULL) bzero(metricptr, sizeof(struct hc_metrics_lite)); return; } /* What have we got? */ switch (offer) { case 0: /* * Offer == 0 means that there was no MSS on the SYN * segment, in this case we use tcp_mssdflt as * already assigned to t_maxseg above. */ offer = tp->t_maxseg; break; case -1: /* * Offer == -1 means that we didn't receive SYN yet. */ /* FALLTHROUGH */ default: /* * Prevent DoS attack with too small MSS. Round up * to at least minmss. */ offer = max(offer, V_tcp_minmss); } /* * rmx information is now retrieved from tcp_hostcache. */ tcp_hc_get(&inp->inp_inc, &metrics); if (metricptr != NULL) bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); /* * If there's a discovered mtu in tcp hostcache, use it. * Else, use the link mtu. */ if (metrics.rmx_mtu) mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; else { #ifdef INET6 if (isipv6) { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in6_localaddr(&inp->in6p_faddr)) mss = min(mss, V_tcp_v6mssdflt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in_localaddr(inp->inp_faddr)) mss = min(mss, V_tcp_mssdflt); } #endif /* * XXX - The above conditional (mss = maxmtu - min_protoh) * probably violates the TCP spec. * The problem is that, since we don't know the * other end's MSS, we are supposed to use a conservative * default. But, if we do that, then MTU discovery will * never actually take place, because the conservative * default is much less than the MTUs typically seen * on the Internet today. For the moment, we'll sweep * this under the carpet. * * The conservative default might not actually be a problem * if the only case this occurs is when sending an initial * SYN with options and data to a host we've never talked * to before. Then, they will reply with an MSS value which * will get recorded and the new parameters should get * recomputed. For Further Study. */ } mss = min(mss, offer); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ mss = max(mss, 64); tp->t_maxseg = mss; } void tcp_mss(struct tcpcb *tp, int offer) { int mss; uint32_t bufsize; struct inpcb *inp; struct socket *so; struct hc_metrics_lite metrics; struct tcp_ifcap cap; KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); bzero(&cap, sizeof(cap)); tcp_mss_update(tp, offer, -1, &metrics, &cap); mss = tp->t_maxseg; inp = tp->t_inpcb; /* * If there's a pipesize, change the socket buffer to that size, * don't change if sb_hiwat is different than default (then it * has been changed on purpose with setsockopt). * Make the socket buffers an integral number of mss units; * if the mss is larger than the socket buffer, decrease the mss. */ so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) bufsize = metrics.rmx_sendpipe; else bufsize = so->so_snd.sb_hiwat; if (bufsize < mss) mss = bufsize; else { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_snd.sb_hiwat) (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); } SOCKBUF_UNLOCK(&so->so_snd); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ tp->t_maxseg = max(mss, 64); SOCKBUF_LOCK(&so->so_rcv); if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) bufsize = metrics.rmx_recvpipe; else bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_rcv.sb_hiwat) (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); } SOCKBUF_UNLOCK(&so->so_rcv); /* Check the interface for TSO capabilities. */ if (cap.ifcap & CSUM_TSO) { tp->t_flags |= TF_TSO; tp->t_tsomax = cap.tsomax; tp->t_tsomaxsegcount = cap.tsomaxsegcount; tp->t_tsomaxsegsize = cap.tsomaxsegsize; } } /* * Determine the MSS option to send on an outgoing SYN. */ int tcp_mssopt(struct in_conninfo *inc) { int mss = 0; uint32_t thcmtu = 0; uint32_t maxmtu = 0; size_t min_protoh; KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { mss = V_tcp_v6mssdflt; maxmtu = tcp_maxmtu6(inc, NULL); min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = V_tcp_mssdflt; maxmtu = tcp_maxmtu(inc, NULL); min_protoh = sizeof(struct tcpiphdr); } #endif #if defined(INET6) || defined(INET) thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ #endif if (maxmtu && thcmtu) mss = min(maxmtu, thcmtu) - min_protoh; else if (maxmtu || thcmtu) mss = max(maxmtu, thcmtu) - min_protoh; return (mss); } /* * On a partial ack arrives, force the retransmission of the * next unacknowledged segment. Do not clear tp->t_dupacks. * By setting snd_nxt to ti_ack, this forces retransmission timer to * be started again. */ void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) { tcp_seq onxt = tp->snd_nxt; uint32_t ocwnd = tp->snd_cwnd; u_int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tp->t_inpcb); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; tp->snd_nxt = th->th_ack; /* * Set snd_cwnd to one segment beyond acknowledged offset. * (tp->snd_una has not yet been updated when this function is called.) */ tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); tp->t_flags |= TF_ACKNOW; (void) tp->t_fb->tfb_tcp_output(tp); tp->snd_cwnd = ocwnd; if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; /* * Partial window deflation. Relies on fact that tp->snd_una * not updated yet. */ if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); else tp->snd_cwnd = 0; tp->snd_cwnd += maxseg; } int tcp_compute_pipe(struct tcpcb *tp) { return (tp->snd_max - tp->snd_una + tp->sackhint.sack_bytes_rexmit - tp->sackhint.sacked_bytes); }