--- //depot/vendor/freebsd/src/sys/kern/kern_descrip.c 2004/03/28 16:35:52 +++ //depot/user/rwatson/netperf/sys/kern/kern_descrip.c 2004/03/28 17:06:00 @@ -2043,6 +2043,7 @@ { NET_ASSERT_GIANT(); + SOCK_LOCK(so); sorele(so); } @@ -2068,6 +2069,10 @@ } /* We have the last ref so we can proceed without the file lock. */ FILE_UNLOCK(fp); + /* + * XXXRW: With sockets locked, can we push this Giant grab into + * the purely VFS case? + */ mtx_lock(&Giant); if (fp->f_count < 0) panic("fdrop: count < 0"); --- //depot/vendor/freebsd/src/sys/kern/kern_timeout.c 2003/12/07 12:05:33 +++ //depot/user/rwatson/netperf/sys/kern/kern_timeout.c 2004/03/08 08:37:54 @@ -47,6 +47,7 @@ #include #include #include +#include #include static int avg_depth; @@ -58,6 +59,89 @@ static int avg_mpcalls; SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0, "Average number of MP callouts made per softclock call. Units = 1/1000"); + +/*- + * Sampling buffer of function pointers executed by timeouts and callouts. + * This circular buffer wraps when it fills, and uses an inefficient + * sbuf-based sysctl to dump sample data to userspace. Sysctls can select + * to monitor mpsafe and !mpsafe callouts/timeouts as desired. Suggested + * use is: (1) set sample of interest (mpsafe/notmpsafe), (2) reset the + * buffer, (3) do some benchmark/test, (5) disable sampling, (6) dump + * buffer. + * + * XXX: ifdef TIMEOUT_SAMPLING? + */ + +#define MAXFUNC 200000 +static void * func_array[MAXFUNC]; +static int array_off; + +static void +push_cfunc(void *ptr) +{ + + /* XXX */ + func_array[array_off % MAXFUNC] = ptr; + array_off++; +} + +static int +sysctl_cfunc(SYSCTL_HANDLER_ARGS) +{ + struct sbuf sb; + int error, i; + + if (req->newptr != NULL) + return (EINVAL); + + sbuf_new(&sb, NULL, 0, SBUF_AUTOEXTEND); + + for (i = 0; i < MAXFUNC; i++) { + if (func_array[i] == NULL) + break; + sbuf_printf(&sb, "%p ", func_array[i]); + } + sbuf_finish(&sb); + + error = sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb) + 1, req); + + sbuf_delete(&sb); + + return (error); +} + +SYSCTL_PROC(_debug, OID_AUTO, to_cfunc, CTLTYPE_STRING|CTLFLAG_RD, 0, 0, + sysctl_cfunc, "A", "callout/timeout sample"); + +static int +sysctl_cfunc_reset(SYSCTL_HANDLER_ARGS) +{ + int dummy, error; + + dummy = 0; + error = sysctl_handle_int(oidp, &dummy, 0, req); + if (error) + return (error); + + if (dummy != 0) { + bzero(func_array, sizeof(void *) * MAXFUNC); + array_off = 0; + } + + return (0); +} + +SYSCTL_PROC(_debug, OID_AUTO, to_cfunc_reset, CTLTYPE_INT|CTLFLAG_RW, 0, 0, + sysctl_cfunc_reset, "I", "Reset sample"); + +static int cfunc_sample_mpsafe; +static int cfunc_sample_notmpsafe; + +SYSCTL_INT(_debug, OID_AUTO, to_cfunc_mpsafe, CTLFLAG_RW, + &cfunc_sample_mpsafe, 0, "Sample mpsafe callouts"); +SYSCTL_INT(_debug, OID_AUTO, to_cfunc_notmpsafe, CTLFLAG_RW, + &cfunc_sample_notmpsafe, 0, "Sample !mpsafe callouts"); + /* * TODO: * allocate more timeout table slots when table overflows. @@ -216,8 +300,12 @@ if (!(c_flags & CALLOUT_MPSAFE)) { mtx_lock(&Giant); gcalls++; + if (cfunc_sample_mpsafe) + push_cfunc(c_func); } else { mpcalls++; + if (cfunc_sample_notmpsafe) + push_cfunc(c_func); } #ifdef DIAGNOSTIC binuptime(&bt1); --- //depot/vendor/freebsd/src/sys/kern/subr_log.c 2004/02/21 13:16:09 +++ //depot/user/rwatson/netperf/sys/kern/subr_log.c 2004/03/06 09:01:33 @@ -87,7 +87,12 @@ struct callout sc_callout; /* callout to wakeup syslog */ } logsoftc; -int log_open; /* also used in log() */ +/* + * log_mtx protects logsoftc, log_open. Note that log_mtx does *not* + * protect the structures associated with msgbuf, which require Giant. + */ +struct mtx log_mtx; +int log_open; /* also used in log() */ /* Times per second to check for a pending syslog wakeup. */ static int log_wakeups_per_second = 5; @@ -98,17 +103,24 @@ static int logopen(dev_t dev, int flags, int mode, struct thread *td) { - if (log_open) + + mtx_lock(&log_mtx); + if (log_open) { + mtx_unlock(&log_mtx); return (EBUSY); + } log_open = 1; - callout_init(&logsoftc.sc_callout, 0); + callout_init(&logsoftc.sc_callout, CALLOUT_MPSAFE); + mtx_unlock(&log_mtx); fsetown(td->td_proc->p_pid, &logsoftc.sc_sigio); /* signal process only */ + mtx_lock(&log_mtx); if (log_wakeups_per_second < 1) { printf("syslog wakeup is less than one. Adjusting to 1.\n"); log_wakeups_per_second = 1; } callout_reset(&logsoftc.sc_callout, hz / log_wakeups_per_second, logtimeout, NULL); + mtx_unlock(&log_mtx); return (0); } @@ -117,9 +129,11 @@ logclose(dev_t dev, int flag, int mode, struct thread *td) { + mtx_lock(&log_mtx); log_open = 0; callout_stop(&logsoftc.sc_callout); logsoftc.sc_state = 0; + mtx_unlock(&log_mtx); funsetown(&logsoftc.sc_sigio); return (0); } @@ -138,14 +152,18 @@ splx(s); return (EWOULDBLOCK); } + mtx_lock(&log_mtx); logsoftc.sc_state |= LOG_RDWAIT; + mtx_unlock(&log_mtx); if ((error = tsleep(mbp, LOG_RDPRI | PCATCH, "klog", 0))) { splx(s); return (error); } } splx(s); + mtx_lock(&log_mtx); logsoftc.sc_state &= ~LOG_RDWAIT; + mtx_unlock(&log_mtx); while (uio->uio_resid > 0) { l = imin(sizeof(buf), uio->uio_resid); @@ -182,8 +200,11 @@ logtimeout(void *arg) { - if (!log_open) + mtx_lock(&log_mtx); + if (!log_open) { + mtx_unlock(&log_mtx); return; + } if (log_wakeups_per_second < 1) { printf("syslog wakeup is less than one. Adjusting to 1.\n"); log_wakeups_per_second = 1; @@ -191,6 +212,7 @@ if (msgbuftrigger == 0) { callout_reset(&logsoftc.sc_callout, hz / log_wakeups_per_second, logtimeout, NULL); + mtx_unlock(&log_mtx); return; } msgbuftrigger = 0; @@ -203,6 +225,7 @@ } callout_reset(&logsoftc.sc_callout, hz / log_wakeups_per_second, logtimeout, NULL); + mtx_unlock(&log_mtx); } /*ARGSUSED*/ @@ -221,10 +244,12 @@ break; case FIOASYNC: + mtx_lock(&log_mtx); if (*(int *)data) logsoftc.sc_state |= LOG_ASYNC; else logsoftc.sc_state &= ~LOG_ASYNC; + mtx_unlock(&log_mtx); break; case FIOSETOWN: @@ -253,6 +278,7 @@ log_drvinit(void *unused) { + mtx_init(&log_mtx, "log_mtx", NULL, MTX_DEF); make_dev(&log_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "klog"); } --- //depot/vendor/freebsd/src/sys/kern/subr_prf.c 2004/02/18 21:30:32 +++ //depot/user/rwatson/netperf/sys/kern/subr_prf.c 2004/03/06 09:01:33 @@ -87,6 +87,9 @@ size_t remain; }; +/* + * XXXRW: We access subr_log.c's log_open variable unlocked. + */ extern int log_open; static void msglogchar(int c, int pri); --- //depot/vendor/freebsd/src/sys/kern/sys_socket.c 2004/03/28 17:56:03 +++ //depot/user/rwatson/netperf/sys/kern/sys_socket.c 2004/03/28 18:34:51 @@ -81,12 +81,14 @@ NET_LOCK_GIANT(); #ifdef MAC + /* XXX: Socket lock needed here? */ error = mac_check_socket_receive(active_cred, so); if (error) { NET_UNLOCK_GIANT(); return (error); } #endif + NET_LOCK_GIANT(); error = so->so_proto->pr_usrreqs->pru_soreceive(so, 0, uio, 0, 0, 0); NET_UNLOCK_GIANT(); return (error); @@ -106,12 +108,14 @@ NET_LOCK_GIANT(); #ifdef MAC + /* XXX: Socket lock needed here? */ error = mac_check_socket_send(active_cred, so); if (error) { NET_UNLOCK_GIANT(); return (error); } #endif + NET_LOCK_GIANT(); error = so->so_proto->pr_usrreqs->pru_sosend(so, 0, uio, 0, 0, 0, uio->uio_td); NET_UNLOCK_GIANT(); --- //depot/vendor/freebsd/src/sys/kern/uipc_domain.c 2003/09/02 14:00:52 +++ //depot/user/rwatson/netperf/sys/kern/uipc_domain.c 2004/02/29 20:52:43 @@ -134,8 +134,13 @@ if (max_linkhdr < 16) /* XXX */ max_linkhdr = 16; - callout_init(&pffast_callout, 0); - callout_init(&pfslow_callout, 0); + if (debug_mpsafenet) { + callout_init(&pffast_callout, CALLOUT_MPSAFE); + callout_init(&pfslow_callout, CALLOUT_MPSAFE); + } else { + callout_init(&pffast_callout, 0); + callout_init(&pfslow_callout, 0); + } callout_reset(&pffast_callout, 1, pffasttimo, NULL); callout_reset(&pfslow_callout, 1, pfslowtimo, NULL); --- //depot/vendor/freebsd/src/sys/kern/uipc_socket.c 2004/02/29 19:15:33 +++ //depot/user/rwatson/netperf/sys/kern/uipc_socket.c 2004/02/29 21:00:26 @@ -145,6 +145,8 @@ return so; } #endif + SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd"); + SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv"); /* XXX race condition for reentrant kernel */ so->so_gencnt = ++so_gencnt; /* sx_init(&so->so_sxlock, "socket sxlock"); */ @@ -203,6 +205,7 @@ soref(so); error = (*prp->pr_usrreqs->pru_attach)(so, proto, td); if (error) { + SOCK_LOCK(so); so->so_state |= SS_NOFDREF; sorele(so); return (error); @@ -246,6 +249,8 @@ mac_destroy_socket(so); #endif crfree(so->so_cred); + SOCKBUF_LOCK_DESTROY(&so->so_snd); + SOCKBUF_LOCK_DESTROY(&so->so_rcv); /* sx_destroy(&so->so_sxlock); */ uma_zfree(socket_zone, so); --numopensockets; @@ -270,11 +275,13 @@ splx(s); return (error); } + SOCKBUF_LOCK(&so->so_rcv); if (TAILQ_EMPTY(&so->so_comp)) so->so_options |= SO_ACCEPTCONN; if (backlog < 0 || backlog > somaxconn) backlog = somaxconn; so->so_qlimit = backlog; + SOCKBUF_UNLOCK(&so->so_rcv); splx(s); return (0); } @@ -283,14 +290,17 @@ sofree(so) struct socket *so; { - struct socket *head = so->so_head; - int s; - KASSERT(so->so_count == 0, ("socket %p so_count not 0", so)); + SOCK_LOCK_ASSERT(so); - if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) + if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0) { + SOCK_UNLOCK(so); return; - if (head != NULL) { + } + SOCK_UNLOCK(so); + SOCKBUF_LOCK(&so->so_rcv); + if (so->so_head != NULL) { + struct socket *head = so->so_head; if (so->so_state & SS_INCOMP) { TAILQ_REMOVE(&head->so_incomp, so, so_list); head->so_incqlen--; @@ -301,6 +311,7 @@ * accept(2) may hang after select(2) indicated * that the listening socket was ready. */ + /* XXX SOCKBUF_UNLOCK(&so->so_rcv); */ return; } else { panic("sofree: not queued"); @@ -308,13 +319,14 @@ so->so_state &= ~SS_INCOMP; so->so_head = NULL; } + SOCKBUF_UNLOCK(&so->so_rcv); + SOCKBUF_LOCK(&so->so_snd); so->so_snd.sb_flags |= SB_NOINTR; (void)sblock(&so->so_snd, M_WAITOK); - s = splimp(); - socantsendmore(so); - splx(s); + socantsendmore_locked(so); sbunlock(&so->so_snd); sbrelease(&so->so_snd, so); + SOCKBUF_UNLOCK(&so->so_snd); sorflush(so); sodealloc(so); } @@ -354,11 +366,14 @@ (void) soabort(sp); } } + SOCK_LOCK(so); if (so->so_pcb == 0) goto discard; if (so->so_state & SS_ISCONNECTED) { if ((so->so_state & SS_ISDISCONNECTING) == 0) { + SOCK_UNLOCK(so); error = sodisconnect(so); + SOCK_LOCK(so); if (error) goto drop; } @@ -367,7 +382,7 @@ (so->so_state & SS_NBIO)) goto drop; while (so->so_state & SS_ISCONNECTED) { - error = tsleep(&so->so_timeo, + error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH, "soclos", so->so_linger * hz); if (error) break; @@ -376,7 +391,10 @@ } drop: if (so->so_pcb) { - int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); + int error2; + SOCK_UNLOCK(so); + error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so); + SOCK_LOCK(so); if (error == 0) error = error2; } @@ -400,6 +418,7 @@ error = (*so->so_proto->pr_usrreqs->pru_abort)(so); if (error) { + SOCK_LOCK(so); sotryfree(so); /* note: does not decrement the ref count */ return error; } @@ -411,14 +430,12 @@ struct socket *so; struct sockaddr **nam; { - int s = splnet(); int error; if ((so->so_state & SS_NOFDREF) == 0) panic("soaccept: !NOFDREF"); so->so_state &= ~SS_NOFDREF; error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam); - splx(s); return (error); } @@ -428,12 +445,10 @@ struct sockaddr *nam; struct thread *td; { - int s; int error; if (so->so_options & SO_ACCEPTCONN) return (EOPNOTSUPP); - s = splnet(); /* * If protocol is connection-based, can only connect once. * Otherwise, if connected, try to disconnect first. @@ -446,7 +461,6 @@ error = EISCONN; else error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td); - splx(s); return (error); } @@ -455,11 +469,9 @@ struct socket *so1; struct socket *so2; { - int s = splnet(); int error; error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2); - splx(s); return (error); } @@ -467,20 +479,13 @@ sodisconnect(so) struct socket *so; { - int s = splnet(); int error; - if ((so->so_state & SS_ISCONNECTED) == 0) { - error = ENOTCONN; - goto bad; - } - if (so->so_state & SS_ISDISCONNECTING) { - error = EALREADY; - goto bad; - } + if ((so->so_state & SS_ISCONNECTED) == 0) + return ENOTCONN; + if (so->so_state & SS_ISDISCONNECTING) + return EALREADY; error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so); -bad: - splx(s); return (error); } @@ -531,7 +536,7 @@ struct mbuf **mp; struct mbuf *m; long space, len, resid; - int clen = 0, error, s, dontroute, mlen; + int clen = 0, error, dontroute, mlen; int atomic = sosendallatonce(so) || top; #ifdef ZERO_COPY_SOCKETS int cow_send; @@ -563,20 +568,18 @@ td->td_proc->p_stats->p_ru.ru_msgsnd++; if (control) clen = control->m_len; -#define snderr(errno) { error = (errno); splx(s); goto release; } +#define snderr(errno) { error = (errno); goto release; } -restart: + SOCKBUF_LOCK(&so->so_snd); error = sblock(&so->so_snd, SBLOCKWAIT(flags)); if (error) goto out; do { - s = splnet(); if (so->so_state & SS_CANTSENDMORE) snderr(EPIPE); if (so->so_error) { error = so->so_error; so->so_error = 0; - splx(s); goto release; } if ((so->so_state & SS_ISCONNECTED) == 0) { @@ -605,14 +608,11 @@ (atomic || space < so->so_snd.sb_lowat || space < clen)) { if (so->so_state & SS_NBIO) snderr(EWOULDBLOCK); - sbunlock(&so->so_snd); error = sbwait(&so->so_snd); - splx(s); if (error) - goto out; - goto restart; + goto release; + continue; } - splx(s); mp = ⊤ space -= clen; do { @@ -627,10 +627,12 @@ #ifdef ZERO_COPY_SOCKETS cow_send = 0; #endif /* ZERO_COPY_SOCKETS */ + SOCKBUF_UNLOCK(&so->so_snd); if (top == 0) { MGETHDR(m, M_TRYWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; + SOCKBUF_LOCK(&so->so_snd); /* XXX */ goto release; } mlen = MHLEN; @@ -640,6 +642,7 @@ MGET(m, M_TRYWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; + SOCKBUF_LOCK(&so->so_snd); /* XXX */ goto release; } mlen = MLEN; @@ -687,6 +690,7 @@ else #endif /* ZERO_COPY_SOCKETS */ error = uiomove(mtod(m, void *), (int)len, uio); + SOCKBUF_LOCK(&so->so_snd); resid = uio->uio_resid; m->m_len = len; *mp = m; @@ -702,13 +706,12 @@ } while (space > 0 && atomic); if (dontroute) so->so_options |= SO_DONTROUTE; - s = splnet(); /* XXX */ /* * XXX all the SS_CANTSENDMORE checks previously * done could be out of date. We could have recieved * a reset packet in an interrupt or maybe we slept * while doing page faults in uiomove() etc. We could - * probably recheck again inside the splnet() protection + * probably recheck again inside the locking protection * here, but there are probably other places that this * also happens. We must rethink this. */ @@ -726,7 +729,6 @@ /* If there is more to send set PRUS_MORETOCOME */ (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, top, addr, control, td); - splx(s); if (dontroute) so->so_options &= ~SO_DONTROUTE; clen = 0; @@ -741,6 +743,7 @@ release: sbunlock(&so->so_snd); out: + SOCKBUF_UNLOCK(&so->so_snd); if (top) m_freem(top); if (control) @@ -774,7 +777,7 @@ int *flagsp; { struct mbuf *m, **mp; - int flags, len, error, s, offset; + int flags, len, error, offset; struct protosw *pr = so->so_proto; struct mbuf *nextrecord; int moff, type = 0; @@ -832,12 +835,12 @@ if (so->so_state & SS_ISCONFIRMING && uio->uio_resid) (*pr->pr_usrreqs->pru_rcvd)(so, 0); -restart: + SOCKBUF_LOCK(&so->so_rcv); error = sblock(&so->so_rcv, SBLOCKWAIT(flags)); if (error) - return (error); - s = splnet(); + goto out; +restart: m = so->so_rcv.sb_mb; /* * If we have less data than requested, block awaiting more @@ -855,9 +858,8 @@ (so->so_rcv.sb_cc < so->so_rcv.sb_lowat || ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) && m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) { - KASSERT(m != 0 || !so->so_rcv.sb_cc, - ("receive: m == %p so->so_rcv.sb_cc == %u", - m, so->so_rcv.sb_cc)); + KASSERT(!(m == 0 && so->so_rcv.sb_cc), + ("m %p so->so_rcv.sb_cc %u", m, so->so_rcv.sb_cc)); if (so->so_error) { if (m) goto dontblock; @@ -890,14 +892,13 @@ } SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); - sbunlock(&so->so_rcv); error = sbwait(&so->so_rcv); - splx(s); if (error) - return (error); + goto release; goto restart; } dontblock: + KASSERT(error == 0, ("unexpected state, error %u", error)); if (uio->uio_td) uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++; SBLASTRECORDCHK(&so->so_rcv); @@ -906,10 +907,14 @@ if (pr->pr_flags & PR_ADDR) { KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type)); - orig_resid = 0; - if (psa) + if (psa) { *psa = sodupsockaddr(mtod(m, struct sockaddr *), - mp0 == NULL ? M_WAITOK : M_NOWAIT); + M_NOWAIT); + if (*psa == NULL) { + error = ENOMEM; + goto release; + } + } if (flags & MSG_PEEK) { m = m->m_next; } else { @@ -917,30 +922,56 @@ so->so_rcv.sb_mb = m_free(m); m = so->so_rcv.sb_mb; } + orig_resid = 0; } - while (m && m->m_type == MT_CONTROL && error == 0) { - if (flags & MSG_PEEK) { - if (controlp) - *controlp = m_copy(m, 0, m->m_len); - m = m->m_next; - } else { - sbfree(&so->so_rcv, m); - so->so_rcv.sb_mb = m->m_next; - m->m_next = NULL; - if (pr->pr_domain->dom_externalize) - error = - (*pr->pr_domain->dom_externalize)(m, controlp); - else if (controlp) - *controlp = m; - else - m_freem(m); - m = so->so_rcv.sb_mb; + if (m && m->m_type == MT_CONTROL) { + struct mbuf *cm = NULL; + struct mbuf **cme = &cm; + + do { + if (flags & MSG_PEEK) { + if (controlp) { + SOCKBUF_UNLOCK(&so->so_rcv); + *controlp = m_copym(m, 0, m->m_len, + M_TRYWAIT); + SOCKBUF_LOCK(&so->so_rcv); + if (*controlp == NULL) { + error = ENOBUFS; + goto release; + } + controlp = &(*controlp)->m_next; + } + m = m->m_next; + } else { + sbfree(&so->so_rcv, m); + so->so_rcv.sb_mb = m->m_next; + m->m_next = NULL; + if (controlp) { + /* + * Collect mbufs for processing below. + */ + *cme = m; + cme = &(*cme)->m_next; + } else + m_free(m); + m = so->so_rcv.sb_mb; + } + } while (m && m->m_type == MT_CONTROL); + if (cm != NULL) { + if (pr->pr_domain->dom_externalize) { + /* + * NB: drop the lock to avoid potential LORs; + * in particular unix domain sockets grab the + * file descriptor lock which would be a LOR. + */ + SOCKBUF_UNLOCK(&so->so_rcv); + error = (*pr->pr_domain->dom_externalize) + (cm, controlp); + SOCKBUF_LOCK(&so->so_rcv); + } else + m_freem(cm); } - if (controlp) { - orig_resid = 0; - while (*controlp != NULL) - controlp = &(*controlp)->m_next; - } + orig_resid = 0; } if (m) { if ((flags & MSG_PEEK) == 0) { @@ -997,7 +1028,7 @@ if (mp == 0) { SBLASTRECORDCHK(&so->so_rcv); SBLASTMBUFCHK(&so->so_rcv); - splx(s); + SOCKBUF_UNLOCK(&so->so_rcv); #ifdef ZERO_COPY_SOCKETS if (so_zero_copy_receive) { vm_page_t pg; @@ -1021,7 +1052,7 @@ } else #endif /* ZERO_COPY_SOCKETS */ error = uiomove(mtod(m, char *) + moff, (int)len, uio); - s = splnet(); + SOCKBUF_LOCK(&so->so_rcv); if (error) goto release; } else @@ -1102,9 +1133,8 @@ SBLASTMBUFCHK(&so->so_rcv); error = sbwait(&so->so_rcv); if (error) { - sbunlock(&so->so_rcv); - splx(s); - return (0); + error = 0; + goto release; } m = so->so_rcv.sb_mb; if (m) @@ -1137,17 +1167,15 @@ (*pr->pr_usrreqs->pru_rcvd)(so, flags); } if (orig_resid == uio->uio_resid && orig_resid && - (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) { - sbunlock(&so->so_rcv); - splx(s); - goto restart; - } + (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) + goto restart; /* XXX multi-counts msgs */ if (flagsp) *flagsp |= flags; release: sbunlock(&so->so_rcv); - splx(s); +out: + SOCKBUF_UNLOCK(&so->so_rcv); return (error); } @@ -1174,22 +1202,21 @@ { struct sockbuf *sb = &so->so_rcv; struct protosw *pr = so->so_proto; - int s; struct sockbuf asb; + SOCKBUF_LOCK(sb); sb->sb_flags |= SB_NOINTR; (void) sblock(sb, M_WAITOK); - s = splimp(); - socantrcvmore(so); + socantrcvmore_locked(so); sbunlock(sb); asb = *sb; /* - * Invalidate/clear most of the sockbuf structure, but keep - * its selinfo structure valid. + * Invalidate/clear most of the sockbuf structure, but leave + * selinfo and mutex data unchanged. */ bzero(&sb->sb_startzero, sizeof(*sb) - offsetof(struct sockbuf, sb_startzero)); - splx(s); + SOCKBUF_UNLOCK(sb); if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose) (*pr->pr_domain->dom_dispose)(asb.sb_mb); @@ -1207,6 +1234,7 @@ struct so_accf *af = so->so_accf; int error = 0; +/* XXX locking */ /* do not set/remove accept filters on non listen sockets */ if ((so->so_options & SO_ACCEPTCONN) == 0) { error = EINVAL; @@ -1803,7 +1831,6 @@ { struct socket *so = kn->kn_fp->f_data; struct sockbuf *sb; - int s; switch (kn->kn_filter) { case EVFILT_READ: @@ -1821,10 +1848,10 @@ return (1); } - s = splnet(); + SOCKBUF_LOCK(sb); SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext); sb->sb_flags |= SB_KNOTE; - splx(s); + SOCKBUF_UNLOCK(sb); return (0); } @@ -1832,12 +1859,12 @@ filt_sordetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; - int s = splnet(); + SOCKBUF_LOCK(&so->so_rcv); SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext); if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note)) so->so_rcv.sb_flags &= ~SB_KNOTE; - splx(s); + SOCKBUF_UNLOCK(&so->so_rcv); } /*ARGSUSED*/ @@ -1845,30 +1872,35 @@ filt_soread(struct knote *kn, long hint) { struct socket *so = kn->kn_fp->f_data; + int result; + SOCKBUF_LOCK(&so->so_rcv); /* XXX too conservative? */ kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl; if (so->so_state & SS_CANTRCVMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; - return (1); + result = 1; + } else if (so->so_error) { /* temporary udp error */ + result = 1; + } else if (kn->kn_sfflags & NOTE_LOWAT) { + result = (kn->kn_data >= kn->kn_sdata); + } else { + result = (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); } - if (so->so_error) /* temporary udp error */ - return (1); - if (kn->kn_sfflags & NOTE_LOWAT) - return (kn->kn_data >= kn->kn_sdata); - return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat); + SOCKBUF_UNLOCK(&so->so_rcv); + return (result); } static void filt_sowdetach(struct knote *kn) { struct socket *so = kn->kn_fp->f_data; - int s = splnet(); + SOCKBUF_LOCK(&so->so_snd); SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext); if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note)) so->so_snd.sb_flags &= ~SB_KNOTE; - splx(s); + SOCKBUF_UNLOCK(&so->so_snd); } /*ARGSUSED*/ @@ -1876,21 +1908,26 @@ filt_sowrite(struct knote *kn, long hint) { struct socket *so = kn->kn_fp->f_data; + int result; + SOCKBUF_LOCK(&so->so_snd); /* XXX too conservative? */ kn->kn_data = sbspace(&so->so_snd); if (so->so_state & SS_CANTSENDMORE) { kn->kn_flags |= EV_EOF; kn->kn_fflags = so->so_error; - return (1); + result = 1; + } else if (so->so_error) { /* temporary udp error */ + result = 1; + } else if (((so->so_state & SS_ISCONNECTED) == 0) && + (so->so_proto->pr_flags & PR_CONNREQUIRED)) { + result = 0; + } else if (kn->kn_sfflags & NOTE_LOWAT) { + result = (kn->kn_data >= kn->kn_sdata); + } else { + result = (kn->kn_data >= so->so_snd.sb_lowat); } - if (so->so_error) /* temporary udp error */ - return (1); - if (((so->so_state & SS_ISCONNECTED) == 0) && - (so->so_proto->pr_flags & PR_CONNREQUIRED)) - return (0); - if (kn->kn_sfflags & NOTE_LOWAT) - return (kn->kn_data >= kn->kn_sdata); - return (kn->kn_data >= so->so_snd.sb_lowat); + SOCKBUF_UNLOCK(&so->so_snd); + return (result); } /*ARGSUSED*/ --- //depot/vendor/freebsd/src/sys/kern/uipc_socket2.c 2004/03/22 02:21:05 +++ //depot/user/rwatson/netperf/sys/kern/uipc_socket2.c 2004/03/22 08:18:59 @@ -108,9 +108,14 @@ soisconnecting(so) register struct socket *so; { + int need_lock = !SOCK_OWNED(so); + if (need_lock) + SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); so->so_state |= SS_ISCONNECTING; + if (need_lock) + SOCK_UNLOCK(so); } void @@ -118,56 +123,79 @@ struct socket *so; { struct socket *head = so->so_head; + int need_lock = !SOCK_OWNED(so); + if (need_lock) + SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); so->so_state |= SS_ISCONNECTED; if (head && (so->so_state & SS_INCOMP)) { - if ((so->so_options & SO_ACCEPTFILTER) != 0) { + if ((so->so_options & SO_ACCEPTFILTER) == 0) { + if (need_lock) + SOCK_UNLOCK(so); + SOCK_LOCK(head); + TAILQ_REMOVE(&head->so_incomp, so, so_list); + head->so_incqlen--; + so->so_state &= ~SS_INCOMP; + TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); + head->so_qlen++; + so->so_state |= SS_COMP; + sorwakeup_locked(head); + wakeup_one(&head->so_timeo); + SOCK_UNLOCK(head); + } else { +/* XXX locking */ so->so_upcall = head->so_accf->so_accept_filter->accf_callback; so->so_upcallarg = head->so_accf->so_accept_filter_arg; so->so_rcv.sb_flags |= SB_UPCALL; so->so_options &= ~SO_ACCEPTFILTER; so->so_upcall(so, so->so_upcallarg, M_TRYWAIT); - return; } - TAILQ_REMOVE(&head->so_incomp, so, so_list); - head->so_incqlen--; - so->so_state &= ~SS_INCOMP; - TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); - head->so_qlen++; - so->so_state |= SS_COMP; - sorwakeup(head); - wakeup_one(&head->so_timeo); } else { wakeup(&so->so_timeo); + SOCK_UNLOCK(so); sorwakeup(so); sowwakeup(so); } + if (!need_lock) + SOCK_LOCK(so); } void soisdisconnecting(so) register struct socket *so; { + int need_lock = !SOCK_OWNED(so); + if (need_lock) + SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTING; so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); wakeup(&so->so_timeo); + SOCK_UNLOCK(so); sowwakeup(so); sorwakeup(so); + if (!need_lock) + SOCK_LOCK(so); } void soisdisconnected(so) register struct socket *so; { + int need_lock = !SOCK_OWNED(so); + if (need_lock) + SOCK_LOCK(so); so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); wakeup(&so->so_timeo); + SOCK_UNLOCK(so); sbdrop(&so->so_snd, so->so_snd.sb_cc); sowwakeup(so); sorwakeup(so); + if (!need_lock) + SOCK_LOCK(so); } /* @@ -186,8 +214,12 @@ int connstatus; { register struct socket *so; + int over; - if (head->so_qlen > 3 * head->so_qlimit / 2) + SOCK_LOCK(head); + over = (head->so_qlen > 3 * head->so_qlimit / 2); + SOCK_UNLOCK(head); + if (over) return ((struct socket *)0); so = soalloc(M_NOWAIT); if (so == NULL) @@ -205,12 +237,13 @@ #ifdef MAC mac_create_socket_from_socket(head, so); #endif + if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) || (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) { sodealloc(so); return ((struct socket *)0); } - + SOCKBUF_LOCK(&head->so_rcv); if (connstatus) { TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); so->so_state |= SS_COMP; @@ -225,10 +258,11 @@ so->so_state |= SS_INCOMP; head->so_incqlen++; } + SOCKBUF_UNLOCK(&head->so_rcv); if (connstatus) { + so->so_state |= connstatus; sorwakeup(head); wakeup(&head->so_timeo); - so->so_state |= connstatus; } return (so); } @@ -253,6 +287,16 @@ } void +socantsendmore_locked(so) + struct socket *so; +{ + SOCKBUF_LOCK_ASSERT(&so->so_snd); + + so->so_state |= SS_CANTSENDMORE; + sowwakeup(so); +} + +void socantrcvmore(so) struct socket *so; { @@ -261,6 +305,16 @@ sorwakeup(so); } +void +socantrcvmore_locked(so) + struct socket *so; +{ + SOCKBUF_LOCK_ASSERT(&so->so_rcv); + + so->so_state |= SS_CANTRCVMORE; + sorwakeup_locked(so); +} + /* * Wait for data to arrive at/drain from a socket buffer. */ @@ -268,9 +322,10 @@ sbwait(sb) struct sockbuf *sb; { + SOCKBUF_LOCK_ASSERT(sb); sb->sb_flags |= SB_WAIT; - return (tsleep(&sb->sb_cc, + return (msleep(&sb->sb_cc, &sb->sb_mtx, (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait", sb->sb_timeo)); } @@ -285,9 +340,11 @@ { int error; + SOCKBUF_LOCK_ASSERT(sb); + while (sb->sb_flags & SB_LOCK) { sb->sb_flags |= SB_WANT; - error = tsleep(&sb->sb_flags, + error = msleep(&sb->sb_flags, &sb->sb_mtx, (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH, "sblock", 0); if (error) @@ -298,29 +355,71 @@ } /* + * The part of sowakeup that must be done while + * holding the sockbuf lock. + */ +static __inline void +sowakeup_under_lock(struct socket *so, struct sockbuf *sb) +{ + SOCKBUF_LOCK_ASSERT(sb); + + selwakeuppri(&sb->sb_sel, PSOCK); + sb->sb_flags &= ~SB_SEL; + if (sb->sb_flags & SB_WAIT) { + sb->sb_flags &= ~SB_WAIT; + wakeup(&sb->sb_cc); + } +} + +/* * Wakeup processes waiting on a socket buffer. * Do asynchronous notification via SIGIO * if the socket has the SS_ASYNC flag set. + * + * The caller is assumed to hold the necessary + * sockbuf lock. */ void +sowakeup_locked(so, sb) + register struct socket *so; + register struct sockbuf *sb; +{ + + sowakeup_under_lock(so, sb); + + if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) + pgsigio(&so->so_sigio, SIGIO, 0); + if (sb->sb_flags & SB_UPCALL) + (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); + if (sb->sb_flags & SB_AIO) /* XXX locking */ + aio_swake(so, sb); + KNOTE(&sb->sb_sel.si_note, 0); /* XXX locking? */ +} + +/* + * Wakeup processes waiting on a socket buffer. + * Do asynchronous notification via SIGIO + * if the socket has the SS_ASYNC flag set. + * + * The caller does not hold the sockbuf lock. + */ +void sowakeup(so, sb) register struct socket *so; register struct sockbuf *sb; { - selwakeuppri(&sb->sb_sel, PSOCK); - sb->sb_flags &= ~SB_SEL; - if (sb->sb_flags & SB_WAIT) { - sb->sb_flags &= ~SB_WAIT; - wakeup(&sb->sb_cc); - } + SOCKBUF_LOCK(sb); + sowakeup_under_lock(so, sb); + SOCKBUF_UNLOCK(sb); + if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) pgsigio(&so->so_sigio, SIGIO, 0); if (sb->sb_flags & SB_UPCALL) (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT); - if (sb->sb_flags & SB_AIO) + if (sb->sb_flags & SB_AIO) /* XXX locking */ aio_swake(so, sb); - KNOTE(&sb->sb_sel.si_note, 0); + KNOTE(&sb->sb_sel.si_note, 0); /* XXX locking? */ } /* @@ -480,6 +579,8 @@ { struct mbuf *m = sb->sb_mb; + SOCKBUF_LOCK_ASSERT(sb); + while (m && m->m_nextpkt) m = m->m_nextpkt; @@ -499,6 +600,8 @@ struct mbuf *m = sb->sb_mb; struct mbuf *n; + SOCKBUF_LOCK_ASSERT(sb); + while (m && m->m_nextpkt) m = m->m_nextpkt; @@ -535,7 +638,7 @@ * discarded and mbufs are compacted where possible. */ void -sbappend(sb, m) +sbappend_locked(sb, m) struct sockbuf *sb; struct mbuf *m; { @@ -543,6 +646,9 @@ if (m == 0) return; + + SOCKBUF_LOCK_ASSERT(sb); + SBLASTRECORDCHK(sb); n = sb->sb_mb; if (n) { @@ -550,7 +656,7 @@ n = n->m_nextpkt; do { if (n->m_flags & M_EOR) { - sbappendrecord(sb, m); /* XXXXXX!!!! */ + sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); @@ -563,7 +669,7 @@ if ((n = sb->sb_lastrecord) != NULL) { do { if (n->m_flags & M_EOR) { - sbappendrecord(sb, m); /* XXXXXX!!!! */ + sbappendrecord_locked(sb, m); /* XXXXXX!!!! */ return; } } while (n->m_next && (n = n->m_next)); @@ -580,13 +686,33 @@ } /* + * Append mbuf chain m to the last record in the + * socket buffer sb. The additional space associated + * the mbuf chain is recorded in sb. Empty mbufs are + * discarded and mbufs are compacted where possible. + */ +void +sbappend(sb, m) + struct sockbuf *sb; + struct mbuf *m; +{ + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + sbappend_locked(sb, m); + SOCKBUF_UNLOCK(sb); + } else + sbappend_locked(sb, m); +} + +/* * This version of sbappend() should only be used when the caller * absolutely knows that there will never be more than one record * in the socket buffer, that is, a stream protocol (such as TCP). */ void -sbappendstream(struct sockbuf *sb, struct mbuf *m) +sbappendstream_locked(struct sockbuf *sb, struct mbuf *m) { + SOCKBUF_LOCK_ASSERT(sb); KASSERT(m->m_nextpkt == NULL,("sbappendstream 0")); KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1")); @@ -599,6 +725,22 @@ SBLASTRECORDCHK(sb); } +/* + * This version of sbappend() should only be used when the caller + * absolutely knows that there will never be more than one record + * in the socket buffer, that is, a stream protocol (such as TCP). + */ +void +sbappendstream(struct sockbuf *sb, struct mbuf *m) +{ + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + sbappendstream_locked(sb, m); + SOCKBUF_UNLOCK(sb); + } else + sbappendstream_locked(sb, m); +} + #ifdef SOCKBUF_DEBUG void sbcheck(sb) @@ -608,6 +750,8 @@ struct mbuf *n = 0; u_long len = 0, mbcnt = 0; + SOCKBUF_LOCK_ASSERT(sb); + for (m = sb->sb_mb; m; m = n) { n = m->m_nextpkt; for (; m; m = m->m_next) { @@ -630,12 +774,14 @@ * begins a new record. */ void -sbappendrecord(sb, m0) +sbappendrecord_locked(sb, m0) register struct sockbuf *sb; register struct mbuf *m0; { register struct mbuf *m; + SOCKBUF_LOCK_ASSERT(sb); + if (m0 == 0) return; m = sb->sb_mb; @@ -663,18 +809,37 @@ } /* + * As above, except the mbuf chain + * begins a new record. + */ +void +sbappendrecord(sb, m0) + register struct sockbuf *sb; + register struct mbuf *m0; +{ + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + sbappendrecord_locked(sb, m0); + SOCKBUF_UNLOCK(sb); + } else + sbappendrecord_locked(sb, m0); +} + +/* * As above except that OOB data * is inserted at the beginning of the sockbuf, * but after any other OOB data. */ void -sbinsertoob(sb, m0) +sbinsertoob_locked(sb, m0) register struct sockbuf *sb; register struct mbuf *m0; { register struct mbuf *m; register struct mbuf **mp; + SOCKBUF_LOCK_ASSERT(sb); + if (m0 == 0) return; for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) { @@ -709,13 +874,31 @@ } /* + * As above except that OOB data + * is inserted at the beginning of the sockbuf, + * but after any other OOB data. + */ +void +sbinsertoob(sb, m0) + register struct sockbuf *sb; + register struct mbuf *m0; +{ + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + sbinsertoob_locked(sb, m0); + SOCKBUF_UNLOCK(sb); + } else + sbinsertoob_locked(sb, m0); +} + +/* * Append address and data, and optionally, control (ancillary) data * to the receive queue of a socket. If present, * m0 must include a packet header with total length. * Returns 0 if no space in sockbuf or insufficient mbufs. */ int -sbappendaddr(sb, asa, m0, control) +sbappendaddr_locked(sb, asa, m0, control) struct sockbuf *sb; struct sockaddr *asa; struct mbuf *m0, *control; @@ -723,11 +906,14 @@ struct mbuf *m, *n, *nlast; int space = asa->sa_len; + SOCKBUF_LOCK_ASSERT(sb); + if (m0 && (m0->m_flags & M_PKTHDR) == 0) panic("sbappendaddr"); if (m0) space += m0->m_pkthdr.len; space += m_length(control, &n); + if (space > sbspace(sb)) return (0); #if MSIZE <= 256 @@ -749,25 +935,50 @@ sballoc(sb, n); nlast = n; SBLINKRECORD(sb, m); + sb->sb_mbtail = nlast; - sb->sb_mbtail = nlast; SBLASTMBUFCHK(sb); - SBLASTRECORDCHK(sb); return (1); } +/* + * Append address and data, and optionally, control (ancillary) data + * to the receive queue of a socket. If present, + * m0 must include a packet header with total length. + * Returns 0 if no space in sockbuf or insufficient mbufs. + */ +int +sbappendaddr(sb, asa, m0, control) + struct sockbuf *sb; + struct sockaddr *asa; + struct mbuf *m0, *control; +{ + int retval; + + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + retval = sbappendaddr_locked(sb, asa, m0, control); + SOCKBUF_UNLOCK(sb); + } else + retval = sbappendaddr_locked(sb, asa, m0, control); + return (retval); +} + int -sbappendcontrol(sb, m0, control) +sbappendcontrol_locked(sb, m0, control) struct sockbuf *sb; struct mbuf *control, *m0; { struct mbuf *m, *n, *mlast; int space; + SOCKBUF_LOCK_ASSERT(sb); + if (control == 0) panic("sbappendcontrol"); space = m_length(control, &n) + m_length(m0, NULL); + if (space > sbspace(sb)) return (0); n->m_next = m0; /* concatenate data to control */ @@ -779,14 +990,30 @@ sballoc(sb, m); mlast = m; SBLINKRECORD(sb, control); + sb->sb_mbtail = mlast; - sb->sb_mbtail = mlast; SBLASTMBUFCHK(sb); + SBLASTRECORDCHK(sb); - SBLASTRECORDCHK(sb); return (1); } +int +sbappendcontrol(sb, m0, control) + struct sockbuf *sb; + struct mbuf *control, *m0; +{ + int retval; + + if (!SOCKBUF_OWNED(sb)) { + SOCKBUF_LOCK(sb); + retval = sbappendcontrol(sb, m0, control); + SOCKBUF_UNLOCK(sb); + } else + retval = sbappendcontrol(sb, m0, control); + return (retval); +} + /* * Compress mbuf chain m into the socket * buffer sb following mbuf n. If n @@ -800,6 +1027,8 @@ register int eor = 0; register struct mbuf *o; + SOCKBUF_LOCK_ASSERT(sb); + while (m) { eor |= m->m_flags & M_EOR; if (m->m_len == 0 && @@ -868,7 +1097,8 @@ sbdrop(sb, (int)sb->sb_cc); } if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt) - panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); + panic("sbflush: cc %u || mb %p || mbcnt %u", + sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt); } /* @@ -881,7 +1111,11 @@ { register struct mbuf *m; struct mbuf *next; + int need_lock = !SOCKBUF_OWNED(sb); + if (need_lock) + SOCKBUF_LOCK(sb); + next = (m = sb->sb_mb) ? m->m_nextpkt : 0; while (len > 0) { if (m == 0) { @@ -925,6 +1159,9 @@ } else if (m->m_nextpkt == NULL) { sb->sb_lastrecord = m; } + + if (need_lock) + SOCKBUF_UNLOCK(sb); } /* @@ -936,7 +1173,11 @@ register struct sockbuf *sb; { register struct mbuf *m; + int need_lock = !SOCKBUF_OWNED(sb); + if (need_lock) + SOCKBUF_LOCK(sb); + m = sb->sb_mb; if (m) { sb->sb_mb = m->m_nextpkt; @@ -946,6 +1187,9 @@ } while (m); } SB_EMPTY_FIXUP(sb); + + if (need_lock) + SOCKBUF_UNLOCK(sb); } /* --- //depot/vendor/freebsd/src/sys/kern/uipc_syscalls.c 2004/03/28 18:25:22 +++ //depot/user/rwatson/netperf/sys/kern/uipc_syscalls.c 2004/03/28 18:34:51 @@ -178,6 +178,7 @@ if ((error = fgetsock(td, fd, &so, NULL)) != 0) goto done2; #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_bind(td->td_ucred, so, sa); if (error) goto done1; @@ -211,6 +212,7 @@ NET_LOCK_GIANT(); if ((error = fgetsock(td, uap->s, &so, NULL)) == 0) { #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_listen(td->td_ucred, so); if (error) goto done; @@ -265,7 +267,9 @@ if (error) goto done2; s = splnet(); + SOCK_LOCK(head); if ((head->so_options & SO_ACCEPTCONN) == 0) { + SOCK_UNLOCK(head); splx(s); error = EINVAL; goto done; @@ -279,9 +283,10 @@ head->so_error = EWOULDBLOCK; break; } - error = tsleep(&head->so_timeo, PSOCK | PCATCH, + error = msleep(&head->so_timeo, SOCK_MTX(head), PSOCK | PCATCH, "accept", 0); if (error) { + SOCK_UNLOCK(head); splx(s); goto done; } @@ -289,6 +294,7 @@ if (head->so_error) { error = head->so_error; head->so_error = 0; + SOCK_UNLOCK(head); splx(s); goto done; } @@ -298,11 +304,13 @@ * ready to be accepted. Remove it from the queue prior to * allocating the file descriptor for it since falloc() may * block allowing another process to accept the connection - * instead. + * instead. The reference previously owned by the socket queue + * is now thread-local, letting us release the lock on the head. */ so = TAILQ_FIRST(&head->so_comp); TAILQ_REMOVE(&head->so_comp, so, so_list); head->so_qlen--; + SOCK_UNLOCK(head); error = falloc(td, &nfp, &fd); if (error) { @@ -310,20 +318,27 @@ * Probably ran out of file descriptors. Put the * unaccepted connection back onto the queue and * do another wakeup so some other process might - * have a chance at it. + * have a chance at it. Note that strict ordering + * is lost. */ + SOCK_LOCK(head); TAILQ_INSERT_HEAD(&head->so_comp, so, so_list); head->so_qlen++; wakeup_one(&head->so_timeo); + SOCK_UNLOCK(head); splx(s); goto done; } /* An extra reference on `nfp' has been held for us by falloc(). */ td->td_retval[0] = fd; + /* XXX lock? */ /* connection has been removed from the listen queue */ KNOTE(&head->so_rcv.sb_sel.si_note, 0); + /* + * XXXRW: so should be locked to modify so_state here? + */ so->so_state &= ~SS_COMP; so->so_head = NULL; pgid = fgetown(&head->so_sigio); @@ -481,6 +496,7 @@ goto done1; } #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_connect(td->td_ucred, so, sa); if (error) goto bad; @@ -493,8 +509,10 @@ goto done1; } s = splnet(); + SOCK_LOCK(so); while ((so->so_state & SS_ISCONNECTING) && so->so_error == 0) { - error = tsleep(&so->so_timeo, PSOCK | PCATCH, "connec", 0); + error = msleep(&so->so_timeo, SOCK_MTX(so), PSOCK | PCATCH, + "connec", 0); if (error) { if (error == EINTR || error == ERESTART) interrupted = 1; @@ -505,6 +523,7 @@ error = so->so_error; so->so_error = 0; } + SOCK_UNLOCK(so); splx(s); bad: if (!interrupted) @@ -700,6 +719,7 @@ goto bad2; #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_send(td->td_ucred, so); if (error) goto bad; @@ -943,6 +963,7 @@ } #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_receive(td->td_ucred, so); if (error) { fputsock(so); @@ -1741,6 +1762,7 @@ } #ifdef MAC + /* XXXRW: MAC requires socket lock? */ error = mac_check_socket_send(td->td_ucred, so); if (error) goto done; @@ -1780,7 +1802,9 @@ /* * Protect against multiple writers to the socket. */ + SOCKBUF_LOCK(&so->so_snd); (void) sblock(&so->so_snd, M_WAITOK); + SOCKBUF_UNLOCK(&so->so_snd); /* * Loop through the pages in the file, starting with the requested @@ -1814,14 +1838,17 @@ * Optimize the non-blocking case by looking at the socket space * before going to the extra work of constituting the sf_buf. */ + SOCKBUF_LOCK(&so->so_snd); if ((so->so_state & SS_NBIO) && sbspace(&so->so_snd) <= 0) { if (so->so_state & SS_CANTSENDMORE) error = EPIPE; else error = EAGAIN; sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); goto done; } + SOCKBUF_UNLOCK(&so->so_snd); VM_OBJECT_LOCK(obj); /* * Attempt to look up the page. @@ -1910,7 +1937,9 @@ } vm_page_unlock_queues(); VM_OBJECT_UNLOCK(obj); + SOCKBUF_LOCK(&so->so_snd); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); goto done; } vm_page_unlock_queues(); @@ -1926,7 +1955,9 @@ if (pg->wire_count == 0 && pg->object == NULL) vm_page_free(pg); vm_page_unlock_queues(); + SOCKBUF_LOCK(&so->so_snd); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); error = EINTR; goto done; } @@ -1963,6 +1994,7 @@ * Add the buffer to the socket buffer chain. */ s = splnet(); + SOCKBUF_LOCK(&so->so_snd); retry_space: /* * Make sure that the socket is still able to take more data. @@ -1984,6 +2016,7 @@ } m_freem(m); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); splx(s); goto done; } @@ -1996,6 +2029,7 @@ if (so->so_state & SS_NBIO) { m_freem(m); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); splx(s); error = EAGAIN; goto done; @@ -2015,14 +2049,20 @@ goto retry_space; } error = (*so->so_proto->pr_usrreqs->pru_send)(so, 0, m, 0, 0, td); + /* XXX: Why release and re-grab? */ + SOCKBUF_UNLOCK(&so->so_snd); splx(s); if (error) { + SOCKBUF_LOCK(&so->so_snd); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); goto done; } headersent = 1; } + SOCKBUF_LOCK(&so->so_snd); sbunlock(&so->so_snd); + SOCKBUF_UNLOCK(&so->so_snd); /* * Send trailers. Wimp out and use writev(2). --- //depot/vendor/freebsd/src/sys/kern/uipc_usrreq.c 2004/02/29 19:15:33 +++ //depot/user/rwatson/netperf/sys/kern/uipc_usrreq.c 2004/02/29 21:00:26 @@ -85,6 +85,39 @@ static struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; static ino_t unp_ino; /* prototype for fake inode numbers */ +static struct mtx unp_mtx; +#define UNP_HEAD_LOCK_INIT() \ + mtx_init(&unp_mtx, "unp head", NULL, MTX_DEF) +#define UNP_HEAD_LOCK() mtx_lock(&unp_mtx) +#define UNP_HEAD_UNLOCK() mtx_unlock(&unp_mtx) +#define UNP_HEAD_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED) + +/* NB: DUPOK is to cover the connect2 case XXX */ +#define UNP_LOCK_INIT(_unp) \ + mtx_init(&(_unp)->unp_mtx, "unp", NULL, MTX_DEF | MTX_DUPOK) +#define UNP_LOCK_DESTROY(_unp) mtx_destroy(&(_unp)->unp_mtx) +#define UNP_LOCK(_unp) mtx_lock(&(_unp)->unp_mtx) +#define UNP_UNLOCK(_unp) mtx_unlock(&(_unp)->unp_mtx) +#define UNP_LOCK_ASSERT(_unp) mtx_assert(&(_unp)->unp_mtx, MA_OWNED) + +/* + * A unp lock is always preceded by locking the head. + * Since this occurs often we define convenienc macros + * for entry, exist, and validation in lower-level routines. + */ +#define UNP_ENTER(_unp) do { \ + UNP_HEAD_LOCK(); \ + UNP_LOCK(_unp); \ +} while (0) +#define UNP_EXIT(_unp) do { \ + UNP_UNLOCK(_unp); \ + UNP_HEAD_UNLOCK(); \ +} while (0) +#define UNP_ASSERT(_unp) do { \ + UNP_HEAD_LOCK_ASSERT(); \ + UNP_LOCK_ASSERT(_unp); \ +} while (0) + static int unp_attach(struct socket *); static void unp_detach(struct unpcb *); static int unp_bind(struct unpcb *,struct sockaddr *, struct thread *); @@ -107,8 +140,10 @@ if (unp == 0) return (EINVAL); + UNP_ENTER(unp); unp_drop(unp, ECONNABORTED); - unp_detach(unp); + unp_detach(unp); /* NB: unlocks unp + head */ + SOCK_LOCK(so); sotryfree(so); return (0); } @@ -117,6 +152,7 @@ uipc_accept(struct socket *so, struct sockaddr **nam) { struct unpcb *unp = sotounpcb(so); + struct sockaddr *sa; if (unp == 0) return (EINVAL); @@ -126,13 +162,14 @@ * if it was bound and we are still connected * (our peer may have closed already!). */ - if (unp->unp_conn && unp->unp_conn->unp_addr) { - *nam = sodupsockaddr( - (struct sockaddr *)unp->unp_conn->unp_addr, M_WAITOK); - } else { - *nam = sodupsockaddr((struct sockaddr *)&sun_noname, - M_WAITOK); - } + *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); + UNP_ENTER(unp); + if (unp->unp_conn && unp->unp_conn->unp_addr) + sa = (struct sockaddr *) unp->unp_conn->unp_addr; + else + sa = &sun_noname; + bcopy(sa, *nam, sa->sa_len); + UNP_EXIT(unp); return (0); } @@ -153,7 +190,6 @@ if (unp == 0) return (EINVAL); - return (unp_bind(unp, nam, td)); } @@ -161,21 +197,30 @@ uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { struct unpcb *unp = sotounpcb(so); + int retval; - if (unp == 0) - return (EINVAL); - return (unp_connect(so, nam, curthread)); + if (unp != 0) { + UNP_ENTER(unp); + retval = unp_connect(so, nam, curthread); + UNP_EXIT(unp); + } else + retval = EINVAL; + return (retval); } static int uipc_connect2(struct socket *so1, struct socket *so2) { struct unpcb *unp = sotounpcb(so1); + int retval; - if (unp == 0) - return (EINVAL); - - return (unp_connect2(so1, so2)); + if (unp != 0) { + UNP_ENTER(unp); + retval = unp_connect2(so1, so2); + UNP_EXIT(unp); + } else + retval = EINVAL; + return (retval); } /* control is EOPNOTSUPP */ @@ -187,8 +232,8 @@ if (unp == 0) return (EINVAL); - - unp_detach(unp); + UNP_ENTER(unp); + unp_detach(unp); /* NB: unlocks unp + head */ return (0); } @@ -197,41 +242,52 @@ { struct unpcb *unp = sotounpcb(so); - if (unp == 0) + if (unp != 0) { + UNP_ENTER(unp); + unp_disconnect(unp); + UNP_EXIT(unp); + return (0); + } else return (EINVAL); - unp_disconnect(unp); - return (0); } static int uipc_listen(struct socket *so, struct thread *td) { struct unpcb *unp = sotounpcb(so); + int retval; - if (unp == 0 || unp->unp_vnode == 0) + if (unp != 0 && unp->unp_vnode != 0) { + UNP_ENTER(unp); + retval = unp_listen(unp, td); + UNP_EXIT(unp); + } else return (EINVAL); - return (unp_listen(unp, td)); + return (retval); } static int uipc_peeraddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp = sotounpcb(so); + struct sockaddr *sa; if (unp == 0) return (EINVAL); + *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); + UNP_ENTER(unp); if (unp->unp_conn && unp->unp_conn->unp_addr) - *nam = sodupsockaddr( - (struct sockaddr *)unp->unp_conn->unp_addr, M_WAITOK); + sa = (struct sockaddr *) unp->unp_conn->unp_addr; else { /* * XXX: It seems that this test always fails even when * connection is established. So, this else clause is * added as workaround to return PF_LOCAL sockaddr. */ - *nam = sodupsockaddr((struct sockaddr *)&sun_noname, - M_WAITOK); + sa = &sun_noname; } + bcopy(sa, *nam, sa->sa_len); + UNP_EXIT(unp); return (0); } @@ -244,15 +300,27 @@ if (unp == 0) return (EINVAL); + /* + * Reorder locks to avoid LORs. Note that we + * delay re-locking so_rcv to below so it can + * be done only once. + */ + SOCKBUF_UNLOCK(&so->so_rcv); + UNP_ENTER(unp); switch (so->so_type) { case SOCK_DGRAM: panic("uipc_rcvd DGRAM?"); /*NOTREACHED*/ case SOCK_STREAM: - if (unp->unp_conn == 0) + if (unp->unp_conn == 0) { + SOCKBUF_LOCK(&so->so_rcv); break; + } so2 = unp->unp_conn->unp_socket; + /* NB: careful of order here */ + SOCKBUF_LOCK(&so2->so_snd); + SOCKBUF_LOCK(&so->so_rcv); /* * Adjust backpressure on sender * and wakeup any waiting to write. @@ -264,12 +332,14 @@ (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat, newhiwat, RLIM_INFINITY); unp->unp_cc = so->so_rcv.sb_cc; - sowwakeup(so2); + sowwakeup_locked(so2); + SOCKBUF_UNLOCK(&so2->so_snd); break; default: panic("uipc_rcvd unknown socktype"); } + UNP_EXIT(unp); return (0); } @@ -296,6 +366,13 @@ if (control && (error = unp_internalize(&control, td))) goto release; + /* + * Reorder locks to avoid LORs. + */ + SOCKBUF_UNLOCK(&so->so_snd); + UNP_ENTER(unp); + SOCKBUF_LOCK(&so->so_snd); + switch (so->so_type) { case SOCK_DGRAM: { @@ -306,7 +383,9 @@ error = EISCONN; break; } + SOCKBUF_UNLOCK(&so->so_snd); error = unp_connect(so, nam, td); + SOCKBUF_LOCK(&so->so_snd); if (error) break; } else { @@ -320,13 +399,14 @@ from = (struct sockaddr *)unp->unp_addr; else from = &sun_noname; - if (sbappendaddr(&so2->so_rcv, from, m, control)) { - sorwakeup(so2); + SOCKBUF_LOCK(&so2->so_rcv); + if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { + sorwakeup_locked(so2); m = 0; control = 0; - } else { + } else error = ENOBUFS; - } + SOCKBUF_UNLOCK(&so2->so_rcv); if (nam) unp_disconnect(unp); break; @@ -340,7 +420,9 @@ */ if ((so->so_state & SS_ISCONNECTED) == 0) { if (nam) { + SOCKBUF_UNLOCK(&so->so_snd); error = unp_connect(so, nam, td); + SOCKBUF_LOCK(&so->so_snd); if (error) break; /* XXX */ } else { @@ -356,17 +438,17 @@ if (unp->unp_conn == 0) panic("uipc_send connected but no connection?"); so2 = unp->unp_conn->unp_socket; + SOCKBUF_LOCK(&so2->so_rcv); /* * Send to paired receive port, and then reduce * send buffer hiwater marks to maintain backpressure. * Wake up readers. */ if (control) { - if (sbappendcontrol(&so2->so_rcv, m, control)) + if (sbappendcontrol_locked(&so2->so_rcv, m, control)) control = 0; - } else { - sbappend(&so2->so_rcv, m); - } + } else + sbappend_locked(&so2->so_rcv, m); so->so_snd.sb_mbmax -= so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt; unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt; @@ -375,7 +457,8 @@ (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat, newhiwat, RLIM_INFINITY); unp->unp_conn->unp_cc = so2->so_rcv.sb_cc; - sorwakeup(so2); + sorwakeup_locked(so2); + SOCKBUF_UNLOCK(&so2->so_rcv); m = 0; break; @@ -388,13 +471,13 @@ * a SHUTDOWN. */ if (flags & PRUS_EOF) { - socantsendmore(so); + socantsendmore_locked(so); unp_shutdown(unp); } + UNP_EXIT(unp); if (control && error != 0) - unp_dispose(control); - + unp_dispose(control); /* XXX need head lock? */ release: if (control) m_freem(control); @@ -411,6 +494,7 @@ if (unp == 0) return (EINVAL); + UNP_ENTER(unp); sb->st_blksize = so->so_snd.sb_hiwat; if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) { so2 = unp->unp_conn->unp_socket; @@ -420,6 +504,7 @@ if (unp->unp_ino == 0) unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino; sb->st_ino = unp->unp_ino; + UNP_EXIT(unp); return (0); } @@ -430,8 +515,11 @@ if (unp == 0) return (EINVAL); + UNP_ENTER(unp); + /* XXX socket lock? */ socantsendmore(so); unp_shutdown(unp); + UNP_EXIT(unp); return (0); } @@ -439,15 +527,18 @@ uipc_sockaddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp = sotounpcb(so); + struct sockaddr *sa; if (unp == 0) return (EINVAL); + *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); + UNP_ENTER(unp); if (unp->unp_addr) - *nam = sodupsockaddr((struct sockaddr *)unp->unp_addr, - M_WAITOK); + sa = (struct sockaddr *) unp->unp_addr; else - *nam = sodupsockaddr((struct sockaddr *)&sun_noname, - M_WAITOK); + sa = &sun_noname; + bcopy(sa, *nam, sa->sa_len); + UNP_EXIT(unp); return (0); } @@ -469,6 +560,7 @@ switch (sopt->sopt_dir) { case SOPT_GET: + UNP_ENTER(unp); switch (sopt->sopt_name) { case LOCAL_PEERCRED: if (unp->unp_flags & UNP_HAVEPC) @@ -485,6 +577,7 @@ error = EOPNOTSUPP; break; } + UNP_EXIT(unp); break; case SOPT_SET: default: @@ -557,8 +650,13 @@ unp_count++; LIST_INIT(&unp->unp_refs); unp->unp_socket = so; + UNP_LOCK_INIT(unp); + + UNP_HEAD_LOCK(); LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead, unp, unp_link); + UNP_HEAD_UNLOCK(); + so->so_pcb = unp; return (0); } @@ -567,18 +665,25 @@ unp_detach(unp) register struct unpcb *unp; { + struct vnode *vp; + + UNP_ASSERT(unp); + LIST_REMOVE(unp, unp_link); unp->unp_gencnt = ++unp_gencnt; --unp_count; - if (unp->unp_vnode) { + if ((vp = unp->unp_vnode)) { unp->unp_vnode->v_socket = 0; - vrele(unp->unp_vnode); unp->unp_vnode = 0; } if (unp->unp_conn) unp_disconnect(unp); - while (!LIST_EMPTY(&unp->unp_refs)) - unp_drop(LIST_FIRST(&unp->unp_refs), ECONNRESET); + while (!LIST_EMPTY(&unp->unp_refs)) { + struct unpcb *ref = LIST_FIRST(&unp->unp_refs); + UNP_LOCK(ref); + unp_drop(ref, ECONNRESET); + UNP_UNLOCK(ref); + } soisdisconnected(unp->unp_socket); unp->unp_socket->so_pcb = 0; if (unp_rights) { @@ -594,7 +699,11 @@ } if (unp->unp_addr) FREE(unp->unp_addr, M_SONAME); + UNP_LOCK_DESTROY(unp); + UNP_HEAD_UNLOCK(); uma_zfree(unp_zone, unp); + if (vp) + vrele(vp); } static int @@ -621,15 +730,14 @@ buf = malloc(namelen + 1, M_TEMP, M_WAITOK); strlcpy(buf, soun->sun_path, namelen + 1); + mtx_lock(&Giant); restart: NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE, buf, td); /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ error = namei(&nd); - if (error) { - free(buf, M_TEMP); - return (error); - } + if (error) + goto done; vp = nd.ni_vp; if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); @@ -639,14 +747,12 @@ vput(nd.ni_dvp); if (vp != NULL) { vrele(vp); - free(buf, M_TEMP); - return (EADDRINUSE); + error = EADDRINUSE; + goto done; } error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); - if (error) { - free(buf, M_TEMP); - return (error); - } + if (error) + goto done; goto restart; } VATTR_NULL(&vattr); @@ -662,18 +768,21 @@ } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); - if (error) { - free(buf, M_TEMP); - return (error); - } + if (error) + goto done; vp = nd.ni_vp; + soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); + UNP_ENTER(unp); vp->v_socket = unp->unp_socket; unp->unp_vnode = vp; - unp->unp_addr = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); + unp->unp_addr = soun; + UNP_EXIT(unp); VOP_UNLOCK(vp, 0, td); vn_finished_write(mp); +done: + mtx_unlock(&Giant); free(buf, M_TEMP); - return (0); + return (error); } static int @@ -685,10 +794,14 @@ register struct sockaddr_un *soun = (struct sockaddr_un *)nam; register struct vnode *vp; register struct socket *so2, *so3; - struct unpcb *unp, *unp2, *unp3; + struct unpcb *unp = sotounpcb(so); + struct unpcb *unp2, *unp3; int error, len; struct nameidata nd; char buf[SOCK_MAXADDRLEN]; + struct sockaddr *sa; + + UNP_ASSERT(unp); len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); if (len <= 0) @@ -696,9 +809,17 @@ strlcpy(buf, soun->sun_path, len + 1); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td); + /* drop locks across namei */ + UNP_EXIT(unp); + mtx_lock(&Giant); error = namei(&nd); + if (!error) + sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); + else + sa = NULL; + UNP_ENTER(unp); if (error) - return (error); + goto bad2; vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (vp->v_type != VSOCK) { @@ -718,19 +839,30 @@ goto bad; } if (so->so_proto->pr_flags & PR_CONNREQUIRED) { - if ((so2->so_options & SO_ACCEPTCONN) == 0 || - (so3 = sonewconn(so2, 0)) == 0) { + if (so2->so_options & SO_ACCEPTCONN) { + /* + * NB: drop locks here so unp_attach is entered + * w/o locks; this avoids a recursive lock + * of the head and holding sleep locks across + * a (potentially) blocking malloc. + */ + UNP_EXIT(unp); + so3 = sonewconn(so2, 0); + UNP_ENTER(unp); + } else + so3 = 0; + if (so3 == 0) { error = ECONNREFUSED; goto bad; } unp = sotounpcb(so); unp2 = sotounpcb(so2); unp3 = sotounpcb(so3); - if (unp2->unp_addr) - unp3->unp_addr = (struct sockaddr_un *) - sodupsockaddr((struct sockaddr *)unp2->unp_addr, - M_WAITOK); - + if (unp2->unp_addr) { + bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); + unp3->unp_addr = (struct sockaddr_un *) sa; + sa = NULL; + } /* * unp_peercred management: * @@ -762,6 +894,10 @@ error = unp_connect2(so, so2); bad: vput(vp); +bad2: + if (sa) + free(sa, M_SONAME); + mtx_unlock(&Giant); return (error); } @@ -773,9 +909,12 @@ register struct unpcb *unp = sotounpcb(so); register struct unpcb *unp2; + UNP_ASSERT(unp); + if (so2->so_type != so->so_type) return (EPROTOTYPE); unp2 = sotounpcb(so2); + UNP_LOCK(unp2); unp->unp_conn = unp2; switch (so->so_type) { @@ -793,6 +932,7 @@ default: panic("unp_connect2"); } + UNP_UNLOCK(unp2); return (0); } @@ -802,6 +942,8 @@ { register struct unpcb *unp2 = unp->unp_conn; + UNP_ASSERT(unp); + if (unp2 == 0) return; unp->unp_conn = 0; @@ -814,10 +956,13 @@ case SOCK_STREAM: soisdisconnected(unp->unp_socket); + UNP_LOCK(unp2); unp2->unp_conn = 0; soisdisconnected(unp2->unp_socket); + UNP_UNLOCK(unp2); break; } + return; } #ifdef notdef @@ -860,8 +1005,10 @@ * OK, now we're committed to doing something. */ xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK); + UNP_HEAD_LOCK(); gencnt = unp_gencnt; n = unp_count; + UNP_HEAD_UNLOCK(); xug->xug_len = sizeof *xug; xug->xug_count = n; @@ -875,6 +1022,7 @@ unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); + UNP_HEAD_LOCK(); for (unp = LIST_FIRST(head), i = 0; unp && i < n; unp = LIST_NEXT(unp, unp_link)) { if (unp->unp_gencnt <= gencnt) { @@ -884,6 +1032,7 @@ unp_list[i++] = unp; } } + UNP_HEAD_UNLOCK(); n = i; /* in case we lost some during malloc */ error = 0; @@ -941,6 +1090,8 @@ { struct socket *so; + UNP_ASSERT(unp); + if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn && (so = unp->unp_conn->unp_socket)) socantrcvmore(so); @@ -953,6 +1104,8 @@ { struct socket *so = unp->unp_socket; + UNP_ASSERT(unp); + so->so_error = errno; unp_disconnect(unp); } @@ -1104,6 +1257,8 @@ uma_zone_set_max(unp_zone, nmbclusters); LIST_INIT(&unp_dhead); LIST_INIT(&unp_shead); + + UNP_HEAD_LOCK_INIT(); } static int @@ -1260,6 +1415,8 @@ struct file **extra_ref, **fpp; int nunref, i; + UNP_HEAD_LOCK_ASSERT(); /* NB: this serializes entry */ + if (unp_gcing) return; unp_gcing = 1; @@ -1454,6 +1611,7 @@ struct unpcb *unp; struct thread *td; { + UNP_ASSERT(unp); cru2x(td->td_ucred, &unp->unp_peercred); unp->unp_flags |= UNP_HAVEPCCACHED; --- //depot/vendor/freebsd/src/sys/net/bpf.c 2004/02/29 07:35:30 +++ //depot/user/rwatson/netperf/sys/net/bpf.c 2004/03/01 16:13:52 @@ -557,7 +557,7 @@ struct ifnet *ifp; struct mbuf *m; int error; - static struct sockaddr dst; + struct sockaddr dst; int datlen; if (d->bd_bif == 0) @@ -568,6 +568,7 @@ if (uio->uio_resid == 0) return (0); + bzero(&dst, sizeof(dst)); error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen); if (error) return (error); @@ -583,12 +584,10 @@ mac_create_mbuf_from_bpfdesc(d, m); BPFD_UNLOCK(d); #endif - mtx_lock(&Giant); + /* NB: the driver frees the mbuf */ + NET_LOCK_GIANT(); error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0); - mtx_unlock(&Giant); - /* - * The driver frees the mbuf. - */ + NET_UNLOCK_GIANT(); return (error); } --- //depot/vendor/freebsd/src/sys/net/if.c 2004/03/12 18:35:54 +++ //depot/user/rwatson/netperf/sys/net/if.c 2004/03/12 22:29:12 @@ -659,6 +659,7 @@ /* * Create a clone network interface. + * XXXRW: Locking? */ int if_clone_create(char *name, int len) @@ -731,6 +732,7 @@ /* * Destroy a clone network interface. + * XXXRW: Locking? */ int if_clone_destroy(const char *name) @@ -768,6 +770,7 @@ /* * Look up a network interface cloner. + * XXXRW: Locking? */ static struct if_clone * if_clone_lookup(const char *name, int *unitp) @@ -809,6 +812,7 @@ /* * Register a network interface cloner. + * XXXRW: Locking? */ void if_clone_attach(struct if_clone *ifc) @@ -851,6 +855,7 @@ /* * Unregister a network interface cloner. + * XXXRW: Locking? */ void if_clone_detach(struct if_clone *ifc) @@ -863,6 +868,7 @@ /* * Provide list of interface cloners to userspace. + * XXXRW: Locking? */ static int if_clone_list(struct if_clonereq *ifcr) --- //depot/vendor/freebsd/src/sys/net/if_gif.c 2004/03/22 07:46:21 +++ //depot/user/rwatson/netperf/sys/net/if_gif.c 2004/03/22 08:18:59 @@ -88,9 +88,14 @@ * XXX: See comment blow on gif_called. * XXX: Per-softc locking is still required. */ +/* + * XXXRW: Note that gif_mtx only protects global gif-related data, not + * per-softc data. See also netinet/in_gif.c for locking needs. + */ static struct mtx gif_mtx; static MALLOC_DEFINE(M_GIF, "gif", "Generic Tunnel Interface"); static LIST_HEAD(, gif_softc) gif_softc_list; +static int gif_called = 0; /* * XXX: gif_called is a recursion counter to prevent misconfiguration to @@ -363,6 +368,10 @@ * XXX: this mechanism may introduce another problem about * mutual exclusion of the variable CALLED, especially if we * use kernel thread. + * XXXRW: use atomic operations to improve performance. + * XXXRW: This doesn't handle parallel entry into the gif code, + * since the counter is global and not packet-local. Use an + * m_tag? */ mtx_lock(&gif_mtx); if (++gif_called > max_gif_nesting) { @@ -414,6 +423,10 @@ } end: + /* + * This is completely bogus. Shouldn't this be 'gif_called--', + * and this is not MPSAFE. + */ mtx_lock(&gif_mtx); gif_called = 0; /* reset recursion counter */ mtx_unlock(&gif_mtx); @@ -489,6 +502,9 @@ } /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */ +/* + * XXXRW: per-gif softc locking required. + */ int gif_ioctl(ifp, cmd, data) struct ifnet *ifp; @@ -745,8 +761,9 @@ int s; int error = 0; - s = splnet(); - + /* + * XXXRW: per-gif softc locking required. + */ mtx_lock(&gif_mtx); LIST_FOREACH(sc2, &gif_softc_list, gif_list) { if (sc2 == sc) @@ -775,6 +792,9 @@ } mtx_unlock(&gif_mtx); + /* + * XXXRW: Lock gif softc fields. + */ /* XXX we can detach from both, but be polite just in case */ if (sc->gif_psrc) switch (sc->gif_psrc->sa_family) { --- //depot/vendor/freebsd/src/sys/net/if_gre.c 2004/03/22 08:06:54 +++ //depot/user/rwatson/netperf/sys/net/if_gre.c 2004/03/22 08:18:59 @@ -93,7 +93,8 @@ /* * gre_mtx protects all global variables in if_gre.c. - * XXX: gre_softc data not protected yet. + * + * XXXRW: It does not protect softc-specific data. */ struct mtx gre_mtx; static MALLOC_DEFINE(M_GRE, GRENAME, "Generic Routing Encapsulation"); --- //depot/vendor/freebsd/src/sys/net/if_gre.h 2004/03/22 08:06:54 +++ //depot/user/rwatson/netperf/sys/net/if_gre.h 2004/03/22 08:18:59 @@ -54,6 +54,12 @@ WCCP_V2 } wccp_ver_t; +/* + * XXXRW: softc fields need locking. + * + * XXXRW: gre's notion of a 'called' count is not MP-safe, as it assumes + * only one packet can be processed at a time. + */ struct gre_softc { struct ifnet sc_if; LIST_ENTRY(gre_softc) sc_list; --- //depot/vendor/freebsd/src/sys/net/if_spppsubr.c 2004/03/13 17:35:36 +++ //depot/user/rwatson/netperf/sys/net/if_spppsubr.c 2004/03/13 21:52:56 @@ -92,12 +92,8 @@ #include #if defined(__FreeBSD__) && __FreeBSD__ >= 3 -# define UNTIMEOUT(fun, arg, handle) untimeout(fun, arg, handle) -# define TIMEOUT(fun, arg1, arg2, handle) handle = timeout(fun, arg1, arg2) # define IOCTL_CMD_T u_long #else -# define UNTIMEOUT(fun, arg, handle) untimeout(fun, arg) -# define TIMEOUT(fun, arg1, arg2, handle) timeout(fun, arg1, arg2) # define IOCTL_CMD_T int #endif @@ -259,10 +255,11 @@ void (*scr)(struct sppp *sp); }; +struct mtx sppp_mtx; +MTX_SYSINIT(sppp_mtx, &sppp_mtx, "sppp_mtx", MTX_DEF); + static struct sppp *spppq; -#if defined(__FreeBSD__) && __FreeBSD__ >= 3 -static struct callout_handle keepalive_ch; -#endif +static struct callout keepalive_callout; #if defined(__FreeBSD__) && __FreeBSD__ >= 3 && __FreeBSD_version < 501113 #define SPP_FMT "%s%d: " @@ -960,13 +957,18 @@ { struct sppp *sp = (struct sppp*) ifp; + mtx_lock(&sppp_mtx); /* Initialize keepalive handler. */ - if (spppq != NULL) - TIMEOUT(sppp_keepalive, 0, hz * 10, keepalive_ch); + if (spppq == NULL) { + callout_init(&keepalive_callout, 0); + callout_reset(&keepalive_callout, hz * 10, sppp_keepalive, + NULL); + } /* Insert new entry into the keepalive list. */ sp->pp_next = spppq; spppq = sp; + mtx_unlock(&sppp_mtx); sp->pp_if.if_mtu = PP_MTU; sp->pp_if.if_flags = IFF_POINTOPOINT | IFF_MULTICAST; @@ -1012,6 +1014,7 @@ struct sppp **q, *p, *sp = (struct sppp*) ifp; int i; + mtx_lock(&sppp_mtx); /* Remove the entry from the keepalive list. */ for (q = &spppq; (p = *q); q = &p->pp_next) if (p == sp) { @@ -1020,12 +1023,13 @@ } /* Stop keepalive handler. */ - if (spppq != NULL) - UNTIMEOUT(sppp_keepalive, 0, keepalive_ch); + if (spppq == NULL) + callout_stop(&keepalive_callout); + mtx_unlock(&sppp_mtx); for (i = 0; i < IDX_COUNT; i++) - UNTIMEOUT((cps[i])->TO, (void *)sp, sp->ch[i]); - UNTIMEOUT(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); + untimeout((cps[i])->TO, (void *)sp, sp->ch[i]); + untimeout(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); mtx_destroy(&sp->pp_cpq.ifq_mtx); mtx_destroy(&sp->pp_fastq.ifq_mtx); } @@ -2004,8 +2008,8 @@ case STATE_STOPPING: sppp_cp_send(sp, cp->proto, TERM_REQ, ++sp->pp_seq[cp->protoidx], 0, 0); - TIMEOUT(cp->TO, (void *)sp, sp->lcp.timeout, - sp->ch[cp->protoidx]); + sp->ch[cp->protoidx] = timeout(cp->TO, (void *)sp, + sp->lcp.timeout); break; case STATE_REQ_SENT: case STATE_ACK_RCVD: @@ -2015,8 +2019,8 @@ break; case STATE_ACK_SENT: (cp->scr)(sp); - TIMEOUT(cp->TO, (void *)sp, sp->lcp.timeout, - sp->ch[cp->protoidx]); + sp->ch[cp->protoidx] = timeout(cp->TO, (void *)sp, + sp->lcp.timeout); break; } @@ -2032,7 +2036,7 @@ { sp->state[cp->protoidx] = newstate; - UNTIMEOUT(cp->TO, (void *)sp, sp->ch[cp->protoidx]); + untimeout(cp->TO, (void *)sp, sp->ch[cp->protoidx]); switch (newstate) { case STATE_INITIAL: case STATE_STARTING: @@ -2045,8 +2049,8 @@ case STATE_REQ_SENT: case STATE_ACK_RCVD: case STATE_ACK_SENT: - TIMEOUT(cp->TO, (void *)sp, sp->lcp.timeout, - sp->ch[cp->protoidx]); + sp->ch[cp->protoidx] = timeout(cp->TO, (void *)sp, + sp->lcp.timeout); break; } } @@ -4142,7 +4146,7 @@ * a number between 300 and 810 seconds. */ i = 300 + ((unsigned)(random() & 0xff00) >> 7); - TIMEOUT(chap.TO, (void *)sp, i * hz, sp->ch[IDX_CHAP]); + sp->ch[IDX_CHAP] = timeout(chap.TO, (void *)sp, i * hz); } if (debug) { @@ -4186,7 +4190,7 @@ if (debug) log(LOG_DEBUG, SPP_FMT "chap tld\n", SPP_ARGS(ifp)); - UNTIMEOUT(chap.TO, (void *)sp, sp->ch[IDX_CHAP]); + untimeout(chap.TO, (void *)sp, sp->ch[IDX_CHAP]); sp->lcp.protos &= ~(1 << IDX_CHAP); lcp.Close(sp); @@ -4322,7 +4326,7 @@ /* ack and nak are his authproto */ case PAP_ACK: - UNTIMEOUT(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); + untimeout(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); if (debug) { log(LOG_DEBUG, SPP_FMT "pap success", SPP_ARGS(ifp)); @@ -4351,7 +4355,7 @@ break; case PAP_NAK: - UNTIMEOUT(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); + untimeout(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); if (debug) { log(LOG_INFO, SPP_FMT "pap failure", SPP_ARGS(ifp)); @@ -4408,8 +4412,8 @@ if (sp->myauth.proto == PPP_PAP) { /* we are peer, send a request, and start a timer */ pap.scr(sp); - TIMEOUT(sppp_pap_my_TO, (void *)sp, sp->lcp.timeout, - sp->pap_my_to_ch); + sp->pap_my_to_ch = timeout(sppp_pap_my_TO, (void *)sp, + sp->lcp.timeout); } } @@ -4512,8 +4516,8 @@ if (debug) log(LOG_DEBUG, SPP_FMT "pap tld\n", SPP_ARGS(ifp)); - UNTIMEOUT(pap.TO, (void *)sp, sp->ch[IDX_PAP]); - UNTIMEOUT(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); + untimeout(pap.TO, (void *)sp, sp->ch[IDX_PAP]); + untimeout(sppp_pap_my_TO, (void *)sp, sp->pap_my_to_ch); sp->lcp.protos &= ~(1 << IDX_PAP); lcp.Close(sp); @@ -4640,7 +4644,12 @@ struct sppp *sp; int s; + /* + * XXXRW: It would be nice to avoid calling all this stuff while + * holding sppp_mtx, or we risk lock order reversals. + */ s = splimp(); + mtx_lock(&sppp_mtx); for (sp=spppq; sp; sp=sp->pp_next) { struct ifnet *ifp = &sp->pp_if; @@ -4679,8 +4688,9 @@ sp->lcp.echoid, 4, &nmagic); } } + callout_reset(&keepalive_callout, hz * 10, sppp_keepalive, NULL); + mtx_unlock(&sppp_mtx); splx(s); - TIMEOUT(sppp_keepalive, 0, hz * 10, keepalive_ch); } /* --- //depot/vendor/freebsd/src/sys/net/if_stf.c 2004/03/09 12:30:36 +++ //depot/user/rwatson/netperf/sys/net/if_stf.c 2004/03/12 06:09:05 @@ -136,13 +136,11 @@ #define sc_ro __sc_ro46.__sc_ro4 const struct encaptab *encap_cookie; LIST_ENTRY(stf_softc) sc_list; /* all stf's are linked */ + struct mtx sc_mtx; /* protect sc_ro */ }; /* * All mutable global variables in if_stf.c are protected by stf_mtx. - * XXXRW: Note that mutable fields in the softc are not currently locked: - * in particular, sc_ro needs to be protected from concurrent entrance - * of stf_output(). */ static struct mtx stf_mtx; static LIST_HEAD(, stf_softc) stf_softc_list; @@ -196,6 +194,7 @@ free(sc, M_STF); return (ENOMEM); } + mtx_init(&sc->sc_mtx, "stf sc_mtx", NULL, MTX_DEF); sc->sc_if.if_mtu = IPV6_MMTU; sc->sc_if.if_ioctl = stf_ioctl; @@ -220,6 +219,7 @@ bpfdetach(&sc->sc_if); if_detach(&sc->sc_if); + mtx_destroy(&sc->sc_mtx); free(sc, M_STF); } @@ -391,9 +391,10 @@ struct ip *ip; struct ip6_hdr *ip6; struct in6_ifaddr *ia6; -#ifdef MAC + struct route ro; int error; +#ifdef MAC error = mac_check_ifnet_transmit(ifp, m); if (error) { m_freem(m); @@ -495,9 +496,7 @@ else ip_ecn_ingress(ECN_NOCARE, &ip->ip_tos, &tos); - /* - * XXXRW: Locking of sc_ro required. - */ + mtx_lock(&sc->sc_mtx); dst4 = (struct sockaddr_in *)&sc->sc_ro.ro_dst; if (dst4->sin_family != AF_INET || bcmp(&dst4->sin_addr, &ip->ip_dst, sizeof(ip->ip_dst)) != 0) { @@ -516,12 +515,21 @@ if (sc->sc_ro.ro_rt == NULL) { m_freem(m); ifp->if_oerrors++; + mtx_unlock(&sc->sc_mtx); return ENETUNREACH; } } + /* + * XXXRW: Holding mutex over call to ip_output(): potential lock + * order issue? Hard to resolve cleanly with the current route + * caching model, as we have to synchronize access to shared softc + * state. + */ ifp->if_opackets++; - return ip_output(m, NULL, &sc->sc_ro, 0, NULL, NULL); + error = ip_output(m, NULL, &ro, 0, NULL, NULL); + mtx_unlock(&sc->sc_mtx); + return (error); } static int --- //depot/vendor/freebsd/src/sys/net/if_tap.c 2004/03/18 06:20:27 +++ //depot/user/rwatson/netperf/sys/net/if_tap.c 2004/03/19 03:01:30 @@ -112,6 +112,10 @@ * All global variables in if_tap.c are locked with tapmtx, with the * exception of tapdebug, which is accessed unlocked; tapclones is * static at runtime. + * + * XXXRW: si_flags appears not to be protected from concurrent access, + * and is written at run-time. + * XXXRW: si_drv1 is also used for test-and-set, and isn't synchronized. */ static struct mtx tapmtx; static int tapdebug = 0; /* debug flag */ @@ -161,6 +165,7 @@ * The EBUSY algorithm here can't quite atomically * guarantee that this is race-free since we have to * release the tap mtx to deregister the clone handler. + * XXXRW: is this true? */ mtx_lock(&tapmtx); SLIST_FOREACH(tp, &taphead, tap_next) { @@ -692,6 +697,7 @@ case SIOCSIFADDR: /* set MAC address of the remote side */ mtx_lock(&tp->tap_mtx); + /* XXXRW: Does this actually do anything? */ bcopy(data, tp->ether_addr, sizeof(tp->ether_addr)); mtx_unlock(&tp->tap_mtx); break; @@ -746,6 +752,7 @@ if (flag & IO_NDELAY) return (EWOULDBLOCK); + /* This looks like a wanna-be condition variable. */ mtx_lock(&tp->tap_mtx); tp->tap_flags |= TAP_RWAIT; mtx_unlock(&tp->tap_mtx); --- //depot/vendor/freebsd/src/sys/net/if_tun.c 2004/03/16 17:15:30 +++ //depot/user/rwatson/netperf/sys/net/if_tun.c 2004/03/21 18:28:18 @@ -55,6 +55,17 @@ #include +/* + * tun_list is protected by global tunmtx. Other mutable fields are + * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is + * static for the duration of a tunnel interface. + * + * XXXRW: we allocate si_drv1 for the dev_t on demand, rather than when + * the dev_t is instantiated. Nothing serializes the test/set of that + * field. + * + * XXXRW: what serializes access to si_flags? + */ struct tun_softc { TAILQ_ENTRY(tun_softc) tun_list; dev_t tun_dev; @@ -82,11 +93,18 @@ struct ifnet tun_if; /* the interface */ struct sigio *tun_sigio; /* information for async I/O */ struct selinfo tun_rsel; /* read select */ + struct mtx tun_mtx; /* protect mutable softc fields */ }; #define TUNDEBUG if (tundebug) if_printf #define TUNNAME "tun" +/* + * All mutable global variables in if_tun are locked using tunmtx, with + * the exception of tundebug, which is used unlocked, and tunclones, + * which is static after setup. + */ +static struct mtx tunmtx; static MALLOC_DEFINE(M_TUN, TUNNAME, "Tunnel Interface"); static int tundebug = 0; static struct clonedevs *tunclones; @@ -109,6 +127,9 @@ static d_ioctl_t tunioctl; static d_poll_t tunpoll; +/* + * XXXRW: can remove D_NEEDGIANT? Probably not because of si_drv1 for now. + */ static struct cdevsw tun_cdevsw = { .d_version = D_VERSION, .d_flags = D_PSEUDO | D_NEEDGIANT, @@ -147,15 +168,32 @@ } } +static void +tun_destroy(struct tun_softc *tp) +{ + dev_t dev; + + /* Unlocked read. */ + KASSERT((tp->tun_flags & TUN_OPEN) == 0, + ("tununits is out of sync - unit %d", tp->tun_if.if_dunit)); + + dev = tp->tun_dev; + bpfdetach(&tp->tun_if); + if_detach(&tp->tun_if); + destroy_dev(dev); + mtx_destroy(&tp->tun_mtx); + free(tp, M_TUN); +} + static int tunmodevent(module_t mod, int type, void *data) { static eventhandler_tag tag; struct tun_softc *tp; - dev_t dev; switch (type) { case MOD_LOAD: + mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF); clone_setup(&tunclones); tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000); if (tag == NULL) @@ -164,19 +202,16 @@ case MOD_UNLOAD: EVENTHANDLER_DEREGISTER(dev_clone, tag); - while (!TAILQ_EMPTY(&tunhead)) { - tp = TAILQ_FIRST(&tunhead); - KASSERT((tp->tun_flags & TUN_OPEN) == 0, - ("tununits is out of sync - unit %d", - tp->tun_if.if_dunit)); + mtx_lock(&tunmtx); + while ((tp = TAILQ_FIRST(&tunhead)) != NULL) { TAILQ_REMOVE(&tunhead, tp, tun_list); - dev = tp->tun_dev; - bpfdetach(&tp->tun_if); - if_detach(&tp->tun_if); - destroy_dev(dev); - free(tp, M_TUN); + mtx_unlock(&tunmtx); + tun_destroy(tp); + mtx_lock(&tunmtx); } + mtx_unlock(&tunmtx); clone_cleanup(&tunclones); + mtx_destroy(&tunmtx); break; } return 0; @@ -195,12 +230,16 @@ { struct tun_softc *tp = ifp->if_softc; + mtx_lock(&tp->tun_mtx); if (tp->tun_flags & TUN_RWAIT) { tp->tun_flags &= ~TUN_RWAIT; wakeup(tp); } - if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) + if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) { + mtx_unlock(&tp->tun_mtx); pgsigio(&tp->tun_sigio, SIGIO, 0); + } else + mtx_unlock(&tp->tun_mtx); selwakeuppri(&tp->tun_rsel, PZERO + 1); } @@ -213,9 +252,12 @@ dev->si_flags &= ~SI_CHEAPCLONE; MALLOC(sc, struct tun_softc *, sizeof(*sc), M_TUN, M_WAITOK | M_ZERO); + mtx_init(&sc->tun_mtx, "tun_mtx", NULL, MTX_DEF); sc->tun_flags = TUN_INITED; sc->tun_dev = dev; + mtx_lock(&tunmtx); TAILQ_INSERT_TAIL(&tunhead, sc, tun_list); + mtx_unlock(&tunmtx); ifp = &sc->tun_if; if_initname(ifp, TUNNAME, dev2unit(dev)); @@ -238,17 +280,31 @@ struct ifnet *ifp; struct tun_softc *tp; + /* + * XXXRW: Non-atomic test and set of dev->si_drv1 requires + * synchronization. + */ tp = dev->si_drv1; if (!tp) { tuncreate(dev); tp = dev->si_drv1; } - if (tp->tun_pid != 0 && tp->tun_pid != td->td_proc->p_pid) + /* + * XXXRW: This use of tun_pid is subject to error due to the + * fact that a reference to the tunnel can live beyond the + * death of the process that created it. Can we replace this + * with a simple busy flag? + */ + mtx_lock(&tp->tun_mtx); + if (tp->tun_pid != 0 && tp->tun_pid != td->td_proc->p_pid) { + mtx_unlock(&tp->tun_mtx); return (EBUSY); + } tp->tun_pid = td->td_proc->p_pid; tp->tun_flags |= TUN_OPEN; + mtx_unlock(&tp->tun_mtx); ifp = &tp->tun_if; TUNDEBUG(ifp, "open\n"); @@ -269,8 +325,10 @@ tp = dev->si_drv1; ifp = &tp->tun_if; + mtx_lock(&tp->tun_mtx); tp->tun_flags &= ~TUN_OPEN; tp->tun_pid = 0; + mtx_unlock(&tp->tun_mtx); /* * junk all pending output @@ -290,6 +348,7 @@ /* find internet addresses and delete routes */ TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) if (ifa->ifa_addr->sa_family == AF_INET) + /* Unlocked read. */ rtinit(ifa, (int)RTM_DELETE, tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0); ifp->if_flags &= ~IFF_RUNNING; @@ -314,6 +373,9 @@ ifp->if_flags |= IFF_UP | IFF_RUNNING; getmicrotime(&ifp->if_lastchange); + /* + * XXXRW: interface locking. + */ for (ifa = TAILQ_FIRST(&ifp->if_addrhead); ifa; ifa = TAILQ_NEXT(ifa, ifa_link)) { if (ifa->ifa_addr == NULL) @@ -325,12 +387,14 @@ struct sockaddr_in *si; si = (struct sockaddr_in *)ifa->ifa_addr; + mtx_lock(&tp->tun_mtx); if (si->sin_addr.s_addr) tp->tun_flags |= TUN_IASET; si = (struct sockaddr_in *)ifa->ifa_dstaddr; if (si && si->sin_addr.s_addr) tp->tun_flags |= TUN_DSTADDR; + mtx_unlock(&tp->tun_mtx); } #endif } @@ -353,9 +417,11 @@ switch(cmd) { case SIOCGIFSTATUS: ifs = (struct ifstat *)data; + mtx_lock(&tp->tun_mtx); if (tp->tun_pid) sprintf(ifs->ascii + strlen(ifs->ascii), "\tOpened by PID %d\n", tp->tun_pid); + mtx_unlock(&tp->tun_mtx); break; case SIOCSIFADDR: error = tuninit(ifp); @@ -391,6 +457,7 @@ struct rtentry *rt) { struct tun_softc *tp = ifp->if_softc; + u_short cached_tun_flags; #ifdef MAC int error; #endif @@ -405,7 +472,11 @@ } #endif - if ((tp->tun_flags & TUN_READY) != TUN_READY) { + /* Could be unlocked read? */ + mtx_lock(&tp->tun_mtx); + cached_tun_flags = tp->tun_flags; + mtx_unlock(&tp->tun_mtx); + if ((cached_tun_flags & TUN_READY) != TUN_READY) { TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); m_freem (m0); return (EHOSTDOWN); @@ -430,7 +501,7 @@ } /* prepend sockaddr? this may abort if the mbuf allocation fails */ - if (tp->tun_flags & TUN_LMODE) { + if (cached_tun_flags & TUN_LMODE) { /* allocate space for sockaddr */ M_PREPEND(m0, dst->sa_len, M_DONTWAIT); @@ -444,7 +515,7 @@ } } - if (tp->tun_flags & TUN_IFHEAD) { + if (cached_tun_flags & TUN_IFHEAD) { /* Prepend the address family */ M_PREPEND(m0, 4, M_DONTWAIT); @@ -509,21 +580,28 @@ *(int *)data = tundebug; break; case TUNSLMODE: + mtx_lock(&tp->tun_mtx); if (*(int *)data) { tp->tun_flags |= TUN_LMODE; tp->tun_flags &= ~TUN_IFHEAD; } else tp->tun_flags &= ~TUN_LMODE; + mtx_unlock(&tp->tun_mtx); break; case TUNSIFHEAD: + mtx_lock(&tp->tun_mtx); if (*(int *)data) { tp->tun_flags |= TUN_IFHEAD; tp->tun_flags &= ~TUN_LMODE; } else tp->tun_flags &= ~TUN_IFHEAD; + mtx_unlock(&tp->tun_mtx); break; case TUNGIFHEAD: + /* Could be unlocked read? */ + mtx_lock(&tp->tun_mtx); *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0; + mtx_unlock(&tp->tun_mtx); break; case TUNSIFMODE: /* deny this if UP */ @@ -542,15 +620,19 @@ } break; case TUNSIFPID: + mtx_lock(&tp->tun_mtx); tp->tun_pid = curthread->td_proc->p_pid; + mtx_unlock(&tp->tun_mtx); break; case FIONBIO: break; case FIOASYNC: + mtx_lock(&tp->tun_mtx); if (*(int *)data) tp->tun_flags |= TUN_ASYNC; else tp->tun_flags &= ~TUN_ASYNC; + mtx_unlock(&tp->tun_mtx); break; case FIONREAD: s = splimp(); @@ -597,12 +679,15 @@ int error=0, len, s; TUNDEBUG (ifp, "read\n"); + mtx_lock(&tp->tun_mtx); if ((tp->tun_flags & TUN_READY) != TUN_READY) { + mtx_unlock(&tp->tun_mtx); TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); return (EHOSTDOWN); } tp->tun_flags &= ~TUN_RWAIT; + mtx_unlock(&tp->tun_mtx); s = splimp(); do { @@ -612,7 +697,9 @@ splx(s); return (EWOULDBLOCK); } + mtx_lock(&tp->tun_mtx); tp->tun_flags |= TUN_RWAIT; + mtx_unlock(&tp->tun_mtx); if((error = tsleep(tp, PCATCH | (PZERO + 1), "tunread", 0)) != 0) { splx(s); @@ -699,14 +786,19 @@ mac_create_mbuf_from_ifnet(ifp, top); #endif + /* Could be unlocked read? */ + mtx_lock(&tp->tun_mtx); if (tp->tun_flags & TUN_IFHEAD) { + mtx_unlock(&tp->tun_mtx); if (top->m_len < sizeof(family) && (top = m_pullup(top, sizeof(family))) == NULL) return (ENOBUFS); family = ntohl(*mtod(top, u_int32_t *)); m_adj(top, sizeof(family)); - } else + } else { + mtx_unlock(&tp->tun_mtx); family = AF_INET; + } BPF_MTAP2(ifp, &family, sizeof(family), top); --- //depot/vendor/freebsd/src/sys/net/raw_cb.c 2003/02/18 21:53:12 +++ //depot/user/rwatson/netperf/sys/net/raw_cb.c 2004/03/11 18:54:06 @@ -36,7 +36,9 @@ #include #include +#include #include +#include #include #include #include @@ -53,10 +55,11 @@ * redo address binding to allow wildcards */ +struct mtx rawcb_mtx; struct rawcb_list_head rawcb_list; -static u_long raw_sendspace = RAWSNDQ; -static u_long raw_recvspace = RAWRCVQ; +static const u_long raw_sendspace = RAWSNDQ; +static const u_long raw_recvspace = RAWRCVQ; /* * Allocate a control block and a nominal amount @@ -83,7 +86,9 @@ rp->rcb_socket = so; rp->rcb_proto.sp_family = so->so_proto->pr_domain->dom_family; rp->rcb_proto.sp_protocol = proto; + mtx_lock(&rawcb_mtx); LIST_INSERT_HEAD(&rawcb_list, rp, list); + mtx_unlock(&rawcb_mtx); return (0); } @@ -97,6 +102,7 @@ { struct socket *so = rp->rcb_socket; + SOCK_LOCK(so); so->so_pcb = 0; sotryfree(so); LIST_REMOVE(rp, list); --- //depot/vendor/freebsd/src/sys/net/raw_cb.h 2002/03/19 13:55:39 +++ //depot/user/rwatson/netperf/sys/net/raw_cb.h 2004/03/11 18:54:06 @@ -61,6 +61,7 @@ #ifdef _KERNEL extern LIST_HEAD(rawcb_list_head, rawcb) rawcb_list; +extern struct mtx rawcb_mtx; /* protosw entries */ pr_ctlinput_t raw_ctlinput; --- //depot/vendor/freebsd/src/sys/net/raw_usrreq.c 2004/02/29 19:15:33 +++ //depot/user/rwatson/netperf/sys/net/raw_usrreq.c 2004/03/13 21:36:41 @@ -35,9 +35,12 @@ */ #include +#include #include #include #include +#include +#include #include #include #include @@ -47,12 +50,15 @@ #include +MTX_SYSINIT(rawcb_mtx, &rawcb_mtx, "rawcb_mtx", MTX_DEF); + /* * Initialize raw connection block q. */ void raw_init() { + LIST_INIT(&rawcb_list); } @@ -75,7 +81,12 @@ register struct mbuf *m = m0; struct socket *last; + /* + * XXXRW: Potential lock order issues due to holding the + * rawcb_mtx across all this stuff. Need to revisit. + */ last = 0; + mtx_lock(&rawcb_mtx); LIST_FOREACH(rp, &rawcb_list, list) { if (rp->rcb_proto.sp_family != proto->sp_family) continue; @@ -120,6 +131,7 @@ } } else m_freem(m); + mtx_unlock(&rawcb_mtx); } /*ARGSUSED*/ @@ -143,8 +155,12 @@ if (rp == 0) return EINVAL; raw_disconnect(rp); - sotryfree(so); - soisdisconnected(so); /* XXX huh? called after the sofree()? */ + SOCK_LOCK(so); + if (so->so_count != 0) { + soisdisconnected(so); + SOCK_UNLOCK(so); + } else + sofree(so); return 0; } --- //depot/vendor/freebsd/src/sys/net/rtsock.c 2003/11/20 12:10:44 +++ //depot/user/rwatson/netperf/sys/net/rtsock.c 2004/03/11 18:56:12 @@ -55,10 +55,18 @@ MALLOC_DEFINE(M_RTABLE, "routetbl", "routing tables"); /* NB: these are not modified */ +/* + * XXXRW: It would be really nice to add const to these, but that may + * not be possible due to where they are passed in. We might need + * to const-poison a whole boatload of APIs...? + */ static struct sockaddr route_dst = { 2, PF_ROUTE, }; static struct sockaddr route_src = { 2, PF_ROUTE, }; static struct sockaddr sa_zero = { sizeof(sa_zero), AF_INET, }; +/* + * XXXRW: These fields are locked by RTSOCK_LOCK(). + */ static struct { int ip_count; /* attacked w/ AF_INET */ int ip6_count; /* attached w/ AF_INET6 */ --- //depot/vendor/freebsd/src/sys/netatalk/aarp.c 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/aarp.c 2004/03/21 21:14:09 @@ -36,11 +36,20 @@ #define AARPTAB_SIZE (AARPTAB_BSIZ * AARPTAB_NB) static struct aarptab aarptab[AARPTAB_SIZE]; +static struct mtx aarptab_mtx; +MTX_SYSINIT(aarptab_mtx, &aarptab_mtx, "aarptab_mtx", MTX_DEF); + +#define AARPTAB_LOCK() mtx_lock(&aarptab_mtx) +#define AARPTAB_UNLOCK() mtx_unlock(&aarptab_mtx) +#define AARPTAB_LOCK_ASSERT() mtx_assert(&aarptab_mtx, MA_OWNED) +#define AARPTAB_UNLOCK_ASSERT() mtx_assert(&aarptab_mtx, MA_NOTOWNED) + #define AARPTAB_HASH(a) \ ((((a).s_net << 8) + (a).s_node) % AARPTAB_NB) #define AARPTAB_LOOK(aat, addr) { \ int n; \ + AARPTAB_LOCK_ASSERT(); \ aat = &aarptab[ AARPTAB_HASH(addr) * AARPTAB_BSIZ ]; \ for (n = 0; n < AARPTAB_BSIZ; n++, aat++) \ if (aat->aat_ataddr.s_net == (addr).s_net && \ @@ -54,6 +63,9 @@ #define AARPT_KILLC 20 #define AARPT_KILLI 3 +/* + * XXXRW: wot? + */ # if !defined(__FreeBSD__) extern u_char etherbroadcastaddr[6]; # endif /* __FreeBSD__ */ @@ -63,7 +75,7 @@ }; /* - * Not used? + * XXXRW: unused? */ u_char at_org_code[ 3 ] = { 0x08, 0x00, 0x07, @@ -72,6 +84,9 @@ 0x00, 0x00, 0x00, }; +/* + * XXXRW: Make use callouts, not timeouts. + */ static struct callout_handle aarptimer_ch = CALLOUT_HANDLE_INITIALIZER(&aarptimer_ch); @@ -79,20 +94,20 @@ aarptimer(void *ignored) { struct aarptab *aat; - int i, s; + int i; aarptimer_ch = timeout(aarptimer, (caddr_t)0, AARPT_AGE * hz); aat = aarptab; + AARPTAB_LOCK(); for (i = 0; i < AARPTAB_SIZE; i++, aat++) { if (aat->aat_flags == 0 || (aat->aat_flags & ATF_PERM)) continue; if (++aat->aat_timer < ((aat->aat_flags & ATF_COM) ? AARPT_KILLC : AARPT_KILLI)) continue; - s = splimp(); aarptfree(aat); - splx(s); } + AARPTAB_UNLOCK(); } /* @@ -130,6 +145,7 @@ struct llc *llc; struct sockaddr sa; + AARPTAB_UNLOCK_ASSERT(); if ((m = m_gethdr(M_DONTWAIT, MT_DATA)) == NULL) { return; } @@ -214,7 +230,6 @@ { struct at_ifaddr *aa; struct aarptab *aat; - int s; if (at_broadcast(destsat)) { m->m_flags |= M_BCAST; @@ -231,7 +246,7 @@ return (1); } - s = splimp(); + AARPTAB_LOCK(); AARPTAB_LOOK(aat, destsat->sat_addr); if (aat == NULL) { /* No entry */ aat = aarptnew(&destsat->sat_addr); @@ -239,8 +254,8 @@ panic("aarpresolve: no free entry"); } aat->aat_hold = m; + AARPTAB_UNLOCK(); aarpwhohas(ac, destsat); - splx(s); return (0); } /* found an entry */ @@ -248,7 +263,7 @@ if (aat->aat_flags & ATF_COM) { /* entry is COMplete */ bcopy((caddr_t)aat->aat_enaddr, (caddr_t)desten, sizeof(aat->aat_enaddr)); - splx(s); + AARPTAB_UNLOCK(); return (1); } /* entry has not completed */ @@ -256,8 +271,8 @@ m_freem(aat->aat_hold); } aat->aat_hold = m; + AARPTAB_UNLOCK(); aarpwhohas(ac, destsat); - splx(s); return (0); } @@ -390,6 +405,7 @@ } } + AARPTAB_LOCK(); /* XXXRW */ AARPTAB_LOOK(aat, spa); if (aat != NULL) { if (op == AARPOP_PROBE) { @@ -399,6 +415,7 @@ * to arp for him. */ aarptfree(aat); + AARPTAB_UNLOCK(); m_freem(m); return; } @@ -409,12 +426,14 @@ if (aat->aat_hold) { struct mbuf *mhold = aat->aat_hold; aat->aat_hold = NULL; + AARPTAB_UNLOCK(); sat.sat_len = sizeof(struct sockaddr_at); sat.sat_family = AF_APPLETALK; sat.sat_addr = spa; (*ac->ac_if.if_output)(&ac->ac_if, mhold, (struct sockaddr *)&sat, NULL); /* XXX */ - } + } else + AARPTAB_UNLOCK(); } else if ((tpa.s_net == ma.s_net) && (tpa.s_node == ma.s_node) && (op != AARPOP_PROBE) @@ -422,7 +441,9 @@ bcopy((caddr_t)ea->aarp_sha, (caddr_t)aat->aat_enaddr, sizeof(ea->aarp_sha)); aat->aat_flags |= ATF_COM; - } + AARPTAB_UNLOCK(); + } else + AARPTAB_UNLOCK(); /* * Don't respond to responses, and never respond if we're @@ -477,6 +498,7 @@ aarptfree(struct aarptab *aat) { + AARPTAB_LOCK_ASSERT(); if (aat->aat_hold) m_freem(aat->aat_hold); aat->aat_hold = NULL; @@ -485,7 +507,7 @@ aat->aat_ataddr.s_node = 0; } - struct aarptab * +struct aarptab * aarptnew(addr) struct at_addr *addr; { @@ -494,6 +516,7 @@ struct aarptab *aat, *aato = NULL; static int first = 1; + AARPTAB_LOCK_ASSERT(); if (first) { first = 0; aarptimer_ch = timeout(aarptimer, (caddr_t)0, hz); @@ -625,11 +648,16 @@ struct aarptab *aat; int i; + /* + * XXXRW: Should grab mutex before untimeout? + */ untimeout(aarptimer, 0, aarptimer_ch); + AARPTAB_LOCK(); for (i = 0, aat = aarptab; i < AARPTAB_SIZE; i++, aat++) { if (aat->aat_hold) { m_freem(aat->aat_hold); aat->aat_hold = NULL; } } + AARPTAB_UNLOCK(); } --- //depot/vendor/freebsd/src/sys/netatalk/at_control.c 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/at_control.c 2004/03/21 21:14:09 @@ -21,6 +21,9 @@ #include #include +/* + * XXXRW: Requires synchronization. + */ struct at_ifaddr *at_ifaddr_list; static int aa_dorangeroute(struct ifaddr *ifa, --- //depot/vendor/freebsd/src/sys/netatalk/at_rmx.c 2004/03/21 20:01:33 +++ //depot/user/rwatson/netperf/sys/netatalk/at_rmx.c 2004/03/21 20:08:45 @@ -40,11 +40,23 @@ int at_inithead(void **head, int off); -static char hexbuf[256]; +/* + * XXXRW: hexdump was a static global variable, but I moved it into the + * stack rather than stick a mutex around it. 256 bytes is smaller than + * it used to be, but this still might be a problem. Needs to be + * revisited. Should probably just use the new hexdump(9). + * + * XXXRW: All this appears to be present just so as to printf debugging + * information. Assuming that this code is known to work, we could just + * scrap all this. In fact, this code isn't even used as it stands, it's + * here for debugging purposes only and requires modifications to + * at_proto.c. + */ static char * prsockaddr(void *v) { + static char hexbuf[256]; char *bp = &hexbuf[0]; u_char *cp = v; --- //depot/vendor/freebsd/src/sys/netatalk/ddp_input.c 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/ddp_input.c 2004/03/21 21:14:09 @@ -29,6 +29,12 @@ static volatile int ddp_forward = 1; static volatile int ddp_firewall = 0; static struct ddpstat ddpstat; + +/* + * XXXRW: If we're going to keep this cached route data, we'll need to lock it + * down, and change later function-local use of it to grab an extra reference + * after deciding it is useful. + */ static struct route forwro; static void ddp_input(struct mbuf *, struct ifnet *, struct elaphdr *, int); --- //depot/vendor/freebsd/src/sys/netatalk/ddp_pcb.c 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/ddp_pcb.c 2004/03/22 05:14:14 @@ -22,12 +22,18 @@ #include #include +struct mtx ddp_list_mtx; static struct ddpcb *ddp_ports[ ATPORT_LAST ]; -struct ddpcb *ddpcb_list = NULL; +struct ddpcb *ddpcb_list = NULL; void at_sockaddr(struct ddpcb *ddp, struct sockaddr **addr) { + + /* + * Prevent modification of ddp during copy of addr. + */ + DDP_LOCK_ASSERT(ddp); *addr = sodupsockaddr((struct sockaddr *)&ddp->ddp_lsat, M_NOWAIT); } @@ -38,6 +44,12 @@ struct at_ifaddr *aa; struct ddpcb *ddpp; + /* + * We read and write both the ddp passed in, and also ddp_ports. + */ + DDP_LIST_XLOCK_ASSERT(); + DDP_LOCK_ASSERT(ddp); + if (ddp->ddp_lsat.sat_port != ATADDR_ANYPORT) { /* shouldn't be bound */ return (EINVAL); } @@ -134,6 +146,9 @@ struct ifnet *ifp; u_short hintnet = 0, net; + DDP_LIST_XLOCK_ASSERT(); + DDP_LOCK_ASSERT(ddp); + if (sat->sat_family != AF_APPLETALK) { return (EAFNOSUPPORT); } @@ -222,6 +237,9 @@ void at_pcbdisconnect(struct ddpcb *ddp) { + + DDP_LOCK_ASSERT(ddp); + ddp->ddp_fsat.sat_addr.s_net = ATADDR_ANYNET; ddp->ddp_fsat.sat_addr.s_node = ATADDR_ANYNODE; ddp->ddp_fsat.sat_port = ATADDR_ANYPORT; @@ -233,8 +251,17 @@ struct ddpcb *ddp; MALLOC(ddp, struct ddpcb *, sizeof *ddp, M_PCB, M_WAITOK | M_ZERO); + DDP_LOCK_INIT(ddp); ddp->ddp_lsat.sat_port = ATADDR_ANYPORT; + /* + * XXXRW: Is this unlocked assignment payer for socket and + * back-pointer something that needs to be protected? + */ + ddp->ddp_socket = so; + so->so_pcb = (caddr_t)ddp; + + DDP_LIST_XLOCK(); ddp->ddp_next = ddpcb_list; ddp->ddp_prev = NULL; ddp->ddp_pprev = NULL; @@ -243,15 +270,21 @@ ddpcb_list->ddp_prev = ddp; } ddpcb_list = ddp; + DDP_LIST_XUNLOCK(); - ddp->ddp_socket = so; - so->so_pcb = (caddr_t)ddp; - return (0); + return(0); } void at_pcbdetach(struct socket *so, struct ddpcb *ddp) { + + /* + * We modify ddp, ddp_ports, and the global list. + */ + DDP_LIST_XLOCK_ASSERT(); + DDP_LOCK_ASSERT(ddp); + soisdisconnected(so); so->so_pcb = NULL; sotryfree(so); @@ -281,6 +314,8 @@ if (ddp->ddp_next) { ddp->ddp_next->ddp_prev = ddp->ddp_prev; } + DDP_UNLOCK(ddp); + DDP_LOCK_DESTROY(ddp); FREE(ddp, M_PCB); } @@ -296,6 +331,8 @@ { struct ddpcb *ddp; + DDP_LIST_SLOCK_ASSERT(); + /* * Check for bad ports. */ @@ -308,11 +345,13 @@ * the interface? */ for (ddp = ddp_ports[ to->sat_port - 1 ]; ddp; ddp = ddp->ddp_pnext) { + DDP_LOCK(ddp); /* XXX should we handle 0.YY? */ /* XXXX.YY to socket on destination interface */ if (to->sat_addr.s_net == ddp->ddp_lsat.sat_addr.s_net && to->sat_addr.s_node == ddp->ddp_lsat.sat_addr.s_node) { + DDP_UNLOCK(ddp); break; } @@ -320,6 +359,7 @@ if (to->sat_addr.s_node == ATADDR_BCAST && (to->sat_addr.s_net == 0 || to->sat_addr.s_net == ddp->ddp_lsat.sat_addr.s_net) && ddp->ddp_lsat.sat_addr.s_net == AA_SAT(aa)->sat_addr.s_net) { + DDP_UNLOCK(ddp); break; } @@ -330,8 +370,10 @@ ntohs(aa->aa_firstnet) && ntohs(ddp->ddp_lsat.sat_addr.s_net) <= ntohs(aa->aa_lastnet)) { + DDP_UNLOCK(ddp); break; } + DDP_UNLOCK(ddp); } return (ddp); } --- //depot/vendor/freebsd/src/sys/netatalk/ddp_pcb.h 2004/03/18 23:25:31 +++ //depot/user/rwatson/netperf/sys/netatalk/ddp_pcb.h 2004/03/21 20:20:48 @@ -17,4 +17,23 @@ struct thread *td); void at_sockaddr(struct ddpcb *ddp, struct sockaddr **addr); +/* Lock macros for per-pcb locks. */ +#define DDP_LOCK_INIT(ddp) mtx_init(&(ddp)->ddp_mtx, "ddp_mtx", \ + NULL, MTX_DEF) +#define DDP_LOCK_DESTROY(ddp) mtx_destroy(&(ddp)->ddp_mtx) +#define DDP_LOCK(ddp) mtx_lock(&(ddp)->ddp_mtx) +#define DDP_UNLOCK(ddp) mtx_unlock(&(ddp)->ddp_mtx) +#define DDP_LOCK_ASSERT(ddp) mtx_assert(&(ddp)->ddp_mtx, MA_OWNED) + +/* Lock macros for global pcb list lock. */ +#define DDP_LIST_LOCK_INIT() mtx_init(&ddp_list_mtx, "ddp_list_mtx", \ + NULL, MTX_DEF) +#define DDP_LIST_LOCK_DESTROY() mtx_destroy(&ddp_list_mtx) +#define DDP_LIST_XLOCK() mtx_lock(&ddp_list_mtx) +#define DDP_LIST_XUNLOCK() mtx_unlock(&ddp_list_mtx) +#define DDP_LIST_XLOCK_ASSERT() mtx_assert(&ddp_list_mtx, MA_OWNED) +#define DDP_LIST_SLOCK() mtx_lock(&ddp_list_mtx) +#define DDP_LIST_SUNLOCK() mtx_unlock(&ddp_list_mtx) +#define DDP_LIST_SLOCK_ASSERT() mtx_assert(&ddp_list_mtx, MA_OWNED) + #endif --- //depot/vendor/freebsd/src/sys/netatalk/ddp_usrreq.c 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/ddp_usrreq.c 2004/03/21 21:14:09 @@ -22,6 +22,9 @@ #include #include +/* + * XXXRW: These structures are currently not mutable. + */ static u_long ddp_sendspace = DDP_MAXSZ; /* Max ddp size + 1 (ddp_type) */ static u_long ddp_recvspace = 10 * (587 + sizeof(struct sockaddr_at)); @@ -32,36 +35,38 @@ { struct ddpcb *ddp; int error = 0; - int s; + ddp = sotoddpcb(so); + if (ddp != NULL) + return (EINVAL); - ddp = sotoddpcb(so); - if (ddp != NULL) { - return (EINVAL); - } + /* + * Allocate socket buffer space first so that it's present + * before first use. + */ + error = soreserve(so, ddp_sendspace, ddp_recvspace); + if (error) + return (error); - s = splnet(); + DDP_LIST_XLOCK(); error = at_pcballoc(so); - splx(s); - if (error) { - return (error); - } - return (soreserve(so, ddp_sendspace, ddp_recvspace)); + DDP_LIST_XUNLOCK(); + return (error); } static int ddp_detach(struct socket *so) { struct ddpcb *ddp; - int s; ddp = sotoddpcb(so); - if (ddp == NULL) { + if (ddp == NULL) return (EINVAL); - } - s = splnet(); + + DDP_LIST_XLOCK(); + DDP_LOCK(ddp); at_pcbdetach(so, ddp); - splx(s); + DDP_LIST_XUNLOCK(); return (0); } @@ -70,15 +75,16 @@ { struct ddpcb *ddp; int error = 0; - int s; ddp = sotoddpcb(so); if (ddp == NULL) { return (EINVAL); } - s = splnet(); + DDP_LIST_XLOCK(); + DDP_LOCK(ddp); error = at_pcbsetaddr(ddp, nam, td); - splx(s); + DDP_UNLOCK(ddp); + DDP_LIST_XUNLOCK(); return (error); } @@ -87,20 +93,22 @@ { struct ddpcb *ddp; int error = 0; - int s; ddp = sotoddpcb(so); if (ddp == NULL) { return (EINVAL); } + DDP_LIST_XLOCK(); + DDP_LOCK(ddp); if (ddp->ddp_fsat.sat_port != ATADDR_ANYPORT) { + DDP_UNLOCK(ddp); return (EISCONN); } - s = splnet(); - error = at_pcbconnect(ddp, nam, td); - splx(s); + error = at_pcbconnect( ddp, nam, td ); + DDP_UNLOCK(ddp); + DDP_LIST_XUNLOCK(); if (error == 0) soisconnected(so); return (error); @@ -111,20 +119,20 @@ { struct ddpcb *ddp; - int s; ddp = sotoddpcb(so); if (ddp == NULL) { return (EINVAL); } + DDP_LOCK(ddp); if (ddp->ddp_fsat.sat_addr.s_node == ATADDR_ANYNODE) { + DDP_UNLOCK(ddp); return (ENOTCONN); } - s = splnet(); at_pcbdisconnect(ddp); ddp->ddp_fsat.sat_addr.s_node = ATADDR_ANYNODE; - splx(s); + DDP_UNLOCK(ddp); soisdisconnected(so); return (0); } @@ -142,13 +150,19 @@ return (0); } +/* + * XXXRW: If an explicit address is specified, then we temporarily change + * the address on the pcb for sending. This is inefficient because it + * requires us to perform global rather than pcb-local operations. It + * may also create a race if other users of the socket are simultaneously + * sending. + */ static int ddp_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr, struct mbuf *control, struct thread *td) { struct ddpcb *ddp; int error = 0; - int s; ddp = sotoddpcb(so); if (ddp == NULL) { @@ -160,28 +174,29 @@ } if (addr != NULL) { + DDP_LIST_XLOCK(); + DDP_LOCK(ddp); if (ddp->ddp_fsat.sat_port != ATADDR_ANYPORT) { - return (EISCONN); + error = EISCONN; + goto out; } - s = splnet(); error = at_pcbconnect(ddp, addr, td); - splx(s); - if (error) { - return (error); + if (error == 0) { + error = ddp_output(m, so); + at_pcbdisconnect(ddp); } +out: + DDP_UNLOCK(ddp); + DDP_LIST_XUNLOCK(); } else { - if (ddp->ddp_fsat.sat_port == ATADDR_ANYPORT) { - return (ENOTCONN); - } + DDP_LOCK(ddp); + if (ddp->ddp_fsat.sat_port == ATADDR_ANYPORT) + error = ENOTCONN; + else + error = ddp_output(m, so); + DDP_UNLOCK(ddp); } - - s = splnet(); - error = ddp_output(m, so); - if (addr != NULL) { - at_pcbdisconnect(ddp); - } - splx(s); return (error); } @@ -189,29 +204,29 @@ ddp_abort(struct socket *so) { struct ddpcb *ddp; - int s; ddp = sotoddpcb(so); if (ddp == NULL) { return (EINVAL); } soisdisconnected(so); - s = splnet(); + DDP_LIST_XLOCK(); + DDP_LOCK(ddp); at_pcbdetach(so, ddp); - splx(s); + DDP_LIST_XUNLOCK(); return (0); } void ddp_init(void) { - atintrq1.ifq_maxlen = IFQ_MAXLEN; atintrq2.ifq_maxlen = IFQ_MAXLEN; aarpintrq.ifq_maxlen = IFQ_MAXLEN; mtx_init(&atintrq1.ifq_mtx, "at1_inq", NULL, MTX_DEF); mtx_init(&atintrq2.ifq_mtx, "at2_inq", NULL, MTX_DEF); mtx_init(&aarpintrq.ifq_mtx, "aarp_inq", NULL, MTX_DEF); + DDP_LIST_LOCK_INIT(); netisr_register(NETISR_ATALK1, at1intr, &atintrq1, 0); netisr_register(NETISR_ATALK2, at2intr, &atintrq2, 0); netisr_register(NETISR_AARP, aarpintr, &aarpintrq, 0); @@ -226,6 +241,7 @@ for (ddp = ddpcb_list; ddp != NULL; ddp = ddp->ddp_next) { at_pcbdetach(ddp->ddp_socket, ddp); } + DDP_LIST_LOCK_DESTROY(); } #endif @@ -244,7 +260,9 @@ if (ddp == NULL) { return (EINVAL); } + DDP_LOCK(ddp); at_sockaddr(ddp, nam); + DDP_UNLOCK(ddp); return (0); } --- //depot/vendor/freebsd/src/sys/netatalk/ddp_var.h 2004/03/21 20:56:26 +++ //depot/user/rwatson/netperf/sys/netatalk/ddp_var.h 2004/03/21 21:14:09 @@ -13,6 +13,7 @@ struct socket *ddp_socket; struct ddpcb *ddp_prev, *ddp_next; struct ddpcb *ddp_pprev, *ddp_pnext; + struct mtx ddp_mtx; }; #define sotoddpcb(so) ((struct ddpcb *)(so)->so_pcb) @@ -34,5 +35,6 @@ extern int ddp_cksum; extern struct ddpcb *ddpcb_list; extern struct pr_usrreqs ddp_usrreqs; +extern struct mtx ddp_list_mtx; #endif #endif /* _NETATALK_DDP_VAR_H_ */ --- //depot/vendor/freebsd/src/sys/netatm/atm_socket.c 2003/10/31 10:36:05 +++ //depot/user/rwatson/netperf/sys/netatm/atm_socket.c 2004/02/28 14:29:37 @@ -173,6 +173,7 @@ /* * Break links and free control blocks */ + SOCK_LOCK(so); so->so_pcb = NULL; sotryfree(so); --- //depot/vendor/freebsd/src/sys/netgraph/ng_socket.c 2004/01/27 14:05:28 +++ //depot/user/rwatson/netperf/sys/netgraph/ng_socket.c 2004/03/13 19:02:19 @@ -153,6 +153,9 @@ SYSCTL_INT(_net_graph, OID_AUTO, recvspace, CTLFLAG_RW, &ngpdg_recvspace , 0, "Maximum space for incoming Netgraph datagrams"); +/* + * XXXRW: Locking? + */ /* List of all sockets */ static LIST_HEAD(, ngpcb) ngsocklist; --- //depot/vendor/freebsd/src/sys/netinet/if_ether.c 2004/03/21 10:56:10 +++ //depot/user/rwatson/netperf/sys/netinet/if_ether.c 2004/03/21 17:49:17 @@ -102,6 +102,11 @@ #define la_timer la_rt->rt_rmx.rmx_expire /* deletion time in seconds */ }; +/* + * XXXRW: Need to document (and/or fix) locking for this. We always + * seem to hold a lock (and assert) when referencing this list, but it's + * not clear it's always the same lock. + */ static LIST_HEAD(, llinfo_arp) llinfo_arp; static struct ifqueue arpintrq; --- //depot/vendor/freebsd/src/sys/netinet/igmp.c 2003/08/28 15:15:24 +++ //depot/user/rwatson/netperf/sys/netinet/igmp.c 2004/03/11 19:06:01 @@ -84,10 +84,28 @@ SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_RW, &igmpstat, igmpstat, ""); +/* + * igmp_mtx protects all mutable global variables in igmp.c, as well as + * the data fields in struct router_info. In general, a router_info + * structure will be valid as long as the referencing struct in_multi is + * valid, so no reference counting is used. We allow unlocked reads of + * router_info data when accessed via an in_multi read-only. + */ +static struct mtx igmp_mtx; static SLIST_HEAD(, router_info) router_info_head; static int igmp_timers_are_running; + +/* + * XXXRW: can we define these such that these can be made const? In any + * case, these shouldn't be changed after igmp_init() and therefore don't + * need locking. + */ static u_long igmp_all_hosts_group; static u_long igmp_all_rtrs_group; + +/* + * XXXRW: These variables make me vaguely nervous. + */ static struct mbuf *router_alert; static struct route igmprt; @@ -112,6 +130,7 @@ /* * Construct a Router Alert option to use in outgoing packets + * XXXRW: This might actually need a MAC label. */ MGET(router_alert, M_DONTWAIT, MT_DATA); ra = mtod(router_alert, struct ipoption *); @@ -122,6 +141,7 @@ ra->ipopt_list[3] = 0x00; router_alert->m_len = sizeof(ra->ipopt_dst) + ra->ipopt_list[1]; + mtx_init(&igmp_mtx, "igmp_mtx", NULL, MTX_DEF); SLIST_INIT(&router_info_head); } @@ -130,6 +150,7 @@ { struct router_info *rti; + mtx_assert(&igmp_mtx, MA_OWNED); IGMP_PRINTF("[igmp.c, _find_rti] --> entering \n"); SLIST_FOREACH(rti, &router_info_head, rti_list) { if (rti->rti_ifp == ifp) { @@ -138,6 +159,9 @@ return rti; } } + /* + * XXXRW: return value of malloc not checked, despite M_NOWAIT. + */ MALLOC(rti, struct router_info *, sizeof *rti, M_IGMP, M_NOWAIT); rti->rti_ifp = ifp; rti->rti_type = IGMP_V2_ROUTER; @@ -201,7 +225,6 @@ timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; if (timer == 0) timer = 1; - rti = find_rti(ifp); /* * In the IGMPv2 specification, there are 3 states and a flag. @@ -228,8 +251,11 @@ * value in RFC 1112. */ + mtx_lock(&igmp_mtx); + rti = find_rti(ifp); rti->rti_type = IGMP_V1_ROUTER; rti->rti_time = 0; + mtx_unlock(&igmp_mtx); timer = IGMP_MAX_HOST_REPORT_DELAY * PR_FASTHZ; @@ -348,7 +374,9 @@ inm->inm_timer = 0; inm->inm_state = IGMP_OTHERMEMBER; } else { + mtx_lock(&igmp_mtx); inm->inm_rti = find_rti(inm->inm_ifp); + mtx_unlock(&igmp_mtx); igmp_sendpkt(inm, inm->inm_rti->rti_type, 0); inm->inm_timer = IGMP_RANDOM_DELAY( IGMP_MAX_HOST_REPORT_DELAY*PR_FASTHZ); @@ -408,6 +436,7 @@ struct router_info *rti; IGMP_PRINTF("[igmp.c,_slowtimo] -- > entering \n"); + mtx_lock(&igmp_mtx); SLIST_FOREACH(rti, &router_info_head, rti_list) { if (rti->rti_type == IGMP_V1_ROUTER) { rti->rti_time++; @@ -415,6 +444,7 @@ rti->rti_type = IGMP_V2_ROUTER; } } + mtx_unlock(&igmp_mtx); IGMP_PRINTF("[igmp.c,_slowtimo] -- > exiting \n"); splx(s); } --- //depot/vendor/freebsd/src/sys/netinet/in_gif.c 2003/10/29 07:10:52 +++ //depot/user/rwatson/netperf/sys/netinet/in_gif.c 2004/03/09 15:48:47 @@ -174,6 +174,9 @@ } bcopy(&iphdr, mtod(m, struct ip *), sizeof(struct ip)); + /* + * XXXRW: locking of gif's softc. + */ if (dst->sin_family != sin_dst->sin_family || dst->sin_addr.s_addr != sin_dst->sin_addr.s_addr) { /* cache route doesn't match */ @@ -320,6 +323,10 @@ case 0: case 127: case 255: return 0; } + + /* + * XXXRW: Lock in_ifaddrhead walking. + */ /* reject packets with broadcast on source */ TAILQ_FOREACH(ia4, &in_ifaddrhead, ia_link) { if ((ia4->ia_ifa.ifa_ifp->if_flags & IFF_BROADCAST) == 0) @@ -328,6 +335,7 @@ return 0; } + /* XXXRW: unlocked read. */ /* ingress filters on outer source */ if ((sc->gif_if.if_flags & IFF_LINK2) == 0 && ifp) { struct sockaddr_in sin; @@ -383,6 +391,11 @@ in_gif_attach(sc) struct gif_softc *sc; { + + /* + * XXXRW: Technically, NULL can also be returned for ENOMEM, + * not just EEXIST. + */ sc->encap_cookie4 = encap_attach_func(AF_INET, -1, gif_encapcheck, &in_gif_protosw, sc); if (sc->encap_cookie4 == NULL) --- //depot/vendor/freebsd/src/sys/netinet/in_pcb.c 2004/03/27 13:10:42 +++ //depot/user/rwatson/netperf/sys/netinet/in_pcb.c 2004/03/28 07:24:45 @@ -687,6 +687,7 @@ inp->inp_gencnt = ++ipi->ipi_gencnt; in_pcbremlists(inp); if (so) { + SOCK_LOCK(so); so->so_pcb = 0; sotryfree(so); } --- //depot/vendor/freebsd/src/sys/netinet/in_pcb.h 2004/03/27 13:10:42 +++ //depot/user/rwatson/netperf/sys/netinet/in_pcb.h 2004/03/28 07:24:45 @@ -248,9 +248,14 @@ #define INP_LOCK(inp) mtx_lock(&(inp)->inp_mtx) #define INP_UNLOCK(inp) mtx_unlock(&(inp)->inp_mtx) #ifndef INET6 -#define INP_LOCK_ASSERT(inp) mtx_assert(&(inp)->inp_mtx, MA_OWNED) +#define INP_LOCK_ASSERT(inp) do { \ + mtx_assert(&(inp)->inp_mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) #else -#define INP_LOCK_ASSERT(inp) +#define INP_LOCK_ASSERT(inp) do { \ + NET_ASSERT_GIANT(); \ +} while (0) #endif #define INP_INFO_LOCK_INIT(ipi, d) \ @@ -260,11 +265,21 @@ #define INP_INFO_RUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_mtx) #define INP_INFO_WUNLOCK(ipi) mtx_unlock(&(ipi)->ipi_mtx) #ifndef INET6 -#define INP_INFO_RLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_mtx, MA_OWNED) -#define INP_INFO_WLOCK_ASSERT(ipi) mtx_assert(&(ipi)->ipi_mtx, MA_OWNED) +#define INP_INFO_RLOCK_ASSERT(ipi) do { \ + mtx_assert(&(ipi)->ipi_mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) +#define INP_INFO_WLOCK_ASSERT(ipi) do { \ + mtx_assert(&(ipi)->ipi_mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) #else -#define INP_INFO_RLOCK_ASSERT(ipi) -#define INP_INFO_WLOCK_ASSERT(ipi) +#define INP_INFO_RLOCK_ASSERT(ipi) do { \ + NET_ASSERT_GIANT(); \ +} while (0) +#define INP_INFO_WLOCK_ASSERT(ipi) do { \ + NET_ASSERT_GIANT(); \ +} while (0) #endif #define INP_PCBHASH(faddr, lport, fport, mask) \ --- //depot/vendor/freebsd/src/sys/netinet/in_proto.c 2004/02/25 11:55:45 +++ //depot/user/rwatson/netperf/sys/netinet/in_proto.c 2004/03/12 22:12:41 @@ -182,7 +182,7 @@ { SOCK_RAW, &inetdomain, IPPROTO_IPV4, PR_ATOMIC|PR_ADDR|PR_LASTHDR, encap4_input, 0, 0, rip_ctloutput, 0, - encap_init, 0, 0, 0, + encap_init, 0, 0, 0, &rip_usrreqs }, { SOCK_RAW, &inetdomain, IPPROTO_MOBILE, PR_ATOMIC|PR_ADDR|PR_LASTHDR, --- //depot/vendor/freebsd/src/sys/netinet/ip_divert.c 2004/03/27 13:10:42 +++ //depot/user/rwatson/netperf/sys/netinet/ip_divert.c 2004/03/28 07:24:45 @@ -223,20 +223,6 @@ sizeof(divsrc.sin_zero)); } - /* - * XXX sbappendaddr must be protected by Giant until - * we have locking at the socket layer. When entered - * from below we come in w/o Giant and must take it - * here. Unfortunately we cannot tell whether we're - * entering from above (already holding Giant), - * below (potentially without Giant), or otherwise - * (e.g. from tcp_syncache through a timeout) so we - * have to grab it regardless. This causes a LOR with - * the tcp lock, at least, and possibly others. For - * the moment we're ignoring this. Once sockets are - * locked this cruft can be removed. - */ - mtx_lock(&Giant); /* Put packet on socket queue, if any */ sa = NULL; nport = htons((u_int16_t)divert_info(mtag)); @@ -258,7 +244,6 @@ INP_UNLOCK(inp); } INP_INFO_RUNLOCK(&divcbinfo); - mtx_unlock(&Giant); if (sa == NULL) { m_freem(m); ipstat.ips_noproto++; --- //depot/vendor/freebsd/src/sys/netinet/ip_dummynet.c 2004/03/02 17:35:21 +++ //depot/user/rwatson/netperf/sys/netinet/ip_dummynet.c 2004/03/03 20:00:06 @@ -171,7 +171,10 @@ #define DUMMYNET_LOCK_DESTROY() mtx_destroy(&dummynet_mtx) #define DUMMYNET_LOCK() mtx_lock(&dummynet_mtx) #define DUMMYNET_UNLOCK() mtx_unlock(&dummynet_mtx) -#define DUMMYNET_LOCK_ASSERT() mtx_assert(&dummynet_mtx, MA_OWNED) +#define DUMMYNET_LOCK_ASSERT() do { \ + mtx_assert(&dummynet_mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) static int config_pipe(struct dn_pipe *p); static int ip_dn_ctl(struct sockopt *sopt); --- //depot/vendor/freebsd/src/sys/netinet/ip_encap.c 2004/03/09 18:50:38 +++ //depot/user/rwatson/netperf/sys/netinet/ip_encap.c 2004/03/09 19:47:45 @@ -1,4 +1,4 @@ -/* $FreeBSD: src/sys/netinet/ip_encap.c,v 1.19 2004/03/10 02:48:50 rwatson Exp $ */ +/* $FreeBSD: src/sys/netinet/ip_encap.c,v 1.18 2003/06/01 09:20:38 phk Exp $ */ /* $KAME: ip_encap.c,v 1.41 2001/03/15 08:35:08 itojun Exp $ */ /* @@ -106,8 +106,7 @@ LIST_HEAD(, encaptab) encaptab = LIST_HEAD_INITIALIZER(&encaptab); /* - * We currently keey encap_init() for source code compatibility reasons -- - * it's referenced by KAME pieces in netinet6. + * XXXRW: encap_init() was entirely useless, so I deleted it. */ void encap_init() @@ -185,6 +184,10 @@ } mtx_unlock(&encapmtx); + /* + * XXXRW: Need drain mechanism to prevent the encapsulation + * entry from being released while in use. + */ if (match) { /* found a match, "match" has the best one */ psw = match->psw; @@ -255,6 +258,10 @@ } mtx_unlock(&encapmtx); + /* + * XXXRW: Need drain mechanism so the encap entry isn't freed + * while in use. + */ if (match) { /* found a match */ psw = (const struct ip6protosw *)match->psw; --- //depot/vendor/freebsd/src/sys/netinet/ip_encap.h 2002/03/19 13:30:39 +++ //depot/user/rwatson/netperf/sys/netinet/ip_encap.h 2004/03/11 22:46:13 @@ -35,6 +35,15 @@ #ifdef _KERNEL +/* + * This structure is entirely static after registration, and other than + * its entry in the encapsulation table, requires no locking. The chain + * field is locked using the global encapmtx. + * + * XXXRW: Need to add a refcount/drain mechanism so that encapsulation + * entries can't be removed while in use. This likely requires a + * refcount and cv to wait for it to drain, or an sx lock. + */ struct encaptab { LIST_ENTRY(encaptab) chain; int af; --- //depot/vendor/freebsd/src/sys/netinet/ip_fastfwd.c 2004/02/25 11:55:45 +++ //depot/user/rwatson/netperf/sys/netinet/ip_fastfwd.c 2004/02/28 14:29:37 @@ -609,6 +609,7 @@ sizeof(struct sockaddr_in *), M_NOWAIT); if (mtag == NULL) { + /* XXX statistic */ if (ro.ro_rt) RTFREE(ro.ro_rt); goto drop; --- //depot/vendor/freebsd/src/sys/netinet/ip_fw2.c 2004/02/25 11:55:45 +++ //depot/user/rwatson/netperf/sys/netinet/ip_fw2.c 2004/02/29 22:51:24 @@ -122,7 +122,10 @@ #define IPFW_LOCK_DESTROY(_chain) mtx_destroy(&(_chain)->mtx) #define IPFW_LOCK(_chain) mtx_lock(&(_chain)->mtx) #define IPFW_UNLOCK(_chain) mtx_unlock(&(_chain)->mtx) -#define IPFW_LOCK_ASSERT(_chain) mtx_assert(&(_chain)->mtx, MA_OWNED) +#define IPFW_LOCK_ASSERT(_chain) do { \ + mtx_assert(&(_chain)->mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) /* * list of rules for layer 3 @@ -1296,7 +1299,8 @@ } static int -check_uidgid(ipfw_insn_u32 *insn, +check_uidgid(struct ip_fw_chain *chain, + ipfw_insn_u32 *insn, int proto, struct ifnet *oif, struct in_addr dst_ip, u_int16_t dst_port, struct in_addr src_ip, u_int16_t src_port) @@ -1317,7 +1321,10 @@ match = 0; - INP_INFO_RLOCK(pi); /* XXX LOR with IPFW */ + /* NB: reorder to avoid LOR between IPFW and inp */ + IPFW_UNLOCK(chain); + INP_INFO_RLOCK(pi); + IPFW_LOCK(chain); pcb = (oif) ? in_pcblookup_hash(pi, dst_ip, htons(dst_port), @@ -1657,7 +1664,7 @@ break; if (proto == IPPROTO_TCP || proto == IPPROTO_UDP) - match = check_uidgid( + match = check_uidgid(chain, (ipfw_insn_u32 *)cmd, proto, oif, dst_ip, dst_port, --- //depot/vendor/freebsd/src/sys/netinet/ip_id.c 2004/02/25 19:55:40 +++ //depot/user/rwatson/netperf/sys/netinet/ip_id.c 2004/03/11 19:03:57 @@ -79,6 +79,9 @@ 2729 }; +/* + * XXXRW: Locking? + */ static u_int16_t ru_x; static u_int16_t ru_seed, ru_seed2; static u_int16_t ru_a, ru_b; --- //depot/vendor/freebsd/src/sys/netinet/ip_input.c 2004/03/28 15:15:25 +++ //depot/user/rwatson/netperf/sys/netinet/ip_input.c 2004/03/28 17:06:00 @@ -931,8 +931,10 @@ /* attach next hop info for TCP */ struct m_tag *mtag = m_tag_get(PACKET_TAG_IPFORWARD, sizeof(struct sockaddr_in *), M_NOWAIT); - if (mtag == NULL) + if (mtag == NULL) { + /* XXX statistic */ goto bad; + } *(struct sockaddr_in **)(mtag+1) = args.next_hop; m_tag_prepend(m, mtag); } @@ -1854,6 +1856,7 @@ struct m_tag *mtag = m_tag_get(PACKET_TAG_IPFORWARD, sizeof(struct sockaddr_in *), M_NOWAIT); if (mtag == NULL) { + /* XXX statistic */ m_freem(m); return; } --- //depot/vendor/freebsd/src/sys/netinet/ip_mroute.c 2004/03/07 23:51:05 +++ //depot/user/rwatson/netperf/sys/netinet/ip_mroute.c 2004/03/09 11:24:20 @@ -103,7 +103,10 @@ static struct mtx mfc_mtx; #define MFC_LOCK() mtx_lock(&mfc_mtx) #define MFC_UNLOCK() mtx_unlock(&mfc_mtx) -#define MFC_LOCK_ASSERT() mtx_assert(&mfc_mtx, MA_OWNED) +#define MFC_LOCK_ASSERT() do { \ + mtx_assert(&mfc_mtx, MA_OWNED); \ + NET_ASSERT_GIANT(); \ +} while (0) #define MFC_LOCK_INIT() mtx_init(&mfc_mtx, "mroute mfc table", NULL, MTX_DEF) #define MFC_LOCK_DESTROY() mtx_destroy(&mfc_mtx) @@ -1303,13 +1306,10 @@ socket_send(struct socket *s, struct mbuf *mm, struct sockaddr_in *src) { if (s) { - mtx_lock(&Giant); /* XXX until sockets are locked */ if (sbappendaddr(&s->so_rcv, (struct sockaddr *)src, mm, NULL) != 0) { sorwakeup(s); - mtx_unlock(&Giant); return 0; } - mtx_unlock(&Giant); } m_freem(mm); return -1; --- //depot/vendor/freebsd/src/sys/netinet/ip_output.c 2004/03/25 00:50:40 +++ //depot/user/rwatson/netperf/sys/netinet/ip_output.c 2004/03/28 07:24:45 @@ -161,6 +161,12 @@ M_ASSERTPKTHDR(m); + /* + * When packet comes from dummynet restore state from + * previous processing instead of the header. Yech! + * + * XXX add conditional compilation? + */ args.next_hop = ip_claim_next_hop(m); dummytag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); if (dummytag != NULL) { @@ -181,6 +187,12 @@ ifp = dt->ifp; } +#ifdef IPSEC + /* XXXRW: so is not defined -- merge error from netperf+sockets? */ + so = ipsec_getsocket(m); + (void)ipsec_setsocket(m, NULL); +#endif /*IPSEC*/ + if (ro == NULL) { ro = &iproute; bzero(ro, sizeof (*ro)); @@ -875,6 +887,7 @@ PACKET_TAG_IPFORWARD, sizeof(struct sockaddr_in *), M_NOWAIT); if (mtag == NULL) { + /* XXX statistic */ error = ENOBUFS; goto bad; } @@ -893,6 +906,7 @@ CSUM_IP_CHECKED | CSUM_IP_VALID; ip->ip_len = htons(ip->ip_len); ip->ip_off = htons(ip->ip_off); + /* XXX netisr_queue(NETISR_IP, m); */ ip_input(m); goto done; } --- //depot/vendor/freebsd/src/sys/netinet/raw_ip.c 2004/03/27 12:45:55 +++ //depot/user/rwatson/netperf/sys/netinet/raw_ip.c 2004/03/28 07:24:45 @@ -90,6 +90,9 @@ * so leave them not initialized and rely on BSS being set to 0. */ +/* + * XXXRW: Locking for mrouter bits? + */ /* The socket used to communicate with the multicast routing daemon. */ struct socket *ip_mrouter; --- //depot/vendor/freebsd/src/sys/netinet/tcp_debug.c 2004/02/25 11:55:45 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_debug.c 2004/02/28 14:29:37 @@ -54,6 +54,7 @@ #include #include #include +#include #include #include --- //depot/vendor/freebsd/src/sys/netinet/tcp_input.c 2004/03/01 11:15:19 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_input.c 2004/03/01 16:04:44 @@ -428,7 +428,7 @@ struct tcpopt to; /* options in this segment */ struct rmxp_tao tao; /* our TAO cache entry */ int headlocked = 0; - struct sockaddr_in *next_hop = NULL; + struct sockaddr_in *next_hop; int rstreason; /* For badport_bandlim accounting purposes */ struct ip6_hdr *ip6 = NULL; @@ -1166,6 +1166,7 @@ acked = th->th_ack - tp->snd_una; tcpstat.tcps_rcvackpack++; tcpstat.tcps_rcvackbyte += acked; + SOCKBUF_LOCK(&so->so_snd); sbdrop(&so->so_snd, acked); if (SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) @@ -1203,7 +1204,8 @@ tp->t_rxtcur, tcp_timer_rexmt, tp); - sowwakeup(so); + sowwakeup_locked(so); + SOCKBUF_UNLOCK(&so->so_snd); if (so->so_snd.sb_cc) (void) tcp_output(tp); goto check_delack; @@ -2093,6 +2095,7 @@ incr = incr * incr / cw; tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<snd_scale); } + SOCKBUF_LOCK(&so->so_snd); if (acked > so->so_snd.sb_cc) { tp->snd_wnd -= so->so_snd.sb_cc; sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); @@ -2102,7 +2105,8 @@ tp->snd_wnd -= acked; ourfinisacked = 0; } - sowwakeup(so); + sowwakeup_locked(so); + SOCKBUF_UNLOCK(&so->so_snd); /* detect una wraparound */ if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && SEQ_GT(tp->snd_una, tp->snd_recover) && --- //depot/vendor/freebsd/src/sys/netinet/tcp_subr.c 2004/02/28 07:15:35 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_subr.c 2004/03/05 23:20:26 @@ -560,6 +560,7 @@ #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ + int callout_flag; tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) @@ -573,11 +574,17 @@ tcp_mssdflt; /* Set up our timeouts. */ - callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, 0); - callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, 0); - callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, 0); - callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, 0); - callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, 0); + /* + * XXXRW: Are these actually MPSAFE? I think so, but need to + * review the timed wait code, as it has some list variables, + * etc, that are global. + */ + callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0; + callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag); + callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag); + callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag); + callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag); + callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag); if (tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); @@ -1542,7 +1549,7 @@ /* * Move a TCP connection into TIME_WAIT state. - * tcbinfo is unlocked. + * tcbinfo is locked. * inp is locked, and is unlocked before returning. */ void @@ -1554,6 +1561,11 @@ int tw_time, acknow; struct socket *so; + INP_INFO_WLOCK_ASSERT(&tcbinfo); +#if 0 + INP_LOCK_ASSERT(tp); +#endif + tw = uma_zalloc(tcptw_zone, M_NOWAIT); if (tw == NULL) { tw = tcp_timer_2msl_tw(1); @@ -1604,13 +1616,19 @@ } tcp_discardcb(tp); so = inp->inp_socket; + SOCK_LOCK(so); so->so_pcb = NULL; tw->tw_cred = crhold(so->so_cred); tw->tw_so_options = so->so_options; + sotryfree(so); /* NB: drops lock */ + inp->inp_socket = NULL; if (acknow) tcp_twrespond(tw, TH_ACK); +#if 0 + /* XXXRW: Sam removed this, need to check why. */ sotryfree(so); inp->inp_socket = NULL; +#endif inp->inp_ppcb = (caddr_t)tw; inp->inp_vflag |= INP_TIMEWAIT; tcp_timer_2msl_reset(tw, tw_time); @@ -1686,6 +1704,8 @@ int isipv6 = inp->inp_inc.inc_isipv6; #endif + INP_LOCK_ASSERT(inp); + m = m_gethdr(M_DONTWAIT, MT_HEADER); if (m == NULL) return (ENOBUFS); --- //depot/vendor/freebsd/src/sys/netinet/tcp_syncache.c 2004/03/27 13:10:42 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_syncache.c 2004/03/28 07:24:45 @@ -540,7 +540,7 @@ struct socket *so; struct tcpcb *tp; - GIANT_REQUIRED; /* XXX until socket locking */ + NET_ASSERT_GIANT(); INP_INFO_WLOCK_ASSERT(&tcbinfo); /* --- //depot/vendor/freebsd/src/sys/netinet/tcp_timer.c 2003/11/20 12:10:44 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_timer.c 2004/03/06 21:55:09 @@ -273,6 +273,9 @@ } } +/* + * XXXRW: This doesn't look MPSAFE. + */ void tcp_timer_2msl_reset(struct tcptw *tw, int timeo) { @@ -287,6 +290,9 @@ LIST_INSERT_BEFORE(tw_tail, tw, tw_2msl); } +/* + * XXXRW: This doesn't look MPSAFE. + */ void tcp_timer_2msl_stop(struct tcptw *tw) { @@ -295,6 +301,9 @@ LIST_REMOVE(tw, tw_2msl); } +/* + * XXXRW: This doesn't look MPSAFE. + */ struct tcptw * tcp_timer_2msl_tw(int reuse) { --- //depot/vendor/freebsd/src/sys/netinet/tcp_usrreq.c 2004/03/28 07:50:52 +++ //depot/user/rwatson/netperf/sys/netinet/tcp_usrreq.c 2004/03/28 17:06:00 @@ -120,7 +120,6 @@ static int tcp_usr_attach(struct socket *so, int proto, struct thread *td) { - int s = splnet(); int error; struct inpcb *inp; struct tcpcb *tp = 0; @@ -146,11 +145,71 @@ out: TCPDEBUG2(PRU_ATTACH); INP_INFO_WUNLOCK(&tcbinfo); - splx(s); return error; } /* + * Common code to setup and teardown locking. Most + * code begins with a COMMON_START macro and finishes + * with COMMON_END. You indicate whether the inpcb + * and enclosing head are to be locked read or write + * and whether there is an existing sockbuf lock that + * needs to be re-ordered. + */ +#define INI_NOLOCK 0 /* no head lock */ +#define INI_READ 1 /* read head lock */ +#define INI_WRITE 2 /* write head lock */ +#define SBI_NONE 0 /* no sockbuf lock to reorder */ +#define SBI_SND 1 /* reorder so->so_snd lock */ +#define SBI_RCV 2 /* reorder so->so_rcv lock */ + +#define COMMON_START0(_headrw, _sbrw) do { \ + if (_sbrw == SBI_SND) \ + SOCKBUF_UNLOCK(&so->so_snd); \ + else if (_sbrw == SBI_RCV) \ + SOCKBUF_UNLOCK(&so->so_rcv); \ + if (_headrw == INI_READ) \ + INP_INFO_RLOCK(&tcbinfo); \ + else if (_headrw == INI_WRITE) \ + INP_INFO_WLOCK(&tcbinfo); \ + inp = sotoinpcb(so); \ + if (inp == 0) { \ + if (_sbrw == SBI_SND) \ + SOCKBUF_LOCK(&so->so_snd); \ + else if (_sbrw == SBI_RCV) \ + SOCKBUF_LOCK(&so->so_rcv); \ + if (_headrw == INI_READ) \ + INP_INFO_RUNLOCK(&tcbinfo); \ + else if (_headrw == INI_WRITE) \ + INP_INFO_WUNLOCK(&tcbinfo); \ + return EINVAL; \ + } \ + INP_LOCK(inp); \ + if (_sbrw == SBI_SND) \ + SOCKBUF_LOCK(&so->so_snd); \ + else if (_sbrw == SBI_RCV) \ + SOCKBUF_LOCK(&so->so_rcv); \ + if (_headrw == INI_READ) \ + INP_INFO_RUNLOCK(&tcbinfo); \ + tp = intotcpcb(inp); \ + TCPDEBUG1(); \ +} while(0) + +#define COMMON_START(_headrw, _sbrw) do { \ + TCPDEBUG0; \ + COMMON_START0(_headrw, _sbrw); \ +} while (0) + +#define COMMON_END(_headrw, req) \ + TCPDEBUG2(req); \ + do { \ + if (tp) \ + INP_UNLOCK(inp); \ + if (_headrw == INI_WRITE) \ + INP_INFO_WUNLOCK(&tcbinfo); \ + } while(0) + +/* * pru_detach() detaches the TCP protocol from the socket. * If the protocol state is non-embryonic, then can't * do this directly: have to initiate a pru_disconnect(), @@ -160,85 +219,28 @@ static int