Unbreak the tree, revert commitid aZ8fm4iaUnTCc0ul

This reverts the commit protecting the list and hashes in the PCB tables
with a mutex since the build of sysctl(8) breaks, as found by kettenis.

ok sthen

diff --git a/sys/kern/kern_sysctl.c b/sys/kern/kern_sysctl.c
index 85be1df..0f8d053 100644 (file)
--- a/sys/kern/kern_sysctl.c
+++ b/sys/kern/kern_sysctl.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: kern_sysctl.c,v 1.400 2022/03/14 17:23:00 bluhm Exp $ */
+/*     $OpenBSD: kern_sysctl.c,v 1.401 2022/03/14 22:38:43 tb Exp $    */
 /*     $NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $     */
 
 /*-
@@ -1366,24 +1366,16 @@ sysctl_file(int *name, u_int namelen, char *where, size_t *sizep,
                        struct inpcb *inp;
 
                        NET_LOCK();
-                       mtx_enter(&tcbtable.inpt_mtx);
                        TAILQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue)
                                FILLSO(inp->inp_socket);
-                       mtx_leave(&tcbtable.inpt_mtx);
-                       mtx_enter(&udbtable.inpt_mtx);
                        TAILQ_FOREACH(inp, &udbtable.inpt_queue, inp_queue)
                                FILLSO(inp->inp_socket);
-                       mtx_leave(&udbtable.inpt_mtx);
-                       mtx_enter(&rawcbtable.inpt_mtx);
                        TAILQ_FOREACH(inp, &rawcbtable.inpt_queue, inp_queue)
                                FILLSO(inp->inp_socket);
-                       mtx_leave(&rawcbtable.inpt_mtx);
 #ifdef INET6
-                       mtx_enter(&rawin6pcbtable.inpt_mtx);
                        TAILQ_FOREACH(inp, &rawin6pcbtable.inpt_queue,
                            inp_queue)
                                FILLSO(inp->inp_socket);
-                       mtx_leave(&rawin6pcbtable.inpt_mtx);
 #endif
                        NET_UNLOCK();
                }

diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c
index 8cddcee..561a8d2 100644 (file)
--- a/sys/netinet/in_pcb.c
+++ b/sys/netinet/in_pcb.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: in_pcb.c,v 1.260 2022/03/14 17:23:00 bluhm Exp $      */
+/*     $OpenBSD: in_pcb.c,v 1.261 2022/03/14 22:38:43 tb Exp $ */
 /*     $NetBSD: in_pcb.c,v 1.25 1996/02/13 23:41:53 christos Exp $     */
 
 /*
@@ -120,8 +120,7 @@ struct baddynamicports baddynamicports;
 struct baddynamicports rootonlyports;
 struct pool inpcb_pool;
 
-void   in_pcbrehash_locked(struct inpcb *);
-int    in_pcbresize(struct inpcbtable *, int);
+int in_pcbresize (struct inpcbtable *, int);
 
 #define        INPCBHASH_LOADFACTOR(_x)        (((_x) * 3) / 4)
 
@@ -174,7 +173,7 @@ in_pcblhash(struct inpcbtable *table, int rdom, u_short lport)
 void
 in_pcbinit(struct inpcbtable *table, int hashsize)
 {
-       mtx_init(&table->inpt_mtx, IPL_SOFTNET);
+
        TAILQ_INIT(&table->inpt_queue);
        table->inpt_hashtbl = hashinit(hashsize, M_PCB, M_WAITOK,
            &table->inpt_mask);
@@ -253,7 +252,6 @@ in_pcballoc(struct socket *so, struct inpcbtable *table)
        inp->inp_cksum6 = -1;
 #endif /* INET6 */
 
-       mtx_enter(&table->inpt_mtx);
        if (table->inpt_count++ > INPCBHASH_LOADFACTOR(table->inpt_size))
                (void)in_pcbresize(table, table->inpt_size * 2);
        TAILQ_INSERT_HEAD(&table->inpt_queue, inp, inp_queue);
@@ -270,8 +268,6 @@ in_pcballoc(struct socket *so, struct inpcbtable *table)
                    &inp->inp_faddr, inp->inp_fport,
                    &inp->inp_laddr, inp->inp_lport);
        LIST_INSERT_HEAD(head, inp, inp_hash);
-       mtx_leave(&table->inpt_mtx);
-
        so->so_pcb = inp;
 
        return (0);
@@ -560,7 +556,6 @@ void
 in_pcbdetach(struct inpcb *inp)
 {
        struct socket *so = inp->inp_socket;
-       struct inpcbtable *table = inp->inp_table;
 
        NET_ASSERT_LOCKED();
 
@@ -590,13 +585,10 @@ in_pcbdetach(struct inpcb *inp)
                pf_inp_unlink(inp);
        }
 #endif
-       mtx_enter(&table->inpt_mtx);
        LIST_REMOVE(inp, inp_lhash);
        LIST_REMOVE(inp, inp_hash);
-       TAILQ_REMOVE(&table->inpt_queue, inp, inp_queue);
-       table->inpt_count--;
-       mtx_leave(&table->inpt_mtx);
-
+       TAILQ_REMOVE(&inp->inp_table->inpt_queue, inp, inp_queue);
+       inp->inp_table->inpt_count--;
        in_pcbunref(inp);
 }
 
@@ -669,25 +661,20 @@ void
 in_pcbnotifyall(struct inpcbtable *table, struct sockaddr *dst, u_int rtable,
     int errno, void (*notify)(struct inpcb *, int))
 {
-       SIMPLEQ_HEAD(, inpcb) inpcblist;
-       struct inpcb *inp;
+       struct inpcb *inp, *ninp;
        struct in_addr faddr;
        u_int rdomain;
 
-       NET_ASSERT_WLOCKED();
+       NET_ASSERT_LOCKED();
 
        if (dst->sa_family != AF_INET)
                return;
        faddr = satosin(dst)->sin_addr;
        if (faddr.s_addr == INADDR_ANY)
                return;
-       if (notify == NULL)
-               return;
 
-       SIMPLEQ_INIT(&inpcblist);
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
-       TAILQ_FOREACH(inp, &table->inpt_queue, inp_queue) {
+       TAILQ_FOREACH_SAFE(inp, &table->inpt_queue, inp_queue, ninp) {
 #ifdef INET6
                if (inp->inp_flags & INP_IPV6)
                        continue;
@@ -697,15 +684,8 @@ in_pcbnotifyall(struct inpcbtable *table, struct sockaddr *dst, u_int rtable,
                    inp->inp_socket == NULL) {
                        continue;
                }
-               in_pcbref(inp);
-               SIMPLEQ_INSERT_TAIL(&inpcblist, inp, inp_notify);
-       }
-       mtx_leave(&table->inpt_mtx);
-
-       while ((inp = SIMPLEQ_FIRST(&inpcblist)) != NULL) {
-               SIMPLEQ_REMOVE_HEAD(&inpcblist, inp_notify);
-               (*notify)(inp, errno);
-               in_pcbunref(inp);
+               if (notify)
+                       (*notify)(inp, errno);
        }
 }
 
@@ -779,7 +759,6 @@ in_pcblookup_local(struct inpcbtable *table, void *laddrp, u_int lport_arg,
        u_int rdomain;
 
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
        head = in_pcblhash(table, rdomain, lport);
        LIST_FOREACH(inp, head, inp_lhash) {
                if (rtable_l2(inp->inp_rtableid) != rdomain)
@@ -830,8 +809,6 @@ in_pcblookup_local(struct inpcbtable *table, void *laddrp, u_int lport_arg,
                                break;
                }
        }
-       mtx_leave(&table->inpt_mtx);
-
        return (match);
 }
 
@@ -998,22 +975,11 @@ in_pcbselsrc(struct in_addr **insrc, struct sockaddr_in *sin,
 
 void
 in_pcbrehash(struct inpcb *inp)
-{
-       struct inpcbtable *table = inp->inp_table;
-
-       mtx_enter(&table->inpt_mtx);
-       in_pcbrehash_locked(inp);
-       mtx_leave(&table->inpt_mtx);
-}
-
-void
-in_pcbrehash_locked(struct inpcb *inp)
 {
        struct inpcbtable *table = inp->inp_table;
        struct inpcbhead *head;
 
        NET_ASSERT_LOCKED();
-       MUTEX_ASSERT_LOCKED(&table->inpt_mtx);
 
        LIST_REMOVE(inp, inp_lhash);
        head = in_pcblhash(table, inp->inp_rtableid, inp->inp_lport);
@@ -1040,8 +1006,6 @@ in_pcbresize(struct inpcbtable *table, int hashsize)
        void *nhashtbl, *nlhashtbl, *ohashtbl, *olhashtbl;
        struct inpcb *inp;
 
-       MUTEX_ASSERT_LOCKED(&table->inpt_mtx);
-
        ohashtbl = table->inpt_hashtbl;
        olhashtbl = table->inpt_lhashtbl;
        osize = table->inpt_size;
@@ -1063,7 +1027,7 @@ in_pcbresize(struct inpcbtable *table, int hashsize)
        arc4random_buf(&table->inpt_lkey, sizeof(table->inpt_lkey));
 
        TAILQ_FOREACH(inp, &table->inpt_queue, inp_queue) {
-               in_pcbrehash_locked(inp);
+               in_pcbrehash(inp);
        }
        hashfree(ohashtbl, osize, M_PCB);
        hashfree(olhashtbl, osize, M_PCB);
@@ -1094,7 +1058,6 @@ in_pcbhashlookup(struct inpcbtable *table, struct in_addr faddr,
        u_int rdomain;
 
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
        head = in_pcbhash(table, rdomain, &faddr, fport, &laddr, lport);
        LIST_FOREACH(inp, head, inp_hash) {
 #ifdef INET6
@@ -1117,7 +1080,6 @@ in_pcbhashlookup(struct inpcbtable *table, struct in_addr faddr,
                        break;
                }
        }
-       mtx_leave(&table->inpt_mtx);
 #ifdef DIAGNOSTIC
        if (inp == NULL && in_pcbnotifymiss) {
                printf("%s: faddr=%08x fport=%d laddr=%08x lport=%d rdom=%u\n",
@@ -1179,7 +1141,6 @@ in_pcblookup_listen(struct inpcbtable *table, struct in_addr laddr,
 #endif
 
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
        head = in_pcbhash(table, rdomain, &zeroin_addr, 0, key1, lport);
        LIST_FOREACH(inp, head, inp_hash) {
 #ifdef INET6
@@ -1216,7 +1177,6 @@ in_pcblookup_listen(struct inpcbtable *table, struct in_addr laddr,
                LIST_REMOVE(inp, inp_hash);
                LIST_INSERT_HEAD(head, inp, inp_hash);
        }
-       mtx_leave(&table->inpt_mtx);
 #ifdef DIAGNOSTIC
        if (inp == NULL && in_pcbnotifymiss) {
                printf("%s: laddr=%08x lport=%d rdom=%u\n",

diff --git a/sys/netinet/in_pcb.h b/sys/netinet/in_pcb.h
index 1db1bfb..1f72d73 100644 (file)
--- a/sys/netinet/in_pcb.h
+++ b/sys/netinet/in_pcb.h
@@ -1,4 +1,4 @@
-/*     $OpenBSD: in_pcb.h,v 1.124 2022/03/14 17:23:00 bluhm Exp $      */
+/*     $OpenBSD: in_pcb.h,v 1.125 2022/03/14 22:38:43 tb Exp $ */
 /*     $NetBSD: in_pcb.h,v 1.14 1996/02/13 23:42:00 christos Exp $     */
 
 /*
@@ -73,13 +73,6 @@
 
 #include <crypto/siphash.h>
 
-/*
- * Locks used to protect struct members in this file:
- *     I       immutable after creation
- *     N       net lock
- *     t       inpt_mtx                pcb table mutex
- */
-
 struct pf_state_key;
 
 union inpaddru {
@@ -98,11 +91,10 @@ union inpaddru {
  * control block.
  */
 struct inpcb {
-       LIST_ENTRY(inpcb) inp_hash;             /* [t] local and foreign hash */
-       LIST_ENTRY(inpcb) inp_lhash;            /* [t] local port hash */
-       TAILQ_ENTRY(inpcb) inp_queue;           /* [t] inet PCB queue */
-       SIMPLEQ_ENTRY(inpcb) inp_notify;        /* [N] queue to notify PCB */
-       struct    inpcbtable *inp_table;        /* [I] inet queue/hash table */
+       LIST_ENTRY(inpcb) inp_hash;             /* local and foreign hash */
+       LIST_ENTRY(inpcb) inp_lhash;            /* local port hash */
+       TAILQ_ENTRY(inpcb) inp_queue;           /* inet PCB queue */
+       struct    inpcbtable *inp_table;        /* inet queue/hash table */
        union     inpaddru inp_faddru;          /* Foreign address. */
        union     inpaddru inp_laddru;          /* Local address. */
 #define        inp_faddr       inp_faddru.iau_a4u.inaddr
@@ -162,13 +154,12 @@ struct inpcb {
 LIST_HEAD(inpcbhead, inpcb);
 
 struct inpcbtable {
-       struct mutex inpt_mtx;                  /* protect queue and hash */
-       TAILQ_HEAD(inpthead, inpcb) inpt_queue; /* [t] inet PCB queue */
-       struct  inpcbhead *inpt_hashtbl;        /* [t] local and foreign hash */
-       struct  inpcbhead *inpt_lhashtbl;       /* [t] local port hash */
-       SIPHASH_KEY inpt_key, inpt_lkey;        /* [t] secrets for hashes */
-       u_long  inpt_mask, inpt_lmask;          /* [t] hash masks */
-       int     inpt_count, inpt_size;          /* [t] queue count, hash size */
+       TAILQ_HEAD(inpthead, inpcb) inpt_queue; /* inet PCB queue */
+       struct  inpcbhead *inpt_hashtbl;        /* local and foreign hash */
+       struct  inpcbhead *inpt_lhashtbl;       /* local port hash */
+       SIPHASH_KEY inpt_key, inpt_lkey;        /* secrets for hashes */
+       u_long  inpt_mask, inpt_lmask;          /* hash masks */
+       int     inpt_count, inpt_size;          /* queue count, hash size */
 };
 
 /* flags in inp_flags: */

diff --git a/sys/netinet/raw_ip.c b/sys/netinet/raw_ip.c
index a25bef9..3dbf927 100644 (file)
--- a/sys/netinet/raw_ip.c
+++ b/sys/netinet/raw_ip.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: raw_ip.c,v 1.122 2022/03/14 17:23:00 bluhm Exp $      */
+/*     $OpenBSD: raw_ip.c,v 1.123 2022/03/14 22:38:43 tb Exp $ */
 /*     $NetBSD: raw_ip.c,v 1.25 1996/02/18 18:58:33 christos Exp $     */
 
 /*
@@ -151,7 +151,6 @@ rip_input(struct mbuf **mp, int *offp, int proto, int af)
        }
 #endif
        NET_ASSERT_LOCKED();
-       mtx_enter(&rawcbtable.inpt_mtx);
        TAILQ_FOREACH(inp, &rawcbtable.inpt_queue, inp_queue) {
                if (inp->inp_socket->so_state & SS_CANTRCVMORE)
                        continue;
@@ -191,8 +190,6 @@ rip_input(struct mbuf **mp, int *offp, int proto, int af)
                }
                last = inp;
        }
-       mtx_leave(&rawcbtable.inpt_mtx);
-
        if (last) {
                if (last->inp_flags & INP_CONTROLOPTS ||
                    last->inp_socket->so_options & SO_TIMESTAMP)

diff --git a/sys/netinet/udp_usrreq.c b/sys/netinet/udp_usrreq.c
index 76a65b8..b60e5bc 100644 (file)
--- a/sys/netinet/udp_usrreq.c
+++ b/sys/netinet/udp_usrreq.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: udp_usrreq.c,v 1.273 2022/03/14 17:23:00 bluhm Exp $  */
+/*     $OpenBSD: udp_usrreq.c,v 1.274 2022/03/14 22:38:43 tb Exp $     */
 /*     $NetBSD: udp_usrreq.c,v 1.28 1996/03/16 23:54:03 christos Exp $ */
 
 /*
@@ -365,7 +365,6 @@ udp_input(struct mbuf **mp, int *offp, int proto, int af)
                 */
                last = NULL;
                NET_ASSERT_LOCKED();
-               mtx_enter(&udbtable.inpt_mtx);
                TAILQ_FOREACH(inp, &udbtable.inpt_queue, inp_queue) {
                        if (inp->inp_socket->so_state & SS_CANTRCVMORE)
                                continue;
@@ -441,7 +440,6 @@ udp_input(struct mbuf **mp, int *offp, int proto, int af)
                            SO_REUSEADDR)) == 0)
                                break;
                }
-               mtx_leave(&udbtable.inpt_mtx);
 
                if (last == NULL) {
                        /*

diff --git a/sys/netinet6/in6_pcb.c b/sys/netinet6/in6_pcb.c
index e539008..65e5784 100644 (file)
--- a/sys/netinet6/in6_pcb.c
+++ b/sys/netinet6/in6_pcb.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: in6_pcb.c,v 1.114 2022/03/14 17:23:00 bluhm Exp $     */
+/*     $OpenBSD: in6_pcb.c,v 1.115 2022/03/14 22:38:43 tb Exp $        */
 
 /*
  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
@@ -369,15 +369,14 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
     uint fport_arg, const struct sockaddr_in6 *src, uint lport_arg,
     u_int rtable, int cmd, void *cmdarg, void (*notify)(struct inpcb *, int))
 {
-       SIMPLEQ_HEAD(, inpcb) inpcblist;
-       struct inpcb *inp;
+       struct inpcb *inp, *ninp;
        u_short fport = fport_arg, lport = lport_arg;
        struct sockaddr_in6 sa6_src;
        int errno;
        u_int32_t flowinfo;
        u_int rdomain;
 
-       NET_ASSERT_WLOCKED();
+       NET_ASSERT_LOCKED();
 
        if ((unsigned)cmd >= PRC_NCMDS)
                return;
@@ -415,13 +414,9 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
                        notify = in_rtchange;
        }
        errno = inet6ctlerrmap[cmd];
-       if (notify == NULL)
-               return;
 
-       SIMPLEQ_INIT(&inpcblist);
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
-       TAILQ_FOREACH(inp, &table->inpt_queue, inp_queue) {
+       TAILQ_FOREACH_SAFE(inp, &table->inpt_queue, inp_queue, ninp) {
                if ((inp->inp_flags & INP_IPV6) == 0)
                        continue;
 
@@ -493,15 +488,8 @@ in6_pcbnotify(struct inpcbtable *table, struct sockaddr_in6 *dst,
                        continue;
                }
          do_notify:
-               in_pcbref(inp);
-               SIMPLEQ_INSERT_TAIL(&inpcblist, inp, inp_notify);
-       }
-       mtx_leave(&table->inpt_mtx);
-
-       while ((inp = SIMPLEQ_FIRST(&inpcblist)) != NULL) {
-               SIMPLEQ_REMOVE_HEAD(&inpcblist, inp_notify);
-               (*notify)(inp, errno);
-               in_pcbunref(inp);
+               if (notify)
+                       (*notify)(inp, errno);
        }
 }
 
@@ -516,7 +504,6 @@ in6_pcbhashlookup(struct inpcbtable *table, const struct in6_addr *faddr,
        u_int rdomain;
 
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
        head = in6_pcbhash(table, rdomain, faddr, fport, laddr, lport);
        LIST_FOREACH(inp, head, inp_hash) {
                if (!(inp->inp_flags & INP_IPV6))
@@ -537,7 +524,6 @@ in6_pcbhashlookup(struct inpcbtable *table, const struct in6_addr *faddr,
                        break;
                }
        }
-       mtx_leave(&table->inpt_mtx);
 #ifdef DIAGNOSTIC
        if (inp == NULL && in_pcbnotifymiss) {
                printf("%s: faddr= fport=%d laddr= lport=%d rdom=%u\n",
@@ -588,7 +574,6 @@ in6_pcblookup_listen(struct inpcbtable *table, struct in6_addr *laddr,
 #endif
 
        rdomain = rtable_l2(rtable);
-       mtx_enter(&table->inpt_mtx);
        head = in6_pcbhash(table, rdomain, &zeroin6_addr, 0, key1, lport);
        LIST_FOREACH(inp, head, inp_hash) {
                if (!(inp->inp_flags & INP_IPV6))
@@ -621,7 +606,6 @@ in6_pcblookup_listen(struct inpcbtable *table, struct in6_addr *laddr,
                LIST_REMOVE(inp, inp_hash);
                LIST_INSERT_HEAD(head, inp, inp_hash);
        }
-       mtx_leave(&table->inpt_mtx);
 #ifdef DIAGNOSTIC
        if (inp == NULL && in_pcbnotifymiss) {
                printf("%s: laddr= lport=%d rdom=%u\n",

diff --git a/sys/netinet6/raw_ip6.c b/sys/netinet6/raw_ip6.c
index 3633485..6b588d6 100644 (file)
--- a/sys/netinet6/raw_ip6.c
+++ b/sys/netinet6/raw_ip6.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: raw_ip6.c,v 1.143 2022/03/14 17:23:00 bluhm Exp $     */
+/*     $OpenBSD: raw_ip6.c,v 1.144 2022/03/14 22:38:43 tb Exp $        */
 /*     $KAME: raw_ip6.c,v 1.69 2001/03/04 15:55:44 itojun Exp $        */
 
 /*
@@ -157,7 +157,6 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
        }
 #endif
        NET_ASSERT_LOCKED();
-       mtx_enter(&rawin6pcbtable.inpt_mtx);
        TAILQ_FOREACH(in6p, &rawin6pcbtable.inpt_queue, inp_queue) {
                if (in6p->inp_socket->so_state & SS_CANTRCVMORE)
                        continue;
@@ -181,10 +180,8 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
 
                        IP6_EXTHDR_GET(icmp6, struct icmp6_hdr *, m, *offp,
                            sizeof(*icmp6));
-                       if (icmp6 == NULL) {
-                               mtx_leave(&rawin6pcbtable.inpt_mtx);
+                       if (icmp6 == NULL)
                                return IPPROTO_DONE;
-                       }
                        if (ICMP6_FILTER_WILLBLOCK(icmp6->icmp6_type,
                            in6p->inp_icmp6filt))
                                continue;
@@ -227,8 +224,6 @@ rip6_input(struct mbuf **mp, int *offp, int proto, int af)
                }
                last = in6p;
        }
-       mtx_leave(&rawin6pcbtable.inpt_mtx);
-
        if (last) {
                if (last->inp_flags & IN6P_CONTROLOPTS)
                        ip6_savecontrol(last, m, &opts);