2 * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
3 * Authors: Doug Rabson <dfr@rabson.org>
4 * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
32 #include <sys/fcntl.h>
33 #include <sys/kernel.h>
34 #include <sys/limits.h>
36 #include <sys/lockf.h>
37 #include <sys/malloc.h>
38 #include <sys/mount.h>
39 #include <sys/mutex.h>
41 #include <sys/syslog.h>
42 #include <sys/systm.h>
43 #include <sys/unistd.h>
44 #include <sys/vnode.h>
46 #include <rpc/rpcclnt.h>
47 #include <nfs/nfsproto.h>
48 #include <nfsclient/nfs.h>
49 #include <nfsclient/nfsnode.h>
50 #include <nfsclient/nfsmount.h>
52 #include <nlm/nlm_prot.h>
56 * We need to keep track of the svid values used for F_FLOCK locks.
58 struct nlm_file_svid {
59 int ns_refs; /* thread count + 1 if active */
60 int ns_svid; /* on-the-wire SVID for this file */
61 struct ucred *ns_ucred; /* creds to use for lock recovery */
62 void *ns_id; /* local struct file pointer */
63 bool_t ns_active; /* TRUE if we own a lock */
64 LIST_ENTRY(nlm_file_svid) ns_link;
66 LIST_HEAD(nlm_file_svid_list, nlm_file_svid);
68 #define NLM_SVID_HASH_SIZE 256
69 struct nlm_file_svid_list nlm_file_svids[NLM_SVID_HASH_SIZE];
71 struct mtx nlm_svid_lock;
72 static struct unrhdr *nlm_svid_allocator;
73 static volatile u_int nlm_xid = 1;
75 static int nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
76 rpcvers_t vers, struct timeval *timo, int retries,
77 struct vnode *vp, int op, struct flock *fl, int flags,
78 int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim);
79 static int nlm_clearlock(struct nlm_host *host, struct rpc_callextra *ext,
80 rpcvers_t vers, struct timeval *timo, int retries,
81 struct vnode *vp, int op, struct flock *fl, int flags,
82 int svid, size_t fhlen, void *fh, off_t size);
83 static int nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
84 rpcvers_t vers, struct timeval *timo, int retries,
85 struct vnode *vp, int op, struct flock *fl, int flags,
86 int svid, size_t fhlen, void *fh, off_t size);
87 static int nlm_map_status(nlm4_stats stat);
88 static struct nlm_file_svid *nlm_find_svid(void *id);
89 static void nlm_free_svid(struct nlm_file_svid *nf);
90 static int nlm_init_lock(struct flock *fl, int flags, int svid,
91 rpcvers_t vers, size_t fhlen, void *fh, off_t size,
92 struct nlm4_lock *lock, char oh_space[32]);
95 nlm_client_init(void *dummy)
99 mtx_init(&nlm_svid_lock, "NLM svid lock", NULL, MTX_DEF);
100 nlm_svid_allocator = new_unrhdr(PID_MAX + 2, INT_MAX, &nlm_svid_lock);
101 for (i = 0; i < NLM_SVID_HASH_SIZE; i++)
102 LIST_INIT(&nlm_file_svids[i]);
104 SYSINIT(nlm_client_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_client_init, NULL);
107 nlm_msg(struct thread *td, const char *server, const char *msg, int error)
111 p = td ? td->td_proc : NULL;
113 tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n", server,
116 tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
121 struct nlm_feedback_arg {
123 struct nfsmount *nf_nmp;
127 nlm_down(struct nlm_feedback_arg *nf, struct thread *td,
128 const char *msg, int error)
130 struct nfsmount *nmp = nf->nf_nmp;
134 mtx_lock(&nmp->nm_mtx);
135 if (!(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
136 nmp->nm_state |= NFSSTA_LOCKTIMEO;
137 mtx_unlock(&nmp->nm_mtx);
138 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
141 mtx_unlock(&nmp->nm_mtx);
144 nf->nf_printed = TRUE;
145 nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
149 nlm_up(struct nlm_feedback_arg *nf, struct thread *td,
152 struct nfsmount *nmp = nf->nf_nmp;
157 nlm_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
159 mtx_lock(&nmp->nm_mtx);
160 if (nmp->nm_state & NFSSTA_LOCKTIMEO) {
161 nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
162 mtx_unlock(&nmp->nm_mtx);
163 vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
166 mtx_unlock(&nmp->nm_mtx);
171 nlm_feedback(int type, int proc, void *arg)
173 struct thread *td = curthread;
174 struct nlm_feedback_arg *nf = (struct nlm_feedback_arg *) arg;
177 case FEEDBACK_REXMIT2:
178 case FEEDBACK_RECONNECT:
179 nlm_down(nf, td, "lockd not responding", 0);
183 nlm_up(nf, td, "lockd is alive again");
190 * NFS advisory byte-level locks.
193 nlm_advlock_internal(struct vnode *vp, void *id, int op, struct flock *fl,
194 int flags, bool_t reclaim, bool_t unlock_vp)
196 struct thread *td = curthread;
197 struct nfsmount *nmp;
203 struct sockaddr_storage ss;
204 char servername[MNAMELEN];
208 struct nlm_host *host;
209 struct rpc_callextra ext;
210 struct nlm_feedback_arg nf;
213 struct nlm_file_svid *ns;
217 ASSERT_VOP_LOCKED(vp, "nlm_advlock_1");
220 * Push any pending writes to the server and flush our cache
221 * so that if we are contending with another machine for a
222 * file, we get whatever they wrote and vice-versa.
224 if (op == F_SETLK || op == F_UNLCK)
225 nfs_vinvalbuf(vp, V_SAVE, td, 1);
228 nmp = VFSTONFS(vp->v_mount);
231 memcpy(&ss, sa, sa->sa_len);
232 sa = (struct sockaddr *) &ss;
233 strcpy(servername, nmp->nm_hostname);
234 fhlen = np->n_fhsize;
235 memcpy(&fh.fh_bytes, np->n_fhp, fhlen);
236 timo.tv_sec = nmp->nm_timeo / NFS_HZ;
237 timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * (1000000 / NFS_HZ);
243 if (nmp->nm_flag & NFSMNT_SOFT)
244 retries = nmp->nm_retry;
249 VOP_UNLOCK(vp, 0, td);
252 * We need to switch to mount-point creds so that we can send
253 * packets from a privileged port.
256 td->td_ucred = vp->v_mount->mnt_cred;
258 host = nlm_find_host_by_name(servername, sa, vers);
259 auth = authunix_create(cred);
260 memset(&ext, 0, sizeof(ext));
262 nf.nf_printed = FALSE;
266 ext.rc_feedback = nlm_feedback;
267 ext.rc_feedback_arg = &nf;
270 if (flags & F_FLOCK) {
271 ns = nlm_find_svid(id);
272 KASSERT(fl->l_start == 0 && fl->l_len == 0,
273 ("F_FLOCK lock requests must be whole-file locks"));
276 * Remember the creds used for locking in case
277 * we need to recover the lock later.
279 ns->ns_ucred = crdup(cred);
282 } else if (flags & F_REMOTE) {
284 * If we are recovering after a server restart or
285 * trashing locks on a force unmount, use the same
290 svid = ((struct proc *) id)->p_pid;
295 if ((flags & (F_FLOCK|F_WAIT)) == (F_FLOCK|F_WAIT)
296 && fl->l_type == F_WRLCK) {
298 * The semantics for flock(2) require that any
299 * shared lock on the file must be released
300 * before an exclusive lock is granted. The
301 * local locking code interprets this by
302 * unlocking the file before sleeping on a
303 * blocked exclusive lock request. We
304 * approximate this by first attempting
305 * non-blocking and if that fails, we unlock
306 * the file and block.
308 error = nlm_setlock(host, &ext, vers, &timo, retries,
309 vp, F_SETLK, fl, flags & ~F_WAIT,
310 svid, fhlen, &fh.fh_bytes, size, reclaim);
311 if (error == EAGAIN) {
312 fl->l_type = F_UNLCK;
313 error = nlm_clearlock(host, &ext, vers, &timo,
314 retries, vp, F_UNLCK, fl, flags,
315 svid, fhlen, &fh.fh_bytes, size);
316 fl->l_type = F_WRLCK;
318 mtx_lock(&nlm_svid_lock);
321 ns->ns_active = FALSE;
323 mtx_unlock(&nlm_svid_lock);
325 error = nlm_setlock(host, &ext, vers,
326 &timo, retries, vp, F_SETLK, fl,
327 flags, svid, fhlen, &fh.fh_bytes,
332 error = nlm_setlock(host, &ext, vers, &timo, retries,
333 vp, op, fl, flags, svid, fhlen, &fh.fh_bytes,
337 mtx_lock(&nlm_svid_lock);
338 if (!ns->ns_active) {
340 * Add one to the reference count to
341 * hold onto the SVID for the lifetime
342 * of the lock. Note that since
343 * F_FLOCK only supports whole-file
344 * locks, there can only be one active
345 * lock for this SVID.
348 ns->ns_active = TRUE;
350 mtx_unlock(&nlm_svid_lock);
355 error = nlm_clearlock(host, &ext, vers, &timo, retries,
356 vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
358 mtx_lock(&nlm_svid_lock);
361 ns->ns_active = FALSE;
363 mtx_unlock(&nlm_svid_lock);
368 error = nlm_getlock(host, &ext, vers, &timo, retries,
369 vp, op, fl, flags, svid, fhlen, &fh.fh_bytes, size);
383 nlm_host_release(host);
389 nlm_advlock(struct vop_advlock_args *ap)
392 return (nlm_advlock_internal(ap->a_vp, ap->a_id, ap->a_op, ap->a_fl,
393 ap->a_flags, FALSE, TRUE));
397 * Set the creds of td to the creds of the given lock's owner. The new
398 * creds reference count will be incremented via crhold. The caller is
399 * responsible for calling crfree and restoring td's original creds.
402 nlm_set_creds_for_lock(struct thread *td, struct flock *fl)
405 struct nlm_file_svid *ns;
410 if (fl->l_pid > PID_MAX) {
412 * If this was originally a F_FLOCK-style lock, we
413 * recorded the creds used when it was originally
414 * locked in the nlm_file_svid structure.
416 mtx_lock(&nlm_svid_lock);
417 for (i = 0; i < NLM_SVID_HASH_SIZE; i++) {
418 for (ns = LIST_FIRST(&nlm_file_svids[i]); ns;
419 ns = LIST_NEXT(ns, ns_link)) {
420 if (ns->ns_svid == fl->l_pid) {
421 cred = crhold(ns->ns_ucred);
426 mtx_unlock(&nlm_svid_lock);
429 * This lock is owned by a process. Get a reference to
432 p = pfind(fl->l_pid);
434 cred = crhold(p->p_ucred);
440 * If we can't find a cred, fall back on the recovery
444 cred = crhold(td->td_ucred);
451 nlm_reclaim_free_lock(struct vnode *vp, struct flock *fl, void *arg)
454 struct thread *td = curthread;
455 struct ucred *oldcred;
459 newfl.l_type = F_UNLCK;
461 oldcred = td->td_ucred;
462 nlm_set_creds_for_lock(td, &newfl);
464 error = nlm_advlock_internal(vp, NULL, F_UNLCK, &newfl, F_REMOTE,
467 crfree(td->td_ucred);
468 td->td_ucred = oldcred;
474 nlm_reclaim(struct vop_reclaim_args *ap)
477 nlm_cancel_wait(ap->a_vp);
478 lf_iteratelocks_vnode(ap->a_vp, nlm_reclaim_free_lock, NULL);
482 struct nlm_recovery_context {
483 struct nlm_host *nr_host; /* host we are recovering */
484 int nr_state; /* remote NSM state for recovery */
488 nlm_client_recover_lock(struct vnode *vp, struct flock *fl, void *arg)
490 struct nlm_recovery_context *nr = (struct nlm_recovery_context *) arg;
491 struct thread *td = curthread;
492 struct ucred *oldcred;
496 * If the remote NSM state changes during recovery, the host
497 * must have rebooted a second time. In that case, we must
498 * restart the recovery.
500 state = nlm_host_get_state(nr->nr_host);
501 if (nr->nr_state != state)
504 error = vn_lock(vp, LK_SHARED, td);
508 oldcred = td->td_ucred;
509 nlm_set_creds_for_lock(td, fl);
511 error = nlm_advlock_internal(vp, NULL, F_SETLK, fl, F_REMOTE,
514 crfree(td->td_ucred);
515 td->td_ucred = oldcred;
521 nlm_client_recovery(struct nlm_host *host)
523 struct nlm_recovery_context nr;
526 sysid = NLM_SYSID_CLIENT | nlm_host_get_sysid(host);
529 nr.nr_state = nlm_host_get_state(host);
530 error = lf_iteratelocks_sysid(sysid,
531 nlm_client_recover_lock, &nr);
532 } while (error == ERESTART);
536 nlm_convert_to_nlm_lock(struct nlm_lock *dst, struct nlm4_lock *src)
539 dst->caller_name = src->caller_name;
542 dst->svid = src->svid;
543 dst->l_offset = src->l_offset;
544 dst->l_len = src->l_len;
548 nlm_convert_to_nlm4_holder(struct nlm4_holder *dst, struct nlm_holder *src)
551 dst->exclusive = src->exclusive;
552 dst->svid = src->svid;
554 dst->l_offset = src->l_offset;
555 dst->l_len = src->l_len;
559 nlm_convert_to_nlm4_res(struct nlm4_res *dst, struct nlm_res *src)
561 dst->cookie = src->cookie;
562 dst->stat.stat = (enum nlm4_stats) src->stat.stat;
565 static enum clnt_stat
566 nlm_test_rpc(rpcvers_t vers, nlm4_testargs *args, nlm4_testres *res, CLIENT *client,
567 struct rpc_callextra *ext, struct timeval timo)
569 if (vers == NLM_VERS4) {
570 return nlm4_test_4(args, res, client, ext, timo);
576 args1.cookie = args->cookie;
577 args1.exclusive = args->exclusive;
578 nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
579 memset(&res1, 0, sizeof(res1));
581 stat = nlm_test_1(&args1, &res1, client, ext, timo);
583 if (stat == RPC_SUCCESS) {
584 res->cookie = res1.cookie;
585 res->stat.stat = (enum nlm4_stats) res1.stat.stat;
586 if (res1.stat.stat == nlm_denied)
587 nlm_convert_to_nlm4_holder(
588 &res->stat.nlm4_testrply_u.holder,
589 &res1.stat.nlm_testrply_u.holder);
596 static enum clnt_stat
597 nlm_lock_rpc(rpcvers_t vers, nlm4_lockargs *args, nlm4_res *res, CLIENT *client,
598 struct rpc_callextra *ext, struct timeval timo)
600 if (vers == NLM_VERS4) {
601 return nlm4_lock_4(args, res, client, ext, timo);
607 args1.cookie = args->cookie;
608 args1.block = args->block;
609 args1.exclusive = args->exclusive;
610 nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
611 args1.reclaim = args->reclaim;
612 args1.state = args->state;
613 memset(&res1, 0, sizeof(res1));
615 stat = nlm_lock_1(&args1, &res1, client, ext, timo);
617 if (stat == RPC_SUCCESS) {
618 nlm_convert_to_nlm4_res(res, &res1);
625 static enum clnt_stat
626 nlm_cancel_rpc(rpcvers_t vers, nlm4_cancargs *args, nlm4_res *res, CLIENT *client,
627 struct rpc_callextra *ext, struct timeval timo)
629 if (vers == NLM_VERS4) {
630 return nlm4_cancel_4(args, res, client, ext, timo);
636 args1.cookie = args->cookie;
637 args1.block = args->block;
638 args1.exclusive = args->exclusive;
639 nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
640 memset(&res1, 0, sizeof(res1));
642 stat = nlm_cancel_1(&args1, &res1, client, ext, timo);
644 if (stat == RPC_SUCCESS) {
645 nlm_convert_to_nlm4_res(res, &res1);
652 static enum clnt_stat
653 nlm_unlock_rpc(rpcvers_t vers, nlm4_unlockargs *args, nlm4_res *res, CLIENT *client,
654 struct rpc_callextra *ext, struct timeval timo)
656 if (vers == NLM_VERS4) {
657 return nlm4_unlock_4(args, res, client, ext, timo);
659 nlm_unlockargs args1;
663 args1.cookie = args->cookie;
664 nlm_convert_to_nlm_lock(&args1.alock, &args->alock);
665 memset(&res1, 0, sizeof(res1));
667 stat = nlm_unlock_1(&args1, &res1, client, ext, timo);
669 if (stat == RPC_SUCCESS) {
670 nlm_convert_to_nlm4_res(res, &res1);
678 * Called after a lock request (set or clear) succeeded. We record the
679 * details in the local lock manager. Note that since the remote
680 * server has granted the lock, we can be sure that it doesn't
681 * conflict with any other locks we have in the local lock manager.
683 * Since it is possible that host may also make NLM client requests to
684 * our NLM server, we use a different sysid value to record our own
687 * Note that since it is possible for us to receive replies from the
688 * server in a different order than the locks were granted (e.g. if
689 * many local threads are contending for the same lock), we must use a
690 * blocking operation when registering with the local lock manager.
691 * We expect that any actual wait will be rare and short hence we
692 * ignore signals for this.
695 nlm_record_lock(struct vnode *vp, int op, struct flock *fl,
696 int svid, int sysid, off_t size)
698 struct vop_advlockasync_args a;
706 a.a_flags = F_REMOTE|F_WAIT|F_NOINTR;
709 newfl.l_start = fl->l_start;
710 newfl.l_len = fl->l_len;
711 newfl.l_type = fl->l_type;
712 newfl.l_whence = fl->l_whence;
714 newfl.l_sysid = NLM_SYSID_CLIENT | sysid;
716 error = lf_advlockasync(&a, &vp->v_lockf, size);
717 KASSERT(error == 0, ("Failed to register NFS lock locally - error=%d",
722 nlm_setlock(struct nlm_host *host, struct rpc_callextra *ext,
723 rpcvers_t vers, struct timeval *timo, int retries,
724 struct vnode *vp, int op, struct flock *fl, int flags,
725 int svid, size_t fhlen, void *fh, off_t size, bool_t reclaim)
727 struct nlm4_lockargs args;
733 int retry, block, exclusive;
734 void *wait_handle = NULL;
737 memset(&args, 0, sizeof(args));
738 memset(&res, 0, sizeof(res));
740 block = (flags & F_WAIT) ? TRUE : FALSE;
741 exclusive = (fl->l_type == F_WRLCK);
743 error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
744 &args.alock, oh_space);
748 args.exclusive = exclusive;
749 args.reclaim = reclaim;
750 args.state = nlm_nsm_state;
754 client = nlm_host_get_rpc(host);
756 return (ENOLCK); /* XXX retry? */
759 wait_handle = nlm_register_wait_lock(&args.alock, vp);
761 xid = atomic_fetchadd_int(&nlm_xid, 1);
762 args.cookie.n_len = sizeof(xid);
763 args.cookie.n_bytes = (char*) &xid;
765 stat = nlm_lock_rpc(vers, &args, &res, client, ext, *timo);
767 CLNT_RELEASE(client);
769 if (stat != RPC_SUCCESS) {
771 nlm_deregister_wait_lock(wait_handle);
782 xdr_free((xdrproc_t) xdr_nlm4_res, &res);
784 if (block && res.stat.stat != nlm4_blocked)
785 nlm_deregister_wait_lock(wait_handle);
787 if (res.stat.stat == nlm4_denied_grace_period) {
789 * The server has recently rebooted and is
790 * giving old clients a change to reclaim
791 * their locks. Wait for a few seconds and try
794 error = tsleep(&args, PCATCH, "nlmgrace", retry);
795 if (error && error != EWOULDBLOCK)
803 if (block && res.stat.stat == nlm4_blocked) {
805 * The server should call us back with a
806 * granted message when the lock succeeds. In
807 * order to deal with broken servers, lost
808 * granted messages and server reboots, we
809 * will also re-try every few seconds.
811 error = nlm_wait_lock(wait_handle, retry);
812 if (error == EWOULDBLOCK) {
820 * We need to call the server to
821 * cancel our lock request.
823 nlm4_cancargs cancel;
825 memset(&cancel, 0, sizeof(cancel));
827 xid = atomic_fetchadd_int(&nlm_xid, 1);
828 cancel.cookie.n_len = sizeof(xid);
829 cancel.cookie.n_bytes = (char*) &xid;
830 cancel.block = block;
831 cancel.exclusive = exclusive;
832 cancel.alock = args.alock;
835 client = nlm_host_get_rpc(host);
840 stat = nlm_cancel_rpc(vers, &cancel,
841 &res, client, ext, *timo);
843 CLNT_RELEASE(client);
845 if (stat != RPC_SUCCESS) {
851 * reboots. This means
857 pause("nlmcancel", 10*hz);
859 } while (stat != RPC_SUCCESS);
864 xdr_free((xdrproc_t) xdr_nlm4_res, &res);
866 switch (res.stat.stat) {
878 case nlm4_denied_grace_period:
881 * recently rebooted -
907 error = nlm_map_status(res.stat.stat);
910 if (!error && !reclaim) {
911 nlm_record_lock(vp, op, fl, args.alock.svid,
912 nlm_host_get_sysid(host), size);
913 nlm_host_monitor(host, 0);
921 nlm_clearlock(struct nlm_host *host, struct rpc_callextra *ext,
922 rpcvers_t vers, struct timeval *timo, int retries,
923 struct vnode *vp, int op, struct flock *fl, int flags,
924 int svid, size_t fhlen, void *fh, off_t size)
926 struct nlm4_unlockargs args;
934 memset(&args, 0, sizeof(args));
935 memset(&res, 0, sizeof(res));
937 error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
938 &args.alock, oh_space);
943 client = nlm_host_get_rpc(host);
945 return (ENOLCK); /* XXX retry? */
947 xid = atomic_fetchadd_int(&nlm_xid, 1);
948 args.cookie.n_len = sizeof(xid);
949 args.cookie.n_bytes = (char*) &xid;
951 stat = nlm_unlock_rpc(vers, &args, &res, client, ext, *timo);
953 CLNT_RELEASE(client);
955 if (stat != RPC_SUCCESS) {
966 xdr_free((xdrproc_t) xdr_nlm4_res, &res);
968 if (res.stat.stat == nlm4_denied_grace_period) {
970 * The server has recently rebooted and is
971 * giving old clients a change to reclaim
972 * their locks. Wait for a few seconds and try
975 error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
976 if (error && error != EWOULDBLOCK)
982 * If we are being called via nlm_reclaim (which will
983 * use the F_REMOTE flag), don't record the lock
984 * operation in the local lock manager since the vnode
987 if (!(flags & F_REMOTE))
988 nlm_record_lock(vp, op, fl, args.alock.svid,
989 nlm_host_get_sysid(host), size);
996 nlm_getlock(struct nlm_host *host, struct rpc_callextra *ext,
997 rpcvers_t vers, struct timeval *timo, int retries,
998 struct vnode *vp, int op, struct flock *fl, int flags,
999 int svid, size_t fhlen, void *fh, off_t size)
1001 struct nlm4_testargs args;
1003 struct nlm4_testres res;
1006 enum clnt_stat stat;
1010 KASSERT(!(flags & F_FLOCK), ("unexpected F_FLOCK for F_GETLK"));
1012 memset(&args, 0, sizeof(args));
1013 memset(&res, 0, sizeof(res));
1015 exclusive = (fl->l_type == F_WRLCK);
1017 error = nlm_init_lock(fl, flags, svid, vers, fhlen, fh, size,
1018 &args.alock, oh_space);
1021 args.exclusive = exclusive;
1024 client = nlm_host_get_rpc(host);
1026 return (ENOLCK); /* XXX retry? */
1028 xid = atomic_fetchadd_int(&nlm_xid, 1);
1029 args.cookie.n_len = sizeof(xid);
1030 args.cookie.n_bytes = (char*) &xid;
1032 stat = nlm_test_rpc(vers, &args, &res, client, ext, *timo);
1034 CLNT_RELEASE(client);
1036 if (stat != RPC_SUCCESS) {
1044 if (res.stat.stat == nlm4_denied_grace_period) {
1046 * The server has recently rebooted and is
1047 * giving old clients a change to reclaim
1048 * their locks. Wait for a few seconds and try
1051 xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1052 error = tsleep(&args, PCATCH, "nlmgrace", 5*hz);
1053 if (error && error != EWOULDBLOCK)
1058 if (res.stat.stat == nlm4_denied) {
1059 struct nlm4_holder *h =
1060 &res.stat.nlm4_testrply_u.holder;
1061 fl->l_start = h->l_offset;
1062 fl->l_len = h->l_len;
1063 fl->l_pid = h->svid;
1065 fl->l_type = F_WRLCK;
1067 fl->l_type = F_RDLCK;
1068 fl->l_whence = SEEK_SET;
1071 fl->l_type = F_UNLCK;
1074 xdr_free((xdrproc_t) xdr_nlm4_testres, &res);
1081 nlm_map_status(nlm4_stats stat)
1090 case nlm4_denied_nolocks:
1113 static struct nlm_file_svid *
1114 nlm_find_svid(void *id)
1116 struct nlm_file_svid *ns, *newns;
1119 h = (((uintptr_t) id) >> 7) % NLM_SVID_HASH_SIZE;
1121 mtx_lock(&nlm_svid_lock);
1122 LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1123 if (ns->ns_id == id) {
1128 mtx_unlock(&nlm_svid_lock);
1130 int svid = alloc_unr(nlm_svid_allocator);
1131 newns = malloc(sizeof(struct nlm_file_svid), M_NLM,
1135 newns->ns_svid = svid;
1136 newns->ns_ucred = NULL;
1137 newns->ns_active = FALSE;
1140 * We need to check for a race with some other
1141 * thread allocating a svid for this file.
1143 mtx_lock(&nlm_svid_lock);
1144 LIST_FOREACH(ns, &nlm_file_svids[h], ns_link) {
1145 if (ns->ns_id == id) {
1151 mtx_unlock(&nlm_svid_lock);
1152 free_unr(nlm_svid_allocator, newns->ns_svid);
1155 LIST_INSERT_HEAD(&nlm_file_svids[h], newns,
1158 mtx_unlock(&nlm_svid_lock);
1166 nlm_free_svid(struct nlm_file_svid *ns)
1169 mtx_lock(&nlm_svid_lock);
1172 KASSERT(!ns->ns_active, ("Freeing active SVID"));
1173 LIST_REMOVE(ns, ns_link);
1174 mtx_unlock(&nlm_svid_lock);
1175 free_unr(nlm_svid_allocator, ns->ns_svid);
1177 crfree(ns->ns_ucred);
1180 mtx_unlock(&nlm_svid_lock);
1185 nlm_init_lock(struct flock *fl, int flags, int svid,
1186 rpcvers_t vers, size_t fhlen, void *fh, off_t size,
1187 struct nlm4_lock *lock, char oh_space[32])
1192 if (fl->l_whence == SEEK_END) {
1194 || (fl->l_start > 0 && size > OFF_MAX - fl->l_start))
1196 start = size + fl->l_start;
1197 } else if (fl->l_whence == SEEK_SET || fl->l_whence == SEEK_CUR) {
1198 start = fl->l_start;
1204 if (fl->l_len < 0) {
1213 if (vers == NLM_VERS) {
1215 * Enforce range limits on V1 locks
1217 if (start > 0xffffffffLL || len > 0xffffffffLL)
1221 snprintf(oh_space, 32, "%d@%s", svid, hostname);
1222 oh_len = strlen(oh_space);
1224 memset(lock, 0, sizeof(*lock));
1225 lock->caller_name = hostname;
1226 lock->fh.n_len = fhlen;
1227 lock->fh.n_bytes = fh;
1228 lock->oh.n_len = oh_len;
1229 lock->oh.n_bytes = oh_space;
1231 lock->l_offset = start;