2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
48 #include "current_stateid.h"
53 #define NFSDDBG_FACILITY NFSDDBG_PROC
55 #define all_ones {{~0,~0},~0}
56 static const stateid_t one_stateid = {
58 .si_opaque = all_ones,
60 static const stateid_t zero_stateid = {
63 static const stateid_t currentstateid = {
66 static const stateid_t close_stateid = {
67 .si_generation = 0xffffffffU,
70 static u64 current_sessionid = 1;
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
75 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
77 /* forward declarations */
78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
84 * Currently used for the del_recall_lru and file hash table. In an
85 * effort to decrease the scope of the client_mutex, this spinlock may
86 * eventually cover more:
88 static DEFINE_SPINLOCK(state_lock);
90 enum nfsd4_st_mutex_lock_subclass {
91 OPEN_STATEID_MUTEX = 0,
92 LOCK_STATEID_MUTEX = 1,
96 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
97 * the refcount on the open stateid to drop.
99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
101 static struct kmem_cache *openowner_slab;
102 static struct kmem_cache *lockowner_slab;
103 static struct kmem_cache *file_slab;
104 static struct kmem_cache *stateid_slab;
105 static struct kmem_cache *deleg_slab;
106 static struct kmem_cache *odstate_slab;
108 static void free_session(struct nfsd4_session *);
110 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
111 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
113 static bool is_session_dead(struct nfsd4_session *ses)
115 return ses->se_flags & NFS4_SESSION_DEAD;
118 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
120 if (atomic_read(&ses->se_ref) > ref_held_by_me)
121 return nfserr_jukebox;
122 ses->se_flags |= NFS4_SESSION_DEAD;
126 static bool is_client_expired(struct nfs4_client *clp)
128 return clp->cl_time == 0;
131 static __be32 get_client_locked(struct nfs4_client *clp)
133 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
135 lockdep_assert_held(&nn->client_lock);
137 if (is_client_expired(clp))
138 return nfserr_expired;
139 atomic_inc(&clp->cl_refcount);
143 /* must be called under the client_lock */
145 renew_client_locked(struct nfs4_client *clp)
147 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
149 if (is_client_expired(clp)) {
151 printk("%s: client (clientid %08x/%08x) already expired\n",
153 clp->cl_clientid.cl_boot,
154 clp->cl_clientid.cl_id);
158 dprintk("renewing client (clientid %08x/%08x)\n",
159 clp->cl_clientid.cl_boot,
160 clp->cl_clientid.cl_id);
161 list_move_tail(&clp->cl_lru, &nn->client_lru);
162 clp->cl_time = get_seconds();
165 static void put_client_renew_locked(struct nfs4_client *clp)
167 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
169 lockdep_assert_held(&nn->client_lock);
171 if (!atomic_dec_and_test(&clp->cl_refcount))
173 if (!is_client_expired(clp))
174 renew_client_locked(clp);
177 static void put_client_renew(struct nfs4_client *clp)
179 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
181 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
183 if (!is_client_expired(clp))
184 renew_client_locked(clp);
185 spin_unlock(&nn->client_lock);
188 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
192 if (is_session_dead(ses))
193 return nfserr_badsession;
194 status = get_client_locked(ses->se_client);
197 atomic_inc(&ses->se_ref);
201 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
203 struct nfs4_client *clp = ses->se_client;
204 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
206 lockdep_assert_held(&nn->client_lock);
208 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
210 put_client_renew_locked(clp);
213 static void nfsd4_put_session(struct nfsd4_session *ses)
215 struct nfs4_client *clp = ses->se_client;
216 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
218 spin_lock(&nn->client_lock);
219 nfsd4_put_session_locked(ses);
220 spin_unlock(&nn->client_lock);
223 static struct nfsd4_blocked_lock *
224 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
227 struct nfsd4_blocked_lock *cur, *found = NULL;
229 spin_lock(&nn->blocked_locks_lock);
230 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
231 if (fh_match(fh, &cur->nbl_fh)) {
232 list_del_init(&cur->nbl_list);
233 list_del_init(&cur->nbl_lru);
238 spin_unlock(&nn->blocked_locks_lock);
240 posix_unblock_lock(&found->nbl_lock);
244 static struct nfsd4_blocked_lock *
245 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
248 struct nfsd4_blocked_lock *nbl;
250 nbl = find_blocked_lock(lo, fh, nn);
252 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
254 fh_copy_shallow(&nbl->nbl_fh, fh);
255 locks_init_lock(&nbl->nbl_lock);
256 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
257 &nfsd4_cb_notify_lock_ops,
258 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
265 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
267 locks_release_private(&nbl->nbl_lock);
272 remove_blocked_locks(struct nfs4_lockowner *lo)
274 struct nfs4_client *clp = lo->lo_owner.so_client;
275 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
276 struct nfsd4_blocked_lock *nbl;
279 /* Dequeue all blocked locks */
280 spin_lock(&nn->blocked_locks_lock);
281 while (!list_empty(&lo->lo_blocked)) {
282 nbl = list_first_entry(&lo->lo_blocked,
283 struct nfsd4_blocked_lock,
285 list_del_init(&nbl->nbl_list);
286 list_move(&nbl->nbl_lru, &reaplist);
288 spin_unlock(&nn->blocked_locks_lock);
291 while (!list_empty(&reaplist)) {
292 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
294 list_del_init(&nbl->nbl_lru);
295 posix_unblock_lock(&nbl->nbl_lock);
296 free_blocked_lock(nbl);
301 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
304 * Since this is just an optimization, we don't try very hard if it
305 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
306 * just quit trying on anything else.
308 switch (task->tk_status) {
310 rpc_delay(task, 1 * HZ);
318 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
320 struct nfsd4_blocked_lock *nbl = container_of(cb,
321 struct nfsd4_blocked_lock, nbl_cb);
323 free_blocked_lock(nbl);
326 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
327 .done = nfsd4_cb_notify_lock_done,
328 .release = nfsd4_cb_notify_lock_release,
331 static inline struct nfs4_stateowner *
332 nfs4_get_stateowner(struct nfs4_stateowner *sop)
334 atomic_inc(&sop->so_count);
339 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
341 return (sop->so_owner.len == owner->len) &&
342 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
345 static struct nfs4_openowner *
346 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
347 struct nfs4_client *clp)
349 struct nfs4_stateowner *so;
351 lockdep_assert_held(&clp->cl_lock);
353 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
355 if (!so->so_is_open_owner)
357 if (same_owner_str(so, &open->op_owner))
358 return openowner(nfs4_get_stateowner(so));
363 static struct nfs4_openowner *
364 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
365 struct nfs4_client *clp)
367 struct nfs4_openowner *oo;
369 spin_lock(&clp->cl_lock);
370 oo = find_openstateowner_str_locked(hashval, open, clp);
371 spin_unlock(&clp->cl_lock);
376 opaque_hashval(const void *ptr, int nbytes)
378 unsigned char *cptr = (unsigned char *) ptr;
388 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
390 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
392 kmem_cache_free(file_slab, fp);
396 put_nfs4_file(struct nfs4_file *fi)
398 might_lock(&state_lock);
400 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
401 hlist_del_rcu(&fi->fi_hash);
402 spin_unlock(&state_lock);
403 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
404 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
405 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
410 __nfs4_get_fd(struct nfs4_file *f, int oflag)
412 if (f->fi_fds[oflag])
413 return get_file(f->fi_fds[oflag]);
418 find_writeable_file_locked(struct nfs4_file *f)
422 lockdep_assert_held(&f->fi_lock);
424 ret = __nfs4_get_fd(f, O_WRONLY);
426 ret = __nfs4_get_fd(f, O_RDWR);
431 find_writeable_file(struct nfs4_file *f)
435 spin_lock(&f->fi_lock);
436 ret = find_writeable_file_locked(f);
437 spin_unlock(&f->fi_lock);
442 static struct file *find_readable_file_locked(struct nfs4_file *f)
446 lockdep_assert_held(&f->fi_lock);
448 ret = __nfs4_get_fd(f, O_RDONLY);
450 ret = __nfs4_get_fd(f, O_RDWR);
455 find_readable_file(struct nfs4_file *f)
459 spin_lock(&f->fi_lock);
460 ret = find_readable_file_locked(f);
461 spin_unlock(&f->fi_lock);
467 find_any_file(struct nfs4_file *f)
471 spin_lock(&f->fi_lock);
472 ret = __nfs4_get_fd(f, O_RDWR);
474 ret = __nfs4_get_fd(f, O_WRONLY);
476 ret = __nfs4_get_fd(f, O_RDONLY);
478 spin_unlock(&f->fi_lock);
482 static atomic_long_t num_delegations;
483 unsigned long max_delegations;
486 * Open owner state (share locks)
489 /* hash tables for lock and open owners */
490 #define OWNER_HASH_BITS 8
491 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
492 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
494 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
498 ret = opaque_hashval(ownername->data, ownername->len);
499 return ret & OWNER_HASH_MASK;
502 /* hash table for nfs4_file */
503 #define FILE_HASH_BITS 8
504 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
506 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
508 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
511 static unsigned int file_hashval(struct knfsd_fh *fh)
513 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
516 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
519 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
521 lockdep_assert_held(&fp->fi_lock);
523 if (access & NFS4_SHARE_ACCESS_WRITE)
524 atomic_inc(&fp->fi_access[O_WRONLY]);
525 if (access & NFS4_SHARE_ACCESS_READ)
526 atomic_inc(&fp->fi_access[O_RDONLY]);
530 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
532 lockdep_assert_held(&fp->fi_lock);
534 /* Does this access mode make sense? */
535 if (access & ~NFS4_SHARE_ACCESS_BOTH)
538 /* Does it conflict with a deny mode already set? */
539 if ((access & fp->fi_share_deny) != 0)
540 return nfserr_share_denied;
542 __nfs4_file_get_access(fp, access);
546 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
548 /* Common case is that there is no deny mode. */
550 /* Does this deny mode make sense? */
551 if (deny & ~NFS4_SHARE_DENY_BOTH)
554 if ((deny & NFS4_SHARE_DENY_READ) &&
555 atomic_read(&fp->fi_access[O_RDONLY]))
556 return nfserr_share_denied;
558 if ((deny & NFS4_SHARE_DENY_WRITE) &&
559 atomic_read(&fp->fi_access[O_WRONLY]))
560 return nfserr_share_denied;
565 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
567 might_lock(&fp->fi_lock);
569 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
570 struct file *f1 = NULL;
571 struct file *f2 = NULL;
573 swap(f1, fp->fi_fds[oflag]);
574 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
575 swap(f2, fp->fi_fds[O_RDWR]);
576 spin_unlock(&fp->fi_lock);
584 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
586 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
588 if (access & NFS4_SHARE_ACCESS_WRITE)
589 __nfs4_file_put_access(fp, O_WRONLY);
590 if (access & NFS4_SHARE_ACCESS_READ)
591 __nfs4_file_put_access(fp, O_RDONLY);
595 * Allocate a new open/delegation state counter. This is needed for
596 * pNFS for proper return on close semantics.
598 * Note that we only allocate it for pNFS-enabled exports, otherwise
599 * all pointers to struct nfs4_clnt_odstate are always NULL.
601 static struct nfs4_clnt_odstate *
602 alloc_clnt_odstate(struct nfs4_client *clp)
604 struct nfs4_clnt_odstate *co;
606 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
609 refcount_set(&co->co_odcount, 1);
615 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
617 struct nfs4_file *fp = co->co_file;
619 lockdep_assert_held(&fp->fi_lock);
620 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
624 get_clnt_odstate(struct nfs4_clnt_odstate *co)
627 refcount_inc(&co->co_odcount);
631 put_clnt_odstate(struct nfs4_clnt_odstate *co)
633 struct nfs4_file *fp;
639 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
640 list_del(&co->co_perfile);
641 spin_unlock(&fp->fi_lock);
643 nfsd4_return_all_file_layouts(co->co_client, fp);
644 kmem_cache_free(odstate_slab, co);
648 static struct nfs4_clnt_odstate *
649 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
651 struct nfs4_clnt_odstate *co;
652 struct nfs4_client *cl;
659 spin_lock(&fp->fi_lock);
660 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
661 if (co->co_client == cl) {
662 get_clnt_odstate(co);
668 hash_clnt_odstate_locked(new);
670 spin_unlock(&fp->fi_lock);
674 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
675 void (*sc_free)(struct nfs4_stid *))
677 struct nfs4_stid *stid;
680 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
684 idr_preload(GFP_KERNEL);
685 spin_lock(&cl->cl_lock);
686 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
687 spin_unlock(&cl->cl_lock);
692 stid->sc_free = sc_free;
693 stid->sc_client = cl;
694 stid->sc_stateid.si_opaque.so_id = new_id;
695 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
696 /* Will be incremented before return to client: */
697 refcount_set(&stid->sc_count, 1);
698 spin_lock_init(&stid->sc_lock);
701 * It shouldn't be a problem to reuse an opaque stateid value.
702 * I don't think it is for 4.1. But with 4.0 I worry that, for
703 * example, a stray write retransmission could be accepted by
704 * the server when it should have been rejected. Therefore,
705 * adopt a trick from the sctp code to attempt to maximize the
706 * amount of time until an id is reused, by ensuring they always
707 * "increase" (mod INT_MAX):
711 kmem_cache_free(slab, stid);
715 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
717 struct nfs4_stid *stid;
719 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
723 return openlockstateid(stid);
726 static void nfs4_free_deleg(struct nfs4_stid *stid)
728 kmem_cache_free(deleg_slab, stid);
729 atomic_long_dec(&num_delegations);
733 * When we recall a delegation, we should be careful not to hand it
734 * out again straight away.
735 * To ensure this we keep a pair of bloom filters ('new' and 'old')
736 * in which the filehandles of recalled delegations are "stored".
737 * If a filehandle appear in either filter, a delegation is blocked.
738 * When a delegation is recalled, the filehandle is stored in the "new"
740 * Every 30 seconds we swap the filters and clear the "new" one,
741 * unless both are empty of course.
743 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
744 * low 3 bytes as hash-table indices.
746 * 'blocked_delegations_lock', which is always taken in block_delegations(),
747 * is used to manage concurrent access. Testing does not need the lock
748 * except when swapping the two filters.
750 static DEFINE_SPINLOCK(blocked_delegations_lock);
751 static struct bloom_pair {
752 int entries, old_entries;
754 int new; /* index into 'set' */
755 DECLARE_BITMAP(set[2], 256);
756 } blocked_delegations;
758 static int delegation_blocked(struct knfsd_fh *fh)
761 struct bloom_pair *bd = &blocked_delegations;
763 if (bd->entries == 0)
765 if (seconds_since_boot() - bd->swap_time > 30) {
766 spin_lock(&blocked_delegations_lock);
767 if (seconds_since_boot() - bd->swap_time > 30) {
768 bd->entries -= bd->old_entries;
769 bd->old_entries = bd->entries;
770 memset(bd->set[bd->new], 0,
773 bd->swap_time = seconds_since_boot();
775 spin_unlock(&blocked_delegations_lock);
777 hash = jhash(&fh->fh_base, fh->fh_size, 0);
778 if (test_bit(hash&255, bd->set[0]) &&
779 test_bit((hash>>8)&255, bd->set[0]) &&
780 test_bit((hash>>16)&255, bd->set[0]))
783 if (test_bit(hash&255, bd->set[1]) &&
784 test_bit((hash>>8)&255, bd->set[1]) &&
785 test_bit((hash>>16)&255, bd->set[1]))
791 static void block_delegations(struct knfsd_fh *fh)
794 struct bloom_pair *bd = &blocked_delegations;
796 hash = jhash(&fh->fh_base, fh->fh_size, 0);
798 spin_lock(&blocked_delegations_lock);
799 __set_bit(hash&255, bd->set[bd->new]);
800 __set_bit((hash>>8)&255, bd->set[bd->new]);
801 __set_bit((hash>>16)&255, bd->set[bd->new]);
802 if (bd->entries == 0)
803 bd->swap_time = seconds_since_boot();
805 spin_unlock(&blocked_delegations_lock);
808 static struct nfs4_delegation *
809 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
810 struct nfs4_clnt_odstate *odstate)
812 struct nfs4_delegation *dp;
815 dprintk("NFSD alloc_init_deleg\n");
816 n = atomic_long_inc_return(&num_delegations);
817 if (n < 0 || n > max_delegations)
819 if (delegation_blocked(¤t_fh->fh_handle))
821 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
826 * delegation seqid's are never incremented. The 4.1 special
827 * meaning of seqid 0 isn't meaningful, really, but let's avoid
828 * 0 anyway just for consistency and use 1:
830 dp->dl_stid.sc_stateid.si_generation = 1;
831 INIT_LIST_HEAD(&dp->dl_perfile);
832 INIT_LIST_HEAD(&dp->dl_perclnt);
833 INIT_LIST_HEAD(&dp->dl_recall_lru);
834 dp->dl_clnt_odstate = odstate;
835 get_clnt_odstate(odstate);
836 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
838 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
839 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
842 atomic_long_dec(&num_delegations);
847 nfs4_put_stid(struct nfs4_stid *s)
849 struct nfs4_file *fp = s->sc_file;
850 struct nfs4_client *clp = s->sc_client;
852 might_lock(&clp->cl_lock);
854 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
855 wake_up_all(&close_wq);
858 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
859 spin_unlock(&clp->cl_lock);
866 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
868 stateid_t *src = &stid->sc_stateid;
870 spin_lock(&stid->sc_lock);
871 if (unlikely(++src->si_generation == 0))
872 src->si_generation = 1;
873 memcpy(dst, src, sizeof(*dst));
874 spin_unlock(&stid->sc_lock);
877 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
879 struct file *filp = NULL;
881 spin_lock(&fp->fi_lock);
882 if (fp->fi_deleg_file && --fp->fi_delegees == 0)
883 swap(filp, fp->fi_deleg_file);
884 spin_unlock(&fp->fi_lock);
887 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
892 void nfs4_unhash_stid(struct nfs4_stid *s)
898 * nfs4_get_existing_delegation - Discover if this delegation already exists
899 * @clp: a pointer to the nfs4_client we're granting a delegation to
900 * @fp: a pointer to the nfs4_file we're granting a delegation on
903 * On success: NULL if an existing delegation was not found.
905 * On error: -EAGAIN if one was previously granted to this nfs4_client
906 * for this nfs4_file.
911 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
913 struct nfs4_delegation *searchdp = NULL;
914 struct nfs4_client *searchclp = NULL;
916 lockdep_assert_held(&state_lock);
917 lockdep_assert_held(&fp->fi_lock);
919 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
920 searchclp = searchdp->dl_stid.sc_client;
921 if (clp == searchclp) {
929 * hash_delegation_locked - Add a delegation to the appropriate lists
930 * @dp: a pointer to the nfs4_delegation we are adding.
931 * @fp: a pointer to the nfs4_file we're granting a delegation on
934 * On success: NULL if the delegation was successfully hashed.
936 * On error: -EAGAIN if one was previously granted to this
937 * nfs4_client for this nfs4_file. Delegation is not hashed.
942 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
945 struct nfs4_client *clp = dp->dl_stid.sc_client;
947 lockdep_assert_held(&state_lock);
948 lockdep_assert_held(&fp->fi_lock);
950 status = nfs4_get_existing_delegation(clp, fp);
954 refcount_inc(&dp->dl_stid.sc_count);
955 dp->dl_stid.sc_type = NFS4_DELEG_STID;
956 list_add(&dp->dl_perfile, &fp->fi_delegations);
957 list_add(&dp->dl_perclnt, &clp->cl_delegations);
962 unhash_delegation_locked(struct nfs4_delegation *dp)
964 struct nfs4_file *fp = dp->dl_stid.sc_file;
966 lockdep_assert_held(&state_lock);
968 if (list_empty(&dp->dl_perfile))
971 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
972 /* Ensure that deleg break won't try to requeue it */
974 spin_lock(&fp->fi_lock);
975 list_del_init(&dp->dl_perclnt);
976 list_del_init(&dp->dl_recall_lru);
977 list_del_init(&dp->dl_perfile);
978 spin_unlock(&fp->fi_lock);
982 static void destroy_delegation(struct nfs4_delegation *dp)
986 spin_lock(&state_lock);
987 unhashed = unhash_delegation_locked(dp);
988 spin_unlock(&state_lock);
990 put_clnt_odstate(dp->dl_clnt_odstate);
991 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
992 nfs4_put_stid(&dp->dl_stid);
996 static void revoke_delegation(struct nfs4_delegation *dp)
998 struct nfs4_client *clp = dp->dl_stid.sc_client;
1000 WARN_ON(!list_empty(&dp->dl_recall_lru));
1002 put_clnt_odstate(dp->dl_clnt_odstate);
1003 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1005 if (clp->cl_minorversion == 0)
1006 nfs4_put_stid(&dp->dl_stid);
1008 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1009 spin_lock(&clp->cl_lock);
1010 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1011 spin_unlock(&clp->cl_lock);
1019 static unsigned int clientid_hashval(u32 id)
1021 return id & CLIENT_HASH_MASK;
1024 static unsigned int clientstr_hashval(const char *name)
1026 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1030 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1031 * st_{access,deny}_bmap field of the stateid, in order to track not
1032 * only what share bits are currently in force, but also what
1033 * combinations of share bits previous opens have used. This allows us
1034 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1035 * return an error if the client attempt to downgrade to a combination
1036 * of share bits not explicable by closing some of its previous opens.
1038 * XXX: This enforcement is actually incomplete, since we don't keep
1039 * track of access/deny bit combinations; so, e.g., we allow:
1041 * OPEN allow read, deny write
1042 * OPEN allow both, deny none
1043 * DOWNGRADE allow read, deny none
1045 * which we should reject.
1048 bmap_to_share_mode(unsigned long bmap) {
1050 unsigned int access = 0;
1052 for (i = 1; i < 4; i++) {
1053 if (test_bit(i, &bmap))
1059 /* set share access for a given stateid */
1061 set_access(u32 access, struct nfs4_ol_stateid *stp)
1063 unsigned char mask = 1 << access;
1065 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1066 stp->st_access_bmap |= mask;
1069 /* clear share access for a given stateid */
1071 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1073 unsigned char mask = 1 << access;
1075 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1076 stp->st_access_bmap &= ~mask;
1079 /* test whether a given stateid has access */
1081 test_access(u32 access, struct nfs4_ol_stateid *stp)
1083 unsigned char mask = 1 << access;
1085 return (bool)(stp->st_access_bmap & mask);
1088 /* set share deny for a given stateid */
1090 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1092 unsigned char mask = 1 << deny;
1094 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1095 stp->st_deny_bmap |= mask;
1098 /* clear share deny for a given stateid */
1100 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1102 unsigned char mask = 1 << deny;
1104 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1105 stp->st_deny_bmap &= ~mask;
1108 /* test whether a given stateid is denying specific access */
1110 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1112 unsigned char mask = 1 << deny;
1114 return (bool)(stp->st_deny_bmap & mask);
1117 static int nfs4_access_to_omode(u32 access)
1119 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1120 case NFS4_SHARE_ACCESS_READ:
1122 case NFS4_SHARE_ACCESS_WRITE:
1124 case NFS4_SHARE_ACCESS_BOTH:
1132 * A stateid that had a deny mode associated with it is being released
1133 * or downgraded. Recalculate the deny mode on the file.
1136 recalculate_deny_mode(struct nfs4_file *fp)
1138 struct nfs4_ol_stateid *stp;
1140 spin_lock(&fp->fi_lock);
1141 fp->fi_share_deny = 0;
1142 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1143 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1144 spin_unlock(&fp->fi_lock);
1148 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1151 bool change = false;
1153 for (i = 1; i < 4; i++) {
1154 if ((i & deny) != i) {
1160 /* Recalculate per-file deny mode if there was a change */
1162 recalculate_deny_mode(stp->st_stid.sc_file);
1165 /* release all access and file references for a given stateid */
1167 release_all_access(struct nfs4_ol_stateid *stp)
1170 struct nfs4_file *fp = stp->st_stid.sc_file;
1172 if (fp && stp->st_deny_bmap != 0)
1173 recalculate_deny_mode(fp);
1175 for (i = 1; i < 4; i++) {
1176 if (test_access(i, stp))
1177 nfs4_file_put_access(stp->st_stid.sc_file, i);
1178 clear_access(i, stp);
1182 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1184 kfree(sop->so_owner.data);
1185 sop->so_ops->so_free(sop);
1188 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1190 struct nfs4_client *clp = sop->so_client;
1192 might_lock(&clp->cl_lock);
1194 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1196 sop->so_ops->so_unhash(sop);
1197 spin_unlock(&clp->cl_lock);
1198 nfs4_free_stateowner(sop);
1201 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1203 struct nfs4_file *fp = stp->st_stid.sc_file;
1205 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1207 if (list_empty(&stp->st_perfile))
1210 spin_lock(&fp->fi_lock);
1211 list_del_init(&stp->st_perfile);
1212 spin_unlock(&fp->fi_lock);
1213 list_del(&stp->st_perstateowner);
1217 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1219 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1221 put_clnt_odstate(stp->st_clnt_odstate);
1222 release_all_access(stp);
1223 if (stp->st_stateowner)
1224 nfs4_put_stateowner(stp->st_stateowner);
1225 kmem_cache_free(stateid_slab, stid);
1228 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1230 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1231 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1234 file = find_any_file(stp->st_stid.sc_file);
1236 filp_close(file, (fl_owner_t)lo);
1237 nfs4_free_ol_stateid(stid);
1241 * Put the persistent reference to an already unhashed generic stateid, while
1242 * holding the cl_lock. If it's the last reference, then put it onto the
1243 * reaplist for later destruction.
1245 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1246 struct list_head *reaplist)
1248 struct nfs4_stid *s = &stp->st_stid;
1249 struct nfs4_client *clp = s->sc_client;
1251 lockdep_assert_held(&clp->cl_lock);
1253 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1255 if (!refcount_dec_and_test(&s->sc_count)) {
1256 wake_up_all(&close_wq);
1260 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1261 list_add(&stp->st_locks, reaplist);
1264 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1266 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1268 list_del_init(&stp->st_locks);
1269 nfs4_unhash_stid(&stp->st_stid);
1270 return unhash_ol_stateid(stp);
1273 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1275 struct nfs4_client *clp = stp->st_stid.sc_client;
1278 spin_lock(&clp->cl_lock);
1279 unhashed = unhash_lock_stateid(stp);
1280 spin_unlock(&clp->cl_lock);
1282 nfs4_put_stid(&stp->st_stid);
1285 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1287 struct nfs4_client *clp = lo->lo_owner.so_client;
1289 lockdep_assert_held(&clp->cl_lock);
1291 list_del_init(&lo->lo_owner.so_strhash);
1295 * Free a list of generic stateids that were collected earlier after being
1299 free_ol_stateid_reaplist(struct list_head *reaplist)
1301 struct nfs4_ol_stateid *stp;
1302 struct nfs4_file *fp;
1306 while (!list_empty(reaplist)) {
1307 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1309 list_del(&stp->st_locks);
1310 fp = stp->st_stid.sc_file;
1311 stp->st_stid.sc_free(&stp->st_stid);
1317 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1318 struct list_head *reaplist)
1320 struct nfs4_ol_stateid *stp;
1322 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1324 while (!list_empty(&open_stp->st_locks)) {
1325 stp = list_entry(open_stp->st_locks.next,
1326 struct nfs4_ol_stateid, st_locks);
1327 WARN_ON(!unhash_lock_stateid(stp));
1328 put_ol_stateid_locked(stp, reaplist);
1332 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1333 struct list_head *reaplist)
1337 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1339 unhashed = unhash_ol_stateid(stp);
1340 release_open_stateid_locks(stp, reaplist);
1344 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1346 LIST_HEAD(reaplist);
1348 spin_lock(&stp->st_stid.sc_client->cl_lock);
1349 if (unhash_open_stateid(stp, &reaplist))
1350 put_ol_stateid_locked(stp, &reaplist);
1351 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1352 free_ol_stateid_reaplist(&reaplist);
1355 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1357 struct nfs4_client *clp = oo->oo_owner.so_client;
1359 lockdep_assert_held(&clp->cl_lock);
1361 list_del_init(&oo->oo_owner.so_strhash);
1362 list_del_init(&oo->oo_perclient);
1365 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1367 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1369 struct nfs4_ol_stateid *s;
1371 spin_lock(&nn->client_lock);
1372 s = oo->oo_last_closed_stid;
1374 list_del_init(&oo->oo_close_lru);
1375 oo->oo_last_closed_stid = NULL;
1377 spin_unlock(&nn->client_lock);
1379 nfs4_put_stid(&s->st_stid);
1382 static void release_openowner(struct nfs4_openowner *oo)
1384 struct nfs4_ol_stateid *stp;
1385 struct nfs4_client *clp = oo->oo_owner.so_client;
1386 struct list_head reaplist;
1388 INIT_LIST_HEAD(&reaplist);
1390 spin_lock(&clp->cl_lock);
1391 unhash_openowner_locked(oo);
1392 while (!list_empty(&oo->oo_owner.so_stateids)) {
1393 stp = list_first_entry(&oo->oo_owner.so_stateids,
1394 struct nfs4_ol_stateid, st_perstateowner);
1395 if (unhash_open_stateid(stp, &reaplist))
1396 put_ol_stateid_locked(stp, &reaplist);
1398 spin_unlock(&clp->cl_lock);
1399 free_ol_stateid_reaplist(&reaplist);
1400 release_last_closed_stateid(oo);
1401 nfs4_put_stateowner(&oo->oo_owner);
1405 hash_sessionid(struct nfs4_sessionid *sessionid)
1407 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1409 return sid->sequence % SESSION_HASH_SIZE;
1412 #ifdef CONFIG_SUNRPC_DEBUG
1414 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1416 u32 *ptr = (u32 *)(&sessionid->data[0]);
1417 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1427 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1428 * won't be used for replay.
1430 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1432 struct nfs4_stateowner *so = cstate->replay_owner;
1434 if (nfserr == nfserr_replay_me)
1437 if (!seqid_mutating_err(ntohl(nfserr))) {
1438 nfsd4_cstate_clear_replay(cstate);
1443 if (so->so_is_open_owner)
1444 release_last_closed_stateid(openowner(so));
1450 gen_sessionid(struct nfsd4_session *ses)
1452 struct nfs4_client *clp = ses->se_client;
1453 struct nfsd4_sessionid *sid;
1455 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1456 sid->clientid = clp->cl_clientid;
1457 sid->sequence = current_sessionid++;
1462 * The protocol defines ca_maxresponssize_cached to include the size of
1463 * the rpc header, but all we need to cache is the data starting after
1464 * the end of the initial SEQUENCE operation--the rest we regenerate
1465 * each time. Therefore we can advertise a ca_maxresponssize_cached
1466 * value that is the number of bytes in our cache plus a few additional
1467 * bytes. In order to stay on the safe side, and not promise more than
1468 * we can cache, those additional bytes must be the minimum possible: 24
1469 * bytes of rpc header (xid through accept state, with AUTH_NULL
1470 * verifier), 12 for the compound header (with zero-length tag), and 44
1471 * for the SEQUENCE op response:
1473 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1476 free_session_slots(struct nfsd4_session *ses)
1480 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1481 free_svc_cred(&ses->se_slots[i]->sl_cred);
1482 kfree(ses->se_slots[i]);
1487 * We don't actually need to cache the rpc and session headers, so we
1488 * can allocate a little less for each slot:
1490 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1494 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1497 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1498 return size + sizeof(struct nfsd4_slot);
1502 * XXX: If we run out of reserved DRC memory we could (up to a point)
1503 * re-negotiate active sessions and reduce their slot usage to make
1504 * room for new connections. For now we just fail the create session.
1506 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1508 u32 slotsize = slot_bytes(ca);
1509 u32 num = ca->maxreqs;
1512 spin_lock(&nfsd_drc_lock);
1513 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1514 nfsd_drc_max_mem - nfsd_drc_mem_used);
1516 * Never use more than a third of the remaining memory,
1517 * unless it's the only way to give this client a slot:
1519 avail = clamp_t(int, avail, slotsize, avail/3);
1520 num = min_t(int, num, avail / slotsize);
1521 nfsd_drc_mem_used += num * slotsize;
1522 spin_unlock(&nfsd_drc_lock);
1527 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1529 int slotsize = slot_bytes(ca);
1531 spin_lock(&nfsd_drc_lock);
1532 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1533 spin_unlock(&nfsd_drc_lock);
1536 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1537 struct nfsd4_channel_attrs *battrs)
1539 int numslots = fattrs->maxreqs;
1540 int slotsize = slot_bytes(fattrs);
1541 struct nfsd4_session *new;
1544 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1545 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1546 mem = numslots * sizeof(struct nfsd4_slot *);
1548 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1551 /* allocate each struct nfsd4_slot and data cache in one piece */
1552 for (i = 0; i < numslots; i++) {
1553 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1554 if (!new->se_slots[i])
1558 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1559 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1564 kfree(new->se_slots[i]);
1569 static void free_conn(struct nfsd4_conn *c)
1571 svc_xprt_put(c->cn_xprt);
1575 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1577 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1578 struct nfs4_client *clp = c->cn_session->se_client;
1580 spin_lock(&clp->cl_lock);
1581 if (!list_empty(&c->cn_persession)) {
1582 list_del(&c->cn_persession);
1585 nfsd4_probe_callback(clp);
1586 spin_unlock(&clp->cl_lock);
1589 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1591 struct nfsd4_conn *conn;
1593 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1596 svc_xprt_get(rqstp->rq_xprt);
1597 conn->cn_xprt = rqstp->rq_xprt;
1598 conn->cn_flags = flags;
1599 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1603 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1605 conn->cn_session = ses;
1606 list_add(&conn->cn_persession, &ses->se_conns);
1609 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1611 struct nfs4_client *clp = ses->se_client;
1613 spin_lock(&clp->cl_lock);
1614 __nfsd4_hash_conn(conn, ses);
1615 spin_unlock(&clp->cl_lock);
1618 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1620 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1621 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1624 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1628 nfsd4_hash_conn(conn, ses);
1629 ret = nfsd4_register_conn(conn);
1631 /* oops; xprt is already down: */
1632 nfsd4_conn_lost(&conn->cn_xpt_user);
1633 /* We may have gained or lost a callback channel: */
1634 nfsd4_probe_callback_sync(ses->se_client);
1637 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1639 u32 dir = NFS4_CDFC4_FORE;
1641 if (cses->flags & SESSION4_BACK_CHAN)
1642 dir |= NFS4_CDFC4_BACK;
1643 return alloc_conn(rqstp, dir);
1646 /* must be called under client_lock */
1647 static void nfsd4_del_conns(struct nfsd4_session *s)
1649 struct nfs4_client *clp = s->se_client;
1650 struct nfsd4_conn *c;
1652 spin_lock(&clp->cl_lock);
1653 while (!list_empty(&s->se_conns)) {
1654 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1655 list_del_init(&c->cn_persession);
1656 spin_unlock(&clp->cl_lock);
1658 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1661 spin_lock(&clp->cl_lock);
1663 spin_unlock(&clp->cl_lock);
1666 static void __free_session(struct nfsd4_session *ses)
1668 free_session_slots(ses);
1672 static void free_session(struct nfsd4_session *ses)
1674 nfsd4_del_conns(ses);
1675 nfsd4_put_drc_mem(&ses->se_fchannel);
1676 __free_session(ses);
1679 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1682 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1684 new->se_client = clp;
1687 INIT_LIST_HEAD(&new->se_conns);
1689 new->se_cb_seq_nr = 1;
1690 new->se_flags = cses->flags;
1691 new->se_cb_prog = cses->callback_prog;
1692 new->se_cb_sec = cses->cb_sec;
1693 atomic_set(&new->se_ref, 0);
1694 idx = hash_sessionid(&new->se_sessionid);
1695 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1696 spin_lock(&clp->cl_lock);
1697 list_add(&new->se_perclnt, &clp->cl_sessions);
1698 spin_unlock(&clp->cl_lock);
1701 struct sockaddr *sa = svc_addr(rqstp);
1703 * This is a little silly; with sessions there's no real
1704 * use for the callback address. Use the peer address
1705 * as a reasonable default for now, but consider fixing
1706 * the rpc client not to require an address in the
1709 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1710 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1714 /* caller must hold client_lock */
1715 static struct nfsd4_session *
1716 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1718 struct nfsd4_session *elem;
1720 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1722 lockdep_assert_held(&nn->client_lock);
1724 dump_sessionid(__func__, sessionid);
1725 idx = hash_sessionid(sessionid);
1726 /* Search in the appropriate list */
1727 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1728 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1729 NFS4_MAX_SESSIONID_LEN)) {
1734 dprintk("%s: session not found\n", __func__);
1738 static struct nfsd4_session *
1739 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1742 struct nfsd4_session *session;
1743 __be32 status = nfserr_badsession;
1745 session = __find_in_sessionid_hashtbl(sessionid, net);
1748 status = nfsd4_get_session_locked(session);
1756 /* caller must hold client_lock */
1758 unhash_session(struct nfsd4_session *ses)
1760 struct nfs4_client *clp = ses->se_client;
1761 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1763 lockdep_assert_held(&nn->client_lock);
1765 list_del(&ses->se_hash);
1766 spin_lock(&ses->se_client->cl_lock);
1767 list_del(&ses->se_perclnt);
1768 spin_unlock(&ses->se_client->cl_lock);
1771 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1773 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1776 * We're assuming the clid was not given out from a boot
1777 * precisely 2^32 (about 136 years) before this one. That seems
1778 * a safe assumption:
1780 if (clid->cl_boot == (u32)nn->boot_time)
1782 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1783 clid->cl_boot, clid->cl_id, nn->boot_time);
1788 * XXX Should we use a slab cache ?
1789 * This type of memory management is somewhat inefficient, but we use it
1790 * anyway since SETCLIENTID is not a common operation.
1792 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1794 struct nfs4_client *clp;
1797 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1800 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1801 if (clp->cl_name.data == NULL)
1803 clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1804 OWNER_HASH_SIZE, GFP_KERNEL);
1805 if (!clp->cl_ownerstr_hashtbl)
1806 goto err_no_hashtbl;
1807 for (i = 0; i < OWNER_HASH_SIZE; i++)
1808 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1809 clp->cl_name.len = name.len;
1810 INIT_LIST_HEAD(&clp->cl_sessions);
1811 idr_init(&clp->cl_stateids);
1812 atomic_set(&clp->cl_refcount, 0);
1813 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1814 INIT_LIST_HEAD(&clp->cl_idhash);
1815 INIT_LIST_HEAD(&clp->cl_openowners);
1816 INIT_LIST_HEAD(&clp->cl_delegations);
1817 INIT_LIST_HEAD(&clp->cl_lru);
1818 INIT_LIST_HEAD(&clp->cl_revoked);
1819 #ifdef CONFIG_NFSD_PNFS
1820 INIT_LIST_HEAD(&clp->cl_lo_states);
1822 spin_lock_init(&clp->cl_lock);
1823 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1826 kfree(clp->cl_name.data);
1833 free_client(struct nfs4_client *clp)
1835 while (!list_empty(&clp->cl_sessions)) {
1836 struct nfsd4_session *ses;
1837 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1839 list_del(&ses->se_perclnt);
1840 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1843 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1844 free_svc_cred(&clp->cl_cred);
1845 kfree(clp->cl_ownerstr_hashtbl);
1846 kfree(clp->cl_name.data);
1847 idr_destroy(&clp->cl_stateids);
1851 /* must be called under the client_lock */
1853 unhash_client_locked(struct nfs4_client *clp)
1855 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1856 struct nfsd4_session *ses;
1858 lockdep_assert_held(&nn->client_lock);
1860 /* Mark the client as expired! */
1862 /* Make it invisible */
1863 if (!list_empty(&clp->cl_idhash)) {
1864 list_del_init(&clp->cl_idhash);
1865 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1866 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1868 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1870 list_del_init(&clp->cl_lru);
1871 spin_lock(&clp->cl_lock);
1872 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1873 list_del_init(&ses->se_hash);
1874 spin_unlock(&clp->cl_lock);
1878 unhash_client(struct nfs4_client *clp)
1880 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1882 spin_lock(&nn->client_lock);
1883 unhash_client_locked(clp);
1884 spin_unlock(&nn->client_lock);
1887 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1889 if (atomic_read(&clp->cl_refcount))
1890 return nfserr_jukebox;
1891 unhash_client_locked(clp);
1896 __destroy_client(struct nfs4_client *clp)
1899 struct nfs4_openowner *oo;
1900 struct nfs4_delegation *dp;
1901 struct list_head reaplist;
1903 INIT_LIST_HEAD(&reaplist);
1904 spin_lock(&state_lock);
1905 while (!list_empty(&clp->cl_delegations)) {
1906 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1907 WARN_ON(!unhash_delegation_locked(dp));
1908 list_add(&dp->dl_recall_lru, &reaplist);
1910 spin_unlock(&state_lock);
1911 while (!list_empty(&reaplist)) {
1912 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1913 list_del_init(&dp->dl_recall_lru);
1914 put_clnt_odstate(dp->dl_clnt_odstate);
1915 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1916 nfs4_put_stid(&dp->dl_stid);
1918 while (!list_empty(&clp->cl_revoked)) {
1919 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1920 list_del_init(&dp->dl_recall_lru);
1921 nfs4_put_stid(&dp->dl_stid);
1923 while (!list_empty(&clp->cl_openowners)) {
1924 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1925 nfs4_get_stateowner(&oo->oo_owner);
1926 release_openowner(oo);
1928 for (i = 0; i < OWNER_HASH_SIZE; i++) {
1929 struct nfs4_stateowner *so, *tmp;
1931 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1933 /* Should be no openowners at this point */
1934 WARN_ON_ONCE(so->so_is_open_owner);
1935 remove_blocked_locks(lockowner(so));
1938 nfsd4_return_all_client_layouts(clp);
1939 nfsd4_shutdown_callback(clp);
1940 if (clp->cl_cb_conn.cb_xprt)
1941 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1946 destroy_client(struct nfs4_client *clp)
1949 __destroy_client(clp);
1952 static void expire_client(struct nfs4_client *clp)
1955 nfsd4_client_record_remove(clp);
1956 __destroy_client(clp);
1959 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1961 memcpy(target->cl_verifier.data, source->data,
1962 sizeof(target->cl_verifier.data));
1965 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1967 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1968 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1971 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1973 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1974 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1976 if ((source->cr_principal && ! target->cr_principal) ||
1977 (source->cr_raw_principal && ! target->cr_raw_principal))
1980 target->cr_flavor = source->cr_flavor;
1981 target->cr_uid = source->cr_uid;
1982 target->cr_gid = source->cr_gid;
1983 target->cr_group_info = source->cr_group_info;
1984 get_group_info(target->cr_group_info);
1985 target->cr_gss_mech = source->cr_gss_mech;
1986 if (source->cr_gss_mech)
1987 gss_mech_get(source->cr_gss_mech);
1992 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1994 if (o1->len < o2->len)
1996 if (o1->len > o2->len)
1998 return memcmp(o1->data, o2->data, o1->len);
2001 static int same_name(const char *n1, const char *n2)
2003 return 0 == memcmp(n1, n2, HEXDIR_LEN);
2007 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2009 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2013 same_clid(clientid_t *cl1, clientid_t *cl2)
2015 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2018 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2022 if (g1->ngroups != g2->ngroups)
2024 for (i=0; i<g1->ngroups; i++)
2025 if (!gid_eq(g1->gid[i], g2->gid[i]))
2031 * RFC 3530 language requires clid_inuse be returned when the
2032 * "principal" associated with a requests differs from that previously
2033 * used. We use uid, gid's, and gss principal string as our best
2034 * approximation. We also don't want to allow non-gss use of a client
2035 * established using gss: in theory cr_principal should catch that
2036 * change, but in practice cr_principal can be null even in the gss case
2037 * since gssd doesn't always pass down a principal string.
2039 static bool is_gss_cred(struct svc_cred *cr)
2041 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2042 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2047 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2049 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2050 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2051 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2052 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2054 if (cr1->cr_principal == cr2->cr_principal)
2056 if (!cr1->cr_principal || !cr2->cr_principal)
2058 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2061 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2063 struct svc_cred *cr = &rqstp->rq_cred;
2066 if (!cr->cr_gss_mech)
2068 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2069 return service == RPC_GSS_SVC_INTEGRITY ||
2070 service == RPC_GSS_SVC_PRIVACY;
2073 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2075 struct svc_cred *cr = &rqstp->rq_cred;
2077 if (!cl->cl_mach_cred)
2079 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2081 if (!svc_rqst_integrity_protected(rqstp))
2083 if (cl->cl_cred.cr_raw_principal)
2084 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2085 cr->cr_raw_principal);
2086 if (!cr->cr_principal)
2088 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2091 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2096 * This is opaque to client, so no need to byte-swap. Use
2097 * __force to keep sparse happy
2099 verf[0] = (__force __be32)get_seconds();
2100 verf[1] = (__force __be32)nn->clverifier_counter++;
2101 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2104 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2106 clp->cl_clientid.cl_boot = nn->boot_time;
2107 clp->cl_clientid.cl_id = nn->clientid_counter++;
2108 gen_confirm(clp, nn);
2111 static struct nfs4_stid *
2112 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2114 struct nfs4_stid *ret;
2116 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2117 if (!ret || !ret->sc_type)
2122 static struct nfs4_stid *
2123 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2125 struct nfs4_stid *s;
2127 spin_lock(&cl->cl_lock);
2128 s = find_stateid_locked(cl, t);
2130 if (typemask & s->sc_type)
2131 refcount_inc(&s->sc_count);
2135 spin_unlock(&cl->cl_lock);
2139 static struct nfs4_client *create_client(struct xdr_netobj name,
2140 struct svc_rqst *rqstp, nfs4_verifier *verf)
2142 struct nfs4_client *clp;
2143 struct sockaddr *sa = svc_addr(rqstp);
2145 struct net *net = SVC_NET(rqstp);
2147 clp = alloc_client(name);
2151 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2156 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2157 clp->cl_time = get_seconds();
2158 clear_bit(0, &clp->cl_cb_slot_busy);
2159 copy_verf(clp, verf);
2160 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2161 clp->cl_cb_session = NULL;
2167 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2169 struct rb_node **new = &(root->rb_node), *parent = NULL;
2170 struct nfs4_client *clp;
2173 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2176 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2177 new = &((*new)->rb_left);
2179 new = &((*new)->rb_right);
2182 rb_link_node(&new_clp->cl_namenode, parent, new);
2183 rb_insert_color(&new_clp->cl_namenode, root);
2186 static struct nfs4_client *
2187 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2190 struct rb_node *node = root->rb_node;
2191 struct nfs4_client *clp;
2194 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2195 cmp = compare_blob(&clp->cl_name, name);
2197 node = node->rb_left;
2199 node = node->rb_right;
2207 add_to_unconfirmed(struct nfs4_client *clp)
2209 unsigned int idhashval;
2210 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2212 lockdep_assert_held(&nn->client_lock);
2214 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2215 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2216 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2217 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2218 renew_client_locked(clp);
2222 move_to_confirmed(struct nfs4_client *clp)
2224 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2225 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2227 lockdep_assert_held(&nn->client_lock);
2229 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2230 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2231 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2232 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2233 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2234 renew_client_locked(clp);
2237 static struct nfs4_client *
2238 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2240 struct nfs4_client *clp;
2241 unsigned int idhashval = clientid_hashval(clid->cl_id);
2243 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2244 if (same_clid(&clp->cl_clientid, clid)) {
2245 if ((bool)clp->cl_minorversion != sessions)
2247 renew_client_locked(clp);
2254 static struct nfs4_client *
2255 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2257 struct list_head *tbl = nn->conf_id_hashtbl;
2259 lockdep_assert_held(&nn->client_lock);
2260 return find_client_in_id_table(tbl, clid, sessions);
2263 static struct nfs4_client *
2264 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2266 struct list_head *tbl = nn->unconf_id_hashtbl;
2268 lockdep_assert_held(&nn->client_lock);
2269 return find_client_in_id_table(tbl, clid, sessions);
2272 static bool clp_used_exchangeid(struct nfs4_client *clp)
2274 return clp->cl_exchange_flags != 0;
2277 static struct nfs4_client *
2278 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2280 lockdep_assert_held(&nn->client_lock);
2281 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2284 static struct nfs4_client *
2285 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2287 lockdep_assert_held(&nn->client_lock);
2288 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2292 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2294 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2295 struct sockaddr *sa = svc_addr(rqstp);
2296 u32 scopeid = rpc_get_scope_id(sa);
2297 unsigned short expected_family;
2299 /* Currently, we only support tcp and tcp6 for the callback channel */
2300 if (se->se_callback_netid_len == 3 &&
2301 !memcmp(se->se_callback_netid_val, "tcp", 3))
2302 expected_family = AF_INET;
2303 else if (se->se_callback_netid_len == 4 &&
2304 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2305 expected_family = AF_INET6;
2309 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2310 se->se_callback_addr_len,
2311 (struct sockaddr *)&conn->cb_addr,
2312 sizeof(conn->cb_addr));
2314 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2317 if (conn->cb_addr.ss_family == AF_INET6)
2318 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2320 conn->cb_prog = se->se_callback_prog;
2321 conn->cb_ident = se->se_callback_ident;
2322 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2325 conn->cb_addr.ss_family = AF_UNSPEC;
2326 conn->cb_addrlen = 0;
2327 dprintk("NFSD: this client (clientid %08x/%08x) "
2328 "will not receive delegations\n",
2329 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2335 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2338 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2340 struct xdr_buf *buf = resp->xdr.buf;
2341 struct nfsd4_slot *slot = resp->cstate.slot;
2344 dprintk("--> %s slot %p\n", __func__, slot);
2346 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2347 slot->sl_opcnt = resp->opcnt;
2348 slot->sl_status = resp->cstate.status;
2349 free_svc_cred(&slot->sl_cred);
2350 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2352 if (!nfsd4_cache_this(resp)) {
2353 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2356 slot->sl_flags |= NFSD4_SLOT_CACHED;
2358 base = resp->cstate.data_offset;
2359 slot->sl_datalen = buf->len - base;
2360 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2361 WARN(1, "%s: sessions DRC could not cache compound\n",
2367 * Encode the replay sequence operation from the slot values.
2368 * If cachethis is FALSE encode the uncached rep error on the next
2369 * operation which sets resp->p and increments resp->opcnt for
2370 * nfs4svc_encode_compoundres.
2374 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2375 struct nfsd4_compoundres *resp)
2377 struct nfsd4_op *op;
2378 struct nfsd4_slot *slot = resp->cstate.slot;
2380 /* Encode the replayed sequence operation */
2381 op = &args->ops[resp->opcnt - 1];
2382 nfsd4_encode_operation(resp, op);
2384 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2386 if (args->opcnt == 1) {
2388 * The original operation wasn't a solo sequence--we
2389 * always cache those--so this retry must not match the
2392 op->status = nfserr_seq_false_retry;
2394 op = &args->ops[resp->opcnt++];
2395 op->status = nfserr_retry_uncached_rep;
2396 nfsd4_encode_operation(resp, op);
2402 * The sequence operation is not cached because we can use the slot and
2406 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2407 struct nfsd4_sequence *seq)
2409 struct nfsd4_slot *slot = resp->cstate.slot;
2410 struct xdr_stream *xdr = &resp->xdr;
2414 dprintk("--> %s slot %p\n", __func__, slot);
2416 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2420 p = xdr_reserve_space(xdr, slot->sl_datalen);
2423 return nfserr_serverfault;
2425 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2426 xdr_commit_encode(xdr);
2428 resp->opcnt = slot->sl_opcnt;
2429 return slot->sl_status;
2433 * Set the exchange_id flags returned by the server.
2436 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2438 #ifdef CONFIG_NFSD_PNFS
2439 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2441 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2444 /* Referrals are supported, Migration is not. */
2445 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2447 /* set the wire flags to return to client. */
2448 clid->flags = new->cl_exchange_flags;
2451 static bool client_has_openowners(struct nfs4_client *clp)
2453 struct nfs4_openowner *oo;
2455 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2456 if (!list_empty(&oo->oo_owner.so_stateids))
2462 static bool client_has_state(struct nfs4_client *clp)
2464 return client_has_openowners(clp)
2465 #ifdef CONFIG_NFSD_PNFS
2466 || !list_empty(&clp->cl_lo_states)
2468 || !list_empty(&clp->cl_delegations)
2469 || !list_empty(&clp->cl_sessions);
2473 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2474 union nfsd4_op_u *u)
2476 struct nfsd4_exchange_id *exid = &u->exchange_id;
2477 struct nfs4_client *conf, *new;
2478 struct nfs4_client *unconf = NULL;
2480 char addr_str[INET6_ADDRSTRLEN];
2481 nfs4_verifier verf = exid->verifier;
2482 struct sockaddr *sa = svc_addr(rqstp);
2483 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2484 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2486 rpc_ntop(sa, addr_str, sizeof(addr_str));
2487 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2488 "ip_addr=%s flags %x, spa_how %d\n",
2489 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2490 addr_str, exid->flags, exid->spa_how);
2492 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2493 return nfserr_inval;
2495 new = create_client(exid->clname, rqstp, &verf);
2497 return nfserr_jukebox;
2499 switch (exid->spa_how) {
2501 exid->spo_must_enforce[0] = 0;
2502 exid->spo_must_enforce[1] = (
2503 1 << (OP_BIND_CONN_TO_SESSION - 32) |
2504 1 << (OP_EXCHANGE_ID - 32) |
2505 1 << (OP_CREATE_SESSION - 32) |
2506 1 << (OP_DESTROY_SESSION - 32) |
2507 1 << (OP_DESTROY_CLIENTID - 32));
2509 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2510 1 << (OP_OPEN_DOWNGRADE) |
2512 1 << (OP_DELEGRETURN));
2514 exid->spo_must_allow[1] &= (
2515 1 << (OP_TEST_STATEID - 32) |
2516 1 << (OP_FREE_STATEID - 32));
2517 if (!svc_rqst_integrity_protected(rqstp)) {
2518 status = nfserr_inval;
2522 * Sometimes userspace doesn't give us a principal.
2523 * Which is a bug, really. Anyway, we can't enforce
2524 * MACH_CRED in that case, better to give up now:
2526 if (!new->cl_cred.cr_principal &&
2527 !new->cl_cred.cr_raw_principal) {
2528 status = nfserr_serverfault;
2531 new->cl_mach_cred = true;
2534 default: /* checked by xdr code */
2537 status = nfserr_encr_alg_unsupp;
2541 /* Cases below refer to rfc 5661 section 18.35.4: */
2542 spin_lock(&nn->client_lock);
2543 conf = find_confirmed_client_by_name(&exid->clname, nn);
2545 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2546 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2549 if (!clp_used_exchangeid(conf)) { /* buggy client */
2550 status = nfserr_inval;
2553 if (!nfsd4_mach_creds_match(conf, rqstp)) {
2554 status = nfserr_wrong_cred;
2557 if (!creds_match) { /* case 9 */
2558 status = nfserr_perm;
2561 if (!verfs_match) { /* case 8 */
2562 status = nfserr_not_same;
2566 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2569 if (!creds_match) { /* case 3 */
2570 if (client_has_state(conf)) {
2571 status = nfserr_clid_inuse;
2576 if (verfs_match) { /* case 2 */
2577 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2580 /* case 5, client reboot */
2585 if (update) { /* case 7 */
2586 status = nfserr_noent;
2590 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2591 if (unconf) /* case 4, possible retry or client restart */
2592 unhash_client_locked(unconf);
2594 /* case 1 (normal case) */
2597 status = mark_client_expired_locked(conf);
2601 new->cl_minorversion = cstate->minorversion;
2602 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2603 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2606 add_to_unconfirmed(new);
2609 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2610 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2612 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2613 nfsd4_set_ex_flags(conf, exid);
2615 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2616 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2620 spin_unlock(&nn->client_lock);
2625 expire_client(unconf);
2630 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2632 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2635 /* The slot is in use, and no response has been sent. */
2637 if (seqid == slot_seqid)
2638 return nfserr_jukebox;
2640 return nfserr_seq_misordered;
2642 /* Note unsigned 32-bit arithmetic handles wraparound: */
2643 if (likely(seqid == slot_seqid + 1))
2645 if (seqid == slot_seqid)
2646 return nfserr_replay_cache;
2647 return nfserr_seq_misordered;
2651 * Cache the create session result into the create session single DRC
2652 * slot cache by saving the xdr structure. sl_seqid has been set.
2653 * Do this for solo or embedded create session operations.
2656 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2657 struct nfsd4_clid_slot *slot, __be32 nfserr)
2659 slot->sl_status = nfserr;
2660 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2664 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2665 struct nfsd4_clid_slot *slot)
2667 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2668 return slot->sl_status;
2671 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2672 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2673 1 + /* MIN tag is length with zero, only length */ \
2674 3 + /* version, opcount, opcode */ \
2675 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2676 /* seqid, slotID, slotID, cache */ \
2677 4 ) * sizeof(__be32))
2679 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2680 2 + /* verifier: AUTH_NULL, length 0 */\
2682 1 + /* MIN tag is length with zero, only length */ \
2683 3 + /* opcount, opcode, opstatus*/ \
2684 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2685 /* seqid, slotID, slotID, slotID, status */ \
2686 5 ) * sizeof(__be32))
2688 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2690 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2692 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2693 return nfserr_toosmall;
2694 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2695 return nfserr_toosmall;
2696 ca->headerpadsz = 0;
2697 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2698 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2699 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2700 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2701 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2702 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2704 * Note decreasing slot size below client's request may make it
2705 * difficult for client to function correctly, whereas
2706 * decreasing the number of slots will (just?) affect
2707 * performance. When short on memory we therefore prefer to
2708 * decrease number of slots instead of their size. Clients that
2709 * request larger slots than they need will get poor results:
2711 ca->maxreqs = nfsd4_get_drc_mem(ca);
2713 return nfserr_jukebox;
2719 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2720 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2722 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2723 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2725 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2726 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2728 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2729 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2730 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2731 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2734 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2736 ca->headerpadsz = 0;
2738 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2739 return nfserr_toosmall;
2740 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2741 return nfserr_toosmall;
2742 ca->maxresp_cached = 0;
2744 return nfserr_toosmall;
2749 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2751 switch (cbs->flavor) {
2757 * GSS case: the spec doesn't allow us to return this
2758 * error. But it also doesn't allow us not to support
2760 * I'd rather this fail hard than return some error the
2761 * client might think it can already handle:
2763 return nfserr_encr_alg_unsupp;
2768 nfsd4_create_session(struct svc_rqst *rqstp,
2769 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2771 struct nfsd4_create_session *cr_ses = &u->create_session;
2772 struct sockaddr *sa = svc_addr(rqstp);
2773 struct nfs4_client *conf, *unconf;
2774 struct nfs4_client *old = NULL;
2775 struct nfsd4_session *new;
2776 struct nfsd4_conn *conn;
2777 struct nfsd4_clid_slot *cs_slot = NULL;
2779 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2781 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2782 return nfserr_inval;
2783 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2786 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2789 status = check_backchannel_attrs(&cr_ses->back_channel);
2791 goto out_release_drc_mem;
2792 status = nfserr_jukebox;
2793 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2795 goto out_release_drc_mem;
2796 conn = alloc_conn_from_crses(rqstp, cr_ses);
2798 goto out_free_session;
2800 spin_lock(&nn->client_lock);
2801 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2802 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2803 WARN_ON_ONCE(conf && unconf);
2806 status = nfserr_wrong_cred;
2807 if (!nfsd4_mach_creds_match(conf, rqstp))
2809 cs_slot = &conf->cl_cs_slot;
2810 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2812 if (status == nfserr_replay_cache)
2813 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2816 } else if (unconf) {
2817 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2818 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2819 status = nfserr_clid_inuse;
2822 status = nfserr_wrong_cred;
2823 if (!nfsd4_mach_creds_match(unconf, rqstp))
2825 cs_slot = &unconf->cl_cs_slot;
2826 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2828 /* an unconfirmed replay returns misordered */
2829 status = nfserr_seq_misordered;
2832 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2834 status = mark_client_expired_locked(old);
2840 move_to_confirmed(unconf);
2843 status = nfserr_stale_clientid;
2847 /* Persistent sessions are not supported */
2848 cr_ses->flags &= ~SESSION4_PERSIST;
2849 /* Upshifting from TCP to RDMA is not supported */
2850 cr_ses->flags &= ~SESSION4_RDMA;
2852 init_session(rqstp, new, conf, cr_ses);
2853 nfsd4_get_session_locked(new);
2855 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2856 NFS4_MAX_SESSIONID_LEN);
2857 cs_slot->sl_seqid++;
2858 cr_ses->seqid = cs_slot->sl_seqid;
2860 /* cache solo and embedded create sessions under the client_lock */
2861 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2862 spin_unlock(&nn->client_lock);
2863 /* init connection and backchannel */
2864 nfsd4_init_conn(rqstp, conn, new);
2865 nfsd4_put_session(new);
2870 spin_unlock(&nn->client_lock);
2875 __free_session(new);
2876 out_release_drc_mem:
2877 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2881 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2884 case NFS4_CDFC4_FORE:
2885 case NFS4_CDFC4_BACK:
2887 case NFS4_CDFC4_FORE_OR_BOTH:
2888 case NFS4_CDFC4_BACK_OR_BOTH:
2889 *dir = NFS4_CDFC4_BOTH;
2892 return nfserr_inval;
2895 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2896 struct nfsd4_compound_state *cstate,
2897 union nfsd4_op_u *u)
2899 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2900 struct nfsd4_session *session = cstate->session;
2901 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2904 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2907 spin_lock(&nn->client_lock);
2908 session->se_cb_prog = bc->bc_cb_program;
2909 session->se_cb_sec = bc->bc_cb_sec;
2910 spin_unlock(&nn->client_lock);
2912 nfsd4_probe_callback(session->se_client);
2917 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2918 struct nfsd4_compound_state *cstate,
2919 union nfsd4_op_u *u)
2921 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2923 struct nfsd4_conn *conn;
2924 struct nfsd4_session *session;
2925 struct net *net = SVC_NET(rqstp);
2926 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2928 if (!nfsd4_last_compound_op(rqstp))
2929 return nfserr_not_only_op;
2930 spin_lock(&nn->client_lock);
2931 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2932 spin_unlock(&nn->client_lock);
2934 goto out_no_session;
2935 status = nfserr_wrong_cred;
2936 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2938 status = nfsd4_map_bcts_dir(&bcts->dir);
2941 conn = alloc_conn(rqstp, bcts->dir);
2942 status = nfserr_jukebox;
2945 nfsd4_init_conn(rqstp, conn, session);
2948 nfsd4_put_session(session);
2953 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2957 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2961 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2962 union nfsd4_op_u *u)
2964 struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2965 struct nfsd4_session *ses;
2967 int ref_held_by_me = 0;
2968 struct net *net = SVC_NET(r);
2969 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2971 status = nfserr_not_only_op;
2972 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2973 if (!nfsd4_last_compound_op(r))
2977 dump_sessionid(__func__, &sessionid->sessionid);
2978 spin_lock(&nn->client_lock);
2979 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2981 goto out_client_lock;
2982 status = nfserr_wrong_cred;
2983 if (!nfsd4_mach_creds_match(ses->se_client, r))
2984 goto out_put_session;
2985 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2987 goto out_put_session;
2988 unhash_session(ses);
2989 spin_unlock(&nn->client_lock);
2991 nfsd4_probe_callback_sync(ses->se_client);
2993 spin_lock(&nn->client_lock);
2996 nfsd4_put_session_locked(ses);
2998 spin_unlock(&nn->client_lock);
3003 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3005 struct nfsd4_conn *c;
3007 list_for_each_entry(c, &s->se_conns, cn_persession) {
3008 if (c->cn_xprt == xpt) {
3015 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3017 struct nfs4_client *clp = ses->se_client;
3018 struct nfsd4_conn *c;
3019 __be32 status = nfs_ok;
3022 spin_lock(&clp->cl_lock);
3023 c = __nfsd4_find_conn(new->cn_xprt, ses);
3026 status = nfserr_conn_not_bound_to_session;
3027 if (clp->cl_mach_cred)
3029 __nfsd4_hash_conn(new, ses);
3030 spin_unlock(&clp->cl_lock);
3031 ret = nfsd4_register_conn(new);
3033 /* oops; xprt is already down: */
3034 nfsd4_conn_lost(&new->cn_xpt_user);
3037 spin_unlock(&clp->cl_lock);
3042 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3044 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3046 return args->opcnt > session->se_fchannel.maxops;
3049 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3050 struct nfsd4_session *session)
3052 struct xdr_buf *xb = &rqstp->rq_arg;
3054 return xb->len > session->se_fchannel.maxreq_sz;
3057 static bool replay_matches_cache(struct svc_rqst *rqstp,
3058 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3060 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3062 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3063 (bool)seq->cachethis)
3066 * If there's an error than the reply can have fewer ops than
3067 * the call. But if we cached a reply with *more* ops than the
3068 * call you're sending us now, then this new call is clearly not
3069 * really a replay of the old one:
3071 if (slot->sl_opcnt < argp->opcnt)
3073 /* This is the only check explicitly called by spec: */
3074 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3077 * There may be more comparisons we could actually do, but the
3078 * spec doesn't require us to catch every case where the calls
3079 * don't match (that would require caching the call as well as
3080 * the reply), so we don't bother.
3086 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3087 union nfsd4_op_u *u)
3089 struct nfsd4_sequence *seq = &u->sequence;
3090 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3091 struct xdr_stream *xdr = &resp->xdr;
3092 struct nfsd4_session *session;
3093 struct nfs4_client *clp;
3094 struct nfsd4_slot *slot;
3095 struct nfsd4_conn *conn;
3098 struct net *net = SVC_NET(rqstp);
3099 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3101 if (resp->opcnt != 1)
3102 return nfserr_sequence_pos;
3105 * Will be either used or freed by nfsd4_sequence_check_conn
3108 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3110 return nfserr_jukebox;
3112 spin_lock(&nn->client_lock);
3113 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3115 goto out_no_session;
3116 clp = session->se_client;
3118 status = nfserr_too_many_ops;
3119 if (nfsd4_session_too_many_ops(rqstp, session))
3120 goto out_put_session;
3122 status = nfserr_req_too_big;
3123 if (nfsd4_request_too_big(rqstp, session))
3124 goto out_put_session;
3126 status = nfserr_badslot;
3127 if (seq->slotid >= session->se_fchannel.maxreqs)
3128 goto out_put_session;
3130 slot = session->se_slots[seq->slotid];
3131 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3133 /* We do not negotiate the number of slots yet, so set the
3134 * maxslots to the session maxreqs which is used to encode
3135 * sr_highest_slotid and the sr_target_slot id to maxslots */
3136 seq->maxslots = session->se_fchannel.maxreqs;
3138 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3139 slot->sl_flags & NFSD4_SLOT_INUSE);
3140 if (status == nfserr_replay_cache) {
3141 status = nfserr_seq_misordered;
3142 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3143 goto out_put_session;
3144 status = nfserr_seq_false_retry;
3145 if (!replay_matches_cache(rqstp, seq, slot))
3146 goto out_put_session;
3147 cstate->slot = slot;
3148 cstate->session = session;
3150 /* Return the cached reply status and set cstate->status
3151 * for nfsd4_proc_compound processing */
3152 status = nfsd4_replay_cache_entry(resp, seq);
3153 cstate->status = nfserr_replay_cache;
3157 goto out_put_session;
3159 status = nfsd4_sequence_check_conn(conn, session);
3162 goto out_put_session;
3164 buflen = (seq->cachethis) ?
3165 session->se_fchannel.maxresp_cached :
3166 session->se_fchannel.maxresp_sz;
3167 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3169 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3170 goto out_put_session;
3171 svc_reserve(rqstp, buflen);
3174 /* Success! bump slot seqid */
3175 slot->sl_seqid = seq->seqid;
3176 slot->sl_flags |= NFSD4_SLOT_INUSE;
3178 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3180 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3182 cstate->slot = slot;
3183 cstate->session = session;
3187 switch (clp->cl_cb_state) {
3189 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3191 case NFSD4_CB_FAULT:
3192 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3195 seq->status_flags = 0;
3197 if (!list_empty(&clp->cl_revoked))
3198 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3202 spin_unlock(&nn->client_lock);
3205 nfsd4_put_session_locked(session);
3206 goto out_no_session;
3210 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3212 struct nfsd4_compound_state *cs = &resp->cstate;
3214 if (nfsd4_has_session(cs)) {
3215 if (cs->status != nfserr_replay_cache) {
3216 nfsd4_store_cache_entry(resp);
3217 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3219 /* Drop session reference that was taken in nfsd4_sequence() */
3220 nfsd4_put_session(cs->session);
3222 put_client_renew(cs->clp);
3226 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3227 struct nfsd4_compound_state *cstate,
3228 union nfsd4_op_u *u)
3230 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3231 struct nfs4_client *conf, *unconf;
3232 struct nfs4_client *clp = NULL;
3234 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3236 spin_lock(&nn->client_lock);
3237 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3238 conf = find_confirmed_client(&dc->clientid, true, nn);
3239 WARN_ON_ONCE(conf && unconf);
3242 if (client_has_state(conf)) {
3243 status = nfserr_clientid_busy;
3246 status = mark_client_expired_locked(conf);
3253 status = nfserr_stale_clientid;
3256 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3258 status = nfserr_wrong_cred;
3261 unhash_client_locked(clp);
3263 spin_unlock(&nn->client_lock);
3270 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3271 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3273 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3276 if (rc->rca_one_fs) {
3277 if (!cstate->current_fh.fh_dentry)
3278 return nfserr_nofilehandle;
3280 * We don't take advantage of the rca_one_fs case.
3281 * That's OK, it's optional, we can safely ignore it.
3286 status = nfserr_complete_already;
3287 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3288 &cstate->session->se_client->cl_flags))
3291 status = nfserr_stale_clientid;
3292 if (is_client_expired(cstate->session->se_client))
3294 * The following error isn't really legal.
3295 * But we only get here if the client just explicitly
3296 * destroyed the client. Surely it no longer cares what
3297 * error it gets back on an operation for the dead
3303 nfsd4_client_record_create(cstate->session->se_client);
3309 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3310 union nfsd4_op_u *u)
3312 struct nfsd4_setclientid *setclid = &u->setclientid;
3313 struct xdr_netobj clname = setclid->se_name;
3314 nfs4_verifier clverifier = setclid->se_verf;
3315 struct nfs4_client *conf, *new;
3316 struct nfs4_client *unconf = NULL;
3318 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3320 new = create_client(clname, rqstp, &clverifier);
3322 return nfserr_jukebox;
3323 /* Cases below refer to rfc 3530 section 14.2.33: */
3324 spin_lock(&nn->client_lock);
3325 conf = find_confirmed_client_by_name(&clname, nn);
3326 if (conf && client_has_state(conf)) {
3328 status = nfserr_clid_inuse;
3329 if (clp_used_exchangeid(conf))
3331 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3332 char addr_str[INET6_ADDRSTRLEN];
3333 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3335 dprintk("NFSD: setclientid: string in use by client "
3336 "at %s\n", addr_str);
3340 unconf = find_unconfirmed_client_by_name(&clname, nn);
3342 unhash_client_locked(unconf);
3343 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3344 /* case 1: probable callback update */
3345 copy_clid(new, conf);
3346 gen_confirm(new, nn);
3347 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3349 new->cl_minorversion = 0;
3350 gen_callback(new, setclid, rqstp);
3351 add_to_unconfirmed(new);
3352 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3353 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3354 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3358 spin_unlock(&nn->client_lock);
3362 expire_client(unconf);
3368 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3369 struct nfsd4_compound_state *cstate,
3370 union nfsd4_op_u *u)
3372 struct nfsd4_setclientid_confirm *setclientid_confirm =
3373 &u->setclientid_confirm;
3374 struct nfs4_client *conf, *unconf;
3375 struct nfs4_client *old = NULL;
3376 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3377 clientid_t * clid = &setclientid_confirm->sc_clientid;
3379 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3381 if (STALE_CLIENTID(clid, nn))
3382 return nfserr_stale_clientid;
3384 spin_lock(&nn->client_lock);
3385 conf = find_confirmed_client(clid, false, nn);
3386 unconf = find_unconfirmed_client(clid, false, nn);
3388 * We try hard to give out unique clientid's, so if we get an
3389 * attempt to confirm the same clientid with a different cred,
3390 * the client may be buggy; this should never happen.
3392 * Nevertheless, RFC 7530 recommends INUSE for this case:
3394 status = nfserr_clid_inuse;
3395 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3397 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3399 /* cases below refer to rfc 3530 section 14.2.34: */
3400 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3401 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3402 /* case 2: probable retransmit */
3404 } else /* case 4: client hasn't noticed we rebooted yet? */
3405 status = nfserr_stale_clientid;
3409 if (conf) { /* case 1: callback update */
3411 unhash_client_locked(old);
3412 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3413 } else { /* case 3: normal case; new or rebooted client */
3414 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3416 status = nfserr_clid_inuse;
3417 if (client_has_state(old)
3418 && !same_creds(&unconf->cl_cred,
3421 status = mark_client_expired_locked(old);
3427 move_to_confirmed(unconf);
3430 get_client_locked(conf);
3431 spin_unlock(&nn->client_lock);
3432 nfsd4_probe_callback(conf);
3433 spin_lock(&nn->client_lock);
3434 put_client_renew_locked(conf);
3436 spin_unlock(&nn->client_lock);
3442 static struct nfs4_file *nfsd4_alloc_file(void)
3444 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3447 /* OPEN Share state helper functions */
3448 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3449 struct nfs4_file *fp)
3451 lockdep_assert_held(&state_lock);
3453 refcount_set(&fp->fi_ref, 1);
3454 spin_lock_init(&fp->fi_lock);
3455 INIT_LIST_HEAD(&fp->fi_stateids);
3456 INIT_LIST_HEAD(&fp->fi_delegations);
3457 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3458 fh_copy_shallow(&fp->fi_fhandle, fh);
3459 fp->fi_deleg_file = NULL;
3460 fp->fi_had_conflict = false;
3461 fp->fi_share_deny = 0;
3462 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3463 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3464 #ifdef CONFIG_NFSD_PNFS
3465 INIT_LIST_HEAD(&fp->fi_lo_states);
3466 atomic_set(&fp->fi_lo_recalls, 0);
3468 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3472 nfsd4_free_slabs(void)
3474 kmem_cache_destroy(odstate_slab);
3475 kmem_cache_destroy(openowner_slab);
3476 kmem_cache_destroy(lockowner_slab);
3477 kmem_cache_destroy(file_slab);
3478 kmem_cache_destroy(stateid_slab);
3479 kmem_cache_destroy(deleg_slab);
3483 nfsd4_init_slabs(void)
3485 openowner_slab = kmem_cache_create("nfsd4_openowners",
3486 sizeof(struct nfs4_openowner), 0, 0, NULL);
3487 if (openowner_slab == NULL)
3489 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3490 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3491 if (lockowner_slab == NULL)
3492 goto out_free_openowner_slab;
3493 file_slab = kmem_cache_create("nfsd4_files",
3494 sizeof(struct nfs4_file), 0, 0, NULL);
3495 if (file_slab == NULL)
3496 goto out_free_lockowner_slab;
3497 stateid_slab = kmem_cache_create("nfsd4_stateids",
3498 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3499 if (stateid_slab == NULL)
3500 goto out_free_file_slab;
3501 deleg_slab = kmem_cache_create("nfsd4_delegations",
3502 sizeof(struct nfs4_delegation), 0, 0, NULL);
3503 if (deleg_slab == NULL)
3504 goto out_free_stateid_slab;
3505 odstate_slab = kmem_cache_create("nfsd4_odstate",
3506 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3507 if (odstate_slab == NULL)
3508 goto out_free_deleg_slab;
3511 out_free_deleg_slab:
3512 kmem_cache_destroy(deleg_slab);
3513 out_free_stateid_slab:
3514 kmem_cache_destroy(stateid_slab);
3516 kmem_cache_destroy(file_slab);
3517 out_free_lockowner_slab:
3518 kmem_cache_destroy(lockowner_slab);
3519 out_free_openowner_slab:
3520 kmem_cache_destroy(openowner_slab);
3522 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3526 static void init_nfs4_replay(struct nfs4_replay *rp)
3528 rp->rp_status = nfserr_serverfault;
3530 rp->rp_buf = rp->rp_ibuf;
3531 mutex_init(&rp->rp_mutex);
3534 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3535 struct nfs4_stateowner *so)
3537 if (!nfsd4_has_session(cstate)) {
3538 mutex_lock(&so->so_replay.rp_mutex);
3539 cstate->replay_owner = nfs4_get_stateowner(so);
3543 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3545 struct nfs4_stateowner *so = cstate->replay_owner;
3548 cstate->replay_owner = NULL;
3549 mutex_unlock(&so->so_replay.rp_mutex);
3550 nfs4_put_stateowner(so);
3554 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3556 struct nfs4_stateowner *sop;
3558 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3562 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3563 if (!sop->so_owner.data) {
3564 kmem_cache_free(slab, sop);
3567 sop->so_owner.len = owner->len;
3569 INIT_LIST_HEAD(&sop->so_stateids);
3570 sop->so_client = clp;
3571 init_nfs4_replay(&sop->so_replay);
3572 atomic_set(&sop->so_count, 1);
3576 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3578 lockdep_assert_held(&clp->cl_lock);
3580 list_add(&oo->oo_owner.so_strhash,
3581 &clp->cl_ownerstr_hashtbl[strhashval]);
3582 list_add(&oo->oo_perclient, &clp->cl_openowners);
3585 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3587 unhash_openowner_locked(openowner(so));
3590 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3592 struct nfs4_openowner *oo = openowner(so);
3594 kmem_cache_free(openowner_slab, oo);
3597 static const struct nfs4_stateowner_operations openowner_ops = {
3598 .so_unhash = nfs4_unhash_openowner,
3599 .so_free = nfs4_free_openowner,
3602 static struct nfs4_ol_stateid *
3603 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3605 struct nfs4_ol_stateid *local, *ret = NULL;
3606 struct nfs4_openowner *oo = open->op_openowner;
3608 lockdep_assert_held(&fp->fi_lock);
3610 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3611 /* ignore lock owners */
3612 if (local->st_stateowner->so_is_open_owner == 0)
3614 if (local->st_stateowner != &oo->oo_owner)
3616 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3618 refcount_inc(&ret->st_stid.sc_count);
3626 nfsd4_verify_open_stid(struct nfs4_stid *s)
3628 __be32 ret = nfs_ok;
3630 switch (s->sc_type) {
3634 case NFS4_CLOSED_STID:
3635 case NFS4_CLOSED_DELEG_STID:
3636 ret = nfserr_bad_stateid;
3638 case NFS4_REVOKED_DELEG_STID:
3639 ret = nfserr_deleg_revoked;
3644 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3646 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3650 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3651 ret = nfsd4_verify_open_stid(&stp->st_stid);
3653 mutex_unlock(&stp->st_mutex);
3657 static struct nfs4_ol_stateid *
3658 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3660 struct nfs4_ol_stateid *stp;
3662 spin_lock(&fp->fi_lock);
3663 stp = nfsd4_find_existing_open(fp, open);
3664 spin_unlock(&fp->fi_lock);
3665 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3667 nfs4_put_stid(&stp->st_stid);
3672 static struct nfs4_openowner *
3673 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3674 struct nfsd4_compound_state *cstate)
3676 struct nfs4_client *clp = cstate->clp;
3677 struct nfs4_openowner *oo, *ret;
3679 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3682 oo->oo_owner.so_ops = &openowner_ops;
3683 oo->oo_owner.so_is_open_owner = 1;
3684 oo->oo_owner.so_seqid = open->op_seqid;
3686 if (nfsd4_has_session(cstate))
3687 oo->oo_flags |= NFS4_OO_CONFIRMED;
3689 oo->oo_last_closed_stid = NULL;
3690 INIT_LIST_HEAD(&oo->oo_close_lru);
3691 spin_lock(&clp->cl_lock);
3692 ret = find_openstateowner_str_locked(strhashval, open, clp);
3694 hash_openowner(oo, clp, strhashval);
3697 nfs4_free_stateowner(&oo->oo_owner);
3699 spin_unlock(&clp->cl_lock);
3703 static struct nfs4_ol_stateid *
3704 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3707 struct nfs4_openowner *oo = open->op_openowner;
3708 struct nfs4_ol_stateid *retstp = NULL;
3709 struct nfs4_ol_stateid *stp;
3712 /* We are moving these outside of the spinlocks to avoid the warnings */
3713 mutex_init(&stp->st_mutex);
3714 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3717 spin_lock(&oo->oo_owner.so_client->cl_lock);
3718 spin_lock(&fp->fi_lock);
3720 retstp = nfsd4_find_existing_open(fp, open);
3724 open->op_stp = NULL;
3725 refcount_inc(&stp->st_stid.sc_count);
3726 stp->st_stid.sc_type = NFS4_OPEN_STID;
3727 INIT_LIST_HEAD(&stp->st_locks);
3728 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3730 stp->st_stid.sc_file = fp;
3731 stp->st_access_bmap = 0;
3732 stp->st_deny_bmap = 0;
3733 stp->st_openstp = NULL;
3734 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3735 list_add(&stp->st_perfile, &fp->fi_stateids);
3738 spin_unlock(&fp->fi_lock);
3739 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3741 /* Handle races with CLOSE */
3742 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3743 nfs4_put_stid(&retstp->st_stid);
3746 /* To keep mutex tracking happy */
3747 mutex_unlock(&stp->st_mutex);
3754 * In the 4.0 case we need to keep the owners around a little while to handle
3755 * CLOSE replay. We still do need to release any file access that is held by
3756 * them before returning however.
3759 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3761 struct nfs4_ol_stateid *last;
3762 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3763 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3766 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3769 * We know that we hold one reference via nfsd4_close, and another
3770 * "persistent" reference for the client. If the refcount is higher
3771 * than 2, then there are still calls in progress that are using this
3772 * stateid. We can't put the sc_file reference until they are finished.
3773 * Wait for the refcount to drop to 2. Since it has been unhashed,
3774 * there should be no danger of the refcount going back up again at
3777 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3779 release_all_access(s);
3780 if (s->st_stid.sc_file) {
3781 put_nfs4_file(s->st_stid.sc_file);
3782 s->st_stid.sc_file = NULL;
3785 spin_lock(&nn->client_lock);
3786 last = oo->oo_last_closed_stid;
3787 oo->oo_last_closed_stid = s;
3788 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3789 oo->oo_time = get_seconds();
3790 spin_unlock(&nn->client_lock);
3792 nfs4_put_stid(&last->st_stid);
3795 /* search file_hashtbl[] for file */
3796 static struct nfs4_file *
3797 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3799 struct nfs4_file *fp;
3801 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3802 if (fh_match(&fp->fi_fhandle, fh)) {
3803 if (refcount_inc_not_zero(&fp->fi_ref))
3811 find_file(struct knfsd_fh *fh)
3813 struct nfs4_file *fp;
3814 unsigned int hashval = file_hashval(fh);
3817 fp = find_file_locked(fh, hashval);
3822 static struct nfs4_file *
3823 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3825 struct nfs4_file *fp;
3826 unsigned int hashval = file_hashval(fh);
3829 fp = find_file_locked(fh, hashval);
3834 spin_lock(&state_lock);
3835 fp = find_file_locked(fh, hashval);
3836 if (likely(fp == NULL)) {
3837 nfsd4_init_file(fh, hashval, new);
3840 spin_unlock(&state_lock);
3846 * Called to check deny when READ with all zero stateid or
3847 * WRITE with all zero or all one stateid
3850 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3852 struct nfs4_file *fp;
3853 __be32 ret = nfs_ok;
3855 fp = find_file(¤t_fh->fh_handle);
3858 /* Check for conflicting share reservations */
3859 spin_lock(&fp->fi_lock);
3860 if (fp->fi_share_deny & deny_type)
3861 ret = nfserr_locked;
3862 spin_unlock(&fp->fi_lock);
3867 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3869 struct nfs4_delegation *dp = cb_to_delegation(cb);
3870 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3873 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3876 * We can't do this in nfsd_break_deleg_cb because it is
3877 * already holding inode->i_lock.
3879 * If the dl_time != 0, then we know that it has already been
3880 * queued for a lease break. Don't queue it again.
3882 spin_lock(&state_lock);
3883 if (dp->dl_time == 0) {
3884 dp->dl_time = get_seconds();
3885 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3887 spin_unlock(&state_lock);
3890 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3891 struct rpc_task *task)
3893 struct nfs4_delegation *dp = cb_to_delegation(cb);
3895 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3898 switch (task->tk_status) {
3902 case -NFS4ERR_BAD_STATEID:
3904 * Race: client probably got cb_recall before open reply
3905 * granting delegation.
3907 if (dp->dl_retries--) {
3908 rpc_delay(task, 2 * HZ);
3917 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3919 struct nfs4_delegation *dp = cb_to_delegation(cb);
3921 nfs4_put_stid(&dp->dl_stid);
3924 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3925 .prepare = nfsd4_cb_recall_prepare,
3926 .done = nfsd4_cb_recall_done,
3927 .release = nfsd4_cb_recall_release,
3930 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3933 * We're assuming the state code never drops its reference
3934 * without first removing the lease. Since we're in this lease
3935 * callback (and since the lease code is serialized by the kernel
3936 * lock) we know the server hasn't removed the lease yet, we know
3937 * it's safe to take a reference.
3939 refcount_inc(&dp->dl_stid.sc_count);
3940 nfsd4_run_cb(&dp->dl_recall);
3943 /* Called from break_lease() with i_lock held. */
3945 nfsd_break_deleg_cb(struct file_lock *fl)
3948 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3949 struct nfs4_delegation *dp;
3952 WARN(1, "(%p)->fl_owner NULL\n", fl);
3955 if (fp->fi_had_conflict) {
3956 WARN(1, "duplicate break on %p\n", fp);
3960 * We don't want the locks code to timeout the lease for us;
3961 * we'll remove it ourself if a delegation isn't returned
3964 fl->fl_break_time = 0;
3966 spin_lock(&fp->fi_lock);
3967 fp->fi_had_conflict = true;
3969 * If there are no delegations on the list, then return true
3970 * so that the lease code will go ahead and delete it.
3972 if (list_empty(&fp->fi_delegations))
3975 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3976 nfsd_break_one_deleg(dp);
3977 spin_unlock(&fp->fi_lock);
3982 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3983 struct list_head *dispose)
3986 return lease_modify(onlist, arg, dispose);
3991 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3992 .lm_break = nfsd_break_deleg_cb,
3993 .lm_change = nfsd_change_deleg_cb,
3996 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3998 if (nfsd4_has_session(cstate))
4000 if (seqid == so->so_seqid - 1)
4001 return nfserr_replay_me;
4002 if (seqid == so->so_seqid)
4004 return nfserr_bad_seqid;
4007 static __be32 lookup_clientid(clientid_t *clid,
4008 struct nfsd4_compound_state *cstate,
4009 struct nfsd_net *nn)
4011 struct nfs4_client *found;
4014 found = cstate->clp;
4015 if (!same_clid(&found->cl_clientid, clid))
4016 return nfserr_stale_clientid;
4020 if (STALE_CLIENTID(clid, nn))
4021 return nfserr_stale_clientid;
4024 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4025 * cached already then we know this is for is for v4.0 and "sessions"
4028 WARN_ON_ONCE(cstate->session);
4029 spin_lock(&nn->client_lock);
4030 found = find_confirmed_client(clid, false, nn);
4032 spin_unlock(&nn->client_lock);
4033 return nfserr_expired;
4035 atomic_inc(&found->cl_refcount);
4036 spin_unlock(&nn->client_lock);
4038 /* Cache the nfs4_client in cstate! */
4039 cstate->clp = found;
4044 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4045 struct nfsd4_open *open, struct nfsd_net *nn)
4047 clientid_t *clientid = &open->op_clientid;
4048 struct nfs4_client *clp = NULL;
4049 unsigned int strhashval;
4050 struct nfs4_openowner *oo = NULL;
4053 if (STALE_CLIENTID(&open->op_clientid, nn))
4054 return nfserr_stale_clientid;
4056 * In case we need it later, after we've already created the
4057 * file and don't want to risk a further failure:
4059 open->op_file = nfsd4_alloc_file();
4060 if (open->op_file == NULL)
4061 return nfserr_jukebox;
4063 status = lookup_clientid(clientid, cstate, nn);
4068 strhashval = ownerstr_hashval(&open->op_owner);
4069 oo = find_openstateowner_str(strhashval, open, clp);
4070 open->op_openowner = oo;
4074 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4075 /* Replace unconfirmed owners without checking for replay. */
4076 release_openowner(oo);
4077 open->op_openowner = NULL;
4080 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4085 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4087 return nfserr_jukebox;
4088 open->op_openowner = oo;
4090 open->op_stp = nfs4_alloc_open_stateid(clp);
4092 return nfserr_jukebox;
4094 if (nfsd4_has_session(cstate) &&
4095 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4096 open->op_odstate = alloc_clnt_odstate(clp);
4097 if (!open->op_odstate)
4098 return nfserr_jukebox;
4104 static inline __be32
4105 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4107 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4108 return nfserr_openmode;
4113 static int share_access_to_flags(u32 share_access)
4115 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4118 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4120 struct nfs4_stid *ret;
4122 ret = find_stateid_by_type(cl, s,
4123 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4126 return delegstateid(ret);
4129 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4131 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4132 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4136 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4137 struct nfs4_delegation **dp)
4140 __be32 status = nfserr_bad_stateid;
4141 struct nfs4_delegation *deleg;
4143 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4146 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4147 nfs4_put_stid(&deleg->dl_stid);
4148 if (cl->cl_minorversion)
4149 status = nfserr_deleg_revoked;
4152 flags = share_access_to_flags(open->op_share_access);
4153 status = nfs4_check_delegmode(deleg, flags);
4155 nfs4_put_stid(&deleg->dl_stid);
4160 if (!nfsd4_is_deleg_cur(open))
4164 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4168 static inline int nfs4_access_to_access(u32 nfs4_access)
4172 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4173 flags |= NFSD_MAY_READ;
4174 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4175 flags |= NFSD_MAY_WRITE;
4179 static inline __be32
4180 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4181 struct nfsd4_open *open)
4183 struct iattr iattr = {
4184 .ia_valid = ATTR_SIZE,
4187 if (!open->op_truncate)
4189 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4190 return nfserr_inval;
4191 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4194 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4195 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4196 struct nfsd4_open *open)
4198 struct file *filp = NULL;
4200 int oflag = nfs4_access_to_omode(open->op_share_access);
4201 int access = nfs4_access_to_access(open->op_share_access);
4202 unsigned char old_access_bmap, old_deny_bmap;
4204 spin_lock(&fp->fi_lock);
4207 * Are we trying to set a deny mode that would conflict with
4210 status = nfs4_file_check_deny(fp, open->op_share_deny);
4211 if (status != nfs_ok) {
4212 spin_unlock(&fp->fi_lock);
4216 /* set access to the file */
4217 status = nfs4_file_get_access(fp, open->op_share_access);
4218 if (status != nfs_ok) {
4219 spin_unlock(&fp->fi_lock);
4223 /* Set access bits in stateid */
4224 old_access_bmap = stp->st_access_bmap;
4225 set_access(open->op_share_access, stp);
4227 /* Set new deny mask */
4228 old_deny_bmap = stp->st_deny_bmap;
4229 set_deny(open->op_share_deny, stp);
4230 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4232 if (!fp->fi_fds[oflag]) {
4233 spin_unlock(&fp->fi_lock);
4234 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4236 goto out_put_access;
4237 spin_lock(&fp->fi_lock);
4238 if (!fp->fi_fds[oflag]) {
4239 fp->fi_fds[oflag] = filp;
4243 spin_unlock(&fp->fi_lock);
4247 status = nfsd4_truncate(rqstp, cur_fh, open);
4249 goto out_put_access;
4253 stp->st_access_bmap = old_access_bmap;
4254 nfs4_file_put_access(fp, open->op_share_access);
4255 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4260 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4263 unsigned char old_deny_bmap = stp->st_deny_bmap;
4265 if (!test_access(open->op_share_access, stp))
4266 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4268 /* test and set deny mode */
4269 spin_lock(&fp->fi_lock);
4270 status = nfs4_file_check_deny(fp, open->op_share_deny);
4271 if (status == nfs_ok) {
4272 set_deny(open->op_share_deny, stp);
4273 fp->fi_share_deny |=
4274 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4276 spin_unlock(&fp->fi_lock);
4278 if (status != nfs_ok)
4281 status = nfsd4_truncate(rqstp, cur_fh, open);
4282 if (status != nfs_ok)
4283 reset_union_bmap_deny(old_deny_bmap, stp);
4287 /* Should we give out recallable state?: */
4288 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4290 if (clp->cl_cb_state == NFSD4_CB_UP)
4293 * In the sessions case, since we don't have to establish a
4294 * separate connection for callbacks, we assume it's OK
4295 * until we hear otherwise:
4297 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4300 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4302 struct file_lock *fl;
4304 fl = locks_alloc_lock();
4307 fl->fl_lmops = &nfsd_lease_mng_ops;
4308 fl->fl_flags = FL_DELEG;
4309 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4310 fl->fl_end = OFFSET_MAX;
4311 fl->fl_owner = (fl_owner_t)fp;
4312 fl->fl_pid = current->tgid;
4317 * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4318 * @dp: a pointer to the nfs4_delegation we're adding.
4321 * On success: Return code will be 0 on success.
4323 * On error: -EAGAIN if there was an existing delegation.
4324 * nonzero if there is an error in other cases.
4328 static int nfs4_setlease(struct nfs4_delegation *dp)
4330 struct nfs4_file *fp = dp->dl_stid.sc_file;
4331 struct file_lock *fl;
4335 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4338 filp = find_readable_file(fp);
4340 /* We should always have a readable file here */
4342 locks_free_lock(fl);
4346 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4348 locks_free_lock(fl);
4351 spin_lock(&state_lock);
4352 spin_lock(&fp->fi_lock);
4353 /* Did the lease get broken before we took the lock? */
4355 if (fp->fi_had_conflict)
4358 if (fp->fi_deleg_file) {
4359 status = hash_delegation_locked(dp, fp);
4362 fp->fi_deleg_file = filp;
4363 fp->fi_delegees = 0;
4364 status = hash_delegation_locked(dp, fp);
4365 spin_unlock(&fp->fi_lock);
4366 spin_unlock(&state_lock);
4368 /* Should never happen, this is a new fi_deleg_file */
4374 spin_unlock(&fp->fi_lock);
4375 spin_unlock(&state_lock);
4381 static struct nfs4_delegation *
4382 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4383 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4386 struct nfs4_delegation *dp;
4388 if (fp->fi_had_conflict)
4389 return ERR_PTR(-EAGAIN);
4391 spin_lock(&state_lock);
4392 spin_lock(&fp->fi_lock);
4393 status = nfs4_get_existing_delegation(clp, fp);
4394 spin_unlock(&fp->fi_lock);
4395 spin_unlock(&state_lock);
4398 return ERR_PTR(status);
4400 dp = alloc_init_deleg(clp, fh, odstate);
4402 return ERR_PTR(-ENOMEM);
4405 spin_lock(&state_lock);
4406 spin_lock(&fp->fi_lock);
4407 dp->dl_stid.sc_file = fp;
4408 if (!fp->fi_deleg_file) {
4409 spin_unlock(&fp->fi_lock);
4410 spin_unlock(&state_lock);
4411 status = nfs4_setlease(dp);
4414 if (fp->fi_had_conflict) {
4418 status = hash_delegation_locked(dp, fp);
4420 spin_unlock(&fp->fi_lock);
4421 spin_unlock(&state_lock);
4424 put_clnt_odstate(dp->dl_clnt_odstate);
4425 nfs4_put_stid(&dp->dl_stid);
4426 return ERR_PTR(status);
4431 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4433 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4434 if (status == -EAGAIN)
4435 open->op_why_no_deleg = WND4_CONTENTION;
4437 open->op_why_no_deleg = WND4_RESOURCE;
4438 switch (open->op_deleg_want) {
4439 case NFS4_SHARE_WANT_READ_DELEG:
4440 case NFS4_SHARE_WANT_WRITE_DELEG:
4441 case NFS4_SHARE_WANT_ANY_DELEG:
4443 case NFS4_SHARE_WANT_CANCEL:
4444 open->op_why_no_deleg = WND4_CANCELLED;
4446 case NFS4_SHARE_WANT_NO_DELEG:
4453 * Attempt to hand out a delegation.
4455 * Note we don't support write delegations, and won't until the vfs has
4456 * proper support for them.
4459 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4460 struct nfs4_ol_stateid *stp)
4462 struct nfs4_delegation *dp;
4463 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4464 struct nfs4_client *clp = stp->st_stid.sc_client;
4468 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4469 open->op_recall = 0;
4470 switch (open->op_claim_type) {
4471 case NFS4_OPEN_CLAIM_PREVIOUS:
4473 open->op_recall = 1;
4474 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4477 case NFS4_OPEN_CLAIM_NULL:
4478 case NFS4_OPEN_CLAIM_FH:
4480 * Let's not give out any delegations till everyone's
4481 * had the chance to reclaim theirs, *and* until
4482 * NLM locks have all been reclaimed:
4484 if (locks_in_grace(clp->net))
4486 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4489 * Also, if the file was opened for write or
4490 * create, there's a good chance the client's
4491 * about to write to it, resulting in an
4492 * immediate recall (since we don't support
4493 * write delegations):
4495 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4497 if (open->op_create == NFS4_OPEN_CREATE)
4503 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4507 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4509 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4510 STATEID_VAL(&dp->dl_stid.sc_stateid));
4511 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4512 nfs4_put_stid(&dp->dl_stid);
4515 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4516 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4517 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4518 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4519 open->op_recall = 1;
4522 /* 4.1 client asking for a delegation? */
4523 if (open->op_deleg_want)
4524 nfsd4_open_deleg_none_ext(open, status);
4528 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4529 struct nfs4_delegation *dp)
4531 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4532 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4533 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4534 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4535 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4536 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4537 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4538 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4540 /* Otherwise the client must be confused wanting a delegation
4541 * it already has, therefore we don't return
4542 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4547 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4549 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4550 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4551 struct nfs4_file *fp = NULL;
4552 struct nfs4_ol_stateid *stp = NULL;
4553 struct nfs4_delegation *dp = NULL;
4555 bool new_stp = false;
4558 * Lookup file; if found, lookup stateid and check open request,
4559 * and check for delegations in the process of being recalled.
4560 * If not found, create the nfs4_file struct
4562 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4563 if (fp != open->op_file) {
4564 status = nfs4_check_deleg(cl, open, &dp);
4567 stp = nfsd4_find_and_lock_existing_open(fp, open);
4569 open->op_file = NULL;
4570 status = nfserr_bad_stateid;
4571 if (nfsd4_is_deleg_cur(open))
4576 stp = init_open_stateid(fp, open);
4582 * OPEN the file, or upgrade an existing OPEN.
4583 * If truncate fails, the OPEN fails.
4585 * stp is already locked.
4588 /* Stateid was found, this is an OPEN upgrade */
4589 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4591 mutex_unlock(&stp->st_mutex);
4595 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4597 stp->st_stid.sc_type = NFS4_CLOSED_STID;
4598 release_open_stateid(stp);
4599 mutex_unlock(&stp->st_mutex);
4603 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4605 if (stp->st_clnt_odstate == open->op_odstate)
4606 open->op_odstate = NULL;
4609 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4610 mutex_unlock(&stp->st_mutex);
4612 if (nfsd4_has_session(&resp->cstate)) {
4613 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4614 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4615 open->op_why_no_deleg = WND4_NOT_WANTED;
4621 * Attempt to hand out a delegation. No error return, because the
4622 * OPEN succeeds even if we fail.
4624 nfs4_open_delegation(current_fh, open, stp);
4628 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4629 STATEID_VAL(&stp->st_stid.sc_stateid));
4631 /* 4.1 client trying to upgrade/downgrade delegation? */
4632 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4633 open->op_deleg_want)
4634 nfsd4_deleg_xgrade_none_ext(open, dp);
4638 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4639 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4641 * To finish the open response, we just need to set the rflags.
4643 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4644 if (nfsd4_has_session(&resp->cstate))
4645 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4646 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4647 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4650 nfs4_put_stid(&dp->dl_stid);
4652 nfs4_put_stid(&stp->st_stid);
4657 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4658 struct nfsd4_open *open)
4660 if (open->op_openowner) {
4661 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4663 nfsd4_cstate_assign_replay(cstate, so);
4664 nfs4_put_stateowner(so);
4667 kmem_cache_free(file_slab, open->op_file);
4669 nfs4_put_stid(&open->op_stp->st_stid);
4670 if (open->op_odstate)
4671 kmem_cache_free(odstate_slab, open->op_odstate);
4675 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4676 union nfsd4_op_u *u)
4678 clientid_t *clid = &u->renew;
4679 struct nfs4_client *clp;
4681 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4683 dprintk("process_renew(%08x/%08x): starting\n",
4684 clid->cl_boot, clid->cl_id);
4685 status = lookup_clientid(clid, cstate, nn);
4689 status = nfserr_cb_path_down;
4690 if (!list_empty(&clp->cl_delegations)
4691 && clp->cl_cb_state != NFSD4_CB_UP)
4699 nfsd4_end_grace(struct nfsd_net *nn)
4701 /* do nothing if grace period already ended */
4702 if (nn->grace_ended)
4705 dprintk("NFSD: end of grace period\n");
4706 nn->grace_ended = true;
4708 * If the server goes down again right now, an NFSv4
4709 * client will still be allowed to reclaim after it comes back up,
4710 * even if it hasn't yet had a chance to reclaim state this time.
4713 nfsd4_record_grace_done(nn);
4715 * At this point, NFSv4 clients can still reclaim. But if the
4716 * server crashes, any that have not yet reclaimed will be out
4717 * of luck on the next boot.
4719 * (NFSv4.1+ clients are considered to have reclaimed once they
4720 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4721 * have reclaimed after their first OPEN.)
4723 locks_end_grace(&nn->nfsd4_manager);
4725 * At this point, and once lockd and/or any other containers
4726 * exit their grace period, further reclaims will fail and
4727 * regular locking can resume.
4732 nfs4_laundromat(struct nfsd_net *nn)
4734 struct nfs4_client *clp;
4735 struct nfs4_openowner *oo;
4736 struct nfs4_delegation *dp;
4737 struct nfs4_ol_stateid *stp;
4738 struct nfsd4_blocked_lock *nbl;
4739 struct list_head *pos, *next, reaplist;
4740 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4741 time_t t, new_timeo = nn->nfsd4_lease;
4743 dprintk("NFSD: laundromat service - starting\n");
4744 nfsd4_end_grace(nn);
4745 INIT_LIST_HEAD(&reaplist);
4746 spin_lock(&nn->client_lock);
4747 list_for_each_safe(pos, next, &nn->client_lru) {
4748 clp = list_entry(pos, struct nfs4_client, cl_lru);
4749 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4750 t = clp->cl_time - cutoff;
4751 new_timeo = min(new_timeo, t);
4754 if (mark_client_expired_locked(clp)) {
4755 dprintk("NFSD: client in use (clientid %08x)\n",
4756 clp->cl_clientid.cl_id);
4759 list_add(&clp->cl_lru, &reaplist);
4761 spin_unlock(&nn->client_lock);
4762 list_for_each_safe(pos, next, &reaplist) {
4763 clp = list_entry(pos, struct nfs4_client, cl_lru);
4764 dprintk("NFSD: purging unused client (clientid %08x)\n",
4765 clp->cl_clientid.cl_id);
4766 list_del_init(&clp->cl_lru);
4769 spin_lock(&state_lock);
4770 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4771 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4772 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4773 t = dp->dl_time - cutoff;
4774 new_timeo = min(new_timeo, t);
4777 WARN_ON(!unhash_delegation_locked(dp));
4778 list_add(&dp->dl_recall_lru, &reaplist);
4780 spin_unlock(&state_lock);
4781 while (!list_empty(&reaplist)) {
4782 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4784 list_del_init(&dp->dl_recall_lru);
4785 revoke_delegation(dp);
4788 spin_lock(&nn->client_lock);
4789 while (!list_empty(&nn->close_lru)) {
4790 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4792 if (time_after((unsigned long)oo->oo_time,
4793 (unsigned long)cutoff)) {
4794 t = oo->oo_time - cutoff;
4795 new_timeo = min(new_timeo, t);
4798 list_del_init(&oo->oo_close_lru);
4799 stp = oo->oo_last_closed_stid;
4800 oo->oo_last_closed_stid = NULL;
4801 spin_unlock(&nn->client_lock);
4802 nfs4_put_stid(&stp->st_stid);
4803 spin_lock(&nn->client_lock);
4805 spin_unlock(&nn->client_lock);
4808 * It's possible for a client to try and acquire an already held lock
4809 * that is being held for a long time, and then lose interest in it.
4810 * So, we clean out any un-revisited request after a lease period
4811 * under the assumption that the client is no longer interested.
4813 * RFC5661, sec. 9.6 states that the client must not rely on getting
4814 * notifications and must continue to poll for locks, even when the
4815 * server supports them. Thus this shouldn't lead to clients blocking
4816 * indefinitely once the lock does become free.
4818 BUG_ON(!list_empty(&reaplist));
4819 spin_lock(&nn->blocked_locks_lock);
4820 while (!list_empty(&nn->blocked_locks_lru)) {
4821 nbl = list_first_entry(&nn->blocked_locks_lru,
4822 struct nfsd4_blocked_lock, nbl_lru);
4823 if (time_after((unsigned long)nbl->nbl_time,
4824 (unsigned long)cutoff)) {
4825 t = nbl->nbl_time - cutoff;
4826 new_timeo = min(new_timeo, t);
4829 list_move(&nbl->nbl_lru, &reaplist);
4830 list_del_init(&nbl->nbl_list);
4832 spin_unlock(&nn->blocked_locks_lock);
4834 while (!list_empty(&reaplist)) {
4835 nbl = list_first_entry(&reaplist,
4836 struct nfsd4_blocked_lock, nbl_lru);
4837 list_del_init(&nbl->nbl_lru);
4838 posix_unblock_lock(&nbl->nbl_lock);
4839 free_blocked_lock(nbl);
4842 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4846 static struct workqueue_struct *laundry_wq;
4847 static void laundromat_main(struct work_struct *);
4850 laundromat_main(struct work_struct *laundry)
4853 struct delayed_work *dwork = to_delayed_work(laundry);
4854 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4857 t = nfs4_laundromat(nn);
4858 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4859 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4862 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4864 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4865 return nfserr_bad_stateid;
4870 access_permit_read(struct nfs4_ol_stateid *stp)
4872 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4873 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4874 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4878 access_permit_write(struct nfs4_ol_stateid *stp)
4880 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4881 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4885 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4887 __be32 status = nfserr_openmode;
4889 /* For lock stateid's, we test the parent open, not the lock: */
4890 if (stp->st_openstp)
4891 stp = stp->st_openstp;
4892 if ((flags & WR_STATE) && !access_permit_write(stp))
4894 if ((flags & RD_STATE) && !access_permit_read(stp))
4901 static inline __be32
4902 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4904 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4906 else if (opens_in_grace(net)) {
4907 /* Answer in remaining cases depends on existence of
4908 * conflicting state; so we must wait out the grace period. */
4909 return nfserr_grace;
4910 } else if (flags & WR_STATE)
4911 return nfs4_share_conflict(current_fh,
4912 NFS4_SHARE_DENY_WRITE);
4913 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4914 return nfs4_share_conflict(current_fh,
4915 NFS4_SHARE_DENY_READ);
4919 * Allow READ/WRITE during grace period on recovered state only for files
4920 * that are not able to provide mandatory locking.
4923 grace_disallows_io(struct net *net, struct inode *inode)
4925 return opens_in_grace(net) && mandatory_lock(inode);
4928 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4931 * When sessions are used the stateid generation number is ignored
4934 if (has_session && in->si_generation == 0)
4937 if (in->si_generation == ref->si_generation)
4940 /* If the client sends us a stateid from the future, it's buggy: */
4941 if (nfsd4_stateid_generation_after(in, ref))
4942 return nfserr_bad_stateid;
4944 * However, we could see a stateid from the past, even from a
4945 * non-buggy client. For example, if the client sends a lock
4946 * while some IO is outstanding, the lock may bump si_generation
4947 * while the IO is still in flight. The client could avoid that
4948 * situation by waiting for responses on all the IO requests,
4949 * but better performance may result in retrying IO that
4950 * receives an old_stateid error if requests are rarely
4951 * reordered in flight:
4953 return nfserr_old_stateid;
4956 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4960 spin_lock(&s->sc_lock);
4961 ret = nfsd4_verify_open_stid(s);
4963 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4964 spin_unlock(&s->sc_lock);
4968 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4970 if (ols->st_stateowner->so_is_open_owner &&
4971 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4972 return nfserr_bad_stateid;
4976 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4978 struct nfs4_stid *s;
4979 __be32 status = nfserr_bad_stateid;
4981 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4982 CLOSE_STATEID(stateid))
4984 /* Client debugging aid. */
4985 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4986 char addr_str[INET6_ADDRSTRLEN];
4987 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4989 pr_warn_ratelimited("NFSD: client %s testing state ID "
4990 "with incorrect client ID\n", addr_str);
4993 spin_lock(&cl->cl_lock);
4994 s = find_stateid_locked(cl, stateid);
4997 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5000 switch (s->sc_type) {
5001 case NFS4_DELEG_STID:
5004 case NFS4_REVOKED_DELEG_STID:
5005 status = nfserr_deleg_revoked;
5007 case NFS4_OPEN_STID:
5008 case NFS4_LOCK_STID:
5009 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5012 printk("unknown stateid type %x\n", s->sc_type);
5014 case NFS4_CLOSED_STID:
5015 case NFS4_CLOSED_DELEG_STID:
5016 status = nfserr_bad_stateid;
5019 spin_unlock(&cl->cl_lock);
5024 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5025 stateid_t *stateid, unsigned char typemask,
5026 struct nfs4_stid **s, struct nfsd_net *nn)
5029 bool return_revoked = false;
5032 * only return revoked delegations if explicitly asked.
5033 * otherwise we report revoked or bad_stateid status.
5035 if (typemask & NFS4_REVOKED_DELEG_STID)
5036 return_revoked = true;
5037 else if (typemask & NFS4_DELEG_STID)
5038 typemask |= NFS4_REVOKED_DELEG_STID;
5040 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5041 CLOSE_STATEID(stateid))
5042 return nfserr_bad_stateid;
5043 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5044 if (status == nfserr_stale_clientid) {
5045 if (cstate->session)
5046 return nfserr_bad_stateid;
5047 return nfserr_stale_stateid;
5051 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5053 return nfserr_bad_stateid;
5054 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5056 if (cstate->minorversion)
5057 return nfserr_deleg_revoked;
5058 return nfserr_bad_stateid;
5063 static struct file *
5064 nfs4_find_file(struct nfs4_stid *s, int flags)
5069 switch (s->sc_type) {
5070 case NFS4_DELEG_STID:
5071 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5073 return get_file(s->sc_file->fi_deleg_file);
5074 case NFS4_OPEN_STID:
5075 case NFS4_LOCK_STID:
5076 if (flags & RD_STATE)
5077 return find_readable_file(s->sc_file);
5079 return find_writeable_file(s->sc_file);
5087 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5091 status = nfsd4_check_openowner_confirmed(ols);
5094 return nfs4_check_openmode(ols, flags);
5098 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5099 struct file **filpp, bool *tmp_file, int flags)
5101 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5105 file = nfs4_find_file(s, flags);
5107 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5108 acc | NFSD_MAY_OWNER_OVERRIDE);
5116 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5128 * Checks for stateid operations
5131 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5132 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5133 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5135 struct inode *ino = d_inode(fhp->fh_dentry);
5136 struct net *net = SVC_NET(rqstp);
5137 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5138 struct nfs4_stid *s = NULL;
5146 if (grace_disallows_io(net, ino))
5147 return nfserr_grace;
5149 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5150 status = check_special_stateids(net, fhp, stateid, flags);
5154 status = nfsd4_lookup_stateid(cstate, stateid,
5155 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5159 status = nfsd4_stid_check_stateid_generation(stateid, s,
5160 nfsd4_has_session(cstate));
5164 switch (s->sc_type) {
5165 case NFS4_DELEG_STID:
5166 status = nfs4_check_delegmode(delegstateid(s), flags);
5168 case NFS4_OPEN_STID:
5169 case NFS4_LOCK_STID:
5170 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5173 status = nfserr_bad_stateid;
5178 status = nfs4_check_fh(fhp, s);
5181 if (!status && filpp)
5182 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5190 * Test if the stateid is valid
5193 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5194 union nfsd4_op_u *u)
5196 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5197 struct nfsd4_test_stateid_id *stateid;
5198 struct nfs4_client *cl = cstate->session->se_client;
5200 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5201 stateid->ts_id_status =
5202 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5208 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5210 struct nfs4_ol_stateid *stp = openlockstateid(s);
5213 ret = nfsd4_lock_ol_stateid(stp);
5217 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5221 ret = nfserr_locks_held;
5222 if (check_for_locks(stp->st_stid.sc_file,
5223 lockowner(stp->st_stateowner)))
5226 release_lock_stateid(stp);
5230 mutex_unlock(&stp->st_mutex);
5237 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5238 union nfsd4_op_u *u)
5240 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5241 stateid_t *stateid = &free_stateid->fr_stateid;
5242 struct nfs4_stid *s;
5243 struct nfs4_delegation *dp;
5244 struct nfs4_client *cl = cstate->session->se_client;
5245 __be32 ret = nfserr_bad_stateid;
5247 spin_lock(&cl->cl_lock);
5248 s = find_stateid_locked(cl, stateid);
5251 spin_lock(&s->sc_lock);
5252 switch (s->sc_type) {
5253 case NFS4_DELEG_STID:
5254 ret = nfserr_locks_held;
5256 case NFS4_OPEN_STID:
5257 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5260 ret = nfserr_locks_held;
5262 case NFS4_LOCK_STID:
5263 spin_unlock(&s->sc_lock);
5264 refcount_inc(&s->sc_count);
5265 spin_unlock(&cl->cl_lock);
5266 ret = nfsd4_free_lock_stateid(stateid, s);
5268 case NFS4_REVOKED_DELEG_STID:
5269 spin_unlock(&s->sc_lock);
5270 dp = delegstateid(s);
5271 list_del_init(&dp->dl_recall_lru);
5272 spin_unlock(&cl->cl_lock);
5276 /* Default falls through and returns nfserr_bad_stateid */
5278 spin_unlock(&s->sc_lock);
5280 spin_unlock(&cl->cl_lock);
5288 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5289 RD_STATE : WR_STATE;
5292 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5294 struct svc_fh *current_fh = &cstate->current_fh;
5295 struct nfs4_stateowner *sop = stp->st_stateowner;
5298 status = nfsd4_check_seqid(cstate, sop, seqid);
5301 status = nfsd4_lock_ol_stateid(stp);
5302 if (status != nfs_ok)
5304 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5305 if (status == nfs_ok)
5306 status = nfs4_check_fh(current_fh, &stp->st_stid);
5307 if (status != nfs_ok)
5308 mutex_unlock(&stp->st_mutex);
5313 * Checks for sequence id mutating operations.
5316 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5317 stateid_t *stateid, char typemask,
5318 struct nfs4_ol_stateid **stpp,
5319 struct nfsd_net *nn)
5322 struct nfs4_stid *s;
5323 struct nfs4_ol_stateid *stp = NULL;
5325 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5326 seqid, STATEID_VAL(stateid));
5329 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5332 stp = openlockstateid(s);
5333 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5335 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5339 nfs4_put_stid(&stp->st_stid);
5343 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5344 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5347 struct nfs4_openowner *oo;
5348 struct nfs4_ol_stateid *stp;
5350 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5351 NFS4_OPEN_STID, &stp, nn);
5354 oo = openowner(stp->st_stateowner);
5355 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5356 mutex_unlock(&stp->st_mutex);
5357 nfs4_put_stid(&stp->st_stid);
5358 return nfserr_bad_stateid;
5365 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5366 union nfsd4_op_u *u)
5368 struct nfsd4_open_confirm *oc = &u->open_confirm;
5370 struct nfs4_openowner *oo;
5371 struct nfs4_ol_stateid *stp;
5372 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5374 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5375 cstate->current_fh.fh_dentry);
5377 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5381 status = nfs4_preprocess_seqid_op(cstate,
5382 oc->oc_seqid, &oc->oc_req_stateid,
5383 NFS4_OPEN_STID, &stp, nn);
5386 oo = openowner(stp->st_stateowner);
5387 status = nfserr_bad_stateid;
5388 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5389 mutex_unlock(&stp->st_mutex);
5392 oo->oo_flags |= NFS4_OO_CONFIRMED;
5393 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5394 mutex_unlock(&stp->st_mutex);
5395 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5396 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5398 nfsd4_client_record_create(oo->oo_owner.so_client);
5401 nfs4_put_stid(&stp->st_stid);
5403 nfsd4_bump_seqid(cstate, status);
5407 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5409 if (!test_access(access, stp))
5411 nfs4_file_put_access(stp->st_stid.sc_file, access);
5412 clear_access(access, stp);
5415 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5417 switch (to_access) {
5418 case NFS4_SHARE_ACCESS_READ:
5419 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5420 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5422 case NFS4_SHARE_ACCESS_WRITE:
5423 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5424 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5426 case NFS4_SHARE_ACCESS_BOTH:
5434 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5435 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5437 struct nfsd4_open_downgrade *od = &u->open_downgrade;
5439 struct nfs4_ol_stateid *stp;
5440 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5442 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5443 cstate->current_fh.fh_dentry);
5445 /* We don't yet support WANT bits: */
5446 if (od->od_deleg_want)
5447 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5450 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5451 &od->od_stateid, &stp, nn);
5454 status = nfserr_inval;
5455 if (!test_access(od->od_share_access, stp)) {
5456 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5457 stp->st_access_bmap, od->od_share_access);
5460 if (!test_deny(od->od_share_deny, stp)) {
5461 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5462 stp->st_deny_bmap, od->od_share_deny);
5465 nfs4_stateid_downgrade(stp, od->od_share_access);
5466 reset_union_bmap_deny(od->od_share_deny, stp);
5467 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5470 mutex_unlock(&stp->st_mutex);
5471 nfs4_put_stid(&stp->st_stid);
5473 nfsd4_bump_seqid(cstate, status);
5477 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5479 struct nfs4_client *clp = s->st_stid.sc_client;
5481 LIST_HEAD(reaplist);
5483 spin_lock(&clp->cl_lock);
5484 unhashed = unhash_open_stateid(s, &reaplist);
5486 if (clp->cl_minorversion) {
5488 put_ol_stateid_locked(s, &reaplist);
5489 spin_unlock(&clp->cl_lock);
5490 free_ol_stateid_reaplist(&reaplist);
5492 spin_unlock(&clp->cl_lock);
5493 free_ol_stateid_reaplist(&reaplist);
5495 move_to_close_lru(s, clp->net);
5500 * nfs4_unlock_state() called after encode
5503 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5504 union nfsd4_op_u *u)
5506 struct nfsd4_close *close = &u->close;
5508 struct nfs4_ol_stateid *stp;
5509 struct net *net = SVC_NET(rqstp);
5510 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5512 dprintk("NFSD: nfsd4_close on file %pd\n",
5513 cstate->current_fh.fh_dentry);
5515 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5517 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5519 nfsd4_bump_seqid(cstate, status);
5523 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5526 * Technically we don't _really_ have to increment or copy it, since
5527 * it should just be gone after this operation and we clobber the
5528 * copied value below, but we continue to do so here just to ensure
5529 * that racing ops see that there was a state change.
5531 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5533 nfsd4_close_open_stateid(stp);
5534 mutex_unlock(&stp->st_mutex);
5536 /* v4.1+ suggests that we send a special stateid in here, since the
5537 * clients should just ignore this anyway. Since this is not useful
5538 * for v4.0 clients either, we set it to the special close_stateid
5541 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5543 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5545 /* put reference from nfs4_preprocess_seqid_op */
5546 nfs4_put_stid(&stp->st_stid);
5552 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5553 union nfsd4_op_u *u)
5555 struct nfsd4_delegreturn *dr = &u->delegreturn;
5556 struct nfs4_delegation *dp;
5557 stateid_t *stateid = &dr->dr_stateid;
5558 struct nfs4_stid *s;
5560 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5562 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5565 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5568 dp = delegstateid(s);
5569 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5573 destroy_delegation(dp);
5575 nfs4_put_stid(&dp->dl_stid);
5581 end_offset(u64 start, u64 len)
5586 return end >= start ? end: NFS4_MAX_UINT64;
5589 /* last octet in a range */
5591 last_byte_offset(u64 start, u64 len)
5597 return end > start ? end - 1: NFS4_MAX_UINT64;
5601 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5602 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5603 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5604 * locking, this prevents us from being completely protocol-compliant. The
5605 * real solution to this problem is to start using unsigned file offsets in
5606 * the VFS, but this is a very deep change!
5609 nfs4_transform_lock_offset(struct file_lock *lock)
5611 if (lock->fl_start < 0)
5612 lock->fl_start = OFFSET_MAX;
5613 if (lock->fl_end < 0)
5614 lock->fl_end = OFFSET_MAX;
5618 nfsd4_fl_get_owner(fl_owner_t owner)
5620 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5622 nfs4_get_stateowner(&lo->lo_owner);
5627 nfsd4_fl_put_owner(fl_owner_t owner)
5629 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5632 nfs4_put_stateowner(&lo->lo_owner);
5636 nfsd4_lm_notify(struct file_lock *fl)
5638 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
5639 struct net *net = lo->lo_owner.so_client->net;
5640 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5641 struct nfsd4_blocked_lock *nbl = container_of(fl,
5642 struct nfsd4_blocked_lock, nbl_lock);
5645 /* An empty list means that something else is going to be using it */
5646 spin_lock(&nn->blocked_locks_lock);
5647 if (!list_empty(&nbl->nbl_list)) {
5648 list_del_init(&nbl->nbl_list);
5649 list_del_init(&nbl->nbl_lru);
5652 spin_unlock(&nn->blocked_locks_lock);
5655 nfsd4_run_cb(&nbl->nbl_cb);
5658 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5659 .lm_notify = nfsd4_lm_notify,
5660 .lm_get_owner = nfsd4_fl_get_owner,
5661 .lm_put_owner = nfsd4_fl_put_owner,
5665 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5667 struct nfs4_lockowner *lo;
5669 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5670 lo = (struct nfs4_lockowner *) fl->fl_owner;
5671 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5672 lo->lo_owner.so_owner.len, GFP_KERNEL);
5673 if (!deny->ld_owner.data)
5674 /* We just don't care that much */
5676 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5677 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5680 deny->ld_owner.len = 0;
5681 deny->ld_owner.data = NULL;
5682 deny->ld_clientid.cl_boot = 0;
5683 deny->ld_clientid.cl_id = 0;
5685 deny->ld_start = fl->fl_start;
5686 deny->ld_length = NFS4_MAX_UINT64;
5687 if (fl->fl_end != NFS4_MAX_UINT64)
5688 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5689 deny->ld_type = NFS4_READ_LT;
5690 if (fl->fl_type != F_RDLCK)
5691 deny->ld_type = NFS4_WRITE_LT;
5694 static struct nfs4_lockowner *
5695 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5697 unsigned int strhashval = ownerstr_hashval(owner);
5698 struct nfs4_stateowner *so;
5700 lockdep_assert_held(&clp->cl_lock);
5702 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5704 if (so->so_is_open_owner)
5706 if (same_owner_str(so, owner))
5707 return lockowner(nfs4_get_stateowner(so));
5712 static struct nfs4_lockowner *
5713 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5715 struct nfs4_lockowner *lo;
5717 spin_lock(&clp->cl_lock);
5718 lo = find_lockowner_str_locked(clp, owner);
5719 spin_unlock(&clp->cl_lock);
5723 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5725 unhash_lockowner_locked(lockowner(sop));
5728 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5730 struct nfs4_lockowner *lo = lockowner(sop);
5732 kmem_cache_free(lockowner_slab, lo);
5735 static const struct nfs4_stateowner_operations lockowner_ops = {
5736 .so_unhash = nfs4_unhash_lockowner,
5737 .so_free = nfs4_free_lockowner,
5741 * Alloc a lock owner structure.
5742 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5745 * strhashval = ownerstr_hashval
5747 static struct nfs4_lockowner *
5748 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5749 struct nfs4_ol_stateid *open_stp,
5750 struct nfsd4_lock *lock)
5752 struct nfs4_lockowner *lo, *ret;
5754 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5757 INIT_LIST_HEAD(&lo->lo_blocked);
5758 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5759 lo->lo_owner.so_is_open_owner = 0;
5760 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5761 lo->lo_owner.so_ops = &lockowner_ops;
5762 spin_lock(&clp->cl_lock);
5763 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5765 list_add(&lo->lo_owner.so_strhash,
5766 &clp->cl_ownerstr_hashtbl[strhashval]);
5769 nfs4_free_stateowner(&lo->lo_owner);
5771 spin_unlock(&clp->cl_lock);
5775 static struct nfs4_ol_stateid *
5776 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5778 struct nfs4_ol_stateid *lst;
5779 struct nfs4_client *clp = lo->lo_owner.so_client;
5781 lockdep_assert_held(&clp->cl_lock);
5783 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5784 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5786 if (lst->st_stid.sc_file == fp) {
5787 refcount_inc(&lst->st_stid.sc_count);
5794 static struct nfs4_ol_stateid *
5795 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5796 struct nfs4_file *fp, struct inode *inode,
5797 struct nfs4_ol_stateid *open_stp)
5799 struct nfs4_client *clp = lo->lo_owner.so_client;
5800 struct nfs4_ol_stateid *retstp;
5802 mutex_init(&stp->st_mutex);
5803 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5805 spin_lock(&clp->cl_lock);
5806 spin_lock(&fp->fi_lock);
5807 retstp = find_lock_stateid(lo, fp);
5811 refcount_inc(&stp->st_stid.sc_count);
5812 stp->st_stid.sc_type = NFS4_LOCK_STID;
5813 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5815 stp->st_stid.sc_file = fp;
5816 stp->st_access_bmap = 0;
5817 stp->st_deny_bmap = open_stp->st_deny_bmap;
5818 stp->st_openstp = open_stp;
5819 list_add(&stp->st_locks, &open_stp->st_locks);
5820 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5821 list_add(&stp->st_perfile, &fp->fi_stateids);
5823 spin_unlock(&fp->fi_lock);
5824 spin_unlock(&clp->cl_lock);
5826 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5827 nfs4_put_stid(&retstp->st_stid);
5830 /* To keep mutex tracking happy */
5831 mutex_unlock(&stp->st_mutex);
5837 static struct nfs4_ol_stateid *
5838 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5839 struct inode *inode, struct nfs4_ol_stateid *ost,
5842 struct nfs4_stid *ns = NULL;
5843 struct nfs4_ol_stateid *lst;
5844 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5845 struct nfs4_client *clp = oo->oo_owner.so_client;
5848 spin_lock(&clp->cl_lock);
5849 lst = find_lock_stateid(lo, fi);
5850 spin_unlock(&clp->cl_lock);
5852 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5854 nfs4_put_stid(&lst->st_stid);
5856 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5860 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5861 if (lst == openlockstateid(ns))
5870 check_lock_length(u64 offset, u64 length)
5872 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5873 (length > ~offset)));
5876 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5878 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5880 lockdep_assert_held(&fp->fi_lock);
5882 if (test_access(access, lock_stp))
5884 __nfs4_file_get_access(fp, access);
5885 set_access(access, lock_stp);
5889 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5890 struct nfs4_ol_stateid *ost,
5891 struct nfsd4_lock *lock,
5892 struct nfs4_ol_stateid **plst, bool *new)
5895 struct nfs4_file *fi = ost->st_stid.sc_file;
5896 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5897 struct nfs4_client *cl = oo->oo_owner.so_client;
5898 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5899 struct nfs4_lockowner *lo;
5900 struct nfs4_ol_stateid *lst;
5901 unsigned int strhashval;
5903 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5905 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5906 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5908 return nfserr_jukebox;
5910 /* with an existing lockowner, seqids must be the same */
5911 status = nfserr_bad_seqid;
5912 if (!cstate->minorversion &&
5913 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5917 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5919 status = nfserr_jukebox;
5926 nfs4_put_stateowner(&lo->lo_owner);
5934 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5935 union nfsd4_op_u *u)
5937 struct nfsd4_lock *lock = &u->lock;
5938 struct nfs4_openowner *open_sop = NULL;
5939 struct nfs4_lockowner *lock_sop = NULL;
5940 struct nfs4_ol_stateid *lock_stp = NULL;
5941 struct nfs4_ol_stateid *open_stp = NULL;
5942 struct nfs4_file *fp;
5943 struct file *filp = NULL;
5944 struct nfsd4_blocked_lock *nbl = NULL;
5945 struct file_lock *file_lock = NULL;
5946 struct file_lock *conflock = NULL;
5951 unsigned char fl_type;
5952 unsigned int fl_flags = FL_POSIX;
5953 struct net *net = SVC_NET(rqstp);
5954 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5956 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5957 (long long) lock->lk_offset,
5958 (long long) lock->lk_length);
5960 if (check_lock_length(lock->lk_offset, lock->lk_length))
5961 return nfserr_inval;
5963 if ((status = fh_verify(rqstp, &cstate->current_fh,
5964 S_IFREG, NFSD_MAY_LOCK))) {
5965 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5969 if (lock->lk_is_new) {
5970 if (nfsd4_has_session(cstate))
5971 /* See rfc 5661 18.10.3: given clientid is ignored: */
5972 memcpy(&lock->lk_new_clientid,
5973 &cstate->session->se_client->cl_clientid,
5974 sizeof(clientid_t));
5976 status = nfserr_stale_clientid;
5977 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5980 /* validate and update open stateid and open seqid */
5981 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5982 lock->lk_new_open_seqid,
5983 &lock->lk_new_open_stateid,
5987 mutex_unlock(&open_stp->st_mutex);
5988 open_sop = openowner(open_stp->st_stateowner);
5989 status = nfserr_bad_stateid;
5990 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5991 &lock->lk_new_clientid))
5993 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5996 status = nfs4_preprocess_seqid_op(cstate,
5997 lock->lk_old_lock_seqid,
5998 &lock->lk_old_lock_stateid,
5999 NFS4_LOCK_STID, &lock_stp, nn);
6003 lock_sop = lockowner(lock_stp->st_stateowner);
6005 lkflg = setlkflg(lock->lk_type);
6006 status = nfs4_check_openmode(lock_stp, lkflg);
6010 status = nfserr_grace;
6011 if (locks_in_grace(net) && !lock->lk_reclaim)
6013 status = nfserr_no_grace;
6014 if (!locks_in_grace(net) && lock->lk_reclaim)
6017 fp = lock_stp->st_stid.sc_file;
6018 switch (lock->lk_type) {
6020 if (nfsd4_has_session(cstate))
6021 fl_flags |= FL_SLEEP;
6024 spin_lock(&fp->fi_lock);
6025 filp = find_readable_file_locked(fp);
6027 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6028 spin_unlock(&fp->fi_lock);
6031 case NFS4_WRITEW_LT:
6032 if (nfsd4_has_session(cstate))
6033 fl_flags |= FL_SLEEP;
6036 spin_lock(&fp->fi_lock);
6037 filp = find_writeable_file_locked(fp);
6039 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6040 spin_unlock(&fp->fi_lock);
6044 status = nfserr_inval;
6049 status = nfserr_openmode;
6053 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6055 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6056 status = nfserr_jukebox;
6060 file_lock = &nbl->nbl_lock;
6061 file_lock->fl_type = fl_type;
6062 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6063 file_lock->fl_pid = current->tgid;
6064 file_lock->fl_file = filp;
6065 file_lock->fl_flags = fl_flags;
6066 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6067 file_lock->fl_start = lock->lk_offset;
6068 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6069 nfs4_transform_lock_offset(file_lock);
6071 conflock = locks_alloc_lock();
6073 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6074 status = nfserr_jukebox;
6078 if (fl_flags & FL_SLEEP) {
6079 nbl->nbl_time = jiffies;
6080 spin_lock(&nn->blocked_locks_lock);
6081 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6082 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6083 spin_unlock(&nn->blocked_locks_lock);
6086 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6088 case 0: /* success! */
6089 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6092 case FILE_LOCK_DEFERRED:
6095 case -EAGAIN: /* conflock holds conflicting lock */
6096 status = nfserr_denied;
6097 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6098 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6101 status = nfserr_deadlock;
6104 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6105 status = nfserrno(err);
6110 /* dequeue it if we queued it before */
6111 if (fl_flags & FL_SLEEP) {
6112 spin_lock(&nn->blocked_locks_lock);
6113 list_del_init(&nbl->nbl_list);
6114 list_del_init(&nbl->nbl_lru);
6115 spin_unlock(&nn->blocked_locks_lock);
6117 free_blocked_lock(nbl);
6122 /* Bump seqid manually if the 4.0 replay owner is openowner */
6123 if (cstate->replay_owner &&
6124 cstate->replay_owner != &lock_sop->lo_owner &&
6125 seqid_mutating_err(ntohl(status)))
6126 lock_sop->lo_owner.so_seqid++;
6129 * If this is a new, never-before-used stateid, and we are
6130 * returning an error, then just go ahead and release it.
6133 release_lock_stateid(lock_stp);
6135 mutex_unlock(&lock_stp->st_mutex);
6137 nfs4_put_stid(&lock_stp->st_stid);
6140 nfs4_put_stid(&open_stp->st_stid);
6141 nfsd4_bump_seqid(cstate, status);
6143 locks_free_lock(conflock);
6148 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6149 * so we do a temporary open here just to get an open file to pass to
6150 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6153 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6156 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6158 err = nfserrno(vfs_test_lock(file, lock));
6168 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6169 union nfsd4_op_u *u)
6171 struct nfsd4_lockt *lockt = &u->lockt;
6172 struct file_lock *file_lock = NULL;
6173 struct nfs4_lockowner *lo = NULL;
6175 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6177 if (locks_in_grace(SVC_NET(rqstp)))
6178 return nfserr_grace;
6180 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6181 return nfserr_inval;
6183 if (!nfsd4_has_session(cstate)) {
6184 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6189 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6192 file_lock = locks_alloc_lock();
6194 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6195 status = nfserr_jukebox;
6199 switch (lockt->lt_type) {
6202 file_lock->fl_type = F_RDLCK;
6205 case NFS4_WRITEW_LT:
6206 file_lock->fl_type = F_WRLCK;
6209 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6210 status = nfserr_inval;
6214 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6216 file_lock->fl_owner = (fl_owner_t)lo;
6217 file_lock->fl_pid = current->tgid;
6218 file_lock->fl_flags = FL_POSIX;
6220 file_lock->fl_start = lockt->lt_offset;
6221 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6223 nfs4_transform_lock_offset(file_lock);
6225 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6229 if (file_lock->fl_type != F_UNLCK) {
6230 status = nfserr_denied;
6231 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6235 nfs4_put_stateowner(&lo->lo_owner);
6237 locks_free_lock(file_lock);
6242 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6243 union nfsd4_op_u *u)
6245 struct nfsd4_locku *locku = &u->locku;
6246 struct nfs4_ol_stateid *stp;
6247 struct file *filp = NULL;
6248 struct file_lock *file_lock = NULL;
6251 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6253 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6254 (long long) locku->lu_offset,
6255 (long long) locku->lu_length);
6257 if (check_lock_length(locku->lu_offset, locku->lu_length))
6258 return nfserr_inval;
6260 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6261 &locku->lu_stateid, NFS4_LOCK_STID,
6265 filp = find_any_file(stp->st_stid.sc_file);
6267 status = nfserr_lock_range;
6270 file_lock = locks_alloc_lock();
6272 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6273 status = nfserr_jukebox;
6277 file_lock->fl_type = F_UNLCK;
6278 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6279 file_lock->fl_pid = current->tgid;
6280 file_lock->fl_file = filp;
6281 file_lock->fl_flags = FL_POSIX;
6282 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6283 file_lock->fl_start = locku->lu_offset;
6285 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6287 nfs4_transform_lock_offset(file_lock);
6289 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6291 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6294 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6298 mutex_unlock(&stp->st_mutex);
6299 nfs4_put_stid(&stp->st_stid);
6301 nfsd4_bump_seqid(cstate, status);
6303 locks_free_lock(file_lock);
6307 status = nfserrno(err);
6313 * true: locks held by lockowner
6314 * false: no locks held by lockowner
6317 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6319 struct file_lock *fl;
6321 struct file *filp = find_any_file(fp);
6322 struct inode *inode;
6323 struct file_lock_context *flctx;
6326 /* Any valid lock stateid should have some sort of access */
6331 inode = file_inode(filp);
6332 flctx = inode->i_flctx;
6334 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6335 spin_lock(&flctx->flc_lock);
6336 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6337 if (fl->fl_owner == (fl_owner_t)lowner) {
6342 spin_unlock(&flctx->flc_lock);
6349 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6350 struct nfsd4_compound_state *cstate,
6351 union nfsd4_op_u *u)
6353 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6354 clientid_t *clid = &rlockowner->rl_clientid;
6355 struct nfs4_stateowner *sop;
6356 struct nfs4_lockowner *lo = NULL;
6357 struct nfs4_ol_stateid *stp;
6358 struct xdr_netobj *owner = &rlockowner->rl_owner;
6359 unsigned int hashval = ownerstr_hashval(owner);
6361 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6362 struct nfs4_client *clp;
6363 LIST_HEAD (reaplist);
6365 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6366 clid->cl_boot, clid->cl_id);
6368 status = lookup_clientid(clid, cstate, nn);
6373 /* Find the matching lock stateowner */
6374 spin_lock(&clp->cl_lock);
6375 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6378 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6381 /* see if there are still any locks associated with it */
6382 lo = lockowner(sop);
6383 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6384 if (check_for_locks(stp->st_stid.sc_file, lo)) {
6385 status = nfserr_locks_held;
6386 spin_unlock(&clp->cl_lock);
6391 nfs4_get_stateowner(sop);
6395 spin_unlock(&clp->cl_lock);
6399 unhash_lockowner_locked(lo);
6400 while (!list_empty(&lo->lo_owner.so_stateids)) {
6401 stp = list_first_entry(&lo->lo_owner.so_stateids,
6402 struct nfs4_ol_stateid,
6404 WARN_ON(!unhash_lock_stateid(stp));
6405 put_ol_stateid_locked(stp, &reaplist);
6407 spin_unlock(&clp->cl_lock);
6408 free_ol_stateid_reaplist(&reaplist);
6409 remove_blocked_locks(lo);
6410 nfs4_put_stateowner(&lo->lo_owner);
6415 static inline struct nfs4_client_reclaim *
6418 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6422 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6424 struct nfs4_client_reclaim *crp;
6426 crp = nfsd4_find_reclaim_client(name, nn);
6427 return (crp && crp->cr_clp);
6431 * failure => all reset bets are off, nfserr_no_grace...
6433 struct nfs4_client_reclaim *
6434 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6436 unsigned int strhashval;
6437 struct nfs4_client_reclaim *crp;
6439 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6440 crp = alloc_reclaim();
6442 strhashval = clientstr_hashval(name);
6443 INIT_LIST_HEAD(&crp->cr_strhash);
6444 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6445 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6447 nn->reclaim_str_hashtbl_size++;
6453 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6455 list_del(&crp->cr_strhash);
6457 nn->reclaim_str_hashtbl_size--;
6461 nfs4_release_reclaim(struct nfsd_net *nn)
6463 struct nfs4_client_reclaim *crp = NULL;
6466 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6467 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6468 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6469 struct nfs4_client_reclaim, cr_strhash);
6470 nfs4_remove_reclaim_record(crp, nn);
6473 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6477 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6478 struct nfs4_client_reclaim *
6479 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6481 unsigned int strhashval;
6482 struct nfs4_client_reclaim *crp = NULL;
6484 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6486 strhashval = clientstr_hashval(recdir);
6487 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6488 if (same_name(crp->cr_recdir, recdir)) {
6496 * Called from OPEN. Look for clientid in reclaim list.
6499 nfs4_check_open_reclaim(clientid_t *clid,
6500 struct nfsd4_compound_state *cstate,
6501 struct nfsd_net *nn)
6505 /* find clientid in conf_id_hashtbl */
6506 status = lookup_clientid(clid, cstate, nn);
6508 return nfserr_reclaim_bad;
6510 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6511 return nfserr_no_grace;
6513 if (nfsd4_client_record_check(cstate->clp))
6514 return nfserr_reclaim_bad;
6519 #ifdef CONFIG_NFSD_FAULT_INJECTION
6521 put_client(struct nfs4_client *clp)
6523 atomic_dec(&clp->cl_refcount);
6526 static struct nfs4_client *
6527 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6529 struct nfs4_client *clp;
6530 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6533 if (!nfsd_netns_ready(nn))
6536 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6537 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6544 nfsd_inject_print_clients(void)
6546 struct nfs4_client *clp;
6548 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6550 char buf[INET6_ADDRSTRLEN];
6552 if (!nfsd_netns_ready(nn))
6555 spin_lock(&nn->client_lock);
6556 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6557 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6558 pr_info("NFS Client: %s\n", buf);
6561 spin_unlock(&nn->client_lock);
6567 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6570 struct nfs4_client *clp;
6571 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6574 if (!nfsd_netns_ready(nn))
6577 spin_lock(&nn->client_lock);
6578 clp = nfsd_find_client(addr, addr_size);
6580 if (mark_client_expired_locked(clp) == nfs_ok)
6585 spin_unlock(&nn->client_lock);
6594 nfsd_inject_forget_clients(u64 max)
6597 struct nfs4_client *clp, *next;
6598 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6600 LIST_HEAD(reaplist);
6602 if (!nfsd_netns_ready(nn))
6605 spin_lock(&nn->client_lock);
6606 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6607 if (mark_client_expired_locked(clp) == nfs_ok) {
6608 list_add(&clp->cl_lru, &reaplist);
6609 if (max != 0 && ++count >= max)
6613 spin_unlock(&nn->client_lock);
6615 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6621 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6624 char buf[INET6_ADDRSTRLEN];
6625 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6626 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6630 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6631 struct list_head *collect)
6633 struct nfs4_client *clp = lst->st_stid.sc_client;
6634 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6640 lockdep_assert_held(&nn->client_lock);
6641 atomic_inc(&clp->cl_refcount);
6642 list_add(&lst->st_locks, collect);
6645 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6646 struct list_head *collect,
6647 bool (*func)(struct nfs4_ol_stateid *))
6649 struct nfs4_openowner *oop;
6650 struct nfs4_ol_stateid *stp, *st_next;
6651 struct nfs4_ol_stateid *lst, *lst_next;
6654 spin_lock(&clp->cl_lock);
6655 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6656 list_for_each_entry_safe(stp, st_next,
6657 &oop->oo_owner.so_stateids, st_perstateowner) {
6658 list_for_each_entry_safe(lst, lst_next,
6659 &stp->st_locks, st_locks) {
6662 nfsd_inject_add_lock_to_list(lst,
6667 * Despite the fact that these functions deal
6668 * with 64-bit integers for "count", we must
6669 * ensure that it doesn't blow up the
6670 * clp->cl_refcount. Throw a warning if we
6671 * start to approach INT_MAX here.
6673 WARN_ON_ONCE(count == (INT_MAX / 2));
6680 spin_unlock(&clp->cl_lock);
6686 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6689 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6693 nfsd_print_client_locks(struct nfs4_client *clp)
6695 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6696 nfsd_print_count(clp, count, "locked files");
6701 nfsd_inject_print_locks(void)
6703 struct nfs4_client *clp;
6705 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6708 if (!nfsd_netns_ready(nn))
6711 spin_lock(&nn->client_lock);
6712 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6713 count += nfsd_print_client_locks(clp);
6714 spin_unlock(&nn->client_lock);
6720 nfsd_reap_locks(struct list_head *reaplist)
6722 struct nfs4_client *clp;
6723 struct nfs4_ol_stateid *stp, *next;
6725 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6726 list_del_init(&stp->st_locks);
6727 clp = stp->st_stid.sc_client;
6728 nfs4_put_stid(&stp->st_stid);
6734 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6736 unsigned int count = 0;
6737 struct nfs4_client *clp;
6738 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6740 LIST_HEAD(reaplist);
6742 if (!nfsd_netns_ready(nn))
6745 spin_lock(&nn->client_lock);
6746 clp = nfsd_find_client(addr, addr_size);
6748 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6749 spin_unlock(&nn->client_lock);
6750 nfsd_reap_locks(&reaplist);
6755 nfsd_inject_forget_locks(u64 max)
6758 struct nfs4_client *clp;
6759 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6761 LIST_HEAD(reaplist);
6763 if (!nfsd_netns_ready(nn))
6766 spin_lock(&nn->client_lock);
6767 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6768 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6769 if (max != 0 && count >= max)
6772 spin_unlock(&nn->client_lock);
6773 nfsd_reap_locks(&reaplist);
6778 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6779 struct list_head *collect,
6780 void (*func)(struct nfs4_openowner *))
6782 struct nfs4_openowner *oop, *next;
6783 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6787 lockdep_assert_held(&nn->client_lock);
6789 spin_lock(&clp->cl_lock);
6790 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6794 atomic_inc(&clp->cl_refcount);
6795 list_add(&oop->oo_perclient, collect);
6800 * Despite the fact that these functions deal with
6801 * 64-bit integers for "count", we must ensure that
6802 * it doesn't blow up the clp->cl_refcount. Throw a
6803 * warning if we start to approach INT_MAX here.
6805 WARN_ON_ONCE(count == (INT_MAX / 2));
6809 spin_unlock(&clp->cl_lock);
6815 nfsd_print_client_openowners(struct nfs4_client *clp)
6817 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6819 nfsd_print_count(clp, count, "openowners");
6824 nfsd_collect_client_openowners(struct nfs4_client *clp,
6825 struct list_head *collect, u64 max)
6827 return nfsd_foreach_client_openowner(clp, max, collect,
6828 unhash_openowner_locked);
6832 nfsd_inject_print_openowners(void)
6834 struct nfs4_client *clp;
6836 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6839 if (!nfsd_netns_ready(nn))
6842 spin_lock(&nn->client_lock);
6843 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6844 count += nfsd_print_client_openowners(clp);
6845 spin_unlock(&nn->client_lock);
6851 nfsd_reap_openowners(struct list_head *reaplist)
6853 struct nfs4_client *clp;
6854 struct nfs4_openowner *oop, *next;
6856 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6857 list_del_init(&oop->oo_perclient);
6858 clp = oop->oo_owner.so_client;
6859 release_openowner(oop);
6865 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6868 unsigned int count = 0;
6869 struct nfs4_client *clp;
6870 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6872 LIST_HEAD(reaplist);
6874 if (!nfsd_netns_ready(nn))
6877 spin_lock(&nn->client_lock);
6878 clp = nfsd_find_client(addr, addr_size);
6880 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6881 spin_unlock(&nn->client_lock);
6882 nfsd_reap_openowners(&reaplist);
6887 nfsd_inject_forget_openowners(u64 max)
6890 struct nfs4_client *clp;
6891 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6893 LIST_HEAD(reaplist);
6895 if (!nfsd_netns_ready(nn))
6898 spin_lock(&nn->client_lock);
6899 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6900 count += nfsd_collect_client_openowners(clp, &reaplist,
6902 if (max != 0 && count >= max)
6905 spin_unlock(&nn->client_lock);
6906 nfsd_reap_openowners(&reaplist);
6910 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6911 struct list_head *victims)
6913 struct nfs4_delegation *dp, *next;
6914 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6918 lockdep_assert_held(&nn->client_lock);
6920 spin_lock(&state_lock);
6921 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6924 * It's not safe to mess with delegations that have a
6925 * non-zero dl_time. They might have already been broken
6926 * and could be processed by the laundromat outside of
6927 * the state_lock. Just leave them be.
6929 if (dp->dl_time != 0)
6932 atomic_inc(&clp->cl_refcount);
6933 WARN_ON(!unhash_delegation_locked(dp));
6934 list_add(&dp->dl_recall_lru, victims);
6938 * Despite the fact that these functions deal with
6939 * 64-bit integers for "count", we must ensure that
6940 * it doesn't blow up the clp->cl_refcount. Throw a
6941 * warning if we start to approach INT_MAX here.
6943 WARN_ON_ONCE(count == (INT_MAX / 2));
6947 spin_unlock(&state_lock);
6952 nfsd_print_client_delegations(struct nfs4_client *clp)
6954 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6956 nfsd_print_count(clp, count, "delegations");
6961 nfsd_inject_print_delegations(void)
6963 struct nfs4_client *clp;
6965 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6968 if (!nfsd_netns_ready(nn))
6971 spin_lock(&nn->client_lock);
6972 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6973 count += nfsd_print_client_delegations(clp);
6974 spin_unlock(&nn->client_lock);
6980 nfsd_forget_delegations(struct list_head *reaplist)
6982 struct nfs4_client *clp;
6983 struct nfs4_delegation *dp, *next;
6985 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6986 list_del_init(&dp->dl_recall_lru);
6987 clp = dp->dl_stid.sc_client;
6988 revoke_delegation(dp);
6994 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6998 struct nfs4_client *clp;
6999 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7001 LIST_HEAD(reaplist);
7003 if (!nfsd_netns_ready(nn))
7006 spin_lock(&nn->client_lock);
7007 clp = nfsd_find_client(addr, addr_size);
7009 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7010 spin_unlock(&nn->client_lock);
7012 nfsd_forget_delegations(&reaplist);
7017 nfsd_inject_forget_delegations(u64 max)
7020 struct nfs4_client *clp;
7021 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7023 LIST_HEAD(reaplist);
7025 if (!nfsd_netns_ready(nn))
7028 spin_lock(&nn->client_lock);
7029 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7030 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7031 if (max != 0 && count >= max)
7034 spin_unlock(&nn->client_lock);
7035 nfsd_forget_delegations(&reaplist);
7040 nfsd_recall_delegations(struct list_head *reaplist)
7042 struct nfs4_client *clp;
7043 struct nfs4_delegation *dp, *next;
7045 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7046 list_del_init(&dp->dl_recall_lru);
7047 clp = dp->dl_stid.sc_client;
7049 * We skipped all entries that had a zero dl_time before,
7050 * so we can now reset the dl_time back to 0. If a delegation
7051 * break comes in now, then it won't make any difference since
7052 * we're recalling it either way.
7054 spin_lock(&state_lock);
7056 spin_unlock(&state_lock);
7057 nfsd_break_one_deleg(dp);
7063 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7067 struct nfs4_client *clp;
7068 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7070 LIST_HEAD(reaplist);
7072 if (!nfsd_netns_ready(nn))
7075 spin_lock(&nn->client_lock);
7076 clp = nfsd_find_client(addr, addr_size);
7078 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7079 spin_unlock(&nn->client_lock);
7081 nfsd_recall_delegations(&reaplist);
7086 nfsd_inject_recall_delegations(u64 max)
7089 struct nfs4_client *clp, *next;
7090 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7092 LIST_HEAD(reaplist);
7094 if (!nfsd_netns_ready(nn))
7097 spin_lock(&nn->client_lock);
7098 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7099 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7100 if (max != 0 && ++count >= max)
7103 spin_unlock(&nn->client_lock);
7104 nfsd_recall_delegations(&reaplist);
7107 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7110 * Since the lifetime of a delegation isn't limited to that of an open, a
7111 * client may quite reasonably hang on to a delegation as long as it has
7112 * the inode cached. This becomes an obvious problem the first time a
7113 * client's inode cache approaches the size of the server's total memory.
7115 * For now we avoid this problem by imposing a hard limit on the number
7116 * of delegations, which varies according to the server's memory size.
7119 set_max_delegations(void)
7122 * Allow at most 4 delegations per megabyte of RAM. Quick
7123 * estimates suggest that in the worst case (where every delegation
7124 * is for a different inode), a delegation could take about 1.5K,
7125 * giving a worst case usage of about 6% of memory.
7127 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7130 static int nfs4_state_create_net(struct net *net)
7132 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7135 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7136 CLIENT_HASH_SIZE, GFP_KERNEL);
7137 if (!nn->conf_id_hashtbl)
7139 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7140 CLIENT_HASH_SIZE, GFP_KERNEL);
7141 if (!nn->unconf_id_hashtbl)
7143 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
7144 SESSION_HASH_SIZE, GFP_KERNEL);
7145 if (!nn->sessionid_hashtbl)
7148 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7149 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7150 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7152 for (i = 0; i < SESSION_HASH_SIZE; i++)
7153 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7154 nn->conf_name_tree = RB_ROOT;
7155 nn->unconf_name_tree = RB_ROOT;
7156 nn->boot_time = get_seconds();
7157 nn->grace_ended = false;
7158 nn->nfsd4_manager.block_opens = true;
7159 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7160 INIT_LIST_HEAD(&nn->client_lru);
7161 INIT_LIST_HEAD(&nn->close_lru);
7162 INIT_LIST_HEAD(&nn->del_recall_lru);
7163 spin_lock_init(&nn->client_lock);
7165 spin_lock_init(&nn->blocked_locks_lock);
7166 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7168 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7174 kfree(nn->unconf_id_hashtbl);
7176 kfree(nn->conf_id_hashtbl);
7182 nfs4_state_destroy_net(struct net *net)
7185 struct nfs4_client *clp = NULL;
7186 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7188 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7189 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7190 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7191 destroy_client(clp);
7195 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7197 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7198 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7199 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7200 destroy_client(clp);
7204 kfree(nn->sessionid_hashtbl);
7205 kfree(nn->unconf_id_hashtbl);
7206 kfree(nn->conf_id_hashtbl);
7211 nfs4_state_start_net(struct net *net)
7213 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7216 ret = nfs4_state_create_net(net);
7219 locks_start_grace(net, &nn->nfsd4_manager);
7220 nfsd4_client_tracking_init(net);
7221 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7222 nn->nfsd4_grace, net->ns.inum);
7223 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7227 /* initialization to perform when the nfsd service is started: */
7230 nfs4_state_start(void)
7234 ret = set_callback_cred();
7238 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7239 if (laundry_wq == NULL) {
7241 goto out_cleanup_cred;
7243 ret = nfsd4_create_callback_queue();
7245 goto out_free_laundry;
7247 set_max_delegations();
7251 destroy_workqueue(laundry_wq);
7253 cleanup_callback_cred();
7258 nfs4_state_shutdown_net(struct net *net)
7260 struct nfs4_delegation *dp = NULL;
7261 struct list_head *pos, *next, reaplist;
7262 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7264 cancel_delayed_work_sync(&nn->laundromat_work);
7265 locks_end_grace(&nn->nfsd4_manager);
7267 INIT_LIST_HEAD(&reaplist);
7268 spin_lock(&state_lock);
7269 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7270 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7271 WARN_ON(!unhash_delegation_locked(dp));
7272 list_add(&dp->dl_recall_lru, &reaplist);
7274 spin_unlock(&state_lock);
7275 list_for_each_safe(pos, next, &reaplist) {
7276 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7277 list_del_init(&dp->dl_recall_lru);
7278 put_clnt_odstate(dp->dl_clnt_odstate);
7279 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7280 nfs4_put_stid(&dp->dl_stid);
7283 nfsd4_client_tracking_exit(net);
7284 nfs4_state_destroy_net(net);
7288 nfs4_state_shutdown(void)
7290 destroy_workqueue(laundry_wq);
7291 nfsd4_destroy_callback_queue();
7292 cleanup_callback_cred();
7296 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7298 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7299 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7303 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7305 if (cstate->minorversion) {
7306 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7307 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7312 clear_current_stateid(struct nfsd4_compound_state *cstate)
7314 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7318 * functions to set current state id
7321 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7322 union nfsd4_op_u *u)
7324 put_stateid(cstate, &u->open_downgrade.od_stateid);
7328 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7329 union nfsd4_op_u *u)
7331 put_stateid(cstate, &u->open.op_stateid);
7335 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7336 union nfsd4_op_u *u)
7338 put_stateid(cstate, &u->close.cl_stateid);
7342 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7343 union nfsd4_op_u *u)
7345 put_stateid(cstate, &u->lock.lk_resp_stateid);
7349 * functions to consume current state id
7353 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7354 union nfsd4_op_u *u)
7356 get_stateid(cstate, &u->open_downgrade.od_stateid);
7360 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7361 union nfsd4_op_u *u)
7363 get_stateid(cstate, &u->delegreturn.dr_stateid);
7367 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7368 union nfsd4_op_u *u)
7370 get_stateid(cstate, &u->free_stateid.fr_stateid);
7374 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7375 union nfsd4_op_u *u)
7377 get_stateid(cstate, &u->setattr.sa_stateid);
7381 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7382 union nfsd4_op_u *u)
7384 get_stateid(cstate, &u->close.cl_stateid);
7388 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7389 union nfsd4_op_u *u)
7391 get_stateid(cstate, &u->locku.lu_stateid);
7395 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7396 union nfsd4_op_u *u)
7398 get_stateid(cstate, &u->read.rd_stateid);
7402 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7403 union nfsd4_op_u *u)
7405 get_stateid(cstate, &u->write.wr_stateid);