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 *client_slab;
102 static struct kmem_cache *openowner_slab;
103 static struct kmem_cache *lockowner_slab;
104 static struct kmem_cache *file_slab;
105 static struct kmem_cache *stateid_slab;
106 static struct kmem_cache *deleg_slab;
107 static struct kmem_cache *odstate_slab;
109 static void free_session(struct nfsd4_session *);
111 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
112 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
114 static bool is_session_dead(struct nfsd4_session *ses)
116 return ses->se_flags & NFS4_SESSION_DEAD;
119 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
121 if (atomic_read(&ses->se_ref) > ref_held_by_me)
122 return nfserr_jukebox;
123 ses->se_flags |= NFS4_SESSION_DEAD;
127 static bool is_client_expired(struct nfs4_client *clp)
129 return clp->cl_time == 0;
132 static __be32 get_client_locked(struct nfs4_client *clp)
134 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
136 lockdep_assert_held(&nn->client_lock);
138 if (is_client_expired(clp))
139 return nfserr_expired;
140 atomic_inc(&clp->cl_refcount);
144 /* must be called under the client_lock */
146 renew_client_locked(struct nfs4_client *clp)
148 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
150 if (is_client_expired(clp)) {
152 printk("%s: client (clientid %08x/%08x) already expired\n",
154 clp->cl_clientid.cl_boot,
155 clp->cl_clientid.cl_id);
159 dprintk("renewing client (clientid %08x/%08x)\n",
160 clp->cl_clientid.cl_boot,
161 clp->cl_clientid.cl_id);
162 list_move_tail(&clp->cl_lru, &nn->client_lru);
163 clp->cl_time = get_seconds();
166 static void put_client_renew_locked(struct nfs4_client *clp)
168 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
170 lockdep_assert_held(&nn->client_lock);
172 if (!atomic_dec_and_test(&clp->cl_refcount))
174 if (!is_client_expired(clp))
175 renew_client_locked(clp);
178 static void put_client_renew(struct nfs4_client *clp)
180 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
182 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
184 if (!is_client_expired(clp))
185 renew_client_locked(clp);
186 spin_unlock(&nn->client_lock);
189 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
193 if (is_session_dead(ses))
194 return nfserr_badsession;
195 status = get_client_locked(ses->se_client);
198 atomic_inc(&ses->se_ref);
202 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
204 struct nfs4_client *clp = ses->se_client;
205 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207 lockdep_assert_held(&nn->client_lock);
209 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
211 put_client_renew_locked(clp);
214 static void nfsd4_put_session(struct nfsd4_session *ses)
216 struct nfs4_client *clp = ses->se_client;
217 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
219 spin_lock(&nn->client_lock);
220 nfsd4_put_session_locked(ses);
221 spin_unlock(&nn->client_lock);
224 static struct nfsd4_blocked_lock *
225 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
228 struct nfsd4_blocked_lock *cur, *found = NULL;
230 spin_lock(&nn->blocked_locks_lock);
231 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
232 if (fh_match(fh, &cur->nbl_fh)) {
233 list_del_init(&cur->nbl_list);
234 list_del_init(&cur->nbl_lru);
239 spin_unlock(&nn->blocked_locks_lock);
241 posix_unblock_lock(&found->nbl_lock);
245 static struct nfsd4_blocked_lock *
246 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
249 struct nfsd4_blocked_lock *nbl;
251 nbl = find_blocked_lock(lo, fh, nn);
253 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
255 INIT_LIST_HEAD(&nbl->nbl_list);
256 INIT_LIST_HEAD(&nbl->nbl_lru);
257 fh_copy_shallow(&nbl->nbl_fh, fh);
258 locks_init_lock(&nbl->nbl_lock);
259 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
260 &nfsd4_cb_notify_lock_ops,
261 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
268 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
270 locks_release_private(&nbl->nbl_lock);
275 remove_blocked_locks(struct nfs4_lockowner *lo)
277 struct nfs4_client *clp = lo->lo_owner.so_client;
278 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
279 struct nfsd4_blocked_lock *nbl;
282 /* Dequeue all blocked locks */
283 spin_lock(&nn->blocked_locks_lock);
284 while (!list_empty(&lo->lo_blocked)) {
285 nbl = list_first_entry(&lo->lo_blocked,
286 struct nfsd4_blocked_lock,
288 list_del_init(&nbl->nbl_list);
289 list_move(&nbl->nbl_lru, &reaplist);
291 spin_unlock(&nn->blocked_locks_lock);
294 while (!list_empty(&reaplist)) {
295 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
297 list_del_init(&nbl->nbl_lru);
298 posix_unblock_lock(&nbl->nbl_lock);
299 free_blocked_lock(nbl);
304 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
307 * Since this is just an optimization, we don't try very hard if it
308 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
309 * just quit trying on anything else.
311 switch (task->tk_status) {
313 rpc_delay(task, 1 * HZ);
321 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
323 struct nfsd4_blocked_lock *nbl = container_of(cb,
324 struct nfsd4_blocked_lock, nbl_cb);
326 free_blocked_lock(nbl);
329 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
330 .done = nfsd4_cb_notify_lock_done,
331 .release = nfsd4_cb_notify_lock_release,
334 static inline struct nfs4_stateowner *
335 nfs4_get_stateowner(struct nfs4_stateowner *sop)
337 atomic_inc(&sop->so_count);
342 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
344 return (sop->so_owner.len == owner->len) &&
345 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
348 static struct nfs4_openowner *
349 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
350 struct nfs4_client *clp)
352 struct nfs4_stateowner *so;
354 lockdep_assert_held(&clp->cl_lock);
356 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
358 if (!so->so_is_open_owner)
360 if (same_owner_str(so, &open->op_owner))
361 return openowner(nfs4_get_stateowner(so));
366 static struct nfs4_openowner *
367 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
368 struct nfs4_client *clp)
370 struct nfs4_openowner *oo;
372 spin_lock(&clp->cl_lock);
373 oo = find_openstateowner_str_locked(hashval, open, clp);
374 spin_unlock(&clp->cl_lock);
379 opaque_hashval(const void *ptr, int nbytes)
381 unsigned char *cptr = (unsigned char *) ptr;
391 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
393 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
395 kmem_cache_free(file_slab, fp);
399 put_nfs4_file(struct nfs4_file *fi)
401 might_lock(&state_lock);
403 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
404 hlist_del_rcu(&fi->fi_hash);
405 spin_unlock(&state_lock);
406 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
407 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
408 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
413 __nfs4_get_fd(struct nfs4_file *f, int oflag)
415 if (f->fi_fds[oflag])
416 return get_file(f->fi_fds[oflag]);
421 find_writeable_file_locked(struct nfs4_file *f)
425 lockdep_assert_held(&f->fi_lock);
427 ret = __nfs4_get_fd(f, O_WRONLY);
429 ret = __nfs4_get_fd(f, O_RDWR);
434 find_writeable_file(struct nfs4_file *f)
438 spin_lock(&f->fi_lock);
439 ret = find_writeable_file_locked(f);
440 spin_unlock(&f->fi_lock);
445 static struct file *find_readable_file_locked(struct nfs4_file *f)
449 lockdep_assert_held(&f->fi_lock);
451 ret = __nfs4_get_fd(f, O_RDONLY);
453 ret = __nfs4_get_fd(f, O_RDWR);
458 find_readable_file(struct nfs4_file *f)
462 spin_lock(&f->fi_lock);
463 ret = find_readable_file_locked(f);
464 spin_unlock(&f->fi_lock);
470 find_any_file(struct nfs4_file *f)
476 spin_lock(&f->fi_lock);
477 ret = __nfs4_get_fd(f, O_RDWR);
479 ret = __nfs4_get_fd(f, O_WRONLY);
481 ret = __nfs4_get_fd(f, O_RDONLY);
483 spin_unlock(&f->fi_lock);
487 static atomic_long_t num_delegations;
488 unsigned long max_delegations;
491 * Open owner state (share locks)
494 /* hash tables for lock and open owners */
495 #define OWNER_HASH_BITS 8
496 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
497 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
499 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
503 ret = opaque_hashval(ownername->data, ownername->len);
504 return ret & OWNER_HASH_MASK;
507 /* hash table for nfs4_file */
508 #define FILE_HASH_BITS 8
509 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
511 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
513 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
516 static unsigned int file_hashval(struct knfsd_fh *fh)
518 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
521 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
524 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
526 lockdep_assert_held(&fp->fi_lock);
528 if (access & NFS4_SHARE_ACCESS_WRITE)
529 atomic_inc(&fp->fi_access[O_WRONLY]);
530 if (access & NFS4_SHARE_ACCESS_READ)
531 atomic_inc(&fp->fi_access[O_RDONLY]);
535 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
537 lockdep_assert_held(&fp->fi_lock);
539 /* Does this access mode make sense? */
540 if (access & ~NFS4_SHARE_ACCESS_BOTH)
543 /* Does it conflict with a deny mode already set? */
544 if ((access & fp->fi_share_deny) != 0)
545 return nfserr_share_denied;
547 __nfs4_file_get_access(fp, access);
551 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
553 /* Common case is that there is no deny mode. */
555 /* Does this deny mode make sense? */
556 if (deny & ~NFS4_SHARE_DENY_BOTH)
559 if ((deny & NFS4_SHARE_DENY_READ) &&
560 atomic_read(&fp->fi_access[O_RDONLY]))
561 return nfserr_share_denied;
563 if ((deny & NFS4_SHARE_DENY_WRITE) &&
564 atomic_read(&fp->fi_access[O_WRONLY]))
565 return nfserr_share_denied;
570 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
572 might_lock(&fp->fi_lock);
574 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
575 struct file *f1 = NULL;
576 struct file *f2 = NULL;
578 swap(f1, fp->fi_fds[oflag]);
579 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
580 swap(f2, fp->fi_fds[O_RDWR]);
581 spin_unlock(&fp->fi_lock);
589 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
591 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
593 if (access & NFS4_SHARE_ACCESS_WRITE)
594 __nfs4_file_put_access(fp, O_WRONLY);
595 if (access & NFS4_SHARE_ACCESS_READ)
596 __nfs4_file_put_access(fp, O_RDONLY);
600 * Allocate a new open/delegation state counter. This is needed for
601 * pNFS for proper return on close semantics.
603 * Note that we only allocate it for pNFS-enabled exports, otherwise
604 * all pointers to struct nfs4_clnt_odstate are always NULL.
606 static struct nfs4_clnt_odstate *
607 alloc_clnt_odstate(struct nfs4_client *clp)
609 struct nfs4_clnt_odstate *co;
611 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
614 refcount_set(&co->co_odcount, 1);
620 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
622 struct nfs4_file *fp = co->co_file;
624 lockdep_assert_held(&fp->fi_lock);
625 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
629 get_clnt_odstate(struct nfs4_clnt_odstate *co)
632 refcount_inc(&co->co_odcount);
636 put_clnt_odstate(struct nfs4_clnt_odstate *co)
638 struct nfs4_file *fp;
644 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
645 list_del(&co->co_perfile);
646 spin_unlock(&fp->fi_lock);
648 nfsd4_return_all_file_layouts(co->co_client, fp);
649 kmem_cache_free(odstate_slab, co);
653 static struct nfs4_clnt_odstate *
654 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
656 struct nfs4_clnt_odstate *co;
657 struct nfs4_client *cl;
664 spin_lock(&fp->fi_lock);
665 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
666 if (co->co_client == cl) {
667 get_clnt_odstate(co);
673 hash_clnt_odstate_locked(new);
675 spin_unlock(&fp->fi_lock);
679 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
680 void (*sc_free)(struct nfs4_stid *))
682 struct nfs4_stid *stid;
685 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
689 idr_preload(GFP_KERNEL);
690 spin_lock(&cl->cl_lock);
691 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
692 spin_unlock(&cl->cl_lock);
697 stid->sc_free = sc_free;
698 stid->sc_client = cl;
699 stid->sc_stateid.si_opaque.so_id = new_id;
700 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
701 /* Will be incremented before return to client: */
702 refcount_set(&stid->sc_count, 1);
703 spin_lock_init(&stid->sc_lock);
706 * It shouldn't be a problem to reuse an opaque stateid value.
707 * I don't think it is for 4.1. But with 4.0 I worry that, for
708 * example, a stray write retransmission could be accepted by
709 * the server when it should have been rejected. Therefore,
710 * adopt a trick from the sctp code to attempt to maximize the
711 * amount of time until an id is reused, by ensuring they always
712 * "increase" (mod INT_MAX):
716 kmem_cache_free(slab, stid);
720 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
722 struct nfs4_stid *stid;
724 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
728 return openlockstateid(stid);
731 static void nfs4_free_deleg(struct nfs4_stid *stid)
733 kmem_cache_free(deleg_slab, stid);
734 atomic_long_dec(&num_delegations);
738 * When we recall a delegation, we should be careful not to hand it
739 * out again straight away.
740 * To ensure this we keep a pair of bloom filters ('new' and 'old')
741 * in which the filehandles of recalled delegations are "stored".
742 * If a filehandle appear in either filter, a delegation is blocked.
743 * When a delegation is recalled, the filehandle is stored in the "new"
745 * Every 30 seconds we swap the filters and clear the "new" one,
746 * unless both are empty of course.
748 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
749 * low 3 bytes as hash-table indices.
751 * 'blocked_delegations_lock', which is always taken in block_delegations(),
752 * is used to manage concurrent access. Testing does not need the lock
753 * except when swapping the two filters.
755 static DEFINE_SPINLOCK(blocked_delegations_lock);
756 static struct bloom_pair {
757 int entries, old_entries;
759 int new; /* index into 'set' */
760 DECLARE_BITMAP(set[2], 256);
761 } blocked_delegations;
763 static int delegation_blocked(struct knfsd_fh *fh)
766 struct bloom_pair *bd = &blocked_delegations;
768 if (bd->entries == 0)
770 if (seconds_since_boot() - bd->swap_time > 30) {
771 spin_lock(&blocked_delegations_lock);
772 if (seconds_since_boot() - bd->swap_time > 30) {
773 bd->entries -= bd->old_entries;
774 bd->old_entries = bd->entries;
775 memset(bd->set[bd->new], 0,
778 bd->swap_time = seconds_since_boot();
780 spin_unlock(&blocked_delegations_lock);
782 hash = jhash(&fh->fh_base, fh->fh_size, 0);
783 if (test_bit(hash&255, bd->set[0]) &&
784 test_bit((hash>>8)&255, bd->set[0]) &&
785 test_bit((hash>>16)&255, bd->set[0]))
788 if (test_bit(hash&255, bd->set[1]) &&
789 test_bit((hash>>8)&255, bd->set[1]) &&
790 test_bit((hash>>16)&255, bd->set[1]))
796 static void block_delegations(struct knfsd_fh *fh)
799 struct bloom_pair *bd = &blocked_delegations;
801 hash = jhash(&fh->fh_base, fh->fh_size, 0);
803 spin_lock(&blocked_delegations_lock);
804 __set_bit(hash&255, bd->set[bd->new]);
805 __set_bit((hash>>8)&255, bd->set[bd->new]);
806 __set_bit((hash>>16)&255, bd->set[bd->new]);
807 if (bd->entries == 0)
808 bd->swap_time = seconds_since_boot();
810 spin_unlock(&blocked_delegations_lock);
813 static struct nfs4_delegation *
814 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
815 struct svc_fh *current_fh,
816 struct nfs4_clnt_odstate *odstate)
818 struct nfs4_delegation *dp;
821 dprintk("NFSD alloc_init_deleg\n");
822 n = atomic_long_inc_return(&num_delegations);
823 if (n < 0 || n > max_delegations)
825 if (delegation_blocked(¤t_fh->fh_handle))
827 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
832 * delegation seqid's are never incremented. The 4.1 special
833 * meaning of seqid 0 isn't meaningful, really, but let's avoid
834 * 0 anyway just for consistency and use 1:
836 dp->dl_stid.sc_stateid.si_generation = 1;
837 INIT_LIST_HEAD(&dp->dl_perfile);
838 INIT_LIST_HEAD(&dp->dl_perclnt);
839 INIT_LIST_HEAD(&dp->dl_recall_lru);
840 dp->dl_clnt_odstate = odstate;
841 get_clnt_odstate(odstate);
842 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
844 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
845 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
847 dp->dl_stid.sc_file = fp;
850 atomic_long_dec(&num_delegations);
855 nfs4_put_stid(struct nfs4_stid *s)
857 struct nfs4_file *fp = s->sc_file;
858 struct nfs4_client *clp = s->sc_client;
860 might_lock(&clp->cl_lock);
862 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
863 wake_up_all(&close_wq);
866 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
867 spin_unlock(&clp->cl_lock);
874 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
876 stateid_t *src = &stid->sc_stateid;
878 spin_lock(&stid->sc_lock);
879 if (unlikely(++src->si_generation == 0))
880 src->si_generation = 1;
881 memcpy(dst, src, sizeof(*dst));
882 spin_unlock(&stid->sc_lock);
885 static void put_deleg_file(struct nfs4_file *fp)
887 struct file *filp = NULL;
889 spin_lock(&fp->fi_lock);
890 if (--fp->fi_delegees == 0)
891 swap(filp, fp->fi_deleg_file);
892 spin_unlock(&fp->fi_lock);
898 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
900 struct nfs4_file *fp = dp->dl_stid.sc_file;
901 struct file *filp = fp->fi_deleg_file;
903 WARN_ON_ONCE(!fp->fi_delegees);
905 vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
909 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
911 put_clnt_odstate(dp->dl_clnt_odstate);
912 nfs4_unlock_deleg_lease(dp);
913 nfs4_put_stid(&dp->dl_stid);
916 void nfs4_unhash_stid(struct nfs4_stid *s)
922 * nfs4_delegation_exists - Discover if this delegation already exists
923 * @clp: a pointer to the nfs4_client we're granting a delegation to
924 * @fp: a pointer to the nfs4_file we're granting a delegation on
927 * On success: true iff an existing delegation is found
931 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
933 struct nfs4_delegation *searchdp = NULL;
934 struct nfs4_client *searchclp = NULL;
936 lockdep_assert_held(&state_lock);
937 lockdep_assert_held(&fp->fi_lock);
939 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
940 searchclp = searchdp->dl_stid.sc_client;
941 if (clp == searchclp) {
949 * hash_delegation_locked - Add a delegation to the appropriate lists
950 * @dp: a pointer to the nfs4_delegation we are adding.
951 * @fp: a pointer to the nfs4_file we're granting a delegation on
954 * On success: NULL if the delegation was successfully hashed.
956 * On error: -EAGAIN if one was previously granted to this
957 * nfs4_client for this nfs4_file. Delegation is not hashed.
962 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
964 struct nfs4_client *clp = dp->dl_stid.sc_client;
966 lockdep_assert_held(&state_lock);
967 lockdep_assert_held(&fp->fi_lock);
969 if (nfs4_delegation_exists(clp, fp))
971 refcount_inc(&dp->dl_stid.sc_count);
972 dp->dl_stid.sc_type = NFS4_DELEG_STID;
973 list_add(&dp->dl_perfile, &fp->fi_delegations);
974 list_add(&dp->dl_perclnt, &clp->cl_delegations);
978 static bool delegation_hashed(struct nfs4_delegation *dp)
980 return !(list_empty(&dp->dl_perfile));
984 unhash_delegation_locked(struct nfs4_delegation *dp)
986 struct nfs4_file *fp = dp->dl_stid.sc_file;
988 lockdep_assert_held(&state_lock);
990 if (!delegation_hashed(dp))
993 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
994 /* Ensure that deleg break won't try to requeue it */
996 spin_lock(&fp->fi_lock);
997 list_del_init(&dp->dl_perclnt);
998 list_del_init(&dp->dl_recall_lru);
999 list_del_init(&dp->dl_perfile);
1000 spin_unlock(&fp->fi_lock);
1004 static void destroy_delegation(struct nfs4_delegation *dp)
1008 spin_lock(&state_lock);
1009 unhashed = unhash_delegation_locked(dp);
1010 spin_unlock(&state_lock);
1012 destroy_unhashed_deleg(dp);
1015 static void revoke_delegation(struct nfs4_delegation *dp)
1017 struct nfs4_client *clp = dp->dl_stid.sc_client;
1019 WARN_ON(!list_empty(&dp->dl_recall_lru));
1021 if (clp->cl_minorversion) {
1022 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1023 refcount_inc(&dp->dl_stid.sc_count);
1024 spin_lock(&clp->cl_lock);
1025 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1026 spin_unlock(&clp->cl_lock);
1028 destroy_unhashed_deleg(dp);
1035 static unsigned int clientid_hashval(u32 id)
1037 return id & CLIENT_HASH_MASK;
1040 static unsigned int clientstr_hashval(const char *name)
1042 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1046 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1047 * st_{access,deny}_bmap field of the stateid, in order to track not
1048 * only what share bits are currently in force, but also what
1049 * combinations of share bits previous opens have used. This allows us
1050 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1051 * return an error if the client attempt to downgrade to a combination
1052 * of share bits not explicable by closing some of its previous opens.
1054 * XXX: This enforcement is actually incomplete, since we don't keep
1055 * track of access/deny bit combinations; so, e.g., we allow:
1057 * OPEN allow read, deny write
1058 * OPEN allow both, deny none
1059 * DOWNGRADE allow read, deny none
1061 * which we should reject.
1064 bmap_to_share_mode(unsigned long bmap) {
1066 unsigned int access = 0;
1068 for (i = 1; i < 4; i++) {
1069 if (test_bit(i, &bmap))
1075 /* set share access for a given stateid */
1077 set_access(u32 access, struct nfs4_ol_stateid *stp)
1079 unsigned char mask = 1 << access;
1081 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1082 stp->st_access_bmap |= mask;
1085 /* clear share access for a given stateid */
1087 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1089 unsigned char mask = 1 << access;
1091 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1092 stp->st_access_bmap &= ~mask;
1095 /* test whether a given stateid has access */
1097 test_access(u32 access, struct nfs4_ol_stateid *stp)
1099 unsigned char mask = 1 << access;
1101 return (bool)(stp->st_access_bmap & mask);
1104 /* set share deny for a given stateid */
1106 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1108 unsigned char mask = 1 << deny;
1110 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1111 stp->st_deny_bmap |= mask;
1114 /* clear share deny for a given stateid */
1116 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1118 unsigned char mask = 1 << deny;
1120 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1121 stp->st_deny_bmap &= ~mask;
1124 /* test whether a given stateid is denying specific access */
1126 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1128 unsigned char mask = 1 << deny;
1130 return (bool)(stp->st_deny_bmap & mask);
1133 static int nfs4_access_to_omode(u32 access)
1135 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1136 case NFS4_SHARE_ACCESS_READ:
1138 case NFS4_SHARE_ACCESS_WRITE:
1140 case NFS4_SHARE_ACCESS_BOTH:
1148 * A stateid that had a deny mode associated with it is being released
1149 * or downgraded. Recalculate the deny mode on the file.
1152 recalculate_deny_mode(struct nfs4_file *fp)
1154 struct nfs4_ol_stateid *stp;
1156 spin_lock(&fp->fi_lock);
1157 fp->fi_share_deny = 0;
1158 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1159 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1160 spin_unlock(&fp->fi_lock);
1164 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1167 bool change = false;
1169 for (i = 1; i < 4; i++) {
1170 if ((i & deny) != i) {
1176 /* Recalculate per-file deny mode if there was a change */
1178 recalculate_deny_mode(stp->st_stid.sc_file);
1181 /* release all access and file references for a given stateid */
1183 release_all_access(struct nfs4_ol_stateid *stp)
1186 struct nfs4_file *fp = stp->st_stid.sc_file;
1188 if (fp && stp->st_deny_bmap != 0)
1189 recalculate_deny_mode(fp);
1191 for (i = 1; i < 4; i++) {
1192 if (test_access(i, stp))
1193 nfs4_file_put_access(stp->st_stid.sc_file, i);
1194 clear_access(i, stp);
1198 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1200 kfree(sop->so_owner.data);
1201 sop->so_ops->so_free(sop);
1204 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1206 struct nfs4_client *clp = sop->so_client;
1208 might_lock(&clp->cl_lock);
1210 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1212 sop->so_ops->so_unhash(sop);
1213 spin_unlock(&clp->cl_lock);
1214 nfs4_free_stateowner(sop);
1218 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1220 return list_empty(&stp->st_perfile);
1223 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1225 struct nfs4_file *fp = stp->st_stid.sc_file;
1227 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1229 if (list_empty(&stp->st_perfile))
1232 spin_lock(&fp->fi_lock);
1233 list_del_init(&stp->st_perfile);
1234 spin_unlock(&fp->fi_lock);
1235 list_del(&stp->st_perstateowner);
1239 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1241 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1243 put_clnt_odstate(stp->st_clnt_odstate);
1244 release_all_access(stp);
1245 if (stp->st_stateowner)
1246 nfs4_put_stateowner(stp->st_stateowner);
1247 kmem_cache_free(stateid_slab, stid);
1250 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1252 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1253 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1256 file = find_any_file(stp->st_stid.sc_file);
1258 filp_close(file, (fl_owner_t)lo);
1259 nfs4_free_ol_stateid(stid);
1263 * Put the persistent reference to an already unhashed generic stateid, while
1264 * holding the cl_lock. If it's the last reference, then put it onto the
1265 * reaplist for later destruction.
1267 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1268 struct list_head *reaplist)
1270 struct nfs4_stid *s = &stp->st_stid;
1271 struct nfs4_client *clp = s->sc_client;
1273 lockdep_assert_held(&clp->cl_lock);
1275 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1277 if (!refcount_dec_and_test(&s->sc_count)) {
1278 wake_up_all(&close_wq);
1282 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1283 list_add(&stp->st_locks, reaplist);
1286 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1288 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1290 if (!unhash_ol_stateid(stp))
1292 list_del_init(&stp->st_locks);
1293 nfs4_unhash_stid(&stp->st_stid);
1297 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1299 struct nfs4_client *clp = stp->st_stid.sc_client;
1302 spin_lock(&clp->cl_lock);
1303 unhashed = unhash_lock_stateid(stp);
1304 spin_unlock(&clp->cl_lock);
1306 nfs4_put_stid(&stp->st_stid);
1309 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1311 struct nfs4_client *clp = lo->lo_owner.so_client;
1313 lockdep_assert_held(&clp->cl_lock);
1315 list_del_init(&lo->lo_owner.so_strhash);
1319 * Free a list of generic stateids that were collected earlier after being
1323 free_ol_stateid_reaplist(struct list_head *reaplist)
1325 struct nfs4_ol_stateid *stp;
1326 struct nfs4_file *fp;
1330 while (!list_empty(reaplist)) {
1331 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1333 list_del(&stp->st_locks);
1334 fp = stp->st_stid.sc_file;
1335 stp->st_stid.sc_free(&stp->st_stid);
1341 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1342 struct list_head *reaplist)
1344 struct nfs4_ol_stateid *stp;
1346 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1348 while (!list_empty(&open_stp->st_locks)) {
1349 stp = list_entry(open_stp->st_locks.next,
1350 struct nfs4_ol_stateid, st_locks);
1351 WARN_ON(!unhash_lock_stateid(stp));
1352 put_ol_stateid_locked(stp, reaplist);
1356 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1357 struct list_head *reaplist)
1359 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1361 if (!unhash_ol_stateid(stp))
1363 release_open_stateid_locks(stp, reaplist);
1367 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1369 LIST_HEAD(reaplist);
1371 spin_lock(&stp->st_stid.sc_client->cl_lock);
1372 if (unhash_open_stateid(stp, &reaplist))
1373 put_ol_stateid_locked(stp, &reaplist);
1374 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1375 free_ol_stateid_reaplist(&reaplist);
1378 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1380 struct nfs4_client *clp = oo->oo_owner.so_client;
1382 lockdep_assert_held(&clp->cl_lock);
1384 list_del_init(&oo->oo_owner.so_strhash);
1385 list_del_init(&oo->oo_perclient);
1388 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1390 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1392 struct nfs4_ol_stateid *s;
1394 spin_lock(&nn->client_lock);
1395 s = oo->oo_last_closed_stid;
1397 list_del_init(&oo->oo_close_lru);
1398 oo->oo_last_closed_stid = NULL;
1400 spin_unlock(&nn->client_lock);
1402 nfs4_put_stid(&s->st_stid);
1405 static void release_openowner(struct nfs4_openowner *oo)
1407 struct nfs4_ol_stateid *stp;
1408 struct nfs4_client *clp = oo->oo_owner.so_client;
1409 struct list_head reaplist;
1411 INIT_LIST_HEAD(&reaplist);
1413 spin_lock(&clp->cl_lock);
1414 unhash_openowner_locked(oo);
1415 while (!list_empty(&oo->oo_owner.so_stateids)) {
1416 stp = list_first_entry(&oo->oo_owner.so_stateids,
1417 struct nfs4_ol_stateid, st_perstateowner);
1418 if (unhash_open_stateid(stp, &reaplist))
1419 put_ol_stateid_locked(stp, &reaplist);
1421 spin_unlock(&clp->cl_lock);
1422 free_ol_stateid_reaplist(&reaplist);
1423 release_last_closed_stateid(oo);
1424 nfs4_put_stateowner(&oo->oo_owner);
1428 hash_sessionid(struct nfs4_sessionid *sessionid)
1430 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1432 return sid->sequence % SESSION_HASH_SIZE;
1435 #ifdef CONFIG_SUNRPC_DEBUG
1437 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1439 u32 *ptr = (u32 *)(&sessionid->data[0]);
1440 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1444 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1450 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1451 * won't be used for replay.
1453 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1455 struct nfs4_stateowner *so = cstate->replay_owner;
1457 if (nfserr == nfserr_replay_me)
1460 if (!seqid_mutating_err(ntohl(nfserr))) {
1461 nfsd4_cstate_clear_replay(cstate);
1466 if (so->so_is_open_owner)
1467 release_last_closed_stateid(openowner(so));
1473 gen_sessionid(struct nfsd4_session *ses)
1475 struct nfs4_client *clp = ses->se_client;
1476 struct nfsd4_sessionid *sid;
1478 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1479 sid->clientid = clp->cl_clientid;
1480 sid->sequence = current_sessionid++;
1485 * The protocol defines ca_maxresponssize_cached to include the size of
1486 * the rpc header, but all we need to cache is the data starting after
1487 * the end of the initial SEQUENCE operation--the rest we regenerate
1488 * each time. Therefore we can advertise a ca_maxresponssize_cached
1489 * value that is the number of bytes in our cache plus a few additional
1490 * bytes. In order to stay on the safe side, and not promise more than
1491 * we can cache, those additional bytes must be the minimum possible: 24
1492 * bytes of rpc header (xid through accept state, with AUTH_NULL
1493 * verifier), 12 for the compound header (with zero-length tag), and 44
1494 * for the SEQUENCE op response:
1496 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1499 free_session_slots(struct nfsd4_session *ses)
1503 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1504 free_svc_cred(&ses->se_slots[i]->sl_cred);
1505 kfree(ses->se_slots[i]);
1510 * We don't actually need to cache the rpc and session headers, so we
1511 * can allocate a little less for each slot:
1513 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1517 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1520 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1521 return size + sizeof(struct nfsd4_slot);
1525 * XXX: If we run out of reserved DRC memory we could (up to a point)
1526 * re-negotiate active sessions and reduce their slot usage to make
1527 * room for new connections. For now we just fail the create session.
1529 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1531 u32 slotsize = slot_bytes(ca);
1532 u32 num = ca->maxreqs;
1533 unsigned long avail, total_avail;
1535 spin_lock(&nfsd_drc_lock);
1536 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1537 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1539 * Never use more than a third of the remaining memory,
1540 * unless it's the only way to give this client a slot:
1542 avail = clamp_t(unsigned long, avail, slotsize, total_avail/3);
1543 num = min_t(int, num, avail / slotsize);
1544 nfsd_drc_mem_used += num * slotsize;
1545 spin_unlock(&nfsd_drc_lock);
1550 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1552 int slotsize = slot_bytes(ca);
1554 spin_lock(&nfsd_drc_lock);
1555 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1556 spin_unlock(&nfsd_drc_lock);
1559 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1560 struct nfsd4_channel_attrs *battrs)
1562 int numslots = fattrs->maxreqs;
1563 int slotsize = slot_bytes(fattrs);
1564 struct nfsd4_session *new;
1567 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1568 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1569 mem = numslots * sizeof(struct nfsd4_slot *);
1571 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1574 /* allocate each struct nfsd4_slot and data cache in one piece */
1575 for (i = 0; i < numslots; i++) {
1576 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1577 if (!new->se_slots[i])
1581 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1582 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1587 kfree(new->se_slots[i]);
1592 static void free_conn(struct nfsd4_conn *c)
1594 svc_xprt_put(c->cn_xprt);
1598 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1600 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1601 struct nfs4_client *clp = c->cn_session->se_client;
1603 spin_lock(&clp->cl_lock);
1604 if (!list_empty(&c->cn_persession)) {
1605 list_del(&c->cn_persession);
1608 nfsd4_probe_callback(clp);
1609 spin_unlock(&clp->cl_lock);
1612 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1614 struct nfsd4_conn *conn;
1616 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1619 svc_xprt_get(rqstp->rq_xprt);
1620 conn->cn_xprt = rqstp->rq_xprt;
1621 conn->cn_flags = flags;
1622 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1626 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1628 conn->cn_session = ses;
1629 list_add(&conn->cn_persession, &ses->se_conns);
1632 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1634 struct nfs4_client *clp = ses->se_client;
1636 spin_lock(&clp->cl_lock);
1637 __nfsd4_hash_conn(conn, ses);
1638 spin_unlock(&clp->cl_lock);
1641 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1643 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1644 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1647 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1651 nfsd4_hash_conn(conn, ses);
1652 ret = nfsd4_register_conn(conn);
1654 /* oops; xprt is already down: */
1655 nfsd4_conn_lost(&conn->cn_xpt_user);
1656 /* We may have gained or lost a callback channel: */
1657 nfsd4_probe_callback_sync(ses->se_client);
1660 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1662 u32 dir = NFS4_CDFC4_FORE;
1664 if (cses->flags & SESSION4_BACK_CHAN)
1665 dir |= NFS4_CDFC4_BACK;
1666 return alloc_conn(rqstp, dir);
1669 /* must be called under client_lock */
1670 static void nfsd4_del_conns(struct nfsd4_session *s)
1672 struct nfs4_client *clp = s->se_client;
1673 struct nfsd4_conn *c;
1675 spin_lock(&clp->cl_lock);
1676 while (!list_empty(&s->se_conns)) {
1677 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1678 list_del_init(&c->cn_persession);
1679 spin_unlock(&clp->cl_lock);
1681 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1684 spin_lock(&clp->cl_lock);
1686 spin_unlock(&clp->cl_lock);
1689 static void __free_session(struct nfsd4_session *ses)
1691 free_session_slots(ses);
1695 static void free_session(struct nfsd4_session *ses)
1697 nfsd4_del_conns(ses);
1698 nfsd4_put_drc_mem(&ses->se_fchannel);
1699 __free_session(ses);
1702 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1705 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1707 new->se_client = clp;
1710 INIT_LIST_HEAD(&new->se_conns);
1712 new->se_cb_seq_nr = 1;
1713 new->se_flags = cses->flags;
1714 new->se_cb_prog = cses->callback_prog;
1715 new->se_cb_sec = cses->cb_sec;
1716 atomic_set(&new->se_ref, 0);
1717 idx = hash_sessionid(&new->se_sessionid);
1718 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1719 spin_lock(&clp->cl_lock);
1720 list_add(&new->se_perclnt, &clp->cl_sessions);
1721 spin_unlock(&clp->cl_lock);
1724 struct sockaddr *sa = svc_addr(rqstp);
1726 * This is a little silly; with sessions there's no real
1727 * use for the callback address. Use the peer address
1728 * as a reasonable default for now, but consider fixing
1729 * the rpc client not to require an address in the
1732 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1733 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1737 /* caller must hold client_lock */
1738 static struct nfsd4_session *
1739 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1741 struct nfsd4_session *elem;
1743 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1745 lockdep_assert_held(&nn->client_lock);
1747 dump_sessionid(__func__, sessionid);
1748 idx = hash_sessionid(sessionid);
1749 /* Search in the appropriate list */
1750 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1751 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1752 NFS4_MAX_SESSIONID_LEN)) {
1757 dprintk("%s: session not found\n", __func__);
1761 static struct nfsd4_session *
1762 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1765 struct nfsd4_session *session;
1766 __be32 status = nfserr_badsession;
1768 session = __find_in_sessionid_hashtbl(sessionid, net);
1771 status = nfsd4_get_session_locked(session);
1779 /* caller must hold client_lock */
1781 unhash_session(struct nfsd4_session *ses)
1783 struct nfs4_client *clp = ses->se_client;
1784 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1786 lockdep_assert_held(&nn->client_lock);
1788 list_del(&ses->se_hash);
1789 spin_lock(&ses->se_client->cl_lock);
1790 list_del(&ses->se_perclnt);
1791 spin_unlock(&ses->se_client->cl_lock);
1794 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1796 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1799 * We're assuming the clid was not given out from a boot
1800 * precisely 2^32 (about 136 years) before this one. That seems
1801 * a safe assumption:
1803 if (clid->cl_boot == (u32)nn->boot_time)
1805 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1806 clid->cl_boot, clid->cl_id, nn->boot_time);
1811 * XXX Should we use a slab cache ?
1812 * This type of memory management is somewhat inefficient, but we use it
1813 * anyway since SETCLIENTID is not a common operation.
1815 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1817 struct nfs4_client *clp;
1820 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1823 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1824 if (clp->cl_name.data == NULL)
1826 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1827 sizeof(struct list_head),
1829 if (!clp->cl_ownerstr_hashtbl)
1830 goto err_no_hashtbl;
1831 for (i = 0; i < OWNER_HASH_SIZE; i++)
1832 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1833 clp->cl_name.len = name.len;
1834 INIT_LIST_HEAD(&clp->cl_sessions);
1835 idr_init(&clp->cl_stateids);
1836 atomic_set(&clp->cl_refcount, 0);
1837 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1838 INIT_LIST_HEAD(&clp->cl_idhash);
1839 INIT_LIST_HEAD(&clp->cl_openowners);
1840 INIT_LIST_HEAD(&clp->cl_delegations);
1841 INIT_LIST_HEAD(&clp->cl_lru);
1842 INIT_LIST_HEAD(&clp->cl_revoked);
1843 #ifdef CONFIG_NFSD_PNFS
1844 INIT_LIST_HEAD(&clp->cl_lo_states);
1846 spin_lock_init(&clp->cl_lock);
1847 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1850 kfree(clp->cl_name.data);
1852 kmem_cache_free(client_slab, clp);
1857 free_client(struct nfs4_client *clp)
1859 while (!list_empty(&clp->cl_sessions)) {
1860 struct nfsd4_session *ses;
1861 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1863 list_del(&ses->se_perclnt);
1864 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1867 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1868 free_svc_cred(&clp->cl_cred);
1869 kfree(clp->cl_ownerstr_hashtbl);
1870 kfree(clp->cl_name.data);
1871 idr_destroy(&clp->cl_stateids);
1872 kmem_cache_free(client_slab, clp);
1875 /* must be called under the client_lock */
1877 unhash_client_locked(struct nfs4_client *clp)
1879 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1880 struct nfsd4_session *ses;
1882 lockdep_assert_held(&nn->client_lock);
1884 /* Mark the client as expired! */
1886 /* Make it invisible */
1887 if (!list_empty(&clp->cl_idhash)) {
1888 list_del_init(&clp->cl_idhash);
1889 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1890 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1892 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1894 list_del_init(&clp->cl_lru);
1895 spin_lock(&clp->cl_lock);
1896 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1897 list_del_init(&ses->se_hash);
1898 spin_unlock(&clp->cl_lock);
1902 unhash_client(struct nfs4_client *clp)
1904 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1906 spin_lock(&nn->client_lock);
1907 unhash_client_locked(clp);
1908 spin_unlock(&nn->client_lock);
1911 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1913 if (atomic_read(&clp->cl_refcount))
1914 return nfserr_jukebox;
1915 unhash_client_locked(clp);
1920 __destroy_client(struct nfs4_client *clp)
1923 struct nfs4_openowner *oo;
1924 struct nfs4_delegation *dp;
1925 struct list_head reaplist;
1927 INIT_LIST_HEAD(&reaplist);
1928 spin_lock(&state_lock);
1929 while (!list_empty(&clp->cl_delegations)) {
1930 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1931 WARN_ON(!unhash_delegation_locked(dp));
1932 list_add(&dp->dl_recall_lru, &reaplist);
1934 spin_unlock(&state_lock);
1935 while (!list_empty(&reaplist)) {
1936 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1937 list_del_init(&dp->dl_recall_lru);
1938 destroy_unhashed_deleg(dp);
1940 while (!list_empty(&clp->cl_revoked)) {
1941 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1942 list_del_init(&dp->dl_recall_lru);
1943 nfs4_put_stid(&dp->dl_stid);
1945 while (!list_empty(&clp->cl_openowners)) {
1946 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1947 nfs4_get_stateowner(&oo->oo_owner);
1948 release_openowner(oo);
1950 for (i = 0; i < OWNER_HASH_SIZE; i++) {
1951 struct nfs4_stateowner *so, *tmp;
1953 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1955 /* Should be no openowners at this point */
1956 WARN_ON_ONCE(so->so_is_open_owner);
1957 remove_blocked_locks(lockowner(so));
1960 nfsd4_return_all_client_layouts(clp);
1961 nfsd4_shutdown_callback(clp);
1962 if (clp->cl_cb_conn.cb_xprt)
1963 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1968 destroy_client(struct nfs4_client *clp)
1971 __destroy_client(clp);
1974 static void expire_client(struct nfs4_client *clp)
1977 nfsd4_client_record_remove(clp);
1978 __destroy_client(clp);
1981 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1983 memcpy(target->cl_verifier.data, source->data,
1984 sizeof(target->cl_verifier.data));
1987 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1989 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1990 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1993 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1995 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1996 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1998 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
1999 if ((source->cr_principal && !target->cr_principal) ||
2000 (source->cr_raw_principal && !target->cr_raw_principal) ||
2001 (source->cr_targ_princ && !target->cr_targ_princ))
2004 target->cr_flavor = source->cr_flavor;
2005 target->cr_uid = source->cr_uid;
2006 target->cr_gid = source->cr_gid;
2007 target->cr_group_info = source->cr_group_info;
2008 get_group_info(target->cr_group_info);
2009 target->cr_gss_mech = source->cr_gss_mech;
2010 if (source->cr_gss_mech)
2011 gss_mech_get(source->cr_gss_mech);
2016 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2018 if (o1->len < o2->len)
2020 if (o1->len > o2->len)
2022 return memcmp(o1->data, o2->data, o1->len);
2025 static int same_name(const char *n1, const char *n2)
2027 return 0 == memcmp(n1, n2, HEXDIR_LEN);
2031 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2033 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2037 same_clid(clientid_t *cl1, clientid_t *cl2)
2039 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2042 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2046 if (g1->ngroups != g2->ngroups)
2048 for (i=0; i<g1->ngroups; i++)
2049 if (!gid_eq(g1->gid[i], g2->gid[i]))
2055 * RFC 3530 language requires clid_inuse be returned when the
2056 * "principal" associated with a requests differs from that previously
2057 * used. We use uid, gid's, and gss principal string as our best
2058 * approximation. We also don't want to allow non-gss use of a client
2059 * established using gss: in theory cr_principal should catch that
2060 * change, but in practice cr_principal can be null even in the gss case
2061 * since gssd doesn't always pass down a principal string.
2063 static bool is_gss_cred(struct svc_cred *cr)
2065 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2066 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2071 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2073 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2074 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2075 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2076 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2078 /* XXX: check that cr_targ_princ fields match ? */
2079 if (cr1->cr_principal == cr2->cr_principal)
2081 if (!cr1->cr_principal || !cr2->cr_principal)
2083 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2086 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2088 struct svc_cred *cr = &rqstp->rq_cred;
2091 if (!cr->cr_gss_mech)
2093 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2094 return service == RPC_GSS_SVC_INTEGRITY ||
2095 service == RPC_GSS_SVC_PRIVACY;
2098 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2100 struct svc_cred *cr = &rqstp->rq_cred;
2102 if (!cl->cl_mach_cred)
2104 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2106 if (!svc_rqst_integrity_protected(rqstp))
2108 if (cl->cl_cred.cr_raw_principal)
2109 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2110 cr->cr_raw_principal);
2111 if (!cr->cr_principal)
2113 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2116 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2121 * This is opaque to client, so no need to byte-swap. Use
2122 * __force to keep sparse happy
2124 verf[0] = (__force __be32)get_seconds();
2125 verf[1] = (__force __be32)nn->clverifier_counter++;
2126 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2129 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2131 clp->cl_clientid.cl_boot = nn->boot_time;
2132 clp->cl_clientid.cl_id = nn->clientid_counter++;
2133 gen_confirm(clp, nn);
2136 static struct nfs4_stid *
2137 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2139 struct nfs4_stid *ret;
2141 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2142 if (!ret || !ret->sc_type)
2147 static struct nfs4_stid *
2148 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2150 struct nfs4_stid *s;
2152 spin_lock(&cl->cl_lock);
2153 s = find_stateid_locked(cl, t);
2155 if (typemask & s->sc_type)
2156 refcount_inc(&s->sc_count);
2160 spin_unlock(&cl->cl_lock);
2164 static struct nfs4_client *create_client(struct xdr_netobj name,
2165 struct svc_rqst *rqstp, nfs4_verifier *verf)
2167 struct nfs4_client *clp;
2168 struct sockaddr *sa = svc_addr(rqstp);
2170 struct net *net = SVC_NET(rqstp);
2172 clp = alloc_client(name);
2176 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2181 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2182 clp->cl_time = get_seconds();
2183 clear_bit(0, &clp->cl_cb_slot_busy);
2184 copy_verf(clp, verf);
2185 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2186 clp->cl_cb_session = NULL;
2192 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2194 struct rb_node **new = &(root->rb_node), *parent = NULL;
2195 struct nfs4_client *clp;
2198 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2201 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2202 new = &((*new)->rb_left);
2204 new = &((*new)->rb_right);
2207 rb_link_node(&new_clp->cl_namenode, parent, new);
2208 rb_insert_color(&new_clp->cl_namenode, root);
2211 static struct nfs4_client *
2212 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2215 struct rb_node *node = root->rb_node;
2216 struct nfs4_client *clp;
2219 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2220 cmp = compare_blob(&clp->cl_name, name);
2222 node = node->rb_left;
2224 node = node->rb_right;
2232 add_to_unconfirmed(struct nfs4_client *clp)
2234 unsigned int idhashval;
2235 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2237 lockdep_assert_held(&nn->client_lock);
2239 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2240 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2241 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2242 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2243 renew_client_locked(clp);
2247 move_to_confirmed(struct nfs4_client *clp)
2249 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2250 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2252 lockdep_assert_held(&nn->client_lock);
2254 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2255 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2256 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2257 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2258 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2259 renew_client_locked(clp);
2262 static struct nfs4_client *
2263 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2265 struct nfs4_client *clp;
2266 unsigned int idhashval = clientid_hashval(clid->cl_id);
2268 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2269 if (same_clid(&clp->cl_clientid, clid)) {
2270 if ((bool)clp->cl_minorversion != sessions)
2272 renew_client_locked(clp);
2279 static struct nfs4_client *
2280 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2282 struct list_head *tbl = nn->conf_id_hashtbl;
2284 lockdep_assert_held(&nn->client_lock);
2285 return find_client_in_id_table(tbl, clid, sessions);
2288 static struct nfs4_client *
2289 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2291 struct list_head *tbl = nn->unconf_id_hashtbl;
2293 lockdep_assert_held(&nn->client_lock);
2294 return find_client_in_id_table(tbl, clid, sessions);
2297 static bool clp_used_exchangeid(struct nfs4_client *clp)
2299 return clp->cl_exchange_flags != 0;
2302 static struct nfs4_client *
2303 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2305 lockdep_assert_held(&nn->client_lock);
2306 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2309 static struct nfs4_client *
2310 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2312 lockdep_assert_held(&nn->client_lock);
2313 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2317 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2319 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2320 struct sockaddr *sa = svc_addr(rqstp);
2321 u32 scopeid = rpc_get_scope_id(sa);
2322 unsigned short expected_family;
2324 /* Currently, we only support tcp and tcp6 for the callback channel */
2325 if (se->se_callback_netid_len == 3 &&
2326 !memcmp(se->se_callback_netid_val, "tcp", 3))
2327 expected_family = AF_INET;
2328 else if (se->se_callback_netid_len == 4 &&
2329 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2330 expected_family = AF_INET6;
2334 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2335 se->se_callback_addr_len,
2336 (struct sockaddr *)&conn->cb_addr,
2337 sizeof(conn->cb_addr));
2339 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2342 if (conn->cb_addr.ss_family == AF_INET6)
2343 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2345 conn->cb_prog = se->se_callback_prog;
2346 conn->cb_ident = se->se_callback_ident;
2347 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2350 conn->cb_addr.ss_family = AF_UNSPEC;
2351 conn->cb_addrlen = 0;
2352 dprintk("NFSD: this client (clientid %08x/%08x) "
2353 "will not receive delegations\n",
2354 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2360 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2363 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2365 struct xdr_buf *buf = resp->xdr.buf;
2366 struct nfsd4_slot *slot = resp->cstate.slot;
2369 dprintk("--> %s slot %p\n", __func__, slot);
2371 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2372 slot->sl_opcnt = resp->opcnt;
2373 slot->sl_status = resp->cstate.status;
2374 free_svc_cred(&slot->sl_cred);
2375 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2377 if (!nfsd4_cache_this(resp)) {
2378 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2381 slot->sl_flags |= NFSD4_SLOT_CACHED;
2383 base = resp->cstate.data_offset;
2384 slot->sl_datalen = buf->len - base;
2385 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2386 WARN(1, "%s: sessions DRC could not cache compound\n",
2392 * Encode the replay sequence operation from the slot values.
2393 * If cachethis is FALSE encode the uncached rep error on the next
2394 * operation which sets resp->p and increments resp->opcnt for
2395 * nfs4svc_encode_compoundres.
2399 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2400 struct nfsd4_compoundres *resp)
2402 struct nfsd4_op *op;
2403 struct nfsd4_slot *slot = resp->cstate.slot;
2405 /* Encode the replayed sequence operation */
2406 op = &args->ops[resp->opcnt - 1];
2407 nfsd4_encode_operation(resp, op);
2409 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2411 if (args->opcnt == 1) {
2413 * The original operation wasn't a solo sequence--we
2414 * always cache those--so this retry must not match the
2417 op->status = nfserr_seq_false_retry;
2419 op = &args->ops[resp->opcnt++];
2420 op->status = nfserr_retry_uncached_rep;
2421 nfsd4_encode_operation(resp, op);
2427 * The sequence operation is not cached because we can use the slot and
2431 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2432 struct nfsd4_sequence *seq)
2434 struct nfsd4_slot *slot = resp->cstate.slot;
2435 struct xdr_stream *xdr = &resp->xdr;
2439 dprintk("--> %s slot %p\n", __func__, slot);
2441 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2445 p = xdr_reserve_space(xdr, slot->sl_datalen);
2448 return nfserr_serverfault;
2450 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2451 xdr_commit_encode(xdr);
2453 resp->opcnt = slot->sl_opcnt;
2454 return slot->sl_status;
2458 * Set the exchange_id flags returned by the server.
2461 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2463 #ifdef CONFIG_NFSD_PNFS
2464 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2466 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2469 /* Referrals are supported, Migration is not. */
2470 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2472 /* set the wire flags to return to client. */
2473 clid->flags = new->cl_exchange_flags;
2476 static bool client_has_openowners(struct nfs4_client *clp)
2478 struct nfs4_openowner *oo;
2480 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2481 if (!list_empty(&oo->oo_owner.so_stateids))
2487 static bool client_has_state(struct nfs4_client *clp)
2489 return client_has_openowners(clp)
2490 #ifdef CONFIG_NFSD_PNFS
2491 || !list_empty(&clp->cl_lo_states)
2493 || !list_empty(&clp->cl_delegations)
2494 || !list_empty(&clp->cl_sessions);
2498 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2499 union nfsd4_op_u *u)
2501 struct nfsd4_exchange_id *exid = &u->exchange_id;
2502 struct nfs4_client *conf, *new;
2503 struct nfs4_client *unconf = NULL;
2505 char addr_str[INET6_ADDRSTRLEN];
2506 nfs4_verifier verf = exid->verifier;
2507 struct sockaddr *sa = svc_addr(rqstp);
2508 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2509 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2511 rpc_ntop(sa, addr_str, sizeof(addr_str));
2512 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2513 "ip_addr=%s flags %x, spa_how %d\n",
2514 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2515 addr_str, exid->flags, exid->spa_how);
2517 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2518 return nfserr_inval;
2520 new = create_client(exid->clname, rqstp, &verf);
2522 return nfserr_jukebox;
2524 switch (exid->spa_how) {
2526 exid->spo_must_enforce[0] = 0;
2527 exid->spo_must_enforce[1] = (
2528 1 << (OP_BIND_CONN_TO_SESSION - 32) |
2529 1 << (OP_EXCHANGE_ID - 32) |
2530 1 << (OP_CREATE_SESSION - 32) |
2531 1 << (OP_DESTROY_SESSION - 32) |
2532 1 << (OP_DESTROY_CLIENTID - 32));
2534 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2535 1 << (OP_OPEN_DOWNGRADE) |
2537 1 << (OP_DELEGRETURN));
2539 exid->spo_must_allow[1] &= (
2540 1 << (OP_TEST_STATEID - 32) |
2541 1 << (OP_FREE_STATEID - 32));
2542 if (!svc_rqst_integrity_protected(rqstp)) {
2543 status = nfserr_inval;
2547 * Sometimes userspace doesn't give us a principal.
2548 * Which is a bug, really. Anyway, we can't enforce
2549 * MACH_CRED in that case, better to give up now:
2551 if (!new->cl_cred.cr_principal &&
2552 !new->cl_cred.cr_raw_principal) {
2553 status = nfserr_serverfault;
2556 new->cl_mach_cred = true;
2559 default: /* checked by xdr code */
2562 status = nfserr_encr_alg_unsupp;
2566 /* Cases below refer to rfc 5661 section 18.35.4: */
2567 spin_lock(&nn->client_lock);
2568 conf = find_confirmed_client_by_name(&exid->clname, nn);
2570 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2571 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2574 if (!clp_used_exchangeid(conf)) { /* buggy client */
2575 status = nfserr_inval;
2578 if (!nfsd4_mach_creds_match(conf, rqstp)) {
2579 status = nfserr_wrong_cred;
2582 if (!creds_match) { /* case 9 */
2583 status = nfserr_perm;
2586 if (!verfs_match) { /* case 8 */
2587 status = nfserr_not_same;
2591 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2594 if (!creds_match) { /* case 3 */
2595 if (client_has_state(conf)) {
2596 status = nfserr_clid_inuse;
2601 if (verfs_match) { /* case 2 */
2602 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2605 /* case 5, client reboot */
2610 if (update) { /* case 7 */
2611 status = nfserr_noent;
2615 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2616 if (unconf) /* case 4, possible retry or client restart */
2617 unhash_client_locked(unconf);
2619 /* case 1 (normal case) */
2622 status = mark_client_expired_locked(conf);
2626 new->cl_minorversion = cstate->minorversion;
2627 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2628 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2631 add_to_unconfirmed(new);
2634 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2635 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2637 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2638 nfsd4_set_ex_flags(conf, exid);
2640 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2641 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2645 spin_unlock(&nn->client_lock);
2650 expire_client(unconf);
2655 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2657 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2660 /* The slot is in use, and no response has been sent. */
2662 if (seqid == slot_seqid)
2663 return nfserr_jukebox;
2665 return nfserr_seq_misordered;
2667 /* Note unsigned 32-bit arithmetic handles wraparound: */
2668 if (likely(seqid == slot_seqid + 1))
2670 if (seqid == slot_seqid)
2671 return nfserr_replay_cache;
2672 return nfserr_seq_misordered;
2676 * Cache the create session result into the create session single DRC
2677 * slot cache by saving the xdr structure. sl_seqid has been set.
2678 * Do this for solo or embedded create session operations.
2681 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2682 struct nfsd4_clid_slot *slot, __be32 nfserr)
2684 slot->sl_status = nfserr;
2685 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2689 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2690 struct nfsd4_clid_slot *slot)
2692 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2693 return slot->sl_status;
2696 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2697 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2698 1 + /* MIN tag is length with zero, only length */ \
2699 3 + /* version, opcount, opcode */ \
2700 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2701 /* seqid, slotID, slotID, cache */ \
2702 4 ) * sizeof(__be32))
2704 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2705 2 + /* verifier: AUTH_NULL, length 0 */\
2707 1 + /* MIN tag is length with zero, only length */ \
2708 3 + /* opcount, opcode, opstatus*/ \
2709 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2710 /* seqid, slotID, slotID, slotID, status */ \
2711 5 ) * sizeof(__be32))
2713 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2715 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2717 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2718 return nfserr_toosmall;
2719 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2720 return nfserr_toosmall;
2721 ca->headerpadsz = 0;
2722 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2723 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2724 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2725 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2726 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2727 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2729 * Note decreasing slot size below client's request may make it
2730 * difficult for client to function correctly, whereas
2731 * decreasing the number of slots will (just?) affect
2732 * performance. When short on memory we therefore prefer to
2733 * decrease number of slots instead of their size. Clients that
2734 * request larger slots than they need will get poor results:
2736 ca->maxreqs = nfsd4_get_drc_mem(ca);
2738 return nfserr_jukebox;
2744 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2745 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2747 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2748 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2750 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2751 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2753 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2754 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2755 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2756 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2759 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2761 ca->headerpadsz = 0;
2763 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2764 return nfserr_toosmall;
2765 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2766 return nfserr_toosmall;
2767 ca->maxresp_cached = 0;
2769 return nfserr_toosmall;
2774 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2776 switch (cbs->flavor) {
2782 * GSS case: the spec doesn't allow us to return this
2783 * error. But it also doesn't allow us not to support
2785 * I'd rather this fail hard than return some error the
2786 * client might think it can already handle:
2788 return nfserr_encr_alg_unsupp;
2793 nfsd4_create_session(struct svc_rqst *rqstp,
2794 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2796 struct nfsd4_create_session *cr_ses = &u->create_session;
2797 struct sockaddr *sa = svc_addr(rqstp);
2798 struct nfs4_client *conf, *unconf;
2799 struct nfs4_client *old = NULL;
2800 struct nfsd4_session *new;
2801 struct nfsd4_conn *conn;
2802 struct nfsd4_clid_slot *cs_slot = NULL;
2804 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2806 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2807 return nfserr_inval;
2808 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2811 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2814 status = check_backchannel_attrs(&cr_ses->back_channel);
2816 goto out_release_drc_mem;
2817 status = nfserr_jukebox;
2818 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2820 goto out_release_drc_mem;
2821 conn = alloc_conn_from_crses(rqstp, cr_ses);
2823 goto out_free_session;
2825 spin_lock(&nn->client_lock);
2826 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2827 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2828 WARN_ON_ONCE(conf && unconf);
2831 status = nfserr_wrong_cred;
2832 if (!nfsd4_mach_creds_match(conf, rqstp))
2834 cs_slot = &conf->cl_cs_slot;
2835 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2837 if (status == nfserr_replay_cache)
2838 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2841 } else if (unconf) {
2842 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2843 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2844 status = nfserr_clid_inuse;
2847 status = nfserr_wrong_cred;
2848 if (!nfsd4_mach_creds_match(unconf, rqstp))
2850 cs_slot = &unconf->cl_cs_slot;
2851 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2853 /* an unconfirmed replay returns misordered */
2854 status = nfserr_seq_misordered;
2857 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2859 status = mark_client_expired_locked(old);
2865 move_to_confirmed(unconf);
2868 status = nfserr_stale_clientid;
2872 /* Persistent sessions are not supported */
2873 cr_ses->flags &= ~SESSION4_PERSIST;
2874 /* Upshifting from TCP to RDMA is not supported */
2875 cr_ses->flags &= ~SESSION4_RDMA;
2877 init_session(rqstp, new, conf, cr_ses);
2878 nfsd4_get_session_locked(new);
2880 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2881 NFS4_MAX_SESSIONID_LEN);
2882 cs_slot->sl_seqid++;
2883 cr_ses->seqid = cs_slot->sl_seqid;
2885 /* cache solo and embedded create sessions under the client_lock */
2886 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2887 spin_unlock(&nn->client_lock);
2888 /* init connection and backchannel */
2889 nfsd4_init_conn(rqstp, conn, new);
2890 nfsd4_put_session(new);
2895 spin_unlock(&nn->client_lock);
2900 __free_session(new);
2901 out_release_drc_mem:
2902 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2906 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2909 case NFS4_CDFC4_FORE:
2910 case NFS4_CDFC4_BACK:
2912 case NFS4_CDFC4_FORE_OR_BOTH:
2913 case NFS4_CDFC4_BACK_OR_BOTH:
2914 *dir = NFS4_CDFC4_BOTH;
2917 return nfserr_inval;
2920 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2921 struct nfsd4_compound_state *cstate,
2922 union nfsd4_op_u *u)
2924 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2925 struct nfsd4_session *session = cstate->session;
2926 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2929 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2932 spin_lock(&nn->client_lock);
2933 session->se_cb_prog = bc->bc_cb_program;
2934 session->se_cb_sec = bc->bc_cb_sec;
2935 spin_unlock(&nn->client_lock);
2937 nfsd4_probe_callback(session->se_client);
2942 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2943 struct nfsd4_compound_state *cstate,
2944 union nfsd4_op_u *u)
2946 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2948 struct nfsd4_conn *conn;
2949 struct nfsd4_session *session;
2950 struct net *net = SVC_NET(rqstp);
2951 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2953 if (!nfsd4_last_compound_op(rqstp))
2954 return nfserr_not_only_op;
2955 spin_lock(&nn->client_lock);
2956 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2957 spin_unlock(&nn->client_lock);
2959 goto out_no_session;
2960 status = nfserr_wrong_cred;
2961 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2963 status = nfsd4_map_bcts_dir(&bcts->dir);
2966 conn = alloc_conn(rqstp, bcts->dir);
2967 status = nfserr_jukebox;
2970 nfsd4_init_conn(rqstp, conn, session);
2973 nfsd4_put_session(session);
2978 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
2980 if (!cstate->session)
2982 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
2986 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2987 union nfsd4_op_u *u)
2989 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
2990 struct nfsd4_session *ses;
2992 int ref_held_by_me = 0;
2993 struct net *net = SVC_NET(r);
2994 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2996 status = nfserr_not_only_op;
2997 if (nfsd4_compound_in_session(cstate, sessionid)) {
2998 if (!nfsd4_last_compound_op(r))
3002 dump_sessionid(__func__, sessionid);
3003 spin_lock(&nn->client_lock);
3004 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3006 goto out_client_lock;
3007 status = nfserr_wrong_cred;
3008 if (!nfsd4_mach_creds_match(ses->se_client, r))
3009 goto out_put_session;
3010 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3012 goto out_put_session;
3013 unhash_session(ses);
3014 spin_unlock(&nn->client_lock);
3016 nfsd4_probe_callback_sync(ses->se_client);
3018 spin_lock(&nn->client_lock);
3021 nfsd4_put_session_locked(ses);
3023 spin_unlock(&nn->client_lock);
3028 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3030 struct nfsd4_conn *c;
3032 list_for_each_entry(c, &s->se_conns, cn_persession) {
3033 if (c->cn_xprt == xpt) {
3040 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3042 struct nfs4_client *clp = ses->se_client;
3043 struct nfsd4_conn *c;
3044 __be32 status = nfs_ok;
3047 spin_lock(&clp->cl_lock);
3048 c = __nfsd4_find_conn(new->cn_xprt, ses);
3051 status = nfserr_conn_not_bound_to_session;
3052 if (clp->cl_mach_cred)
3054 __nfsd4_hash_conn(new, ses);
3055 spin_unlock(&clp->cl_lock);
3056 ret = nfsd4_register_conn(new);
3058 /* oops; xprt is already down: */
3059 nfsd4_conn_lost(&new->cn_xpt_user);
3062 spin_unlock(&clp->cl_lock);
3067 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3069 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3071 return args->opcnt > session->se_fchannel.maxops;
3074 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3075 struct nfsd4_session *session)
3077 struct xdr_buf *xb = &rqstp->rq_arg;
3079 return xb->len > session->se_fchannel.maxreq_sz;
3082 static bool replay_matches_cache(struct svc_rqst *rqstp,
3083 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3085 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3087 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3088 (bool)seq->cachethis)
3091 * If there's an error then the reply can have fewer ops than
3094 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3097 * But if we cached a reply with *more* ops than the call you're
3098 * sending us now, then this new call is clearly not really a
3099 * replay of the old one:
3101 if (slot->sl_opcnt > argp->opcnt)
3103 /* This is the only check explicitly called by spec: */
3104 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3107 * There may be more comparisons we could actually do, but the
3108 * spec doesn't require us to catch every case where the calls
3109 * don't match (that would require caching the call as well as
3110 * the reply), so we don't bother.
3116 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3117 union nfsd4_op_u *u)
3119 struct nfsd4_sequence *seq = &u->sequence;
3120 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3121 struct xdr_stream *xdr = &resp->xdr;
3122 struct nfsd4_session *session;
3123 struct nfs4_client *clp;
3124 struct nfsd4_slot *slot;
3125 struct nfsd4_conn *conn;
3128 struct net *net = SVC_NET(rqstp);
3129 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3131 if (resp->opcnt != 1)
3132 return nfserr_sequence_pos;
3135 * Will be either used or freed by nfsd4_sequence_check_conn
3138 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3140 return nfserr_jukebox;
3142 spin_lock(&nn->client_lock);
3143 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3145 goto out_no_session;
3146 clp = session->se_client;
3148 status = nfserr_too_many_ops;
3149 if (nfsd4_session_too_many_ops(rqstp, session))
3150 goto out_put_session;
3152 status = nfserr_req_too_big;
3153 if (nfsd4_request_too_big(rqstp, session))
3154 goto out_put_session;
3156 status = nfserr_badslot;
3157 if (seq->slotid >= session->se_fchannel.maxreqs)
3158 goto out_put_session;
3160 slot = session->se_slots[seq->slotid];
3161 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3163 /* We do not negotiate the number of slots yet, so set the
3164 * maxslots to the session maxreqs which is used to encode
3165 * sr_highest_slotid and the sr_target_slot id to maxslots */
3166 seq->maxslots = session->se_fchannel.maxreqs;
3168 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3169 slot->sl_flags & NFSD4_SLOT_INUSE);
3170 if (status == nfserr_replay_cache) {
3171 status = nfserr_seq_misordered;
3172 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3173 goto out_put_session;
3174 status = nfserr_seq_false_retry;
3175 if (!replay_matches_cache(rqstp, seq, slot))
3176 goto out_put_session;
3177 cstate->slot = slot;
3178 cstate->session = session;
3180 /* Return the cached reply status and set cstate->status
3181 * for nfsd4_proc_compound processing */
3182 status = nfsd4_replay_cache_entry(resp, seq);
3183 cstate->status = nfserr_replay_cache;
3187 goto out_put_session;
3189 status = nfsd4_sequence_check_conn(conn, session);
3192 goto out_put_session;
3194 buflen = (seq->cachethis) ?
3195 session->se_fchannel.maxresp_cached :
3196 session->se_fchannel.maxresp_sz;
3197 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3199 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3200 goto out_put_session;
3201 svc_reserve(rqstp, buflen);
3204 /* Success! bump slot seqid */
3205 slot->sl_seqid = seq->seqid;
3206 slot->sl_flags |= NFSD4_SLOT_INUSE;
3208 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3210 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3212 cstate->slot = slot;
3213 cstate->session = session;
3217 switch (clp->cl_cb_state) {
3219 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3221 case NFSD4_CB_FAULT:
3222 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3225 seq->status_flags = 0;
3227 if (!list_empty(&clp->cl_revoked))
3228 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3232 spin_unlock(&nn->client_lock);
3235 nfsd4_put_session_locked(session);
3236 goto out_no_session;
3240 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3242 struct nfsd4_compound_state *cs = &resp->cstate;
3244 if (nfsd4_has_session(cs)) {
3245 if (cs->status != nfserr_replay_cache) {
3246 nfsd4_store_cache_entry(resp);
3247 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3249 /* Drop session reference that was taken in nfsd4_sequence() */
3250 nfsd4_put_session(cs->session);
3252 put_client_renew(cs->clp);
3256 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3257 struct nfsd4_compound_state *cstate,
3258 union nfsd4_op_u *u)
3260 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3261 struct nfs4_client *conf, *unconf;
3262 struct nfs4_client *clp = NULL;
3264 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3266 spin_lock(&nn->client_lock);
3267 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3268 conf = find_confirmed_client(&dc->clientid, true, nn);
3269 WARN_ON_ONCE(conf && unconf);
3272 if (client_has_state(conf)) {
3273 status = nfserr_clientid_busy;
3276 status = mark_client_expired_locked(conf);
3283 status = nfserr_stale_clientid;
3286 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3288 status = nfserr_wrong_cred;
3291 unhash_client_locked(clp);
3293 spin_unlock(&nn->client_lock);
3300 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3301 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3303 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3306 if (rc->rca_one_fs) {
3307 if (!cstate->current_fh.fh_dentry)
3308 return nfserr_nofilehandle;
3310 * We don't take advantage of the rca_one_fs case.
3311 * That's OK, it's optional, we can safely ignore it.
3316 status = nfserr_complete_already;
3317 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3318 &cstate->session->se_client->cl_flags))
3321 status = nfserr_stale_clientid;
3322 if (is_client_expired(cstate->session->se_client))
3324 * The following error isn't really legal.
3325 * But we only get here if the client just explicitly
3326 * destroyed the client. Surely it no longer cares what
3327 * error it gets back on an operation for the dead
3333 nfsd4_client_record_create(cstate->session->se_client);
3339 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3340 union nfsd4_op_u *u)
3342 struct nfsd4_setclientid *setclid = &u->setclientid;
3343 struct xdr_netobj clname = setclid->se_name;
3344 nfs4_verifier clverifier = setclid->se_verf;
3345 struct nfs4_client *conf, *new;
3346 struct nfs4_client *unconf = NULL;
3348 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3350 new = create_client(clname, rqstp, &clverifier);
3352 return nfserr_jukebox;
3353 /* Cases below refer to rfc 3530 section 14.2.33: */
3354 spin_lock(&nn->client_lock);
3355 conf = find_confirmed_client_by_name(&clname, nn);
3356 if (conf && client_has_state(conf)) {
3358 status = nfserr_clid_inuse;
3359 if (clp_used_exchangeid(conf))
3361 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3362 char addr_str[INET6_ADDRSTRLEN];
3363 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3365 dprintk("NFSD: setclientid: string in use by client "
3366 "at %s\n", addr_str);
3370 unconf = find_unconfirmed_client_by_name(&clname, nn);
3372 unhash_client_locked(unconf);
3373 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3374 /* case 1: probable callback update */
3375 copy_clid(new, conf);
3376 gen_confirm(new, nn);
3377 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3379 new->cl_minorversion = 0;
3380 gen_callback(new, setclid, rqstp);
3381 add_to_unconfirmed(new);
3382 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3383 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3384 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3388 spin_unlock(&nn->client_lock);
3392 expire_client(unconf);
3398 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3399 struct nfsd4_compound_state *cstate,
3400 union nfsd4_op_u *u)
3402 struct nfsd4_setclientid_confirm *setclientid_confirm =
3403 &u->setclientid_confirm;
3404 struct nfs4_client *conf, *unconf;
3405 struct nfs4_client *old = NULL;
3406 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3407 clientid_t * clid = &setclientid_confirm->sc_clientid;
3409 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3411 if (STALE_CLIENTID(clid, nn))
3412 return nfserr_stale_clientid;
3414 spin_lock(&nn->client_lock);
3415 conf = find_confirmed_client(clid, false, nn);
3416 unconf = find_unconfirmed_client(clid, false, nn);
3418 * We try hard to give out unique clientid's, so if we get an
3419 * attempt to confirm the same clientid with a different cred,
3420 * the client may be buggy; this should never happen.
3422 * Nevertheless, RFC 7530 recommends INUSE for this case:
3424 status = nfserr_clid_inuse;
3425 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3427 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3429 /* cases below refer to rfc 3530 section 14.2.34: */
3430 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3431 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3432 /* case 2: probable retransmit */
3434 } else /* case 4: client hasn't noticed we rebooted yet? */
3435 status = nfserr_stale_clientid;
3439 if (conf) { /* case 1: callback update */
3441 unhash_client_locked(old);
3442 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3443 } else { /* case 3: normal case; new or rebooted client */
3444 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3446 status = nfserr_clid_inuse;
3447 if (client_has_state(old)
3448 && !same_creds(&unconf->cl_cred,
3453 status = mark_client_expired_locked(old);
3459 move_to_confirmed(unconf);
3462 get_client_locked(conf);
3463 spin_unlock(&nn->client_lock);
3464 nfsd4_probe_callback(conf);
3465 spin_lock(&nn->client_lock);
3466 put_client_renew_locked(conf);
3468 spin_unlock(&nn->client_lock);
3474 static struct nfs4_file *nfsd4_alloc_file(void)
3476 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3479 /* OPEN Share state helper functions */
3480 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3481 struct nfs4_file *fp)
3483 lockdep_assert_held(&state_lock);
3485 refcount_set(&fp->fi_ref, 1);
3486 spin_lock_init(&fp->fi_lock);
3487 INIT_LIST_HEAD(&fp->fi_stateids);
3488 INIT_LIST_HEAD(&fp->fi_delegations);
3489 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3490 fh_copy_shallow(&fp->fi_fhandle, fh);
3491 fp->fi_deleg_file = NULL;
3492 fp->fi_had_conflict = false;
3493 fp->fi_share_deny = 0;
3494 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3495 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3496 #ifdef CONFIG_NFSD_PNFS
3497 INIT_LIST_HEAD(&fp->fi_lo_states);
3498 atomic_set(&fp->fi_lo_recalls, 0);
3500 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3504 nfsd4_free_slabs(void)
3506 kmem_cache_destroy(client_slab);
3507 kmem_cache_destroy(openowner_slab);
3508 kmem_cache_destroy(lockowner_slab);
3509 kmem_cache_destroy(file_slab);
3510 kmem_cache_destroy(stateid_slab);
3511 kmem_cache_destroy(deleg_slab);
3512 kmem_cache_destroy(odstate_slab);
3516 nfsd4_init_slabs(void)
3518 client_slab = kmem_cache_create("nfsd4_clients",
3519 sizeof(struct nfs4_client), 0, 0, NULL);
3520 if (client_slab == NULL)
3522 openowner_slab = kmem_cache_create("nfsd4_openowners",
3523 sizeof(struct nfs4_openowner), 0, 0, NULL);
3524 if (openowner_slab == NULL)
3525 goto out_free_client_slab;
3526 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3527 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3528 if (lockowner_slab == NULL)
3529 goto out_free_openowner_slab;
3530 file_slab = kmem_cache_create("nfsd4_files",
3531 sizeof(struct nfs4_file), 0, 0, NULL);
3532 if (file_slab == NULL)
3533 goto out_free_lockowner_slab;
3534 stateid_slab = kmem_cache_create("nfsd4_stateids",
3535 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3536 if (stateid_slab == NULL)
3537 goto out_free_file_slab;
3538 deleg_slab = kmem_cache_create("nfsd4_delegations",
3539 sizeof(struct nfs4_delegation), 0, 0, NULL);
3540 if (deleg_slab == NULL)
3541 goto out_free_stateid_slab;
3542 odstate_slab = kmem_cache_create("nfsd4_odstate",
3543 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3544 if (odstate_slab == NULL)
3545 goto out_free_deleg_slab;
3548 out_free_deleg_slab:
3549 kmem_cache_destroy(deleg_slab);
3550 out_free_stateid_slab:
3551 kmem_cache_destroy(stateid_slab);
3553 kmem_cache_destroy(file_slab);
3554 out_free_lockowner_slab:
3555 kmem_cache_destroy(lockowner_slab);
3556 out_free_openowner_slab:
3557 kmem_cache_destroy(openowner_slab);
3558 out_free_client_slab:
3559 kmem_cache_destroy(client_slab);
3561 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3565 static void init_nfs4_replay(struct nfs4_replay *rp)
3567 rp->rp_status = nfserr_serverfault;
3569 rp->rp_buf = rp->rp_ibuf;
3570 mutex_init(&rp->rp_mutex);
3573 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3574 struct nfs4_stateowner *so)
3576 if (!nfsd4_has_session(cstate)) {
3577 mutex_lock(&so->so_replay.rp_mutex);
3578 cstate->replay_owner = nfs4_get_stateowner(so);
3582 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3584 struct nfs4_stateowner *so = cstate->replay_owner;
3587 cstate->replay_owner = NULL;
3588 mutex_unlock(&so->so_replay.rp_mutex);
3589 nfs4_put_stateowner(so);
3593 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3595 struct nfs4_stateowner *sop;
3597 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3601 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3602 if (!sop->so_owner.data) {
3603 kmem_cache_free(slab, sop);
3606 sop->so_owner.len = owner->len;
3608 INIT_LIST_HEAD(&sop->so_stateids);
3609 sop->so_client = clp;
3610 init_nfs4_replay(&sop->so_replay);
3611 atomic_set(&sop->so_count, 1);
3615 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3617 lockdep_assert_held(&clp->cl_lock);
3619 list_add(&oo->oo_owner.so_strhash,
3620 &clp->cl_ownerstr_hashtbl[strhashval]);
3621 list_add(&oo->oo_perclient, &clp->cl_openowners);
3624 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3626 unhash_openowner_locked(openowner(so));
3629 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3631 struct nfs4_openowner *oo = openowner(so);
3633 kmem_cache_free(openowner_slab, oo);
3636 static const struct nfs4_stateowner_operations openowner_ops = {
3637 .so_unhash = nfs4_unhash_openowner,
3638 .so_free = nfs4_free_openowner,
3641 static struct nfs4_ol_stateid *
3642 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3644 struct nfs4_ol_stateid *local, *ret = NULL;
3645 struct nfs4_openowner *oo = open->op_openowner;
3647 lockdep_assert_held(&fp->fi_lock);
3649 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3650 /* ignore lock owners */
3651 if (local->st_stateowner->so_is_open_owner == 0)
3653 if (local->st_stateowner != &oo->oo_owner)
3655 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3657 refcount_inc(&ret->st_stid.sc_count);
3665 nfsd4_verify_open_stid(struct nfs4_stid *s)
3667 __be32 ret = nfs_ok;
3669 switch (s->sc_type) {
3673 case NFS4_CLOSED_STID:
3674 case NFS4_CLOSED_DELEG_STID:
3675 ret = nfserr_bad_stateid;
3677 case NFS4_REVOKED_DELEG_STID:
3678 ret = nfserr_deleg_revoked;
3683 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3685 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3689 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3690 ret = nfsd4_verify_open_stid(&stp->st_stid);
3692 mutex_unlock(&stp->st_mutex);
3696 static struct nfs4_ol_stateid *
3697 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3699 struct nfs4_ol_stateid *stp;
3701 spin_lock(&fp->fi_lock);
3702 stp = nfsd4_find_existing_open(fp, open);
3703 spin_unlock(&fp->fi_lock);
3704 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3706 nfs4_put_stid(&stp->st_stid);
3711 static struct nfs4_openowner *
3712 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3713 struct nfsd4_compound_state *cstate)
3715 struct nfs4_client *clp = cstate->clp;
3716 struct nfs4_openowner *oo, *ret;
3718 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3721 oo->oo_owner.so_ops = &openowner_ops;
3722 oo->oo_owner.so_is_open_owner = 1;
3723 oo->oo_owner.so_seqid = open->op_seqid;
3725 if (nfsd4_has_session(cstate))
3726 oo->oo_flags |= NFS4_OO_CONFIRMED;
3728 oo->oo_last_closed_stid = NULL;
3729 INIT_LIST_HEAD(&oo->oo_close_lru);
3730 spin_lock(&clp->cl_lock);
3731 ret = find_openstateowner_str_locked(strhashval, open, clp);
3733 hash_openowner(oo, clp, strhashval);
3736 nfs4_free_stateowner(&oo->oo_owner);
3738 spin_unlock(&clp->cl_lock);
3742 static struct nfs4_ol_stateid *
3743 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3746 struct nfs4_openowner *oo = open->op_openowner;
3747 struct nfs4_ol_stateid *retstp = NULL;
3748 struct nfs4_ol_stateid *stp;
3751 /* We are moving these outside of the spinlocks to avoid the warnings */
3752 mutex_init(&stp->st_mutex);
3753 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3756 spin_lock(&oo->oo_owner.so_client->cl_lock);
3757 spin_lock(&fp->fi_lock);
3759 retstp = nfsd4_find_existing_open(fp, open);
3763 open->op_stp = NULL;
3764 refcount_inc(&stp->st_stid.sc_count);
3765 stp->st_stid.sc_type = NFS4_OPEN_STID;
3766 INIT_LIST_HEAD(&stp->st_locks);
3767 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3769 stp->st_stid.sc_file = fp;
3770 stp->st_access_bmap = 0;
3771 stp->st_deny_bmap = 0;
3772 stp->st_openstp = NULL;
3773 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3774 list_add(&stp->st_perfile, &fp->fi_stateids);
3777 spin_unlock(&fp->fi_lock);
3778 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3780 /* Handle races with CLOSE */
3781 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3782 nfs4_put_stid(&retstp->st_stid);
3785 /* To keep mutex tracking happy */
3786 mutex_unlock(&stp->st_mutex);
3793 * In the 4.0 case we need to keep the owners around a little while to handle
3794 * CLOSE replay. We still do need to release any file access that is held by
3795 * them before returning however.
3798 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3800 struct nfs4_ol_stateid *last;
3801 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3802 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3805 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3808 * We know that we hold one reference via nfsd4_close, and another
3809 * "persistent" reference for the client. If the refcount is higher
3810 * than 2, then there are still calls in progress that are using this
3811 * stateid. We can't put the sc_file reference until they are finished.
3812 * Wait for the refcount to drop to 2. Since it has been unhashed,
3813 * there should be no danger of the refcount going back up again at
3816 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3818 release_all_access(s);
3819 if (s->st_stid.sc_file) {
3820 put_nfs4_file(s->st_stid.sc_file);
3821 s->st_stid.sc_file = NULL;
3824 spin_lock(&nn->client_lock);
3825 last = oo->oo_last_closed_stid;
3826 oo->oo_last_closed_stid = s;
3827 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3828 oo->oo_time = get_seconds();
3829 spin_unlock(&nn->client_lock);
3831 nfs4_put_stid(&last->st_stid);
3834 /* search file_hashtbl[] for file */
3835 static struct nfs4_file *
3836 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3838 struct nfs4_file *fp;
3840 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3841 if (fh_match(&fp->fi_fhandle, fh)) {
3842 if (refcount_inc_not_zero(&fp->fi_ref))
3850 find_file(struct knfsd_fh *fh)
3852 struct nfs4_file *fp;
3853 unsigned int hashval = file_hashval(fh);
3856 fp = find_file_locked(fh, hashval);
3861 static struct nfs4_file *
3862 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3864 struct nfs4_file *fp;
3865 unsigned int hashval = file_hashval(fh);
3868 fp = find_file_locked(fh, hashval);
3873 spin_lock(&state_lock);
3874 fp = find_file_locked(fh, hashval);
3875 if (likely(fp == NULL)) {
3876 nfsd4_init_file(fh, hashval, new);
3879 spin_unlock(&state_lock);
3885 * Called to check deny when READ with all zero stateid or
3886 * WRITE with all zero or all one stateid
3889 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3891 struct nfs4_file *fp;
3892 __be32 ret = nfs_ok;
3894 fp = find_file(¤t_fh->fh_handle);
3897 /* Check for conflicting share reservations */
3898 spin_lock(&fp->fi_lock);
3899 if (fp->fi_share_deny & deny_type)
3900 ret = nfserr_locked;
3901 spin_unlock(&fp->fi_lock);
3906 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3908 struct nfs4_delegation *dp = cb_to_delegation(cb);
3909 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3912 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3915 * We can't do this in nfsd_break_deleg_cb because it is
3916 * already holding inode->i_lock.
3918 * If the dl_time != 0, then we know that it has already been
3919 * queued for a lease break. Don't queue it again.
3921 spin_lock(&state_lock);
3922 if (delegation_hashed(dp) && dp->dl_time == 0) {
3923 dp->dl_time = get_seconds();
3924 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3926 spin_unlock(&state_lock);
3929 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3930 struct rpc_task *task)
3932 struct nfs4_delegation *dp = cb_to_delegation(cb);
3934 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3937 switch (task->tk_status) {
3941 case -NFS4ERR_BAD_STATEID:
3943 * Race: client probably got cb_recall before open reply
3944 * granting delegation.
3946 if (dp->dl_retries--) {
3947 rpc_delay(task, 2 * HZ);
3956 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3958 struct nfs4_delegation *dp = cb_to_delegation(cb);
3960 nfs4_put_stid(&dp->dl_stid);
3963 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3964 .prepare = nfsd4_cb_recall_prepare,
3965 .done = nfsd4_cb_recall_done,
3966 .release = nfsd4_cb_recall_release,
3969 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3972 * We're assuming the state code never drops its reference
3973 * without first removing the lease. Since we're in this lease
3974 * callback (and since the lease code is serialized by the
3975 * i_lock) we know the server hasn't removed the lease yet, and
3976 * we know it's safe to take a reference.
3978 refcount_inc(&dp->dl_stid.sc_count);
3979 nfsd4_run_cb(&dp->dl_recall);
3982 /* Called from break_lease() with i_lock held. */
3984 nfsd_break_deleg_cb(struct file_lock *fl)
3987 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
3988 struct nfs4_file *fp = dp->dl_stid.sc_file;
3991 * We don't want the locks code to timeout the lease for us;
3992 * we'll remove it ourself if a delegation isn't returned
3995 fl->fl_break_time = 0;
3997 spin_lock(&fp->fi_lock);
3998 fp->fi_had_conflict = true;
3999 nfsd_break_one_deleg(dp);
4000 spin_unlock(&fp->fi_lock);
4005 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4006 struct list_head *dispose)
4009 return lease_modify(onlist, arg, dispose);
4014 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4015 .lm_break = nfsd_break_deleg_cb,
4016 .lm_change = nfsd_change_deleg_cb,
4019 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4021 if (nfsd4_has_session(cstate))
4023 if (seqid == so->so_seqid - 1)
4024 return nfserr_replay_me;
4025 if (seqid == so->so_seqid)
4027 return nfserr_bad_seqid;
4030 static __be32 lookup_clientid(clientid_t *clid,
4031 struct nfsd4_compound_state *cstate,
4032 struct nfsd_net *nn)
4034 struct nfs4_client *found;
4037 found = cstate->clp;
4038 if (!same_clid(&found->cl_clientid, clid))
4039 return nfserr_stale_clientid;
4043 if (STALE_CLIENTID(clid, nn))
4044 return nfserr_stale_clientid;
4047 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4048 * cached already then we know this is for is for v4.0 and "sessions"
4051 WARN_ON_ONCE(cstate->session);
4052 spin_lock(&nn->client_lock);
4053 found = find_confirmed_client(clid, false, nn);
4055 spin_unlock(&nn->client_lock);
4056 return nfserr_expired;
4058 atomic_inc(&found->cl_refcount);
4059 spin_unlock(&nn->client_lock);
4061 /* Cache the nfs4_client in cstate! */
4062 cstate->clp = found;
4067 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4068 struct nfsd4_open *open, struct nfsd_net *nn)
4070 clientid_t *clientid = &open->op_clientid;
4071 struct nfs4_client *clp = NULL;
4072 unsigned int strhashval;
4073 struct nfs4_openowner *oo = NULL;
4076 if (STALE_CLIENTID(&open->op_clientid, nn))
4077 return nfserr_stale_clientid;
4079 * In case we need it later, after we've already created the
4080 * file and don't want to risk a further failure:
4082 open->op_file = nfsd4_alloc_file();
4083 if (open->op_file == NULL)
4084 return nfserr_jukebox;
4086 status = lookup_clientid(clientid, cstate, nn);
4091 strhashval = ownerstr_hashval(&open->op_owner);
4092 oo = find_openstateowner_str(strhashval, open, clp);
4093 open->op_openowner = oo;
4097 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4098 /* Replace unconfirmed owners without checking for replay. */
4099 release_openowner(oo);
4100 open->op_openowner = NULL;
4103 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4108 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4110 return nfserr_jukebox;
4111 open->op_openowner = oo;
4113 open->op_stp = nfs4_alloc_open_stateid(clp);
4115 return nfserr_jukebox;
4117 if (nfsd4_has_session(cstate) &&
4118 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4119 open->op_odstate = alloc_clnt_odstate(clp);
4120 if (!open->op_odstate)
4121 return nfserr_jukebox;
4127 static inline __be32
4128 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4130 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4131 return nfserr_openmode;
4136 static int share_access_to_flags(u32 share_access)
4138 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4141 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4143 struct nfs4_stid *ret;
4145 ret = find_stateid_by_type(cl, s,
4146 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4149 return delegstateid(ret);
4152 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4154 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4155 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4159 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4160 struct nfs4_delegation **dp)
4163 __be32 status = nfserr_bad_stateid;
4164 struct nfs4_delegation *deleg;
4166 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4169 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4170 nfs4_put_stid(&deleg->dl_stid);
4171 if (cl->cl_minorversion)
4172 status = nfserr_deleg_revoked;
4175 flags = share_access_to_flags(open->op_share_access);
4176 status = nfs4_check_delegmode(deleg, flags);
4178 nfs4_put_stid(&deleg->dl_stid);
4183 if (!nfsd4_is_deleg_cur(open))
4187 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4191 static inline int nfs4_access_to_access(u32 nfs4_access)
4195 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4196 flags |= NFSD_MAY_READ;
4197 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4198 flags |= NFSD_MAY_WRITE;
4202 static inline __be32
4203 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4204 struct nfsd4_open *open)
4206 struct iattr iattr = {
4207 .ia_valid = ATTR_SIZE,
4210 if (!open->op_truncate)
4212 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4213 return nfserr_inval;
4214 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4217 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4218 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4219 struct nfsd4_open *open)
4221 struct file *filp = NULL;
4223 int oflag = nfs4_access_to_omode(open->op_share_access);
4224 int access = nfs4_access_to_access(open->op_share_access);
4225 unsigned char old_access_bmap, old_deny_bmap;
4227 spin_lock(&fp->fi_lock);
4230 * Are we trying to set a deny mode that would conflict with
4233 status = nfs4_file_check_deny(fp, open->op_share_deny);
4234 if (status != nfs_ok) {
4235 spin_unlock(&fp->fi_lock);
4239 /* set access to the file */
4240 status = nfs4_file_get_access(fp, open->op_share_access);
4241 if (status != nfs_ok) {
4242 spin_unlock(&fp->fi_lock);
4246 /* Set access bits in stateid */
4247 old_access_bmap = stp->st_access_bmap;
4248 set_access(open->op_share_access, stp);
4250 /* Set new deny mask */
4251 old_deny_bmap = stp->st_deny_bmap;
4252 set_deny(open->op_share_deny, stp);
4253 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4255 if (!fp->fi_fds[oflag]) {
4256 spin_unlock(&fp->fi_lock);
4257 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4259 goto out_put_access;
4260 spin_lock(&fp->fi_lock);
4261 if (!fp->fi_fds[oflag]) {
4262 fp->fi_fds[oflag] = filp;
4266 spin_unlock(&fp->fi_lock);
4270 status = nfsd4_truncate(rqstp, cur_fh, open);
4272 goto out_put_access;
4276 stp->st_access_bmap = old_access_bmap;
4277 nfs4_file_put_access(fp, open->op_share_access);
4278 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4283 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)
4286 unsigned char old_deny_bmap = stp->st_deny_bmap;
4288 if (!test_access(open->op_share_access, stp))
4289 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4291 /* test and set deny mode */
4292 spin_lock(&fp->fi_lock);
4293 status = nfs4_file_check_deny(fp, open->op_share_deny);
4294 if (status == nfs_ok) {
4295 set_deny(open->op_share_deny, stp);
4296 fp->fi_share_deny |=
4297 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4299 spin_unlock(&fp->fi_lock);
4301 if (status != nfs_ok)
4304 status = nfsd4_truncate(rqstp, cur_fh, open);
4305 if (status != nfs_ok)
4306 reset_union_bmap_deny(old_deny_bmap, stp);
4310 /* Should we give out recallable state?: */
4311 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4313 if (clp->cl_cb_state == NFSD4_CB_UP)
4316 * In the sessions case, since we don't have to establish a
4317 * separate connection for callbacks, we assume it's OK
4318 * until we hear otherwise:
4320 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4323 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4326 struct file_lock *fl;
4328 fl = locks_alloc_lock();
4331 fl->fl_lmops = &nfsd_lease_mng_ops;
4332 fl->fl_flags = FL_DELEG;
4333 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4334 fl->fl_end = OFFSET_MAX;
4335 fl->fl_owner = (fl_owner_t)dp;
4336 fl->fl_pid = current->tgid;
4337 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file;
4341 static struct nfs4_delegation *
4342 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4343 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4346 struct nfs4_delegation *dp;
4348 struct file_lock *fl;
4351 * The fi_had_conflict and nfs_get_existing_delegation checks
4352 * here are just optimizations; we'll need to recheck them at
4355 if (fp->fi_had_conflict)
4356 return ERR_PTR(-EAGAIN);
4358 filp = find_readable_file(fp);
4360 /* We should always have a readable file here */
4362 return ERR_PTR(-EBADF);
4364 spin_lock(&state_lock);
4365 spin_lock(&fp->fi_lock);
4366 if (nfs4_delegation_exists(clp, fp))
4368 else if (!fp->fi_deleg_file) {
4369 fp->fi_deleg_file = filp;
4370 /* increment early to prevent fi_deleg_file from being
4372 fp->fi_delegees = 1;
4376 spin_unlock(&fp->fi_lock);
4377 spin_unlock(&state_lock);
4381 return ERR_PTR(status);
4384 dp = alloc_init_deleg(clp, fp, fh, odstate);
4388 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4390 goto out_clnt_odstate;
4392 status = vfs_setlease(fp->fi_deleg_file, fl->fl_type, &fl, NULL);
4394 locks_free_lock(fl);
4396 goto out_clnt_odstate;
4398 spin_lock(&state_lock);
4399 spin_lock(&fp->fi_lock);
4400 if (fp->fi_had_conflict)
4403 status = hash_delegation_locked(dp, fp);
4404 spin_unlock(&fp->fi_lock);
4405 spin_unlock(&state_lock);
4412 vfs_setlease(fp->fi_deleg_file, F_UNLCK, NULL, (void **)&dp);
4414 put_clnt_odstate(dp->dl_clnt_odstate);
4415 nfs4_put_stid(&dp->dl_stid);
4418 return ERR_PTR(status);
4421 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4423 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4424 if (status == -EAGAIN)
4425 open->op_why_no_deleg = WND4_CONTENTION;
4427 open->op_why_no_deleg = WND4_RESOURCE;
4428 switch (open->op_deleg_want) {
4429 case NFS4_SHARE_WANT_READ_DELEG:
4430 case NFS4_SHARE_WANT_WRITE_DELEG:
4431 case NFS4_SHARE_WANT_ANY_DELEG:
4433 case NFS4_SHARE_WANT_CANCEL:
4434 open->op_why_no_deleg = WND4_CANCELLED;
4436 case NFS4_SHARE_WANT_NO_DELEG:
4443 * Attempt to hand out a delegation.
4445 * Note we don't support write delegations, and won't until the vfs has
4446 * proper support for them.
4449 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4450 struct nfs4_ol_stateid *stp)
4452 struct nfs4_delegation *dp;
4453 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4454 struct nfs4_client *clp = stp->st_stid.sc_client;
4458 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4459 open->op_recall = 0;
4460 switch (open->op_claim_type) {
4461 case NFS4_OPEN_CLAIM_PREVIOUS:
4463 open->op_recall = 1;
4464 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4467 case NFS4_OPEN_CLAIM_NULL:
4468 case NFS4_OPEN_CLAIM_FH:
4470 * Let's not give out any delegations till everyone's
4471 * had the chance to reclaim theirs, *and* until
4472 * NLM locks have all been reclaimed:
4474 if (locks_in_grace(clp->net))
4476 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4479 * Also, if the file was opened for write or
4480 * create, there's a good chance the client's
4481 * about to write to it, resulting in an
4482 * immediate recall (since we don't support
4483 * write delegations):
4485 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4487 if (open->op_create == NFS4_OPEN_CREATE)
4493 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4497 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4499 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4500 STATEID_VAL(&dp->dl_stid.sc_stateid));
4501 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4502 nfs4_put_stid(&dp->dl_stid);
4505 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4506 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4507 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4508 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4509 open->op_recall = 1;
4512 /* 4.1 client asking for a delegation? */
4513 if (open->op_deleg_want)
4514 nfsd4_open_deleg_none_ext(open, status);
4518 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4519 struct nfs4_delegation *dp)
4521 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4522 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4523 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4524 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4525 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4526 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4527 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4528 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4530 /* Otherwise the client must be confused wanting a delegation
4531 * it already has, therefore we don't return
4532 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4537 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4539 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4540 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4541 struct nfs4_file *fp = NULL;
4542 struct nfs4_ol_stateid *stp = NULL;
4543 struct nfs4_delegation *dp = NULL;
4545 bool new_stp = false;
4548 * Lookup file; if found, lookup stateid and check open request,
4549 * and check for delegations in the process of being recalled.
4550 * If not found, create the nfs4_file struct
4552 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4553 if (fp != open->op_file) {
4554 status = nfs4_check_deleg(cl, open, &dp);
4557 stp = nfsd4_find_and_lock_existing_open(fp, open);
4559 open->op_file = NULL;
4560 status = nfserr_bad_stateid;
4561 if (nfsd4_is_deleg_cur(open))
4566 stp = init_open_stateid(fp, open);
4572 * OPEN the file, or upgrade an existing OPEN.
4573 * If truncate fails, the OPEN fails.
4575 * stp is already locked.
4578 /* Stateid was found, this is an OPEN upgrade */
4579 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4581 mutex_unlock(&stp->st_mutex);
4585 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4587 stp->st_stid.sc_type = NFS4_CLOSED_STID;
4588 release_open_stateid(stp);
4589 mutex_unlock(&stp->st_mutex);
4593 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4595 if (stp->st_clnt_odstate == open->op_odstate)
4596 open->op_odstate = NULL;
4599 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4600 mutex_unlock(&stp->st_mutex);
4602 if (nfsd4_has_session(&resp->cstate)) {
4603 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4604 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4605 open->op_why_no_deleg = WND4_NOT_WANTED;
4611 * Attempt to hand out a delegation. No error return, because the
4612 * OPEN succeeds even if we fail.
4614 nfs4_open_delegation(current_fh, open, stp);
4618 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4619 STATEID_VAL(&stp->st_stid.sc_stateid));
4621 /* 4.1 client trying to upgrade/downgrade delegation? */
4622 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4623 open->op_deleg_want)
4624 nfsd4_deleg_xgrade_none_ext(open, dp);
4628 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4629 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4631 * To finish the open response, we just need to set the rflags.
4633 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4634 if (nfsd4_has_session(&resp->cstate))
4635 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4636 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4637 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4640 nfs4_put_stid(&dp->dl_stid);
4642 nfs4_put_stid(&stp->st_stid);
4647 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4648 struct nfsd4_open *open)
4650 if (open->op_openowner) {
4651 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4653 nfsd4_cstate_assign_replay(cstate, so);
4654 nfs4_put_stateowner(so);
4657 kmem_cache_free(file_slab, open->op_file);
4659 nfs4_put_stid(&open->op_stp->st_stid);
4660 if (open->op_odstate)
4661 kmem_cache_free(odstate_slab, open->op_odstate);
4665 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4666 union nfsd4_op_u *u)
4668 clientid_t *clid = &u->renew;
4669 struct nfs4_client *clp;
4671 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4673 dprintk("process_renew(%08x/%08x): starting\n",
4674 clid->cl_boot, clid->cl_id);
4675 status = lookup_clientid(clid, cstate, nn);
4679 status = nfserr_cb_path_down;
4680 if (!list_empty(&clp->cl_delegations)
4681 && clp->cl_cb_state != NFSD4_CB_UP)
4689 nfsd4_end_grace(struct nfsd_net *nn)
4691 /* do nothing if grace period already ended */
4692 if (nn->grace_ended)
4695 dprintk("NFSD: end of grace period\n");
4696 nn->grace_ended = true;
4698 * If the server goes down again right now, an NFSv4
4699 * client will still be allowed to reclaim after it comes back up,
4700 * even if it hasn't yet had a chance to reclaim state this time.
4703 nfsd4_record_grace_done(nn);
4705 * At this point, NFSv4 clients can still reclaim. But if the
4706 * server crashes, any that have not yet reclaimed will be out
4707 * of luck on the next boot.
4709 * (NFSv4.1+ clients are considered to have reclaimed once they
4710 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4711 * have reclaimed after their first OPEN.)
4713 locks_end_grace(&nn->nfsd4_manager);
4715 * At this point, and once lockd and/or any other containers
4716 * exit their grace period, further reclaims will fail and
4717 * regular locking can resume.
4722 * If we've waited a lease period but there are still clients trying to
4723 * reclaim, wait a little longer to give them a chance to finish.
4725 static bool clients_still_reclaiming(struct nfsd_net *nn)
4727 unsigned long now = get_seconds();
4728 unsigned long double_grace_period_end = nn->boot_time +
4729 2 * nn->nfsd4_lease;
4731 if (!nn->somebody_reclaimed)
4733 nn->somebody_reclaimed = false;
4735 * If we've given them *two* lease times to reclaim, and they're
4736 * still not done, give up:
4738 if (time_after(now, double_grace_period_end))
4744 nfs4_laundromat(struct nfsd_net *nn)
4746 struct nfs4_client *clp;
4747 struct nfs4_openowner *oo;
4748 struct nfs4_delegation *dp;
4749 struct nfs4_ol_stateid *stp;
4750 struct nfsd4_blocked_lock *nbl;
4751 struct list_head *pos, *next, reaplist;
4752 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4753 time_t t, new_timeo = nn->nfsd4_lease;
4755 dprintk("NFSD: laundromat service - starting\n");
4757 if (clients_still_reclaiming(nn)) {
4761 nfsd4_end_grace(nn);
4762 INIT_LIST_HEAD(&reaplist);
4763 spin_lock(&nn->client_lock);
4764 list_for_each_safe(pos, next, &nn->client_lru) {
4765 clp = list_entry(pos, struct nfs4_client, cl_lru);
4766 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4767 t = clp->cl_time - cutoff;
4768 new_timeo = min(new_timeo, t);
4771 if (mark_client_expired_locked(clp)) {
4772 dprintk("NFSD: client in use (clientid %08x)\n",
4773 clp->cl_clientid.cl_id);
4776 list_add(&clp->cl_lru, &reaplist);
4778 spin_unlock(&nn->client_lock);
4779 list_for_each_safe(pos, next, &reaplist) {
4780 clp = list_entry(pos, struct nfs4_client, cl_lru);
4781 dprintk("NFSD: purging unused client (clientid %08x)\n",
4782 clp->cl_clientid.cl_id);
4783 list_del_init(&clp->cl_lru);
4786 spin_lock(&state_lock);
4787 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4788 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4789 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4790 t = dp->dl_time - cutoff;
4791 new_timeo = min(new_timeo, t);
4794 WARN_ON(!unhash_delegation_locked(dp));
4795 list_add(&dp->dl_recall_lru, &reaplist);
4797 spin_unlock(&state_lock);
4798 while (!list_empty(&reaplist)) {
4799 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4801 list_del_init(&dp->dl_recall_lru);
4802 revoke_delegation(dp);
4805 spin_lock(&nn->client_lock);
4806 while (!list_empty(&nn->close_lru)) {
4807 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4809 if (time_after((unsigned long)oo->oo_time,
4810 (unsigned long)cutoff)) {
4811 t = oo->oo_time - cutoff;
4812 new_timeo = min(new_timeo, t);
4815 list_del_init(&oo->oo_close_lru);
4816 stp = oo->oo_last_closed_stid;
4817 oo->oo_last_closed_stid = NULL;
4818 spin_unlock(&nn->client_lock);
4819 nfs4_put_stid(&stp->st_stid);
4820 spin_lock(&nn->client_lock);
4822 spin_unlock(&nn->client_lock);
4825 * It's possible for a client to try and acquire an already held lock
4826 * that is being held for a long time, and then lose interest in it.
4827 * So, we clean out any un-revisited request after a lease period
4828 * under the assumption that the client is no longer interested.
4830 * RFC5661, sec. 9.6 states that the client must not rely on getting
4831 * notifications and must continue to poll for locks, even when the
4832 * server supports them. Thus this shouldn't lead to clients blocking
4833 * indefinitely once the lock does become free.
4835 BUG_ON(!list_empty(&reaplist));
4836 spin_lock(&nn->blocked_locks_lock);
4837 while (!list_empty(&nn->blocked_locks_lru)) {
4838 nbl = list_first_entry(&nn->blocked_locks_lru,
4839 struct nfsd4_blocked_lock, nbl_lru);
4840 if (time_after((unsigned long)nbl->nbl_time,
4841 (unsigned long)cutoff)) {
4842 t = nbl->nbl_time - cutoff;
4843 new_timeo = min(new_timeo, t);
4846 list_move(&nbl->nbl_lru, &reaplist);
4847 list_del_init(&nbl->nbl_list);
4849 spin_unlock(&nn->blocked_locks_lock);
4851 while (!list_empty(&reaplist)) {
4852 nbl = list_first_entry(&reaplist,
4853 struct nfsd4_blocked_lock, nbl_lru);
4854 list_del_init(&nbl->nbl_lru);
4855 posix_unblock_lock(&nbl->nbl_lock);
4856 free_blocked_lock(nbl);
4859 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4863 static struct workqueue_struct *laundry_wq;
4864 static void laundromat_main(struct work_struct *);
4867 laundromat_main(struct work_struct *laundry)
4870 struct delayed_work *dwork = to_delayed_work(laundry);
4871 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4874 t = nfs4_laundromat(nn);
4875 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4876 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4879 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4881 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4882 return nfserr_bad_stateid;
4887 access_permit_read(struct nfs4_ol_stateid *stp)
4889 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4890 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4891 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4895 access_permit_write(struct nfs4_ol_stateid *stp)
4897 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4898 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4902 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4904 __be32 status = nfserr_openmode;
4906 /* For lock stateid's, we test the parent open, not the lock: */
4907 if (stp->st_openstp)
4908 stp = stp->st_openstp;
4909 if ((flags & WR_STATE) && !access_permit_write(stp))
4911 if ((flags & RD_STATE) && !access_permit_read(stp))
4918 static inline __be32
4919 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4921 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4923 else if (opens_in_grace(net)) {
4924 /* Answer in remaining cases depends on existence of
4925 * conflicting state; so we must wait out the grace period. */
4926 return nfserr_grace;
4927 } else if (flags & WR_STATE)
4928 return nfs4_share_conflict(current_fh,
4929 NFS4_SHARE_DENY_WRITE);
4930 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4931 return nfs4_share_conflict(current_fh,
4932 NFS4_SHARE_DENY_READ);
4936 * Allow READ/WRITE during grace period on recovered state only for files
4937 * that are not able to provide mandatory locking.
4940 grace_disallows_io(struct net *net, struct inode *inode)
4942 return opens_in_grace(net) && mandatory_lock(inode);
4945 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4948 * When sessions are used the stateid generation number is ignored
4951 if (has_session && in->si_generation == 0)
4954 if (in->si_generation == ref->si_generation)
4957 /* If the client sends us a stateid from the future, it's buggy: */
4958 if (nfsd4_stateid_generation_after(in, ref))
4959 return nfserr_bad_stateid;
4961 * However, we could see a stateid from the past, even from a
4962 * non-buggy client. For example, if the client sends a lock
4963 * while some IO is outstanding, the lock may bump si_generation
4964 * while the IO is still in flight. The client could avoid that
4965 * situation by waiting for responses on all the IO requests,
4966 * but better performance may result in retrying IO that
4967 * receives an old_stateid error if requests are rarely
4968 * reordered in flight:
4970 return nfserr_old_stateid;
4973 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4977 spin_lock(&s->sc_lock);
4978 ret = nfsd4_verify_open_stid(s);
4980 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4981 spin_unlock(&s->sc_lock);
4985 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4987 if (ols->st_stateowner->so_is_open_owner &&
4988 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4989 return nfserr_bad_stateid;
4993 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4995 struct nfs4_stid *s;
4996 __be32 status = nfserr_bad_stateid;
4998 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4999 CLOSE_STATEID(stateid))
5001 /* Client debugging aid. */
5002 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
5003 char addr_str[INET6_ADDRSTRLEN];
5004 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
5006 pr_warn_ratelimited("NFSD: client %s testing state ID "
5007 "with incorrect client ID\n", addr_str);
5010 spin_lock(&cl->cl_lock);
5011 s = find_stateid_locked(cl, stateid);
5014 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5017 switch (s->sc_type) {
5018 case NFS4_DELEG_STID:
5021 case NFS4_REVOKED_DELEG_STID:
5022 status = nfserr_deleg_revoked;
5024 case NFS4_OPEN_STID:
5025 case NFS4_LOCK_STID:
5026 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5029 printk("unknown stateid type %x\n", s->sc_type);
5031 case NFS4_CLOSED_STID:
5032 case NFS4_CLOSED_DELEG_STID:
5033 status = nfserr_bad_stateid;
5036 spin_unlock(&cl->cl_lock);
5041 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5042 stateid_t *stateid, unsigned char typemask,
5043 struct nfs4_stid **s, struct nfsd_net *nn)
5046 bool return_revoked = false;
5049 * only return revoked delegations if explicitly asked.
5050 * otherwise we report revoked or bad_stateid status.
5052 if (typemask & NFS4_REVOKED_DELEG_STID)
5053 return_revoked = true;
5054 else if (typemask & NFS4_DELEG_STID)
5055 typemask |= NFS4_REVOKED_DELEG_STID;
5057 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5058 CLOSE_STATEID(stateid))
5059 return nfserr_bad_stateid;
5060 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5061 if (status == nfserr_stale_clientid) {
5062 if (cstate->session)
5063 return nfserr_bad_stateid;
5064 return nfserr_stale_stateid;
5068 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5070 return nfserr_bad_stateid;
5071 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5073 if (cstate->minorversion)
5074 return nfserr_deleg_revoked;
5075 return nfserr_bad_stateid;
5080 static struct file *
5081 nfs4_find_file(struct nfs4_stid *s, int flags)
5086 switch (s->sc_type) {
5087 case NFS4_DELEG_STID:
5088 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5090 return get_file(s->sc_file->fi_deleg_file);
5091 case NFS4_OPEN_STID:
5092 case NFS4_LOCK_STID:
5093 if (flags & RD_STATE)
5094 return find_readable_file(s->sc_file);
5096 return find_writeable_file(s->sc_file);
5104 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5108 status = nfsd4_check_openowner_confirmed(ols);
5111 return nfs4_check_openmode(ols, flags);
5115 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5116 struct file **filpp, bool *tmp_file, int flags)
5118 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5122 file = nfs4_find_file(s, flags);
5124 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5125 acc | NFSD_MAY_OWNER_OVERRIDE);
5133 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5145 * Checks for stateid operations
5148 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5149 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5150 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5152 struct inode *ino = d_inode(fhp->fh_dentry);
5153 struct net *net = SVC_NET(rqstp);
5154 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5155 struct nfs4_stid *s = NULL;
5163 if (grace_disallows_io(net, ino))
5164 return nfserr_grace;
5166 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5167 status = check_special_stateids(net, fhp, stateid, flags);
5171 status = nfsd4_lookup_stateid(cstate, stateid,
5172 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5176 status = nfsd4_stid_check_stateid_generation(stateid, s,
5177 nfsd4_has_session(cstate));
5181 switch (s->sc_type) {
5182 case NFS4_DELEG_STID:
5183 status = nfs4_check_delegmode(delegstateid(s), flags);
5185 case NFS4_OPEN_STID:
5186 case NFS4_LOCK_STID:
5187 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5190 status = nfserr_bad_stateid;
5195 status = nfs4_check_fh(fhp, s);
5198 if (!status && filpp)
5199 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5207 * Test if the stateid is valid
5210 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5211 union nfsd4_op_u *u)
5213 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5214 struct nfsd4_test_stateid_id *stateid;
5215 struct nfs4_client *cl = cstate->session->se_client;
5217 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5218 stateid->ts_id_status =
5219 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5225 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5227 struct nfs4_ol_stateid *stp = openlockstateid(s);
5230 ret = nfsd4_lock_ol_stateid(stp);
5234 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5238 ret = nfserr_locks_held;
5239 if (check_for_locks(stp->st_stid.sc_file,
5240 lockowner(stp->st_stateowner)))
5243 release_lock_stateid(stp);
5247 mutex_unlock(&stp->st_mutex);
5254 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5255 union nfsd4_op_u *u)
5257 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5258 stateid_t *stateid = &free_stateid->fr_stateid;
5259 struct nfs4_stid *s;
5260 struct nfs4_delegation *dp;
5261 struct nfs4_client *cl = cstate->session->se_client;
5262 __be32 ret = nfserr_bad_stateid;
5264 spin_lock(&cl->cl_lock);
5265 s = find_stateid_locked(cl, stateid);
5268 spin_lock(&s->sc_lock);
5269 switch (s->sc_type) {
5270 case NFS4_DELEG_STID:
5271 ret = nfserr_locks_held;
5273 case NFS4_OPEN_STID:
5274 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5277 ret = nfserr_locks_held;
5279 case NFS4_LOCK_STID:
5280 spin_unlock(&s->sc_lock);
5281 refcount_inc(&s->sc_count);
5282 spin_unlock(&cl->cl_lock);
5283 ret = nfsd4_free_lock_stateid(stateid, s);
5285 case NFS4_REVOKED_DELEG_STID:
5286 spin_unlock(&s->sc_lock);
5287 dp = delegstateid(s);
5288 list_del_init(&dp->dl_recall_lru);
5289 spin_unlock(&cl->cl_lock);
5293 /* Default falls through and returns nfserr_bad_stateid */
5295 spin_unlock(&s->sc_lock);
5297 spin_unlock(&cl->cl_lock);
5305 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5306 RD_STATE : WR_STATE;
5309 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5311 struct svc_fh *current_fh = &cstate->current_fh;
5312 struct nfs4_stateowner *sop = stp->st_stateowner;
5315 status = nfsd4_check_seqid(cstate, sop, seqid);
5318 status = nfsd4_lock_ol_stateid(stp);
5319 if (status != nfs_ok)
5321 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5322 if (status == nfs_ok)
5323 status = nfs4_check_fh(current_fh, &stp->st_stid);
5324 if (status != nfs_ok)
5325 mutex_unlock(&stp->st_mutex);
5330 * Checks for sequence id mutating operations.
5333 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5334 stateid_t *stateid, char typemask,
5335 struct nfs4_ol_stateid **stpp,
5336 struct nfsd_net *nn)
5339 struct nfs4_stid *s;
5340 struct nfs4_ol_stateid *stp = NULL;
5342 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5343 seqid, STATEID_VAL(stateid));
5346 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5349 stp = openlockstateid(s);
5350 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5352 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5356 nfs4_put_stid(&stp->st_stid);
5360 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5361 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5364 struct nfs4_openowner *oo;
5365 struct nfs4_ol_stateid *stp;
5367 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5368 NFS4_OPEN_STID, &stp, nn);
5371 oo = openowner(stp->st_stateowner);
5372 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5373 mutex_unlock(&stp->st_mutex);
5374 nfs4_put_stid(&stp->st_stid);
5375 return nfserr_bad_stateid;
5382 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5383 union nfsd4_op_u *u)
5385 struct nfsd4_open_confirm *oc = &u->open_confirm;
5387 struct nfs4_openowner *oo;
5388 struct nfs4_ol_stateid *stp;
5389 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5391 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5392 cstate->current_fh.fh_dentry);
5394 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5398 status = nfs4_preprocess_seqid_op(cstate,
5399 oc->oc_seqid, &oc->oc_req_stateid,
5400 NFS4_OPEN_STID, &stp, nn);
5403 oo = openowner(stp->st_stateowner);
5404 status = nfserr_bad_stateid;
5405 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5406 mutex_unlock(&stp->st_mutex);
5409 oo->oo_flags |= NFS4_OO_CONFIRMED;
5410 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5411 mutex_unlock(&stp->st_mutex);
5412 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5413 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5415 nfsd4_client_record_create(oo->oo_owner.so_client);
5418 nfs4_put_stid(&stp->st_stid);
5420 nfsd4_bump_seqid(cstate, status);
5424 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5426 if (!test_access(access, stp))
5428 nfs4_file_put_access(stp->st_stid.sc_file, access);
5429 clear_access(access, stp);
5432 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5434 switch (to_access) {
5435 case NFS4_SHARE_ACCESS_READ:
5436 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5437 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5439 case NFS4_SHARE_ACCESS_WRITE:
5440 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5441 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5443 case NFS4_SHARE_ACCESS_BOTH:
5451 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5452 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5454 struct nfsd4_open_downgrade *od = &u->open_downgrade;
5456 struct nfs4_ol_stateid *stp;
5457 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5459 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5460 cstate->current_fh.fh_dentry);
5462 /* We don't yet support WANT bits: */
5463 if (od->od_deleg_want)
5464 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5467 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5468 &od->od_stateid, &stp, nn);
5471 status = nfserr_inval;
5472 if (!test_access(od->od_share_access, stp)) {
5473 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5474 stp->st_access_bmap, od->od_share_access);
5477 if (!test_deny(od->od_share_deny, stp)) {
5478 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5479 stp->st_deny_bmap, od->od_share_deny);
5482 nfs4_stateid_downgrade(stp, od->od_share_access);
5483 reset_union_bmap_deny(od->od_share_deny, stp);
5484 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5487 mutex_unlock(&stp->st_mutex);
5488 nfs4_put_stid(&stp->st_stid);
5490 nfsd4_bump_seqid(cstate, status);
5494 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5496 struct nfs4_client *clp = s->st_stid.sc_client;
5498 LIST_HEAD(reaplist);
5500 spin_lock(&clp->cl_lock);
5501 unhashed = unhash_open_stateid(s, &reaplist);
5503 if (clp->cl_minorversion) {
5505 put_ol_stateid_locked(s, &reaplist);
5506 spin_unlock(&clp->cl_lock);
5507 free_ol_stateid_reaplist(&reaplist);
5509 spin_unlock(&clp->cl_lock);
5510 free_ol_stateid_reaplist(&reaplist);
5512 move_to_close_lru(s, clp->net);
5517 * nfs4_unlock_state() called after encode
5520 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5521 union nfsd4_op_u *u)
5523 struct nfsd4_close *close = &u->close;
5525 struct nfs4_ol_stateid *stp;
5526 struct net *net = SVC_NET(rqstp);
5527 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5529 dprintk("NFSD: nfsd4_close on file %pd\n",
5530 cstate->current_fh.fh_dentry);
5532 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5534 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5536 nfsd4_bump_seqid(cstate, status);
5540 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5543 * Technically we don't _really_ have to increment or copy it, since
5544 * it should just be gone after this operation and we clobber the
5545 * copied value below, but we continue to do so here just to ensure
5546 * that racing ops see that there was a state change.
5548 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5550 nfsd4_close_open_stateid(stp);
5551 mutex_unlock(&stp->st_mutex);
5553 /* v4.1+ suggests that we send a special stateid in here, since the
5554 * clients should just ignore this anyway. Since this is not useful
5555 * for v4.0 clients either, we set it to the special close_stateid
5558 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
5560 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
5562 /* put reference from nfs4_preprocess_seqid_op */
5563 nfs4_put_stid(&stp->st_stid);
5569 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5570 union nfsd4_op_u *u)
5572 struct nfsd4_delegreturn *dr = &u->delegreturn;
5573 struct nfs4_delegation *dp;
5574 stateid_t *stateid = &dr->dr_stateid;
5575 struct nfs4_stid *s;
5577 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5579 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5582 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5585 dp = delegstateid(s);
5586 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5590 destroy_delegation(dp);
5592 nfs4_put_stid(&dp->dl_stid);
5598 end_offset(u64 start, u64 len)
5603 return end >= start ? end: NFS4_MAX_UINT64;
5606 /* last octet in a range */
5608 last_byte_offset(u64 start, u64 len)
5614 return end > start ? end - 1: NFS4_MAX_UINT64;
5618 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5619 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5620 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5621 * locking, this prevents us from being completely protocol-compliant. The
5622 * real solution to this problem is to start using unsigned file offsets in
5623 * the VFS, but this is a very deep change!
5626 nfs4_transform_lock_offset(struct file_lock *lock)
5628 if (lock->fl_start < 0)
5629 lock->fl_start = OFFSET_MAX;
5630 if (lock->fl_end < 0)
5631 lock->fl_end = OFFSET_MAX;
5635 nfsd4_fl_get_owner(fl_owner_t owner)
5637 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5639 nfs4_get_stateowner(&lo->lo_owner);
5644 nfsd4_fl_put_owner(fl_owner_t owner)
5646 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5649 nfs4_put_stateowner(&lo->lo_owner);
5653 nfsd4_lm_notify(struct file_lock *fl)
5655 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
5656 struct net *net = lo->lo_owner.so_client->net;
5657 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5658 struct nfsd4_blocked_lock *nbl = container_of(fl,
5659 struct nfsd4_blocked_lock, nbl_lock);
5662 /* An empty list means that something else is going to be using it */
5663 spin_lock(&nn->blocked_locks_lock);
5664 if (!list_empty(&nbl->nbl_list)) {
5665 list_del_init(&nbl->nbl_list);
5666 list_del_init(&nbl->nbl_lru);
5669 spin_unlock(&nn->blocked_locks_lock);
5672 nfsd4_run_cb(&nbl->nbl_cb);
5675 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5676 .lm_notify = nfsd4_lm_notify,
5677 .lm_get_owner = nfsd4_fl_get_owner,
5678 .lm_put_owner = nfsd4_fl_put_owner,
5682 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5684 struct nfs4_lockowner *lo;
5686 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5687 lo = (struct nfs4_lockowner *) fl->fl_owner;
5688 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5689 lo->lo_owner.so_owner.len, GFP_KERNEL);
5690 if (!deny->ld_owner.data)
5691 /* We just don't care that much */
5693 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5694 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5697 deny->ld_owner.len = 0;
5698 deny->ld_owner.data = NULL;
5699 deny->ld_clientid.cl_boot = 0;
5700 deny->ld_clientid.cl_id = 0;
5702 deny->ld_start = fl->fl_start;
5703 deny->ld_length = NFS4_MAX_UINT64;
5704 if (fl->fl_end != NFS4_MAX_UINT64)
5705 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5706 deny->ld_type = NFS4_READ_LT;
5707 if (fl->fl_type != F_RDLCK)
5708 deny->ld_type = NFS4_WRITE_LT;
5711 static struct nfs4_lockowner *
5712 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5714 unsigned int strhashval = ownerstr_hashval(owner);
5715 struct nfs4_stateowner *so;
5717 lockdep_assert_held(&clp->cl_lock);
5719 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5721 if (so->so_is_open_owner)
5723 if (same_owner_str(so, owner))
5724 return lockowner(nfs4_get_stateowner(so));
5729 static struct nfs4_lockowner *
5730 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5732 struct nfs4_lockowner *lo;
5734 spin_lock(&clp->cl_lock);
5735 lo = find_lockowner_str_locked(clp, owner);
5736 spin_unlock(&clp->cl_lock);
5740 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5742 unhash_lockowner_locked(lockowner(sop));
5745 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5747 struct nfs4_lockowner *lo = lockowner(sop);
5749 kmem_cache_free(lockowner_slab, lo);
5752 static const struct nfs4_stateowner_operations lockowner_ops = {
5753 .so_unhash = nfs4_unhash_lockowner,
5754 .so_free = nfs4_free_lockowner,
5758 * Alloc a lock owner structure.
5759 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5762 * strhashval = ownerstr_hashval
5764 static struct nfs4_lockowner *
5765 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5766 struct nfs4_ol_stateid *open_stp,
5767 struct nfsd4_lock *lock)
5769 struct nfs4_lockowner *lo, *ret;
5771 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5774 INIT_LIST_HEAD(&lo->lo_blocked);
5775 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5776 lo->lo_owner.so_is_open_owner = 0;
5777 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5778 lo->lo_owner.so_ops = &lockowner_ops;
5779 spin_lock(&clp->cl_lock);
5780 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5782 list_add(&lo->lo_owner.so_strhash,
5783 &clp->cl_ownerstr_hashtbl[strhashval]);
5786 nfs4_free_stateowner(&lo->lo_owner);
5788 spin_unlock(&clp->cl_lock);
5792 static struct nfs4_ol_stateid *
5793 find_lock_stateid(const struct nfs4_lockowner *lo,
5794 const struct nfs4_ol_stateid *ost)
5796 struct nfs4_ol_stateid *lst;
5798 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
5800 /* If ost is not hashed, ost->st_locks will not be valid */
5801 if (!nfs4_ol_stateid_unhashed(ost))
5802 list_for_each_entry(lst, &ost->st_locks, st_locks) {
5803 if (lst->st_stateowner == &lo->lo_owner) {
5804 refcount_inc(&lst->st_stid.sc_count);
5811 static struct nfs4_ol_stateid *
5812 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5813 struct nfs4_file *fp, struct inode *inode,
5814 struct nfs4_ol_stateid *open_stp)
5816 struct nfs4_client *clp = lo->lo_owner.so_client;
5817 struct nfs4_ol_stateid *retstp;
5819 mutex_init(&stp->st_mutex);
5820 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5822 spin_lock(&clp->cl_lock);
5823 if (nfs4_ol_stateid_unhashed(open_stp))
5825 retstp = find_lock_stateid(lo, open_stp);
5828 refcount_inc(&stp->st_stid.sc_count);
5829 stp->st_stid.sc_type = NFS4_LOCK_STID;
5830 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5832 stp->st_stid.sc_file = fp;
5833 stp->st_access_bmap = 0;
5834 stp->st_deny_bmap = open_stp->st_deny_bmap;
5835 stp->st_openstp = open_stp;
5836 spin_lock(&fp->fi_lock);
5837 list_add(&stp->st_locks, &open_stp->st_locks);
5838 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5839 list_add(&stp->st_perfile, &fp->fi_stateids);
5840 spin_unlock(&fp->fi_lock);
5841 spin_unlock(&clp->cl_lock);
5844 spin_unlock(&clp->cl_lock);
5845 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5846 nfs4_put_stid(&retstp->st_stid);
5849 /* To keep mutex tracking happy */
5850 mutex_unlock(&stp->st_mutex);
5853 spin_unlock(&clp->cl_lock);
5854 mutex_unlock(&stp->st_mutex);
5858 static struct nfs4_ol_stateid *
5859 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5860 struct inode *inode, struct nfs4_ol_stateid *ost,
5863 struct nfs4_stid *ns = NULL;
5864 struct nfs4_ol_stateid *lst;
5865 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5866 struct nfs4_client *clp = oo->oo_owner.so_client;
5869 spin_lock(&clp->cl_lock);
5870 lst = find_lock_stateid(lo, ost);
5871 spin_unlock(&clp->cl_lock);
5873 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5875 nfs4_put_stid(&lst->st_stid);
5877 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5881 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5882 if (lst == openlockstateid(ns))
5891 check_lock_length(u64 offset, u64 length)
5893 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5894 (length > ~offset)));
5897 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5899 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5901 lockdep_assert_held(&fp->fi_lock);
5903 if (test_access(access, lock_stp))
5905 __nfs4_file_get_access(fp, access);
5906 set_access(access, lock_stp);
5910 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5911 struct nfs4_ol_stateid *ost,
5912 struct nfsd4_lock *lock,
5913 struct nfs4_ol_stateid **plst, bool *new)
5916 struct nfs4_file *fi = ost->st_stid.sc_file;
5917 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5918 struct nfs4_client *cl = oo->oo_owner.so_client;
5919 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5920 struct nfs4_lockowner *lo;
5921 struct nfs4_ol_stateid *lst;
5922 unsigned int strhashval;
5924 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5926 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5927 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5929 return nfserr_jukebox;
5931 /* with an existing lockowner, seqids must be the same */
5932 status = nfserr_bad_seqid;
5933 if (!cstate->minorversion &&
5934 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5938 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5940 status = nfserr_jukebox;
5947 nfs4_put_stateowner(&lo->lo_owner);
5955 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5956 union nfsd4_op_u *u)
5958 struct nfsd4_lock *lock = &u->lock;
5959 struct nfs4_openowner *open_sop = NULL;
5960 struct nfs4_lockowner *lock_sop = NULL;
5961 struct nfs4_ol_stateid *lock_stp = NULL;
5962 struct nfs4_ol_stateid *open_stp = NULL;
5963 struct nfs4_file *fp;
5964 struct file *filp = NULL;
5965 struct nfsd4_blocked_lock *nbl = NULL;
5966 struct file_lock *file_lock = NULL;
5967 struct file_lock *conflock = NULL;
5972 unsigned char fl_type;
5973 unsigned int fl_flags = FL_POSIX;
5974 struct net *net = SVC_NET(rqstp);
5975 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5977 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5978 (long long) lock->lk_offset,
5979 (long long) lock->lk_length);
5981 if (check_lock_length(lock->lk_offset, lock->lk_length))
5982 return nfserr_inval;
5984 if ((status = fh_verify(rqstp, &cstate->current_fh,
5985 S_IFREG, NFSD_MAY_LOCK))) {
5986 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5990 if (lock->lk_is_new) {
5991 if (nfsd4_has_session(cstate))
5992 /* See rfc 5661 18.10.3: given clientid is ignored: */
5993 memcpy(&lock->lk_new_clientid,
5994 &cstate->session->se_client->cl_clientid,
5995 sizeof(clientid_t));
5997 status = nfserr_stale_clientid;
5998 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6001 /* validate and update open stateid and open seqid */
6002 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6003 lock->lk_new_open_seqid,
6004 &lock->lk_new_open_stateid,
6008 mutex_unlock(&open_stp->st_mutex);
6009 open_sop = openowner(open_stp->st_stateowner);
6010 status = nfserr_bad_stateid;
6011 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6012 &lock->lk_new_clientid))
6014 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6017 status = nfs4_preprocess_seqid_op(cstate,
6018 lock->lk_old_lock_seqid,
6019 &lock->lk_old_lock_stateid,
6020 NFS4_LOCK_STID, &lock_stp, nn);
6024 lock_sop = lockowner(lock_stp->st_stateowner);
6026 lkflg = setlkflg(lock->lk_type);
6027 status = nfs4_check_openmode(lock_stp, lkflg);
6031 status = nfserr_grace;
6032 if (locks_in_grace(net) && !lock->lk_reclaim)
6034 status = nfserr_no_grace;
6035 if (!locks_in_grace(net) && lock->lk_reclaim)
6038 fp = lock_stp->st_stid.sc_file;
6039 switch (lock->lk_type) {
6041 if (nfsd4_has_session(cstate))
6042 fl_flags |= FL_SLEEP;
6045 spin_lock(&fp->fi_lock);
6046 filp = find_readable_file_locked(fp);
6048 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6049 spin_unlock(&fp->fi_lock);
6052 case NFS4_WRITEW_LT:
6053 if (nfsd4_has_session(cstate))
6054 fl_flags |= FL_SLEEP;
6057 spin_lock(&fp->fi_lock);
6058 filp = find_writeable_file_locked(fp);
6060 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6061 spin_unlock(&fp->fi_lock);
6065 status = nfserr_inval;
6070 status = nfserr_openmode;
6074 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6076 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6077 status = nfserr_jukebox;
6081 file_lock = &nbl->nbl_lock;
6082 file_lock->fl_type = fl_type;
6083 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6084 file_lock->fl_pid = current->tgid;
6085 file_lock->fl_file = filp;
6086 file_lock->fl_flags = fl_flags;
6087 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6088 file_lock->fl_start = lock->lk_offset;
6089 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6090 nfs4_transform_lock_offset(file_lock);
6092 conflock = locks_alloc_lock();
6094 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6095 status = nfserr_jukebox;
6099 if (fl_flags & FL_SLEEP) {
6100 nbl->nbl_time = get_seconds();
6101 spin_lock(&nn->blocked_locks_lock);
6102 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6103 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6104 spin_unlock(&nn->blocked_locks_lock);
6107 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6109 case 0: /* success! */
6110 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6112 if (lock->lk_reclaim)
6113 nn->somebody_reclaimed = true;
6115 case FILE_LOCK_DEFERRED:
6118 case -EAGAIN: /* conflock holds conflicting lock */
6119 status = nfserr_denied;
6120 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6121 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6124 status = nfserr_deadlock;
6127 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6128 status = nfserrno(err);
6133 /* dequeue it if we queued it before */
6134 if (fl_flags & FL_SLEEP) {
6135 spin_lock(&nn->blocked_locks_lock);
6136 list_del_init(&nbl->nbl_list);
6137 list_del_init(&nbl->nbl_lru);
6138 spin_unlock(&nn->blocked_locks_lock);
6140 free_blocked_lock(nbl);
6145 /* Bump seqid manually if the 4.0 replay owner is openowner */
6146 if (cstate->replay_owner &&
6147 cstate->replay_owner != &lock_sop->lo_owner &&
6148 seqid_mutating_err(ntohl(status)))
6149 lock_sop->lo_owner.so_seqid++;
6152 * If this is a new, never-before-used stateid, and we are
6153 * returning an error, then just go ahead and release it.
6156 release_lock_stateid(lock_stp);
6158 mutex_unlock(&lock_stp->st_mutex);
6160 nfs4_put_stid(&lock_stp->st_stid);
6163 nfs4_put_stid(&open_stp->st_stid);
6164 nfsd4_bump_seqid(cstate, status);
6166 locks_free_lock(conflock);
6171 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6172 * so we do a temporary open here just to get an open file to pass to
6173 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6176 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6179 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6181 err = nfserrno(vfs_test_lock(file, lock));
6191 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6192 union nfsd4_op_u *u)
6194 struct nfsd4_lockt *lockt = &u->lockt;
6195 struct file_lock *file_lock = NULL;
6196 struct nfs4_lockowner *lo = NULL;
6198 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6200 if (locks_in_grace(SVC_NET(rqstp)))
6201 return nfserr_grace;
6203 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6204 return nfserr_inval;
6206 if (!nfsd4_has_session(cstate)) {
6207 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6212 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6215 file_lock = locks_alloc_lock();
6217 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6218 status = nfserr_jukebox;
6222 switch (lockt->lt_type) {
6225 file_lock->fl_type = F_RDLCK;
6228 case NFS4_WRITEW_LT:
6229 file_lock->fl_type = F_WRLCK;
6232 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6233 status = nfserr_inval;
6237 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6239 file_lock->fl_owner = (fl_owner_t)lo;
6240 file_lock->fl_pid = current->tgid;
6241 file_lock->fl_flags = FL_POSIX;
6243 file_lock->fl_start = lockt->lt_offset;
6244 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6246 nfs4_transform_lock_offset(file_lock);
6248 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6252 if (file_lock->fl_type != F_UNLCK) {
6253 status = nfserr_denied;
6254 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6258 nfs4_put_stateowner(&lo->lo_owner);
6260 locks_free_lock(file_lock);
6265 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6266 union nfsd4_op_u *u)
6268 struct nfsd4_locku *locku = &u->locku;
6269 struct nfs4_ol_stateid *stp;
6270 struct file *filp = NULL;
6271 struct file_lock *file_lock = NULL;
6274 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6276 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6277 (long long) locku->lu_offset,
6278 (long long) locku->lu_length);
6280 if (check_lock_length(locku->lu_offset, locku->lu_length))
6281 return nfserr_inval;
6283 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6284 &locku->lu_stateid, NFS4_LOCK_STID,
6288 filp = find_any_file(stp->st_stid.sc_file);
6290 status = nfserr_lock_range;
6293 file_lock = locks_alloc_lock();
6295 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6296 status = nfserr_jukebox;
6300 file_lock->fl_type = F_UNLCK;
6301 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6302 file_lock->fl_pid = current->tgid;
6303 file_lock->fl_file = filp;
6304 file_lock->fl_flags = FL_POSIX;
6305 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6306 file_lock->fl_start = locku->lu_offset;
6308 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6310 nfs4_transform_lock_offset(file_lock);
6312 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6314 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6317 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6321 mutex_unlock(&stp->st_mutex);
6322 nfs4_put_stid(&stp->st_stid);
6324 nfsd4_bump_seqid(cstate, status);
6326 locks_free_lock(file_lock);
6330 status = nfserrno(err);
6336 * true: locks held by lockowner
6337 * false: no locks held by lockowner
6340 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6342 struct file_lock *fl;
6344 struct file *filp = find_any_file(fp);
6345 struct inode *inode;
6346 struct file_lock_context *flctx;
6349 /* Any valid lock stateid should have some sort of access */
6354 inode = locks_inode(filp);
6355 flctx = inode->i_flctx;
6357 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6358 spin_lock(&flctx->flc_lock);
6359 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6360 if (fl->fl_owner == (fl_owner_t)lowner) {
6365 spin_unlock(&flctx->flc_lock);
6372 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6373 struct nfsd4_compound_state *cstate,
6374 union nfsd4_op_u *u)
6376 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6377 clientid_t *clid = &rlockowner->rl_clientid;
6378 struct nfs4_stateowner *sop;
6379 struct nfs4_lockowner *lo = NULL;
6380 struct nfs4_ol_stateid *stp;
6381 struct xdr_netobj *owner = &rlockowner->rl_owner;
6382 unsigned int hashval = ownerstr_hashval(owner);
6384 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6385 struct nfs4_client *clp;
6386 LIST_HEAD (reaplist);
6388 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6389 clid->cl_boot, clid->cl_id);
6391 status = lookup_clientid(clid, cstate, nn);
6396 /* Find the matching lock stateowner */
6397 spin_lock(&clp->cl_lock);
6398 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6401 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6404 if (atomic_read(&sop->so_count) != 1) {
6405 spin_unlock(&clp->cl_lock);
6406 return nfserr_locks_held;
6409 lo = lockowner(sop);
6410 nfs4_get_stateowner(sop);
6414 spin_unlock(&clp->cl_lock);
6418 unhash_lockowner_locked(lo);
6419 while (!list_empty(&lo->lo_owner.so_stateids)) {
6420 stp = list_first_entry(&lo->lo_owner.so_stateids,
6421 struct nfs4_ol_stateid,
6423 WARN_ON(!unhash_lock_stateid(stp));
6424 put_ol_stateid_locked(stp, &reaplist);
6426 spin_unlock(&clp->cl_lock);
6427 free_ol_stateid_reaplist(&reaplist);
6428 remove_blocked_locks(lo);
6429 nfs4_put_stateowner(&lo->lo_owner);
6434 static inline struct nfs4_client_reclaim *
6437 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6441 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6443 struct nfs4_client_reclaim *crp;
6445 crp = nfsd4_find_reclaim_client(name, nn);
6446 return (crp && crp->cr_clp);
6450 * failure => all reset bets are off, nfserr_no_grace...
6452 struct nfs4_client_reclaim *
6453 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6455 unsigned int strhashval;
6456 struct nfs4_client_reclaim *crp;
6458 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6459 crp = alloc_reclaim();
6461 strhashval = clientstr_hashval(name);
6462 INIT_LIST_HEAD(&crp->cr_strhash);
6463 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6464 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6466 nn->reclaim_str_hashtbl_size++;
6472 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6474 list_del(&crp->cr_strhash);
6476 nn->reclaim_str_hashtbl_size--;
6480 nfs4_release_reclaim(struct nfsd_net *nn)
6482 struct nfs4_client_reclaim *crp = NULL;
6485 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6486 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6487 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6488 struct nfs4_client_reclaim, cr_strhash);
6489 nfs4_remove_reclaim_record(crp, nn);
6492 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6496 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6497 struct nfs4_client_reclaim *
6498 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6500 unsigned int strhashval;
6501 struct nfs4_client_reclaim *crp = NULL;
6503 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6505 strhashval = clientstr_hashval(recdir);
6506 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6507 if (same_name(crp->cr_recdir, recdir)) {
6515 * Called from OPEN. Look for clientid in reclaim list.
6518 nfs4_check_open_reclaim(clientid_t *clid,
6519 struct nfsd4_compound_state *cstate,
6520 struct nfsd_net *nn)
6524 /* find clientid in conf_id_hashtbl */
6525 status = lookup_clientid(clid, cstate, nn);
6527 return nfserr_reclaim_bad;
6529 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6530 return nfserr_no_grace;
6532 if (nfsd4_client_record_check(cstate->clp))
6533 return nfserr_reclaim_bad;
6538 #ifdef CONFIG_NFSD_FAULT_INJECTION
6540 put_client(struct nfs4_client *clp)
6542 atomic_dec(&clp->cl_refcount);
6545 static struct nfs4_client *
6546 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6548 struct nfs4_client *clp;
6549 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6552 if (!nfsd_netns_ready(nn))
6555 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6556 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6563 nfsd_inject_print_clients(void)
6565 struct nfs4_client *clp;
6567 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6569 char buf[INET6_ADDRSTRLEN];
6571 if (!nfsd_netns_ready(nn))
6574 spin_lock(&nn->client_lock);
6575 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6576 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6577 pr_info("NFS Client: %s\n", buf);
6580 spin_unlock(&nn->client_lock);
6586 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6589 struct nfs4_client *clp;
6590 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6593 if (!nfsd_netns_ready(nn))
6596 spin_lock(&nn->client_lock);
6597 clp = nfsd_find_client(addr, addr_size);
6599 if (mark_client_expired_locked(clp) == nfs_ok)
6604 spin_unlock(&nn->client_lock);
6613 nfsd_inject_forget_clients(u64 max)
6616 struct nfs4_client *clp, *next;
6617 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6619 LIST_HEAD(reaplist);
6621 if (!nfsd_netns_ready(nn))
6624 spin_lock(&nn->client_lock);
6625 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6626 if (mark_client_expired_locked(clp) == nfs_ok) {
6627 list_add(&clp->cl_lru, &reaplist);
6628 if (max != 0 && ++count >= max)
6632 spin_unlock(&nn->client_lock);
6634 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6640 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6643 char buf[INET6_ADDRSTRLEN];
6644 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6645 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6649 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6650 struct list_head *collect)
6652 struct nfs4_client *clp = lst->st_stid.sc_client;
6653 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6659 lockdep_assert_held(&nn->client_lock);
6660 atomic_inc(&clp->cl_refcount);
6661 list_add(&lst->st_locks, collect);
6664 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6665 struct list_head *collect,
6666 bool (*func)(struct nfs4_ol_stateid *))
6668 struct nfs4_openowner *oop;
6669 struct nfs4_ol_stateid *stp, *st_next;
6670 struct nfs4_ol_stateid *lst, *lst_next;
6673 spin_lock(&clp->cl_lock);
6674 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6675 list_for_each_entry_safe(stp, st_next,
6676 &oop->oo_owner.so_stateids, st_perstateowner) {
6677 list_for_each_entry_safe(lst, lst_next,
6678 &stp->st_locks, st_locks) {
6681 nfsd_inject_add_lock_to_list(lst,
6686 * Despite the fact that these functions deal
6687 * with 64-bit integers for "count", we must
6688 * ensure that it doesn't blow up the
6689 * clp->cl_refcount. Throw a warning if we
6690 * start to approach INT_MAX here.
6692 WARN_ON_ONCE(count == (INT_MAX / 2));
6699 spin_unlock(&clp->cl_lock);
6705 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6708 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6712 nfsd_print_client_locks(struct nfs4_client *clp)
6714 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6715 nfsd_print_count(clp, count, "locked files");
6720 nfsd_inject_print_locks(void)
6722 struct nfs4_client *clp;
6724 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6727 if (!nfsd_netns_ready(nn))
6730 spin_lock(&nn->client_lock);
6731 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6732 count += nfsd_print_client_locks(clp);
6733 spin_unlock(&nn->client_lock);
6739 nfsd_reap_locks(struct list_head *reaplist)
6741 struct nfs4_client *clp;
6742 struct nfs4_ol_stateid *stp, *next;
6744 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6745 list_del_init(&stp->st_locks);
6746 clp = stp->st_stid.sc_client;
6747 nfs4_put_stid(&stp->st_stid);
6753 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6755 unsigned int count = 0;
6756 struct nfs4_client *clp;
6757 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6759 LIST_HEAD(reaplist);
6761 if (!nfsd_netns_ready(nn))
6764 spin_lock(&nn->client_lock);
6765 clp = nfsd_find_client(addr, addr_size);
6767 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6768 spin_unlock(&nn->client_lock);
6769 nfsd_reap_locks(&reaplist);
6774 nfsd_inject_forget_locks(u64 max)
6777 struct nfs4_client *clp;
6778 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6780 LIST_HEAD(reaplist);
6782 if (!nfsd_netns_ready(nn))
6785 spin_lock(&nn->client_lock);
6786 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6787 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6788 if (max != 0 && count >= max)
6791 spin_unlock(&nn->client_lock);
6792 nfsd_reap_locks(&reaplist);
6797 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6798 struct list_head *collect,
6799 void (*func)(struct nfs4_openowner *))
6801 struct nfs4_openowner *oop, *next;
6802 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6806 lockdep_assert_held(&nn->client_lock);
6808 spin_lock(&clp->cl_lock);
6809 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6813 atomic_inc(&clp->cl_refcount);
6814 list_add(&oop->oo_perclient, collect);
6819 * Despite the fact that these functions deal with
6820 * 64-bit integers for "count", we must ensure that
6821 * it doesn't blow up the clp->cl_refcount. Throw a
6822 * warning if we start to approach INT_MAX here.
6824 WARN_ON_ONCE(count == (INT_MAX / 2));
6828 spin_unlock(&clp->cl_lock);
6834 nfsd_print_client_openowners(struct nfs4_client *clp)
6836 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6838 nfsd_print_count(clp, count, "openowners");
6843 nfsd_collect_client_openowners(struct nfs4_client *clp,
6844 struct list_head *collect, u64 max)
6846 return nfsd_foreach_client_openowner(clp, max, collect,
6847 unhash_openowner_locked);
6851 nfsd_inject_print_openowners(void)
6853 struct nfs4_client *clp;
6855 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6858 if (!nfsd_netns_ready(nn))
6861 spin_lock(&nn->client_lock);
6862 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6863 count += nfsd_print_client_openowners(clp);
6864 spin_unlock(&nn->client_lock);
6870 nfsd_reap_openowners(struct list_head *reaplist)
6872 struct nfs4_client *clp;
6873 struct nfs4_openowner *oop, *next;
6875 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6876 list_del_init(&oop->oo_perclient);
6877 clp = oop->oo_owner.so_client;
6878 release_openowner(oop);
6884 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6887 unsigned int count = 0;
6888 struct nfs4_client *clp;
6889 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6891 LIST_HEAD(reaplist);
6893 if (!nfsd_netns_ready(nn))
6896 spin_lock(&nn->client_lock);
6897 clp = nfsd_find_client(addr, addr_size);
6899 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6900 spin_unlock(&nn->client_lock);
6901 nfsd_reap_openowners(&reaplist);
6906 nfsd_inject_forget_openowners(u64 max)
6909 struct nfs4_client *clp;
6910 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6912 LIST_HEAD(reaplist);
6914 if (!nfsd_netns_ready(nn))
6917 spin_lock(&nn->client_lock);
6918 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6919 count += nfsd_collect_client_openowners(clp, &reaplist,
6921 if (max != 0 && count >= max)
6924 spin_unlock(&nn->client_lock);
6925 nfsd_reap_openowners(&reaplist);
6929 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6930 struct list_head *victims)
6932 struct nfs4_delegation *dp, *next;
6933 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6937 lockdep_assert_held(&nn->client_lock);
6939 spin_lock(&state_lock);
6940 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6943 * It's not safe to mess with delegations that have a
6944 * non-zero dl_time. They might have already been broken
6945 * and could be processed by the laundromat outside of
6946 * the state_lock. Just leave them be.
6948 if (dp->dl_time != 0)
6951 atomic_inc(&clp->cl_refcount);
6952 WARN_ON(!unhash_delegation_locked(dp));
6953 list_add(&dp->dl_recall_lru, victims);
6957 * Despite the fact that these functions deal with
6958 * 64-bit integers for "count", we must ensure that
6959 * it doesn't blow up the clp->cl_refcount. Throw a
6960 * warning if we start to approach INT_MAX here.
6962 WARN_ON_ONCE(count == (INT_MAX / 2));
6966 spin_unlock(&state_lock);
6971 nfsd_print_client_delegations(struct nfs4_client *clp)
6973 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6975 nfsd_print_count(clp, count, "delegations");
6980 nfsd_inject_print_delegations(void)
6982 struct nfs4_client *clp;
6984 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6987 if (!nfsd_netns_ready(nn))
6990 spin_lock(&nn->client_lock);
6991 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6992 count += nfsd_print_client_delegations(clp);
6993 spin_unlock(&nn->client_lock);
6999 nfsd_forget_delegations(struct list_head *reaplist)
7001 struct nfs4_client *clp;
7002 struct nfs4_delegation *dp, *next;
7004 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7005 list_del_init(&dp->dl_recall_lru);
7006 clp = dp->dl_stid.sc_client;
7007 revoke_delegation(dp);
7013 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7017 struct nfs4_client *clp;
7018 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7020 LIST_HEAD(reaplist);
7022 if (!nfsd_netns_ready(nn))
7025 spin_lock(&nn->client_lock);
7026 clp = nfsd_find_client(addr, addr_size);
7028 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7029 spin_unlock(&nn->client_lock);
7031 nfsd_forget_delegations(&reaplist);
7036 nfsd_inject_forget_delegations(u64 max)
7039 struct nfs4_client *clp;
7040 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7042 LIST_HEAD(reaplist);
7044 if (!nfsd_netns_ready(nn))
7047 spin_lock(&nn->client_lock);
7048 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7049 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7050 if (max != 0 && count >= max)
7053 spin_unlock(&nn->client_lock);
7054 nfsd_forget_delegations(&reaplist);
7059 nfsd_recall_delegations(struct list_head *reaplist)
7061 struct nfs4_client *clp;
7062 struct nfs4_delegation *dp, *next;
7064 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7065 list_del_init(&dp->dl_recall_lru);
7066 clp = dp->dl_stid.sc_client;
7068 * We skipped all entries that had a zero dl_time before,
7069 * so we can now reset the dl_time back to 0. If a delegation
7070 * break comes in now, then it won't make any difference since
7071 * we're recalling it either way.
7073 spin_lock(&state_lock);
7075 spin_unlock(&state_lock);
7076 nfsd_break_one_deleg(dp);
7082 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7086 struct nfs4_client *clp;
7087 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7089 LIST_HEAD(reaplist);
7091 if (!nfsd_netns_ready(nn))
7094 spin_lock(&nn->client_lock);
7095 clp = nfsd_find_client(addr, addr_size);
7097 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7098 spin_unlock(&nn->client_lock);
7100 nfsd_recall_delegations(&reaplist);
7105 nfsd_inject_recall_delegations(u64 max)
7108 struct nfs4_client *clp, *next;
7109 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7111 LIST_HEAD(reaplist);
7113 if (!nfsd_netns_ready(nn))
7116 spin_lock(&nn->client_lock);
7117 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7118 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7119 if (max != 0 && ++count >= max)
7122 spin_unlock(&nn->client_lock);
7123 nfsd_recall_delegations(&reaplist);
7126 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7129 * Since the lifetime of a delegation isn't limited to that of an open, a
7130 * client may quite reasonably hang on to a delegation as long as it has
7131 * the inode cached. This becomes an obvious problem the first time a
7132 * client's inode cache approaches the size of the server's total memory.
7134 * For now we avoid this problem by imposing a hard limit on the number
7135 * of delegations, which varies according to the server's memory size.
7138 set_max_delegations(void)
7141 * Allow at most 4 delegations per megabyte of RAM. Quick
7142 * estimates suggest that in the worst case (where every delegation
7143 * is for a different inode), a delegation could take about 1.5K,
7144 * giving a worst case usage of about 6% of memory.
7146 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7149 static int nfs4_state_create_net(struct net *net)
7151 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7154 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7155 sizeof(struct list_head),
7157 if (!nn->conf_id_hashtbl)
7159 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7160 sizeof(struct list_head),
7162 if (!nn->unconf_id_hashtbl)
7164 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7165 sizeof(struct list_head),
7167 if (!nn->sessionid_hashtbl)
7170 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7171 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7172 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7174 for (i = 0; i < SESSION_HASH_SIZE; i++)
7175 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7176 nn->conf_name_tree = RB_ROOT;
7177 nn->unconf_name_tree = RB_ROOT;
7178 nn->boot_time = get_seconds();
7179 nn->grace_ended = false;
7180 nn->nfsd4_manager.block_opens = true;
7181 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7182 INIT_LIST_HEAD(&nn->client_lru);
7183 INIT_LIST_HEAD(&nn->close_lru);
7184 INIT_LIST_HEAD(&nn->del_recall_lru);
7185 spin_lock_init(&nn->client_lock);
7187 spin_lock_init(&nn->blocked_locks_lock);
7188 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7190 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7196 kfree(nn->unconf_id_hashtbl);
7198 kfree(nn->conf_id_hashtbl);
7204 nfs4_state_destroy_net(struct net *net)
7207 struct nfs4_client *clp = NULL;
7208 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7210 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7211 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7212 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7213 destroy_client(clp);
7217 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7219 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7220 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7221 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7222 destroy_client(clp);
7226 kfree(nn->sessionid_hashtbl);
7227 kfree(nn->unconf_id_hashtbl);
7228 kfree(nn->conf_id_hashtbl);
7233 nfs4_state_start_net(struct net *net)
7235 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7238 ret = nfs4_state_create_net(net);
7241 locks_start_grace(net, &nn->nfsd4_manager);
7242 nfsd4_client_tracking_init(net);
7243 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7244 nn->nfsd4_grace, net->ns.inum);
7245 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7249 /* initialization to perform when the nfsd service is started: */
7252 nfs4_state_start(void)
7256 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7257 if (laundry_wq == NULL) {
7261 ret = nfsd4_create_callback_queue();
7263 goto out_free_laundry;
7265 set_max_delegations();
7269 destroy_workqueue(laundry_wq);
7275 nfs4_state_shutdown_net(struct net *net)
7277 struct nfs4_delegation *dp = NULL;
7278 struct list_head *pos, *next, reaplist;
7279 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7281 cancel_delayed_work_sync(&nn->laundromat_work);
7282 locks_end_grace(&nn->nfsd4_manager);
7284 INIT_LIST_HEAD(&reaplist);
7285 spin_lock(&state_lock);
7286 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7287 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7288 WARN_ON(!unhash_delegation_locked(dp));
7289 list_add(&dp->dl_recall_lru, &reaplist);
7291 spin_unlock(&state_lock);
7292 list_for_each_safe(pos, next, &reaplist) {
7293 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7294 list_del_init(&dp->dl_recall_lru);
7295 destroy_unhashed_deleg(dp);
7298 nfsd4_client_tracking_exit(net);
7299 nfs4_state_destroy_net(net);
7303 nfs4_state_shutdown(void)
7305 destroy_workqueue(laundry_wq);
7306 nfsd4_destroy_callback_queue();
7310 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7312 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7313 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7317 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7319 if (cstate->minorversion) {
7320 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7321 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7326 clear_current_stateid(struct nfsd4_compound_state *cstate)
7328 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7332 * functions to set current state id
7335 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7336 union nfsd4_op_u *u)
7338 put_stateid(cstate, &u->open_downgrade.od_stateid);
7342 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7343 union nfsd4_op_u *u)
7345 put_stateid(cstate, &u->open.op_stateid);
7349 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7350 union nfsd4_op_u *u)
7352 put_stateid(cstate, &u->close.cl_stateid);
7356 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7357 union nfsd4_op_u *u)
7359 put_stateid(cstate, &u->lock.lk_resp_stateid);
7363 * functions to consume current state id
7367 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7368 union nfsd4_op_u *u)
7370 get_stateid(cstate, &u->open_downgrade.od_stateid);
7374 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7375 union nfsd4_op_u *u)
7377 get_stateid(cstate, &u->delegreturn.dr_stateid);
7381 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7382 union nfsd4_op_u *u)
7384 get_stateid(cstate, &u->free_stateid.fr_stateid);
7388 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7389 union nfsd4_op_u *u)
7391 get_stateid(cstate, &u->setattr.sa_stateid);
7395 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7396 union nfsd4_op_u *u)
7398 get_stateid(cstate, &u->close.cl_stateid);
7402 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7403 union nfsd4_op_u *u)
7405 get_stateid(cstate, &u->locku.lu_stateid);
7409 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7410 union nfsd4_op_u *u)
7412 get_stateid(cstate, &u->read.rd_stateid);
7416 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7417 union nfsd4_op_u *u)
7419 get_stateid(cstate, &u->write.wr_stateid);