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);
91 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
92 * the refcount on the open stateid to drop.
94 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
96 static struct kmem_cache *openowner_slab;
97 static struct kmem_cache *lockowner_slab;
98 static struct kmem_cache *file_slab;
99 static struct kmem_cache *stateid_slab;
100 static struct kmem_cache *deleg_slab;
101 static struct kmem_cache *odstate_slab;
103 static void free_session(struct nfsd4_session *);
105 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
106 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
108 static bool is_session_dead(struct nfsd4_session *ses)
110 return ses->se_flags & NFS4_SESSION_DEAD;
113 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
115 if (atomic_read(&ses->se_ref) > ref_held_by_me)
116 return nfserr_jukebox;
117 ses->se_flags |= NFS4_SESSION_DEAD;
121 static bool is_client_expired(struct nfs4_client *clp)
123 return clp->cl_time == 0;
126 static __be32 get_client_locked(struct nfs4_client *clp)
128 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
130 lockdep_assert_held(&nn->client_lock);
132 if (is_client_expired(clp))
133 return nfserr_expired;
134 atomic_inc(&clp->cl_refcount);
138 /* must be called under the client_lock */
140 renew_client_locked(struct nfs4_client *clp)
142 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
144 if (is_client_expired(clp)) {
146 printk("%s: client (clientid %08x/%08x) already expired\n",
148 clp->cl_clientid.cl_boot,
149 clp->cl_clientid.cl_id);
153 dprintk("renewing client (clientid %08x/%08x)\n",
154 clp->cl_clientid.cl_boot,
155 clp->cl_clientid.cl_id);
156 list_move_tail(&clp->cl_lru, &nn->client_lru);
157 clp->cl_time = get_seconds();
160 static void put_client_renew_locked(struct nfs4_client *clp)
162 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
164 lockdep_assert_held(&nn->client_lock);
166 if (!atomic_dec_and_test(&clp->cl_refcount))
168 if (!is_client_expired(clp))
169 renew_client_locked(clp);
172 static void put_client_renew(struct nfs4_client *clp)
174 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
176 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
178 if (!is_client_expired(clp))
179 renew_client_locked(clp);
180 spin_unlock(&nn->client_lock);
183 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
187 if (is_session_dead(ses))
188 return nfserr_badsession;
189 status = get_client_locked(ses->se_client);
192 atomic_inc(&ses->se_ref);
196 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
198 struct nfs4_client *clp = ses->se_client;
199 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
201 lockdep_assert_held(&nn->client_lock);
203 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
205 put_client_renew_locked(clp);
208 static void nfsd4_put_session(struct nfsd4_session *ses)
210 struct nfs4_client *clp = ses->se_client;
211 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
213 spin_lock(&nn->client_lock);
214 nfsd4_put_session_locked(ses);
215 spin_unlock(&nn->client_lock);
218 static struct nfsd4_blocked_lock *
219 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
222 struct nfsd4_blocked_lock *cur, *found = NULL;
224 spin_lock(&nn->blocked_locks_lock);
225 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
226 if (fh_match(fh, &cur->nbl_fh)) {
227 list_del_init(&cur->nbl_list);
228 list_del_init(&cur->nbl_lru);
233 spin_unlock(&nn->blocked_locks_lock);
235 posix_unblock_lock(&found->nbl_lock);
239 static struct nfsd4_blocked_lock *
240 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
243 struct nfsd4_blocked_lock *nbl;
245 nbl = find_blocked_lock(lo, fh, nn);
247 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
249 INIT_LIST_HEAD(&nbl->nbl_list);
250 INIT_LIST_HEAD(&nbl->nbl_lru);
251 fh_copy_shallow(&nbl->nbl_fh, fh);
252 locks_init_lock(&nbl->nbl_lock);
253 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
254 &nfsd4_cb_notify_lock_ops,
255 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
262 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
264 locks_release_private(&nbl->nbl_lock);
269 remove_blocked_locks(struct nfs4_lockowner *lo)
271 struct nfs4_client *clp = lo->lo_owner.so_client;
272 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
273 struct nfsd4_blocked_lock *nbl;
276 /* Dequeue all blocked locks */
277 spin_lock(&nn->blocked_locks_lock);
278 while (!list_empty(&lo->lo_blocked)) {
279 nbl = list_first_entry(&lo->lo_blocked,
280 struct nfsd4_blocked_lock,
282 list_del_init(&nbl->nbl_list);
283 list_move(&nbl->nbl_lru, &reaplist);
285 spin_unlock(&nn->blocked_locks_lock);
288 while (!list_empty(&reaplist)) {
289 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
291 list_del_init(&nbl->nbl_lru);
292 posix_unblock_lock(&nbl->nbl_lock);
293 free_blocked_lock(nbl);
298 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
301 * Since this is just an optimization, we don't try very hard if it
302 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
303 * just quit trying on anything else.
305 switch (task->tk_status) {
307 rpc_delay(task, 1 * HZ);
315 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
317 struct nfsd4_blocked_lock *nbl = container_of(cb,
318 struct nfsd4_blocked_lock, nbl_cb);
320 free_blocked_lock(nbl);
323 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
324 .done = nfsd4_cb_notify_lock_done,
325 .release = nfsd4_cb_notify_lock_release,
328 static inline struct nfs4_stateowner *
329 nfs4_get_stateowner(struct nfs4_stateowner *sop)
331 atomic_inc(&sop->so_count);
336 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
338 return (sop->so_owner.len == owner->len) &&
339 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
342 static struct nfs4_openowner *
343 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
344 struct nfs4_client *clp)
346 struct nfs4_stateowner *so;
348 lockdep_assert_held(&clp->cl_lock);
350 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
352 if (!so->so_is_open_owner)
354 if (same_owner_str(so, &open->op_owner))
355 return openowner(nfs4_get_stateowner(so));
360 static struct nfs4_openowner *
361 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
362 struct nfs4_client *clp)
364 struct nfs4_openowner *oo;
366 spin_lock(&clp->cl_lock);
367 oo = find_openstateowner_str_locked(hashval, open, clp);
368 spin_unlock(&clp->cl_lock);
373 opaque_hashval(const void *ptr, int nbytes)
375 unsigned char *cptr = (unsigned char *) ptr;
385 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
387 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
389 kmem_cache_free(file_slab, fp);
393 put_nfs4_file(struct nfs4_file *fi)
395 might_lock(&state_lock);
397 if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
398 hlist_del_rcu(&fi->fi_hash);
399 spin_unlock(&state_lock);
400 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
401 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
402 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
407 __nfs4_get_fd(struct nfs4_file *f, int oflag)
409 if (f->fi_fds[oflag])
410 return get_file(f->fi_fds[oflag]);
415 find_writeable_file_locked(struct nfs4_file *f)
419 lockdep_assert_held(&f->fi_lock);
421 ret = __nfs4_get_fd(f, O_WRONLY);
423 ret = __nfs4_get_fd(f, O_RDWR);
428 find_writeable_file(struct nfs4_file *f)
432 spin_lock(&f->fi_lock);
433 ret = find_writeable_file_locked(f);
434 spin_unlock(&f->fi_lock);
439 static struct file *find_readable_file_locked(struct nfs4_file *f)
443 lockdep_assert_held(&f->fi_lock);
445 ret = __nfs4_get_fd(f, O_RDONLY);
447 ret = __nfs4_get_fd(f, O_RDWR);
452 find_readable_file(struct nfs4_file *f)
456 spin_lock(&f->fi_lock);
457 ret = find_readable_file_locked(f);
458 spin_unlock(&f->fi_lock);
464 find_any_file(struct nfs4_file *f)
468 spin_lock(&f->fi_lock);
469 ret = __nfs4_get_fd(f, O_RDWR);
471 ret = __nfs4_get_fd(f, O_WRONLY);
473 ret = __nfs4_get_fd(f, O_RDONLY);
475 spin_unlock(&f->fi_lock);
479 static atomic_long_t num_delegations;
480 unsigned long max_delegations;
483 * Open owner state (share locks)
486 /* hash tables for lock and open owners */
487 #define OWNER_HASH_BITS 8
488 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
489 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
491 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
495 ret = opaque_hashval(ownername->data, ownername->len);
496 return ret & OWNER_HASH_MASK;
499 /* hash table for nfs4_file */
500 #define FILE_HASH_BITS 8
501 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
503 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
505 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
508 static unsigned int file_hashval(struct knfsd_fh *fh)
510 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
513 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
516 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
518 lockdep_assert_held(&fp->fi_lock);
520 if (access & NFS4_SHARE_ACCESS_WRITE)
521 atomic_inc(&fp->fi_access[O_WRONLY]);
522 if (access & NFS4_SHARE_ACCESS_READ)
523 atomic_inc(&fp->fi_access[O_RDONLY]);
527 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
529 lockdep_assert_held(&fp->fi_lock);
531 /* Does this access mode make sense? */
532 if (access & ~NFS4_SHARE_ACCESS_BOTH)
535 /* Does it conflict with a deny mode already set? */
536 if ((access & fp->fi_share_deny) != 0)
537 return nfserr_share_denied;
539 __nfs4_file_get_access(fp, access);
543 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
545 /* Common case is that there is no deny mode. */
547 /* Does this deny mode make sense? */
548 if (deny & ~NFS4_SHARE_DENY_BOTH)
551 if ((deny & NFS4_SHARE_DENY_READ) &&
552 atomic_read(&fp->fi_access[O_RDONLY]))
553 return nfserr_share_denied;
555 if ((deny & NFS4_SHARE_DENY_WRITE) &&
556 atomic_read(&fp->fi_access[O_WRONLY]))
557 return nfserr_share_denied;
562 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
564 might_lock(&fp->fi_lock);
566 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
567 struct file *f1 = NULL;
568 struct file *f2 = NULL;
570 swap(f1, fp->fi_fds[oflag]);
571 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
572 swap(f2, fp->fi_fds[O_RDWR]);
573 spin_unlock(&fp->fi_lock);
581 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
583 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
585 if (access & NFS4_SHARE_ACCESS_WRITE)
586 __nfs4_file_put_access(fp, O_WRONLY);
587 if (access & NFS4_SHARE_ACCESS_READ)
588 __nfs4_file_put_access(fp, O_RDONLY);
592 * Allocate a new open/delegation state counter. This is needed for
593 * pNFS for proper return on close semantics.
595 * Note that we only allocate it for pNFS-enabled exports, otherwise
596 * all pointers to struct nfs4_clnt_odstate are always NULL.
598 static struct nfs4_clnt_odstate *
599 alloc_clnt_odstate(struct nfs4_client *clp)
601 struct nfs4_clnt_odstate *co;
603 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
606 atomic_set(&co->co_odcount, 1);
612 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
614 struct nfs4_file *fp = co->co_file;
616 lockdep_assert_held(&fp->fi_lock);
617 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
621 get_clnt_odstate(struct nfs4_clnt_odstate *co)
624 atomic_inc(&co->co_odcount);
628 put_clnt_odstate(struct nfs4_clnt_odstate *co)
630 struct nfs4_file *fp;
636 if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
637 list_del(&co->co_perfile);
638 spin_unlock(&fp->fi_lock);
640 nfsd4_return_all_file_layouts(co->co_client, fp);
641 kmem_cache_free(odstate_slab, co);
645 static struct nfs4_clnt_odstate *
646 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
648 struct nfs4_clnt_odstate *co;
649 struct nfs4_client *cl;
656 spin_lock(&fp->fi_lock);
657 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
658 if (co->co_client == cl) {
659 get_clnt_odstate(co);
665 hash_clnt_odstate_locked(new);
667 spin_unlock(&fp->fi_lock);
671 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
672 void (*sc_free)(struct nfs4_stid *))
674 struct nfs4_stid *stid;
677 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
681 idr_preload(GFP_KERNEL);
682 spin_lock(&cl->cl_lock);
683 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
684 spin_unlock(&cl->cl_lock);
689 stid->sc_free = sc_free;
690 stid->sc_client = cl;
691 stid->sc_stateid.si_opaque.so_id = new_id;
692 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
693 /* Will be incremented before return to client: */
694 atomic_set(&stid->sc_count, 1);
695 spin_lock_init(&stid->sc_lock);
698 * It shouldn't be a problem to reuse an opaque stateid value.
699 * I don't think it is for 4.1. But with 4.0 I worry that, for
700 * example, a stray write retransmission could be accepted by
701 * the server when it should have been rejected. Therefore,
702 * adopt a trick from the sctp code to attempt to maximize the
703 * amount of time until an id is reused, by ensuring they always
704 * "increase" (mod INT_MAX):
708 kmem_cache_free(slab, stid);
712 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
714 struct nfs4_stid *stid;
716 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
720 return openlockstateid(stid);
723 static void nfs4_free_deleg(struct nfs4_stid *stid)
725 kmem_cache_free(deleg_slab, stid);
726 atomic_long_dec(&num_delegations);
730 * When we recall a delegation, we should be careful not to hand it
731 * out again straight away.
732 * To ensure this we keep a pair of bloom filters ('new' and 'old')
733 * in which the filehandles of recalled delegations are "stored".
734 * If a filehandle appear in either filter, a delegation is blocked.
735 * When a delegation is recalled, the filehandle is stored in the "new"
737 * Every 30 seconds we swap the filters and clear the "new" one,
738 * unless both are empty of course.
740 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
741 * low 3 bytes as hash-table indices.
743 * 'blocked_delegations_lock', which is always taken in block_delegations(),
744 * is used to manage concurrent access. Testing does not need the lock
745 * except when swapping the two filters.
747 static DEFINE_SPINLOCK(blocked_delegations_lock);
748 static struct bloom_pair {
749 int entries, old_entries;
751 int new; /* index into 'set' */
752 DECLARE_BITMAP(set[2], 256);
753 } blocked_delegations;
755 static int delegation_blocked(struct knfsd_fh *fh)
758 struct bloom_pair *bd = &blocked_delegations;
760 if (bd->entries == 0)
762 if (seconds_since_boot() - bd->swap_time > 30) {
763 spin_lock(&blocked_delegations_lock);
764 if (seconds_since_boot() - bd->swap_time > 30) {
765 bd->entries -= bd->old_entries;
766 bd->old_entries = bd->entries;
767 memset(bd->set[bd->new], 0,
770 bd->swap_time = seconds_since_boot();
772 spin_unlock(&blocked_delegations_lock);
774 hash = jhash(&fh->fh_base, fh->fh_size, 0);
775 if (test_bit(hash&255, bd->set[0]) &&
776 test_bit((hash>>8)&255, bd->set[0]) &&
777 test_bit((hash>>16)&255, bd->set[0]))
780 if (test_bit(hash&255, bd->set[1]) &&
781 test_bit((hash>>8)&255, bd->set[1]) &&
782 test_bit((hash>>16)&255, bd->set[1]))
788 static void block_delegations(struct knfsd_fh *fh)
791 struct bloom_pair *bd = &blocked_delegations;
793 hash = jhash(&fh->fh_base, fh->fh_size, 0);
795 spin_lock(&blocked_delegations_lock);
796 __set_bit(hash&255, bd->set[bd->new]);
797 __set_bit((hash>>8)&255, bd->set[bd->new]);
798 __set_bit((hash>>16)&255, bd->set[bd->new]);
799 if (bd->entries == 0)
800 bd->swap_time = seconds_since_boot();
802 spin_unlock(&blocked_delegations_lock);
805 static struct nfs4_delegation *
806 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
807 struct nfs4_clnt_odstate *odstate)
809 struct nfs4_delegation *dp;
812 dprintk("NFSD alloc_init_deleg\n");
813 n = atomic_long_inc_return(&num_delegations);
814 if (n < 0 || n > max_delegations)
816 if (delegation_blocked(¤t_fh->fh_handle))
818 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
823 * delegation seqid's are never incremented. The 4.1 special
824 * meaning of seqid 0 isn't meaningful, really, but let's avoid
825 * 0 anyway just for consistency and use 1:
827 dp->dl_stid.sc_stateid.si_generation = 1;
828 INIT_LIST_HEAD(&dp->dl_perfile);
829 INIT_LIST_HEAD(&dp->dl_perclnt);
830 INIT_LIST_HEAD(&dp->dl_recall_lru);
831 dp->dl_clnt_odstate = odstate;
832 get_clnt_odstate(odstate);
833 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
835 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
836 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
839 atomic_long_dec(&num_delegations);
844 nfs4_put_stid(struct nfs4_stid *s)
846 struct nfs4_file *fp = s->sc_file;
847 struct nfs4_client *clp = s->sc_client;
849 might_lock(&clp->cl_lock);
851 if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
852 wake_up_all(&close_wq);
855 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
856 spin_unlock(&clp->cl_lock);
863 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
865 stateid_t *src = &stid->sc_stateid;
867 spin_lock(&stid->sc_lock);
868 if (unlikely(++src->si_generation == 0))
869 src->si_generation = 1;
870 memcpy(dst, src, sizeof(*dst));
871 spin_unlock(&stid->sc_lock);
874 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
876 struct file *filp = NULL;
878 spin_lock(&fp->fi_lock);
879 if (fp->fi_deleg_file && --fp->fi_delegees == 0)
880 swap(filp, fp->fi_deleg_file);
881 spin_unlock(&fp->fi_lock);
884 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
889 void nfs4_unhash_stid(struct nfs4_stid *s)
895 * nfs4_get_existing_delegation - Discover if this delegation already exists
896 * @clp: a pointer to the nfs4_client we're granting a delegation to
897 * @fp: a pointer to the nfs4_file we're granting a delegation on
900 * On success: NULL if an existing delegation was not found.
902 * On error: -EAGAIN if one was previously granted to this nfs4_client
903 * for this nfs4_file.
908 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
910 struct nfs4_delegation *searchdp = NULL;
911 struct nfs4_client *searchclp = NULL;
913 lockdep_assert_held(&state_lock);
914 lockdep_assert_held(&fp->fi_lock);
916 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
917 searchclp = searchdp->dl_stid.sc_client;
918 if (clp == searchclp) {
926 * hash_delegation_locked - Add a delegation to the appropriate lists
927 * @dp: a pointer to the nfs4_delegation we are adding.
928 * @fp: a pointer to the nfs4_file we're granting a delegation on
931 * On success: NULL if the delegation was successfully hashed.
933 * On error: -EAGAIN if one was previously granted to this
934 * nfs4_client for this nfs4_file. Delegation is not hashed.
939 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
942 struct nfs4_client *clp = dp->dl_stid.sc_client;
944 lockdep_assert_held(&state_lock);
945 lockdep_assert_held(&fp->fi_lock);
947 status = nfs4_get_existing_delegation(clp, fp);
951 atomic_inc(&dp->dl_stid.sc_count);
952 dp->dl_stid.sc_type = NFS4_DELEG_STID;
953 list_add(&dp->dl_perfile, &fp->fi_delegations);
954 list_add(&dp->dl_perclnt, &clp->cl_delegations);
958 static bool delegation_hashed(struct nfs4_delegation *dp)
960 return !(list_empty(&dp->dl_perfile));
964 unhash_delegation_locked(struct nfs4_delegation *dp)
966 struct nfs4_file *fp = dp->dl_stid.sc_file;
968 lockdep_assert_held(&state_lock);
970 if (!delegation_hashed(dp))
973 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
974 /* Ensure that deleg break won't try to requeue it */
976 spin_lock(&fp->fi_lock);
977 list_del_init(&dp->dl_perclnt);
978 list_del_init(&dp->dl_recall_lru);
979 list_del_init(&dp->dl_perfile);
980 spin_unlock(&fp->fi_lock);
984 static void destroy_delegation(struct nfs4_delegation *dp)
988 spin_lock(&state_lock);
989 unhashed = unhash_delegation_locked(dp);
990 spin_unlock(&state_lock);
992 put_clnt_odstate(dp->dl_clnt_odstate);
993 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
994 nfs4_put_stid(&dp->dl_stid);
998 static void revoke_delegation(struct nfs4_delegation *dp)
1000 struct nfs4_client *clp = dp->dl_stid.sc_client;
1002 WARN_ON(!list_empty(&dp->dl_recall_lru));
1004 put_clnt_odstate(dp->dl_clnt_odstate);
1005 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1007 if (clp->cl_minorversion == 0)
1008 nfs4_put_stid(&dp->dl_stid);
1010 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1011 spin_lock(&clp->cl_lock);
1012 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1013 spin_unlock(&clp->cl_lock);
1021 static unsigned int clientid_hashval(u32 id)
1023 return id & CLIENT_HASH_MASK;
1026 static unsigned int clientstr_hashval(const char *name)
1028 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1032 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1033 * st_{access,deny}_bmap field of the stateid, in order to track not
1034 * only what share bits are currently in force, but also what
1035 * combinations of share bits previous opens have used. This allows us
1036 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1037 * return an error if the client attempt to downgrade to a combination
1038 * of share bits not explicable by closing some of its previous opens.
1040 * XXX: This enforcement is actually incomplete, since we don't keep
1041 * track of access/deny bit combinations; so, e.g., we allow:
1043 * OPEN allow read, deny write
1044 * OPEN allow both, deny none
1045 * DOWNGRADE allow read, deny none
1047 * which we should reject.
1050 bmap_to_share_mode(unsigned long bmap) {
1052 unsigned int access = 0;
1054 for (i = 1; i < 4; i++) {
1055 if (test_bit(i, &bmap))
1061 /* set share access for a given stateid */
1063 set_access(u32 access, struct nfs4_ol_stateid *stp)
1065 unsigned char mask = 1 << access;
1067 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1068 stp->st_access_bmap |= mask;
1071 /* clear share access for a given stateid */
1073 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1075 unsigned char mask = 1 << access;
1077 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1078 stp->st_access_bmap &= ~mask;
1081 /* test whether a given stateid has access */
1083 test_access(u32 access, struct nfs4_ol_stateid *stp)
1085 unsigned char mask = 1 << access;
1087 return (bool)(stp->st_access_bmap & mask);
1090 /* set share deny for a given stateid */
1092 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1094 unsigned char mask = 1 << deny;
1096 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1097 stp->st_deny_bmap |= mask;
1100 /* clear share deny for a given stateid */
1102 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1104 unsigned char mask = 1 << deny;
1106 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1107 stp->st_deny_bmap &= ~mask;
1110 /* test whether a given stateid is denying specific access */
1112 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1114 unsigned char mask = 1 << deny;
1116 return (bool)(stp->st_deny_bmap & mask);
1119 static int nfs4_access_to_omode(u32 access)
1121 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1122 case NFS4_SHARE_ACCESS_READ:
1124 case NFS4_SHARE_ACCESS_WRITE:
1126 case NFS4_SHARE_ACCESS_BOTH:
1134 * A stateid that had a deny mode associated with it is being released
1135 * or downgraded. Recalculate the deny mode on the file.
1138 recalculate_deny_mode(struct nfs4_file *fp)
1140 struct nfs4_ol_stateid *stp;
1142 spin_lock(&fp->fi_lock);
1143 fp->fi_share_deny = 0;
1144 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1145 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1146 spin_unlock(&fp->fi_lock);
1150 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1153 bool change = false;
1155 for (i = 1; i < 4; i++) {
1156 if ((i & deny) != i) {
1162 /* Recalculate per-file deny mode if there was a change */
1164 recalculate_deny_mode(stp->st_stid.sc_file);
1167 /* release all access and file references for a given stateid */
1169 release_all_access(struct nfs4_ol_stateid *stp)
1172 struct nfs4_file *fp = stp->st_stid.sc_file;
1174 if (fp && stp->st_deny_bmap != 0)
1175 recalculate_deny_mode(fp);
1177 for (i = 1; i < 4; i++) {
1178 if (test_access(i, stp))
1179 nfs4_file_put_access(stp->st_stid.sc_file, i);
1180 clear_access(i, stp);
1184 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1186 kfree(sop->so_owner.data);
1187 sop->so_ops->so_free(sop);
1190 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1192 struct nfs4_client *clp = sop->so_client;
1194 might_lock(&clp->cl_lock);
1196 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1198 sop->so_ops->so_unhash(sop);
1199 spin_unlock(&clp->cl_lock);
1200 nfs4_free_stateowner(sop);
1203 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1205 struct nfs4_file *fp = stp->st_stid.sc_file;
1207 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1209 if (list_empty(&stp->st_perfile))
1212 spin_lock(&fp->fi_lock);
1213 list_del_init(&stp->st_perfile);
1214 spin_unlock(&fp->fi_lock);
1215 list_del(&stp->st_perstateowner);
1219 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1221 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1223 put_clnt_odstate(stp->st_clnt_odstate);
1224 release_all_access(stp);
1225 if (stp->st_stateowner)
1226 nfs4_put_stateowner(stp->st_stateowner);
1227 kmem_cache_free(stateid_slab, stid);
1230 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1232 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1233 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1236 file = find_any_file(stp->st_stid.sc_file);
1238 filp_close(file, (fl_owner_t)lo);
1239 nfs4_free_ol_stateid(stid);
1243 * Put the persistent reference to an already unhashed generic stateid, while
1244 * holding the cl_lock. If it's the last reference, then put it onto the
1245 * reaplist for later destruction.
1247 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1248 struct list_head *reaplist)
1250 struct nfs4_stid *s = &stp->st_stid;
1251 struct nfs4_client *clp = s->sc_client;
1253 lockdep_assert_held(&clp->cl_lock);
1255 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1257 if (!atomic_dec_and_test(&s->sc_count)) {
1258 wake_up_all(&close_wq);
1262 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1263 list_add(&stp->st_locks, reaplist);
1266 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1268 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1270 list_del_init(&stp->st_locks);
1271 nfs4_unhash_stid(&stp->st_stid);
1272 return unhash_ol_stateid(stp);
1275 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1277 struct nfs4_client *clp = stp->st_stid.sc_client;
1280 spin_lock(&clp->cl_lock);
1281 unhashed = unhash_lock_stateid(stp);
1282 spin_unlock(&clp->cl_lock);
1284 nfs4_put_stid(&stp->st_stid);
1287 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1289 struct nfs4_client *clp = lo->lo_owner.so_client;
1291 lockdep_assert_held(&clp->cl_lock);
1293 list_del_init(&lo->lo_owner.so_strhash);
1297 * Free a list of generic stateids that were collected earlier after being
1301 free_ol_stateid_reaplist(struct list_head *reaplist)
1303 struct nfs4_ol_stateid *stp;
1304 struct nfs4_file *fp;
1308 while (!list_empty(reaplist)) {
1309 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1311 list_del(&stp->st_locks);
1312 fp = stp->st_stid.sc_file;
1313 stp->st_stid.sc_free(&stp->st_stid);
1319 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1320 struct list_head *reaplist)
1322 struct nfs4_ol_stateid *stp;
1324 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1326 while (!list_empty(&open_stp->st_locks)) {
1327 stp = list_entry(open_stp->st_locks.next,
1328 struct nfs4_ol_stateid, st_locks);
1329 WARN_ON(!unhash_lock_stateid(stp));
1330 put_ol_stateid_locked(stp, reaplist);
1334 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1335 struct list_head *reaplist)
1339 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1341 unhashed = unhash_ol_stateid(stp);
1342 release_open_stateid_locks(stp, reaplist);
1346 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1348 LIST_HEAD(reaplist);
1350 spin_lock(&stp->st_stid.sc_client->cl_lock);
1351 if (unhash_open_stateid(stp, &reaplist))
1352 put_ol_stateid_locked(stp, &reaplist);
1353 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1354 free_ol_stateid_reaplist(&reaplist);
1357 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1359 struct nfs4_client *clp = oo->oo_owner.so_client;
1361 lockdep_assert_held(&clp->cl_lock);
1363 list_del_init(&oo->oo_owner.so_strhash);
1364 list_del_init(&oo->oo_perclient);
1367 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1369 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1371 struct nfs4_ol_stateid *s;
1373 spin_lock(&nn->client_lock);
1374 s = oo->oo_last_closed_stid;
1376 list_del_init(&oo->oo_close_lru);
1377 oo->oo_last_closed_stid = NULL;
1379 spin_unlock(&nn->client_lock);
1381 nfs4_put_stid(&s->st_stid);
1384 static void release_openowner(struct nfs4_openowner *oo)
1386 struct nfs4_ol_stateid *stp;
1387 struct nfs4_client *clp = oo->oo_owner.so_client;
1388 struct list_head reaplist;
1390 INIT_LIST_HEAD(&reaplist);
1392 spin_lock(&clp->cl_lock);
1393 unhash_openowner_locked(oo);
1394 while (!list_empty(&oo->oo_owner.so_stateids)) {
1395 stp = list_first_entry(&oo->oo_owner.so_stateids,
1396 struct nfs4_ol_stateid, st_perstateowner);
1397 if (unhash_open_stateid(stp, &reaplist))
1398 put_ol_stateid_locked(stp, &reaplist);
1400 spin_unlock(&clp->cl_lock);
1401 free_ol_stateid_reaplist(&reaplist);
1402 release_last_closed_stateid(oo);
1403 nfs4_put_stateowner(&oo->oo_owner);
1407 hash_sessionid(struct nfs4_sessionid *sessionid)
1409 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1411 return sid->sequence % SESSION_HASH_SIZE;
1414 #ifdef CONFIG_SUNRPC_DEBUG
1416 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1418 u32 *ptr = (u32 *)(&sessionid->data[0]);
1419 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1423 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1429 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1430 * won't be used for replay.
1432 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1434 struct nfs4_stateowner *so = cstate->replay_owner;
1436 if (nfserr == nfserr_replay_me)
1439 if (!seqid_mutating_err(ntohl(nfserr))) {
1440 nfsd4_cstate_clear_replay(cstate);
1445 if (so->so_is_open_owner)
1446 release_last_closed_stateid(openowner(so));
1452 gen_sessionid(struct nfsd4_session *ses)
1454 struct nfs4_client *clp = ses->se_client;
1455 struct nfsd4_sessionid *sid;
1457 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1458 sid->clientid = clp->cl_clientid;
1459 sid->sequence = current_sessionid++;
1464 * The protocol defines ca_maxresponssize_cached to include the size of
1465 * the rpc header, but all we need to cache is the data starting after
1466 * the end of the initial SEQUENCE operation--the rest we regenerate
1467 * each time. Therefore we can advertise a ca_maxresponssize_cached
1468 * value that is the number of bytes in our cache plus a few additional
1469 * bytes. In order to stay on the safe side, and not promise more than
1470 * we can cache, those additional bytes must be the minimum possible: 24
1471 * bytes of rpc header (xid through accept state, with AUTH_NULL
1472 * verifier), 12 for the compound header (with zero-length tag), and 44
1473 * for the SEQUENCE op response:
1475 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1478 free_session_slots(struct nfsd4_session *ses)
1482 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1483 free_svc_cred(&ses->se_slots[i]->sl_cred);
1484 kfree(ses->se_slots[i]);
1489 * We don't actually need to cache the rpc and session headers, so we
1490 * can allocate a little less for each slot:
1492 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1496 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1499 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1500 return size + sizeof(struct nfsd4_slot);
1504 * XXX: If we run out of reserved DRC memory we could (up to a point)
1505 * re-negotiate active sessions and reduce their slot usage to make
1506 * room for new connections. For now we just fail the create session.
1508 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1510 u32 slotsize = slot_bytes(ca);
1511 u32 num = ca->maxreqs;
1512 unsigned long avail, total_avail;
1514 spin_lock(&nfsd_drc_lock);
1515 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1516 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1518 * Never use more than a third of the remaining memory,
1519 * unless it's the only way to give this client a slot:
1521 avail = clamp_t(unsigned long, avail, slotsize, total_avail/3);
1522 num = min_t(int, num, avail / slotsize);
1523 nfsd_drc_mem_used += num * slotsize;
1524 spin_unlock(&nfsd_drc_lock);
1529 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1531 int slotsize = slot_bytes(ca);
1533 spin_lock(&nfsd_drc_lock);
1534 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1535 spin_unlock(&nfsd_drc_lock);
1538 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1539 struct nfsd4_channel_attrs *battrs)
1541 int numslots = fattrs->maxreqs;
1542 int slotsize = slot_bytes(fattrs);
1543 struct nfsd4_session *new;
1546 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1547 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1548 mem = numslots * sizeof(struct nfsd4_slot *);
1550 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1553 /* allocate each struct nfsd4_slot and data cache in one piece */
1554 for (i = 0; i < numslots; i++) {
1555 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1556 if (!new->se_slots[i])
1560 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1561 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1566 kfree(new->se_slots[i]);
1571 static void free_conn(struct nfsd4_conn *c)
1573 svc_xprt_put(c->cn_xprt);
1577 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1579 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1580 struct nfs4_client *clp = c->cn_session->se_client;
1582 spin_lock(&clp->cl_lock);
1583 if (!list_empty(&c->cn_persession)) {
1584 list_del(&c->cn_persession);
1587 nfsd4_probe_callback(clp);
1588 spin_unlock(&clp->cl_lock);
1591 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1593 struct nfsd4_conn *conn;
1595 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1598 svc_xprt_get(rqstp->rq_xprt);
1599 conn->cn_xprt = rqstp->rq_xprt;
1600 conn->cn_flags = flags;
1601 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1605 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1607 conn->cn_session = ses;
1608 list_add(&conn->cn_persession, &ses->se_conns);
1611 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1613 struct nfs4_client *clp = ses->se_client;
1615 spin_lock(&clp->cl_lock);
1616 __nfsd4_hash_conn(conn, ses);
1617 spin_unlock(&clp->cl_lock);
1620 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1622 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1623 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1626 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1630 nfsd4_hash_conn(conn, ses);
1631 ret = nfsd4_register_conn(conn);
1633 /* oops; xprt is already down: */
1634 nfsd4_conn_lost(&conn->cn_xpt_user);
1635 /* We may have gained or lost a callback channel: */
1636 nfsd4_probe_callback_sync(ses->se_client);
1639 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1641 u32 dir = NFS4_CDFC4_FORE;
1643 if (cses->flags & SESSION4_BACK_CHAN)
1644 dir |= NFS4_CDFC4_BACK;
1645 return alloc_conn(rqstp, dir);
1648 /* must be called under client_lock */
1649 static void nfsd4_del_conns(struct nfsd4_session *s)
1651 struct nfs4_client *clp = s->se_client;
1652 struct nfsd4_conn *c;
1654 spin_lock(&clp->cl_lock);
1655 while (!list_empty(&s->se_conns)) {
1656 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1657 list_del_init(&c->cn_persession);
1658 spin_unlock(&clp->cl_lock);
1660 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1663 spin_lock(&clp->cl_lock);
1665 spin_unlock(&clp->cl_lock);
1668 static void __free_session(struct nfsd4_session *ses)
1670 free_session_slots(ses);
1674 static void free_session(struct nfsd4_session *ses)
1676 nfsd4_del_conns(ses);
1677 nfsd4_put_drc_mem(&ses->se_fchannel);
1678 __free_session(ses);
1681 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1684 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1686 new->se_client = clp;
1689 INIT_LIST_HEAD(&new->se_conns);
1691 new->se_cb_seq_nr = 1;
1692 new->se_flags = cses->flags;
1693 new->se_cb_prog = cses->callback_prog;
1694 new->se_cb_sec = cses->cb_sec;
1695 atomic_set(&new->se_ref, 0);
1696 idx = hash_sessionid(&new->se_sessionid);
1697 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1698 spin_lock(&clp->cl_lock);
1699 list_add(&new->se_perclnt, &clp->cl_sessions);
1700 spin_unlock(&clp->cl_lock);
1703 struct sockaddr *sa = svc_addr(rqstp);
1705 * This is a little silly; with sessions there's no real
1706 * use for the callback address. Use the peer address
1707 * as a reasonable default for now, but consider fixing
1708 * the rpc client not to require an address in the
1711 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1712 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1716 /* caller must hold client_lock */
1717 static struct nfsd4_session *
1718 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1720 struct nfsd4_session *elem;
1722 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1724 lockdep_assert_held(&nn->client_lock);
1726 dump_sessionid(__func__, sessionid);
1727 idx = hash_sessionid(sessionid);
1728 /* Search in the appropriate list */
1729 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1730 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1731 NFS4_MAX_SESSIONID_LEN)) {
1736 dprintk("%s: session not found\n", __func__);
1740 static struct nfsd4_session *
1741 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1744 struct nfsd4_session *session;
1745 __be32 status = nfserr_badsession;
1747 session = __find_in_sessionid_hashtbl(sessionid, net);
1750 status = nfsd4_get_session_locked(session);
1758 /* caller must hold client_lock */
1760 unhash_session(struct nfsd4_session *ses)
1762 struct nfs4_client *clp = ses->se_client;
1763 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1765 lockdep_assert_held(&nn->client_lock);
1767 list_del(&ses->se_hash);
1768 spin_lock(&ses->se_client->cl_lock);
1769 list_del(&ses->se_perclnt);
1770 spin_unlock(&ses->se_client->cl_lock);
1773 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1775 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1778 * We're assuming the clid was not given out from a boot
1779 * precisely 2^32 (about 136 years) before this one. That seems
1780 * a safe assumption:
1782 if (clid->cl_boot == (u32)nn->boot_time)
1784 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1785 clid->cl_boot, clid->cl_id, nn->boot_time);
1790 * XXX Should we use a slab cache ?
1791 * This type of memory management is somewhat inefficient, but we use it
1792 * anyway since SETCLIENTID is not a common operation.
1794 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1796 struct nfs4_client *clp;
1799 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1802 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1803 if (clp->cl_name.data == NULL)
1805 clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1806 OWNER_HASH_SIZE, GFP_KERNEL);
1807 if (!clp->cl_ownerstr_hashtbl)
1808 goto err_no_hashtbl;
1809 for (i = 0; i < OWNER_HASH_SIZE; i++)
1810 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1811 clp->cl_name.len = name.len;
1812 INIT_LIST_HEAD(&clp->cl_sessions);
1813 idr_init(&clp->cl_stateids);
1814 atomic_set(&clp->cl_refcount, 0);
1815 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1816 INIT_LIST_HEAD(&clp->cl_idhash);
1817 INIT_LIST_HEAD(&clp->cl_openowners);
1818 INIT_LIST_HEAD(&clp->cl_delegations);
1819 INIT_LIST_HEAD(&clp->cl_lru);
1820 INIT_LIST_HEAD(&clp->cl_revoked);
1821 #ifdef CONFIG_NFSD_PNFS
1822 INIT_LIST_HEAD(&clp->cl_lo_states);
1824 spin_lock_init(&clp->cl_lock);
1825 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1828 kfree(clp->cl_name.data);
1835 free_client(struct nfs4_client *clp)
1837 while (!list_empty(&clp->cl_sessions)) {
1838 struct nfsd4_session *ses;
1839 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1841 list_del(&ses->se_perclnt);
1842 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1845 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1846 free_svc_cred(&clp->cl_cred);
1847 kfree(clp->cl_ownerstr_hashtbl);
1848 kfree(clp->cl_name.data);
1849 idr_destroy(&clp->cl_stateids);
1853 /* must be called under the client_lock */
1855 unhash_client_locked(struct nfs4_client *clp)
1857 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1858 struct nfsd4_session *ses;
1860 lockdep_assert_held(&nn->client_lock);
1862 /* Mark the client as expired! */
1864 /* Make it invisible */
1865 if (!list_empty(&clp->cl_idhash)) {
1866 list_del_init(&clp->cl_idhash);
1867 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1868 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1870 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1872 list_del_init(&clp->cl_lru);
1873 spin_lock(&clp->cl_lock);
1874 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1875 list_del_init(&ses->se_hash);
1876 spin_unlock(&clp->cl_lock);
1880 unhash_client(struct nfs4_client *clp)
1882 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1884 spin_lock(&nn->client_lock);
1885 unhash_client_locked(clp);
1886 spin_unlock(&nn->client_lock);
1889 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1891 if (atomic_read(&clp->cl_refcount))
1892 return nfserr_jukebox;
1893 unhash_client_locked(clp);
1898 __destroy_client(struct nfs4_client *clp)
1901 struct nfs4_openowner *oo;
1902 struct nfs4_delegation *dp;
1903 struct list_head reaplist;
1905 INIT_LIST_HEAD(&reaplist);
1906 spin_lock(&state_lock);
1907 while (!list_empty(&clp->cl_delegations)) {
1908 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1909 WARN_ON(!unhash_delegation_locked(dp));
1910 list_add(&dp->dl_recall_lru, &reaplist);
1912 spin_unlock(&state_lock);
1913 while (!list_empty(&reaplist)) {
1914 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1915 list_del_init(&dp->dl_recall_lru);
1916 put_clnt_odstate(dp->dl_clnt_odstate);
1917 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1918 nfs4_put_stid(&dp->dl_stid);
1920 while (!list_empty(&clp->cl_revoked)) {
1921 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1922 list_del_init(&dp->dl_recall_lru);
1923 nfs4_put_stid(&dp->dl_stid);
1925 while (!list_empty(&clp->cl_openowners)) {
1926 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1927 nfs4_get_stateowner(&oo->oo_owner);
1928 release_openowner(oo);
1930 for (i = 0; i < OWNER_HASH_SIZE; i++) {
1931 struct nfs4_stateowner *so, *tmp;
1933 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1935 /* Should be no openowners at this point */
1936 WARN_ON_ONCE(so->so_is_open_owner);
1937 remove_blocked_locks(lockowner(so));
1940 nfsd4_return_all_client_layouts(clp);
1941 nfsd4_shutdown_callback(clp);
1942 if (clp->cl_cb_conn.cb_xprt)
1943 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1948 destroy_client(struct nfs4_client *clp)
1951 __destroy_client(clp);
1954 static void expire_client(struct nfs4_client *clp)
1957 nfsd4_client_record_remove(clp);
1958 __destroy_client(clp);
1961 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1963 memcpy(target->cl_verifier.data, source->data,
1964 sizeof(target->cl_verifier.data));
1967 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1969 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1970 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1973 int strdup_if_nonnull(char **target, char *source)
1976 *target = kstrdup(source, GFP_KERNEL);
1984 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1988 ret = strdup_if_nonnull(&target->cr_principal, source->cr_principal);
1991 ret = strdup_if_nonnull(&target->cr_raw_principal,
1992 source->cr_raw_principal);
1995 target->cr_flavor = source->cr_flavor;
1996 target->cr_uid = source->cr_uid;
1997 target->cr_gid = source->cr_gid;
1998 target->cr_group_info = source->cr_group_info;
1999 get_group_info(target->cr_group_info);
2000 target->cr_gss_mech = source->cr_gss_mech;
2001 if (source->cr_gss_mech)
2002 gss_mech_get(source->cr_gss_mech);
2007 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2009 if (o1->len < o2->len)
2011 if (o1->len > o2->len)
2013 return memcmp(o1->data, o2->data, o1->len);
2016 static int same_name(const char *n1, const char *n2)
2018 return 0 == memcmp(n1, n2, HEXDIR_LEN);
2022 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2024 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2028 same_clid(clientid_t *cl1, clientid_t *cl2)
2030 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2033 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2037 if (g1->ngroups != g2->ngroups)
2039 for (i=0; i<g1->ngroups; i++)
2040 if (!gid_eq(g1->gid[i], g2->gid[i]))
2046 * RFC 3530 language requires clid_inuse be returned when the
2047 * "principal" associated with a requests differs from that previously
2048 * used. We use uid, gid's, and gss principal string as our best
2049 * approximation. We also don't want to allow non-gss use of a client
2050 * established using gss: in theory cr_principal should catch that
2051 * change, but in practice cr_principal can be null even in the gss case
2052 * since gssd doesn't always pass down a principal string.
2054 static bool is_gss_cred(struct svc_cred *cr)
2056 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2057 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2062 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2064 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2065 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2066 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2067 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2069 if (cr1->cr_principal == cr2->cr_principal)
2071 if (!cr1->cr_principal || !cr2->cr_principal)
2073 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2076 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2078 struct svc_cred *cr = &rqstp->rq_cred;
2081 if (!cr->cr_gss_mech)
2083 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2084 return service == RPC_GSS_SVC_INTEGRITY ||
2085 service == RPC_GSS_SVC_PRIVACY;
2088 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2090 struct svc_cred *cr = &rqstp->rq_cred;
2092 if (!cl->cl_mach_cred)
2094 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2096 if (!svc_rqst_integrity_protected(rqstp))
2098 if (cl->cl_cred.cr_raw_principal)
2099 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2100 cr->cr_raw_principal);
2101 if (!cr->cr_principal)
2103 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2106 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2111 * This is opaque to client, so no need to byte-swap. Use
2112 * __force to keep sparse happy
2114 verf[0] = (__force __be32)get_seconds();
2115 verf[1] = (__force __be32)nn->clverifier_counter++;
2116 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2119 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2121 clp->cl_clientid.cl_boot = nn->boot_time;
2122 clp->cl_clientid.cl_id = nn->clientid_counter++;
2123 gen_confirm(clp, nn);
2126 static struct nfs4_stid *
2127 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2129 struct nfs4_stid *ret;
2131 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2132 if (!ret || !ret->sc_type)
2137 static struct nfs4_stid *
2138 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2140 struct nfs4_stid *s;
2142 spin_lock(&cl->cl_lock);
2143 s = find_stateid_locked(cl, t);
2145 if (typemask & s->sc_type)
2146 atomic_inc(&s->sc_count);
2150 spin_unlock(&cl->cl_lock);
2154 static struct nfs4_client *create_client(struct xdr_netobj name,
2155 struct svc_rqst *rqstp, nfs4_verifier *verf)
2157 struct nfs4_client *clp;
2158 struct sockaddr *sa = svc_addr(rqstp);
2160 struct net *net = SVC_NET(rqstp);
2162 clp = alloc_client(name);
2166 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2171 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2172 clp->cl_time = get_seconds();
2173 clear_bit(0, &clp->cl_cb_slot_busy);
2174 copy_verf(clp, verf);
2175 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2176 clp->cl_cb_session = NULL;
2182 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2184 struct rb_node **new = &(root->rb_node), *parent = NULL;
2185 struct nfs4_client *clp;
2188 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2191 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2192 new = &((*new)->rb_left);
2194 new = &((*new)->rb_right);
2197 rb_link_node(&new_clp->cl_namenode, parent, new);
2198 rb_insert_color(&new_clp->cl_namenode, root);
2201 static struct nfs4_client *
2202 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2205 struct rb_node *node = root->rb_node;
2206 struct nfs4_client *clp;
2209 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2210 cmp = compare_blob(&clp->cl_name, name);
2212 node = node->rb_left;
2214 node = node->rb_right;
2222 add_to_unconfirmed(struct nfs4_client *clp)
2224 unsigned int idhashval;
2225 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2227 lockdep_assert_held(&nn->client_lock);
2229 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2230 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2231 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2232 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2233 renew_client_locked(clp);
2237 move_to_confirmed(struct nfs4_client *clp)
2239 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2240 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2242 lockdep_assert_held(&nn->client_lock);
2244 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2245 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2246 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2247 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2248 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2249 renew_client_locked(clp);
2252 static struct nfs4_client *
2253 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2255 struct nfs4_client *clp;
2256 unsigned int idhashval = clientid_hashval(clid->cl_id);
2258 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2259 if (same_clid(&clp->cl_clientid, clid)) {
2260 if ((bool)clp->cl_minorversion != sessions)
2262 renew_client_locked(clp);
2269 static struct nfs4_client *
2270 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2272 struct list_head *tbl = nn->conf_id_hashtbl;
2274 lockdep_assert_held(&nn->client_lock);
2275 return find_client_in_id_table(tbl, clid, sessions);
2278 static struct nfs4_client *
2279 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2281 struct list_head *tbl = nn->unconf_id_hashtbl;
2283 lockdep_assert_held(&nn->client_lock);
2284 return find_client_in_id_table(tbl, clid, sessions);
2287 static bool clp_used_exchangeid(struct nfs4_client *clp)
2289 return clp->cl_exchange_flags != 0;
2292 static struct nfs4_client *
2293 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2295 lockdep_assert_held(&nn->client_lock);
2296 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2299 static struct nfs4_client *
2300 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2302 lockdep_assert_held(&nn->client_lock);
2303 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2307 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2309 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2310 struct sockaddr *sa = svc_addr(rqstp);
2311 u32 scopeid = rpc_get_scope_id(sa);
2312 unsigned short expected_family;
2314 /* Currently, we only support tcp and tcp6 for the callback channel */
2315 if (se->se_callback_netid_len == 3 &&
2316 !memcmp(se->se_callback_netid_val, "tcp", 3))
2317 expected_family = AF_INET;
2318 else if (se->se_callback_netid_len == 4 &&
2319 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2320 expected_family = AF_INET6;
2324 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2325 se->se_callback_addr_len,
2326 (struct sockaddr *)&conn->cb_addr,
2327 sizeof(conn->cb_addr));
2329 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2332 if (conn->cb_addr.ss_family == AF_INET6)
2333 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2335 conn->cb_prog = se->se_callback_prog;
2336 conn->cb_ident = se->se_callback_ident;
2337 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2340 conn->cb_addr.ss_family = AF_UNSPEC;
2341 conn->cb_addrlen = 0;
2342 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2343 "will not receive delegations\n",
2344 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2350 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2353 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2355 struct xdr_buf *buf = resp->xdr.buf;
2356 struct nfsd4_slot *slot = resp->cstate.slot;
2359 dprintk("--> %s slot %p\n", __func__, slot);
2361 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2362 slot->sl_opcnt = resp->opcnt;
2363 slot->sl_status = resp->cstate.status;
2364 free_svc_cred(&slot->sl_cred);
2365 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2367 if (!nfsd4_cache_this(resp)) {
2368 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2371 slot->sl_flags |= NFSD4_SLOT_CACHED;
2373 base = resp->cstate.data_offset;
2374 slot->sl_datalen = buf->len - base;
2375 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2376 WARN(1, "%s: sessions DRC could not cache compound\n",
2382 * Encode the replay sequence operation from the slot values.
2383 * If cachethis is FALSE encode the uncached rep error on the next
2384 * operation which sets resp->p and increments resp->opcnt for
2385 * nfs4svc_encode_compoundres.
2389 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2390 struct nfsd4_compoundres *resp)
2392 struct nfsd4_op *op;
2393 struct nfsd4_slot *slot = resp->cstate.slot;
2395 /* Encode the replayed sequence operation */
2396 op = &args->ops[resp->opcnt - 1];
2397 nfsd4_encode_operation(resp, op);
2399 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2401 if (args->opcnt == 1) {
2403 * The original operation wasn't a solo sequence--we
2404 * always cache those--so this retry must not match the
2407 op->status = nfserr_seq_false_retry;
2409 op = &args->ops[resp->opcnt++];
2410 op->status = nfserr_retry_uncached_rep;
2411 nfsd4_encode_operation(resp, op);
2417 * The sequence operation is not cached because we can use the slot and
2421 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2422 struct nfsd4_sequence *seq)
2424 struct nfsd4_slot *slot = resp->cstate.slot;
2425 struct xdr_stream *xdr = &resp->xdr;
2429 dprintk("--> %s slot %p\n", __func__, slot);
2431 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2435 p = xdr_reserve_space(xdr, slot->sl_datalen);
2438 return nfserr_serverfault;
2440 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2441 xdr_commit_encode(xdr);
2443 resp->opcnt = slot->sl_opcnt;
2444 return slot->sl_status;
2448 * Set the exchange_id flags returned by the server.
2451 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2453 #ifdef CONFIG_NFSD_PNFS
2454 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2456 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2459 /* Referrals are supported, Migration is not. */
2460 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2462 /* set the wire flags to return to client. */
2463 clid->flags = new->cl_exchange_flags;
2466 static bool client_has_openowners(struct nfs4_client *clp)
2468 struct nfs4_openowner *oo;
2470 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2471 if (!list_empty(&oo->oo_owner.so_stateids))
2477 static bool client_has_state(struct nfs4_client *clp)
2479 return client_has_openowners(clp)
2480 #ifdef CONFIG_NFSD_PNFS
2481 || !list_empty(&clp->cl_lo_states)
2483 || !list_empty(&clp->cl_delegations)
2484 || !list_empty(&clp->cl_sessions);
2488 nfsd4_exchange_id(struct svc_rqst *rqstp,
2489 struct nfsd4_compound_state *cstate,
2490 struct nfsd4_exchange_id *exid)
2492 struct nfs4_client *conf, *new;
2493 struct nfs4_client *unconf = NULL;
2495 char addr_str[INET6_ADDRSTRLEN];
2496 nfs4_verifier verf = exid->verifier;
2497 struct sockaddr *sa = svc_addr(rqstp);
2498 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2499 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2501 rpc_ntop(sa, addr_str, sizeof(addr_str));
2502 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2503 "ip_addr=%s flags %x, spa_how %d\n",
2504 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2505 addr_str, exid->flags, exid->spa_how);
2507 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2508 return nfserr_inval;
2510 new = create_client(exid->clname, rqstp, &verf);
2512 return nfserr_jukebox;
2514 switch (exid->spa_how) {
2516 exid->spo_must_enforce[0] = 0;
2517 exid->spo_must_enforce[1] = (
2518 1 << (OP_BIND_CONN_TO_SESSION - 32) |
2519 1 << (OP_EXCHANGE_ID - 32) |
2520 1 << (OP_CREATE_SESSION - 32) |
2521 1 << (OP_DESTROY_SESSION - 32) |
2522 1 << (OP_DESTROY_CLIENTID - 32));
2524 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2525 1 << (OP_OPEN_DOWNGRADE) |
2527 1 << (OP_DELEGRETURN));
2529 exid->spo_must_allow[1] &= (
2530 1 << (OP_TEST_STATEID - 32) |
2531 1 << (OP_FREE_STATEID - 32));
2532 if (!svc_rqst_integrity_protected(rqstp)) {
2533 status = nfserr_inval;
2537 * Sometimes userspace doesn't give us a principal.
2538 * Which is a bug, really. Anyway, we can't enforce
2539 * MACH_CRED in that case, better to give up now:
2541 if (!new->cl_cred.cr_principal &&
2542 !new->cl_cred.cr_raw_principal) {
2543 status = nfserr_serverfault;
2546 new->cl_mach_cred = true;
2549 default: /* checked by xdr code */
2552 status = nfserr_encr_alg_unsupp;
2556 /* Cases below refer to rfc 5661 section 18.35.4: */
2557 spin_lock(&nn->client_lock);
2558 conf = find_confirmed_client_by_name(&exid->clname, nn);
2560 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2561 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2564 if (!clp_used_exchangeid(conf)) { /* buggy client */
2565 status = nfserr_inval;
2568 if (!nfsd4_mach_creds_match(conf, rqstp)) {
2569 status = nfserr_wrong_cred;
2572 if (!creds_match) { /* case 9 */
2573 status = nfserr_perm;
2576 if (!verfs_match) { /* case 8 */
2577 status = nfserr_not_same;
2581 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2584 if (!creds_match) { /* case 3 */
2585 if (client_has_state(conf)) {
2586 status = nfserr_clid_inuse;
2591 if (verfs_match) { /* case 2 */
2592 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2595 /* case 5, client reboot */
2600 if (update) { /* case 7 */
2601 status = nfserr_noent;
2605 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2606 if (unconf) /* case 4, possible retry or client restart */
2607 unhash_client_locked(unconf);
2609 /* case 1 (normal case) */
2612 status = mark_client_expired_locked(conf);
2616 new->cl_minorversion = cstate->minorversion;
2617 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2618 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2621 add_to_unconfirmed(new);
2624 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2625 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2627 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2628 nfsd4_set_ex_flags(conf, exid);
2630 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2631 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2635 spin_unlock(&nn->client_lock);
2640 expire_client(unconf);
2645 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2647 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2650 /* The slot is in use, and no response has been sent. */
2652 if (seqid == slot_seqid)
2653 return nfserr_jukebox;
2655 return nfserr_seq_misordered;
2657 /* Note unsigned 32-bit arithmetic handles wraparound: */
2658 if (likely(seqid == slot_seqid + 1))
2660 if (seqid == slot_seqid)
2661 return nfserr_replay_cache;
2662 return nfserr_seq_misordered;
2666 * Cache the create session result into the create session single DRC
2667 * slot cache by saving the xdr structure. sl_seqid has been set.
2668 * Do this for solo or embedded create session operations.
2671 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2672 struct nfsd4_clid_slot *slot, __be32 nfserr)
2674 slot->sl_status = nfserr;
2675 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2679 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2680 struct nfsd4_clid_slot *slot)
2682 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2683 return slot->sl_status;
2686 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2687 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2688 1 + /* MIN tag is length with zero, only length */ \
2689 3 + /* version, opcount, opcode */ \
2690 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2691 /* seqid, slotID, slotID, cache */ \
2692 4 ) * sizeof(__be32))
2694 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2695 2 + /* verifier: AUTH_NULL, length 0 */\
2697 1 + /* MIN tag is length with zero, only length */ \
2698 3 + /* opcount, opcode, opstatus*/ \
2699 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2700 /* seqid, slotID, slotID, slotID, status */ \
2701 5 ) * sizeof(__be32))
2703 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2705 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2707 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2708 return nfserr_toosmall;
2709 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2710 return nfserr_toosmall;
2711 ca->headerpadsz = 0;
2712 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2713 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2714 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2715 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2716 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2717 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2719 * Note decreasing slot size below client's request may make it
2720 * difficult for client to function correctly, whereas
2721 * decreasing the number of slots will (just?) affect
2722 * performance. When short on memory we therefore prefer to
2723 * decrease number of slots instead of their size. Clients that
2724 * request larger slots than they need will get poor results:
2726 ca->maxreqs = nfsd4_get_drc_mem(ca);
2728 return nfserr_jukebox;
2734 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2735 * These are based on similar macros in linux/sunrpc/msg_prot.h .
2737 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2738 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2740 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2741 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2743 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2744 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2745 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2746 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2749 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2751 ca->headerpadsz = 0;
2753 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2754 return nfserr_toosmall;
2755 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2756 return nfserr_toosmall;
2757 ca->maxresp_cached = 0;
2759 return nfserr_toosmall;
2764 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2766 switch (cbs->flavor) {
2772 * GSS case: the spec doesn't allow us to return this
2773 * error. But it also doesn't allow us not to support
2775 * I'd rather this fail hard than return some error the
2776 * client might think it can already handle:
2778 return nfserr_encr_alg_unsupp;
2783 nfsd4_create_session(struct svc_rqst *rqstp,
2784 struct nfsd4_compound_state *cstate,
2785 struct nfsd4_create_session *cr_ses)
2787 struct sockaddr *sa = svc_addr(rqstp);
2788 struct nfs4_client *conf, *unconf;
2789 struct nfs4_client *old = NULL;
2790 struct nfsd4_session *new;
2791 struct nfsd4_conn *conn;
2792 struct nfsd4_clid_slot *cs_slot = NULL;
2794 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2796 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2797 return nfserr_inval;
2798 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2801 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2804 status = check_backchannel_attrs(&cr_ses->back_channel);
2806 goto out_release_drc_mem;
2807 status = nfserr_jukebox;
2808 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2810 goto out_release_drc_mem;
2811 conn = alloc_conn_from_crses(rqstp, cr_ses);
2813 goto out_free_session;
2815 spin_lock(&nn->client_lock);
2816 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2817 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2818 WARN_ON_ONCE(conf && unconf);
2821 status = nfserr_wrong_cred;
2822 if (!nfsd4_mach_creds_match(conf, rqstp))
2824 cs_slot = &conf->cl_cs_slot;
2825 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2827 if (status == nfserr_replay_cache)
2828 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2831 } else if (unconf) {
2832 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2833 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2834 status = nfserr_clid_inuse;
2837 status = nfserr_wrong_cred;
2838 if (!nfsd4_mach_creds_match(unconf, rqstp))
2840 cs_slot = &unconf->cl_cs_slot;
2841 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2843 /* an unconfirmed replay returns misordered */
2844 status = nfserr_seq_misordered;
2847 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2849 status = mark_client_expired_locked(old);
2855 move_to_confirmed(unconf);
2858 status = nfserr_stale_clientid;
2862 /* Persistent sessions are not supported */
2863 cr_ses->flags &= ~SESSION4_PERSIST;
2864 /* Upshifting from TCP to RDMA is not supported */
2865 cr_ses->flags &= ~SESSION4_RDMA;
2867 init_session(rqstp, new, conf, cr_ses);
2868 nfsd4_get_session_locked(new);
2870 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2871 NFS4_MAX_SESSIONID_LEN);
2872 cs_slot->sl_seqid++;
2873 cr_ses->seqid = cs_slot->sl_seqid;
2875 /* cache solo and embedded create sessions under the client_lock */
2876 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2877 spin_unlock(&nn->client_lock);
2878 /* init connection and backchannel */
2879 nfsd4_init_conn(rqstp, conn, new);
2880 nfsd4_put_session(new);
2885 spin_unlock(&nn->client_lock);
2890 __free_session(new);
2891 out_release_drc_mem:
2892 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2896 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2899 case NFS4_CDFC4_FORE:
2900 case NFS4_CDFC4_BACK:
2902 case NFS4_CDFC4_FORE_OR_BOTH:
2903 case NFS4_CDFC4_BACK_OR_BOTH:
2904 *dir = NFS4_CDFC4_BOTH;
2907 return nfserr_inval;
2910 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2912 struct nfsd4_session *session = cstate->session;
2913 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2916 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2919 spin_lock(&nn->client_lock);
2920 session->se_cb_prog = bc->bc_cb_program;
2921 session->se_cb_sec = bc->bc_cb_sec;
2922 spin_unlock(&nn->client_lock);
2924 nfsd4_probe_callback(session->se_client);
2929 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2930 struct nfsd4_compound_state *cstate,
2931 struct nfsd4_bind_conn_to_session *bcts)
2934 struct nfsd4_conn *conn;
2935 struct nfsd4_session *session;
2936 struct net *net = SVC_NET(rqstp);
2937 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2939 if (!nfsd4_last_compound_op(rqstp))
2940 return nfserr_not_only_op;
2941 spin_lock(&nn->client_lock);
2942 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2943 spin_unlock(&nn->client_lock);
2945 goto out_no_session;
2946 status = nfserr_wrong_cred;
2947 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2949 status = nfsd4_map_bcts_dir(&bcts->dir);
2952 conn = alloc_conn(rqstp, bcts->dir);
2953 status = nfserr_jukebox;
2956 nfsd4_init_conn(rqstp, conn, session);
2959 nfsd4_put_session(session);
2964 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2968 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2972 nfsd4_destroy_session(struct svc_rqst *r,
2973 struct nfsd4_compound_state *cstate,
2974 struct nfsd4_destroy_session *sessionid)
2976 struct nfsd4_session *ses;
2978 int ref_held_by_me = 0;
2979 struct net *net = SVC_NET(r);
2980 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2982 status = nfserr_not_only_op;
2983 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2984 if (!nfsd4_last_compound_op(r))
2988 dump_sessionid(__func__, &sessionid->sessionid);
2989 spin_lock(&nn->client_lock);
2990 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2992 goto out_client_lock;
2993 status = nfserr_wrong_cred;
2994 if (!nfsd4_mach_creds_match(ses->se_client, r))
2995 goto out_put_session;
2996 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2998 goto out_put_session;
2999 unhash_session(ses);
3000 spin_unlock(&nn->client_lock);
3002 nfsd4_probe_callback_sync(ses->se_client);
3004 spin_lock(&nn->client_lock);
3007 nfsd4_put_session_locked(ses);
3009 spin_unlock(&nn->client_lock);
3014 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3016 struct nfsd4_conn *c;
3018 list_for_each_entry(c, &s->se_conns, cn_persession) {
3019 if (c->cn_xprt == xpt) {
3026 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3028 struct nfs4_client *clp = ses->se_client;
3029 struct nfsd4_conn *c;
3030 __be32 status = nfs_ok;
3033 spin_lock(&clp->cl_lock);
3034 c = __nfsd4_find_conn(new->cn_xprt, ses);
3037 status = nfserr_conn_not_bound_to_session;
3038 if (clp->cl_mach_cred)
3040 __nfsd4_hash_conn(new, ses);
3041 spin_unlock(&clp->cl_lock);
3042 ret = nfsd4_register_conn(new);
3044 /* oops; xprt is already down: */
3045 nfsd4_conn_lost(&new->cn_xpt_user);
3048 spin_unlock(&clp->cl_lock);
3053 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3055 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3057 return args->opcnt > session->se_fchannel.maxops;
3060 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3061 struct nfsd4_session *session)
3063 struct xdr_buf *xb = &rqstp->rq_arg;
3065 return xb->len > session->se_fchannel.maxreq_sz;
3068 static bool replay_matches_cache(struct svc_rqst *rqstp,
3069 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3071 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3073 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3074 (bool)seq->cachethis)
3077 * If there's an error then the reply can have fewer ops than
3080 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3083 * But if we cached a reply with *more* ops than the call you're
3084 * sending us now, then this new call is clearly not really a
3085 * replay of the old one:
3087 if (slot->sl_opcnt > argp->opcnt)
3089 /* This is the only check explicitly called by spec: */
3090 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3093 * There may be more comparisons we could actually do, but the
3094 * spec doesn't require us to catch every case where the calls
3095 * don't match (that would require caching the call as well as
3096 * the reply), so we don't bother.
3102 nfsd4_sequence(struct svc_rqst *rqstp,
3103 struct nfsd4_compound_state *cstate,
3104 struct nfsd4_sequence *seq)
3106 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3107 struct xdr_stream *xdr = &resp->xdr;
3108 struct nfsd4_session *session;
3109 struct nfs4_client *clp;
3110 struct nfsd4_slot *slot;
3111 struct nfsd4_conn *conn;
3114 struct net *net = SVC_NET(rqstp);
3115 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3117 if (resp->opcnt != 1)
3118 return nfserr_sequence_pos;
3121 * Will be either used or freed by nfsd4_sequence_check_conn
3124 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3126 return nfserr_jukebox;
3128 spin_lock(&nn->client_lock);
3129 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3131 goto out_no_session;
3132 clp = session->se_client;
3134 status = nfserr_too_many_ops;
3135 if (nfsd4_session_too_many_ops(rqstp, session))
3136 goto out_put_session;
3138 status = nfserr_req_too_big;
3139 if (nfsd4_request_too_big(rqstp, session))
3140 goto out_put_session;
3142 status = nfserr_badslot;
3143 if (seq->slotid >= session->se_fchannel.maxreqs)
3144 goto out_put_session;
3146 slot = session->se_slots[seq->slotid];
3147 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3149 /* We do not negotiate the number of slots yet, so set the
3150 * maxslots to the session maxreqs which is used to encode
3151 * sr_highest_slotid and the sr_target_slot id to maxslots */
3152 seq->maxslots = session->se_fchannel.maxreqs;
3154 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3155 slot->sl_flags & NFSD4_SLOT_INUSE);
3156 if (status == nfserr_replay_cache) {
3157 status = nfserr_seq_misordered;
3158 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3159 goto out_put_session;
3160 status = nfserr_seq_false_retry;
3161 if (!replay_matches_cache(rqstp, seq, slot))
3162 goto out_put_session;
3163 cstate->slot = slot;
3164 cstate->session = session;
3166 /* Return the cached reply status and set cstate->status
3167 * for nfsd4_proc_compound processing */
3168 status = nfsd4_replay_cache_entry(resp, seq);
3169 cstate->status = nfserr_replay_cache;
3173 goto out_put_session;
3175 status = nfsd4_sequence_check_conn(conn, session);
3178 goto out_put_session;
3180 buflen = (seq->cachethis) ?
3181 session->se_fchannel.maxresp_cached :
3182 session->se_fchannel.maxresp_sz;
3183 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3185 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3186 goto out_put_session;
3187 svc_reserve(rqstp, buflen);
3190 /* Success! bump slot seqid */
3191 slot->sl_seqid = seq->seqid;
3192 slot->sl_flags |= NFSD4_SLOT_INUSE;
3194 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3196 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3198 cstate->slot = slot;
3199 cstate->session = session;
3203 switch (clp->cl_cb_state) {
3205 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3207 case NFSD4_CB_FAULT:
3208 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3211 seq->status_flags = 0;
3213 if (!list_empty(&clp->cl_revoked))
3214 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3218 spin_unlock(&nn->client_lock);
3221 nfsd4_put_session_locked(session);
3222 goto out_no_session;
3226 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3228 struct nfsd4_compound_state *cs = &resp->cstate;
3230 if (nfsd4_has_session(cs)) {
3231 if (cs->status != nfserr_replay_cache) {
3232 nfsd4_store_cache_entry(resp);
3233 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3235 /* Drop session reference that was taken in nfsd4_sequence() */
3236 nfsd4_put_session(cs->session);
3238 put_client_renew(cs->clp);
3242 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3244 struct nfs4_client *conf, *unconf;
3245 struct nfs4_client *clp = NULL;
3247 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3249 spin_lock(&nn->client_lock);
3250 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3251 conf = find_confirmed_client(&dc->clientid, true, nn);
3252 WARN_ON_ONCE(conf && unconf);
3255 if (client_has_state(conf)) {
3256 status = nfserr_clientid_busy;
3259 status = mark_client_expired_locked(conf);
3266 status = nfserr_stale_clientid;
3269 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3271 status = nfserr_wrong_cred;
3274 unhash_client_locked(clp);
3276 spin_unlock(&nn->client_lock);
3283 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3287 if (rc->rca_one_fs) {
3288 if (!cstate->current_fh.fh_dentry)
3289 return nfserr_nofilehandle;
3291 * We don't take advantage of the rca_one_fs case.
3292 * That's OK, it's optional, we can safely ignore it.
3297 status = nfserr_complete_already;
3298 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3299 &cstate->session->se_client->cl_flags))
3302 status = nfserr_stale_clientid;
3303 if (is_client_expired(cstate->session->se_client))
3305 * The following error isn't really legal.
3306 * But we only get here if the client just explicitly
3307 * destroyed the client. Surely it no longer cares what
3308 * error it gets back on an operation for the dead
3314 nfsd4_client_record_create(cstate->session->se_client);
3320 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3321 struct nfsd4_setclientid *setclid)
3323 struct xdr_netobj clname = setclid->se_name;
3324 nfs4_verifier clverifier = setclid->se_verf;
3325 struct nfs4_client *conf, *new;
3326 struct nfs4_client *unconf = NULL;
3328 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3330 new = create_client(clname, rqstp, &clverifier);
3332 return nfserr_jukebox;
3333 /* Cases below refer to rfc 3530 section 14.2.33: */
3334 spin_lock(&nn->client_lock);
3335 conf = find_confirmed_client_by_name(&clname, nn);
3336 if (conf && client_has_state(conf)) {
3338 status = nfserr_clid_inuse;
3339 if (clp_used_exchangeid(conf))
3341 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3342 char addr_str[INET6_ADDRSTRLEN];
3343 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3345 dprintk("NFSD: setclientid: string in use by client "
3346 "at %s\n", addr_str);
3350 unconf = find_unconfirmed_client_by_name(&clname, nn);
3352 unhash_client_locked(unconf);
3353 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3354 /* case 1: probable callback update */
3355 copy_clid(new, conf);
3356 gen_confirm(new, nn);
3357 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3359 new->cl_minorversion = 0;
3360 gen_callback(new, setclid, rqstp);
3361 add_to_unconfirmed(new);
3362 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3363 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3364 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3368 spin_unlock(&nn->client_lock);
3372 expire_client(unconf);
3378 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3379 struct nfsd4_compound_state *cstate,
3380 struct nfsd4_setclientid_confirm *setclientid_confirm)
3382 struct nfs4_client *conf, *unconf;
3383 struct nfs4_client *old = NULL;
3384 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3385 clientid_t * clid = &setclientid_confirm->sc_clientid;
3387 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3389 if (STALE_CLIENTID(clid, nn))
3390 return nfserr_stale_clientid;
3392 spin_lock(&nn->client_lock);
3393 conf = find_confirmed_client(clid, false, nn);
3394 unconf = find_unconfirmed_client(clid, false, nn);
3396 * We try hard to give out unique clientid's, so if we get an
3397 * attempt to confirm the same clientid with a different cred,
3398 * the client may be buggy; this should never happen.
3400 * Nevertheless, RFC 7530 recommends INUSE for this case:
3402 status = nfserr_clid_inuse;
3403 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3405 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3407 /* cases below refer to rfc 3530 section 14.2.34: */
3408 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3409 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3410 /* case 2: probable retransmit */
3412 } else /* case 4: client hasn't noticed we rebooted yet? */
3413 status = nfserr_stale_clientid;
3417 if (conf) { /* case 1: callback update */
3419 unhash_client_locked(old);
3420 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3421 } else { /* case 3: normal case; new or rebooted client */
3422 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3424 status = nfserr_clid_inuse;
3425 if (client_has_state(old)
3426 && !same_creds(&unconf->cl_cred,
3429 status = mark_client_expired_locked(old);
3435 move_to_confirmed(unconf);
3438 get_client_locked(conf);
3439 spin_unlock(&nn->client_lock);
3440 nfsd4_probe_callback(conf);
3441 spin_lock(&nn->client_lock);
3442 put_client_renew_locked(conf);
3444 spin_unlock(&nn->client_lock);
3450 static struct nfs4_file *nfsd4_alloc_file(void)
3452 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3455 /* OPEN Share state helper functions */
3456 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3457 struct nfs4_file *fp)
3459 lockdep_assert_held(&state_lock);
3461 atomic_set(&fp->fi_ref, 1);
3462 spin_lock_init(&fp->fi_lock);
3463 INIT_LIST_HEAD(&fp->fi_stateids);
3464 INIT_LIST_HEAD(&fp->fi_delegations);
3465 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3466 fh_copy_shallow(&fp->fi_fhandle, fh);
3467 fp->fi_deleg_file = NULL;
3468 fp->fi_had_conflict = false;
3469 fp->fi_share_deny = 0;
3470 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3471 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3472 #ifdef CONFIG_NFSD_PNFS
3473 INIT_LIST_HEAD(&fp->fi_lo_states);
3474 atomic_set(&fp->fi_lo_recalls, 0);
3476 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3480 nfsd4_free_slabs(void)
3482 kmem_cache_destroy(odstate_slab);
3483 kmem_cache_destroy(openowner_slab);
3484 kmem_cache_destroy(lockowner_slab);
3485 kmem_cache_destroy(file_slab);
3486 kmem_cache_destroy(stateid_slab);
3487 kmem_cache_destroy(deleg_slab);
3491 nfsd4_init_slabs(void)
3493 openowner_slab = kmem_cache_create("nfsd4_openowners",
3494 sizeof(struct nfs4_openowner), 0, 0, NULL);
3495 if (openowner_slab == NULL)
3497 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3498 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3499 if (lockowner_slab == NULL)
3500 goto out_free_openowner_slab;
3501 file_slab = kmem_cache_create("nfsd4_files",
3502 sizeof(struct nfs4_file), 0, 0, NULL);
3503 if (file_slab == NULL)
3504 goto out_free_lockowner_slab;
3505 stateid_slab = kmem_cache_create("nfsd4_stateids",
3506 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3507 if (stateid_slab == NULL)
3508 goto out_free_file_slab;
3509 deleg_slab = kmem_cache_create("nfsd4_delegations",
3510 sizeof(struct nfs4_delegation), 0, 0, NULL);
3511 if (deleg_slab == NULL)
3512 goto out_free_stateid_slab;
3513 odstate_slab = kmem_cache_create("nfsd4_odstate",
3514 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3515 if (odstate_slab == NULL)
3516 goto out_free_deleg_slab;
3519 out_free_deleg_slab:
3520 kmem_cache_destroy(deleg_slab);
3521 out_free_stateid_slab:
3522 kmem_cache_destroy(stateid_slab);
3524 kmem_cache_destroy(file_slab);
3525 out_free_lockowner_slab:
3526 kmem_cache_destroy(lockowner_slab);
3527 out_free_openowner_slab:
3528 kmem_cache_destroy(openowner_slab);
3530 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3534 static void init_nfs4_replay(struct nfs4_replay *rp)
3536 rp->rp_status = nfserr_serverfault;
3538 rp->rp_buf = rp->rp_ibuf;
3539 mutex_init(&rp->rp_mutex);
3542 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3543 struct nfs4_stateowner *so)
3545 if (!nfsd4_has_session(cstate)) {
3546 mutex_lock(&so->so_replay.rp_mutex);
3547 cstate->replay_owner = nfs4_get_stateowner(so);
3551 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3553 struct nfs4_stateowner *so = cstate->replay_owner;
3556 cstate->replay_owner = NULL;
3557 mutex_unlock(&so->so_replay.rp_mutex);
3558 nfs4_put_stateowner(so);
3562 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3564 struct nfs4_stateowner *sop;
3566 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3570 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3571 if (!sop->so_owner.data) {
3572 kmem_cache_free(slab, sop);
3575 sop->so_owner.len = owner->len;
3577 INIT_LIST_HEAD(&sop->so_stateids);
3578 sop->so_client = clp;
3579 init_nfs4_replay(&sop->so_replay);
3580 atomic_set(&sop->so_count, 1);
3584 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3586 lockdep_assert_held(&clp->cl_lock);
3588 list_add(&oo->oo_owner.so_strhash,
3589 &clp->cl_ownerstr_hashtbl[strhashval]);
3590 list_add(&oo->oo_perclient, &clp->cl_openowners);
3593 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3595 unhash_openowner_locked(openowner(so));
3598 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3600 struct nfs4_openowner *oo = openowner(so);
3602 kmem_cache_free(openowner_slab, oo);
3605 static const struct nfs4_stateowner_operations openowner_ops = {
3606 .so_unhash = nfs4_unhash_openowner,
3607 .so_free = nfs4_free_openowner,
3610 static struct nfs4_ol_stateid *
3611 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3613 struct nfs4_ol_stateid *local, *ret = NULL;
3614 struct nfs4_openowner *oo = open->op_openowner;
3616 lockdep_assert_held(&fp->fi_lock);
3618 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3619 /* ignore lock owners */
3620 if (local->st_stateowner->so_is_open_owner == 0)
3622 if (local->st_stateowner != &oo->oo_owner)
3624 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3626 atomic_inc(&ret->st_stid.sc_count);
3634 nfsd4_verify_open_stid(struct nfs4_stid *s)
3636 __be32 ret = nfs_ok;
3638 switch (s->sc_type) {
3641 case NFS4_CLOSED_STID:
3642 case NFS4_CLOSED_DELEG_STID:
3643 ret = nfserr_bad_stateid;
3645 case NFS4_REVOKED_DELEG_STID:
3646 ret = nfserr_deleg_revoked;
3651 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3653 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3657 mutex_lock(&stp->st_mutex);
3658 ret = nfsd4_verify_open_stid(&stp->st_stid);
3660 mutex_unlock(&stp->st_mutex);
3664 static struct nfs4_ol_stateid *
3665 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3667 struct nfs4_ol_stateid *stp;
3669 spin_lock(&fp->fi_lock);
3670 stp = nfsd4_find_existing_open(fp, open);
3671 spin_unlock(&fp->fi_lock);
3672 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3674 nfs4_put_stid(&stp->st_stid);
3679 static struct nfs4_openowner *
3680 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3681 struct nfsd4_compound_state *cstate)
3683 struct nfs4_client *clp = cstate->clp;
3684 struct nfs4_openowner *oo, *ret;
3686 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3689 oo->oo_owner.so_ops = &openowner_ops;
3690 oo->oo_owner.so_is_open_owner = 1;
3691 oo->oo_owner.so_seqid = open->op_seqid;
3693 if (nfsd4_has_session(cstate))
3694 oo->oo_flags |= NFS4_OO_CONFIRMED;
3696 oo->oo_last_closed_stid = NULL;
3697 INIT_LIST_HEAD(&oo->oo_close_lru);
3698 spin_lock(&clp->cl_lock);
3699 ret = find_openstateowner_str_locked(strhashval, open, clp);
3701 hash_openowner(oo, clp, strhashval);
3704 nfs4_free_stateowner(&oo->oo_owner);
3706 spin_unlock(&clp->cl_lock);
3710 static struct nfs4_ol_stateid *
3711 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3714 struct nfs4_openowner *oo = open->op_openowner;
3715 struct nfs4_ol_stateid *retstp = NULL;
3716 struct nfs4_ol_stateid *stp;
3719 /* We are moving these outside of the spinlocks to avoid the warnings */
3720 mutex_init(&stp->st_mutex);
3721 mutex_lock(&stp->st_mutex);
3724 spin_lock(&oo->oo_owner.so_client->cl_lock);
3725 spin_lock(&fp->fi_lock);
3727 retstp = nfsd4_find_existing_open(fp, open);
3731 open->op_stp = NULL;
3732 atomic_inc(&stp->st_stid.sc_count);
3733 stp->st_stid.sc_type = NFS4_OPEN_STID;
3734 INIT_LIST_HEAD(&stp->st_locks);
3735 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3737 stp->st_stid.sc_file = fp;
3738 stp->st_access_bmap = 0;
3739 stp->st_deny_bmap = 0;
3740 stp->st_openstp = NULL;
3741 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3742 list_add(&stp->st_perfile, &fp->fi_stateids);
3745 spin_unlock(&fp->fi_lock);
3746 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3748 /* Handle races with CLOSE */
3749 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3750 nfs4_put_stid(&retstp->st_stid);
3753 /* To keep mutex tracking happy */
3754 mutex_unlock(&stp->st_mutex);
3761 * In the 4.0 case we need to keep the owners around a little while to handle
3762 * CLOSE replay. We still do need to release any file access that is held by
3763 * them before returning however.
3766 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3768 struct nfs4_ol_stateid *last;
3769 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3770 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3773 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3776 * We know that we hold one reference via nfsd4_close, and another
3777 * "persistent" reference for the client. If the refcount is higher
3778 * than 2, then there are still calls in progress that are using this
3779 * stateid. We can't put the sc_file reference until they are finished.
3780 * Wait for the refcount to drop to 2. Since it has been unhashed,
3781 * there should be no danger of the refcount going back up again at
3784 wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3786 release_all_access(s);
3787 if (s->st_stid.sc_file) {
3788 put_nfs4_file(s->st_stid.sc_file);
3789 s->st_stid.sc_file = NULL;
3792 spin_lock(&nn->client_lock);
3793 last = oo->oo_last_closed_stid;
3794 oo->oo_last_closed_stid = s;
3795 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3796 oo->oo_time = get_seconds();
3797 spin_unlock(&nn->client_lock);
3799 nfs4_put_stid(&last->st_stid);
3802 /* search file_hashtbl[] for file */
3803 static struct nfs4_file *
3804 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3806 struct nfs4_file *fp;
3808 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3809 if (fh_match(&fp->fi_fhandle, fh)) {
3810 if (atomic_inc_not_zero(&fp->fi_ref))
3818 find_file(struct knfsd_fh *fh)
3820 struct nfs4_file *fp;
3821 unsigned int hashval = file_hashval(fh);
3824 fp = find_file_locked(fh, hashval);
3829 static struct nfs4_file *
3830 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3832 struct nfs4_file *fp;
3833 unsigned int hashval = file_hashval(fh);
3836 fp = find_file_locked(fh, hashval);
3841 spin_lock(&state_lock);
3842 fp = find_file_locked(fh, hashval);
3843 if (likely(fp == NULL)) {
3844 nfsd4_init_file(fh, hashval, new);
3847 spin_unlock(&state_lock);
3853 * Called to check deny when READ with all zero stateid or
3854 * WRITE with all zero or all one stateid
3857 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3859 struct nfs4_file *fp;
3860 __be32 ret = nfs_ok;
3862 fp = find_file(¤t_fh->fh_handle);
3865 /* Check for conflicting share reservations */
3866 spin_lock(&fp->fi_lock);
3867 if (fp->fi_share_deny & deny_type)
3868 ret = nfserr_locked;
3869 spin_unlock(&fp->fi_lock);
3874 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3876 struct nfs4_delegation *dp = cb_to_delegation(cb);
3877 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3880 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3883 * We can't do this in nfsd_break_deleg_cb because it is
3884 * already holding inode->i_lock.
3886 * If the dl_time != 0, then we know that it has already been
3887 * queued for a lease break. Don't queue it again.
3889 spin_lock(&state_lock);
3890 if (delegation_hashed(dp) && dp->dl_time == 0) {
3891 dp->dl_time = get_seconds();
3892 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3894 spin_unlock(&state_lock);
3897 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3898 struct rpc_task *task)
3900 struct nfs4_delegation *dp = cb_to_delegation(cb);
3902 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3905 switch (task->tk_status) {
3909 case -NFS4ERR_BAD_STATEID:
3911 * Race: client probably got cb_recall before open reply
3912 * granting delegation.
3914 if (dp->dl_retries--) {
3915 rpc_delay(task, 2 * HZ);
3924 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3926 struct nfs4_delegation *dp = cb_to_delegation(cb);
3928 nfs4_put_stid(&dp->dl_stid);
3931 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3932 .prepare = nfsd4_cb_recall_prepare,
3933 .done = nfsd4_cb_recall_done,
3934 .release = nfsd4_cb_recall_release,
3937 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3940 * We're assuming the state code never drops its reference
3941 * without first removing the lease. Since we're in this lease
3942 * callback (and since the lease code is serialized by the kernel
3943 * lock) we know the server hasn't removed the lease yet, we know
3944 * it's safe to take a reference.
3946 atomic_inc(&dp->dl_stid.sc_count);
3947 nfsd4_run_cb(&dp->dl_recall);
3950 /* Called from break_lease() with i_lock held. */
3952 nfsd_break_deleg_cb(struct file_lock *fl)
3955 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3956 struct nfs4_delegation *dp;
3959 WARN(1, "(%p)->fl_owner NULL\n", fl);
3962 if (fp->fi_had_conflict) {
3963 WARN(1, "duplicate break on %p\n", fp);
3967 * We don't want the locks code to timeout the lease for us;
3968 * we'll remove it ourself if a delegation isn't returned
3971 fl->fl_break_time = 0;
3973 spin_lock(&fp->fi_lock);
3974 fp->fi_had_conflict = true;
3976 * If there are no delegations on the list, then return true
3977 * so that the lease code will go ahead and delete it.
3979 if (list_empty(&fp->fi_delegations))
3982 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3983 nfsd_break_one_deleg(dp);
3984 spin_unlock(&fp->fi_lock);
3989 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3990 struct list_head *dispose)
3993 return lease_modify(onlist, arg, dispose);
3998 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3999 .lm_break = nfsd_break_deleg_cb,
4000 .lm_change = nfsd_change_deleg_cb,
4003 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4005 if (nfsd4_has_session(cstate))
4007 if (seqid == so->so_seqid - 1)
4008 return nfserr_replay_me;
4009 if (seqid == so->so_seqid)
4011 return nfserr_bad_seqid;
4014 static __be32 lookup_clientid(clientid_t *clid,
4015 struct nfsd4_compound_state *cstate,
4016 struct nfsd_net *nn)
4018 struct nfs4_client *found;
4021 found = cstate->clp;
4022 if (!same_clid(&found->cl_clientid, clid))
4023 return nfserr_stale_clientid;
4027 if (STALE_CLIENTID(clid, nn))
4028 return nfserr_stale_clientid;
4031 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4032 * cached already then we know this is for is for v4.0 and "sessions"
4035 WARN_ON_ONCE(cstate->session);
4036 spin_lock(&nn->client_lock);
4037 found = find_confirmed_client(clid, false, nn);
4039 spin_unlock(&nn->client_lock);
4040 return nfserr_expired;
4042 atomic_inc(&found->cl_refcount);
4043 spin_unlock(&nn->client_lock);
4045 /* Cache the nfs4_client in cstate! */
4046 cstate->clp = found;
4051 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4052 struct nfsd4_open *open, struct nfsd_net *nn)
4054 clientid_t *clientid = &open->op_clientid;
4055 struct nfs4_client *clp = NULL;
4056 unsigned int strhashval;
4057 struct nfs4_openowner *oo = NULL;
4060 if (STALE_CLIENTID(&open->op_clientid, nn))
4061 return nfserr_stale_clientid;
4063 * In case we need it later, after we've already created the
4064 * file and don't want to risk a further failure:
4066 open->op_file = nfsd4_alloc_file();
4067 if (open->op_file == NULL)
4068 return nfserr_jukebox;
4070 status = lookup_clientid(clientid, cstate, nn);
4075 strhashval = ownerstr_hashval(&open->op_owner);
4076 oo = find_openstateowner_str(strhashval, open, clp);
4077 open->op_openowner = oo;
4081 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4082 /* Replace unconfirmed owners without checking for replay. */
4083 release_openowner(oo);
4084 open->op_openowner = NULL;
4087 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4092 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4094 return nfserr_jukebox;
4095 open->op_openowner = oo;
4097 open->op_stp = nfs4_alloc_open_stateid(clp);
4099 return nfserr_jukebox;
4101 if (nfsd4_has_session(cstate) &&
4102 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4103 open->op_odstate = alloc_clnt_odstate(clp);
4104 if (!open->op_odstate)
4105 return nfserr_jukebox;
4111 static inline __be32
4112 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4114 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4115 return nfserr_openmode;
4120 static int share_access_to_flags(u32 share_access)
4122 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4125 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4127 struct nfs4_stid *ret;
4129 ret = find_stateid_by_type(cl, s,
4130 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4133 return delegstateid(ret);
4136 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4138 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4139 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4143 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4144 struct nfs4_delegation **dp)
4147 __be32 status = nfserr_bad_stateid;
4148 struct nfs4_delegation *deleg;
4150 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4153 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4154 nfs4_put_stid(&deleg->dl_stid);
4155 if (cl->cl_minorversion)
4156 status = nfserr_deleg_revoked;
4159 flags = share_access_to_flags(open->op_share_access);
4160 status = nfs4_check_delegmode(deleg, flags);
4162 nfs4_put_stid(&deleg->dl_stid);
4167 if (!nfsd4_is_deleg_cur(open))
4171 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4175 static inline int nfs4_access_to_access(u32 nfs4_access)
4179 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4180 flags |= NFSD_MAY_READ;
4181 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4182 flags |= NFSD_MAY_WRITE;
4186 static inline __be32
4187 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4188 struct nfsd4_open *open)
4190 struct iattr iattr = {
4191 .ia_valid = ATTR_SIZE,
4194 if (!open->op_truncate)
4196 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4197 return nfserr_inval;
4198 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4201 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4202 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4203 struct nfsd4_open *open)
4205 struct file *filp = NULL;
4207 int oflag = nfs4_access_to_omode(open->op_share_access);
4208 int access = nfs4_access_to_access(open->op_share_access);
4209 unsigned char old_access_bmap, old_deny_bmap;
4211 spin_lock(&fp->fi_lock);
4214 * Are we trying to set a deny mode that would conflict with
4217 status = nfs4_file_check_deny(fp, open->op_share_deny);
4218 if (status != nfs_ok) {
4219 spin_unlock(&fp->fi_lock);
4223 /* set access to the file */
4224 status = nfs4_file_get_access(fp, open->op_share_access);
4225 if (status != nfs_ok) {
4226 spin_unlock(&fp->fi_lock);
4230 /* Set access bits in stateid */
4231 old_access_bmap = stp->st_access_bmap;
4232 set_access(open->op_share_access, stp);
4234 /* Set new deny mask */
4235 old_deny_bmap = stp->st_deny_bmap;
4236 set_deny(open->op_share_deny, stp);
4237 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4239 if (!fp->fi_fds[oflag]) {
4240 spin_unlock(&fp->fi_lock);
4241 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4243 goto out_put_access;
4244 spin_lock(&fp->fi_lock);
4245 if (!fp->fi_fds[oflag]) {
4246 fp->fi_fds[oflag] = filp;
4250 spin_unlock(&fp->fi_lock);
4254 status = nfsd4_truncate(rqstp, cur_fh, open);
4256 goto out_put_access;
4260 stp->st_access_bmap = old_access_bmap;
4261 nfs4_file_put_access(fp, open->op_share_access);
4262 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4267 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)
4270 unsigned char old_deny_bmap = stp->st_deny_bmap;
4272 if (!test_access(open->op_share_access, stp))
4273 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4275 /* test and set deny mode */
4276 spin_lock(&fp->fi_lock);
4277 status = nfs4_file_check_deny(fp, open->op_share_deny);
4278 if (status == nfs_ok) {
4279 set_deny(open->op_share_deny, stp);
4280 fp->fi_share_deny |=
4281 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4283 spin_unlock(&fp->fi_lock);
4285 if (status != nfs_ok)
4288 status = nfsd4_truncate(rqstp, cur_fh, open);
4289 if (status != nfs_ok)
4290 reset_union_bmap_deny(old_deny_bmap, stp);
4294 /* Should we give out recallable state?: */
4295 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4297 if (clp->cl_cb_state == NFSD4_CB_UP)
4300 * In the sessions case, since we don't have to establish a
4301 * separate connection for callbacks, we assume it's OK
4302 * until we hear otherwise:
4304 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4307 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4309 struct file_lock *fl;
4311 fl = locks_alloc_lock();
4314 fl->fl_lmops = &nfsd_lease_mng_ops;
4315 fl->fl_flags = FL_DELEG;
4316 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4317 fl->fl_end = OFFSET_MAX;
4318 fl->fl_owner = (fl_owner_t)fp;
4319 fl->fl_pid = current->tgid;
4324 * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4325 * @dp: a pointer to the nfs4_delegation we're adding.
4328 * On success: Return code will be 0 on success.
4330 * On error: -EAGAIN if there was an existing delegation.
4331 * nonzero if there is an error in other cases.
4335 static int nfs4_setlease(struct nfs4_delegation *dp)
4337 struct nfs4_file *fp = dp->dl_stid.sc_file;
4338 struct file_lock *fl;
4342 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4345 filp = find_readable_file(fp);
4347 /* We should always have a readable file here */
4349 locks_free_lock(fl);
4353 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4355 locks_free_lock(fl);
4358 spin_lock(&state_lock);
4359 spin_lock(&fp->fi_lock);
4360 /* Did the lease get broken before we took the lock? */
4362 if (fp->fi_had_conflict)
4365 if (fp->fi_deleg_file) {
4366 status = hash_delegation_locked(dp, fp);
4369 fp->fi_deleg_file = filp;
4370 fp->fi_delegees = 0;
4371 status = hash_delegation_locked(dp, fp);
4372 spin_unlock(&fp->fi_lock);
4373 spin_unlock(&state_lock);
4375 /* Should never happen, this is a new fi_deleg_file */
4381 spin_unlock(&fp->fi_lock);
4382 spin_unlock(&state_lock);
4388 static struct nfs4_delegation *
4389 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4390 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4393 struct nfs4_delegation *dp;
4395 if (fp->fi_had_conflict)
4396 return ERR_PTR(-EAGAIN);
4398 spin_lock(&state_lock);
4399 spin_lock(&fp->fi_lock);
4400 status = nfs4_get_existing_delegation(clp, fp);
4401 spin_unlock(&fp->fi_lock);
4402 spin_unlock(&state_lock);
4405 return ERR_PTR(status);
4407 dp = alloc_init_deleg(clp, fh, odstate);
4409 return ERR_PTR(-ENOMEM);
4412 spin_lock(&state_lock);
4413 spin_lock(&fp->fi_lock);
4414 dp->dl_stid.sc_file = fp;
4415 if (!fp->fi_deleg_file) {
4416 spin_unlock(&fp->fi_lock);
4417 spin_unlock(&state_lock);
4418 status = nfs4_setlease(dp);
4421 if (fp->fi_had_conflict) {
4425 status = hash_delegation_locked(dp, fp);
4427 spin_unlock(&fp->fi_lock);
4428 spin_unlock(&state_lock);
4431 put_clnt_odstate(dp->dl_clnt_odstate);
4432 nfs4_put_stid(&dp->dl_stid);
4433 return ERR_PTR(status);
4438 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4440 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4441 if (status == -EAGAIN)
4442 open->op_why_no_deleg = WND4_CONTENTION;
4444 open->op_why_no_deleg = WND4_RESOURCE;
4445 switch (open->op_deleg_want) {
4446 case NFS4_SHARE_WANT_READ_DELEG:
4447 case NFS4_SHARE_WANT_WRITE_DELEG:
4448 case NFS4_SHARE_WANT_ANY_DELEG:
4450 case NFS4_SHARE_WANT_CANCEL:
4451 open->op_why_no_deleg = WND4_CANCELLED;
4453 case NFS4_SHARE_WANT_NO_DELEG:
4460 * Attempt to hand out a delegation.
4462 * Note we don't support write delegations, and won't until the vfs has
4463 * proper support for them.
4466 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4467 struct nfs4_ol_stateid *stp)
4469 struct nfs4_delegation *dp;
4470 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4471 struct nfs4_client *clp = stp->st_stid.sc_client;
4475 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4476 open->op_recall = 0;
4477 switch (open->op_claim_type) {
4478 case NFS4_OPEN_CLAIM_PREVIOUS:
4480 open->op_recall = 1;
4481 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4484 case NFS4_OPEN_CLAIM_NULL:
4485 case NFS4_OPEN_CLAIM_FH:
4487 * Let's not give out any delegations till everyone's
4488 * had the chance to reclaim theirs, *and* until
4489 * NLM locks have all been reclaimed:
4491 if (locks_in_grace(clp->net))
4493 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4496 * Also, if the file was opened for write or
4497 * create, there's a good chance the client's
4498 * about to write to it, resulting in an
4499 * immediate recall (since we don't support
4500 * write delegations):
4502 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4504 if (open->op_create == NFS4_OPEN_CREATE)
4510 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4514 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4516 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4517 STATEID_VAL(&dp->dl_stid.sc_stateid));
4518 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4519 nfs4_put_stid(&dp->dl_stid);
4522 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4523 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4524 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4525 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4526 open->op_recall = 1;
4529 /* 4.1 client asking for a delegation? */
4530 if (open->op_deleg_want)
4531 nfsd4_open_deleg_none_ext(open, status);
4535 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4536 struct nfs4_delegation *dp)
4538 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4539 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4540 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4541 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4542 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4543 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4544 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4545 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4547 /* Otherwise the client must be confused wanting a delegation
4548 * it already has, therefore we don't return
4549 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4554 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4556 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4557 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4558 struct nfs4_file *fp = NULL;
4559 struct nfs4_ol_stateid *stp = NULL;
4560 struct nfs4_delegation *dp = NULL;
4562 bool new_stp = false;
4565 * Lookup file; if found, lookup stateid and check open request,
4566 * and check for delegations in the process of being recalled.
4567 * If not found, create the nfs4_file struct
4569 fp = find_or_add_file(open->op_file, ¤t_fh->fh_handle);
4570 if (fp != open->op_file) {
4571 status = nfs4_check_deleg(cl, open, &dp);
4574 stp = nfsd4_find_and_lock_existing_open(fp, open);
4576 open->op_file = NULL;
4577 status = nfserr_bad_stateid;
4578 if (nfsd4_is_deleg_cur(open))
4583 stp = init_open_stateid(fp, open);
4589 * OPEN the file, or upgrade an existing OPEN.
4590 * If truncate fails, the OPEN fails.
4592 * stp is already locked.
4595 /* Stateid was found, this is an OPEN upgrade */
4596 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4598 mutex_unlock(&stp->st_mutex);
4602 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4604 stp->st_stid.sc_type = NFS4_CLOSED_STID;
4605 release_open_stateid(stp);
4606 mutex_unlock(&stp->st_mutex);
4610 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4612 if (stp->st_clnt_odstate == open->op_odstate)
4613 open->op_odstate = NULL;
4616 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4617 mutex_unlock(&stp->st_mutex);
4619 if (nfsd4_has_session(&resp->cstate)) {
4620 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4621 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4622 open->op_why_no_deleg = WND4_NOT_WANTED;
4628 * Attempt to hand out a delegation. No error return, because the
4629 * OPEN succeeds even if we fail.
4631 nfs4_open_delegation(current_fh, open, stp);
4635 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4636 STATEID_VAL(&stp->st_stid.sc_stateid));
4638 /* 4.1 client trying to upgrade/downgrade delegation? */
4639 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4640 open->op_deleg_want)
4641 nfsd4_deleg_xgrade_none_ext(open, dp);
4645 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4646 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4648 * To finish the open response, we just need to set the rflags.
4650 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4651 if (nfsd4_has_session(&resp->cstate))
4652 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4653 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4654 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4657 nfs4_put_stid(&dp->dl_stid);
4659 nfs4_put_stid(&stp->st_stid);
4664 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4665 struct nfsd4_open *open)
4667 if (open->op_openowner) {
4668 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4670 nfsd4_cstate_assign_replay(cstate, so);
4671 nfs4_put_stateowner(so);
4674 kmem_cache_free(file_slab, open->op_file);
4676 nfs4_put_stid(&open->op_stp->st_stid);
4677 if (open->op_odstate)
4678 kmem_cache_free(odstate_slab, open->op_odstate);
4682 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4685 struct nfs4_client *clp;
4687 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4689 dprintk("process_renew(%08x/%08x): starting\n",
4690 clid->cl_boot, clid->cl_id);
4691 status = lookup_clientid(clid, cstate, nn);
4695 status = nfserr_cb_path_down;
4696 if (!list_empty(&clp->cl_delegations)
4697 && clp->cl_cb_state != NFSD4_CB_UP)
4705 nfsd4_end_grace(struct nfsd_net *nn)
4707 /* do nothing if grace period already ended */
4708 if (nn->grace_ended)
4711 dprintk("NFSD: end of grace period\n");
4712 nn->grace_ended = true;
4714 * If the server goes down again right now, an NFSv4
4715 * client will still be allowed to reclaim after it comes back up,
4716 * even if it hasn't yet had a chance to reclaim state this time.
4719 nfsd4_record_grace_done(nn);
4721 * At this point, NFSv4 clients can still reclaim. But if the
4722 * server crashes, any that have not yet reclaimed will be out
4723 * of luck on the next boot.
4725 * (NFSv4.1+ clients are considered to have reclaimed once they
4726 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4727 * have reclaimed after their first OPEN.)
4729 locks_end_grace(&nn->nfsd4_manager);
4731 * At this point, and once lockd and/or any other containers
4732 * exit their grace period, further reclaims will fail and
4733 * regular locking can resume.
4738 nfs4_laundromat(struct nfsd_net *nn)
4740 struct nfs4_client *clp;
4741 struct nfs4_openowner *oo;
4742 struct nfs4_delegation *dp;
4743 struct nfs4_ol_stateid *stp;
4744 struct nfsd4_blocked_lock *nbl;
4745 struct list_head *pos, *next, reaplist;
4746 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4747 time_t t, new_timeo = nn->nfsd4_lease;
4749 dprintk("NFSD: laundromat service - starting\n");
4750 nfsd4_end_grace(nn);
4751 INIT_LIST_HEAD(&reaplist);
4752 spin_lock(&nn->client_lock);
4753 list_for_each_safe(pos, next, &nn->client_lru) {
4754 clp = list_entry(pos, struct nfs4_client, cl_lru);
4755 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4756 t = clp->cl_time - cutoff;
4757 new_timeo = min(new_timeo, t);
4760 if (mark_client_expired_locked(clp)) {
4761 dprintk("NFSD: client in use (clientid %08x)\n",
4762 clp->cl_clientid.cl_id);
4765 list_add(&clp->cl_lru, &reaplist);
4767 spin_unlock(&nn->client_lock);
4768 list_for_each_safe(pos, next, &reaplist) {
4769 clp = list_entry(pos, struct nfs4_client, cl_lru);
4770 dprintk("NFSD: purging unused client (clientid %08x)\n",
4771 clp->cl_clientid.cl_id);
4772 list_del_init(&clp->cl_lru);
4775 spin_lock(&state_lock);
4776 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4777 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4778 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4779 t = dp->dl_time - cutoff;
4780 new_timeo = min(new_timeo, t);
4783 WARN_ON(!unhash_delegation_locked(dp));
4784 list_add(&dp->dl_recall_lru, &reaplist);
4786 spin_unlock(&state_lock);
4787 while (!list_empty(&reaplist)) {
4788 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4790 list_del_init(&dp->dl_recall_lru);
4791 revoke_delegation(dp);
4794 spin_lock(&nn->client_lock);
4795 while (!list_empty(&nn->close_lru)) {
4796 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4798 if (time_after((unsigned long)oo->oo_time,
4799 (unsigned long)cutoff)) {
4800 t = oo->oo_time - cutoff;
4801 new_timeo = min(new_timeo, t);
4804 list_del_init(&oo->oo_close_lru);
4805 stp = oo->oo_last_closed_stid;
4806 oo->oo_last_closed_stid = NULL;
4807 spin_unlock(&nn->client_lock);
4808 nfs4_put_stid(&stp->st_stid);
4809 spin_lock(&nn->client_lock);
4811 spin_unlock(&nn->client_lock);
4814 * It's possible for a client to try and acquire an already held lock
4815 * that is being held for a long time, and then lose interest in it.
4816 * So, we clean out any un-revisited request after a lease period
4817 * under the assumption that the client is no longer interested.
4819 * RFC5661, sec. 9.6 states that the client must not rely on getting
4820 * notifications and must continue to poll for locks, even when the
4821 * server supports them. Thus this shouldn't lead to clients blocking
4822 * indefinitely once the lock does become free.
4824 BUG_ON(!list_empty(&reaplist));
4825 spin_lock(&nn->blocked_locks_lock);
4826 while (!list_empty(&nn->blocked_locks_lru)) {
4827 nbl = list_first_entry(&nn->blocked_locks_lru,
4828 struct nfsd4_blocked_lock, nbl_lru);
4829 if (time_after((unsigned long)nbl->nbl_time,
4830 (unsigned long)cutoff)) {
4831 t = nbl->nbl_time - cutoff;
4832 new_timeo = min(new_timeo, t);
4835 list_move(&nbl->nbl_lru, &reaplist);
4836 list_del_init(&nbl->nbl_list);
4838 spin_unlock(&nn->blocked_locks_lock);
4840 while (!list_empty(&reaplist)) {
4841 nbl = list_first_entry(&reaplist,
4842 struct nfsd4_blocked_lock, nbl_lru);
4843 list_del_init(&nbl->nbl_lru);
4844 posix_unblock_lock(&nbl->nbl_lock);
4845 free_blocked_lock(nbl);
4848 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4852 static struct workqueue_struct *laundry_wq;
4853 static void laundromat_main(struct work_struct *);
4856 laundromat_main(struct work_struct *laundry)
4859 struct delayed_work *dwork = to_delayed_work(laundry);
4860 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4863 t = nfs4_laundromat(nn);
4864 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4865 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4868 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4870 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4871 return nfserr_bad_stateid;
4876 access_permit_read(struct nfs4_ol_stateid *stp)
4878 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4879 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4880 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4884 access_permit_write(struct nfs4_ol_stateid *stp)
4886 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4887 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4891 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4893 __be32 status = nfserr_openmode;
4895 /* For lock stateid's, we test the parent open, not the lock: */
4896 if (stp->st_openstp)
4897 stp = stp->st_openstp;
4898 if ((flags & WR_STATE) && !access_permit_write(stp))
4900 if ((flags & RD_STATE) && !access_permit_read(stp))
4907 static inline __be32
4908 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4910 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4912 else if (opens_in_grace(net)) {
4913 /* Answer in remaining cases depends on existence of
4914 * conflicting state; so we must wait out the grace period. */
4915 return nfserr_grace;
4916 } else if (flags & WR_STATE)
4917 return nfs4_share_conflict(current_fh,
4918 NFS4_SHARE_DENY_WRITE);
4919 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4920 return nfs4_share_conflict(current_fh,
4921 NFS4_SHARE_DENY_READ);
4925 * Allow READ/WRITE during grace period on recovered state only for files
4926 * that are not able to provide mandatory locking.
4929 grace_disallows_io(struct net *net, struct inode *inode)
4931 return opens_in_grace(net) && mandatory_lock(inode);
4934 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4937 * When sessions are used the stateid generation number is ignored
4940 if (has_session && in->si_generation == 0)
4943 if (in->si_generation == ref->si_generation)
4946 /* If the client sends us a stateid from the future, it's buggy: */
4947 if (nfsd4_stateid_generation_after(in, ref))
4948 return nfserr_bad_stateid;
4950 * However, we could see a stateid from the past, even from a
4951 * non-buggy client. For example, if the client sends a lock
4952 * while some IO is outstanding, the lock may bump si_generation
4953 * while the IO is still in flight. The client could avoid that
4954 * situation by waiting for responses on all the IO requests,
4955 * but better performance may result in retrying IO that
4956 * receives an old_stateid error if requests are rarely
4957 * reordered in flight:
4959 return nfserr_old_stateid;
4962 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4964 if (ols->st_stateowner->so_is_open_owner &&
4965 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4966 return nfserr_bad_stateid;
4970 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4972 struct nfs4_stid *s;
4973 __be32 status = nfserr_bad_stateid;
4975 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4976 CLOSE_STATEID(stateid))
4978 /* Client debugging aid. */
4979 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4980 char addr_str[INET6_ADDRSTRLEN];
4981 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4983 pr_warn_ratelimited("NFSD: client %s testing state ID "
4984 "with incorrect client ID\n", addr_str);
4987 spin_lock(&cl->cl_lock);
4988 s = find_stateid_locked(cl, stateid);
4991 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4994 switch (s->sc_type) {
4995 case NFS4_DELEG_STID:
4998 case NFS4_REVOKED_DELEG_STID:
4999 status = nfserr_deleg_revoked;
5001 case NFS4_OPEN_STID:
5002 case NFS4_LOCK_STID:
5003 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5006 printk("unknown stateid type %x\n", s->sc_type);
5008 case NFS4_CLOSED_STID:
5009 case NFS4_CLOSED_DELEG_STID:
5010 status = nfserr_bad_stateid;
5013 spin_unlock(&cl->cl_lock);
5018 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5019 stateid_t *stateid, unsigned char typemask,
5020 struct nfs4_stid **s, struct nfsd_net *nn)
5023 bool return_revoked = false;
5026 * only return revoked delegations if explicitly asked.
5027 * otherwise we report revoked or bad_stateid status.
5029 if (typemask & NFS4_REVOKED_DELEG_STID)
5030 return_revoked = true;
5031 else if (typemask & NFS4_DELEG_STID)
5032 typemask |= NFS4_REVOKED_DELEG_STID;
5034 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5035 CLOSE_STATEID(stateid))
5036 return nfserr_bad_stateid;
5037 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5038 if (status == nfserr_stale_clientid) {
5039 if (cstate->session)
5040 return nfserr_bad_stateid;
5041 return nfserr_stale_stateid;
5045 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5047 return nfserr_bad_stateid;
5048 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5050 if (cstate->minorversion)
5051 return nfserr_deleg_revoked;
5052 return nfserr_bad_stateid;
5057 static struct file *
5058 nfs4_find_file(struct nfs4_stid *s, int flags)
5063 switch (s->sc_type) {
5064 case NFS4_DELEG_STID:
5065 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5067 return get_file(s->sc_file->fi_deleg_file);
5068 case NFS4_OPEN_STID:
5069 case NFS4_LOCK_STID:
5070 if (flags & RD_STATE)
5071 return find_readable_file(s->sc_file);
5073 return find_writeable_file(s->sc_file);
5081 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5085 status = nfsd4_check_openowner_confirmed(ols);
5088 return nfs4_check_openmode(ols, flags);
5092 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5093 struct file **filpp, bool *tmp_file, int flags)
5095 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5099 file = nfs4_find_file(s, flags);
5101 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5102 acc | NFSD_MAY_OWNER_OVERRIDE);
5110 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5122 * Checks for stateid operations
5125 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5126 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5127 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5129 struct inode *ino = d_inode(fhp->fh_dentry);
5130 struct net *net = SVC_NET(rqstp);
5131 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5132 struct nfs4_stid *s = NULL;
5140 if (grace_disallows_io(net, ino))
5141 return nfserr_grace;
5143 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5144 status = check_special_stateids(net, fhp, stateid, flags);
5148 status = nfsd4_lookup_stateid(cstate, stateid,
5149 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5153 status = check_stateid_generation(stateid, &s->sc_stateid,
5154 nfsd4_has_session(cstate));
5158 switch (s->sc_type) {
5159 case NFS4_DELEG_STID:
5160 status = nfs4_check_delegmode(delegstateid(s), flags);
5162 case NFS4_OPEN_STID:
5163 case NFS4_LOCK_STID:
5164 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5167 status = nfserr_bad_stateid;
5172 status = nfs4_check_fh(fhp, s);
5175 if (!status && filpp)
5176 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5184 * Test if the stateid is valid
5187 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5188 struct nfsd4_test_stateid *test_stateid)
5190 struct nfsd4_test_stateid_id *stateid;
5191 struct nfs4_client *cl = cstate->session->se_client;
5193 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5194 stateid->ts_id_status =
5195 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5201 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5203 struct nfs4_ol_stateid *stp = openlockstateid(s);
5206 mutex_lock(&stp->st_mutex);
5208 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5212 ret = nfserr_locks_held;
5213 if (check_for_locks(stp->st_stid.sc_file,
5214 lockowner(stp->st_stateowner)))
5217 release_lock_stateid(stp);
5221 mutex_unlock(&stp->st_mutex);
5227 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5228 struct nfsd4_free_stateid *free_stateid)
5230 stateid_t *stateid = &free_stateid->fr_stateid;
5231 struct nfs4_stid *s;
5232 struct nfs4_delegation *dp;
5233 struct nfs4_client *cl = cstate->session->se_client;
5234 __be32 ret = nfserr_bad_stateid;
5236 spin_lock(&cl->cl_lock);
5237 s = find_stateid_locked(cl, stateid);
5240 switch (s->sc_type) {
5241 case NFS4_DELEG_STID:
5242 ret = nfserr_locks_held;
5244 case NFS4_OPEN_STID:
5245 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5248 ret = nfserr_locks_held;
5250 case NFS4_LOCK_STID:
5251 atomic_inc(&s->sc_count);
5252 spin_unlock(&cl->cl_lock);
5253 ret = nfsd4_free_lock_stateid(stateid, s);
5255 case NFS4_REVOKED_DELEG_STID:
5256 dp = delegstateid(s);
5257 list_del_init(&dp->dl_recall_lru);
5258 spin_unlock(&cl->cl_lock);
5262 /* Default falls through and returns nfserr_bad_stateid */
5265 spin_unlock(&cl->cl_lock);
5273 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5274 RD_STATE : WR_STATE;
5277 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5279 struct svc_fh *current_fh = &cstate->current_fh;
5280 struct nfs4_stateowner *sop = stp->st_stateowner;
5283 status = nfsd4_check_seqid(cstate, sop, seqid);
5286 status = nfsd4_lock_ol_stateid(stp);
5287 if (status != nfs_ok)
5289 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5290 if (status == nfs_ok)
5291 status = nfs4_check_fh(current_fh, &stp->st_stid);
5292 if (status != nfs_ok)
5293 mutex_unlock(&stp->st_mutex);
5298 * Checks for sequence id mutating operations.
5301 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5302 stateid_t *stateid, char typemask,
5303 struct nfs4_ol_stateid **stpp,
5304 struct nfsd_net *nn)
5307 struct nfs4_stid *s;
5308 struct nfs4_ol_stateid *stp = NULL;
5310 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5311 seqid, STATEID_VAL(stateid));
5314 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5317 stp = openlockstateid(s);
5318 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5320 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5324 nfs4_put_stid(&stp->st_stid);
5328 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5329 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5332 struct nfs4_openowner *oo;
5333 struct nfs4_ol_stateid *stp;
5335 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5336 NFS4_OPEN_STID, &stp, nn);
5339 oo = openowner(stp->st_stateowner);
5340 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5341 mutex_unlock(&stp->st_mutex);
5342 nfs4_put_stid(&stp->st_stid);
5343 return nfserr_bad_stateid;
5350 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5351 struct nfsd4_open_confirm *oc)
5354 struct nfs4_openowner *oo;
5355 struct nfs4_ol_stateid *stp;
5356 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5358 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5359 cstate->current_fh.fh_dentry);
5361 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5365 status = nfs4_preprocess_seqid_op(cstate,
5366 oc->oc_seqid, &oc->oc_req_stateid,
5367 NFS4_OPEN_STID, &stp, nn);
5370 oo = openowner(stp->st_stateowner);
5371 status = nfserr_bad_stateid;
5372 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5373 mutex_unlock(&stp->st_mutex);
5376 oo->oo_flags |= NFS4_OO_CONFIRMED;
5377 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5378 mutex_unlock(&stp->st_mutex);
5379 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5380 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5382 nfsd4_client_record_create(oo->oo_owner.so_client);
5385 nfs4_put_stid(&stp->st_stid);
5387 nfsd4_bump_seqid(cstate, status);
5391 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5393 if (!test_access(access, stp))
5395 nfs4_file_put_access(stp->st_stid.sc_file, access);
5396 clear_access(access, stp);
5399 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5401 switch (to_access) {
5402 case NFS4_SHARE_ACCESS_READ:
5403 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5404 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5406 case NFS4_SHARE_ACCESS_WRITE:
5407 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5408 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5410 case NFS4_SHARE_ACCESS_BOTH:
5418 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5419 struct nfsd4_compound_state *cstate,
5420 struct nfsd4_open_downgrade *od)
5423 struct nfs4_ol_stateid *stp;
5424 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5426 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5427 cstate->current_fh.fh_dentry);
5429 /* We don't yet support WANT bits: */
5430 if (od->od_deleg_want)
5431 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5434 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5435 &od->od_stateid, &stp, nn);
5438 status = nfserr_inval;
5439 if (!test_access(od->od_share_access, stp)) {
5440 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5441 stp->st_access_bmap, od->od_share_access);
5444 if (!test_deny(od->od_share_deny, stp)) {
5445 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5446 stp->st_deny_bmap, od->od_share_deny);
5449 nfs4_stateid_downgrade(stp, od->od_share_access);
5450 reset_union_bmap_deny(od->od_share_deny, stp);
5451 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5454 mutex_unlock(&stp->st_mutex);
5455 nfs4_put_stid(&stp->st_stid);
5457 nfsd4_bump_seqid(cstate, status);
5461 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5463 struct nfs4_client *clp = s->st_stid.sc_client;
5465 LIST_HEAD(reaplist);
5467 spin_lock(&clp->cl_lock);
5468 unhashed = unhash_open_stateid(s, &reaplist);
5470 if (clp->cl_minorversion) {
5472 put_ol_stateid_locked(s, &reaplist);
5473 spin_unlock(&clp->cl_lock);
5474 free_ol_stateid_reaplist(&reaplist);
5476 spin_unlock(&clp->cl_lock);
5477 free_ol_stateid_reaplist(&reaplist);
5479 move_to_close_lru(s, clp->net);
5484 * nfs4_unlock_state() called after encode
5487 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5488 struct nfsd4_close *close)
5491 struct nfs4_ol_stateid *stp;
5492 struct net *net = SVC_NET(rqstp);
5493 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5495 dprintk("NFSD: nfsd4_close on file %pd\n",
5496 cstate->current_fh.fh_dentry);
5498 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5500 NFS4_OPEN_STID|NFS4_CLOSED_STID,
5502 nfsd4_bump_seqid(cstate, status);
5506 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5507 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5509 nfsd4_close_open_stateid(stp);
5510 mutex_unlock(&stp->st_mutex);
5512 /* See RFC5661 sectionm 18.2.4 */
5513 if (stp->st_stid.sc_client->cl_minorversion)
5514 memcpy(&close->cl_stateid, &close_stateid,
5515 sizeof(close->cl_stateid));
5517 /* put reference from nfs4_preprocess_seqid_op */
5518 nfs4_put_stid(&stp->st_stid);
5524 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5525 struct nfsd4_delegreturn *dr)
5527 struct nfs4_delegation *dp;
5528 stateid_t *stateid = &dr->dr_stateid;
5529 struct nfs4_stid *s;
5531 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5533 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5536 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5539 dp = delegstateid(s);
5540 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5544 destroy_delegation(dp);
5546 nfs4_put_stid(&dp->dl_stid);
5552 end_offset(u64 start, u64 len)
5557 return end >= start ? end: NFS4_MAX_UINT64;
5560 /* last octet in a range */
5562 last_byte_offset(u64 start, u64 len)
5568 return end > start ? end - 1: NFS4_MAX_UINT64;
5572 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5573 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5574 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
5575 * locking, this prevents us from being completely protocol-compliant. The
5576 * real solution to this problem is to start using unsigned file offsets in
5577 * the VFS, but this is a very deep change!
5580 nfs4_transform_lock_offset(struct file_lock *lock)
5582 if (lock->fl_start < 0)
5583 lock->fl_start = OFFSET_MAX;
5584 if (lock->fl_end < 0)
5585 lock->fl_end = OFFSET_MAX;
5589 nfsd4_fl_get_owner(fl_owner_t owner)
5591 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5593 nfs4_get_stateowner(&lo->lo_owner);
5598 nfsd4_fl_put_owner(fl_owner_t owner)
5600 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5603 nfs4_put_stateowner(&lo->lo_owner);
5607 nfsd4_lm_notify(struct file_lock *fl)
5609 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
5610 struct net *net = lo->lo_owner.so_client->net;
5611 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5612 struct nfsd4_blocked_lock *nbl = container_of(fl,
5613 struct nfsd4_blocked_lock, nbl_lock);
5616 /* An empty list means that something else is going to be using it */
5617 spin_lock(&nn->blocked_locks_lock);
5618 if (!list_empty(&nbl->nbl_list)) {
5619 list_del_init(&nbl->nbl_list);
5620 list_del_init(&nbl->nbl_lru);
5623 spin_unlock(&nn->blocked_locks_lock);
5626 nfsd4_run_cb(&nbl->nbl_cb);
5629 static const struct lock_manager_operations nfsd_posix_mng_ops = {
5630 .lm_notify = nfsd4_lm_notify,
5631 .lm_get_owner = nfsd4_fl_get_owner,
5632 .lm_put_owner = nfsd4_fl_put_owner,
5636 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5638 struct nfs4_lockowner *lo;
5640 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5641 lo = (struct nfs4_lockowner *) fl->fl_owner;
5642 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5643 lo->lo_owner.so_owner.len, GFP_KERNEL);
5644 if (!deny->ld_owner.data)
5645 /* We just don't care that much */
5647 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5648 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5651 deny->ld_owner.len = 0;
5652 deny->ld_owner.data = NULL;
5653 deny->ld_clientid.cl_boot = 0;
5654 deny->ld_clientid.cl_id = 0;
5656 deny->ld_start = fl->fl_start;
5657 deny->ld_length = NFS4_MAX_UINT64;
5658 if (fl->fl_end != NFS4_MAX_UINT64)
5659 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5660 deny->ld_type = NFS4_READ_LT;
5661 if (fl->fl_type != F_RDLCK)
5662 deny->ld_type = NFS4_WRITE_LT;
5665 static struct nfs4_lockowner *
5666 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5668 unsigned int strhashval = ownerstr_hashval(owner);
5669 struct nfs4_stateowner *so;
5671 lockdep_assert_held(&clp->cl_lock);
5673 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5675 if (so->so_is_open_owner)
5677 if (same_owner_str(so, owner))
5678 return lockowner(nfs4_get_stateowner(so));
5683 static struct nfs4_lockowner *
5684 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5686 struct nfs4_lockowner *lo;
5688 spin_lock(&clp->cl_lock);
5689 lo = find_lockowner_str_locked(clp, owner);
5690 spin_unlock(&clp->cl_lock);
5694 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5696 unhash_lockowner_locked(lockowner(sop));
5699 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5701 struct nfs4_lockowner *lo = lockowner(sop);
5703 kmem_cache_free(lockowner_slab, lo);
5706 static const struct nfs4_stateowner_operations lockowner_ops = {
5707 .so_unhash = nfs4_unhash_lockowner,
5708 .so_free = nfs4_free_lockowner,
5712 * Alloc a lock owner structure.
5713 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5716 * strhashval = ownerstr_hashval
5718 static struct nfs4_lockowner *
5719 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5720 struct nfs4_ol_stateid *open_stp,
5721 struct nfsd4_lock *lock)
5723 struct nfs4_lockowner *lo, *ret;
5725 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5728 INIT_LIST_HEAD(&lo->lo_blocked);
5729 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5730 lo->lo_owner.so_is_open_owner = 0;
5731 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5732 lo->lo_owner.so_ops = &lockowner_ops;
5733 spin_lock(&clp->cl_lock);
5734 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5736 list_add(&lo->lo_owner.so_strhash,
5737 &clp->cl_ownerstr_hashtbl[strhashval]);
5740 nfs4_free_stateowner(&lo->lo_owner);
5742 spin_unlock(&clp->cl_lock);
5747 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5748 struct nfs4_file *fp, struct inode *inode,
5749 struct nfs4_ol_stateid *open_stp)
5751 struct nfs4_client *clp = lo->lo_owner.so_client;
5753 lockdep_assert_held(&clp->cl_lock);
5755 atomic_inc(&stp->st_stid.sc_count);
5756 stp->st_stid.sc_type = NFS4_LOCK_STID;
5757 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5759 stp->st_stid.sc_file = fp;
5760 stp->st_access_bmap = 0;
5761 stp->st_deny_bmap = open_stp->st_deny_bmap;
5762 stp->st_openstp = open_stp;
5763 mutex_init(&stp->st_mutex);
5764 list_add(&stp->st_locks, &open_stp->st_locks);
5765 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5766 spin_lock(&fp->fi_lock);
5767 list_add(&stp->st_perfile, &fp->fi_stateids);
5768 spin_unlock(&fp->fi_lock);
5771 static struct nfs4_ol_stateid *
5772 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5774 struct nfs4_ol_stateid *lst;
5775 struct nfs4_client *clp = lo->lo_owner.so_client;
5777 lockdep_assert_held(&clp->cl_lock);
5779 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5780 if (lst->st_stid.sc_file == fp) {
5781 atomic_inc(&lst->st_stid.sc_count);
5788 static struct nfs4_ol_stateid *
5789 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5790 struct inode *inode, struct nfs4_ol_stateid *ost,
5793 struct nfs4_stid *ns = NULL;
5794 struct nfs4_ol_stateid *lst;
5795 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5796 struct nfs4_client *clp = oo->oo_owner.so_client;
5798 spin_lock(&clp->cl_lock);
5799 lst = find_lock_stateid(lo, fi);
5801 spin_unlock(&clp->cl_lock);
5802 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5806 spin_lock(&clp->cl_lock);
5807 lst = find_lock_stateid(lo, fi);
5809 lst = openlockstateid(ns);
5810 init_lock_stateid(lst, lo, fi, inode, ost);
5815 spin_unlock(&clp->cl_lock);
5822 check_lock_length(u64 offset, u64 length)
5824 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5825 (length > ~offset)));
5828 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5830 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5832 lockdep_assert_held(&fp->fi_lock);
5834 if (test_access(access, lock_stp))
5836 __nfs4_file_get_access(fp, access);
5837 set_access(access, lock_stp);
5841 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5842 struct nfs4_ol_stateid *ost,
5843 struct nfsd4_lock *lock,
5844 struct nfs4_ol_stateid **plst, bool *new)
5847 struct nfs4_file *fi = ost->st_stid.sc_file;
5848 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5849 struct nfs4_client *cl = oo->oo_owner.so_client;
5850 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5851 struct nfs4_lockowner *lo;
5852 struct nfs4_ol_stateid *lst;
5853 unsigned int strhashval;
5856 lo = find_lockowner_str(cl, &lock->lk_new_owner);
5858 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5859 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5861 return nfserr_jukebox;
5863 /* with an existing lockowner, seqids must be the same */
5864 status = nfserr_bad_seqid;
5865 if (!cstate->minorversion &&
5866 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5871 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5873 status = nfserr_jukebox;
5877 mutex_lock(&lst->st_mutex);
5879 /* See if it's still hashed to avoid race with FREE_STATEID */
5880 spin_lock(&cl->cl_lock);
5881 hashed = !list_empty(&lst->st_perfile);
5882 spin_unlock(&cl->cl_lock);
5885 mutex_unlock(&lst->st_mutex);
5886 nfs4_put_stid(&lst->st_stid);
5892 nfs4_put_stateowner(&lo->lo_owner);
5900 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5901 struct nfsd4_lock *lock)
5903 struct nfs4_openowner *open_sop = NULL;
5904 struct nfs4_lockowner *lock_sop = NULL;
5905 struct nfs4_ol_stateid *lock_stp = NULL;
5906 struct nfs4_ol_stateid *open_stp = NULL;
5907 struct nfs4_file *fp;
5908 struct file *filp = NULL;
5909 struct nfsd4_blocked_lock *nbl = NULL;
5910 struct file_lock *file_lock = NULL;
5911 struct file_lock *conflock = NULL;
5916 unsigned char fl_type;
5917 unsigned int fl_flags = FL_POSIX;
5918 struct net *net = SVC_NET(rqstp);
5919 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5921 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5922 (long long) lock->lk_offset,
5923 (long long) lock->lk_length);
5925 if (check_lock_length(lock->lk_offset, lock->lk_length))
5926 return nfserr_inval;
5928 if ((status = fh_verify(rqstp, &cstate->current_fh,
5929 S_IFREG, NFSD_MAY_LOCK))) {
5930 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5934 if (lock->lk_is_new) {
5935 if (nfsd4_has_session(cstate))
5936 /* See rfc 5661 18.10.3: given clientid is ignored: */
5937 memcpy(&lock->lk_new_clientid,
5938 &cstate->session->se_client->cl_clientid,
5939 sizeof(clientid_t));
5941 status = nfserr_stale_clientid;
5942 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5945 /* validate and update open stateid and open seqid */
5946 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5947 lock->lk_new_open_seqid,
5948 &lock->lk_new_open_stateid,
5952 mutex_unlock(&open_stp->st_mutex);
5953 open_sop = openowner(open_stp->st_stateowner);
5954 status = nfserr_bad_stateid;
5955 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5956 &lock->lk_new_clientid))
5958 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5961 status = nfs4_preprocess_seqid_op(cstate,
5962 lock->lk_old_lock_seqid,
5963 &lock->lk_old_lock_stateid,
5964 NFS4_LOCK_STID, &lock_stp, nn);
5968 lock_sop = lockowner(lock_stp->st_stateowner);
5970 lkflg = setlkflg(lock->lk_type);
5971 status = nfs4_check_openmode(lock_stp, lkflg);
5975 status = nfserr_grace;
5976 if (locks_in_grace(net) && !lock->lk_reclaim)
5978 status = nfserr_no_grace;
5979 if (!locks_in_grace(net) && lock->lk_reclaim)
5982 fp = lock_stp->st_stid.sc_file;
5983 switch (lock->lk_type) {
5985 if (nfsd4_has_session(cstate))
5986 fl_flags |= FL_SLEEP;
5989 spin_lock(&fp->fi_lock);
5990 filp = find_readable_file_locked(fp);
5992 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5993 spin_unlock(&fp->fi_lock);
5996 case NFS4_WRITEW_LT:
5997 if (nfsd4_has_session(cstate))
5998 fl_flags |= FL_SLEEP;
6001 spin_lock(&fp->fi_lock);
6002 filp = find_writeable_file_locked(fp);
6004 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6005 spin_unlock(&fp->fi_lock);
6009 status = nfserr_inval;
6014 status = nfserr_openmode;
6018 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6020 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6021 status = nfserr_jukebox;
6025 file_lock = &nbl->nbl_lock;
6026 file_lock->fl_type = fl_type;
6027 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6028 file_lock->fl_pid = current->tgid;
6029 file_lock->fl_file = filp;
6030 file_lock->fl_flags = fl_flags;
6031 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6032 file_lock->fl_start = lock->lk_offset;
6033 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6034 nfs4_transform_lock_offset(file_lock);
6036 conflock = locks_alloc_lock();
6038 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6039 status = nfserr_jukebox;
6043 if (fl_flags & FL_SLEEP) {
6044 nbl->nbl_time = get_seconds();
6045 spin_lock(&nn->blocked_locks_lock);
6046 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6047 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6048 spin_unlock(&nn->blocked_locks_lock);
6051 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6053 case 0: /* success! */
6054 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6057 case FILE_LOCK_DEFERRED:
6060 case -EAGAIN: /* conflock holds conflicting lock */
6061 status = nfserr_denied;
6062 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6063 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6066 status = nfserr_deadlock;
6069 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6070 status = nfserrno(err);
6075 /* dequeue it if we queued it before */
6076 if (fl_flags & FL_SLEEP) {
6077 spin_lock(&nn->blocked_locks_lock);
6078 list_del_init(&nbl->nbl_list);
6079 list_del_init(&nbl->nbl_lru);
6080 spin_unlock(&nn->blocked_locks_lock);
6082 free_blocked_lock(nbl);
6087 /* Bump seqid manually if the 4.0 replay owner is openowner */
6088 if (cstate->replay_owner &&
6089 cstate->replay_owner != &lock_sop->lo_owner &&
6090 seqid_mutating_err(ntohl(status)))
6091 lock_sop->lo_owner.so_seqid++;
6093 mutex_unlock(&lock_stp->st_mutex);
6096 * If this is a new, never-before-used stateid, and we are
6097 * returning an error, then just go ahead and release it.
6100 release_lock_stateid(lock_stp);
6102 nfs4_put_stid(&lock_stp->st_stid);
6105 nfs4_put_stid(&open_stp->st_stid);
6106 nfsd4_bump_seqid(cstate, status);
6108 locks_free_lock(conflock);
6113 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6114 * so we do a temporary open here just to get an open file to pass to
6115 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6118 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6121 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6123 err = nfserrno(vfs_test_lock(file, lock));
6133 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6134 struct nfsd4_lockt *lockt)
6136 struct file_lock *file_lock = NULL;
6137 struct nfs4_lockowner *lo = NULL;
6139 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6141 if (locks_in_grace(SVC_NET(rqstp)))
6142 return nfserr_grace;
6144 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6145 return nfserr_inval;
6147 if (!nfsd4_has_session(cstate)) {
6148 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6153 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6156 file_lock = locks_alloc_lock();
6158 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6159 status = nfserr_jukebox;
6163 switch (lockt->lt_type) {
6166 file_lock->fl_type = F_RDLCK;
6169 case NFS4_WRITEW_LT:
6170 file_lock->fl_type = F_WRLCK;
6173 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6174 status = nfserr_inval;
6178 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6180 file_lock->fl_owner = (fl_owner_t)lo;
6181 file_lock->fl_pid = current->tgid;
6182 file_lock->fl_flags = FL_POSIX;
6184 file_lock->fl_start = lockt->lt_offset;
6185 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6187 nfs4_transform_lock_offset(file_lock);
6189 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6193 if (file_lock->fl_type != F_UNLCK) {
6194 status = nfserr_denied;
6195 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6199 nfs4_put_stateowner(&lo->lo_owner);
6201 locks_free_lock(file_lock);
6206 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6207 struct nfsd4_locku *locku)
6209 struct nfs4_ol_stateid *stp;
6210 struct file *filp = NULL;
6211 struct file_lock *file_lock = NULL;
6214 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6216 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6217 (long long) locku->lu_offset,
6218 (long long) locku->lu_length);
6220 if (check_lock_length(locku->lu_offset, locku->lu_length))
6221 return nfserr_inval;
6223 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6224 &locku->lu_stateid, NFS4_LOCK_STID,
6228 filp = find_any_file(stp->st_stid.sc_file);
6230 status = nfserr_lock_range;
6233 file_lock = locks_alloc_lock();
6235 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6236 status = nfserr_jukebox;
6240 file_lock->fl_type = F_UNLCK;
6241 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6242 file_lock->fl_pid = current->tgid;
6243 file_lock->fl_file = filp;
6244 file_lock->fl_flags = FL_POSIX;
6245 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6246 file_lock->fl_start = locku->lu_offset;
6248 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6250 nfs4_transform_lock_offset(file_lock);
6252 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6254 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6257 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6261 mutex_unlock(&stp->st_mutex);
6262 nfs4_put_stid(&stp->st_stid);
6264 nfsd4_bump_seqid(cstate, status);
6266 locks_free_lock(file_lock);
6270 status = nfserrno(err);
6276 * true: locks held by lockowner
6277 * false: no locks held by lockowner
6280 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6282 struct file_lock *fl;
6284 struct file *filp = find_any_file(fp);
6285 struct inode *inode;
6286 struct file_lock_context *flctx;
6289 /* Any valid lock stateid should have some sort of access */
6294 inode = file_inode(filp);
6295 flctx = inode->i_flctx;
6297 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6298 spin_lock(&flctx->flc_lock);
6299 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6300 if (fl->fl_owner == (fl_owner_t)lowner) {
6305 spin_unlock(&flctx->flc_lock);
6312 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6313 struct nfsd4_compound_state *cstate,
6314 struct nfsd4_release_lockowner *rlockowner)
6316 clientid_t *clid = &rlockowner->rl_clientid;
6317 struct nfs4_stateowner *sop;
6318 struct nfs4_lockowner *lo = NULL;
6319 struct nfs4_ol_stateid *stp;
6320 struct xdr_netobj *owner = &rlockowner->rl_owner;
6321 unsigned int hashval = ownerstr_hashval(owner);
6323 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6324 struct nfs4_client *clp;
6325 LIST_HEAD (reaplist);
6327 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6328 clid->cl_boot, clid->cl_id);
6330 status = lookup_clientid(clid, cstate, nn);
6335 /* Find the matching lock stateowner */
6336 spin_lock(&clp->cl_lock);
6337 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6340 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6343 /* see if there are still any locks associated with it */
6344 lo = lockowner(sop);
6345 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6346 if (check_for_locks(stp->st_stid.sc_file, lo)) {
6347 status = nfserr_locks_held;
6348 spin_unlock(&clp->cl_lock);
6353 nfs4_get_stateowner(sop);
6357 spin_unlock(&clp->cl_lock);
6361 unhash_lockowner_locked(lo);
6362 while (!list_empty(&lo->lo_owner.so_stateids)) {
6363 stp = list_first_entry(&lo->lo_owner.so_stateids,
6364 struct nfs4_ol_stateid,
6366 WARN_ON(!unhash_lock_stateid(stp));
6367 put_ol_stateid_locked(stp, &reaplist);
6369 spin_unlock(&clp->cl_lock);
6370 free_ol_stateid_reaplist(&reaplist);
6371 remove_blocked_locks(lo);
6372 nfs4_put_stateowner(&lo->lo_owner);
6377 static inline struct nfs4_client_reclaim *
6380 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6384 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6386 struct nfs4_client_reclaim *crp;
6388 crp = nfsd4_find_reclaim_client(name, nn);
6389 return (crp && crp->cr_clp);
6393 * failure => all reset bets are off, nfserr_no_grace...
6395 struct nfs4_client_reclaim *
6396 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6398 unsigned int strhashval;
6399 struct nfs4_client_reclaim *crp;
6401 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6402 crp = alloc_reclaim();
6404 strhashval = clientstr_hashval(name);
6405 INIT_LIST_HEAD(&crp->cr_strhash);
6406 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6407 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6409 nn->reclaim_str_hashtbl_size++;
6415 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6417 list_del(&crp->cr_strhash);
6419 nn->reclaim_str_hashtbl_size--;
6423 nfs4_release_reclaim(struct nfsd_net *nn)
6425 struct nfs4_client_reclaim *crp = NULL;
6428 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6429 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6430 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6431 struct nfs4_client_reclaim, cr_strhash);
6432 nfs4_remove_reclaim_record(crp, nn);
6435 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6439 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6440 struct nfs4_client_reclaim *
6441 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6443 unsigned int strhashval;
6444 struct nfs4_client_reclaim *crp = NULL;
6446 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6448 strhashval = clientstr_hashval(recdir);
6449 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6450 if (same_name(crp->cr_recdir, recdir)) {
6458 * Called from OPEN. Look for clientid in reclaim list.
6461 nfs4_check_open_reclaim(clientid_t *clid,
6462 struct nfsd4_compound_state *cstate,
6463 struct nfsd_net *nn)
6467 /* find clientid in conf_id_hashtbl */
6468 status = lookup_clientid(clid, cstate, nn);
6470 return nfserr_reclaim_bad;
6472 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6473 return nfserr_no_grace;
6475 if (nfsd4_client_record_check(cstate->clp))
6476 return nfserr_reclaim_bad;
6481 #ifdef CONFIG_NFSD_FAULT_INJECTION
6483 put_client(struct nfs4_client *clp)
6485 atomic_dec(&clp->cl_refcount);
6488 static struct nfs4_client *
6489 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6491 struct nfs4_client *clp;
6492 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6495 if (!nfsd_netns_ready(nn))
6498 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6499 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6506 nfsd_inject_print_clients(void)
6508 struct nfs4_client *clp;
6510 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6512 char buf[INET6_ADDRSTRLEN];
6514 if (!nfsd_netns_ready(nn))
6517 spin_lock(&nn->client_lock);
6518 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6519 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6520 pr_info("NFS Client: %s\n", buf);
6523 spin_unlock(&nn->client_lock);
6529 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6532 struct nfs4_client *clp;
6533 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6536 if (!nfsd_netns_ready(nn))
6539 spin_lock(&nn->client_lock);
6540 clp = nfsd_find_client(addr, addr_size);
6542 if (mark_client_expired_locked(clp) == nfs_ok)
6547 spin_unlock(&nn->client_lock);
6556 nfsd_inject_forget_clients(u64 max)
6559 struct nfs4_client *clp, *next;
6560 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6562 LIST_HEAD(reaplist);
6564 if (!nfsd_netns_ready(nn))
6567 spin_lock(&nn->client_lock);
6568 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6569 if (mark_client_expired_locked(clp) == nfs_ok) {
6570 list_add(&clp->cl_lru, &reaplist);
6571 if (max != 0 && ++count >= max)
6575 spin_unlock(&nn->client_lock);
6577 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6583 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6586 char buf[INET6_ADDRSTRLEN];
6587 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6588 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6592 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6593 struct list_head *collect)
6595 struct nfs4_client *clp = lst->st_stid.sc_client;
6596 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6602 lockdep_assert_held(&nn->client_lock);
6603 atomic_inc(&clp->cl_refcount);
6604 list_add(&lst->st_locks, collect);
6607 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6608 struct list_head *collect,
6609 bool (*func)(struct nfs4_ol_stateid *))
6611 struct nfs4_openowner *oop;
6612 struct nfs4_ol_stateid *stp, *st_next;
6613 struct nfs4_ol_stateid *lst, *lst_next;
6616 spin_lock(&clp->cl_lock);
6617 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6618 list_for_each_entry_safe(stp, st_next,
6619 &oop->oo_owner.so_stateids, st_perstateowner) {
6620 list_for_each_entry_safe(lst, lst_next,
6621 &stp->st_locks, st_locks) {
6624 nfsd_inject_add_lock_to_list(lst,
6629 * Despite the fact that these functions deal
6630 * with 64-bit integers for "count", we must
6631 * ensure that it doesn't blow up the
6632 * clp->cl_refcount. Throw a warning if we
6633 * start to approach INT_MAX here.
6635 WARN_ON_ONCE(count == (INT_MAX / 2));
6642 spin_unlock(&clp->cl_lock);
6648 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6651 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6655 nfsd_print_client_locks(struct nfs4_client *clp)
6657 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6658 nfsd_print_count(clp, count, "locked files");
6663 nfsd_inject_print_locks(void)
6665 struct nfs4_client *clp;
6667 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6670 if (!nfsd_netns_ready(nn))
6673 spin_lock(&nn->client_lock);
6674 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6675 count += nfsd_print_client_locks(clp);
6676 spin_unlock(&nn->client_lock);
6682 nfsd_reap_locks(struct list_head *reaplist)
6684 struct nfs4_client *clp;
6685 struct nfs4_ol_stateid *stp, *next;
6687 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6688 list_del_init(&stp->st_locks);
6689 clp = stp->st_stid.sc_client;
6690 nfs4_put_stid(&stp->st_stid);
6696 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6698 unsigned int count = 0;
6699 struct nfs4_client *clp;
6700 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6702 LIST_HEAD(reaplist);
6704 if (!nfsd_netns_ready(nn))
6707 spin_lock(&nn->client_lock);
6708 clp = nfsd_find_client(addr, addr_size);
6710 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6711 spin_unlock(&nn->client_lock);
6712 nfsd_reap_locks(&reaplist);
6717 nfsd_inject_forget_locks(u64 max)
6720 struct nfs4_client *clp;
6721 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6723 LIST_HEAD(reaplist);
6725 if (!nfsd_netns_ready(nn))
6728 spin_lock(&nn->client_lock);
6729 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6730 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6731 if (max != 0 && count >= max)
6734 spin_unlock(&nn->client_lock);
6735 nfsd_reap_locks(&reaplist);
6740 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6741 struct list_head *collect,
6742 void (*func)(struct nfs4_openowner *))
6744 struct nfs4_openowner *oop, *next;
6745 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6749 lockdep_assert_held(&nn->client_lock);
6751 spin_lock(&clp->cl_lock);
6752 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6756 atomic_inc(&clp->cl_refcount);
6757 list_add(&oop->oo_perclient, collect);
6762 * Despite the fact that these functions deal with
6763 * 64-bit integers for "count", we must ensure that
6764 * it doesn't blow up the clp->cl_refcount. Throw a
6765 * warning if we start to approach INT_MAX here.
6767 WARN_ON_ONCE(count == (INT_MAX / 2));
6771 spin_unlock(&clp->cl_lock);
6777 nfsd_print_client_openowners(struct nfs4_client *clp)
6779 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6781 nfsd_print_count(clp, count, "openowners");
6786 nfsd_collect_client_openowners(struct nfs4_client *clp,
6787 struct list_head *collect, u64 max)
6789 return nfsd_foreach_client_openowner(clp, max, collect,
6790 unhash_openowner_locked);
6794 nfsd_inject_print_openowners(void)
6796 struct nfs4_client *clp;
6798 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6801 if (!nfsd_netns_ready(nn))
6804 spin_lock(&nn->client_lock);
6805 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6806 count += nfsd_print_client_openowners(clp);
6807 spin_unlock(&nn->client_lock);
6813 nfsd_reap_openowners(struct list_head *reaplist)
6815 struct nfs4_client *clp;
6816 struct nfs4_openowner *oop, *next;
6818 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6819 list_del_init(&oop->oo_perclient);
6820 clp = oop->oo_owner.so_client;
6821 release_openowner(oop);
6827 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6830 unsigned int count = 0;
6831 struct nfs4_client *clp;
6832 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6834 LIST_HEAD(reaplist);
6836 if (!nfsd_netns_ready(nn))
6839 spin_lock(&nn->client_lock);
6840 clp = nfsd_find_client(addr, addr_size);
6842 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6843 spin_unlock(&nn->client_lock);
6844 nfsd_reap_openowners(&reaplist);
6849 nfsd_inject_forget_openowners(u64 max)
6852 struct nfs4_client *clp;
6853 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6855 LIST_HEAD(reaplist);
6857 if (!nfsd_netns_ready(nn))
6860 spin_lock(&nn->client_lock);
6861 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6862 count += nfsd_collect_client_openowners(clp, &reaplist,
6864 if (max != 0 && count >= max)
6867 spin_unlock(&nn->client_lock);
6868 nfsd_reap_openowners(&reaplist);
6872 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6873 struct list_head *victims)
6875 struct nfs4_delegation *dp, *next;
6876 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6880 lockdep_assert_held(&nn->client_lock);
6882 spin_lock(&state_lock);
6883 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6886 * It's not safe to mess with delegations that have a
6887 * non-zero dl_time. They might have already been broken
6888 * and could be processed by the laundromat outside of
6889 * the state_lock. Just leave them be.
6891 if (dp->dl_time != 0)
6894 atomic_inc(&clp->cl_refcount);
6895 WARN_ON(!unhash_delegation_locked(dp));
6896 list_add(&dp->dl_recall_lru, victims);
6900 * Despite the fact that these functions deal with
6901 * 64-bit integers for "count", we must ensure that
6902 * it doesn't blow up the clp->cl_refcount. Throw a
6903 * warning if we start to approach INT_MAX here.
6905 WARN_ON_ONCE(count == (INT_MAX / 2));
6909 spin_unlock(&state_lock);
6914 nfsd_print_client_delegations(struct nfs4_client *clp)
6916 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6918 nfsd_print_count(clp, count, "delegations");
6923 nfsd_inject_print_delegations(void)
6925 struct nfs4_client *clp;
6927 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6930 if (!nfsd_netns_ready(nn))
6933 spin_lock(&nn->client_lock);
6934 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6935 count += nfsd_print_client_delegations(clp);
6936 spin_unlock(&nn->client_lock);
6942 nfsd_forget_delegations(struct list_head *reaplist)
6944 struct nfs4_client *clp;
6945 struct nfs4_delegation *dp, *next;
6947 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6948 list_del_init(&dp->dl_recall_lru);
6949 clp = dp->dl_stid.sc_client;
6950 revoke_delegation(dp);
6956 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6960 struct nfs4_client *clp;
6961 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6963 LIST_HEAD(reaplist);
6965 if (!nfsd_netns_ready(nn))
6968 spin_lock(&nn->client_lock);
6969 clp = nfsd_find_client(addr, addr_size);
6971 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6972 spin_unlock(&nn->client_lock);
6974 nfsd_forget_delegations(&reaplist);
6979 nfsd_inject_forget_delegations(u64 max)
6982 struct nfs4_client *clp;
6983 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6985 LIST_HEAD(reaplist);
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_find_all_delegations(clp, max - count, &reaplist);
6993 if (max != 0 && count >= max)
6996 spin_unlock(&nn->client_lock);
6997 nfsd_forget_delegations(&reaplist);
7002 nfsd_recall_delegations(struct list_head *reaplist)
7004 struct nfs4_client *clp;
7005 struct nfs4_delegation *dp, *next;
7007 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7008 list_del_init(&dp->dl_recall_lru);
7009 clp = dp->dl_stid.sc_client;
7011 * We skipped all entries that had a zero dl_time before,
7012 * so we can now reset the dl_time back to 0. If a delegation
7013 * break comes in now, then it won't make any difference since
7014 * we're recalling it either way.
7016 spin_lock(&state_lock);
7018 spin_unlock(&state_lock);
7019 nfsd_break_one_deleg(dp);
7025 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7029 struct nfs4_client *clp;
7030 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7032 LIST_HEAD(reaplist);
7034 if (!nfsd_netns_ready(nn))
7037 spin_lock(&nn->client_lock);
7038 clp = nfsd_find_client(addr, addr_size);
7040 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7041 spin_unlock(&nn->client_lock);
7043 nfsd_recall_delegations(&reaplist);
7048 nfsd_inject_recall_delegations(u64 max)
7051 struct nfs4_client *clp, *next;
7052 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7054 LIST_HEAD(reaplist);
7056 if (!nfsd_netns_ready(nn))
7059 spin_lock(&nn->client_lock);
7060 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7061 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7062 if (max != 0 && ++count >= max)
7065 spin_unlock(&nn->client_lock);
7066 nfsd_recall_delegations(&reaplist);
7069 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7072 * Since the lifetime of a delegation isn't limited to that of an open, a
7073 * client may quite reasonably hang on to a delegation as long as it has
7074 * the inode cached. This becomes an obvious problem the first time a
7075 * client's inode cache approaches the size of the server's total memory.
7077 * For now we avoid this problem by imposing a hard limit on the number
7078 * of delegations, which varies according to the server's memory size.
7081 set_max_delegations(void)
7084 * Allow at most 4 delegations per megabyte of RAM. Quick
7085 * estimates suggest that in the worst case (where every delegation
7086 * is for a different inode), a delegation could take about 1.5K,
7087 * giving a worst case usage of about 6% of memory.
7089 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7092 static int nfs4_state_create_net(struct net *net)
7094 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7097 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7098 CLIENT_HASH_SIZE, GFP_KERNEL);
7099 if (!nn->conf_id_hashtbl)
7101 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7102 CLIENT_HASH_SIZE, GFP_KERNEL);
7103 if (!nn->unconf_id_hashtbl)
7105 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
7106 SESSION_HASH_SIZE, GFP_KERNEL);
7107 if (!nn->sessionid_hashtbl)
7110 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7111 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7112 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7114 for (i = 0; i < SESSION_HASH_SIZE; i++)
7115 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7116 nn->conf_name_tree = RB_ROOT;
7117 nn->unconf_name_tree = RB_ROOT;
7118 nn->boot_time = get_seconds();
7119 nn->grace_ended = false;
7120 nn->nfsd4_manager.block_opens = true;
7121 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7122 INIT_LIST_HEAD(&nn->client_lru);
7123 INIT_LIST_HEAD(&nn->close_lru);
7124 INIT_LIST_HEAD(&nn->del_recall_lru);
7125 spin_lock_init(&nn->client_lock);
7127 spin_lock_init(&nn->blocked_locks_lock);
7128 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7130 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7136 kfree(nn->unconf_id_hashtbl);
7138 kfree(nn->conf_id_hashtbl);
7144 nfs4_state_destroy_net(struct net *net)
7147 struct nfs4_client *clp = NULL;
7148 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7150 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7151 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7152 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7153 destroy_client(clp);
7157 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7159 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7160 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7161 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7162 destroy_client(clp);
7166 kfree(nn->sessionid_hashtbl);
7167 kfree(nn->unconf_id_hashtbl);
7168 kfree(nn->conf_id_hashtbl);
7173 nfs4_state_start_net(struct net *net)
7175 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7178 ret = nfs4_state_create_net(net);
7181 locks_start_grace(net, &nn->nfsd4_manager);
7182 nfsd4_client_tracking_init(net);
7183 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
7184 nn->nfsd4_grace, net);
7185 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7189 /* initialization to perform when the nfsd service is started: */
7192 nfs4_state_start(void)
7196 ret = set_callback_cred();
7200 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7201 if (laundry_wq == NULL) {
7203 goto out_cleanup_cred;
7205 ret = nfsd4_create_callback_queue();
7207 goto out_free_laundry;
7209 set_max_delegations();
7213 destroy_workqueue(laundry_wq);
7215 cleanup_callback_cred();
7220 nfs4_state_shutdown_net(struct net *net)
7222 struct nfs4_delegation *dp = NULL;
7223 struct list_head *pos, *next, reaplist;
7224 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7226 cancel_delayed_work_sync(&nn->laundromat_work);
7227 locks_end_grace(&nn->nfsd4_manager);
7229 INIT_LIST_HEAD(&reaplist);
7230 spin_lock(&state_lock);
7231 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7232 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7233 WARN_ON(!unhash_delegation_locked(dp));
7234 list_add(&dp->dl_recall_lru, &reaplist);
7236 spin_unlock(&state_lock);
7237 list_for_each_safe(pos, next, &reaplist) {
7238 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7239 list_del_init(&dp->dl_recall_lru);
7240 put_clnt_odstate(dp->dl_clnt_odstate);
7241 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7242 nfs4_put_stid(&dp->dl_stid);
7245 nfsd4_client_tracking_exit(net);
7246 nfs4_state_destroy_net(net);
7250 nfs4_state_shutdown(void)
7252 destroy_workqueue(laundry_wq);
7253 nfsd4_destroy_callback_queue();
7254 cleanup_callback_cred();
7258 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7260 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7261 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7265 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7267 if (cstate->minorversion) {
7268 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7269 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7274 clear_current_stateid(struct nfsd4_compound_state *cstate)
7276 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7280 * functions to set current state id
7283 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7285 put_stateid(cstate, &odp->od_stateid);
7289 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
7291 put_stateid(cstate, &open->op_stateid);
7295 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7297 put_stateid(cstate, &close->cl_stateid);
7301 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
7303 put_stateid(cstate, &lock->lk_resp_stateid);
7307 * functions to consume current state id
7311 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
7313 get_stateid(cstate, &odp->od_stateid);
7317 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
7319 get_stateid(cstate, &drp->dr_stateid);
7323 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
7325 get_stateid(cstate, &fsp->fr_stateid);
7329 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
7331 get_stateid(cstate, &setattr->sa_stateid);
7335 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
7337 get_stateid(cstate, &close->cl_stateid);
7341 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
7343 get_stateid(cstate, &locku->lu_stateid);
7347 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
7349 get_stateid(cstate, &read->rd_stateid);
7353 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
7355 get_stateid(cstate, &write->wr_stateid);