Merge tag 'fsnotify_for_v4.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / fs / nfs / delegation.c
1 /*
2  * linux/fs/nfs/delegation.c
3  *
4  * Copyright (C) 2004 Trond Myklebust
5  *
6  * NFS file delegation management
7  *
8  */
9 #include <linux/completion.h>
10 #include <linux/kthread.h>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/iversion.h>
16
17 #include <linux/nfs4.h>
18 #include <linux/nfs_fs.h>
19 #include <linux/nfs_xdr.h>
20
21 #include "nfs4_fs.h"
22 #include "nfs4session.h"
23 #include "delegation.h"
24 #include "internal.h"
25 #include "nfs4trace.h"
26
27 static void nfs_free_delegation(struct nfs_delegation *delegation)
28 {
29         if (delegation->cred) {
30                 put_rpccred(delegation->cred);
31                 delegation->cred = NULL;
32         }
33         kfree_rcu(delegation, rcu);
34 }
35
36 /**
37  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
38  * @delegation: delegation to process
39  *
40  */
41 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
42 {
43         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
44 }
45
46 static bool
47 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
48                 fmode_t flags)
49 {
50         if (delegation != NULL && (delegation->type & flags) == flags &&
51             !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
52             !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
53                 return true;
54         return false;
55 }
56
57 static int
58 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
59 {
60         struct nfs_delegation *delegation;
61         int ret = 0;
62
63         flags &= FMODE_READ|FMODE_WRITE;
64         rcu_read_lock();
65         delegation = rcu_dereference(NFS_I(inode)->delegation);
66         if (nfs4_is_valid_delegation(delegation, flags)) {
67                 if (mark)
68                         nfs_mark_delegation_referenced(delegation);
69                 ret = 1;
70         }
71         rcu_read_unlock();
72         return ret;
73 }
74 /**
75  * nfs_have_delegation - check if inode has a delegation, mark it
76  * NFS_DELEGATION_REFERENCED if there is one.
77  * @inode: inode to check
78  * @flags: delegation types to check for
79  *
80  * Returns one if inode has the indicated delegation, otherwise zero.
81  */
82 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
83 {
84         return nfs4_do_check_delegation(inode, flags, true);
85 }
86
87 /*
88  * nfs4_check_delegation - check if inode has a delegation, do not mark
89  * NFS_DELEGATION_REFERENCED if it has one.
90  */
91 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
92 {
93         return nfs4_do_check_delegation(inode, flags, false);
94 }
95
96 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state, const nfs4_stateid *stateid)
97 {
98         struct inode *inode = state->inode;
99         struct file_lock *fl;
100         struct file_lock_context *flctx = inode->i_flctx;
101         struct list_head *list;
102         int status = 0;
103
104         if (flctx == NULL)
105                 goto out;
106
107         list = &flctx->flc_posix;
108         spin_lock(&flctx->flc_lock);
109 restart:
110         list_for_each_entry(fl, list, fl_list) {
111                 if (nfs_file_open_context(fl->fl_file) != ctx)
112                         continue;
113                 spin_unlock(&flctx->flc_lock);
114                 status = nfs4_lock_delegation_recall(fl, state, stateid);
115                 if (status < 0)
116                         goto out;
117                 spin_lock(&flctx->flc_lock);
118         }
119         if (list == &flctx->flc_posix) {
120                 list = &flctx->flc_flock;
121                 goto restart;
122         }
123         spin_unlock(&flctx->flc_lock);
124 out:
125         return status;
126 }
127
128 static int nfs_delegation_claim_opens(struct inode *inode,
129                 const nfs4_stateid *stateid, fmode_t type)
130 {
131         struct nfs_inode *nfsi = NFS_I(inode);
132         struct nfs_open_context *ctx;
133         struct nfs4_state_owner *sp;
134         struct nfs4_state *state;
135         unsigned int seq;
136         int err;
137
138 again:
139         spin_lock(&inode->i_lock);
140         list_for_each_entry(ctx, &nfsi->open_files, list) {
141                 state = ctx->state;
142                 if (state == NULL)
143                         continue;
144                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
145                         continue;
146                 if (!nfs4_valid_open_stateid(state))
147                         continue;
148                 if (!nfs4_stateid_match(&state->stateid, stateid))
149                         continue;
150                 get_nfs_open_context(ctx);
151                 spin_unlock(&inode->i_lock);
152                 sp = state->owner;
153                 /* Block nfs4_proc_unlck */
154                 mutex_lock(&sp->so_delegreturn_mutex);
155                 seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
156                 err = nfs4_open_delegation_recall(ctx, state, stateid, type);
157                 if (!err)
158                         err = nfs_delegation_claim_locks(ctx, state, stateid);
159                 if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
160                         err = -EAGAIN;
161                 mutex_unlock(&sp->so_delegreturn_mutex);
162                 put_nfs_open_context(ctx);
163                 if (err != 0)
164                         return err;
165                 goto again;
166         }
167         spin_unlock(&inode->i_lock);
168         return 0;
169 }
170
171 /**
172  * nfs_inode_reclaim_delegation - process a delegation reclaim request
173  * @inode: inode to process
174  * @cred: credential to use for request
175  * @type: delegation type
176  * @stateid: delegation stateid
177  * @pagemod_limit: write delegation "space_limit"
178  *
179  */
180 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred,
181                                   fmode_t type,
182                                   const nfs4_stateid *stateid,
183                                   unsigned long pagemod_limit)
184 {
185         struct nfs_delegation *delegation;
186         struct rpc_cred *oldcred = NULL;
187
188         rcu_read_lock();
189         delegation = rcu_dereference(NFS_I(inode)->delegation);
190         if (delegation != NULL) {
191                 spin_lock(&delegation->lock);
192                 if (delegation->inode != NULL) {
193                         nfs4_stateid_copy(&delegation->stateid, stateid);
194                         delegation->type = type;
195                         delegation->pagemod_limit = pagemod_limit;
196                         oldcred = delegation->cred;
197                         delegation->cred = get_rpccred(cred);
198                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
199                                   &delegation->flags);
200                         spin_unlock(&delegation->lock);
201                         rcu_read_unlock();
202                         put_rpccred(oldcred);
203                         trace_nfs4_reclaim_delegation(inode, type);
204                         return;
205                 }
206                 /* We appear to have raced with a delegation return. */
207                 spin_unlock(&delegation->lock);
208         }
209         rcu_read_unlock();
210         nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
211 }
212
213 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
214 {
215         int res = 0;
216
217         if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
218                 res = nfs4_proc_delegreturn(inode,
219                                 delegation->cred,
220                                 &delegation->stateid,
221                                 issync);
222         nfs_free_delegation(delegation);
223         return res;
224 }
225
226 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
227 {
228         struct inode *inode = NULL;
229
230         spin_lock(&delegation->lock);
231         if (delegation->inode != NULL)
232                 inode = igrab(delegation->inode);
233         spin_unlock(&delegation->lock);
234         return inode;
235 }
236
237 static struct nfs_delegation *
238 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
239 {
240         struct nfs_delegation *ret = NULL;
241         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
242
243         if (delegation == NULL)
244                 goto out;
245         spin_lock(&delegation->lock);
246         if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
247                 ret = delegation;
248         spin_unlock(&delegation->lock);
249 out:
250         return ret;
251 }
252
253 static struct nfs_delegation *
254 nfs_start_delegation_return(struct nfs_inode *nfsi)
255 {
256         struct nfs_delegation *delegation;
257
258         rcu_read_lock();
259         delegation = nfs_start_delegation_return_locked(nfsi);
260         rcu_read_unlock();
261         return delegation;
262 }
263
264 static void
265 nfs_abort_delegation_return(struct nfs_delegation *delegation,
266                 struct nfs_client *clp)
267 {
268
269         spin_lock(&delegation->lock);
270         clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
271         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
272         spin_unlock(&delegation->lock);
273         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
274 }
275
276 static struct nfs_delegation *
277 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
278                 struct nfs_delegation *delegation,
279                 struct nfs_client *clp)
280 {
281         struct nfs_delegation *deleg_cur =
282                 rcu_dereference_protected(nfsi->delegation,
283                                 lockdep_is_held(&clp->cl_lock));
284
285         if (deleg_cur == NULL || delegation != deleg_cur)
286                 return NULL;
287
288         spin_lock(&delegation->lock);
289         set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
290         list_del_rcu(&delegation->super_list);
291         delegation->inode = NULL;
292         rcu_assign_pointer(nfsi->delegation, NULL);
293         spin_unlock(&delegation->lock);
294         return delegation;
295 }
296
297 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
298                 struct nfs_delegation *delegation,
299                 struct nfs_server *server)
300 {
301         struct nfs_client *clp = server->nfs_client;
302
303         spin_lock(&clp->cl_lock);
304         delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
305         spin_unlock(&clp->cl_lock);
306         return delegation;
307 }
308
309 static struct nfs_delegation *
310 nfs_inode_detach_delegation(struct inode *inode)
311 {
312         struct nfs_inode *nfsi = NFS_I(inode);
313         struct nfs_server *server = NFS_SERVER(inode);
314         struct nfs_delegation *delegation;
315
316         delegation = nfs_start_delegation_return(nfsi);
317         if (delegation == NULL)
318                 return NULL;
319         return nfs_detach_delegation(nfsi, delegation, server);
320 }
321
322 static void
323 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
324                 const struct nfs_delegation *update)
325 {
326         if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
327                 delegation->stateid.seqid = update->stateid.seqid;
328                 smp_wmb();
329                 delegation->type = update->type;
330         }
331 }
332
333 /**
334  * nfs_inode_set_delegation - set up a delegation on an inode
335  * @inode: inode to which delegation applies
336  * @cred: cred to use for subsequent delegation processing
337  * @type: delegation type
338  * @stateid: delegation stateid
339  * @pagemod_limit: write delegation "space_limit"
340  *
341  * Returns zero on success, or a negative errno value.
342  */
343 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred,
344                                   fmode_t type,
345                                   const nfs4_stateid *stateid,
346                                   unsigned long pagemod_limit)
347 {
348         struct nfs_server *server = NFS_SERVER(inode);
349         struct nfs_client *clp = server->nfs_client;
350         struct nfs_inode *nfsi = NFS_I(inode);
351         struct nfs_delegation *delegation, *old_delegation;
352         struct nfs_delegation *freeme = NULL;
353         int status = 0;
354
355         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
356         if (delegation == NULL)
357                 return -ENOMEM;
358         nfs4_stateid_copy(&delegation->stateid, stateid);
359         delegation->type = type;
360         delegation->pagemod_limit = pagemod_limit;
361         delegation->change_attr = inode_peek_iversion_raw(inode);
362         delegation->cred = get_rpccred(cred);
363         delegation->inode = inode;
364         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
365         spin_lock_init(&delegation->lock);
366
367         spin_lock(&clp->cl_lock);
368         old_delegation = rcu_dereference_protected(nfsi->delegation,
369                                         lockdep_is_held(&clp->cl_lock));
370         if (old_delegation != NULL) {
371                 /* Is this an update of the existing delegation? */
372                 if (nfs4_stateid_match_other(&old_delegation->stateid,
373                                         &delegation->stateid)) {
374                         nfs_update_inplace_delegation(old_delegation,
375                                         delegation);
376                         goto out;
377                 }
378                 /*
379                  * Deal with broken servers that hand out two
380                  * delegations for the same file.
381                  * Allow for upgrades to a WRITE delegation, but
382                  * nothing else.
383                  */
384                 dfprintk(FILE, "%s: server %s handed out "
385                                 "a duplicate delegation!\n",
386                                 __func__, clp->cl_hostname);
387                 if (delegation->type == old_delegation->type ||
388                     !(delegation->type & FMODE_WRITE)) {
389                         freeme = delegation;
390                         delegation = NULL;
391                         goto out;
392                 }
393                 if (test_and_set_bit(NFS_DELEGATION_RETURNING,
394                                         &old_delegation->flags))
395                         goto out;
396                 freeme = nfs_detach_delegation_locked(nfsi,
397                                 old_delegation, clp);
398                 if (freeme == NULL)
399                         goto out;
400         }
401         list_add_tail_rcu(&delegation->super_list, &server->delegations);
402         rcu_assign_pointer(nfsi->delegation, delegation);
403         delegation = NULL;
404
405         trace_nfs4_set_delegation(inode, type);
406
407         spin_lock(&inode->i_lock);
408         if (NFS_I(inode)->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME))
409                 NFS_I(inode)->cache_validity |= NFS_INO_REVAL_FORCED;
410         spin_unlock(&inode->i_lock);
411 out:
412         spin_unlock(&clp->cl_lock);
413         if (delegation != NULL)
414                 nfs_free_delegation(delegation);
415         if (freeme != NULL)
416                 nfs_do_return_delegation(inode, freeme, 0);
417         return status;
418 }
419
420 /*
421  * Basic procedure for returning a delegation to the server
422  */
423 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
424 {
425         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
426         struct nfs_inode *nfsi = NFS_I(inode);
427         int err = 0;
428
429         if (delegation == NULL)
430                 return 0;
431         do {
432                 if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
433                         break;
434                 err = nfs_delegation_claim_opens(inode, &delegation->stateid,
435                                 delegation->type);
436                 if (!issync || err != -EAGAIN)
437                         break;
438                 /*
439                  * Guard against state recovery
440                  */
441                 err = nfs4_wait_clnt_recover(clp);
442         } while (err == 0);
443
444         if (err) {
445                 nfs_abort_delegation_return(delegation, clp);
446                 goto out;
447         }
448         if (!nfs_detach_delegation(nfsi, delegation, NFS_SERVER(inode)))
449                 goto out;
450
451         err = nfs_do_return_delegation(inode, delegation, issync);
452 out:
453         return err;
454 }
455
456 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
457 {
458         bool ret = false;
459
460         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
461                 goto out;
462         if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
463                 ret = true;
464         if (test_and_clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) && !ret) {
465                 struct inode *inode;
466
467                 spin_lock(&delegation->lock);
468                 inode = delegation->inode;
469                 if (inode && list_empty(&NFS_I(inode)->open_files))
470                         ret = true;
471                 spin_unlock(&delegation->lock);
472         }
473 out:
474         return ret;
475 }
476
477 /**
478  * nfs_client_return_marked_delegations - return previously marked delegations
479  * @clp: nfs_client to process
480  *
481  * Note that this function is designed to be called by the state
482  * manager thread. For this reason, it cannot flush the dirty data,
483  * since that could deadlock in case of a state recovery error.
484  *
485  * Returns zero on success, or a negative errno value.
486  */
487 int nfs_client_return_marked_delegations(struct nfs_client *clp)
488 {
489         struct nfs_delegation *delegation;
490         struct nfs_delegation *prev;
491         struct nfs_server *server;
492         struct inode *inode;
493         struct inode *place_holder = NULL;
494         struct nfs_delegation *place_holder_deleg = NULL;
495         int err = 0;
496
497 restart:
498         /*
499          * To avoid quadratic looping we hold a reference
500          * to an inode place_holder.  Each time we restart, we
501          * list nfs_servers from the server of that inode, and
502          * delegation in the server from the delegations of that
503          * inode.
504          * prev is an RCU-protected pointer to a delegation which
505          * wasn't marked for return and might be a good choice for
506          * the next place_holder.
507          */
508         rcu_read_lock();
509         prev = NULL;
510         if (place_holder)
511                 server = NFS_SERVER(place_holder);
512         else
513                 server = list_entry_rcu(clp->cl_superblocks.next,
514                                         struct nfs_server, client_link);
515         list_for_each_entry_from_rcu(server, &clp->cl_superblocks, client_link) {
516                 delegation = NULL;
517                 if (place_holder && server == NFS_SERVER(place_holder))
518                         delegation = rcu_dereference(NFS_I(place_holder)->delegation);
519                 if (!delegation || delegation != place_holder_deleg)
520                         delegation = list_entry_rcu(server->delegations.next,
521                                                     struct nfs_delegation, super_list);
522                 list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
523                         struct inode *to_put = NULL;
524
525                         if (!nfs_delegation_need_return(delegation)) {
526                                 prev = delegation;
527                                 continue;
528                         }
529                         if (!nfs_sb_active(server->super))
530                                 break; /* continue in outer loop */
531
532                         if (prev) {
533                                 struct inode *tmp;
534
535                                 tmp = nfs_delegation_grab_inode(prev);
536                                 if (tmp) {
537                                         to_put = place_holder;
538                                         place_holder = tmp;
539                                         place_holder_deleg = prev;
540                                 }
541                         }
542
543                         inode = nfs_delegation_grab_inode(delegation);
544                         if (inode == NULL) {
545                                 rcu_read_unlock();
546                                 if (to_put)
547                                         iput(to_put);
548                                 nfs_sb_deactive(server->super);
549                                 goto restart;
550                         }
551                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
552                         rcu_read_unlock();
553
554                         if (to_put)
555                                 iput(to_put);
556
557                         err = nfs_end_delegation_return(inode, delegation, 0);
558                         iput(inode);
559                         nfs_sb_deactive(server->super);
560                         cond_resched();
561                         if (!err)
562                                 goto restart;
563                         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
564                         if (place_holder)
565                                 iput(place_holder);
566                         return err;
567                 }
568         }
569         rcu_read_unlock();
570         if (place_holder)
571                 iput(place_holder);
572         return 0;
573 }
574
575 /**
576  * nfs_inode_return_delegation_noreclaim - return delegation, don't reclaim opens
577  * @inode: inode to process
578  *
579  * Does not protect against delegation reclaims, therefore really only safe
580  * to be called from nfs4_clear_inode().
581  */
582 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
583 {
584         struct nfs_delegation *delegation;
585
586         delegation = nfs_inode_detach_delegation(inode);
587         if (delegation != NULL)
588                 nfs_do_return_delegation(inode, delegation, 1);
589 }
590
591 /**
592  * nfs_inode_return_delegation - synchronously return a delegation
593  * @inode: inode to process
594  *
595  * This routine will always flush any dirty data to disk on the
596  * assumption that if we need to return the delegation, then
597  * we should stop caching.
598  *
599  * Returns zero on success, or a negative errno value.
600  */
601 int nfs4_inode_return_delegation(struct inode *inode)
602 {
603         struct nfs_inode *nfsi = NFS_I(inode);
604         struct nfs_delegation *delegation;
605         int err = 0;
606
607         nfs_wb_all(inode);
608         delegation = nfs_start_delegation_return(nfsi);
609         if (delegation != NULL)
610                 err = nfs_end_delegation_return(inode, delegation, 1);
611         return err;
612 }
613
614 /**
615  * nfs4_inode_make_writeable
616  * @inode: pointer to inode
617  *
618  * Make the inode writeable by returning the delegation if necessary
619  *
620  * Returns zero on success, or a negative errno value.
621  */
622 int nfs4_inode_make_writeable(struct inode *inode)
623 {
624         if (!nfs4_has_session(NFS_SERVER(inode)->nfs_client) ||
625             !nfs4_check_delegation(inode, FMODE_WRITE))
626                 return nfs4_inode_return_delegation(inode);
627         return 0;
628 }
629
630 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
631                 struct nfs_delegation *delegation)
632 {
633         set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
634         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
635 }
636
637 static void nfs_mark_return_delegation(struct nfs_server *server,
638                 struct nfs_delegation *delegation)
639 {
640         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
641         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
642 }
643
644 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
645 {
646         struct nfs_delegation *delegation;
647         bool ret = false;
648
649         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
650                 nfs_mark_return_delegation(server, delegation);
651                 ret = true;
652         }
653         return ret;
654 }
655
656 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
657 {
658         struct nfs_server *server;
659
660         rcu_read_lock();
661         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
662                 nfs_server_mark_return_all_delegations(server);
663         rcu_read_unlock();
664 }
665
666 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
667 {
668         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
669                 nfs4_schedule_state_manager(clp);
670 }
671
672 /**
673  * nfs_expire_all_delegations
674  * @clp: client to process
675  *
676  */
677 void nfs_expire_all_delegations(struct nfs_client *clp)
678 {
679         nfs_client_mark_return_all_delegations(clp);
680         nfs_delegation_run_state_manager(clp);
681 }
682
683 /**
684  * nfs_super_return_all_delegations - return delegations for one superblock
685  * @sb: sb to process
686  *
687  */
688 void nfs_server_return_all_delegations(struct nfs_server *server)
689 {
690         struct nfs_client *clp = server->nfs_client;
691         bool need_wait;
692
693         if (clp == NULL)
694                 return;
695
696         rcu_read_lock();
697         need_wait = nfs_server_mark_return_all_delegations(server);
698         rcu_read_unlock();
699
700         if (need_wait) {
701                 nfs4_schedule_state_manager(clp);
702                 nfs4_wait_clnt_recover(clp);
703         }
704 }
705
706 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
707                                                  fmode_t flags)
708 {
709         struct nfs_delegation *delegation;
710
711         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
712                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
713                         continue;
714                 if (delegation->type & flags)
715                         nfs_mark_return_if_closed_delegation(server, delegation);
716         }
717 }
718
719 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
720                                                         fmode_t flags)
721 {
722         struct nfs_server *server;
723
724         rcu_read_lock();
725         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
726                 nfs_mark_return_unused_delegation_types(server, flags);
727         rcu_read_unlock();
728 }
729
730 static void nfs_mark_delegation_revoked(struct nfs_server *server,
731                 struct nfs_delegation *delegation)
732 {
733         set_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
734         delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
735         nfs_mark_return_delegation(server, delegation);
736 }
737
738 static bool nfs_revoke_delegation(struct inode *inode,
739                 const nfs4_stateid *stateid)
740 {
741         struct nfs_delegation *delegation;
742         nfs4_stateid tmp;
743         bool ret = false;
744
745         rcu_read_lock();
746         delegation = rcu_dereference(NFS_I(inode)->delegation);
747         if (delegation == NULL)
748                 goto out;
749         if (stateid == NULL) {
750                 nfs4_stateid_copy(&tmp, &delegation->stateid);
751                 stateid = &tmp;
752         } else if (!nfs4_stateid_match(stateid, &delegation->stateid))
753                 goto out;
754         nfs_mark_delegation_revoked(NFS_SERVER(inode), delegation);
755         ret = true;
756 out:
757         rcu_read_unlock();
758         if (ret)
759                 nfs_inode_find_state_and_recover(inode, stateid);
760         return ret;
761 }
762
763 void nfs_remove_bad_delegation(struct inode *inode,
764                 const nfs4_stateid *stateid)
765 {
766         struct nfs_delegation *delegation;
767
768         if (!nfs_revoke_delegation(inode, stateid))
769                 return;
770         delegation = nfs_inode_detach_delegation(inode);
771         if (delegation)
772                 nfs_free_delegation(delegation);
773 }
774 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
775
776 /**
777  * nfs_expire_unused_delegation_types
778  * @clp: client to process
779  * @flags: delegation types to expire
780  *
781  */
782 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
783 {
784         nfs_client_mark_return_unused_delegation_types(clp, flags);
785         nfs_delegation_run_state_manager(clp);
786 }
787
788 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
789 {
790         struct nfs_delegation *delegation;
791
792         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
793                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
794                         continue;
795                 nfs_mark_return_if_closed_delegation(server, delegation);
796         }
797 }
798
799 /**
800  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
801  * @clp: nfs_client to process
802  *
803  */
804 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
805 {
806         struct nfs_server *server;
807
808         rcu_read_lock();
809         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
810                 nfs_mark_return_unreferenced_delegations(server);
811         rcu_read_unlock();
812
813         nfs_delegation_run_state_manager(clp);
814 }
815
816 /**
817  * nfs_async_inode_return_delegation - asynchronously return a delegation
818  * @inode: inode to process
819  * @stateid: state ID information
820  *
821  * Returns zero on success, or a negative errno value.
822  */
823 int nfs_async_inode_return_delegation(struct inode *inode,
824                                       const nfs4_stateid *stateid)
825 {
826         struct nfs_server *server = NFS_SERVER(inode);
827         struct nfs_client *clp = server->nfs_client;
828         struct nfs_delegation *delegation;
829
830         rcu_read_lock();
831         delegation = rcu_dereference(NFS_I(inode)->delegation);
832         if (delegation == NULL)
833                 goto out_enoent;
834         if (stateid != NULL &&
835             !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
836                 goto out_enoent;
837         nfs_mark_return_delegation(server, delegation);
838         rcu_read_unlock();
839
840         nfs_delegation_run_state_manager(clp);
841         return 0;
842 out_enoent:
843         rcu_read_unlock();
844         return -ENOENT;
845 }
846
847 static struct inode *
848 nfs_delegation_find_inode_server(struct nfs_server *server,
849                                  const struct nfs_fh *fhandle)
850 {
851         struct nfs_delegation *delegation;
852         struct inode *res = NULL;
853
854         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
855                 spin_lock(&delegation->lock);
856                 if (delegation->inode != NULL &&
857                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
858                         res = igrab(delegation->inode);
859                         spin_unlock(&delegation->lock);
860                         if (res != NULL)
861                                 return res;
862                         return ERR_PTR(-EAGAIN);
863                 }
864                 spin_unlock(&delegation->lock);
865         }
866         return ERR_PTR(-ENOENT);
867 }
868
869 /**
870  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
871  * @clp: client state handle
872  * @fhandle: filehandle from a delegation recall
873  *
874  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
875  * cannot be found.
876  */
877 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
878                                         const struct nfs_fh *fhandle)
879 {
880         struct nfs_server *server;
881         struct inode *res;
882
883         rcu_read_lock();
884         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
885                 res = nfs_delegation_find_inode_server(server, fhandle);
886                 if (res != ERR_PTR(-ENOENT))
887                         return res;
888         }
889         rcu_read_unlock();
890         return ERR_PTR(-ENOENT);
891 }
892
893 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
894 {
895         struct nfs_delegation *delegation;
896
897         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
898                 /*
899                  * If the delegation may have been admin revoked, then we
900                  * cannot reclaim it.
901                  */
902                 if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
903                         continue;
904                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
905         }
906 }
907
908 /**
909  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
910  * @clp: nfs_client to process
911  *
912  */
913 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
914 {
915         struct nfs_server *server;
916
917         rcu_read_lock();
918         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
919                 nfs_delegation_mark_reclaim_server(server);
920         rcu_read_unlock();
921 }
922
923 /**
924  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
925  * @clp: nfs_client to process
926  *
927  */
928 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
929 {
930         struct nfs_delegation *delegation;
931         struct nfs_server *server;
932         struct inode *inode;
933
934 restart:
935         rcu_read_lock();
936         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
937                 list_for_each_entry_rcu(delegation, &server->delegations,
938                                                                 super_list) {
939                         if (test_bit(NFS_DELEGATION_RETURNING,
940                                                 &delegation->flags))
941                                 continue;
942                         if (test_bit(NFS_DELEGATION_NEED_RECLAIM,
943                                                 &delegation->flags) == 0)
944                                 continue;
945                         if (!nfs_sb_active(server->super))
946                                 break; /* continue in outer loop */
947                         inode = nfs_delegation_grab_inode(delegation);
948                         if (inode == NULL) {
949                                 rcu_read_unlock();
950                                 nfs_sb_deactive(server->super);
951                                 goto restart;
952                         }
953                         delegation = nfs_start_delegation_return_locked(NFS_I(inode));
954                         rcu_read_unlock();
955                         if (delegation != NULL) {
956                                 delegation = nfs_detach_delegation(NFS_I(inode),
957                                         delegation, server);
958                                 if (delegation != NULL)
959                                         nfs_free_delegation(delegation);
960                         }
961                         iput(inode);
962                         nfs_sb_deactive(server->super);
963                         cond_resched();
964                         goto restart;
965                 }
966         }
967         rcu_read_unlock();
968 }
969
970 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
971 {
972         return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
973                                 BIT(NFS4CLNT_LEASE_EXPIRED) |
974                                 BIT(NFS4CLNT_SESSION_RESET))) != 0;
975 }
976
977 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
978             struct nfs_delegation *delegation)
979 {
980         if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
981                 return;
982         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
983         set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
984         set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
985 }
986
987 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
988                 struct inode *inode)
989 {
990         struct nfs_delegation *delegation;
991
992         rcu_read_lock();
993         delegation = rcu_dereference(NFS_I(inode)->delegation);
994         if (delegation)
995                 nfs_mark_test_expired_delegation(server, delegation);
996         rcu_read_unlock();
997
998 }
999
1000 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1001 {
1002         struct nfs_delegation *delegation;
1003
1004         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1005                 nfs_mark_test_expired_delegation(server, delegation);
1006 }
1007
1008 /**
1009  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1010  * @clp: nfs_client to process
1011  *
1012  * Iterates through all the delegations associated with this server and
1013  * marks them as needing to be checked for validity.
1014  */
1015 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1016 {
1017         struct nfs_server *server;
1018
1019         rcu_read_lock();
1020         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1021                 nfs_delegation_mark_test_expired_server(server);
1022         rcu_read_unlock();
1023 }
1024
1025 /**
1026  * nfs_reap_expired_delegations - reap expired delegations
1027  * @clp: nfs_client to process
1028  *
1029  * Iterates through all the delegations associated with this server and
1030  * checks if they have may have been revoked. This function is usually
1031  * expected to be called in cases where the server may have lost its
1032  * lease.
1033  */
1034 void nfs_reap_expired_delegations(struct nfs_client *clp)
1035 {
1036         const struct nfs4_minor_version_ops *ops = clp->cl_mvops;
1037         struct nfs_delegation *delegation;
1038         struct nfs_server *server;
1039         struct inode *inode;
1040         struct rpc_cred *cred;
1041         nfs4_stateid stateid;
1042
1043 restart:
1044         rcu_read_lock();
1045         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1046                 list_for_each_entry_rcu(delegation, &server->delegations,
1047                                                                 super_list) {
1048                         if (test_bit(NFS_DELEGATION_RETURNING,
1049                                                 &delegation->flags))
1050                                 continue;
1051                         if (test_bit(NFS_DELEGATION_TEST_EXPIRED,
1052                                                 &delegation->flags) == 0)
1053                                 continue;
1054                         if (!nfs_sb_active(server->super))
1055                                 break; /* continue in outer loop */
1056                         inode = nfs_delegation_grab_inode(delegation);
1057                         if (inode == NULL) {
1058                                 rcu_read_unlock();
1059                                 nfs_sb_deactive(server->super);
1060                                 goto restart;
1061                         }
1062                         cred = get_rpccred_rcu(delegation->cred);
1063                         nfs4_stateid_copy(&stateid, &delegation->stateid);
1064                         clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1065                         rcu_read_unlock();
1066                         if (cred != NULL &&
1067                             ops->test_and_free_expired(server, &stateid, cred) < 0) {
1068                                 nfs_revoke_delegation(inode, &stateid);
1069                                 nfs_inode_find_state_and_recover(inode, &stateid);
1070                         }
1071                         put_rpccred(cred);
1072                         if (nfs4_server_rebooted(clp)) {
1073                                 nfs_inode_mark_test_expired_delegation(server,inode);
1074                                 iput(inode);
1075                                 nfs_sb_deactive(server->super);
1076                                 return;
1077                         }
1078                         iput(inode);
1079                         nfs_sb_deactive(server->super);
1080                         cond_resched();
1081                         goto restart;
1082                 }
1083         }
1084         rcu_read_unlock();
1085 }
1086
1087 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1088                 const nfs4_stateid *stateid)
1089 {
1090         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1091         struct nfs_delegation *delegation;
1092         bool found = false;
1093
1094         rcu_read_lock();
1095         delegation = rcu_dereference(NFS_I(inode)->delegation);
1096         if (delegation &&
1097             nfs4_stateid_match_other(&delegation->stateid, stateid)) {
1098                 nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1099                 found = true;
1100         }
1101         rcu_read_unlock();
1102         if (found)
1103                 nfs4_schedule_state_manager(clp);
1104 }
1105
1106 /**
1107  * nfs_delegations_present - check for existence of delegations
1108  * @clp: client state handle
1109  *
1110  * Returns one if there are any nfs_delegation structures attached
1111  * to this nfs_client.
1112  */
1113 int nfs_delegations_present(struct nfs_client *clp)
1114 {
1115         struct nfs_server *server;
1116         int ret = 0;
1117
1118         rcu_read_lock();
1119         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1120                 if (!list_empty(&server->delegations)) {
1121                         ret = 1;
1122                         break;
1123                 }
1124         rcu_read_unlock();
1125         return ret;
1126 }
1127
1128 /**
1129  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1130  * @dst: stateid to refresh
1131  * @inode: inode to check
1132  *
1133  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1134  * that matches our delegation stateid. Otherwise "false" is returned.
1135  */
1136 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1137 {
1138         struct nfs_delegation *delegation;
1139         bool ret = false;
1140         if (!inode)
1141                 goto out;
1142
1143         rcu_read_lock();
1144         delegation = rcu_dereference(NFS_I(inode)->delegation);
1145         if (delegation != NULL &&
1146             nfs4_stateid_match_other(dst, &delegation->stateid)) {
1147                 dst->seqid = delegation->stateid.seqid;
1148                 return ret;
1149         }
1150         rcu_read_unlock();
1151 out:
1152         return ret;
1153 }
1154
1155 /**
1156  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1157  * @inode: inode to check
1158  * @flags: delegation type requirement
1159  * @dst: stateid data structure to fill in
1160  * @cred: optional argument to retrieve credential
1161  *
1162  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1163  * otherwise "false" is returned.
1164  */
1165 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1166                 nfs4_stateid *dst, struct rpc_cred **cred)
1167 {
1168         struct nfs_inode *nfsi = NFS_I(inode);
1169         struct nfs_delegation *delegation;
1170         bool ret;
1171
1172         flags &= FMODE_READ|FMODE_WRITE;
1173         rcu_read_lock();
1174         delegation = rcu_dereference(nfsi->delegation);
1175         ret = nfs4_is_valid_delegation(delegation, flags);
1176         if (ret) {
1177                 nfs4_stateid_copy(dst, &delegation->stateid);
1178                 nfs_mark_delegation_referenced(delegation);
1179                 if (cred)
1180                         *cred = get_rpccred(delegation->cred);
1181         }
1182         rcu_read_unlock();
1183         return ret;
1184 }
1185
1186 /**
1187  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1188  * @inode: inode to check
1189  *
1190  * This function checks the number of outstanding writes to the file
1191  * against the delegation 'space_limit' field to see if
1192  * the spec requires us to flush the file on close.
1193  */
1194 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1195 {
1196         struct nfs_inode *nfsi = NFS_I(inode);
1197         struct nfs_delegation *delegation;
1198         bool ret = true;
1199
1200         rcu_read_lock();
1201         delegation = rcu_dereference(nfsi->delegation);
1202         if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1203                 goto out;
1204         if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1205                 ret = false;
1206 out:
1207         rcu_read_unlock();
1208         return ret;
1209 }