NFSv4: Fix the underestimation of delegation XDR space reservation
[pandora-kernel.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
16 #include <linux/nfs4.h>
17 #include <linux/nfs_fs.h>
18 #include <linux/nfs_xdr.h>
19
20 #include "nfs4_fs.h"
21 #include "delegation.h"
22 #include "internal.h"
23
24 static void nfs_free_delegation(struct nfs_delegation *delegation)
25 {
26         if (delegation->cred) {
27                 put_rpccred(delegation->cred);
28                 delegation->cred = NULL;
29         }
30         kfree_rcu(delegation, rcu);
31 }
32
33 /**
34  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
35  * @delegation: delegation to process
36  *
37  */
38 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
39 {
40         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
41 }
42
43 /**
44  * nfs_have_delegation - check if inode has a delegation
45  * @inode: inode to check
46  * @flags: delegation types to check for
47  *
48  * Returns one if inode has the indicated delegation, otherwise zero.
49  */
50 int nfs_have_delegation(struct inode *inode, fmode_t flags)
51 {
52         struct nfs_delegation *delegation;
53         int ret = 0;
54
55         flags &= FMODE_READ|FMODE_WRITE;
56         rcu_read_lock();
57         delegation = rcu_dereference(NFS_I(inode)->delegation);
58         if (delegation != NULL && (delegation->type & flags) == flags) {
59                 nfs_mark_delegation_referenced(delegation);
60                 ret = 1;
61         }
62         rcu_read_unlock();
63         return ret;
64 }
65
66 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
67 {
68         struct inode *inode = state->inode;
69         struct file_lock *fl;
70         int status = 0;
71
72         if (inode->i_flock == NULL)
73                 goto out;
74
75         /* Protect inode->i_flock using the file locks lock */
76         lock_flocks();
77         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
78                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
79                         continue;
80                 if (nfs_file_open_context(fl->fl_file) != ctx)
81                         continue;
82                 unlock_flocks();
83                 status = nfs4_lock_delegation_recall(state, fl);
84                 if (status < 0)
85                         goto out;
86                 lock_flocks();
87         }
88         unlock_flocks();
89 out:
90         return status;
91 }
92
93 static int nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
94 {
95         struct nfs_inode *nfsi = NFS_I(inode);
96         struct nfs_open_context *ctx;
97         struct nfs4_state *state;
98         int err;
99
100 again:
101         spin_lock(&inode->i_lock);
102         list_for_each_entry(ctx, &nfsi->open_files, list) {
103                 state = ctx->state;
104                 if (state == NULL)
105                         continue;
106                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
107                         continue;
108                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
109                         continue;
110                 get_nfs_open_context(ctx);
111                 spin_unlock(&inode->i_lock);
112                 err = nfs4_open_delegation_recall(ctx, state, stateid);
113                 if (err >= 0)
114                         err = nfs_delegation_claim_locks(ctx, state);
115                 put_nfs_open_context(ctx);
116                 if (err != 0)
117                         return err;
118                 goto again;
119         }
120         spin_unlock(&inode->i_lock);
121         return 0;
122 }
123
124 /**
125  * nfs_inode_reclaim_delegation - process a delegation reclaim request
126  * @inode: inode to process
127  * @cred: credential to use for request
128  * @res: new delegation state from server
129  *
130  */
131 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred,
132                                   struct nfs_openres *res)
133 {
134         struct nfs_delegation *delegation;
135         struct rpc_cred *oldcred = NULL;
136
137         rcu_read_lock();
138         delegation = rcu_dereference(NFS_I(inode)->delegation);
139         if (delegation != NULL) {
140                 spin_lock(&delegation->lock);
141                 if (delegation->inode != NULL) {
142                         memcpy(delegation->stateid.data, res->delegation.data,
143                                sizeof(delegation->stateid.data));
144                         delegation->type = res->delegation_type;
145                         delegation->maxsize = res->maxsize;
146                         oldcred = delegation->cred;
147                         delegation->cred = get_rpccred(cred);
148                         clear_bit(NFS_DELEGATION_NEED_RECLAIM,
149                                   &delegation->flags);
150                         NFS_I(inode)->delegation_state = delegation->type;
151                         spin_unlock(&delegation->lock);
152                         rcu_read_unlock();
153                         put_rpccred(oldcred);
154                 } else {
155                         /* We appear to have raced with a delegation return. */
156                         spin_unlock(&delegation->lock);
157                         rcu_read_unlock();
158                         nfs_inode_set_delegation(inode, cred, res);
159                 }
160         } else {
161                 rcu_read_unlock();
162         }
163 }
164
165 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
166 {
167         int res = 0;
168
169         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
170         nfs_free_delegation(delegation);
171         return res;
172 }
173
174 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
175 {
176         struct inode *inode = NULL;
177
178         spin_lock(&delegation->lock);
179         if (delegation->inode != NULL)
180                 inode = igrab(delegation->inode);
181         spin_unlock(&delegation->lock);
182         return inode;
183 }
184
185 static struct nfs_delegation *
186 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
187                              struct nfs_server *server)
188 {
189         struct nfs_delegation *delegation =
190                 rcu_dereference_protected(nfsi->delegation,
191                                 lockdep_is_held(&server->nfs_client->cl_lock));
192
193         if (delegation == NULL)
194                 goto nomatch;
195
196         spin_lock(&delegation->lock);
197         list_del_rcu(&delegation->super_list);
198         delegation->inode = NULL;
199         nfsi->delegation_state = 0;
200         rcu_assign_pointer(nfsi->delegation, NULL);
201         spin_unlock(&delegation->lock);
202         return delegation;
203 nomatch:
204         return NULL;
205 }
206
207 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
208                                                     struct nfs_server *server)
209 {
210         struct nfs_client *clp = server->nfs_client;
211         struct nfs_delegation *delegation;
212
213         spin_lock(&clp->cl_lock);
214         delegation = nfs_detach_delegation_locked(nfsi, server);
215         spin_unlock(&clp->cl_lock);
216         return delegation;
217 }
218
219 /**
220  * nfs_inode_set_delegation - set up a delegation on an inode
221  * @inode: inode to which delegation applies
222  * @cred: cred to use for subsequent delegation processing
223  * @res: new delegation state from server
224  *
225  * Returns zero on success, or a negative errno value.
226  */
227 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
228 {
229         struct nfs_server *server = NFS_SERVER(inode);
230         struct nfs_client *clp = server->nfs_client;
231         struct nfs_inode *nfsi = NFS_I(inode);
232         struct nfs_delegation *delegation, *old_delegation;
233         struct nfs_delegation *freeme = NULL;
234         int status = 0;
235
236         delegation = kmalloc(sizeof(*delegation), GFP_NOFS);
237         if (delegation == NULL)
238                 return -ENOMEM;
239         memcpy(delegation->stateid.data, res->delegation.data,
240                         sizeof(delegation->stateid.data));
241         delegation->type = res->delegation_type;
242         delegation->maxsize = res->maxsize;
243         delegation->change_attr = inode->i_version;
244         delegation->cred = get_rpccred(cred);
245         delegation->inode = inode;
246         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
247         spin_lock_init(&delegation->lock);
248
249         spin_lock(&clp->cl_lock);
250         old_delegation = rcu_dereference_protected(nfsi->delegation,
251                                         lockdep_is_held(&clp->cl_lock));
252         if (old_delegation != NULL) {
253                 if (memcmp(&delegation->stateid, &old_delegation->stateid,
254                                         sizeof(old_delegation->stateid)) == 0 &&
255                                 delegation->type == old_delegation->type) {
256                         goto out;
257                 }
258                 /*
259                  * Deal with broken servers that hand out two
260                  * delegations for the same file.
261                  */
262                 dfprintk(FILE, "%s: server %s handed out "
263                                 "a duplicate delegation!\n",
264                                 __func__, clp->cl_hostname);
265                 if (delegation->type <= old_delegation->type) {
266                         freeme = delegation;
267                         delegation = NULL;
268                         goto out;
269                 }
270                 freeme = nfs_detach_delegation_locked(nfsi, server);
271         }
272         list_add_rcu(&delegation->super_list, &server->delegations);
273         nfsi->delegation_state = delegation->type;
274         rcu_assign_pointer(nfsi->delegation, delegation);
275         delegation = NULL;
276
277         /* Ensure we revalidate the attributes and page cache! */
278         spin_lock(&inode->i_lock);
279         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
280         spin_unlock(&inode->i_lock);
281
282 out:
283         spin_unlock(&clp->cl_lock);
284         if (delegation != NULL)
285                 nfs_free_delegation(delegation);
286         if (freeme != NULL)
287                 nfs_do_return_delegation(inode, freeme, 0);
288         return status;
289 }
290
291 /*
292  * Basic procedure for returning a delegation to the server
293  */
294 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
295 {
296         struct nfs_inode *nfsi = NFS_I(inode);
297         int err;
298
299         /*
300          * Guard against new delegated open/lock/unlock calls and against
301          * state recovery
302          */
303         down_write(&nfsi->rwsem);
304         err = nfs_delegation_claim_opens(inode, &delegation->stateid);
305         up_write(&nfsi->rwsem);
306         if (err)
307                 goto out;
308
309         err = nfs_do_return_delegation(inode, delegation, issync);
310 out:
311         return err;
312 }
313
314 /**
315  * nfs_client_return_marked_delegations - return previously marked delegations
316  * @clp: nfs_client to process
317  *
318  * Returns zero on success, or a negative errno value.
319  */
320 int nfs_client_return_marked_delegations(struct nfs_client *clp)
321 {
322         struct nfs_delegation *delegation;
323         struct nfs_server *server;
324         struct inode *inode;
325         int err = 0;
326
327 restart:
328         rcu_read_lock();
329         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
330                 list_for_each_entry_rcu(delegation, &server->delegations,
331                                                                 super_list) {
332                         if (!test_and_clear_bit(NFS_DELEGATION_RETURN,
333                                                         &delegation->flags))
334                                 continue;
335                         inode = nfs_delegation_grab_inode(delegation);
336                         if (inode == NULL)
337                                 continue;
338                         delegation = nfs_detach_delegation(NFS_I(inode),
339                                                                 server);
340                         rcu_read_unlock();
341
342                         if (delegation != NULL) {
343                                 filemap_flush(inode->i_mapping);
344                                 err = __nfs_inode_return_delegation(inode,
345                                                                 delegation, 0);
346                         }
347                         iput(inode);
348                         if (!err)
349                                 goto restart;
350                         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
351                         return err;
352                 }
353         }
354         rcu_read_unlock();
355         return 0;
356 }
357
358 /**
359  * nfs_inode_return_delegation_noreclaim - return delegation, don't reclaim opens
360  * @inode: inode to process
361  *
362  * Does not protect against delegation reclaims, therefore really only safe
363  * to be called from nfs4_clear_inode().
364  */
365 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
366 {
367         struct nfs_server *server = NFS_SERVER(inode);
368         struct nfs_inode *nfsi = NFS_I(inode);
369         struct nfs_delegation *delegation;
370
371         if (rcu_access_pointer(nfsi->delegation) != NULL) {
372                 delegation = nfs_detach_delegation(nfsi, server);
373                 if (delegation != NULL)
374                         nfs_do_return_delegation(inode, delegation, 0);
375         }
376 }
377
378 /**
379  * nfs_inode_return_delegation - synchronously return a delegation
380  * @inode: inode to process
381  *
382  * Returns zero on success, or a negative errno value.
383  */
384 int nfs_inode_return_delegation(struct inode *inode)
385 {
386         struct nfs_server *server = NFS_SERVER(inode);
387         struct nfs_inode *nfsi = NFS_I(inode);
388         struct nfs_delegation *delegation;
389         int err = 0;
390
391         if (rcu_access_pointer(nfsi->delegation) != NULL) {
392                 delegation = nfs_detach_delegation(nfsi, server);
393                 if (delegation != NULL) {
394                         nfs_wb_all(inode);
395                         err = __nfs_inode_return_delegation(inode, delegation, 1);
396                 }
397         }
398         return err;
399 }
400
401 static void nfs_mark_return_delegation(struct nfs_server *server,
402                 struct nfs_delegation *delegation)
403 {
404         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
405         set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
406 }
407
408 /**
409  * nfs_super_return_all_delegations - return delegations for one superblock
410  * @sb: sb to process
411  *
412  */
413 void nfs_super_return_all_delegations(struct super_block *sb)
414 {
415         struct nfs_server *server = NFS_SB(sb);
416         struct nfs_client *clp = server->nfs_client;
417         struct nfs_delegation *delegation;
418
419         if (clp == NULL)
420                 return;
421
422         rcu_read_lock();
423         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
424                 spin_lock(&delegation->lock);
425                 set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
426                 spin_unlock(&delegation->lock);
427         }
428         rcu_read_unlock();
429
430         if (nfs_client_return_marked_delegations(clp) != 0)
431                 nfs4_schedule_state_manager(clp);
432 }
433
434 static void nfs_mark_return_all_delegation_types(struct nfs_server *server,
435                                                  fmode_t flags)
436 {
437         struct nfs_delegation *delegation;
438
439         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
440                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
441                         continue;
442                 if (delegation->type & flags)
443                         nfs_mark_return_delegation(server, delegation);
444         }
445 }
446
447 static void nfs_client_mark_return_all_delegation_types(struct nfs_client *clp,
448                                                         fmode_t flags)
449 {
450         struct nfs_server *server;
451
452         rcu_read_lock();
453         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
454                 nfs_mark_return_all_delegation_types(server, flags);
455         rcu_read_unlock();
456 }
457
458 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
459 {
460         nfs_client_mark_return_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
461 }
462
463 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
464 {
465         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
466                 nfs4_schedule_state_manager(clp);
467 }
468
469 void nfs_remove_bad_delegation(struct inode *inode)
470 {
471         struct nfs_delegation *delegation;
472
473         delegation = nfs_detach_delegation(NFS_I(inode), NFS_SERVER(inode));
474         if (delegation) {
475                 nfs_inode_find_state_and_recover(inode, &delegation->stateid);
476                 nfs_free_delegation(delegation);
477         }
478 }
479
480 /**
481  * nfs_expire_all_delegation_types
482  * @clp: client to process
483  * @flags: delegation types to expire
484  *
485  */
486 void nfs_expire_all_delegation_types(struct nfs_client *clp, fmode_t flags)
487 {
488         nfs_client_mark_return_all_delegation_types(clp, flags);
489         nfs_delegation_run_state_manager(clp);
490 }
491
492 /**
493  * nfs_expire_all_delegations
494  * @clp: client to process
495  *
496  */
497 void nfs_expire_all_delegations(struct nfs_client *clp)
498 {
499         nfs_expire_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
500 }
501
502 /**
503  * nfs_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
504  * @clp: client to process
505  *
506  */
507 void nfs_handle_cb_pathdown(struct nfs_client *clp)
508 {
509         if (clp == NULL)
510                 return;
511         nfs_client_mark_return_all_delegations(clp);
512 }
513
514 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
515 {
516         struct nfs_delegation *delegation;
517
518         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
519                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
520                         continue;
521                 nfs_mark_return_delegation(server, delegation);
522         }
523 }
524
525 /**
526  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
527  * @clp: nfs_client to process
528  *
529  */
530 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
531 {
532         struct nfs_server *server;
533
534         rcu_read_lock();
535         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
536                 nfs_mark_return_unreferenced_delegations(server);
537         rcu_read_unlock();
538
539         nfs_delegation_run_state_manager(clp);
540 }
541
542 /**
543  * nfs_async_inode_return_delegation - asynchronously return a delegation
544  * @inode: inode to process
545  * @stateid: state ID information from CB_RECALL arguments
546  *
547  * Returns zero on success, or a negative errno value.
548  */
549 int nfs_async_inode_return_delegation(struct inode *inode,
550                                       const nfs4_stateid *stateid)
551 {
552         struct nfs_server *server = NFS_SERVER(inode);
553         struct nfs_client *clp = server->nfs_client;
554         struct nfs_delegation *delegation;
555
556         rcu_read_lock();
557         delegation = rcu_dereference(NFS_I(inode)->delegation);
558
559         if (!clp->cl_mvops->validate_stateid(delegation, stateid)) {
560                 rcu_read_unlock();
561                 return -ENOENT;
562         }
563         nfs_mark_return_delegation(server, delegation);
564         rcu_read_unlock();
565
566         nfs_delegation_run_state_manager(clp);
567         return 0;
568 }
569
570 static struct inode *
571 nfs_delegation_find_inode_server(struct nfs_server *server,
572                                  const struct nfs_fh *fhandle)
573 {
574         struct nfs_delegation *delegation;
575         struct inode *res = NULL;
576
577         list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
578                 spin_lock(&delegation->lock);
579                 if (delegation->inode != NULL &&
580                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
581                         res = igrab(delegation->inode);
582                 }
583                 spin_unlock(&delegation->lock);
584                 if (res != NULL)
585                         break;
586         }
587         return res;
588 }
589
590 /**
591  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
592  * @clp: client state handle
593  * @fhandle: filehandle from a delegation recall
594  *
595  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
596  * cannot be found.
597  */
598 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
599                                         const struct nfs_fh *fhandle)
600 {
601         struct nfs_server *server;
602         struct inode *res = NULL;
603
604         rcu_read_lock();
605         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
606                 res = nfs_delegation_find_inode_server(server, fhandle);
607                 if (res != NULL)
608                         break;
609         }
610         rcu_read_unlock();
611         return res;
612 }
613
614 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
615 {
616         struct nfs_delegation *delegation;
617
618         list_for_each_entry_rcu(delegation, &server->delegations, super_list)
619                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
620 }
621
622 /**
623  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
624  * @clp: nfs_client to process
625  *
626  */
627 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
628 {
629         struct nfs_server *server;
630
631         rcu_read_lock();
632         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
633                 nfs_delegation_mark_reclaim_server(server);
634         rcu_read_unlock();
635 }
636
637 /**
638  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
639  * @clp: nfs_client to process
640  *
641  */
642 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
643 {
644         struct nfs_delegation *delegation;
645         struct nfs_server *server;
646         struct inode *inode;
647
648 restart:
649         rcu_read_lock();
650         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
651                 list_for_each_entry_rcu(delegation, &server->delegations,
652                                                                 super_list) {
653                         if (test_bit(NFS_DELEGATION_NEED_RECLAIM,
654                                                 &delegation->flags) == 0)
655                                 continue;
656                         inode = nfs_delegation_grab_inode(delegation);
657                         if (inode == NULL)
658                                 continue;
659                         delegation = nfs_detach_delegation(NFS_I(inode),
660                                                                 server);
661                         rcu_read_unlock();
662
663                         if (delegation != NULL)
664                                 nfs_free_delegation(delegation);
665                         iput(inode);
666                         goto restart;
667                 }
668         }
669         rcu_read_unlock();
670 }
671
672 /**
673  * nfs_delegations_present - check for existence of delegations
674  * @clp: client state handle
675  *
676  * Returns one if there are any nfs_delegation structures attached
677  * to this nfs_client.
678  */
679 int nfs_delegations_present(struct nfs_client *clp)
680 {
681         struct nfs_server *server;
682         int ret = 0;
683
684         rcu_read_lock();
685         list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
686                 if (!list_empty(&server->delegations)) {
687                         ret = 1;
688                         break;
689                 }
690         rcu_read_unlock();
691         return ret;
692 }
693
694 /**
695  * nfs4_copy_delegation_stateid - Copy inode's state ID information
696  * @dst: stateid data structure to fill in
697  * @inode: inode to check
698  *
699  * Returns one and fills in "dst->data" * if inode had a delegation,
700  * otherwise zero is returned.
701  */
702 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
703 {
704         struct nfs_inode *nfsi = NFS_I(inode);
705         struct nfs_delegation *delegation;
706         int ret = 0;
707
708         rcu_read_lock();
709         delegation = rcu_dereference(nfsi->delegation);
710         if (delegation != NULL) {
711                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
712                 ret = 1;
713         }
714         rcu_read_unlock();
715         return ret;
716 }