- patches.suse/slab-handle-memoryless-nodes-v2a.patch: Refresh.
[linux-flexiantxendom0-3.2.10.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/smp_lock.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_do_free_delegation(struct nfs_delegation *delegation)
25 {
26         kfree(delegation);
27 }
28
29 static void nfs_free_delegation_callback(struct rcu_head *head)
30 {
31         struct nfs_delegation *delegation = container_of(head, struct nfs_delegation, rcu);
32
33         nfs_do_free_delegation(delegation);
34 }
35
36 static void nfs_free_delegation(struct nfs_delegation *delegation)
37 {
38         struct rpc_cred *cred;
39
40         cred = rcu_dereference(delegation->cred);
41         rcu_assign_pointer(delegation->cred, NULL);
42         call_rcu(&delegation->rcu, nfs_free_delegation_callback);
43         if (cred)
44                 put_rpccred(cred);
45 }
46
47 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
48 {
49         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
50 }
51
52 int nfs_have_delegation(struct inode *inode, fmode_t flags)
53 {
54         struct nfs_delegation *delegation;
55         int ret = 0;
56
57         flags &= FMODE_READ|FMODE_WRITE;
58         rcu_read_lock();
59         delegation = rcu_dereference(NFS_I(inode)->delegation);
60         if (delegation != NULL && (delegation->type & flags) == flags) {
61                 nfs_mark_delegation_referenced(delegation);
62                 ret = 1;
63         }
64         rcu_read_unlock();
65         return ret;
66 }
67
68 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
69 {
70         struct inode *inode = state->inode;
71         struct file_lock *fl;
72         int status = 0;
73
74         if (inode->i_flock == NULL)
75                 goto out;
76
77         /* Protect inode->i_flock using the BKL */
78         lock_kernel();
79         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
80                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
81                         continue;
82                 if (nfs_file_open_context(fl->fl_file) != ctx)
83                         continue;
84                 unlock_kernel();
85                 status = nfs4_lock_delegation_recall(state, fl);
86                 if (status < 0)
87                         goto out;
88                 lock_kernel();
89         }
90         unlock_kernel();
91 out:
92         return status;
93 }
94
95 static int nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
96 {
97         struct nfs_inode *nfsi = NFS_I(inode);
98         struct nfs_open_context *ctx;
99         struct nfs4_state *state;
100         int err;
101
102 again:
103         spin_lock(&inode->i_lock);
104         list_for_each_entry(ctx, &nfsi->open_files, list) {
105                 state = ctx->state;
106                 if (state == NULL)
107                         continue;
108                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
109                         continue;
110                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
111                         continue;
112                 get_nfs_open_context(ctx);
113                 spin_unlock(&inode->i_lock);
114                 err = nfs4_open_delegation_recall(ctx, state, stateid);
115                 if (err >= 0)
116                         err = nfs_delegation_claim_locks(ctx, state);
117                 put_nfs_open_context(ctx);
118                 if (err != 0)
119                         return err;
120                 goto again;
121         }
122         spin_unlock(&inode->i_lock);
123         return 0;
124 }
125
126 /*
127  * Set up a delegation on an inode
128  */
129 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
130 {
131         struct nfs_delegation *delegation = NFS_I(inode)->delegation;
132         struct rpc_cred *oldcred;
133
134         if (delegation == NULL)
135                 return;
136         memcpy(delegation->stateid.data, res->delegation.data,
137                         sizeof(delegation->stateid.data));
138         delegation->type = res->delegation_type;
139         delegation->maxsize = res->maxsize;
140         oldcred = delegation->cred;
141         delegation->cred = get_rpccred(cred);
142         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
143         NFS_I(inode)->delegation_state = delegation->type;
144         smp_wmb();
145         put_rpccred(oldcred);
146 }
147
148 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
149 {
150         int res = 0;
151
152         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
153         nfs_free_delegation(delegation);
154         return res;
155 }
156
157 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
158 {
159         struct inode *inode = NULL;
160
161         spin_lock(&delegation->lock);
162         if (delegation->inode != NULL)
163                 inode = igrab(delegation->inode);
164         spin_unlock(&delegation->lock);
165         return inode;
166 }
167
168 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
169 {
170         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
171
172         if (delegation == NULL)
173                 goto nomatch;
174         spin_lock(&delegation->lock);
175         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
176                                 sizeof(delegation->stateid.data)) != 0)
177                 goto nomatch_unlock;
178         list_del_rcu(&delegation->super_list);
179         delegation->inode = NULL;
180         nfsi->delegation_state = 0;
181         rcu_assign_pointer(nfsi->delegation, NULL);
182         spin_unlock(&delegation->lock);
183         return delegation;
184 nomatch_unlock:
185         spin_unlock(&delegation->lock);
186 nomatch:
187         return NULL;
188 }
189
190 /*
191  * Set up a delegation on an inode
192  */
193 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
194 {
195         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
196         struct nfs_inode *nfsi = NFS_I(inode);
197         struct nfs_delegation *delegation;
198         struct nfs_delegation *freeme = NULL;
199         int status = 0;
200
201         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
202         if (delegation == NULL)
203                 return -ENOMEM;
204         memcpy(delegation->stateid.data, res->delegation.data,
205                         sizeof(delegation->stateid.data));
206         delegation->type = res->delegation_type;
207         delegation->maxsize = res->maxsize;
208         delegation->change_attr = nfsi->change_attr;
209         delegation->cred = get_rpccred(cred);
210         delegation->inode = inode;
211         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
212         spin_lock_init(&delegation->lock);
213
214         spin_lock(&clp->cl_lock);
215         if (rcu_dereference(nfsi->delegation) != NULL) {
216                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
217                                         sizeof(delegation->stateid)) == 0 &&
218                                 delegation->type == nfsi->delegation->type) {
219                         goto out;
220                 }
221                 /*
222                  * Deal with broken servers that hand out two
223                  * delegations for the same file.
224                  */
225                 dfprintk(FILE, "%s: server %s handed out "
226                                 "a duplicate delegation!\n",
227                                 __func__, clp->cl_hostname);
228                 if (delegation->type <= nfsi->delegation->type) {
229                         freeme = delegation;
230                         delegation = NULL;
231                         goto out;
232                 }
233                 freeme = nfs_detach_delegation_locked(nfsi, NULL);
234         }
235         list_add_rcu(&delegation->super_list, &clp->cl_delegations);
236         nfsi->delegation_state = delegation->type;
237         rcu_assign_pointer(nfsi->delegation, delegation);
238         delegation = NULL;
239
240         /* Ensure we revalidate the attributes and page cache! */
241         spin_lock(&inode->i_lock);
242         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
243         spin_unlock(&inode->i_lock);
244
245 out:
246         spin_unlock(&clp->cl_lock);
247         if (delegation != NULL)
248                 nfs_free_delegation(delegation);
249         if (freeme != NULL)
250                 nfs_do_return_delegation(inode, freeme, 0);
251         return status;
252 }
253
254 /* Sync all data to disk upon delegation return */
255 static void nfs_msync_inode(struct inode *inode)
256 {
257         filemap_fdatawrite(inode->i_mapping);
258         nfs_wb_all(inode);
259         filemap_fdatawait(inode->i_mapping);
260 }
261
262 /*
263  * Basic procedure for returning a delegation to the server
264  */
265 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
266 {
267         struct nfs_inode *nfsi = NFS_I(inode);
268         int err;
269
270         /*
271          * Guard against new delegated open/lock/unlock calls and against
272          * state recovery
273          */
274         down_write(&nfsi->rwsem);
275         err = nfs_delegation_claim_opens(inode, &delegation->stateid);
276         up_write(&nfsi->rwsem);
277         if (err)
278                 goto out;
279
280         err = nfs_do_return_delegation(inode, delegation, issync);
281 out:
282         return err;
283 }
284
285 /*
286  * Return all delegations that have been marked for return
287  */
288 int nfs_client_return_marked_delegations(struct nfs_client *clp)
289 {
290         struct nfs_delegation *delegation;
291         struct inode *inode;
292         int err = 0;
293
294 restart:
295         rcu_read_lock();
296         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
297                 if (!test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
298                         continue;
299                 inode = nfs_delegation_grab_inode(delegation);
300                 if (inode == NULL)
301                         continue;
302                 spin_lock(&clp->cl_lock);
303                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
304                 spin_unlock(&clp->cl_lock);
305                 rcu_read_unlock();
306                 if (delegation != NULL) {
307                         filemap_flush(inode->i_mapping);
308                         err = __nfs_inode_return_delegation(inode, delegation, 0);
309                 }
310                 iput(inode);
311                 if (!err)
312                         goto restart;
313                 set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
314                 return err;
315         }
316         rcu_read_unlock();
317         return 0;
318 }
319
320 /*
321  * This function returns the delegation without reclaiming opens
322  * or protecting against delegation reclaims.
323  * It is therefore really only safe to be called from
324  * nfs4_clear_inode()
325  */
326 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
327 {
328         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
329         struct nfs_inode *nfsi = NFS_I(inode);
330         struct nfs_delegation *delegation;
331
332         if (rcu_dereference(nfsi->delegation) != NULL) {
333                 spin_lock(&clp->cl_lock);
334                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
335                 spin_unlock(&clp->cl_lock);
336                 if (delegation != NULL)
337                         nfs_do_return_delegation(inode, delegation, 0);
338         }
339 }
340
341 int nfs_inode_return_delegation(struct inode *inode)
342 {
343         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
344         struct nfs_inode *nfsi = NFS_I(inode);
345         struct nfs_delegation *delegation;
346         int err = 0;
347
348         if (rcu_dereference(nfsi->delegation) != NULL) {
349                 spin_lock(&clp->cl_lock);
350                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
351                 spin_unlock(&clp->cl_lock);
352                 if (delegation != NULL) {
353                         nfs_msync_inode(inode);
354                         err = __nfs_inode_return_delegation(inode, delegation, 1);
355                 }
356         }
357         return err;
358 }
359
360 static void nfs_mark_return_delegation(struct nfs_client *clp, struct nfs_delegation *delegation)
361 {
362         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
363         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
364 }
365
366 /*
367  * Return all delegations associated to a super block
368  */
369 void nfs_super_return_all_delegations(struct super_block *sb)
370 {
371         struct nfs_client *clp = NFS_SB(sb)->nfs_client;
372         struct nfs_delegation *delegation;
373
374         if (clp == NULL)
375                 return;
376         rcu_read_lock();
377         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
378                 spin_lock(&delegation->lock);
379                 if (delegation->inode != NULL && delegation->inode->i_sb == sb)
380                         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
381                 spin_unlock(&delegation->lock);
382         }
383         rcu_read_unlock();
384         if (nfs_client_return_marked_delegations(clp) != 0)
385                 nfs4_schedule_state_manager(clp);
386 }
387
388 static
389 void nfs_client_mark_return_all_delegation_types(struct nfs_client *clp, fmode_t flags)
390 {
391         struct nfs_delegation *delegation;
392
393         rcu_read_lock();
394         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
395                 if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
396                         continue;
397                 if (delegation->type & flags)
398                         nfs_mark_return_delegation(clp, delegation);
399         }
400         rcu_read_unlock();
401 }
402
403 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
404 {
405         nfs_client_mark_return_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
406 }
407
408 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
409 {
410         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
411                 nfs4_schedule_state_manager(clp);
412 }
413
414 void nfs_expire_all_delegation_types(struct nfs_client *clp, fmode_t flags)
415 {
416         nfs_client_mark_return_all_delegation_types(clp, flags);
417         nfs_delegation_run_state_manager(clp);
418 }
419
420 void nfs_expire_all_delegations(struct nfs_client *clp)
421 {
422         nfs_expire_all_delegation_types(clp, FMODE_READ|FMODE_WRITE);
423 }
424
425 /*
426  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
427  */
428 void nfs_handle_cb_pathdown(struct nfs_client *clp)
429 {
430         if (clp == NULL)
431                 return;
432         nfs_client_mark_return_all_delegations(clp);
433 }
434
435 static void nfs_client_mark_return_unreferenced_delegations(struct nfs_client *clp)
436 {
437         struct nfs_delegation *delegation;
438
439         rcu_read_lock();
440         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
441                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
442                         continue;
443                 nfs_mark_return_delegation(clp, delegation);
444         }
445         rcu_read_unlock();
446 }
447
448 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
449 {
450         nfs_client_mark_return_unreferenced_delegations(clp);
451         nfs_delegation_run_state_manager(clp);
452 }
453
454 /*
455  * Asynchronous delegation recall!
456  */
457 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid,
458                                       int (*validate_stateid)(struct nfs_delegation *delegation,
459                                                               const nfs4_stateid *stateid))
460 {
461         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
462         struct nfs_delegation *delegation;
463
464         rcu_read_lock();
465         delegation = rcu_dereference(NFS_I(inode)->delegation);
466
467         if (!validate_stateid(delegation, stateid)) {
468                 rcu_read_unlock();
469                 return -ENOENT;
470         }
471
472         nfs_mark_return_delegation(clp, delegation);
473         rcu_read_unlock();
474         nfs_delegation_run_state_manager(clp);
475         return 0;
476 }
477
478 /*
479  * Retrieve the inode associated with a delegation
480  */
481 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
482 {
483         struct nfs_delegation *delegation;
484         struct inode *res = NULL;
485         rcu_read_lock();
486         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
487                 spin_lock(&delegation->lock);
488                 if (delegation->inode != NULL &&
489                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
490                         res = igrab(delegation->inode);
491                 }
492                 spin_unlock(&delegation->lock);
493                 if (res != NULL)
494                         break;
495         }
496         rcu_read_unlock();
497         return res;
498 }
499
500 /*
501  * Mark all delegations as needing to be reclaimed
502  */
503 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
504 {
505         struct nfs_delegation *delegation;
506         rcu_read_lock();
507         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
508                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
509         rcu_read_unlock();
510 }
511
512 /*
513  * Reap all unclaimed delegations after reboot recovery is done
514  */
515 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
516 {
517         struct nfs_delegation *delegation;
518         struct inode *inode;
519 restart:
520         rcu_read_lock();
521         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
522                 if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) == 0)
523                         continue;
524                 inode = nfs_delegation_grab_inode(delegation);
525                 if (inode == NULL)
526                         continue;
527                 spin_lock(&clp->cl_lock);
528                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
529                 spin_unlock(&clp->cl_lock);
530                 rcu_read_unlock();
531                 if (delegation != NULL)
532                         nfs_free_delegation(delegation);
533                 iput(inode);
534                 goto restart;
535         }
536         rcu_read_unlock();
537 }
538
539 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
540 {
541         struct nfs_inode *nfsi = NFS_I(inode);
542         struct nfs_delegation *delegation;
543         int ret = 0;
544
545         rcu_read_lock();
546         delegation = rcu_dereference(nfsi->delegation);
547         if (delegation != NULL) {
548                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
549                 ret = 1;
550         }
551         rcu_read_unlock();
552         return ret;
553 }