Merge branch 'for-2.6.31' of git://fieldses.org/git/linux-nfsd
[linux-2.6] / 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/spinlock.h>
14
15 #include <linux/nfs4.h>
16 #include <linux/nfs_fs.h>
17 #include <linux/nfs_xdr.h>
18
19 #include "nfs4_fs.h"
20 #include "delegation.h"
21 #include "internal.h"
22
23 static void nfs_do_free_delegation(struct nfs_delegation *delegation)
24 {
25         kfree(delegation);
26 }
27
28 static void nfs_free_delegation_callback(struct rcu_head *head)
29 {
30         struct nfs_delegation *delegation = container_of(head, struct nfs_delegation, rcu);
31
32         nfs_do_free_delegation(delegation);
33 }
34
35 static void nfs_free_delegation(struct nfs_delegation *delegation)
36 {
37         struct rpc_cred *cred;
38
39         cred = rcu_dereference(delegation->cred);
40         rcu_assign_pointer(delegation->cred, NULL);
41         call_rcu(&delegation->rcu, nfs_free_delegation_callback);
42         if (cred)
43                 put_rpccred(cred);
44 }
45
46 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
47 {
48         set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
49 }
50
51 int nfs_have_delegation(struct inode *inode, fmode_t flags)
52 {
53         struct nfs_delegation *delegation;
54         int ret = 0;
55
56         flags &= FMODE_READ|FMODE_WRITE;
57         rcu_read_lock();
58         delegation = rcu_dereference(NFS_I(inode)->delegation);
59         if (delegation != NULL && (delegation->type & flags) == flags) {
60                 nfs_mark_delegation_referenced(delegation);
61                 ret = 1;
62         }
63         rcu_read_unlock();
64         return ret;
65 }
66
67 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
68 {
69         struct inode *inode = state->inode;
70         struct file_lock *fl;
71         int status = 0;
72
73         if (inode->i_flock == NULL)
74                 goto out;
75
76         /* Protect inode->i_flock using the BKL */
77         lock_kernel();
78         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
79                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
80                         continue;
81                 if (nfs_file_open_context(fl->fl_file) != ctx)
82                         continue;
83                 unlock_kernel();
84                 status = nfs4_lock_delegation_recall(state, fl);
85                 if (status < 0)
86                         goto out;
87                 lock_kernel();
88         }
89         unlock_kernel();
90 out:
91         return status;
92 }
93
94 static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
95 {
96         struct nfs_inode *nfsi = NFS_I(inode);
97         struct nfs_open_context *ctx;
98         struct nfs4_state *state;
99         int err;
100
101 again:
102         spin_lock(&inode->i_lock);
103         list_for_each_entry(ctx, &nfsi->open_files, list) {
104                 state = ctx->state;
105                 if (state == NULL)
106                         continue;
107                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
108                         continue;
109                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
110                         continue;
111                 get_nfs_open_context(ctx);
112                 spin_unlock(&inode->i_lock);
113                 err = nfs4_open_delegation_recall(ctx, state, stateid);
114                 if (err >= 0)
115                         err = nfs_delegation_claim_locks(ctx, state);
116                 put_nfs_open_context(ctx);
117                 if (err != 0)
118                         return;
119                 goto again;
120         }
121         spin_unlock(&inode->i_lock);
122 }
123
124 /*
125  * Set up a delegation on an inode
126  */
127 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
128 {
129         struct nfs_delegation *delegation = NFS_I(inode)->delegation;
130         struct rpc_cred *oldcred;
131
132         if (delegation == NULL)
133                 return;
134         memcpy(delegation->stateid.data, res->delegation.data,
135                         sizeof(delegation->stateid.data));
136         delegation->type = res->delegation_type;
137         delegation->maxsize = res->maxsize;
138         oldcred = delegation->cred;
139         delegation->cred = get_rpccred(cred);
140         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
141         NFS_I(inode)->delegation_state = delegation->type;
142         smp_wmb();
143         put_rpccred(oldcred);
144 }
145
146 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
147 {
148         int res = 0;
149
150         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
151         nfs_free_delegation(delegation);
152         return res;
153 }
154
155 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
156 {
157         struct inode *inode = NULL;
158
159         spin_lock(&delegation->lock);
160         if (delegation->inode != NULL)
161                 inode = igrab(delegation->inode);
162         spin_unlock(&delegation->lock);
163         return inode;
164 }
165
166 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
167 {
168         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
169
170         if (delegation == NULL)
171                 goto nomatch;
172         spin_lock(&delegation->lock);
173         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
174                                 sizeof(delegation->stateid.data)) != 0)
175                 goto nomatch_unlock;
176         list_del_rcu(&delegation->super_list);
177         delegation->inode = NULL;
178         nfsi->delegation_state = 0;
179         rcu_assign_pointer(nfsi->delegation, NULL);
180         spin_unlock(&delegation->lock);
181         return delegation;
182 nomatch_unlock:
183         spin_unlock(&delegation->lock);
184 nomatch:
185         return NULL;
186 }
187
188 /*
189  * Set up a delegation on an inode
190  */
191 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
192 {
193         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
194         struct nfs_inode *nfsi = NFS_I(inode);
195         struct nfs_delegation *delegation;
196         struct nfs_delegation *freeme = NULL;
197         int status = 0;
198
199         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
200         if (delegation == NULL)
201                 return -ENOMEM;
202         memcpy(delegation->stateid.data, res->delegation.data,
203                         sizeof(delegation->stateid.data));
204         delegation->type = res->delegation_type;
205         delegation->maxsize = res->maxsize;
206         delegation->change_attr = nfsi->change_attr;
207         delegation->cred = get_rpccred(cred);
208         delegation->inode = inode;
209         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
210         spin_lock_init(&delegation->lock);
211
212         spin_lock(&clp->cl_lock);
213         if (rcu_dereference(nfsi->delegation) != NULL) {
214                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
215                                         sizeof(delegation->stateid)) == 0 &&
216                                 delegation->type == nfsi->delegation->type) {
217                         goto out;
218                 }
219                 /*
220                  * Deal with broken servers that hand out two
221                  * delegations for the same file.
222                  */
223                 dfprintk(FILE, "%s: server %s handed out "
224                                 "a duplicate delegation!\n",
225                                 __func__, clp->cl_hostname);
226                 if (delegation->type <= nfsi->delegation->type) {
227                         freeme = delegation;
228                         delegation = NULL;
229                         goto out;
230                 }
231                 freeme = nfs_detach_delegation_locked(nfsi, NULL);
232         }
233         list_add_rcu(&delegation->super_list, &clp->cl_delegations);
234         nfsi->delegation_state = delegation->type;
235         rcu_assign_pointer(nfsi->delegation, delegation);
236         delegation = NULL;
237
238         /* Ensure we revalidate the attributes and page cache! */
239         spin_lock(&inode->i_lock);
240         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
241         spin_unlock(&inode->i_lock);
242
243 out:
244         spin_unlock(&clp->cl_lock);
245         if (delegation != NULL)
246                 nfs_free_delegation(delegation);
247         if (freeme != NULL)
248                 nfs_do_return_delegation(inode, freeme, 0);
249         return status;
250 }
251
252 /* Sync all data to disk upon delegation return */
253 static void nfs_msync_inode(struct inode *inode)
254 {
255         filemap_fdatawrite(inode->i_mapping);
256         nfs_wb_all(inode);
257         filemap_fdatawait(inode->i_mapping);
258 }
259
260 /*
261  * Basic procedure for returning a delegation to the server
262  */
263 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
264 {
265         struct nfs_inode *nfsi = NFS_I(inode);
266
267         nfs_msync_inode(inode);
268         /*
269          * Guard against new delegated open/lock/unlock calls and against
270          * state recovery
271          */
272         down_write(&nfsi->rwsem);
273         nfs_delegation_claim_opens(inode, &delegation->stateid);
274         up_write(&nfsi->rwsem);
275         nfs_msync_inode(inode);
276
277         return nfs_do_return_delegation(inode, delegation, 1);
278 }
279
280 /*
281  * Return all delegations that have been marked for return
282  */
283 void nfs_client_return_marked_delegations(struct nfs_client *clp)
284 {
285         struct nfs_delegation *delegation;
286         struct inode *inode;
287
288 restart:
289         rcu_read_lock();
290         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
291                 if (!test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
292                         continue;
293                 inode = nfs_delegation_grab_inode(delegation);
294                 if (inode == NULL)
295                         continue;
296                 spin_lock(&clp->cl_lock);
297                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
298                 spin_unlock(&clp->cl_lock);
299                 rcu_read_unlock();
300                 if (delegation != NULL)
301                         __nfs_inode_return_delegation(inode, delegation);
302                 iput(inode);
303                 goto restart;
304         }
305         rcu_read_unlock();
306 }
307
308 /*
309  * This function returns the delegation without reclaiming opens
310  * or protecting against delegation reclaims.
311  * It is therefore really only safe to be called from
312  * nfs4_clear_inode()
313  */
314 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
315 {
316         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
317         struct nfs_inode *nfsi = NFS_I(inode);
318         struct nfs_delegation *delegation;
319
320         if (rcu_dereference(nfsi->delegation) != NULL) {
321                 spin_lock(&clp->cl_lock);
322                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
323                 spin_unlock(&clp->cl_lock);
324                 if (delegation != NULL)
325                         nfs_do_return_delegation(inode, delegation, 0);
326         }
327 }
328
329 int nfs_inode_return_delegation(struct inode *inode)
330 {
331         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
332         struct nfs_inode *nfsi = NFS_I(inode);
333         struct nfs_delegation *delegation;
334         int err = 0;
335
336         if (rcu_dereference(nfsi->delegation) != NULL) {
337                 spin_lock(&clp->cl_lock);
338                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
339                 spin_unlock(&clp->cl_lock);
340                 if (delegation != NULL)
341                         err = __nfs_inode_return_delegation(inode, delegation);
342         }
343         return err;
344 }
345
346 static void nfs_mark_return_delegation(struct nfs_client *clp, struct nfs_delegation *delegation)
347 {
348         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
349         set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
350 }
351
352 /*
353  * Return all delegations associated to a super block
354  */
355 void nfs_super_return_all_delegations(struct super_block *sb)
356 {
357         struct nfs_client *clp = NFS_SB(sb)->nfs_client;
358         struct nfs_delegation *delegation;
359
360         if (clp == NULL)
361                 return;
362         rcu_read_lock();
363         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
364                 spin_lock(&delegation->lock);
365                 if (delegation->inode != NULL && delegation->inode->i_sb == sb)
366                         set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
367                 spin_unlock(&delegation->lock);
368         }
369         rcu_read_unlock();
370         nfs_client_return_marked_delegations(clp);
371 }
372
373 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
374 {
375         struct nfs_delegation *delegation;
376
377         rcu_read_lock();
378         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
379                 set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
380                 set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
381         }
382         rcu_read_unlock();
383 }
384
385 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
386 {
387         if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
388                 nfs4_schedule_state_manager(clp);
389 }
390
391 void nfs_expire_all_delegations(struct nfs_client *clp)
392 {
393         nfs_client_mark_return_all_delegations(clp);
394         nfs_delegation_run_state_manager(clp);
395 }
396
397 /*
398  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
399  */
400 void nfs_handle_cb_pathdown(struct nfs_client *clp)
401 {
402         if (clp == NULL)
403                 return;
404         nfs_client_mark_return_all_delegations(clp);
405 }
406
407 static void nfs_client_mark_return_unreferenced_delegations(struct nfs_client *clp)
408 {
409         struct nfs_delegation *delegation;
410
411         rcu_read_lock();
412         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
413                 if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
414                         continue;
415                 set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
416                 set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
417         }
418         rcu_read_unlock();
419 }
420
421 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
422 {
423         nfs_client_mark_return_unreferenced_delegations(clp);
424         nfs_delegation_run_state_manager(clp);
425 }
426
427 /*
428  * Asynchronous delegation recall!
429  */
430 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid)
431 {
432         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
433         struct nfs_delegation *delegation;
434
435         rcu_read_lock();
436         delegation = rcu_dereference(NFS_I(inode)->delegation);
437         if (delegation == NULL || memcmp(delegation->stateid.data, stateid->data,
438                                 sizeof(delegation->stateid.data)) != 0) {
439                 rcu_read_unlock();
440                 return -ENOENT;
441         }
442         nfs_mark_return_delegation(clp, delegation);
443         rcu_read_unlock();
444         nfs_delegation_run_state_manager(clp);
445         return 0;
446 }
447
448 /*
449  * Retrieve the inode associated with a delegation
450  */
451 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
452 {
453         struct nfs_delegation *delegation;
454         struct inode *res = NULL;
455         rcu_read_lock();
456         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
457                 spin_lock(&delegation->lock);
458                 if (delegation->inode != NULL &&
459                     nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
460                         res = igrab(delegation->inode);
461                 }
462                 spin_unlock(&delegation->lock);
463                 if (res != NULL)
464                         break;
465         }
466         rcu_read_unlock();
467         return res;
468 }
469
470 /*
471  * Mark all delegations as needing to be reclaimed
472  */
473 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
474 {
475         struct nfs_delegation *delegation;
476         rcu_read_lock();
477         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
478                 set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
479         rcu_read_unlock();
480 }
481
482 /*
483  * Reap all unclaimed delegations after reboot recovery is done
484  */
485 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
486 {
487         struct nfs_delegation *delegation;
488         struct inode *inode;
489 restart:
490         rcu_read_lock();
491         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
492                 if (test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) == 0)
493                         continue;
494                 inode = nfs_delegation_grab_inode(delegation);
495                 if (inode == NULL)
496                         continue;
497                 spin_lock(&clp->cl_lock);
498                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
499                 spin_unlock(&clp->cl_lock);
500                 rcu_read_unlock();
501                 if (delegation != NULL)
502                         nfs_free_delegation(delegation);
503                 iput(inode);
504                 goto restart;
505         }
506         rcu_read_unlock();
507 }
508
509 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
510 {
511         struct nfs_inode *nfsi = NFS_I(inode);
512         struct nfs_delegation *delegation;
513         int ret = 0;
514
515         rcu_read_lock();
516         delegation = rcu_dereference(nfsi->delegation);
517         if (delegation != NULL) {
518                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
519                 ret = 1;
520         }
521         rcu_read_unlock();
522         return ret;
523 }