Merge git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[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;
72
73         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
74                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
75                         continue;
76                 if (nfs_file_open_context(fl->fl_file) != ctx)
77                         continue;
78                 status = nfs4_lock_delegation_recall(state, fl);
79                 if (status >= 0)
80                         continue;
81                 switch (status) {
82                         default:
83                                 printk(KERN_ERR "%s: unhandled error %d.\n",
84                                                 __func__, status);
85                         case -NFS4ERR_EXPIRED:
86                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
87                         case -NFS4ERR_STALE_CLIENTID:
88                                 nfs4_schedule_state_recovery(NFS_SERVER(inode)->nfs_client);
89                                 goto out_err;
90                 }
91         }
92         return 0;
93 out_err:
94         return status;
95 }
96
97 static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
98 {
99         struct nfs_inode *nfsi = NFS_I(inode);
100         struct nfs_open_context *ctx;
101         struct nfs4_state *state;
102         int err;
103
104 again:
105         spin_lock(&inode->i_lock);
106         list_for_each_entry(ctx, &nfsi->open_files, list) {
107                 state = ctx->state;
108                 if (state == NULL)
109                         continue;
110                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
111                         continue;
112                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
113                         continue;
114                 get_nfs_open_context(ctx);
115                 spin_unlock(&inode->i_lock);
116                 err = nfs4_open_delegation_recall(ctx, state, stateid);
117                 if (err >= 0)
118                         err = nfs_delegation_claim_locks(ctx, state);
119                 put_nfs_open_context(ctx);
120                 if (err != 0)
121                         return;
122                 goto again;
123         }
124         spin_unlock(&inode->i_lock);
125 }
126
127 /*
128  * Set up a delegation on an inode
129  */
130 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
131 {
132         struct nfs_delegation *delegation = NFS_I(inode)->delegation;
133         struct rpc_cred *oldcred;
134
135         if (delegation == NULL)
136                 return;
137         memcpy(delegation->stateid.data, res->delegation.data,
138                         sizeof(delegation->stateid.data));
139         delegation->type = res->delegation_type;
140         delegation->maxsize = res->maxsize;
141         oldcred = delegation->cred;
142         delegation->cred = get_rpccred(cred);
143         clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
144         NFS_I(inode)->delegation_state = delegation->type;
145         smp_wmb();
146         put_rpccred(oldcred);
147 }
148
149 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
150 {
151         int res = 0;
152
153         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
154         nfs_free_delegation(delegation);
155         return res;
156 }
157
158 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
159 {
160         struct inode *inode = NULL;
161
162         spin_lock(&delegation->lock);
163         if (delegation->inode != NULL)
164                 inode = igrab(delegation->inode);
165         spin_unlock(&delegation->lock);
166         return inode;
167 }
168
169 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
170 {
171         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
172
173         if (delegation == NULL)
174                 goto nomatch;
175         spin_lock(&delegation->lock);
176         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
177                                 sizeof(delegation->stateid.data)) != 0)
178                 goto nomatch_unlock;
179         list_del_rcu(&delegation->super_list);
180         delegation->inode = NULL;
181         nfsi->delegation_state = 0;
182         rcu_assign_pointer(nfsi->delegation, NULL);
183         spin_unlock(&delegation->lock);
184         return delegation;
185 nomatch_unlock:
186         spin_unlock(&delegation->lock);
187 nomatch:
188         return NULL;
189 }
190
191 /*
192  * Set up a delegation on an inode
193  */
194 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
195 {
196         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
197         struct nfs_inode *nfsi = NFS_I(inode);
198         struct nfs_delegation *delegation;
199         struct nfs_delegation *freeme = NULL;
200         int status = 0;
201
202         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
203         if (delegation == NULL)
204                 return -ENOMEM;
205         memcpy(delegation->stateid.data, res->delegation.data,
206                         sizeof(delegation->stateid.data));
207         delegation->type = res->delegation_type;
208         delegation->maxsize = res->maxsize;
209         delegation->change_attr = nfsi->change_attr;
210         delegation->cred = get_rpccred(cred);
211         delegation->inode = inode;
212         delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
213         spin_lock_init(&delegation->lock);
214
215         spin_lock(&clp->cl_lock);
216         if (rcu_dereference(nfsi->delegation) != NULL) {
217                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
218                                         sizeof(delegation->stateid)) == 0 &&
219                                 delegation->type == nfsi->delegation->type) {
220                         goto out;
221                 }
222                 /*
223                  * Deal with broken servers that hand out two
224                  * delegations for the same file.
225                  */
226                 dfprintk(FILE, "%s: server %s handed out "
227                                 "a duplicate delegation!\n",
228                                 __func__, clp->cl_hostname);
229                 if (delegation->type <= nfsi->delegation->type) {
230                         freeme = delegation;
231                         delegation = NULL;
232                         goto out;
233                 }
234                 freeme = nfs_detach_delegation_locked(nfsi, NULL);
235         }
236         list_add_rcu(&delegation->super_list, &clp->cl_delegations);
237         nfsi->delegation_state = delegation->type;
238         rcu_assign_pointer(nfsi->delegation, delegation);
239         delegation = NULL;
240
241         /* Ensure we revalidate the attributes and page cache! */
242         spin_lock(&inode->i_lock);
243         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
244         spin_unlock(&inode->i_lock);
245
246 out:
247         spin_unlock(&clp->cl_lock);
248         if (delegation != NULL)
249                 nfs_free_delegation(delegation);
250         if (freeme != NULL)
251                 nfs_do_return_delegation(inode, freeme, 0);
252         return status;
253 }
254
255 /* Sync all data to disk upon delegation return */
256 static void nfs_msync_inode(struct inode *inode)
257 {
258         filemap_fdatawrite(inode->i_mapping);
259         nfs_wb_all(inode);
260         filemap_fdatawait(inode->i_mapping);
261 }
262
263 /*
264  * Basic procedure for returning a delegation to the server
265  */
266 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
267 {
268         struct nfs_inode *nfsi = NFS_I(inode);
269
270         nfs_msync_inode(inode);
271         /* Guard against new delegated open calls */
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 }