Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-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 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
47 {
48         struct inode *inode = state->inode;
49         struct file_lock *fl;
50         int status;
51
52         for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
53                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
54                         continue;
55                 if (nfs_file_open_context(fl->fl_file) != ctx)
56                         continue;
57                 status = nfs4_lock_delegation_recall(state, fl);
58                 if (status >= 0)
59                         continue;
60                 switch (status) {
61                         default:
62                                 printk(KERN_ERR "%s: unhandled error %d.\n",
63                                                 __FUNCTION__, status);
64                         case -NFS4ERR_EXPIRED:
65                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
66                         case -NFS4ERR_STALE_CLIENTID:
67                                 nfs4_schedule_state_recovery(NFS_SERVER(inode)->nfs_client);
68                                 goto out_err;
69                 }
70         }
71         return 0;
72 out_err:
73         return status;
74 }
75
76 static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
77 {
78         struct nfs_inode *nfsi = NFS_I(inode);
79         struct nfs_open_context *ctx;
80         struct nfs4_state *state;
81         int err;
82
83 again:
84         spin_lock(&inode->i_lock);
85         list_for_each_entry(ctx, &nfsi->open_files, list) {
86                 state = ctx->state;
87                 if (state == NULL)
88                         continue;
89                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
90                         continue;
91                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
92                         continue;
93                 get_nfs_open_context(ctx);
94                 spin_unlock(&inode->i_lock);
95                 err = nfs4_open_delegation_recall(ctx, state, stateid);
96                 if (err >= 0)
97                         err = nfs_delegation_claim_locks(ctx, state);
98                 put_nfs_open_context(ctx);
99                 if (err != 0)
100                         return;
101                 goto again;
102         }
103         spin_unlock(&inode->i_lock);
104 }
105
106 /*
107  * Set up a delegation on an inode
108  */
109 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
110 {
111         struct nfs_delegation *delegation = NFS_I(inode)->delegation;
112         struct rpc_cred *oldcred;
113
114         if (delegation == NULL)
115                 return;
116         memcpy(delegation->stateid.data, res->delegation.data,
117                         sizeof(delegation->stateid.data));
118         delegation->type = res->delegation_type;
119         delegation->maxsize = res->maxsize;
120         oldcred = delegation->cred;
121         delegation->cred = get_rpccred(cred);
122         delegation->flags &= ~NFS_DELEGATION_NEED_RECLAIM;
123         NFS_I(inode)->delegation_state = delegation->type;
124         smp_wmb();
125         put_rpccred(oldcred);
126 }
127
128 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
129 {
130         int res = 0;
131
132         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid, issync);
133         nfs_free_delegation(delegation);
134         return res;
135 }
136
137 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
138 {
139         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
140
141         if (delegation == NULL)
142                 goto nomatch;
143         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
144                                 sizeof(delegation->stateid.data)) != 0)
145                 goto nomatch;
146         list_del_rcu(&delegation->super_list);
147         nfsi->delegation_state = 0;
148         rcu_assign_pointer(nfsi->delegation, NULL);
149         return delegation;
150 nomatch:
151         return NULL;
152 }
153
154 /*
155  * Set up a delegation on an inode
156  */
157 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
158 {
159         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
160         struct nfs_inode *nfsi = NFS_I(inode);
161         struct nfs_delegation *delegation;
162         struct nfs_delegation *freeme = NULL;
163         int status = 0;
164
165         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
166         if (delegation == NULL)
167                 return -ENOMEM;
168         memcpy(delegation->stateid.data, res->delegation.data,
169                         sizeof(delegation->stateid.data));
170         delegation->type = res->delegation_type;
171         delegation->maxsize = res->maxsize;
172         delegation->change_attr = nfsi->change_attr;
173         delegation->cred = get_rpccred(cred);
174         delegation->inode = inode;
175
176         spin_lock(&clp->cl_lock);
177         if (rcu_dereference(nfsi->delegation) != NULL) {
178                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
179                                         sizeof(delegation->stateid)) == 0 &&
180                                 delegation->type == nfsi->delegation->type) {
181                         goto out;
182                 }
183                 /*
184                  * Deal with broken servers that hand out two
185                  * delegations for the same file.
186                  */
187                 dfprintk(FILE, "%s: server %s handed out "
188                                 "a duplicate delegation!\n",
189                                 __FUNCTION__, clp->cl_hostname);
190                 if (delegation->type <= nfsi->delegation->type) {
191                         freeme = delegation;
192                         delegation = NULL;
193                         goto out;
194                 }
195                 freeme = nfs_detach_delegation_locked(nfsi, NULL);
196         }
197         list_add_rcu(&delegation->super_list, &clp->cl_delegations);
198         nfsi->delegation_state = delegation->type;
199         rcu_assign_pointer(nfsi->delegation, delegation);
200         delegation = NULL;
201
202         /* Ensure we revalidate the attributes and page cache! */
203         spin_lock(&inode->i_lock);
204         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
205         spin_unlock(&inode->i_lock);
206
207 out:
208         spin_unlock(&clp->cl_lock);
209         if (delegation != NULL)
210                 nfs_free_delegation(delegation);
211         if (freeme != NULL)
212                 nfs_do_return_delegation(inode, freeme, 0);
213         return status;
214 }
215
216 /* Sync all data to disk upon delegation return */
217 static void nfs_msync_inode(struct inode *inode)
218 {
219         filemap_fdatawrite(inode->i_mapping);
220         nfs_wb_all(inode);
221         filemap_fdatawait(inode->i_mapping);
222 }
223
224 /*
225  * Basic procedure for returning a delegation to the server
226  */
227 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
228 {
229         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
230         struct nfs_inode *nfsi = NFS_I(inode);
231
232         nfs_msync_inode(inode);
233         down_read(&clp->cl_sem);
234         /* Guard against new delegated open calls */
235         down_write(&nfsi->rwsem);
236         nfs_delegation_claim_opens(inode, &delegation->stateid);
237         up_write(&nfsi->rwsem);
238         up_read(&clp->cl_sem);
239         nfs_msync_inode(inode);
240
241         return nfs_do_return_delegation(inode, delegation, 1);
242 }
243
244 /*
245  * This function returns the delegation without reclaiming opens
246  * or protecting against delegation reclaims.
247  * It is therefore really only safe to be called from
248  * nfs4_clear_inode()
249  */
250 void nfs_inode_return_delegation_noreclaim(struct inode *inode)
251 {
252         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
253         struct nfs_inode *nfsi = NFS_I(inode);
254         struct nfs_delegation *delegation;
255
256         if (rcu_dereference(nfsi->delegation) != NULL) {
257                 spin_lock(&clp->cl_lock);
258                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
259                 spin_unlock(&clp->cl_lock);
260                 if (delegation != NULL)
261                         nfs_do_return_delegation(inode, delegation, 0);
262         }
263 }
264
265 int nfs_inode_return_delegation(struct inode *inode)
266 {
267         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
268         struct nfs_inode *nfsi = NFS_I(inode);
269         struct nfs_delegation *delegation;
270         int err = 0;
271
272         if (rcu_dereference(nfsi->delegation) != NULL) {
273                 spin_lock(&clp->cl_lock);
274                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
275                 spin_unlock(&clp->cl_lock);
276                 if (delegation != NULL)
277                         err = __nfs_inode_return_delegation(inode, delegation);
278         }
279         return err;
280 }
281
282 /*
283  * Return all delegations associated to a super block
284  */
285 void nfs_return_all_delegations(struct super_block *sb)
286 {
287         struct nfs_client *clp = NFS_SB(sb)->nfs_client;
288         struct nfs_delegation *delegation;
289         struct inode *inode;
290
291         if (clp == NULL)
292                 return;
293 restart:
294         rcu_read_lock();
295         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
296                 if (delegation->inode->i_sb != sb)
297                         continue;
298                 inode = igrab(delegation->inode);
299                 if (inode == NULL)
300                         continue;
301                 spin_lock(&clp->cl_lock);
302                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
303                 spin_unlock(&clp->cl_lock);
304                 rcu_read_unlock();
305                 if (delegation != NULL)
306                         __nfs_inode_return_delegation(inode, delegation);
307                 iput(inode);
308                 goto restart;
309         }
310         rcu_read_unlock();
311 }
312
313 static int nfs_do_expire_all_delegations(void *ptr)
314 {
315         struct nfs_client *clp = ptr;
316         struct nfs_delegation *delegation;
317         struct inode *inode;
318
319         allow_signal(SIGKILL);
320 restart:
321         if (test_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) != 0)
322                 goto out;
323         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0)
324                 goto out;
325         rcu_read_lock();
326         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
327                 inode = igrab(delegation->inode);
328                 if (inode == NULL)
329                         continue;
330                 spin_lock(&clp->cl_lock);
331                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
332                 spin_unlock(&clp->cl_lock);
333                 rcu_read_unlock();
334                 if (delegation)
335                         __nfs_inode_return_delegation(inode, delegation);
336                 iput(inode);
337                 goto restart;
338         }
339         rcu_read_unlock();
340 out:
341         nfs_put_client(clp);
342         module_put_and_exit(0);
343 }
344
345 void nfs_expire_all_delegations(struct nfs_client *clp)
346 {
347         struct task_struct *task;
348
349         __module_get(THIS_MODULE);
350         atomic_inc(&clp->cl_count);
351         task = kthread_run(nfs_do_expire_all_delegations, clp,
352                                 "%s-delegreturn",
353                                 rpc_peeraddr2str(clp->cl_rpcclient,
354                                                         RPC_DISPLAY_ADDR));
355         if (!IS_ERR(task))
356                 return;
357         nfs_put_client(clp);
358         module_put(THIS_MODULE);
359 }
360
361 /*
362  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
363  */
364 void nfs_handle_cb_pathdown(struct nfs_client *clp)
365 {
366         struct nfs_delegation *delegation;
367         struct inode *inode;
368
369         if (clp == NULL)
370                 return;
371 restart:
372         rcu_read_lock();
373         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
374                 inode = igrab(delegation->inode);
375                 if (inode == NULL)
376                         continue;
377                 spin_lock(&clp->cl_lock);
378                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
379                 spin_unlock(&clp->cl_lock);
380                 rcu_read_unlock();
381                 if (delegation != NULL)
382                         __nfs_inode_return_delegation(inode, delegation);
383                 iput(inode);
384                 goto restart;
385         }
386         rcu_read_unlock();
387 }
388
389 struct recall_threadargs {
390         struct inode *inode;
391         struct nfs_client *clp;
392         const nfs4_stateid *stateid;
393
394         struct completion started;
395         int result;
396 };
397
398 static int recall_thread(void *data)
399 {
400         struct recall_threadargs *args = (struct recall_threadargs *)data;
401         struct inode *inode = igrab(args->inode);
402         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
403         struct nfs_inode *nfsi = NFS_I(inode);
404         struct nfs_delegation *delegation;
405
406         daemonize("nfsv4-delegreturn");
407
408         nfs_msync_inode(inode);
409         down_read(&clp->cl_sem);
410         down_write(&nfsi->rwsem);
411         spin_lock(&clp->cl_lock);
412         delegation = nfs_detach_delegation_locked(nfsi, args->stateid);
413         if (delegation != NULL)
414                 args->result = 0;
415         else
416                 args->result = -ENOENT;
417         spin_unlock(&clp->cl_lock);
418         complete(&args->started);
419         nfs_delegation_claim_opens(inode, args->stateid);
420         up_write(&nfsi->rwsem);
421         up_read(&clp->cl_sem);
422         nfs_msync_inode(inode);
423
424         if (delegation != NULL)
425                 nfs_do_return_delegation(inode, delegation, 1);
426         iput(inode);
427         module_put_and_exit(0);
428 }
429
430 /*
431  * Asynchronous delegation recall!
432  */
433 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid)
434 {
435         struct recall_threadargs data = {
436                 .inode = inode,
437                 .stateid = stateid,
438         };
439         int status;
440
441         init_completion(&data.started);
442         __module_get(THIS_MODULE);
443         status = kernel_thread(recall_thread, &data, CLONE_KERNEL);
444         if (status < 0)
445                 goto out_module_put;
446         wait_for_completion(&data.started);
447         return data.result;
448 out_module_put:
449         module_put(THIS_MODULE);
450         return status;
451 }
452
453 /*
454  * Retrieve the inode associated with a delegation
455  */
456 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
457 {
458         struct nfs_delegation *delegation;
459         struct inode *res = NULL;
460         rcu_read_lock();
461         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
462                 if (nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
463                         res = igrab(delegation->inode);
464                         break;
465                 }
466         }
467         rcu_read_unlock();
468         return res;
469 }
470
471 /*
472  * Mark all delegations as needing to be reclaimed
473  */
474 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
475 {
476         struct nfs_delegation *delegation;
477         rcu_read_lock();
478         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
479                 delegation->flags |= NFS_DELEGATION_NEED_RECLAIM;
480         rcu_read_unlock();
481 }
482
483 /*
484  * Reap all unclaimed delegations after reboot recovery is done
485  */
486 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
487 {
488         struct nfs_delegation *delegation;
489 restart:
490         rcu_read_lock();
491         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
492                 if ((delegation->flags & NFS_DELEGATION_NEED_RECLAIM) == 0)
493                         continue;
494                 spin_lock(&clp->cl_lock);
495                 delegation = nfs_detach_delegation_locked(NFS_I(delegation->inode), NULL);
496                 spin_unlock(&clp->cl_lock);
497                 rcu_read_unlock();
498                 if (delegation != NULL)
499                         nfs_free_delegation(delegation);
500                 goto restart;
501         }
502         rcu_read_unlock();
503 }
504
505 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
506 {
507         struct nfs_inode *nfsi = NFS_I(inode);
508         struct nfs_delegation *delegation;
509         int ret = 0;
510
511         rcu_read_lock();
512         delegation = rcu_dereference(nfsi->delegation);
513         if (delegation != NULL) {
514                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
515                 ret = 1;
516         }
517         rcu_read_unlock();
518         return ret;
519 }