NFSv4: Defer inode revalidation when setting up a delegation
[safe/jmp/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_free_delegation(struct nfs_delegation *delegation)
24 {
25         if (delegation->cred)
26                 put_rpccred(delegation->cred);
27         kfree(delegation);
28 }
29
30 static void nfs_free_delegation_callback(struct rcu_head *head)
31 {
32         struct nfs_delegation *delegation = container_of(head, struct nfs_delegation, rcu);
33
34         nfs_free_delegation(delegation);
35 }
36
37 static int nfs_delegation_claim_locks(struct nfs_open_context *ctx, struct nfs4_state *state)
38 {
39         struct inode *inode = state->inode;
40         struct file_lock *fl;
41         int status;
42
43         for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
44                 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
45                         continue;
46                 if ((struct nfs_open_context *)fl->fl_file->private_data != ctx)
47                         continue;
48                 status = nfs4_lock_delegation_recall(state, fl);
49                 if (status >= 0)
50                         continue;
51                 switch (status) {
52                         default:
53                                 printk(KERN_ERR "%s: unhandled error %d.\n",
54                                                 __FUNCTION__, status);
55                         case -NFS4ERR_EXPIRED:
56                                 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
57                         case -NFS4ERR_STALE_CLIENTID:
58                                 nfs4_schedule_state_recovery(NFS_SERVER(inode)->nfs_client);
59                                 goto out_err;
60                 }
61         }
62         return 0;
63 out_err:
64         return status;
65 }
66
67 static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
68 {
69         struct nfs_inode *nfsi = NFS_I(inode);
70         struct nfs_open_context *ctx;
71         struct nfs4_state *state;
72         int err;
73
74 again:
75         spin_lock(&inode->i_lock);
76         list_for_each_entry(ctx, &nfsi->open_files, list) {
77                 state = ctx->state;
78                 if (state == NULL)
79                         continue;
80                 if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
81                         continue;
82                 if (memcmp(state->stateid.data, stateid->data, sizeof(state->stateid.data)) != 0)
83                         continue;
84                 get_nfs_open_context(ctx);
85                 spin_unlock(&inode->i_lock);
86                 err = nfs4_open_delegation_recall(ctx, state, stateid);
87                 if (err >= 0)
88                         err = nfs_delegation_claim_locks(ctx, state);
89                 put_nfs_open_context(ctx);
90                 if (err != 0)
91                         return;
92                 goto again;
93         }
94         spin_unlock(&inode->i_lock);
95 }
96
97 /*
98  * Set up a delegation on an inode
99  */
100 void nfs_inode_reclaim_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
101 {
102         struct nfs_delegation *delegation = NFS_I(inode)->delegation;
103
104         if (delegation == NULL)
105                 return;
106         memcpy(delegation->stateid.data, res->delegation.data,
107                         sizeof(delegation->stateid.data));
108         delegation->type = res->delegation_type;
109         delegation->maxsize = res->maxsize;
110         put_rpccred(cred);
111         delegation->cred = get_rpccred(cred);
112         delegation->flags &= ~NFS_DELEGATION_NEED_RECLAIM;
113         NFS_I(inode)->delegation_state = delegation->type;
114         smp_wmb();
115 }
116
117 /*
118  * Set up a delegation on an inode
119  */
120 int nfs_inode_set_delegation(struct inode *inode, struct rpc_cred *cred, struct nfs_openres *res)
121 {
122         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
123         struct nfs_inode *nfsi = NFS_I(inode);
124         struct nfs_delegation *delegation;
125         int status = 0;
126
127         delegation = kmalloc(sizeof(*delegation), GFP_KERNEL);
128         if (delegation == NULL)
129                 return -ENOMEM;
130         memcpy(delegation->stateid.data, res->delegation.data,
131                         sizeof(delegation->stateid.data));
132         delegation->type = res->delegation_type;
133         delegation->maxsize = res->maxsize;
134         delegation->change_attr = nfsi->change_attr;
135         delegation->cred = get_rpccred(cred);
136         delegation->inode = inode;
137
138         spin_lock(&clp->cl_lock);
139         if (rcu_dereference(nfsi->delegation) == NULL) {
140                 list_add_rcu(&delegation->super_list, &clp->cl_delegations);
141                 nfsi->delegation_state = delegation->type;
142                 rcu_assign_pointer(nfsi->delegation, delegation);
143                 delegation = NULL;
144         } else {
145                 if (memcmp(&delegation->stateid, &nfsi->delegation->stateid,
146                                         sizeof(delegation->stateid)) != 0 ||
147                                 delegation->type != nfsi->delegation->type) {
148                         printk("%s: server %u.%u.%u.%u, handed out a duplicate delegation!\n",
149                                         __FUNCTION__, NIPQUAD(clp->cl_addr.sin_addr));
150                         status = -EIO;
151                 }
152         }
153
154         /* Ensure we revalidate the attributes and page cache! */
155         spin_lock(&inode->i_lock);
156         nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
157         spin_unlock(&inode->i_lock);
158
159         spin_unlock(&clp->cl_lock);
160         kfree(delegation);
161         return status;
162 }
163
164 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
165 {
166         int res = 0;
167
168         res = nfs4_proc_delegreturn(inode, delegation->cred, &delegation->stateid);
169         call_rcu(&delegation->rcu, nfs_free_delegation_callback);
170         return res;
171 }
172
173 /* Sync all data to disk upon delegation return */
174 static void nfs_msync_inode(struct inode *inode)
175 {
176         filemap_fdatawrite(inode->i_mapping);
177         nfs_wb_all(inode);
178         filemap_fdatawait(inode->i_mapping);
179 }
180
181 /*
182  * Basic procedure for returning a delegation to the server
183  */
184 static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
185 {
186         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
187         struct nfs_inode *nfsi = NFS_I(inode);
188
189         nfs_msync_inode(inode);
190         down_read(&clp->cl_sem);
191         /* Guard against new delegated open calls */
192         down_write(&nfsi->rwsem);
193         nfs_delegation_claim_opens(inode, &delegation->stateid);
194         up_write(&nfsi->rwsem);
195         up_read(&clp->cl_sem);
196         nfs_msync_inode(inode);
197
198         return nfs_do_return_delegation(inode, delegation);
199 }
200
201 static struct nfs_delegation *nfs_detach_delegation_locked(struct nfs_inode *nfsi, const nfs4_stateid *stateid)
202 {
203         struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
204
205         if (delegation == NULL)
206                 goto nomatch;
207         if (stateid != NULL && memcmp(delegation->stateid.data, stateid->data,
208                                 sizeof(delegation->stateid.data)) != 0)
209                 goto nomatch;
210         list_del_rcu(&delegation->super_list);
211         nfsi->delegation_state = 0;
212         rcu_assign_pointer(nfsi->delegation, NULL);
213         return delegation;
214 nomatch:
215         return NULL;
216 }
217
218 int nfs_inode_return_delegation(struct inode *inode)
219 {
220         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
221         struct nfs_inode *nfsi = NFS_I(inode);
222         struct nfs_delegation *delegation;
223         int err = 0;
224
225         if (rcu_dereference(nfsi->delegation) != NULL) {
226                 spin_lock(&clp->cl_lock);
227                 delegation = nfs_detach_delegation_locked(nfsi, NULL);
228                 spin_unlock(&clp->cl_lock);
229                 if (delegation != NULL)
230                         err = __nfs_inode_return_delegation(inode, delegation);
231         }
232         return err;
233 }
234
235 /*
236  * Return all delegations associated to a super block
237  */
238 void nfs_return_all_delegations(struct super_block *sb)
239 {
240         struct nfs_client *clp = NFS_SB(sb)->nfs_client;
241         struct nfs_delegation *delegation;
242         struct inode *inode;
243
244         if (clp == NULL)
245                 return;
246 restart:
247         rcu_read_lock();
248         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
249                 if (delegation->inode->i_sb != sb)
250                         continue;
251                 inode = igrab(delegation->inode);
252                 if (inode == NULL)
253                         continue;
254                 spin_lock(&clp->cl_lock);
255                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
256                 spin_unlock(&clp->cl_lock);
257                 rcu_read_unlock();
258                 if (delegation != NULL)
259                         __nfs_inode_return_delegation(inode, delegation);
260                 iput(inode);
261                 goto restart;
262         }
263         rcu_read_unlock();
264 }
265
266 static int nfs_do_expire_all_delegations(void *ptr)
267 {
268         struct nfs_client *clp = ptr;
269         struct nfs_delegation *delegation;
270         struct inode *inode;
271
272         allow_signal(SIGKILL);
273 restart:
274         if (test_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) != 0)
275                 goto out;
276         if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0)
277                 goto out;
278         rcu_read_lock();
279         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
280                 inode = igrab(delegation->inode);
281                 if (inode == NULL)
282                         continue;
283                 spin_lock(&clp->cl_lock);
284                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
285                 spin_unlock(&clp->cl_lock);
286                 rcu_read_unlock();
287                 if (delegation)
288                         __nfs_inode_return_delegation(inode, delegation);
289                 iput(inode);
290                 goto restart;
291         }
292         rcu_read_unlock();
293 out:
294         nfs_put_client(clp);
295         module_put_and_exit(0);
296 }
297
298 void nfs_expire_all_delegations(struct nfs_client *clp)
299 {
300         struct task_struct *task;
301
302         __module_get(THIS_MODULE);
303         atomic_inc(&clp->cl_count);
304         task = kthread_run(nfs_do_expire_all_delegations, clp,
305                         "%u.%u.%u.%u-delegreturn",
306                         NIPQUAD(clp->cl_addr.sin_addr));
307         if (!IS_ERR(task))
308                 return;
309         nfs_put_client(clp);
310         module_put(THIS_MODULE);
311 }
312
313 /*
314  * Return all delegations following an NFS4ERR_CB_PATH_DOWN error.
315  */
316 void nfs_handle_cb_pathdown(struct nfs_client *clp)
317 {
318         struct nfs_delegation *delegation;
319         struct inode *inode;
320
321         if (clp == NULL)
322                 return;
323 restart:
324         rcu_read_lock();
325         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
326                 inode = igrab(delegation->inode);
327                 if (inode == NULL)
328                         continue;
329                 spin_lock(&clp->cl_lock);
330                 delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
331                 spin_unlock(&clp->cl_lock);
332                 rcu_read_unlock();
333                 if (delegation != NULL)
334                         __nfs_inode_return_delegation(inode, delegation);
335                 iput(inode);
336                 goto restart;
337         }
338         rcu_read_unlock();
339 }
340
341 struct recall_threadargs {
342         struct inode *inode;
343         struct nfs_client *clp;
344         const nfs4_stateid *stateid;
345
346         struct completion started;
347         int result;
348 };
349
350 static int recall_thread(void *data)
351 {
352         struct recall_threadargs *args = (struct recall_threadargs *)data;
353         struct inode *inode = igrab(args->inode);
354         struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
355         struct nfs_inode *nfsi = NFS_I(inode);
356         struct nfs_delegation *delegation;
357
358         daemonize("nfsv4-delegreturn");
359
360         nfs_msync_inode(inode);
361         down_read(&clp->cl_sem);
362         down_write(&nfsi->rwsem);
363         spin_lock(&clp->cl_lock);
364         delegation = nfs_detach_delegation_locked(nfsi, args->stateid);
365         if (delegation != NULL)
366                 args->result = 0;
367         else
368                 args->result = -ENOENT;
369         spin_unlock(&clp->cl_lock);
370         complete(&args->started);
371         nfs_delegation_claim_opens(inode, args->stateid);
372         up_write(&nfsi->rwsem);
373         up_read(&clp->cl_sem);
374         nfs_msync_inode(inode);
375
376         if (delegation != NULL)
377                 nfs_do_return_delegation(inode, delegation);
378         iput(inode);
379         module_put_and_exit(0);
380 }
381
382 /*
383  * Asynchronous delegation recall!
384  */
385 int nfs_async_inode_return_delegation(struct inode *inode, const nfs4_stateid *stateid)
386 {
387         struct recall_threadargs data = {
388                 .inode = inode,
389                 .stateid = stateid,
390         };
391         int status;
392
393         init_completion(&data.started);
394         __module_get(THIS_MODULE);
395         status = kernel_thread(recall_thread, &data, CLONE_KERNEL);
396         if (status < 0)
397                 goto out_module_put;
398         wait_for_completion(&data.started);
399         return data.result;
400 out_module_put:
401         module_put(THIS_MODULE);
402         return status;
403 }
404
405 /*
406  * Retrieve the inode associated with a delegation
407  */
408 struct inode *nfs_delegation_find_inode(struct nfs_client *clp, const struct nfs_fh *fhandle)
409 {
410         struct nfs_delegation *delegation;
411         struct inode *res = NULL;
412         rcu_read_lock();
413         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
414                 if (nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
415                         res = igrab(delegation->inode);
416                         break;
417                 }
418         }
419         rcu_read_unlock();
420         return res;
421 }
422
423 /*
424  * Mark all delegations as needing to be reclaimed
425  */
426 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
427 {
428         struct nfs_delegation *delegation;
429         rcu_read_lock();
430         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list)
431                 delegation->flags |= NFS_DELEGATION_NEED_RECLAIM;
432         rcu_read_unlock();
433 }
434
435 /*
436  * Reap all unclaimed delegations after reboot recovery is done
437  */
438 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
439 {
440         struct nfs_delegation *delegation;
441 restart:
442         rcu_read_lock();
443         list_for_each_entry_rcu(delegation, &clp->cl_delegations, super_list) {
444                 if ((delegation->flags & NFS_DELEGATION_NEED_RECLAIM) == 0)
445                         continue;
446                 spin_lock(&clp->cl_lock);
447                 delegation = nfs_detach_delegation_locked(NFS_I(delegation->inode), NULL);
448                 spin_unlock(&clp->cl_lock);
449                 rcu_read_unlock();
450                 if (delegation != NULL)
451                         call_rcu(&delegation->rcu, nfs_free_delegation_callback);
452                 goto restart;
453         }
454         rcu_read_unlock();
455 }
456
457 int nfs4_copy_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
458 {
459         struct nfs_inode *nfsi = NFS_I(inode);
460         struct nfs_delegation *delegation;
461         int ret = 0;
462
463         rcu_read_lock();
464         delegation = rcu_dereference(nfsi->delegation);
465         if (delegation != NULL) {
466                 memcpy(dst->data, delegation->stateid.data, sizeof(dst->data));
467                 ret = 1;
468         }
469         rcu_read_unlock();
470         return ret;
471 }