Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/nfs/nfs4state.c | |
3 | * | |
4 | * Client-side XDR for NFSv4. | |
5 | * | |
6 | * Copyright (c) 2002 The Regents of the University of Michigan. | |
7 | * All rights reserved. | |
8 | * | |
9 | * Kendrick Smith <kmsmith@umich.edu> | |
10 | * | |
11 | * Redistribution and use in source and binary forms, with or without | |
12 | * modification, are permitted provided that the following conditions | |
13 | * are met: | |
14 | * | |
15 | * 1. Redistributions of source code must retain the above copyright | |
16 | * notice, this list of conditions and the following disclaimer. | |
17 | * 2. Redistributions in binary form must reproduce the above copyright | |
18 | * notice, this list of conditions and the following disclaimer in the | |
19 | * documentation and/or other materials provided with the distribution. | |
20 | * 3. Neither the name of the University nor the names of its | |
21 | * contributors may be used to endorse or promote products derived | |
22 | * from this software without specific prior written permission. | |
23 | * | |
24 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED | |
25 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF | |
26 | * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE | |
27 | * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
28 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR | |
29 | * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF | |
30 | * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR | |
31 | * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF | |
32 | * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING | |
33 | * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS | |
34 | * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
35 | * | |
36 | * Implementation of the NFSv4 state model. For the time being, | |
37 | * this is minimal, but will be made much more complex in a | |
38 | * subsequent patch. | |
39 | */ | |
40 | ||
41 | #include <linux/config.h> | |
42 | #include <linux/slab.h> | |
43 | #include <linux/smp_lock.h> | |
44 | #include <linux/nfs_fs.h> | |
45 | #include <linux/nfs_idmap.h> | |
46 | #include <linux/workqueue.h> | |
47 | #include <linux/bitops.h> | |
48 | ||
49 | #include "callback.h" | |
50 | #include "delegation.h" | |
51 | ||
52 | #define OPENOWNER_POOL_SIZE 8 | |
53 | ||
54 | static DEFINE_SPINLOCK(state_spinlock); | |
55 | ||
56 | nfs4_stateid zero_stateid; | |
57 | ||
58 | #if 0 | |
59 | nfs4_stateid one_stateid = | |
60 | { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; | |
61 | #endif | |
62 | ||
63 | static LIST_HEAD(nfs4_clientid_list); | |
64 | ||
65 | static void nfs4_recover_state(void *); | |
66 | extern void nfs4_renew_state(void *); | |
67 | ||
68 | void | |
69 | init_nfsv4_state(struct nfs_server *server) | |
70 | { | |
71 | server->nfs4_state = NULL; | |
72 | INIT_LIST_HEAD(&server->nfs4_siblings); | |
73 | } | |
74 | ||
75 | void | |
76 | destroy_nfsv4_state(struct nfs_server *server) | |
77 | { | |
78 | if (server->mnt_path) { | |
79 | kfree(server->mnt_path); | |
80 | server->mnt_path = NULL; | |
81 | } | |
82 | if (server->nfs4_state) { | |
83 | nfs4_put_client(server->nfs4_state); | |
84 | server->nfs4_state = NULL; | |
85 | } | |
86 | } | |
87 | ||
88 | /* | |
89 | * nfs4_get_client(): returns an empty client structure | |
90 | * nfs4_put_client(): drops reference to client structure | |
91 | * | |
92 | * Since these are allocated/deallocated very rarely, we don't | |
93 | * bother putting them in a slab cache... | |
94 | */ | |
95 | static struct nfs4_client * | |
96 | nfs4_alloc_client(struct in_addr *addr) | |
97 | { | |
98 | struct nfs4_client *clp; | |
99 | ||
100 | if (nfs_callback_up() < 0) | |
101 | return NULL; | |
102 | if ((clp = kmalloc(sizeof(*clp), GFP_KERNEL)) == NULL) { | |
103 | nfs_callback_down(); | |
104 | return NULL; | |
105 | } | |
106 | memset(clp, 0, sizeof(*clp)); | |
107 | memcpy(&clp->cl_addr, addr, sizeof(clp->cl_addr)); | |
108 | init_rwsem(&clp->cl_sem); | |
109 | INIT_LIST_HEAD(&clp->cl_delegations); | |
110 | INIT_LIST_HEAD(&clp->cl_state_owners); | |
111 | INIT_LIST_HEAD(&clp->cl_unused); | |
112 | spin_lock_init(&clp->cl_lock); | |
113 | atomic_set(&clp->cl_count, 1); | |
114 | INIT_WORK(&clp->cl_recoverd, nfs4_recover_state, clp); | |
115 | INIT_WORK(&clp->cl_renewd, nfs4_renew_state, clp); | |
116 | INIT_LIST_HEAD(&clp->cl_superblocks); | |
117 | init_waitqueue_head(&clp->cl_waitq); | |
118 | rpc_init_wait_queue(&clp->cl_rpcwaitq, "NFS4 client"); | |
119 | clp->cl_boot_time = CURRENT_TIME; | |
120 | clp->cl_state = 1 << NFS4CLNT_OK; | |
121 | return clp; | |
122 | } | |
123 | ||
124 | static void | |
125 | nfs4_free_client(struct nfs4_client *clp) | |
126 | { | |
127 | struct nfs4_state_owner *sp; | |
128 | ||
129 | while (!list_empty(&clp->cl_unused)) { | |
130 | sp = list_entry(clp->cl_unused.next, | |
131 | struct nfs4_state_owner, | |
132 | so_list); | |
133 | list_del(&sp->so_list); | |
134 | kfree(sp); | |
135 | } | |
136 | BUG_ON(!list_empty(&clp->cl_state_owners)); | |
137 | if (clp->cl_cred) | |
138 | put_rpccred(clp->cl_cred); | |
139 | nfs_idmap_delete(clp); | |
140 | if (clp->cl_rpcclient) | |
141 | rpc_shutdown_client(clp->cl_rpcclient); | |
142 | kfree(clp); | |
143 | nfs_callback_down(); | |
144 | } | |
145 | ||
146 | static struct nfs4_client *__nfs4_find_client(struct in_addr *addr) | |
147 | { | |
148 | struct nfs4_client *clp; | |
149 | list_for_each_entry(clp, &nfs4_clientid_list, cl_servers) { | |
150 | if (memcmp(&clp->cl_addr, addr, sizeof(clp->cl_addr)) == 0) { | |
151 | atomic_inc(&clp->cl_count); | |
152 | return clp; | |
153 | } | |
154 | } | |
155 | return NULL; | |
156 | } | |
157 | ||
158 | struct nfs4_client *nfs4_find_client(struct in_addr *addr) | |
159 | { | |
160 | struct nfs4_client *clp; | |
161 | spin_lock(&state_spinlock); | |
162 | clp = __nfs4_find_client(addr); | |
163 | spin_unlock(&state_spinlock); | |
164 | return clp; | |
165 | } | |
166 | ||
167 | struct nfs4_client * | |
168 | nfs4_get_client(struct in_addr *addr) | |
169 | { | |
170 | struct nfs4_client *clp, *new = NULL; | |
171 | ||
172 | spin_lock(&state_spinlock); | |
173 | for (;;) { | |
174 | clp = __nfs4_find_client(addr); | |
175 | if (clp != NULL) | |
176 | break; | |
177 | clp = new; | |
178 | if (clp != NULL) { | |
179 | list_add(&clp->cl_servers, &nfs4_clientid_list); | |
180 | new = NULL; | |
181 | break; | |
182 | } | |
183 | spin_unlock(&state_spinlock); | |
184 | new = nfs4_alloc_client(addr); | |
185 | spin_lock(&state_spinlock); | |
186 | if (new == NULL) | |
187 | break; | |
188 | } | |
189 | spin_unlock(&state_spinlock); | |
190 | if (new) | |
191 | nfs4_free_client(new); | |
192 | return clp; | |
193 | } | |
194 | ||
195 | void | |
196 | nfs4_put_client(struct nfs4_client *clp) | |
197 | { | |
198 | if (!atomic_dec_and_lock(&clp->cl_count, &state_spinlock)) | |
199 | return; | |
200 | list_del(&clp->cl_servers); | |
201 | spin_unlock(&state_spinlock); | |
202 | BUG_ON(!list_empty(&clp->cl_superblocks)); | |
203 | wake_up_all(&clp->cl_waitq); | |
204 | rpc_wake_up(&clp->cl_rpcwaitq); | |
205 | nfs4_kill_renewd(clp); | |
206 | nfs4_free_client(clp); | |
207 | } | |
208 | ||
209 | static int __nfs4_init_client(struct nfs4_client *clp) | |
210 | { | |
211 | int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, nfs_callback_tcpport); | |
212 | if (status == 0) | |
213 | status = nfs4_proc_setclientid_confirm(clp); | |
214 | if (status == 0) | |
215 | nfs4_schedule_state_renewal(clp); | |
216 | return status; | |
217 | } | |
218 | ||
219 | int nfs4_init_client(struct nfs4_client *clp) | |
220 | { | |
221 | return nfs4_map_errors(__nfs4_init_client(clp)); | |
222 | } | |
223 | ||
224 | u32 | |
225 | nfs4_alloc_lockowner_id(struct nfs4_client *clp) | |
226 | { | |
227 | return clp->cl_lockowner_id ++; | |
228 | } | |
229 | ||
230 | static struct nfs4_state_owner * | |
231 | nfs4_client_grab_unused(struct nfs4_client *clp, struct rpc_cred *cred) | |
232 | { | |
233 | struct nfs4_state_owner *sp = NULL; | |
234 | ||
235 | if (!list_empty(&clp->cl_unused)) { | |
236 | sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list); | |
237 | atomic_inc(&sp->so_count); | |
238 | sp->so_cred = cred; | |
239 | list_move(&sp->so_list, &clp->cl_state_owners); | |
240 | clp->cl_nunused--; | |
241 | } | |
242 | return sp; | |
243 | } | |
244 | ||
245 | static struct nfs4_state_owner * | |
246 | nfs4_find_state_owner(struct nfs4_client *clp, struct rpc_cred *cred) | |
247 | { | |
248 | struct nfs4_state_owner *sp, *res = NULL; | |
249 | ||
250 | list_for_each_entry(sp, &clp->cl_state_owners, so_list) { | |
251 | if (sp->so_cred != cred) | |
252 | continue; | |
253 | atomic_inc(&sp->so_count); | |
254 | /* Move to the head of the list */ | |
255 | list_move(&sp->so_list, &clp->cl_state_owners); | |
256 | res = sp; | |
257 | break; | |
258 | } | |
259 | return res; | |
260 | } | |
261 | ||
262 | /* | |
263 | * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to | |
264 | * create a new state_owner. | |
265 | * | |
266 | */ | |
267 | static struct nfs4_state_owner * | |
268 | nfs4_alloc_state_owner(void) | |
269 | { | |
270 | struct nfs4_state_owner *sp; | |
271 | ||
272 | sp = kmalloc(sizeof(*sp),GFP_KERNEL); | |
273 | if (!sp) | |
274 | return NULL; | |
275 | init_MUTEX(&sp->so_sema); | |
276 | sp->so_seqid = 0; /* arbitrary */ | |
277 | INIT_LIST_HEAD(&sp->so_states); | |
278 | INIT_LIST_HEAD(&sp->so_delegations); | |
279 | atomic_set(&sp->so_count, 1); | |
280 | return sp; | |
281 | } | |
282 | ||
283 | void | |
284 | nfs4_drop_state_owner(struct nfs4_state_owner *sp) | |
285 | { | |
286 | struct nfs4_client *clp = sp->so_client; | |
287 | spin_lock(&clp->cl_lock); | |
288 | list_del_init(&sp->so_list); | |
289 | spin_unlock(&clp->cl_lock); | |
290 | } | |
291 | ||
292 | /* | |
293 | * Note: must be called with clp->cl_sem held in order to prevent races | |
294 | * with reboot recovery! | |
295 | */ | |
296 | struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred) | |
297 | { | |
298 | struct nfs4_client *clp = server->nfs4_state; | |
299 | struct nfs4_state_owner *sp, *new; | |
300 | ||
301 | get_rpccred(cred); | |
302 | new = nfs4_alloc_state_owner(); | |
303 | spin_lock(&clp->cl_lock); | |
304 | sp = nfs4_find_state_owner(clp, cred); | |
305 | if (sp == NULL) | |
306 | sp = nfs4_client_grab_unused(clp, cred); | |
307 | if (sp == NULL && new != NULL) { | |
308 | list_add(&new->so_list, &clp->cl_state_owners); | |
309 | new->so_client = clp; | |
310 | new->so_id = nfs4_alloc_lockowner_id(clp); | |
311 | new->so_cred = cred; | |
312 | sp = new; | |
313 | new = NULL; | |
314 | } | |
315 | spin_unlock(&clp->cl_lock); | |
316 | if (new) | |
317 | kfree(new); | |
318 | if (sp != NULL) | |
319 | return sp; | |
320 | put_rpccred(cred); | |
321 | return NULL; | |
322 | } | |
323 | ||
324 | /* | |
325 | * Must be called with clp->cl_sem held in order to avoid races | |
326 | * with state recovery... | |
327 | */ | |
328 | void nfs4_put_state_owner(struct nfs4_state_owner *sp) | |
329 | { | |
330 | struct nfs4_client *clp = sp->so_client; | |
331 | struct rpc_cred *cred = sp->so_cred; | |
332 | ||
333 | if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock)) | |
334 | return; | |
335 | if (clp->cl_nunused >= OPENOWNER_POOL_SIZE) | |
336 | goto out_free; | |
337 | if (list_empty(&sp->so_list)) | |
338 | goto out_free; | |
339 | list_move(&sp->so_list, &clp->cl_unused); | |
340 | clp->cl_nunused++; | |
341 | spin_unlock(&clp->cl_lock); | |
342 | put_rpccred(cred); | |
343 | cred = NULL; | |
344 | return; | |
345 | out_free: | |
346 | list_del(&sp->so_list); | |
347 | spin_unlock(&clp->cl_lock); | |
348 | put_rpccred(cred); | |
349 | kfree(sp); | |
350 | } | |
351 | ||
352 | static struct nfs4_state * | |
353 | nfs4_alloc_open_state(void) | |
354 | { | |
355 | struct nfs4_state *state; | |
356 | ||
357 | state = kmalloc(sizeof(*state), GFP_KERNEL); | |
358 | if (!state) | |
359 | return NULL; | |
360 | state->state = 0; | |
361 | state->nreaders = 0; | |
362 | state->nwriters = 0; | |
363 | state->flags = 0; | |
364 | memset(state->stateid.data, 0, sizeof(state->stateid.data)); | |
365 | atomic_set(&state->count, 1); | |
366 | INIT_LIST_HEAD(&state->lock_states); | |
367 | init_MUTEX(&state->lock_sema); | |
368 | rwlock_init(&state->state_lock); | |
369 | return state; | |
370 | } | |
371 | ||
372 | static struct nfs4_state * | |
373 | __nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode) | |
374 | { | |
375 | struct nfs_inode *nfsi = NFS_I(inode); | |
376 | struct nfs4_state *state; | |
377 | ||
378 | mode &= (FMODE_READ|FMODE_WRITE); | |
379 | list_for_each_entry(state, &nfsi->open_states, inode_states) { | |
380 | if (state->owner->so_cred != cred) | |
381 | continue; | |
382 | if ((mode & FMODE_READ) != 0 && state->nreaders == 0) | |
383 | continue; | |
384 | if ((mode & FMODE_WRITE) != 0 && state->nwriters == 0) | |
385 | continue; | |
386 | if ((state->state & mode) != mode) | |
387 | continue; | |
388 | atomic_inc(&state->count); | |
389 | if (mode & FMODE_READ) | |
390 | state->nreaders++; | |
391 | if (mode & FMODE_WRITE) | |
392 | state->nwriters++; | |
393 | return state; | |
394 | } | |
395 | return NULL; | |
396 | } | |
397 | ||
398 | static struct nfs4_state * | |
399 | __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner) | |
400 | { | |
401 | struct nfs_inode *nfsi = NFS_I(inode); | |
402 | struct nfs4_state *state; | |
403 | ||
404 | list_for_each_entry(state, &nfsi->open_states, inode_states) { | |
405 | /* Is this in the process of being freed? */ | |
406 | if (state->nreaders == 0 && state->nwriters == 0) | |
407 | continue; | |
408 | if (state->owner == owner) { | |
409 | atomic_inc(&state->count); | |
410 | return state; | |
411 | } | |
412 | } | |
413 | return NULL; | |
414 | } | |
415 | ||
416 | struct nfs4_state * | |
417 | nfs4_find_state(struct inode *inode, struct rpc_cred *cred, mode_t mode) | |
418 | { | |
419 | struct nfs4_state *state; | |
420 | ||
421 | spin_lock(&inode->i_lock); | |
422 | state = __nfs4_find_state(inode, cred, mode); | |
423 | spin_unlock(&inode->i_lock); | |
424 | return state; | |
425 | } | |
426 | ||
427 | static void | |
428 | nfs4_free_open_state(struct nfs4_state *state) | |
429 | { | |
430 | kfree(state); | |
431 | } | |
432 | ||
433 | struct nfs4_state * | |
434 | nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner) | |
435 | { | |
436 | struct nfs4_state *state, *new; | |
437 | struct nfs_inode *nfsi = NFS_I(inode); | |
438 | ||
439 | spin_lock(&inode->i_lock); | |
440 | state = __nfs4_find_state_byowner(inode, owner); | |
441 | spin_unlock(&inode->i_lock); | |
442 | if (state) | |
443 | goto out; | |
444 | new = nfs4_alloc_open_state(); | |
445 | spin_lock(&inode->i_lock); | |
446 | state = __nfs4_find_state_byowner(inode, owner); | |
447 | if (state == NULL && new != NULL) { | |
448 | state = new; | |
449 | /* Caller *must* be holding owner->so_sem */ | |
450 | /* Note: The reclaim code dictates that we add stateless | |
451 | * and read-only stateids to the end of the list */ | |
452 | list_add_tail(&state->open_states, &owner->so_states); | |
453 | state->owner = owner; | |
454 | atomic_inc(&owner->so_count); | |
455 | list_add(&state->inode_states, &nfsi->open_states); | |
456 | state->inode = igrab(inode); | |
457 | spin_unlock(&inode->i_lock); | |
458 | } else { | |
459 | spin_unlock(&inode->i_lock); | |
460 | if (new) | |
461 | nfs4_free_open_state(new); | |
462 | } | |
463 | out: | |
464 | return state; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Beware! Caller must be holding exactly one | |
469 | * reference to clp->cl_sem and owner->so_sema! | |
470 | */ | |
471 | void nfs4_put_open_state(struct nfs4_state *state) | |
472 | { | |
473 | struct inode *inode = state->inode; | |
474 | struct nfs4_state_owner *owner = state->owner; | |
475 | ||
476 | if (!atomic_dec_and_lock(&state->count, &inode->i_lock)) | |
477 | return; | |
478 | if (!list_empty(&state->inode_states)) | |
479 | list_del(&state->inode_states); | |
480 | spin_unlock(&inode->i_lock); | |
481 | list_del(&state->open_states); | |
482 | iput(inode); | |
483 | BUG_ON (state->state != 0); | |
484 | nfs4_free_open_state(state); | |
485 | nfs4_put_state_owner(owner); | |
486 | } | |
487 | ||
488 | /* | |
489 | * Beware! Caller must be holding no references to clp->cl_sem! | |
490 | * of owner->so_sema! | |
491 | */ | |
492 | void nfs4_close_state(struct nfs4_state *state, mode_t mode) | |
493 | { | |
494 | struct inode *inode = state->inode; | |
495 | struct nfs4_state_owner *owner = state->owner; | |
496 | struct nfs4_client *clp = owner->so_client; | |
497 | int newstate; | |
498 | ||
499 | atomic_inc(&owner->so_count); | |
500 | down_read(&clp->cl_sem); | |
501 | down(&owner->so_sema); | |
502 | /* Protect against nfs4_find_state() */ | |
503 | spin_lock(&inode->i_lock); | |
504 | if (mode & FMODE_READ) | |
505 | state->nreaders--; | |
506 | if (mode & FMODE_WRITE) | |
507 | state->nwriters--; | |
508 | if (state->nwriters == 0) { | |
509 | if (state->nreaders == 0) | |
510 | list_del_init(&state->inode_states); | |
511 | /* See reclaim code */ | |
512 | list_move_tail(&state->open_states, &owner->so_states); | |
513 | } | |
514 | spin_unlock(&inode->i_lock); | |
515 | newstate = 0; | |
516 | if (state->state != 0) { | |
517 | if (state->nreaders) | |
518 | newstate |= FMODE_READ; | |
519 | if (state->nwriters) | |
520 | newstate |= FMODE_WRITE; | |
521 | if (state->state == newstate) | |
522 | goto out; | |
523 | if (nfs4_do_close(inode, state, newstate) == -EINPROGRESS) | |
524 | return; | |
525 | } | |
526 | out: | |
527 | nfs4_put_open_state(state); | |
528 | up(&owner->so_sema); | |
529 | nfs4_put_state_owner(owner); | |
530 | up_read(&clp->cl_sem); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Search the state->lock_states for an existing lock_owner | |
535 | * that is compatible with current->files | |
536 | */ | |
537 | static struct nfs4_lock_state * | |
538 | __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner) | |
539 | { | |
540 | struct nfs4_lock_state *pos; | |
541 | list_for_each_entry(pos, &state->lock_states, ls_locks) { | |
542 | if (pos->ls_owner != fl_owner) | |
543 | continue; | |
544 | atomic_inc(&pos->ls_count); | |
545 | return pos; | |
546 | } | |
547 | return NULL; | |
548 | } | |
549 | ||
550 | struct nfs4_lock_state * | |
551 | nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner) | |
552 | { | |
553 | struct nfs4_lock_state *lsp; | |
554 | read_lock(&state->state_lock); | |
555 | lsp = __nfs4_find_lock_state(state, fl_owner); | |
556 | read_unlock(&state->state_lock); | |
557 | return lsp; | |
558 | } | |
559 | ||
560 | /* | |
561 | * Return a compatible lock_state. If no initialized lock_state structure | |
562 | * exists, return an uninitialized one. | |
563 | * | |
564 | * The caller must be holding state->lock_sema | |
565 | */ | |
566 | static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner) | |
567 | { | |
568 | struct nfs4_lock_state *lsp; | |
569 | struct nfs4_client *clp = state->owner->so_client; | |
570 | ||
571 | lsp = kmalloc(sizeof(*lsp), GFP_KERNEL); | |
572 | if (lsp == NULL) | |
573 | return NULL; | |
574 | lsp->ls_flags = 0; | |
575 | lsp->ls_seqid = 0; /* arbitrary */ | |
576 | lsp->ls_id = -1; | |
577 | memset(lsp->ls_stateid.data, 0, sizeof(lsp->ls_stateid.data)); | |
578 | atomic_set(&lsp->ls_count, 1); | |
579 | lsp->ls_owner = fl_owner; | |
580 | INIT_LIST_HEAD(&lsp->ls_locks); | |
581 | spin_lock(&clp->cl_lock); | |
582 | lsp->ls_id = nfs4_alloc_lockowner_id(clp); | |
583 | spin_unlock(&clp->cl_lock); | |
584 | return lsp; | |
585 | } | |
586 | ||
587 | /* | |
588 | * Return a compatible lock_state. If no initialized lock_state structure | |
589 | * exists, return an uninitialized one. | |
590 | * | |
591 | * The caller must be holding state->lock_sema and clp->cl_sem | |
592 | */ | |
593 | struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner) | |
594 | { | |
595 | struct nfs4_lock_state * lsp; | |
596 | ||
597 | lsp = nfs4_find_lock_state(state, owner); | |
598 | if (lsp == NULL) | |
599 | lsp = nfs4_alloc_lock_state(state, owner); | |
600 | return lsp; | |
601 | } | |
602 | ||
603 | /* | |
604 | * Byte-range lock aware utility to initialize the stateid of read/write | |
605 | * requests. | |
606 | */ | |
607 | void | |
608 | nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner) | |
609 | { | |
610 | if (test_bit(LK_STATE_IN_USE, &state->flags)) { | |
611 | struct nfs4_lock_state *lsp; | |
612 | ||
613 | lsp = nfs4_find_lock_state(state, fl_owner); | |
614 | if (lsp) { | |
615 | memcpy(dst, &lsp->ls_stateid, sizeof(*dst)); | |
616 | nfs4_put_lock_state(lsp); | |
617 | return; | |
618 | } | |
619 | } | |
620 | memcpy(dst, &state->stateid, sizeof(*dst)); | |
621 | } | |
622 | ||
623 | /* | |
624 | * Called with state->lock_sema and clp->cl_sem held. | |
625 | */ | |
626 | void nfs4_increment_lock_seqid(int status, struct nfs4_lock_state *lsp) | |
627 | { | |
628 | if (status == NFS_OK || seqid_mutating_err(-status)) | |
629 | lsp->ls_seqid++; | |
630 | } | |
631 | ||
632 | /* | |
633 | * Check to see if the request lock (type FL_UNLK) effects the fl lock. | |
634 | * | |
635 | * fl and request must have the same posix owner | |
636 | * | |
637 | * return: | |
638 | * 0 -> fl not effected by request | |
639 | * 1 -> fl consumed by request | |
640 | */ | |
641 | ||
642 | static int | |
643 | nfs4_check_unlock(struct file_lock *fl, struct file_lock *request) | |
644 | { | |
645 | if (fl->fl_start >= request->fl_start && fl->fl_end <= request->fl_end) | |
646 | return 1; | |
647 | return 0; | |
648 | } | |
649 | ||
650 | /* | |
651 | * Post an initialized lock_state on the state->lock_states list. | |
652 | */ | |
653 | void nfs4_notify_setlk(struct nfs4_state *state, struct file_lock *request, struct nfs4_lock_state *lsp) | |
654 | { | |
655 | if (!list_empty(&lsp->ls_locks)) | |
656 | return; | |
657 | atomic_inc(&lsp->ls_count); | |
658 | write_lock(&state->state_lock); | |
659 | list_add(&lsp->ls_locks, &state->lock_states); | |
660 | set_bit(LK_STATE_IN_USE, &state->flags); | |
661 | write_unlock(&state->state_lock); | |
662 | } | |
663 | ||
664 | /* | |
665 | * to decide to 'reap' lock state: | |
666 | * 1) search i_flock for file_locks with fl.lock_state = to ls. | |
667 | * 2) determine if unlock will consume found lock. | |
668 | * if so, reap | |
669 | * | |
670 | * else, don't reap. | |
671 | * | |
672 | */ | |
673 | void | |
674 | nfs4_notify_unlck(struct nfs4_state *state, struct file_lock *request, struct nfs4_lock_state *lsp) | |
675 | { | |
676 | struct inode *inode = state->inode; | |
677 | struct file_lock *fl; | |
678 | ||
679 | for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { | |
680 | if (!(fl->fl_flags & FL_POSIX)) | |
681 | continue; | |
682 | if (fl->fl_owner != lsp->ls_owner) | |
683 | continue; | |
684 | /* Exit if we find at least one lock which is not consumed */ | |
685 | if (nfs4_check_unlock(fl,request) == 0) | |
686 | return; | |
687 | } | |
688 | ||
689 | write_lock(&state->state_lock); | |
690 | list_del_init(&lsp->ls_locks); | |
691 | if (list_empty(&state->lock_states)) | |
692 | clear_bit(LK_STATE_IN_USE, &state->flags); | |
693 | write_unlock(&state->state_lock); | |
694 | nfs4_put_lock_state(lsp); | |
695 | } | |
696 | ||
697 | /* | |
698 | * Release reference to lock_state, and free it if we see that | |
699 | * it is no longer in use | |
700 | */ | |
701 | void | |
702 | nfs4_put_lock_state(struct nfs4_lock_state *lsp) | |
703 | { | |
704 | if (!atomic_dec_and_test(&lsp->ls_count)) | |
705 | return; | |
706 | BUG_ON (!list_empty(&lsp->ls_locks)); | |
707 | kfree(lsp); | |
708 | } | |
709 | ||
710 | /* | |
711 | * Called with sp->so_sema and clp->cl_sem held. | |
712 | * | |
713 | * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or | |
714 | * failed with a seqid incrementing error - | |
715 | * see comments nfs_fs.h:seqid_mutating_error() | |
716 | */ | |
717 | void nfs4_increment_seqid(int status, struct nfs4_state_owner *sp) | |
718 | { | |
719 | if (status == NFS_OK || seqid_mutating_err(-status)) | |
720 | sp->so_seqid++; | |
721 | /* If the server returns BAD_SEQID, unhash state_owner here */ | |
722 | if (status == -NFS4ERR_BAD_SEQID) | |
723 | nfs4_drop_state_owner(sp); | |
724 | } | |
725 | ||
726 | static int reclaimer(void *); | |
727 | struct reclaimer_args { | |
728 | struct nfs4_client *clp; | |
729 | struct completion complete; | |
730 | }; | |
731 | ||
732 | /* | |
733 | * State recovery routine | |
734 | */ | |
735 | void | |
736 | nfs4_recover_state(void *data) | |
737 | { | |
738 | struct nfs4_client *clp = (struct nfs4_client *)data; | |
739 | struct reclaimer_args args = { | |
740 | .clp = clp, | |
741 | }; | |
742 | might_sleep(); | |
743 | ||
744 | init_completion(&args.complete); | |
745 | ||
746 | if (kernel_thread(reclaimer, &args, CLONE_KERNEL) < 0) | |
747 | goto out_failed_clear; | |
748 | wait_for_completion(&args.complete); | |
749 | return; | |
750 | out_failed_clear: | |
751 | set_bit(NFS4CLNT_OK, &clp->cl_state); | |
752 | wake_up_all(&clp->cl_waitq); | |
753 | rpc_wake_up(&clp->cl_rpcwaitq); | |
754 | } | |
755 | ||
756 | /* | |
757 | * Schedule a state recovery attempt | |
758 | */ | |
759 | void | |
760 | nfs4_schedule_state_recovery(struct nfs4_client *clp) | |
761 | { | |
762 | if (!clp) | |
763 | return; | |
764 | if (test_and_clear_bit(NFS4CLNT_OK, &clp->cl_state)) | |
765 | schedule_work(&clp->cl_recoverd); | |
766 | } | |
767 | ||
768 | static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state) | |
769 | { | |
770 | struct inode *inode = state->inode; | |
771 | struct file_lock *fl; | |
772 | int status = 0; | |
773 | ||
774 | for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) { | |
775 | if (!(fl->fl_flags & FL_POSIX)) | |
776 | continue; | |
777 | if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state) | |
778 | continue; | |
779 | status = ops->recover_lock(state, fl); | |
780 | if (status >= 0) | |
781 | continue; | |
782 | switch (status) { | |
783 | default: | |
784 | printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n", | |
785 | __FUNCTION__, status); | |
786 | case -NFS4ERR_EXPIRED: | |
787 | case -NFS4ERR_NO_GRACE: | |
788 | case -NFS4ERR_RECLAIM_BAD: | |
789 | case -NFS4ERR_RECLAIM_CONFLICT: | |
790 | /* kill_proc(fl->fl_owner, SIGLOST, 1); */ | |
791 | break; | |
792 | case -NFS4ERR_STALE_CLIENTID: | |
793 | goto out_err; | |
794 | } | |
795 | } | |
796 | return 0; | |
797 | out_err: | |
798 | return status; | |
799 | } | |
800 | ||
801 | static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp) | |
802 | { | |
803 | struct nfs4_state *state; | |
804 | struct nfs4_lock_state *lock; | |
805 | int status = 0; | |
806 | ||
807 | /* Note: we rely on the sp->so_states list being ordered | |
808 | * so that we always reclaim open(O_RDWR) and/or open(O_WRITE) | |
809 | * states first. | |
810 | * This is needed to ensure that the server won't give us any | |
811 | * read delegations that we have to return if, say, we are | |
812 | * recovering after a network partition or a reboot from a | |
813 | * server that doesn't support a grace period. | |
814 | */ | |
815 | list_for_each_entry(state, &sp->so_states, open_states) { | |
816 | if (state->state == 0) | |
817 | continue; | |
818 | status = ops->recover_open(sp, state); | |
819 | list_for_each_entry(lock, &state->lock_states, ls_locks) | |
820 | lock->ls_flags &= ~NFS_LOCK_INITIALIZED; | |
821 | if (status >= 0) { | |
822 | status = nfs4_reclaim_locks(ops, state); | |
823 | if (status < 0) | |
824 | goto out_err; | |
825 | list_for_each_entry(lock, &state->lock_states, ls_locks) { | |
826 | if (!(lock->ls_flags & NFS_LOCK_INITIALIZED)) | |
827 | printk("%s: Lock reclaim failed!\n", | |
828 | __FUNCTION__); | |
829 | } | |
830 | continue; | |
831 | } | |
832 | switch (status) { | |
833 | default: | |
834 | printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n", | |
835 | __FUNCTION__, status); | |
836 | case -ENOENT: | |
837 | case -NFS4ERR_RECLAIM_BAD: | |
838 | case -NFS4ERR_RECLAIM_CONFLICT: | |
839 | /* | |
840 | * Open state on this file cannot be recovered | |
841 | * All we can do is revert to using the zero stateid. | |
842 | */ | |
843 | memset(state->stateid.data, 0, | |
844 | sizeof(state->stateid.data)); | |
845 | /* Mark the file as being 'closed' */ | |
846 | state->state = 0; | |
847 | break; | |
848 | case -NFS4ERR_EXPIRED: | |
849 | case -NFS4ERR_NO_GRACE: | |
850 | case -NFS4ERR_STALE_CLIENTID: | |
851 | goto out_err; | |
852 | } | |
853 | } | |
854 | return 0; | |
855 | out_err: | |
856 | return status; | |
857 | } | |
858 | ||
859 | static int reclaimer(void *ptr) | |
860 | { | |
861 | struct reclaimer_args *args = (struct reclaimer_args *)ptr; | |
862 | struct nfs4_client *clp = args->clp; | |
863 | struct nfs4_state_owner *sp; | |
864 | struct nfs4_state_recovery_ops *ops; | |
865 | int status = 0; | |
866 | ||
867 | daemonize("%u.%u.%u.%u-reclaim", NIPQUAD(clp->cl_addr)); | |
868 | allow_signal(SIGKILL); | |
869 | ||
870 | atomic_inc(&clp->cl_count); | |
871 | complete(&args->complete); | |
872 | ||
873 | /* Ensure exclusive access to NFSv4 state */ | |
874 | lock_kernel(); | |
875 | down_write(&clp->cl_sem); | |
876 | /* Are there any NFS mounts out there? */ | |
877 | if (list_empty(&clp->cl_superblocks)) | |
878 | goto out; | |
879 | restart_loop: | |
880 | status = nfs4_proc_renew(clp); | |
881 | switch (status) { | |
882 | case 0: | |
883 | case -NFS4ERR_CB_PATH_DOWN: | |
884 | goto out; | |
885 | case -NFS4ERR_STALE_CLIENTID: | |
886 | case -NFS4ERR_LEASE_MOVED: | |
887 | ops = &nfs4_reboot_recovery_ops; | |
888 | break; | |
889 | default: | |
890 | ops = &nfs4_network_partition_recovery_ops; | |
891 | }; | |
892 | status = __nfs4_init_client(clp); | |
893 | if (status) | |
894 | goto out_error; | |
895 | /* Mark all delegations for reclaim */ | |
896 | nfs_delegation_mark_reclaim(clp); | |
897 | /* Note: list is protected by exclusive lock on cl->cl_sem */ | |
898 | list_for_each_entry(sp, &clp->cl_state_owners, so_list) { | |
899 | status = nfs4_reclaim_open_state(ops, sp); | |
900 | if (status < 0) { | |
901 | if (status == -NFS4ERR_NO_GRACE) { | |
902 | ops = &nfs4_network_partition_recovery_ops; | |
903 | status = nfs4_reclaim_open_state(ops, sp); | |
904 | } | |
905 | if (status == -NFS4ERR_STALE_CLIENTID) | |
906 | goto restart_loop; | |
907 | if (status == -NFS4ERR_EXPIRED) | |
908 | goto restart_loop; | |
909 | } | |
910 | } | |
911 | nfs_delegation_reap_unclaimed(clp); | |
912 | out: | |
913 | set_bit(NFS4CLNT_OK, &clp->cl_state); | |
914 | up_write(&clp->cl_sem); | |
915 | unlock_kernel(); | |
916 | wake_up_all(&clp->cl_waitq); | |
917 | rpc_wake_up(&clp->cl_rpcwaitq); | |
918 | if (status == -NFS4ERR_CB_PATH_DOWN) | |
919 | nfs_handle_cb_pathdown(clp); | |
920 | nfs4_put_client(clp); | |
921 | return 0; | |
922 | out_error: | |
923 | printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n", | |
924 | NIPQUAD(clp->cl_addr.s_addr), -status); | |
925 | goto out; | |
926 | } | |
927 | ||
928 | /* | |
929 | * Local variables: | |
930 | * c-basic-offset: 8 | |
931 | * End: | |
932 | */ |