xref: /openbmc/linux/fs/nfs/nfs4state.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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/slab.h>
42 #include <linux/smp_lock.h>
43 #include <linux/nfs_fs.h>
44 #include <linux/nfs_idmap.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/workqueue.h>
48 #include <linux/bitops.h>
49 
50 #include "nfs4_fs.h"
51 #include "callback.h"
52 #include "delegation.h"
53 #include "internal.h"
54 
55 #define OPENOWNER_POOL_SIZE	8
56 
57 const nfs4_stateid zero_stateid;
58 
59 static LIST_HEAD(nfs4_clientid_list);
60 
61 static int nfs4_init_client(struct nfs_client *clp, struct rpc_cred *cred)
62 {
63 	int status = nfs4_proc_setclientid(clp, NFS4_CALLBACK,
64 			nfs_callback_tcpport, cred);
65 	if (status == 0)
66 		status = nfs4_proc_setclientid_confirm(clp, cred);
67 	if (status == 0)
68 		nfs4_schedule_state_renewal(clp);
69 	return status;
70 }
71 
72 u32
73 nfs4_alloc_lockowner_id(struct nfs_client *clp)
74 {
75 	return clp->cl_lockowner_id ++;
76 }
77 
78 static struct nfs4_state_owner *
79 nfs4_client_grab_unused(struct nfs_client *clp, struct rpc_cred *cred)
80 {
81 	struct nfs4_state_owner *sp = NULL;
82 
83 	if (!list_empty(&clp->cl_unused)) {
84 		sp = list_entry(clp->cl_unused.next, struct nfs4_state_owner, so_list);
85 		atomic_inc(&sp->so_count);
86 		sp->so_cred = cred;
87 		list_move(&sp->so_list, &clp->cl_state_owners);
88 		clp->cl_nunused--;
89 	}
90 	return sp;
91 }
92 
93 struct rpc_cred *nfs4_get_renew_cred(struct nfs_client *clp)
94 {
95 	struct nfs4_state_owner *sp;
96 	struct rpc_cred *cred = NULL;
97 
98 	list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
99 		if (list_empty(&sp->so_states))
100 			continue;
101 		cred = get_rpccred(sp->so_cred);
102 		break;
103 	}
104 	return cred;
105 }
106 
107 struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
108 {
109 	struct nfs4_state_owner *sp;
110 
111 	if (!list_empty(&clp->cl_state_owners)) {
112 		sp = list_entry(clp->cl_state_owners.next,
113 				struct nfs4_state_owner, so_list);
114 		return get_rpccred(sp->so_cred);
115 	}
116 	return NULL;
117 }
118 
119 static struct nfs4_state_owner *
120 nfs4_find_state_owner(struct nfs_client *clp, struct rpc_cred *cred)
121 {
122 	struct nfs4_state_owner *sp, *res = NULL;
123 
124 	list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
125 		if (sp->so_cred != cred)
126 			continue;
127 		atomic_inc(&sp->so_count);
128 		/* Move to the head of the list */
129 		list_move(&sp->so_list, &clp->cl_state_owners);
130 		res = sp;
131 		break;
132 	}
133 	return res;
134 }
135 
136 /*
137  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
138  * create a new state_owner.
139  *
140  */
141 static struct nfs4_state_owner *
142 nfs4_alloc_state_owner(void)
143 {
144 	struct nfs4_state_owner *sp;
145 
146 	sp = kzalloc(sizeof(*sp),GFP_KERNEL);
147 	if (!sp)
148 		return NULL;
149 	spin_lock_init(&sp->so_lock);
150 	INIT_LIST_HEAD(&sp->so_states);
151 	INIT_LIST_HEAD(&sp->so_delegations);
152 	rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
153 	sp->so_seqid.sequence = &sp->so_sequence;
154 	spin_lock_init(&sp->so_sequence.lock);
155 	INIT_LIST_HEAD(&sp->so_sequence.list);
156 	atomic_set(&sp->so_count, 1);
157 	return sp;
158 }
159 
160 void
161 nfs4_drop_state_owner(struct nfs4_state_owner *sp)
162 {
163 	struct nfs_client *clp = sp->so_client;
164 	spin_lock(&clp->cl_lock);
165 	list_del_init(&sp->so_list);
166 	spin_unlock(&clp->cl_lock);
167 }
168 
169 /*
170  * Note: must be called with clp->cl_sem held in order to prevent races
171  *       with reboot recovery!
172  */
173 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, struct rpc_cred *cred)
174 {
175 	struct nfs_client *clp = server->nfs_client;
176 	struct nfs4_state_owner *sp, *new;
177 
178 	get_rpccred(cred);
179 	new = nfs4_alloc_state_owner();
180 	spin_lock(&clp->cl_lock);
181 	sp = nfs4_find_state_owner(clp, cred);
182 	if (sp == NULL)
183 		sp = nfs4_client_grab_unused(clp, cred);
184 	if (sp == NULL && new != NULL) {
185 		list_add(&new->so_list, &clp->cl_state_owners);
186 		new->so_client = clp;
187 		new->so_id = nfs4_alloc_lockowner_id(clp);
188 		new->so_cred = cred;
189 		sp = new;
190 		new = NULL;
191 	}
192 	spin_unlock(&clp->cl_lock);
193 	kfree(new);
194 	if (sp != NULL)
195 		return sp;
196 	put_rpccred(cred);
197 	return NULL;
198 }
199 
200 /*
201  * Must be called with clp->cl_sem held in order to avoid races
202  * with state recovery...
203  */
204 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
205 {
206 	struct nfs_client *clp = sp->so_client;
207 	struct rpc_cred *cred = sp->so_cred;
208 
209 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
210 		return;
211 	if (clp->cl_nunused >= OPENOWNER_POOL_SIZE)
212 		goto out_free;
213 	if (list_empty(&sp->so_list))
214 		goto out_free;
215 	list_move(&sp->so_list, &clp->cl_unused);
216 	clp->cl_nunused++;
217 	spin_unlock(&clp->cl_lock);
218 	put_rpccred(cred);
219 	cred = NULL;
220 	return;
221 out_free:
222 	list_del(&sp->so_list);
223 	spin_unlock(&clp->cl_lock);
224 	put_rpccred(cred);
225 	kfree(sp);
226 }
227 
228 static struct nfs4_state *
229 nfs4_alloc_open_state(void)
230 {
231 	struct nfs4_state *state;
232 
233 	state = kzalloc(sizeof(*state), GFP_KERNEL);
234 	if (!state)
235 		return NULL;
236 	atomic_set(&state->count, 1);
237 	INIT_LIST_HEAD(&state->lock_states);
238 	spin_lock_init(&state->state_lock);
239 	return state;
240 }
241 
242 void
243 nfs4_state_set_mode_locked(struct nfs4_state *state, mode_t mode)
244 {
245 	if (state->state == mode)
246 		return;
247 	/* NB! List reordering - see the reclaim code for why.  */
248 	if ((mode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
249 		if (mode & FMODE_WRITE)
250 			list_move(&state->open_states, &state->owner->so_states);
251 		else
252 			list_move_tail(&state->open_states, &state->owner->so_states);
253 	}
254 	if (mode == 0)
255 		list_del_init(&state->inode_states);
256 	state->state = mode;
257 }
258 
259 static struct nfs4_state *
260 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
261 {
262 	struct nfs_inode *nfsi = NFS_I(inode);
263 	struct nfs4_state *state;
264 
265 	list_for_each_entry(state, &nfsi->open_states, inode_states) {
266 		/* Is this in the process of being freed? */
267 		if (state->state == 0)
268 			continue;
269 		if (state->owner == owner) {
270 			atomic_inc(&state->count);
271 			return state;
272 		}
273 	}
274 	return NULL;
275 }
276 
277 static void
278 nfs4_free_open_state(struct nfs4_state *state)
279 {
280 	kfree(state);
281 }
282 
283 struct nfs4_state *
284 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
285 {
286 	struct nfs4_state *state, *new;
287 	struct nfs_inode *nfsi = NFS_I(inode);
288 
289 	spin_lock(&inode->i_lock);
290 	state = __nfs4_find_state_byowner(inode, owner);
291 	spin_unlock(&inode->i_lock);
292 	if (state)
293 		goto out;
294 	new = nfs4_alloc_open_state();
295 	spin_lock(&owner->so_lock);
296 	spin_lock(&inode->i_lock);
297 	state = __nfs4_find_state_byowner(inode, owner);
298 	if (state == NULL && new != NULL) {
299 		state = new;
300 		state->owner = owner;
301 		atomic_inc(&owner->so_count);
302 		list_add(&state->inode_states, &nfsi->open_states);
303 		state->inode = igrab(inode);
304 		spin_unlock(&inode->i_lock);
305 		/* Note: The reclaim code dictates that we add stateless
306 		 * and read-only stateids to the end of the list */
307 		list_add_tail(&state->open_states, &owner->so_states);
308 		spin_unlock(&owner->so_lock);
309 	} else {
310 		spin_unlock(&inode->i_lock);
311 		spin_unlock(&owner->so_lock);
312 		if (new)
313 			nfs4_free_open_state(new);
314 	}
315 out:
316 	return state;
317 }
318 
319 /*
320  * Beware! Caller must be holding exactly one
321  * reference to clp->cl_sem!
322  */
323 void nfs4_put_open_state(struct nfs4_state *state)
324 {
325 	struct inode *inode = state->inode;
326 	struct nfs4_state_owner *owner = state->owner;
327 
328 	if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
329 		return;
330 	spin_lock(&inode->i_lock);
331 	if (!list_empty(&state->inode_states))
332 		list_del(&state->inode_states);
333 	list_del(&state->open_states);
334 	spin_unlock(&inode->i_lock);
335 	spin_unlock(&owner->so_lock);
336 	iput(inode);
337 	nfs4_free_open_state(state);
338 	nfs4_put_state_owner(owner);
339 }
340 
341 /*
342  * Close the current file.
343  */
344 void nfs4_close_state(struct nfs4_state *state, mode_t mode)
345 {
346 	struct inode *inode = state->inode;
347 	struct nfs4_state_owner *owner = state->owner;
348 	int oldstate, newstate = 0;
349 
350 	atomic_inc(&owner->so_count);
351 	/* Protect against nfs4_find_state() */
352 	spin_lock(&owner->so_lock);
353 	spin_lock(&inode->i_lock);
354 	switch (mode & (FMODE_READ | FMODE_WRITE)) {
355 		case FMODE_READ:
356 			state->n_rdonly--;
357 			break;
358 		case FMODE_WRITE:
359 			state->n_wronly--;
360 			break;
361 		case FMODE_READ|FMODE_WRITE:
362 			state->n_rdwr--;
363 	}
364 	oldstate = newstate = state->state;
365 	if (state->n_rdwr == 0) {
366 		if (state->n_rdonly == 0)
367 			newstate &= ~FMODE_READ;
368 		if (state->n_wronly == 0)
369 			newstate &= ~FMODE_WRITE;
370 	}
371 	if (test_bit(NFS_DELEGATED_STATE, &state->flags)) {
372 		nfs4_state_set_mode_locked(state, newstate);
373 		oldstate = newstate;
374 	}
375 	spin_unlock(&inode->i_lock);
376 	spin_unlock(&owner->so_lock);
377 
378 	if (oldstate != newstate && nfs4_do_close(inode, state) == 0)
379 		return;
380 	nfs4_put_open_state(state);
381 	nfs4_put_state_owner(owner);
382 }
383 
384 /*
385  * Search the state->lock_states for an existing lock_owner
386  * that is compatible with current->files
387  */
388 static struct nfs4_lock_state *
389 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
390 {
391 	struct nfs4_lock_state *pos;
392 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
393 		if (pos->ls_owner != fl_owner)
394 			continue;
395 		atomic_inc(&pos->ls_count);
396 		return pos;
397 	}
398 	return NULL;
399 }
400 
401 /*
402  * Return a compatible lock_state. If no initialized lock_state structure
403  * exists, return an uninitialized one.
404  *
405  */
406 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
407 {
408 	struct nfs4_lock_state *lsp;
409 	struct nfs_client *clp = state->owner->so_client;
410 
411 	lsp = kzalloc(sizeof(*lsp), GFP_KERNEL);
412 	if (lsp == NULL)
413 		return NULL;
414 	lsp->ls_seqid.sequence = &state->owner->so_sequence;
415 	atomic_set(&lsp->ls_count, 1);
416 	lsp->ls_owner = fl_owner;
417 	spin_lock(&clp->cl_lock);
418 	lsp->ls_id = nfs4_alloc_lockowner_id(clp);
419 	spin_unlock(&clp->cl_lock);
420 	INIT_LIST_HEAD(&lsp->ls_locks);
421 	return lsp;
422 }
423 
424 /*
425  * Return a compatible lock_state. If no initialized lock_state structure
426  * exists, return an uninitialized one.
427  *
428  * The caller must be holding clp->cl_sem
429  */
430 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
431 {
432 	struct nfs4_lock_state *lsp, *new = NULL;
433 
434 	for(;;) {
435 		spin_lock(&state->state_lock);
436 		lsp = __nfs4_find_lock_state(state, owner);
437 		if (lsp != NULL)
438 			break;
439 		if (new != NULL) {
440 			new->ls_state = state;
441 			list_add(&new->ls_locks, &state->lock_states);
442 			set_bit(LK_STATE_IN_USE, &state->flags);
443 			lsp = new;
444 			new = NULL;
445 			break;
446 		}
447 		spin_unlock(&state->state_lock);
448 		new = nfs4_alloc_lock_state(state, owner);
449 		if (new == NULL)
450 			return NULL;
451 	}
452 	spin_unlock(&state->state_lock);
453 	kfree(new);
454 	return lsp;
455 }
456 
457 /*
458  * Release reference to lock_state, and free it if we see that
459  * it is no longer in use
460  */
461 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
462 {
463 	struct nfs4_state *state;
464 
465 	if (lsp == NULL)
466 		return;
467 	state = lsp->ls_state;
468 	if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
469 		return;
470 	list_del(&lsp->ls_locks);
471 	if (list_empty(&state->lock_states))
472 		clear_bit(LK_STATE_IN_USE, &state->flags);
473 	spin_unlock(&state->state_lock);
474 	kfree(lsp);
475 }
476 
477 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
478 {
479 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
480 
481 	dst->fl_u.nfs4_fl.owner = lsp;
482 	atomic_inc(&lsp->ls_count);
483 }
484 
485 static void nfs4_fl_release_lock(struct file_lock *fl)
486 {
487 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
488 }
489 
490 static struct file_lock_operations nfs4_fl_lock_ops = {
491 	.fl_copy_lock = nfs4_fl_copy_lock,
492 	.fl_release_private = nfs4_fl_release_lock,
493 };
494 
495 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
496 {
497 	struct nfs4_lock_state *lsp;
498 
499 	if (fl->fl_ops != NULL)
500 		return 0;
501 	lsp = nfs4_get_lock_state(state, fl->fl_owner);
502 	if (lsp == NULL)
503 		return -ENOMEM;
504 	fl->fl_u.nfs4_fl.owner = lsp;
505 	fl->fl_ops = &nfs4_fl_lock_ops;
506 	return 0;
507 }
508 
509 /*
510  * Byte-range lock aware utility to initialize the stateid of read/write
511  * requests.
512  */
513 void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner)
514 {
515 	struct nfs4_lock_state *lsp;
516 
517 	memcpy(dst, &state->stateid, sizeof(*dst));
518 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
519 		return;
520 
521 	spin_lock(&state->state_lock);
522 	lsp = __nfs4_find_lock_state(state, fl_owner);
523 	if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
524 		memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
525 	spin_unlock(&state->state_lock);
526 	nfs4_put_lock_state(lsp);
527 }
528 
529 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter)
530 {
531 	struct rpc_sequence *sequence = counter->sequence;
532 	struct nfs_seqid *new;
533 
534 	new = kmalloc(sizeof(*new), GFP_KERNEL);
535 	if (new != NULL) {
536 		new->sequence = counter;
537 		spin_lock(&sequence->lock);
538 		list_add_tail(&new->list, &sequence->list);
539 		spin_unlock(&sequence->lock);
540 	}
541 	return new;
542 }
543 
544 void nfs_free_seqid(struct nfs_seqid *seqid)
545 {
546 	struct rpc_sequence *sequence = seqid->sequence->sequence;
547 
548 	spin_lock(&sequence->lock);
549 	list_del(&seqid->list);
550 	spin_unlock(&sequence->lock);
551 	rpc_wake_up(&sequence->wait);
552 	kfree(seqid);
553 }
554 
555 /*
556  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
557  * failed with a seqid incrementing error -
558  * see comments nfs_fs.h:seqid_mutating_error()
559  */
560 static inline void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
561 {
562 	switch (status) {
563 		case 0:
564 			break;
565 		case -NFS4ERR_BAD_SEQID:
566 		case -NFS4ERR_STALE_CLIENTID:
567 		case -NFS4ERR_STALE_STATEID:
568 		case -NFS4ERR_BAD_STATEID:
569 		case -NFS4ERR_BADXDR:
570 		case -NFS4ERR_RESOURCE:
571 		case -NFS4ERR_NOFILEHANDLE:
572 			/* Non-seqid mutating errors */
573 			return;
574 	};
575 	/*
576 	 * Note: no locking needed as we are guaranteed to be first
577 	 * on the sequence list
578 	 */
579 	seqid->sequence->counter++;
580 }
581 
582 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
583 {
584 	if (status == -NFS4ERR_BAD_SEQID) {
585 		struct nfs4_state_owner *sp = container_of(seqid->sequence,
586 				struct nfs4_state_owner, so_seqid);
587 		nfs4_drop_state_owner(sp);
588 	}
589 	return nfs_increment_seqid(status, seqid);
590 }
591 
592 /*
593  * Increment the seqid if the LOCK/LOCKU succeeded, or
594  * failed with a seqid incrementing error -
595  * see comments nfs_fs.h:seqid_mutating_error()
596  */
597 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
598 {
599 	return nfs_increment_seqid(status, seqid);
600 }
601 
602 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
603 {
604 	struct rpc_sequence *sequence = seqid->sequence->sequence;
605 	int status = 0;
606 
607 	if (sequence->list.next == &seqid->list)
608 		goto out;
609 	spin_lock(&sequence->lock);
610 	if (sequence->list.next != &seqid->list) {
611 		rpc_sleep_on(&sequence->wait, task, NULL, NULL);
612 		status = -EAGAIN;
613 	}
614 	spin_unlock(&sequence->lock);
615 out:
616 	return status;
617 }
618 
619 static int reclaimer(void *);
620 
621 static inline void nfs4_clear_recover_bit(struct nfs_client *clp)
622 {
623 	smp_mb__before_clear_bit();
624 	clear_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state);
625 	smp_mb__after_clear_bit();
626 	wake_up_bit(&clp->cl_state, NFS4CLNT_STATE_RECOVER);
627 	rpc_wake_up(&clp->cl_rpcwaitq);
628 }
629 
630 /*
631  * State recovery routine
632  */
633 static void nfs4_recover_state(struct nfs_client *clp)
634 {
635 	struct task_struct *task;
636 
637 	__module_get(THIS_MODULE);
638 	atomic_inc(&clp->cl_count);
639 	task = kthread_run(reclaimer, clp, "%u.%u.%u.%u-reclaim",
640 			NIPQUAD(clp->cl_addr.sin_addr));
641 	if (!IS_ERR(task))
642 		return;
643 	nfs4_clear_recover_bit(clp);
644 	nfs_put_client(clp);
645 	module_put(THIS_MODULE);
646 }
647 
648 /*
649  * Schedule a state recovery attempt
650  */
651 void nfs4_schedule_state_recovery(struct nfs_client *clp)
652 {
653 	if (!clp)
654 		return;
655 	if (test_and_set_bit(NFS4CLNT_STATE_RECOVER, &clp->cl_state) == 0)
656 		nfs4_recover_state(clp);
657 }
658 
659 static int nfs4_reclaim_locks(struct nfs4_state_recovery_ops *ops, struct nfs4_state *state)
660 {
661 	struct inode *inode = state->inode;
662 	struct file_lock *fl;
663 	int status = 0;
664 
665 	for (fl = inode->i_flock; fl != 0; fl = fl->fl_next) {
666 		if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
667 			continue;
668 		if (((struct nfs_open_context *)fl->fl_file->private_data)->state != state)
669 			continue;
670 		status = ops->recover_lock(state, fl);
671 		if (status >= 0)
672 			continue;
673 		switch (status) {
674 			default:
675 				printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
676 						__FUNCTION__, status);
677 			case -NFS4ERR_EXPIRED:
678 			case -NFS4ERR_NO_GRACE:
679 			case -NFS4ERR_RECLAIM_BAD:
680 			case -NFS4ERR_RECLAIM_CONFLICT:
681 				/* kill_proc(fl->fl_pid, SIGLOST, 1); */
682 				break;
683 			case -NFS4ERR_STALE_CLIENTID:
684 				goto out_err;
685 		}
686 	}
687 	return 0;
688 out_err:
689 	return status;
690 }
691 
692 static int nfs4_reclaim_open_state(struct nfs4_state_recovery_ops *ops, struct nfs4_state_owner *sp)
693 {
694 	struct nfs4_state *state;
695 	struct nfs4_lock_state *lock;
696 	int status = 0;
697 
698 	/* Note: we rely on the sp->so_states list being ordered
699 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
700 	 * states first.
701 	 * This is needed to ensure that the server won't give us any
702 	 * read delegations that we have to return if, say, we are
703 	 * recovering after a network partition or a reboot from a
704 	 * server that doesn't support a grace period.
705 	 */
706 	list_for_each_entry(state, &sp->so_states, open_states) {
707 		if (state->state == 0)
708 			continue;
709 		status = ops->recover_open(sp, state);
710 		if (status >= 0) {
711 			status = nfs4_reclaim_locks(ops, state);
712 			if (status < 0)
713 				goto out_err;
714 			list_for_each_entry(lock, &state->lock_states, ls_locks) {
715 				if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
716 					printk("%s: Lock reclaim failed!\n",
717 							__FUNCTION__);
718 			}
719 			continue;
720 		}
721 		switch (status) {
722 			default:
723 				printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
724 						__FUNCTION__, status);
725 			case -ENOENT:
726 			case -NFS4ERR_RECLAIM_BAD:
727 			case -NFS4ERR_RECLAIM_CONFLICT:
728 				/*
729 				 * Open state on this file cannot be recovered
730 				 * All we can do is revert to using the zero stateid.
731 				 */
732 				memset(state->stateid.data, 0,
733 					sizeof(state->stateid.data));
734 				/* Mark the file as being 'closed' */
735 				state->state = 0;
736 				break;
737 			case -NFS4ERR_EXPIRED:
738 			case -NFS4ERR_NO_GRACE:
739 			case -NFS4ERR_STALE_CLIENTID:
740 				goto out_err;
741 		}
742 	}
743 	return 0;
744 out_err:
745 	return status;
746 }
747 
748 static void nfs4_state_mark_reclaim(struct nfs_client *clp)
749 {
750 	struct nfs4_state_owner *sp;
751 	struct nfs4_state *state;
752 	struct nfs4_lock_state *lock;
753 
754 	/* Reset all sequence ids to zero */
755 	list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
756 		sp->so_seqid.counter = 0;
757 		sp->so_seqid.flags = 0;
758 		spin_lock(&sp->so_lock);
759 		list_for_each_entry(state, &sp->so_states, open_states) {
760 			list_for_each_entry(lock, &state->lock_states, ls_locks) {
761 				lock->ls_seqid.counter = 0;
762 				lock->ls_seqid.flags = 0;
763 				lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
764 			}
765 		}
766 		spin_unlock(&sp->so_lock);
767 	}
768 }
769 
770 static int reclaimer(void *ptr)
771 {
772 	struct nfs_client *clp = ptr;
773 	struct nfs4_state_owner *sp;
774 	struct nfs4_state_recovery_ops *ops;
775 	struct rpc_cred *cred;
776 	int status = 0;
777 
778 	allow_signal(SIGKILL);
779 
780 	/* Ensure exclusive access to NFSv4 state */
781 	lock_kernel();
782 	down_write(&clp->cl_sem);
783 	/* Are there any NFS mounts out there? */
784 	if (list_empty(&clp->cl_superblocks))
785 		goto out;
786 restart_loop:
787 	ops = &nfs4_network_partition_recovery_ops;
788 	/* Are there any open files on this volume? */
789 	cred = nfs4_get_renew_cred(clp);
790 	if (cred != NULL) {
791 		/* Yes there are: try to renew the old lease */
792 		status = nfs4_proc_renew(clp, cred);
793 		switch (status) {
794 			case 0:
795 			case -NFS4ERR_CB_PATH_DOWN:
796 				put_rpccred(cred);
797 				goto out;
798 			case -NFS4ERR_STALE_CLIENTID:
799 			case -NFS4ERR_LEASE_MOVED:
800 				ops = &nfs4_reboot_recovery_ops;
801 		}
802 	} else {
803 		/* "reboot" to ensure we clear all state on the server */
804 		clp->cl_boot_time = CURRENT_TIME;
805 		cred = nfs4_get_setclientid_cred(clp);
806 	}
807 	/* We're going to have to re-establish a clientid */
808 	nfs4_state_mark_reclaim(clp);
809 	status = -ENOENT;
810 	if (cred != NULL) {
811 		status = nfs4_init_client(clp, cred);
812 		put_rpccred(cred);
813 	}
814 	if (status)
815 		goto out_error;
816 	/* Mark all delegations for reclaim */
817 	nfs_delegation_mark_reclaim(clp);
818 	/* Note: list is protected by exclusive lock on cl->cl_sem */
819 	list_for_each_entry(sp, &clp->cl_state_owners, so_list) {
820 		status = nfs4_reclaim_open_state(ops, sp);
821 		if (status < 0) {
822 			if (status == -NFS4ERR_NO_GRACE) {
823 				ops = &nfs4_network_partition_recovery_ops;
824 				status = nfs4_reclaim_open_state(ops, sp);
825 			}
826 			if (status == -NFS4ERR_STALE_CLIENTID)
827 				goto restart_loop;
828 			if (status == -NFS4ERR_EXPIRED)
829 				goto restart_loop;
830 		}
831 	}
832 	nfs_delegation_reap_unclaimed(clp);
833 out:
834 	up_write(&clp->cl_sem);
835 	unlock_kernel();
836 	if (status == -NFS4ERR_CB_PATH_DOWN)
837 		nfs_handle_cb_pathdown(clp);
838 	nfs4_clear_recover_bit(clp);
839 	nfs_put_client(clp);
840 	module_put_and_exit(0);
841 	return 0;
842 out_error:
843 	printk(KERN_WARNING "Error: state recovery failed on NFSv4 server %u.%u.%u.%u with error %d\n",
844 				NIPQUAD(clp->cl_addr.sin_addr), -status);
845 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
846 	goto out;
847 }
848 
849 /*
850  * Local variables:
851  *  c-basic-offset: 8
852  * End:
853  */
854