xref: /linux/fs/nfs/delegation.c (revision c9933d494c54f72290831191c09bb8488bfd5905)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * linux/fs/nfs/delegation.c
4  *
5  * Copyright (C) 2004 Trond Myklebust
6  *
7  * NFS file delegation management
8  *
9  */
10 #include <linux/completion.h>
11 #include <linux/kthread.h>
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <linux/spinlock.h>
16 #include <linux/iversion.h>
17 
18 #include <linux/nfs4.h>
19 #include <linux/nfs_fs.h>
20 #include <linux/nfs_xdr.h>
21 
22 #include "nfs4_fs.h"
23 #include "nfs4session.h"
24 #include "delegation.h"
25 #include "internal.h"
26 #include "nfs4trace.h"
27 
28 #define NFS_DEFAULT_DELEGATION_WATERMARK (5000U)
29 
30 static atomic_long_t nfs_active_delegations;
31 static unsigned nfs_delegation_watermark = NFS_DEFAULT_DELEGATION_WATERMARK;
32 
33 static void __nfs_free_delegation(struct nfs_delegation *delegation)
34 {
35 	put_cred(delegation->cred);
36 	delegation->cred = NULL;
37 	kfree_rcu(delegation, rcu);
38 }
39 
40 static void nfs_mark_delegation_revoked(struct nfs_delegation *delegation)
41 {
42 	if (!test_and_set_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
43 		delegation->stateid.type = NFS4_INVALID_STATEID_TYPE;
44 		atomic_long_dec(&nfs_active_delegations);
45 		if (!test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
46 			nfs_clear_verifier_delegated(delegation->inode);
47 	}
48 }
49 
50 static struct nfs_delegation *nfs_get_delegation(struct nfs_delegation *delegation)
51 {
52 	refcount_inc(&delegation->refcount);
53 	return delegation;
54 }
55 
56 static void nfs_put_delegation(struct nfs_delegation *delegation)
57 {
58 	if (refcount_dec_and_test(&delegation->refcount))
59 		__nfs_free_delegation(delegation);
60 }
61 
62 static void nfs_free_delegation(struct nfs_delegation *delegation)
63 {
64 	nfs_mark_delegation_revoked(delegation);
65 	nfs_put_delegation(delegation);
66 }
67 
68 /**
69  * nfs_mark_delegation_referenced - set delegation's REFERENCED flag
70  * @delegation: delegation to process
71  *
72  */
73 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
74 {
75 	set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
76 }
77 
78 static void nfs_mark_return_delegation(struct nfs_server *server,
79 				       struct nfs_delegation *delegation)
80 {
81 	set_bit(NFS_DELEGATION_RETURN, &delegation->flags);
82 	set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
83 }
84 
85 static bool
86 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
87 		fmode_t flags)
88 {
89 	if (delegation != NULL && (delegation->type & flags) == flags &&
90 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
91 	    !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
92 		return true;
93 	return false;
94 }
95 
96 struct nfs_delegation *nfs4_get_valid_delegation(const struct inode *inode)
97 {
98 	struct nfs_delegation *delegation;
99 
100 	delegation = rcu_dereference(NFS_I(inode)->delegation);
101 	if (nfs4_is_valid_delegation(delegation, 0))
102 		return delegation;
103 	return NULL;
104 }
105 
106 static int
107 nfs4_do_check_delegation(struct inode *inode, fmode_t flags, bool mark)
108 {
109 	struct nfs_delegation *delegation;
110 	int ret = 0;
111 
112 	flags &= FMODE_READ|FMODE_WRITE;
113 	rcu_read_lock();
114 	delegation = rcu_dereference(NFS_I(inode)->delegation);
115 	if (nfs4_is_valid_delegation(delegation, flags)) {
116 		if (mark)
117 			nfs_mark_delegation_referenced(delegation);
118 		ret = 1;
119 	}
120 	rcu_read_unlock();
121 	return ret;
122 }
123 /**
124  * nfs4_have_delegation - check if inode has a delegation, mark it
125  * NFS_DELEGATION_REFERENCED if there is one.
126  * @inode: inode to check
127  * @flags: delegation types to check for
128  *
129  * Returns one if inode has the indicated delegation, otherwise zero.
130  */
131 int nfs4_have_delegation(struct inode *inode, fmode_t flags)
132 {
133 	return nfs4_do_check_delegation(inode, flags, true);
134 }
135 
136 /*
137  * nfs4_check_delegation - check if inode has a delegation, do not mark
138  * NFS_DELEGATION_REFERENCED if it has one.
139  */
140 int nfs4_check_delegation(struct inode *inode, fmode_t flags)
141 {
142 	return nfs4_do_check_delegation(inode, flags, false);
143 }
144 
145 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
146 {
147 	struct inode *inode = state->inode;
148 	struct file_lock *fl;
149 	struct file_lock_context *flctx = inode->i_flctx;
150 	struct list_head *list;
151 	int status = 0;
152 
153 	if (flctx == NULL)
154 		goto out;
155 
156 	list = &flctx->flc_posix;
157 	spin_lock(&flctx->flc_lock);
158 restart:
159 	list_for_each_entry(fl, list, fl_list) {
160 		if (nfs_file_open_context(fl->fl_file)->state != state)
161 			continue;
162 		spin_unlock(&flctx->flc_lock);
163 		status = nfs4_lock_delegation_recall(fl, state, stateid);
164 		if (status < 0)
165 			goto out;
166 		spin_lock(&flctx->flc_lock);
167 	}
168 	if (list == &flctx->flc_posix) {
169 		list = &flctx->flc_flock;
170 		goto restart;
171 	}
172 	spin_unlock(&flctx->flc_lock);
173 out:
174 	return status;
175 }
176 
177 static int nfs_delegation_claim_opens(struct inode *inode,
178 		const nfs4_stateid *stateid, fmode_t type)
179 {
180 	struct nfs_inode *nfsi = NFS_I(inode);
181 	struct nfs_open_context *ctx;
182 	struct nfs4_state_owner *sp;
183 	struct nfs4_state *state;
184 	unsigned int seq;
185 	int err;
186 
187 again:
188 	rcu_read_lock();
189 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
190 		state = ctx->state;
191 		if (state == NULL)
192 			continue;
193 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
194 			continue;
195 		if (!nfs4_valid_open_stateid(state))
196 			continue;
197 		if (!nfs4_stateid_match(&state->stateid, stateid))
198 			continue;
199 		if (!get_nfs_open_context(ctx))
200 			continue;
201 		rcu_read_unlock();
202 		sp = state->owner;
203 		/* Block nfs4_proc_unlck */
204 		mutex_lock(&sp->so_delegreturn_mutex);
205 		seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
206 		err = nfs4_open_delegation_recall(ctx, state, stateid);
207 		if (!err)
208 			err = nfs_delegation_claim_locks(state, stateid);
209 		if (!err && read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
210 			err = -EAGAIN;
211 		mutex_unlock(&sp->so_delegreturn_mutex);
212 		put_nfs_open_context(ctx);
213 		if (err != 0)
214 			return err;
215 		goto again;
216 	}
217 	rcu_read_unlock();
218 	return 0;
219 }
220 
221 /**
222  * nfs_inode_reclaim_delegation - process a delegation reclaim request
223  * @inode: inode to process
224  * @cred: credential to use for request
225  * @type: delegation type
226  * @stateid: delegation stateid
227  * @pagemod_limit: write delegation "space_limit"
228  *
229  */
230 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
231 				  fmode_t type,
232 				  const nfs4_stateid *stateid,
233 				  unsigned long pagemod_limit)
234 {
235 	struct nfs_delegation *delegation;
236 	const struct cred *oldcred = NULL;
237 
238 	rcu_read_lock();
239 	delegation = rcu_dereference(NFS_I(inode)->delegation);
240 	if (delegation != NULL) {
241 		spin_lock(&delegation->lock);
242 		if (nfs4_is_valid_delegation(delegation, 0)) {
243 			nfs4_stateid_copy(&delegation->stateid, stateid);
244 			delegation->type = type;
245 			delegation->pagemod_limit = pagemod_limit;
246 			oldcred = delegation->cred;
247 			delegation->cred = get_cred(cred);
248 			clear_bit(NFS_DELEGATION_NEED_RECLAIM,
249 				  &delegation->flags);
250 			spin_unlock(&delegation->lock);
251 			rcu_read_unlock();
252 			put_cred(oldcred);
253 			trace_nfs4_reclaim_delegation(inode, type);
254 			return;
255 		}
256 		/* We appear to have raced with a delegation return. */
257 		spin_unlock(&delegation->lock);
258 	}
259 	rcu_read_unlock();
260 	nfs_inode_set_delegation(inode, cred, type, stateid, pagemod_limit);
261 }
262 
263 static int nfs_do_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
264 {
265 	const struct cred *cred;
266 	int res = 0;
267 
268 	if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
269 		spin_lock(&delegation->lock);
270 		cred = get_cred(delegation->cred);
271 		spin_unlock(&delegation->lock);
272 		res = nfs4_proc_delegreturn(inode, cred,
273 				&delegation->stateid,
274 				issync);
275 		put_cred(cred);
276 	}
277 	return res;
278 }
279 
280 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
281 {
282 	struct inode *inode = NULL;
283 
284 	spin_lock(&delegation->lock);
285 	if (delegation->inode != NULL)
286 		inode = igrab(delegation->inode);
287 	if (!inode)
288 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
289 	spin_unlock(&delegation->lock);
290 	return inode;
291 }
292 
293 static struct nfs_delegation *
294 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
295 {
296 	struct nfs_delegation *ret = NULL;
297 	struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
298 
299 	if (delegation == NULL)
300 		goto out;
301 	spin_lock(&delegation->lock);
302 	if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
303 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
304 		/* Refcount matched in nfs_end_delegation_return() */
305 		ret = nfs_get_delegation(delegation);
306 	}
307 	spin_unlock(&delegation->lock);
308 	if (ret)
309 		nfs_clear_verifier_delegated(&nfsi->vfs_inode);
310 out:
311 	return ret;
312 }
313 
314 static struct nfs_delegation *
315 nfs_start_delegation_return(struct nfs_inode *nfsi)
316 {
317 	struct nfs_delegation *delegation;
318 
319 	rcu_read_lock();
320 	delegation = nfs_start_delegation_return_locked(nfsi);
321 	rcu_read_unlock();
322 	return delegation;
323 }
324 
325 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
326 					struct nfs_client *clp, int err)
327 {
328 
329 	spin_lock(&delegation->lock);
330 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
331 	if (err == -EAGAIN) {
332 		set_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
333 		set_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state);
334 	}
335 	spin_unlock(&delegation->lock);
336 }
337 
338 static struct nfs_delegation *
339 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
340 		struct nfs_delegation *delegation,
341 		struct nfs_client *clp)
342 {
343 	struct nfs_delegation *deleg_cur =
344 		rcu_dereference_protected(nfsi->delegation,
345 				lockdep_is_held(&clp->cl_lock));
346 
347 	if (deleg_cur == NULL || delegation != deleg_cur)
348 		return NULL;
349 
350 	spin_lock(&delegation->lock);
351 	if (!delegation->inode) {
352 		spin_unlock(&delegation->lock);
353 		return NULL;
354 	}
355 	list_del_rcu(&delegation->super_list);
356 	delegation->inode = NULL;
357 	rcu_assign_pointer(nfsi->delegation, NULL);
358 	spin_unlock(&delegation->lock);
359 	return delegation;
360 }
361 
362 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
363 		struct nfs_delegation *delegation,
364 		struct nfs_server *server)
365 {
366 	struct nfs_client *clp = server->nfs_client;
367 
368 	spin_lock(&clp->cl_lock);
369 	delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
370 	spin_unlock(&clp->cl_lock);
371 	return delegation;
372 }
373 
374 static struct nfs_delegation *
375 nfs_inode_detach_delegation(struct inode *inode)
376 {
377 	struct nfs_inode *nfsi = NFS_I(inode);
378 	struct nfs_server *server = NFS_SERVER(inode);
379 	struct nfs_delegation *delegation;
380 
381 	rcu_read_lock();
382 	delegation = rcu_dereference(nfsi->delegation);
383 	if (delegation != NULL)
384 		delegation = nfs_detach_delegation(nfsi, delegation, server);
385 	rcu_read_unlock();
386 	return delegation;
387 }
388 
389 static void
390 nfs_update_delegation_cred(struct nfs_delegation *delegation,
391 		const struct cred *cred)
392 {
393 	const struct cred *old;
394 
395 	if (cred_fscmp(delegation->cred, cred) != 0) {
396 		old = xchg(&delegation->cred, get_cred(cred));
397 		put_cred(old);
398 	}
399 }
400 
401 static void
402 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
403 		const struct nfs_delegation *update)
404 {
405 	if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
406 		delegation->stateid.seqid = update->stateid.seqid;
407 		smp_wmb();
408 		delegation->type = update->type;
409 		delegation->pagemod_limit = update->pagemod_limit;
410 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
411 			delegation->change_attr = update->change_attr;
412 			nfs_update_delegation_cred(delegation, update->cred);
413 			/* smp_mb__before_atomic() is implicit due to xchg() */
414 			clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
415 			atomic_long_inc(&nfs_active_delegations);
416 		}
417 	}
418 }
419 
420 /**
421  * nfs_inode_set_delegation - set up a delegation on an inode
422  * @inode: inode to which delegation applies
423  * @cred: cred to use for subsequent delegation processing
424  * @type: delegation type
425  * @stateid: delegation stateid
426  * @pagemod_limit: write delegation "space_limit"
427  *
428  * Returns zero on success, or a negative errno value.
429  */
430 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
431 				  fmode_t type,
432 				  const nfs4_stateid *stateid,
433 				  unsigned long pagemod_limit)
434 {
435 	struct nfs_server *server = NFS_SERVER(inode);
436 	struct nfs_client *clp = server->nfs_client;
437 	struct nfs_inode *nfsi = NFS_I(inode);
438 	struct nfs_delegation *delegation, *old_delegation;
439 	struct nfs_delegation *freeme = NULL;
440 	int status = 0;
441 
442 	delegation = kmalloc(sizeof(*delegation), GFP_KERNEL_ACCOUNT);
443 	if (delegation == NULL)
444 		return -ENOMEM;
445 	nfs4_stateid_copy(&delegation->stateid, stateid);
446 	refcount_set(&delegation->refcount, 1);
447 	delegation->type = type;
448 	delegation->pagemod_limit = pagemod_limit;
449 	delegation->change_attr = inode_peek_iversion_raw(inode);
450 	delegation->cred = get_cred(cred);
451 	delegation->inode = inode;
452 	delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
453 	spin_lock_init(&delegation->lock);
454 
455 	spin_lock(&clp->cl_lock);
456 	old_delegation = rcu_dereference_protected(nfsi->delegation,
457 					lockdep_is_held(&clp->cl_lock));
458 	if (old_delegation == NULL)
459 		goto add_new;
460 	/* Is this an update of the existing delegation? */
461 	if (nfs4_stateid_match_other(&old_delegation->stateid,
462 				&delegation->stateid)) {
463 		spin_lock(&old_delegation->lock);
464 		nfs_update_inplace_delegation(old_delegation,
465 				delegation);
466 		spin_unlock(&old_delegation->lock);
467 		goto out;
468 	}
469 	if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
470 		/*
471 		 * Deal with broken servers that hand out two
472 		 * delegations for the same file.
473 		 * Allow for upgrades to a WRITE delegation, but
474 		 * nothing else.
475 		 */
476 		dfprintk(FILE, "%s: server %s handed out "
477 				"a duplicate delegation!\n",
478 				__func__, clp->cl_hostname);
479 		if (delegation->type == old_delegation->type ||
480 		    !(delegation->type & FMODE_WRITE)) {
481 			freeme = delegation;
482 			delegation = NULL;
483 			goto out;
484 		}
485 		if (test_and_set_bit(NFS_DELEGATION_RETURNING,
486 					&old_delegation->flags))
487 			goto out;
488 	}
489 	freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
490 	if (freeme == NULL)
491 		goto out;
492 add_new:
493 	/*
494 	 * If we didn't revalidate the change attribute before setting
495 	 * the delegation, then pre-emptively ask for a full attribute
496 	 * cache revalidation.
497 	 */
498 	spin_lock(&inode->i_lock);
499 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
500 		nfs_set_cache_invalid(inode,
501 			NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
502 			NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
503 			NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
504 			NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
505 			NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
506 			NFS_INO_INVALID_XATTR);
507 	spin_unlock(&inode->i_lock);
508 
509 	list_add_tail_rcu(&delegation->super_list, &server->delegations);
510 	rcu_assign_pointer(nfsi->delegation, delegation);
511 	delegation = NULL;
512 
513 	atomic_long_inc(&nfs_active_delegations);
514 
515 	trace_nfs4_set_delegation(inode, type);
516 out:
517 	spin_unlock(&clp->cl_lock);
518 	if (delegation != NULL)
519 		__nfs_free_delegation(delegation);
520 	if (freeme != NULL) {
521 		nfs_do_return_delegation(inode, freeme, 0);
522 		nfs_free_delegation(freeme);
523 	}
524 	return status;
525 }
526 
527 /*
528  * Basic procedure for returning a delegation to the server
529  */
530 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
531 {
532 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
533 	unsigned int mode = O_WRONLY | O_RDWR;
534 	int err = 0;
535 
536 	if (delegation == NULL)
537 		return 0;
538 
539 	if (!issync)
540 		mode |= O_NONBLOCK;
541 	/* Recall of any remaining application leases */
542 	err = break_lease(inode, mode);
543 
544 	while (err == 0) {
545 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
546 			break;
547 		err = nfs_delegation_claim_opens(inode, &delegation->stateid,
548 				delegation->type);
549 		if (!issync || err != -EAGAIN)
550 			break;
551 		/*
552 		 * Guard against state recovery
553 		 */
554 		err = nfs4_wait_clnt_recover(clp);
555 	}
556 
557 	if (err) {
558 		nfs_abort_delegation_return(delegation, clp, err);
559 		goto out;
560 	}
561 
562 	err = nfs_do_return_delegation(inode, delegation, issync);
563 out:
564 	/* Refcount matched in nfs_start_delegation_return_locked() */
565 	nfs_put_delegation(delegation);
566 	return err;
567 }
568 
569 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
570 {
571 	bool ret = false;
572 
573 	if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
574 		ret = true;
575 	else if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags)) {
576 		struct inode *inode;
577 
578 		spin_lock(&delegation->lock);
579 		inode = delegation->inode;
580 		if (inode && list_empty(&NFS_I(inode)->open_files))
581 			ret = true;
582 		spin_unlock(&delegation->lock);
583 	}
584 	if (ret)
585 		clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
586 	if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
587 	    test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
588 	    test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
589 		ret = false;
590 
591 	return ret;
592 }
593 
594 static int nfs_server_return_marked_delegations(struct nfs_server *server,
595 		void __always_unused *data)
596 {
597 	struct nfs_delegation *delegation;
598 	struct nfs_delegation *prev;
599 	struct inode *inode;
600 	struct inode *place_holder = NULL;
601 	struct nfs_delegation *place_holder_deleg = NULL;
602 	int err = 0;
603 
604 restart:
605 	/*
606 	 * To avoid quadratic looping we hold a reference
607 	 * to an inode place_holder.  Each time we restart, we
608 	 * list delegation in the server from the delegations
609 	 * of that inode.
610 	 * prev is an RCU-protected pointer to a delegation which
611 	 * wasn't marked for return and might be a good choice for
612 	 * the next place_holder.
613 	 */
614 	prev = NULL;
615 	delegation = NULL;
616 	rcu_read_lock();
617 	if (place_holder)
618 		delegation = rcu_dereference(NFS_I(place_holder)->delegation);
619 	if (!delegation || delegation != place_holder_deleg)
620 		delegation = list_entry_rcu(server->delegations.next,
621 					    struct nfs_delegation, super_list);
622 	list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
623 		struct inode *to_put = NULL;
624 
625 		if (test_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags))
626 			continue;
627 		if (!nfs_delegation_need_return(delegation)) {
628 			if (nfs4_is_valid_delegation(delegation, 0))
629 				prev = delegation;
630 			continue;
631 		}
632 
633 		if (prev) {
634 			struct inode *tmp = nfs_delegation_grab_inode(prev);
635 			if (tmp) {
636 				to_put = place_holder;
637 				place_holder = tmp;
638 				place_holder_deleg = prev;
639 			}
640 		}
641 
642 		inode = nfs_delegation_grab_inode(delegation);
643 		if (inode == NULL) {
644 			rcu_read_unlock();
645 			iput(to_put);
646 			goto restart;
647 		}
648 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
649 		rcu_read_unlock();
650 
651 		iput(to_put);
652 
653 		err = nfs_end_delegation_return(inode, delegation, 0);
654 		iput(inode);
655 		cond_resched();
656 		if (!err)
657 			goto restart;
658 		set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
659 		goto out;
660 	}
661 	rcu_read_unlock();
662 out:
663 	iput(place_holder);
664 	return err;
665 }
666 
667 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
668 {
669 	struct nfs_delegation *d;
670 	bool ret = false;
671 
672 	list_for_each_entry_rcu (d, &server->delegations, super_list) {
673 		if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
674 			continue;
675 		nfs_mark_return_delegation(server, d);
676 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
677 		ret = true;
678 	}
679 	return ret;
680 }
681 
682 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
683 {
684 	struct nfs_server *server;
685 	bool ret = false;
686 
687 	if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
688 		goto out;
689 	rcu_read_lock();
690 	list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
691 		if (nfs_server_clear_delayed_delegations(server))
692 			ret = true;
693 	}
694 	rcu_read_unlock();
695 out:
696 	return ret;
697 }
698 
699 /**
700  * nfs_client_return_marked_delegations - return previously marked delegations
701  * @clp: nfs_client to process
702  *
703  * Note that this function is designed to be called by the state
704  * manager thread. For this reason, it cannot flush the dirty data,
705  * since that could deadlock in case of a state recovery error.
706  *
707  * Returns zero on success, or a negative errno value.
708  */
709 int nfs_client_return_marked_delegations(struct nfs_client *clp)
710 {
711 	int err = nfs_client_for_each_server(
712 		clp, nfs_server_return_marked_delegations, NULL);
713 	if (err)
714 		return err;
715 	/* If a return was delayed, sleep to prevent hard looping */
716 	if (nfs_client_clear_delayed_delegations(clp))
717 		ssleep(1);
718 	return 0;
719 }
720 
721 /**
722  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
723  * @inode: inode to process
724  *
725  * Does not protect against delegation reclaims, therefore really only safe
726  * to be called from nfs4_clear_inode(). Guaranteed to always free
727  * the delegation structure.
728  */
729 void nfs_inode_evict_delegation(struct inode *inode)
730 {
731 	struct nfs_delegation *delegation;
732 
733 	delegation = nfs_inode_detach_delegation(inode);
734 	if (delegation != NULL) {
735 		set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
736 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
737 		nfs_do_return_delegation(inode, delegation, 1);
738 		nfs_free_delegation(delegation);
739 	}
740 }
741 
742 /**
743  * nfs4_inode_return_delegation - synchronously return a delegation
744  * @inode: inode to process
745  *
746  * This routine will always flush any dirty data to disk on the
747  * assumption that if we need to return the delegation, then
748  * we should stop caching.
749  *
750  * Returns zero on success, or a negative errno value.
751  */
752 int nfs4_inode_return_delegation(struct inode *inode)
753 {
754 	struct nfs_inode *nfsi = NFS_I(inode);
755 	struct nfs_delegation *delegation;
756 
757 	delegation = nfs_start_delegation_return(nfsi);
758 	if (delegation != NULL) {
759 		/* Synchronous recall of any application leases */
760 		break_lease(inode, O_WRONLY | O_RDWR);
761 		if (S_ISREG(inode->i_mode))
762 			nfs_wb_all(inode);
763 		return nfs_end_delegation_return(inode, delegation, 1);
764 	}
765 	return 0;
766 }
767 
768 /**
769  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
770  * @inode: inode to process
771  *
772  * This routine is called on file close in order to determine if the
773  * inode delegation needs to be returned immediately.
774  */
775 void nfs4_inode_return_delegation_on_close(struct inode *inode)
776 {
777 	struct nfs_delegation *delegation;
778 	struct nfs_delegation *ret = NULL;
779 
780 	if (!inode)
781 		return;
782 	rcu_read_lock();
783 	delegation = nfs4_get_valid_delegation(inode);
784 	if (!delegation)
785 		goto out;
786 	if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
787 	    atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
788 		spin_lock(&delegation->lock);
789 		if (delegation->inode &&
790 		    list_empty(&NFS_I(inode)->open_files) &&
791 		    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
792 			clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
793 			/* Refcount matched in nfs_end_delegation_return() */
794 			ret = nfs_get_delegation(delegation);
795 		}
796 		spin_unlock(&delegation->lock);
797 		if (ret)
798 			nfs_clear_verifier_delegated(inode);
799 	}
800 out:
801 	rcu_read_unlock();
802 	nfs_end_delegation_return(inode, ret, 0);
803 }
804 
805 /**
806  * nfs4_inode_make_writeable
807  * @inode: pointer to inode
808  *
809  * Make the inode writeable by returning the delegation if necessary
810  *
811  * Returns zero on success, or a negative errno value.
812  */
813 int nfs4_inode_make_writeable(struct inode *inode)
814 {
815 	struct nfs_delegation *delegation;
816 
817 	rcu_read_lock();
818 	delegation = nfs4_get_valid_delegation(inode);
819 	if (delegation == NULL ||
820 	    (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
821 	     (delegation->type & FMODE_WRITE))) {
822 		rcu_read_unlock();
823 		return 0;
824 	}
825 	rcu_read_unlock();
826 	return nfs4_inode_return_delegation(inode);
827 }
828 
829 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
830 		struct nfs_delegation *delegation)
831 {
832 	set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
833 	set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
834 }
835 
836 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
837 {
838 	struct nfs_delegation *delegation;
839 	bool ret = false;
840 
841 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
842 		nfs_mark_return_delegation(server, delegation);
843 		ret = true;
844 	}
845 	return ret;
846 }
847 
848 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
849 {
850 	struct nfs_server *server;
851 
852 	rcu_read_lock();
853 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
854 		nfs_server_mark_return_all_delegations(server);
855 	rcu_read_unlock();
856 }
857 
858 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
859 {
860 	if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
861 		nfs4_schedule_state_manager(clp);
862 }
863 
864 /**
865  * nfs_expire_all_delegations
866  * @clp: client to process
867  *
868  */
869 void nfs_expire_all_delegations(struct nfs_client *clp)
870 {
871 	nfs_client_mark_return_all_delegations(clp);
872 	nfs_delegation_run_state_manager(clp);
873 }
874 
875 /**
876  * nfs_server_return_all_delegations - return delegations for one superblock
877  * @server: pointer to nfs_server to process
878  *
879  */
880 void nfs_server_return_all_delegations(struct nfs_server *server)
881 {
882 	struct nfs_client *clp = server->nfs_client;
883 	bool need_wait;
884 
885 	if (clp == NULL)
886 		return;
887 
888 	rcu_read_lock();
889 	need_wait = nfs_server_mark_return_all_delegations(server);
890 	rcu_read_unlock();
891 
892 	if (need_wait) {
893 		nfs4_schedule_state_manager(clp);
894 		nfs4_wait_clnt_recover(clp);
895 	}
896 }
897 
898 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
899 						 fmode_t flags)
900 {
901 	struct nfs_delegation *delegation;
902 
903 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
904 		if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
905 			continue;
906 		if (delegation->type & flags)
907 			nfs_mark_return_if_closed_delegation(server, delegation);
908 	}
909 }
910 
911 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
912 							fmode_t flags)
913 {
914 	struct nfs_server *server;
915 
916 	rcu_read_lock();
917 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
918 		nfs_mark_return_unused_delegation_types(server, flags);
919 	rcu_read_unlock();
920 }
921 
922 static void nfs_revoke_delegation(struct inode *inode,
923 		const nfs4_stateid *stateid)
924 {
925 	struct nfs_delegation *delegation;
926 	nfs4_stateid tmp;
927 	bool ret = false;
928 
929 	rcu_read_lock();
930 	delegation = rcu_dereference(NFS_I(inode)->delegation);
931 	if (delegation == NULL)
932 		goto out;
933 	if (stateid == NULL) {
934 		nfs4_stateid_copy(&tmp, &delegation->stateid);
935 		stateid = &tmp;
936 	} else {
937 		if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
938 			goto out;
939 		spin_lock(&delegation->lock);
940 		if (stateid->seqid) {
941 			if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
942 				spin_unlock(&delegation->lock);
943 				goto out;
944 			}
945 			delegation->stateid.seqid = stateid->seqid;
946 		}
947 		spin_unlock(&delegation->lock);
948 	}
949 	nfs_mark_delegation_revoked(delegation);
950 	ret = true;
951 out:
952 	rcu_read_unlock();
953 	if (ret)
954 		nfs_inode_find_state_and_recover(inode, stateid);
955 }
956 
957 void nfs_remove_bad_delegation(struct inode *inode,
958 		const nfs4_stateid *stateid)
959 {
960 	nfs_revoke_delegation(inode, stateid);
961 }
962 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
963 
964 void nfs_delegation_mark_returned(struct inode *inode,
965 		const nfs4_stateid *stateid)
966 {
967 	struct nfs_delegation *delegation;
968 
969 	if (!inode)
970 		return;
971 
972 	rcu_read_lock();
973 	delegation = rcu_dereference(NFS_I(inode)->delegation);
974 	if (!delegation)
975 		goto out_rcu_unlock;
976 
977 	spin_lock(&delegation->lock);
978 	if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
979 		goto out_spin_unlock;
980 	if (stateid->seqid) {
981 		/* If delegation->stateid is newer, dont mark as returned */
982 		if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
983 			goto out_clear_returning;
984 		if (delegation->stateid.seqid != stateid->seqid)
985 			delegation->stateid.seqid = stateid->seqid;
986 	}
987 
988 	nfs_mark_delegation_revoked(delegation);
989 
990 out_clear_returning:
991 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
992 out_spin_unlock:
993 	spin_unlock(&delegation->lock);
994 out_rcu_unlock:
995 	rcu_read_unlock();
996 
997 	nfs_inode_find_state_and_recover(inode, stateid);
998 }
999 
1000 /**
1001  * nfs_expire_unused_delegation_types
1002  * @clp: client to process
1003  * @flags: delegation types to expire
1004  *
1005  */
1006 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1007 {
1008 	nfs_client_mark_return_unused_delegation_types(clp, flags);
1009 	nfs_delegation_run_state_manager(clp);
1010 }
1011 
1012 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1013 {
1014 	struct nfs_delegation *delegation;
1015 
1016 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1017 		if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1018 			continue;
1019 		nfs_mark_return_if_closed_delegation(server, delegation);
1020 	}
1021 }
1022 
1023 /**
1024  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1025  * @clp: nfs_client to process
1026  *
1027  */
1028 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1029 {
1030 	struct nfs_server *server;
1031 
1032 	rcu_read_lock();
1033 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1034 		nfs_mark_return_unreferenced_delegations(server);
1035 	rcu_read_unlock();
1036 
1037 	nfs_delegation_run_state_manager(clp);
1038 }
1039 
1040 /**
1041  * nfs_async_inode_return_delegation - asynchronously return a delegation
1042  * @inode: inode to process
1043  * @stateid: state ID information
1044  *
1045  * Returns zero on success, or a negative errno value.
1046  */
1047 int nfs_async_inode_return_delegation(struct inode *inode,
1048 				      const nfs4_stateid *stateid)
1049 {
1050 	struct nfs_server *server = NFS_SERVER(inode);
1051 	struct nfs_client *clp = server->nfs_client;
1052 	struct nfs_delegation *delegation;
1053 
1054 	rcu_read_lock();
1055 	delegation = nfs4_get_valid_delegation(inode);
1056 	if (delegation == NULL)
1057 		goto out_enoent;
1058 	if (stateid != NULL &&
1059 	    !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1060 		goto out_enoent;
1061 	nfs_mark_return_delegation(server, delegation);
1062 	rcu_read_unlock();
1063 
1064 	/* If there are any application leases or delegations, recall them */
1065 	break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1066 
1067 	nfs_delegation_run_state_manager(clp);
1068 	return 0;
1069 out_enoent:
1070 	rcu_read_unlock();
1071 	return -ENOENT;
1072 }
1073 
1074 static struct inode *
1075 nfs_delegation_find_inode_server(struct nfs_server *server,
1076 				 const struct nfs_fh *fhandle)
1077 {
1078 	struct nfs_delegation *delegation;
1079 	struct super_block *freeme = NULL;
1080 	struct inode *res = NULL;
1081 
1082 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1083 		spin_lock(&delegation->lock);
1084 		if (delegation->inode != NULL &&
1085 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1086 		    nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1087 			if (nfs_sb_active(server->super)) {
1088 				freeme = server->super;
1089 				res = igrab(delegation->inode);
1090 			}
1091 			spin_unlock(&delegation->lock);
1092 			if (res != NULL)
1093 				return res;
1094 			if (freeme) {
1095 				rcu_read_unlock();
1096 				nfs_sb_deactive(freeme);
1097 				rcu_read_lock();
1098 			}
1099 			return ERR_PTR(-EAGAIN);
1100 		}
1101 		spin_unlock(&delegation->lock);
1102 	}
1103 	return ERR_PTR(-ENOENT);
1104 }
1105 
1106 /**
1107  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1108  * @clp: client state handle
1109  * @fhandle: filehandle from a delegation recall
1110  *
1111  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1112  * cannot be found.
1113  */
1114 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1115 					const struct nfs_fh *fhandle)
1116 {
1117 	struct nfs_server *server;
1118 	struct inode *res;
1119 
1120 	rcu_read_lock();
1121 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1122 		res = nfs_delegation_find_inode_server(server, fhandle);
1123 		if (res != ERR_PTR(-ENOENT)) {
1124 			rcu_read_unlock();
1125 			return res;
1126 		}
1127 	}
1128 	rcu_read_unlock();
1129 	return ERR_PTR(-ENOENT);
1130 }
1131 
1132 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1133 {
1134 	struct nfs_delegation *delegation;
1135 
1136 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1137 		/*
1138 		 * If the delegation may have been admin revoked, then we
1139 		 * cannot reclaim it.
1140 		 */
1141 		if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1142 			continue;
1143 		set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1144 	}
1145 }
1146 
1147 /**
1148  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1149  * @clp: nfs_client to process
1150  *
1151  */
1152 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1153 {
1154 	struct nfs_server *server;
1155 
1156 	rcu_read_lock();
1157 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1158 		nfs_delegation_mark_reclaim_server(server);
1159 	rcu_read_unlock();
1160 }
1161 
1162 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1163 		void __always_unused *data)
1164 {
1165 	struct nfs_delegation *delegation;
1166 	struct inode *inode;
1167 restart:
1168 	rcu_read_lock();
1169 restart_locked:
1170 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1171 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1172 					&delegation->flags) ||
1173 		    test_bit(NFS_DELEGATION_RETURNING,
1174 					&delegation->flags) ||
1175 		    test_bit(NFS_DELEGATION_NEED_RECLAIM,
1176 					&delegation->flags) == 0)
1177 			continue;
1178 		inode = nfs_delegation_grab_inode(delegation);
1179 		if (inode == NULL)
1180 			goto restart_locked;
1181 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1182 		rcu_read_unlock();
1183 		if (delegation != NULL) {
1184 			if (nfs_detach_delegation(NFS_I(inode), delegation,
1185 						server) != NULL)
1186 				nfs_free_delegation(delegation);
1187 			/* Match nfs_start_delegation_return_locked */
1188 			nfs_put_delegation(delegation);
1189 		}
1190 		iput(inode);
1191 		cond_resched();
1192 		goto restart;
1193 	}
1194 	rcu_read_unlock();
1195 	return 0;
1196 }
1197 
1198 /**
1199  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1200  * @clp: nfs_client to process
1201  *
1202  */
1203 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1204 {
1205 	nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1206 			NULL);
1207 }
1208 
1209 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1210 {
1211 	return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1212 				BIT(NFS4CLNT_LEASE_EXPIRED) |
1213 				BIT(NFS4CLNT_SESSION_RESET))) != 0;
1214 }
1215 
1216 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1217 	    struct nfs_delegation *delegation)
1218 {
1219 	if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1220 		return;
1221 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1222 	set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1223 	set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1224 }
1225 
1226 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1227 		struct inode *inode)
1228 {
1229 	struct nfs_delegation *delegation;
1230 
1231 	rcu_read_lock();
1232 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1233 	if (delegation)
1234 		nfs_mark_test_expired_delegation(server, delegation);
1235 	rcu_read_unlock();
1236 
1237 }
1238 
1239 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1240 {
1241 	struct nfs_delegation *delegation;
1242 
1243 	list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1244 		nfs_mark_test_expired_delegation(server, delegation);
1245 }
1246 
1247 /**
1248  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1249  * @clp: nfs_client to process
1250  *
1251  * Iterates through all the delegations associated with this server and
1252  * marks them as needing to be checked for validity.
1253  */
1254 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1255 {
1256 	struct nfs_server *server;
1257 
1258 	rcu_read_lock();
1259 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1260 		nfs_delegation_mark_test_expired_server(server);
1261 	rcu_read_unlock();
1262 }
1263 
1264 /**
1265  * nfs_test_expired_all_delegations - test all delegations for a client
1266  * @clp: nfs_client to process
1267  *
1268  * Helper for handling "recallable state revoked" status from server.
1269  */
1270 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1271 {
1272 	nfs_mark_test_expired_all_delegations(clp);
1273 	nfs4_schedule_state_manager(clp);
1274 }
1275 
1276 static void
1277 nfs_delegation_test_free_expired(struct inode *inode,
1278 		nfs4_stateid *stateid,
1279 		const struct cred *cred)
1280 {
1281 	struct nfs_server *server = NFS_SERVER(inode);
1282 	const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1283 	int status;
1284 
1285 	if (!cred)
1286 		return;
1287 	status = ops->test_and_free_expired(server, stateid, cred);
1288 	if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1289 		nfs_remove_bad_delegation(inode, stateid);
1290 }
1291 
1292 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1293 		void __always_unused *data)
1294 {
1295 	struct nfs_delegation *delegation;
1296 	struct inode *inode;
1297 	const struct cred *cred;
1298 	nfs4_stateid stateid;
1299 restart:
1300 	rcu_read_lock();
1301 restart_locked:
1302 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1303 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1304 					&delegation->flags) ||
1305 		    test_bit(NFS_DELEGATION_RETURNING,
1306 					&delegation->flags) ||
1307 		    test_bit(NFS_DELEGATION_TEST_EXPIRED,
1308 					&delegation->flags) == 0)
1309 			continue;
1310 		inode = nfs_delegation_grab_inode(delegation);
1311 		if (inode == NULL)
1312 			goto restart_locked;
1313 		spin_lock(&delegation->lock);
1314 		cred = get_cred_rcu(delegation->cred);
1315 		nfs4_stateid_copy(&stateid, &delegation->stateid);
1316 		spin_unlock(&delegation->lock);
1317 		clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1318 		rcu_read_unlock();
1319 		nfs_delegation_test_free_expired(inode, &stateid, cred);
1320 		put_cred(cred);
1321 		if (!nfs4_server_rebooted(server->nfs_client)) {
1322 			iput(inode);
1323 			cond_resched();
1324 			goto restart;
1325 		}
1326 		nfs_inode_mark_test_expired_delegation(server,inode);
1327 		iput(inode);
1328 		return -EAGAIN;
1329 	}
1330 	rcu_read_unlock();
1331 	return 0;
1332 }
1333 
1334 /**
1335  * nfs_reap_expired_delegations - reap expired delegations
1336  * @clp: nfs_client to process
1337  *
1338  * Iterates through all the delegations associated with this server and
1339  * checks if they have may have been revoked. This function is usually
1340  * expected to be called in cases where the server may have lost its
1341  * lease.
1342  */
1343 void nfs_reap_expired_delegations(struct nfs_client *clp)
1344 {
1345 	nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1346 			NULL);
1347 }
1348 
1349 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1350 		const nfs4_stateid *stateid)
1351 {
1352 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1353 	struct nfs_delegation *delegation;
1354 	bool found = false;
1355 
1356 	rcu_read_lock();
1357 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1358 	if (delegation &&
1359 	    nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1360 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1361 		nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1362 		found = true;
1363 	}
1364 	rcu_read_unlock();
1365 	if (found)
1366 		nfs4_schedule_state_manager(clp);
1367 }
1368 
1369 /**
1370  * nfs_delegations_present - check for existence of delegations
1371  * @clp: client state handle
1372  *
1373  * Returns one if there are any nfs_delegation structures attached
1374  * to this nfs_client.
1375  */
1376 int nfs_delegations_present(struct nfs_client *clp)
1377 {
1378 	struct nfs_server *server;
1379 	int ret = 0;
1380 
1381 	rcu_read_lock();
1382 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1383 		if (!list_empty(&server->delegations)) {
1384 			ret = 1;
1385 			break;
1386 		}
1387 	rcu_read_unlock();
1388 	return ret;
1389 }
1390 
1391 /**
1392  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1393  * @dst: stateid to refresh
1394  * @inode: inode to check
1395  *
1396  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1397  * that matches our delegation stateid. Otherwise "false" is returned.
1398  */
1399 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1400 {
1401 	struct nfs_delegation *delegation;
1402 	bool ret = false;
1403 	if (!inode)
1404 		goto out;
1405 
1406 	rcu_read_lock();
1407 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1408 	if (delegation != NULL &&
1409 	    nfs4_stateid_match_other(dst, &delegation->stateid) &&
1410 	    nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1411 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1412 		dst->seqid = delegation->stateid.seqid;
1413 		ret = true;
1414 	}
1415 	rcu_read_unlock();
1416 out:
1417 	return ret;
1418 }
1419 
1420 /**
1421  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1422  * @inode: inode to check
1423  * @flags: delegation type requirement
1424  * @dst: stateid data structure to fill in
1425  * @cred: optional argument to retrieve credential
1426  *
1427  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1428  * otherwise "false" is returned.
1429  */
1430 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1431 		nfs4_stateid *dst, const struct cred **cred)
1432 {
1433 	struct nfs_inode *nfsi = NFS_I(inode);
1434 	struct nfs_delegation *delegation;
1435 	bool ret = false;
1436 
1437 	flags &= FMODE_READ|FMODE_WRITE;
1438 	rcu_read_lock();
1439 	delegation = rcu_dereference(nfsi->delegation);
1440 	if (!delegation)
1441 		goto out;
1442 	spin_lock(&delegation->lock);
1443 	ret = nfs4_is_valid_delegation(delegation, flags);
1444 	if (ret) {
1445 		nfs4_stateid_copy(dst, &delegation->stateid);
1446 		nfs_mark_delegation_referenced(delegation);
1447 		if (cred)
1448 			*cred = get_cred(delegation->cred);
1449 	}
1450 	spin_unlock(&delegation->lock);
1451 out:
1452 	rcu_read_unlock();
1453 	return ret;
1454 }
1455 
1456 /**
1457  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1458  * @inode: inode to check
1459  *
1460  * This function checks the number of outstanding writes to the file
1461  * against the delegation 'space_limit' field to see if
1462  * the spec requires us to flush the file on close.
1463  */
1464 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1465 {
1466 	struct nfs_inode *nfsi = NFS_I(inode);
1467 	struct nfs_delegation *delegation;
1468 	bool ret = true;
1469 
1470 	rcu_read_lock();
1471 	delegation = rcu_dereference(nfsi->delegation);
1472 	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1473 		goto out;
1474 	if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1475 		ret = false;
1476 out:
1477 	rcu_read_unlock();
1478 	return ret;
1479 }
1480 
1481 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);
1482