xref: /linux/fs/nfs/delegation.c (revision a3a02a52bcfcbcc4a637d4b68bf1bc391c9fad02)
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 nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
86 				     fmode_t type)
87 {
88 	if (delegation != NULL && (delegation->type & type) == type &&
89 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
90 	    !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
91 		return true;
92 	return false;
93 }
94 
95 struct nfs_delegation *nfs4_get_valid_delegation(const struct inode *inode)
96 {
97 	struct nfs_delegation *delegation;
98 
99 	delegation = rcu_dereference(NFS_I(inode)->delegation);
100 	if (nfs4_is_valid_delegation(delegation, 0))
101 		return delegation;
102 	return NULL;
103 }
104 
105 static int nfs4_do_check_delegation(struct inode *inode, fmode_t type,
106 				    int flags, bool mark)
107 {
108 	struct nfs_delegation *delegation;
109 	int ret = 0;
110 
111 	type &= FMODE_READ|FMODE_WRITE;
112 	rcu_read_lock();
113 	delegation = rcu_dereference(NFS_I(inode)->delegation);
114 	if (nfs4_is_valid_delegation(delegation, type)) {
115 		if (mark)
116 			nfs_mark_delegation_referenced(delegation);
117 		ret = 1;
118 		if ((flags & NFS_DELEGATION_FLAG_TIME) &&
119 		    !test_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags))
120 			ret = 0;
121 	}
122 	rcu_read_unlock();
123 	return ret;
124 }
125 /**
126  * nfs4_have_delegation - check if inode has a delegation, mark it
127  * NFS_DELEGATION_REFERENCED if there is one.
128  * @inode: inode to check
129  * @type: delegation types to check for
130  * @flags: various modifiers
131  *
132  * Returns one if inode has the indicated delegation, otherwise zero.
133  */
134 int nfs4_have_delegation(struct inode *inode, fmode_t type, int flags)
135 {
136 	return nfs4_do_check_delegation(inode, type, flags, true);
137 }
138 
139 /*
140  * nfs4_check_delegation - check if inode has a delegation, do not mark
141  * NFS_DELEGATION_REFERENCED if it has one.
142  */
143 int nfs4_check_delegation(struct inode *inode, fmode_t type)
144 {
145 	return nfs4_do_check_delegation(inode, type, 0, false);
146 }
147 
148 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
149 {
150 	struct inode *inode = state->inode;
151 	struct file_lock *fl;
152 	struct file_lock_context *flctx = locks_inode_context(inode);
153 	struct list_head *list;
154 	int status = 0;
155 
156 	if (flctx == NULL)
157 		goto out;
158 
159 	list = &flctx->flc_posix;
160 	spin_lock(&flctx->flc_lock);
161 restart:
162 	for_each_file_lock(fl, list) {
163 		if (nfs_file_open_context(fl->c.flc_file)->state != state)
164 			continue;
165 		spin_unlock(&flctx->flc_lock);
166 		status = nfs4_lock_delegation_recall(fl, state, stateid);
167 		if (status < 0)
168 			goto out;
169 		spin_lock(&flctx->flc_lock);
170 	}
171 	if (list == &flctx->flc_posix) {
172 		list = &flctx->flc_flock;
173 		goto restart;
174 	}
175 	spin_unlock(&flctx->flc_lock);
176 out:
177 	return status;
178 }
179 
180 static int nfs_delegation_claim_opens(struct inode *inode,
181 		const nfs4_stateid *stateid, fmode_t type)
182 {
183 	struct nfs_inode *nfsi = NFS_I(inode);
184 	struct nfs_open_context *ctx;
185 	struct nfs4_state_owner *sp;
186 	struct nfs4_state *state;
187 	int err;
188 
189 again:
190 	rcu_read_lock();
191 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
192 		state = ctx->state;
193 		if (state == NULL)
194 			continue;
195 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
196 			continue;
197 		if (!nfs4_valid_open_stateid(state))
198 			continue;
199 		if (!nfs4_stateid_match(&state->stateid, stateid))
200 			continue;
201 		if (!get_nfs_open_context(ctx))
202 			continue;
203 		rcu_read_unlock();
204 		sp = state->owner;
205 		/* Block nfs4_proc_unlck */
206 		mutex_lock(&sp->so_delegreturn_mutex);
207 		err = nfs4_open_delegation_recall(ctx, state, stateid);
208 		if (!err)
209 			err = nfs_delegation_claim_locks(state, stateid);
210 		mutex_unlock(&sp->so_delegreturn_mutex);
211 		put_nfs_open_context(ctx);
212 		if (err != 0)
213 			return err;
214 		goto again;
215 	}
216 	rcu_read_unlock();
217 	return 0;
218 }
219 
220 /**
221  * nfs_inode_reclaim_delegation - process a delegation reclaim request
222  * @inode: inode to process
223  * @cred: credential to use for request
224  * @type: delegation type
225  * @stateid: delegation stateid
226  * @pagemod_limit: write delegation "space_limit"
227  * @deleg_type: raw delegation type
228  *
229  */
230 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
231 				  fmode_t type, const nfs4_stateid *stateid,
232 				  unsigned long pagemod_limit, u32 deleg_type)
233 {
234 	struct nfs_delegation *delegation;
235 	const struct cred *oldcred = NULL;
236 
237 	rcu_read_lock();
238 	delegation = rcu_dereference(NFS_I(inode)->delegation);
239 	if (delegation != NULL) {
240 		spin_lock(&delegation->lock);
241 		nfs4_stateid_copy(&delegation->stateid, stateid);
242 		delegation->type = type;
243 		delegation->pagemod_limit = pagemod_limit;
244 		oldcred = delegation->cred;
245 		delegation->cred = get_cred(cred);
246 		switch (deleg_type) {
247 		case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
248 		case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
249 			set_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
250 			break;
251 		default:
252 			clear_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
253 		}
254 		clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
255 		if (test_and_clear_bit(NFS_DELEGATION_REVOKED,
256 				       &delegation->flags))
257 			atomic_long_inc(&nfs_active_delegations);
258 		spin_unlock(&delegation->lock);
259 		rcu_read_unlock();
260 		put_cred(oldcred);
261 		trace_nfs4_reclaim_delegation(inode, type);
262 	} else {
263 		rcu_read_unlock();
264 		nfs_inode_set_delegation(inode, cred, type, stateid,
265 					 pagemod_limit, deleg_type);
266 	}
267 }
268 
269 static int nfs_do_return_delegation(struct inode *inode,
270 				    struct nfs_delegation *delegation,
271 				    int issync)
272 {
273 	const struct cred *cred;
274 	int res = 0;
275 
276 	if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
277 		spin_lock(&delegation->lock);
278 		cred = get_cred(delegation->cred);
279 		spin_unlock(&delegation->lock);
280 		res = nfs4_proc_delegreturn(inode, cred, &delegation->stateid,
281 					    delegation, issync);
282 		put_cred(cred);
283 	}
284 	return res;
285 }
286 
287 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
288 {
289 	struct inode *inode = NULL;
290 
291 	spin_lock(&delegation->lock);
292 	if (delegation->inode != NULL)
293 		inode = igrab(delegation->inode);
294 	if (!inode)
295 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
296 	spin_unlock(&delegation->lock);
297 	return inode;
298 }
299 
300 static struct nfs_delegation *
301 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
302 {
303 	struct nfs_delegation *ret = NULL;
304 	struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
305 
306 	if (delegation == NULL)
307 		goto out;
308 	spin_lock(&delegation->lock);
309 	if (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
310 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
311 		/* Refcount matched in nfs_end_delegation_return() */
312 		ret = nfs_get_delegation(delegation);
313 	}
314 	spin_unlock(&delegation->lock);
315 	if (ret)
316 		nfs_clear_verifier_delegated(&nfsi->vfs_inode);
317 out:
318 	return ret;
319 }
320 
321 static struct nfs_delegation *
322 nfs_start_delegation_return(struct nfs_inode *nfsi)
323 {
324 	struct nfs_delegation *delegation;
325 
326 	rcu_read_lock();
327 	delegation = nfs_start_delegation_return_locked(nfsi);
328 	rcu_read_unlock();
329 	return delegation;
330 }
331 
332 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
333 					struct nfs_client *clp, int err)
334 {
335 
336 	spin_lock(&delegation->lock);
337 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
338 	if (err == -EAGAIN) {
339 		set_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
340 		set_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state);
341 	}
342 	spin_unlock(&delegation->lock);
343 }
344 
345 static struct nfs_delegation *
346 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
347 		struct nfs_delegation *delegation,
348 		struct nfs_client *clp)
349 {
350 	struct nfs_delegation *deleg_cur =
351 		rcu_dereference_protected(nfsi->delegation,
352 				lockdep_is_held(&clp->cl_lock));
353 
354 	if (deleg_cur == NULL || delegation != deleg_cur)
355 		return NULL;
356 
357 	spin_lock(&delegation->lock);
358 	if (!delegation->inode) {
359 		spin_unlock(&delegation->lock);
360 		return NULL;
361 	}
362 	list_del_rcu(&delegation->super_list);
363 	delegation->inode = NULL;
364 	rcu_assign_pointer(nfsi->delegation, NULL);
365 	spin_unlock(&delegation->lock);
366 	return delegation;
367 }
368 
369 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
370 		struct nfs_delegation *delegation,
371 		struct nfs_server *server)
372 {
373 	struct nfs_client *clp = server->nfs_client;
374 
375 	spin_lock(&clp->cl_lock);
376 	delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
377 	spin_unlock(&clp->cl_lock);
378 	return delegation;
379 }
380 
381 static struct nfs_delegation *
382 nfs_inode_detach_delegation(struct inode *inode)
383 {
384 	struct nfs_inode *nfsi = NFS_I(inode);
385 	struct nfs_server *server = NFS_SERVER(inode);
386 	struct nfs_delegation *delegation;
387 
388 	rcu_read_lock();
389 	delegation = rcu_dereference(nfsi->delegation);
390 	if (delegation != NULL)
391 		delegation = nfs_detach_delegation(nfsi, delegation, server);
392 	rcu_read_unlock();
393 	return delegation;
394 }
395 
396 static void
397 nfs_update_delegation_cred(struct nfs_delegation *delegation,
398 		const struct cred *cred)
399 {
400 	const struct cred *old;
401 
402 	if (cred_fscmp(delegation->cred, cred) != 0) {
403 		old = xchg(&delegation->cred, get_cred(cred));
404 		put_cred(old);
405 	}
406 }
407 
408 static void
409 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
410 		const struct nfs_delegation *update)
411 {
412 	if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
413 		delegation->stateid.seqid = update->stateid.seqid;
414 		smp_wmb();
415 		delegation->type = update->type;
416 		delegation->pagemod_limit = update->pagemod_limit;
417 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
418 			delegation->change_attr = update->change_attr;
419 			nfs_update_delegation_cred(delegation, update->cred);
420 			/* smp_mb__before_atomic() is implicit due to xchg() */
421 			clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
422 			atomic_long_inc(&nfs_active_delegations);
423 		}
424 	}
425 }
426 
427 /**
428  * nfs_inode_set_delegation - set up a delegation on an inode
429  * @inode: inode to which delegation applies
430  * @cred: cred to use for subsequent delegation processing
431  * @type: delegation type
432  * @stateid: delegation stateid
433  * @pagemod_limit: write delegation "space_limit"
434  * @deleg_type: raw delegation type
435  *
436  * Returns zero on success, or a negative errno value.
437  */
438 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
439 			     fmode_t type, const nfs4_stateid *stateid,
440 			     unsigned long pagemod_limit, u32 deleg_type)
441 {
442 	struct nfs_server *server = NFS_SERVER(inode);
443 	struct nfs_client *clp = server->nfs_client;
444 	struct nfs_inode *nfsi = NFS_I(inode);
445 	struct nfs_delegation *delegation, *old_delegation;
446 	struct nfs_delegation *freeme = NULL;
447 	int status = 0;
448 
449 	delegation = kmalloc(sizeof(*delegation), GFP_KERNEL_ACCOUNT);
450 	if (delegation == NULL)
451 		return -ENOMEM;
452 	nfs4_stateid_copy(&delegation->stateid, stateid);
453 	refcount_set(&delegation->refcount, 1);
454 	delegation->type = type;
455 	delegation->pagemod_limit = pagemod_limit;
456 	delegation->change_attr = inode_peek_iversion_raw(inode);
457 	delegation->cred = get_cred(cred);
458 	delegation->inode = inode;
459 	delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
460 	switch (deleg_type) {
461 	case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
462 	case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
463 		delegation->flags |= BIT(NFS_DELEGATION_DELEGTIME);
464 	}
465 	delegation->test_gen = 0;
466 	spin_lock_init(&delegation->lock);
467 
468 	spin_lock(&clp->cl_lock);
469 	old_delegation = rcu_dereference_protected(nfsi->delegation,
470 					lockdep_is_held(&clp->cl_lock));
471 	if (old_delegation == NULL)
472 		goto add_new;
473 	/* Is this an update of the existing delegation? */
474 	if (nfs4_stateid_match_other(&old_delegation->stateid,
475 				&delegation->stateid)) {
476 		spin_lock(&old_delegation->lock);
477 		nfs_update_inplace_delegation(old_delegation,
478 				delegation);
479 		spin_unlock(&old_delegation->lock);
480 		goto out;
481 	}
482 	if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
483 		/*
484 		 * Deal with broken servers that hand out two
485 		 * delegations for the same file.
486 		 * Allow for upgrades to a WRITE delegation, but
487 		 * nothing else.
488 		 */
489 		dfprintk(FILE, "%s: server %s handed out "
490 				"a duplicate delegation!\n",
491 				__func__, clp->cl_hostname);
492 		if (delegation->type == old_delegation->type ||
493 		    !(delegation->type & FMODE_WRITE)) {
494 			freeme = delegation;
495 			delegation = NULL;
496 			goto out;
497 		}
498 		if (test_and_set_bit(NFS_DELEGATION_RETURNING,
499 					&old_delegation->flags))
500 			goto out;
501 	}
502 	freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
503 	if (freeme == NULL)
504 		goto out;
505 add_new:
506 	/*
507 	 * If we didn't revalidate the change attribute before setting
508 	 * the delegation, then pre-emptively ask for a full attribute
509 	 * cache revalidation.
510 	 */
511 	spin_lock(&inode->i_lock);
512 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
513 		nfs_set_cache_invalid(inode,
514 			NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
515 			NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
516 			NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
517 			NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
518 			NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
519 			NFS_INO_INVALID_XATTR);
520 	spin_unlock(&inode->i_lock);
521 
522 	list_add_tail_rcu(&delegation->super_list, &server->delegations);
523 	rcu_assign_pointer(nfsi->delegation, delegation);
524 	delegation = NULL;
525 
526 	atomic_long_inc(&nfs_active_delegations);
527 
528 	trace_nfs4_set_delegation(inode, type);
529 
530 	/* If we hold writebacks and have delegated mtime then update */
531 	if (deleg_type == NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG &&
532 	    nfs_have_writebacks(inode))
533 		nfs_update_delegated_mtime(inode);
534 out:
535 	spin_unlock(&clp->cl_lock);
536 	if (delegation != NULL)
537 		__nfs_free_delegation(delegation);
538 	if (freeme != NULL) {
539 		nfs_do_return_delegation(inode, freeme, 0);
540 		nfs_free_delegation(freeme);
541 	}
542 	return status;
543 }
544 
545 /*
546  * Basic procedure for returning a delegation to the server
547  */
548 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
549 {
550 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
551 	unsigned int mode = O_WRONLY | O_RDWR;
552 	int err = 0;
553 
554 	if (delegation == NULL)
555 		return 0;
556 
557 	if (!issync)
558 		mode |= O_NONBLOCK;
559 	/* Recall of any remaining application leases */
560 	err = break_lease(inode, mode);
561 
562 	while (err == 0) {
563 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
564 			break;
565 		err = nfs_delegation_claim_opens(inode, &delegation->stateid,
566 				delegation->type);
567 		if (!issync || err != -EAGAIN)
568 			break;
569 		/*
570 		 * Guard against state recovery
571 		 */
572 		err = nfs4_wait_clnt_recover(clp);
573 	}
574 
575 	if (err) {
576 		nfs_abort_delegation_return(delegation, clp, err);
577 		goto out;
578 	}
579 
580 	err = nfs_do_return_delegation(inode, delegation, issync);
581 out:
582 	/* Refcount matched in nfs_start_delegation_return_locked() */
583 	nfs_put_delegation(delegation);
584 	return err;
585 }
586 
587 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
588 {
589 	bool ret = false;
590 
591 	if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
592 		ret = true;
593 	else if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags)) {
594 		struct inode *inode;
595 
596 		spin_lock(&delegation->lock);
597 		inode = delegation->inode;
598 		if (inode && list_empty(&NFS_I(inode)->open_files))
599 			ret = true;
600 		spin_unlock(&delegation->lock);
601 	}
602 	if (ret)
603 		clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
604 	if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
605 	    test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
606 	    test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
607 		ret = false;
608 
609 	return ret;
610 }
611 
612 static int nfs_server_return_marked_delegations(struct nfs_server *server,
613 		void __always_unused *data)
614 {
615 	struct nfs_delegation *delegation;
616 	struct nfs_delegation *prev;
617 	struct inode *inode;
618 	struct inode *place_holder = NULL;
619 	struct nfs_delegation *place_holder_deleg = NULL;
620 	int err = 0;
621 
622 restart:
623 	/*
624 	 * To avoid quadratic looping we hold a reference
625 	 * to an inode place_holder.  Each time we restart, we
626 	 * list delegation in the server from the delegations
627 	 * of that inode.
628 	 * prev is an RCU-protected pointer to a delegation which
629 	 * wasn't marked for return and might be a good choice for
630 	 * the next place_holder.
631 	 */
632 	prev = NULL;
633 	delegation = NULL;
634 	rcu_read_lock();
635 	if (place_holder)
636 		delegation = rcu_dereference(NFS_I(place_holder)->delegation);
637 	if (!delegation || delegation != place_holder_deleg)
638 		delegation = list_entry_rcu(server->delegations.next,
639 					    struct nfs_delegation, super_list);
640 	list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
641 		struct inode *to_put = NULL;
642 
643 		if (test_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags))
644 			continue;
645 		if (!nfs_delegation_need_return(delegation)) {
646 			if (nfs4_is_valid_delegation(delegation, 0))
647 				prev = delegation;
648 			continue;
649 		}
650 
651 		if (prev) {
652 			struct inode *tmp = nfs_delegation_grab_inode(prev);
653 			if (tmp) {
654 				to_put = place_holder;
655 				place_holder = tmp;
656 				place_holder_deleg = prev;
657 			}
658 		}
659 
660 		inode = nfs_delegation_grab_inode(delegation);
661 		if (inode == NULL) {
662 			rcu_read_unlock();
663 			iput(to_put);
664 			goto restart;
665 		}
666 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
667 		rcu_read_unlock();
668 
669 		iput(to_put);
670 
671 		err = nfs_end_delegation_return(inode, delegation, 0);
672 		iput(inode);
673 		cond_resched();
674 		if (!err)
675 			goto restart;
676 		set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
677 		goto out;
678 	}
679 	rcu_read_unlock();
680 out:
681 	iput(place_holder);
682 	return err;
683 }
684 
685 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
686 {
687 	struct nfs_delegation *d;
688 	bool ret = false;
689 
690 	list_for_each_entry_rcu (d, &server->delegations, super_list) {
691 		if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
692 			continue;
693 		nfs_mark_return_delegation(server, d);
694 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
695 		ret = true;
696 	}
697 	return ret;
698 }
699 
700 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
701 {
702 	struct nfs_server *server;
703 	bool ret = false;
704 
705 	if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
706 		goto out;
707 	rcu_read_lock();
708 	list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
709 		if (nfs_server_clear_delayed_delegations(server))
710 			ret = true;
711 	}
712 	rcu_read_unlock();
713 out:
714 	return ret;
715 }
716 
717 /**
718  * nfs_client_return_marked_delegations - return previously marked delegations
719  * @clp: nfs_client to process
720  *
721  * Note that this function is designed to be called by the state
722  * manager thread. For this reason, it cannot flush the dirty data,
723  * since that could deadlock in case of a state recovery error.
724  *
725  * Returns zero on success, or a negative errno value.
726  */
727 int nfs_client_return_marked_delegations(struct nfs_client *clp)
728 {
729 	int err = nfs_client_for_each_server(
730 		clp, nfs_server_return_marked_delegations, NULL);
731 	if (err)
732 		return err;
733 	/* If a return was delayed, sleep to prevent hard looping */
734 	if (nfs_client_clear_delayed_delegations(clp))
735 		ssleep(1);
736 	return 0;
737 }
738 
739 /**
740  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
741  * @inode: inode to process
742  *
743  * Does not protect against delegation reclaims, therefore really only safe
744  * to be called from nfs4_clear_inode(). Guaranteed to always free
745  * the delegation structure.
746  */
747 void nfs_inode_evict_delegation(struct inode *inode)
748 {
749 	struct nfs_delegation *delegation;
750 
751 	delegation = nfs_inode_detach_delegation(inode);
752 	if (delegation != NULL) {
753 		set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
754 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
755 		nfs_do_return_delegation(inode, delegation, 1);
756 		nfs_free_delegation(delegation);
757 	}
758 }
759 
760 /**
761  * nfs4_inode_return_delegation - synchronously return a delegation
762  * @inode: inode to process
763  *
764  * This routine will always flush any dirty data to disk on the
765  * assumption that if we need to return the delegation, then
766  * we should stop caching.
767  *
768  * Returns zero on success, or a negative errno value.
769  */
770 int nfs4_inode_return_delegation(struct inode *inode)
771 {
772 	struct nfs_inode *nfsi = NFS_I(inode);
773 	struct nfs_delegation *delegation;
774 
775 	delegation = nfs_start_delegation_return(nfsi);
776 	if (delegation != NULL) {
777 		/* Synchronous recall of any application leases */
778 		break_lease(inode, O_WRONLY | O_RDWR);
779 		if (S_ISREG(inode->i_mode))
780 			nfs_wb_all(inode);
781 		return nfs_end_delegation_return(inode, delegation, 1);
782 	}
783 	return 0;
784 }
785 
786 /**
787  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
788  * @inode: inode to process
789  *
790  * This routine is called on file close in order to determine if the
791  * inode delegation needs to be returned immediately.
792  */
793 void nfs4_inode_return_delegation_on_close(struct inode *inode)
794 {
795 	struct nfs_delegation *delegation;
796 	struct nfs_delegation *ret = NULL;
797 
798 	if (!inode)
799 		return;
800 	rcu_read_lock();
801 	delegation = nfs4_get_valid_delegation(inode);
802 	if (!delegation)
803 		goto out;
804 	if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
805 	    atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
806 		spin_lock(&delegation->lock);
807 		if (delegation->inode &&
808 		    list_empty(&NFS_I(inode)->open_files) &&
809 		    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
810 			clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
811 			/* Refcount matched in nfs_end_delegation_return() */
812 			ret = nfs_get_delegation(delegation);
813 		}
814 		spin_unlock(&delegation->lock);
815 		if (ret)
816 			nfs_clear_verifier_delegated(inode);
817 	}
818 out:
819 	rcu_read_unlock();
820 	nfs_end_delegation_return(inode, ret, 0);
821 }
822 
823 /**
824  * nfs4_inode_make_writeable
825  * @inode: pointer to inode
826  *
827  * Make the inode writeable by returning the delegation if necessary
828  *
829  * Returns zero on success, or a negative errno value.
830  */
831 int nfs4_inode_make_writeable(struct inode *inode)
832 {
833 	struct nfs_delegation *delegation;
834 
835 	rcu_read_lock();
836 	delegation = nfs4_get_valid_delegation(inode);
837 	if (delegation == NULL ||
838 	    (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
839 	     (delegation->type & FMODE_WRITE))) {
840 		rcu_read_unlock();
841 		return 0;
842 	}
843 	rcu_read_unlock();
844 	return nfs4_inode_return_delegation(inode);
845 }
846 
847 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
848 		struct nfs_delegation *delegation)
849 {
850 	set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
851 	set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
852 }
853 
854 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
855 {
856 	struct nfs_delegation *delegation;
857 	bool ret = false;
858 
859 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
860 		nfs_mark_return_delegation(server, delegation);
861 		ret = true;
862 	}
863 	return ret;
864 }
865 
866 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
867 {
868 	struct nfs_server *server;
869 
870 	rcu_read_lock();
871 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
872 		nfs_server_mark_return_all_delegations(server);
873 	rcu_read_unlock();
874 }
875 
876 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
877 {
878 	if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
879 		nfs4_schedule_state_manager(clp);
880 }
881 
882 /**
883  * nfs_expire_all_delegations
884  * @clp: client to process
885  *
886  */
887 void nfs_expire_all_delegations(struct nfs_client *clp)
888 {
889 	nfs_client_mark_return_all_delegations(clp);
890 	nfs_delegation_run_state_manager(clp);
891 }
892 
893 /**
894  * nfs_server_return_all_delegations - return delegations for one superblock
895  * @server: pointer to nfs_server to process
896  *
897  */
898 void nfs_server_return_all_delegations(struct nfs_server *server)
899 {
900 	struct nfs_client *clp = server->nfs_client;
901 	bool need_wait;
902 
903 	if (clp == NULL)
904 		return;
905 
906 	rcu_read_lock();
907 	need_wait = nfs_server_mark_return_all_delegations(server);
908 	rcu_read_unlock();
909 
910 	if (need_wait) {
911 		nfs4_schedule_state_manager(clp);
912 		nfs4_wait_clnt_recover(clp);
913 	}
914 }
915 
916 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
917 						 fmode_t flags)
918 {
919 	struct nfs_delegation *delegation;
920 
921 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
922 		if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
923 			continue;
924 		if (delegation->type & flags)
925 			nfs_mark_return_if_closed_delegation(server, delegation);
926 	}
927 }
928 
929 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
930 							fmode_t flags)
931 {
932 	struct nfs_server *server;
933 
934 	rcu_read_lock();
935 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
936 		nfs_mark_return_unused_delegation_types(server, flags);
937 	rcu_read_unlock();
938 }
939 
940 static void nfs_revoke_delegation(struct inode *inode,
941 		const nfs4_stateid *stateid)
942 {
943 	struct nfs_delegation *delegation;
944 	nfs4_stateid tmp;
945 	bool ret = false;
946 
947 	rcu_read_lock();
948 	delegation = rcu_dereference(NFS_I(inode)->delegation);
949 	if (delegation == NULL)
950 		goto out;
951 	if (stateid == NULL) {
952 		nfs4_stateid_copy(&tmp, &delegation->stateid);
953 		stateid = &tmp;
954 	} else {
955 		if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
956 			goto out;
957 		spin_lock(&delegation->lock);
958 		if (stateid->seqid) {
959 			if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
960 				spin_unlock(&delegation->lock);
961 				goto out;
962 			}
963 			delegation->stateid.seqid = stateid->seqid;
964 		}
965 		spin_unlock(&delegation->lock);
966 	}
967 	nfs_mark_delegation_revoked(delegation);
968 	ret = true;
969 out:
970 	rcu_read_unlock();
971 	if (ret)
972 		nfs_inode_find_state_and_recover(inode, stateid);
973 }
974 
975 void nfs_remove_bad_delegation(struct inode *inode,
976 		const nfs4_stateid *stateid)
977 {
978 	nfs_revoke_delegation(inode, stateid);
979 }
980 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
981 
982 void nfs_delegation_mark_returned(struct inode *inode,
983 		const nfs4_stateid *stateid)
984 {
985 	struct nfs_delegation *delegation;
986 
987 	if (!inode)
988 		return;
989 
990 	rcu_read_lock();
991 	delegation = rcu_dereference(NFS_I(inode)->delegation);
992 	if (!delegation)
993 		goto out_rcu_unlock;
994 
995 	spin_lock(&delegation->lock);
996 	if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
997 		goto out_spin_unlock;
998 	if (stateid->seqid) {
999 		/* If delegation->stateid is newer, dont mark as returned */
1000 		if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
1001 			goto out_clear_returning;
1002 		if (delegation->stateid.seqid != stateid->seqid)
1003 			delegation->stateid.seqid = stateid->seqid;
1004 	}
1005 
1006 	nfs_mark_delegation_revoked(delegation);
1007 
1008 out_clear_returning:
1009 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1010 out_spin_unlock:
1011 	spin_unlock(&delegation->lock);
1012 out_rcu_unlock:
1013 	rcu_read_unlock();
1014 
1015 	nfs_inode_find_state_and_recover(inode, stateid);
1016 }
1017 
1018 /**
1019  * nfs_expire_unused_delegation_types
1020  * @clp: client to process
1021  * @flags: delegation types to expire
1022  *
1023  */
1024 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1025 {
1026 	nfs_client_mark_return_unused_delegation_types(clp, flags);
1027 	nfs_delegation_run_state_manager(clp);
1028 }
1029 
1030 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1031 {
1032 	struct nfs_delegation *delegation;
1033 
1034 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1035 		if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1036 			continue;
1037 		nfs_mark_return_if_closed_delegation(server, delegation);
1038 	}
1039 }
1040 
1041 /**
1042  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1043  * @clp: nfs_client to process
1044  *
1045  */
1046 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1047 {
1048 	struct nfs_server *server;
1049 
1050 	rcu_read_lock();
1051 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1052 		nfs_mark_return_unreferenced_delegations(server);
1053 	rcu_read_unlock();
1054 
1055 	nfs_delegation_run_state_manager(clp);
1056 }
1057 
1058 /**
1059  * nfs_async_inode_return_delegation - asynchronously return a delegation
1060  * @inode: inode to process
1061  * @stateid: state ID information
1062  *
1063  * Returns zero on success, or a negative errno value.
1064  */
1065 int nfs_async_inode_return_delegation(struct inode *inode,
1066 				      const nfs4_stateid *stateid)
1067 {
1068 	struct nfs_server *server = NFS_SERVER(inode);
1069 	struct nfs_client *clp = server->nfs_client;
1070 	struct nfs_delegation *delegation;
1071 
1072 	rcu_read_lock();
1073 	delegation = nfs4_get_valid_delegation(inode);
1074 	if (delegation == NULL)
1075 		goto out_enoent;
1076 	if (stateid != NULL &&
1077 	    !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1078 		goto out_enoent;
1079 	nfs_mark_return_delegation(server, delegation);
1080 	rcu_read_unlock();
1081 
1082 	/* If there are any application leases or delegations, recall them */
1083 	break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1084 
1085 	nfs_delegation_run_state_manager(clp);
1086 	return 0;
1087 out_enoent:
1088 	rcu_read_unlock();
1089 	return -ENOENT;
1090 }
1091 
1092 static struct inode *
1093 nfs_delegation_find_inode_server(struct nfs_server *server,
1094 				 const struct nfs_fh *fhandle)
1095 {
1096 	struct nfs_delegation *delegation;
1097 	struct super_block *freeme = NULL;
1098 	struct inode *res = NULL;
1099 
1100 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1101 		spin_lock(&delegation->lock);
1102 		if (delegation->inode != NULL &&
1103 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1104 		    nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1105 			if (nfs_sb_active(server->super)) {
1106 				freeme = server->super;
1107 				res = igrab(delegation->inode);
1108 			}
1109 			spin_unlock(&delegation->lock);
1110 			if (res != NULL)
1111 				return res;
1112 			if (freeme) {
1113 				rcu_read_unlock();
1114 				nfs_sb_deactive(freeme);
1115 				rcu_read_lock();
1116 			}
1117 			return ERR_PTR(-EAGAIN);
1118 		}
1119 		spin_unlock(&delegation->lock);
1120 	}
1121 	return ERR_PTR(-ENOENT);
1122 }
1123 
1124 /**
1125  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1126  * @clp: client state handle
1127  * @fhandle: filehandle from a delegation recall
1128  *
1129  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1130  * cannot be found.
1131  */
1132 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1133 					const struct nfs_fh *fhandle)
1134 {
1135 	struct nfs_server *server;
1136 	struct inode *res;
1137 
1138 	rcu_read_lock();
1139 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1140 		res = nfs_delegation_find_inode_server(server, fhandle);
1141 		if (res != ERR_PTR(-ENOENT)) {
1142 			rcu_read_unlock();
1143 			return res;
1144 		}
1145 	}
1146 	rcu_read_unlock();
1147 	return ERR_PTR(-ENOENT);
1148 }
1149 
1150 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1151 {
1152 	struct nfs_delegation *delegation;
1153 
1154 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1155 		/*
1156 		 * If the delegation may have been admin revoked, then we
1157 		 * cannot reclaim it.
1158 		 */
1159 		if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1160 			continue;
1161 		set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1162 	}
1163 }
1164 
1165 /**
1166  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1167  * @clp: nfs_client to process
1168  *
1169  */
1170 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1171 {
1172 	struct nfs_server *server;
1173 
1174 	rcu_read_lock();
1175 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1176 		nfs_delegation_mark_reclaim_server(server);
1177 	rcu_read_unlock();
1178 }
1179 
1180 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1181 		void __always_unused *data)
1182 {
1183 	struct nfs_delegation *delegation;
1184 	struct inode *inode;
1185 restart:
1186 	rcu_read_lock();
1187 restart_locked:
1188 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1189 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1190 					&delegation->flags) ||
1191 		    test_bit(NFS_DELEGATION_RETURNING,
1192 					&delegation->flags) ||
1193 		    test_bit(NFS_DELEGATION_NEED_RECLAIM,
1194 					&delegation->flags) == 0)
1195 			continue;
1196 		inode = nfs_delegation_grab_inode(delegation);
1197 		if (inode == NULL)
1198 			goto restart_locked;
1199 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1200 		rcu_read_unlock();
1201 		if (delegation != NULL) {
1202 			if (nfs_detach_delegation(NFS_I(inode), delegation,
1203 						server) != NULL)
1204 				nfs_free_delegation(delegation);
1205 			/* Match nfs_start_delegation_return_locked */
1206 			nfs_put_delegation(delegation);
1207 		}
1208 		iput(inode);
1209 		cond_resched();
1210 		goto restart;
1211 	}
1212 	rcu_read_unlock();
1213 	return 0;
1214 }
1215 
1216 /**
1217  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1218  * @clp: nfs_client to process
1219  *
1220  */
1221 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1222 {
1223 	nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1224 			NULL);
1225 }
1226 
1227 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1228 {
1229 	return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1230 				BIT(NFS4CLNT_LEASE_EXPIRED) |
1231 				BIT(NFS4CLNT_SESSION_RESET))) != 0;
1232 }
1233 
1234 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1235 	    struct nfs_delegation *delegation)
1236 {
1237 	if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1238 		return;
1239 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1240 	set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1241 	set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1242 }
1243 
1244 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1245 		struct inode *inode)
1246 {
1247 	struct nfs_delegation *delegation;
1248 
1249 	rcu_read_lock();
1250 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1251 	if (delegation)
1252 		nfs_mark_test_expired_delegation(server, delegation);
1253 	rcu_read_unlock();
1254 
1255 }
1256 
1257 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1258 {
1259 	struct nfs_delegation *delegation;
1260 
1261 	list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1262 		nfs_mark_test_expired_delegation(server, delegation);
1263 }
1264 
1265 /**
1266  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1267  * @clp: nfs_client to process
1268  *
1269  * Iterates through all the delegations associated with this server and
1270  * marks them as needing to be checked for validity.
1271  */
1272 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1273 {
1274 	struct nfs_server *server;
1275 
1276 	rcu_read_lock();
1277 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1278 		nfs_delegation_mark_test_expired_server(server);
1279 	rcu_read_unlock();
1280 }
1281 
1282 /**
1283  * nfs_test_expired_all_delegations - test all delegations for a client
1284  * @clp: nfs_client to process
1285  *
1286  * Helper for handling "recallable state revoked" status from server.
1287  */
1288 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1289 {
1290 	nfs_mark_test_expired_all_delegations(clp);
1291 	nfs4_schedule_state_manager(clp);
1292 }
1293 
1294 static void
1295 nfs_delegation_test_free_expired(struct inode *inode,
1296 		nfs4_stateid *stateid,
1297 		const struct cred *cred)
1298 {
1299 	struct nfs_server *server = NFS_SERVER(inode);
1300 	const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1301 	int status;
1302 
1303 	if (!cred)
1304 		return;
1305 	status = ops->test_and_free_expired(server, stateid, cred);
1306 	if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1307 		nfs_remove_bad_delegation(inode, stateid);
1308 }
1309 
1310 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1311 		void __always_unused *data)
1312 {
1313 	struct nfs_delegation *delegation;
1314 	struct inode *inode;
1315 	const struct cred *cred;
1316 	nfs4_stateid stateid;
1317 	unsigned long gen = ++server->delegation_gen;
1318 
1319 restart:
1320 	rcu_read_lock();
1321 restart_locked:
1322 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1323 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1324 					&delegation->flags) ||
1325 		    test_bit(NFS_DELEGATION_RETURNING,
1326 					&delegation->flags) ||
1327 		    test_bit(NFS_DELEGATION_TEST_EXPIRED,
1328 					&delegation->flags) == 0 ||
1329 			delegation->test_gen == gen)
1330 			continue;
1331 		inode = nfs_delegation_grab_inode(delegation);
1332 		if (inode == NULL)
1333 			goto restart_locked;
1334 		spin_lock(&delegation->lock);
1335 		cred = get_cred_rcu(delegation->cred);
1336 		nfs4_stateid_copy(&stateid, &delegation->stateid);
1337 		spin_unlock(&delegation->lock);
1338 		delegation->test_gen = gen;
1339 		clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1340 		rcu_read_unlock();
1341 		nfs_delegation_test_free_expired(inode, &stateid, cred);
1342 		put_cred(cred);
1343 		if (!nfs4_server_rebooted(server->nfs_client)) {
1344 			iput(inode);
1345 			cond_resched();
1346 			goto restart;
1347 		}
1348 		nfs_inode_mark_test_expired_delegation(server,inode);
1349 		iput(inode);
1350 		return -EAGAIN;
1351 	}
1352 	rcu_read_unlock();
1353 	return 0;
1354 }
1355 
1356 /**
1357  * nfs_reap_expired_delegations - reap expired delegations
1358  * @clp: nfs_client to process
1359  *
1360  * Iterates through all the delegations associated with this server and
1361  * checks if they have may have been revoked. This function is usually
1362  * expected to be called in cases where the server may have lost its
1363  * lease.
1364  */
1365 void nfs_reap_expired_delegations(struct nfs_client *clp)
1366 {
1367 	nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1368 			NULL);
1369 }
1370 
1371 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1372 		const nfs4_stateid *stateid)
1373 {
1374 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1375 	struct nfs_delegation *delegation;
1376 	bool found = false;
1377 
1378 	rcu_read_lock();
1379 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1380 	if (delegation &&
1381 	    nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1382 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1383 		nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1384 		found = true;
1385 	}
1386 	rcu_read_unlock();
1387 	if (found)
1388 		nfs4_schedule_state_manager(clp);
1389 }
1390 
1391 /**
1392  * nfs_delegations_present - check for existence of delegations
1393  * @clp: client state handle
1394  *
1395  * Returns one if there are any nfs_delegation structures attached
1396  * to this nfs_client.
1397  */
1398 int nfs_delegations_present(struct nfs_client *clp)
1399 {
1400 	struct nfs_server *server;
1401 	int ret = 0;
1402 
1403 	rcu_read_lock();
1404 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1405 		if (!list_empty(&server->delegations)) {
1406 			ret = 1;
1407 			break;
1408 		}
1409 	rcu_read_unlock();
1410 	return ret;
1411 }
1412 
1413 /**
1414  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1415  * @dst: stateid to refresh
1416  * @inode: inode to check
1417  *
1418  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1419  * that matches our delegation stateid. Otherwise "false" is returned.
1420  */
1421 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1422 {
1423 	struct nfs_delegation *delegation;
1424 	bool ret = false;
1425 	if (!inode)
1426 		goto out;
1427 
1428 	rcu_read_lock();
1429 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1430 	if (delegation != NULL &&
1431 	    nfs4_stateid_match_other(dst, &delegation->stateid) &&
1432 	    nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1433 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1434 		dst->seqid = delegation->stateid.seqid;
1435 		ret = true;
1436 	}
1437 	rcu_read_unlock();
1438 out:
1439 	return ret;
1440 }
1441 
1442 /**
1443  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1444  * @inode: inode to check
1445  * @flags: delegation type requirement
1446  * @dst: stateid data structure to fill in
1447  * @cred: optional argument to retrieve credential
1448  *
1449  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1450  * otherwise "false" is returned.
1451  */
1452 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1453 		nfs4_stateid *dst, const struct cred **cred)
1454 {
1455 	struct nfs_inode *nfsi = NFS_I(inode);
1456 	struct nfs_delegation *delegation;
1457 	bool ret = false;
1458 
1459 	flags &= FMODE_READ|FMODE_WRITE;
1460 	rcu_read_lock();
1461 	delegation = rcu_dereference(nfsi->delegation);
1462 	if (!delegation)
1463 		goto out;
1464 	spin_lock(&delegation->lock);
1465 	ret = nfs4_is_valid_delegation(delegation, flags);
1466 	if (ret) {
1467 		nfs4_stateid_copy(dst, &delegation->stateid);
1468 		nfs_mark_delegation_referenced(delegation);
1469 		if (cred)
1470 			*cred = get_cred(delegation->cred);
1471 	}
1472 	spin_unlock(&delegation->lock);
1473 out:
1474 	rcu_read_unlock();
1475 	return ret;
1476 }
1477 
1478 /**
1479  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1480  * @inode: inode to check
1481  *
1482  * This function checks the number of outstanding writes to the file
1483  * against the delegation 'space_limit' field to see if
1484  * the spec requires us to flush the file on close.
1485  */
1486 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1487 {
1488 	struct nfs_inode *nfsi = NFS_I(inode);
1489 	struct nfs_delegation *delegation;
1490 	bool ret = true;
1491 
1492 	rcu_read_lock();
1493 	delegation = rcu_dereference(nfsi->delegation);
1494 	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1495 		goto out;
1496 	if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1497 		ret = false;
1498 out:
1499 	rcu_read_unlock();
1500 	return ret;
1501 }
1502 
1503 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);
1504