xref: /linux/fs/nfs/delegation.c (revision 7ef60108069b7e3cc66432304e1dd197d5c0a9b5)
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 		inode = nfs_delegation_grab_inode(delegation);
651 		if (inode == NULL)
652 			continue;
653 
654 		if (prev) {
655 			struct inode *tmp = nfs_delegation_grab_inode(prev);
656 			if (tmp) {
657 				to_put = place_holder;
658 				place_holder = tmp;
659 				place_holder_deleg = prev;
660 			}
661 		}
662 
663 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
664 		rcu_read_unlock();
665 
666 		iput(to_put);
667 
668 		err = nfs_end_delegation_return(inode, delegation, 0);
669 		iput(inode);
670 		cond_resched();
671 		if (!err)
672 			goto restart;
673 		set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
674 		goto out;
675 	}
676 	rcu_read_unlock();
677 out:
678 	iput(place_holder);
679 	return err;
680 }
681 
682 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
683 {
684 	struct nfs_delegation *d;
685 	bool ret = false;
686 
687 	list_for_each_entry_rcu (d, &server->delegations, super_list) {
688 		if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
689 			continue;
690 		nfs_mark_return_delegation(server, d);
691 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
692 		ret = true;
693 	}
694 	return ret;
695 }
696 
697 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
698 {
699 	struct nfs_server *server;
700 	bool ret = false;
701 
702 	if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
703 		goto out;
704 	rcu_read_lock();
705 	list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
706 		if (nfs_server_clear_delayed_delegations(server))
707 			ret = true;
708 	}
709 	rcu_read_unlock();
710 out:
711 	return ret;
712 }
713 
714 /**
715  * nfs_client_return_marked_delegations - return previously marked delegations
716  * @clp: nfs_client to process
717  *
718  * Note that this function is designed to be called by the state
719  * manager thread. For this reason, it cannot flush the dirty data,
720  * since that could deadlock in case of a state recovery error.
721  *
722  * Returns zero on success, or a negative errno value.
723  */
724 int nfs_client_return_marked_delegations(struct nfs_client *clp)
725 {
726 	int err = nfs_client_for_each_server(
727 		clp, nfs_server_return_marked_delegations, NULL);
728 	if (err)
729 		return err;
730 	/* If a return was delayed, sleep to prevent hard looping */
731 	if (nfs_client_clear_delayed_delegations(clp))
732 		ssleep(1);
733 	return 0;
734 }
735 
736 /**
737  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
738  * @inode: inode to process
739  *
740  * Does not protect against delegation reclaims, therefore really only safe
741  * to be called from nfs4_clear_inode(). Guaranteed to always free
742  * the delegation structure.
743  */
744 void nfs_inode_evict_delegation(struct inode *inode)
745 {
746 	struct nfs_delegation *delegation;
747 
748 	delegation = nfs_inode_detach_delegation(inode);
749 	if (delegation != NULL) {
750 		set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
751 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
752 		nfs_do_return_delegation(inode, delegation, 1);
753 		nfs_free_delegation(delegation);
754 	}
755 }
756 
757 /**
758  * nfs4_inode_return_delegation - synchronously return a delegation
759  * @inode: inode to process
760  *
761  * This routine will always flush any dirty data to disk on the
762  * assumption that if we need to return the delegation, then
763  * we should stop caching.
764  *
765  * Returns zero on success, or a negative errno value.
766  */
767 int nfs4_inode_return_delegation(struct inode *inode)
768 {
769 	struct nfs_inode *nfsi = NFS_I(inode);
770 	struct nfs_delegation *delegation;
771 
772 	delegation = nfs_start_delegation_return(nfsi);
773 	if (delegation != NULL) {
774 		/* Synchronous recall of any application leases */
775 		break_lease(inode, O_WRONLY | O_RDWR);
776 		if (S_ISREG(inode->i_mode))
777 			nfs_wb_all(inode);
778 		return nfs_end_delegation_return(inode, delegation, 1);
779 	}
780 	return 0;
781 }
782 
783 /**
784  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
785  * @inode: inode to process
786  *
787  * This routine is called on file close in order to determine if the
788  * inode delegation needs to be returned immediately.
789  */
790 void nfs4_inode_return_delegation_on_close(struct inode *inode)
791 {
792 	struct nfs_delegation *delegation;
793 	struct nfs_delegation *ret = NULL;
794 
795 	if (!inode)
796 		return;
797 	rcu_read_lock();
798 	delegation = nfs4_get_valid_delegation(inode);
799 	if (!delegation)
800 		goto out;
801 	if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
802 	    atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
803 		spin_lock(&delegation->lock);
804 		if (delegation->inode &&
805 		    list_empty(&NFS_I(inode)->open_files) &&
806 		    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
807 			clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
808 			/* Refcount matched in nfs_end_delegation_return() */
809 			ret = nfs_get_delegation(delegation);
810 		}
811 		spin_unlock(&delegation->lock);
812 		if (ret)
813 			nfs_clear_verifier_delegated(inode);
814 	}
815 out:
816 	rcu_read_unlock();
817 	nfs_end_delegation_return(inode, ret, 0);
818 }
819 
820 /**
821  * nfs4_inode_make_writeable
822  * @inode: pointer to inode
823  *
824  * Make the inode writeable by returning the delegation if necessary
825  *
826  * Returns zero on success, or a negative errno value.
827  */
828 int nfs4_inode_make_writeable(struct inode *inode)
829 {
830 	struct nfs_delegation *delegation;
831 
832 	rcu_read_lock();
833 	delegation = nfs4_get_valid_delegation(inode);
834 	if (delegation == NULL ||
835 	    (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
836 	     (delegation->type & FMODE_WRITE))) {
837 		rcu_read_unlock();
838 		return 0;
839 	}
840 	rcu_read_unlock();
841 	return nfs4_inode_return_delegation(inode);
842 }
843 
844 static void nfs_mark_return_if_closed_delegation(struct nfs_server *server,
845 		struct nfs_delegation *delegation)
846 {
847 	set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
848 	set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
849 }
850 
851 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
852 {
853 	struct nfs_delegation *delegation;
854 	bool ret = false;
855 
856 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
857 		nfs_mark_return_delegation(server, delegation);
858 		ret = true;
859 	}
860 	return ret;
861 }
862 
863 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
864 {
865 	struct nfs_server *server;
866 
867 	rcu_read_lock();
868 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
869 		nfs_server_mark_return_all_delegations(server);
870 	rcu_read_unlock();
871 }
872 
873 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
874 {
875 	if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
876 		nfs4_schedule_state_manager(clp);
877 }
878 
879 /**
880  * nfs_expire_all_delegations
881  * @clp: client to process
882  *
883  */
884 void nfs_expire_all_delegations(struct nfs_client *clp)
885 {
886 	nfs_client_mark_return_all_delegations(clp);
887 	nfs_delegation_run_state_manager(clp);
888 }
889 
890 /**
891  * nfs_server_return_all_delegations - return delegations for one superblock
892  * @server: pointer to nfs_server to process
893  *
894  */
895 void nfs_server_return_all_delegations(struct nfs_server *server)
896 {
897 	struct nfs_client *clp = server->nfs_client;
898 	bool need_wait;
899 
900 	if (clp == NULL)
901 		return;
902 
903 	rcu_read_lock();
904 	need_wait = nfs_server_mark_return_all_delegations(server);
905 	rcu_read_unlock();
906 
907 	if (need_wait) {
908 		nfs4_schedule_state_manager(clp);
909 		nfs4_wait_clnt_recover(clp);
910 	}
911 }
912 
913 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
914 						 fmode_t flags)
915 {
916 	struct nfs_delegation *delegation;
917 
918 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
919 		if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
920 			continue;
921 		if (delegation->type & flags)
922 			nfs_mark_return_if_closed_delegation(server, delegation);
923 	}
924 }
925 
926 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
927 							fmode_t flags)
928 {
929 	struct nfs_server *server;
930 
931 	rcu_read_lock();
932 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
933 		nfs_mark_return_unused_delegation_types(server, flags);
934 	rcu_read_unlock();
935 }
936 
937 static void nfs_revoke_delegation(struct inode *inode,
938 		const nfs4_stateid *stateid)
939 {
940 	struct nfs_delegation *delegation;
941 	nfs4_stateid tmp;
942 	bool ret = false;
943 
944 	rcu_read_lock();
945 	delegation = rcu_dereference(NFS_I(inode)->delegation);
946 	if (delegation == NULL)
947 		goto out;
948 	if (stateid == NULL) {
949 		nfs4_stateid_copy(&tmp, &delegation->stateid);
950 		stateid = &tmp;
951 	} else {
952 		if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
953 			goto out;
954 		spin_lock(&delegation->lock);
955 		if (stateid->seqid) {
956 			if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
957 				spin_unlock(&delegation->lock);
958 				goto out;
959 			}
960 			delegation->stateid.seqid = stateid->seqid;
961 		}
962 		spin_unlock(&delegation->lock);
963 	}
964 	nfs_mark_delegation_revoked(delegation);
965 	ret = true;
966 out:
967 	rcu_read_unlock();
968 	if (ret)
969 		nfs_inode_find_state_and_recover(inode, stateid);
970 }
971 
972 void nfs_remove_bad_delegation(struct inode *inode,
973 		const nfs4_stateid *stateid)
974 {
975 	nfs_revoke_delegation(inode, stateid);
976 }
977 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
978 
979 void nfs_delegation_mark_returned(struct inode *inode,
980 		const nfs4_stateid *stateid)
981 {
982 	struct nfs_delegation *delegation;
983 
984 	if (!inode)
985 		return;
986 
987 	rcu_read_lock();
988 	delegation = rcu_dereference(NFS_I(inode)->delegation);
989 	if (!delegation)
990 		goto out_rcu_unlock;
991 
992 	spin_lock(&delegation->lock);
993 	if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
994 		goto out_spin_unlock;
995 	if (stateid->seqid) {
996 		/* If delegation->stateid is newer, dont mark as returned */
997 		if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
998 			goto out_clear_returning;
999 		if (delegation->stateid.seqid != stateid->seqid)
1000 			delegation->stateid.seqid = stateid->seqid;
1001 	}
1002 
1003 	nfs_mark_delegation_revoked(delegation);
1004 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1005 	spin_unlock(&delegation->lock);
1006 	if (nfs_detach_delegation(NFS_I(inode), delegation, NFS_SERVER(inode)))
1007 		nfs_put_delegation(delegation);
1008 	goto out_rcu_unlock;
1009 
1010 out_clear_returning:
1011 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1012 out_spin_unlock:
1013 	spin_unlock(&delegation->lock);
1014 out_rcu_unlock:
1015 	rcu_read_unlock();
1016 
1017 	nfs_inode_find_state_and_recover(inode, stateid);
1018 }
1019 
1020 /**
1021  * nfs_expire_unused_delegation_types
1022  * @clp: client to process
1023  * @flags: delegation types to expire
1024  *
1025  */
1026 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1027 {
1028 	nfs_client_mark_return_unused_delegation_types(clp, flags);
1029 	nfs_delegation_run_state_manager(clp);
1030 }
1031 
1032 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1033 {
1034 	struct nfs_delegation *delegation;
1035 
1036 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1037 		if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1038 			continue;
1039 		nfs_mark_return_if_closed_delegation(server, delegation);
1040 	}
1041 }
1042 
1043 /**
1044  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1045  * @clp: nfs_client to process
1046  *
1047  */
1048 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1049 {
1050 	struct nfs_server *server;
1051 
1052 	rcu_read_lock();
1053 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1054 		nfs_mark_return_unreferenced_delegations(server);
1055 	rcu_read_unlock();
1056 
1057 	nfs_delegation_run_state_manager(clp);
1058 }
1059 
1060 /**
1061  * nfs_async_inode_return_delegation - asynchronously return a delegation
1062  * @inode: inode to process
1063  * @stateid: state ID information
1064  *
1065  * Returns zero on success, or a negative errno value.
1066  */
1067 int nfs_async_inode_return_delegation(struct inode *inode,
1068 				      const nfs4_stateid *stateid)
1069 {
1070 	struct nfs_server *server = NFS_SERVER(inode);
1071 	struct nfs_client *clp = server->nfs_client;
1072 	struct nfs_delegation *delegation;
1073 
1074 	rcu_read_lock();
1075 	delegation = nfs4_get_valid_delegation(inode);
1076 	if (delegation == NULL)
1077 		goto out_enoent;
1078 	if (stateid != NULL &&
1079 	    !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1080 		goto out_enoent;
1081 	nfs_mark_return_delegation(server, delegation);
1082 	rcu_read_unlock();
1083 
1084 	/* If there are any application leases or delegations, recall them */
1085 	break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1086 
1087 	nfs_delegation_run_state_manager(clp);
1088 	return 0;
1089 out_enoent:
1090 	rcu_read_unlock();
1091 	return -ENOENT;
1092 }
1093 
1094 static struct inode *
1095 nfs_delegation_find_inode_server(struct nfs_server *server,
1096 				 const struct nfs_fh *fhandle)
1097 {
1098 	struct nfs_delegation *delegation;
1099 	struct super_block *freeme = NULL;
1100 	struct inode *res = NULL;
1101 
1102 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1103 		spin_lock(&delegation->lock);
1104 		if (delegation->inode != NULL &&
1105 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1106 		    nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1107 			if (nfs_sb_active(server->super)) {
1108 				freeme = server->super;
1109 				res = igrab(delegation->inode);
1110 			}
1111 			spin_unlock(&delegation->lock);
1112 			if (res != NULL)
1113 				return res;
1114 			if (freeme) {
1115 				rcu_read_unlock();
1116 				nfs_sb_deactive(freeme);
1117 				rcu_read_lock();
1118 			}
1119 			return ERR_PTR(-EAGAIN);
1120 		}
1121 		spin_unlock(&delegation->lock);
1122 	}
1123 	return ERR_PTR(-ENOENT);
1124 }
1125 
1126 /**
1127  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1128  * @clp: client state handle
1129  * @fhandle: filehandle from a delegation recall
1130  *
1131  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1132  * cannot be found.
1133  */
1134 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1135 					const struct nfs_fh *fhandle)
1136 {
1137 	struct nfs_server *server;
1138 	struct inode *res;
1139 
1140 	rcu_read_lock();
1141 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1142 		res = nfs_delegation_find_inode_server(server, fhandle);
1143 		if (res != ERR_PTR(-ENOENT)) {
1144 			rcu_read_unlock();
1145 			return res;
1146 		}
1147 	}
1148 	rcu_read_unlock();
1149 	return ERR_PTR(-ENOENT);
1150 }
1151 
1152 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1153 {
1154 	struct nfs_delegation *delegation;
1155 
1156 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1157 		/*
1158 		 * If the delegation may have been admin revoked, then we
1159 		 * cannot reclaim it.
1160 		 */
1161 		if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1162 			continue;
1163 		set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1164 	}
1165 }
1166 
1167 /**
1168  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1169  * @clp: nfs_client to process
1170  *
1171  */
1172 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1173 {
1174 	struct nfs_server *server;
1175 
1176 	rcu_read_lock();
1177 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1178 		nfs_delegation_mark_reclaim_server(server);
1179 	rcu_read_unlock();
1180 }
1181 
1182 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1183 		void __always_unused *data)
1184 {
1185 	struct nfs_delegation *delegation;
1186 	struct inode *inode;
1187 restart:
1188 	rcu_read_lock();
1189 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1190 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1191 					&delegation->flags) ||
1192 		    test_bit(NFS_DELEGATION_RETURNING,
1193 					&delegation->flags) ||
1194 		    test_bit(NFS_DELEGATION_NEED_RECLAIM,
1195 					&delegation->flags) == 0)
1196 			continue;
1197 		inode = nfs_delegation_grab_inode(delegation);
1198 		if (inode == NULL)
1199 			continue;
1200 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1201 		rcu_read_unlock();
1202 		if (delegation != NULL) {
1203 			if (nfs_detach_delegation(NFS_I(inode), delegation,
1204 						server) != NULL)
1205 				nfs_free_delegation(delegation);
1206 			/* Match nfs_start_delegation_return_locked */
1207 			nfs_put_delegation(delegation);
1208 		}
1209 		iput(inode);
1210 		cond_resched();
1211 		goto restart;
1212 	}
1213 	rcu_read_unlock();
1214 	return 0;
1215 }
1216 
1217 /**
1218  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1219  * @clp: nfs_client to process
1220  *
1221  */
1222 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1223 {
1224 	nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1225 			NULL);
1226 }
1227 
1228 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1229 {
1230 	return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1231 				BIT(NFS4CLNT_LEASE_EXPIRED) |
1232 				BIT(NFS4CLNT_SESSION_RESET))) != 0;
1233 }
1234 
1235 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1236 	    struct nfs_delegation *delegation)
1237 {
1238 	if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1239 		return;
1240 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1241 	set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1242 	set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1243 }
1244 
1245 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1246 		struct inode *inode)
1247 {
1248 	struct nfs_delegation *delegation;
1249 
1250 	rcu_read_lock();
1251 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1252 	if (delegation)
1253 		nfs_mark_test_expired_delegation(server, delegation);
1254 	rcu_read_unlock();
1255 
1256 }
1257 
1258 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1259 {
1260 	struct nfs_delegation *delegation;
1261 
1262 	list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1263 		nfs_mark_test_expired_delegation(server, delegation);
1264 }
1265 
1266 /**
1267  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1268  * @clp: nfs_client to process
1269  *
1270  * Iterates through all the delegations associated with this server and
1271  * marks them as needing to be checked for validity.
1272  */
1273 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1274 {
1275 	struct nfs_server *server;
1276 
1277 	rcu_read_lock();
1278 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1279 		nfs_delegation_mark_test_expired_server(server);
1280 	rcu_read_unlock();
1281 }
1282 
1283 /**
1284  * nfs_test_expired_all_delegations - test all delegations for a client
1285  * @clp: nfs_client to process
1286  *
1287  * Helper for handling "recallable state revoked" status from server.
1288  */
1289 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1290 {
1291 	nfs_mark_test_expired_all_delegations(clp);
1292 	nfs4_schedule_state_manager(clp);
1293 }
1294 
1295 static void
1296 nfs_delegation_test_free_expired(struct inode *inode,
1297 		nfs4_stateid *stateid,
1298 		const struct cred *cred)
1299 {
1300 	struct nfs_server *server = NFS_SERVER(inode);
1301 	const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1302 	int status;
1303 
1304 	if (!cred)
1305 		return;
1306 	status = ops->test_and_free_expired(server, stateid, cred);
1307 	if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1308 		nfs_remove_bad_delegation(inode, stateid);
1309 }
1310 
1311 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1312 		void __always_unused *data)
1313 {
1314 	struct nfs_delegation *delegation;
1315 	struct inode *inode;
1316 	const struct cred *cred;
1317 	nfs4_stateid stateid;
1318 	unsigned long gen = ++server->delegation_gen;
1319 
1320 restart:
1321 	rcu_read_lock();
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 			continue;
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