xref: /linux/fs/nfs/delegation.c (revision 7375bbad467e9b1b101d591a458b49a0f3896641)
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(NFS4SERV_DELEGRETURN, &server->delegation_flags);
83 	set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
84 }
85 
86 static bool nfs4_is_valid_delegation(const struct nfs_delegation *delegation,
87 				     fmode_t type)
88 {
89 	if (delegation != NULL && (delegation->type & type) == type &&
90 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
91 	    !test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
92 		return true;
93 	return false;
94 }
95 
96 struct nfs_delegation *nfs4_get_valid_delegation(const struct inode *inode)
97 {
98 	struct nfs_delegation *delegation;
99 
100 	delegation = rcu_dereference(NFS_I(inode)->delegation);
101 	if (nfs4_is_valid_delegation(delegation, 0))
102 		return delegation;
103 	return NULL;
104 }
105 
106 static int nfs4_do_check_delegation(struct inode *inode, fmode_t type,
107 				    int flags, bool mark)
108 {
109 	struct nfs_delegation *delegation;
110 	int ret = 0;
111 
112 	type &= FMODE_READ|FMODE_WRITE;
113 	rcu_read_lock();
114 	delegation = rcu_dereference(NFS_I(inode)->delegation);
115 	if (nfs4_is_valid_delegation(delegation, type)) {
116 		if (mark)
117 			nfs_mark_delegation_referenced(delegation);
118 		ret = 1;
119 		if ((flags & NFS_DELEGATION_FLAG_TIME) &&
120 		    !test_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags))
121 			ret = 0;
122 	}
123 	rcu_read_unlock();
124 	return ret;
125 }
126 /**
127  * nfs4_have_delegation - check if inode has a delegation, mark it
128  * NFS_DELEGATION_REFERENCED if there is one.
129  * @inode: inode to check
130  * @type: delegation types to check for
131  * @flags: various modifiers
132  *
133  * Returns one if inode has the indicated delegation, otherwise zero.
134  */
135 int nfs4_have_delegation(struct inode *inode, fmode_t type, int flags)
136 {
137 	return nfs4_do_check_delegation(inode, type, flags, true);
138 }
139 
140 /*
141  * nfs4_check_delegation - check if inode has a delegation, do not mark
142  * NFS_DELEGATION_REFERENCED if it has one.
143  */
144 int nfs4_check_delegation(struct inode *inode, fmode_t type)
145 {
146 	return nfs4_do_check_delegation(inode, type, 0, false);
147 }
148 
149 static int nfs_delegation_claim_locks(struct nfs4_state *state, const nfs4_stateid *stateid)
150 {
151 	struct inode *inode = state->inode;
152 	struct file_lock *fl;
153 	struct file_lock_context *flctx = locks_inode_context(inode);
154 	struct list_head *list;
155 	int status = 0;
156 
157 	if (flctx == NULL)
158 		goto out;
159 
160 	list = &flctx->flc_posix;
161 	spin_lock(&flctx->flc_lock);
162 restart:
163 	for_each_file_lock(fl, list) {
164 		if (nfs_file_open_context(fl->c.flc_file)->state != state)
165 			continue;
166 		spin_unlock(&flctx->flc_lock);
167 		status = nfs4_lock_delegation_recall(fl, state, stateid);
168 		if (status < 0)
169 			goto out;
170 		spin_lock(&flctx->flc_lock);
171 	}
172 	if (list == &flctx->flc_posix) {
173 		list = &flctx->flc_flock;
174 		goto restart;
175 	}
176 	spin_unlock(&flctx->flc_lock);
177 out:
178 	return status;
179 }
180 
181 static int nfs_delegation_claim_opens(struct inode *inode,
182 		const nfs4_stateid *stateid, fmode_t type)
183 {
184 	struct nfs_inode *nfsi = NFS_I(inode);
185 	struct nfs_open_context *ctx;
186 	struct nfs4_state_owner *sp;
187 	struct nfs4_state *state;
188 	int err;
189 
190 again:
191 	rcu_read_lock();
192 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
193 		state = ctx->state;
194 		if (state == NULL)
195 			continue;
196 		if (!test_bit(NFS_DELEGATED_STATE, &state->flags))
197 			continue;
198 		if (!nfs4_valid_open_stateid(state))
199 			continue;
200 		if (!nfs4_stateid_match(&state->stateid, stateid))
201 			continue;
202 		if (!get_nfs_open_context(ctx))
203 			continue;
204 		rcu_read_unlock();
205 		sp = state->owner;
206 		/* Block nfs4_proc_unlck */
207 		mutex_lock(&sp->so_delegreturn_mutex);
208 		err = nfs4_open_delegation_recall(ctx, state, stateid);
209 		if (!err)
210 			err = nfs_delegation_claim_locks(state, stateid);
211 		mutex_unlock(&sp->so_delegreturn_mutex);
212 		put_nfs_open_context(ctx);
213 		if (err != 0)
214 			return err;
215 		goto again;
216 	}
217 	rcu_read_unlock();
218 	return 0;
219 }
220 
221 /**
222  * nfs_inode_reclaim_delegation - process a delegation reclaim request
223  * @inode: inode to process
224  * @cred: credential to use for request
225  * @type: delegation type
226  * @stateid: delegation stateid
227  * @pagemod_limit: write delegation "space_limit"
228  * @deleg_type: raw delegation type
229  *
230  */
231 void nfs_inode_reclaim_delegation(struct inode *inode, const struct cred *cred,
232 				  fmode_t type, const nfs4_stateid *stateid,
233 				  unsigned long pagemod_limit, u32 deleg_type)
234 {
235 	struct nfs_delegation *delegation;
236 	const struct cred *oldcred = NULL;
237 
238 	rcu_read_lock();
239 	delegation = rcu_dereference(NFS_I(inode)->delegation);
240 	if (!delegation) {
241 		rcu_read_unlock();
242 		nfs_inode_set_delegation(inode, cred, type, stateid,
243 					 pagemod_limit, deleg_type);
244 		return;
245 	}
246 
247 	spin_lock(&delegation->lock);
248 	nfs4_stateid_copy(&delegation->stateid, stateid);
249 	delegation->type = type;
250 	delegation->pagemod_limit = pagemod_limit;
251 	oldcred = delegation->cred;
252 	delegation->cred = get_cred(cred);
253 	switch (deleg_type) {
254 	case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
255 	case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
256 		set_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
257 		break;
258 	default:
259 		clear_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
260 	}
261 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
262 	if (test_and_clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
263 		atomic_long_inc(&nfs_active_delegations);
264 	spin_unlock(&delegation->lock);
265 	rcu_read_unlock();
266 	put_cred(oldcred);
267 	trace_nfs4_reclaim_delegation(inode, type);
268 }
269 
270 static int nfs_do_return_delegation(struct inode *inode,
271 				    struct nfs_delegation *delegation,
272 				    int issync)
273 {
274 	const struct cred *cred;
275 	int res = 0;
276 
277 	if (!test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
278 		spin_lock(&delegation->lock);
279 		cred = get_cred(delegation->cred);
280 		spin_unlock(&delegation->lock);
281 		res = nfs4_proc_delegreturn(inode, cred, &delegation->stateid,
282 					    delegation, issync);
283 		put_cred(cred);
284 	}
285 	return res;
286 }
287 
288 static struct inode *nfs_delegation_grab_inode(struct nfs_delegation *delegation)
289 {
290 	struct inode *inode = NULL;
291 
292 	spin_lock(&delegation->lock);
293 	if (delegation->inode != NULL)
294 		inode = igrab(delegation->inode);
295 	if (!inode)
296 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
297 	spin_unlock(&delegation->lock);
298 	return inode;
299 }
300 
301 static struct nfs_delegation *
302 nfs_start_delegation_return_locked(struct nfs_inode *nfsi)
303 {
304 	struct nfs_delegation *ret = NULL;
305 	struct nfs_delegation *delegation = rcu_dereference(nfsi->delegation);
306 
307 	if (delegation == NULL)
308 		goto out;
309 	spin_lock(&delegation->lock);
310 	if (delegation->inode &&
311 	    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
312 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
313 		/* Refcount matched in nfs_end_delegation_return() */
314 		ret = nfs_get_delegation(delegation);
315 	}
316 	spin_unlock(&delegation->lock);
317 	if (ret)
318 		nfs_clear_verifier_delegated(&nfsi->vfs_inode);
319 out:
320 	return ret;
321 }
322 
323 static struct nfs_delegation *
324 nfs_start_delegation_return(struct nfs_inode *nfsi)
325 {
326 	struct nfs_delegation *delegation;
327 
328 	rcu_read_lock();
329 	delegation = nfs_start_delegation_return_locked(nfsi);
330 	rcu_read_unlock();
331 	return delegation;
332 }
333 
334 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
335 					struct nfs_server *server, int err)
336 {
337 	spin_lock(&delegation->lock);
338 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
339 	if (err == -EAGAIN) {
340 		set_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
341 		set_bit(NFS4SERV_DELEGRETURN_DELAYED,
342 			&server->delegation_flags);
343 		set_bit(NFS4CLNT_DELEGRETURN_DELAYED,
344 			&server->nfs_client->cl_state);
345 	}
346 	spin_unlock(&delegation->lock);
347 }
348 
349 static struct nfs_delegation *
350 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
351 		struct nfs_delegation *delegation,
352 		struct nfs_client *clp)
353 {
354 	struct nfs_delegation *deleg_cur =
355 		rcu_dereference_protected(nfsi->delegation,
356 				lockdep_is_held(&clp->cl_lock));
357 
358 	trace_nfs4_detach_delegation(&nfsi->vfs_inode, delegation->type);
359 
360 	if (deleg_cur == NULL || delegation != deleg_cur)
361 		return NULL;
362 
363 	spin_lock(&delegation->lock);
364 	if (!delegation->inode) {
365 		spin_unlock(&delegation->lock);
366 		return NULL;
367 	}
368 	list_del_rcu(&delegation->super_list);
369 	delegation->inode = NULL;
370 	rcu_assign_pointer(nfsi->delegation, NULL);
371 	spin_unlock(&delegation->lock);
372 	return delegation;
373 }
374 
375 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
376 		struct nfs_delegation *delegation,
377 		struct nfs_server *server)
378 {
379 	struct nfs_client *clp = server->nfs_client;
380 
381 	spin_lock(&clp->cl_lock);
382 	delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
383 	spin_unlock(&clp->cl_lock);
384 	return delegation;
385 }
386 
387 static struct nfs_delegation *
388 nfs_inode_detach_delegation(struct inode *inode)
389 {
390 	struct nfs_inode *nfsi = NFS_I(inode);
391 	struct nfs_server *server = NFS_SERVER(inode);
392 	struct nfs_delegation *delegation;
393 
394 	rcu_read_lock();
395 	delegation = rcu_dereference(nfsi->delegation);
396 	if (delegation != NULL)
397 		delegation = nfs_detach_delegation(nfsi, delegation, server);
398 	rcu_read_unlock();
399 	return delegation;
400 }
401 
402 static void
403 nfs_update_delegation_cred(struct nfs_delegation *delegation,
404 		const struct cred *cred)
405 {
406 	const struct cred *old;
407 
408 	if (cred_fscmp(delegation->cred, cred) != 0) {
409 		old = xchg(&delegation->cred, get_cred(cred));
410 		put_cred(old);
411 	}
412 }
413 
414 static void
415 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
416 		const struct nfs_delegation *update)
417 {
418 	if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
419 		delegation->stateid.seqid = update->stateid.seqid;
420 		smp_wmb();
421 		delegation->type = update->type;
422 		delegation->pagemod_limit = update->pagemod_limit;
423 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
424 			delegation->change_attr = update->change_attr;
425 			nfs_update_delegation_cred(delegation, update->cred);
426 			/* smp_mb__before_atomic() is implicit due to xchg() */
427 			clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
428 			atomic_long_inc(&nfs_active_delegations);
429 		}
430 	}
431 }
432 
433 /**
434  * nfs_inode_set_delegation - set up a delegation on an inode
435  * @inode: inode to which delegation applies
436  * @cred: cred to use for subsequent delegation processing
437  * @type: delegation type
438  * @stateid: delegation stateid
439  * @pagemod_limit: write delegation "space_limit"
440  * @deleg_type: raw delegation type
441  *
442  * Returns zero on success, or a negative errno value.
443  */
444 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
445 			     fmode_t type, const nfs4_stateid *stateid,
446 			     unsigned long pagemod_limit, u32 deleg_type)
447 {
448 	struct nfs_server *server = NFS_SERVER(inode);
449 	struct nfs_client *clp = server->nfs_client;
450 	struct nfs_inode *nfsi = NFS_I(inode);
451 	struct nfs_delegation *delegation, *old_delegation;
452 	struct nfs_delegation *freeme = NULL;
453 	int status = 0;
454 
455 	delegation = kmalloc(sizeof(*delegation), GFP_KERNEL_ACCOUNT);
456 	if (delegation == NULL)
457 		return -ENOMEM;
458 	nfs4_stateid_copy(&delegation->stateid, stateid);
459 	refcount_set(&delegation->refcount, 1);
460 	delegation->type = type;
461 	delegation->pagemod_limit = pagemod_limit;
462 	delegation->change_attr = inode_peek_iversion_raw(inode);
463 	delegation->cred = get_cred(cred);
464 	delegation->inode = inode;
465 	delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
466 	switch (deleg_type) {
467 	case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
468 	case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
469 		delegation->flags |= BIT(NFS_DELEGATION_DELEGTIME);
470 	}
471 	delegation->test_gen = 0;
472 	spin_lock_init(&delegation->lock);
473 
474 	spin_lock(&clp->cl_lock);
475 	old_delegation = rcu_dereference_protected(nfsi->delegation,
476 					lockdep_is_held(&clp->cl_lock));
477 	if (old_delegation == NULL)
478 		goto add_new;
479 	/* Is this an update of the existing delegation? */
480 	if (nfs4_stateid_match_other(&old_delegation->stateid,
481 				&delegation->stateid)) {
482 		spin_lock(&old_delegation->lock);
483 		nfs_update_inplace_delegation(old_delegation,
484 				delegation);
485 		spin_unlock(&old_delegation->lock);
486 		goto out;
487 	}
488 	if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
489 		/*
490 		 * Deal with broken servers that hand out two
491 		 * delegations for the same file.
492 		 * Allow for upgrades to a WRITE delegation, but
493 		 * nothing else.
494 		 */
495 		dfprintk(FILE, "%s: server %s handed out "
496 				"a duplicate delegation!\n",
497 				__func__, clp->cl_hostname);
498 		if (delegation->type == old_delegation->type ||
499 		    !(delegation->type & FMODE_WRITE)) {
500 			freeme = delegation;
501 			delegation = NULL;
502 			goto out;
503 		}
504 		if (test_and_set_bit(NFS_DELEGATION_RETURNING,
505 					&old_delegation->flags))
506 			goto out;
507 	}
508 	freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
509 	if (freeme == NULL)
510 		goto out;
511 add_new:
512 	/*
513 	 * If we didn't revalidate the change attribute before setting
514 	 * the delegation, then pre-emptively ask for a full attribute
515 	 * cache revalidation.
516 	 */
517 	spin_lock(&inode->i_lock);
518 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
519 		nfs_set_cache_invalid(inode,
520 			NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
521 			NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
522 			NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
523 			NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
524 			NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
525 			NFS_INO_INVALID_XATTR);
526 	spin_unlock(&inode->i_lock);
527 
528 	list_add_tail_rcu(&delegation->super_list, &server->delegations);
529 	rcu_assign_pointer(nfsi->delegation, delegation);
530 	delegation = NULL;
531 
532 	atomic_long_inc(&nfs_active_delegations);
533 
534 	trace_nfs4_set_delegation(inode, type);
535 
536 	/* If we hold writebacks and have delegated mtime then update */
537 	if (deleg_type == NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG &&
538 	    nfs_have_writebacks(inode))
539 		nfs_update_delegated_mtime(inode);
540 out:
541 	spin_unlock(&clp->cl_lock);
542 	if (delegation != NULL)
543 		__nfs_free_delegation(delegation);
544 	if (freeme != NULL) {
545 		nfs_do_return_delegation(inode, freeme, 0);
546 		nfs_free_delegation(freeme);
547 	}
548 	return status;
549 }
550 
551 /*
552  * Basic procedure for returning a delegation to the server
553  */
554 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
555 {
556 	struct nfs_server *server = NFS_SERVER(inode);
557 	unsigned int mode = O_WRONLY | O_RDWR;
558 	int err = 0;
559 
560 	if (delegation == NULL)
561 		return 0;
562 
563 	if (!issync)
564 		mode |= O_NONBLOCK;
565 	/* Recall of any remaining application leases */
566 	err = break_lease(inode, mode);
567 
568 	while (err == 0) {
569 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
570 			break;
571 		err = nfs_delegation_claim_opens(inode, &delegation->stateid,
572 				delegation->type);
573 		if (!issync || err != -EAGAIN)
574 			break;
575 		/*
576 		 * Guard against state recovery
577 		 */
578 		err = nfs4_wait_clnt_recover(server->nfs_client);
579 	}
580 
581 	if (err) {
582 		nfs_abort_delegation_return(delegation, server, err);
583 		goto out;
584 	}
585 
586 	err = nfs_do_return_delegation(inode, delegation, issync);
587 out:
588 	/* Refcount matched in nfs_start_delegation_return_locked() */
589 	nfs_put_delegation(delegation);
590 	return err;
591 }
592 
593 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
594 {
595 	bool ret = false;
596 
597 	trace_nfs_delegation_need_return(delegation);
598 
599 	if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
600 		ret = true;
601 	if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
602 	    test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
603 	    test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
604 		ret = false;
605 
606 	return ret;
607 }
608 
609 static int nfs_server_return_marked_delegations(struct nfs_server *server,
610 		void __always_unused *data)
611 {
612 	struct nfs_delegation *delegation;
613 	struct nfs_delegation *prev;
614 	struct inode *inode;
615 	struct inode *place_holder = NULL;
616 	struct nfs_delegation *place_holder_deleg = NULL;
617 	int err = 0;
618 
619 	if (!test_and_clear_bit(NFS4SERV_DELEGRETURN,
620 				&server->delegation_flags))
621 		return 0;
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(NFS4SERV_DELEGRETURN, &server->delegation_flags);
674 		set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
675 		goto out;
676 	}
677 	rcu_read_unlock();
678 out:
679 	iput(place_holder);
680 	return err;
681 }
682 
683 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
684 {
685 	struct nfs_delegation *d;
686 	bool ret = false;
687 
688 	if (!test_and_clear_bit(NFS4SERV_DELEGRETURN_DELAYED,
689 				&server->delegation_flags))
690 		goto out;
691 	list_for_each_entry_rcu (d, &server->delegations, super_list) {
692 		if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
693 			continue;
694 		nfs_mark_return_delegation(server, d);
695 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
696 		ret = true;
697 	}
698 out:
699 	return ret;
700 }
701 
702 static bool nfs_client_clear_delayed_delegations(struct nfs_client *clp)
703 {
704 	struct nfs_server *server;
705 	bool ret = false;
706 
707 	if (!test_and_clear_bit(NFS4CLNT_DELEGRETURN_DELAYED, &clp->cl_state))
708 		goto out;
709 	rcu_read_lock();
710 	list_for_each_entry_rcu (server, &clp->cl_superblocks, client_link) {
711 		if (nfs_server_clear_delayed_delegations(server))
712 			ret = true;
713 	}
714 	rcu_read_unlock();
715 out:
716 	return ret;
717 }
718 
719 /**
720  * nfs_client_return_marked_delegations - return previously marked delegations
721  * @clp: nfs_client to process
722  *
723  * Note that this function is designed to be called by the state
724  * manager thread. For this reason, it cannot flush the dirty data,
725  * since that could deadlock in case of a state recovery error.
726  *
727  * Returns zero on success, or a negative errno value.
728  */
729 int nfs_client_return_marked_delegations(struct nfs_client *clp)
730 {
731 	int err = nfs_client_for_each_server(
732 		clp, nfs_server_return_marked_delegations, NULL);
733 	if (err)
734 		return err;
735 	/* If a return was delayed, sleep to prevent hard looping */
736 	if (nfs_client_clear_delayed_delegations(clp))
737 		ssleep(1);
738 	return 0;
739 }
740 
741 /**
742  * nfs_inode_evict_delegation - return delegation, don't reclaim opens
743  * @inode: inode to process
744  *
745  * Does not protect against delegation reclaims, therefore really only safe
746  * to be called from nfs4_clear_inode(). Guaranteed to always free
747  * the delegation structure.
748  */
749 void nfs_inode_evict_delegation(struct inode *inode)
750 {
751 	struct nfs_delegation *delegation;
752 
753 	delegation = nfs_inode_detach_delegation(inode);
754 	if (delegation != NULL) {
755 		set_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
756 		set_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags);
757 		nfs_do_return_delegation(inode, delegation, 1);
758 		nfs_free_delegation(delegation);
759 	}
760 }
761 
762 /**
763  * nfs4_inode_return_delegation - synchronously return a delegation
764  * @inode: inode to process
765  *
766  * This routine will always flush any dirty data to disk on the
767  * assumption that if we need to return the delegation, then
768  * we should stop caching.
769  *
770  * Returns zero on success, or a negative errno value.
771  */
772 int nfs4_inode_return_delegation(struct inode *inode)
773 {
774 	struct nfs_inode *nfsi = NFS_I(inode);
775 	struct nfs_delegation *delegation;
776 
777 	delegation = nfs_start_delegation_return(nfsi);
778 	if (delegation != NULL) {
779 		/* Synchronous recall of any application leases */
780 		break_lease(inode, O_WRONLY | O_RDWR);
781 		if (S_ISREG(inode->i_mode))
782 			nfs_wb_all(inode);
783 		return nfs_end_delegation_return(inode, delegation, 1);
784 	}
785 	return 0;
786 }
787 
788 /**
789  * nfs4_inode_set_return_delegation_on_close - asynchronously return a delegation
790  * @inode: inode to process
791  *
792  * This routine is called to request that the delegation be returned as soon
793  * as the file is closed. If the file is already closed, the delegation is
794  * immediately returned.
795  */
796 void nfs4_inode_set_return_delegation_on_close(struct inode *inode)
797 {
798 	struct nfs_delegation *delegation;
799 	struct nfs_delegation *ret = NULL;
800 
801 	if (!inode)
802 		return;
803 	rcu_read_lock();
804 	delegation = nfs4_get_valid_delegation(inode);
805 	if (!delegation)
806 		goto out;
807 	spin_lock(&delegation->lock);
808 	if (!delegation->inode)
809 		goto out_unlock;
810 	if (list_empty(&NFS_I(inode)->open_files) &&
811 	    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
812 		/* Refcount matched in nfs_end_delegation_return() */
813 		ret = nfs_get_delegation(delegation);
814 	} else
815 		set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
816 out_unlock:
817 	spin_unlock(&delegation->lock);
818 	if (ret)
819 		nfs_clear_verifier_delegated(inode);
820 out:
821 	rcu_read_unlock();
822 	nfs_end_delegation_return(inode, ret, 0);
823 }
824 
825 /**
826  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
827  * @inode: inode to process
828  *
829  * This routine is called on file close in order to determine if the
830  * inode delegation needs to be returned immediately.
831  */
832 void nfs4_inode_return_delegation_on_close(struct inode *inode)
833 {
834 	struct nfs_delegation *delegation;
835 	struct nfs_delegation *ret = NULL;
836 
837 	if (!inode)
838 		return;
839 	rcu_read_lock();
840 	delegation = nfs4_get_valid_delegation(inode);
841 	if (!delegation)
842 		goto out;
843 	if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
844 	    atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
845 		spin_lock(&delegation->lock);
846 		if (delegation->inode &&
847 		    list_empty(&NFS_I(inode)->open_files) &&
848 		    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
849 			clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
850 			/* Refcount matched in nfs_end_delegation_return() */
851 			ret = nfs_get_delegation(delegation);
852 		}
853 		spin_unlock(&delegation->lock);
854 		if (ret)
855 			nfs_clear_verifier_delegated(inode);
856 	}
857 out:
858 	rcu_read_unlock();
859 	nfs_end_delegation_return(inode, ret, 0);
860 }
861 
862 /**
863  * nfs4_inode_make_writeable
864  * @inode: pointer to inode
865  *
866  * Make the inode writeable by returning the delegation if necessary
867  *
868  * Returns zero on success, or a negative errno value.
869  */
870 int nfs4_inode_make_writeable(struct inode *inode)
871 {
872 	struct nfs_delegation *delegation;
873 
874 	rcu_read_lock();
875 	delegation = nfs4_get_valid_delegation(inode);
876 	if (delegation == NULL ||
877 	    (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
878 	     (delegation->type & FMODE_WRITE))) {
879 		rcu_read_unlock();
880 		return 0;
881 	}
882 	rcu_read_unlock();
883 	return nfs4_inode_return_delegation(inode);
884 }
885 
886 static void
887 nfs_mark_return_if_closed_delegation(struct nfs_server *server,
888 				     struct nfs_delegation *delegation)
889 {
890 	struct inode *inode;
891 
892 	if (test_bit(NFS_DELEGATION_RETURN, &delegation->flags) ||
893 	    test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags))
894 		return;
895 	spin_lock(&delegation->lock);
896 	inode = delegation->inode;
897 	if (!inode)
898 		goto out;
899 	if (list_empty(&NFS_I(inode)->open_files))
900 		nfs_mark_return_delegation(server, delegation);
901 	else
902 		set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
903 out:
904 	spin_unlock(&delegation->lock);
905 }
906 
907 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
908 {
909 	struct nfs_delegation *delegation;
910 	bool ret = false;
911 
912 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
913 		nfs_mark_return_delegation(server, delegation);
914 		ret = true;
915 	}
916 	return ret;
917 }
918 
919 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
920 {
921 	struct nfs_server *server;
922 
923 	rcu_read_lock();
924 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
925 		nfs_server_mark_return_all_delegations(server);
926 	rcu_read_unlock();
927 }
928 
929 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
930 {
931 	if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
932 		nfs4_schedule_state_manager(clp);
933 }
934 
935 /**
936  * nfs_expire_all_delegations
937  * @clp: client to process
938  *
939  */
940 void nfs_expire_all_delegations(struct nfs_client *clp)
941 {
942 	nfs_client_mark_return_all_delegations(clp);
943 	nfs_delegation_run_state_manager(clp);
944 }
945 
946 /**
947  * nfs_server_return_all_delegations - return delegations for one superblock
948  * @server: pointer to nfs_server to process
949  *
950  */
951 void nfs_server_return_all_delegations(struct nfs_server *server)
952 {
953 	struct nfs_client *clp = server->nfs_client;
954 	bool need_wait;
955 
956 	if (clp == NULL)
957 		return;
958 
959 	rcu_read_lock();
960 	need_wait = nfs_server_mark_return_all_delegations(server);
961 	rcu_read_unlock();
962 
963 	if (need_wait) {
964 		nfs4_schedule_state_manager(clp);
965 		nfs4_wait_clnt_recover(clp);
966 	}
967 }
968 
969 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
970 						 fmode_t flags)
971 {
972 	struct nfs_delegation *delegation;
973 
974 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
975 		if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
976 			continue;
977 		if (delegation->type & flags)
978 			nfs_mark_return_if_closed_delegation(server, delegation);
979 	}
980 }
981 
982 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
983 							fmode_t flags)
984 {
985 	struct nfs_server *server;
986 
987 	rcu_read_lock();
988 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
989 		nfs_mark_return_unused_delegation_types(server, flags);
990 	rcu_read_unlock();
991 }
992 
993 static void nfs_revoke_delegation(struct inode *inode,
994 		const nfs4_stateid *stateid)
995 {
996 	struct nfs_delegation *delegation;
997 	nfs4_stateid tmp;
998 	bool ret = false;
999 
1000 	rcu_read_lock();
1001 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1002 	if (delegation == NULL)
1003 		goto out;
1004 	if (stateid == NULL) {
1005 		nfs4_stateid_copy(&tmp, &delegation->stateid);
1006 		stateid = &tmp;
1007 	} else {
1008 		if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
1009 			goto out;
1010 		spin_lock(&delegation->lock);
1011 		if (stateid->seqid) {
1012 			if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
1013 				spin_unlock(&delegation->lock);
1014 				goto out;
1015 			}
1016 			delegation->stateid.seqid = stateid->seqid;
1017 		}
1018 		spin_unlock(&delegation->lock);
1019 	}
1020 	nfs_mark_delegation_revoked(delegation);
1021 	ret = true;
1022 out:
1023 	rcu_read_unlock();
1024 	if (ret)
1025 		nfs_inode_find_state_and_recover(inode, stateid);
1026 }
1027 
1028 void nfs_delegation_mark_returned(struct inode *inode,
1029 		const nfs4_stateid *stateid)
1030 {
1031 	struct nfs_delegation *delegation;
1032 
1033 	if (!inode)
1034 		return;
1035 
1036 	rcu_read_lock();
1037 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1038 	if (!delegation)
1039 		goto out_rcu_unlock;
1040 
1041 	spin_lock(&delegation->lock);
1042 	if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
1043 		goto out_spin_unlock;
1044 	if (stateid->seqid) {
1045 		/* If delegation->stateid is newer, dont mark as returned */
1046 		if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
1047 			goto out_clear_returning;
1048 		if (delegation->stateid.seqid != stateid->seqid)
1049 			delegation->stateid.seqid = stateid->seqid;
1050 	}
1051 
1052 	nfs_mark_delegation_revoked(delegation);
1053 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1054 	spin_unlock(&delegation->lock);
1055 	if (nfs_detach_delegation(NFS_I(inode), delegation, NFS_SERVER(inode)))
1056 		nfs_put_delegation(delegation);
1057 	goto out_rcu_unlock;
1058 
1059 out_clear_returning:
1060 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1061 out_spin_unlock:
1062 	spin_unlock(&delegation->lock);
1063 out_rcu_unlock:
1064 	rcu_read_unlock();
1065 
1066 	nfs_inode_find_state_and_recover(inode, stateid);
1067 }
1068 
1069 /**
1070  * nfs_remove_bad_delegation - handle delegations that are unusable
1071  * @inode: inode to process
1072  * @stateid: the delegation's stateid
1073  *
1074  * If the server ACK-ed our FREE_STATEID then clean
1075  * up the delegation, else mark and keep the revoked state.
1076  */
1077 void nfs_remove_bad_delegation(struct inode *inode,
1078 		const nfs4_stateid *stateid)
1079 {
1080 	if (stateid && stateid->type == NFS4_FREED_STATEID_TYPE)
1081 		nfs_delegation_mark_returned(inode, stateid);
1082 	else
1083 		nfs_revoke_delegation(inode, stateid);
1084 }
1085 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
1086 
1087 /**
1088  * nfs_expire_unused_delegation_types
1089  * @clp: client to process
1090  * @flags: delegation types to expire
1091  *
1092  */
1093 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1094 {
1095 	nfs_client_mark_return_unused_delegation_types(clp, flags);
1096 	nfs_delegation_run_state_manager(clp);
1097 }
1098 
1099 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1100 {
1101 	struct nfs_delegation *delegation;
1102 
1103 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1104 		if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1105 			continue;
1106 		nfs_mark_return_if_closed_delegation(server, delegation);
1107 	}
1108 }
1109 
1110 /**
1111  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1112  * @clp: nfs_client to process
1113  *
1114  */
1115 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1116 {
1117 	struct nfs_server *server;
1118 
1119 	rcu_read_lock();
1120 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1121 		nfs_mark_return_unreferenced_delegations(server);
1122 	rcu_read_unlock();
1123 
1124 	nfs_delegation_run_state_manager(clp);
1125 }
1126 
1127 /**
1128  * nfs_async_inode_return_delegation - asynchronously return a delegation
1129  * @inode: inode to process
1130  * @stateid: state ID information
1131  *
1132  * Returns zero on success, or a negative errno value.
1133  */
1134 int nfs_async_inode_return_delegation(struct inode *inode,
1135 				      const nfs4_stateid *stateid)
1136 {
1137 	struct nfs_server *server = NFS_SERVER(inode);
1138 	struct nfs_client *clp = server->nfs_client;
1139 	struct nfs_delegation *delegation;
1140 
1141 	rcu_read_lock();
1142 	delegation = nfs4_get_valid_delegation(inode);
1143 	if (delegation == NULL)
1144 		goto out_enoent;
1145 	if (stateid != NULL &&
1146 	    !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1147 		goto out_enoent;
1148 	nfs_mark_return_delegation(server, delegation);
1149 	rcu_read_unlock();
1150 
1151 	/* If there are any application leases or delegations, recall them */
1152 	break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1153 
1154 	nfs_delegation_run_state_manager(clp);
1155 	return 0;
1156 out_enoent:
1157 	rcu_read_unlock();
1158 	return -ENOENT;
1159 }
1160 
1161 static struct inode *
1162 nfs_delegation_find_inode_server(struct nfs_server *server,
1163 				 const struct nfs_fh *fhandle)
1164 {
1165 	struct nfs_delegation *delegation;
1166 	struct super_block *freeme = NULL;
1167 	struct inode *res = NULL;
1168 
1169 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1170 		spin_lock(&delegation->lock);
1171 		if (delegation->inode != NULL &&
1172 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1173 		    nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1174 			if (nfs_sb_active(server->super)) {
1175 				freeme = server->super;
1176 				res = igrab(delegation->inode);
1177 			}
1178 			spin_unlock(&delegation->lock);
1179 			if (res != NULL)
1180 				return res;
1181 			if (freeme) {
1182 				rcu_read_unlock();
1183 				nfs_sb_deactive(freeme);
1184 				rcu_read_lock();
1185 			}
1186 			return ERR_PTR(-EAGAIN);
1187 		}
1188 		spin_unlock(&delegation->lock);
1189 	}
1190 	return ERR_PTR(-ENOENT);
1191 }
1192 
1193 /**
1194  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1195  * @clp: client state handle
1196  * @fhandle: filehandle from a delegation recall
1197  *
1198  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1199  * cannot be found.
1200  */
1201 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1202 					const struct nfs_fh *fhandle)
1203 {
1204 	struct nfs_server *server;
1205 	struct inode *res;
1206 
1207 	rcu_read_lock();
1208 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1209 		res = nfs_delegation_find_inode_server(server, fhandle);
1210 		if (res != ERR_PTR(-ENOENT)) {
1211 			rcu_read_unlock();
1212 			return res;
1213 		}
1214 	}
1215 	rcu_read_unlock();
1216 	return ERR_PTR(-ENOENT);
1217 }
1218 
1219 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1220 {
1221 	struct nfs_delegation *delegation;
1222 
1223 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1224 		/*
1225 		 * If the delegation may have been admin revoked, then we
1226 		 * cannot reclaim it.
1227 		 */
1228 		if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1229 			continue;
1230 		set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1231 	}
1232 }
1233 
1234 /**
1235  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1236  * @clp: nfs_client to process
1237  *
1238  */
1239 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1240 {
1241 	struct nfs_server *server;
1242 
1243 	rcu_read_lock();
1244 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1245 		nfs_delegation_mark_reclaim_server(server);
1246 	rcu_read_unlock();
1247 }
1248 
1249 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1250 		void __always_unused *data)
1251 {
1252 	struct nfs_delegation *delegation;
1253 	struct inode *inode;
1254 restart:
1255 	rcu_read_lock();
1256 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1257 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1258 					&delegation->flags) ||
1259 		    test_bit(NFS_DELEGATION_RETURNING,
1260 					&delegation->flags) ||
1261 		    test_bit(NFS_DELEGATION_NEED_RECLAIM,
1262 					&delegation->flags) == 0)
1263 			continue;
1264 		inode = nfs_delegation_grab_inode(delegation);
1265 		if (inode == NULL)
1266 			continue;
1267 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1268 		rcu_read_unlock();
1269 		if (delegation != NULL) {
1270 			if (nfs_detach_delegation(NFS_I(inode), delegation,
1271 						server) != NULL)
1272 				nfs_free_delegation(delegation);
1273 			/* Match nfs_start_delegation_return_locked */
1274 			nfs_put_delegation(delegation);
1275 		}
1276 		iput(inode);
1277 		cond_resched();
1278 		goto restart;
1279 	}
1280 	rcu_read_unlock();
1281 	return 0;
1282 }
1283 
1284 /**
1285  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1286  * @clp: nfs_client to process
1287  *
1288  */
1289 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1290 {
1291 	nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1292 			NULL);
1293 }
1294 
1295 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1296 {
1297 	return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1298 				BIT(NFS4CLNT_LEASE_EXPIRED) |
1299 				BIT(NFS4CLNT_SESSION_RESET))) != 0;
1300 }
1301 
1302 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1303 	    struct nfs_delegation *delegation)
1304 {
1305 	if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1306 		return;
1307 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1308 	set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1309 	set_bit(NFS4SERV_DELEGATION_EXPIRED, &server->delegation_flags);
1310 	set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1311 }
1312 
1313 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1314 		struct inode *inode)
1315 {
1316 	struct nfs_delegation *delegation;
1317 
1318 	rcu_read_lock();
1319 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1320 	if (delegation)
1321 		nfs_mark_test_expired_delegation(server, delegation);
1322 	rcu_read_unlock();
1323 
1324 }
1325 
1326 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1327 {
1328 	struct nfs_delegation *delegation;
1329 
1330 	list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1331 		nfs_mark_test_expired_delegation(server, delegation);
1332 }
1333 
1334 /**
1335  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1336  * @clp: nfs_client to process
1337  *
1338  * Iterates through all the delegations associated with this server and
1339  * marks them as needing to be checked for validity.
1340  */
1341 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1342 {
1343 	struct nfs_server *server;
1344 
1345 	rcu_read_lock();
1346 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1347 		nfs_delegation_mark_test_expired_server(server);
1348 	rcu_read_unlock();
1349 }
1350 
1351 /**
1352  * nfs_test_expired_all_delegations - test all delegations for a client
1353  * @clp: nfs_client to process
1354  *
1355  * Helper for handling "recallable state revoked" status from server.
1356  */
1357 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1358 {
1359 	nfs_mark_test_expired_all_delegations(clp);
1360 	nfs4_schedule_state_manager(clp);
1361 }
1362 
1363 static void
1364 nfs_delegation_test_free_expired(struct inode *inode,
1365 		nfs4_stateid *stateid,
1366 		const struct cred *cred)
1367 {
1368 	struct nfs_server *server = NFS_SERVER(inode);
1369 	const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1370 	int status;
1371 
1372 	if (!cred)
1373 		return;
1374 	status = ops->test_and_free_expired(server, stateid, cred);
1375 	if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1376 		nfs_remove_bad_delegation(inode, stateid);
1377 }
1378 
1379 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1380 		void __always_unused *data)
1381 {
1382 	struct nfs_delegation *delegation;
1383 	struct inode *inode;
1384 	const struct cred *cred;
1385 	nfs4_stateid stateid;
1386 	unsigned long gen = ++server->delegation_gen;
1387 
1388 	if (!test_and_clear_bit(NFS4SERV_DELEGATION_EXPIRED,
1389 				&server->delegation_flags))
1390 		return 0;
1391 restart:
1392 	rcu_read_lock();
1393 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1394 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1395 					&delegation->flags) ||
1396 		    test_bit(NFS_DELEGATION_RETURNING,
1397 					&delegation->flags) ||
1398 		    test_bit(NFS_DELEGATION_TEST_EXPIRED,
1399 					&delegation->flags) == 0 ||
1400 			delegation->test_gen == gen)
1401 			continue;
1402 		inode = nfs_delegation_grab_inode(delegation);
1403 		if (inode == NULL)
1404 			continue;
1405 		spin_lock(&delegation->lock);
1406 		cred = get_cred_rcu(delegation->cred);
1407 		nfs4_stateid_copy(&stateid, &delegation->stateid);
1408 		spin_unlock(&delegation->lock);
1409 		delegation->test_gen = gen;
1410 		clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1411 		rcu_read_unlock();
1412 		nfs_delegation_test_free_expired(inode, &stateid, cred);
1413 		put_cred(cred);
1414 		if (!nfs4_server_rebooted(server->nfs_client)) {
1415 			iput(inode);
1416 			cond_resched();
1417 			goto restart;
1418 		}
1419 		nfs_inode_mark_test_expired_delegation(server,inode);
1420 		set_bit(NFS4SERV_DELEGATION_EXPIRED, &server->delegation_flags);
1421 		set_bit(NFS4CLNT_DELEGATION_EXPIRED,
1422 			&server->nfs_client->cl_state);
1423 		iput(inode);
1424 		return -EAGAIN;
1425 	}
1426 	rcu_read_unlock();
1427 	return 0;
1428 }
1429 
1430 /**
1431  * nfs_reap_expired_delegations - reap expired delegations
1432  * @clp: nfs_client to process
1433  *
1434  * Iterates through all the delegations associated with this server and
1435  * checks if they have may have been revoked. This function is usually
1436  * expected to be called in cases where the server may have lost its
1437  * lease.
1438  */
1439 void nfs_reap_expired_delegations(struct nfs_client *clp)
1440 {
1441 	nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1442 			NULL);
1443 }
1444 
1445 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1446 		const nfs4_stateid *stateid)
1447 {
1448 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1449 	struct nfs_delegation *delegation;
1450 	bool found = false;
1451 
1452 	rcu_read_lock();
1453 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1454 	if (delegation &&
1455 	    nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1456 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1457 		nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1458 		found = true;
1459 	}
1460 	rcu_read_unlock();
1461 	if (found)
1462 		nfs4_schedule_state_manager(clp);
1463 }
1464 
1465 /**
1466  * nfs_delegations_present - check for existence of delegations
1467  * @clp: client state handle
1468  *
1469  * Returns one if there are any nfs_delegation structures attached
1470  * to this nfs_client.
1471  */
1472 int nfs_delegations_present(struct nfs_client *clp)
1473 {
1474 	struct nfs_server *server;
1475 	int ret = 0;
1476 
1477 	rcu_read_lock();
1478 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1479 		if (!list_empty(&server->delegations)) {
1480 			ret = 1;
1481 			break;
1482 		}
1483 	rcu_read_unlock();
1484 	return ret;
1485 }
1486 
1487 /**
1488  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1489  * @dst: stateid to refresh
1490  * @inode: inode to check
1491  *
1492  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1493  * that matches our delegation stateid. Otherwise "false" is returned.
1494  */
1495 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1496 {
1497 	struct nfs_delegation *delegation;
1498 	bool ret = false;
1499 	if (!inode)
1500 		goto out;
1501 
1502 	rcu_read_lock();
1503 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1504 	if (delegation != NULL &&
1505 	    nfs4_stateid_match_other(dst, &delegation->stateid) &&
1506 	    nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1507 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1508 		dst->seqid = delegation->stateid.seqid;
1509 		ret = true;
1510 	}
1511 	rcu_read_unlock();
1512 out:
1513 	return ret;
1514 }
1515 
1516 /**
1517  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1518  * @inode: inode to check
1519  * @flags: delegation type requirement
1520  * @dst: stateid data structure to fill in
1521  * @cred: optional argument to retrieve credential
1522  *
1523  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1524  * otherwise "false" is returned.
1525  */
1526 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1527 		nfs4_stateid *dst, const struct cred **cred)
1528 {
1529 	struct nfs_inode *nfsi = NFS_I(inode);
1530 	struct nfs_delegation *delegation;
1531 	bool ret = false;
1532 
1533 	flags &= FMODE_READ|FMODE_WRITE;
1534 	rcu_read_lock();
1535 	delegation = rcu_dereference(nfsi->delegation);
1536 	if (!delegation)
1537 		goto out;
1538 	spin_lock(&delegation->lock);
1539 	ret = nfs4_is_valid_delegation(delegation, flags);
1540 	if (ret) {
1541 		nfs4_stateid_copy(dst, &delegation->stateid);
1542 		nfs_mark_delegation_referenced(delegation);
1543 		if (cred)
1544 			*cred = get_cred(delegation->cred);
1545 	}
1546 	spin_unlock(&delegation->lock);
1547 out:
1548 	rcu_read_unlock();
1549 	return ret;
1550 }
1551 
1552 /**
1553  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1554  * @inode: inode to check
1555  *
1556  * This function checks the number of outstanding writes to the file
1557  * against the delegation 'space_limit' field to see if
1558  * the spec requires us to flush the file on close.
1559  */
1560 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1561 {
1562 	struct nfs_inode *nfsi = NFS_I(inode);
1563 	struct nfs_delegation *delegation;
1564 	bool ret = true;
1565 
1566 	rcu_read_lock();
1567 	delegation = rcu_dereference(nfsi->delegation);
1568 	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1569 		goto out;
1570 	if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1571 		ret = false;
1572 out:
1573 	rcu_read_unlock();
1574 	return ret;
1575 }
1576 
1577 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);
1578