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 
__nfs_free_delegation(struct nfs_delegation * delegation)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 
nfs_mark_delegation_revoked(struct nfs_delegation * delegation)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 
nfs_get_delegation(struct nfs_delegation * delegation)50 static struct nfs_delegation *nfs_get_delegation(struct nfs_delegation *delegation)
51 {
52 	refcount_inc(&delegation->refcount);
53 	return delegation;
54 }
55 
nfs_put_delegation(struct nfs_delegation * delegation)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 
nfs_free_delegation(struct nfs_delegation * delegation)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  */
nfs_mark_delegation_referenced(struct nfs_delegation * delegation)73 void nfs_mark_delegation_referenced(struct nfs_delegation *delegation)
74 {
75 	set_bit(NFS_DELEGATION_REFERENCED, &delegation->flags);
76 }
77 
nfs_mark_return_delegation(struct nfs_server * server,struct nfs_delegation * delegation)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 
nfs4_is_valid_delegation(const struct nfs_delegation * delegation,fmode_t type)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 
nfs4_get_valid_delegation(const struct inode * inode)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 
nfs4_do_check_delegation(struct inode * inode,fmode_t type,int flags,bool mark)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  */
nfs4_have_delegation(struct inode * inode,fmode_t type,int flags)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  */
nfs4_check_delegation(struct inode * inode,fmode_t type)144 int nfs4_check_delegation(struct inode *inode, fmode_t type)
145 {
146 	return nfs4_do_check_delegation(inode, type, 0, false);
147 }
148 
nfs_delegation_claim_locks(struct nfs4_state * state,const nfs4_stateid * stateid)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 
nfs_delegation_claim_opens(struct inode * inode,const nfs4_stateid * stateid,fmode_t type)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  */
nfs_inode_reclaim_delegation(struct inode * inode,const struct cred * cred,fmode_t type,const nfs4_stateid * stateid,unsigned long pagemod_limit,u32 deleg_type)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 != NULL) {
241 		spin_lock(&delegation->lock);
242 		nfs4_stateid_copy(&delegation->stateid, stateid);
243 		delegation->type = type;
244 		delegation->pagemod_limit = pagemod_limit;
245 		oldcred = delegation->cred;
246 		delegation->cred = get_cred(cred);
247 		switch (deleg_type) {
248 		case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
249 		case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
250 			set_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
251 			break;
252 		default:
253 			clear_bit(NFS_DELEGATION_DELEGTIME, &delegation->flags);
254 		}
255 		clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
256 		if (test_and_clear_bit(NFS_DELEGATION_REVOKED,
257 				       &delegation->flags))
258 			atomic_long_inc(&nfs_active_delegations);
259 		spin_unlock(&delegation->lock);
260 		rcu_read_unlock();
261 		put_cred(oldcred);
262 		trace_nfs4_reclaim_delegation(inode, type);
263 	} else {
264 		rcu_read_unlock();
265 		nfs_inode_set_delegation(inode, cred, type, stateid,
266 					 pagemod_limit, deleg_type);
267 	}
268 }
269 
nfs_do_return_delegation(struct inode * inode,struct nfs_delegation * delegation,int issync)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 
nfs_delegation_grab_inode(struct nfs_delegation * delegation)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 *
nfs_start_delegation_return_locked(struct nfs_inode * nfsi)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 (!test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
311 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags);
312 		/* Refcount matched in nfs_end_delegation_return() */
313 		ret = nfs_get_delegation(delegation);
314 	}
315 	spin_unlock(&delegation->lock);
316 	if (ret)
317 		nfs_clear_verifier_delegated(&nfsi->vfs_inode);
318 out:
319 	return ret;
320 }
321 
322 static struct nfs_delegation *
nfs_start_delegation_return(struct nfs_inode * nfsi)323 nfs_start_delegation_return(struct nfs_inode *nfsi)
324 {
325 	struct nfs_delegation *delegation;
326 
327 	rcu_read_lock();
328 	delegation = nfs_start_delegation_return_locked(nfsi);
329 	rcu_read_unlock();
330 	return delegation;
331 }
332 
nfs_abort_delegation_return(struct nfs_delegation * delegation,struct nfs_server * server,int err)333 static void nfs_abort_delegation_return(struct nfs_delegation *delegation,
334 					struct nfs_server *server, int err)
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(NFS4SERV_DELEGRETURN_DELAYED,
341 			&server->delegation_flags);
342 		set_bit(NFS4CLNT_DELEGRETURN_DELAYED,
343 			&server->nfs_client->cl_state);
344 	}
345 	spin_unlock(&delegation->lock);
346 }
347 
348 static struct nfs_delegation *
nfs_detach_delegation_locked(struct nfs_inode * nfsi,struct nfs_delegation * delegation,struct nfs_client * clp)349 nfs_detach_delegation_locked(struct nfs_inode *nfsi,
350 		struct nfs_delegation *delegation,
351 		struct nfs_client *clp)
352 {
353 	struct nfs_delegation *deleg_cur =
354 		rcu_dereference_protected(nfsi->delegation,
355 				lockdep_is_held(&clp->cl_lock));
356 
357 	if (deleg_cur == NULL || delegation != deleg_cur)
358 		return NULL;
359 
360 	spin_lock(&delegation->lock);
361 	if (!delegation->inode) {
362 		spin_unlock(&delegation->lock);
363 		return NULL;
364 	}
365 	list_del_rcu(&delegation->super_list);
366 	delegation->inode = NULL;
367 	rcu_assign_pointer(nfsi->delegation, NULL);
368 	spin_unlock(&delegation->lock);
369 	return delegation;
370 }
371 
nfs_detach_delegation(struct nfs_inode * nfsi,struct nfs_delegation * delegation,struct nfs_server * server)372 static struct nfs_delegation *nfs_detach_delegation(struct nfs_inode *nfsi,
373 		struct nfs_delegation *delegation,
374 		struct nfs_server *server)
375 {
376 	struct nfs_client *clp = server->nfs_client;
377 
378 	spin_lock(&clp->cl_lock);
379 	delegation = nfs_detach_delegation_locked(nfsi, delegation, clp);
380 	spin_unlock(&clp->cl_lock);
381 	return delegation;
382 }
383 
384 static struct nfs_delegation *
nfs_inode_detach_delegation(struct inode * inode)385 nfs_inode_detach_delegation(struct inode *inode)
386 {
387 	struct nfs_inode *nfsi = NFS_I(inode);
388 	struct nfs_server *server = NFS_SERVER(inode);
389 	struct nfs_delegation *delegation;
390 
391 	rcu_read_lock();
392 	delegation = rcu_dereference(nfsi->delegation);
393 	if (delegation != NULL)
394 		delegation = nfs_detach_delegation(nfsi, delegation, server);
395 	rcu_read_unlock();
396 	return delegation;
397 }
398 
399 static void
nfs_update_delegation_cred(struct nfs_delegation * delegation,const struct cred * cred)400 nfs_update_delegation_cred(struct nfs_delegation *delegation,
401 		const struct cred *cred)
402 {
403 	const struct cred *old;
404 
405 	if (cred_fscmp(delegation->cred, cred) != 0) {
406 		old = xchg(&delegation->cred, get_cred(cred));
407 		put_cred(old);
408 	}
409 }
410 
411 static void
nfs_update_inplace_delegation(struct nfs_delegation * delegation,const struct nfs_delegation * update)412 nfs_update_inplace_delegation(struct nfs_delegation *delegation,
413 		const struct nfs_delegation *update)
414 {
415 	if (nfs4_stateid_is_newer(&update->stateid, &delegation->stateid)) {
416 		delegation->stateid.seqid = update->stateid.seqid;
417 		smp_wmb();
418 		delegation->type = update->type;
419 		delegation->pagemod_limit = update->pagemod_limit;
420 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
421 			delegation->change_attr = update->change_attr;
422 			nfs_update_delegation_cred(delegation, update->cred);
423 			/* smp_mb__before_atomic() is implicit due to xchg() */
424 			clear_bit(NFS_DELEGATION_REVOKED, &delegation->flags);
425 			atomic_long_inc(&nfs_active_delegations);
426 		}
427 	}
428 }
429 
430 /**
431  * nfs_inode_set_delegation - set up a delegation on an inode
432  * @inode: inode to which delegation applies
433  * @cred: cred to use for subsequent delegation processing
434  * @type: delegation type
435  * @stateid: delegation stateid
436  * @pagemod_limit: write delegation "space_limit"
437  * @deleg_type: raw delegation type
438  *
439  * Returns zero on success, or a negative errno value.
440  */
nfs_inode_set_delegation(struct inode * inode,const struct cred * cred,fmode_t type,const nfs4_stateid * stateid,unsigned long pagemod_limit,u32 deleg_type)441 int nfs_inode_set_delegation(struct inode *inode, const struct cred *cred,
442 			     fmode_t type, const nfs4_stateid *stateid,
443 			     unsigned long pagemod_limit, u32 deleg_type)
444 {
445 	struct nfs_server *server = NFS_SERVER(inode);
446 	struct nfs_client *clp = server->nfs_client;
447 	struct nfs_inode *nfsi = NFS_I(inode);
448 	struct nfs_delegation *delegation, *old_delegation;
449 	struct nfs_delegation *freeme = NULL;
450 	int status = 0;
451 
452 	delegation = kmalloc(sizeof(*delegation), GFP_KERNEL_ACCOUNT);
453 	if (delegation == NULL)
454 		return -ENOMEM;
455 	nfs4_stateid_copy(&delegation->stateid, stateid);
456 	refcount_set(&delegation->refcount, 1);
457 	delegation->type = type;
458 	delegation->pagemod_limit = pagemod_limit;
459 	delegation->change_attr = inode_peek_iversion_raw(inode);
460 	delegation->cred = get_cred(cred);
461 	delegation->inode = inode;
462 	delegation->flags = 1<<NFS_DELEGATION_REFERENCED;
463 	switch (deleg_type) {
464 	case NFS4_OPEN_DELEGATE_READ_ATTRS_DELEG:
465 	case NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG:
466 		delegation->flags |= BIT(NFS_DELEGATION_DELEGTIME);
467 	}
468 	delegation->test_gen = 0;
469 	spin_lock_init(&delegation->lock);
470 
471 	spin_lock(&clp->cl_lock);
472 	old_delegation = rcu_dereference_protected(nfsi->delegation,
473 					lockdep_is_held(&clp->cl_lock));
474 	if (old_delegation == NULL)
475 		goto add_new;
476 	/* Is this an update of the existing delegation? */
477 	if (nfs4_stateid_match_other(&old_delegation->stateid,
478 				&delegation->stateid)) {
479 		spin_lock(&old_delegation->lock);
480 		nfs_update_inplace_delegation(old_delegation,
481 				delegation);
482 		spin_unlock(&old_delegation->lock);
483 		goto out;
484 	}
485 	if (!test_bit(NFS_DELEGATION_REVOKED, &old_delegation->flags)) {
486 		/*
487 		 * Deal with broken servers that hand out two
488 		 * delegations for the same file.
489 		 * Allow for upgrades to a WRITE delegation, but
490 		 * nothing else.
491 		 */
492 		dfprintk(FILE, "%s: server %s handed out "
493 				"a duplicate delegation!\n",
494 				__func__, clp->cl_hostname);
495 		if (delegation->type == old_delegation->type ||
496 		    !(delegation->type & FMODE_WRITE)) {
497 			freeme = delegation;
498 			delegation = NULL;
499 			goto out;
500 		}
501 		if (test_and_set_bit(NFS_DELEGATION_RETURNING,
502 					&old_delegation->flags))
503 			goto out;
504 	}
505 	freeme = nfs_detach_delegation_locked(nfsi, old_delegation, clp);
506 	if (freeme == NULL)
507 		goto out;
508 add_new:
509 	/*
510 	 * If we didn't revalidate the change attribute before setting
511 	 * the delegation, then pre-emptively ask for a full attribute
512 	 * cache revalidation.
513 	 */
514 	spin_lock(&inode->i_lock);
515 	if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_CHANGE)
516 		nfs_set_cache_invalid(inode,
517 			NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
518 			NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
519 			NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_NLINK |
520 			NFS_INO_INVALID_OTHER | NFS_INO_INVALID_DATA |
521 			NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL |
522 			NFS_INO_INVALID_XATTR);
523 	spin_unlock(&inode->i_lock);
524 
525 	list_add_tail_rcu(&delegation->super_list, &server->delegations);
526 	rcu_assign_pointer(nfsi->delegation, delegation);
527 	delegation = NULL;
528 
529 	atomic_long_inc(&nfs_active_delegations);
530 
531 	trace_nfs4_set_delegation(inode, type);
532 
533 	/* If we hold writebacks and have delegated mtime then update */
534 	if (deleg_type == NFS4_OPEN_DELEGATE_WRITE_ATTRS_DELEG &&
535 	    nfs_have_writebacks(inode))
536 		nfs_update_delegated_mtime(inode);
537 out:
538 	spin_unlock(&clp->cl_lock);
539 	if (delegation != NULL)
540 		__nfs_free_delegation(delegation);
541 	if (freeme != NULL) {
542 		nfs_do_return_delegation(inode, freeme, 0);
543 		nfs_free_delegation(freeme);
544 	}
545 	return status;
546 }
547 
548 /*
549  * Basic procedure for returning a delegation to the server
550  */
nfs_end_delegation_return(struct inode * inode,struct nfs_delegation * delegation,int issync)551 static int nfs_end_delegation_return(struct inode *inode, struct nfs_delegation *delegation, int issync)
552 {
553 	struct nfs_server *server = NFS_SERVER(inode);
554 	unsigned int mode = O_WRONLY | O_RDWR;
555 	int err = 0;
556 
557 	if (delegation == NULL)
558 		return 0;
559 
560 	if (!issync)
561 		mode |= O_NONBLOCK;
562 	/* Recall of any remaining application leases */
563 	err = break_lease(inode, mode);
564 
565 	while (err == 0) {
566 		if (test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
567 			break;
568 		err = nfs_delegation_claim_opens(inode, &delegation->stateid,
569 				delegation->type);
570 		if (!issync || err != -EAGAIN)
571 			break;
572 		/*
573 		 * Guard against state recovery
574 		 */
575 		err = nfs4_wait_clnt_recover(server->nfs_client);
576 	}
577 
578 	if (err) {
579 		nfs_abort_delegation_return(delegation, server, err);
580 		goto out;
581 	}
582 
583 	err = nfs_do_return_delegation(inode, delegation, issync);
584 out:
585 	/* Refcount matched in nfs_start_delegation_return_locked() */
586 	nfs_put_delegation(delegation);
587 	return err;
588 }
589 
nfs_delegation_need_return(struct nfs_delegation * delegation)590 static bool nfs_delegation_need_return(struct nfs_delegation *delegation)
591 {
592 	bool ret = false;
593 
594 	if (test_and_clear_bit(NFS_DELEGATION_RETURN, &delegation->flags))
595 		ret = true;
596 	if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags) ||
597 	    test_bit(NFS_DELEGATION_RETURN_DELAYED, &delegation->flags) ||
598 	    test_bit(NFS_DELEGATION_REVOKED, &delegation->flags))
599 		ret = false;
600 
601 	return ret;
602 }
603 
nfs_server_return_marked_delegations(struct nfs_server * server,void __always_unused * data)604 static int nfs_server_return_marked_delegations(struct nfs_server *server,
605 		void __always_unused *data)
606 {
607 	struct nfs_delegation *delegation;
608 	struct nfs_delegation *prev;
609 	struct inode *inode;
610 	struct inode *place_holder = NULL;
611 	struct nfs_delegation *place_holder_deleg = NULL;
612 	int err = 0;
613 
614 	if (!test_and_clear_bit(NFS4SERV_DELEGRETURN,
615 				&server->delegation_flags))
616 		return 0;
617 restart:
618 	/*
619 	 * To avoid quadratic looping we hold a reference
620 	 * to an inode place_holder.  Each time we restart, we
621 	 * list delegation in the server from the delegations
622 	 * of that inode.
623 	 * prev is an RCU-protected pointer to a delegation which
624 	 * wasn't marked for return and might be a good choice for
625 	 * the next place_holder.
626 	 */
627 	prev = NULL;
628 	delegation = NULL;
629 	rcu_read_lock();
630 	if (place_holder)
631 		delegation = rcu_dereference(NFS_I(place_holder)->delegation);
632 	if (!delegation || delegation != place_holder_deleg)
633 		delegation = list_entry_rcu(server->delegations.next,
634 					    struct nfs_delegation, super_list);
635 	list_for_each_entry_from_rcu(delegation, &server->delegations, super_list) {
636 		struct inode *to_put = NULL;
637 
638 		if (test_bit(NFS_DELEGATION_INODE_FREEING, &delegation->flags))
639 			continue;
640 		if (!nfs_delegation_need_return(delegation)) {
641 			if (nfs4_is_valid_delegation(delegation, 0))
642 				prev = delegation;
643 			continue;
644 		}
645 		inode = nfs_delegation_grab_inode(delegation);
646 		if (inode == NULL)
647 			continue;
648 
649 		if (prev) {
650 			struct inode *tmp = nfs_delegation_grab_inode(prev);
651 			if (tmp) {
652 				to_put = place_holder;
653 				place_holder = tmp;
654 				place_holder_deleg = prev;
655 			}
656 		}
657 
658 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
659 		rcu_read_unlock();
660 
661 		iput(to_put);
662 
663 		err = nfs_end_delegation_return(inode, delegation, 0);
664 		iput(inode);
665 		cond_resched();
666 		if (!err)
667 			goto restart;
668 		set_bit(NFS4SERV_DELEGRETURN, &server->delegation_flags);
669 		set_bit(NFS4CLNT_DELEGRETURN, &server->nfs_client->cl_state);
670 		goto out;
671 	}
672 	rcu_read_unlock();
673 out:
674 	iput(place_holder);
675 	return err;
676 }
677 
nfs_server_clear_delayed_delegations(struct nfs_server * server)678 static bool nfs_server_clear_delayed_delegations(struct nfs_server *server)
679 {
680 	struct nfs_delegation *d;
681 	bool ret = false;
682 
683 	if (!test_and_clear_bit(NFS4SERV_DELEGRETURN_DELAYED,
684 				&server->delegation_flags))
685 		goto out;
686 	list_for_each_entry_rcu (d, &server->delegations, super_list) {
687 		if (!test_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags))
688 			continue;
689 		nfs_mark_return_delegation(server, d);
690 		clear_bit(NFS_DELEGATION_RETURN_DELAYED, &d->flags);
691 		ret = true;
692 	}
693 out:
694 	return ret;
695 }
696 
nfs_client_clear_delayed_delegations(struct nfs_client * clp)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  */
nfs_client_return_marked_delegations(struct nfs_client * clp)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  */
nfs_inode_evict_delegation(struct inode * inode)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  */
nfs4_inode_return_delegation(struct inode * inode)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_set_return_delegation_on_close - asynchronously return a delegation
785  * @inode: inode to process
786  *
787  * This routine is called to request that the delegation be returned as soon
788  * as the file is closed. If the file is already closed, the delegation is
789  * immediately returned.
790  */
nfs4_inode_set_return_delegation_on_close(struct inode * inode)791 void nfs4_inode_set_return_delegation_on_close(struct inode *inode)
792 {
793 	struct nfs_delegation *delegation;
794 	struct nfs_delegation *ret = NULL;
795 
796 	if (!inode)
797 		return;
798 	rcu_read_lock();
799 	delegation = nfs4_get_valid_delegation(inode);
800 	if (!delegation)
801 		goto out;
802 	spin_lock(&delegation->lock);
803 	if (!delegation->inode)
804 		goto out_unlock;
805 	if (list_empty(&NFS_I(inode)->open_files) &&
806 	    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
807 		/* Refcount matched in nfs_end_delegation_return() */
808 		ret = nfs_get_delegation(delegation);
809 	} else
810 		set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
811 out_unlock:
812 	spin_unlock(&delegation->lock);
813 	if (ret)
814 		nfs_clear_verifier_delegated(inode);
815 out:
816 	rcu_read_unlock();
817 	nfs_end_delegation_return(inode, ret, 0);
818 }
819 
820 /**
821  * nfs4_inode_return_delegation_on_close - asynchronously return a delegation
822  * @inode: inode to process
823  *
824  * This routine is called on file close in order to determine if the
825  * inode delegation needs to be returned immediately.
826  */
nfs4_inode_return_delegation_on_close(struct inode * inode)827 void nfs4_inode_return_delegation_on_close(struct inode *inode)
828 {
829 	struct nfs_delegation *delegation;
830 	struct nfs_delegation *ret = NULL;
831 
832 	if (!inode)
833 		return;
834 	rcu_read_lock();
835 	delegation = nfs4_get_valid_delegation(inode);
836 	if (!delegation)
837 		goto out;
838 	if (test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags) ||
839 	    atomic_long_read(&nfs_active_delegations) >= nfs_delegation_watermark) {
840 		spin_lock(&delegation->lock);
841 		if (delegation->inode &&
842 		    list_empty(&NFS_I(inode)->open_files) &&
843 		    !test_and_set_bit(NFS_DELEGATION_RETURNING, &delegation->flags)) {
844 			clear_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
845 			/* Refcount matched in nfs_end_delegation_return() */
846 			ret = nfs_get_delegation(delegation);
847 		}
848 		spin_unlock(&delegation->lock);
849 		if (ret)
850 			nfs_clear_verifier_delegated(inode);
851 	}
852 out:
853 	rcu_read_unlock();
854 	nfs_end_delegation_return(inode, ret, 0);
855 }
856 
857 /**
858  * nfs4_inode_make_writeable
859  * @inode: pointer to inode
860  *
861  * Make the inode writeable by returning the delegation if necessary
862  *
863  * Returns zero on success, or a negative errno value.
864  */
nfs4_inode_make_writeable(struct inode * inode)865 int nfs4_inode_make_writeable(struct inode *inode)
866 {
867 	struct nfs_delegation *delegation;
868 
869 	rcu_read_lock();
870 	delegation = nfs4_get_valid_delegation(inode);
871 	if (delegation == NULL ||
872 	    (nfs4_has_session(NFS_SERVER(inode)->nfs_client) &&
873 	     (delegation->type & FMODE_WRITE))) {
874 		rcu_read_unlock();
875 		return 0;
876 	}
877 	rcu_read_unlock();
878 	return nfs4_inode_return_delegation(inode);
879 }
880 
881 static void
nfs_mark_return_if_closed_delegation(struct nfs_server * server,struct nfs_delegation * delegation)882 nfs_mark_return_if_closed_delegation(struct nfs_server *server,
883 				     struct nfs_delegation *delegation)
884 {
885 	struct inode *inode;
886 
887 	if (test_bit(NFS_DELEGATION_RETURN, &delegation->flags) ||
888 	    test_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags))
889 		return;
890 	spin_lock(&delegation->lock);
891 	inode = delegation->inode;
892 	if (!inode)
893 		goto out;
894 	if (list_empty(&NFS_I(inode)->open_files))
895 		nfs_mark_return_delegation(server, delegation);
896 	else
897 		set_bit(NFS_DELEGATION_RETURN_IF_CLOSED, &delegation->flags);
898 out:
899 	spin_unlock(&delegation->lock);
900 }
901 
nfs_server_mark_return_all_delegations(struct nfs_server * server)902 static bool nfs_server_mark_return_all_delegations(struct nfs_server *server)
903 {
904 	struct nfs_delegation *delegation;
905 	bool ret = false;
906 
907 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
908 		nfs_mark_return_delegation(server, delegation);
909 		ret = true;
910 	}
911 	return ret;
912 }
913 
nfs_client_mark_return_all_delegations(struct nfs_client * clp)914 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)
915 {
916 	struct nfs_server *server;
917 
918 	rcu_read_lock();
919 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
920 		nfs_server_mark_return_all_delegations(server);
921 	rcu_read_unlock();
922 }
923 
nfs_delegation_run_state_manager(struct nfs_client * clp)924 static void nfs_delegation_run_state_manager(struct nfs_client *clp)
925 {
926 	if (test_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state))
927 		nfs4_schedule_state_manager(clp);
928 }
929 
930 /**
931  * nfs_expire_all_delegations
932  * @clp: client to process
933  *
934  */
nfs_expire_all_delegations(struct nfs_client * clp)935 void nfs_expire_all_delegations(struct nfs_client *clp)
936 {
937 	nfs_client_mark_return_all_delegations(clp);
938 	nfs_delegation_run_state_manager(clp);
939 }
940 
941 /**
942  * nfs_server_return_all_delegations - return delegations for one superblock
943  * @server: pointer to nfs_server to process
944  *
945  */
nfs_server_return_all_delegations(struct nfs_server * server)946 void nfs_server_return_all_delegations(struct nfs_server *server)
947 {
948 	struct nfs_client *clp = server->nfs_client;
949 	bool need_wait;
950 
951 	if (clp == NULL)
952 		return;
953 
954 	rcu_read_lock();
955 	need_wait = nfs_server_mark_return_all_delegations(server);
956 	rcu_read_unlock();
957 
958 	if (need_wait) {
959 		nfs4_schedule_state_manager(clp);
960 		nfs4_wait_clnt_recover(clp);
961 	}
962 }
963 
nfs_mark_return_unused_delegation_types(struct nfs_server * server,fmode_t flags)964 static void nfs_mark_return_unused_delegation_types(struct nfs_server *server,
965 						 fmode_t flags)
966 {
967 	struct nfs_delegation *delegation;
968 
969 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
970 		if ((delegation->type == (FMODE_READ|FMODE_WRITE)) && !(flags & FMODE_WRITE))
971 			continue;
972 		if (delegation->type & flags)
973 			nfs_mark_return_if_closed_delegation(server, delegation);
974 	}
975 }
976 
nfs_client_mark_return_unused_delegation_types(struct nfs_client * clp,fmode_t flags)977 static void nfs_client_mark_return_unused_delegation_types(struct nfs_client *clp,
978 							fmode_t flags)
979 {
980 	struct nfs_server *server;
981 
982 	rcu_read_lock();
983 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
984 		nfs_mark_return_unused_delegation_types(server, flags);
985 	rcu_read_unlock();
986 }
987 
nfs_revoke_delegation(struct inode * inode,const nfs4_stateid * stateid)988 static void nfs_revoke_delegation(struct inode *inode,
989 		const nfs4_stateid *stateid)
990 {
991 	struct nfs_delegation *delegation;
992 	nfs4_stateid tmp;
993 	bool ret = false;
994 
995 	rcu_read_lock();
996 	delegation = rcu_dereference(NFS_I(inode)->delegation);
997 	if (delegation == NULL)
998 		goto out;
999 	if (stateid == NULL) {
1000 		nfs4_stateid_copy(&tmp, &delegation->stateid);
1001 		stateid = &tmp;
1002 	} else {
1003 		if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
1004 			goto out;
1005 		spin_lock(&delegation->lock);
1006 		if (stateid->seqid) {
1007 			if (nfs4_stateid_is_newer(&delegation->stateid, stateid)) {
1008 				spin_unlock(&delegation->lock);
1009 				goto out;
1010 			}
1011 			delegation->stateid.seqid = stateid->seqid;
1012 		}
1013 		spin_unlock(&delegation->lock);
1014 	}
1015 	nfs_mark_delegation_revoked(delegation);
1016 	ret = true;
1017 out:
1018 	rcu_read_unlock();
1019 	if (ret)
1020 		nfs_inode_find_state_and_recover(inode, stateid);
1021 }
1022 
nfs_remove_bad_delegation(struct inode * inode,const nfs4_stateid * stateid)1023 void nfs_remove_bad_delegation(struct inode *inode,
1024 		const nfs4_stateid *stateid)
1025 {
1026 	nfs_revoke_delegation(inode, stateid);
1027 }
1028 EXPORT_SYMBOL_GPL(nfs_remove_bad_delegation);
1029 
nfs_delegation_mark_returned(struct inode * inode,const nfs4_stateid * stateid)1030 void nfs_delegation_mark_returned(struct inode *inode,
1031 		const nfs4_stateid *stateid)
1032 {
1033 	struct nfs_delegation *delegation;
1034 
1035 	if (!inode)
1036 		return;
1037 
1038 	rcu_read_lock();
1039 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1040 	if (!delegation)
1041 		goto out_rcu_unlock;
1042 
1043 	spin_lock(&delegation->lock);
1044 	if (!nfs4_stateid_match_other(stateid, &delegation->stateid))
1045 		goto out_spin_unlock;
1046 	if (stateid->seqid) {
1047 		/* If delegation->stateid is newer, dont mark as returned */
1048 		if (nfs4_stateid_is_newer(&delegation->stateid, stateid))
1049 			goto out_clear_returning;
1050 		if (delegation->stateid.seqid != stateid->seqid)
1051 			delegation->stateid.seqid = stateid->seqid;
1052 	}
1053 
1054 	nfs_mark_delegation_revoked(delegation);
1055 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1056 	spin_unlock(&delegation->lock);
1057 	if (nfs_detach_delegation(NFS_I(inode), delegation, NFS_SERVER(inode)))
1058 		nfs_put_delegation(delegation);
1059 	goto out_rcu_unlock;
1060 
1061 out_clear_returning:
1062 	clear_bit(NFS_DELEGATION_RETURNING, &delegation->flags);
1063 out_spin_unlock:
1064 	spin_unlock(&delegation->lock);
1065 out_rcu_unlock:
1066 	rcu_read_unlock();
1067 
1068 	nfs_inode_find_state_and_recover(inode, stateid);
1069 }
1070 
1071 /**
1072  * nfs_expire_unused_delegation_types
1073  * @clp: client to process
1074  * @flags: delegation types to expire
1075  *
1076  */
nfs_expire_unused_delegation_types(struct nfs_client * clp,fmode_t flags)1077 void nfs_expire_unused_delegation_types(struct nfs_client *clp, fmode_t flags)
1078 {
1079 	nfs_client_mark_return_unused_delegation_types(clp, flags);
1080 	nfs_delegation_run_state_manager(clp);
1081 }
1082 
nfs_mark_return_unreferenced_delegations(struct nfs_server * server)1083 static void nfs_mark_return_unreferenced_delegations(struct nfs_server *server)
1084 {
1085 	struct nfs_delegation *delegation;
1086 
1087 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1088 		if (test_and_clear_bit(NFS_DELEGATION_REFERENCED, &delegation->flags))
1089 			continue;
1090 		nfs_mark_return_if_closed_delegation(server, delegation);
1091 	}
1092 }
1093 
1094 /**
1095  * nfs_expire_unreferenced_delegations - Eliminate unused delegations
1096  * @clp: nfs_client to process
1097  *
1098  */
nfs_expire_unreferenced_delegations(struct nfs_client * clp)1099 void nfs_expire_unreferenced_delegations(struct nfs_client *clp)
1100 {
1101 	struct nfs_server *server;
1102 
1103 	rcu_read_lock();
1104 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1105 		nfs_mark_return_unreferenced_delegations(server);
1106 	rcu_read_unlock();
1107 
1108 	nfs_delegation_run_state_manager(clp);
1109 }
1110 
1111 /**
1112  * nfs_async_inode_return_delegation - asynchronously return a delegation
1113  * @inode: inode to process
1114  * @stateid: state ID information
1115  *
1116  * Returns zero on success, or a negative errno value.
1117  */
nfs_async_inode_return_delegation(struct inode * inode,const nfs4_stateid * stateid)1118 int nfs_async_inode_return_delegation(struct inode *inode,
1119 				      const nfs4_stateid *stateid)
1120 {
1121 	struct nfs_server *server = NFS_SERVER(inode);
1122 	struct nfs_client *clp = server->nfs_client;
1123 	struct nfs_delegation *delegation;
1124 
1125 	rcu_read_lock();
1126 	delegation = nfs4_get_valid_delegation(inode);
1127 	if (delegation == NULL)
1128 		goto out_enoent;
1129 	if (stateid != NULL &&
1130 	    !clp->cl_mvops->match_stateid(&delegation->stateid, stateid))
1131 		goto out_enoent;
1132 	nfs_mark_return_delegation(server, delegation);
1133 	rcu_read_unlock();
1134 
1135 	/* If there are any application leases or delegations, recall them */
1136 	break_lease(inode, O_WRONLY | O_RDWR | O_NONBLOCK);
1137 
1138 	nfs_delegation_run_state_manager(clp);
1139 	return 0;
1140 out_enoent:
1141 	rcu_read_unlock();
1142 	return -ENOENT;
1143 }
1144 
1145 static struct inode *
nfs_delegation_find_inode_server(struct nfs_server * server,const struct nfs_fh * fhandle)1146 nfs_delegation_find_inode_server(struct nfs_server *server,
1147 				 const struct nfs_fh *fhandle)
1148 {
1149 	struct nfs_delegation *delegation;
1150 	struct super_block *freeme = NULL;
1151 	struct inode *res = NULL;
1152 
1153 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1154 		spin_lock(&delegation->lock);
1155 		if (delegation->inode != NULL &&
1156 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
1157 		    nfs_compare_fh(fhandle, &NFS_I(delegation->inode)->fh) == 0) {
1158 			if (nfs_sb_active(server->super)) {
1159 				freeme = server->super;
1160 				res = igrab(delegation->inode);
1161 			}
1162 			spin_unlock(&delegation->lock);
1163 			if (res != NULL)
1164 				return res;
1165 			if (freeme) {
1166 				rcu_read_unlock();
1167 				nfs_sb_deactive(freeme);
1168 				rcu_read_lock();
1169 			}
1170 			return ERR_PTR(-EAGAIN);
1171 		}
1172 		spin_unlock(&delegation->lock);
1173 	}
1174 	return ERR_PTR(-ENOENT);
1175 }
1176 
1177 /**
1178  * nfs_delegation_find_inode - retrieve the inode associated with a delegation
1179  * @clp: client state handle
1180  * @fhandle: filehandle from a delegation recall
1181  *
1182  * Returns pointer to inode matching "fhandle," or NULL if a matching inode
1183  * cannot be found.
1184  */
nfs_delegation_find_inode(struct nfs_client * clp,const struct nfs_fh * fhandle)1185 struct inode *nfs_delegation_find_inode(struct nfs_client *clp,
1186 					const struct nfs_fh *fhandle)
1187 {
1188 	struct nfs_server *server;
1189 	struct inode *res;
1190 
1191 	rcu_read_lock();
1192 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1193 		res = nfs_delegation_find_inode_server(server, fhandle);
1194 		if (res != ERR_PTR(-ENOENT)) {
1195 			rcu_read_unlock();
1196 			return res;
1197 		}
1198 	}
1199 	rcu_read_unlock();
1200 	return ERR_PTR(-ENOENT);
1201 }
1202 
nfs_delegation_mark_reclaim_server(struct nfs_server * server)1203 static void nfs_delegation_mark_reclaim_server(struct nfs_server *server)
1204 {
1205 	struct nfs_delegation *delegation;
1206 
1207 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1208 		/*
1209 		 * If the delegation may have been admin revoked, then we
1210 		 * cannot reclaim it.
1211 		 */
1212 		if (test_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags))
1213 			continue;
1214 		set_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1215 	}
1216 }
1217 
1218 /**
1219  * nfs_delegation_mark_reclaim - mark all delegations as needing to be reclaimed
1220  * @clp: nfs_client to process
1221  *
1222  */
nfs_delegation_mark_reclaim(struct nfs_client * clp)1223 void nfs_delegation_mark_reclaim(struct nfs_client *clp)
1224 {
1225 	struct nfs_server *server;
1226 
1227 	rcu_read_lock();
1228 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1229 		nfs_delegation_mark_reclaim_server(server);
1230 	rcu_read_unlock();
1231 }
1232 
nfs_server_reap_unclaimed_delegations(struct nfs_server * server,void __always_unused * data)1233 static int nfs_server_reap_unclaimed_delegations(struct nfs_server *server,
1234 		void __always_unused *data)
1235 {
1236 	struct nfs_delegation *delegation;
1237 	struct inode *inode;
1238 restart:
1239 	rcu_read_lock();
1240 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1241 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1242 					&delegation->flags) ||
1243 		    test_bit(NFS_DELEGATION_RETURNING,
1244 					&delegation->flags) ||
1245 		    test_bit(NFS_DELEGATION_NEED_RECLAIM,
1246 					&delegation->flags) == 0)
1247 			continue;
1248 		inode = nfs_delegation_grab_inode(delegation);
1249 		if (inode == NULL)
1250 			continue;
1251 		delegation = nfs_start_delegation_return_locked(NFS_I(inode));
1252 		rcu_read_unlock();
1253 		if (delegation != NULL) {
1254 			if (nfs_detach_delegation(NFS_I(inode), delegation,
1255 						server) != NULL)
1256 				nfs_free_delegation(delegation);
1257 			/* Match nfs_start_delegation_return_locked */
1258 			nfs_put_delegation(delegation);
1259 		}
1260 		iput(inode);
1261 		cond_resched();
1262 		goto restart;
1263 	}
1264 	rcu_read_unlock();
1265 	return 0;
1266 }
1267 
1268 /**
1269  * nfs_delegation_reap_unclaimed - reap unclaimed delegations after reboot recovery is done
1270  * @clp: nfs_client to process
1271  *
1272  */
nfs_delegation_reap_unclaimed(struct nfs_client * clp)1273 void nfs_delegation_reap_unclaimed(struct nfs_client *clp)
1274 {
1275 	nfs_client_for_each_server(clp, nfs_server_reap_unclaimed_delegations,
1276 			NULL);
1277 }
1278 
nfs4_server_rebooted(const struct nfs_client * clp)1279 static inline bool nfs4_server_rebooted(const struct nfs_client *clp)
1280 {
1281 	return (clp->cl_state & (BIT(NFS4CLNT_CHECK_LEASE) |
1282 				BIT(NFS4CLNT_LEASE_EXPIRED) |
1283 				BIT(NFS4CLNT_SESSION_RESET))) != 0;
1284 }
1285 
nfs_mark_test_expired_delegation(struct nfs_server * server,struct nfs_delegation * delegation)1286 static void nfs_mark_test_expired_delegation(struct nfs_server *server,
1287 	    struct nfs_delegation *delegation)
1288 {
1289 	if (delegation->stateid.type == NFS4_INVALID_STATEID_TYPE)
1290 		return;
1291 	clear_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags);
1292 	set_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1293 	set_bit(NFS4SERV_DELEGATION_EXPIRED, &server->delegation_flags);
1294 	set_bit(NFS4CLNT_DELEGATION_EXPIRED, &server->nfs_client->cl_state);
1295 }
1296 
nfs_inode_mark_test_expired_delegation(struct nfs_server * server,struct inode * inode)1297 static void nfs_inode_mark_test_expired_delegation(struct nfs_server *server,
1298 		struct inode *inode)
1299 {
1300 	struct nfs_delegation *delegation;
1301 
1302 	rcu_read_lock();
1303 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1304 	if (delegation)
1305 		nfs_mark_test_expired_delegation(server, delegation);
1306 	rcu_read_unlock();
1307 
1308 }
1309 
nfs_delegation_mark_test_expired_server(struct nfs_server * server)1310 static void nfs_delegation_mark_test_expired_server(struct nfs_server *server)
1311 {
1312 	struct nfs_delegation *delegation;
1313 
1314 	list_for_each_entry_rcu(delegation, &server->delegations, super_list)
1315 		nfs_mark_test_expired_delegation(server, delegation);
1316 }
1317 
1318 /**
1319  * nfs_mark_test_expired_all_delegations - mark all delegations for testing
1320  * @clp: nfs_client to process
1321  *
1322  * Iterates through all the delegations associated with this server and
1323  * marks them as needing to be checked for validity.
1324  */
nfs_mark_test_expired_all_delegations(struct nfs_client * clp)1325 void nfs_mark_test_expired_all_delegations(struct nfs_client *clp)
1326 {
1327 	struct nfs_server *server;
1328 
1329 	rcu_read_lock();
1330 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1331 		nfs_delegation_mark_test_expired_server(server);
1332 	rcu_read_unlock();
1333 }
1334 
1335 /**
1336  * nfs_test_expired_all_delegations - test all delegations for a client
1337  * @clp: nfs_client to process
1338  *
1339  * Helper for handling "recallable state revoked" status from server.
1340  */
nfs_test_expired_all_delegations(struct nfs_client * clp)1341 void nfs_test_expired_all_delegations(struct nfs_client *clp)
1342 {
1343 	nfs_mark_test_expired_all_delegations(clp);
1344 	nfs4_schedule_state_manager(clp);
1345 }
1346 
1347 static void
nfs_delegation_test_free_expired(struct inode * inode,nfs4_stateid * stateid,const struct cred * cred)1348 nfs_delegation_test_free_expired(struct inode *inode,
1349 		nfs4_stateid *stateid,
1350 		const struct cred *cred)
1351 {
1352 	struct nfs_server *server = NFS_SERVER(inode);
1353 	const struct nfs4_minor_version_ops *ops = server->nfs_client->cl_mvops;
1354 	int status;
1355 
1356 	if (!cred)
1357 		return;
1358 	status = ops->test_and_free_expired(server, stateid, cred);
1359 	if (status == -NFS4ERR_EXPIRED || status == -NFS4ERR_BAD_STATEID)
1360 		nfs_remove_bad_delegation(inode, stateid);
1361 }
1362 
nfs_server_reap_expired_delegations(struct nfs_server * server,void __always_unused * data)1363 static int nfs_server_reap_expired_delegations(struct nfs_server *server,
1364 		void __always_unused *data)
1365 {
1366 	struct nfs_delegation *delegation;
1367 	struct inode *inode;
1368 	const struct cred *cred;
1369 	nfs4_stateid stateid;
1370 	unsigned long gen = ++server->delegation_gen;
1371 
1372 	if (!test_and_clear_bit(NFS4SERV_DELEGATION_EXPIRED,
1373 				&server->delegation_flags))
1374 		return 0;
1375 restart:
1376 	rcu_read_lock();
1377 	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
1378 		if (test_bit(NFS_DELEGATION_INODE_FREEING,
1379 					&delegation->flags) ||
1380 		    test_bit(NFS_DELEGATION_RETURNING,
1381 					&delegation->flags) ||
1382 		    test_bit(NFS_DELEGATION_TEST_EXPIRED,
1383 					&delegation->flags) == 0 ||
1384 			delegation->test_gen == gen)
1385 			continue;
1386 		inode = nfs_delegation_grab_inode(delegation);
1387 		if (inode == NULL)
1388 			continue;
1389 		spin_lock(&delegation->lock);
1390 		cred = get_cred_rcu(delegation->cred);
1391 		nfs4_stateid_copy(&stateid, &delegation->stateid);
1392 		spin_unlock(&delegation->lock);
1393 		delegation->test_gen = gen;
1394 		clear_bit(NFS_DELEGATION_TEST_EXPIRED, &delegation->flags);
1395 		rcu_read_unlock();
1396 		nfs_delegation_test_free_expired(inode, &stateid, cred);
1397 		put_cred(cred);
1398 		if (!nfs4_server_rebooted(server->nfs_client)) {
1399 			iput(inode);
1400 			cond_resched();
1401 			goto restart;
1402 		}
1403 		nfs_inode_mark_test_expired_delegation(server,inode);
1404 		set_bit(NFS4SERV_DELEGATION_EXPIRED, &server->delegation_flags);
1405 		set_bit(NFS4CLNT_DELEGATION_EXPIRED,
1406 			&server->nfs_client->cl_state);
1407 		iput(inode);
1408 		return -EAGAIN;
1409 	}
1410 	rcu_read_unlock();
1411 	return 0;
1412 }
1413 
1414 /**
1415  * nfs_reap_expired_delegations - reap expired delegations
1416  * @clp: nfs_client to process
1417  *
1418  * Iterates through all the delegations associated with this server and
1419  * checks if they have may have been revoked. This function is usually
1420  * expected to be called in cases where the server may have lost its
1421  * lease.
1422  */
nfs_reap_expired_delegations(struct nfs_client * clp)1423 void nfs_reap_expired_delegations(struct nfs_client *clp)
1424 {
1425 	nfs_client_for_each_server(clp, nfs_server_reap_expired_delegations,
1426 			NULL);
1427 }
1428 
nfs_inode_find_delegation_state_and_recover(struct inode * inode,const nfs4_stateid * stateid)1429 void nfs_inode_find_delegation_state_and_recover(struct inode *inode,
1430 		const nfs4_stateid *stateid)
1431 {
1432 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1433 	struct nfs_delegation *delegation;
1434 	bool found = false;
1435 
1436 	rcu_read_lock();
1437 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1438 	if (delegation &&
1439 	    nfs4_stateid_match_or_older(&delegation->stateid, stateid) &&
1440 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1441 		nfs_mark_test_expired_delegation(NFS_SERVER(inode), delegation);
1442 		found = true;
1443 	}
1444 	rcu_read_unlock();
1445 	if (found)
1446 		nfs4_schedule_state_manager(clp);
1447 }
1448 
1449 /**
1450  * nfs_delegations_present - check for existence of delegations
1451  * @clp: client state handle
1452  *
1453  * Returns one if there are any nfs_delegation structures attached
1454  * to this nfs_client.
1455  */
nfs_delegations_present(struct nfs_client * clp)1456 int nfs_delegations_present(struct nfs_client *clp)
1457 {
1458 	struct nfs_server *server;
1459 	int ret = 0;
1460 
1461 	rcu_read_lock();
1462 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1463 		if (!list_empty(&server->delegations)) {
1464 			ret = 1;
1465 			break;
1466 		}
1467 	rcu_read_unlock();
1468 	return ret;
1469 }
1470 
1471 /**
1472  * nfs4_refresh_delegation_stateid - Update delegation stateid seqid
1473  * @dst: stateid to refresh
1474  * @inode: inode to check
1475  *
1476  * Returns "true" and updates "dst->seqid" * if inode had a delegation
1477  * that matches our delegation stateid. Otherwise "false" is returned.
1478  */
nfs4_refresh_delegation_stateid(nfs4_stateid * dst,struct inode * inode)1479 bool nfs4_refresh_delegation_stateid(nfs4_stateid *dst, struct inode *inode)
1480 {
1481 	struct nfs_delegation *delegation;
1482 	bool ret = false;
1483 	if (!inode)
1484 		goto out;
1485 
1486 	rcu_read_lock();
1487 	delegation = rcu_dereference(NFS_I(inode)->delegation);
1488 	if (delegation != NULL &&
1489 	    nfs4_stateid_match_other(dst, &delegation->stateid) &&
1490 	    nfs4_stateid_is_newer(&delegation->stateid, dst) &&
1491 	    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags)) {
1492 		dst->seqid = delegation->stateid.seqid;
1493 		ret = true;
1494 	}
1495 	rcu_read_unlock();
1496 out:
1497 	return ret;
1498 }
1499 
1500 /**
1501  * nfs4_copy_delegation_stateid - Copy inode's state ID information
1502  * @inode: inode to check
1503  * @flags: delegation type requirement
1504  * @dst: stateid data structure to fill in
1505  * @cred: optional argument to retrieve credential
1506  *
1507  * Returns "true" and fills in "dst->data" * if inode had a delegation,
1508  * otherwise "false" is returned.
1509  */
nfs4_copy_delegation_stateid(struct inode * inode,fmode_t flags,nfs4_stateid * dst,const struct cred ** cred)1510 bool nfs4_copy_delegation_stateid(struct inode *inode, fmode_t flags,
1511 		nfs4_stateid *dst, const struct cred **cred)
1512 {
1513 	struct nfs_inode *nfsi = NFS_I(inode);
1514 	struct nfs_delegation *delegation;
1515 	bool ret = false;
1516 
1517 	flags &= FMODE_READ|FMODE_WRITE;
1518 	rcu_read_lock();
1519 	delegation = rcu_dereference(nfsi->delegation);
1520 	if (!delegation)
1521 		goto out;
1522 	spin_lock(&delegation->lock);
1523 	ret = nfs4_is_valid_delegation(delegation, flags);
1524 	if (ret) {
1525 		nfs4_stateid_copy(dst, &delegation->stateid);
1526 		nfs_mark_delegation_referenced(delegation);
1527 		if (cred)
1528 			*cred = get_cred(delegation->cred);
1529 	}
1530 	spin_unlock(&delegation->lock);
1531 out:
1532 	rcu_read_unlock();
1533 	return ret;
1534 }
1535 
1536 /**
1537  * nfs4_delegation_flush_on_close - Check if we must flush file on close
1538  * @inode: inode to check
1539  *
1540  * This function checks the number of outstanding writes to the file
1541  * against the delegation 'space_limit' field to see if
1542  * the spec requires us to flush the file on close.
1543  */
nfs4_delegation_flush_on_close(const struct inode * inode)1544 bool nfs4_delegation_flush_on_close(const struct inode *inode)
1545 {
1546 	struct nfs_inode *nfsi = NFS_I(inode);
1547 	struct nfs_delegation *delegation;
1548 	bool ret = true;
1549 
1550 	rcu_read_lock();
1551 	delegation = rcu_dereference(nfsi->delegation);
1552 	if (delegation == NULL || !(delegation->type & FMODE_WRITE))
1553 		goto out;
1554 	if (atomic_long_read(&nfsi->nrequests) < delegation->pagemod_limit)
1555 		ret = false;
1556 out:
1557 	rcu_read_unlock();
1558 	return ret;
1559 }
1560 
1561 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);
1562