1 /*
2  *  fs/nfs/nfs4state.c
3  *
4  *  Client-side XDR for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *
11  *  Redistribution and use in source and binary forms, with or without
12  *  modification, are permitted provided that the following conditions
13  *  are met:
14  *
15  *  1. Redistributions of source code must retain the above copyright
16  *     notice, this list of conditions and the following disclaimer.
17  *  2. Redistributions in binary form must reproduce the above copyright
18  *     notice, this list of conditions and the following disclaimer in the
19  *     documentation and/or other materials provided with the distribution.
20  *  3. Neither the name of the University nor the names of its
21  *     contributors may be used to endorse or promote products derived
22  *     from this software without specific prior written permission.
23  *
24  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
25  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
26  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
27  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
32  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
33  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
34  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Implementation of the NFSv4 state model.  For the time being,
37  * this is minimal, but will be made much more complex in a
38  * subsequent patch.
39  */
40 
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
43 #include <linux/fs.h>
44 #include <linux/nfs_fs.h>
45 #include <linux/kthread.h>
46 #include <linux/module.h>
47 #include <linux/random.h>
48 #include <linux/ratelimit.h>
49 #include <linux/workqueue.h>
50 #include <linux/bitops.h>
51 #include <linux/jiffies.h>
52 #include <linux/sched/mm.h>
53 
54 #include <linux/sunrpc/clnt.h>
55 
56 #include "nfs4_fs.h"
57 #include "callback.h"
58 #include "delegation.h"
59 #include "internal.h"
60 #include "nfs4idmap.h"
61 #include "nfs4session.h"
62 #include "pnfs.h"
63 #include "netns.h"
64 #include "nfs4trace.h"
65 
66 #define NFSDBG_FACILITY		NFSDBG_STATE
67 
68 #define OPENOWNER_POOL_SIZE	8
69 
70 const nfs4_stateid zero_stateid = {
71 	{ .data = { 0 } },
72 	.type = NFS4_SPECIAL_STATEID_TYPE,
73 };
74 const nfs4_stateid invalid_stateid = {
75 	{
76 		/* Funky initialiser keeps older gcc versions happy */
77 		.data = { 0xff, 0xff, 0xff, 0xff, 0 },
78 	},
79 	.type = NFS4_INVALID_STATEID_TYPE,
80 };
81 
82 const nfs4_stateid current_stateid = {
83 	{
84 		/* Funky initialiser keeps older gcc versions happy */
85 		.data = { 0x0, 0x0, 0x0, 0x1, 0 },
86 	},
87 	.type = NFS4_SPECIAL_STATEID_TYPE,
88 };
89 
90 static DEFINE_MUTEX(nfs_clid_init_mutex);
91 
nfs4_setup_state_renewal(struct nfs_client * clp)92 static int nfs4_setup_state_renewal(struct nfs_client *clp)
93 {
94 	int status;
95 	struct nfs_fsinfo fsinfo;
96 
97 	if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
98 		nfs4_schedule_state_renewal(clp);
99 		return 0;
100 	}
101 
102 	status = nfs4_proc_get_lease_time(clp, &fsinfo);
103 	if (status == 0) {
104 		nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
105 		nfs4_schedule_state_renewal(clp);
106 	}
107 
108 	return status;
109 }
110 
nfs4_init_clientid(struct nfs_client * clp,const struct cred * cred)111 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
112 {
113 	struct nfs4_setclientid_res clid = {
114 		.clientid = clp->cl_clientid,
115 		.confirm = clp->cl_confirm,
116 	};
117 	unsigned short port;
118 	int status;
119 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
120 
121 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
122 		goto do_confirm;
123 	port = nn->nfs_callback_tcpport;
124 	if (clp->cl_addr.ss_family == AF_INET6)
125 		port = nn->nfs_callback_tcpport6;
126 
127 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
128 	if (status != 0)
129 		goto out;
130 	clp->cl_clientid = clid.clientid;
131 	clp->cl_confirm = clid.confirm;
132 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
133 do_confirm:
134 	status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
135 	if (status != 0)
136 		goto out;
137 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
138 	nfs4_setup_state_renewal(clp);
139 out:
140 	return status;
141 }
142 
143 /**
144  * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
145  *
146  * @clp: nfs_client under test
147  * @result: OUT: found nfs_client, or clp
148  * @cred: credential to use for trunking test
149  *
150  * Returns zero, a negative errno, or a negative NFS4ERR status.
151  * If zero is returned, an nfs_client pointer is planted in
152  * "result".
153  *
154  * Note: The returned client may not yet be marked ready.
155  */
nfs40_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result,const struct cred * cred)156 int nfs40_discover_server_trunking(struct nfs_client *clp,
157 				   struct nfs_client **result,
158 				   const struct cred *cred)
159 {
160 	struct nfs4_setclientid_res clid = {
161 		.clientid = clp->cl_clientid,
162 		.confirm = clp->cl_confirm,
163 	};
164 	struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
165 	unsigned short port;
166 	int status;
167 
168 	port = nn->nfs_callback_tcpport;
169 	if (clp->cl_addr.ss_family == AF_INET6)
170 		port = nn->nfs_callback_tcpport6;
171 
172 	status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
173 	if (status != 0)
174 		goto out;
175 	clp->cl_clientid = clid.clientid;
176 	clp->cl_confirm = clid.confirm;
177 
178 	status = nfs40_walk_client_list(clp, result, cred);
179 	if (status == 0) {
180 		/* Sustain the lease, even if it's empty.  If the clientid4
181 		 * goes stale it's of no use for trunking discovery. */
182 		nfs4_schedule_state_renewal(*result);
183 
184 		/* If the client state need to recover, do it. */
185 		if (clp->cl_state)
186 			nfs4_schedule_state_manager(clp);
187 	}
188 out:
189 	return status;
190 }
191 
nfs4_get_machine_cred(struct nfs_client * clp)192 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
193 {
194 	return get_cred(rpc_machine_cred());
195 }
196 
nfs4_root_machine_cred(struct nfs_client * clp)197 static void nfs4_root_machine_cred(struct nfs_client *clp)
198 {
199 
200 	/* Force root creds instead of machine */
201 	clp->cl_principal = NULL;
202 	clp->cl_rpcclient->cl_principal = NULL;
203 }
204 
205 static const struct cred *
nfs4_get_renew_cred_server_locked(struct nfs_server * server)206 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
207 {
208 	const struct cred *cred = NULL;
209 	struct nfs4_state_owner *sp;
210 	struct rb_node *pos;
211 
212 	for (pos = rb_first(&server->state_owners);
213 	     pos != NULL;
214 	     pos = rb_next(pos)) {
215 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
216 		if (list_empty(&sp->so_states))
217 			continue;
218 		cred = get_cred(sp->so_cred);
219 		break;
220 	}
221 	return cred;
222 }
223 
224 /**
225  * nfs4_get_renew_cred - Acquire credential for a renew operation
226  * @clp: client state handle
227  *
228  * Returns an rpc_cred with reference count bumped, or NULL.
229  * Caller must hold clp->cl_lock.
230  */
nfs4_get_renew_cred(struct nfs_client * clp)231 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
232 {
233 	const struct cred *cred = NULL;
234 	struct nfs_server *server;
235 
236 	/* Use machine credentials if available */
237 	cred = nfs4_get_machine_cred(clp);
238 	if (cred != NULL)
239 		goto out;
240 
241 	spin_lock(&clp->cl_lock);
242 	rcu_read_lock();
243 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
244 		cred = nfs4_get_renew_cred_server_locked(server);
245 		if (cred != NULL)
246 			break;
247 	}
248 	rcu_read_unlock();
249 	spin_unlock(&clp->cl_lock);
250 
251 out:
252 	return cred;
253 }
254 
nfs4_end_drain_slot_table(struct nfs4_slot_table * tbl)255 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
256 {
257 	if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
258 		spin_lock(&tbl->slot_tbl_lock);
259 		nfs41_wake_slot_table(tbl);
260 		spin_unlock(&tbl->slot_tbl_lock);
261 	}
262 }
263 
nfs4_end_drain_session(struct nfs_client * clp)264 static void nfs4_end_drain_session(struct nfs_client *clp)
265 {
266 	struct nfs4_session *ses = clp->cl_session;
267 
268 	if (clp->cl_slot_tbl) {
269 		nfs4_end_drain_slot_table(clp->cl_slot_tbl);
270 		return;
271 	}
272 
273 	if (ses != NULL) {
274 		nfs4_end_drain_slot_table(&ses->bc_slot_table);
275 		nfs4_end_drain_slot_table(&ses->fc_slot_table);
276 	}
277 }
278 
nfs4_drain_slot_tbl(struct nfs4_slot_table * tbl)279 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
280 {
281 	set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
282 	spin_lock(&tbl->slot_tbl_lock);
283 	if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
284 		reinit_completion(&tbl->complete);
285 		spin_unlock(&tbl->slot_tbl_lock);
286 		return wait_for_completion_interruptible(&tbl->complete);
287 	}
288 	spin_unlock(&tbl->slot_tbl_lock);
289 	return 0;
290 }
291 
nfs4_begin_drain_session(struct nfs_client * clp)292 static int nfs4_begin_drain_session(struct nfs_client *clp)
293 {
294 	struct nfs4_session *ses = clp->cl_session;
295 	int ret;
296 
297 	if (clp->cl_slot_tbl)
298 		return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
299 
300 	/* back channel */
301 	ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
302 	if (ret)
303 		return ret;
304 	/* fore channel */
305 	return nfs4_drain_slot_tbl(&ses->fc_slot_table);
306 }
307 
308 #if defined(CONFIG_NFS_V4_1)
309 
nfs41_finish_session_reset(struct nfs_client * clp)310 static void nfs41_finish_session_reset(struct nfs_client *clp)
311 {
312 	clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
313 	clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
314 	/* create_session negotiated new slot table */
315 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
316 	nfs4_setup_state_renewal(clp);
317 }
318 
nfs41_init_clientid(struct nfs_client * clp,const struct cred * cred)319 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
320 {
321 	int status;
322 
323 	if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
324 		goto do_confirm;
325 	status = nfs4_proc_exchange_id(clp, cred);
326 	if (status != 0)
327 		goto out;
328 	set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
329 do_confirm:
330 	status = nfs4_proc_create_session(clp, cred);
331 	if (status != 0)
332 		goto out;
333 	nfs41_finish_session_reset(clp);
334 	nfs_mark_client_ready(clp, NFS_CS_READY);
335 out:
336 	return status;
337 }
338 
339 /**
340  * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
341  *
342  * @clp: nfs_client under test
343  * @result: OUT: found nfs_client, or clp
344  * @cred: credential to use for trunking test
345  *
346  * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
347  * If NFS4_OK is returned, an nfs_client pointer is planted in
348  * "result".
349  *
350  * Note: The returned client may not yet be marked ready.
351  */
nfs41_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result,const struct cred * cred)352 int nfs41_discover_server_trunking(struct nfs_client *clp,
353 				   struct nfs_client **result,
354 				   const struct cred *cred)
355 {
356 	int status;
357 
358 	status = nfs4_proc_exchange_id(clp, cred);
359 	if (status != NFS4_OK)
360 		return status;
361 
362 	status = nfs41_walk_client_list(clp, result, cred);
363 	if (status < 0)
364 		return status;
365 	if (clp != *result)
366 		return 0;
367 
368 	/*
369 	 * Purge state if the client id was established in a prior
370 	 * instance and the client id could not have arrived on the
371 	 * server via Transparent State Migration.
372 	 */
373 	if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
374 		if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
375 			set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
376 		else
377 			set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
378 	}
379 	nfs4_schedule_state_manager(clp);
380 	status = nfs_wait_client_init_complete(clp);
381 	if (status < 0)
382 		nfs_put_client(clp);
383 	return status;
384 }
385 
386 #endif /* CONFIG_NFS_V4_1 */
387 
388 /**
389  * nfs4_get_clid_cred - Acquire credential for a setclientid operation
390  * @clp: client state handle
391  *
392  * Returns a cred with reference count bumped, or NULL.
393  */
nfs4_get_clid_cred(struct nfs_client * clp)394 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
395 {
396 	const struct cred *cred;
397 
398 	cred = nfs4_get_machine_cred(clp);
399 	return cred;
400 }
401 
402 static struct nfs4_state_owner *
nfs4_find_state_owner_locked(struct nfs_server * server,const struct cred * cred)403 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
404 {
405 	struct rb_node **p = &server->state_owners.rb_node,
406 		       *parent = NULL;
407 	struct nfs4_state_owner *sp;
408 	int cmp;
409 
410 	while (*p != NULL) {
411 		parent = *p;
412 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
413 		cmp = cred_fscmp(cred, sp->so_cred);
414 
415 		if (cmp < 0)
416 			p = &parent->rb_left;
417 		else if (cmp > 0)
418 			p = &parent->rb_right;
419 		else {
420 			if (!list_empty(&sp->so_lru))
421 				list_del_init(&sp->so_lru);
422 			atomic_inc(&sp->so_count);
423 			return sp;
424 		}
425 	}
426 	return NULL;
427 }
428 
429 static struct nfs4_state_owner *
nfs4_insert_state_owner_locked(struct nfs4_state_owner * new)430 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
431 {
432 	struct nfs_server *server = new->so_server;
433 	struct rb_node **p = &server->state_owners.rb_node,
434 		       *parent = NULL;
435 	struct nfs4_state_owner *sp;
436 	int cmp;
437 
438 	while (*p != NULL) {
439 		parent = *p;
440 		sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
441 		cmp = cred_fscmp(new->so_cred, sp->so_cred);
442 
443 		if (cmp < 0)
444 			p = &parent->rb_left;
445 		else if (cmp > 0)
446 			p = &parent->rb_right;
447 		else {
448 			if (!list_empty(&sp->so_lru))
449 				list_del_init(&sp->so_lru);
450 			atomic_inc(&sp->so_count);
451 			return sp;
452 		}
453 	}
454 	rb_link_node(&new->so_server_node, parent, p);
455 	rb_insert_color(&new->so_server_node, &server->state_owners);
456 	return new;
457 }
458 
459 static void
nfs4_remove_state_owner_locked(struct nfs4_state_owner * sp)460 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
461 {
462 	struct nfs_server *server = sp->so_server;
463 
464 	if (!RB_EMPTY_NODE(&sp->so_server_node))
465 		rb_erase(&sp->so_server_node, &server->state_owners);
466 }
467 
468 static void
nfs4_init_seqid_counter(struct nfs_seqid_counter * sc)469 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
470 {
471 	sc->create_time = ktime_get();
472 	sc->flags = 0;
473 	sc->counter = 0;
474 	spin_lock_init(&sc->lock);
475 	INIT_LIST_HEAD(&sc->list);
476 	rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
477 }
478 
479 static void
nfs4_destroy_seqid_counter(struct nfs_seqid_counter * sc)480 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
481 {
482 	rpc_destroy_wait_queue(&sc->wait);
483 }
484 
485 /*
486  * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
487  * create a new state_owner.
488  *
489  */
490 static struct nfs4_state_owner *
nfs4_alloc_state_owner(struct nfs_server * server,const struct cred * cred,gfp_t gfp_flags)491 nfs4_alloc_state_owner(struct nfs_server *server,
492 		const struct cred *cred,
493 		gfp_t gfp_flags)
494 {
495 	struct nfs4_state_owner *sp;
496 
497 	sp = kzalloc(sizeof(*sp), gfp_flags);
498 	if (!sp)
499 		return NULL;
500 	sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
501 						gfp_flags);
502 	if (sp->so_seqid.owner_id < 0) {
503 		kfree(sp);
504 		return NULL;
505 	}
506 	sp->so_server = server;
507 	sp->so_cred = get_cred(cred);
508 	spin_lock_init(&sp->so_lock);
509 	INIT_LIST_HEAD(&sp->so_states);
510 	nfs4_init_seqid_counter(&sp->so_seqid);
511 	atomic_set(&sp->so_count, 1);
512 	INIT_LIST_HEAD(&sp->so_lru);
513 	seqcount_spinlock_init(&sp->so_reclaim_seqcount, &sp->so_lock);
514 	mutex_init(&sp->so_delegreturn_mutex);
515 	return sp;
516 }
517 
518 static void
nfs4_reset_state_owner(struct nfs4_state_owner * sp)519 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
520 {
521 	/* This state_owner is no longer usable, but must
522 	 * remain in place so that state recovery can find it
523 	 * and the opens associated with it.
524 	 * It may also be used for new 'open' request to
525 	 * return a delegation to the server.
526 	 * So update the 'create_time' so that it looks like
527 	 * a new state_owner.  This will cause the server to
528 	 * request an OPEN_CONFIRM to start a new sequence.
529 	 */
530 	sp->so_seqid.create_time = ktime_get();
531 }
532 
nfs4_free_state_owner(struct nfs4_state_owner * sp)533 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
534 {
535 	nfs4_destroy_seqid_counter(&sp->so_seqid);
536 	put_cred(sp->so_cred);
537 	ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
538 	kfree(sp);
539 }
540 
nfs4_gc_state_owners(struct nfs_server * server)541 static void nfs4_gc_state_owners(struct nfs_server *server)
542 {
543 	struct nfs_client *clp = server->nfs_client;
544 	struct nfs4_state_owner *sp, *tmp;
545 	unsigned long time_min, time_max;
546 	LIST_HEAD(doomed);
547 
548 	spin_lock(&clp->cl_lock);
549 	time_max = jiffies;
550 	time_min = (long)time_max - (long)clp->cl_lease_time;
551 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
552 		/* NB: LRU is sorted so that oldest is at the head */
553 		if (time_in_range(sp->so_expires, time_min, time_max))
554 			break;
555 		list_move(&sp->so_lru, &doomed);
556 		nfs4_remove_state_owner_locked(sp);
557 	}
558 	spin_unlock(&clp->cl_lock);
559 
560 	list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
561 		list_del(&sp->so_lru);
562 		nfs4_free_state_owner(sp);
563 	}
564 }
565 
566 /**
567  * nfs4_get_state_owner - Look up a state owner given a credential
568  * @server: nfs_server to search
569  * @cred: RPC credential to match
570  * @gfp_flags: allocation mode
571  *
572  * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
573  */
nfs4_get_state_owner(struct nfs_server * server,const struct cred * cred,gfp_t gfp_flags)574 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
575 					      const struct cred *cred,
576 					      gfp_t gfp_flags)
577 {
578 	struct nfs_client *clp = server->nfs_client;
579 	struct nfs4_state_owner *sp, *new;
580 
581 	spin_lock(&clp->cl_lock);
582 	sp = nfs4_find_state_owner_locked(server, cred);
583 	spin_unlock(&clp->cl_lock);
584 	if (sp != NULL)
585 		goto out;
586 	new = nfs4_alloc_state_owner(server, cred, gfp_flags);
587 	if (new == NULL)
588 		goto out;
589 	spin_lock(&clp->cl_lock);
590 	sp = nfs4_insert_state_owner_locked(new);
591 	spin_unlock(&clp->cl_lock);
592 	if (sp != new)
593 		nfs4_free_state_owner(new);
594 out:
595 	nfs4_gc_state_owners(server);
596 	return sp;
597 }
598 
599 /**
600  * nfs4_put_state_owner - Release a nfs4_state_owner
601  * @sp: state owner data to release
602  *
603  * Note that we keep released state owners on an LRU
604  * list.
605  * This caches valid state owners so that they can be
606  * reused, to avoid the OPEN_CONFIRM on minor version 0.
607  * It also pins the uniquifier of dropped state owners for
608  * a while, to ensure that those state owner names are
609  * never reused.
610  */
nfs4_put_state_owner(struct nfs4_state_owner * sp)611 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
612 {
613 	struct nfs_server *server = sp->so_server;
614 	struct nfs_client *clp = server->nfs_client;
615 
616 	if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
617 		return;
618 
619 	sp->so_expires = jiffies;
620 	list_add_tail(&sp->so_lru, &server->state_owners_lru);
621 	spin_unlock(&clp->cl_lock);
622 }
623 
624 /**
625  * nfs4_purge_state_owners - Release all cached state owners
626  * @server: nfs_server with cached state owners to release
627  * @head: resulting list of state owners
628  *
629  * Called at umount time.  Remaining state owners will be on
630  * the LRU with ref count of zero.
631  * Note that the state owners are not freed, but are added
632  * to the list @head, which can later be used as an argument
633  * to nfs4_free_state_owners.
634  */
nfs4_purge_state_owners(struct nfs_server * server,struct list_head * head)635 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
636 {
637 	struct nfs_client *clp = server->nfs_client;
638 	struct nfs4_state_owner *sp, *tmp;
639 
640 	spin_lock(&clp->cl_lock);
641 	list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
642 		list_move(&sp->so_lru, head);
643 		nfs4_remove_state_owner_locked(sp);
644 	}
645 	spin_unlock(&clp->cl_lock);
646 }
647 
648 /**
649  * nfs4_free_state_owners - Release all cached state owners
650  * @head: resulting list of state owners
651  *
652  * Frees a list of state owners that was generated by
653  * nfs4_purge_state_owners
654  */
nfs4_free_state_owners(struct list_head * head)655 void nfs4_free_state_owners(struct list_head *head)
656 {
657 	struct nfs4_state_owner *sp, *tmp;
658 
659 	list_for_each_entry_safe(sp, tmp, head, so_lru) {
660 		list_del(&sp->so_lru);
661 		nfs4_free_state_owner(sp);
662 	}
663 }
664 
665 static struct nfs4_state *
nfs4_alloc_open_state(void)666 nfs4_alloc_open_state(void)
667 {
668 	struct nfs4_state *state;
669 
670 	state = kzalloc(sizeof(*state), GFP_KERNEL_ACCOUNT);
671 	if (!state)
672 		return NULL;
673 	refcount_set(&state->count, 1);
674 	INIT_LIST_HEAD(&state->lock_states);
675 	spin_lock_init(&state->state_lock);
676 	seqlock_init(&state->seqlock);
677 	init_waitqueue_head(&state->waitq);
678 	return state;
679 }
680 
681 void
nfs4_state_set_mode_locked(struct nfs4_state * state,fmode_t fmode)682 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
683 {
684 	if (state->state == fmode)
685 		return;
686 	/* NB! List reordering - see the reclaim code for why.  */
687 	if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
688 		if (fmode & FMODE_WRITE)
689 			list_move(&state->open_states, &state->owner->so_states);
690 		else
691 			list_move_tail(&state->open_states, &state->owner->so_states);
692 	}
693 	state->state = fmode;
694 }
695 
696 static struct nfs4_state *
__nfs4_find_state_byowner(struct inode * inode,struct nfs4_state_owner * owner)697 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
698 {
699 	struct nfs_inode *nfsi = NFS_I(inode);
700 	struct nfs4_state *state;
701 
702 	list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
703 		if (state->owner != owner)
704 			continue;
705 		if (!nfs4_valid_open_stateid(state))
706 			continue;
707 		if (refcount_inc_not_zero(&state->count))
708 			return state;
709 	}
710 	return NULL;
711 }
712 
713 static void
nfs4_free_open_state(struct nfs4_state * state)714 nfs4_free_open_state(struct nfs4_state *state)
715 {
716 	kfree_rcu(state, rcu_head);
717 }
718 
719 struct nfs4_state *
nfs4_get_open_state(struct inode * inode,struct nfs4_state_owner * owner)720 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
721 {
722 	struct nfs4_state *state, *new;
723 	struct nfs_inode *nfsi = NFS_I(inode);
724 
725 	rcu_read_lock();
726 	state = __nfs4_find_state_byowner(inode, owner);
727 	rcu_read_unlock();
728 	if (state)
729 		goto out;
730 	new = nfs4_alloc_open_state();
731 	spin_lock(&owner->so_lock);
732 	spin_lock(&inode->i_lock);
733 	state = __nfs4_find_state_byowner(inode, owner);
734 	if (state == NULL && new != NULL) {
735 		state = new;
736 		state->owner = owner;
737 		atomic_inc(&owner->so_count);
738 		ihold(inode);
739 		state->inode = inode;
740 		list_add_rcu(&state->inode_states, &nfsi->open_states);
741 		spin_unlock(&inode->i_lock);
742 		/* Note: The reclaim code dictates that we add stateless
743 		 * and read-only stateids to the end of the list */
744 		list_add_tail(&state->open_states, &owner->so_states);
745 		spin_unlock(&owner->so_lock);
746 	} else {
747 		spin_unlock(&inode->i_lock);
748 		spin_unlock(&owner->so_lock);
749 		if (new)
750 			nfs4_free_open_state(new);
751 	}
752 out:
753 	return state;
754 }
755 
nfs4_put_open_state(struct nfs4_state * state)756 void nfs4_put_open_state(struct nfs4_state *state)
757 {
758 	struct inode *inode = state->inode;
759 	struct nfs4_state_owner *owner = state->owner;
760 
761 	if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
762 		return;
763 	spin_lock(&inode->i_lock);
764 	list_del_rcu(&state->inode_states);
765 	list_del(&state->open_states);
766 	spin_unlock(&inode->i_lock);
767 	spin_unlock(&owner->so_lock);
768 	nfs4_inode_return_delegation_on_close(inode);
769 	iput(inode);
770 	nfs4_free_open_state(state);
771 	nfs4_put_state_owner(owner);
772 }
773 
774 /*
775  * Close the current file.
776  */
__nfs4_close(struct nfs4_state * state,fmode_t fmode,gfp_t gfp_mask,int wait)777 static void __nfs4_close(struct nfs4_state *state,
778 		fmode_t fmode, gfp_t gfp_mask, int wait)
779 {
780 	struct nfs4_state_owner *owner = state->owner;
781 	int call_close = 0;
782 	fmode_t newstate;
783 
784 	atomic_inc(&owner->so_count);
785 	/* Protect against nfs4_find_state() */
786 	spin_lock(&owner->so_lock);
787 	switch (fmode & (FMODE_READ | FMODE_WRITE)) {
788 		case FMODE_READ:
789 			state->n_rdonly--;
790 			break;
791 		case FMODE_WRITE:
792 			state->n_wronly--;
793 			break;
794 		case FMODE_READ|FMODE_WRITE:
795 			state->n_rdwr--;
796 	}
797 	newstate = FMODE_READ|FMODE_WRITE;
798 	if (state->n_rdwr == 0) {
799 		if (state->n_rdonly == 0) {
800 			newstate &= ~FMODE_READ;
801 			call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
802 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
803 		}
804 		if (state->n_wronly == 0) {
805 			newstate &= ~FMODE_WRITE;
806 			call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
807 			call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
808 		}
809 		if (newstate == 0)
810 			clear_bit(NFS_DELEGATED_STATE, &state->flags);
811 	}
812 	nfs4_state_set_mode_locked(state, newstate);
813 	spin_unlock(&owner->so_lock);
814 
815 	if (!call_close) {
816 		nfs4_put_open_state(state);
817 		nfs4_put_state_owner(owner);
818 	} else
819 		nfs4_do_close(state, gfp_mask, wait);
820 }
821 
nfs4_close_state(struct nfs4_state * state,fmode_t fmode)822 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
823 {
824 	__nfs4_close(state, fmode, GFP_KERNEL, 0);
825 }
826 
nfs4_close_sync(struct nfs4_state * state,fmode_t fmode)827 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
828 {
829 	__nfs4_close(state, fmode, GFP_KERNEL, 1);
830 }
831 
832 /*
833  * Search the state->lock_states for an existing lock_owner
834  * that is compatible with either of the given owners.
835  * If the second is non-zero, then the first refers to a Posix-lock
836  * owner (current->files) and the second refers to a flock/OFD
837  * owner (struct file*).  In that case, prefer a match for the first
838  * owner.
839  * If both sorts of locks are held on the one file we cannot know
840  * which stateid was intended to be used, so a "correct" choice cannot
841  * be made.  Failing that, a "consistent" choice is preferable.  The
842  * consistent choice we make is to prefer the first owner, that of a
843  * Posix lock.
844  */
845 static struct nfs4_lock_state *
__nfs4_find_lock_state(struct nfs4_state * state,fl_owner_t fl_owner,fl_owner_t fl_owner2)846 __nfs4_find_lock_state(struct nfs4_state *state,
847 		       fl_owner_t fl_owner, fl_owner_t fl_owner2)
848 {
849 	struct nfs4_lock_state *pos, *ret = NULL;
850 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
851 		if (pos->ls_owner == fl_owner) {
852 			ret = pos;
853 			break;
854 		}
855 		if (pos->ls_owner == fl_owner2)
856 			ret = pos;
857 	}
858 	if (ret)
859 		refcount_inc(&ret->ls_count);
860 	return ret;
861 }
862 
863 /*
864  * Return a compatible lock_state. If no initialized lock_state structure
865  * exists, return an uninitialized one.
866  *
867  */
nfs4_alloc_lock_state(struct nfs4_state * state,fl_owner_t fl_owner)868 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
869 {
870 	struct nfs4_lock_state *lsp;
871 	struct nfs_server *server = state->owner->so_server;
872 
873 	lsp = kzalloc(sizeof(*lsp), GFP_KERNEL_ACCOUNT);
874 	if (lsp == NULL)
875 		return NULL;
876 	nfs4_init_seqid_counter(&lsp->ls_seqid);
877 	refcount_set(&lsp->ls_count, 1);
878 	lsp->ls_state = state;
879 	lsp->ls_owner = fl_owner;
880 	lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id,
881 						0, 0, GFP_KERNEL_ACCOUNT);
882 	if (lsp->ls_seqid.owner_id < 0)
883 		goto out_free;
884 	INIT_LIST_HEAD(&lsp->ls_locks);
885 	return lsp;
886 out_free:
887 	kfree(lsp);
888 	return NULL;
889 }
890 
nfs4_free_lock_state(struct nfs_server * server,struct nfs4_lock_state * lsp)891 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
892 {
893 	ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
894 	nfs4_destroy_seqid_counter(&lsp->ls_seqid);
895 	kfree(lsp);
896 }
897 
898 /*
899  * Return a compatible lock_state. If no initialized lock_state structure
900  * exists, return an uninitialized one.
901  *
902  */
nfs4_get_lock_state(struct nfs4_state * state,fl_owner_t owner)903 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
904 {
905 	struct nfs4_lock_state *lsp, *new = NULL;
906 
907 	for(;;) {
908 		spin_lock(&state->state_lock);
909 		lsp = __nfs4_find_lock_state(state, owner, NULL);
910 		if (lsp != NULL)
911 			break;
912 		if (new != NULL) {
913 			list_add(&new->ls_locks, &state->lock_states);
914 			set_bit(LK_STATE_IN_USE, &state->flags);
915 			lsp = new;
916 			new = NULL;
917 			break;
918 		}
919 		spin_unlock(&state->state_lock);
920 		new = nfs4_alloc_lock_state(state, owner);
921 		if (new == NULL)
922 			return NULL;
923 	}
924 	spin_unlock(&state->state_lock);
925 	if (new != NULL)
926 		nfs4_free_lock_state(state->owner->so_server, new);
927 	return lsp;
928 }
929 
930 /*
931  * Release reference to lock_state, and free it if we see that
932  * it is no longer in use
933  */
nfs4_put_lock_state(struct nfs4_lock_state * lsp)934 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
935 {
936 	struct nfs_server *server;
937 	struct nfs4_state *state;
938 
939 	if (lsp == NULL)
940 		return;
941 	state = lsp->ls_state;
942 	if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
943 		return;
944 	list_del(&lsp->ls_locks);
945 	if (list_empty(&state->lock_states))
946 		clear_bit(LK_STATE_IN_USE, &state->flags);
947 	spin_unlock(&state->state_lock);
948 	server = state->owner->so_server;
949 	if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
950 		struct nfs_client *clp = server->nfs_client;
951 
952 		clp->cl_mvops->free_lock_state(server, lsp);
953 	} else
954 		nfs4_free_lock_state(server, lsp);
955 }
956 
nfs4_fl_copy_lock(struct file_lock * dst,struct file_lock * src)957 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
958 {
959 	struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
960 
961 	dst->fl_u.nfs4_fl.owner = lsp;
962 	refcount_inc(&lsp->ls_count);
963 }
964 
nfs4_fl_release_lock(struct file_lock * fl)965 static void nfs4_fl_release_lock(struct file_lock *fl)
966 {
967 	nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
968 }
969 
970 static const struct file_lock_operations nfs4_fl_lock_ops = {
971 	.fl_copy_lock = nfs4_fl_copy_lock,
972 	.fl_release_private = nfs4_fl_release_lock,
973 };
974 
nfs4_set_lock_state(struct nfs4_state * state,struct file_lock * fl)975 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
976 {
977 	struct nfs4_lock_state *lsp;
978 
979 	if (fl->fl_ops != NULL)
980 		return 0;
981 	lsp = nfs4_get_lock_state(state, fl->fl_owner);
982 	if (lsp == NULL)
983 		return -ENOMEM;
984 	fl->fl_u.nfs4_fl.owner = lsp;
985 	fl->fl_ops = &nfs4_fl_lock_ops;
986 	return 0;
987 }
988 
nfs4_copy_lock_stateid(nfs4_stateid * dst,struct nfs4_state * state,const struct nfs_lock_context * l_ctx)989 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
990 		struct nfs4_state *state,
991 		const struct nfs_lock_context *l_ctx)
992 {
993 	struct nfs4_lock_state *lsp;
994 	fl_owner_t fl_owner, fl_flock_owner;
995 	int ret = -ENOENT;
996 
997 	if (l_ctx == NULL)
998 		goto out;
999 
1000 	if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1001 		goto out;
1002 
1003 	fl_owner = l_ctx->lockowner;
1004 	fl_flock_owner = l_ctx->open_context->flock_owner;
1005 
1006 	spin_lock(&state->state_lock);
1007 	lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1008 	if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1009 		ret = -EIO;
1010 	else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1011 		nfs4_stateid_copy(dst, &lsp->ls_stateid);
1012 		ret = 0;
1013 	}
1014 	spin_unlock(&state->state_lock);
1015 	nfs4_put_lock_state(lsp);
1016 out:
1017 	return ret;
1018 }
1019 
nfs4_copy_open_stateid(nfs4_stateid * dst,struct nfs4_state * state)1020 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1021 {
1022 	bool ret;
1023 	const nfs4_stateid *src;
1024 	int seq;
1025 
1026 	do {
1027 		ret = false;
1028 		src = &zero_stateid;
1029 		seq = read_seqbegin(&state->seqlock);
1030 		if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1031 			src = &state->open_stateid;
1032 			ret = true;
1033 		}
1034 		nfs4_stateid_copy(dst, src);
1035 	} while (read_seqretry(&state->seqlock, seq));
1036 	return ret;
1037 }
1038 
1039 /*
1040  * Byte-range lock aware utility to initialize the stateid of read/write
1041  * requests.
1042  */
nfs4_select_rw_stateid(struct nfs4_state * state,fmode_t fmode,const struct nfs_lock_context * l_ctx,nfs4_stateid * dst,const struct cred ** cred)1043 int nfs4_select_rw_stateid(struct nfs4_state *state,
1044 		fmode_t fmode, const struct nfs_lock_context *l_ctx,
1045 		nfs4_stateid *dst, const struct cred **cred)
1046 {
1047 	int ret;
1048 
1049 	if (!nfs4_valid_open_stateid(state))
1050 		return -EIO;
1051 	if (cred != NULL)
1052 		*cred = NULL;
1053 	ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1054 	if (ret == -EIO)
1055 		/* A lost lock - don't even consider delegations */
1056 		goto out;
1057 	/* returns true if delegation stateid found and copied */
1058 	if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1059 		ret = 0;
1060 		goto out;
1061 	}
1062 	if (ret != -ENOENT)
1063 		/* nfs4_copy_delegation_stateid() didn't over-write
1064 		 * dst, so it still has the lock stateid which we now
1065 		 * choose to use.
1066 		 */
1067 		goto out;
1068 	ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1069 out:
1070 	if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1071 		dst->seqid = 0;
1072 	return ret;
1073 }
1074 
nfs_alloc_seqid(struct nfs_seqid_counter * counter,gfp_t gfp_mask)1075 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1076 {
1077 	struct nfs_seqid *new;
1078 
1079 	new = kmalloc(sizeof(*new), gfp_mask);
1080 	if (new == NULL)
1081 		return ERR_PTR(-ENOMEM);
1082 	new->sequence = counter;
1083 	INIT_LIST_HEAD(&new->list);
1084 	new->task = NULL;
1085 	return new;
1086 }
1087 
nfs_release_seqid(struct nfs_seqid * seqid)1088 void nfs_release_seqid(struct nfs_seqid *seqid)
1089 {
1090 	struct nfs_seqid_counter *sequence;
1091 
1092 	if (seqid == NULL || list_empty(&seqid->list))
1093 		return;
1094 	sequence = seqid->sequence;
1095 	spin_lock(&sequence->lock);
1096 	list_del_init(&seqid->list);
1097 	if (!list_empty(&sequence->list)) {
1098 		struct nfs_seqid *next;
1099 
1100 		next = list_first_entry(&sequence->list,
1101 				struct nfs_seqid, list);
1102 		rpc_wake_up_queued_task(&sequence->wait, next->task);
1103 	}
1104 	spin_unlock(&sequence->lock);
1105 }
1106 
nfs_free_seqid(struct nfs_seqid * seqid)1107 void nfs_free_seqid(struct nfs_seqid *seqid)
1108 {
1109 	nfs_release_seqid(seqid);
1110 	kfree(seqid);
1111 }
1112 
1113 /*
1114  * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1115  * failed with a seqid incrementing error -
1116  * see comments nfs4.h:seqid_mutating_error()
1117  */
nfs_increment_seqid(int status,struct nfs_seqid * seqid)1118 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1119 {
1120 	switch (status) {
1121 		case 0:
1122 			break;
1123 		case -NFS4ERR_BAD_SEQID:
1124 			if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1125 				return;
1126 			pr_warn_ratelimited("NFS: v4 server returned a bad"
1127 					" sequence-id error on an"
1128 					" unconfirmed sequence %p!\n",
1129 					seqid->sequence);
1130 			return;
1131 		case -NFS4ERR_STALE_CLIENTID:
1132 		case -NFS4ERR_STALE_STATEID:
1133 		case -NFS4ERR_BAD_STATEID:
1134 		case -NFS4ERR_BADXDR:
1135 		case -NFS4ERR_RESOURCE:
1136 		case -NFS4ERR_NOFILEHANDLE:
1137 		case -NFS4ERR_MOVED:
1138 			/* Non-seqid mutating errors */
1139 			return;
1140 	}
1141 	/*
1142 	 * Note: no locking needed as we are guaranteed to be first
1143 	 * on the sequence list
1144 	 */
1145 	seqid->sequence->counter++;
1146 }
1147 
nfs_increment_open_seqid(int status,struct nfs_seqid * seqid)1148 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1149 {
1150 	struct nfs4_state_owner *sp;
1151 
1152 	if (seqid == NULL)
1153 		return;
1154 
1155 	sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1156 	if (status == -NFS4ERR_BAD_SEQID)
1157 		nfs4_reset_state_owner(sp);
1158 	if (!nfs4_has_session(sp->so_server->nfs_client))
1159 		nfs_increment_seqid(status, seqid);
1160 }
1161 
1162 /*
1163  * Increment the seqid if the LOCK/LOCKU succeeded, or
1164  * failed with a seqid incrementing error -
1165  * see comments nfs4.h:seqid_mutating_error()
1166  */
nfs_increment_lock_seqid(int status,struct nfs_seqid * seqid)1167 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1168 {
1169 	if (seqid != NULL)
1170 		nfs_increment_seqid(status, seqid);
1171 }
1172 
nfs_wait_on_sequence(struct nfs_seqid * seqid,struct rpc_task * task)1173 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1174 {
1175 	struct nfs_seqid_counter *sequence;
1176 	int status = 0;
1177 
1178 	if (seqid == NULL)
1179 		goto out;
1180 	sequence = seqid->sequence;
1181 	spin_lock(&sequence->lock);
1182 	seqid->task = task;
1183 	if (list_empty(&seqid->list))
1184 		list_add_tail(&seqid->list, &sequence->list);
1185 	if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1186 		goto unlock;
1187 	rpc_sleep_on(&sequence->wait, task, NULL);
1188 	status = -EAGAIN;
1189 unlock:
1190 	spin_unlock(&sequence->lock);
1191 out:
1192 	return status;
1193 }
1194 
1195 static int nfs4_run_state_manager(void *);
1196 
nfs4_clear_state_manager_bit(struct nfs_client * clp)1197 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1198 {
1199 	clear_and_wake_up_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1200 	rpc_wake_up(&clp->cl_rpcwaitq);
1201 }
1202 
1203 /*
1204  * Schedule the nfs_client asynchronous state management routine
1205  */
nfs4_schedule_state_manager(struct nfs_client * clp)1206 void nfs4_schedule_state_manager(struct nfs_client *clp)
1207 {
1208 	struct task_struct *task;
1209 	char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1210 	struct rpc_clnt *cl = clp->cl_rpcclient;
1211 
1212 	while (cl != cl->cl_parent)
1213 		cl = cl->cl_parent;
1214 
1215 	set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1216 	if (test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) != 0) {
1217 		wake_up_var(&clp->cl_state);
1218 		return;
1219 	}
1220 	set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1221 	__module_get(THIS_MODULE);
1222 	refcount_inc(&clp->cl_count);
1223 
1224 	/* The rcu_read_lock() is not strictly necessary, as the state
1225 	 * manager is the only thread that ever changes the rpc_xprt
1226 	 * after it's initialized.  At this point, we're single threaded. */
1227 	rcu_read_lock();
1228 	snprintf(buf, sizeof(buf), "%s-manager",
1229 			rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1230 	rcu_read_unlock();
1231 	task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1232 	if (IS_ERR(task)) {
1233 		printk(KERN_ERR "%s: kthread_run: %ld\n",
1234 			__func__, PTR_ERR(task));
1235 		nfs4_clear_state_manager_bit(clp);
1236 		clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1237 		nfs_put_client(clp);
1238 		module_put(THIS_MODULE);
1239 	}
1240 }
1241 
1242 /*
1243  * Schedule a lease recovery attempt
1244  */
nfs4_schedule_lease_recovery(struct nfs_client * clp)1245 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1246 {
1247 	if (!clp)
1248 		return;
1249 	if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1250 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1251 	dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1252 			clp->cl_hostname);
1253 	nfs4_schedule_state_manager(clp);
1254 }
1255 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1256 
1257 /**
1258  * nfs4_schedule_migration_recovery - trigger migration recovery
1259  *
1260  * @server: FSID that is migrating
1261  *
1262  * Returns zero if recovery has started, otherwise a negative NFS4ERR
1263  * value is returned.
1264  */
nfs4_schedule_migration_recovery(const struct nfs_server * server)1265 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1266 {
1267 	struct nfs_client *clp = server->nfs_client;
1268 
1269 	if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1270 		pr_err("NFS: volatile file handles not supported (server %s)\n",
1271 				clp->cl_hostname);
1272 		return -NFS4ERR_IO;
1273 	}
1274 
1275 	if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1276 		return -NFS4ERR_IO;
1277 
1278 	dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1279 			__func__,
1280 			(unsigned long long)server->fsid.major,
1281 			(unsigned long long)server->fsid.minor,
1282 			clp->cl_hostname);
1283 
1284 	set_bit(NFS_MIG_IN_TRANSITION,
1285 			&((struct nfs_server *)server)->mig_status);
1286 	set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1287 
1288 	nfs4_schedule_state_manager(clp);
1289 	return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1292 
1293 /**
1294  * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1295  *
1296  * @clp: server to check for moved leases
1297  *
1298  */
nfs4_schedule_lease_moved_recovery(struct nfs_client * clp)1299 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1300 {
1301 	dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1302 		__func__, clp->cl_clientid, clp->cl_hostname);
1303 
1304 	set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1305 	nfs4_schedule_state_manager(clp);
1306 }
1307 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1308 
nfs4_wait_clnt_recover(struct nfs_client * clp)1309 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1310 {
1311 	int res;
1312 
1313 	might_sleep();
1314 
1315 	refcount_inc(&clp->cl_count);
1316 	res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1317 				 nfs_wait_bit_killable, TASK_KILLABLE);
1318 	if (res)
1319 		goto out;
1320 	if (clp->cl_cons_state < 0)
1321 		res = clp->cl_cons_state;
1322 out:
1323 	nfs_put_client(clp);
1324 	return res;
1325 }
1326 
nfs4_client_recover_expired_lease(struct nfs_client * clp)1327 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1328 {
1329 	unsigned int loop;
1330 	int ret;
1331 
1332 	for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1333 		ret = nfs4_wait_clnt_recover(clp);
1334 		if (ret != 0)
1335 			break;
1336 		if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1337 		    !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1338 			break;
1339 		nfs4_schedule_state_manager(clp);
1340 		ret = -EIO;
1341 	}
1342 	return ret;
1343 }
1344 
1345 /*
1346  * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1347  * @clp: client to process
1348  *
1349  * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1350  * resend of the SETCLIENTID and hence re-establish the
1351  * callback channel. Then return all existing delegations.
1352  */
nfs40_handle_cb_pathdown(struct nfs_client * clp)1353 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1354 {
1355 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1356 	nfs_expire_all_delegations(clp);
1357 	dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1358 			clp->cl_hostname);
1359 }
1360 
nfs4_schedule_path_down_recovery(struct nfs_client * clp)1361 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1362 {
1363 	nfs40_handle_cb_pathdown(clp);
1364 	nfs4_schedule_state_manager(clp);
1365 }
1366 
nfs4_state_mark_reclaim_reboot(struct nfs_client * clp,struct nfs4_state * state)1367 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1368 {
1369 
1370 	if (!nfs4_valid_open_stateid(state))
1371 		return 0;
1372 	set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1373 	/* Don't recover state that expired before the reboot */
1374 	if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1375 		clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1376 		return 0;
1377 	}
1378 	set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1379 	set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1380 	return 1;
1381 }
1382 
nfs4_state_mark_reclaim_nograce(struct nfs_client * clp,struct nfs4_state * state)1383 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1384 {
1385 	if (!nfs4_valid_open_stateid(state))
1386 		return 0;
1387 	set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1388 	clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1389 	set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1390 	set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1391 	return 1;
1392 }
1393 
nfs4_schedule_stateid_recovery(const struct nfs_server * server,struct nfs4_state * state)1394 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1395 {
1396 	struct nfs_client *clp = server->nfs_client;
1397 
1398 	if (!nfs4_state_mark_reclaim_nograce(clp, state))
1399 		return -EBADF;
1400 	nfs_inode_find_delegation_state_and_recover(state->inode,
1401 			&state->stateid);
1402 	dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1403 			clp->cl_hostname);
1404 	nfs4_schedule_state_manager(clp);
1405 	return 0;
1406 }
1407 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1408 
1409 static struct nfs4_lock_state *
nfs_state_find_lock_state_by_stateid(struct nfs4_state * state,const nfs4_stateid * stateid)1410 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1411 		const nfs4_stateid *stateid)
1412 {
1413 	struct nfs4_lock_state *pos;
1414 
1415 	list_for_each_entry(pos, &state->lock_states, ls_locks) {
1416 		if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1417 			continue;
1418 		if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1419 			return pos;
1420 	}
1421 	return NULL;
1422 }
1423 
nfs_state_lock_state_matches_stateid(struct nfs4_state * state,const nfs4_stateid * stateid)1424 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1425 		const nfs4_stateid *stateid)
1426 {
1427 	bool found = false;
1428 
1429 	if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1430 		spin_lock(&state->state_lock);
1431 		if (nfs_state_find_lock_state_by_stateid(state, stateid))
1432 			found = true;
1433 		spin_unlock(&state->state_lock);
1434 	}
1435 	return found;
1436 }
1437 
nfs_inode_find_state_and_recover(struct inode * inode,const nfs4_stateid * stateid)1438 void nfs_inode_find_state_and_recover(struct inode *inode,
1439 		const nfs4_stateid *stateid)
1440 {
1441 	struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1442 	struct nfs_inode *nfsi = NFS_I(inode);
1443 	struct nfs_open_context *ctx;
1444 	struct nfs4_state *state;
1445 	bool found = false;
1446 
1447 	rcu_read_lock();
1448 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1449 		state = ctx->state;
1450 		if (state == NULL)
1451 			continue;
1452 		if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1453 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1454 			found = true;
1455 			continue;
1456 		}
1457 		if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1458 		    nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1459 		    nfs4_state_mark_reclaim_nograce(clp, state)) {
1460 			found = true;
1461 			continue;
1462 		}
1463 		if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1464 		    nfs4_state_mark_reclaim_nograce(clp, state))
1465 			found = true;
1466 	}
1467 	rcu_read_unlock();
1468 
1469 	nfs_inode_find_delegation_state_and_recover(inode, stateid);
1470 	if (found)
1471 		nfs4_schedule_state_manager(clp);
1472 }
1473 
nfs4_state_mark_open_context_bad(struct nfs4_state * state,int err)1474 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1475 {
1476 	struct inode *inode = state->inode;
1477 	struct nfs_inode *nfsi = NFS_I(inode);
1478 	struct nfs_open_context *ctx;
1479 
1480 	rcu_read_lock();
1481 	list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1482 		if (ctx->state != state)
1483 			continue;
1484 		set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1485 		pr_warn("NFSv4: state recovery failed for open file %pd2, "
1486 				"error = %d\n", ctx->dentry, err);
1487 	}
1488 	rcu_read_unlock();
1489 }
1490 
nfs4_state_mark_recovery_failed(struct nfs4_state * state,int error)1491 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1492 {
1493 	set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1494 	nfs4_state_mark_open_context_bad(state, error);
1495 }
1496 
1497 
nfs4_reclaim_locks(struct nfs4_state * state,const struct nfs4_state_recovery_ops * ops)1498 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1499 {
1500 	struct inode *inode = state->inode;
1501 	struct nfs_inode *nfsi = NFS_I(inode);
1502 	struct file_lock *fl;
1503 	struct nfs4_lock_state *lsp;
1504 	int status = 0;
1505 	struct file_lock_context *flctx = inode->i_flctx;
1506 	struct list_head *list;
1507 
1508 	if (flctx == NULL)
1509 		return 0;
1510 
1511 	list = &flctx->flc_posix;
1512 
1513 	/* Guard against delegation returns and new lock/unlock calls */
1514 	down_write(&nfsi->rwsem);
1515 	spin_lock(&flctx->flc_lock);
1516 restart:
1517 	list_for_each_entry(fl, list, fl_list) {
1518 		if (nfs_file_open_context(fl->fl_file)->state != state)
1519 			continue;
1520 		spin_unlock(&flctx->flc_lock);
1521 		status = ops->recover_lock(state, fl);
1522 		switch (status) {
1523 		case 0:
1524 			break;
1525 		case -ETIMEDOUT:
1526 		case -ESTALE:
1527 		case -NFS4ERR_ADMIN_REVOKED:
1528 		case -NFS4ERR_STALE_STATEID:
1529 		case -NFS4ERR_BAD_STATEID:
1530 		case -NFS4ERR_EXPIRED:
1531 		case -NFS4ERR_NO_GRACE:
1532 		case -NFS4ERR_STALE_CLIENTID:
1533 		case -NFS4ERR_BADSESSION:
1534 		case -NFS4ERR_BADSLOT:
1535 		case -NFS4ERR_BAD_HIGH_SLOT:
1536 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1537 			goto out;
1538 		default:
1539 			pr_err("NFS: %s: unhandled error %d\n",
1540 					__func__, status);
1541 			fallthrough;
1542 		case -ENOMEM:
1543 		case -NFS4ERR_DENIED:
1544 		case -NFS4ERR_RECLAIM_BAD:
1545 		case -NFS4ERR_RECLAIM_CONFLICT:
1546 			lsp = fl->fl_u.nfs4_fl.owner;
1547 			if (lsp)
1548 				set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1549 			status = 0;
1550 		}
1551 		spin_lock(&flctx->flc_lock);
1552 	}
1553 	if (list == &flctx->flc_posix) {
1554 		list = &flctx->flc_flock;
1555 		goto restart;
1556 	}
1557 	spin_unlock(&flctx->flc_lock);
1558 out:
1559 	up_write(&nfsi->rwsem);
1560 	return status;
1561 }
1562 
1563 #ifdef CONFIG_NFS_V4_2
nfs42_complete_copies(struct nfs4_state_owner * sp,struct nfs4_state * state)1564 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1565 {
1566 	struct nfs4_copy_state *copy;
1567 
1568 	if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1569 		!test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1570 		return;
1571 
1572 	spin_lock(&sp->so_server->nfs_client->cl_lock);
1573 	list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1574 		if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1575 				!nfs4_stateid_match_other(&state->stateid,
1576 				&copy->parent_dst_state->stateid)))
1577 				continue;
1578 		copy->flags = 1;
1579 		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1580 				&state->flags)) {
1581 			clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1582 			complete(&copy->completion);
1583 		}
1584 	}
1585 	list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1586 		if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1587 				!nfs4_stateid_match_other(&state->stateid,
1588 				&copy->parent_src_state->stateid)))
1589 				continue;
1590 		copy->flags = 1;
1591 		if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1592 				&state->flags))
1593 			complete(&copy->completion);
1594 	}
1595 	spin_unlock(&sp->so_server->nfs_client->cl_lock);
1596 }
1597 #else /* !CONFIG_NFS_V4_2 */
nfs42_complete_copies(struct nfs4_state_owner * sp,struct nfs4_state * state)1598 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1599 					 struct nfs4_state *state)
1600 {
1601 }
1602 #endif /* CONFIG_NFS_V4_2 */
1603 
__nfs4_reclaim_open_state(struct nfs4_state_owner * sp,struct nfs4_state * state,const struct nfs4_state_recovery_ops * ops,int * lost_locks)1604 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1605 				     const struct nfs4_state_recovery_ops *ops,
1606 				     int *lost_locks)
1607 {
1608 	struct nfs4_lock_state *lock;
1609 	int status;
1610 
1611 	status = ops->recover_open(sp, state);
1612 	if (status < 0)
1613 		return status;
1614 
1615 	status = nfs4_reclaim_locks(state, ops);
1616 	if (status < 0)
1617 		return status;
1618 
1619 	if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1620 		spin_lock(&state->state_lock);
1621 		list_for_each_entry(lock, &state->lock_states, ls_locks) {
1622 			trace_nfs4_state_lock_reclaim(state, lock);
1623 			if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1624 				*lost_locks += 1;
1625 		}
1626 		spin_unlock(&state->state_lock);
1627 	}
1628 
1629 	nfs42_complete_copies(sp, state);
1630 	clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1631 	return status;
1632 }
1633 
nfs4_reclaim_open_state(struct nfs4_state_owner * sp,const struct nfs4_state_recovery_ops * ops,int * lost_locks)1634 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp,
1635 				   const struct nfs4_state_recovery_ops *ops,
1636 				   int *lost_locks)
1637 {
1638 	struct nfs4_state *state;
1639 	unsigned int loop = 0;
1640 	int status = 0;
1641 #ifdef CONFIG_NFS_V4_2
1642 	bool found_ssc_copy_state = false;
1643 #endif /* CONFIG_NFS_V4_2 */
1644 
1645 	/* Note: we rely on the sp->so_states list being ordered
1646 	 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1647 	 * states first.
1648 	 * This is needed to ensure that the server won't give us any
1649 	 * read delegations that we have to return if, say, we are
1650 	 * recovering after a network partition or a reboot from a
1651 	 * server that doesn't support a grace period.
1652 	 */
1653 	spin_lock(&sp->so_lock);
1654 	raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1655 restart:
1656 	list_for_each_entry(state, &sp->so_states, open_states) {
1657 		if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1658 			continue;
1659 		if (!nfs4_valid_open_stateid(state))
1660 			continue;
1661 		if (state->state == 0)
1662 			continue;
1663 #ifdef CONFIG_NFS_V4_2
1664 		if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1665 			nfs4_state_mark_recovery_failed(state, -EIO);
1666 			found_ssc_copy_state = true;
1667 			continue;
1668 		}
1669 #endif /* CONFIG_NFS_V4_2 */
1670 		refcount_inc(&state->count);
1671 		spin_unlock(&sp->so_lock);
1672 		status = __nfs4_reclaim_open_state(sp, state, ops, lost_locks);
1673 
1674 		switch (status) {
1675 		default:
1676 			if (status >= 0) {
1677 				loop = 0;
1678 				break;
1679 			}
1680 			printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1681 			fallthrough;
1682 		case -ENOENT:
1683 		case -ENOMEM:
1684 		case -EACCES:
1685 		case -EROFS:
1686 		case -EIO:
1687 		case -ESTALE:
1688 			/* Open state on this file cannot be recovered */
1689 			nfs4_state_mark_recovery_failed(state, status);
1690 			break;
1691 		case -EAGAIN:
1692 			ssleep(1);
1693 			if (loop++ < 10) {
1694 				set_bit(ops->state_flag_bit, &state->flags);
1695 				break;
1696 			}
1697 			fallthrough;
1698 		case -NFS4ERR_ADMIN_REVOKED:
1699 		case -NFS4ERR_STALE_STATEID:
1700 		case -NFS4ERR_OLD_STATEID:
1701 		case -NFS4ERR_BAD_STATEID:
1702 		case -NFS4ERR_RECLAIM_BAD:
1703 		case -NFS4ERR_RECLAIM_CONFLICT:
1704 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1705 			break;
1706 		case -NFS4ERR_EXPIRED:
1707 		case -NFS4ERR_NO_GRACE:
1708 			nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1709 			fallthrough;
1710 		case -NFS4ERR_STALE_CLIENTID:
1711 		case -NFS4ERR_BADSESSION:
1712 		case -NFS4ERR_BADSLOT:
1713 		case -NFS4ERR_BAD_HIGH_SLOT:
1714 		case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1715 		case -ETIMEDOUT:
1716 			goto out_err;
1717 		}
1718 		nfs4_put_open_state(state);
1719 		spin_lock(&sp->so_lock);
1720 		goto restart;
1721 	}
1722 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1723 	spin_unlock(&sp->so_lock);
1724 #ifdef CONFIG_NFS_V4_2
1725 	if (found_ssc_copy_state)
1726 		return -EIO;
1727 #endif /* CONFIG_NFS_V4_2 */
1728 	return 0;
1729 out_err:
1730 	nfs4_put_open_state(state);
1731 	spin_lock(&sp->so_lock);
1732 	raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1733 	spin_unlock(&sp->so_lock);
1734 	return status;
1735 }
1736 
nfs4_clear_open_state(struct nfs4_state * state)1737 static void nfs4_clear_open_state(struct nfs4_state *state)
1738 {
1739 	struct nfs4_lock_state *lock;
1740 
1741 	clear_bit(NFS_DELEGATED_STATE, &state->flags);
1742 	clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1743 	clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1744 	clear_bit(NFS_O_RDWR_STATE, &state->flags);
1745 	spin_lock(&state->state_lock);
1746 	list_for_each_entry(lock, &state->lock_states, ls_locks) {
1747 		lock->ls_seqid.flags = 0;
1748 		clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1749 	}
1750 	spin_unlock(&state->state_lock);
1751 }
1752 
nfs4_reset_seqids(struct nfs_server * server,int (* mark_reclaim)(struct nfs_client * clp,struct nfs4_state * state))1753 static void nfs4_reset_seqids(struct nfs_server *server,
1754 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1755 {
1756 	struct nfs_client *clp = server->nfs_client;
1757 	struct nfs4_state_owner *sp;
1758 	struct rb_node *pos;
1759 	struct nfs4_state *state;
1760 
1761 	spin_lock(&clp->cl_lock);
1762 	for (pos = rb_first(&server->state_owners);
1763 	     pos != NULL;
1764 	     pos = rb_next(pos)) {
1765 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1766 		sp->so_seqid.flags = 0;
1767 		spin_lock(&sp->so_lock);
1768 		list_for_each_entry(state, &sp->so_states, open_states) {
1769 			if (mark_reclaim(clp, state))
1770 				nfs4_clear_open_state(state);
1771 		}
1772 		spin_unlock(&sp->so_lock);
1773 	}
1774 	spin_unlock(&clp->cl_lock);
1775 }
1776 
nfs4_state_mark_reclaim_helper(struct nfs_client * clp,int (* mark_reclaim)(struct nfs_client * clp,struct nfs4_state * state))1777 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1778 	int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1779 {
1780 	struct nfs_server *server;
1781 
1782 	rcu_read_lock();
1783 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1784 		nfs4_reset_seqids(server, mark_reclaim);
1785 	rcu_read_unlock();
1786 }
1787 
nfs4_state_start_reclaim_reboot(struct nfs_client * clp)1788 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1789 {
1790 	/* Mark all delegations for reclaim */
1791 	nfs_delegation_mark_reclaim(clp);
1792 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1793 }
1794 
nfs4_reclaim_complete(struct nfs_client * clp,const struct nfs4_state_recovery_ops * ops,const struct cred * cred)1795 static int nfs4_reclaim_complete(struct nfs_client *clp,
1796 				 const struct nfs4_state_recovery_ops *ops,
1797 				 const struct cred *cred)
1798 {
1799 	/* Notify the server we're done reclaiming our state */
1800 	if (ops->reclaim_complete)
1801 		return ops->reclaim_complete(clp, cred);
1802 	return 0;
1803 }
1804 
nfs4_clear_reclaim_server(struct nfs_server * server)1805 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1806 {
1807 	struct nfs_client *clp = server->nfs_client;
1808 	struct nfs4_state_owner *sp;
1809 	struct rb_node *pos;
1810 	struct nfs4_state *state;
1811 
1812 	spin_lock(&clp->cl_lock);
1813 	for (pos = rb_first(&server->state_owners);
1814 	     pos != NULL;
1815 	     pos = rb_next(pos)) {
1816 		sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1817 		spin_lock(&sp->so_lock);
1818 		list_for_each_entry(state, &sp->so_states, open_states) {
1819 			if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1820 						&state->flags))
1821 				continue;
1822 			nfs4_state_mark_reclaim_nograce(clp, state);
1823 		}
1824 		spin_unlock(&sp->so_lock);
1825 	}
1826 	spin_unlock(&clp->cl_lock);
1827 }
1828 
nfs4_state_clear_reclaim_reboot(struct nfs_client * clp)1829 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1830 {
1831 	struct nfs_server *server;
1832 
1833 	if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1834 		return 0;
1835 
1836 	rcu_read_lock();
1837 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1838 		nfs4_clear_reclaim_server(server);
1839 	rcu_read_unlock();
1840 
1841 	nfs_delegation_reap_unclaimed(clp);
1842 	return 1;
1843 }
1844 
nfs4_state_end_reclaim_reboot(struct nfs_client * clp)1845 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1846 {
1847 	const struct nfs4_state_recovery_ops *ops;
1848 	const struct cred *cred;
1849 	int err;
1850 
1851 	if (!nfs4_state_clear_reclaim_reboot(clp))
1852 		return;
1853 	ops = clp->cl_mvops->reboot_recovery_ops;
1854 	cred = nfs4_get_clid_cred(clp);
1855 	err = nfs4_reclaim_complete(clp, ops, cred);
1856 	put_cred(cred);
1857 	if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1858 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1859 }
1860 
nfs4_state_start_reclaim_nograce(struct nfs_client * clp)1861 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1862 {
1863 	nfs_mark_test_expired_all_delegations(clp);
1864 	nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1865 }
1866 
nfs4_recovery_handle_error(struct nfs_client * clp,int error)1867 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1868 {
1869 	switch (error) {
1870 	case 0:
1871 		break;
1872 	case -NFS4ERR_CB_PATH_DOWN:
1873 		nfs40_handle_cb_pathdown(clp);
1874 		break;
1875 	case -NFS4ERR_NO_GRACE:
1876 		nfs4_state_end_reclaim_reboot(clp);
1877 		break;
1878 	case -NFS4ERR_STALE_CLIENTID:
1879 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1880 		nfs4_state_start_reclaim_reboot(clp);
1881 		break;
1882 	case -NFS4ERR_EXPIRED:
1883 		set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1884 		nfs4_state_start_reclaim_nograce(clp);
1885 		break;
1886 	case -NFS4ERR_BADSESSION:
1887 	case -NFS4ERR_BADSLOT:
1888 	case -NFS4ERR_BAD_HIGH_SLOT:
1889 	case -NFS4ERR_DEADSESSION:
1890 	case -NFS4ERR_SEQ_FALSE_RETRY:
1891 	case -NFS4ERR_SEQ_MISORDERED:
1892 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1893 		/* Zero session reset errors */
1894 		break;
1895 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1896 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1897 		break;
1898 	default:
1899 		dprintk("%s: failed to handle error %d for server %s\n",
1900 				__func__, error, clp->cl_hostname);
1901 		return error;
1902 	}
1903 	dprintk("%s: handled error %d for server %s\n", __func__, error,
1904 			clp->cl_hostname);
1905 	return 0;
1906 }
1907 
nfs4_do_reclaim(struct nfs_client * clp,const struct nfs4_state_recovery_ops * ops)1908 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1909 {
1910 	struct nfs4_state_owner *sp;
1911 	struct nfs_server *server;
1912 	struct rb_node *pos;
1913 	LIST_HEAD(freeme);
1914 	int status = 0;
1915 	int lost_locks = 0;
1916 
1917 restart:
1918 	rcu_read_lock();
1919 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1920 		nfs4_purge_state_owners(server, &freeme);
1921 		spin_lock(&clp->cl_lock);
1922 		for (pos = rb_first(&server->state_owners);
1923 		     pos != NULL;
1924 		     pos = rb_next(pos)) {
1925 			sp = rb_entry(pos,
1926 				struct nfs4_state_owner, so_server_node);
1927 			if (!test_and_clear_bit(ops->owner_flag_bit,
1928 							&sp->so_flags))
1929 				continue;
1930 			if (!atomic_inc_not_zero(&sp->so_count))
1931 				continue;
1932 			spin_unlock(&clp->cl_lock);
1933 			rcu_read_unlock();
1934 
1935 			status = nfs4_reclaim_open_state(sp, ops, &lost_locks);
1936 			if (status < 0) {
1937 				if (lost_locks)
1938 					pr_warn("NFS: %s: lost %d locks\n",
1939 						clp->cl_hostname, lost_locks);
1940 				set_bit(ops->owner_flag_bit, &sp->so_flags);
1941 				nfs4_put_state_owner(sp);
1942 				status = nfs4_recovery_handle_error(clp, status);
1943 				return (status != 0) ? status : -EAGAIN;
1944 			}
1945 
1946 			nfs4_put_state_owner(sp);
1947 			goto restart;
1948 		}
1949 		spin_unlock(&clp->cl_lock);
1950 	}
1951 	rcu_read_unlock();
1952 	nfs4_free_state_owners(&freeme);
1953 	if (lost_locks)
1954 		pr_warn("NFS: %s: lost %d locks\n",
1955 			clp->cl_hostname, lost_locks);
1956 	return 0;
1957 }
1958 
nfs4_check_lease(struct nfs_client * clp)1959 static int nfs4_check_lease(struct nfs_client *clp)
1960 {
1961 	const struct cred *cred;
1962 	const struct nfs4_state_maintenance_ops *ops =
1963 		clp->cl_mvops->state_renewal_ops;
1964 	int status;
1965 
1966 	/* Is the client already known to have an expired lease? */
1967 	if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1968 		return 0;
1969 	cred = ops->get_state_renewal_cred(clp);
1970 	if (cred == NULL) {
1971 		cred = nfs4_get_clid_cred(clp);
1972 		status = -ENOKEY;
1973 		if (cred == NULL)
1974 			goto out;
1975 	}
1976 	status = ops->renew_lease(clp, cred);
1977 	put_cred(cred);
1978 	if (status == -ETIMEDOUT) {
1979 		set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1980 		return 0;
1981 	}
1982 out:
1983 	return nfs4_recovery_handle_error(clp, status);
1984 }
1985 
1986 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1987  * and for recoverable errors on EXCHANGE_ID for v4.1
1988  */
nfs4_handle_reclaim_lease_error(struct nfs_client * clp,int status)1989 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1990 {
1991 	switch (status) {
1992 	case -NFS4ERR_SEQ_MISORDERED:
1993 		if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1994 			return -ESERVERFAULT;
1995 		/* Lease confirmation error: retry after purging the lease */
1996 		ssleep(1);
1997 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1998 		break;
1999 	case -NFS4ERR_STALE_CLIENTID:
2000 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2001 		nfs4_state_start_reclaim_reboot(clp);
2002 		break;
2003 	case -NFS4ERR_CLID_INUSE:
2004 		pr_err("NFS: Server %s reports our clientid is in use\n",
2005 			clp->cl_hostname);
2006 		nfs_mark_client_ready(clp, -EPERM);
2007 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2008 		return -EPERM;
2009 	case -EACCES:
2010 	case -NFS4ERR_DELAY:
2011 	case -EAGAIN:
2012 		ssleep(1);
2013 		break;
2014 
2015 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2016 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2017 			nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2018 		dprintk("%s: exit with error %d for server %s\n",
2019 				__func__, -EPROTONOSUPPORT, clp->cl_hostname);
2020 		return -EPROTONOSUPPORT;
2021 	case -ENOSPC:
2022 		if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2023 			nfs_mark_client_ready(clp, -EIO);
2024 		return -EIO;
2025 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2026 				 * in nfs4_exchange_id */
2027 	default:
2028 		dprintk("%s: exit with error %d for server %s\n", __func__,
2029 				status, clp->cl_hostname);
2030 		return status;
2031 	}
2032 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2033 	dprintk("%s: handled error %d for server %s\n", __func__, status,
2034 			clp->cl_hostname);
2035 	return 0;
2036 }
2037 
nfs4_establish_lease(struct nfs_client * clp)2038 static int nfs4_establish_lease(struct nfs_client *clp)
2039 {
2040 	const struct cred *cred;
2041 	const struct nfs4_state_recovery_ops *ops =
2042 		clp->cl_mvops->reboot_recovery_ops;
2043 	int status;
2044 
2045 	status = nfs4_begin_drain_session(clp);
2046 	if (status != 0)
2047 		return status;
2048 	cred = nfs4_get_clid_cred(clp);
2049 	if (cred == NULL)
2050 		return -ENOENT;
2051 	status = ops->establish_clid(clp, cred);
2052 	put_cred(cred);
2053 	if (status != 0)
2054 		return status;
2055 	pnfs_destroy_all_layouts(clp);
2056 	return 0;
2057 }
2058 
2059 /*
2060  * Returns zero or a negative errno.  NFS4ERR values are converted
2061  * to local errno values.
2062  */
nfs4_reclaim_lease(struct nfs_client * clp)2063 static int nfs4_reclaim_lease(struct nfs_client *clp)
2064 {
2065 	int status;
2066 
2067 	status = nfs4_establish_lease(clp);
2068 	if (status < 0)
2069 		return nfs4_handle_reclaim_lease_error(clp, status);
2070 	if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2071 		nfs4_state_start_reclaim_nograce(clp);
2072 	if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2073 		set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2074 	clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2075 	clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2076 	return 0;
2077 }
2078 
nfs4_purge_lease(struct nfs_client * clp)2079 static int nfs4_purge_lease(struct nfs_client *clp)
2080 {
2081 	int status;
2082 
2083 	status = nfs4_establish_lease(clp);
2084 	if (status < 0)
2085 		return nfs4_handle_reclaim_lease_error(clp, status);
2086 	clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2087 	set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2088 	nfs4_state_start_reclaim_nograce(clp);
2089 	return 0;
2090 }
2091 
2092 /*
2093  * Try remote migration of one FSID from a source server to a
2094  * destination server.  The source server provides a list of
2095  * potential destinations.
2096  *
2097  * Returns zero or a negative NFS4ERR status code.
2098  */
nfs4_try_migration(struct nfs_server * server,const struct cred * cred)2099 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2100 {
2101 	struct nfs_client *clp = server->nfs_client;
2102 	struct nfs4_fs_locations *locations = NULL;
2103 	struct inode *inode;
2104 	struct page *page;
2105 	int status, result;
2106 
2107 	dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2108 			(unsigned long long)server->fsid.major,
2109 			(unsigned long long)server->fsid.minor,
2110 			clp->cl_hostname);
2111 
2112 	result = 0;
2113 	page = alloc_page(GFP_KERNEL);
2114 	locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2115 	if (page == NULL || locations == NULL) {
2116 		dprintk("<-- %s: no memory\n", __func__);
2117 		goto out;
2118 	}
2119 	locations->fattr = nfs_alloc_fattr();
2120 	if (locations->fattr == NULL) {
2121 		dprintk("<-- %s: no memory\n", __func__);
2122 		goto out;
2123 	}
2124 
2125 	inode = d_inode(server->super->s_root);
2126 	result = nfs4_proc_get_locations(server, NFS_FH(inode), locations,
2127 					 page, cred);
2128 	if (result) {
2129 		dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2130 			__func__, result);
2131 		goto out;
2132 	}
2133 
2134 	result = -NFS4ERR_NXIO;
2135 	if (!locations->nlocations)
2136 		goto out;
2137 
2138 	if (!(locations->fattr->valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2139 		dprintk("<-- %s: No fs_locations data, migration skipped\n",
2140 			__func__);
2141 		goto out;
2142 	}
2143 
2144 	status = nfs4_begin_drain_session(clp);
2145 	if (status != 0) {
2146 		result = status;
2147 		goto out;
2148 	}
2149 
2150 	status = nfs4_replace_transport(server, locations);
2151 	if (status != 0) {
2152 		dprintk("<-- %s: failed to replace transport: %d\n",
2153 			__func__, status);
2154 		goto out;
2155 	}
2156 
2157 	result = 0;
2158 	dprintk("<-- %s: migration succeeded\n", __func__);
2159 
2160 out:
2161 	if (page != NULL)
2162 		__free_page(page);
2163 	if (locations != NULL)
2164 		kfree(locations->fattr);
2165 	kfree(locations);
2166 	if (result) {
2167 		pr_err("NFS: migration recovery failed (server %s)\n",
2168 				clp->cl_hostname);
2169 		set_bit(NFS_MIG_FAILED, &server->mig_status);
2170 	}
2171 	return result;
2172 }
2173 
2174 /*
2175  * Returns zero or a negative NFS4ERR status code.
2176  */
nfs4_handle_migration(struct nfs_client * clp)2177 static int nfs4_handle_migration(struct nfs_client *clp)
2178 {
2179 	const struct nfs4_state_maintenance_ops *ops =
2180 				clp->cl_mvops->state_renewal_ops;
2181 	struct nfs_server *server;
2182 	const struct cred *cred;
2183 
2184 	dprintk("%s: migration reported on \"%s\"\n", __func__,
2185 			clp->cl_hostname);
2186 
2187 	cred = ops->get_state_renewal_cred(clp);
2188 	if (cred == NULL)
2189 		return -NFS4ERR_NOENT;
2190 
2191 	clp->cl_mig_gen++;
2192 restart:
2193 	rcu_read_lock();
2194 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2195 		int status;
2196 
2197 		if (server->mig_gen == clp->cl_mig_gen)
2198 			continue;
2199 		server->mig_gen = clp->cl_mig_gen;
2200 
2201 		if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2202 						&server->mig_status))
2203 			continue;
2204 
2205 		rcu_read_unlock();
2206 		status = nfs4_try_migration(server, cred);
2207 		if (status < 0) {
2208 			put_cred(cred);
2209 			return status;
2210 		}
2211 		goto restart;
2212 	}
2213 	rcu_read_unlock();
2214 	put_cred(cred);
2215 	return 0;
2216 }
2217 
2218 /*
2219  * Test each nfs_server on the clp's cl_superblocks list to see
2220  * if it's moved to another server.  Stop when the server no longer
2221  * returns NFS4ERR_LEASE_MOVED.
2222  */
nfs4_handle_lease_moved(struct nfs_client * clp)2223 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2224 {
2225 	const struct nfs4_state_maintenance_ops *ops =
2226 				clp->cl_mvops->state_renewal_ops;
2227 	struct nfs_server *server;
2228 	const struct cred *cred;
2229 
2230 	dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2231 			clp->cl_hostname);
2232 
2233 	cred = ops->get_state_renewal_cred(clp);
2234 	if (cred == NULL)
2235 		return -NFS4ERR_NOENT;
2236 
2237 	clp->cl_mig_gen++;
2238 restart:
2239 	rcu_read_lock();
2240 	list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2241 		struct inode *inode;
2242 		int status;
2243 
2244 		if (server->mig_gen == clp->cl_mig_gen)
2245 			continue;
2246 		server->mig_gen = clp->cl_mig_gen;
2247 
2248 		rcu_read_unlock();
2249 
2250 		inode = d_inode(server->super->s_root);
2251 		status = nfs4_proc_fsid_present(inode, cred);
2252 		if (status != -NFS4ERR_MOVED)
2253 			goto restart;	/* wasn't this one */
2254 		if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2255 			goto restart;	/* there are more */
2256 		goto out;
2257 	}
2258 	rcu_read_unlock();
2259 
2260 out:
2261 	put_cred(cred);
2262 	return 0;
2263 }
2264 
2265 /**
2266  * nfs4_discover_server_trunking - Detect server IP address trunking
2267  *
2268  * @clp: nfs_client under test
2269  * @result: OUT: found nfs_client, or clp
2270  *
2271  * Returns zero or a negative errno.  If zero is returned,
2272  * an nfs_client pointer is planted in "result".
2273  *
2274  * Note: since we are invoked in process context, and
2275  * not from inside the state manager, we cannot use
2276  * nfs4_handle_reclaim_lease_error().
2277  */
nfs4_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result)2278 int nfs4_discover_server_trunking(struct nfs_client *clp,
2279 				  struct nfs_client **result)
2280 {
2281 	const struct nfs4_state_recovery_ops *ops =
2282 				clp->cl_mvops->reboot_recovery_ops;
2283 	struct rpc_clnt *clnt;
2284 	const struct cred *cred;
2285 	int i, status;
2286 
2287 	dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2288 
2289 	clnt = clp->cl_rpcclient;
2290 	i = 0;
2291 
2292 	mutex_lock(&nfs_clid_init_mutex);
2293 again:
2294 	status  = -ENOENT;
2295 	cred = nfs4_get_clid_cred(clp);
2296 	if (cred == NULL)
2297 		goto out_unlock;
2298 
2299 	status = ops->detect_trunking(clp, result, cred);
2300 	put_cred(cred);
2301 	switch (status) {
2302 	case 0:
2303 	case -EINTR:
2304 	case -ERESTARTSYS:
2305 		break;
2306 	case -ETIMEDOUT:
2307 		if (clnt->cl_softrtry)
2308 			break;
2309 		fallthrough;
2310 	case -NFS4ERR_DELAY:
2311 	case -EAGAIN:
2312 		ssleep(1);
2313 		fallthrough;
2314 	case -NFS4ERR_STALE_CLIENTID:
2315 		dprintk("NFS: %s after status %d, retrying\n",
2316 			__func__, status);
2317 		goto again;
2318 	case -EACCES:
2319 		if (i++ == 0) {
2320 			nfs4_root_machine_cred(clp);
2321 			goto again;
2322 		}
2323 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2324 			break;
2325 		fallthrough;
2326 	case -NFS4ERR_CLID_INUSE:
2327 	case -NFS4ERR_WRONGSEC:
2328 		/* No point in retrying if we already used RPC_AUTH_UNIX */
2329 		if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2330 			status = -EPERM;
2331 			break;
2332 		}
2333 		clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2334 		if (IS_ERR(clnt)) {
2335 			status = PTR_ERR(clnt);
2336 			break;
2337 		}
2338 		/* Note: this is safe because we haven't yet marked the
2339 		 * client as ready, so we are the only user of
2340 		 * clp->cl_rpcclient
2341 		 */
2342 		clnt = xchg(&clp->cl_rpcclient, clnt);
2343 		rpc_shutdown_client(clnt);
2344 		clnt = clp->cl_rpcclient;
2345 		goto again;
2346 
2347 	case -NFS4ERR_MINOR_VERS_MISMATCH:
2348 		status = -EPROTONOSUPPORT;
2349 		break;
2350 
2351 	case -EKEYEXPIRED:
2352 	case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2353 				 * in nfs4_exchange_id */
2354 		status = -EKEYEXPIRED;
2355 		break;
2356 	default:
2357 		pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2358 				__func__, status);
2359 		status = -EIO;
2360 	}
2361 
2362 out_unlock:
2363 	mutex_unlock(&nfs_clid_init_mutex);
2364 	dprintk("NFS: %s: status = %d\n", __func__, status);
2365 	return status;
2366 }
2367 
2368 #ifdef CONFIG_NFS_V4_1
nfs4_schedule_session_recovery(struct nfs4_session * session,int err)2369 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2370 {
2371 	struct nfs_client *clp = session->clp;
2372 
2373 	switch (err) {
2374 	default:
2375 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2376 		break;
2377 	case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2378 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2379 	}
2380 	nfs4_schedule_state_manager(clp);
2381 }
2382 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2383 
nfs41_notify_server(struct nfs_client * clp)2384 void nfs41_notify_server(struct nfs_client *clp)
2385 {
2386 	/* Use CHECK_LEASE to ping the server with a SEQUENCE */
2387 	set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2388 	nfs4_schedule_state_manager(clp);
2389 }
2390 
nfs4_reset_all_state(struct nfs_client * clp)2391 static void nfs4_reset_all_state(struct nfs_client *clp)
2392 {
2393 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2394 		set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2395 		clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2396 		nfs4_state_start_reclaim_nograce(clp);
2397 		dprintk("%s: scheduling reset of all state for server %s!\n",
2398 				__func__, clp->cl_hostname);
2399 		nfs4_schedule_state_manager(clp);
2400 	}
2401 }
2402 
nfs41_handle_server_reboot(struct nfs_client * clp)2403 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2404 {
2405 	if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2406 		nfs4_state_start_reclaim_reboot(clp);
2407 		dprintk("%s: server %s rebooted!\n", __func__,
2408 				clp->cl_hostname);
2409 		nfs4_schedule_state_manager(clp);
2410 	}
2411 }
2412 
nfs41_handle_all_state_revoked(struct nfs_client * clp)2413 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2414 {
2415 	nfs4_reset_all_state(clp);
2416 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2417 }
2418 
nfs41_handle_some_state_revoked(struct nfs_client * clp)2419 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2420 {
2421 	nfs4_state_start_reclaim_nograce(clp);
2422 	nfs4_schedule_state_manager(clp);
2423 
2424 	dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2425 }
2426 
nfs41_handle_recallable_state_revoked(struct nfs_client * clp)2427 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2428 {
2429 	/* FIXME: For now, we destroy all layouts. */
2430 	pnfs_destroy_all_layouts(clp);
2431 	nfs_test_expired_all_delegations(clp);
2432 	dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2433 			clp->cl_hostname);
2434 }
2435 
nfs41_handle_backchannel_fault(struct nfs_client * clp)2436 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2437 {
2438 	set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2439 	nfs4_schedule_state_manager(clp);
2440 
2441 	dprintk("%s: server %s declared a backchannel fault\n", __func__,
2442 			clp->cl_hostname);
2443 }
2444 
nfs41_handle_cb_path_down(struct nfs_client * clp)2445 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2446 {
2447 	if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2448 		&clp->cl_state) == 0)
2449 		nfs4_schedule_state_manager(clp);
2450 }
2451 
nfs41_handle_sequence_flag_errors(struct nfs_client * clp,u32 flags,bool recovery)2452 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2453 		bool recovery)
2454 {
2455 	if (!flags)
2456 		return;
2457 
2458 	dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2459 		__func__, clp->cl_hostname, clp->cl_clientid, flags);
2460 	/*
2461 	 * If we're called from the state manager thread, then assume we're
2462 	 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2463 	 * Those flags are expected to remain set until we're done
2464 	 * recovering (see RFC5661, section 18.46.3).
2465 	 */
2466 	if (recovery)
2467 		goto out_recovery;
2468 
2469 	if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2470 		nfs41_handle_server_reboot(clp);
2471 	if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2472 		nfs41_handle_all_state_revoked(clp);
2473 	if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2474 			    SEQ4_STATUS_ADMIN_STATE_REVOKED))
2475 		nfs41_handle_some_state_revoked(clp);
2476 	if (flags & SEQ4_STATUS_LEASE_MOVED)
2477 		nfs4_schedule_lease_moved_recovery(clp);
2478 	if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2479 		nfs41_handle_recallable_state_revoked(clp);
2480 out_recovery:
2481 	if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2482 		nfs41_handle_backchannel_fault(clp);
2483 	else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2484 				SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2485 		nfs41_handle_cb_path_down(clp);
2486 }
2487 
nfs4_reset_session(struct nfs_client * clp)2488 static int nfs4_reset_session(struct nfs_client *clp)
2489 {
2490 	const struct cred *cred;
2491 	int status;
2492 
2493 	if (!nfs4_has_session(clp))
2494 		return 0;
2495 	status = nfs4_begin_drain_session(clp);
2496 	if (status != 0)
2497 		return status;
2498 	cred = nfs4_get_clid_cred(clp);
2499 	status = nfs4_proc_destroy_session(clp->cl_session, cred);
2500 	switch (status) {
2501 	case 0:
2502 	case -NFS4ERR_BADSESSION:
2503 	case -NFS4ERR_DEADSESSION:
2504 		break;
2505 	case -NFS4ERR_BACK_CHAN_BUSY:
2506 	case -NFS4ERR_DELAY:
2507 		set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2508 		status = 0;
2509 		ssleep(1);
2510 		goto out;
2511 	default:
2512 		status = nfs4_recovery_handle_error(clp, status);
2513 		goto out;
2514 	}
2515 
2516 	memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2517 	status = nfs4_proc_create_session(clp, cred);
2518 	if (status) {
2519 		dprintk("%s: session reset failed with status %d for server %s!\n",
2520 			__func__, status, clp->cl_hostname);
2521 		status = nfs4_handle_reclaim_lease_error(clp, status);
2522 		goto out;
2523 	}
2524 	nfs41_finish_session_reset(clp);
2525 	dprintk("%s: session reset was successful for server %s!\n",
2526 			__func__, clp->cl_hostname);
2527 out:
2528 	put_cred(cred);
2529 	return status;
2530 }
2531 
nfs4_bind_conn_to_session(struct nfs_client * clp)2532 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2533 {
2534 	const struct cred *cred;
2535 	int ret;
2536 
2537 	if (!nfs4_has_session(clp))
2538 		return 0;
2539 	ret = nfs4_begin_drain_session(clp);
2540 	if (ret != 0)
2541 		return ret;
2542 	cred = nfs4_get_clid_cred(clp);
2543 	ret = nfs4_proc_bind_conn_to_session(clp, cred);
2544 	put_cred(cred);
2545 	clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2546 	switch (ret) {
2547 	case 0:
2548 		dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2549 			__func__, clp->cl_hostname);
2550 		break;
2551 	case -NFS4ERR_DELAY:
2552 		ssleep(1);
2553 		set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2554 		break;
2555 	default:
2556 		return nfs4_recovery_handle_error(clp, ret);
2557 	}
2558 	return 0;
2559 }
2560 
nfs4_layoutreturn_any_run(struct nfs_client * clp)2561 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2562 {
2563 	int iomode = 0;
2564 
2565 	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2566 		iomode += IOMODE_READ;
2567 	if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2568 		iomode += IOMODE_RW;
2569 	/* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2570 	if (iomode) {
2571 		pnfs_layout_return_unused_byclid(clp, iomode);
2572 		set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2573 	}
2574 }
2575 #else /* CONFIG_NFS_V4_1 */
nfs4_reset_session(struct nfs_client * clp)2576 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2577 
nfs4_bind_conn_to_session(struct nfs_client * clp)2578 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2579 {
2580 	return 0;
2581 }
2582 
nfs4_layoutreturn_any_run(struct nfs_client * clp)2583 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2584 {
2585 }
2586 #endif /* CONFIG_NFS_V4_1 */
2587 
nfs4_state_manager(struct nfs_client * clp)2588 static void nfs4_state_manager(struct nfs_client *clp)
2589 {
2590 	unsigned int memflags;
2591 	int status = 0;
2592 	const char *section = "", *section_sep = "";
2593 
2594 	/*
2595 	 * State recovery can deadlock if the direct reclaim code tries
2596 	 * start NFS writeback. So ensure memory allocations are all
2597 	 * GFP_NOFS.
2598 	 */
2599 	memflags = memalloc_nofs_save();
2600 
2601 	/* Ensure exclusive access to NFSv4 state */
2602 	do {
2603 		trace_nfs4_state_mgr(clp);
2604 		clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2605 		if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2606 			section = "purge state";
2607 			status = nfs4_purge_lease(clp);
2608 			if (status < 0)
2609 				goto out_error;
2610 			continue;
2611 		}
2612 
2613 		if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2614 			section = "lease expired";
2615 			/* We're going to have to re-establish a clientid */
2616 			status = nfs4_reclaim_lease(clp);
2617 			if (status < 0)
2618 				goto out_error;
2619 			continue;
2620 		}
2621 
2622 		/* Initialize or reset the session */
2623 		if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2624 			section = "reset session";
2625 			status = nfs4_reset_session(clp);
2626 			if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2627 				continue;
2628 			if (status < 0)
2629 				goto out_error;
2630 		}
2631 
2632 		/* Send BIND_CONN_TO_SESSION */
2633 		if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2634 				&clp->cl_state)) {
2635 			section = "bind conn to session";
2636 			status = nfs4_bind_conn_to_session(clp);
2637 			if (status < 0)
2638 				goto out_error;
2639 			continue;
2640 		}
2641 
2642 		if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2643 			section = "check lease";
2644 			status = nfs4_check_lease(clp);
2645 			if (status < 0)
2646 				goto out_error;
2647 			continue;
2648 		}
2649 
2650 		if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2651 			section = "migration";
2652 			status = nfs4_handle_migration(clp);
2653 			if (status < 0)
2654 				goto out_error;
2655 		}
2656 
2657 		if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2658 			section = "lease moved";
2659 			status = nfs4_handle_lease_moved(clp);
2660 			if (status < 0)
2661 				goto out_error;
2662 		}
2663 
2664 		/* First recover reboot state... */
2665 		if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2666 			section = "reclaim reboot";
2667 			status = nfs4_do_reclaim(clp,
2668 				clp->cl_mvops->reboot_recovery_ops);
2669 			if (status == -EAGAIN)
2670 				continue;
2671 			if (status < 0)
2672 				goto out_error;
2673 			nfs4_state_end_reclaim_reboot(clp);
2674 		}
2675 
2676 		/* Detect expired delegations... */
2677 		if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2678 			section = "detect expired delegations";
2679 			nfs_reap_expired_delegations(clp);
2680 			continue;
2681 		}
2682 
2683 		/* Now recover expired state... */
2684 		if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2685 			section = "reclaim nograce";
2686 			status = nfs4_do_reclaim(clp,
2687 				clp->cl_mvops->nograce_recovery_ops);
2688 			if (status == -EAGAIN)
2689 				continue;
2690 			if (status < 0)
2691 				goto out_error;
2692 			clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2693 		}
2694 
2695 		memalloc_nofs_restore(memflags);
2696 		nfs4_end_drain_session(clp);
2697 		nfs4_clear_state_manager_bit(clp);
2698 
2699 		if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2700 			if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2701 				nfs_client_return_marked_delegations(clp);
2702 				set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2703 			}
2704 			nfs4_layoutreturn_any_run(clp);
2705 			clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2706 		}
2707 
2708 		return;
2709 
2710 	} while (refcount_read(&clp->cl_count) > 1 && !signalled());
2711 	goto out_drain;
2712 
2713 out_error:
2714 	if (strlen(section))
2715 		section_sep = ": ";
2716 	trace_nfs4_state_mgr_failed(clp, section, status);
2717 	pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2718 			" with error %d\n", section_sep, section,
2719 			clp->cl_hostname, -status);
2720 	ssleep(1);
2721 out_drain:
2722 	memalloc_nofs_restore(memflags);
2723 	nfs4_end_drain_session(clp);
2724 	nfs4_clear_state_manager_bit(clp);
2725 }
2726 
nfs4_run_state_manager(void * ptr)2727 static int nfs4_run_state_manager(void *ptr)
2728 {
2729 	struct nfs_client *clp = ptr;
2730 	struct rpc_clnt *cl = clp->cl_rpcclient;
2731 
2732 	while (cl != cl->cl_parent)
2733 		cl = cl->cl_parent;
2734 
2735 	allow_signal(SIGKILL);
2736 again:
2737 	set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
2738 	nfs4_state_manager(clp);
2739 	if (atomic_read(&cl->cl_swapper)) {
2740 		wait_var_event_interruptible(&clp->cl_state,
2741 					     test_bit(NFS4CLNT_RUN_MANAGER,
2742 						      &clp->cl_state));
2743 		if (atomic_read(&cl->cl_swapper) &&
2744 		    test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state))
2745 			goto again;
2746 		/* Either no longer a swapper, or were signalled */
2747 	}
2748 	clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2749 
2750 	if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2751 	    test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2752 	    !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state))
2753 		goto again;
2754 
2755 	nfs_put_client(clp);
2756 	module_put_and_kthread_exit(0);
2757 	return 0;
2758 }
2759