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 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp);
71
72 const nfs4_stateid zero_stateid = {
73 { .data = { 0 } },
74 .type = NFS4_SPECIAL_STATEID_TYPE,
75 };
76 const nfs4_stateid invalid_stateid = {
77 {
78 /* Funky initialiser keeps older gcc versions happy */
79 .data = { 0xff, 0xff, 0xff, 0xff, 0 },
80 },
81 .type = NFS4_INVALID_STATEID_TYPE,
82 };
83
84 const nfs4_stateid current_stateid = {
85 {
86 /* Funky initialiser keeps older gcc versions happy */
87 .data = { 0x0, 0x0, 0x0, 0x1, 0 },
88 },
89 .type = NFS4_SPECIAL_STATEID_TYPE,
90 };
91
92 static DEFINE_MUTEX(nfs_clid_init_mutex);
93
nfs4_setup_state_renewal(struct nfs_client * clp)94 static int nfs4_setup_state_renewal(struct nfs_client *clp)
95 {
96 int status;
97 struct nfs_fsinfo fsinfo;
98
99 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
100 nfs4_schedule_state_renewal(clp);
101 return 0;
102 }
103
104 status = nfs4_proc_get_lease_time(clp, &fsinfo);
105 if (status == 0) {
106 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ);
107 nfs4_schedule_state_renewal(clp);
108 }
109
110 return status;
111 }
112
nfs4_init_clientid(struct nfs_client * clp,const struct cred * cred)113 int nfs4_init_clientid(struct nfs_client *clp, const struct cred *cred)
114 {
115 struct nfs4_setclientid_res clid = {
116 .clientid = clp->cl_clientid,
117 .confirm = clp->cl_confirm,
118 };
119 unsigned short port;
120 int status;
121 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
122
123 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
124 goto do_confirm;
125 port = nn->nfs_callback_tcpport;
126 if (clp->cl_addr.ss_family == AF_INET6)
127 port = nn->nfs_callback_tcpport6;
128
129 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
130 if (status != 0)
131 goto out;
132 clp->cl_clientid = clid.clientid;
133 clp->cl_confirm = clid.confirm;
134 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
135 do_confirm:
136 status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
137 if (status != 0)
138 goto out;
139 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
140 nfs4_setup_state_renewal(clp);
141 out:
142 return status;
143 }
144
145 /**
146 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
147 *
148 * @clp: nfs_client under test
149 * @result: OUT: found nfs_client, or clp
150 * @cred: credential to use for trunking test
151 *
152 * Returns zero, a negative errno, or a negative NFS4ERR status.
153 * If zero is returned, an nfs_client pointer is planted in
154 * "result".
155 *
156 * Note: The returned client may not yet be marked ready.
157 */
nfs40_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result,const struct cred * cred)158 int nfs40_discover_server_trunking(struct nfs_client *clp,
159 struct nfs_client **result,
160 const struct cred *cred)
161 {
162 struct nfs4_setclientid_res clid = {
163 .clientid = clp->cl_clientid,
164 .confirm = clp->cl_confirm,
165 };
166 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
167 unsigned short port;
168 int status;
169
170 port = nn->nfs_callback_tcpport;
171 if (clp->cl_addr.ss_family == AF_INET6)
172 port = nn->nfs_callback_tcpport6;
173
174 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
175 if (status != 0)
176 goto out;
177 clp->cl_clientid = clid.clientid;
178 clp->cl_confirm = clid.confirm;
179
180 status = nfs40_walk_client_list(clp, result, cred);
181 if (status == 0) {
182 /* Sustain the lease, even if it's empty. If the clientid4
183 * goes stale it's of no use for trunking discovery. */
184 nfs4_schedule_state_renewal(*result);
185
186 /* If the client state need to recover, do it. */
187 if (clp->cl_state)
188 nfs4_schedule_state_manager(clp);
189 }
190 out:
191 return status;
192 }
193
nfs4_get_machine_cred(struct nfs_client * clp)194 const struct cred *nfs4_get_machine_cred(struct nfs_client *clp)
195 {
196 return get_cred(rpc_machine_cred());
197 }
198
nfs4_root_machine_cred(struct nfs_client * clp)199 static void nfs4_root_machine_cred(struct nfs_client *clp)
200 {
201
202 /* Force root creds instead of machine */
203 clp->cl_principal = NULL;
204 clp->cl_rpcclient->cl_principal = NULL;
205 }
206
207 static const struct cred *
nfs4_get_renew_cred_server_locked(struct nfs_server * server)208 nfs4_get_renew_cred_server_locked(struct nfs_server *server)
209 {
210 const struct cred *cred = NULL;
211 struct nfs4_state_owner *sp;
212 struct rb_node *pos;
213
214 for (pos = rb_first(&server->state_owners);
215 pos != NULL;
216 pos = rb_next(pos)) {
217 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
218 if (list_empty(&sp->so_states))
219 continue;
220 cred = get_cred(sp->so_cred);
221 break;
222 }
223 return cred;
224 }
225
226 /**
227 * nfs4_get_renew_cred - Acquire credential for a renew operation
228 * @clp: client state handle
229 *
230 * Returns an rpc_cred with reference count bumped, or NULL.
231 * Caller must hold clp->cl_lock.
232 */
nfs4_get_renew_cred(struct nfs_client * clp)233 const struct cred *nfs4_get_renew_cred(struct nfs_client *clp)
234 {
235 const struct cred *cred = NULL;
236 struct nfs_server *server;
237
238 /* Use machine credentials if available */
239 cred = nfs4_get_machine_cred(clp);
240 if (cred != NULL)
241 goto out;
242
243 spin_lock(&clp->cl_lock);
244 rcu_read_lock();
245 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
246 cred = nfs4_get_renew_cred_server_locked(server);
247 if (cred != NULL)
248 break;
249 }
250 rcu_read_unlock();
251 spin_unlock(&clp->cl_lock);
252
253 out:
254 return cred;
255 }
256
nfs4_end_drain_slot_table(struct nfs4_slot_table * tbl)257 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
258 {
259 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
260 spin_lock(&tbl->slot_tbl_lock);
261 nfs41_wake_slot_table(tbl);
262 spin_unlock(&tbl->slot_tbl_lock);
263 }
264 }
265
nfs4_end_drain_session(struct nfs_client * clp)266 static void nfs4_end_drain_session(struct nfs_client *clp)
267 {
268 struct nfs4_session *ses = clp->cl_session;
269
270 if (clp->cl_slot_tbl) {
271 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
272 return;
273 }
274
275 if (ses != NULL) {
276 nfs4_end_drain_slot_table(&ses->bc_slot_table);
277 nfs4_end_drain_slot_table(&ses->fc_slot_table);
278 }
279 }
280
nfs4_drain_slot_tbl(struct nfs4_slot_table * tbl)281 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
282 {
283 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
284 spin_lock(&tbl->slot_tbl_lock);
285 if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
286 reinit_completion(&tbl->complete);
287 spin_unlock(&tbl->slot_tbl_lock);
288 return wait_for_completion_interruptible(&tbl->complete);
289 }
290 spin_unlock(&tbl->slot_tbl_lock);
291 return 0;
292 }
293
nfs4_begin_drain_session(struct nfs_client * clp)294 static int nfs4_begin_drain_session(struct nfs_client *clp)
295 {
296 struct nfs4_session *ses = clp->cl_session;
297 int ret;
298
299 if (clp->cl_slot_tbl)
300 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
301
302 /* back channel */
303 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
304 if (ret)
305 return ret;
306 /* fore channel */
307 return nfs4_drain_slot_tbl(&ses->fc_slot_table);
308 }
309
310 #if defined(CONFIG_NFS_V4_1)
311
nfs41_finish_session_reset(struct nfs_client * clp)312 static void nfs41_finish_session_reset(struct nfs_client *clp)
313 {
314 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
315 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
316 /* create_session negotiated new slot table */
317 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
318 nfs4_setup_state_renewal(clp);
319 }
320
nfs41_init_clientid(struct nfs_client * clp,const struct cred * cred)321 int nfs41_init_clientid(struct nfs_client *clp, const struct cred *cred)
322 {
323 int status;
324
325 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
326 goto do_confirm;
327 status = nfs4_proc_exchange_id(clp, cred);
328 if (status != 0)
329 goto out;
330 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
331 do_confirm:
332 status = nfs4_proc_create_session(clp, cred);
333 if (status != 0)
334 goto out;
335 if (!(clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R))
336 nfs4_state_start_reclaim_reboot(clp);
337 nfs41_finish_session_reset(clp);
338 nfs_mark_client_ready(clp, NFS_CS_READY);
339 out:
340 return status;
341 }
342
343 /**
344 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
345 *
346 * @clp: nfs_client under test
347 * @result: OUT: found nfs_client, or clp
348 * @cred: credential to use for trunking test
349 *
350 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
351 * If NFS4_OK is returned, an nfs_client pointer is planted in
352 * "result".
353 *
354 * Note: The returned client may not yet be marked ready.
355 */
nfs41_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result,const struct cred * cred)356 int nfs41_discover_server_trunking(struct nfs_client *clp,
357 struct nfs_client **result,
358 const struct cred *cred)
359 {
360 int status;
361
362 status = nfs4_proc_exchange_id(clp, cred);
363 if (status != NFS4_OK)
364 return status;
365
366 status = nfs41_walk_client_list(clp, result, cred);
367 if (status < 0)
368 return status;
369 if (clp != *result)
370 return 0;
371
372 /*
373 * Purge state if the client id was established in a prior
374 * instance and the client id could not have arrived on the
375 * server via Transparent State Migration.
376 */
377 if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R) {
378 if (!test_bit(NFS_CS_TSM_POSSIBLE, &clp->cl_flags))
379 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
380 else
381 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
382 }
383 nfs4_schedule_state_manager(clp);
384 status = nfs_wait_client_init_complete(clp);
385 if (status < 0)
386 nfs_put_client(clp);
387 return status;
388 }
389
390 #endif /* CONFIG_NFS_V4_1 */
391
392 /**
393 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
394 * @clp: client state handle
395 *
396 * Returns a cred with reference count bumped, or NULL.
397 */
nfs4_get_clid_cred(struct nfs_client * clp)398 const struct cred *nfs4_get_clid_cred(struct nfs_client *clp)
399 {
400 const struct cred *cred;
401
402 cred = nfs4_get_machine_cred(clp);
403 return cred;
404 }
405
406 static struct nfs4_state_owner *
nfs4_find_state_owner_locked(struct nfs_server * server,const struct cred * cred)407 nfs4_find_state_owner_locked(struct nfs_server *server, const struct cred *cred)
408 {
409 struct rb_node **p = &server->state_owners.rb_node,
410 *parent = NULL;
411 struct nfs4_state_owner *sp;
412 int cmp;
413
414 while (*p != NULL) {
415 parent = *p;
416 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
417 cmp = cred_fscmp(cred, sp->so_cred);
418
419 if (cmp < 0)
420 p = &parent->rb_left;
421 else if (cmp > 0)
422 p = &parent->rb_right;
423 else {
424 if (!list_empty(&sp->so_lru))
425 list_del_init(&sp->so_lru);
426 atomic_inc(&sp->so_count);
427 return sp;
428 }
429 }
430 return NULL;
431 }
432
433 static struct nfs4_state_owner *
nfs4_insert_state_owner_locked(struct nfs4_state_owner * new)434 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
435 {
436 struct nfs_server *server = new->so_server;
437 struct rb_node **p = &server->state_owners.rb_node,
438 *parent = NULL;
439 struct nfs4_state_owner *sp;
440 int cmp;
441
442 while (*p != NULL) {
443 parent = *p;
444 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
445 cmp = cred_fscmp(new->so_cred, sp->so_cred);
446
447 if (cmp < 0)
448 p = &parent->rb_left;
449 else if (cmp > 0)
450 p = &parent->rb_right;
451 else {
452 if (!list_empty(&sp->so_lru))
453 list_del_init(&sp->so_lru);
454 atomic_inc(&sp->so_count);
455 return sp;
456 }
457 }
458 rb_link_node(&new->so_server_node, parent, p);
459 rb_insert_color(&new->so_server_node, &server->state_owners);
460 return new;
461 }
462
463 static void
nfs4_remove_state_owner_locked(struct nfs4_state_owner * sp)464 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
465 {
466 struct nfs_server *server = sp->so_server;
467
468 if (!RB_EMPTY_NODE(&sp->so_server_node))
469 rb_erase(&sp->so_server_node, &server->state_owners);
470 }
471
472 static void
nfs4_init_seqid_counter(struct nfs_seqid_counter * sc)473 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
474 {
475 sc->create_time = ktime_get();
476 sc->flags = 0;
477 sc->counter = 0;
478 spin_lock_init(&sc->lock);
479 INIT_LIST_HEAD(&sc->list);
480 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
481 }
482
483 static void
nfs4_destroy_seqid_counter(struct nfs_seqid_counter * sc)484 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
485 {
486 rpc_destroy_wait_queue(&sc->wait);
487 }
488
489 /*
490 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
491 * create a new state_owner.
492 *
493 */
494 static struct nfs4_state_owner *
nfs4_alloc_state_owner(struct nfs_server * server,const struct cred * cred,gfp_t gfp_flags)495 nfs4_alloc_state_owner(struct nfs_server *server,
496 const struct cred *cred,
497 gfp_t gfp_flags)
498 {
499 struct nfs4_state_owner *sp;
500
501 sp = kzalloc(sizeof(*sp), gfp_flags);
502 if (!sp)
503 return NULL;
504 sp->so_seqid.owner_id = ida_simple_get(&server->openowner_id, 0, 0,
505 gfp_flags);
506 if (sp->so_seqid.owner_id < 0) {
507 kfree(sp);
508 return NULL;
509 }
510 sp->so_server = server;
511 sp->so_cred = get_cred(cred);
512 spin_lock_init(&sp->so_lock);
513 INIT_LIST_HEAD(&sp->so_states);
514 nfs4_init_seqid_counter(&sp->so_seqid);
515 atomic_set(&sp->so_count, 1);
516 INIT_LIST_HEAD(&sp->so_lru);
517 seqcount_spinlock_init(&sp->so_reclaim_seqcount, &sp->so_lock);
518 mutex_init(&sp->so_delegreturn_mutex);
519 return sp;
520 }
521
522 static void
nfs4_reset_state_owner(struct nfs4_state_owner * sp)523 nfs4_reset_state_owner(struct nfs4_state_owner *sp)
524 {
525 /* This state_owner is no longer usable, but must
526 * remain in place so that state recovery can find it
527 * and the opens associated with it.
528 * It may also be used for new 'open' request to
529 * return a delegation to the server.
530 * So update the 'create_time' so that it looks like
531 * a new state_owner. This will cause the server to
532 * request an OPEN_CONFIRM to start a new sequence.
533 */
534 sp->so_seqid.create_time = ktime_get();
535 }
536
nfs4_free_state_owner(struct nfs4_state_owner * sp)537 static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
538 {
539 nfs4_destroy_seqid_counter(&sp->so_seqid);
540 put_cred(sp->so_cred);
541 ida_simple_remove(&sp->so_server->openowner_id, sp->so_seqid.owner_id);
542 kfree(sp);
543 }
544
nfs4_gc_state_owners(struct nfs_server * server)545 static void nfs4_gc_state_owners(struct nfs_server *server)
546 {
547 struct nfs_client *clp = server->nfs_client;
548 struct nfs4_state_owner *sp, *tmp;
549 unsigned long time_min, time_max;
550 LIST_HEAD(doomed);
551
552 spin_lock(&clp->cl_lock);
553 time_max = jiffies;
554 time_min = (long)time_max - (long)clp->cl_lease_time;
555 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
556 /* NB: LRU is sorted so that oldest is at the head */
557 if (time_in_range(sp->so_expires, time_min, time_max))
558 break;
559 list_move(&sp->so_lru, &doomed);
560 nfs4_remove_state_owner_locked(sp);
561 }
562 spin_unlock(&clp->cl_lock);
563
564 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
565 list_del(&sp->so_lru);
566 nfs4_free_state_owner(sp);
567 }
568 }
569
570 /**
571 * nfs4_get_state_owner - Look up a state owner given a credential
572 * @server: nfs_server to search
573 * @cred: RPC credential to match
574 * @gfp_flags: allocation mode
575 *
576 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
577 */
nfs4_get_state_owner(struct nfs_server * server,const struct cred * cred,gfp_t gfp_flags)578 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
579 const struct cred *cred,
580 gfp_t gfp_flags)
581 {
582 struct nfs_client *clp = server->nfs_client;
583 struct nfs4_state_owner *sp, *new;
584
585 spin_lock(&clp->cl_lock);
586 sp = nfs4_find_state_owner_locked(server, cred);
587 spin_unlock(&clp->cl_lock);
588 if (sp != NULL)
589 goto out;
590 new = nfs4_alloc_state_owner(server, cred, gfp_flags);
591 if (new == NULL)
592 goto out;
593 spin_lock(&clp->cl_lock);
594 sp = nfs4_insert_state_owner_locked(new);
595 spin_unlock(&clp->cl_lock);
596 if (sp != new)
597 nfs4_free_state_owner(new);
598 out:
599 nfs4_gc_state_owners(server);
600 return sp;
601 }
602
603 /**
604 * nfs4_put_state_owner - Release a nfs4_state_owner
605 * @sp: state owner data to release
606 *
607 * Note that we keep released state owners on an LRU
608 * list.
609 * This caches valid state owners so that they can be
610 * reused, to avoid the OPEN_CONFIRM on minor version 0.
611 * It also pins the uniquifier of dropped state owners for
612 * a while, to ensure that those state owner names are
613 * never reused.
614 */
nfs4_put_state_owner(struct nfs4_state_owner * sp)615 void nfs4_put_state_owner(struct nfs4_state_owner *sp)
616 {
617 struct nfs_server *server = sp->so_server;
618 struct nfs_client *clp = server->nfs_client;
619
620 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
621 return;
622
623 sp->so_expires = jiffies;
624 list_add_tail(&sp->so_lru, &server->state_owners_lru);
625 spin_unlock(&clp->cl_lock);
626 }
627
628 /**
629 * nfs4_purge_state_owners - Release all cached state owners
630 * @server: nfs_server with cached state owners to release
631 * @head: resulting list of state owners
632 *
633 * Called at umount time. Remaining state owners will be on
634 * the LRU with ref count of zero.
635 * Note that the state owners are not freed, but are added
636 * to the list @head, which can later be used as an argument
637 * to nfs4_free_state_owners.
638 */
nfs4_purge_state_owners(struct nfs_server * server,struct list_head * head)639 void nfs4_purge_state_owners(struct nfs_server *server, struct list_head *head)
640 {
641 struct nfs_client *clp = server->nfs_client;
642 struct nfs4_state_owner *sp, *tmp;
643
644 spin_lock(&clp->cl_lock);
645 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
646 list_move(&sp->so_lru, head);
647 nfs4_remove_state_owner_locked(sp);
648 }
649 spin_unlock(&clp->cl_lock);
650 }
651
652 /**
653 * nfs4_free_state_owners - Release all cached state owners
654 * @head: resulting list of state owners
655 *
656 * Frees a list of state owners that was generated by
657 * nfs4_purge_state_owners
658 */
nfs4_free_state_owners(struct list_head * head)659 void nfs4_free_state_owners(struct list_head *head)
660 {
661 struct nfs4_state_owner *sp, *tmp;
662
663 list_for_each_entry_safe(sp, tmp, head, so_lru) {
664 list_del(&sp->so_lru);
665 nfs4_free_state_owner(sp);
666 }
667 }
668
669 static struct nfs4_state *
nfs4_alloc_open_state(void)670 nfs4_alloc_open_state(void)
671 {
672 struct nfs4_state *state;
673
674 state = kzalloc(sizeof(*state), GFP_NOFS);
675 if (!state)
676 return NULL;
677 refcount_set(&state->count, 1);
678 INIT_LIST_HEAD(&state->lock_states);
679 spin_lock_init(&state->state_lock);
680 seqlock_init(&state->seqlock);
681 init_waitqueue_head(&state->waitq);
682 return state;
683 }
684
685 void
nfs4_state_set_mode_locked(struct nfs4_state * state,fmode_t fmode)686 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
687 {
688 if (state->state == fmode)
689 return;
690 /* NB! List reordering - see the reclaim code for why. */
691 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
692 if (fmode & FMODE_WRITE)
693 list_move(&state->open_states, &state->owner->so_states);
694 else
695 list_move_tail(&state->open_states, &state->owner->so_states);
696 }
697 state->state = fmode;
698 }
699
700 static struct nfs4_state *
__nfs4_find_state_byowner(struct inode * inode,struct nfs4_state_owner * owner)701 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
702 {
703 struct nfs_inode *nfsi = NFS_I(inode);
704 struct nfs4_state *state;
705
706 list_for_each_entry_rcu(state, &nfsi->open_states, inode_states) {
707 if (state->owner != owner)
708 continue;
709 if (!nfs4_valid_open_stateid(state))
710 continue;
711 if (refcount_inc_not_zero(&state->count))
712 return state;
713 }
714 return NULL;
715 }
716
717 static void
nfs4_free_open_state(struct nfs4_state * state)718 nfs4_free_open_state(struct nfs4_state *state)
719 {
720 kfree_rcu(state, rcu_head);
721 }
722
723 struct nfs4_state *
nfs4_get_open_state(struct inode * inode,struct nfs4_state_owner * owner)724 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
725 {
726 struct nfs4_state *state, *new;
727 struct nfs_inode *nfsi = NFS_I(inode);
728
729 rcu_read_lock();
730 state = __nfs4_find_state_byowner(inode, owner);
731 rcu_read_unlock();
732 if (state)
733 goto out;
734 new = nfs4_alloc_open_state();
735 spin_lock(&owner->so_lock);
736 spin_lock(&inode->i_lock);
737 state = __nfs4_find_state_byowner(inode, owner);
738 if (state == NULL && new != NULL) {
739 state = new;
740 state->owner = owner;
741 atomic_inc(&owner->so_count);
742 ihold(inode);
743 state->inode = inode;
744 list_add_rcu(&state->inode_states, &nfsi->open_states);
745 spin_unlock(&inode->i_lock);
746 /* Note: The reclaim code dictates that we add stateless
747 * and read-only stateids to the end of the list */
748 list_add_tail(&state->open_states, &owner->so_states);
749 spin_unlock(&owner->so_lock);
750 } else {
751 spin_unlock(&inode->i_lock);
752 spin_unlock(&owner->so_lock);
753 if (new)
754 nfs4_free_open_state(new);
755 }
756 out:
757 return state;
758 }
759
nfs4_put_open_state(struct nfs4_state * state)760 void nfs4_put_open_state(struct nfs4_state *state)
761 {
762 struct inode *inode = state->inode;
763 struct nfs4_state_owner *owner = state->owner;
764
765 if (!refcount_dec_and_lock(&state->count, &owner->so_lock))
766 return;
767 spin_lock(&inode->i_lock);
768 list_del_rcu(&state->inode_states);
769 list_del(&state->open_states);
770 spin_unlock(&inode->i_lock);
771 spin_unlock(&owner->so_lock);
772 nfs4_inode_return_delegation_on_close(inode);
773 iput(inode);
774 nfs4_free_open_state(state);
775 nfs4_put_state_owner(owner);
776 }
777
778 /*
779 * Close the current file.
780 */
__nfs4_close(struct nfs4_state * state,fmode_t fmode,gfp_t gfp_mask,int wait)781 static void __nfs4_close(struct nfs4_state *state,
782 fmode_t fmode, gfp_t gfp_mask, int wait)
783 {
784 struct nfs4_state_owner *owner = state->owner;
785 int call_close = 0;
786 fmode_t newstate;
787
788 atomic_inc(&owner->so_count);
789 /* Protect against nfs4_find_state() */
790 spin_lock(&owner->so_lock);
791 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
792 case FMODE_READ:
793 state->n_rdonly--;
794 break;
795 case FMODE_WRITE:
796 state->n_wronly--;
797 break;
798 case FMODE_READ|FMODE_WRITE:
799 state->n_rdwr--;
800 }
801 newstate = FMODE_READ|FMODE_WRITE;
802 if (state->n_rdwr == 0) {
803 if (state->n_rdonly == 0) {
804 newstate &= ~FMODE_READ;
805 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
806 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
807 }
808 if (state->n_wronly == 0) {
809 newstate &= ~FMODE_WRITE;
810 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
811 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
812 }
813 if (newstate == 0)
814 clear_bit(NFS_DELEGATED_STATE, &state->flags);
815 }
816 nfs4_state_set_mode_locked(state, newstate);
817 spin_unlock(&owner->so_lock);
818
819 if (!call_close) {
820 nfs4_put_open_state(state);
821 nfs4_put_state_owner(owner);
822 } else
823 nfs4_do_close(state, gfp_mask, wait);
824 }
825
nfs4_close_state(struct nfs4_state * state,fmode_t fmode)826 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
827 {
828 __nfs4_close(state, fmode, GFP_KERNEL, 0);
829 }
830
nfs4_close_sync(struct nfs4_state * state,fmode_t fmode)831 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
832 {
833 __nfs4_close(state, fmode, GFP_KERNEL, 1);
834 }
835
836 /*
837 * Search the state->lock_states for an existing lock_owner
838 * that is compatible with either of the given owners.
839 * If the second is non-zero, then the first refers to a Posix-lock
840 * owner (current->files) and the second refers to a flock/OFD
841 * owner (struct file*). In that case, prefer a match for the first
842 * owner.
843 * If both sorts of locks are held on the one file we cannot know
844 * which stateid was intended to be used, so a "correct" choice cannot
845 * be made. Failing that, a "consistent" choice is preferable. The
846 * consistent choice we make is to prefer the first owner, that of a
847 * Posix lock.
848 */
849 static struct nfs4_lock_state *
__nfs4_find_lock_state(struct nfs4_state * state,fl_owner_t fl_owner,fl_owner_t fl_owner2)850 __nfs4_find_lock_state(struct nfs4_state *state,
851 fl_owner_t fl_owner, fl_owner_t fl_owner2)
852 {
853 struct nfs4_lock_state *pos, *ret = NULL;
854 list_for_each_entry(pos, &state->lock_states, ls_locks) {
855 if (pos->ls_owner == fl_owner) {
856 ret = pos;
857 break;
858 }
859 if (pos->ls_owner == fl_owner2)
860 ret = pos;
861 }
862 if (ret)
863 refcount_inc(&ret->ls_count);
864 return ret;
865 }
866
867 /*
868 * Return a compatible lock_state. If no initialized lock_state structure
869 * exists, return an uninitialized one.
870 *
871 */
nfs4_alloc_lock_state(struct nfs4_state * state,fl_owner_t fl_owner)872 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
873 {
874 struct nfs4_lock_state *lsp;
875 struct nfs_server *server = state->owner->so_server;
876
877 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
878 if (lsp == NULL)
879 return NULL;
880 nfs4_init_seqid_counter(&lsp->ls_seqid);
881 refcount_set(&lsp->ls_count, 1);
882 lsp->ls_state = state;
883 lsp->ls_owner = fl_owner;
884 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
885 if (lsp->ls_seqid.owner_id < 0)
886 goto out_free;
887 INIT_LIST_HEAD(&lsp->ls_locks);
888 return lsp;
889 out_free:
890 kfree(lsp);
891 return NULL;
892 }
893
nfs4_free_lock_state(struct nfs_server * server,struct nfs4_lock_state * lsp)894 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
895 {
896 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
897 nfs4_destroy_seqid_counter(&lsp->ls_seqid);
898 kfree(lsp);
899 }
900
901 /*
902 * Return a compatible lock_state. If no initialized lock_state structure
903 * exists, return an uninitialized one.
904 *
905 */
nfs4_get_lock_state(struct nfs4_state * state,fl_owner_t owner)906 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
907 {
908 struct nfs4_lock_state *lsp, *new = NULL;
909
910 for(;;) {
911 spin_lock(&state->state_lock);
912 lsp = __nfs4_find_lock_state(state, owner, NULL);
913 if (lsp != NULL)
914 break;
915 if (new != NULL) {
916 list_add(&new->ls_locks, &state->lock_states);
917 set_bit(LK_STATE_IN_USE, &state->flags);
918 lsp = new;
919 new = NULL;
920 break;
921 }
922 spin_unlock(&state->state_lock);
923 new = nfs4_alloc_lock_state(state, owner);
924 if (new == NULL)
925 return NULL;
926 }
927 spin_unlock(&state->state_lock);
928 if (new != NULL)
929 nfs4_free_lock_state(state->owner->so_server, new);
930 return lsp;
931 }
932
933 /*
934 * Release reference to lock_state, and free it if we see that
935 * it is no longer in use
936 */
nfs4_put_lock_state(struct nfs4_lock_state * lsp)937 void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
938 {
939 struct nfs_server *server;
940 struct nfs4_state *state;
941
942 if (lsp == NULL)
943 return;
944 state = lsp->ls_state;
945 if (!refcount_dec_and_lock(&lsp->ls_count, &state->state_lock))
946 return;
947 list_del(&lsp->ls_locks);
948 if (list_empty(&state->lock_states))
949 clear_bit(LK_STATE_IN_USE, &state->flags);
950 spin_unlock(&state->state_lock);
951 server = state->owner->so_server;
952 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
953 struct nfs_client *clp = server->nfs_client;
954
955 clp->cl_mvops->free_lock_state(server, lsp);
956 } else
957 nfs4_free_lock_state(server, lsp);
958 }
959
nfs4_fl_copy_lock(struct file_lock * dst,struct file_lock * src)960 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
961 {
962 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
963
964 dst->fl_u.nfs4_fl.owner = lsp;
965 refcount_inc(&lsp->ls_count);
966 }
967
nfs4_fl_release_lock(struct file_lock * fl)968 static void nfs4_fl_release_lock(struct file_lock *fl)
969 {
970 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
971 }
972
973 static const struct file_lock_operations nfs4_fl_lock_ops = {
974 .fl_copy_lock = nfs4_fl_copy_lock,
975 .fl_release_private = nfs4_fl_release_lock,
976 };
977
nfs4_set_lock_state(struct nfs4_state * state,struct file_lock * fl)978 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
979 {
980 struct nfs4_lock_state *lsp;
981
982 if (fl->fl_ops != NULL)
983 return 0;
984 lsp = nfs4_get_lock_state(state, fl->fl_owner);
985 if (lsp == NULL)
986 return -ENOMEM;
987 fl->fl_u.nfs4_fl.owner = lsp;
988 fl->fl_ops = &nfs4_fl_lock_ops;
989 return 0;
990 }
991
nfs4_copy_lock_stateid(nfs4_stateid * dst,struct nfs4_state * state,const struct nfs_lock_context * l_ctx)992 static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
993 struct nfs4_state *state,
994 const struct nfs_lock_context *l_ctx)
995 {
996 struct nfs4_lock_state *lsp;
997 fl_owner_t fl_owner, fl_flock_owner;
998 int ret = -ENOENT;
999
1000 if (l_ctx == NULL)
1001 goto out;
1002
1003 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
1004 goto out;
1005
1006 fl_owner = l_ctx->lockowner;
1007 fl_flock_owner = l_ctx->open_context->flock_owner;
1008
1009 spin_lock(&state->state_lock);
1010 lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
1011 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
1012 ret = -EIO;
1013 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
1014 nfs4_stateid_copy(dst, &lsp->ls_stateid);
1015 ret = 0;
1016 }
1017 spin_unlock(&state->state_lock);
1018 nfs4_put_lock_state(lsp);
1019 out:
1020 return ret;
1021 }
1022
nfs4_copy_open_stateid(nfs4_stateid * dst,struct nfs4_state * state)1023 bool nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
1024 {
1025 bool ret;
1026 const nfs4_stateid *src;
1027 int seq;
1028
1029 do {
1030 ret = false;
1031 src = &zero_stateid;
1032 seq = read_seqbegin(&state->seqlock);
1033 if (test_bit(NFS_OPEN_STATE, &state->flags)) {
1034 src = &state->open_stateid;
1035 ret = true;
1036 }
1037 nfs4_stateid_copy(dst, src);
1038 } while (read_seqretry(&state->seqlock, seq));
1039 return ret;
1040 }
1041
1042 /*
1043 * Byte-range lock aware utility to initialize the stateid of read/write
1044 * requests.
1045 */
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)1046 int nfs4_select_rw_stateid(struct nfs4_state *state,
1047 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1048 nfs4_stateid *dst, const struct cred **cred)
1049 {
1050 int ret;
1051
1052 if (!nfs4_valid_open_stateid(state))
1053 return -EIO;
1054 if (cred != NULL)
1055 *cred = NULL;
1056 ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1057 if (ret == -EIO)
1058 /* A lost lock - don't even consider delegations */
1059 goto out;
1060 /* returns true if delegation stateid found and copied */
1061 if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1062 ret = 0;
1063 goto out;
1064 }
1065 if (ret != -ENOENT)
1066 /* nfs4_copy_delegation_stateid() didn't over-write
1067 * dst, so it still has the lock stateid which we now
1068 * choose to use.
1069 */
1070 goto out;
1071 ret = nfs4_copy_open_stateid(dst, state) ? 0 : -EAGAIN;
1072 out:
1073 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1074 dst->seqid = 0;
1075 return ret;
1076 }
1077
nfs_alloc_seqid(struct nfs_seqid_counter * counter,gfp_t gfp_mask)1078 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1079 {
1080 struct nfs_seqid *new;
1081
1082 new = kmalloc(sizeof(*new), gfp_mask);
1083 if (new == NULL)
1084 return ERR_PTR(-ENOMEM);
1085 new->sequence = counter;
1086 INIT_LIST_HEAD(&new->list);
1087 new->task = NULL;
1088 return new;
1089 }
1090
nfs_release_seqid(struct nfs_seqid * seqid)1091 void nfs_release_seqid(struct nfs_seqid *seqid)
1092 {
1093 struct nfs_seqid_counter *sequence;
1094
1095 if (seqid == NULL || list_empty(&seqid->list))
1096 return;
1097 sequence = seqid->sequence;
1098 spin_lock(&sequence->lock);
1099 list_del_init(&seqid->list);
1100 if (!list_empty(&sequence->list)) {
1101 struct nfs_seqid *next;
1102
1103 next = list_first_entry(&sequence->list,
1104 struct nfs_seqid, list);
1105 rpc_wake_up_queued_task(&sequence->wait, next->task);
1106 }
1107 spin_unlock(&sequence->lock);
1108 }
1109
nfs_free_seqid(struct nfs_seqid * seqid)1110 void nfs_free_seqid(struct nfs_seqid *seqid)
1111 {
1112 nfs_release_seqid(seqid);
1113 kfree(seqid);
1114 }
1115
1116 /*
1117 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1118 * failed with a seqid incrementing error -
1119 * see comments nfs4.h:seqid_mutating_error()
1120 */
nfs_increment_seqid(int status,struct nfs_seqid * seqid)1121 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1122 {
1123 switch (status) {
1124 case 0:
1125 break;
1126 case -NFS4ERR_BAD_SEQID:
1127 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1128 return;
1129 pr_warn_ratelimited("NFS: v4 server returned a bad"
1130 " sequence-id error on an"
1131 " unconfirmed sequence %p!\n",
1132 seqid->sequence);
1133 return;
1134 case -NFS4ERR_STALE_CLIENTID:
1135 case -NFS4ERR_STALE_STATEID:
1136 case -NFS4ERR_BAD_STATEID:
1137 case -NFS4ERR_BADXDR:
1138 case -NFS4ERR_RESOURCE:
1139 case -NFS4ERR_NOFILEHANDLE:
1140 case -NFS4ERR_MOVED:
1141 /* Non-seqid mutating errors */
1142 return;
1143 }
1144 /*
1145 * Note: no locking needed as we are guaranteed to be first
1146 * on the sequence list
1147 */
1148 seqid->sequence->counter++;
1149 }
1150
nfs_increment_open_seqid(int status,struct nfs_seqid * seqid)1151 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1152 {
1153 struct nfs4_state_owner *sp;
1154
1155 if (seqid == NULL)
1156 return;
1157
1158 sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1159 if (status == -NFS4ERR_BAD_SEQID)
1160 nfs4_reset_state_owner(sp);
1161 if (!nfs4_has_session(sp->so_server->nfs_client))
1162 nfs_increment_seqid(status, seqid);
1163 }
1164
1165 /*
1166 * Increment the seqid if the LOCK/LOCKU succeeded, or
1167 * failed with a seqid incrementing error -
1168 * see comments nfs4.h:seqid_mutating_error()
1169 */
nfs_increment_lock_seqid(int status,struct nfs_seqid * seqid)1170 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1171 {
1172 if (seqid != NULL)
1173 nfs_increment_seqid(status, seqid);
1174 }
1175
nfs_wait_on_sequence(struct nfs_seqid * seqid,struct rpc_task * task)1176 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1177 {
1178 struct nfs_seqid_counter *sequence;
1179 int status = 0;
1180
1181 if (seqid == NULL)
1182 goto out;
1183 sequence = seqid->sequence;
1184 spin_lock(&sequence->lock);
1185 seqid->task = task;
1186 if (list_empty(&seqid->list))
1187 list_add_tail(&seqid->list, &sequence->list);
1188 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1189 goto unlock;
1190 rpc_sleep_on(&sequence->wait, task, NULL);
1191 status = -EAGAIN;
1192 unlock:
1193 spin_unlock(&sequence->lock);
1194 out:
1195 return status;
1196 }
1197
1198 static int nfs4_run_state_manager(void *);
1199
nfs4_clear_state_manager_bit(struct nfs_client * clp)1200 static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1201 {
1202 smp_mb__before_atomic();
1203 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1204 smp_mb__after_atomic();
1205 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1206 rpc_wake_up(&clp->cl_rpcwaitq);
1207 }
1208
1209 /*
1210 * Schedule the nfs_client asynchronous state management routine
1211 */
nfs4_schedule_state_manager(struct nfs_client * clp)1212 void nfs4_schedule_state_manager(struct nfs_client *clp)
1213 {
1214 struct task_struct *task;
1215 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1216 struct rpc_clnt *clnt = clp->cl_rpcclient;
1217 bool swapon = false;
1218
1219 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
1220
1221 if (atomic_read(&clnt->cl_swapper)) {
1222 swapon = !test_and_set_bit(NFS4CLNT_MANAGER_AVAILABLE,
1223 &clp->cl_state);
1224 if (!swapon) {
1225 wake_up_var(&clp->cl_state);
1226 return;
1227 }
1228 }
1229
1230 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1231 return;
1232
1233 __module_get(THIS_MODULE);
1234 refcount_inc(&clp->cl_count);
1235
1236 /* The rcu_read_lock() is not strictly necessary, as the state
1237 * manager is the only thread that ever changes the rpc_xprt
1238 * after it's initialized. At this point, we're single threaded. */
1239 rcu_read_lock();
1240 snprintf(buf, sizeof(buf), "%s-manager",
1241 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1242 rcu_read_unlock();
1243 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1244 if (IS_ERR(task)) {
1245 printk(KERN_ERR "%s: kthread_run: %ld\n",
1246 __func__, PTR_ERR(task));
1247 if (!nfs_client_init_is_complete(clp))
1248 nfs_mark_client_ready(clp, PTR_ERR(task));
1249 if (swapon)
1250 clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
1251 nfs4_clear_state_manager_bit(clp);
1252 nfs_put_client(clp);
1253 module_put(THIS_MODULE);
1254 }
1255 }
1256
1257 /*
1258 * Schedule a lease recovery attempt
1259 */
nfs4_schedule_lease_recovery(struct nfs_client * clp)1260 void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1261 {
1262 if (!clp)
1263 return;
1264 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1265 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1266 dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1267 clp->cl_hostname);
1268 nfs4_schedule_state_manager(clp);
1269 }
1270 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1271
1272 /**
1273 * nfs4_schedule_migration_recovery - trigger migration recovery
1274 *
1275 * @server: FSID that is migrating
1276 *
1277 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1278 * value is returned.
1279 */
nfs4_schedule_migration_recovery(const struct nfs_server * server)1280 int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1281 {
1282 struct nfs_client *clp = server->nfs_client;
1283
1284 if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1285 pr_err("NFS: volatile file handles not supported (server %s)\n",
1286 clp->cl_hostname);
1287 return -NFS4ERR_IO;
1288 }
1289
1290 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1291 return -NFS4ERR_IO;
1292
1293 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1294 __func__,
1295 (unsigned long long)server->fsid.major,
1296 (unsigned long long)server->fsid.minor,
1297 clp->cl_hostname);
1298
1299 set_bit(NFS_MIG_IN_TRANSITION,
1300 &((struct nfs_server *)server)->mig_status);
1301 set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1302
1303 nfs4_schedule_state_manager(clp);
1304 return 0;
1305 }
1306 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1307
1308 /**
1309 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1310 *
1311 * @clp: server to check for moved leases
1312 *
1313 */
nfs4_schedule_lease_moved_recovery(struct nfs_client * clp)1314 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1315 {
1316 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1317 __func__, clp->cl_clientid, clp->cl_hostname);
1318
1319 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1320 nfs4_schedule_state_manager(clp);
1321 }
1322 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1323
nfs4_wait_clnt_recover(struct nfs_client * clp)1324 int nfs4_wait_clnt_recover(struct nfs_client *clp)
1325 {
1326 int res;
1327
1328 might_sleep();
1329
1330 refcount_inc(&clp->cl_count);
1331 res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1332 nfs_wait_bit_killable, TASK_KILLABLE);
1333 if (res)
1334 goto out;
1335 if (clp->cl_cons_state < 0)
1336 res = clp->cl_cons_state;
1337 out:
1338 nfs_put_client(clp);
1339 return res;
1340 }
1341
nfs4_client_recover_expired_lease(struct nfs_client * clp)1342 int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1343 {
1344 unsigned int loop;
1345 int ret;
1346
1347 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1348 ret = nfs4_wait_clnt_recover(clp);
1349 if (ret != 0)
1350 break;
1351 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1352 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1353 break;
1354 nfs4_schedule_state_manager(clp);
1355 ret = -EIO;
1356 }
1357 return ret;
1358 }
1359
1360 /*
1361 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1362 * @clp: client to process
1363 *
1364 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1365 * resend of the SETCLIENTID and hence re-establish the
1366 * callback channel. Then return all existing delegations.
1367 */
nfs40_handle_cb_pathdown(struct nfs_client * clp)1368 static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1369 {
1370 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1371 nfs_expire_all_delegations(clp);
1372 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1373 clp->cl_hostname);
1374 }
1375
nfs4_schedule_path_down_recovery(struct nfs_client * clp)1376 void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1377 {
1378 nfs40_handle_cb_pathdown(clp);
1379 nfs4_schedule_state_manager(clp);
1380 }
1381
nfs4_state_mark_reclaim_reboot(struct nfs_client * clp,struct nfs4_state * state)1382 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1383 {
1384
1385 if (!nfs4_valid_open_stateid(state))
1386 return 0;
1387 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1388 /* Don't recover state that expired before the reboot */
1389 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1390 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1391 return 0;
1392 }
1393 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1394 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1395 return 1;
1396 }
1397
nfs4_state_mark_reclaim_nograce(struct nfs_client * clp,struct nfs4_state * state)1398 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1399 {
1400 if (!nfs4_valid_open_stateid(state))
1401 return 0;
1402 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1403 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1404 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1405 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1406 return 1;
1407 }
1408
nfs4_schedule_stateid_recovery(const struct nfs_server * server,struct nfs4_state * state)1409 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1410 {
1411 struct nfs_client *clp = server->nfs_client;
1412
1413 if (!nfs4_state_mark_reclaim_nograce(clp, state))
1414 return -EBADF;
1415 nfs_inode_find_delegation_state_and_recover(state->inode,
1416 &state->stateid);
1417 dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1418 clp->cl_hostname);
1419 nfs4_schedule_state_manager(clp);
1420 return 0;
1421 }
1422 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1423
1424 static struct nfs4_lock_state *
nfs_state_find_lock_state_by_stateid(struct nfs4_state * state,const nfs4_stateid * stateid)1425 nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1426 const nfs4_stateid *stateid)
1427 {
1428 struct nfs4_lock_state *pos;
1429
1430 list_for_each_entry(pos, &state->lock_states, ls_locks) {
1431 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1432 continue;
1433 if (nfs4_stateid_match_or_older(&pos->ls_stateid, stateid))
1434 return pos;
1435 }
1436 return NULL;
1437 }
1438
nfs_state_lock_state_matches_stateid(struct nfs4_state * state,const nfs4_stateid * stateid)1439 static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1440 const nfs4_stateid *stateid)
1441 {
1442 bool found = false;
1443
1444 if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1445 spin_lock(&state->state_lock);
1446 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1447 found = true;
1448 spin_unlock(&state->state_lock);
1449 }
1450 return found;
1451 }
1452
nfs_inode_find_state_and_recover(struct inode * inode,const nfs4_stateid * stateid)1453 void nfs_inode_find_state_and_recover(struct inode *inode,
1454 const nfs4_stateid *stateid)
1455 {
1456 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1457 struct nfs_inode *nfsi = NFS_I(inode);
1458 struct nfs_open_context *ctx;
1459 struct nfs4_state *state;
1460 bool found = false;
1461
1462 rcu_read_lock();
1463 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1464 state = ctx->state;
1465 if (state == NULL)
1466 continue;
1467 if (nfs4_stateid_match_or_older(&state->stateid, stateid) &&
1468 nfs4_state_mark_reclaim_nograce(clp, state)) {
1469 found = true;
1470 continue;
1471 }
1472 if (test_bit(NFS_OPEN_STATE, &state->flags) &&
1473 nfs4_stateid_match_or_older(&state->open_stateid, stateid) &&
1474 nfs4_state_mark_reclaim_nograce(clp, state)) {
1475 found = true;
1476 continue;
1477 }
1478 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1479 nfs4_state_mark_reclaim_nograce(clp, state))
1480 found = true;
1481 }
1482 rcu_read_unlock();
1483
1484 nfs_inode_find_delegation_state_and_recover(inode, stateid);
1485 if (found)
1486 nfs4_schedule_state_manager(clp);
1487 }
1488
nfs4_state_mark_open_context_bad(struct nfs4_state * state,int err)1489 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state, int err)
1490 {
1491 struct inode *inode = state->inode;
1492 struct nfs_inode *nfsi = NFS_I(inode);
1493 struct nfs_open_context *ctx;
1494
1495 rcu_read_lock();
1496 list_for_each_entry_rcu(ctx, &nfsi->open_files, list) {
1497 if (ctx->state != state)
1498 continue;
1499 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1500 pr_warn("NFSv4: state recovery failed for open file %pd2, "
1501 "error = %d\n", ctx->dentry, err);
1502 }
1503 rcu_read_unlock();
1504 }
1505
nfs4_state_mark_recovery_failed(struct nfs4_state * state,int error)1506 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1507 {
1508 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1509 nfs4_state_mark_open_context_bad(state, error);
1510 }
1511
1512
nfs4_reclaim_locks(struct nfs4_state * state,const struct nfs4_state_recovery_ops * ops)1513 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1514 {
1515 struct inode *inode = state->inode;
1516 struct nfs_inode *nfsi = NFS_I(inode);
1517 struct file_lock *fl;
1518 struct nfs4_lock_state *lsp;
1519 int status = 0;
1520 struct file_lock_context *flctx = inode->i_flctx;
1521 struct list_head *list;
1522
1523 if (flctx == NULL)
1524 return 0;
1525
1526 list = &flctx->flc_posix;
1527
1528 /* Guard against delegation returns and new lock/unlock calls */
1529 down_write(&nfsi->rwsem);
1530 spin_lock(&flctx->flc_lock);
1531 restart:
1532 list_for_each_entry(fl, list, fl_list) {
1533 if (nfs_file_open_context(fl->fl_file)->state != state)
1534 continue;
1535 spin_unlock(&flctx->flc_lock);
1536 status = ops->recover_lock(state, fl);
1537 switch (status) {
1538 case 0:
1539 break;
1540 case -ETIMEDOUT:
1541 case -ESTALE:
1542 case -NFS4ERR_ADMIN_REVOKED:
1543 case -NFS4ERR_STALE_STATEID:
1544 case -NFS4ERR_BAD_STATEID:
1545 case -NFS4ERR_EXPIRED:
1546 case -NFS4ERR_NO_GRACE:
1547 case -NFS4ERR_STALE_CLIENTID:
1548 case -NFS4ERR_BADSESSION:
1549 case -NFS4ERR_BADSLOT:
1550 case -NFS4ERR_BAD_HIGH_SLOT:
1551 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1552 goto out;
1553 default:
1554 pr_err("NFS: %s: unhandled error %d\n",
1555 __func__, status);
1556 fallthrough;
1557 case -ENOMEM:
1558 case -NFS4ERR_DENIED:
1559 case -NFS4ERR_RECLAIM_BAD:
1560 case -NFS4ERR_RECLAIM_CONFLICT:
1561 lsp = fl->fl_u.nfs4_fl.owner;
1562 if (lsp)
1563 set_bit(NFS_LOCK_LOST, &lsp->ls_flags);
1564 status = 0;
1565 }
1566 spin_lock(&flctx->flc_lock);
1567 }
1568 if (list == &flctx->flc_posix) {
1569 list = &flctx->flc_flock;
1570 goto restart;
1571 }
1572 spin_unlock(&flctx->flc_lock);
1573 out:
1574 up_write(&nfsi->rwsem);
1575 return status;
1576 }
1577
1578 #ifdef CONFIG_NFS_V4_2
nfs42_complete_copies(struct nfs4_state_owner * sp,struct nfs4_state * state)1579 static void nfs42_complete_copies(struct nfs4_state_owner *sp, struct nfs4_state *state)
1580 {
1581 struct nfs4_copy_state *copy;
1582
1583 if (!test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1584 !test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags))
1585 return;
1586
1587 spin_lock(&sp->so_server->nfs_client->cl_lock);
1588 list_for_each_entry(copy, &sp->so_server->ss_copies, copies) {
1589 if ((test_bit(NFS_CLNT_DST_SSC_COPY_STATE, &state->flags) &&
1590 !nfs4_stateid_match_other(&state->stateid,
1591 ©->parent_dst_state->stateid)))
1592 continue;
1593 copy->flags = 1;
1594 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1595 &state->flags)) {
1596 clear_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags);
1597 complete(©->completion);
1598 }
1599 }
1600 list_for_each_entry(copy, &sp->so_server->ss_copies, src_copies) {
1601 if ((test_bit(NFS_CLNT_SRC_SSC_COPY_STATE, &state->flags) &&
1602 !nfs4_stateid_match_other(&state->stateid,
1603 ©->parent_src_state->stateid)))
1604 continue;
1605 copy->flags = 1;
1606 if (test_and_clear_bit(NFS_CLNT_DST_SSC_COPY_STATE,
1607 &state->flags))
1608 complete(©->completion);
1609 }
1610 spin_unlock(&sp->so_server->nfs_client->cl_lock);
1611 }
1612 #else /* !CONFIG_NFS_V4_2 */
nfs42_complete_copies(struct nfs4_state_owner * sp,struct nfs4_state * state)1613 static inline void nfs42_complete_copies(struct nfs4_state_owner *sp,
1614 struct nfs4_state *state)
1615 {
1616 }
1617 #endif /* CONFIG_NFS_V4_2 */
1618
__nfs4_reclaim_open_state(struct nfs4_state_owner * sp,struct nfs4_state * state,const struct nfs4_state_recovery_ops * ops)1619 static int __nfs4_reclaim_open_state(struct nfs4_state_owner *sp, struct nfs4_state *state,
1620 const struct nfs4_state_recovery_ops *ops)
1621 {
1622 struct nfs4_lock_state *lock;
1623 int status;
1624
1625 status = ops->recover_open(sp, state);
1626 if (status < 0)
1627 return status;
1628
1629 status = nfs4_reclaim_locks(state, ops);
1630 if (status < 0)
1631 return status;
1632
1633 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1634 spin_lock(&state->state_lock);
1635 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1636 trace_nfs4_state_lock_reclaim(state, lock);
1637 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1638 pr_warn_ratelimited("NFS: %s: Lock reclaim failed!\n", __func__);
1639 }
1640 spin_unlock(&state->state_lock);
1641 }
1642
1643 nfs42_complete_copies(sp, state);
1644 clear_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1645 return status;
1646 }
1647
nfs4_reclaim_open_state(struct nfs4_state_owner * sp,const struct nfs4_state_recovery_ops * ops)1648 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1649 {
1650 struct nfs4_state *state;
1651 unsigned int loop = 0;
1652 int status = 0;
1653 #ifdef CONFIG_NFS_V4_2
1654 bool found_ssc_copy_state = false;
1655 #endif /* CONFIG_NFS_V4_2 */
1656
1657 /* Note: we rely on the sp->so_states list being ordered
1658 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1659 * states first.
1660 * This is needed to ensure that the server won't give us any
1661 * read delegations that we have to return if, say, we are
1662 * recovering after a network partition or a reboot from a
1663 * server that doesn't support a grace period.
1664 */
1665 spin_lock(&sp->so_lock);
1666 raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1667 restart:
1668 list_for_each_entry(state, &sp->so_states, open_states) {
1669 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1670 continue;
1671 if (!nfs4_valid_open_stateid(state))
1672 continue;
1673 if (state->state == 0)
1674 continue;
1675 #ifdef CONFIG_NFS_V4_2
1676 if (test_bit(NFS_SRV_SSC_COPY_STATE, &state->flags)) {
1677 nfs4_state_mark_recovery_failed(state, -EIO);
1678 found_ssc_copy_state = true;
1679 continue;
1680 }
1681 #endif /* CONFIG_NFS_V4_2 */
1682 refcount_inc(&state->count);
1683 spin_unlock(&sp->so_lock);
1684 status = __nfs4_reclaim_open_state(sp, state, ops);
1685
1686 switch (status) {
1687 default:
1688 if (status >= 0) {
1689 loop = 0;
1690 break;
1691 }
1692 printk(KERN_ERR "NFS: %s: unhandled error %d\n", __func__, status);
1693 fallthrough;
1694 case -ENOENT:
1695 case -ENOMEM:
1696 case -EACCES:
1697 case -EROFS:
1698 case -EIO:
1699 case -ESTALE:
1700 /* Open state on this file cannot be recovered */
1701 nfs4_state_mark_recovery_failed(state, status);
1702 break;
1703 case -EAGAIN:
1704 ssleep(1);
1705 if (loop++ < 10) {
1706 set_bit(ops->state_flag_bit, &state->flags);
1707 break;
1708 }
1709 fallthrough;
1710 case -NFS4ERR_ADMIN_REVOKED:
1711 case -NFS4ERR_STALE_STATEID:
1712 case -NFS4ERR_OLD_STATEID:
1713 case -NFS4ERR_BAD_STATEID:
1714 case -NFS4ERR_RECLAIM_BAD:
1715 case -NFS4ERR_RECLAIM_CONFLICT:
1716 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1717 break;
1718 case -NFS4ERR_EXPIRED:
1719 case -NFS4ERR_NO_GRACE:
1720 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1721 fallthrough;
1722 case -NFS4ERR_STALE_CLIENTID:
1723 case -NFS4ERR_BADSESSION:
1724 case -NFS4ERR_BADSLOT:
1725 case -NFS4ERR_BAD_HIGH_SLOT:
1726 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1727 case -ETIMEDOUT:
1728 goto out_err;
1729 }
1730 nfs4_put_open_state(state);
1731 spin_lock(&sp->so_lock);
1732 goto restart;
1733 }
1734 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1735 spin_unlock(&sp->so_lock);
1736 #ifdef CONFIG_NFS_V4_2
1737 if (found_ssc_copy_state)
1738 return -EIO;
1739 #endif /* CONFIG_NFS_V4_2 */
1740 return 0;
1741 out_err:
1742 nfs4_put_open_state(state);
1743 spin_lock(&sp->so_lock);
1744 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1745 spin_unlock(&sp->so_lock);
1746 return status;
1747 }
1748
nfs4_clear_open_state(struct nfs4_state * state)1749 static void nfs4_clear_open_state(struct nfs4_state *state)
1750 {
1751 struct nfs4_lock_state *lock;
1752
1753 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1754 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1755 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1756 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1757 spin_lock(&state->state_lock);
1758 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1759 lock->ls_seqid.flags = 0;
1760 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1761 }
1762 spin_unlock(&state->state_lock);
1763 }
1764
nfs4_reset_seqids(struct nfs_server * server,int (* mark_reclaim)(struct nfs_client * clp,struct nfs4_state * state))1765 static void nfs4_reset_seqids(struct nfs_server *server,
1766 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1767 {
1768 struct nfs_client *clp = server->nfs_client;
1769 struct nfs4_state_owner *sp;
1770 struct rb_node *pos;
1771 struct nfs4_state *state;
1772
1773 spin_lock(&clp->cl_lock);
1774 for (pos = rb_first(&server->state_owners);
1775 pos != NULL;
1776 pos = rb_next(pos)) {
1777 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1778 sp->so_seqid.flags = 0;
1779 spin_lock(&sp->so_lock);
1780 list_for_each_entry(state, &sp->so_states, open_states) {
1781 if (mark_reclaim(clp, state))
1782 nfs4_clear_open_state(state);
1783 }
1784 spin_unlock(&sp->so_lock);
1785 }
1786 spin_unlock(&clp->cl_lock);
1787 }
1788
nfs4_state_mark_reclaim_helper(struct nfs_client * clp,int (* mark_reclaim)(struct nfs_client * clp,struct nfs4_state * state))1789 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1790 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1791 {
1792 struct nfs_server *server;
1793
1794 rcu_read_lock();
1795 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1796 nfs4_reset_seqids(server, mark_reclaim);
1797 rcu_read_unlock();
1798 }
1799
nfs4_state_start_reclaim_reboot(struct nfs_client * clp)1800 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1801 {
1802 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1803 /* Mark all delegations for reclaim */
1804 nfs_delegation_mark_reclaim(clp);
1805 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1806 }
1807
nfs4_reclaim_complete(struct nfs_client * clp,const struct nfs4_state_recovery_ops * ops,const struct cred * cred)1808 static int nfs4_reclaim_complete(struct nfs_client *clp,
1809 const struct nfs4_state_recovery_ops *ops,
1810 const struct cred *cred)
1811 {
1812 /* Notify the server we're done reclaiming our state */
1813 if (ops->reclaim_complete)
1814 return ops->reclaim_complete(clp, cred);
1815 return 0;
1816 }
1817
nfs4_clear_reclaim_server(struct nfs_server * server)1818 static void nfs4_clear_reclaim_server(struct nfs_server *server)
1819 {
1820 struct nfs_client *clp = server->nfs_client;
1821 struct nfs4_state_owner *sp;
1822 struct rb_node *pos;
1823 struct nfs4_state *state;
1824
1825 spin_lock(&clp->cl_lock);
1826 for (pos = rb_first(&server->state_owners);
1827 pos != NULL;
1828 pos = rb_next(pos)) {
1829 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1830 spin_lock(&sp->so_lock);
1831 list_for_each_entry(state, &sp->so_states, open_states) {
1832 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1833 &state->flags))
1834 continue;
1835 nfs4_state_mark_reclaim_nograce(clp, state);
1836 }
1837 spin_unlock(&sp->so_lock);
1838 }
1839 spin_unlock(&clp->cl_lock);
1840 }
1841
nfs4_state_clear_reclaim_reboot(struct nfs_client * clp)1842 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1843 {
1844 struct nfs_server *server;
1845
1846 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1847 return 0;
1848
1849 rcu_read_lock();
1850 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1851 nfs4_clear_reclaim_server(server);
1852 rcu_read_unlock();
1853
1854 nfs_delegation_reap_unclaimed(clp);
1855 return 1;
1856 }
1857
nfs4_state_end_reclaim_reboot(struct nfs_client * clp)1858 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1859 {
1860 const struct nfs4_state_recovery_ops *ops;
1861 const struct cred *cred;
1862 int err;
1863
1864 if (!nfs4_state_clear_reclaim_reboot(clp))
1865 return;
1866 ops = clp->cl_mvops->reboot_recovery_ops;
1867 cred = nfs4_get_clid_cred(clp);
1868 err = nfs4_reclaim_complete(clp, ops, cred);
1869 put_cred(cred);
1870 if (err == -NFS4ERR_CONN_NOT_BOUND_TO_SESSION)
1871 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1872 }
1873
nfs4_state_start_reclaim_nograce(struct nfs_client * clp)1874 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1875 {
1876 nfs_mark_test_expired_all_delegations(clp);
1877 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1878 }
1879
nfs4_recovery_handle_error(struct nfs_client * clp,int error)1880 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1881 {
1882 switch (error) {
1883 case 0:
1884 break;
1885 case -NFS4ERR_CB_PATH_DOWN:
1886 nfs40_handle_cb_pathdown(clp);
1887 break;
1888 case -NFS4ERR_NO_GRACE:
1889 nfs4_state_end_reclaim_reboot(clp);
1890 break;
1891 case -NFS4ERR_STALE_CLIENTID:
1892 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1893 nfs4_state_start_reclaim_reboot(clp);
1894 break;
1895 case -NFS4ERR_EXPIRED:
1896 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1897 nfs4_state_start_reclaim_nograce(clp);
1898 break;
1899 case -NFS4ERR_BADSESSION:
1900 case -NFS4ERR_BADSLOT:
1901 case -NFS4ERR_BAD_HIGH_SLOT:
1902 case -NFS4ERR_DEADSESSION:
1903 case -NFS4ERR_SEQ_FALSE_RETRY:
1904 case -NFS4ERR_SEQ_MISORDERED:
1905 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1906 /* Zero session reset errors */
1907 break;
1908 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1909 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1910 break;
1911 default:
1912 dprintk("%s: failed to handle error %d for server %s\n",
1913 __func__, error, clp->cl_hostname);
1914 return error;
1915 }
1916 dprintk("%s: handled error %d for server %s\n", __func__, error,
1917 clp->cl_hostname);
1918 return 0;
1919 }
1920
nfs4_do_reclaim(struct nfs_client * clp,const struct nfs4_state_recovery_ops * ops)1921 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1922 {
1923 struct nfs4_state_owner *sp;
1924 struct nfs_server *server;
1925 struct rb_node *pos;
1926 LIST_HEAD(freeme);
1927 int status = 0;
1928
1929 restart:
1930 rcu_read_lock();
1931 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1932 nfs4_purge_state_owners(server, &freeme);
1933 spin_lock(&clp->cl_lock);
1934 for (pos = rb_first(&server->state_owners);
1935 pos != NULL;
1936 pos = rb_next(pos)) {
1937 sp = rb_entry(pos,
1938 struct nfs4_state_owner, so_server_node);
1939 if (!test_and_clear_bit(ops->owner_flag_bit,
1940 &sp->so_flags))
1941 continue;
1942 if (!atomic_inc_not_zero(&sp->so_count))
1943 continue;
1944 spin_unlock(&clp->cl_lock);
1945 rcu_read_unlock();
1946
1947 status = nfs4_reclaim_open_state(sp, ops);
1948 if (status < 0) {
1949 set_bit(ops->owner_flag_bit, &sp->so_flags);
1950 nfs4_put_state_owner(sp);
1951 status = nfs4_recovery_handle_error(clp, status);
1952 return (status != 0) ? status : -EAGAIN;
1953 }
1954
1955 nfs4_put_state_owner(sp);
1956 goto restart;
1957 }
1958 spin_unlock(&clp->cl_lock);
1959 }
1960 rcu_read_unlock();
1961 nfs4_free_state_owners(&freeme);
1962 return 0;
1963 }
1964
nfs4_check_lease(struct nfs_client * clp)1965 static int nfs4_check_lease(struct nfs_client *clp)
1966 {
1967 const struct cred *cred;
1968 const struct nfs4_state_maintenance_ops *ops =
1969 clp->cl_mvops->state_renewal_ops;
1970 int status;
1971
1972 /* Is the client already known to have an expired lease? */
1973 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1974 return 0;
1975 cred = ops->get_state_renewal_cred(clp);
1976 if (cred == NULL) {
1977 cred = nfs4_get_clid_cred(clp);
1978 status = -ENOKEY;
1979 if (cred == NULL)
1980 goto out;
1981 }
1982 status = ops->renew_lease(clp, cred);
1983 put_cred(cred);
1984 if (status == -ETIMEDOUT) {
1985 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1986 return 0;
1987 }
1988 out:
1989 return nfs4_recovery_handle_error(clp, status);
1990 }
1991
1992 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1993 * and for recoverable errors on EXCHANGE_ID for v4.1
1994 */
nfs4_handle_reclaim_lease_error(struct nfs_client * clp,int status)1995 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1996 {
1997 switch (status) {
1998 case -NFS4ERR_SEQ_MISORDERED:
1999 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
2000 return -ESERVERFAULT;
2001 /* Lease confirmation error: retry after purging the lease */
2002 ssleep(1);
2003 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2004 break;
2005 case -NFS4ERR_STALE_CLIENTID:
2006 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2007 nfs4_state_start_reclaim_reboot(clp);
2008 break;
2009 case -NFS4ERR_CLID_INUSE:
2010 pr_err("NFS: Server %s reports our clientid is in use\n",
2011 clp->cl_hostname);
2012 nfs_mark_client_ready(clp, -EPERM);
2013 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2014 return -EPERM;
2015 case -EACCES:
2016 case -NFS4ERR_DELAY:
2017 case -EAGAIN:
2018 ssleep(1);
2019 break;
2020
2021 case -NFS4ERR_MINOR_VERS_MISMATCH:
2022 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
2023 nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
2024 dprintk("%s: exit with error %d for server %s\n",
2025 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
2026 return -EPROTONOSUPPORT;
2027 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2028 * in nfs4_exchange_id */
2029 default:
2030 dprintk("%s: exit with error %d for server %s\n", __func__,
2031 status, clp->cl_hostname);
2032 return status;
2033 }
2034 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2035 dprintk("%s: handled error %d for server %s\n", __func__, status,
2036 clp->cl_hostname);
2037 return 0;
2038 }
2039
nfs4_establish_lease(struct nfs_client * clp)2040 static int nfs4_establish_lease(struct nfs_client *clp)
2041 {
2042 const struct cred *cred;
2043 const struct nfs4_state_recovery_ops *ops =
2044 clp->cl_mvops->reboot_recovery_ops;
2045 int status;
2046
2047 status = nfs4_begin_drain_session(clp);
2048 if (status != 0)
2049 return status;
2050 cred = nfs4_get_clid_cred(clp);
2051 if (cred == NULL)
2052 return -ENOENT;
2053 status = ops->establish_clid(clp, cred);
2054 put_cred(cred);
2055 if (status != 0)
2056 return status;
2057 pnfs_destroy_all_layouts(clp);
2058 return 0;
2059 }
2060
2061 /*
2062 * Returns zero or a negative errno. NFS4ERR values are converted
2063 * to local errno values.
2064 */
nfs4_reclaim_lease(struct nfs_client * clp)2065 static int nfs4_reclaim_lease(struct nfs_client *clp)
2066 {
2067 int status;
2068
2069 status = nfs4_establish_lease(clp);
2070 if (status < 0)
2071 return nfs4_handle_reclaim_lease_error(clp, status);
2072 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
2073 nfs4_state_start_reclaim_nograce(clp);
2074 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
2075 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
2076 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2077 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2078 return 0;
2079 }
2080
nfs4_purge_lease(struct nfs_client * clp)2081 static int nfs4_purge_lease(struct nfs_client *clp)
2082 {
2083 int status;
2084
2085 status = nfs4_establish_lease(clp);
2086 if (status < 0)
2087 return nfs4_handle_reclaim_lease_error(clp, status);
2088 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2089 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
2090 nfs4_state_start_reclaim_nograce(clp);
2091 return 0;
2092 }
2093
2094 /*
2095 * Try remote migration of one FSID from a source server to a
2096 * destination server. The source server provides a list of
2097 * potential destinations.
2098 *
2099 * Returns zero or a negative NFS4ERR status code.
2100 */
nfs4_try_migration(struct nfs_server * server,const struct cred * cred)2101 static int nfs4_try_migration(struct nfs_server *server, const struct cred *cred)
2102 {
2103 struct nfs_client *clp = server->nfs_client;
2104 struct nfs4_fs_locations *locations = NULL;
2105 struct inode *inode;
2106 struct page *page;
2107 int status, result;
2108
2109 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
2110 (unsigned long long)server->fsid.major,
2111 (unsigned long long)server->fsid.minor,
2112 clp->cl_hostname);
2113
2114 result = 0;
2115 page = alloc_page(GFP_KERNEL);
2116 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
2117 if (page == NULL || locations == NULL) {
2118 dprintk("<-- %s: no memory\n", __func__);
2119 goto out;
2120 }
2121 locations->fattr = nfs_alloc_fattr();
2122 if (locations->fattr == NULL) {
2123 dprintk("<-- %s: no memory\n", __func__);
2124 goto out;
2125 }
2126
2127 inode = d_inode(server->super->s_root);
2128 result = nfs4_proc_get_locations(server, NFS_FH(inode), locations,
2129 page, cred);
2130 if (result) {
2131 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
2132 __func__, result);
2133 goto out;
2134 }
2135
2136 result = -NFS4ERR_NXIO;
2137 if (!locations->nlocations)
2138 goto out;
2139
2140 if (!(locations->fattr->valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
2141 dprintk("<-- %s: No fs_locations data, migration skipped\n",
2142 __func__);
2143 goto out;
2144 }
2145
2146 status = nfs4_begin_drain_session(clp);
2147 if (status != 0) {
2148 result = status;
2149 goto out;
2150 }
2151
2152 status = nfs4_replace_transport(server, locations);
2153 if (status != 0) {
2154 dprintk("<-- %s: failed to replace transport: %d\n",
2155 __func__, status);
2156 goto out;
2157 }
2158
2159 result = 0;
2160 dprintk("<-- %s: migration succeeded\n", __func__);
2161
2162 out:
2163 if (page != NULL)
2164 __free_page(page);
2165 if (locations != NULL)
2166 kfree(locations->fattr);
2167 kfree(locations);
2168 if (result) {
2169 pr_err("NFS: migration recovery failed (server %s)\n",
2170 clp->cl_hostname);
2171 set_bit(NFS_MIG_FAILED, &server->mig_status);
2172 }
2173 return result;
2174 }
2175
2176 /*
2177 * Returns zero or a negative NFS4ERR status code.
2178 */
nfs4_handle_migration(struct nfs_client * clp)2179 static int nfs4_handle_migration(struct nfs_client *clp)
2180 {
2181 const struct nfs4_state_maintenance_ops *ops =
2182 clp->cl_mvops->state_renewal_ops;
2183 struct nfs_server *server;
2184 const struct cred *cred;
2185
2186 dprintk("%s: migration reported on \"%s\"\n", __func__,
2187 clp->cl_hostname);
2188
2189 cred = ops->get_state_renewal_cred(clp);
2190 if (cred == NULL)
2191 return -NFS4ERR_NOENT;
2192
2193 clp->cl_mig_gen++;
2194 restart:
2195 rcu_read_lock();
2196 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2197 int status;
2198
2199 if (server->mig_gen == clp->cl_mig_gen)
2200 continue;
2201 server->mig_gen = clp->cl_mig_gen;
2202
2203 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2204 &server->mig_status))
2205 continue;
2206
2207 rcu_read_unlock();
2208 status = nfs4_try_migration(server, cred);
2209 if (status < 0) {
2210 put_cred(cred);
2211 return status;
2212 }
2213 goto restart;
2214 }
2215 rcu_read_unlock();
2216 put_cred(cred);
2217 return 0;
2218 }
2219
2220 /*
2221 * Test each nfs_server on the clp's cl_superblocks list to see
2222 * if it's moved to another server. Stop when the server no longer
2223 * returns NFS4ERR_LEASE_MOVED.
2224 */
nfs4_handle_lease_moved(struct nfs_client * clp)2225 static int nfs4_handle_lease_moved(struct nfs_client *clp)
2226 {
2227 const struct nfs4_state_maintenance_ops *ops =
2228 clp->cl_mvops->state_renewal_ops;
2229 struct nfs_server *server;
2230 const struct cred *cred;
2231
2232 dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2233 clp->cl_hostname);
2234
2235 cred = ops->get_state_renewal_cred(clp);
2236 if (cred == NULL)
2237 return -NFS4ERR_NOENT;
2238
2239 clp->cl_mig_gen++;
2240 restart:
2241 rcu_read_lock();
2242 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2243 struct inode *inode;
2244 int status;
2245
2246 if (server->mig_gen == clp->cl_mig_gen)
2247 continue;
2248 server->mig_gen = clp->cl_mig_gen;
2249
2250 rcu_read_unlock();
2251
2252 inode = d_inode(server->super->s_root);
2253 status = nfs4_proc_fsid_present(inode, cred);
2254 if (status != -NFS4ERR_MOVED)
2255 goto restart; /* wasn't this one */
2256 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2257 goto restart; /* there are more */
2258 goto out;
2259 }
2260 rcu_read_unlock();
2261
2262 out:
2263 put_cred(cred);
2264 return 0;
2265 }
2266
2267 /**
2268 * nfs4_discover_server_trunking - Detect server IP address trunking
2269 *
2270 * @clp: nfs_client under test
2271 * @result: OUT: found nfs_client, or clp
2272 *
2273 * Returns zero or a negative errno. If zero is returned,
2274 * an nfs_client pointer is planted in "result".
2275 *
2276 * Note: since we are invoked in process context, and
2277 * not from inside the state manager, we cannot use
2278 * nfs4_handle_reclaim_lease_error().
2279 */
nfs4_discover_server_trunking(struct nfs_client * clp,struct nfs_client ** result)2280 int nfs4_discover_server_trunking(struct nfs_client *clp,
2281 struct nfs_client **result)
2282 {
2283 const struct nfs4_state_recovery_ops *ops =
2284 clp->cl_mvops->reboot_recovery_ops;
2285 struct rpc_clnt *clnt;
2286 const struct cred *cred;
2287 int i, status;
2288
2289 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2290
2291 clnt = clp->cl_rpcclient;
2292 i = 0;
2293
2294 mutex_lock(&nfs_clid_init_mutex);
2295 again:
2296 status = -ENOENT;
2297 cred = nfs4_get_clid_cred(clp);
2298 if (cred == NULL)
2299 goto out_unlock;
2300
2301 status = ops->detect_trunking(clp, result, cred);
2302 put_cred(cred);
2303 switch (status) {
2304 case 0:
2305 case -EINTR:
2306 case -ERESTARTSYS:
2307 break;
2308 case -ETIMEDOUT:
2309 if (clnt->cl_softrtry)
2310 break;
2311 fallthrough;
2312 case -NFS4ERR_DELAY:
2313 case -EAGAIN:
2314 ssleep(1);
2315 fallthrough;
2316 case -NFS4ERR_STALE_CLIENTID:
2317 dprintk("NFS: %s after status %d, retrying\n",
2318 __func__, status);
2319 goto again;
2320 case -EACCES:
2321 if (i++ == 0) {
2322 nfs4_root_machine_cred(clp);
2323 goto again;
2324 }
2325 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2326 break;
2327 fallthrough;
2328 case -NFS4ERR_CLID_INUSE:
2329 case -NFS4ERR_WRONGSEC:
2330 /* No point in retrying if we already used RPC_AUTH_UNIX */
2331 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2332 status = -EPERM;
2333 break;
2334 }
2335 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2336 if (IS_ERR(clnt)) {
2337 status = PTR_ERR(clnt);
2338 break;
2339 }
2340 /* Note: this is safe because we haven't yet marked the
2341 * client as ready, so we are the only user of
2342 * clp->cl_rpcclient
2343 */
2344 clnt = xchg(&clp->cl_rpcclient, clnt);
2345 rpc_shutdown_client(clnt);
2346 clnt = clp->cl_rpcclient;
2347 goto again;
2348
2349 case -NFS4ERR_MINOR_VERS_MISMATCH:
2350 status = -EPROTONOSUPPORT;
2351 break;
2352
2353 case -EKEYEXPIRED:
2354 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2355 * in nfs4_exchange_id */
2356 status = -EKEYEXPIRED;
2357 break;
2358 default:
2359 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2360 __func__, status);
2361 status = -EIO;
2362 }
2363
2364 out_unlock:
2365 mutex_unlock(&nfs_clid_init_mutex);
2366 dprintk("NFS: %s: status = %d\n", __func__, status);
2367 return status;
2368 }
2369
2370 #ifdef CONFIG_NFS_V4_1
nfs4_schedule_session_recovery(struct nfs4_session * session,int err)2371 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2372 {
2373 struct nfs_client *clp = session->clp;
2374
2375 switch (err) {
2376 default:
2377 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2378 break;
2379 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2380 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2381 }
2382 nfs4_schedule_state_manager(clp);
2383 }
2384 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2385
nfs41_notify_server(struct nfs_client * clp)2386 void nfs41_notify_server(struct nfs_client *clp)
2387 {
2388 /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2389 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2390 nfs4_schedule_state_manager(clp);
2391 }
2392
nfs4_reset_all_state(struct nfs_client * clp)2393 static void nfs4_reset_all_state(struct nfs_client *clp)
2394 {
2395 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2396 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2397 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2398 nfs4_state_start_reclaim_nograce(clp);
2399 dprintk("%s: scheduling reset of all state for server %s!\n",
2400 __func__, clp->cl_hostname);
2401 nfs4_schedule_state_manager(clp);
2402 }
2403 }
2404
nfs41_handle_server_reboot(struct nfs_client * clp)2405 static void nfs41_handle_server_reboot(struct nfs_client *clp)
2406 {
2407 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2408 nfs4_state_start_reclaim_reboot(clp);
2409 dprintk("%s: server %s rebooted!\n", __func__,
2410 clp->cl_hostname);
2411 nfs4_schedule_state_manager(clp);
2412 }
2413 }
2414
nfs41_handle_all_state_revoked(struct nfs_client * clp)2415 static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2416 {
2417 nfs4_reset_all_state(clp);
2418 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2419 }
2420
nfs41_handle_some_state_revoked(struct nfs_client * clp)2421 static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2422 {
2423 nfs4_state_start_reclaim_nograce(clp);
2424 nfs4_schedule_state_manager(clp);
2425
2426 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2427 }
2428
nfs41_handle_recallable_state_revoked(struct nfs_client * clp)2429 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2430 {
2431 /* FIXME: For now, we destroy all layouts. */
2432 pnfs_destroy_all_layouts(clp);
2433 nfs_test_expired_all_delegations(clp);
2434 dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2435 clp->cl_hostname);
2436 }
2437
nfs41_handle_backchannel_fault(struct nfs_client * clp)2438 static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2439 {
2440 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2441 nfs4_schedule_state_manager(clp);
2442
2443 dprintk("%s: server %s declared a backchannel fault\n", __func__,
2444 clp->cl_hostname);
2445 }
2446
nfs41_handle_cb_path_down(struct nfs_client * clp)2447 static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2448 {
2449 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2450 &clp->cl_state) == 0)
2451 nfs4_schedule_state_manager(clp);
2452 }
2453
nfs41_handle_sequence_flag_errors(struct nfs_client * clp,u32 flags,bool recovery)2454 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2455 bool recovery)
2456 {
2457 if (!flags)
2458 return;
2459
2460 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2461 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2462 /*
2463 * If we're called from the state manager thread, then assume we're
2464 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2465 * Those flags are expected to remain set until we're done
2466 * recovering (see RFC5661, section 18.46.3).
2467 */
2468 if (recovery)
2469 goto out_recovery;
2470
2471 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2472 nfs41_handle_server_reboot(clp);
2473 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2474 nfs41_handle_all_state_revoked(clp);
2475 if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2476 SEQ4_STATUS_ADMIN_STATE_REVOKED))
2477 nfs41_handle_some_state_revoked(clp);
2478 if (flags & SEQ4_STATUS_LEASE_MOVED)
2479 nfs4_schedule_lease_moved_recovery(clp);
2480 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2481 nfs41_handle_recallable_state_revoked(clp);
2482 out_recovery:
2483 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2484 nfs41_handle_backchannel_fault(clp);
2485 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2486 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2487 nfs41_handle_cb_path_down(clp);
2488 }
2489
nfs4_reset_session(struct nfs_client * clp)2490 static int nfs4_reset_session(struct nfs_client *clp)
2491 {
2492 const struct cred *cred;
2493 int status;
2494
2495 if (!nfs4_has_session(clp))
2496 return 0;
2497 status = nfs4_begin_drain_session(clp);
2498 if (status != 0)
2499 return status;
2500 cred = nfs4_get_clid_cred(clp);
2501 status = nfs4_proc_destroy_session(clp->cl_session, cred);
2502 switch (status) {
2503 case 0:
2504 case -NFS4ERR_BADSESSION:
2505 case -NFS4ERR_DEADSESSION:
2506 break;
2507 case -NFS4ERR_BACK_CHAN_BUSY:
2508 case -NFS4ERR_DELAY:
2509 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2510 status = 0;
2511 ssleep(1);
2512 goto out;
2513 default:
2514 status = nfs4_recovery_handle_error(clp, status);
2515 goto out;
2516 }
2517
2518 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2519 status = nfs4_proc_create_session(clp, cred);
2520 if (status) {
2521 dprintk("%s: session reset failed with status %d for server %s!\n",
2522 __func__, status, clp->cl_hostname);
2523 status = nfs4_handle_reclaim_lease_error(clp, status);
2524 goto out;
2525 }
2526 nfs41_finish_session_reset(clp);
2527 dprintk("%s: session reset was successful for server %s!\n",
2528 __func__, clp->cl_hostname);
2529 out:
2530 put_cred(cred);
2531 return status;
2532 }
2533
nfs4_bind_conn_to_session(struct nfs_client * clp)2534 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2535 {
2536 const struct cred *cred;
2537 int ret;
2538
2539 if (!nfs4_has_session(clp))
2540 return 0;
2541 ret = nfs4_begin_drain_session(clp);
2542 if (ret != 0)
2543 return ret;
2544 cred = nfs4_get_clid_cred(clp);
2545 ret = nfs4_proc_bind_conn_to_session(clp, cred);
2546 put_cred(cred);
2547 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2548 switch (ret) {
2549 case 0:
2550 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2551 __func__, clp->cl_hostname);
2552 break;
2553 case -NFS4ERR_DELAY:
2554 ssleep(1);
2555 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2556 break;
2557 default:
2558 return nfs4_recovery_handle_error(clp, ret);
2559 }
2560 return 0;
2561 }
2562
nfs4_layoutreturn_any_run(struct nfs_client * clp)2563 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2564 {
2565 int iomode = 0;
2566
2567 if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_READ, &clp->cl_state))
2568 iomode += IOMODE_READ;
2569 if (test_and_clear_bit(NFS4CLNT_RECALL_ANY_LAYOUT_RW, &clp->cl_state))
2570 iomode += IOMODE_RW;
2571 /* Note: IOMODE_READ + IOMODE_RW == IOMODE_ANY */
2572 if (iomode) {
2573 pnfs_layout_return_unused_byclid(clp, iomode);
2574 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2575 }
2576 }
2577 #else /* CONFIG_NFS_V4_1 */
nfs4_reset_session(struct nfs_client * clp)2578 static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2579
nfs4_bind_conn_to_session(struct nfs_client * clp)2580 static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2581 {
2582 return 0;
2583 }
2584
nfs4_layoutreturn_any_run(struct nfs_client * clp)2585 static void nfs4_layoutreturn_any_run(struct nfs_client *clp)
2586 {
2587 }
2588 #endif /* CONFIG_NFS_V4_1 */
2589
nfs4_state_manager(struct nfs_client * clp)2590 static void nfs4_state_manager(struct nfs_client *clp)
2591 {
2592 unsigned int memflags;
2593 int status = 0;
2594 const char *section = "", *section_sep = "";
2595
2596 /*
2597 * State recovery can deadlock if the direct reclaim code tries
2598 * start NFS writeback. So ensure memory allocations are all
2599 * GFP_NOFS.
2600 */
2601 memflags = memalloc_nofs_save();
2602
2603 /* Ensure exclusive access to NFSv4 state */
2604 do {
2605 trace_nfs4_state_mgr(clp);
2606 clear_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2607 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2608 section = "purge state";
2609 status = nfs4_purge_lease(clp);
2610 if (status < 0)
2611 goto out_error;
2612 continue;
2613 }
2614
2615 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2616 section = "lease expired";
2617 /* We're going to have to re-establish a clientid */
2618 status = nfs4_reclaim_lease(clp);
2619 if (status < 0)
2620 goto out_error;
2621 continue;
2622 }
2623
2624 /* Initialize or reset the session */
2625 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2626 section = "reset session";
2627 status = nfs4_reset_session(clp);
2628 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2629 continue;
2630 if (status < 0)
2631 goto out_error;
2632 }
2633
2634 /* Send BIND_CONN_TO_SESSION */
2635 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2636 &clp->cl_state)) {
2637 section = "bind conn to session";
2638 status = nfs4_bind_conn_to_session(clp);
2639 if (status < 0)
2640 goto out_error;
2641 continue;
2642 }
2643
2644 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2645 section = "check lease";
2646 status = nfs4_check_lease(clp);
2647 if (status < 0)
2648 goto out_error;
2649 continue;
2650 }
2651
2652 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2653 section = "migration";
2654 status = nfs4_handle_migration(clp);
2655 if (status < 0)
2656 goto out_error;
2657 }
2658
2659 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2660 section = "lease moved";
2661 status = nfs4_handle_lease_moved(clp);
2662 if (status < 0)
2663 goto out_error;
2664 }
2665
2666 /* First recover reboot state... */
2667 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2668 section = "reclaim reboot";
2669 status = nfs4_do_reclaim(clp,
2670 clp->cl_mvops->reboot_recovery_ops);
2671 if (status == -EAGAIN)
2672 continue;
2673 if (status < 0)
2674 goto out_error;
2675 nfs4_state_end_reclaim_reboot(clp);
2676 continue;
2677 }
2678
2679 /* Detect expired delegations... */
2680 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2681 section = "detect expired delegations";
2682 nfs_reap_expired_delegations(clp);
2683 continue;
2684 }
2685
2686 /* Now recover expired state... */
2687 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2688 section = "reclaim nograce";
2689 status = nfs4_do_reclaim(clp,
2690 clp->cl_mvops->nograce_recovery_ops);
2691 if (status == -EAGAIN)
2692 continue;
2693 if (status < 0)
2694 goto out_error;
2695 clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
2696 }
2697
2698 memalloc_nofs_restore(memflags);
2699 nfs4_end_drain_session(clp);
2700 nfs4_clear_state_manager_bit(clp);
2701
2702 if (test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2703 !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING,
2704 &clp->cl_state)) {
2705 memflags = memalloc_nofs_save();
2706 continue;
2707 }
2708
2709 if (!test_and_set_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state)) {
2710 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2711 nfs_client_return_marked_delegations(clp);
2712 set_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state);
2713 }
2714 nfs4_layoutreturn_any_run(clp);
2715 clear_bit(NFS4CLNT_RECALL_RUNNING, &clp->cl_state);
2716 }
2717
2718 return;
2719
2720 } while (refcount_read(&clp->cl_count) > 1 && !signalled());
2721 goto out_drain;
2722
2723 out_error:
2724 if (strlen(section))
2725 section_sep = ": ";
2726 trace_nfs4_state_mgr_failed(clp, section, status);
2727 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2728 " with error %d\n", section_sep, section,
2729 clp->cl_hostname, -status);
2730 ssleep(1);
2731 out_drain:
2732 memalloc_nofs_restore(memflags);
2733 nfs4_end_drain_session(clp);
2734 nfs4_clear_state_manager_bit(clp);
2735 }
2736
nfs4_run_state_manager(void * ptr)2737 static int nfs4_run_state_manager(void *ptr)
2738 {
2739 struct nfs_client *clp = ptr;
2740 struct rpc_clnt *cl = clp->cl_rpcclient;
2741
2742 while (cl != cl->cl_parent)
2743 cl = cl->cl_parent;
2744
2745 allow_signal(SIGKILL);
2746 again:
2747 nfs4_state_manager(clp);
2748
2749 if (test_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state) &&
2750 !test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state)) {
2751 wait_var_event_interruptible(&clp->cl_state,
2752 test_bit(NFS4CLNT_RUN_MANAGER,
2753 &clp->cl_state));
2754 if (!atomic_read(&cl->cl_swapper))
2755 clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2756 if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2757 !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2758 goto again;
2759 /* Either no longer a swapper, or were signalled */
2760 clear_bit(NFS4CLNT_MANAGER_AVAILABLE, &clp->cl_state);
2761 }
2762
2763 if (refcount_read(&clp->cl_count) > 1 && !signalled() &&
2764 test_bit(NFS4CLNT_RUN_MANAGER, &clp->cl_state) &&
2765 !test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state))
2766 goto again;
2767
2768 nfs_put_client(clp);
2769 module_put_and_exit(0);
2770 return 0;
2771 }
2772