Loading...
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/nfs_idmap.h>
46#include <linux/kthread.h>
47#include <linux/module.h>
48#include <linux/random.h>
49#include <linux/ratelimit.h>
50#include <linux/workqueue.h>
51#include <linux/bitops.h>
52
53#include "nfs4_fs.h"
54#include "callback.h"
55#include "delegation.h"
56#include "internal.h"
57#include "pnfs.h"
58
59#define OPENOWNER_POOL_SIZE 8
60
61const nfs4_stateid zero_stateid;
62
63static LIST_HEAD(nfs4_clientid_list);
64
65int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
66{
67 struct nfs4_setclientid_res clid = {
68 .clientid = clp->cl_clientid,
69 .confirm = clp->cl_confirm,
70 };
71 unsigned short port;
72 int status;
73
74 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
75 goto do_confirm;
76 port = nfs_callback_tcpport;
77 if (clp->cl_addr.ss_family == AF_INET6)
78 port = nfs_callback_tcpport6;
79
80 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
81 if (status != 0)
82 goto out;
83 clp->cl_clientid = clid.clientid;
84 clp->cl_confirm = clid.confirm;
85 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
86do_confirm:
87 status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
88 if (status != 0)
89 goto out;
90 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
91 nfs4_schedule_state_renewal(clp);
92out:
93 return status;
94}
95
96struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
97{
98 struct rpc_cred *cred = NULL;
99
100 if (clp->cl_machine_cred != NULL)
101 cred = get_rpccred(clp->cl_machine_cred);
102 return cred;
103}
104
105static void nfs4_clear_machine_cred(struct nfs_client *clp)
106{
107 struct rpc_cred *cred;
108
109 spin_lock(&clp->cl_lock);
110 cred = clp->cl_machine_cred;
111 clp->cl_machine_cred = NULL;
112 spin_unlock(&clp->cl_lock);
113 if (cred != NULL)
114 put_rpccred(cred);
115}
116
117static struct rpc_cred *
118nfs4_get_renew_cred_server_locked(struct nfs_server *server)
119{
120 struct rpc_cred *cred = NULL;
121 struct nfs4_state_owner *sp;
122 struct rb_node *pos;
123
124 for (pos = rb_first(&server->state_owners);
125 pos != NULL;
126 pos = rb_next(pos)) {
127 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
128 if (list_empty(&sp->so_states))
129 continue;
130 cred = get_rpccred(sp->so_cred);
131 break;
132 }
133 return cred;
134}
135
136/**
137 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
138 * @clp: client state handle
139 *
140 * Returns an rpc_cred with reference count bumped, or NULL.
141 * Caller must hold clp->cl_lock.
142 */
143struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
144{
145 struct rpc_cred *cred = NULL;
146 struct nfs_server *server;
147
148 rcu_read_lock();
149 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
150 cred = nfs4_get_renew_cred_server_locked(server);
151 if (cred != NULL)
152 break;
153 }
154 rcu_read_unlock();
155 return cred;
156}
157
158#if defined(CONFIG_NFS_V4_1)
159
160static int nfs41_setup_state_renewal(struct nfs_client *clp)
161{
162 int status;
163 struct nfs_fsinfo fsinfo;
164
165 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
166 nfs4_schedule_state_renewal(clp);
167 return 0;
168 }
169
170 status = nfs4_proc_get_lease_time(clp, &fsinfo);
171 if (status == 0) {
172 /* Update lease time and schedule renewal */
173 spin_lock(&clp->cl_lock);
174 clp->cl_lease_time = fsinfo.lease_time * HZ;
175 clp->cl_last_renewal = jiffies;
176 spin_unlock(&clp->cl_lock);
177
178 nfs4_schedule_state_renewal(clp);
179 }
180
181 return status;
182}
183
184/*
185 * Back channel returns NFS4ERR_DELAY for new requests when
186 * NFS4_SESSION_DRAINING is set so there is no work to be done when draining
187 * is ended.
188 */
189static void nfs4_end_drain_session(struct nfs_client *clp)
190{
191 struct nfs4_session *ses = clp->cl_session;
192 int max_slots;
193
194 if (ses == NULL)
195 return;
196 if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) {
197 spin_lock(&ses->fc_slot_table.slot_tbl_lock);
198 max_slots = ses->fc_slot_table.max_slots;
199 while (max_slots--) {
200 struct rpc_task *task;
201
202 task = rpc_wake_up_next(&ses->fc_slot_table.
203 slot_tbl_waitq);
204 if (!task)
205 break;
206 rpc_task_set_priority(task, RPC_PRIORITY_PRIVILEGED);
207 }
208 spin_unlock(&ses->fc_slot_table.slot_tbl_lock);
209 }
210}
211
212static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl)
213{
214 spin_lock(&tbl->slot_tbl_lock);
215 if (tbl->highest_used_slotid != -1) {
216 INIT_COMPLETION(tbl->complete);
217 spin_unlock(&tbl->slot_tbl_lock);
218 return wait_for_completion_interruptible(&tbl->complete);
219 }
220 spin_unlock(&tbl->slot_tbl_lock);
221 return 0;
222}
223
224static int nfs4_begin_drain_session(struct nfs_client *clp)
225{
226 struct nfs4_session *ses = clp->cl_session;
227 int ret = 0;
228
229 set_bit(NFS4_SESSION_DRAINING, &ses->session_state);
230 /* back channel */
231 ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table);
232 if (ret)
233 return ret;
234 /* fore channel */
235 return nfs4_wait_on_slot_tbl(&ses->fc_slot_table);
236}
237
238int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
239{
240 int status;
241
242 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
243 goto do_confirm;
244 nfs4_begin_drain_session(clp);
245 status = nfs4_proc_exchange_id(clp, cred);
246 if (status != 0)
247 goto out;
248 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
249do_confirm:
250 status = nfs4_proc_create_session(clp);
251 if (status != 0)
252 goto out;
253 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
254 nfs41_setup_state_renewal(clp);
255 nfs_mark_client_ready(clp, NFS_CS_READY);
256out:
257 return status;
258}
259
260struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp)
261{
262 struct rpc_cred *cred;
263
264 spin_lock(&clp->cl_lock);
265 cred = nfs4_get_machine_cred_locked(clp);
266 spin_unlock(&clp->cl_lock);
267 return cred;
268}
269
270#endif /* CONFIG_NFS_V4_1 */
271
272static struct rpc_cred *
273nfs4_get_setclientid_cred_server(struct nfs_server *server)
274{
275 struct nfs_client *clp = server->nfs_client;
276 struct rpc_cred *cred = NULL;
277 struct nfs4_state_owner *sp;
278 struct rb_node *pos;
279
280 spin_lock(&clp->cl_lock);
281 pos = rb_first(&server->state_owners);
282 if (pos != NULL) {
283 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
284 cred = get_rpccred(sp->so_cred);
285 }
286 spin_unlock(&clp->cl_lock);
287 return cred;
288}
289
290/**
291 * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation
292 * @clp: client state handle
293 *
294 * Returns an rpc_cred with reference count bumped, or NULL.
295 */
296struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp)
297{
298 struct nfs_server *server;
299 struct rpc_cred *cred;
300
301 spin_lock(&clp->cl_lock);
302 cred = nfs4_get_machine_cred_locked(clp);
303 spin_unlock(&clp->cl_lock);
304 if (cred != NULL)
305 goto out;
306
307 rcu_read_lock();
308 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
309 cred = nfs4_get_setclientid_cred_server(server);
310 if (cred != NULL)
311 break;
312 }
313 rcu_read_unlock();
314
315out:
316 return cred;
317}
318
319static void nfs_alloc_unique_id_locked(struct rb_root *root,
320 struct nfs_unique_id *new,
321 __u64 minval, int maxbits)
322{
323 struct rb_node **p, *parent;
324 struct nfs_unique_id *pos;
325 __u64 mask = ~0ULL;
326
327 if (maxbits < 64)
328 mask = (1ULL << maxbits) - 1ULL;
329
330 /* Ensure distribution is more or less flat */
331 get_random_bytes(&new->id, sizeof(new->id));
332 new->id &= mask;
333 if (new->id < minval)
334 new->id += minval;
335retry:
336 p = &root->rb_node;
337 parent = NULL;
338
339 while (*p != NULL) {
340 parent = *p;
341 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
342
343 if (new->id < pos->id)
344 p = &(*p)->rb_left;
345 else if (new->id > pos->id)
346 p = &(*p)->rb_right;
347 else
348 goto id_exists;
349 }
350 rb_link_node(&new->rb_node, parent, p);
351 rb_insert_color(&new->rb_node, root);
352 return;
353id_exists:
354 for (;;) {
355 new->id++;
356 if (new->id < minval || (new->id & mask) != new->id) {
357 new->id = minval;
358 break;
359 }
360 parent = rb_next(parent);
361 if (parent == NULL)
362 break;
363 pos = rb_entry(parent, struct nfs_unique_id, rb_node);
364 if (new->id < pos->id)
365 break;
366 }
367 goto retry;
368}
369
370static void nfs_free_unique_id(struct rb_root *root, struct nfs_unique_id *id)
371{
372 rb_erase(&id->rb_node, root);
373}
374
375static struct nfs4_state_owner *
376nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
377{
378 struct rb_node **p = &server->state_owners.rb_node,
379 *parent = NULL;
380 struct nfs4_state_owner *sp, *res = NULL;
381
382 while (*p != NULL) {
383 parent = *p;
384 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
385
386 if (server < sp->so_server) {
387 p = &parent->rb_left;
388 continue;
389 }
390 if (server > sp->so_server) {
391 p = &parent->rb_right;
392 continue;
393 }
394 if (cred < sp->so_cred)
395 p = &parent->rb_left;
396 else if (cred > sp->so_cred)
397 p = &parent->rb_right;
398 else {
399 atomic_inc(&sp->so_count);
400 res = sp;
401 break;
402 }
403 }
404 return res;
405}
406
407static struct nfs4_state_owner *
408nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
409{
410 struct nfs_server *server = new->so_server;
411 struct rb_node **p = &server->state_owners.rb_node,
412 *parent = NULL;
413 struct nfs4_state_owner *sp;
414
415 while (*p != NULL) {
416 parent = *p;
417 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
418
419 if (new->so_cred < sp->so_cred)
420 p = &parent->rb_left;
421 else if (new->so_cred > sp->so_cred)
422 p = &parent->rb_right;
423 else {
424 atomic_inc(&sp->so_count);
425 return sp;
426 }
427 }
428 nfs_alloc_unique_id_locked(&server->openowner_id,
429 &new->so_owner_id, 1, 64);
430 rb_link_node(&new->so_server_node, parent, p);
431 rb_insert_color(&new->so_server_node, &server->state_owners);
432 return new;
433}
434
435static void
436nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
437{
438 struct nfs_server *server = sp->so_server;
439
440 if (!RB_EMPTY_NODE(&sp->so_server_node))
441 rb_erase(&sp->so_server_node, &server->state_owners);
442 nfs_free_unique_id(&server->openowner_id, &sp->so_owner_id);
443}
444
445/*
446 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
447 * create a new state_owner.
448 *
449 */
450static struct nfs4_state_owner *
451nfs4_alloc_state_owner(void)
452{
453 struct nfs4_state_owner *sp;
454
455 sp = kzalloc(sizeof(*sp),GFP_NOFS);
456 if (!sp)
457 return NULL;
458 spin_lock_init(&sp->so_lock);
459 INIT_LIST_HEAD(&sp->so_states);
460 rpc_init_wait_queue(&sp->so_sequence.wait, "Seqid_waitqueue");
461 sp->so_seqid.sequence = &sp->so_sequence;
462 spin_lock_init(&sp->so_sequence.lock);
463 INIT_LIST_HEAD(&sp->so_sequence.list);
464 atomic_set(&sp->so_count, 1);
465 return sp;
466}
467
468static void
469nfs4_drop_state_owner(struct nfs4_state_owner *sp)
470{
471 if (!RB_EMPTY_NODE(&sp->so_server_node)) {
472 struct nfs_server *server = sp->so_server;
473 struct nfs_client *clp = server->nfs_client;
474
475 spin_lock(&clp->cl_lock);
476 rb_erase(&sp->so_server_node, &server->state_owners);
477 RB_CLEAR_NODE(&sp->so_server_node);
478 spin_unlock(&clp->cl_lock);
479 }
480}
481
482/**
483 * nfs4_get_state_owner - Look up a state owner given a credential
484 * @server: nfs_server to search
485 * @cred: RPC credential to match
486 *
487 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
488 */
489struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
490 struct rpc_cred *cred)
491{
492 struct nfs_client *clp = server->nfs_client;
493 struct nfs4_state_owner *sp, *new;
494
495 spin_lock(&clp->cl_lock);
496 sp = nfs4_find_state_owner_locked(server, cred);
497 spin_unlock(&clp->cl_lock);
498 if (sp != NULL)
499 return sp;
500 new = nfs4_alloc_state_owner();
501 if (new == NULL)
502 return NULL;
503 new->so_server = server;
504 new->so_cred = cred;
505 spin_lock(&clp->cl_lock);
506 sp = nfs4_insert_state_owner_locked(new);
507 spin_unlock(&clp->cl_lock);
508 if (sp == new)
509 get_rpccred(cred);
510 else {
511 rpc_destroy_wait_queue(&new->so_sequence.wait);
512 kfree(new);
513 }
514 return sp;
515}
516
517/**
518 * nfs4_put_state_owner - Release a nfs4_state_owner
519 * @sp: state owner data to release
520 *
521 */
522void nfs4_put_state_owner(struct nfs4_state_owner *sp)
523{
524 struct nfs_client *clp = sp->so_server->nfs_client;
525 struct rpc_cred *cred = sp->so_cred;
526
527 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
528 return;
529 nfs4_remove_state_owner_locked(sp);
530 spin_unlock(&clp->cl_lock);
531 rpc_destroy_wait_queue(&sp->so_sequence.wait);
532 put_rpccred(cred);
533 kfree(sp);
534}
535
536static struct nfs4_state *
537nfs4_alloc_open_state(void)
538{
539 struct nfs4_state *state;
540
541 state = kzalloc(sizeof(*state), GFP_NOFS);
542 if (!state)
543 return NULL;
544 atomic_set(&state->count, 1);
545 INIT_LIST_HEAD(&state->lock_states);
546 spin_lock_init(&state->state_lock);
547 seqlock_init(&state->seqlock);
548 return state;
549}
550
551void
552nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
553{
554 if (state->state == fmode)
555 return;
556 /* NB! List reordering - see the reclaim code for why. */
557 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
558 if (fmode & FMODE_WRITE)
559 list_move(&state->open_states, &state->owner->so_states);
560 else
561 list_move_tail(&state->open_states, &state->owner->so_states);
562 }
563 state->state = fmode;
564}
565
566static struct nfs4_state *
567__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
568{
569 struct nfs_inode *nfsi = NFS_I(inode);
570 struct nfs4_state *state;
571
572 list_for_each_entry(state, &nfsi->open_states, inode_states) {
573 if (state->owner != owner)
574 continue;
575 if (atomic_inc_not_zero(&state->count))
576 return state;
577 }
578 return NULL;
579}
580
581static void
582nfs4_free_open_state(struct nfs4_state *state)
583{
584 kfree(state);
585}
586
587struct nfs4_state *
588nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
589{
590 struct nfs4_state *state, *new;
591 struct nfs_inode *nfsi = NFS_I(inode);
592
593 spin_lock(&inode->i_lock);
594 state = __nfs4_find_state_byowner(inode, owner);
595 spin_unlock(&inode->i_lock);
596 if (state)
597 goto out;
598 new = nfs4_alloc_open_state();
599 spin_lock(&owner->so_lock);
600 spin_lock(&inode->i_lock);
601 state = __nfs4_find_state_byowner(inode, owner);
602 if (state == NULL && new != NULL) {
603 state = new;
604 state->owner = owner;
605 atomic_inc(&owner->so_count);
606 list_add(&state->inode_states, &nfsi->open_states);
607 ihold(inode);
608 state->inode = inode;
609 spin_unlock(&inode->i_lock);
610 /* Note: The reclaim code dictates that we add stateless
611 * and read-only stateids to the end of the list */
612 list_add_tail(&state->open_states, &owner->so_states);
613 spin_unlock(&owner->so_lock);
614 } else {
615 spin_unlock(&inode->i_lock);
616 spin_unlock(&owner->so_lock);
617 if (new)
618 nfs4_free_open_state(new);
619 }
620out:
621 return state;
622}
623
624void nfs4_put_open_state(struct nfs4_state *state)
625{
626 struct inode *inode = state->inode;
627 struct nfs4_state_owner *owner = state->owner;
628
629 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
630 return;
631 spin_lock(&inode->i_lock);
632 list_del(&state->inode_states);
633 list_del(&state->open_states);
634 spin_unlock(&inode->i_lock);
635 spin_unlock(&owner->so_lock);
636 iput(inode);
637 nfs4_free_open_state(state);
638 nfs4_put_state_owner(owner);
639}
640
641/*
642 * Close the current file.
643 */
644static void __nfs4_close(struct nfs4_state *state,
645 fmode_t fmode, gfp_t gfp_mask, int wait)
646{
647 struct nfs4_state_owner *owner = state->owner;
648 int call_close = 0;
649 fmode_t newstate;
650
651 atomic_inc(&owner->so_count);
652 /* Protect against nfs4_find_state() */
653 spin_lock(&owner->so_lock);
654 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
655 case FMODE_READ:
656 state->n_rdonly--;
657 break;
658 case FMODE_WRITE:
659 state->n_wronly--;
660 break;
661 case FMODE_READ|FMODE_WRITE:
662 state->n_rdwr--;
663 }
664 newstate = FMODE_READ|FMODE_WRITE;
665 if (state->n_rdwr == 0) {
666 if (state->n_rdonly == 0) {
667 newstate &= ~FMODE_READ;
668 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
669 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
670 }
671 if (state->n_wronly == 0) {
672 newstate &= ~FMODE_WRITE;
673 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
674 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
675 }
676 if (newstate == 0)
677 clear_bit(NFS_DELEGATED_STATE, &state->flags);
678 }
679 nfs4_state_set_mode_locked(state, newstate);
680 spin_unlock(&owner->so_lock);
681
682 if (!call_close) {
683 nfs4_put_open_state(state);
684 nfs4_put_state_owner(owner);
685 } else {
686 bool roc = pnfs_roc(state->inode);
687
688 nfs4_do_close(state, gfp_mask, wait, roc);
689 }
690}
691
692void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
693{
694 __nfs4_close(state, fmode, GFP_NOFS, 0);
695}
696
697void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
698{
699 __nfs4_close(state, fmode, GFP_KERNEL, 1);
700}
701
702/*
703 * Search the state->lock_states for an existing lock_owner
704 * that is compatible with current->files
705 */
706static struct nfs4_lock_state *
707__nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
708{
709 struct nfs4_lock_state *pos;
710 list_for_each_entry(pos, &state->lock_states, ls_locks) {
711 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type)
712 continue;
713 switch (pos->ls_owner.lo_type) {
714 case NFS4_POSIX_LOCK_TYPE:
715 if (pos->ls_owner.lo_u.posix_owner != fl_owner)
716 continue;
717 break;
718 case NFS4_FLOCK_LOCK_TYPE:
719 if (pos->ls_owner.lo_u.flock_owner != fl_pid)
720 continue;
721 }
722 atomic_inc(&pos->ls_count);
723 return pos;
724 }
725 return NULL;
726}
727
728/*
729 * Return a compatible lock_state. If no initialized lock_state structure
730 * exists, return an uninitialized one.
731 *
732 */
733static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type)
734{
735 struct nfs4_lock_state *lsp;
736 struct nfs_server *server = state->owner->so_server;
737 struct nfs_client *clp = server->nfs_client;
738
739 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
740 if (lsp == NULL)
741 return NULL;
742 rpc_init_wait_queue(&lsp->ls_sequence.wait, "lock_seqid_waitqueue");
743 spin_lock_init(&lsp->ls_sequence.lock);
744 INIT_LIST_HEAD(&lsp->ls_sequence.list);
745 lsp->ls_seqid.sequence = &lsp->ls_sequence;
746 atomic_set(&lsp->ls_count, 1);
747 lsp->ls_state = state;
748 lsp->ls_owner.lo_type = type;
749 switch (lsp->ls_owner.lo_type) {
750 case NFS4_FLOCK_LOCK_TYPE:
751 lsp->ls_owner.lo_u.flock_owner = fl_pid;
752 break;
753 case NFS4_POSIX_LOCK_TYPE:
754 lsp->ls_owner.lo_u.posix_owner = fl_owner;
755 break;
756 default:
757 kfree(lsp);
758 return NULL;
759 }
760 spin_lock(&clp->cl_lock);
761 nfs_alloc_unique_id_locked(&server->lockowner_id, &lsp->ls_id, 1, 64);
762 spin_unlock(&clp->cl_lock);
763 INIT_LIST_HEAD(&lsp->ls_locks);
764 return lsp;
765}
766
767static void nfs4_free_lock_state(struct nfs4_lock_state *lsp)
768{
769 struct nfs_server *server = lsp->ls_state->owner->so_server;
770 struct nfs_client *clp = server->nfs_client;
771
772 spin_lock(&clp->cl_lock);
773 nfs_free_unique_id(&server->lockowner_id, &lsp->ls_id);
774 spin_unlock(&clp->cl_lock);
775 rpc_destroy_wait_queue(&lsp->ls_sequence.wait);
776 kfree(lsp);
777}
778
779/*
780 * Return a compatible lock_state. If no initialized lock_state structure
781 * exists, return an uninitialized one.
782 *
783 */
784static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type)
785{
786 struct nfs4_lock_state *lsp, *new = NULL;
787
788 for(;;) {
789 spin_lock(&state->state_lock);
790 lsp = __nfs4_find_lock_state(state, owner, pid, type);
791 if (lsp != NULL)
792 break;
793 if (new != NULL) {
794 list_add(&new->ls_locks, &state->lock_states);
795 set_bit(LK_STATE_IN_USE, &state->flags);
796 lsp = new;
797 new = NULL;
798 break;
799 }
800 spin_unlock(&state->state_lock);
801 new = nfs4_alloc_lock_state(state, owner, pid, type);
802 if (new == NULL)
803 return NULL;
804 }
805 spin_unlock(&state->state_lock);
806 if (new != NULL)
807 nfs4_free_lock_state(new);
808 return lsp;
809}
810
811/*
812 * Release reference to lock_state, and free it if we see that
813 * it is no longer in use
814 */
815void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
816{
817 struct nfs4_state *state;
818
819 if (lsp == NULL)
820 return;
821 state = lsp->ls_state;
822 if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
823 return;
824 list_del(&lsp->ls_locks);
825 if (list_empty(&state->lock_states))
826 clear_bit(LK_STATE_IN_USE, &state->flags);
827 spin_unlock(&state->state_lock);
828 if (lsp->ls_flags & NFS_LOCK_INITIALIZED)
829 nfs4_release_lockowner(lsp);
830 nfs4_free_lock_state(lsp);
831}
832
833static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
834{
835 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
836
837 dst->fl_u.nfs4_fl.owner = lsp;
838 atomic_inc(&lsp->ls_count);
839}
840
841static void nfs4_fl_release_lock(struct file_lock *fl)
842{
843 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
844}
845
846static const struct file_lock_operations nfs4_fl_lock_ops = {
847 .fl_copy_lock = nfs4_fl_copy_lock,
848 .fl_release_private = nfs4_fl_release_lock,
849};
850
851int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
852{
853 struct nfs4_lock_state *lsp;
854
855 if (fl->fl_ops != NULL)
856 return 0;
857 if (fl->fl_flags & FL_POSIX)
858 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE);
859 else if (fl->fl_flags & FL_FLOCK)
860 lsp = nfs4_get_lock_state(state, 0, fl->fl_pid, NFS4_FLOCK_LOCK_TYPE);
861 else
862 return -EINVAL;
863 if (lsp == NULL)
864 return -ENOMEM;
865 fl->fl_u.nfs4_fl.owner = lsp;
866 fl->fl_ops = &nfs4_fl_lock_ops;
867 return 0;
868}
869
870/*
871 * Byte-range lock aware utility to initialize the stateid of read/write
872 * requests.
873 */
874void nfs4_copy_stateid(nfs4_stateid *dst, struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid)
875{
876 struct nfs4_lock_state *lsp;
877 int seq;
878
879 do {
880 seq = read_seqbegin(&state->seqlock);
881 memcpy(dst, &state->stateid, sizeof(*dst));
882 } while (read_seqretry(&state->seqlock, seq));
883 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
884 return;
885
886 spin_lock(&state->state_lock);
887 lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE);
888 if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0)
889 memcpy(dst, &lsp->ls_stateid, sizeof(*dst));
890 spin_unlock(&state->state_lock);
891 nfs4_put_lock_state(lsp);
892}
893
894struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
895{
896 struct nfs_seqid *new;
897
898 new = kmalloc(sizeof(*new), gfp_mask);
899 if (new != NULL) {
900 new->sequence = counter;
901 INIT_LIST_HEAD(&new->list);
902 }
903 return new;
904}
905
906void nfs_release_seqid(struct nfs_seqid *seqid)
907{
908 if (!list_empty(&seqid->list)) {
909 struct rpc_sequence *sequence = seqid->sequence->sequence;
910
911 spin_lock(&sequence->lock);
912 list_del_init(&seqid->list);
913 spin_unlock(&sequence->lock);
914 rpc_wake_up(&sequence->wait);
915 }
916}
917
918void nfs_free_seqid(struct nfs_seqid *seqid)
919{
920 nfs_release_seqid(seqid);
921 kfree(seqid);
922}
923
924/*
925 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
926 * failed with a seqid incrementing error -
927 * see comments nfs_fs.h:seqid_mutating_error()
928 */
929static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
930{
931 BUG_ON(list_first_entry(&seqid->sequence->sequence->list, struct nfs_seqid, list) != seqid);
932 switch (status) {
933 case 0:
934 break;
935 case -NFS4ERR_BAD_SEQID:
936 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
937 return;
938 printk(KERN_WARNING "NFS: v4 server returned a bad"
939 " sequence-id error on an"
940 " unconfirmed sequence %p!\n",
941 seqid->sequence);
942 case -NFS4ERR_STALE_CLIENTID:
943 case -NFS4ERR_STALE_STATEID:
944 case -NFS4ERR_BAD_STATEID:
945 case -NFS4ERR_BADXDR:
946 case -NFS4ERR_RESOURCE:
947 case -NFS4ERR_NOFILEHANDLE:
948 /* Non-seqid mutating errors */
949 return;
950 };
951 /*
952 * Note: no locking needed as we are guaranteed to be first
953 * on the sequence list
954 */
955 seqid->sequence->counter++;
956}
957
958void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
959{
960 struct nfs4_state_owner *sp = container_of(seqid->sequence,
961 struct nfs4_state_owner, so_seqid);
962 struct nfs_server *server = sp->so_server;
963
964 if (status == -NFS4ERR_BAD_SEQID)
965 nfs4_drop_state_owner(sp);
966 if (!nfs4_has_session(server->nfs_client))
967 nfs_increment_seqid(status, seqid);
968}
969
970/*
971 * Increment the seqid if the LOCK/LOCKU succeeded, or
972 * failed with a seqid incrementing error -
973 * see comments nfs_fs.h:seqid_mutating_error()
974 */
975void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
976{
977 nfs_increment_seqid(status, seqid);
978}
979
980int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
981{
982 struct rpc_sequence *sequence = seqid->sequence->sequence;
983 int status = 0;
984
985 spin_lock(&sequence->lock);
986 if (list_empty(&seqid->list))
987 list_add_tail(&seqid->list, &sequence->list);
988 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
989 goto unlock;
990 rpc_sleep_on(&sequence->wait, task, NULL);
991 status = -EAGAIN;
992unlock:
993 spin_unlock(&sequence->lock);
994 return status;
995}
996
997static int nfs4_run_state_manager(void *);
998
999static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1000{
1001 smp_mb__before_clear_bit();
1002 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1003 smp_mb__after_clear_bit();
1004 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1005 rpc_wake_up(&clp->cl_rpcwaitq);
1006}
1007
1008/*
1009 * Schedule the nfs_client asynchronous state management routine
1010 */
1011void nfs4_schedule_state_manager(struct nfs_client *clp)
1012{
1013 struct task_struct *task;
1014
1015 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1016 return;
1017 __module_get(THIS_MODULE);
1018 atomic_inc(&clp->cl_count);
1019 task = kthread_run(nfs4_run_state_manager, clp, "%s-manager",
1020 rpc_peeraddr2str(clp->cl_rpcclient,
1021 RPC_DISPLAY_ADDR));
1022 if (!IS_ERR(task))
1023 return;
1024 nfs4_clear_state_manager_bit(clp);
1025 nfs_put_client(clp);
1026 module_put(THIS_MODULE);
1027}
1028
1029/*
1030 * Schedule a lease recovery attempt
1031 */
1032void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1033{
1034 if (!clp)
1035 return;
1036 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1037 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1038 nfs4_schedule_state_manager(clp);
1039}
1040
1041void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1042{
1043 nfs_handle_cb_pathdown(clp);
1044 nfs4_schedule_state_manager(clp);
1045}
1046
1047static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1048{
1049
1050 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1051 /* Don't recover state that expired before the reboot */
1052 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1053 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1054 return 0;
1055 }
1056 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1057 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1058 return 1;
1059}
1060
1061static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1062{
1063 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1064 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1065 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1066 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1067 return 1;
1068}
1069
1070void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1071{
1072 struct nfs_client *clp = server->nfs_client;
1073
1074 nfs4_state_mark_reclaim_nograce(clp, state);
1075 nfs4_schedule_state_manager(clp);
1076}
1077
1078static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1079{
1080 struct inode *inode = state->inode;
1081 struct nfs_inode *nfsi = NFS_I(inode);
1082 struct file_lock *fl;
1083 int status = 0;
1084
1085 if (inode->i_flock == NULL)
1086 return 0;
1087
1088 /* Guard against delegation returns and new lock/unlock calls */
1089 down_write(&nfsi->rwsem);
1090 /* Protect inode->i_flock using the BKL */
1091 lock_flocks();
1092 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1093 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK)))
1094 continue;
1095 if (nfs_file_open_context(fl->fl_file)->state != state)
1096 continue;
1097 unlock_flocks();
1098 status = ops->recover_lock(state, fl);
1099 switch (status) {
1100 case 0:
1101 break;
1102 case -ESTALE:
1103 case -NFS4ERR_ADMIN_REVOKED:
1104 case -NFS4ERR_STALE_STATEID:
1105 case -NFS4ERR_BAD_STATEID:
1106 case -NFS4ERR_EXPIRED:
1107 case -NFS4ERR_NO_GRACE:
1108 case -NFS4ERR_STALE_CLIENTID:
1109 case -NFS4ERR_BADSESSION:
1110 case -NFS4ERR_BADSLOT:
1111 case -NFS4ERR_BAD_HIGH_SLOT:
1112 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1113 goto out;
1114 default:
1115 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1116 __func__, status);
1117 case -ENOMEM:
1118 case -NFS4ERR_DENIED:
1119 case -NFS4ERR_RECLAIM_BAD:
1120 case -NFS4ERR_RECLAIM_CONFLICT:
1121 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1122 status = 0;
1123 }
1124 lock_flocks();
1125 }
1126 unlock_flocks();
1127out:
1128 up_write(&nfsi->rwsem);
1129 return status;
1130}
1131
1132static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1133{
1134 struct nfs4_state *state;
1135 struct nfs4_lock_state *lock;
1136 int status = 0;
1137
1138 /* Note: we rely on the sp->so_states list being ordered
1139 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1140 * states first.
1141 * This is needed to ensure that the server won't give us any
1142 * read delegations that we have to return if, say, we are
1143 * recovering after a network partition or a reboot from a
1144 * server that doesn't support a grace period.
1145 */
1146restart:
1147 spin_lock(&sp->so_lock);
1148 list_for_each_entry(state, &sp->so_states, open_states) {
1149 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1150 continue;
1151 if (state->state == 0)
1152 continue;
1153 atomic_inc(&state->count);
1154 spin_unlock(&sp->so_lock);
1155 status = ops->recover_open(sp, state);
1156 if (status >= 0) {
1157 status = nfs4_reclaim_locks(state, ops);
1158 if (status >= 0) {
1159 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1160 if (!(lock->ls_flags & NFS_LOCK_INITIALIZED))
1161 printk("%s: Lock reclaim failed!\n",
1162 __func__);
1163 }
1164 nfs4_put_open_state(state);
1165 goto restart;
1166 }
1167 }
1168 switch (status) {
1169 default:
1170 printk(KERN_ERR "%s: unhandled error %d. Zeroing state\n",
1171 __func__, status);
1172 case -ENOENT:
1173 case -ENOMEM:
1174 case -ESTALE:
1175 /*
1176 * Open state on this file cannot be recovered
1177 * All we can do is revert to using the zero stateid.
1178 */
1179 memset(state->stateid.data, 0,
1180 sizeof(state->stateid.data));
1181 /* Mark the file as being 'closed' */
1182 state->state = 0;
1183 break;
1184 case -EKEYEXPIRED:
1185 /*
1186 * User RPCSEC_GSS context has expired.
1187 * We cannot recover this stateid now, so
1188 * skip it and allow recovery thread to
1189 * proceed.
1190 */
1191 break;
1192 case -NFS4ERR_ADMIN_REVOKED:
1193 case -NFS4ERR_STALE_STATEID:
1194 case -NFS4ERR_BAD_STATEID:
1195 case -NFS4ERR_RECLAIM_BAD:
1196 case -NFS4ERR_RECLAIM_CONFLICT:
1197 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1198 break;
1199 case -NFS4ERR_EXPIRED:
1200 case -NFS4ERR_NO_GRACE:
1201 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1202 case -NFS4ERR_STALE_CLIENTID:
1203 case -NFS4ERR_BADSESSION:
1204 case -NFS4ERR_BADSLOT:
1205 case -NFS4ERR_BAD_HIGH_SLOT:
1206 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1207 goto out_err;
1208 }
1209 nfs4_put_open_state(state);
1210 goto restart;
1211 }
1212 spin_unlock(&sp->so_lock);
1213 return 0;
1214out_err:
1215 nfs4_put_open_state(state);
1216 return status;
1217}
1218
1219static void nfs4_clear_open_state(struct nfs4_state *state)
1220{
1221 struct nfs4_lock_state *lock;
1222
1223 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1224 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1225 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1226 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1227 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1228 lock->ls_seqid.flags = 0;
1229 lock->ls_flags &= ~NFS_LOCK_INITIALIZED;
1230 }
1231}
1232
1233static void nfs4_reset_seqids(struct nfs_server *server,
1234 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1235{
1236 struct nfs_client *clp = server->nfs_client;
1237 struct nfs4_state_owner *sp;
1238 struct rb_node *pos;
1239 struct nfs4_state *state;
1240
1241 spin_lock(&clp->cl_lock);
1242 for (pos = rb_first(&server->state_owners);
1243 pos != NULL;
1244 pos = rb_next(pos)) {
1245 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1246 sp->so_seqid.flags = 0;
1247 spin_lock(&sp->so_lock);
1248 list_for_each_entry(state, &sp->so_states, open_states) {
1249 if (mark_reclaim(clp, state))
1250 nfs4_clear_open_state(state);
1251 }
1252 spin_unlock(&sp->so_lock);
1253 }
1254 spin_unlock(&clp->cl_lock);
1255}
1256
1257static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1258 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1259{
1260 struct nfs_server *server;
1261
1262 rcu_read_lock();
1263 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1264 nfs4_reset_seqids(server, mark_reclaim);
1265 rcu_read_unlock();
1266}
1267
1268static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1269{
1270 /* Mark all delegations for reclaim */
1271 nfs_delegation_mark_reclaim(clp);
1272 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1273}
1274
1275static void nfs4_reclaim_complete(struct nfs_client *clp,
1276 const struct nfs4_state_recovery_ops *ops)
1277{
1278 /* Notify the server we're done reclaiming our state */
1279 if (ops->reclaim_complete)
1280 (void)ops->reclaim_complete(clp);
1281}
1282
1283static void nfs4_clear_reclaim_server(struct nfs_server *server)
1284{
1285 struct nfs_client *clp = server->nfs_client;
1286 struct nfs4_state_owner *sp;
1287 struct rb_node *pos;
1288 struct nfs4_state *state;
1289
1290 spin_lock(&clp->cl_lock);
1291 for (pos = rb_first(&server->state_owners);
1292 pos != NULL;
1293 pos = rb_next(pos)) {
1294 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1295 spin_lock(&sp->so_lock);
1296 list_for_each_entry(state, &sp->so_states, open_states) {
1297 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1298 &state->flags))
1299 continue;
1300 nfs4_state_mark_reclaim_nograce(clp, state);
1301 }
1302 spin_unlock(&sp->so_lock);
1303 }
1304 spin_unlock(&clp->cl_lock);
1305}
1306
1307static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1308{
1309 struct nfs_server *server;
1310
1311 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1312 return 0;
1313
1314 rcu_read_lock();
1315 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1316 nfs4_clear_reclaim_server(server);
1317 rcu_read_unlock();
1318
1319 nfs_delegation_reap_unclaimed(clp);
1320 return 1;
1321}
1322
1323static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1324{
1325 if (!nfs4_state_clear_reclaim_reboot(clp))
1326 return;
1327 nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops);
1328}
1329
1330static void nfs_delegation_clear_all(struct nfs_client *clp)
1331{
1332 nfs_delegation_mark_reclaim(clp);
1333 nfs_delegation_reap_unclaimed(clp);
1334}
1335
1336static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1337{
1338 nfs_delegation_clear_all(clp);
1339 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1340}
1341
1342static void nfs4_warn_keyexpired(const char *s)
1343{
1344 printk_ratelimited(KERN_WARNING "Error: state manager"
1345 " encountered RPCSEC_GSS session"
1346 " expired against NFSv4 server %s.\n",
1347 s);
1348}
1349
1350static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1351{
1352 switch (error) {
1353 case -NFS4ERR_CB_PATH_DOWN:
1354 nfs_handle_cb_pathdown(clp);
1355 return 0;
1356 case -NFS4ERR_NO_GRACE:
1357 nfs4_state_end_reclaim_reboot(clp);
1358 return 0;
1359 case -NFS4ERR_STALE_CLIENTID:
1360 case -NFS4ERR_LEASE_MOVED:
1361 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1362 nfs4_state_clear_reclaim_reboot(clp);
1363 nfs4_state_start_reclaim_reboot(clp);
1364 break;
1365 case -NFS4ERR_EXPIRED:
1366 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1367 nfs4_state_start_reclaim_nograce(clp);
1368 break;
1369 case -NFS4ERR_BADSESSION:
1370 case -NFS4ERR_BADSLOT:
1371 case -NFS4ERR_BAD_HIGH_SLOT:
1372 case -NFS4ERR_DEADSESSION:
1373 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1374 case -NFS4ERR_SEQ_FALSE_RETRY:
1375 case -NFS4ERR_SEQ_MISORDERED:
1376 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1377 /* Zero session reset errors */
1378 return 0;
1379 case -EKEYEXPIRED:
1380 /* Nothing we can do */
1381 nfs4_warn_keyexpired(clp->cl_hostname);
1382 return 0;
1383 }
1384 return error;
1385}
1386
1387static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1388{
1389 struct nfs4_state_owner *sp;
1390 struct nfs_server *server;
1391 struct rb_node *pos;
1392 int status = 0;
1393
1394restart:
1395 rcu_read_lock();
1396 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1397 spin_lock(&clp->cl_lock);
1398 for (pos = rb_first(&server->state_owners);
1399 pos != NULL;
1400 pos = rb_next(pos)) {
1401 sp = rb_entry(pos,
1402 struct nfs4_state_owner, so_server_node);
1403 if (!test_and_clear_bit(ops->owner_flag_bit,
1404 &sp->so_flags))
1405 continue;
1406 atomic_inc(&sp->so_count);
1407 spin_unlock(&clp->cl_lock);
1408 rcu_read_unlock();
1409
1410 status = nfs4_reclaim_open_state(sp, ops);
1411 if (status < 0) {
1412 set_bit(ops->owner_flag_bit, &sp->so_flags);
1413 nfs4_put_state_owner(sp);
1414 return nfs4_recovery_handle_error(clp, status);
1415 }
1416
1417 nfs4_put_state_owner(sp);
1418 goto restart;
1419 }
1420 spin_unlock(&clp->cl_lock);
1421 }
1422 rcu_read_unlock();
1423 return status;
1424}
1425
1426static int nfs4_check_lease(struct nfs_client *clp)
1427{
1428 struct rpc_cred *cred;
1429 const struct nfs4_state_maintenance_ops *ops =
1430 clp->cl_mvops->state_renewal_ops;
1431 int status = -NFS4ERR_EXPIRED;
1432
1433 /* Is the client already known to have an expired lease? */
1434 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1435 return 0;
1436 spin_lock(&clp->cl_lock);
1437 cred = ops->get_state_renewal_cred_locked(clp);
1438 spin_unlock(&clp->cl_lock);
1439 if (cred == NULL) {
1440 cred = nfs4_get_setclientid_cred(clp);
1441 if (cred == NULL)
1442 goto out;
1443 }
1444 status = ops->renew_lease(clp, cred);
1445 put_rpccred(cred);
1446out:
1447 return nfs4_recovery_handle_error(clp, status);
1448}
1449
1450static int nfs4_reclaim_lease(struct nfs_client *clp)
1451{
1452 struct rpc_cred *cred;
1453 const struct nfs4_state_recovery_ops *ops =
1454 clp->cl_mvops->reboot_recovery_ops;
1455 int status = -ENOENT;
1456
1457 cred = ops->get_clid_cred(clp);
1458 if (cred != NULL) {
1459 status = ops->establish_clid(clp, cred);
1460 put_rpccred(cred);
1461 /* Handle case where the user hasn't set up machine creds */
1462 if (status == -EACCES && cred == clp->cl_machine_cred) {
1463 nfs4_clear_machine_cred(clp);
1464 status = -EAGAIN;
1465 }
1466 if (status == -NFS4ERR_MINOR_VERS_MISMATCH)
1467 status = -EPROTONOSUPPORT;
1468 }
1469 return status;
1470}
1471
1472#ifdef CONFIG_NFS_V4_1
1473void nfs4_schedule_session_recovery(struct nfs4_session *session)
1474{
1475 struct nfs_client *clp = session->clp;
1476
1477 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1478 nfs4_schedule_lease_recovery(clp);
1479}
1480EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
1481
1482void nfs41_handle_recall_slot(struct nfs_client *clp)
1483{
1484 set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1485 nfs4_schedule_state_manager(clp);
1486}
1487
1488static void nfs4_reset_all_state(struct nfs_client *clp)
1489{
1490 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1491 clp->cl_boot_time = CURRENT_TIME;
1492 nfs4_state_start_reclaim_nograce(clp);
1493 nfs4_schedule_state_manager(clp);
1494 }
1495}
1496
1497static void nfs41_handle_server_reboot(struct nfs_client *clp)
1498{
1499 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
1500 nfs4_state_start_reclaim_reboot(clp);
1501 nfs4_schedule_state_manager(clp);
1502 }
1503}
1504
1505static void nfs41_handle_state_revoked(struct nfs_client *clp)
1506{
1507 /* Temporary */
1508 nfs4_reset_all_state(clp);
1509}
1510
1511static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
1512{
1513 /* This will need to handle layouts too */
1514 nfs_expire_all_delegations(clp);
1515}
1516
1517static void nfs41_handle_cb_path_down(struct nfs_client *clp)
1518{
1519 nfs_expire_all_delegations(clp);
1520 if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0)
1521 nfs4_schedule_state_manager(clp);
1522}
1523
1524void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags)
1525{
1526 if (!flags)
1527 return;
1528 else if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
1529 nfs41_handle_server_reboot(clp);
1530 else if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED |
1531 SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
1532 SEQ4_STATUS_ADMIN_STATE_REVOKED |
1533 SEQ4_STATUS_LEASE_MOVED))
1534 nfs41_handle_state_revoked(clp);
1535 else if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
1536 nfs41_handle_recallable_state_revoked(clp);
1537 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
1538 SEQ4_STATUS_BACKCHANNEL_FAULT |
1539 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
1540 nfs41_handle_cb_path_down(clp);
1541}
1542
1543static int nfs4_reset_session(struct nfs_client *clp)
1544{
1545 int status;
1546
1547 nfs4_begin_drain_session(clp);
1548 status = nfs4_proc_destroy_session(clp->cl_session);
1549 if (status && status != -NFS4ERR_BADSESSION &&
1550 status != -NFS4ERR_DEADSESSION) {
1551 status = nfs4_recovery_handle_error(clp, status);
1552 goto out;
1553 }
1554
1555 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
1556 status = nfs4_proc_create_session(clp);
1557 if (status) {
1558 status = nfs4_recovery_handle_error(clp, status);
1559 goto out;
1560 }
1561 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1562 /* create_session negotiated new slot table */
1563 clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state);
1564
1565 /* Let the state manager reestablish state */
1566 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1567 nfs41_setup_state_renewal(clp);
1568out:
1569 return status;
1570}
1571
1572static int nfs4_recall_slot(struct nfs_client *clp)
1573{
1574 struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table;
1575 struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs;
1576 struct nfs4_slot *new, *old;
1577 int i;
1578
1579 nfs4_begin_drain_session(clp);
1580 new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot),
1581 GFP_NOFS);
1582 if (!new)
1583 return -ENOMEM;
1584
1585 spin_lock(&fc_tbl->slot_tbl_lock);
1586 for (i = 0; i < fc_tbl->target_max_slots; i++)
1587 new[i].seq_nr = fc_tbl->slots[i].seq_nr;
1588 old = fc_tbl->slots;
1589 fc_tbl->slots = new;
1590 fc_tbl->max_slots = fc_tbl->target_max_slots;
1591 fc_tbl->target_max_slots = 0;
1592 fc_attrs->max_reqs = fc_tbl->max_slots;
1593 spin_unlock(&fc_tbl->slot_tbl_lock);
1594
1595 kfree(old);
1596 nfs4_end_drain_session(clp);
1597 return 0;
1598}
1599
1600#else /* CONFIG_NFS_V4_1 */
1601static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
1602static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; }
1603static int nfs4_recall_slot(struct nfs_client *clp) { return 0; }
1604#endif /* CONFIG_NFS_V4_1 */
1605
1606/* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors
1607 * on EXCHANGE_ID for v4.1
1608 */
1609static void nfs4_set_lease_expired(struct nfs_client *clp, int status)
1610{
1611 switch (status) {
1612 case -NFS4ERR_CLID_INUSE:
1613 case -NFS4ERR_STALE_CLIENTID:
1614 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1615 break;
1616 case -NFS4ERR_DELAY:
1617 case -ETIMEDOUT:
1618 case -EAGAIN:
1619 ssleep(1);
1620 break;
1621
1622 case -EKEYEXPIRED:
1623 nfs4_warn_keyexpired(clp->cl_hostname);
1624 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1625 * in nfs4_exchange_id */
1626 default:
1627 return;
1628 }
1629 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1630}
1631
1632static void nfs4_state_manager(struct nfs_client *clp)
1633{
1634 int status = 0;
1635
1636 /* Ensure exclusive access to NFSv4 state */
1637 do {
1638 if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
1639 /* We're going to have to re-establish a clientid */
1640 status = nfs4_reclaim_lease(clp);
1641 if (status) {
1642 nfs4_set_lease_expired(clp, status);
1643 if (test_bit(NFS4CLNT_LEASE_EXPIRED,
1644 &clp->cl_state))
1645 continue;
1646 if (clp->cl_cons_state ==
1647 NFS_CS_SESSION_INITING)
1648 nfs_mark_client_ready(clp, status);
1649 goto out_error;
1650 }
1651 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1652
1653 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH,
1654 &clp->cl_state))
1655 nfs4_state_start_reclaim_nograce(clp);
1656 else
1657 set_bit(NFS4CLNT_RECLAIM_REBOOT,
1658 &clp->cl_state);
1659
1660 pnfs_destroy_all_layouts(clp);
1661 }
1662
1663 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
1664 status = nfs4_check_lease(clp);
1665 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1666 continue;
1667 if (status < 0 && status != -NFS4ERR_CB_PATH_DOWN)
1668 goto out_error;
1669 }
1670
1671 /* Initialize or reset the session */
1672 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)
1673 && nfs4_has_session(clp)) {
1674 status = nfs4_reset_session(clp);
1675 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1676 continue;
1677 if (status < 0)
1678 goto out_error;
1679 }
1680
1681 /* First recover reboot state... */
1682 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
1683 status = nfs4_do_reclaim(clp,
1684 clp->cl_mvops->reboot_recovery_ops);
1685 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1686 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state))
1687 continue;
1688 nfs4_state_end_reclaim_reboot(clp);
1689 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1690 continue;
1691 if (status < 0)
1692 goto out_error;
1693 }
1694
1695 /* Now recover expired state... */
1696 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
1697 status = nfs4_do_reclaim(clp,
1698 clp->cl_mvops->nograce_recovery_ops);
1699 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) ||
1700 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) ||
1701 test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1702 continue;
1703 if (status < 0)
1704 goto out_error;
1705 }
1706
1707 nfs4_end_drain_session(clp);
1708 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
1709 nfs_client_return_marked_delegations(clp);
1710 continue;
1711 }
1712 /* Recall session slots */
1713 if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state)
1714 && nfs4_has_session(clp)) {
1715 status = nfs4_recall_slot(clp);
1716 if (status < 0)
1717 goto out_error;
1718 continue;
1719 }
1720
1721
1722 nfs4_clear_state_manager_bit(clp);
1723 /* Did we race with an attempt to give us more work? */
1724 if (clp->cl_state == 0)
1725 break;
1726 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1727 break;
1728 } while (atomic_read(&clp->cl_count) > 1);
1729 return;
1730out_error:
1731 printk(KERN_WARNING "Error: state manager failed on NFSv4 server %s"
1732 " with error %d\n", clp->cl_hostname, -status);
1733 nfs4_end_drain_session(clp);
1734 nfs4_clear_state_manager_bit(clp);
1735}
1736
1737static int nfs4_run_state_manager(void *ptr)
1738{
1739 struct nfs_client *clp = ptr;
1740
1741 allow_signal(SIGKILL);
1742 nfs4_state_manager(clp);
1743 nfs_put_client(clp);
1744 module_put_and_exit(0);
1745 return 0;
1746}
1747
1748/*
1749 * Local variables:
1750 * c-basic-offset: 8
1751 * End:
1752 */
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
53#include <linux/sunrpc/clnt.h>
54
55#include "nfs4_fs.h"
56#include "callback.h"
57#include "delegation.h"
58#include "internal.h"
59#include "nfs4idmap.h"
60#include "nfs4session.h"
61#include "pnfs.h"
62#include "netns.h"
63
64#define NFSDBG_FACILITY NFSDBG_STATE
65
66#define OPENOWNER_POOL_SIZE 8
67
68const nfs4_stateid zero_stateid = {
69 { .data = { 0 } },
70 .type = NFS4_SPECIAL_STATEID_TYPE,
71};
72static DEFINE_MUTEX(nfs_clid_init_mutex);
73
74int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
75{
76 struct nfs4_setclientid_res clid = {
77 .clientid = clp->cl_clientid,
78 .confirm = clp->cl_confirm,
79 };
80 unsigned short port;
81 int status;
82 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
83
84 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
85 goto do_confirm;
86 port = nn->nfs_callback_tcpport;
87 if (clp->cl_addr.ss_family == AF_INET6)
88 port = nn->nfs_callback_tcpport6;
89
90 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
91 if (status != 0)
92 goto out;
93 clp->cl_clientid = clid.clientid;
94 clp->cl_confirm = clid.confirm;
95 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
96do_confirm:
97 status = nfs4_proc_setclientid_confirm(clp, &clid, cred);
98 if (status != 0)
99 goto out;
100 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
101 nfs4_schedule_state_renewal(clp);
102out:
103 return status;
104}
105
106/**
107 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0)
108 *
109 * @clp: nfs_client under test
110 * @result: OUT: found nfs_client, or clp
111 * @cred: credential to use for trunking test
112 *
113 * Returns zero, a negative errno, or a negative NFS4ERR status.
114 * If zero is returned, an nfs_client pointer is planted in
115 * "result".
116 *
117 * Note: The returned client may not yet be marked ready.
118 */
119int nfs40_discover_server_trunking(struct nfs_client *clp,
120 struct nfs_client **result,
121 struct rpc_cred *cred)
122{
123 struct nfs4_setclientid_res clid = {
124 .clientid = clp->cl_clientid,
125 .confirm = clp->cl_confirm,
126 };
127 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id);
128 unsigned short port;
129 int status;
130
131 port = nn->nfs_callback_tcpport;
132 if (clp->cl_addr.ss_family == AF_INET6)
133 port = nn->nfs_callback_tcpport6;
134
135 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid);
136 if (status != 0)
137 goto out;
138 clp->cl_clientid = clid.clientid;
139 clp->cl_confirm = clid.confirm;
140
141 status = nfs40_walk_client_list(clp, result, cred);
142 if (status == 0) {
143 /* Sustain the lease, even if it's empty. If the clientid4
144 * goes stale it's of no use for trunking discovery. */
145 nfs4_schedule_state_renewal(*result);
146 }
147out:
148 return status;
149}
150
151struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp)
152{
153 struct rpc_cred *cred = NULL;
154
155 if (clp->cl_machine_cred != NULL)
156 cred = get_rpccred(clp->cl_machine_cred);
157 return cred;
158}
159
160static void nfs4_root_machine_cred(struct nfs_client *clp)
161{
162 struct rpc_cred *cred, *new;
163
164 new = rpc_lookup_machine_cred(NULL);
165 spin_lock(&clp->cl_lock);
166 cred = clp->cl_machine_cred;
167 clp->cl_machine_cred = new;
168 spin_unlock(&clp->cl_lock);
169 if (cred != NULL)
170 put_rpccred(cred);
171}
172
173static struct rpc_cred *
174nfs4_get_renew_cred_server_locked(struct nfs_server *server)
175{
176 struct rpc_cred *cred = NULL;
177 struct nfs4_state_owner *sp;
178 struct rb_node *pos;
179
180 for (pos = rb_first(&server->state_owners);
181 pos != NULL;
182 pos = rb_next(pos)) {
183 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
184 if (list_empty(&sp->so_states))
185 continue;
186 cred = get_rpccred(sp->so_cred);
187 break;
188 }
189 return cred;
190}
191
192/**
193 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation
194 * @clp: client state handle
195 *
196 * Returns an rpc_cred with reference count bumped, or NULL.
197 * Caller must hold clp->cl_lock.
198 */
199struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp)
200{
201 struct rpc_cred *cred = NULL;
202 struct nfs_server *server;
203
204 /* Use machine credentials if available */
205 cred = nfs4_get_machine_cred_locked(clp);
206 if (cred != NULL)
207 goto out;
208
209 rcu_read_lock();
210 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
211 cred = nfs4_get_renew_cred_server_locked(server);
212 if (cred != NULL)
213 break;
214 }
215 rcu_read_unlock();
216
217out:
218 return cred;
219}
220
221static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl)
222{
223 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) {
224 spin_lock(&tbl->slot_tbl_lock);
225 nfs41_wake_slot_table(tbl);
226 spin_unlock(&tbl->slot_tbl_lock);
227 }
228}
229
230static void nfs4_end_drain_session(struct nfs_client *clp)
231{
232 struct nfs4_session *ses = clp->cl_session;
233
234 if (clp->cl_slot_tbl) {
235 nfs4_end_drain_slot_table(clp->cl_slot_tbl);
236 return;
237 }
238
239 if (ses != NULL) {
240 nfs4_end_drain_slot_table(&ses->bc_slot_table);
241 nfs4_end_drain_slot_table(&ses->fc_slot_table);
242 }
243}
244
245static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl)
246{
247 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state);
248 spin_lock(&tbl->slot_tbl_lock);
249 if (tbl->highest_used_slotid != NFS4_NO_SLOT) {
250 reinit_completion(&tbl->complete);
251 spin_unlock(&tbl->slot_tbl_lock);
252 return wait_for_completion_interruptible(&tbl->complete);
253 }
254 spin_unlock(&tbl->slot_tbl_lock);
255 return 0;
256}
257
258static int nfs4_begin_drain_session(struct nfs_client *clp)
259{
260 struct nfs4_session *ses = clp->cl_session;
261 int ret = 0;
262
263 if (clp->cl_slot_tbl)
264 return nfs4_drain_slot_tbl(clp->cl_slot_tbl);
265
266 /* back channel */
267 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table);
268 if (ret)
269 return ret;
270 /* fore channel */
271 return nfs4_drain_slot_tbl(&ses->fc_slot_table);
272}
273
274#if defined(CONFIG_NFS_V4_1)
275
276static int nfs41_setup_state_renewal(struct nfs_client *clp)
277{
278 int status;
279 struct nfs_fsinfo fsinfo;
280 unsigned long now;
281
282 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) {
283 nfs4_schedule_state_renewal(clp);
284 return 0;
285 }
286
287 now = jiffies;
288 status = nfs4_proc_get_lease_time(clp, &fsinfo);
289 if (status == 0) {
290 nfs4_set_lease_period(clp, fsinfo.lease_time * HZ, now);
291 nfs4_schedule_state_renewal(clp);
292 }
293
294 return status;
295}
296
297static void nfs41_finish_session_reset(struct nfs_client *clp)
298{
299 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
300 clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
301 /* create_session negotiated new slot table */
302 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
303 nfs41_setup_state_renewal(clp);
304}
305
306int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred)
307{
308 int status;
309
310 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state))
311 goto do_confirm;
312 status = nfs4_proc_exchange_id(clp, cred);
313 if (status != 0)
314 goto out;
315 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
316do_confirm:
317 status = nfs4_proc_create_session(clp, cred);
318 if (status != 0)
319 goto out;
320 nfs41_finish_session_reset(clp);
321 nfs_mark_client_ready(clp, NFS_CS_READY);
322out:
323 return status;
324}
325
326/**
327 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1)
328 *
329 * @clp: nfs_client under test
330 * @result: OUT: found nfs_client, or clp
331 * @cred: credential to use for trunking test
332 *
333 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status.
334 * If NFS4_OK is returned, an nfs_client pointer is planted in
335 * "result".
336 *
337 * Note: The returned client may not yet be marked ready.
338 */
339int nfs41_discover_server_trunking(struct nfs_client *clp,
340 struct nfs_client **result,
341 struct rpc_cred *cred)
342{
343 int status;
344
345 status = nfs4_proc_exchange_id(clp, cred);
346 if (status != NFS4_OK)
347 return status;
348
349 status = nfs41_walk_client_list(clp, result, cred);
350 if (status < 0)
351 return status;
352 if (clp != *result)
353 return 0;
354
355 /* Purge state if the client id was established in a prior instance */
356 if (clp->cl_exchange_flags & EXCHGID4_FLAG_CONFIRMED_R)
357 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
358 else
359 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
360 nfs4_schedule_state_manager(clp);
361 status = nfs_wait_client_init_complete(clp);
362 if (status < 0)
363 nfs_put_client(clp);
364 return status;
365}
366
367#endif /* CONFIG_NFS_V4_1 */
368
369/**
370 * nfs4_get_clid_cred - Acquire credential for a setclientid operation
371 * @clp: client state handle
372 *
373 * Returns an rpc_cred with reference count bumped, or NULL.
374 */
375struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp)
376{
377 struct rpc_cred *cred;
378
379 spin_lock(&clp->cl_lock);
380 cred = nfs4_get_machine_cred_locked(clp);
381 spin_unlock(&clp->cl_lock);
382 return cred;
383}
384
385static struct nfs4_state_owner *
386nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred)
387{
388 struct rb_node **p = &server->state_owners.rb_node,
389 *parent = NULL;
390 struct nfs4_state_owner *sp;
391
392 while (*p != NULL) {
393 parent = *p;
394 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
395
396 if (cred < sp->so_cred)
397 p = &parent->rb_left;
398 else if (cred > sp->so_cred)
399 p = &parent->rb_right;
400 else {
401 if (!list_empty(&sp->so_lru))
402 list_del_init(&sp->so_lru);
403 atomic_inc(&sp->so_count);
404 return sp;
405 }
406 }
407 return NULL;
408}
409
410static struct nfs4_state_owner *
411nfs4_insert_state_owner_locked(struct nfs4_state_owner *new)
412{
413 struct nfs_server *server = new->so_server;
414 struct rb_node **p = &server->state_owners.rb_node,
415 *parent = NULL;
416 struct nfs4_state_owner *sp;
417 int err;
418
419 while (*p != NULL) {
420 parent = *p;
421 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node);
422
423 if (new->so_cred < sp->so_cred)
424 p = &parent->rb_left;
425 else if (new->so_cred > sp->so_cred)
426 p = &parent->rb_right;
427 else {
428 if (!list_empty(&sp->so_lru))
429 list_del_init(&sp->so_lru);
430 atomic_inc(&sp->so_count);
431 return sp;
432 }
433 }
434 err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id);
435 if (err)
436 return ERR_PTR(err);
437 rb_link_node(&new->so_server_node, parent, p);
438 rb_insert_color(&new->so_server_node, &server->state_owners);
439 return new;
440}
441
442static void
443nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp)
444{
445 struct nfs_server *server = sp->so_server;
446
447 if (!RB_EMPTY_NODE(&sp->so_server_node))
448 rb_erase(&sp->so_server_node, &server->state_owners);
449 ida_remove(&server->openowner_id, sp->so_seqid.owner_id);
450}
451
452static void
453nfs4_init_seqid_counter(struct nfs_seqid_counter *sc)
454{
455 sc->create_time = ktime_get();
456 sc->flags = 0;
457 sc->counter = 0;
458 spin_lock_init(&sc->lock);
459 INIT_LIST_HEAD(&sc->list);
460 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue");
461}
462
463static void
464nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc)
465{
466 rpc_destroy_wait_queue(&sc->wait);
467}
468
469/*
470 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to
471 * create a new state_owner.
472 *
473 */
474static struct nfs4_state_owner *
475nfs4_alloc_state_owner(struct nfs_server *server,
476 struct rpc_cred *cred,
477 gfp_t gfp_flags)
478{
479 struct nfs4_state_owner *sp;
480
481 sp = kzalloc(sizeof(*sp), gfp_flags);
482 if (!sp)
483 return NULL;
484 sp->so_server = server;
485 sp->so_cred = get_rpccred(cred);
486 spin_lock_init(&sp->so_lock);
487 INIT_LIST_HEAD(&sp->so_states);
488 nfs4_init_seqid_counter(&sp->so_seqid);
489 atomic_set(&sp->so_count, 1);
490 INIT_LIST_HEAD(&sp->so_lru);
491 seqcount_init(&sp->so_reclaim_seqcount);
492 mutex_init(&sp->so_delegreturn_mutex);
493 return sp;
494}
495
496static void
497nfs4_reset_state_owner(struct nfs4_state_owner *sp)
498{
499 /* This state_owner is no longer usable, but must
500 * remain in place so that state recovery can find it
501 * and the opens associated with it.
502 * It may also be used for new 'open' request to
503 * return a delegation to the server.
504 * So update the 'create_time' so that it looks like
505 * a new state_owner. This will cause the server to
506 * request an OPEN_CONFIRM to start a new sequence.
507 */
508 sp->so_seqid.create_time = ktime_get();
509}
510
511static void nfs4_free_state_owner(struct nfs4_state_owner *sp)
512{
513 nfs4_destroy_seqid_counter(&sp->so_seqid);
514 put_rpccred(sp->so_cred);
515 kfree(sp);
516}
517
518static void nfs4_gc_state_owners(struct nfs_server *server)
519{
520 struct nfs_client *clp = server->nfs_client;
521 struct nfs4_state_owner *sp, *tmp;
522 unsigned long time_min, time_max;
523 LIST_HEAD(doomed);
524
525 spin_lock(&clp->cl_lock);
526 time_max = jiffies;
527 time_min = (long)time_max - (long)clp->cl_lease_time;
528 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
529 /* NB: LRU is sorted so that oldest is at the head */
530 if (time_in_range(sp->so_expires, time_min, time_max))
531 break;
532 list_move(&sp->so_lru, &doomed);
533 nfs4_remove_state_owner_locked(sp);
534 }
535 spin_unlock(&clp->cl_lock);
536
537 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
538 list_del(&sp->so_lru);
539 nfs4_free_state_owner(sp);
540 }
541}
542
543/**
544 * nfs4_get_state_owner - Look up a state owner given a credential
545 * @server: nfs_server to search
546 * @cred: RPC credential to match
547 *
548 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL.
549 */
550struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server,
551 struct rpc_cred *cred,
552 gfp_t gfp_flags)
553{
554 struct nfs_client *clp = server->nfs_client;
555 struct nfs4_state_owner *sp, *new;
556
557 spin_lock(&clp->cl_lock);
558 sp = nfs4_find_state_owner_locked(server, cred);
559 spin_unlock(&clp->cl_lock);
560 if (sp != NULL)
561 goto out;
562 new = nfs4_alloc_state_owner(server, cred, gfp_flags);
563 if (new == NULL)
564 goto out;
565 do {
566 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0)
567 break;
568 spin_lock(&clp->cl_lock);
569 sp = nfs4_insert_state_owner_locked(new);
570 spin_unlock(&clp->cl_lock);
571 } while (sp == ERR_PTR(-EAGAIN));
572 if (sp != new)
573 nfs4_free_state_owner(new);
574out:
575 nfs4_gc_state_owners(server);
576 return sp;
577}
578
579/**
580 * nfs4_put_state_owner - Release a nfs4_state_owner
581 * @sp: state owner data to release
582 *
583 * Note that we keep released state owners on an LRU
584 * list.
585 * This caches valid state owners so that they can be
586 * reused, to avoid the OPEN_CONFIRM on minor version 0.
587 * It also pins the uniquifier of dropped state owners for
588 * a while, to ensure that those state owner names are
589 * never reused.
590 */
591void nfs4_put_state_owner(struct nfs4_state_owner *sp)
592{
593 struct nfs_server *server = sp->so_server;
594 struct nfs_client *clp = server->nfs_client;
595
596 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock))
597 return;
598
599 sp->so_expires = jiffies;
600 list_add_tail(&sp->so_lru, &server->state_owners_lru);
601 spin_unlock(&clp->cl_lock);
602}
603
604/**
605 * nfs4_purge_state_owners - Release all cached state owners
606 * @server: nfs_server with cached state owners to release
607 *
608 * Called at umount time. Remaining state owners will be on
609 * the LRU with ref count of zero.
610 */
611void nfs4_purge_state_owners(struct nfs_server *server)
612{
613 struct nfs_client *clp = server->nfs_client;
614 struct nfs4_state_owner *sp, *tmp;
615 LIST_HEAD(doomed);
616
617 spin_lock(&clp->cl_lock);
618 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) {
619 list_move(&sp->so_lru, &doomed);
620 nfs4_remove_state_owner_locked(sp);
621 }
622 spin_unlock(&clp->cl_lock);
623
624 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) {
625 list_del(&sp->so_lru);
626 nfs4_free_state_owner(sp);
627 }
628}
629
630static struct nfs4_state *
631nfs4_alloc_open_state(void)
632{
633 struct nfs4_state *state;
634
635 state = kzalloc(sizeof(*state), GFP_NOFS);
636 if (!state)
637 return NULL;
638 atomic_set(&state->count, 1);
639 INIT_LIST_HEAD(&state->lock_states);
640 spin_lock_init(&state->state_lock);
641 seqlock_init(&state->seqlock);
642 return state;
643}
644
645void
646nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode)
647{
648 if (state->state == fmode)
649 return;
650 /* NB! List reordering - see the reclaim code for why. */
651 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) {
652 if (fmode & FMODE_WRITE)
653 list_move(&state->open_states, &state->owner->so_states);
654 else
655 list_move_tail(&state->open_states, &state->owner->so_states);
656 }
657 state->state = fmode;
658}
659
660static struct nfs4_state *
661__nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner)
662{
663 struct nfs_inode *nfsi = NFS_I(inode);
664 struct nfs4_state *state;
665
666 list_for_each_entry(state, &nfsi->open_states, inode_states) {
667 if (state->owner != owner)
668 continue;
669 if (!nfs4_valid_open_stateid(state))
670 continue;
671 if (atomic_inc_not_zero(&state->count))
672 return state;
673 }
674 return NULL;
675}
676
677static void
678nfs4_free_open_state(struct nfs4_state *state)
679{
680 kfree(state);
681}
682
683struct nfs4_state *
684nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner)
685{
686 struct nfs4_state *state, *new;
687 struct nfs_inode *nfsi = NFS_I(inode);
688
689 spin_lock(&inode->i_lock);
690 state = __nfs4_find_state_byowner(inode, owner);
691 spin_unlock(&inode->i_lock);
692 if (state)
693 goto out;
694 new = nfs4_alloc_open_state();
695 spin_lock(&owner->so_lock);
696 spin_lock(&inode->i_lock);
697 state = __nfs4_find_state_byowner(inode, owner);
698 if (state == NULL && new != NULL) {
699 state = new;
700 state->owner = owner;
701 atomic_inc(&owner->so_count);
702 list_add(&state->inode_states, &nfsi->open_states);
703 ihold(inode);
704 state->inode = inode;
705 spin_unlock(&inode->i_lock);
706 /* Note: The reclaim code dictates that we add stateless
707 * and read-only stateids to the end of the list */
708 list_add_tail(&state->open_states, &owner->so_states);
709 spin_unlock(&owner->so_lock);
710 } else {
711 spin_unlock(&inode->i_lock);
712 spin_unlock(&owner->so_lock);
713 if (new)
714 nfs4_free_open_state(new);
715 }
716out:
717 return state;
718}
719
720void nfs4_put_open_state(struct nfs4_state *state)
721{
722 struct inode *inode = state->inode;
723 struct nfs4_state_owner *owner = state->owner;
724
725 if (!atomic_dec_and_lock(&state->count, &owner->so_lock))
726 return;
727 spin_lock(&inode->i_lock);
728 list_del(&state->inode_states);
729 list_del(&state->open_states);
730 spin_unlock(&inode->i_lock);
731 spin_unlock(&owner->so_lock);
732 iput(inode);
733 nfs4_free_open_state(state);
734 nfs4_put_state_owner(owner);
735}
736
737/*
738 * Close the current file.
739 */
740static void __nfs4_close(struct nfs4_state *state,
741 fmode_t fmode, gfp_t gfp_mask, int wait)
742{
743 struct nfs4_state_owner *owner = state->owner;
744 int call_close = 0;
745 fmode_t newstate;
746
747 atomic_inc(&owner->so_count);
748 /* Protect against nfs4_find_state() */
749 spin_lock(&owner->so_lock);
750 switch (fmode & (FMODE_READ | FMODE_WRITE)) {
751 case FMODE_READ:
752 state->n_rdonly--;
753 break;
754 case FMODE_WRITE:
755 state->n_wronly--;
756 break;
757 case FMODE_READ|FMODE_WRITE:
758 state->n_rdwr--;
759 }
760 newstate = FMODE_READ|FMODE_WRITE;
761 if (state->n_rdwr == 0) {
762 if (state->n_rdonly == 0) {
763 newstate &= ~FMODE_READ;
764 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags);
765 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
766 }
767 if (state->n_wronly == 0) {
768 newstate &= ~FMODE_WRITE;
769 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags);
770 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags);
771 }
772 if (newstate == 0)
773 clear_bit(NFS_DELEGATED_STATE, &state->flags);
774 }
775 nfs4_state_set_mode_locked(state, newstate);
776 spin_unlock(&owner->so_lock);
777
778 if (!call_close) {
779 nfs4_put_open_state(state);
780 nfs4_put_state_owner(owner);
781 } else
782 nfs4_do_close(state, gfp_mask, wait);
783}
784
785void nfs4_close_state(struct nfs4_state *state, fmode_t fmode)
786{
787 __nfs4_close(state, fmode, GFP_NOFS, 0);
788}
789
790void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode)
791{
792 __nfs4_close(state, fmode, GFP_KERNEL, 1);
793}
794
795/*
796 * Search the state->lock_states for an existing lock_owner
797 * that is compatible with either of the given owners.
798 * If the second is non-zero, then the first refers to a Posix-lock
799 * owner (current->files) and the second refers to a flock/OFD
800 * owner (struct file*). In that case, prefer a match for the first
801 * owner.
802 * If both sorts of locks are held on the one file we cannot know
803 * which stateid was intended to be used, so a "correct" choice cannot
804 * be made. Failing that, a "consistent" choice is preferable. The
805 * consistent choice we make is to prefer the first owner, that of a
806 * Posix lock.
807 */
808static struct nfs4_lock_state *
809__nfs4_find_lock_state(struct nfs4_state *state,
810 fl_owner_t fl_owner, fl_owner_t fl_owner2)
811{
812 struct nfs4_lock_state *pos, *ret = NULL;
813 list_for_each_entry(pos, &state->lock_states, ls_locks) {
814 if (pos->ls_owner == fl_owner) {
815 ret = pos;
816 break;
817 }
818 if (pos->ls_owner == fl_owner2)
819 ret = pos;
820 }
821 if (ret)
822 atomic_inc(&ret->ls_count);
823 return ret;
824}
825
826/*
827 * Return a compatible lock_state. If no initialized lock_state structure
828 * exists, return an uninitialized one.
829 *
830 */
831static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner)
832{
833 struct nfs4_lock_state *lsp;
834 struct nfs_server *server = state->owner->so_server;
835
836 lsp = kzalloc(sizeof(*lsp), GFP_NOFS);
837 if (lsp == NULL)
838 return NULL;
839 nfs4_init_seqid_counter(&lsp->ls_seqid);
840 atomic_set(&lsp->ls_count, 1);
841 lsp->ls_state = state;
842 lsp->ls_owner = fl_owner;
843 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS);
844 if (lsp->ls_seqid.owner_id < 0)
845 goto out_free;
846 INIT_LIST_HEAD(&lsp->ls_locks);
847 return lsp;
848out_free:
849 kfree(lsp);
850 return NULL;
851}
852
853void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp)
854{
855 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id);
856 nfs4_destroy_seqid_counter(&lsp->ls_seqid);
857 kfree(lsp);
858}
859
860/*
861 * Return a compatible lock_state. If no initialized lock_state structure
862 * exists, return an uninitialized one.
863 *
864 */
865static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner)
866{
867 struct nfs4_lock_state *lsp, *new = NULL;
868
869 for(;;) {
870 spin_lock(&state->state_lock);
871 lsp = __nfs4_find_lock_state(state, owner, 0);
872 if (lsp != NULL)
873 break;
874 if (new != NULL) {
875 list_add(&new->ls_locks, &state->lock_states);
876 set_bit(LK_STATE_IN_USE, &state->flags);
877 lsp = new;
878 new = NULL;
879 break;
880 }
881 spin_unlock(&state->state_lock);
882 new = nfs4_alloc_lock_state(state, owner);
883 if (new == NULL)
884 return NULL;
885 }
886 spin_unlock(&state->state_lock);
887 if (new != NULL)
888 nfs4_free_lock_state(state->owner->so_server, new);
889 return lsp;
890}
891
892/*
893 * Release reference to lock_state, and free it if we see that
894 * it is no longer in use
895 */
896void nfs4_put_lock_state(struct nfs4_lock_state *lsp)
897{
898 struct nfs_server *server;
899 struct nfs4_state *state;
900
901 if (lsp == NULL)
902 return;
903 state = lsp->ls_state;
904 if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock))
905 return;
906 list_del(&lsp->ls_locks);
907 if (list_empty(&state->lock_states))
908 clear_bit(LK_STATE_IN_USE, &state->flags);
909 spin_unlock(&state->state_lock);
910 server = state->owner->so_server;
911 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) {
912 struct nfs_client *clp = server->nfs_client;
913
914 clp->cl_mvops->free_lock_state(server, lsp);
915 } else
916 nfs4_free_lock_state(server, lsp);
917}
918
919static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
920{
921 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner;
922
923 dst->fl_u.nfs4_fl.owner = lsp;
924 atomic_inc(&lsp->ls_count);
925}
926
927static void nfs4_fl_release_lock(struct file_lock *fl)
928{
929 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner);
930}
931
932static const struct file_lock_operations nfs4_fl_lock_ops = {
933 .fl_copy_lock = nfs4_fl_copy_lock,
934 .fl_release_private = nfs4_fl_release_lock,
935};
936
937int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl)
938{
939 struct nfs4_lock_state *lsp;
940
941 if (fl->fl_ops != NULL)
942 return 0;
943 lsp = nfs4_get_lock_state(state, fl->fl_owner);
944 if (lsp == NULL)
945 return -ENOMEM;
946 fl->fl_u.nfs4_fl.owner = lsp;
947 fl->fl_ops = &nfs4_fl_lock_ops;
948 return 0;
949}
950
951static int nfs4_copy_lock_stateid(nfs4_stateid *dst,
952 struct nfs4_state *state,
953 const struct nfs_lock_context *l_ctx)
954{
955 struct nfs4_lock_state *lsp;
956 fl_owner_t fl_owner, fl_flock_owner;
957 int ret = -ENOENT;
958
959 if (l_ctx == NULL)
960 goto out;
961
962 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0)
963 goto out;
964
965 fl_owner = l_ctx->lockowner;
966 fl_flock_owner = l_ctx->open_context->flock_owner;
967
968 spin_lock(&state->state_lock);
969 lsp = __nfs4_find_lock_state(state, fl_owner, fl_flock_owner);
970 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags))
971 ret = -EIO;
972 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) {
973 nfs4_stateid_copy(dst, &lsp->ls_stateid);
974 ret = 0;
975 }
976 spin_unlock(&state->state_lock);
977 nfs4_put_lock_state(lsp);
978out:
979 return ret;
980}
981
982static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state)
983{
984 const nfs4_stateid *src;
985 int seq;
986
987 do {
988 src = &zero_stateid;
989 seq = read_seqbegin(&state->seqlock);
990 if (test_bit(NFS_OPEN_STATE, &state->flags))
991 src = &state->open_stateid;
992 nfs4_stateid_copy(dst, src);
993 } while (read_seqretry(&state->seqlock, seq));
994}
995
996/*
997 * Byte-range lock aware utility to initialize the stateid of read/write
998 * requests.
999 */
1000int nfs4_select_rw_stateid(struct nfs4_state *state,
1001 fmode_t fmode, const struct nfs_lock_context *l_ctx,
1002 nfs4_stateid *dst, struct rpc_cred **cred)
1003{
1004 int ret;
1005
1006 if (!nfs4_valid_open_stateid(state))
1007 return -EIO;
1008 if (cred != NULL)
1009 *cred = NULL;
1010 ret = nfs4_copy_lock_stateid(dst, state, l_ctx);
1011 if (ret == -EIO)
1012 /* A lost lock - don't even consider delegations */
1013 goto out;
1014 /* returns true if delegation stateid found and copied */
1015 if (nfs4_copy_delegation_stateid(state->inode, fmode, dst, cred)) {
1016 ret = 0;
1017 goto out;
1018 }
1019 if (ret != -ENOENT)
1020 /* nfs4_copy_delegation_stateid() didn't over-write
1021 * dst, so it still has the lock stateid which we now
1022 * choose to use.
1023 */
1024 goto out;
1025 nfs4_copy_open_stateid(dst, state);
1026 ret = 0;
1027out:
1028 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41))
1029 dst->seqid = 0;
1030 return ret;
1031}
1032
1033struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask)
1034{
1035 struct nfs_seqid *new;
1036
1037 new = kmalloc(sizeof(*new), gfp_mask);
1038 if (new == NULL)
1039 return ERR_PTR(-ENOMEM);
1040 new->sequence = counter;
1041 INIT_LIST_HEAD(&new->list);
1042 new->task = NULL;
1043 return new;
1044}
1045
1046void nfs_release_seqid(struct nfs_seqid *seqid)
1047{
1048 struct nfs_seqid_counter *sequence;
1049
1050 if (seqid == NULL || list_empty(&seqid->list))
1051 return;
1052 sequence = seqid->sequence;
1053 spin_lock(&sequence->lock);
1054 list_del_init(&seqid->list);
1055 if (!list_empty(&sequence->list)) {
1056 struct nfs_seqid *next;
1057
1058 next = list_first_entry(&sequence->list,
1059 struct nfs_seqid, list);
1060 rpc_wake_up_queued_task(&sequence->wait, next->task);
1061 }
1062 spin_unlock(&sequence->lock);
1063}
1064
1065void nfs_free_seqid(struct nfs_seqid *seqid)
1066{
1067 nfs_release_seqid(seqid);
1068 kfree(seqid);
1069}
1070
1071/*
1072 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or
1073 * failed with a seqid incrementing error -
1074 * see comments nfs4.h:seqid_mutating_error()
1075 */
1076static void nfs_increment_seqid(int status, struct nfs_seqid *seqid)
1077{
1078 switch (status) {
1079 case 0:
1080 break;
1081 case -NFS4ERR_BAD_SEQID:
1082 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED)
1083 return;
1084 pr_warn_ratelimited("NFS: v4 server returned a bad"
1085 " sequence-id error on an"
1086 " unconfirmed sequence %p!\n",
1087 seqid->sequence);
1088 case -NFS4ERR_STALE_CLIENTID:
1089 case -NFS4ERR_STALE_STATEID:
1090 case -NFS4ERR_BAD_STATEID:
1091 case -NFS4ERR_BADXDR:
1092 case -NFS4ERR_RESOURCE:
1093 case -NFS4ERR_NOFILEHANDLE:
1094 case -NFS4ERR_MOVED:
1095 /* Non-seqid mutating errors */
1096 return;
1097 };
1098 /*
1099 * Note: no locking needed as we are guaranteed to be first
1100 * on the sequence list
1101 */
1102 seqid->sequence->counter++;
1103}
1104
1105void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid)
1106{
1107 struct nfs4_state_owner *sp;
1108
1109 if (seqid == NULL)
1110 return;
1111
1112 sp = container_of(seqid->sequence, struct nfs4_state_owner, so_seqid);
1113 if (status == -NFS4ERR_BAD_SEQID)
1114 nfs4_reset_state_owner(sp);
1115 if (!nfs4_has_session(sp->so_server->nfs_client))
1116 nfs_increment_seqid(status, seqid);
1117}
1118
1119/*
1120 * Increment the seqid if the LOCK/LOCKU succeeded, or
1121 * failed with a seqid incrementing error -
1122 * see comments nfs4.h:seqid_mutating_error()
1123 */
1124void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid)
1125{
1126 if (seqid != NULL)
1127 nfs_increment_seqid(status, seqid);
1128}
1129
1130int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task)
1131{
1132 struct nfs_seqid_counter *sequence;
1133 int status = 0;
1134
1135 if (seqid == NULL)
1136 goto out;
1137 sequence = seqid->sequence;
1138 spin_lock(&sequence->lock);
1139 seqid->task = task;
1140 if (list_empty(&seqid->list))
1141 list_add_tail(&seqid->list, &sequence->list);
1142 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid)
1143 goto unlock;
1144 rpc_sleep_on(&sequence->wait, task, NULL);
1145 status = -EAGAIN;
1146unlock:
1147 spin_unlock(&sequence->lock);
1148out:
1149 return status;
1150}
1151
1152static int nfs4_run_state_manager(void *);
1153
1154static void nfs4_clear_state_manager_bit(struct nfs_client *clp)
1155{
1156 smp_mb__before_atomic();
1157 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state);
1158 smp_mb__after_atomic();
1159 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING);
1160 rpc_wake_up(&clp->cl_rpcwaitq);
1161}
1162
1163/*
1164 * Schedule the nfs_client asynchronous state management routine
1165 */
1166void nfs4_schedule_state_manager(struct nfs_client *clp)
1167{
1168 struct task_struct *task;
1169 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1];
1170
1171 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
1172 return;
1173 __module_get(THIS_MODULE);
1174 atomic_inc(&clp->cl_count);
1175
1176 /* The rcu_read_lock() is not strictly necessary, as the state
1177 * manager is the only thread that ever changes the rpc_xprt
1178 * after it's initialized. At this point, we're single threaded. */
1179 rcu_read_lock();
1180 snprintf(buf, sizeof(buf), "%s-manager",
1181 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR));
1182 rcu_read_unlock();
1183 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf);
1184 if (IS_ERR(task)) {
1185 printk(KERN_ERR "%s: kthread_run: %ld\n",
1186 __func__, PTR_ERR(task));
1187 nfs4_clear_state_manager_bit(clp);
1188 nfs_put_client(clp);
1189 module_put(THIS_MODULE);
1190 }
1191}
1192
1193/*
1194 * Schedule a lease recovery attempt
1195 */
1196void nfs4_schedule_lease_recovery(struct nfs_client *clp)
1197{
1198 if (!clp)
1199 return;
1200 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1201 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1202 dprintk("%s: scheduling lease recovery for server %s\n", __func__,
1203 clp->cl_hostname);
1204 nfs4_schedule_state_manager(clp);
1205}
1206EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery);
1207
1208/**
1209 * nfs4_schedule_migration_recovery - trigger migration recovery
1210 *
1211 * @server: FSID that is migrating
1212 *
1213 * Returns zero if recovery has started, otherwise a negative NFS4ERR
1214 * value is returned.
1215 */
1216int nfs4_schedule_migration_recovery(const struct nfs_server *server)
1217{
1218 struct nfs_client *clp = server->nfs_client;
1219
1220 if (server->fh_expire_type != NFS4_FH_PERSISTENT) {
1221 pr_err("NFS: volatile file handles not supported (server %s)\n",
1222 clp->cl_hostname);
1223 return -NFS4ERR_IO;
1224 }
1225
1226 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
1227 return -NFS4ERR_IO;
1228
1229 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n",
1230 __func__,
1231 (unsigned long long)server->fsid.major,
1232 (unsigned long long)server->fsid.minor,
1233 clp->cl_hostname);
1234
1235 set_bit(NFS_MIG_IN_TRANSITION,
1236 &((struct nfs_server *)server)->mig_status);
1237 set_bit(NFS4CLNT_MOVED, &clp->cl_state);
1238
1239 nfs4_schedule_state_manager(clp);
1240 return 0;
1241}
1242EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery);
1243
1244/**
1245 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery
1246 *
1247 * @clp: server to check for moved leases
1248 *
1249 */
1250void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp)
1251{
1252 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n",
1253 __func__, clp->cl_clientid, clp->cl_hostname);
1254
1255 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state);
1256 nfs4_schedule_state_manager(clp);
1257}
1258EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery);
1259
1260int nfs4_wait_clnt_recover(struct nfs_client *clp)
1261{
1262 int res;
1263
1264 might_sleep();
1265
1266 atomic_inc(&clp->cl_count);
1267 res = wait_on_bit_action(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING,
1268 nfs_wait_bit_killable, TASK_KILLABLE);
1269 if (res)
1270 goto out;
1271 if (clp->cl_cons_state < 0)
1272 res = clp->cl_cons_state;
1273out:
1274 nfs_put_client(clp);
1275 return res;
1276}
1277
1278int nfs4_client_recover_expired_lease(struct nfs_client *clp)
1279{
1280 unsigned int loop;
1281 int ret;
1282
1283 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) {
1284 ret = nfs4_wait_clnt_recover(clp);
1285 if (ret != 0)
1286 break;
1287 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) &&
1288 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state))
1289 break;
1290 nfs4_schedule_state_manager(clp);
1291 ret = -EIO;
1292 }
1293 return ret;
1294}
1295
1296/*
1297 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN
1298 * @clp: client to process
1299 *
1300 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a
1301 * resend of the SETCLIENTID and hence re-establish the
1302 * callback channel. Then return all existing delegations.
1303 */
1304static void nfs40_handle_cb_pathdown(struct nfs_client *clp)
1305{
1306 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1307 nfs_expire_all_delegations(clp);
1308 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__,
1309 clp->cl_hostname);
1310}
1311
1312void nfs4_schedule_path_down_recovery(struct nfs_client *clp)
1313{
1314 nfs40_handle_cb_pathdown(clp);
1315 nfs4_schedule_state_manager(clp);
1316}
1317
1318static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state)
1319{
1320
1321 if (!nfs4_valid_open_stateid(state))
1322 return 0;
1323 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1324 /* Don't recover state that expired before the reboot */
1325 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) {
1326 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1327 return 0;
1328 }
1329 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags);
1330 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1331 return 1;
1332}
1333
1334int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state)
1335{
1336 if (!nfs4_valid_open_stateid(state))
1337 return 0;
1338 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags);
1339 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags);
1340 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags);
1341 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state);
1342 return 1;
1343}
1344
1345int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state)
1346{
1347 struct nfs_client *clp = server->nfs_client;
1348
1349 if (!nfs4_state_mark_reclaim_nograce(clp, state))
1350 return -EBADF;
1351 dprintk("%s: scheduling stateid recovery for server %s\n", __func__,
1352 clp->cl_hostname);
1353 nfs4_schedule_state_manager(clp);
1354 return 0;
1355}
1356EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery);
1357
1358static struct nfs4_lock_state *
1359nfs_state_find_lock_state_by_stateid(struct nfs4_state *state,
1360 const nfs4_stateid *stateid)
1361{
1362 struct nfs4_lock_state *pos;
1363
1364 list_for_each_entry(pos, &state->lock_states, ls_locks) {
1365 if (!test_bit(NFS_LOCK_INITIALIZED, &pos->ls_flags))
1366 continue;
1367 if (nfs4_stateid_match_other(&pos->ls_stateid, stateid))
1368 return pos;
1369 }
1370 return NULL;
1371}
1372
1373static bool nfs_state_lock_state_matches_stateid(struct nfs4_state *state,
1374 const nfs4_stateid *stateid)
1375{
1376 bool found = false;
1377
1378 if (test_bit(LK_STATE_IN_USE, &state->flags)) {
1379 spin_lock(&state->state_lock);
1380 if (nfs_state_find_lock_state_by_stateid(state, stateid))
1381 found = true;
1382 spin_unlock(&state->state_lock);
1383 }
1384 return found;
1385}
1386
1387void nfs_inode_find_state_and_recover(struct inode *inode,
1388 const nfs4_stateid *stateid)
1389{
1390 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client;
1391 struct nfs_inode *nfsi = NFS_I(inode);
1392 struct nfs_open_context *ctx;
1393 struct nfs4_state *state;
1394 bool found = false;
1395
1396 spin_lock(&inode->i_lock);
1397 list_for_each_entry(ctx, &nfsi->open_files, list) {
1398 state = ctx->state;
1399 if (state == NULL)
1400 continue;
1401 if (nfs4_stateid_match_other(&state->stateid, stateid) &&
1402 nfs4_state_mark_reclaim_nograce(clp, state)) {
1403 found = true;
1404 continue;
1405 }
1406 if (nfs_state_lock_state_matches_stateid(state, stateid) &&
1407 nfs4_state_mark_reclaim_nograce(clp, state))
1408 found = true;
1409 }
1410 spin_unlock(&inode->i_lock);
1411
1412 nfs_inode_find_delegation_state_and_recover(inode, stateid);
1413 if (found)
1414 nfs4_schedule_state_manager(clp);
1415}
1416
1417static void nfs4_state_mark_open_context_bad(struct nfs4_state *state)
1418{
1419 struct inode *inode = state->inode;
1420 struct nfs_inode *nfsi = NFS_I(inode);
1421 struct nfs_open_context *ctx;
1422
1423 spin_lock(&inode->i_lock);
1424 list_for_each_entry(ctx, &nfsi->open_files, list) {
1425 if (ctx->state != state)
1426 continue;
1427 set_bit(NFS_CONTEXT_BAD, &ctx->flags);
1428 }
1429 spin_unlock(&inode->i_lock);
1430}
1431
1432static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error)
1433{
1434 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags);
1435 nfs4_state_mark_open_context_bad(state);
1436}
1437
1438
1439static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops)
1440{
1441 struct inode *inode = state->inode;
1442 struct nfs_inode *nfsi = NFS_I(inode);
1443 struct file_lock *fl;
1444 int status = 0;
1445 struct file_lock_context *flctx = inode->i_flctx;
1446 struct list_head *list;
1447
1448 if (flctx == NULL)
1449 return 0;
1450
1451 list = &flctx->flc_posix;
1452
1453 /* Guard against delegation returns and new lock/unlock calls */
1454 down_write(&nfsi->rwsem);
1455 spin_lock(&flctx->flc_lock);
1456restart:
1457 list_for_each_entry(fl, list, fl_list) {
1458 if (nfs_file_open_context(fl->fl_file)->state != state)
1459 continue;
1460 spin_unlock(&flctx->flc_lock);
1461 status = ops->recover_lock(state, fl);
1462 switch (status) {
1463 case 0:
1464 break;
1465 case -ESTALE:
1466 case -NFS4ERR_ADMIN_REVOKED:
1467 case -NFS4ERR_STALE_STATEID:
1468 case -NFS4ERR_BAD_STATEID:
1469 case -NFS4ERR_EXPIRED:
1470 case -NFS4ERR_NO_GRACE:
1471 case -NFS4ERR_STALE_CLIENTID:
1472 case -NFS4ERR_BADSESSION:
1473 case -NFS4ERR_BADSLOT:
1474 case -NFS4ERR_BAD_HIGH_SLOT:
1475 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1476 goto out;
1477 default:
1478 pr_err("NFS: %s: unhandled error %d\n",
1479 __func__, status);
1480 case -ENOMEM:
1481 case -NFS4ERR_DENIED:
1482 case -NFS4ERR_RECLAIM_BAD:
1483 case -NFS4ERR_RECLAIM_CONFLICT:
1484 /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1485 status = 0;
1486 }
1487 spin_lock(&flctx->flc_lock);
1488 }
1489 if (list == &flctx->flc_posix) {
1490 list = &flctx->flc_flock;
1491 goto restart;
1492 }
1493 spin_unlock(&flctx->flc_lock);
1494out:
1495 up_write(&nfsi->rwsem);
1496 return status;
1497}
1498
1499static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops)
1500{
1501 struct nfs4_state *state;
1502 struct nfs4_lock_state *lock;
1503 int status = 0;
1504
1505 /* Note: we rely on the sp->so_states list being ordered
1506 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE)
1507 * states first.
1508 * This is needed to ensure that the server won't give us any
1509 * read delegations that we have to return if, say, we are
1510 * recovering after a network partition or a reboot from a
1511 * server that doesn't support a grace period.
1512 */
1513 spin_lock(&sp->so_lock);
1514 raw_write_seqcount_begin(&sp->so_reclaim_seqcount);
1515restart:
1516 list_for_each_entry(state, &sp->so_states, open_states) {
1517 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags))
1518 continue;
1519 if (!nfs4_valid_open_stateid(state))
1520 continue;
1521 if (state->state == 0)
1522 continue;
1523 atomic_inc(&state->count);
1524 spin_unlock(&sp->so_lock);
1525 status = ops->recover_open(sp, state);
1526 if (status >= 0) {
1527 status = nfs4_reclaim_locks(state, ops);
1528 if (status >= 0) {
1529 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) {
1530 spin_lock(&state->state_lock);
1531 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1532 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags))
1533 pr_warn_ratelimited("NFS: "
1534 "%s: Lock reclaim "
1535 "failed!\n", __func__);
1536 }
1537 spin_unlock(&state->state_lock);
1538 }
1539 clear_bit(NFS_STATE_RECLAIM_NOGRACE,
1540 &state->flags);
1541 nfs4_put_open_state(state);
1542 spin_lock(&sp->so_lock);
1543 goto restart;
1544 }
1545 }
1546 switch (status) {
1547 default:
1548 printk(KERN_ERR "NFS: %s: unhandled error %d\n",
1549 __func__, status);
1550 case -ENOENT:
1551 case -ENOMEM:
1552 case -EACCES:
1553 case -EROFS:
1554 case -EIO:
1555 case -ESTALE:
1556 /* Open state on this file cannot be recovered */
1557 nfs4_state_mark_recovery_failed(state, status);
1558 break;
1559 case -EAGAIN:
1560 ssleep(1);
1561 case -NFS4ERR_ADMIN_REVOKED:
1562 case -NFS4ERR_STALE_STATEID:
1563 case -NFS4ERR_OLD_STATEID:
1564 case -NFS4ERR_BAD_STATEID:
1565 case -NFS4ERR_RECLAIM_BAD:
1566 case -NFS4ERR_RECLAIM_CONFLICT:
1567 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1568 break;
1569 case -NFS4ERR_EXPIRED:
1570 case -NFS4ERR_NO_GRACE:
1571 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state);
1572 case -NFS4ERR_STALE_CLIENTID:
1573 case -NFS4ERR_BADSESSION:
1574 case -NFS4ERR_BADSLOT:
1575 case -NFS4ERR_BAD_HIGH_SLOT:
1576 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1577 goto out_err;
1578 }
1579 nfs4_put_open_state(state);
1580 spin_lock(&sp->so_lock);
1581 goto restart;
1582 }
1583 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1584 spin_unlock(&sp->so_lock);
1585 return 0;
1586out_err:
1587 nfs4_put_open_state(state);
1588 spin_lock(&sp->so_lock);
1589 raw_write_seqcount_end(&sp->so_reclaim_seqcount);
1590 spin_unlock(&sp->so_lock);
1591 return status;
1592}
1593
1594static void nfs4_clear_open_state(struct nfs4_state *state)
1595{
1596 struct nfs4_lock_state *lock;
1597
1598 clear_bit(NFS_DELEGATED_STATE, &state->flags);
1599 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1600 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1601 clear_bit(NFS_O_RDWR_STATE, &state->flags);
1602 spin_lock(&state->state_lock);
1603 list_for_each_entry(lock, &state->lock_states, ls_locks) {
1604 lock->ls_seqid.flags = 0;
1605 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags);
1606 }
1607 spin_unlock(&state->state_lock);
1608}
1609
1610static void nfs4_reset_seqids(struct nfs_server *server,
1611 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1612{
1613 struct nfs_client *clp = server->nfs_client;
1614 struct nfs4_state_owner *sp;
1615 struct rb_node *pos;
1616 struct nfs4_state *state;
1617
1618 spin_lock(&clp->cl_lock);
1619 for (pos = rb_first(&server->state_owners);
1620 pos != NULL;
1621 pos = rb_next(pos)) {
1622 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1623 sp->so_seqid.flags = 0;
1624 spin_lock(&sp->so_lock);
1625 list_for_each_entry(state, &sp->so_states, open_states) {
1626 if (mark_reclaim(clp, state))
1627 nfs4_clear_open_state(state);
1628 }
1629 spin_unlock(&sp->so_lock);
1630 }
1631 spin_unlock(&clp->cl_lock);
1632}
1633
1634static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp,
1635 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state))
1636{
1637 struct nfs_server *server;
1638
1639 rcu_read_lock();
1640 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1641 nfs4_reset_seqids(server, mark_reclaim);
1642 rcu_read_unlock();
1643}
1644
1645static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp)
1646{
1647 /* Mark all delegations for reclaim */
1648 nfs_delegation_mark_reclaim(clp);
1649 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot);
1650}
1651
1652static void nfs4_reclaim_complete(struct nfs_client *clp,
1653 const struct nfs4_state_recovery_ops *ops,
1654 struct rpc_cred *cred)
1655{
1656 /* Notify the server we're done reclaiming our state */
1657 if (ops->reclaim_complete)
1658 (void)ops->reclaim_complete(clp, cred);
1659}
1660
1661static void nfs4_clear_reclaim_server(struct nfs_server *server)
1662{
1663 struct nfs_client *clp = server->nfs_client;
1664 struct nfs4_state_owner *sp;
1665 struct rb_node *pos;
1666 struct nfs4_state *state;
1667
1668 spin_lock(&clp->cl_lock);
1669 for (pos = rb_first(&server->state_owners);
1670 pos != NULL;
1671 pos = rb_next(pos)) {
1672 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node);
1673 spin_lock(&sp->so_lock);
1674 list_for_each_entry(state, &sp->so_states, open_states) {
1675 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT,
1676 &state->flags))
1677 continue;
1678 nfs4_state_mark_reclaim_nograce(clp, state);
1679 }
1680 spin_unlock(&sp->so_lock);
1681 }
1682 spin_unlock(&clp->cl_lock);
1683}
1684
1685static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp)
1686{
1687 struct nfs_server *server;
1688
1689 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state))
1690 return 0;
1691
1692 rcu_read_lock();
1693 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link)
1694 nfs4_clear_reclaim_server(server);
1695 rcu_read_unlock();
1696
1697 nfs_delegation_reap_unclaimed(clp);
1698 return 1;
1699}
1700
1701static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp)
1702{
1703 const struct nfs4_state_recovery_ops *ops;
1704 struct rpc_cred *cred;
1705
1706 if (!nfs4_state_clear_reclaim_reboot(clp))
1707 return;
1708 ops = clp->cl_mvops->reboot_recovery_ops;
1709 cred = nfs4_get_clid_cred(clp);
1710 nfs4_reclaim_complete(clp, ops, cred);
1711 put_rpccred(cred);
1712}
1713
1714static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp)
1715{
1716 nfs_mark_test_expired_all_delegations(clp);
1717 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce);
1718}
1719
1720static int nfs4_recovery_handle_error(struct nfs_client *clp, int error)
1721{
1722 switch (error) {
1723 case 0:
1724 break;
1725 case -NFS4ERR_CB_PATH_DOWN:
1726 nfs40_handle_cb_pathdown(clp);
1727 break;
1728 case -NFS4ERR_NO_GRACE:
1729 nfs4_state_end_reclaim_reboot(clp);
1730 break;
1731 case -NFS4ERR_STALE_CLIENTID:
1732 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1733 nfs4_state_start_reclaim_reboot(clp);
1734 break;
1735 case -NFS4ERR_EXPIRED:
1736 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1737 nfs4_state_start_reclaim_nograce(clp);
1738 break;
1739 case -NFS4ERR_BADSESSION:
1740 case -NFS4ERR_BADSLOT:
1741 case -NFS4ERR_BAD_HIGH_SLOT:
1742 case -NFS4ERR_DEADSESSION:
1743 case -NFS4ERR_SEQ_FALSE_RETRY:
1744 case -NFS4ERR_SEQ_MISORDERED:
1745 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
1746 /* Zero session reset errors */
1747 break;
1748 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1749 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
1750 break;
1751 default:
1752 dprintk("%s: failed to handle error %d for server %s\n",
1753 __func__, error, clp->cl_hostname);
1754 return error;
1755 }
1756 dprintk("%s: handled error %d for server %s\n", __func__, error,
1757 clp->cl_hostname);
1758 return 0;
1759}
1760
1761static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops)
1762{
1763 struct nfs4_state_owner *sp;
1764 struct nfs_server *server;
1765 struct rb_node *pos;
1766 int status = 0;
1767
1768restart:
1769 rcu_read_lock();
1770 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
1771 nfs4_purge_state_owners(server);
1772 spin_lock(&clp->cl_lock);
1773 for (pos = rb_first(&server->state_owners);
1774 pos != NULL;
1775 pos = rb_next(pos)) {
1776 sp = rb_entry(pos,
1777 struct nfs4_state_owner, so_server_node);
1778 if (!test_and_clear_bit(ops->owner_flag_bit,
1779 &sp->so_flags))
1780 continue;
1781 if (!atomic_inc_not_zero(&sp->so_count))
1782 continue;
1783 spin_unlock(&clp->cl_lock);
1784 rcu_read_unlock();
1785
1786 status = nfs4_reclaim_open_state(sp, ops);
1787 if (status < 0) {
1788 set_bit(ops->owner_flag_bit, &sp->so_flags);
1789 nfs4_put_state_owner(sp);
1790 status = nfs4_recovery_handle_error(clp, status);
1791 return (status != 0) ? status : -EAGAIN;
1792 }
1793
1794 nfs4_put_state_owner(sp);
1795 goto restart;
1796 }
1797 spin_unlock(&clp->cl_lock);
1798 }
1799 rcu_read_unlock();
1800 return 0;
1801}
1802
1803static int nfs4_check_lease(struct nfs_client *clp)
1804{
1805 struct rpc_cred *cred;
1806 const struct nfs4_state_maintenance_ops *ops =
1807 clp->cl_mvops->state_renewal_ops;
1808 int status;
1809
1810 /* Is the client already known to have an expired lease? */
1811 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
1812 return 0;
1813 spin_lock(&clp->cl_lock);
1814 cred = ops->get_state_renewal_cred_locked(clp);
1815 spin_unlock(&clp->cl_lock);
1816 if (cred == NULL) {
1817 cred = nfs4_get_clid_cred(clp);
1818 status = -ENOKEY;
1819 if (cred == NULL)
1820 goto out;
1821 }
1822 status = ops->renew_lease(clp, cred);
1823 put_rpccred(cred);
1824 if (status == -ETIMEDOUT) {
1825 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1826 return 0;
1827 }
1828out:
1829 return nfs4_recovery_handle_error(clp, status);
1830}
1831
1832/* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors
1833 * and for recoverable errors on EXCHANGE_ID for v4.1
1834 */
1835static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status)
1836{
1837 switch (status) {
1838 case -NFS4ERR_SEQ_MISORDERED:
1839 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state))
1840 return -ESERVERFAULT;
1841 /* Lease confirmation error: retry after purging the lease */
1842 ssleep(1);
1843 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1844 break;
1845 case -NFS4ERR_STALE_CLIENTID:
1846 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1847 nfs4_state_start_reclaim_reboot(clp);
1848 break;
1849 case -NFS4ERR_CLID_INUSE:
1850 pr_err("NFS: Server %s reports our clientid is in use\n",
1851 clp->cl_hostname);
1852 nfs_mark_client_ready(clp, -EPERM);
1853 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
1854 return -EPERM;
1855 case -EACCES:
1856 case -NFS4ERR_DELAY:
1857 case -ETIMEDOUT:
1858 case -EAGAIN:
1859 ssleep(1);
1860 break;
1861
1862 case -NFS4ERR_MINOR_VERS_MISMATCH:
1863 if (clp->cl_cons_state == NFS_CS_SESSION_INITING)
1864 nfs_mark_client_ready(clp, -EPROTONOSUPPORT);
1865 dprintk("%s: exit with error %d for server %s\n",
1866 __func__, -EPROTONOSUPPORT, clp->cl_hostname);
1867 return -EPROTONOSUPPORT;
1868 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
1869 * in nfs4_exchange_id */
1870 default:
1871 dprintk("%s: exit with error %d for server %s\n", __func__,
1872 status, clp->cl_hostname);
1873 return status;
1874 }
1875 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1876 dprintk("%s: handled error %d for server %s\n", __func__, status,
1877 clp->cl_hostname);
1878 return 0;
1879}
1880
1881static int nfs4_establish_lease(struct nfs_client *clp)
1882{
1883 struct rpc_cred *cred;
1884 const struct nfs4_state_recovery_ops *ops =
1885 clp->cl_mvops->reboot_recovery_ops;
1886 int status;
1887
1888 nfs4_begin_drain_session(clp);
1889 cred = nfs4_get_clid_cred(clp);
1890 if (cred == NULL)
1891 return -ENOENT;
1892 status = ops->establish_clid(clp, cred);
1893 put_rpccred(cred);
1894 if (status != 0)
1895 return status;
1896 pnfs_destroy_all_layouts(clp);
1897 return 0;
1898}
1899
1900/*
1901 * Returns zero or a negative errno. NFS4ERR values are converted
1902 * to local errno values.
1903 */
1904static int nfs4_reclaim_lease(struct nfs_client *clp)
1905{
1906 int status;
1907
1908 status = nfs4_establish_lease(clp);
1909 if (status < 0)
1910 return nfs4_handle_reclaim_lease_error(clp, status);
1911 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state))
1912 nfs4_state_start_reclaim_nograce(clp);
1913 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state))
1914 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1915 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1916 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1917 return 0;
1918}
1919
1920static int nfs4_purge_lease(struct nfs_client *clp)
1921{
1922 int status;
1923
1924 status = nfs4_establish_lease(clp);
1925 if (status < 0)
1926 return nfs4_handle_reclaim_lease_error(clp, status);
1927 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
1928 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state);
1929 nfs4_state_start_reclaim_nograce(clp);
1930 return 0;
1931}
1932
1933/*
1934 * Try remote migration of one FSID from a source server to a
1935 * destination server. The source server provides a list of
1936 * potential destinations.
1937 *
1938 * Returns zero or a negative NFS4ERR status code.
1939 */
1940static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred)
1941{
1942 struct nfs_client *clp = server->nfs_client;
1943 struct nfs4_fs_locations *locations = NULL;
1944 struct inode *inode;
1945 struct page *page;
1946 int status, result;
1947
1948 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__,
1949 (unsigned long long)server->fsid.major,
1950 (unsigned long long)server->fsid.minor,
1951 clp->cl_hostname);
1952
1953 result = 0;
1954 page = alloc_page(GFP_KERNEL);
1955 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
1956 if (page == NULL || locations == NULL) {
1957 dprintk("<-- %s: no memory\n", __func__);
1958 goto out;
1959 }
1960
1961 inode = d_inode(server->super->s_root);
1962 result = nfs4_proc_get_locations(inode, locations, page, cred);
1963 if (result) {
1964 dprintk("<-- %s: failed to retrieve fs_locations: %d\n",
1965 __func__, result);
1966 goto out;
1967 }
1968
1969 result = -NFS4ERR_NXIO;
1970 if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) {
1971 dprintk("<-- %s: No fs_locations data, migration skipped\n",
1972 __func__);
1973 goto out;
1974 }
1975
1976 nfs4_begin_drain_session(clp);
1977
1978 status = nfs4_replace_transport(server, locations);
1979 if (status != 0) {
1980 dprintk("<-- %s: failed to replace transport: %d\n",
1981 __func__, status);
1982 goto out;
1983 }
1984
1985 result = 0;
1986 dprintk("<-- %s: migration succeeded\n", __func__);
1987
1988out:
1989 if (page != NULL)
1990 __free_page(page);
1991 kfree(locations);
1992 if (result) {
1993 pr_err("NFS: migration recovery failed (server %s)\n",
1994 clp->cl_hostname);
1995 set_bit(NFS_MIG_FAILED, &server->mig_status);
1996 }
1997 return result;
1998}
1999
2000/*
2001 * Returns zero or a negative NFS4ERR status code.
2002 */
2003static int nfs4_handle_migration(struct nfs_client *clp)
2004{
2005 const struct nfs4_state_maintenance_ops *ops =
2006 clp->cl_mvops->state_renewal_ops;
2007 struct nfs_server *server;
2008 struct rpc_cred *cred;
2009
2010 dprintk("%s: migration reported on \"%s\"\n", __func__,
2011 clp->cl_hostname);
2012
2013 spin_lock(&clp->cl_lock);
2014 cred = ops->get_state_renewal_cred_locked(clp);
2015 spin_unlock(&clp->cl_lock);
2016 if (cred == NULL)
2017 return -NFS4ERR_NOENT;
2018
2019 clp->cl_mig_gen++;
2020restart:
2021 rcu_read_lock();
2022 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2023 int status;
2024
2025 if (server->mig_gen == clp->cl_mig_gen)
2026 continue;
2027 server->mig_gen = clp->cl_mig_gen;
2028
2029 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION,
2030 &server->mig_status))
2031 continue;
2032
2033 rcu_read_unlock();
2034 status = nfs4_try_migration(server, cred);
2035 if (status < 0) {
2036 put_rpccred(cred);
2037 return status;
2038 }
2039 goto restart;
2040 }
2041 rcu_read_unlock();
2042 put_rpccred(cred);
2043 return 0;
2044}
2045
2046/*
2047 * Test each nfs_server on the clp's cl_superblocks list to see
2048 * if it's moved to another server. Stop when the server no longer
2049 * returns NFS4ERR_LEASE_MOVED.
2050 */
2051static int nfs4_handle_lease_moved(struct nfs_client *clp)
2052{
2053 const struct nfs4_state_maintenance_ops *ops =
2054 clp->cl_mvops->state_renewal_ops;
2055 struct nfs_server *server;
2056 struct rpc_cred *cred;
2057
2058 dprintk("%s: lease moved reported on \"%s\"\n", __func__,
2059 clp->cl_hostname);
2060
2061 spin_lock(&clp->cl_lock);
2062 cred = ops->get_state_renewal_cred_locked(clp);
2063 spin_unlock(&clp->cl_lock);
2064 if (cred == NULL)
2065 return -NFS4ERR_NOENT;
2066
2067 clp->cl_mig_gen++;
2068restart:
2069 rcu_read_lock();
2070 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) {
2071 struct inode *inode;
2072 int status;
2073
2074 if (server->mig_gen == clp->cl_mig_gen)
2075 continue;
2076 server->mig_gen = clp->cl_mig_gen;
2077
2078 rcu_read_unlock();
2079
2080 inode = d_inode(server->super->s_root);
2081 status = nfs4_proc_fsid_present(inode, cred);
2082 if (status != -NFS4ERR_MOVED)
2083 goto restart; /* wasn't this one */
2084 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED)
2085 goto restart; /* there are more */
2086 goto out;
2087 }
2088 rcu_read_unlock();
2089
2090out:
2091 put_rpccred(cred);
2092 return 0;
2093}
2094
2095/**
2096 * nfs4_discover_server_trunking - Detect server IP address trunking
2097 *
2098 * @clp: nfs_client under test
2099 * @result: OUT: found nfs_client, or clp
2100 *
2101 * Returns zero or a negative errno. If zero is returned,
2102 * an nfs_client pointer is planted in "result".
2103 *
2104 * Note: since we are invoked in process context, and
2105 * not from inside the state manager, we cannot use
2106 * nfs4_handle_reclaim_lease_error().
2107 */
2108int nfs4_discover_server_trunking(struct nfs_client *clp,
2109 struct nfs_client **result)
2110{
2111 const struct nfs4_state_recovery_ops *ops =
2112 clp->cl_mvops->reboot_recovery_ops;
2113 struct rpc_clnt *clnt;
2114 struct rpc_cred *cred;
2115 int i, status;
2116
2117 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname);
2118
2119 clnt = clp->cl_rpcclient;
2120 i = 0;
2121
2122 mutex_lock(&nfs_clid_init_mutex);
2123again:
2124 status = -ENOENT;
2125 cred = nfs4_get_clid_cred(clp);
2126 if (cred == NULL)
2127 goto out_unlock;
2128
2129 status = ops->detect_trunking(clp, result, cred);
2130 put_rpccred(cred);
2131 switch (status) {
2132 case 0:
2133 break;
2134 case -ETIMEDOUT:
2135 if (clnt->cl_softrtry)
2136 break;
2137 case -NFS4ERR_DELAY:
2138 case -EAGAIN:
2139 ssleep(1);
2140 case -NFS4ERR_STALE_CLIENTID:
2141 dprintk("NFS: %s after status %d, retrying\n",
2142 __func__, status);
2143 goto again;
2144 case -EACCES:
2145 if (i++ == 0) {
2146 nfs4_root_machine_cred(clp);
2147 goto again;
2148 }
2149 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX)
2150 break;
2151 case -NFS4ERR_CLID_INUSE:
2152 case -NFS4ERR_WRONGSEC:
2153 /* No point in retrying if we already used RPC_AUTH_UNIX */
2154 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) {
2155 status = -EPERM;
2156 break;
2157 }
2158 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX);
2159 if (IS_ERR(clnt)) {
2160 status = PTR_ERR(clnt);
2161 break;
2162 }
2163 /* Note: this is safe because we haven't yet marked the
2164 * client as ready, so we are the only user of
2165 * clp->cl_rpcclient
2166 */
2167 clnt = xchg(&clp->cl_rpcclient, clnt);
2168 rpc_shutdown_client(clnt);
2169 clnt = clp->cl_rpcclient;
2170 goto again;
2171
2172 case -NFS4ERR_MINOR_VERS_MISMATCH:
2173 status = -EPROTONOSUPPORT;
2174 break;
2175
2176 case -EKEYEXPIRED:
2177 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery
2178 * in nfs4_exchange_id */
2179 status = -EKEYEXPIRED;
2180 break;
2181 default:
2182 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n",
2183 __func__, status);
2184 status = -EIO;
2185 }
2186
2187out_unlock:
2188 mutex_unlock(&nfs_clid_init_mutex);
2189 dprintk("NFS: %s: status = %d\n", __func__, status);
2190 return status;
2191}
2192
2193#ifdef CONFIG_NFS_V4_1
2194void nfs4_schedule_session_recovery(struct nfs4_session *session, int err)
2195{
2196 struct nfs_client *clp = session->clp;
2197
2198 switch (err) {
2199 default:
2200 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2201 break;
2202 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
2203 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2204 }
2205 nfs4_schedule_state_manager(clp);
2206}
2207EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery);
2208
2209void nfs41_notify_server(struct nfs_client *clp)
2210{
2211 /* Use CHECK_LEASE to ping the server with a SEQUENCE */
2212 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
2213 nfs4_schedule_state_manager(clp);
2214}
2215
2216static void nfs4_reset_all_state(struct nfs_client *clp)
2217{
2218 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2219 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state);
2220 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state);
2221 nfs4_state_start_reclaim_nograce(clp);
2222 dprintk("%s: scheduling reset of all state for server %s!\n",
2223 __func__, clp->cl_hostname);
2224 nfs4_schedule_state_manager(clp);
2225 }
2226}
2227
2228static void nfs41_handle_server_reboot(struct nfs_client *clp)
2229{
2230 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) {
2231 nfs4_state_start_reclaim_reboot(clp);
2232 dprintk("%s: server %s rebooted!\n", __func__,
2233 clp->cl_hostname);
2234 nfs4_schedule_state_manager(clp);
2235 }
2236}
2237
2238static void nfs41_handle_all_state_revoked(struct nfs_client *clp)
2239{
2240 nfs4_reset_all_state(clp);
2241 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2242}
2243
2244static void nfs41_handle_some_state_revoked(struct nfs_client *clp)
2245{
2246 nfs4_state_start_reclaim_nograce(clp);
2247 nfs4_schedule_state_manager(clp);
2248
2249 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname);
2250}
2251
2252static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp)
2253{
2254 /* FIXME: For now, we destroy all layouts. */
2255 pnfs_destroy_all_layouts(clp);
2256 /* FIXME: For now, we test all delegations+open state+locks. */
2257 nfs41_handle_some_state_revoked(clp);
2258 dprintk("%s: Recallable state revoked on server %s!\n", __func__,
2259 clp->cl_hostname);
2260}
2261
2262static void nfs41_handle_backchannel_fault(struct nfs_client *clp)
2263{
2264 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2265 nfs4_schedule_state_manager(clp);
2266
2267 dprintk("%s: server %s declared a backchannel fault\n", __func__,
2268 clp->cl_hostname);
2269}
2270
2271static void nfs41_handle_cb_path_down(struct nfs_client *clp)
2272{
2273 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2274 &clp->cl_state) == 0)
2275 nfs4_schedule_state_manager(clp);
2276}
2277
2278void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags,
2279 bool recovery)
2280{
2281 if (!flags)
2282 return;
2283
2284 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n",
2285 __func__, clp->cl_hostname, clp->cl_clientid, flags);
2286 /*
2287 * If we're called from the state manager thread, then assume we're
2288 * already handling the RECLAIM_NEEDED and/or STATE_REVOKED.
2289 * Those flags are expected to remain set until we're done
2290 * recovering (see RFC5661, section 18.46.3).
2291 */
2292 if (recovery)
2293 goto out_recovery;
2294
2295 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED)
2296 nfs41_handle_server_reboot(clp);
2297 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED))
2298 nfs41_handle_all_state_revoked(clp);
2299 if (flags & (SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED |
2300 SEQ4_STATUS_ADMIN_STATE_REVOKED))
2301 nfs41_handle_some_state_revoked(clp);
2302 if (flags & SEQ4_STATUS_LEASE_MOVED)
2303 nfs4_schedule_lease_moved_recovery(clp);
2304 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED)
2305 nfs41_handle_recallable_state_revoked(clp);
2306out_recovery:
2307 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT)
2308 nfs41_handle_backchannel_fault(clp);
2309 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN |
2310 SEQ4_STATUS_CB_PATH_DOWN_SESSION))
2311 nfs41_handle_cb_path_down(clp);
2312}
2313
2314static int nfs4_reset_session(struct nfs_client *clp)
2315{
2316 struct rpc_cred *cred;
2317 int status;
2318
2319 if (!nfs4_has_session(clp))
2320 return 0;
2321 nfs4_begin_drain_session(clp);
2322 cred = nfs4_get_clid_cred(clp);
2323 status = nfs4_proc_destroy_session(clp->cl_session, cred);
2324 switch (status) {
2325 case 0:
2326 case -NFS4ERR_BADSESSION:
2327 case -NFS4ERR_DEADSESSION:
2328 break;
2329 case -NFS4ERR_BACK_CHAN_BUSY:
2330 case -NFS4ERR_DELAY:
2331 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state);
2332 status = 0;
2333 ssleep(1);
2334 goto out;
2335 default:
2336 status = nfs4_recovery_handle_error(clp, status);
2337 goto out;
2338 }
2339
2340 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN);
2341 status = nfs4_proc_create_session(clp, cred);
2342 if (status) {
2343 dprintk("%s: session reset failed with status %d for server %s!\n",
2344 __func__, status, clp->cl_hostname);
2345 status = nfs4_handle_reclaim_lease_error(clp, status);
2346 goto out;
2347 }
2348 nfs41_finish_session_reset(clp);
2349 dprintk("%s: session reset was successful for server %s!\n",
2350 __func__, clp->cl_hostname);
2351out:
2352 if (cred)
2353 put_rpccred(cred);
2354 return status;
2355}
2356
2357static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2358{
2359 struct rpc_cred *cred;
2360 int ret;
2361
2362 if (!nfs4_has_session(clp))
2363 return 0;
2364 nfs4_begin_drain_session(clp);
2365 cred = nfs4_get_clid_cred(clp);
2366 ret = nfs4_proc_bind_conn_to_session(clp, cred);
2367 if (cred)
2368 put_rpccred(cred);
2369 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2370 switch (ret) {
2371 case 0:
2372 dprintk("%s: bind_conn_to_session was successful for server %s!\n",
2373 __func__, clp->cl_hostname);
2374 break;
2375 case -NFS4ERR_DELAY:
2376 ssleep(1);
2377 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state);
2378 break;
2379 default:
2380 return nfs4_recovery_handle_error(clp, ret);
2381 }
2382 return 0;
2383}
2384#else /* CONFIG_NFS_V4_1 */
2385static int nfs4_reset_session(struct nfs_client *clp) { return 0; }
2386
2387static int nfs4_bind_conn_to_session(struct nfs_client *clp)
2388{
2389 return 0;
2390}
2391#endif /* CONFIG_NFS_V4_1 */
2392
2393static void nfs4_state_manager(struct nfs_client *clp)
2394{
2395 int status = 0;
2396 const char *section = "", *section_sep = "";
2397
2398 /* Ensure exclusive access to NFSv4 state */
2399 do {
2400 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) {
2401 section = "purge state";
2402 status = nfs4_purge_lease(clp);
2403 if (status < 0)
2404 goto out_error;
2405 continue;
2406 }
2407
2408 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) {
2409 section = "lease expired";
2410 /* We're going to have to re-establish a clientid */
2411 status = nfs4_reclaim_lease(clp);
2412 if (status < 0)
2413 goto out_error;
2414 continue;
2415 }
2416
2417 /* Initialize or reset the session */
2418 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) {
2419 section = "reset session";
2420 status = nfs4_reset_session(clp);
2421 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state))
2422 continue;
2423 if (status < 0)
2424 goto out_error;
2425 }
2426
2427 /* Send BIND_CONN_TO_SESSION */
2428 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION,
2429 &clp->cl_state)) {
2430 section = "bind conn to session";
2431 status = nfs4_bind_conn_to_session(clp);
2432 if (status < 0)
2433 goto out_error;
2434 continue;
2435 }
2436
2437 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
2438 section = "check lease";
2439 status = nfs4_check_lease(clp);
2440 if (status < 0)
2441 goto out_error;
2442 continue;
2443 }
2444
2445 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) {
2446 section = "migration";
2447 status = nfs4_handle_migration(clp);
2448 if (status < 0)
2449 goto out_error;
2450 }
2451
2452 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) {
2453 section = "lease moved";
2454 status = nfs4_handle_lease_moved(clp);
2455 if (status < 0)
2456 goto out_error;
2457 }
2458
2459 /* First recover reboot state... */
2460 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) {
2461 section = "reclaim reboot";
2462 status = nfs4_do_reclaim(clp,
2463 clp->cl_mvops->reboot_recovery_ops);
2464 if (status == -EAGAIN)
2465 continue;
2466 if (status < 0)
2467 goto out_error;
2468 nfs4_state_end_reclaim_reboot(clp);
2469 }
2470
2471 /* Detect expired delegations... */
2472 if (test_and_clear_bit(NFS4CLNT_DELEGATION_EXPIRED, &clp->cl_state)) {
2473 section = "detect expired delegations";
2474 nfs_reap_expired_delegations(clp);
2475 continue;
2476 }
2477
2478 /* Now recover expired state... */
2479 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) {
2480 section = "reclaim nograce";
2481 status = nfs4_do_reclaim(clp,
2482 clp->cl_mvops->nograce_recovery_ops);
2483 if (status == -EAGAIN)
2484 continue;
2485 if (status < 0)
2486 goto out_error;
2487 }
2488
2489 nfs4_end_drain_session(clp);
2490 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) {
2491 nfs_client_return_marked_delegations(clp);
2492 continue;
2493 }
2494
2495 nfs4_clear_state_manager_bit(clp);
2496 /* Did we race with an attempt to give us more work? */
2497 if (clp->cl_state == 0)
2498 break;
2499 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0)
2500 break;
2501 } while (atomic_read(&clp->cl_count) > 1);
2502 return;
2503out_error:
2504 if (strlen(section))
2505 section_sep = ": ";
2506 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s"
2507 " with error %d\n", section_sep, section,
2508 clp->cl_hostname, -status);
2509 ssleep(1);
2510 nfs4_end_drain_session(clp);
2511 nfs4_clear_state_manager_bit(clp);
2512}
2513
2514static int nfs4_run_state_manager(void *ptr)
2515{
2516 struct nfs_client *clp = ptr;
2517
2518 allow_signal(SIGKILL);
2519 nfs4_state_manager(clp);
2520 nfs_put_client(clp);
2521 module_put_and_exit(0);
2522 return 0;
2523}
2524
2525/*
2526 * Local variables:
2527 * c-basic-offset: 8
2528 * End:
2529 */