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 #include <linux/jiffies.h> 53 54 #include <linux/sunrpc/clnt.h> 55 56 #include "nfs4_fs.h" 57 #include "callback.h" 58 #include "delegation.h" 59 #include "internal.h" 60 #include "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 68 const nfs4_stateid zero_stateid; 69 static DEFINE_MUTEX(nfs_clid_init_mutex); 70 71 int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred) 72 { 73 struct nfs4_setclientid_res clid = { 74 .clientid = clp->cl_clientid, 75 .confirm = clp->cl_confirm, 76 }; 77 unsigned short port; 78 int status; 79 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id); 80 81 if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state)) 82 goto do_confirm; 83 port = nn->nfs_callback_tcpport; 84 if (clp->cl_addr.ss_family == AF_INET6) 85 port = nn->nfs_callback_tcpport6; 86 87 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid); 88 if (status != 0) 89 goto out; 90 clp->cl_clientid = clid.clientid; 91 clp->cl_confirm = clid.confirm; 92 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 93 do_confirm: 94 status = nfs4_proc_setclientid_confirm(clp, &clid, cred); 95 if (status != 0) 96 goto out; 97 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 98 nfs4_schedule_state_renewal(clp); 99 out: 100 return status; 101 } 102 103 /** 104 * nfs40_discover_server_trunking - Detect server IP address trunking (mv0) 105 * 106 * @clp: nfs_client under test 107 * @result: OUT: found nfs_client, or clp 108 * @cred: credential to use for trunking test 109 * 110 * Returns zero, a negative errno, or a negative NFS4ERR status. 111 * If zero is returned, an nfs_client pointer is planted in 112 * "result". 113 * 114 * Note: The returned client may not yet be marked ready. 115 */ 116 int nfs40_discover_server_trunking(struct nfs_client *clp, 117 struct nfs_client **result, 118 struct rpc_cred *cred) 119 { 120 struct nfs4_setclientid_res clid = { 121 .clientid = clp->cl_clientid, 122 .confirm = clp->cl_confirm, 123 }; 124 struct nfs_net *nn = net_generic(clp->cl_net, nfs_net_id); 125 unsigned short port; 126 int status; 127 128 port = nn->nfs_callback_tcpport; 129 if (clp->cl_addr.ss_family == AF_INET6) 130 port = nn->nfs_callback_tcpport6; 131 132 status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid); 133 if (status != 0) 134 goto out; 135 clp->cl_clientid = clid.clientid; 136 clp->cl_confirm = clid.confirm; 137 138 status = nfs40_walk_client_list(clp, result, cred); 139 if (status == 0) { 140 /* Sustain the lease, even if it's empty. If the clientid4 141 * goes stale it's of no use for trunking discovery. */ 142 nfs4_schedule_state_renewal(*result); 143 } 144 out: 145 return status; 146 } 147 148 struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp) 149 { 150 struct rpc_cred *cred = NULL; 151 152 if (clp->cl_machine_cred != NULL) 153 cred = get_rpccred(clp->cl_machine_cred); 154 return cred; 155 } 156 157 static void nfs4_root_machine_cred(struct nfs_client *clp) 158 { 159 struct rpc_cred *cred, *new; 160 161 new = rpc_lookup_machine_cred(NULL); 162 spin_lock(&clp->cl_lock); 163 cred = clp->cl_machine_cred; 164 clp->cl_machine_cred = new; 165 spin_unlock(&clp->cl_lock); 166 if (cred != NULL) 167 put_rpccred(cred); 168 } 169 170 static struct rpc_cred * 171 nfs4_get_renew_cred_server_locked(struct nfs_server *server) 172 { 173 struct rpc_cred *cred = NULL; 174 struct nfs4_state_owner *sp; 175 struct rb_node *pos; 176 177 for (pos = rb_first(&server->state_owners); 178 pos != NULL; 179 pos = rb_next(pos)) { 180 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); 181 if (list_empty(&sp->so_states)) 182 continue; 183 cred = get_rpccred(sp->so_cred); 184 break; 185 } 186 return cred; 187 } 188 189 /** 190 * nfs4_get_renew_cred_locked - Acquire credential for a renew operation 191 * @clp: client state handle 192 * 193 * Returns an rpc_cred with reference count bumped, or NULL. 194 * Caller must hold clp->cl_lock. 195 */ 196 struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp) 197 { 198 struct rpc_cred *cred = NULL; 199 struct nfs_server *server; 200 201 /* Use machine credentials if available */ 202 cred = nfs4_get_machine_cred_locked(clp); 203 if (cred != NULL) 204 goto out; 205 206 rcu_read_lock(); 207 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { 208 cred = nfs4_get_renew_cred_server_locked(server); 209 if (cred != NULL) 210 break; 211 } 212 rcu_read_unlock(); 213 214 out: 215 return cred; 216 } 217 218 static void nfs4_end_drain_slot_table(struct nfs4_slot_table *tbl) 219 { 220 if (test_and_clear_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state)) { 221 spin_lock(&tbl->slot_tbl_lock); 222 nfs41_wake_slot_table(tbl); 223 spin_unlock(&tbl->slot_tbl_lock); 224 } 225 } 226 227 static void nfs4_end_drain_session(struct nfs_client *clp) 228 { 229 struct nfs4_session *ses = clp->cl_session; 230 231 if (clp->cl_slot_tbl) { 232 nfs4_end_drain_slot_table(clp->cl_slot_tbl); 233 return; 234 } 235 236 if (ses != NULL) { 237 nfs4_end_drain_slot_table(&ses->bc_slot_table); 238 nfs4_end_drain_slot_table(&ses->fc_slot_table); 239 } 240 } 241 242 static int nfs4_drain_slot_tbl(struct nfs4_slot_table *tbl) 243 { 244 set_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state); 245 spin_lock(&tbl->slot_tbl_lock); 246 if (tbl->highest_used_slotid != NFS4_NO_SLOT) { 247 reinit_completion(&tbl->complete); 248 spin_unlock(&tbl->slot_tbl_lock); 249 return wait_for_completion_interruptible(&tbl->complete); 250 } 251 spin_unlock(&tbl->slot_tbl_lock); 252 return 0; 253 } 254 255 static int nfs4_begin_drain_session(struct nfs_client *clp) 256 { 257 struct nfs4_session *ses = clp->cl_session; 258 int ret = 0; 259 260 if (clp->cl_slot_tbl) 261 return nfs4_drain_slot_tbl(clp->cl_slot_tbl); 262 263 /* back channel */ 264 ret = nfs4_drain_slot_tbl(&ses->bc_slot_table); 265 if (ret) 266 return ret; 267 /* fore channel */ 268 return nfs4_drain_slot_tbl(&ses->fc_slot_table); 269 } 270 271 #if defined(CONFIG_NFS_V4_1) 272 273 static int nfs41_setup_state_renewal(struct nfs_client *clp) 274 { 275 int status; 276 struct nfs_fsinfo fsinfo; 277 278 if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) { 279 nfs4_schedule_state_renewal(clp); 280 return 0; 281 } 282 283 status = nfs4_proc_get_lease_time(clp, &fsinfo); 284 if (status == 0) { 285 /* Update lease time and schedule renewal */ 286 spin_lock(&clp->cl_lock); 287 clp->cl_lease_time = fsinfo.lease_time * HZ; 288 clp->cl_last_renewal = jiffies; 289 spin_unlock(&clp->cl_lock); 290 291 nfs4_schedule_state_renewal(clp); 292 } 293 294 return status; 295 } 296 297 static 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 } 305 306 int 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 nfs4_begin_drain_session(clp); 313 status = nfs4_proc_exchange_id(clp, cred); 314 if (status != 0) 315 goto out; 316 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 317 do_confirm: 318 status = nfs4_proc_create_session(clp, cred); 319 if (status != 0) 320 goto out; 321 nfs41_finish_session_reset(clp); 322 nfs_mark_client_ready(clp, NFS_CS_READY); 323 out: 324 return status; 325 } 326 327 /** 328 * nfs41_discover_server_trunking - Detect server IP address trunking (mv1) 329 * 330 * @clp: nfs_client under test 331 * @result: OUT: found nfs_client, or clp 332 * @cred: credential to use for trunking test 333 * 334 * Returns NFS4_OK, a negative errno, or a negative NFS4ERR status. 335 * If NFS4_OK is returned, an nfs_client pointer is planted in 336 * "result". 337 * 338 * Note: The returned client may not yet be marked ready. 339 */ 340 int nfs41_discover_server_trunking(struct nfs_client *clp, 341 struct nfs_client **result, 342 struct rpc_cred *cred) 343 { 344 int status; 345 346 status = nfs4_proc_exchange_id(clp, cred); 347 if (status != NFS4_OK) 348 return status; 349 set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 350 351 return nfs41_walk_client_list(clp, result, cred); 352 } 353 354 #endif /* CONFIG_NFS_V4_1 */ 355 356 /** 357 * nfs4_get_clid_cred - Acquire credential for a setclientid operation 358 * @clp: client state handle 359 * 360 * Returns an rpc_cred with reference count bumped, or NULL. 361 */ 362 struct rpc_cred *nfs4_get_clid_cred(struct nfs_client *clp) 363 { 364 struct rpc_cred *cred; 365 366 spin_lock(&clp->cl_lock); 367 cred = nfs4_get_machine_cred_locked(clp); 368 spin_unlock(&clp->cl_lock); 369 return cred; 370 } 371 372 static struct nfs4_state_owner * 373 nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred) 374 { 375 struct rb_node **p = &server->state_owners.rb_node, 376 *parent = NULL; 377 struct nfs4_state_owner *sp; 378 379 while (*p != NULL) { 380 parent = *p; 381 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node); 382 383 if (cred < sp->so_cred) 384 p = &parent->rb_left; 385 else if (cred > sp->so_cred) 386 p = &parent->rb_right; 387 else { 388 if (!list_empty(&sp->so_lru)) 389 list_del_init(&sp->so_lru); 390 atomic_inc(&sp->so_count); 391 return sp; 392 } 393 } 394 return NULL; 395 } 396 397 static struct nfs4_state_owner * 398 nfs4_insert_state_owner_locked(struct nfs4_state_owner *new) 399 { 400 struct nfs_server *server = new->so_server; 401 struct rb_node **p = &server->state_owners.rb_node, 402 *parent = NULL; 403 struct nfs4_state_owner *sp; 404 int err; 405 406 while (*p != NULL) { 407 parent = *p; 408 sp = rb_entry(parent, struct nfs4_state_owner, so_server_node); 409 410 if (new->so_cred < sp->so_cred) 411 p = &parent->rb_left; 412 else if (new->so_cred > sp->so_cred) 413 p = &parent->rb_right; 414 else { 415 if (!list_empty(&sp->so_lru)) 416 list_del_init(&sp->so_lru); 417 atomic_inc(&sp->so_count); 418 return sp; 419 } 420 } 421 err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id); 422 if (err) 423 return ERR_PTR(err); 424 rb_link_node(&new->so_server_node, parent, p); 425 rb_insert_color(&new->so_server_node, &server->state_owners); 426 return new; 427 } 428 429 static void 430 nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp) 431 { 432 struct nfs_server *server = sp->so_server; 433 434 if (!RB_EMPTY_NODE(&sp->so_server_node)) 435 rb_erase(&sp->so_server_node, &server->state_owners); 436 ida_remove(&server->openowner_id, sp->so_seqid.owner_id); 437 } 438 439 static void 440 nfs4_init_seqid_counter(struct nfs_seqid_counter *sc) 441 { 442 sc->create_time = ktime_get(); 443 sc->flags = 0; 444 sc->counter = 0; 445 spin_lock_init(&sc->lock); 446 INIT_LIST_HEAD(&sc->list); 447 rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue"); 448 } 449 450 static void 451 nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc) 452 { 453 rpc_destroy_wait_queue(&sc->wait); 454 } 455 456 /* 457 * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to 458 * create a new state_owner. 459 * 460 */ 461 static struct nfs4_state_owner * 462 nfs4_alloc_state_owner(struct nfs_server *server, 463 struct rpc_cred *cred, 464 gfp_t gfp_flags) 465 { 466 struct nfs4_state_owner *sp; 467 468 sp = kzalloc(sizeof(*sp), gfp_flags); 469 if (!sp) 470 return NULL; 471 sp->so_server = server; 472 sp->so_cred = get_rpccred(cred); 473 spin_lock_init(&sp->so_lock); 474 INIT_LIST_HEAD(&sp->so_states); 475 nfs4_init_seqid_counter(&sp->so_seqid); 476 atomic_set(&sp->so_count, 1); 477 INIT_LIST_HEAD(&sp->so_lru); 478 seqcount_init(&sp->so_reclaim_seqcount); 479 mutex_init(&sp->so_delegreturn_mutex); 480 return sp; 481 } 482 483 static void 484 nfs4_drop_state_owner(struct nfs4_state_owner *sp) 485 { 486 struct rb_node *rb_node = &sp->so_server_node; 487 488 if (!RB_EMPTY_NODE(rb_node)) { 489 struct nfs_server *server = sp->so_server; 490 struct nfs_client *clp = server->nfs_client; 491 492 spin_lock(&clp->cl_lock); 493 if (!RB_EMPTY_NODE(rb_node)) { 494 rb_erase(rb_node, &server->state_owners); 495 RB_CLEAR_NODE(rb_node); 496 } 497 spin_unlock(&clp->cl_lock); 498 } 499 } 500 501 static void nfs4_free_state_owner(struct nfs4_state_owner *sp) 502 { 503 nfs4_destroy_seqid_counter(&sp->so_seqid); 504 put_rpccred(sp->so_cred); 505 kfree(sp); 506 } 507 508 static void nfs4_gc_state_owners(struct nfs_server *server) 509 { 510 struct nfs_client *clp = server->nfs_client; 511 struct nfs4_state_owner *sp, *tmp; 512 unsigned long time_min, time_max; 513 LIST_HEAD(doomed); 514 515 spin_lock(&clp->cl_lock); 516 time_max = jiffies; 517 time_min = (long)time_max - (long)clp->cl_lease_time; 518 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) { 519 /* NB: LRU is sorted so that oldest is at the head */ 520 if (time_in_range(sp->so_expires, time_min, time_max)) 521 break; 522 list_move(&sp->so_lru, &doomed); 523 nfs4_remove_state_owner_locked(sp); 524 } 525 spin_unlock(&clp->cl_lock); 526 527 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) { 528 list_del(&sp->so_lru); 529 nfs4_free_state_owner(sp); 530 } 531 } 532 533 /** 534 * nfs4_get_state_owner - Look up a state owner given a credential 535 * @server: nfs_server to search 536 * @cred: RPC credential to match 537 * 538 * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL. 539 */ 540 struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, 541 struct rpc_cred *cred, 542 gfp_t gfp_flags) 543 { 544 struct nfs_client *clp = server->nfs_client; 545 struct nfs4_state_owner *sp, *new; 546 547 spin_lock(&clp->cl_lock); 548 sp = nfs4_find_state_owner_locked(server, cred); 549 spin_unlock(&clp->cl_lock); 550 if (sp != NULL) 551 goto out; 552 new = nfs4_alloc_state_owner(server, cred, gfp_flags); 553 if (new == NULL) 554 goto out; 555 do { 556 if (ida_pre_get(&server->openowner_id, gfp_flags) == 0) 557 break; 558 spin_lock(&clp->cl_lock); 559 sp = nfs4_insert_state_owner_locked(new); 560 spin_unlock(&clp->cl_lock); 561 } while (sp == ERR_PTR(-EAGAIN)); 562 if (sp != new) 563 nfs4_free_state_owner(new); 564 out: 565 nfs4_gc_state_owners(server); 566 return sp; 567 } 568 569 /** 570 * nfs4_put_state_owner - Release a nfs4_state_owner 571 * @sp: state owner data to release 572 * 573 * Note that we keep released state owners on an LRU 574 * list. 575 * This caches valid state owners so that they can be 576 * reused, to avoid the OPEN_CONFIRM on minor version 0. 577 * It also pins the uniquifier of dropped state owners for 578 * a while, to ensure that those state owner names are 579 * never reused. 580 */ 581 void nfs4_put_state_owner(struct nfs4_state_owner *sp) 582 { 583 struct nfs_server *server = sp->so_server; 584 struct nfs_client *clp = server->nfs_client; 585 586 if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock)) 587 return; 588 589 sp->so_expires = jiffies; 590 list_add_tail(&sp->so_lru, &server->state_owners_lru); 591 spin_unlock(&clp->cl_lock); 592 } 593 594 /** 595 * nfs4_purge_state_owners - Release all cached state owners 596 * @server: nfs_server with cached state owners to release 597 * 598 * Called at umount time. Remaining state owners will be on 599 * the LRU with ref count of zero. 600 */ 601 void nfs4_purge_state_owners(struct nfs_server *server) 602 { 603 struct nfs_client *clp = server->nfs_client; 604 struct nfs4_state_owner *sp, *tmp; 605 LIST_HEAD(doomed); 606 607 spin_lock(&clp->cl_lock); 608 list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) { 609 list_move(&sp->so_lru, &doomed); 610 nfs4_remove_state_owner_locked(sp); 611 } 612 spin_unlock(&clp->cl_lock); 613 614 list_for_each_entry_safe(sp, tmp, &doomed, so_lru) { 615 list_del(&sp->so_lru); 616 nfs4_free_state_owner(sp); 617 } 618 } 619 620 static struct nfs4_state * 621 nfs4_alloc_open_state(void) 622 { 623 struct nfs4_state *state; 624 625 state = kzalloc(sizeof(*state), GFP_NOFS); 626 if (!state) 627 return NULL; 628 atomic_set(&state->count, 1); 629 INIT_LIST_HEAD(&state->lock_states); 630 spin_lock_init(&state->state_lock); 631 seqlock_init(&state->seqlock); 632 return state; 633 } 634 635 void 636 nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode) 637 { 638 if (state->state == fmode) 639 return; 640 /* NB! List reordering - see the reclaim code for why. */ 641 if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) { 642 if (fmode & FMODE_WRITE) 643 list_move(&state->open_states, &state->owner->so_states); 644 else 645 list_move_tail(&state->open_states, &state->owner->so_states); 646 } 647 state->state = fmode; 648 } 649 650 static struct nfs4_state * 651 __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner) 652 { 653 struct nfs_inode *nfsi = NFS_I(inode); 654 struct nfs4_state *state; 655 656 list_for_each_entry(state, &nfsi->open_states, inode_states) { 657 if (state->owner != owner) 658 continue; 659 if (!nfs4_valid_open_stateid(state)) 660 continue; 661 if (atomic_inc_not_zero(&state->count)) 662 return state; 663 } 664 return NULL; 665 } 666 667 static void 668 nfs4_free_open_state(struct nfs4_state *state) 669 { 670 kfree(state); 671 } 672 673 struct nfs4_state * 674 nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner) 675 { 676 struct nfs4_state *state, *new; 677 struct nfs_inode *nfsi = NFS_I(inode); 678 679 spin_lock(&inode->i_lock); 680 state = __nfs4_find_state_byowner(inode, owner); 681 spin_unlock(&inode->i_lock); 682 if (state) 683 goto out; 684 new = nfs4_alloc_open_state(); 685 spin_lock(&owner->so_lock); 686 spin_lock(&inode->i_lock); 687 state = __nfs4_find_state_byowner(inode, owner); 688 if (state == NULL && new != NULL) { 689 state = new; 690 state->owner = owner; 691 atomic_inc(&owner->so_count); 692 list_add(&state->inode_states, &nfsi->open_states); 693 ihold(inode); 694 state->inode = inode; 695 spin_unlock(&inode->i_lock); 696 /* Note: The reclaim code dictates that we add stateless 697 * and read-only stateids to the end of the list */ 698 list_add_tail(&state->open_states, &owner->so_states); 699 spin_unlock(&owner->so_lock); 700 } else { 701 spin_unlock(&inode->i_lock); 702 spin_unlock(&owner->so_lock); 703 if (new) 704 nfs4_free_open_state(new); 705 } 706 out: 707 return state; 708 } 709 710 void nfs4_put_open_state(struct nfs4_state *state) 711 { 712 struct inode *inode = state->inode; 713 struct nfs4_state_owner *owner = state->owner; 714 715 if (!atomic_dec_and_lock(&state->count, &owner->so_lock)) 716 return; 717 spin_lock(&inode->i_lock); 718 list_del(&state->inode_states); 719 list_del(&state->open_states); 720 spin_unlock(&inode->i_lock); 721 spin_unlock(&owner->so_lock); 722 iput(inode); 723 nfs4_free_open_state(state); 724 nfs4_put_state_owner(owner); 725 } 726 727 /* 728 * Close the current file. 729 */ 730 static void __nfs4_close(struct nfs4_state *state, 731 fmode_t fmode, gfp_t gfp_mask, int wait) 732 { 733 struct nfs4_state_owner *owner = state->owner; 734 int call_close = 0; 735 fmode_t newstate; 736 737 atomic_inc(&owner->so_count); 738 /* Protect against nfs4_find_state() */ 739 spin_lock(&owner->so_lock); 740 switch (fmode & (FMODE_READ | FMODE_WRITE)) { 741 case FMODE_READ: 742 state->n_rdonly--; 743 break; 744 case FMODE_WRITE: 745 state->n_wronly--; 746 break; 747 case FMODE_READ|FMODE_WRITE: 748 state->n_rdwr--; 749 } 750 newstate = FMODE_READ|FMODE_WRITE; 751 if (state->n_rdwr == 0) { 752 if (state->n_rdonly == 0) { 753 newstate &= ~FMODE_READ; 754 call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags); 755 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); 756 } 757 if (state->n_wronly == 0) { 758 newstate &= ~FMODE_WRITE; 759 call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags); 760 call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); 761 } 762 if (newstate == 0) 763 clear_bit(NFS_DELEGATED_STATE, &state->flags); 764 } 765 nfs4_state_set_mode_locked(state, newstate); 766 spin_unlock(&owner->so_lock); 767 768 if (!call_close) { 769 nfs4_put_open_state(state); 770 nfs4_put_state_owner(owner); 771 } else 772 nfs4_do_close(state, gfp_mask, wait); 773 } 774 775 void nfs4_close_state(struct nfs4_state *state, fmode_t fmode) 776 { 777 __nfs4_close(state, fmode, GFP_NOFS, 0); 778 } 779 780 void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode) 781 { 782 __nfs4_close(state, fmode, GFP_KERNEL, 1); 783 } 784 785 /* 786 * Search the state->lock_states for an existing lock_owner 787 * that is compatible with current->files 788 */ 789 static struct nfs4_lock_state * 790 __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type) 791 { 792 struct nfs4_lock_state *pos; 793 list_for_each_entry(pos, &state->lock_states, ls_locks) { 794 if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type) 795 continue; 796 switch (pos->ls_owner.lo_type) { 797 case NFS4_POSIX_LOCK_TYPE: 798 if (pos->ls_owner.lo_u.posix_owner != fl_owner) 799 continue; 800 break; 801 case NFS4_FLOCK_LOCK_TYPE: 802 if (pos->ls_owner.lo_u.flock_owner != fl_pid) 803 continue; 804 } 805 atomic_inc(&pos->ls_count); 806 return pos; 807 } 808 return NULL; 809 } 810 811 /* 812 * Return a compatible lock_state. If no initialized lock_state structure 813 * exists, return an uninitialized one. 814 * 815 */ 816 static struct nfs4_lock_state *nfs4_alloc_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type) 817 { 818 struct nfs4_lock_state *lsp; 819 struct nfs_server *server = state->owner->so_server; 820 821 lsp = kzalloc(sizeof(*lsp), GFP_NOFS); 822 if (lsp == NULL) 823 return NULL; 824 nfs4_init_seqid_counter(&lsp->ls_seqid); 825 atomic_set(&lsp->ls_count, 1); 826 lsp->ls_state = state; 827 lsp->ls_owner.lo_type = type; 828 switch (lsp->ls_owner.lo_type) { 829 case NFS4_FLOCK_LOCK_TYPE: 830 lsp->ls_owner.lo_u.flock_owner = fl_pid; 831 break; 832 case NFS4_POSIX_LOCK_TYPE: 833 lsp->ls_owner.lo_u.posix_owner = fl_owner; 834 break; 835 default: 836 goto out_free; 837 } 838 lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS); 839 if (lsp->ls_seqid.owner_id < 0) 840 goto out_free; 841 INIT_LIST_HEAD(&lsp->ls_locks); 842 return lsp; 843 out_free: 844 kfree(lsp); 845 return NULL; 846 } 847 848 void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp) 849 { 850 ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id); 851 nfs4_destroy_seqid_counter(&lsp->ls_seqid); 852 kfree(lsp); 853 } 854 855 /* 856 * Return a compatible lock_state. If no initialized lock_state structure 857 * exists, return an uninitialized one. 858 * 859 */ 860 static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type) 861 { 862 struct nfs4_lock_state *lsp, *new = NULL; 863 864 for(;;) { 865 spin_lock(&state->state_lock); 866 lsp = __nfs4_find_lock_state(state, owner, pid, type); 867 if (lsp != NULL) 868 break; 869 if (new != NULL) { 870 list_add(&new->ls_locks, &state->lock_states); 871 set_bit(LK_STATE_IN_USE, &state->flags); 872 lsp = new; 873 new = NULL; 874 break; 875 } 876 spin_unlock(&state->state_lock); 877 new = nfs4_alloc_lock_state(state, owner, pid, type); 878 if (new == NULL) 879 return NULL; 880 } 881 spin_unlock(&state->state_lock); 882 if (new != NULL) 883 nfs4_free_lock_state(state->owner->so_server, new); 884 return lsp; 885 } 886 887 /* 888 * Release reference to lock_state, and free it if we see that 889 * it is no longer in use 890 */ 891 void nfs4_put_lock_state(struct nfs4_lock_state *lsp) 892 { 893 struct nfs_server *server; 894 struct nfs4_state *state; 895 896 if (lsp == NULL) 897 return; 898 state = lsp->ls_state; 899 if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock)) 900 return; 901 list_del(&lsp->ls_locks); 902 if (list_empty(&state->lock_states)) 903 clear_bit(LK_STATE_IN_USE, &state->flags); 904 spin_unlock(&state->state_lock); 905 server = state->owner->so_server; 906 if (test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags)) { 907 struct nfs_client *clp = server->nfs_client; 908 909 clp->cl_mvops->free_lock_state(server, lsp); 910 } else 911 nfs4_free_lock_state(server, lsp); 912 } 913 914 static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src) 915 { 916 struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner; 917 918 dst->fl_u.nfs4_fl.owner = lsp; 919 atomic_inc(&lsp->ls_count); 920 } 921 922 static void nfs4_fl_release_lock(struct file_lock *fl) 923 { 924 nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner); 925 } 926 927 static const struct file_lock_operations nfs4_fl_lock_ops = { 928 .fl_copy_lock = nfs4_fl_copy_lock, 929 .fl_release_private = nfs4_fl_release_lock, 930 }; 931 932 int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl) 933 { 934 struct nfs4_lock_state *lsp; 935 936 if (fl->fl_ops != NULL) 937 return 0; 938 if (fl->fl_flags & FL_POSIX) 939 lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE); 940 else if (fl->fl_flags & FL_FLOCK) 941 lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid, 942 NFS4_FLOCK_LOCK_TYPE); 943 else 944 return -EINVAL; 945 if (lsp == NULL) 946 return -ENOMEM; 947 fl->fl_u.nfs4_fl.owner = lsp; 948 fl->fl_ops = &nfs4_fl_lock_ops; 949 return 0; 950 } 951 952 static int nfs4_copy_lock_stateid(nfs4_stateid *dst, 953 struct nfs4_state *state, 954 const struct nfs_lockowner *lockowner) 955 { 956 struct nfs4_lock_state *lsp; 957 fl_owner_t fl_owner; 958 pid_t fl_pid; 959 int ret = -ENOENT; 960 961 962 if (lockowner == NULL) 963 goto out; 964 965 if (test_bit(LK_STATE_IN_USE, &state->flags) == 0) 966 goto out; 967 968 fl_owner = lockowner->l_owner; 969 fl_pid = lockowner->l_pid; 970 spin_lock(&state->state_lock); 971 lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE); 972 if (lsp && test_bit(NFS_LOCK_LOST, &lsp->ls_flags)) 973 ret = -EIO; 974 else if (lsp != NULL && test_bit(NFS_LOCK_INITIALIZED, &lsp->ls_flags) != 0) { 975 nfs4_stateid_copy(dst, &lsp->ls_stateid); 976 ret = 0; 977 } 978 spin_unlock(&state->state_lock); 979 nfs4_put_lock_state(lsp); 980 out: 981 return ret; 982 } 983 984 static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state) 985 { 986 const nfs4_stateid *src; 987 int seq; 988 989 do { 990 src = &zero_stateid; 991 seq = read_seqbegin(&state->seqlock); 992 if (test_bit(NFS_OPEN_STATE, &state->flags)) 993 src = &state->open_stateid; 994 nfs4_stateid_copy(dst, src); 995 } while (read_seqretry(&state->seqlock, seq)); 996 } 997 998 /* 999 * Byte-range lock aware utility to initialize the stateid of read/write 1000 * requests. 1001 */ 1002 int nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state, 1003 fmode_t fmode, const struct nfs_lockowner *lockowner) 1004 { 1005 int ret = nfs4_copy_lock_stateid(dst, state, lockowner); 1006 if (ret == -EIO) 1007 /* A lost lock - don't even consider delegations */ 1008 goto out; 1009 /* returns true if delegation stateid found and copied */ 1010 if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) { 1011 ret = 0; 1012 goto out; 1013 } 1014 if (ret != -ENOENT) 1015 /* nfs4_copy_delegation_stateid() didn't over-write 1016 * dst, so it still has the lock stateid which we now 1017 * choose to use. 1018 */ 1019 goto out; 1020 nfs4_copy_open_stateid(dst, state); 1021 ret = 0; 1022 out: 1023 if (nfs_server_capable(state->inode, NFS_CAP_STATEID_NFSV41)) 1024 dst->seqid = 0; 1025 return ret; 1026 } 1027 1028 struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask) 1029 { 1030 struct nfs_seqid *new; 1031 1032 new = kmalloc(sizeof(*new), gfp_mask); 1033 if (new != NULL) { 1034 new->sequence = counter; 1035 INIT_LIST_HEAD(&new->list); 1036 new->task = NULL; 1037 } 1038 return new; 1039 } 1040 1041 void nfs_release_seqid(struct nfs_seqid *seqid) 1042 { 1043 struct nfs_seqid_counter *sequence; 1044 1045 if (list_empty(&seqid->list)) 1046 return; 1047 sequence = seqid->sequence; 1048 spin_lock(&sequence->lock); 1049 list_del_init(&seqid->list); 1050 if (!list_empty(&sequence->list)) { 1051 struct nfs_seqid *next; 1052 1053 next = list_first_entry(&sequence->list, 1054 struct nfs_seqid, list); 1055 rpc_wake_up_queued_task(&sequence->wait, next->task); 1056 } 1057 spin_unlock(&sequence->lock); 1058 } 1059 1060 void nfs_free_seqid(struct nfs_seqid *seqid) 1061 { 1062 nfs_release_seqid(seqid); 1063 kfree(seqid); 1064 } 1065 1066 /* 1067 * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or 1068 * failed with a seqid incrementing error - 1069 * see comments nfs4.h:seqid_mutating_error() 1070 */ 1071 static void nfs_increment_seqid(int status, struct nfs_seqid *seqid) 1072 { 1073 switch (status) { 1074 case 0: 1075 break; 1076 case -NFS4ERR_BAD_SEQID: 1077 if (seqid->sequence->flags & NFS_SEQID_CONFIRMED) 1078 return; 1079 pr_warn_ratelimited("NFS: v4 server returned a bad" 1080 " sequence-id error on an" 1081 " unconfirmed sequence %p!\n", 1082 seqid->sequence); 1083 case -NFS4ERR_STALE_CLIENTID: 1084 case -NFS4ERR_STALE_STATEID: 1085 case -NFS4ERR_BAD_STATEID: 1086 case -NFS4ERR_BADXDR: 1087 case -NFS4ERR_RESOURCE: 1088 case -NFS4ERR_NOFILEHANDLE: 1089 /* Non-seqid mutating errors */ 1090 return; 1091 }; 1092 /* 1093 * Note: no locking needed as we are guaranteed to be first 1094 * on the sequence list 1095 */ 1096 seqid->sequence->counter++; 1097 } 1098 1099 void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid) 1100 { 1101 struct nfs4_state_owner *sp = container_of(seqid->sequence, 1102 struct nfs4_state_owner, so_seqid); 1103 struct nfs_server *server = sp->so_server; 1104 1105 if (status == -NFS4ERR_BAD_SEQID) 1106 nfs4_drop_state_owner(sp); 1107 if (!nfs4_has_session(server->nfs_client)) 1108 nfs_increment_seqid(status, seqid); 1109 } 1110 1111 /* 1112 * Increment the seqid if the LOCK/LOCKU succeeded, or 1113 * failed with a seqid incrementing error - 1114 * see comments nfs4.h:seqid_mutating_error() 1115 */ 1116 void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid) 1117 { 1118 nfs_increment_seqid(status, seqid); 1119 } 1120 1121 int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task) 1122 { 1123 struct nfs_seqid_counter *sequence = seqid->sequence; 1124 int status = 0; 1125 1126 spin_lock(&sequence->lock); 1127 seqid->task = task; 1128 if (list_empty(&seqid->list)) 1129 list_add_tail(&seqid->list, &sequence->list); 1130 if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid) 1131 goto unlock; 1132 rpc_sleep_on(&sequence->wait, task, NULL); 1133 status = -EAGAIN; 1134 unlock: 1135 spin_unlock(&sequence->lock); 1136 return status; 1137 } 1138 1139 static int nfs4_run_state_manager(void *); 1140 1141 static void nfs4_clear_state_manager_bit(struct nfs_client *clp) 1142 { 1143 smp_mb__before_atomic(); 1144 clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state); 1145 smp_mb__after_atomic(); 1146 wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING); 1147 rpc_wake_up(&clp->cl_rpcwaitq); 1148 } 1149 1150 /* 1151 * Schedule the nfs_client asynchronous state management routine 1152 */ 1153 void nfs4_schedule_state_manager(struct nfs_client *clp) 1154 { 1155 struct task_struct *task; 1156 char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1]; 1157 1158 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0) 1159 return; 1160 __module_get(THIS_MODULE); 1161 atomic_inc(&clp->cl_count); 1162 1163 /* The rcu_read_lock() is not strictly necessary, as the state 1164 * manager is the only thread that ever changes the rpc_xprt 1165 * after it's initialized. At this point, we're single threaded. */ 1166 rcu_read_lock(); 1167 snprintf(buf, sizeof(buf), "%s-manager", 1168 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR)); 1169 rcu_read_unlock(); 1170 task = kthread_run(nfs4_run_state_manager, clp, "%s", buf); 1171 if (IS_ERR(task)) { 1172 printk(KERN_ERR "%s: kthread_run: %ld\n", 1173 __func__, PTR_ERR(task)); 1174 nfs4_clear_state_manager_bit(clp); 1175 nfs_put_client(clp); 1176 module_put(THIS_MODULE); 1177 } 1178 } 1179 1180 /* 1181 * Schedule a lease recovery attempt 1182 */ 1183 void nfs4_schedule_lease_recovery(struct nfs_client *clp) 1184 { 1185 if (!clp) 1186 return; 1187 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) 1188 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); 1189 dprintk("%s: scheduling lease recovery for server %s\n", __func__, 1190 clp->cl_hostname); 1191 nfs4_schedule_state_manager(clp); 1192 } 1193 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery); 1194 1195 /** 1196 * nfs4_schedule_migration_recovery - trigger migration recovery 1197 * 1198 * @server: FSID that is migrating 1199 * 1200 * Returns zero if recovery has started, otherwise a negative NFS4ERR 1201 * value is returned. 1202 */ 1203 int nfs4_schedule_migration_recovery(const struct nfs_server *server) 1204 { 1205 struct nfs_client *clp = server->nfs_client; 1206 1207 if (server->fh_expire_type != NFS4_FH_PERSISTENT) { 1208 pr_err("NFS: volatile file handles not supported (server %s)\n", 1209 clp->cl_hostname); 1210 return -NFS4ERR_IO; 1211 } 1212 1213 if (test_bit(NFS_MIG_FAILED, &server->mig_status)) 1214 return -NFS4ERR_IO; 1215 1216 dprintk("%s: scheduling migration recovery for (%llx:%llx) on %s\n", 1217 __func__, 1218 (unsigned long long)server->fsid.major, 1219 (unsigned long long)server->fsid.minor, 1220 clp->cl_hostname); 1221 1222 set_bit(NFS_MIG_IN_TRANSITION, 1223 &((struct nfs_server *)server)->mig_status); 1224 set_bit(NFS4CLNT_MOVED, &clp->cl_state); 1225 1226 nfs4_schedule_state_manager(clp); 1227 return 0; 1228 } 1229 EXPORT_SYMBOL_GPL(nfs4_schedule_migration_recovery); 1230 1231 /** 1232 * nfs4_schedule_lease_moved_recovery - start lease-moved recovery 1233 * 1234 * @clp: server to check for moved leases 1235 * 1236 */ 1237 void nfs4_schedule_lease_moved_recovery(struct nfs_client *clp) 1238 { 1239 dprintk("%s: scheduling lease-moved recovery for client ID %llx on %s\n", 1240 __func__, clp->cl_clientid, clp->cl_hostname); 1241 1242 set_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state); 1243 nfs4_schedule_state_manager(clp); 1244 } 1245 EXPORT_SYMBOL_GPL(nfs4_schedule_lease_moved_recovery); 1246 1247 int nfs4_wait_clnt_recover(struct nfs_client *clp) 1248 { 1249 int res; 1250 1251 might_sleep(); 1252 1253 atomic_inc(&clp->cl_count); 1254 res = wait_on_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING, 1255 nfs_wait_bit_killable, TASK_KILLABLE); 1256 if (res) 1257 goto out; 1258 if (clp->cl_cons_state < 0) 1259 res = clp->cl_cons_state; 1260 out: 1261 nfs_put_client(clp); 1262 return res; 1263 } 1264 1265 int nfs4_client_recover_expired_lease(struct nfs_client *clp) 1266 { 1267 unsigned int loop; 1268 int ret; 1269 1270 for (loop = NFS4_MAX_LOOP_ON_RECOVER; loop != 0; loop--) { 1271 ret = nfs4_wait_clnt_recover(clp); 1272 if (ret != 0) 1273 break; 1274 if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) && 1275 !test_bit(NFS4CLNT_CHECK_LEASE,&clp->cl_state)) 1276 break; 1277 nfs4_schedule_state_manager(clp); 1278 ret = -EIO; 1279 } 1280 return ret; 1281 } 1282 1283 /* 1284 * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN 1285 * @clp: client to process 1286 * 1287 * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a 1288 * resend of the SETCLIENTID and hence re-establish the 1289 * callback channel. Then return all existing delegations. 1290 */ 1291 static void nfs40_handle_cb_pathdown(struct nfs_client *clp) 1292 { 1293 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1294 nfs_expire_all_delegations(clp); 1295 dprintk("%s: handling CB_PATHDOWN recovery for server %s\n", __func__, 1296 clp->cl_hostname); 1297 } 1298 1299 void nfs4_schedule_path_down_recovery(struct nfs_client *clp) 1300 { 1301 nfs40_handle_cb_pathdown(clp); 1302 nfs4_schedule_state_manager(clp); 1303 } 1304 1305 static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state) 1306 { 1307 1308 set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); 1309 /* Don't recover state that expired before the reboot */ 1310 if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) { 1311 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); 1312 return 0; 1313 } 1314 set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags); 1315 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state); 1316 return 1; 1317 } 1318 1319 int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state) 1320 { 1321 set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags); 1322 clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); 1323 set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags); 1324 set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state); 1325 return 1; 1326 } 1327 1328 int nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state) 1329 { 1330 struct nfs_client *clp = server->nfs_client; 1331 1332 if (!nfs4_valid_open_stateid(state)) 1333 return -EBADF; 1334 nfs4_state_mark_reclaim_nograce(clp, state); 1335 dprintk("%s: scheduling stateid recovery for server %s\n", __func__, 1336 clp->cl_hostname); 1337 nfs4_schedule_state_manager(clp); 1338 return 0; 1339 } 1340 EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery); 1341 1342 void nfs_inode_find_state_and_recover(struct inode *inode, 1343 const nfs4_stateid *stateid) 1344 { 1345 struct nfs_client *clp = NFS_SERVER(inode)->nfs_client; 1346 struct nfs_inode *nfsi = NFS_I(inode); 1347 struct nfs_open_context *ctx; 1348 struct nfs4_state *state; 1349 bool found = false; 1350 1351 spin_lock(&inode->i_lock); 1352 list_for_each_entry(ctx, &nfsi->open_files, list) { 1353 state = ctx->state; 1354 if (state == NULL) 1355 continue; 1356 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) 1357 continue; 1358 if (!nfs4_stateid_match(&state->stateid, stateid)) 1359 continue; 1360 nfs4_state_mark_reclaim_nograce(clp, state); 1361 found = true; 1362 } 1363 spin_unlock(&inode->i_lock); 1364 if (found) 1365 nfs4_schedule_state_manager(clp); 1366 } 1367 1368 static void nfs4_state_mark_open_context_bad(struct nfs4_state *state) 1369 { 1370 struct inode *inode = state->inode; 1371 struct nfs_inode *nfsi = NFS_I(inode); 1372 struct nfs_open_context *ctx; 1373 1374 spin_lock(&inode->i_lock); 1375 list_for_each_entry(ctx, &nfsi->open_files, list) { 1376 if (ctx->state != state) 1377 continue; 1378 set_bit(NFS_CONTEXT_BAD, &ctx->flags); 1379 } 1380 spin_unlock(&inode->i_lock); 1381 } 1382 1383 static void nfs4_state_mark_recovery_failed(struct nfs4_state *state, int error) 1384 { 1385 set_bit(NFS_STATE_RECOVERY_FAILED, &state->flags); 1386 nfs4_state_mark_open_context_bad(state); 1387 } 1388 1389 1390 static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops) 1391 { 1392 struct inode *inode = state->inode; 1393 struct nfs_inode *nfsi = NFS_I(inode); 1394 struct file_lock *fl; 1395 int status = 0; 1396 1397 if (inode->i_flock == NULL) 1398 return 0; 1399 1400 /* Guard against delegation returns and new lock/unlock calls */ 1401 down_write(&nfsi->rwsem); 1402 /* Protect inode->i_flock using the BKL */ 1403 spin_lock(&inode->i_lock); 1404 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { 1405 if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK))) 1406 continue; 1407 if (nfs_file_open_context(fl->fl_file)->state != state) 1408 continue; 1409 spin_unlock(&inode->i_lock); 1410 status = ops->recover_lock(state, fl); 1411 switch (status) { 1412 case 0: 1413 break; 1414 case -ESTALE: 1415 case -NFS4ERR_ADMIN_REVOKED: 1416 case -NFS4ERR_STALE_STATEID: 1417 case -NFS4ERR_BAD_STATEID: 1418 case -NFS4ERR_EXPIRED: 1419 case -NFS4ERR_NO_GRACE: 1420 case -NFS4ERR_STALE_CLIENTID: 1421 case -NFS4ERR_BADSESSION: 1422 case -NFS4ERR_BADSLOT: 1423 case -NFS4ERR_BAD_HIGH_SLOT: 1424 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1425 goto out; 1426 default: 1427 printk(KERN_ERR "NFS: %s: unhandled error %d\n", 1428 __func__, status); 1429 case -ENOMEM: 1430 case -NFS4ERR_DENIED: 1431 case -NFS4ERR_RECLAIM_BAD: 1432 case -NFS4ERR_RECLAIM_CONFLICT: 1433 /* kill_proc(fl->fl_pid, SIGLOST, 1); */ 1434 status = 0; 1435 } 1436 spin_lock(&inode->i_lock); 1437 } 1438 spin_unlock(&inode->i_lock); 1439 out: 1440 up_write(&nfsi->rwsem); 1441 return status; 1442 } 1443 1444 static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops) 1445 { 1446 struct nfs4_state *state; 1447 struct nfs4_lock_state *lock; 1448 int status = 0; 1449 1450 /* Note: we rely on the sp->so_states list being ordered 1451 * so that we always reclaim open(O_RDWR) and/or open(O_WRITE) 1452 * states first. 1453 * This is needed to ensure that the server won't give us any 1454 * read delegations that we have to return if, say, we are 1455 * recovering after a network partition or a reboot from a 1456 * server that doesn't support a grace period. 1457 */ 1458 spin_lock(&sp->so_lock); 1459 raw_write_seqcount_begin(&sp->so_reclaim_seqcount); 1460 restart: 1461 list_for_each_entry(state, &sp->so_states, open_states) { 1462 if (!test_and_clear_bit(ops->state_flag_bit, &state->flags)) 1463 continue; 1464 if (!nfs4_valid_open_stateid(state)) 1465 continue; 1466 if (state->state == 0) 1467 continue; 1468 atomic_inc(&state->count); 1469 spin_unlock(&sp->so_lock); 1470 status = ops->recover_open(sp, state); 1471 if (status >= 0) { 1472 status = nfs4_reclaim_locks(state, ops); 1473 if (status >= 0) { 1474 if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) { 1475 spin_lock(&state->state_lock); 1476 list_for_each_entry(lock, &state->lock_states, ls_locks) { 1477 if (!test_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags)) 1478 pr_warn_ratelimited("NFS: " 1479 "%s: Lock reclaim " 1480 "failed!\n", __func__); 1481 } 1482 spin_unlock(&state->state_lock); 1483 } 1484 nfs4_put_open_state(state); 1485 spin_lock(&sp->so_lock); 1486 goto restart; 1487 } 1488 } 1489 switch (status) { 1490 default: 1491 printk(KERN_ERR "NFS: %s: unhandled error %d\n", 1492 __func__, status); 1493 case -ENOENT: 1494 case -ENOMEM: 1495 case -ESTALE: 1496 /* Open state on this file cannot be recovered */ 1497 nfs4_state_mark_recovery_failed(state, status); 1498 break; 1499 case -EAGAIN: 1500 ssleep(1); 1501 case -NFS4ERR_ADMIN_REVOKED: 1502 case -NFS4ERR_STALE_STATEID: 1503 case -NFS4ERR_BAD_STATEID: 1504 case -NFS4ERR_RECLAIM_BAD: 1505 case -NFS4ERR_RECLAIM_CONFLICT: 1506 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state); 1507 break; 1508 case -NFS4ERR_EXPIRED: 1509 case -NFS4ERR_NO_GRACE: 1510 nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state); 1511 case -NFS4ERR_STALE_CLIENTID: 1512 case -NFS4ERR_BADSESSION: 1513 case -NFS4ERR_BADSLOT: 1514 case -NFS4ERR_BAD_HIGH_SLOT: 1515 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1516 goto out_err; 1517 } 1518 nfs4_put_open_state(state); 1519 spin_lock(&sp->so_lock); 1520 goto restart; 1521 } 1522 raw_write_seqcount_end(&sp->so_reclaim_seqcount); 1523 spin_unlock(&sp->so_lock); 1524 return 0; 1525 out_err: 1526 nfs4_put_open_state(state); 1527 spin_lock(&sp->so_lock); 1528 raw_write_seqcount_end(&sp->so_reclaim_seqcount); 1529 spin_unlock(&sp->so_lock); 1530 return status; 1531 } 1532 1533 static void nfs4_clear_open_state(struct nfs4_state *state) 1534 { 1535 struct nfs4_lock_state *lock; 1536 1537 clear_bit(NFS_DELEGATED_STATE, &state->flags); 1538 clear_bit(NFS_O_RDONLY_STATE, &state->flags); 1539 clear_bit(NFS_O_WRONLY_STATE, &state->flags); 1540 clear_bit(NFS_O_RDWR_STATE, &state->flags); 1541 spin_lock(&state->state_lock); 1542 list_for_each_entry(lock, &state->lock_states, ls_locks) { 1543 lock->ls_seqid.flags = 0; 1544 clear_bit(NFS_LOCK_INITIALIZED, &lock->ls_flags); 1545 } 1546 spin_unlock(&state->state_lock); 1547 } 1548 1549 static void nfs4_reset_seqids(struct nfs_server *server, 1550 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state)) 1551 { 1552 struct nfs_client *clp = server->nfs_client; 1553 struct nfs4_state_owner *sp; 1554 struct rb_node *pos; 1555 struct nfs4_state *state; 1556 1557 spin_lock(&clp->cl_lock); 1558 for (pos = rb_first(&server->state_owners); 1559 pos != NULL; 1560 pos = rb_next(pos)) { 1561 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); 1562 sp->so_seqid.flags = 0; 1563 spin_lock(&sp->so_lock); 1564 list_for_each_entry(state, &sp->so_states, open_states) { 1565 if (mark_reclaim(clp, state)) 1566 nfs4_clear_open_state(state); 1567 } 1568 spin_unlock(&sp->so_lock); 1569 } 1570 spin_unlock(&clp->cl_lock); 1571 } 1572 1573 static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, 1574 int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state)) 1575 { 1576 struct nfs_server *server; 1577 1578 rcu_read_lock(); 1579 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) 1580 nfs4_reset_seqids(server, mark_reclaim); 1581 rcu_read_unlock(); 1582 } 1583 1584 static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp) 1585 { 1586 /* Mark all delegations for reclaim */ 1587 nfs_delegation_mark_reclaim(clp); 1588 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot); 1589 } 1590 1591 static void nfs4_reclaim_complete(struct nfs_client *clp, 1592 const struct nfs4_state_recovery_ops *ops, 1593 struct rpc_cred *cred) 1594 { 1595 /* Notify the server we're done reclaiming our state */ 1596 if (ops->reclaim_complete) 1597 (void)ops->reclaim_complete(clp, cred); 1598 } 1599 1600 static void nfs4_clear_reclaim_server(struct nfs_server *server) 1601 { 1602 struct nfs_client *clp = server->nfs_client; 1603 struct nfs4_state_owner *sp; 1604 struct rb_node *pos; 1605 struct nfs4_state *state; 1606 1607 spin_lock(&clp->cl_lock); 1608 for (pos = rb_first(&server->state_owners); 1609 pos != NULL; 1610 pos = rb_next(pos)) { 1611 sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); 1612 spin_lock(&sp->so_lock); 1613 list_for_each_entry(state, &sp->so_states, open_states) { 1614 if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, 1615 &state->flags)) 1616 continue; 1617 nfs4_state_mark_reclaim_nograce(clp, state); 1618 } 1619 spin_unlock(&sp->so_lock); 1620 } 1621 spin_unlock(&clp->cl_lock); 1622 } 1623 1624 static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp) 1625 { 1626 struct nfs_server *server; 1627 1628 if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) 1629 return 0; 1630 1631 rcu_read_lock(); 1632 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) 1633 nfs4_clear_reclaim_server(server); 1634 rcu_read_unlock(); 1635 1636 nfs_delegation_reap_unclaimed(clp); 1637 return 1; 1638 } 1639 1640 static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp) 1641 { 1642 const struct nfs4_state_recovery_ops *ops; 1643 struct rpc_cred *cred; 1644 1645 if (!nfs4_state_clear_reclaim_reboot(clp)) 1646 return; 1647 ops = clp->cl_mvops->reboot_recovery_ops; 1648 cred = nfs4_get_clid_cred(clp); 1649 nfs4_reclaim_complete(clp, ops, cred); 1650 put_rpccred(cred); 1651 } 1652 1653 static void nfs_delegation_clear_all(struct nfs_client *clp) 1654 { 1655 nfs_delegation_mark_reclaim(clp); 1656 nfs_delegation_reap_unclaimed(clp); 1657 } 1658 1659 static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp) 1660 { 1661 nfs_delegation_clear_all(clp); 1662 nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce); 1663 } 1664 1665 static int nfs4_recovery_handle_error(struct nfs_client *clp, int error) 1666 { 1667 switch (error) { 1668 case 0: 1669 break; 1670 case -NFS4ERR_CB_PATH_DOWN: 1671 nfs40_handle_cb_pathdown(clp); 1672 break; 1673 case -NFS4ERR_NO_GRACE: 1674 nfs4_state_end_reclaim_reboot(clp); 1675 break; 1676 case -NFS4ERR_STALE_CLIENTID: 1677 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1678 nfs4_state_clear_reclaim_reboot(clp); 1679 nfs4_state_start_reclaim_reboot(clp); 1680 break; 1681 case -NFS4ERR_EXPIRED: 1682 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1683 nfs4_state_start_reclaim_nograce(clp); 1684 break; 1685 case -NFS4ERR_BADSESSION: 1686 case -NFS4ERR_BADSLOT: 1687 case -NFS4ERR_BAD_HIGH_SLOT: 1688 case -NFS4ERR_DEADSESSION: 1689 case -NFS4ERR_SEQ_FALSE_RETRY: 1690 case -NFS4ERR_SEQ_MISORDERED: 1691 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); 1692 /* Zero session reset errors */ 1693 break; 1694 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 1695 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state); 1696 break; 1697 default: 1698 dprintk("%s: failed to handle error %d for server %s\n", 1699 __func__, error, clp->cl_hostname); 1700 return error; 1701 } 1702 dprintk("%s: handled error %d for server %s\n", __func__, error, 1703 clp->cl_hostname); 1704 return 0; 1705 } 1706 1707 static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops) 1708 { 1709 struct nfs4_state_owner *sp; 1710 struct nfs_server *server; 1711 struct rb_node *pos; 1712 int status = 0; 1713 1714 restart: 1715 rcu_read_lock(); 1716 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { 1717 nfs4_purge_state_owners(server); 1718 spin_lock(&clp->cl_lock); 1719 for (pos = rb_first(&server->state_owners); 1720 pos != NULL; 1721 pos = rb_next(pos)) { 1722 sp = rb_entry(pos, 1723 struct nfs4_state_owner, so_server_node); 1724 if (!test_and_clear_bit(ops->owner_flag_bit, 1725 &sp->so_flags)) 1726 continue; 1727 atomic_inc(&sp->so_count); 1728 spin_unlock(&clp->cl_lock); 1729 rcu_read_unlock(); 1730 1731 status = nfs4_reclaim_open_state(sp, ops); 1732 if (status < 0) { 1733 set_bit(ops->owner_flag_bit, &sp->so_flags); 1734 nfs4_put_state_owner(sp); 1735 return nfs4_recovery_handle_error(clp, status); 1736 } 1737 1738 nfs4_put_state_owner(sp); 1739 goto restart; 1740 } 1741 spin_unlock(&clp->cl_lock); 1742 } 1743 rcu_read_unlock(); 1744 return status; 1745 } 1746 1747 static int nfs4_check_lease(struct nfs_client *clp) 1748 { 1749 struct rpc_cred *cred; 1750 const struct nfs4_state_maintenance_ops *ops = 1751 clp->cl_mvops->state_renewal_ops; 1752 int status; 1753 1754 /* Is the client already known to have an expired lease? */ 1755 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) 1756 return 0; 1757 spin_lock(&clp->cl_lock); 1758 cred = ops->get_state_renewal_cred_locked(clp); 1759 spin_unlock(&clp->cl_lock); 1760 if (cred == NULL) { 1761 cred = nfs4_get_clid_cred(clp); 1762 status = -ENOKEY; 1763 if (cred == NULL) 1764 goto out; 1765 } 1766 status = ops->renew_lease(clp, cred); 1767 put_rpccred(cred); 1768 if (status == -ETIMEDOUT) { 1769 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); 1770 return 0; 1771 } 1772 out: 1773 return nfs4_recovery_handle_error(clp, status); 1774 } 1775 1776 /* Set NFS4CLNT_LEASE_EXPIRED and reclaim reboot state for all v4.0 errors 1777 * and for recoverable errors on EXCHANGE_ID for v4.1 1778 */ 1779 static int nfs4_handle_reclaim_lease_error(struct nfs_client *clp, int status) 1780 { 1781 switch (status) { 1782 case -NFS4ERR_SEQ_MISORDERED: 1783 if (test_and_set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) 1784 return -ESERVERFAULT; 1785 /* Lease confirmation error: retry after purging the lease */ 1786 ssleep(1); 1787 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 1788 break; 1789 case -NFS4ERR_STALE_CLIENTID: 1790 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 1791 nfs4_state_clear_reclaim_reboot(clp); 1792 nfs4_state_start_reclaim_reboot(clp); 1793 break; 1794 case -NFS4ERR_CLID_INUSE: 1795 pr_err("NFS: Server %s reports our clientid is in use\n", 1796 clp->cl_hostname); 1797 nfs_mark_client_ready(clp, -EPERM); 1798 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 1799 return -EPERM; 1800 case -EACCES: 1801 case -NFS4ERR_DELAY: 1802 case -ETIMEDOUT: 1803 case -EAGAIN: 1804 ssleep(1); 1805 break; 1806 1807 case -NFS4ERR_MINOR_VERS_MISMATCH: 1808 if (clp->cl_cons_state == NFS_CS_SESSION_INITING) 1809 nfs_mark_client_ready(clp, -EPROTONOSUPPORT); 1810 dprintk("%s: exit with error %d for server %s\n", 1811 __func__, -EPROTONOSUPPORT, clp->cl_hostname); 1812 return -EPROTONOSUPPORT; 1813 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery 1814 * in nfs4_exchange_id */ 1815 default: 1816 dprintk("%s: exit with error %d for server %s\n", __func__, 1817 status, clp->cl_hostname); 1818 return status; 1819 } 1820 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1821 dprintk("%s: handled error %d for server %s\n", __func__, status, 1822 clp->cl_hostname); 1823 return 0; 1824 } 1825 1826 static int nfs4_establish_lease(struct nfs_client *clp) 1827 { 1828 struct rpc_cred *cred; 1829 const struct nfs4_state_recovery_ops *ops = 1830 clp->cl_mvops->reboot_recovery_ops; 1831 int status; 1832 1833 cred = nfs4_get_clid_cred(clp); 1834 if (cred == NULL) 1835 return -ENOENT; 1836 status = ops->establish_clid(clp, cred); 1837 put_rpccred(cred); 1838 if (status != 0) 1839 return status; 1840 pnfs_destroy_all_layouts(clp); 1841 return 0; 1842 } 1843 1844 /* 1845 * Returns zero or a negative errno. NFS4ERR values are converted 1846 * to local errno values. 1847 */ 1848 static int nfs4_reclaim_lease(struct nfs_client *clp) 1849 { 1850 int status; 1851 1852 status = nfs4_establish_lease(clp); 1853 if (status < 0) 1854 return nfs4_handle_reclaim_lease_error(clp, status); 1855 if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, &clp->cl_state)) 1856 nfs4_state_start_reclaim_nograce(clp); 1857 if (!test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) 1858 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state); 1859 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); 1860 clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1861 return 0; 1862 } 1863 1864 static int nfs4_purge_lease(struct nfs_client *clp) 1865 { 1866 int status; 1867 1868 status = nfs4_establish_lease(clp); 1869 if (status < 0) 1870 return nfs4_handle_reclaim_lease_error(clp, status); 1871 clear_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state); 1872 set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); 1873 nfs4_state_start_reclaim_nograce(clp); 1874 return 0; 1875 } 1876 1877 /* 1878 * Try remote migration of one FSID from a source server to a 1879 * destination server. The source server provides a list of 1880 * potential destinations. 1881 * 1882 * Returns zero or a negative NFS4ERR status code. 1883 */ 1884 static int nfs4_try_migration(struct nfs_server *server, struct rpc_cred *cred) 1885 { 1886 struct nfs_client *clp = server->nfs_client; 1887 struct nfs4_fs_locations *locations = NULL; 1888 struct inode *inode; 1889 struct page *page; 1890 int status, result; 1891 1892 dprintk("--> %s: FSID %llx:%llx on \"%s\"\n", __func__, 1893 (unsigned long long)server->fsid.major, 1894 (unsigned long long)server->fsid.minor, 1895 clp->cl_hostname); 1896 1897 result = 0; 1898 page = alloc_page(GFP_KERNEL); 1899 locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL); 1900 if (page == NULL || locations == NULL) { 1901 dprintk("<-- %s: no memory\n", __func__); 1902 goto out; 1903 } 1904 1905 inode = server->super->s_root->d_inode; 1906 result = nfs4_proc_get_locations(inode, locations, page, cred); 1907 if (result) { 1908 dprintk("<-- %s: failed to retrieve fs_locations: %d\n", 1909 __func__, result); 1910 goto out; 1911 } 1912 1913 result = -NFS4ERR_NXIO; 1914 if (!(locations->fattr.valid & NFS_ATTR_FATTR_V4_LOCATIONS)) { 1915 dprintk("<-- %s: No fs_locations data, migration skipped\n", 1916 __func__); 1917 goto out; 1918 } 1919 1920 nfs4_begin_drain_session(clp); 1921 1922 status = nfs4_replace_transport(server, locations); 1923 if (status != 0) { 1924 dprintk("<-- %s: failed to replace transport: %d\n", 1925 __func__, status); 1926 goto out; 1927 } 1928 1929 result = 0; 1930 dprintk("<-- %s: migration succeeded\n", __func__); 1931 1932 out: 1933 if (page != NULL) 1934 __free_page(page); 1935 kfree(locations); 1936 if (result) { 1937 pr_err("NFS: migration recovery failed (server %s)\n", 1938 clp->cl_hostname); 1939 set_bit(NFS_MIG_FAILED, &server->mig_status); 1940 } 1941 return result; 1942 } 1943 1944 /* 1945 * Returns zero or a negative NFS4ERR status code. 1946 */ 1947 static int nfs4_handle_migration(struct nfs_client *clp) 1948 { 1949 const struct nfs4_state_maintenance_ops *ops = 1950 clp->cl_mvops->state_renewal_ops; 1951 struct nfs_server *server; 1952 struct rpc_cred *cred; 1953 1954 dprintk("%s: migration reported on \"%s\"\n", __func__, 1955 clp->cl_hostname); 1956 1957 spin_lock(&clp->cl_lock); 1958 cred = ops->get_state_renewal_cred_locked(clp); 1959 spin_unlock(&clp->cl_lock); 1960 if (cred == NULL) 1961 return -NFS4ERR_NOENT; 1962 1963 clp->cl_mig_gen++; 1964 restart: 1965 rcu_read_lock(); 1966 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { 1967 int status; 1968 1969 if (server->mig_gen == clp->cl_mig_gen) 1970 continue; 1971 server->mig_gen = clp->cl_mig_gen; 1972 1973 if (!test_and_clear_bit(NFS_MIG_IN_TRANSITION, 1974 &server->mig_status)) 1975 continue; 1976 1977 rcu_read_unlock(); 1978 status = nfs4_try_migration(server, cred); 1979 if (status < 0) { 1980 put_rpccred(cred); 1981 return status; 1982 } 1983 goto restart; 1984 } 1985 rcu_read_unlock(); 1986 put_rpccred(cred); 1987 return 0; 1988 } 1989 1990 /* 1991 * Test each nfs_server on the clp's cl_superblocks list to see 1992 * if it's moved to another server. Stop when the server no longer 1993 * returns NFS4ERR_LEASE_MOVED. 1994 */ 1995 static int nfs4_handle_lease_moved(struct nfs_client *clp) 1996 { 1997 const struct nfs4_state_maintenance_ops *ops = 1998 clp->cl_mvops->state_renewal_ops; 1999 struct nfs_server *server; 2000 struct rpc_cred *cred; 2001 2002 dprintk("%s: lease moved reported on \"%s\"\n", __func__, 2003 clp->cl_hostname); 2004 2005 spin_lock(&clp->cl_lock); 2006 cred = ops->get_state_renewal_cred_locked(clp); 2007 spin_unlock(&clp->cl_lock); 2008 if (cred == NULL) 2009 return -NFS4ERR_NOENT; 2010 2011 clp->cl_mig_gen++; 2012 restart: 2013 rcu_read_lock(); 2014 list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { 2015 struct inode *inode; 2016 int status; 2017 2018 if (server->mig_gen == clp->cl_mig_gen) 2019 continue; 2020 server->mig_gen = clp->cl_mig_gen; 2021 2022 rcu_read_unlock(); 2023 2024 inode = server->super->s_root->d_inode; 2025 status = nfs4_proc_fsid_present(inode, cred); 2026 if (status != -NFS4ERR_MOVED) 2027 goto restart; /* wasn't this one */ 2028 if (nfs4_try_migration(server, cred) == -NFS4ERR_LEASE_MOVED) 2029 goto restart; /* there are more */ 2030 goto out; 2031 } 2032 rcu_read_unlock(); 2033 2034 out: 2035 put_rpccred(cred); 2036 return 0; 2037 } 2038 2039 /** 2040 * nfs4_discover_server_trunking - Detect server IP address trunking 2041 * 2042 * @clp: nfs_client under test 2043 * @result: OUT: found nfs_client, or clp 2044 * 2045 * Returns zero or a negative errno. If zero is returned, 2046 * an nfs_client pointer is planted in "result". 2047 * 2048 * Note: since we are invoked in process context, and 2049 * not from inside the state manager, we cannot use 2050 * nfs4_handle_reclaim_lease_error(). 2051 */ 2052 int nfs4_discover_server_trunking(struct nfs_client *clp, 2053 struct nfs_client **result) 2054 { 2055 const struct nfs4_state_recovery_ops *ops = 2056 clp->cl_mvops->reboot_recovery_ops; 2057 struct rpc_clnt *clnt; 2058 struct rpc_cred *cred; 2059 int i, status; 2060 2061 dprintk("NFS: %s: testing '%s'\n", __func__, clp->cl_hostname); 2062 2063 clnt = clp->cl_rpcclient; 2064 i = 0; 2065 2066 mutex_lock(&nfs_clid_init_mutex); 2067 again: 2068 status = -ENOENT; 2069 cred = nfs4_get_clid_cred(clp); 2070 if (cred == NULL) 2071 goto out_unlock; 2072 2073 status = ops->detect_trunking(clp, result, cred); 2074 put_rpccred(cred); 2075 switch (status) { 2076 case 0: 2077 break; 2078 case -ETIMEDOUT: 2079 if (clnt->cl_softrtry) 2080 break; 2081 case -NFS4ERR_DELAY: 2082 case -EAGAIN: 2083 ssleep(1); 2084 case -NFS4ERR_STALE_CLIENTID: 2085 dprintk("NFS: %s after status %d, retrying\n", 2086 __func__, status); 2087 goto again; 2088 case -EACCES: 2089 if (i++ == 0) { 2090 nfs4_root_machine_cred(clp); 2091 goto again; 2092 } 2093 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) 2094 break; 2095 case -NFS4ERR_CLID_INUSE: 2096 case -NFS4ERR_WRONGSEC: 2097 /* No point in retrying if we already used RPC_AUTH_UNIX */ 2098 if (clnt->cl_auth->au_flavor == RPC_AUTH_UNIX) { 2099 status = -EPERM; 2100 break; 2101 } 2102 clnt = rpc_clone_client_set_auth(clnt, RPC_AUTH_UNIX); 2103 if (IS_ERR(clnt)) { 2104 status = PTR_ERR(clnt); 2105 break; 2106 } 2107 /* Note: this is safe because we haven't yet marked the 2108 * client as ready, so we are the only user of 2109 * clp->cl_rpcclient 2110 */ 2111 clnt = xchg(&clp->cl_rpcclient, clnt); 2112 rpc_shutdown_client(clnt); 2113 clnt = clp->cl_rpcclient; 2114 goto again; 2115 2116 case -NFS4ERR_MINOR_VERS_MISMATCH: 2117 status = -EPROTONOSUPPORT; 2118 break; 2119 2120 case -EKEYEXPIRED: 2121 case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery 2122 * in nfs4_exchange_id */ 2123 status = -EKEYEXPIRED; 2124 break; 2125 default: 2126 pr_warn("NFS: %s unhandled error %d. Exiting with error EIO\n", 2127 __func__, status); 2128 status = -EIO; 2129 } 2130 2131 out_unlock: 2132 mutex_unlock(&nfs_clid_init_mutex); 2133 dprintk("NFS: %s: status = %d\n", __func__, status); 2134 return status; 2135 } 2136 2137 #ifdef CONFIG_NFS_V4_1 2138 void nfs4_schedule_session_recovery(struct nfs4_session *session, int err) 2139 { 2140 struct nfs_client *clp = session->clp; 2141 2142 switch (err) { 2143 default: 2144 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); 2145 break; 2146 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: 2147 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state); 2148 } 2149 nfs4_schedule_lease_recovery(clp); 2150 } 2151 EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery); 2152 2153 static void nfs41_ping_server(struct nfs_client *clp) 2154 { 2155 /* Use CHECK_LEASE to ping the server with a SEQUENCE */ 2156 set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); 2157 nfs4_schedule_state_manager(clp); 2158 } 2159 2160 void nfs41_server_notify_target_slotid_update(struct nfs_client *clp) 2161 { 2162 nfs41_ping_server(clp); 2163 } 2164 2165 void nfs41_server_notify_highest_slotid_update(struct nfs_client *clp) 2166 { 2167 nfs41_ping_server(clp); 2168 } 2169 2170 static void nfs4_reset_all_state(struct nfs_client *clp) 2171 { 2172 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) { 2173 set_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state); 2174 clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); 2175 nfs4_state_start_reclaim_nograce(clp); 2176 dprintk("%s: scheduling reset of all state for server %s!\n", 2177 __func__, clp->cl_hostname); 2178 nfs4_schedule_state_manager(clp); 2179 } 2180 } 2181 2182 static void nfs41_handle_server_reboot(struct nfs_client *clp) 2183 { 2184 if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) { 2185 nfs4_state_start_reclaim_reboot(clp); 2186 dprintk("%s: server %s rebooted!\n", __func__, 2187 clp->cl_hostname); 2188 nfs4_schedule_state_manager(clp); 2189 } 2190 } 2191 2192 static void nfs41_handle_state_revoked(struct nfs_client *clp) 2193 { 2194 nfs4_reset_all_state(clp); 2195 dprintk("%s: state revoked on server %s\n", __func__, clp->cl_hostname); 2196 } 2197 2198 static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp) 2199 { 2200 /* This will need to handle layouts too */ 2201 nfs_expire_all_delegations(clp); 2202 dprintk("%s: Recallable state revoked on server %s!\n", __func__, 2203 clp->cl_hostname); 2204 } 2205 2206 static void nfs41_handle_backchannel_fault(struct nfs_client *clp) 2207 { 2208 nfs_expire_all_delegations(clp); 2209 if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0) 2210 nfs4_schedule_state_manager(clp); 2211 dprintk("%s: server %s declared a backchannel fault\n", __func__, 2212 clp->cl_hostname); 2213 } 2214 2215 static void nfs41_handle_cb_path_down(struct nfs_client *clp) 2216 { 2217 if (test_and_set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, 2218 &clp->cl_state) == 0) 2219 nfs4_schedule_state_manager(clp); 2220 } 2221 2222 void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags) 2223 { 2224 if (!flags) 2225 return; 2226 2227 dprintk("%s: \"%s\" (client ID %llx) flags=0x%08x\n", 2228 __func__, clp->cl_hostname, clp->cl_clientid, flags); 2229 2230 if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED) 2231 nfs41_handle_server_reboot(clp); 2232 if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED | 2233 SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED | 2234 SEQ4_STATUS_ADMIN_STATE_REVOKED)) 2235 nfs41_handle_state_revoked(clp); 2236 if (flags & SEQ4_STATUS_LEASE_MOVED) 2237 nfs4_schedule_lease_moved_recovery(clp); 2238 if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED) 2239 nfs41_handle_recallable_state_revoked(clp); 2240 if (flags & SEQ4_STATUS_BACKCHANNEL_FAULT) 2241 nfs41_handle_backchannel_fault(clp); 2242 else if (flags & (SEQ4_STATUS_CB_PATH_DOWN | 2243 SEQ4_STATUS_CB_PATH_DOWN_SESSION)) 2244 nfs41_handle_cb_path_down(clp); 2245 } 2246 2247 static int nfs4_reset_session(struct nfs_client *clp) 2248 { 2249 struct rpc_cred *cred; 2250 int status; 2251 2252 if (!nfs4_has_session(clp)) 2253 return 0; 2254 nfs4_begin_drain_session(clp); 2255 cred = nfs4_get_clid_cred(clp); 2256 status = nfs4_proc_destroy_session(clp->cl_session, cred); 2257 switch (status) { 2258 case 0: 2259 case -NFS4ERR_BADSESSION: 2260 case -NFS4ERR_DEADSESSION: 2261 break; 2262 case -NFS4ERR_BACK_CHAN_BUSY: 2263 case -NFS4ERR_DELAY: 2264 set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); 2265 status = 0; 2266 ssleep(1); 2267 goto out; 2268 default: 2269 status = nfs4_recovery_handle_error(clp, status); 2270 goto out; 2271 } 2272 2273 memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN); 2274 status = nfs4_proc_create_session(clp, cred); 2275 if (status) { 2276 dprintk("%s: session reset failed with status %d for server %s!\n", 2277 __func__, status, clp->cl_hostname); 2278 status = nfs4_handle_reclaim_lease_error(clp, status); 2279 goto out; 2280 } 2281 nfs41_finish_session_reset(clp); 2282 dprintk("%s: session reset was successful for server %s!\n", 2283 __func__, clp->cl_hostname); 2284 out: 2285 if (cred) 2286 put_rpccred(cred); 2287 return status; 2288 } 2289 2290 static int nfs4_bind_conn_to_session(struct nfs_client *clp) 2291 { 2292 struct rpc_cred *cred; 2293 int ret; 2294 2295 if (!nfs4_has_session(clp)) 2296 return 0; 2297 nfs4_begin_drain_session(clp); 2298 cred = nfs4_get_clid_cred(clp); 2299 ret = nfs4_proc_bind_conn_to_session(clp, cred); 2300 if (cred) 2301 put_rpccred(cred); 2302 clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state); 2303 switch (ret) { 2304 case 0: 2305 dprintk("%s: bind_conn_to_session was successful for server %s!\n", 2306 __func__, clp->cl_hostname); 2307 break; 2308 case -NFS4ERR_DELAY: 2309 ssleep(1); 2310 set_bit(NFS4CLNT_BIND_CONN_TO_SESSION, &clp->cl_state); 2311 break; 2312 default: 2313 return nfs4_recovery_handle_error(clp, ret); 2314 } 2315 return 0; 2316 } 2317 #else /* CONFIG_NFS_V4_1 */ 2318 static int nfs4_reset_session(struct nfs_client *clp) { return 0; } 2319 2320 static int nfs4_bind_conn_to_session(struct nfs_client *clp) 2321 { 2322 return 0; 2323 } 2324 #endif /* CONFIG_NFS_V4_1 */ 2325 2326 static void nfs4_state_manager(struct nfs_client *clp) 2327 { 2328 int status = 0; 2329 const char *section = "", *section_sep = ""; 2330 2331 /* Ensure exclusive access to NFSv4 state */ 2332 do { 2333 if (test_bit(NFS4CLNT_PURGE_STATE, &clp->cl_state)) { 2334 section = "purge state"; 2335 status = nfs4_purge_lease(clp); 2336 if (status < 0) 2337 goto out_error; 2338 continue; 2339 } 2340 2341 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) { 2342 section = "lease expired"; 2343 /* We're going to have to re-establish a clientid */ 2344 status = nfs4_reclaim_lease(clp); 2345 if (status < 0) 2346 goto out_error; 2347 continue; 2348 } 2349 2350 /* Initialize or reset the session */ 2351 if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) { 2352 section = "reset session"; 2353 status = nfs4_reset_session(clp); 2354 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) 2355 continue; 2356 if (status < 0) 2357 goto out_error; 2358 } 2359 2360 /* Send BIND_CONN_TO_SESSION */ 2361 if (test_and_clear_bit(NFS4CLNT_BIND_CONN_TO_SESSION, 2362 &clp->cl_state)) { 2363 section = "bind conn to session"; 2364 status = nfs4_bind_conn_to_session(clp); 2365 if (status < 0) 2366 goto out_error; 2367 continue; 2368 } 2369 2370 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) { 2371 section = "check lease"; 2372 status = nfs4_check_lease(clp); 2373 if (status < 0) 2374 goto out_error; 2375 } 2376 2377 if (test_and_clear_bit(NFS4CLNT_MOVED, &clp->cl_state)) { 2378 section = "migration"; 2379 status = nfs4_handle_migration(clp); 2380 if (status < 0) 2381 goto out_error; 2382 } 2383 2384 if (test_and_clear_bit(NFS4CLNT_LEASE_MOVED, &clp->cl_state)) { 2385 section = "lease moved"; 2386 status = nfs4_handle_lease_moved(clp); 2387 if (status < 0) 2388 goto out_error; 2389 } 2390 2391 /* First recover reboot state... */ 2392 if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) { 2393 section = "reclaim reboot"; 2394 status = nfs4_do_reclaim(clp, 2395 clp->cl_mvops->reboot_recovery_ops); 2396 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) || 2397 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state)) 2398 continue; 2399 nfs4_state_end_reclaim_reboot(clp); 2400 if (test_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) 2401 continue; 2402 if (status < 0) 2403 goto out_error; 2404 } 2405 2406 /* Now recover expired state... */ 2407 if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) { 2408 section = "reclaim nograce"; 2409 status = nfs4_do_reclaim(clp, 2410 clp->cl_mvops->nograce_recovery_ops); 2411 if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) || 2412 test_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) || 2413 test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) 2414 continue; 2415 if (status < 0) 2416 goto out_error; 2417 } 2418 2419 nfs4_end_drain_session(clp); 2420 if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) { 2421 nfs_client_return_marked_delegations(clp); 2422 continue; 2423 } 2424 2425 nfs4_clear_state_manager_bit(clp); 2426 /* Did we race with an attempt to give us more work? */ 2427 if (clp->cl_state == 0) 2428 break; 2429 if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0) 2430 break; 2431 } while (atomic_read(&clp->cl_count) > 1); 2432 return; 2433 out_error: 2434 if (strlen(section)) 2435 section_sep = ": "; 2436 pr_warn_ratelimited("NFS: state manager%s%s failed on NFSv4 server %s" 2437 " with error %d\n", section_sep, section, 2438 clp->cl_hostname, -status); 2439 ssleep(1); 2440 nfs4_end_drain_session(clp); 2441 nfs4_clear_state_manager_bit(clp); 2442 } 2443 2444 static int nfs4_run_state_manager(void *ptr) 2445 { 2446 struct nfs_client *clp = ptr; 2447 2448 allow_signal(SIGKILL); 2449 nfs4_state_manager(clp); 2450 nfs_put_client(clp); 2451 module_put_and_exit(0); 2452 return 0; 2453 } 2454 2455 /* 2456 * Local variables: 2457 * c-basic-offset: 8 2458 * End: 2459 */ 2460