lh | 9ed821d | 2023-04-07 01:36:19 -0700 | [diff] [blame] | 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 "nfs4_fs.h" |
| 55 | #include "callback.h" |
| 56 | #include "delegation.h" |
| 57 | #include "internal.h" |
| 58 | #include "pnfs.h" |
| 59 | |
| 60 | #define OPENOWNER_POOL_SIZE 8 |
| 61 | |
| 62 | const nfs4_stateid zero_stateid; |
| 63 | |
| 64 | static LIST_HEAD(nfs4_clientid_list); |
| 65 | |
| 66 | int nfs4_init_clientid(struct nfs_client *clp, struct rpc_cred *cred) |
| 67 | { |
| 68 | struct nfs4_setclientid_res clid = { |
| 69 | .clientid = clp->cl_clientid, |
| 70 | .confirm = clp->cl_confirm, |
| 71 | }; |
| 72 | unsigned short port; |
| 73 | int status; |
| 74 | |
| 75 | if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state)) |
| 76 | goto do_confirm; |
| 77 | port = nfs_callback_tcpport; |
| 78 | if (clp->cl_addr.ss_family == AF_INET6) |
| 79 | port = nfs_callback_tcpport6; |
| 80 | |
| 81 | status = nfs4_proc_setclientid(clp, NFS4_CALLBACK, port, cred, &clid); |
| 82 | if (status != 0) |
| 83 | goto out; |
| 84 | clp->cl_clientid = clid.clientid; |
| 85 | clp->cl_confirm = clid.confirm; |
| 86 | set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); |
| 87 | do_confirm: |
| 88 | status = nfs4_proc_setclientid_confirm(clp, &clid, cred); |
| 89 | if (status != 0) |
| 90 | goto out; |
| 91 | clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); |
| 92 | nfs4_schedule_state_renewal(clp); |
| 93 | out: |
| 94 | return status; |
| 95 | } |
| 96 | |
| 97 | struct rpc_cred *nfs4_get_machine_cred_locked(struct nfs_client *clp) |
| 98 | { |
| 99 | struct rpc_cred *cred = NULL; |
| 100 | |
| 101 | if (clp->cl_machine_cred != NULL) |
| 102 | cred = get_rpccred(clp->cl_machine_cred); |
| 103 | return cred; |
| 104 | } |
| 105 | |
| 106 | static void nfs4_clear_machine_cred(struct nfs_client *clp) |
| 107 | { |
| 108 | struct rpc_cred *cred; |
| 109 | |
| 110 | spin_lock(&clp->cl_lock); |
| 111 | cred = clp->cl_machine_cred; |
| 112 | clp->cl_machine_cred = NULL; |
| 113 | spin_unlock(&clp->cl_lock); |
| 114 | if (cred != NULL) |
| 115 | put_rpccred(cred); |
| 116 | } |
| 117 | |
| 118 | static struct rpc_cred * |
| 119 | nfs4_get_renew_cred_server_locked(struct nfs_server *server) |
| 120 | { |
| 121 | struct rpc_cred *cred = NULL; |
| 122 | struct nfs4_state_owner *sp; |
| 123 | struct rb_node *pos; |
| 124 | |
| 125 | for (pos = rb_first(&server->state_owners); |
| 126 | pos != NULL; |
| 127 | pos = rb_next(pos)) { |
| 128 | sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); |
| 129 | if (list_empty(&sp->so_states)) |
| 130 | continue; |
| 131 | cred = get_rpccred(sp->so_cred); |
| 132 | break; |
| 133 | } |
| 134 | return cred; |
| 135 | } |
| 136 | |
| 137 | /** |
| 138 | * nfs4_get_renew_cred_locked - Acquire credential for a renew operation |
| 139 | * @clp: client state handle |
| 140 | * |
| 141 | * Returns an rpc_cred with reference count bumped, or NULL. |
| 142 | * Caller must hold clp->cl_lock. |
| 143 | */ |
| 144 | struct rpc_cred *nfs4_get_renew_cred_locked(struct nfs_client *clp) |
| 145 | { |
| 146 | struct rpc_cred *cred = NULL; |
| 147 | struct nfs_server *server; |
| 148 | |
| 149 | /* Use machine credentials if available */ |
| 150 | cred = nfs4_get_machine_cred_locked(clp); |
| 151 | if (cred != NULL) |
| 152 | goto out; |
| 153 | |
| 154 | rcu_read_lock(); |
| 155 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
| 156 | cred = nfs4_get_renew_cred_server_locked(server); |
| 157 | if (cred != NULL) |
| 158 | break; |
| 159 | } |
| 160 | rcu_read_unlock(); |
| 161 | |
| 162 | out: |
| 163 | return cred; |
| 164 | } |
| 165 | |
| 166 | #if defined(CONFIG_NFS_V4_1) |
| 167 | |
| 168 | static int nfs41_setup_state_renewal(struct nfs_client *clp) |
| 169 | { |
| 170 | int status; |
| 171 | struct nfs_fsinfo fsinfo; |
| 172 | |
| 173 | if (!test_bit(NFS_CS_CHECK_LEASE_TIME, &clp->cl_res_state)) { |
| 174 | nfs4_schedule_state_renewal(clp); |
| 175 | return 0; |
| 176 | } |
| 177 | |
| 178 | status = nfs4_proc_get_lease_time(clp, &fsinfo); |
| 179 | if (status == 0) { |
| 180 | /* Update lease time and schedule renewal */ |
| 181 | spin_lock(&clp->cl_lock); |
| 182 | clp->cl_lease_time = fsinfo.lease_time * HZ; |
| 183 | clp->cl_last_renewal = jiffies; |
| 184 | spin_unlock(&clp->cl_lock); |
| 185 | |
| 186 | nfs4_schedule_state_renewal(clp); |
| 187 | } |
| 188 | |
| 189 | return status; |
| 190 | } |
| 191 | |
| 192 | /* |
| 193 | * Back channel returns NFS4ERR_DELAY for new requests when |
| 194 | * NFS4_SESSION_DRAINING is set so there is no work to be done when draining |
| 195 | * is ended. |
| 196 | */ |
| 197 | static void nfs4_end_drain_session(struct nfs_client *clp) |
| 198 | { |
| 199 | struct nfs4_session *ses = clp->cl_session; |
| 200 | struct nfs4_slot_table *tbl; |
| 201 | int max_slots; |
| 202 | |
| 203 | if (ses == NULL) |
| 204 | return; |
| 205 | tbl = &ses->fc_slot_table; |
| 206 | if (test_and_clear_bit(NFS4_SESSION_DRAINING, &ses->session_state)) { |
| 207 | spin_lock(&tbl->slot_tbl_lock); |
| 208 | max_slots = tbl->max_slots; |
| 209 | while (max_slots--) { |
| 210 | if (rpc_wake_up_first(&tbl->slot_tbl_waitq, |
| 211 | nfs4_set_task_privileged, |
| 212 | NULL) == NULL) |
| 213 | break; |
| 214 | } |
| 215 | spin_unlock(&tbl->slot_tbl_lock); |
| 216 | } |
| 217 | } |
| 218 | |
| 219 | static int nfs4_wait_on_slot_tbl(struct nfs4_slot_table *tbl) |
| 220 | { |
| 221 | spin_lock(&tbl->slot_tbl_lock); |
| 222 | if (tbl->highest_used_slotid != NFS4_NO_SLOT) { |
| 223 | INIT_COMPLETION(tbl->complete); |
| 224 | spin_unlock(&tbl->slot_tbl_lock); |
| 225 | return wait_for_completion_interruptible(&tbl->complete); |
| 226 | } |
| 227 | spin_unlock(&tbl->slot_tbl_lock); |
| 228 | return 0; |
| 229 | } |
| 230 | |
| 231 | static int nfs4_begin_drain_session(struct nfs_client *clp) |
| 232 | { |
| 233 | struct nfs4_session *ses = clp->cl_session; |
| 234 | int ret = 0; |
| 235 | |
| 236 | set_bit(NFS4_SESSION_DRAINING, &ses->session_state); |
| 237 | /* back channel */ |
| 238 | ret = nfs4_wait_on_slot_tbl(&ses->bc_slot_table); |
| 239 | if (ret) |
| 240 | return ret; |
| 241 | /* fore channel */ |
| 242 | return nfs4_wait_on_slot_tbl(&ses->fc_slot_table); |
| 243 | } |
| 244 | |
| 245 | int nfs41_init_clientid(struct nfs_client *clp, struct rpc_cred *cred) |
| 246 | { |
| 247 | int status; |
| 248 | |
| 249 | if (test_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state)) |
| 250 | goto do_confirm; |
| 251 | nfs4_begin_drain_session(clp); |
| 252 | status = nfs4_proc_exchange_id(clp, cred); |
| 253 | if (status != 0) |
| 254 | goto out; |
| 255 | set_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); |
| 256 | do_confirm: |
| 257 | status = nfs4_proc_create_session(clp); |
| 258 | if (status != 0) |
| 259 | goto out; |
| 260 | clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); |
| 261 | nfs41_setup_state_renewal(clp); |
| 262 | nfs_mark_client_ready(clp, NFS_CS_READY); |
| 263 | out: |
| 264 | return status; |
| 265 | } |
| 266 | |
| 267 | struct rpc_cred *nfs4_get_exchange_id_cred(struct nfs_client *clp) |
| 268 | { |
| 269 | struct rpc_cred *cred; |
| 270 | |
| 271 | spin_lock(&clp->cl_lock); |
| 272 | cred = nfs4_get_machine_cred_locked(clp); |
| 273 | spin_unlock(&clp->cl_lock); |
| 274 | return cred; |
| 275 | } |
| 276 | |
| 277 | #endif /* CONFIG_NFS_V4_1 */ |
| 278 | |
| 279 | static struct rpc_cred * |
| 280 | nfs4_get_setclientid_cred_server(struct nfs_server *server) |
| 281 | { |
| 282 | struct nfs_client *clp = server->nfs_client; |
| 283 | struct rpc_cred *cred = NULL; |
| 284 | struct nfs4_state_owner *sp; |
| 285 | struct rb_node *pos; |
| 286 | |
| 287 | spin_lock(&clp->cl_lock); |
| 288 | pos = rb_first(&server->state_owners); |
| 289 | if (pos != NULL) { |
| 290 | sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); |
| 291 | cred = get_rpccred(sp->so_cred); |
| 292 | } |
| 293 | spin_unlock(&clp->cl_lock); |
| 294 | return cred; |
| 295 | } |
| 296 | |
| 297 | /** |
| 298 | * nfs4_get_setclientid_cred - Acquire credential for a setclientid operation |
| 299 | * @clp: client state handle |
| 300 | * |
| 301 | * Returns an rpc_cred with reference count bumped, or NULL. |
| 302 | */ |
| 303 | struct rpc_cred *nfs4_get_setclientid_cred(struct nfs_client *clp) |
| 304 | { |
| 305 | struct nfs_server *server; |
| 306 | struct rpc_cred *cred; |
| 307 | |
| 308 | spin_lock(&clp->cl_lock); |
| 309 | cred = nfs4_get_machine_cred_locked(clp); |
| 310 | spin_unlock(&clp->cl_lock); |
| 311 | if (cred != NULL) |
| 312 | goto out; |
| 313 | |
| 314 | rcu_read_lock(); |
| 315 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
| 316 | cred = nfs4_get_setclientid_cred_server(server); |
| 317 | if (cred != NULL) |
| 318 | break; |
| 319 | } |
| 320 | rcu_read_unlock(); |
| 321 | |
| 322 | out: |
| 323 | return cred; |
| 324 | } |
| 325 | |
| 326 | static struct nfs4_state_owner * |
| 327 | nfs4_find_state_owner_locked(struct nfs_server *server, struct rpc_cred *cred) |
| 328 | { |
| 329 | struct rb_node **p = &server->state_owners.rb_node, |
| 330 | *parent = NULL; |
| 331 | struct nfs4_state_owner *sp; |
| 332 | |
| 333 | while (*p != NULL) { |
| 334 | parent = *p; |
| 335 | sp = rb_entry(parent, struct nfs4_state_owner, so_server_node); |
| 336 | |
| 337 | if (cred < sp->so_cred) |
| 338 | p = &parent->rb_left; |
| 339 | else if (cred > sp->so_cred) |
| 340 | p = &parent->rb_right; |
| 341 | else { |
| 342 | if (!list_empty(&sp->so_lru)) |
| 343 | list_del_init(&sp->so_lru); |
| 344 | atomic_inc(&sp->so_count); |
| 345 | return sp; |
| 346 | } |
| 347 | } |
| 348 | return NULL; |
| 349 | } |
| 350 | |
| 351 | static struct nfs4_state_owner * |
| 352 | nfs4_insert_state_owner_locked(struct nfs4_state_owner *new) |
| 353 | { |
| 354 | struct nfs_server *server = new->so_server; |
| 355 | struct rb_node **p = &server->state_owners.rb_node, |
| 356 | *parent = NULL; |
| 357 | struct nfs4_state_owner *sp; |
| 358 | int err; |
| 359 | |
| 360 | while (*p != NULL) { |
| 361 | parent = *p; |
| 362 | sp = rb_entry(parent, struct nfs4_state_owner, so_server_node); |
| 363 | |
| 364 | if (new->so_cred < sp->so_cred) |
| 365 | p = &parent->rb_left; |
| 366 | else if (new->so_cred > sp->so_cred) |
| 367 | p = &parent->rb_right; |
| 368 | else { |
| 369 | if (!list_empty(&sp->so_lru)) |
| 370 | list_del_init(&sp->so_lru); |
| 371 | atomic_inc(&sp->so_count); |
| 372 | return sp; |
| 373 | } |
| 374 | } |
| 375 | err = ida_get_new(&server->openowner_id, &new->so_seqid.owner_id); |
| 376 | if (err) |
| 377 | return ERR_PTR(err); |
| 378 | rb_link_node(&new->so_server_node, parent, p); |
| 379 | rb_insert_color(&new->so_server_node, &server->state_owners); |
| 380 | return new; |
| 381 | } |
| 382 | |
| 383 | static void |
| 384 | nfs4_remove_state_owner_locked(struct nfs4_state_owner *sp) |
| 385 | { |
| 386 | struct nfs_server *server = sp->so_server; |
| 387 | |
| 388 | if (!RB_EMPTY_NODE(&sp->so_server_node)) |
| 389 | rb_erase(&sp->so_server_node, &server->state_owners); |
| 390 | ida_remove(&server->openowner_id, sp->so_seqid.owner_id); |
| 391 | } |
| 392 | |
| 393 | static void |
| 394 | nfs4_init_seqid_counter(struct nfs_seqid_counter *sc) |
| 395 | { |
| 396 | sc->create_time = ktime_get(); |
| 397 | sc->flags = 0; |
| 398 | sc->counter = 0; |
| 399 | spin_lock_init(&sc->lock); |
| 400 | INIT_LIST_HEAD(&sc->list); |
| 401 | rpc_init_wait_queue(&sc->wait, "Seqid_waitqueue"); |
| 402 | } |
| 403 | |
| 404 | static void |
| 405 | nfs4_destroy_seqid_counter(struct nfs_seqid_counter *sc) |
| 406 | { |
| 407 | rpc_destroy_wait_queue(&sc->wait); |
| 408 | } |
| 409 | |
| 410 | /* |
| 411 | * nfs4_alloc_state_owner(): this is called on the OPEN or CREATE path to |
| 412 | * create a new state_owner. |
| 413 | * |
| 414 | */ |
| 415 | static struct nfs4_state_owner * |
| 416 | nfs4_alloc_state_owner(struct nfs_server *server, |
| 417 | struct rpc_cred *cred, |
| 418 | gfp_t gfp_flags) |
| 419 | { |
| 420 | struct nfs4_state_owner *sp; |
| 421 | |
| 422 | sp = kzalloc(sizeof(*sp), gfp_flags); |
| 423 | if (!sp) |
| 424 | return NULL; |
| 425 | sp->so_server = server; |
| 426 | sp->so_cred = get_rpccred(cred); |
| 427 | spin_lock_init(&sp->so_lock); |
| 428 | INIT_LIST_HEAD(&sp->so_states); |
| 429 | nfs4_init_seqid_counter(&sp->so_seqid); |
| 430 | atomic_set(&sp->so_count, 1); |
| 431 | INIT_LIST_HEAD(&sp->so_lru); |
| 432 | return sp; |
| 433 | } |
| 434 | |
| 435 | static void |
| 436 | nfs4_drop_state_owner(struct nfs4_state_owner *sp) |
| 437 | { |
| 438 | struct rb_node *rb_node = &sp->so_server_node; |
| 439 | |
| 440 | if (!RB_EMPTY_NODE(rb_node)) { |
| 441 | struct nfs_server *server = sp->so_server; |
| 442 | struct nfs_client *clp = server->nfs_client; |
| 443 | |
| 444 | spin_lock(&clp->cl_lock); |
| 445 | if (!RB_EMPTY_NODE(rb_node)) { |
| 446 | rb_erase(rb_node, &server->state_owners); |
| 447 | RB_CLEAR_NODE(rb_node); |
| 448 | } |
| 449 | spin_unlock(&clp->cl_lock); |
| 450 | } |
| 451 | } |
| 452 | |
| 453 | static void nfs4_free_state_owner(struct nfs4_state_owner *sp) |
| 454 | { |
| 455 | nfs4_destroy_seqid_counter(&sp->so_seqid); |
| 456 | put_rpccred(sp->so_cred); |
| 457 | kfree(sp); |
| 458 | } |
| 459 | |
| 460 | static void nfs4_gc_state_owners(struct nfs_server *server) |
| 461 | { |
| 462 | struct nfs_client *clp = server->nfs_client; |
| 463 | struct nfs4_state_owner *sp, *tmp; |
| 464 | unsigned long time_min, time_max; |
| 465 | LIST_HEAD(doomed); |
| 466 | |
| 467 | spin_lock(&clp->cl_lock); |
| 468 | time_max = jiffies; |
| 469 | time_min = (long)time_max - (long)clp->cl_lease_time; |
| 470 | list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) { |
| 471 | /* NB: LRU is sorted so that oldest is at the head */ |
| 472 | if (time_in_range(sp->so_expires, time_min, time_max)) |
| 473 | break; |
| 474 | list_move(&sp->so_lru, &doomed); |
| 475 | nfs4_remove_state_owner_locked(sp); |
| 476 | } |
| 477 | spin_unlock(&clp->cl_lock); |
| 478 | |
| 479 | list_for_each_entry_safe(sp, tmp, &doomed, so_lru) { |
| 480 | list_del(&sp->so_lru); |
| 481 | nfs4_free_state_owner(sp); |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | /** |
| 486 | * nfs4_get_state_owner - Look up a state owner given a credential |
| 487 | * @server: nfs_server to search |
| 488 | * @cred: RPC credential to match |
| 489 | * |
| 490 | * Returns a pointer to an instantiated nfs4_state_owner struct, or NULL. |
| 491 | */ |
| 492 | struct nfs4_state_owner *nfs4_get_state_owner(struct nfs_server *server, |
| 493 | struct rpc_cred *cred, |
| 494 | gfp_t gfp_flags) |
| 495 | { |
| 496 | struct nfs_client *clp = server->nfs_client; |
| 497 | struct nfs4_state_owner *sp, *new; |
| 498 | |
| 499 | spin_lock(&clp->cl_lock); |
| 500 | sp = nfs4_find_state_owner_locked(server, cred); |
| 501 | spin_unlock(&clp->cl_lock); |
| 502 | if (sp != NULL) |
| 503 | goto out; |
| 504 | new = nfs4_alloc_state_owner(server, cred, gfp_flags); |
| 505 | if (new == NULL) |
| 506 | goto out; |
| 507 | do { |
| 508 | if (ida_pre_get(&server->openowner_id, gfp_flags) == 0) |
| 509 | break; |
| 510 | spin_lock(&clp->cl_lock); |
| 511 | sp = nfs4_insert_state_owner_locked(new); |
| 512 | spin_unlock(&clp->cl_lock); |
| 513 | } while (sp == ERR_PTR(-EAGAIN)); |
| 514 | if (sp != new) |
| 515 | nfs4_free_state_owner(new); |
| 516 | out: |
| 517 | nfs4_gc_state_owners(server); |
| 518 | return sp; |
| 519 | } |
| 520 | |
| 521 | /** |
| 522 | * nfs4_put_state_owner - Release a nfs4_state_owner |
| 523 | * @sp: state owner data to release |
| 524 | * |
| 525 | * Note that we keep released state owners on an LRU |
| 526 | * list. |
| 527 | * This caches valid state owners so that they can be |
| 528 | * reused, to avoid the OPEN_CONFIRM on minor version 0. |
| 529 | * It also pins the uniquifier of dropped state owners for |
| 530 | * a while, to ensure that those state owner names are |
| 531 | * never reused. |
| 532 | */ |
| 533 | void nfs4_put_state_owner(struct nfs4_state_owner *sp) |
| 534 | { |
| 535 | struct nfs_server *server = sp->so_server; |
| 536 | struct nfs_client *clp = server->nfs_client; |
| 537 | |
| 538 | if (!atomic_dec_and_lock(&sp->so_count, &clp->cl_lock)) |
| 539 | return; |
| 540 | |
| 541 | sp->so_expires = jiffies; |
| 542 | list_add_tail(&sp->so_lru, &server->state_owners_lru); |
| 543 | spin_unlock(&clp->cl_lock); |
| 544 | } |
| 545 | |
| 546 | /** |
| 547 | * nfs4_purge_state_owners - Release all cached state owners |
| 548 | * @server: nfs_server with cached state owners to release |
| 549 | * |
| 550 | * Called at umount time. Remaining state owners will be on |
| 551 | * the LRU with ref count of zero. |
| 552 | */ |
| 553 | void nfs4_purge_state_owners(struct nfs_server *server) |
| 554 | { |
| 555 | struct nfs_client *clp = server->nfs_client; |
| 556 | struct nfs4_state_owner *sp, *tmp; |
| 557 | LIST_HEAD(doomed); |
| 558 | |
| 559 | spin_lock(&clp->cl_lock); |
| 560 | list_for_each_entry_safe(sp, tmp, &server->state_owners_lru, so_lru) { |
| 561 | list_move(&sp->so_lru, &doomed); |
| 562 | nfs4_remove_state_owner_locked(sp); |
| 563 | } |
| 564 | spin_unlock(&clp->cl_lock); |
| 565 | |
| 566 | list_for_each_entry_safe(sp, tmp, &doomed, so_lru) { |
| 567 | list_del(&sp->so_lru); |
| 568 | nfs4_free_state_owner(sp); |
| 569 | } |
| 570 | } |
| 571 | |
| 572 | static struct nfs4_state * |
| 573 | nfs4_alloc_open_state(void) |
| 574 | { |
| 575 | struct nfs4_state *state; |
| 576 | |
| 577 | state = kzalloc(sizeof(*state), GFP_NOFS); |
| 578 | if (!state) |
| 579 | return NULL; |
| 580 | atomic_set(&state->count, 1); |
| 581 | INIT_LIST_HEAD(&state->lock_states); |
| 582 | spin_lock_init(&state->state_lock); |
| 583 | seqlock_init(&state->seqlock); |
| 584 | return state; |
| 585 | } |
| 586 | |
| 587 | void |
| 588 | nfs4_state_set_mode_locked(struct nfs4_state *state, fmode_t fmode) |
| 589 | { |
| 590 | if (state->state == fmode) |
| 591 | return; |
| 592 | /* NB! List reordering - see the reclaim code for why. */ |
| 593 | if ((fmode & FMODE_WRITE) != (state->state & FMODE_WRITE)) { |
| 594 | if (fmode & FMODE_WRITE) |
| 595 | list_move(&state->open_states, &state->owner->so_states); |
| 596 | else |
| 597 | list_move_tail(&state->open_states, &state->owner->so_states); |
| 598 | } |
| 599 | state->state = fmode; |
| 600 | } |
| 601 | |
| 602 | static struct nfs4_state * |
| 603 | __nfs4_find_state_byowner(struct inode *inode, struct nfs4_state_owner *owner) |
| 604 | { |
| 605 | struct nfs_inode *nfsi = NFS_I(inode); |
| 606 | struct nfs4_state *state; |
| 607 | |
| 608 | list_for_each_entry(state, &nfsi->open_states, inode_states) { |
| 609 | if (state->owner != owner) |
| 610 | continue; |
| 611 | if (atomic_inc_not_zero(&state->count)) |
| 612 | return state; |
| 613 | } |
| 614 | return NULL; |
| 615 | } |
| 616 | |
| 617 | static void |
| 618 | nfs4_free_open_state(struct nfs4_state *state) |
| 619 | { |
| 620 | kfree(state); |
| 621 | } |
| 622 | |
| 623 | struct nfs4_state * |
| 624 | nfs4_get_open_state(struct inode *inode, struct nfs4_state_owner *owner) |
| 625 | { |
| 626 | struct nfs4_state *state, *new; |
| 627 | struct nfs_inode *nfsi = NFS_I(inode); |
| 628 | |
| 629 | spin_lock(&inode->i_lock); |
| 630 | state = __nfs4_find_state_byowner(inode, owner); |
| 631 | spin_unlock(&inode->i_lock); |
| 632 | if (state) |
| 633 | goto out; |
| 634 | new = nfs4_alloc_open_state(); |
| 635 | spin_lock(&owner->so_lock); |
| 636 | spin_lock(&inode->i_lock); |
| 637 | state = __nfs4_find_state_byowner(inode, owner); |
| 638 | if (state == NULL && new != NULL) { |
| 639 | state = new; |
| 640 | state->owner = owner; |
| 641 | atomic_inc(&owner->so_count); |
| 642 | list_add(&state->inode_states, &nfsi->open_states); |
| 643 | ihold(inode); |
| 644 | state->inode = inode; |
| 645 | spin_unlock(&inode->i_lock); |
| 646 | /* Note: The reclaim code dictates that we add stateless |
| 647 | * and read-only stateids to the end of the list */ |
| 648 | list_add_tail(&state->open_states, &owner->so_states); |
| 649 | spin_unlock(&owner->so_lock); |
| 650 | } else { |
| 651 | spin_unlock(&inode->i_lock); |
| 652 | spin_unlock(&owner->so_lock); |
| 653 | if (new) |
| 654 | nfs4_free_open_state(new); |
| 655 | } |
| 656 | out: |
| 657 | return state; |
| 658 | } |
| 659 | |
| 660 | void nfs4_put_open_state(struct nfs4_state *state) |
| 661 | { |
| 662 | struct inode *inode = state->inode; |
| 663 | struct nfs4_state_owner *owner = state->owner; |
| 664 | |
| 665 | if (!atomic_dec_and_lock(&state->count, &owner->so_lock)) |
| 666 | return; |
| 667 | spin_lock(&inode->i_lock); |
| 668 | list_del(&state->inode_states); |
| 669 | list_del(&state->open_states); |
| 670 | spin_unlock(&inode->i_lock); |
| 671 | spin_unlock(&owner->so_lock); |
| 672 | iput(inode); |
| 673 | nfs4_free_open_state(state); |
| 674 | nfs4_put_state_owner(owner); |
| 675 | } |
| 676 | |
| 677 | /* |
| 678 | * Close the current file. |
| 679 | */ |
| 680 | static void __nfs4_close(struct nfs4_state *state, |
| 681 | fmode_t fmode, gfp_t gfp_mask, int wait) |
| 682 | { |
| 683 | struct nfs4_state_owner *owner = state->owner; |
| 684 | int call_close = 0; |
| 685 | fmode_t newstate; |
| 686 | |
| 687 | atomic_inc(&owner->so_count); |
| 688 | /* Protect against nfs4_find_state() */ |
| 689 | spin_lock(&owner->so_lock); |
| 690 | switch (fmode & (FMODE_READ | FMODE_WRITE)) { |
| 691 | case FMODE_READ: |
| 692 | state->n_rdonly--; |
| 693 | break; |
| 694 | case FMODE_WRITE: |
| 695 | state->n_wronly--; |
| 696 | break; |
| 697 | case FMODE_READ|FMODE_WRITE: |
| 698 | state->n_rdwr--; |
| 699 | } |
| 700 | newstate = FMODE_READ|FMODE_WRITE; |
| 701 | if (state->n_rdwr == 0) { |
| 702 | if (state->n_rdonly == 0) { |
| 703 | newstate &= ~FMODE_READ; |
| 704 | call_close |= test_bit(NFS_O_RDONLY_STATE, &state->flags); |
| 705 | call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); |
| 706 | } |
| 707 | if (state->n_wronly == 0) { |
| 708 | newstate &= ~FMODE_WRITE; |
| 709 | call_close |= test_bit(NFS_O_WRONLY_STATE, &state->flags); |
| 710 | call_close |= test_bit(NFS_O_RDWR_STATE, &state->flags); |
| 711 | } |
| 712 | if (newstate == 0) |
| 713 | clear_bit(NFS_DELEGATED_STATE, &state->flags); |
| 714 | } |
| 715 | nfs4_state_set_mode_locked(state, newstate); |
| 716 | spin_unlock(&owner->so_lock); |
| 717 | |
| 718 | if (!call_close) { |
| 719 | nfs4_put_open_state(state); |
| 720 | nfs4_put_state_owner(owner); |
| 721 | } else { |
| 722 | bool roc = pnfs_roc(state->inode); |
| 723 | |
| 724 | nfs4_do_close(state, gfp_mask, wait, roc); |
| 725 | } |
| 726 | } |
| 727 | |
| 728 | void nfs4_close_state(struct nfs4_state *state, fmode_t fmode) |
| 729 | { |
| 730 | __nfs4_close(state, fmode, GFP_NOFS, 0); |
| 731 | } |
| 732 | |
| 733 | void nfs4_close_sync(struct nfs4_state *state, fmode_t fmode) |
| 734 | { |
| 735 | __nfs4_close(state, fmode, GFP_KERNEL, 1); |
| 736 | } |
| 737 | |
| 738 | /* |
| 739 | * Search the state->lock_states for an existing lock_owner |
| 740 | * that is compatible with current->files |
| 741 | */ |
| 742 | static struct nfs4_lock_state * |
| 743 | __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_pid, unsigned int type) |
| 744 | { |
| 745 | struct nfs4_lock_state *pos; |
| 746 | list_for_each_entry(pos, &state->lock_states, ls_locks) { |
| 747 | if (type != NFS4_ANY_LOCK_TYPE && pos->ls_owner.lo_type != type) |
| 748 | continue; |
| 749 | switch (pos->ls_owner.lo_type) { |
| 750 | case NFS4_POSIX_LOCK_TYPE: |
| 751 | if (pos->ls_owner.lo_u.posix_owner != fl_owner) |
| 752 | continue; |
| 753 | break; |
| 754 | case NFS4_FLOCK_LOCK_TYPE: |
| 755 | if (pos->ls_owner.lo_u.flock_owner != fl_pid) |
| 756 | continue; |
| 757 | } |
| 758 | atomic_inc(&pos->ls_count); |
| 759 | return pos; |
| 760 | } |
| 761 | return NULL; |
| 762 | } |
| 763 | |
| 764 | /* |
| 765 | * Return a compatible lock_state. If no initialized lock_state structure |
| 766 | * exists, return an uninitialized one. |
| 767 | * |
| 768 | */ |
| 769 | 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) |
| 770 | { |
| 771 | struct nfs4_lock_state *lsp; |
| 772 | struct nfs_server *server = state->owner->so_server; |
| 773 | |
| 774 | lsp = kzalloc(sizeof(*lsp), GFP_NOFS); |
| 775 | if (lsp == NULL) |
| 776 | return NULL; |
| 777 | nfs4_init_seqid_counter(&lsp->ls_seqid); |
| 778 | atomic_set(&lsp->ls_count, 1); |
| 779 | lsp->ls_state = state; |
| 780 | lsp->ls_owner.lo_type = type; |
| 781 | switch (lsp->ls_owner.lo_type) { |
| 782 | case NFS4_FLOCK_LOCK_TYPE: |
| 783 | lsp->ls_owner.lo_u.flock_owner = fl_pid; |
| 784 | break; |
| 785 | case NFS4_POSIX_LOCK_TYPE: |
| 786 | lsp->ls_owner.lo_u.posix_owner = fl_owner; |
| 787 | break; |
| 788 | default: |
| 789 | goto out_free; |
| 790 | } |
| 791 | lsp->ls_seqid.owner_id = ida_simple_get(&server->lockowner_id, 0, 0, GFP_NOFS); |
| 792 | if (lsp->ls_seqid.owner_id < 0) |
| 793 | goto out_free; |
| 794 | INIT_LIST_HEAD(&lsp->ls_locks); |
| 795 | return lsp; |
| 796 | out_free: |
| 797 | kfree(lsp); |
| 798 | return NULL; |
| 799 | } |
| 800 | |
| 801 | void nfs4_free_lock_state(struct nfs_server *server, struct nfs4_lock_state *lsp) |
| 802 | { |
| 803 | ida_simple_remove(&server->lockowner_id, lsp->ls_seqid.owner_id); |
| 804 | nfs4_destroy_seqid_counter(&lsp->ls_seqid); |
| 805 | kfree(lsp); |
| 806 | } |
| 807 | |
| 808 | /* |
| 809 | * Return a compatible lock_state. If no initialized lock_state structure |
| 810 | * exists, return an uninitialized one. |
| 811 | * |
| 812 | */ |
| 813 | static struct nfs4_lock_state *nfs4_get_lock_state(struct nfs4_state *state, fl_owner_t owner, pid_t pid, unsigned int type) |
| 814 | { |
| 815 | struct nfs4_lock_state *lsp, *new = NULL; |
| 816 | |
| 817 | for(;;) { |
| 818 | spin_lock(&state->state_lock); |
| 819 | lsp = __nfs4_find_lock_state(state, owner, pid, type); |
| 820 | if (lsp != NULL) |
| 821 | break; |
| 822 | if (new != NULL) { |
| 823 | list_add(&new->ls_locks, &state->lock_states); |
| 824 | set_bit(LK_STATE_IN_USE, &state->flags); |
| 825 | lsp = new; |
| 826 | new = NULL; |
| 827 | break; |
| 828 | } |
| 829 | spin_unlock(&state->state_lock); |
| 830 | new = nfs4_alloc_lock_state(state, owner, pid, type); |
| 831 | if (new == NULL) |
| 832 | return NULL; |
| 833 | } |
| 834 | spin_unlock(&state->state_lock); |
| 835 | if (new != NULL) |
| 836 | nfs4_free_lock_state(state->owner->so_server, new); |
| 837 | return lsp; |
| 838 | } |
| 839 | |
| 840 | /* |
| 841 | * Release reference to lock_state, and free it if we see that |
| 842 | * it is no longer in use |
| 843 | */ |
| 844 | void nfs4_put_lock_state(struct nfs4_lock_state *lsp) |
| 845 | { |
| 846 | struct nfs4_state *state; |
| 847 | |
| 848 | if (lsp == NULL) |
| 849 | return; |
| 850 | state = lsp->ls_state; |
| 851 | if (!atomic_dec_and_lock(&lsp->ls_count, &state->state_lock)) |
| 852 | return; |
| 853 | list_del(&lsp->ls_locks); |
| 854 | if (list_empty(&state->lock_states)) |
| 855 | clear_bit(LK_STATE_IN_USE, &state->flags); |
| 856 | spin_unlock(&state->state_lock); |
| 857 | if (lsp->ls_flags & NFS_LOCK_INITIALIZED) { |
| 858 | if (nfs4_release_lockowner(lsp) == 0) |
| 859 | return; |
| 860 | } |
| 861 | nfs4_free_lock_state(lsp->ls_state->owner->so_server, lsp); |
| 862 | } |
| 863 | |
| 864 | static void nfs4_fl_copy_lock(struct file_lock *dst, struct file_lock *src) |
| 865 | { |
| 866 | struct nfs4_lock_state *lsp = src->fl_u.nfs4_fl.owner; |
| 867 | |
| 868 | dst->fl_u.nfs4_fl.owner = lsp; |
| 869 | atomic_inc(&lsp->ls_count); |
| 870 | } |
| 871 | |
| 872 | static void nfs4_fl_release_lock(struct file_lock *fl) |
| 873 | { |
| 874 | nfs4_put_lock_state(fl->fl_u.nfs4_fl.owner); |
| 875 | } |
| 876 | |
| 877 | static const struct file_lock_operations nfs4_fl_lock_ops = { |
| 878 | .fl_copy_lock = nfs4_fl_copy_lock, |
| 879 | .fl_release_private = nfs4_fl_release_lock, |
| 880 | }; |
| 881 | |
| 882 | int nfs4_set_lock_state(struct nfs4_state *state, struct file_lock *fl) |
| 883 | { |
| 884 | struct nfs4_lock_state *lsp; |
| 885 | |
| 886 | if (fl->fl_ops != NULL) |
| 887 | return 0; |
| 888 | if (fl->fl_flags & FL_POSIX) |
| 889 | lsp = nfs4_get_lock_state(state, fl->fl_owner, 0, NFS4_POSIX_LOCK_TYPE); |
| 890 | else if (fl->fl_flags & FL_FLOCK) |
| 891 | lsp = nfs4_get_lock_state(state, NULL, fl->fl_pid, |
| 892 | NFS4_FLOCK_LOCK_TYPE); |
| 893 | else |
| 894 | return -EINVAL; |
| 895 | if (lsp == NULL) |
| 896 | return -ENOMEM; |
| 897 | fl->fl_u.nfs4_fl.owner = lsp; |
| 898 | fl->fl_ops = &nfs4_fl_lock_ops; |
| 899 | return 0; |
| 900 | } |
| 901 | |
| 902 | static bool nfs4_copy_lock_stateid(nfs4_stateid *dst, struct nfs4_state *state, |
| 903 | fl_owner_t fl_owner, pid_t fl_pid) |
| 904 | { |
| 905 | struct nfs4_lock_state *lsp; |
| 906 | bool ret = false; |
| 907 | |
| 908 | if (test_bit(LK_STATE_IN_USE, &state->flags) == 0) |
| 909 | goto out; |
| 910 | |
| 911 | spin_lock(&state->state_lock); |
| 912 | lsp = __nfs4_find_lock_state(state, fl_owner, fl_pid, NFS4_ANY_LOCK_TYPE); |
| 913 | if (lsp != NULL && (lsp->ls_flags & NFS_LOCK_INITIALIZED) != 0) { |
| 914 | nfs4_stateid_copy(dst, &lsp->ls_stateid); |
| 915 | ret = true; |
| 916 | } |
| 917 | spin_unlock(&state->state_lock); |
| 918 | nfs4_put_lock_state(lsp); |
| 919 | out: |
| 920 | return ret; |
| 921 | } |
| 922 | |
| 923 | static void nfs4_copy_open_stateid(nfs4_stateid *dst, struct nfs4_state *state) |
| 924 | { |
| 925 | int seq; |
| 926 | |
| 927 | do { |
| 928 | seq = read_seqbegin(&state->seqlock); |
| 929 | nfs4_stateid_copy(dst, &state->stateid); |
| 930 | } while (read_seqretry(&state->seqlock, seq)); |
| 931 | } |
| 932 | |
| 933 | /* |
| 934 | * Byte-range lock aware utility to initialize the stateid of read/write |
| 935 | * requests. |
| 936 | */ |
| 937 | void nfs4_select_rw_stateid(nfs4_stateid *dst, struct nfs4_state *state, |
| 938 | fmode_t fmode, fl_owner_t fl_owner, pid_t fl_pid) |
| 939 | { |
| 940 | if (nfs4_copy_delegation_stateid(dst, state->inode, fmode)) |
| 941 | return; |
| 942 | if (nfs4_copy_lock_stateid(dst, state, fl_owner, fl_pid)) |
| 943 | return; |
| 944 | nfs4_copy_open_stateid(dst, state); |
| 945 | } |
| 946 | |
| 947 | struct nfs_seqid *nfs_alloc_seqid(struct nfs_seqid_counter *counter, gfp_t gfp_mask) |
| 948 | { |
| 949 | struct nfs_seqid *new; |
| 950 | |
| 951 | new = kmalloc(sizeof(*new), gfp_mask); |
| 952 | if (new != NULL) { |
| 953 | new->sequence = counter; |
| 954 | INIT_LIST_HEAD(&new->list); |
| 955 | new->task = NULL; |
| 956 | } |
| 957 | return new; |
| 958 | } |
| 959 | |
| 960 | void nfs_release_seqid(struct nfs_seqid *seqid) |
| 961 | { |
| 962 | struct nfs_seqid_counter *sequence; |
| 963 | |
| 964 | if (list_empty(&seqid->list)) |
| 965 | return; |
| 966 | sequence = seqid->sequence; |
| 967 | spin_lock(&sequence->lock); |
| 968 | list_del_init(&seqid->list); |
| 969 | if (!list_empty(&sequence->list)) { |
| 970 | struct nfs_seqid *next; |
| 971 | |
| 972 | next = list_first_entry(&sequence->list, |
| 973 | struct nfs_seqid, list); |
| 974 | rpc_wake_up_queued_task(&sequence->wait, next->task); |
| 975 | } |
| 976 | spin_unlock(&sequence->lock); |
| 977 | } |
| 978 | |
| 979 | void nfs_free_seqid(struct nfs_seqid *seqid) |
| 980 | { |
| 981 | nfs_release_seqid(seqid); |
| 982 | kfree(seqid); |
| 983 | } |
| 984 | |
| 985 | /* |
| 986 | * Increment the seqid if the OPEN/OPEN_DOWNGRADE/CLOSE succeeded, or |
| 987 | * failed with a seqid incrementing error - |
| 988 | * see comments nfs_fs.h:seqid_mutating_error() |
| 989 | */ |
| 990 | static void nfs_increment_seqid(int status, struct nfs_seqid *seqid) |
| 991 | { |
| 992 | BUG_ON(list_first_entry(&seqid->sequence->list, struct nfs_seqid, list) != seqid); |
| 993 | switch (status) { |
| 994 | case 0: |
| 995 | break; |
| 996 | case -NFS4ERR_BAD_SEQID: |
| 997 | if (seqid->sequence->flags & NFS_SEQID_CONFIRMED) |
| 998 | return; |
| 999 | pr_warn_ratelimited("NFS: v4 server returned a bad" |
| 1000 | " sequence-id error on an" |
| 1001 | " unconfirmed sequence %p!\n", |
| 1002 | seqid->sequence); |
| 1003 | case -NFS4ERR_STALE_CLIENTID: |
| 1004 | case -NFS4ERR_STALE_STATEID: |
| 1005 | case -NFS4ERR_BAD_STATEID: |
| 1006 | case -NFS4ERR_BADXDR: |
| 1007 | case -NFS4ERR_RESOURCE: |
| 1008 | case -NFS4ERR_NOFILEHANDLE: |
| 1009 | /* Non-seqid mutating errors */ |
| 1010 | return; |
| 1011 | }; |
| 1012 | /* |
| 1013 | * Note: no locking needed as we are guaranteed to be first |
| 1014 | * on the sequence list |
| 1015 | */ |
| 1016 | seqid->sequence->counter++; |
| 1017 | } |
| 1018 | |
| 1019 | void nfs_increment_open_seqid(int status, struct nfs_seqid *seqid) |
| 1020 | { |
| 1021 | struct nfs4_state_owner *sp = container_of(seqid->sequence, |
| 1022 | struct nfs4_state_owner, so_seqid); |
| 1023 | struct nfs_server *server = sp->so_server; |
| 1024 | |
| 1025 | if (status == -NFS4ERR_BAD_SEQID) |
| 1026 | nfs4_drop_state_owner(sp); |
| 1027 | if (!nfs4_has_session(server->nfs_client)) |
| 1028 | nfs_increment_seqid(status, seqid); |
| 1029 | } |
| 1030 | |
| 1031 | /* |
| 1032 | * Increment the seqid if the LOCK/LOCKU succeeded, or |
| 1033 | * failed with a seqid incrementing error - |
| 1034 | * see comments nfs_fs.h:seqid_mutating_error() |
| 1035 | */ |
| 1036 | void nfs_increment_lock_seqid(int status, struct nfs_seqid *seqid) |
| 1037 | { |
| 1038 | nfs_increment_seqid(status, seqid); |
| 1039 | } |
| 1040 | |
| 1041 | int nfs_wait_on_sequence(struct nfs_seqid *seqid, struct rpc_task *task) |
| 1042 | { |
| 1043 | struct nfs_seqid_counter *sequence = seqid->sequence; |
| 1044 | int status = 0; |
| 1045 | |
| 1046 | spin_lock(&sequence->lock); |
| 1047 | seqid->task = task; |
| 1048 | if (list_empty(&seqid->list)) |
| 1049 | list_add_tail(&seqid->list, &sequence->list); |
| 1050 | if (list_first_entry(&sequence->list, struct nfs_seqid, list) == seqid) |
| 1051 | goto unlock; |
| 1052 | rpc_sleep_on(&sequence->wait, task, NULL); |
| 1053 | status = -EAGAIN; |
| 1054 | unlock: |
| 1055 | spin_unlock(&sequence->lock); |
| 1056 | return status; |
| 1057 | } |
| 1058 | |
| 1059 | static int nfs4_run_state_manager(void *); |
| 1060 | |
| 1061 | static void nfs4_clear_state_manager_bit(struct nfs_client *clp) |
| 1062 | { |
| 1063 | smp_mb__before_clear_bit(); |
| 1064 | clear_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state); |
| 1065 | smp_mb__after_clear_bit(); |
| 1066 | wake_up_bit(&clp->cl_state, NFS4CLNT_MANAGER_RUNNING); |
| 1067 | rpc_wake_up(&clp->cl_rpcwaitq); |
| 1068 | } |
| 1069 | |
| 1070 | /* |
| 1071 | * Schedule the nfs_client asynchronous state management routine |
| 1072 | */ |
| 1073 | void nfs4_schedule_state_manager(struct nfs_client *clp) |
| 1074 | { |
| 1075 | struct task_struct *task; |
| 1076 | char buf[INET6_ADDRSTRLEN + sizeof("-manager") + 1]; |
| 1077 | |
| 1078 | if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0) |
| 1079 | return; |
| 1080 | __module_get(THIS_MODULE); |
| 1081 | atomic_inc(&clp->cl_count); |
| 1082 | |
| 1083 | /* The rcu_read_lock() is not strictly necessary, as the state |
| 1084 | * manager is the only thread that ever changes the rpc_xprt |
| 1085 | * after it's initialized. At this point, we're single threaded. */ |
| 1086 | rcu_read_lock(); |
| 1087 | snprintf(buf, sizeof(buf), "%s-manager", |
| 1088 | rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_ADDR)); |
| 1089 | rcu_read_unlock(); |
| 1090 | task = kthread_run(nfs4_run_state_manager, clp, buf); |
| 1091 | if (IS_ERR(task)) { |
| 1092 | printk(KERN_ERR "%s: kthread_run: %ld\n", |
| 1093 | __func__, PTR_ERR(task)); |
| 1094 | nfs4_clear_state_manager_bit(clp); |
| 1095 | nfs_put_client(clp); |
| 1096 | module_put(THIS_MODULE); |
| 1097 | } |
| 1098 | } |
| 1099 | |
| 1100 | /* |
| 1101 | * Schedule a lease recovery attempt |
| 1102 | */ |
| 1103 | void nfs4_schedule_lease_recovery(struct nfs_client *clp) |
| 1104 | { |
| 1105 | if (!clp) |
| 1106 | return; |
| 1107 | if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) |
| 1108 | set_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); |
| 1109 | nfs4_schedule_state_manager(clp); |
| 1110 | } |
| 1111 | EXPORT_SYMBOL_GPL(nfs4_schedule_lease_recovery); |
| 1112 | |
| 1113 | /* |
| 1114 | * nfs40_handle_cb_pathdown - return all delegations after NFS4ERR_CB_PATH_DOWN |
| 1115 | * @clp: client to process |
| 1116 | * |
| 1117 | * Set the NFS4CLNT_LEASE_EXPIRED state in order to force a |
| 1118 | * resend of the SETCLIENTID and hence re-establish the |
| 1119 | * callback channel. Then return all existing delegations. |
| 1120 | */ |
| 1121 | static void nfs40_handle_cb_pathdown(struct nfs_client *clp) |
| 1122 | { |
| 1123 | set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); |
| 1124 | nfs_expire_all_delegations(clp); |
| 1125 | } |
| 1126 | |
| 1127 | void nfs4_schedule_path_down_recovery(struct nfs_client *clp) |
| 1128 | { |
| 1129 | nfs40_handle_cb_pathdown(clp); |
| 1130 | nfs4_schedule_state_manager(clp); |
| 1131 | } |
| 1132 | |
| 1133 | static int nfs4_state_mark_reclaim_reboot(struct nfs_client *clp, struct nfs4_state *state) |
| 1134 | { |
| 1135 | |
| 1136 | set_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); |
| 1137 | /* Don't recover state that expired before the reboot */ |
| 1138 | if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags)) { |
| 1139 | clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); |
| 1140 | return 0; |
| 1141 | } |
| 1142 | set_bit(NFS_OWNER_RECLAIM_REBOOT, &state->owner->so_flags); |
| 1143 | set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state); |
| 1144 | return 1; |
| 1145 | } |
| 1146 | |
| 1147 | static int nfs4_state_mark_reclaim_nograce(struct nfs_client *clp, struct nfs4_state *state) |
| 1148 | { |
| 1149 | set_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags); |
| 1150 | clear_bit(NFS_STATE_RECLAIM_REBOOT, &state->flags); |
| 1151 | set_bit(NFS_OWNER_RECLAIM_NOGRACE, &state->owner->so_flags); |
| 1152 | set_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state); |
| 1153 | return 1; |
| 1154 | } |
| 1155 | |
| 1156 | void nfs4_schedule_stateid_recovery(const struct nfs_server *server, struct nfs4_state *state) |
| 1157 | { |
| 1158 | struct nfs_client *clp = server->nfs_client; |
| 1159 | |
| 1160 | nfs4_state_mark_reclaim_nograce(clp, state); |
| 1161 | nfs4_schedule_state_manager(clp); |
| 1162 | } |
| 1163 | EXPORT_SYMBOL_GPL(nfs4_schedule_stateid_recovery); |
| 1164 | |
| 1165 | void nfs_inode_find_state_and_recover(struct inode *inode, |
| 1166 | const nfs4_stateid *stateid) |
| 1167 | { |
| 1168 | struct nfs_client *clp = NFS_SERVER(inode)->nfs_client; |
| 1169 | struct nfs_inode *nfsi = NFS_I(inode); |
| 1170 | struct nfs_open_context *ctx; |
| 1171 | struct nfs4_state *state; |
| 1172 | bool found = false; |
| 1173 | |
| 1174 | spin_lock(&inode->i_lock); |
| 1175 | list_for_each_entry(ctx, &nfsi->open_files, list) { |
| 1176 | state = ctx->state; |
| 1177 | if (state == NULL) |
| 1178 | continue; |
| 1179 | if (!test_bit(NFS_DELEGATED_STATE, &state->flags)) |
| 1180 | continue; |
| 1181 | if (!nfs4_stateid_match(&state->stateid, stateid)) |
| 1182 | continue; |
| 1183 | nfs4_state_mark_reclaim_nograce(clp, state); |
| 1184 | found = true; |
| 1185 | } |
| 1186 | spin_unlock(&inode->i_lock); |
| 1187 | if (found) |
| 1188 | nfs4_schedule_state_manager(clp); |
| 1189 | } |
| 1190 | |
| 1191 | |
| 1192 | static int nfs4_reclaim_locks(struct nfs4_state *state, const struct nfs4_state_recovery_ops *ops) |
| 1193 | { |
| 1194 | struct inode *inode = state->inode; |
| 1195 | struct nfs_inode *nfsi = NFS_I(inode); |
| 1196 | struct file_lock *fl; |
| 1197 | int status = 0; |
| 1198 | |
| 1199 | if (inode->i_flock == NULL) |
| 1200 | return 0; |
| 1201 | |
| 1202 | /* Guard against delegation returns and new lock/unlock calls */ |
| 1203 | down_write(&nfsi->rwsem); |
| 1204 | /* Protect inode->i_flock using the BKL */ |
| 1205 | lock_flocks(); |
| 1206 | for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { |
| 1207 | if (!(fl->fl_flags & (FL_POSIX|FL_FLOCK))) |
| 1208 | continue; |
| 1209 | if (nfs_file_open_context(fl->fl_file)->state != state) |
| 1210 | continue; |
| 1211 | unlock_flocks(); |
| 1212 | status = ops->recover_lock(state, fl); |
| 1213 | switch (status) { |
| 1214 | case 0: |
| 1215 | break; |
| 1216 | case -ESTALE: |
| 1217 | case -NFS4ERR_ADMIN_REVOKED: |
| 1218 | case -NFS4ERR_STALE_STATEID: |
| 1219 | case -NFS4ERR_BAD_STATEID: |
| 1220 | case -NFS4ERR_EXPIRED: |
| 1221 | case -NFS4ERR_NO_GRACE: |
| 1222 | case -NFS4ERR_STALE_CLIENTID: |
| 1223 | case -NFS4ERR_BADSESSION: |
| 1224 | case -NFS4ERR_BADSLOT: |
| 1225 | case -NFS4ERR_BAD_HIGH_SLOT: |
| 1226 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: |
| 1227 | goto out; |
| 1228 | default: |
| 1229 | printk(KERN_ERR "NFS: %s: unhandled error %d. " |
| 1230 | "Zeroing state\n", __func__, status); |
| 1231 | case -ENOMEM: |
| 1232 | case -NFS4ERR_DENIED: |
| 1233 | case -NFS4ERR_RECLAIM_BAD: |
| 1234 | case -NFS4ERR_RECLAIM_CONFLICT: |
| 1235 | /* kill_proc(fl->fl_pid, SIGLOST, 1); */ |
| 1236 | status = 0; |
| 1237 | } |
| 1238 | lock_flocks(); |
| 1239 | } |
| 1240 | unlock_flocks(); |
| 1241 | out: |
| 1242 | up_write(&nfsi->rwsem); |
| 1243 | return status; |
| 1244 | } |
| 1245 | |
| 1246 | static int nfs4_reclaim_open_state(struct nfs4_state_owner *sp, const struct nfs4_state_recovery_ops *ops) |
| 1247 | { |
| 1248 | struct nfs4_state *state; |
| 1249 | struct nfs4_lock_state *lock; |
| 1250 | int status = 0; |
| 1251 | |
| 1252 | /* Note: we rely on the sp->so_states list being ordered |
| 1253 | * so that we always reclaim open(O_RDWR) and/or open(O_WRITE) |
| 1254 | * states first. |
| 1255 | * This is needed to ensure that the server won't give us any |
| 1256 | * read delegations that we have to return if, say, we are |
| 1257 | * recovering after a network partition or a reboot from a |
| 1258 | * server that doesn't support a grace period. |
| 1259 | */ |
| 1260 | restart: |
| 1261 | spin_lock(&sp->so_lock); |
| 1262 | list_for_each_entry(state, &sp->so_states, open_states) { |
| 1263 | if (!test_and_clear_bit(ops->state_flag_bit, &state->flags)) |
| 1264 | continue; |
| 1265 | if (state->state == 0) |
| 1266 | continue; |
| 1267 | atomic_inc(&state->count); |
| 1268 | spin_unlock(&sp->so_lock); |
| 1269 | status = ops->recover_open(sp, state); |
| 1270 | if (status >= 0) { |
| 1271 | status = nfs4_reclaim_locks(state, ops); |
| 1272 | if (status >= 0) { |
| 1273 | spin_lock(&state->state_lock); |
| 1274 | list_for_each_entry(lock, &state->lock_states, ls_locks) { |
| 1275 | if (!(lock->ls_flags & NFS_LOCK_INITIALIZED)) |
| 1276 | pr_warn_ratelimited("NFS: " |
| 1277 | "%s: Lock reclaim " |
| 1278 | "failed!\n", __func__); |
| 1279 | } |
| 1280 | spin_unlock(&state->state_lock); |
| 1281 | nfs4_put_open_state(state); |
| 1282 | clear_bit(NFS_STATE_RECLAIM_NOGRACE, |
| 1283 | &state->flags); |
| 1284 | goto restart; |
| 1285 | } |
| 1286 | } |
| 1287 | switch (status) { |
| 1288 | default: |
| 1289 | printk(KERN_ERR "NFS: %s: unhandled error %d. " |
| 1290 | "Zeroing state\n", __func__, status); |
| 1291 | case -ENOENT: |
| 1292 | case -ENOMEM: |
| 1293 | case -ESTALE: |
| 1294 | /* |
| 1295 | * Open state on this file cannot be recovered |
| 1296 | * All we can do is revert to using the zero stateid. |
| 1297 | */ |
| 1298 | memset(&state->stateid, 0, |
| 1299 | sizeof(state->stateid)); |
| 1300 | /* Mark the file as being 'closed' */ |
| 1301 | state->state = 0; |
| 1302 | break; |
| 1303 | case -NFS4ERR_ADMIN_REVOKED: |
| 1304 | case -NFS4ERR_STALE_STATEID: |
| 1305 | case -NFS4ERR_BAD_STATEID: |
| 1306 | case -NFS4ERR_RECLAIM_BAD: |
| 1307 | case -NFS4ERR_RECLAIM_CONFLICT: |
| 1308 | nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state); |
| 1309 | break; |
| 1310 | case -NFS4ERR_EXPIRED: |
| 1311 | case -NFS4ERR_NO_GRACE: |
| 1312 | nfs4_state_mark_reclaim_nograce(sp->so_server->nfs_client, state); |
| 1313 | case -NFS4ERR_STALE_CLIENTID: |
| 1314 | case -NFS4ERR_BADSESSION: |
| 1315 | case -NFS4ERR_BADSLOT: |
| 1316 | case -NFS4ERR_BAD_HIGH_SLOT: |
| 1317 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: |
| 1318 | goto out_err; |
| 1319 | } |
| 1320 | nfs4_put_open_state(state); |
| 1321 | goto restart; |
| 1322 | } |
| 1323 | spin_unlock(&sp->so_lock); |
| 1324 | return 0; |
| 1325 | out_err: |
| 1326 | nfs4_put_open_state(state); |
| 1327 | return status; |
| 1328 | } |
| 1329 | |
| 1330 | static void nfs4_clear_open_state(struct nfs4_state *state) |
| 1331 | { |
| 1332 | struct nfs4_lock_state *lock; |
| 1333 | |
| 1334 | clear_bit(NFS_DELEGATED_STATE, &state->flags); |
| 1335 | clear_bit(NFS_O_RDONLY_STATE, &state->flags); |
| 1336 | clear_bit(NFS_O_WRONLY_STATE, &state->flags); |
| 1337 | clear_bit(NFS_O_RDWR_STATE, &state->flags); |
| 1338 | spin_lock(&state->state_lock); |
| 1339 | list_for_each_entry(lock, &state->lock_states, ls_locks) { |
| 1340 | lock->ls_seqid.flags = 0; |
| 1341 | lock->ls_flags &= ~NFS_LOCK_INITIALIZED; |
| 1342 | } |
| 1343 | spin_unlock(&state->state_lock); |
| 1344 | } |
| 1345 | |
| 1346 | static void nfs4_reset_seqids(struct nfs_server *server, |
| 1347 | int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state)) |
| 1348 | { |
| 1349 | struct nfs_client *clp = server->nfs_client; |
| 1350 | struct nfs4_state_owner *sp; |
| 1351 | struct rb_node *pos; |
| 1352 | struct nfs4_state *state; |
| 1353 | |
| 1354 | spin_lock(&clp->cl_lock); |
| 1355 | for (pos = rb_first(&server->state_owners); |
| 1356 | pos != NULL; |
| 1357 | pos = rb_next(pos)) { |
| 1358 | sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); |
| 1359 | sp->so_seqid.flags = 0; |
| 1360 | spin_lock(&sp->so_lock); |
| 1361 | list_for_each_entry(state, &sp->so_states, open_states) { |
| 1362 | if (mark_reclaim(clp, state)) |
| 1363 | nfs4_clear_open_state(state); |
| 1364 | } |
| 1365 | spin_unlock(&sp->so_lock); |
| 1366 | } |
| 1367 | spin_unlock(&clp->cl_lock); |
| 1368 | } |
| 1369 | |
| 1370 | static void nfs4_state_mark_reclaim_helper(struct nfs_client *clp, |
| 1371 | int (*mark_reclaim)(struct nfs_client *clp, struct nfs4_state *state)) |
| 1372 | { |
| 1373 | struct nfs_server *server; |
| 1374 | |
| 1375 | rcu_read_lock(); |
| 1376 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) |
| 1377 | nfs4_reset_seqids(server, mark_reclaim); |
| 1378 | rcu_read_unlock(); |
| 1379 | } |
| 1380 | |
| 1381 | static void nfs4_state_start_reclaim_reboot(struct nfs_client *clp) |
| 1382 | { |
| 1383 | /* Mark all delegations for reclaim */ |
| 1384 | nfs_delegation_mark_reclaim(clp); |
| 1385 | nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_reboot); |
| 1386 | } |
| 1387 | |
| 1388 | static void nfs4_reclaim_complete(struct nfs_client *clp, |
| 1389 | const struct nfs4_state_recovery_ops *ops) |
| 1390 | { |
| 1391 | /* Notify the server we're done reclaiming our state */ |
| 1392 | if (ops->reclaim_complete) |
| 1393 | (void)ops->reclaim_complete(clp); |
| 1394 | } |
| 1395 | |
| 1396 | static void nfs4_clear_reclaim_server(struct nfs_server *server) |
| 1397 | { |
| 1398 | struct nfs_client *clp = server->nfs_client; |
| 1399 | struct nfs4_state_owner *sp; |
| 1400 | struct rb_node *pos; |
| 1401 | struct nfs4_state *state; |
| 1402 | |
| 1403 | spin_lock(&clp->cl_lock); |
| 1404 | for (pos = rb_first(&server->state_owners); |
| 1405 | pos != NULL; |
| 1406 | pos = rb_next(pos)) { |
| 1407 | sp = rb_entry(pos, struct nfs4_state_owner, so_server_node); |
| 1408 | spin_lock(&sp->so_lock); |
| 1409 | list_for_each_entry(state, &sp->so_states, open_states) { |
| 1410 | if (!test_and_clear_bit(NFS_STATE_RECLAIM_REBOOT, |
| 1411 | &state->flags)) |
| 1412 | continue; |
| 1413 | nfs4_state_mark_reclaim_nograce(clp, state); |
| 1414 | } |
| 1415 | spin_unlock(&sp->so_lock); |
| 1416 | } |
| 1417 | spin_unlock(&clp->cl_lock); |
| 1418 | } |
| 1419 | |
| 1420 | static int nfs4_state_clear_reclaim_reboot(struct nfs_client *clp) |
| 1421 | { |
| 1422 | struct nfs_server *server; |
| 1423 | |
| 1424 | if (!test_and_clear_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) |
| 1425 | return 0; |
| 1426 | |
| 1427 | rcu_read_lock(); |
| 1428 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) |
| 1429 | nfs4_clear_reclaim_server(server); |
| 1430 | rcu_read_unlock(); |
| 1431 | |
| 1432 | nfs_delegation_reap_unclaimed(clp); |
| 1433 | return 1; |
| 1434 | } |
| 1435 | |
| 1436 | static void nfs4_state_end_reclaim_reboot(struct nfs_client *clp) |
| 1437 | { |
| 1438 | if (!nfs4_state_clear_reclaim_reboot(clp)) |
| 1439 | return; |
| 1440 | nfs4_reclaim_complete(clp, clp->cl_mvops->reboot_recovery_ops); |
| 1441 | } |
| 1442 | |
| 1443 | static void nfs_delegation_clear_all(struct nfs_client *clp) |
| 1444 | { |
| 1445 | nfs_delegation_mark_reclaim(clp); |
| 1446 | nfs_delegation_reap_unclaimed(clp); |
| 1447 | } |
| 1448 | |
| 1449 | static void nfs4_state_start_reclaim_nograce(struct nfs_client *clp) |
| 1450 | { |
| 1451 | nfs_delegation_clear_all(clp); |
| 1452 | nfs4_state_mark_reclaim_helper(clp, nfs4_state_mark_reclaim_nograce); |
| 1453 | } |
| 1454 | |
| 1455 | static int nfs4_recovery_handle_error(struct nfs_client *clp, int error) |
| 1456 | { |
| 1457 | switch (error) { |
| 1458 | case 0: |
| 1459 | break; |
| 1460 | case -NFS4ERR_CB_PATH_DOWN: |
| 1461 | nfs40_handle_cb_pathdown(clp); |
| 1462 | break; |
| 1463 | case -NFS4ERR_NO_GRACE: |
| 1464 | nfs4_state_end_reclaim_reboot(clp); |
| 1465 | break; |
| 1466 | case -NFS4ERR_STALE_CLIENTID: |
| 1467 | case -NFS4ERR_LEASE_MOVED: |
| 1468 | set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); |
| 1469 | nfs4_state_clear_reclaim_reboot(clp); |
| 1470 | nfs4_state_start_reclaim_reboot(clp); |
| 1471 | break; |
| 1472 | case -NFS4ERR_EXPIRED: |
| 1473 | set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); |
| 1474 | nfs4_state_start_reclaim_nograce(clp); |
| 1475 | break; |
| 1476 | case -NFS4ERR_BADSESSION: |
| 1477 | case -NFS4ERR_BADSLOT: |
| 1478 | case -NFS4ERR_BAD_HIGH_SLOT: |
| 1479 | case -NFS4ERR_DEADSESSION: |
| 1480 | case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION: |
| 1481 | case -NFS4ERR_SEQ_FALSE_RETRY: |
| 1482 | case -NFS4ERR_SEQ_MISORDERED: |
| 1483 | set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); |
| 1484 | /* Zero session reset errors */ |
| 1485 | break; |
| 1486 | default: |
| 1487 | return error; |
| 1488 | } |
| 1489 | return 0; |
| 1490 | } |
| 1491 | |
| 1492 | static int nfs4_do_reclaim(struct nfs_client *clp, const struct nfs4_state_recovery_ops *ops) |
| 1493 | { |
| 1494 | struct nfs4_state_owner *sp; |
| 1495 | struct nfs_server *server; |
| 1496 | struct rb_node *pos; |
| 1497 | int status = 0; |
| 1498 | |
| 1499 | restart: |
| 1500 | rcu_read_lock(); |
| 1501 | list_for_each_entry_rcu(server, &clp->cl_superblocks, client_link) { |
| 1502 | nfs4_purge_state_owners(server); |
| 1503 | spin_lock(&clp->cl_lock); |
| 1504 | for (pos = rb_first(&server->state_owners); |
| 1505 | pos != NULL; |
| 1506 | pos = rb_next(pos)) { |
| 1507 | sp = rb_entry(pos, |
| 1508 | struct nfs4_state_owner, so_server_node); |
| 1509 | if (!test_and_clear_bit(ops->owner_flag_bit, |
| 1510 | &sp->so_flags)) |
| 1511 | continue; |
| 1512 | atomic_inc(&sp->so_count); |
| 1513 | spin_unlock(&clp->cl_lock); |
| 1514 | rcu_read_unlock(); |
| 1515 | |
| 1516 | status = nfs4_reclaim_open_state(sp, ops); |
| 1517 | if (status < 0) { |
| 1518 | set_bit(ops->owner_flag_bit, &sp->so_flags); |
| 1519 | nfs4_put_state_owner(sp); |
| 1520 | status = nfs4_recovery_handle_error(clp, status); |
| 1521 | return (status != 0) ? status : -EAGAIN; |
| 1522 | } |
| 1523 | |
| 1524 | nfs4_put_state_owner(sp); |
| 1525 | goto restart; |
| 1526 | } |
| 1527 | spin_unlock(&clp->cl_lock); |
| 1528 | } |
| 1529 | rcu_read_unlock(); |
| 1530 | return 0; |
| 1531 | } |
| 1532 | |
| 1533 | static int nfs4_check_lease(struct nfs_client *clp) |
| 1534 | { |
| 1535 | struct rpc_cred *cred; |
| 1536 | const struct nfs4_state_maintenance_ops *ops = |
| 1537 | clp->cl_mvops->state_renewal_ops; |
| 1538 | int status; |
| 1539 | |
| 1540 | /* Is the client already known to have an expired lease? */ |
| 1541 | if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) |
| 1542 | return 0; |
| 1543 | spin_lock(&clp->cl_lock); |
| 1544 | cred = ops->get_state_renewal_cred_locked(clp); |
| 1545 | spin_unlock(&clp->cl_lock); |
| 1546 | if (cred == NULL) { |
| 1547 | cred = nfs4_get_setclientid_cred(clp); |
| 1548 | status = -ENOKEY; |
| 1549 | if (cred == NULL) |
| 1550 | goto out; |
| 1551 | } |
| 1552 | status = ops->renew_lease(clp, cred); |
| 1553 | put_rpccred(cred); |
| 1554 | out: |
| 1555 | return nfs4_recovery_handle_error(clp, status); |
| 1556 | } |
| 1557 | |
| 1558 | static int nfs4_reclaim_lease(struct nfs_client *clp) |
| 1559 | { |
| 1560 | struct rpc_cred *cred; |
| 1561 | const struct nfs4_state_recovery_ops *ops = |
| 1562 | clp->cl_mvops->reboot_recovery_ops; |
| 1563 | int status = -ENOENT; |
| 1564 | |
| 1565 | cred = ops->get_clid_cred(clp); |
| 1566 | if (cred != NULL) { |
| 1567 | status = ops->establish_clid(clp, cred); |
| 1568 | put_rpccred(cred); |
| 1569 | /* Handle case where the user hasn't set up machine creds */ |
| 1570 | if (status == -EACCES && cred == clp->cl_machine_cred) { |
| 1571 | nfs4_clear_machine_cred(clp); |
| 1572 | status = -EAGAIN; |
| 1573 | } |
| 1574 | if (status == -NFS4ERR_MINOR_VERS_MISMATCH) |
| 1575 | status = -EPROTONOSUPPORT; |
| 1576 | } |
| 1577 | return status; |
| 1578 | } |
| 1579 | |
| 1580 | #ifdef CONFIG_NFS_V4_1 |
| 1581 | void nfs4_schedule_session_recovery(struct nfs4_session *session) |
| 1582 | { |
| 1583 | struct nfs_client *clp = session->clp; |
| 1584 | |
| 1585 | set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); |
| 1586 | nfs4_schedule_lease_recovery(clp); |
| 1587 | } |
| 1588 | EXPORT_SYMBOL_GPL(nfs4_schedule_session_recovery); |
| 1589 | |
| 1590 | void nfs41_handle_recall_slot(struct nfs_client *clp) |
| 1591 | { |
| 1592 | set_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state); |
| 1593 | nfs4_schedule_state_manager(clp); |
| 1594 | } |
| 1595 | |
| 1596 | static void nfs4_reset_all_state(struct nfs_client *clp) |
| 1597 | { |
| 1598 | if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) { |
| 1599 | clp->cl_boot_time = CURRENT_TIME; |
| 1600 | nfs4_state_start_reclaim_nograce(clp); |
| 1601 | nfs4_schedule_state_manager(clp); |
| 1602 | } |
| 1603 | } |
| 1604 | |
| 1605 | static void nfs41_handle_server_reboot(struct nfs_client *clp) |
| 1606 | { |
| 1607 | if (test_and_set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state) == 0) { |
| 1608 | nfs4_state_start_reclaim_reboot(clp); |
| 1609 | nfs4_schedule_state_manager(clp); |
| 1610 | } |
| 1611 | } |
| 1612 | |
| 1613 | static void nfs41_handle_state_revoked(struct nfs_client *clp) |
| 1614 | { |
| 1615 | /* Temporary */ |
| 1616 | nfs4_reset_all_state(clp); |
| 1617 | } |
| 1618 | |
| 1619 | static void nfs41_handle_recallable_state_revoked(struct nfs_client *clp) |
| 1620 | { |
| 1621 | /* This will need to handle layouts too */ |
| 1622 | nfs_expire_all_delegations(clp); |
| 1623 | } |
| 1624 | |
| 1625 | static void nfs41_handle_cb_path_down(struct nfs_client *clp) |
| 1626 | { |
| 1627 | nfs_expire_all_delegations(clp); |
| 1628 | if (test_and_set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) == 0) |
| 1629 | nfs4_schedule_state_manager(clp); |
| 1630 | } |
| 1631 | |
| 1632 | void nfs41_handle_sequence_flag_errors(struct nfs_client *clp, u32 flags) |
| 1633 | { |
| 1634 | if (!flags) |
| 1635 | return; |
| 1636 | if (flags & SEQ4_STATUS_RESTART_RECLAIM_NEEDED) |
| 1637 | nfs41_handle_server_reboot(clp); |
| 1638 | if (flags & (SEQ4_STATUS_EXPIRED_ALL_STATE_REVOKED | |
| 1639 | SEQ4_STATUS_EXPIRED_SOME_STATE_REVOKED | |
| 1640 | SEQ4_STATUS_ADMIN_STATE_REVOKED | |
| 1641 | SEQ4_STATUS_LEASE_MOVED)) |
| 1642 | nfs41_handle_state_revoked(clp); |
| 1643 | if (flags & SEQ4_STATUS_RECALLABLE_STATE_REVOKED) |
| 1644 | nfs41_handle_recallable_state_revoked(clp); |
| 1645 | if (flags & (SEQ4_STATUS_CB_PATH_DOWN | |
| 1646 | SEQ4_STATUS_BACKCHANNEL_FAULT | |
| 1647 | SEQ4_STATUS_CB_PATH_DOWN_SESSION)) |
| 1648 | nfs41_handle_cb_path_down(clp); |
| 1649 | } |
| 1650 | |
| 1651 | static int nfs4_reset_session(struct nfs_client *clp) |
| 1652 | { |
| 1653 | int status; |
| 1654 | |
| 1655 | nfs4_begin_drain_session(clp); |
| 1656 | status = nfs4_proc_destroy_session(clp->cl_session); |
| 1657 | switch (status) { |
| 1658 | case 0: |
| 1659 | case -NFS4ERR_BADSESSION: |
| 1660 | case -NFS4ERR_DEADSESSION: |
| 1661 | break; |
| 1662 | case -NFS4ERR_BACK_CHAN_BUSY: |
| 1663 | case -NFS4ERR_DELAY: |
| 1664 | set_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); |
| 1665 | status = 0; |
| 1666 | ssleep(1); |
| 1667 | goto out; |
| 1668 | default: |
| 1669 | status = nfs4_recovery_handle_error(clp, status); |
| 1670 | goto out; |
| 1671 | } |
| 1672 | |
| 1673 | memset(clp->cl_session->sess_id.data, 0, NFS4_MAX_SESSIONID_LEN); |
| 1674 | status = nfs4_proc_create_session(clp); |
| 1675 | if (status) { |
| 1676 | status = nfs4_recovery_handle_error(clp, status); |
| 1677 | goto out; |
| 1678 | } |
| 1679 | clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state); |
| 1680 | /* create_session negotiated new slot table */ |
| 1681 | clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state); |
| 1682 | |
| 1683 | /* Let the state manager reestablish state */ |
| 1684 | if (!test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) |
| 1685 | nfs41_setup_state_renewal(clp); |
| 1686 | out: |
| 1687 | return status; |
| 1688 | } |
| 1689 | |
| 1690 | static int nfs4_recall_slot(struct nfs_client *clp) |
| 1691 | { |
| 1692 | struct nfs4_slot_table *fc_tbl = &clp->cl_session->fc_slot_table; |
| 1693 | struct nfs4_channel_attrs *fc_attrs = &clp->cl_session->fc_attrs; |
| 1694 | struct nfs4_slot *new, *old; |
| 1695 | int i; |
| 1696 | |
| 1697 | nfs4_begin_drain_session(clp); |
| 1698 | new = kmalloc(fc_tbl->target_max_slots * sizeof(struct nfs4_slot), |
| 1699 | GFP_NOFS); |
| 1700 | if (!new) |
| 1701 | return -ENOMEM; |
| 1702 | |
| 1703 | spin_lock(&fc_tbl->slot_tbl_lock); |
| 1704 | for (i = 0; i < fc_tbl->target_max_slots; i++) |
| 1705 | new[i].seq_nr = fc_tbl->slots[i].seq_nr; |
| 1706 | old = fc_tbl->slots; |
| 1707 | fc_tbl->slots = new; |
| 1708 | fc_tbl->max_slots = fc_tbl->target_max_slots; |
| 1709 | fc_tbl->target_max_slots = 0; |
| 1710 | fc_attrs->max_reqs = fc_tbl->max_slots; |
| 1711 | spin_unlock(&fc_tbl->slot_tbl_lock); |
| 1712 | |
| 1713 | kfree(old); |
| 1714 | nfs4_end_drain_session(clp); |
| 1715 | return 0; |
| 1716 | } |
| 1717 | |
| 1718 | #else /* CONFIG_NFS_V4_1 */ |
| 1719 | static int nfs4_reset_session(struct nfs_client *clp) { return 0; } |
| 1720 | static int nfs4_end_drain_session(struct nfs_client *clp) { return 0; } |
| 1721 | static int nfs4_recall_slot(struct nfs_client *clp) { return 0; } |
| 1722 | #endif /* CONFIG_NFS_V4_1 */ |
| 1723 | |
| 1724 | /* Set NFS4CLNT_LEASE_EXPIRED for all v4.0 errors and for recoverable errors |
| 1725 | * on EXCHANGE_ID for v4.1 |
| 1726 | */ |
| 1727 | static void nfs4_set_lease_expired(struct nfs_client *clp, int status) |
| 1728 | { |
| 1729 | switch (status) { |
| 1730 | case -NFS4ERR_CLID_INUSE: |
| 1731 | case -NFS4ERR_STALE_CLIENTID: |
| 1732 | clear_bit(NFS4CLNT_LEASE_CONFIRM, &clp->cl_state); |
| 1733 | break; |
| 1734 | case -NFS4ERR_DELAY: |
| 1735 | case -ETIMEDOUT: |
| 1736 | case -EAGAIN: |
| 1737 | ssleep(1); |
| 1738 | break; |
| 1739 | |
| 1740 | case -EKEYEXPIRED: |
| 1741 | case -NFS4ERR_NOT_SAME: /* FixMe: implement recovery |
| 1742 | * in nfs4_exchange_id */ |
| 1743 | default: |
| 1744 | return; |
| 1745 | } |
| 1746 | set_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state); |
| 1747 | } |
| 1748 | |
| 1749 | static void nfs4_state_manager(struct nfs_client *clp) |
| 1750 | { |
| 1751 | int status = 0; |
| 1752 | |
| 1753 | /* Ensure exclusive access to NFSv4 state */ |
| 1754 | do { |
| 1755 | if (test_and_clear_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) { |
| 1756 | /* We're going to have to re-establish a clientid */ |
| 1757 | status = nfs4_reclaim_lease(clp); |
| 1758 | if (status) { |
| 1759 | nfs4_set_lease_expired(clp, status); |
| 1760 | if (test_bit(NFS4CLNT_LEASE_EXPIRED, |
| 1761 | &clp->cl_state)) |
| 1762 | continue; |
| 1763 | if (clp->cl_cons_state == |
| 1764 | NFS_CS_SESSION_INITING) |
| 1765 | nfs_mark_client_ready(clp, status); |
| 1766 | goto out_error; |
| 1767 | } |
| 1768 | clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); |
| 1769 | |
| 1770 | if (test_and_clear_bit(NFS4CLNT_SERVER_SCOPE_MISMATCH, |
| 1771 | &clp->cl_state)) |
| 1772 | nfs4_state_start_reclaim_nograce(clp); |
| 1773 | else |
| 1774 | set_bit(NFS4CLNT_RECLAIM_REBOOT, |
| 1775 | &clp->cl_state); |
| 1776 | |
| 1777 | pnfs_destroy_all_layouts(clp); |
| 1778 | } |
| 1779 | |
| 1780 | if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) { |
| 1781 | status = nfs4_check_lease(clp); |
| 1782 | if (status < 0) |
| 1783 | goto out_error; |
| 1784 | if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) |
| 1785 | continue; |
| 1786 | } |
| 1787 | |
| 1788 | /* Initialize or reset the session */ |
| 1789 | if (test_and_clear_bit(NFS4CLNT_SESSION_RESET, &clp->cl_state) |
| 1790 | && nfs4_has_session(clp)) { |
| 1791 | status = nfs4_reset_session(clp); |
| 1792 | if (test_bit(NFS4CLNT_LEASE_EXPIRED, &clp->cl_state)) |
| 1793 | continue; |
| 1794 | if (status < 0) |
| 1795 | goto out_error; |
| 1796 | } |
| 1797 | |
| 1798 | /* First recover reboot state... */ |
| 1799 | if (test_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state)) { |
| 1800 | status = nfs4_do_reclaim(clp, |
| 1801 | clp->cl_mvops->reboot_recovery_ops); |
| 1802 | if (status == -EAGAIN) |
| 1803 | continue; |
| 1804 | if (status < 0) |
| 1805 | goto out_error; |
| 1806 | nfs4_state_end_reclaim_reboot(clp); |
| 1807 | } |
| 1808 | |
| 1809 | /* Now recover expired state... */ |
| 1810 | if (test_and_clear_bit(NFS4CLNT_RECLAIM_NOGRACE, &clp->cl_state)) { |
| 1811 | status = nfs4_do_reclaim(clp, |
| 1812 | clp->cl_mvops->nograce_recovery_ops); |
| 1813 | if (status == -EAGAIN) |
| 1814 | continue; |
| 1815 | if (status < 0) |
| 1816 | goto out_error; |
| 1817 | } |
| 1818 | |
| 1819 | nfs4_end_drain_session(clp); |
| 1820 | if (test_and_clear_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state)) { |
| 1821 | nfs_client_return_marked_delegations(clp); |
| 1822 | continue; |
| 1823 | } |
| 1824 | /* Recall session slots */ |
| 1825 | if (test_and_clear_bit(NFS4CLNT_RECALL_SLOT, &clp->cl_state) |
| 1826 | && nfs4_has_session(clp)) { |
| 1827 | status = nfs4_recall_slot(clp); |
| 1828 | if (status < 0) |
| 1829 | goto out_error; |
| 1830 | continue; |
| 1831 | } |
| 1832 | |
| 1833 | |
| 1834 | nfs4_clear_state_manager_bit(clp); |
| 1835 | /* Did we race with an attempt to give us more work? */ |
| 1836 | if (clp->cl_state == 0) |
| 1837 | break; |
| 1838 | if (test_and_set_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) != 0) |
| 1839 | break; |
| 1840 | } while (atomic_read(&clp->cl_count) > 1); |
| 1841 | return; |
| 1842 | out_error: |
| 1843 | pr_warn_ratelimited("NFS: state manager failed on NFSv4 server %s" |
| 1844 | " with error %d\n", clp->cl_hostname, -status); |
| 1845 | nfs4_end_drain_session(clp); |
| 1846 | nfs4_clear_state_manager_bit(clp); |
| 1847 | } |
| 1848 | |
| 1849 | static int nfs4_run_state_manager(void *ptr) |
| 1850 | { |
| 1851 | struct nfs_client *clp = ptr; |
| 1852 | |
| 1853 | allow_signal(SIGKILL); |
| 1854 | nfs4_state_manager(clp); |
| 1855 | nfs_put_client(clp); |
| 1856 | module_put_and_exit(0); |
| 1857 | return 0; |
| 1858 | } |
| 1859 | |
| 1860 | /* |
| 1861 | * Local variables: |
| 1862 | * c-basic-offset: 8 |
| 1863 | * End: |
| 1864 | */ |