1 /* AFS file locking support 2 * 3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 4 * Written by David Howells (dhowells@redhat.com) 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #include "internal.h" 13 14 #define AFS_LOCK_GRANTED 0 15 #define AFS_LOCK_PENDING 1 16 17 struct workqueue_struct *afs_lock_manager; 18 19 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl); 20 static void afs_fl_release_private(struct file_lock *fl); 21 22 static const struct file_lock_operations afs_lock_ops = { 23 .fl_copy_lock = afs_fl_copy_lock, 24 .fl_release_private = afs_fl_release_private, 25 }; 26 27 /* 28 * if the callback is broken on this vnode, then the lock may now be available 29 */ 30 void afs_lock_may_be_available(struct afs_vnode *vnode) 31 { 32 _enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode); 33 34 queue_delayed_work(afs_lock_manager, &vnode->lock_work, 0); 35 } 36 37 /* 38 * the lock will time out in 5 minutes unless we extend it, so schedule 39 * extension in a bit less than that time 40 */ 41 static void afs_schedule_lock_extension(struct afs_vnode *vnode) 42 { 43 queue_delayed_work(afs_lock_manager, &vnode->lock_work, 44 AFS_LOCKWAIT * HZ / 2); 45 } 46 47 /* 48 * grant one or more locks (readlocks are allowed to jump the queue if the 49 * first lock in the queue is itself a readlock) 50 * - the caller must hold the vnode lock 51 */ 52 static void afs_grant_locks(struct afs_vnode *vnode, struct file_lock *fl) 53 { 54 struct file_lock *p, *_p; 55 56 list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks); 57 if (fl->fl_type == F_RDLCK) { 58 list_for_each_entry_safe(p, _p, &vnode->pending_locks, 59 fl_u.afs.link) { 60 if (p->fl_type == F_RDLCK) { 61 p->fl_u.afs.state = AFS_LOCK_GRANTED; 62 list_move_tail(&p->fl_u.afs.link, 63 &vnode->granted_locks); 64 wake_up(&p->fl_wait); 65 } 66 } 67 } 68 } 69 70 /* 71 * Get a lock on a file 72 */ 73 static int afs_set_lock(struct afs_vnode *vnode, struct key *key, 74 afs_lock_type_t type) 75 { 76 struct afs_fs_cursor fc; 77 int ret; 78 79 _enter("%s{%x:%u.%u},%x,%u", 80 vnode->volume->name, 81 vnode->fid.vid, 82 vnode->fid.vnode, 83 vnode->fid.unique, 84 key_serial(key), type); 85 86 ret = -ERESTARTSYS; 87 if (afs_begin_vnode_operation(&fc, vnode, key)) { 88 while (afs_select_fileserver(&fc)) { 89 fc.cb_break = vnode->cb_break + vnode->cb_s_break; 90 afs_fs_set_lock(&fc, type); 91 } 92 93 afs_check_for_remote_deletion(&fc, fc.vnode); 94 afs_vnode_commit_status(&fc, vnode, fc.cb_break); 95 ret = afs_end_vnode_operation(&fc); 96 } 97 98 _leave(" = %d", ret); 99 return ret; 100 } 101 102 /* 103 * Extend a lock on a file 104 */ 105 static int afs_extend_lock(struct afs_vnode *vnode, struct key *key) 106 { 107 struct afs_fs_cursor fc; 108 int ret; 109 110 _enter("%s{%x:%u.%u},%x", 111 vnode->volume->name, 112 vnode->fid.vid, 113 vnode->fid.vnode, 114 vnode->fid.unique, 115 key_serial(key)); 116 117 ret = -ERESTARTSYS; 118 if (afs_begin_vnode_operation(&fc, vnode, key)) { 119 while (afs_select_current_fileserver(&fc)) { 120 fc.cb_break = vnode->cb_break + vnode->cb_s_break; 121 afs_fs_extend_lock(&fc); 122 } 123 124 afs_check_for_remote_deletion(&fc, fc.vnode); 125 afs_vnode_commit_status(&fc, vnode, fc.cb_break); 126 ret = afs_end_vnode_operation(&fc); 127 } 128 129 _leave(" = %d", ret); 130 return ret; 131 } 132 133 /* 134 * Release a lock on a file 135 */ 136 static int afs_release_lock(struct afs_vnode *vnode, struct key *key) 137 { 138 struct afs_fs_cursor fc; 139 int ret; 140 141 _enter("%s{%x:%u.%u},%x", 142 vnode->volume->name, 143 vnode->fid.vid, 144 vnode->fid.vnode, 145 vnode->fid.unique, 146 key_serial(key)); 147 148 ret = -ERESTARTSYS; 149 if (afs_begin_vnode_operation(&fc, vnode, key)) { 150 while (afs_select_current_fileserver(&fc)) { 151 fc.cb_break = vnode->cb_break + vnode->cb_s_break; 152 afs_fs_release_lock(&fc); 153 } 154 155 afs_check_for_remote_deletion(&fc, fc.vnode); 156 afs_vnode_commit_status(&fc, vnode, fc.cb_break); 157 ret = afs_end_vnode_operation(&fc); 158 } 159 160 _leave(" = %d", ret); 161 return ret; 162 } 163 164 /* 165 * do work for a lock, including: 166 * - probing for a lock we're waiting on but didn't get immediately 167 * - extending a lock that's close to timing out 168 */ 169 void afs_lock_work(struct work_struct *work) 170 { 171 struct afs_vnode *vnode = 172 container_of(work, struct afs_vnode, lock_work.work); 173 struct file_lock *fl; 174 afs_lock_type_t type; 175 struct key *key; 176 int ret; 177 178 _enter("{%x:%u}", vnode->fid.vid, vnode->fid.vnode); 179 180 spin_lock(&vnode->lock); 181 182 if (test_bit(AFS_VNODE_UNLOCKING, &vnode->flags)) { 183 _debug("unlock"); 184 spin_unlock(&vnode->lock); 185 186 /* attempt to release the server lock; if it fails, we just 187 * wait 5 minutes and it'll time out anyway */ 188 ret = afs_release_lock(vnode, vnode->unlock_key); 189 if (ret < 0) 190 printk(KERN_WARNING "AFS:" 191 " Failed to release lock on {%x:%x} error %d\n", 192 vnode->fid.vid, vnode->fid.vnode, ret); 193 194 spin_lock(&vnode->lock); 195 key_put(vnode->unlock_key); 196 vnode->unlock_key = NULL; 197 clear_bit(AFS_VNODE_UNLOCKING, &vnode->flags); 198 } 199 200 /* if we've got a lock, then it must be time to extend that lock as AFS 201 * locks time out after 5 minutes */ 202 if (!list_empty(&vnode->granted_locks)) { 203 _debug("extend"); 204 205 if (test_and_set_bit(AFS_VNODE_LOCKING, &vnode->flags)) 206 BUG(); 207 fl = list_entry(vnode->granted_locks.next, 208 struct file_lock, fl_u.afs.link); 209 key = key_get(afs_file_key(fl->fl_file)); 210 spin_unlock(&vnode->lock); 211 212 ret = afs_extend_lock(vnode, key); 213 clear_bit(AFS_VNODE_LOCKING, &vnode->flags); 214 key_put(key); 215 switch (ret) { 216 case 0: 217 afs_schedule_lock_extension(vnode); 218 break; 219 default: 220 /* ummm... we failed to extend the lock - retry 221 * extension shortly */ 222 printk(KERN_WARNING "AFS:" 223 " Failed to extend lock on {%x:%x} error %d\n", 224 vnode->fid.vid, vnode->fid.vnode, ret); 225 queue_delayed_work(afs_lock_manager, &vnode->lock_work, 226 HZ * 10); 227 break; 228 } 229 _leave(" [extend]"); 230 return; 231 } 232 233 /* if we don't have a granted lock, then we must've been called back by 234 * the server, and so if might be possible to get a lock we're 235 * currently waiting for */ 236 if (!list_empty(&vnode->pending_locks)) { 237 _debug("get"); 238 239 if (test_and_set_bit(AFS_VNODE_LOCKING, &vnode->flags)) 240 BUG(); 241 fl = list_entry(vnode->pending_locks.next, 242 struct file_lock, fl_u.afs.link); 243 key = key_get(afs_file_key(fl->fl_file)); 244 type = (fl->fl_type == F_RDLCK) ? 245 AFS_LOCK_READ : AFS_LOCK_WRITE; 246 spin_unlock(&vnode->lock); 247 248 ret = afs_set_lock(vnode, key, type); 249 clear_bit(AFS_VNODE_LOCKING, &vnode->flags); 250 switch (ret) { 251 case -EWOULDBLOCK: 252 _debug("blocked"); 253 break; 254 case 0: 255 _debug("acquired"); 256 if (type == AFS_LOCK_READ) 257 set_bit(AFS_VNODE_READLOCKED, &vnode->flags); 258 else 259 set_bit(AFS_VNODE_WRITELOCKED, &vnode->flags); 260 ret = AFS_LOCK_GRANTED; 261 default: 262 spin_lock(&vnode->lock); 263 /* the pending lock may have been withdrawn due to a 264 * signal */ 265 if (list_entry(vnode->pending_locks.next, 266 struct file_lock, fl_u.afs.link) == fl) { 267 fl->fl_u.afs.state = ret; 268 if (ret == AFS_LOCK_GRANTED) 269 afs_grant_locks(vnode, fl); 270 else 271 list_del_init(&fl->fl_u.afs.link); 272 wake_up(&fl->fl_wait); 273 spin_unlock(&vnode->lock); 274 } else { 275 _debug("withdrawn"); 276 clear_bit(AFS_VNODE_READLOCKED, &vnode->flags); 277 clear_bit(AFS_VNODE_WRITELOCKED, &vnode->flags); 278 spin_unlock(&vnode->lock); 279 afs_release_lock(vnode, key); 280 if (!list_empty(&vnode->pending_locks)) 281 afs_lock_may_be_available(vnode); 282 } 283 break; 284 } 285 key_put(key); 286 _leave(" [pend]"); 287 return; 288 } 289 290 /* looks like the lock request was withdrawn on a signal */ 291 spin_unlock(&vnode->lock); 292 _leave(" [no locks]"); 293 } 294 295 /* 296 * pass responsibility for the unlocking of a vnode on the server to the 297 * manager thread, lest a pending signal in the calling thread interrupt 298 * AF_RXRPC 299 * - the caller must hold the vnode lock 300 */ 301 static void afs_defer_unlock(struct afs_vnode *vnode, struct key *key) 302 { 303 cancel_delayed_work(&vnode->lock_work); 304 if (!test_and_clear_bit(AFS_VNODE_READLOCKED, &vnode->flags) && 305 !test_and_clear_bit(AFS_VNODE_WRITELOCKED, &vnode->flags)) 306 BUG(); 307 if (test_and_set_bit(AFS_VNODE_UNLOCKING, &vnode->flags)) 308 BUG(); 309 vnode->unlock_key = key_get(key); 310 afs_lock_may_be_available(vnode); 311 } 312 313 /* 314 * request a lock on a file on the server 315 */ 316 static int afs_do_setlk(struct file *file, struct file_lock *fl) 317 { 318 struct inode *inode = file_inode(file); 319 struct afs_vnode *vnode = AFS_FS_I(inode); 320 afs_lock_type_t type; 321 struct key *key = afs_file_key(file); 322 int ret; 323 324 _enter("{%x:%u},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type); 325 326 /* only whole-file locks are supported */ 327 if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX) 328 return -EINVAL; 329 330 fl->fl_ops = &afs_lock_ops; 331 INIT_LIST_HEAD(&fl->fl_u.afs.link); 332 fl->fl_u.afs.state = AFS_LOCK_PENDING; 333 334 type = (fl->fl_type == F_RDLCK) ? AFS_LOCK_READ : AFS_LOCK_WRITE; 335 336 spin_lock(&inode->i_lock); 337 338 /* make sure we've got a callback on this file and that our view of the 339 * data version is up to date */ 340 ret = afs_validate(vnode, key); 341 if (ret < 0) 342 goto error; 343 344 if (vnode->status.lock_count != 0 && !(fl->fl_flags & FL_SLEEP)) { 345 ret = -EAGAIN; 346 goto error; 347 } 348 349 spin_lock(&vnode->lock); 350 351 /* if we've already got a readlock on the server then we can instantly 352 * grant another readlock, irrespective of whether there are any 353 * pending writelocks */ 354 if (type == AFS_LOCK_READ && 355 vnode->flags & (1 << AFS_VNODE_READLOCKED)) { 356 _debug("instant readlock"); 357 ASSERTCMP(vnode->flags & 358 ((1 << AFS_VNODE_LOCKING) | 359 (1 << AFS_VNODE_WRITELOCKED)), ==, 0); 360 ASSERT(!list_empty(&vnode->granted_locks)); 361 goto sharing_existing_lock; 362 } 363 364 /* if there's no-one else with a lock on this vnode, then we need to 365 * ask the server for a lock */ 366 if (list_empty(&vnode->pending_locks) && 367 list_empty(&vnode->granted_locks)) { 368 _debug("not locked"); 369 ASSERTCMP(vnode->flags & 370 ((1 << AFS_VNODE_LOCKING) | 371 (1 << AFS_VNODE_READLOCKED) | 372 (1 << AFS_VNODE_WRITELOCKED)), ==, 0); 373 list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks); 374 set_bit(AFS_VNODE_LOCKING, &vnode->flags); 375 spin_unlock(&vnode->lock); 376 377 ret = afs_set_lock(vnode, key, type); 378 clear_bit(AFS_VNODE_LOCKING, &vnode->flags); 379 switch (ret) { 380 case 0: 381 _debug("acquired"); 382 goto acquired_server_lock; 383 case -EWOULDBLOCK: 384 _debug("would block"); 385 spin_lock(&vnode->lock); 386 ASSERT(list_empty(&vnode->granted_locks)); 387 ASSERTCMP(vnode->pending_locks.next, ==, 388 &fl->fl_u.afs.link); 389 goto wait; 390 default: 391 spin_lock(&vnode->lock); 392 list_del_init(&fl->fl_u.afs.link); 393 spin_unlock(&vnode->lock); 394 goto error; 395 } 396 } 397 398 /* otherwise, we need to wait for a local lock to become available */ 399 _debug("wait local"); 400 list_add_tail(&fl->fl_u.afs.link, &vnode->pending_locks); 401 wait: 402 if (!(fl->fl_flags & FL_SLEEP)) { 403 _debug("noblock"); 404 ret = -EAGAIN; 405 goto abort_attempt; 406 } 407 spin_unlock(&vnode->lock); 408 409 /* now we need to sleep and wait for the lock manager thread to get the 410 * lock from the server */ 411 _debug("sleep"); 412 ret = wait_event_interruptible(fl->fl_wait, 413 fl->fl_u.afs.state <= AFS_LOCK_GRANTED); 414 if (fl->fl_u.afs.state <= AFS_LOCK_GRANTED) { 415 ret = fl->fl_u.afs.state; 416 if (ret < 0) 417 goto error; 418 spin_lock(&vnode->lock); 419 goto given_lock; 420 } 421 422 /* we were interrupted, but someone may still be in the throes of 423 * giving us the lock */ 424 _debug("intr"); 425 ASSERTCMP(ret, ==, -ERESTARTSYS); 426 427 spin_lock(&vnode->lock); 428 if (fl->fl_u.afs.state <= AFS_LOCK_GRANTED) { 429 ret = fl->fl_u.afs.state; 430 if (ret < 0) { 431 spin_unlock(&vnode->lock); 432 goto error; 433 } 434 goto given_lock; 435 } 436 437 abort_attempt: 438 /* we aren't going to get the lock, either because we're unwilling to 439 * wait, or because some signal happened */ 440 _debug("abort"); 441 if (list_empty(&vnode->granted_locks) && 442 vnode->pending_locks.next == &fl->fl_u.afs.link) { 443 if (vnode->pending_locks.prev != &fl->fl_u.afs.link) { 444 /* kick the next pending lock into having a go */ 445 list_del_init(&fl->fl_u.afs.link); 446 afs_lock_may_be_available(vnode); 447 } 448 } else { 449 list_del_init(&fl->fl_u.afs.link); 450 } 451 spin_unlock(&vnode->lock); 452 goto error; 453 454 acquired_server_lock: 455 /* we've acquired a server lock, but it needs to be renewed after 5 456 * mins */ 457 spin_lock(&vnode->lock); 458 afs_schedule_lock_extension(vnode); 459 if (type == AFS_LOCK_READ) 460 set_bit(AFS_VNODE_READLOCKED, &vnode->flags); 461 else 462 set_bit(AFS_VNODE_WRITELOCKED, &vnode->flags); 463 sharing_existing_lock: 464 /* the lock has been granted as far as we're concerned... */ 465 fl->fl_u.afs.state = AFS_LOCK_GRANTED; 466 list_move_tail(&fl->fl_u.afs.link, &vnode->granted_locks); 467 given_lock: 468 /* ... but we do still need to get the VFS's blessing */ 469 ASSERT(!(vnode->flags & (1 << AFS_VNODE_LOCKING))); 470 ASSERT((vnode->flags & ((1 << AFS_VNODE_READLOCKED) | 471 (1 << AFS_VNODE_WRITELOCKED))) != 0); 472 ret = posix_lock_file(file, fl, NULL); 473 if (ret < 0) 474 goto vfs_rejected_lock; 475 spin_unlock(&vnode->lock); 476 477 /* again, make sure we've got a callback on this file and, again, make 478 * sure that our view of the data version is up to date (we ignore 479 * errors incurred here and deal with the consequences elsewhere) */ 480 afs_validate(vnode, key); 481 482 error: 483 spin_unlock(&inode->i_lock); 484 _leave(" = %d", ret); 485 return ret; 486 487 vfs_rejected_lock: 488 /* the VFS rejected the lock we just obtained, so we have to discard 489 * what we just got */ 490 _debug("vfs refused %d", ret); 491 list_del_init(&fl->fl_u.afs.link); 492 if (list_empty(&vnode->granted_locks)) 493 afs_defer_unlock(vnode, key); 494 goto abort_attempt; 495 } 496 497 /* 498 * unlock on a file on the server 499 */ 500 static int afs_do_unlk(struct file *file, struct file_lock *fl) 501 { 502 struct afs_vnode *vnode = AFS_FS_I(file->f_mapping->host); 503 struct key *key = afs_file_key(file); 504 int ret; 505 506 _enter("{%x:%u},%u", vnode->fid.vid, vnode->fid.vnode, fl->fl_type); 507 508 /* only whole-file unlocks are supported */ 509 if (fl->fl_start != 0 || fl->fl_end != OFFSET_MAX) 510 return -EINVAL; 511 512 fl->fl_ops = &afs_lock_ops; 513 INIT_LIST_HEAD(&fl->fl_u.afs.link); 514 fl->fl_u.afs.state = AFS_LOCK_PENDING; 515 516 spin_lock(&vnode->lock); 517 ret = posix_lock_file(file, fl, NULL); 518 if (ret < 0) { 519 spin_unlock(&vnode->lock); 520 _leave(" = %d [vfs]", ret); 521 return ret; 522 } 523 524 /* discard the server lock only if all granted locks are gone */ 525 if (list_empty(&vnode->granted_locks)) 526 afs_defer_unlock(vnode, key); 527 spin_unlock(&vnode->lock); 528 _leave(" = 0"); 529 return 0; 530 } 531 532 /* 533 * return information about a lock we currently hold, if indeed we hold one 534 */ 535 static int afs_do_getlk(struct file *file, struct file_lock *fl) 536 { 537 struct afs_vnode *vnode = AFS_FS_I(file->f_mapping->host); 538 struct key *key = afs_file_key(file); 539 int ret, lock_count; 540 541 _enter(""); 542 543 fl->fl_type = F_UNLCK; 544 545 inode_lock(&vnode->vfs_inode); 546 547 /* check local lock records first */ 548 ret = 0; 549 posix_test_lock(file, fl); 550 if (fl->fl_type == F_UNLCK) { 551 /* no local locks; consult the server */ 552 ret = afs_fetch_status(vnode, key); 553 if (ret < 0) 554 goto error; 555 lock_count = vnode->status.lock_count; 556 if (lock_count) { 557 if (lock_count > 0) 558 fl->fl_type = F_RDLCK; 559 else 560 fl->fl_type = F_WRLCK; 561 fl->fl_start = 0; 562 fl->fl_end = OFFSET_MAX; 563 } 564 } 565 566 error: 567 inode_unlock(&vnode->vfs_inode); 568 _leave(" = %d [%hd]", ret, fl->fl_type); 569 return ret; 570 } 571 572 /* 573 * manage POSIX locks on a file 574 */ 575 int afs_lock(struct file *file, int cmd, struct file_lock *fl) 576 { 577 struct afs_vnode *vnode = AFS_FS_I(file_inode(file)); 578 579 _enter("{%x:%u},%d,{t=%x,fl=%x,r=%Ld:%Ld}", 580 vnode->fid.vid, vnode->fid.vnode, cmd, 581 fl->fl_type, fl->fl_flags, 582 (long long) fl->fl_start, (long long) fl->fl_end); 583 584 /* AFS doesn't support mandatory locks */ 585 if (__mandatory_lock(&vnode->vfs_inode) && fl->fl_type != F_UNLCK) 586 return -ENOLCK; 587 588 if (IS_GETLK(cmd)) 589 return afs_do_getlk(file, fl); 590 if (fl->fl_type == F_UNLCK) 591 return afs_do_unlk(file, fl); 592 return afs_do_setlk(file, fl); 593 } 594 595 /* 596 * manage FLOCK locks on a file 597 */ 598 int afs_flock(struct file *file, int cmd, struct file_lock *fl) 599 { 600 struct afs_vnode *vnode = AFS_FS_I(file_inode(file)); 601 602 _enter("{%x:%u},%d,{t=%x,fl=%x}", 603 vnode->fid.vid, vnode->fid.vnode, cmd, 604 fl->fl_type, fl->fl_flags); 605 606 /* 607 * No BSD flocks over NFS allowed. 608 * Note: we could try to fake a POSIX lock request here by 609 * using ((u32) filp | 0x80000000) or some such as the pid. 610 * Not sure whether that would be unique, though, or whether 611 * that would break in other places. 612 */ 613 if (!(fl->fl_flags & FL_FLOCK)) 614 return -ENOLCK; 615 616 /* we're simulating flock() locks using posix locks on the server */ 617 if (fl->fl_type == F_UNLCK) 618 return afs_do_unlk(file, fl); 619 return afs_do_setlk(file, fl); 620 } 621 622 /* 623 * the POSIX lock management core VFS code copies the lock record and adds the 624 * copy into its own list, so we need to add that copy to the vnode's lock 625 * queue in the same place as the original (which will be deleted shortly 626 * after) 627 */ 628 static void afs_fl_copy_lock(struct file_lock *new, struct file_lock *fl) 629 { 630 _enter(""); 631 632 list_add(&new->fl_u.afs.link, &fl->fl_u.afs.link); 633 } 634 635 /* 636 * need to remove this lock from the vnode queue when it's removed from the 637 * VFS's list 638 */ 639 static void afs_fl_release_private(struct file_lock *fl) 640 { 641 _enter(""); 642 643 list_del_init(&fl->fl_u.afs.link); 644 } 645