1 /* 2 * linux/fs/locks.c 3 * 4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls. 5 * Doug Evans (dje@spiff.uucp), August 07, 1992 6 * 7 * Deadlock detection added. 8 * FIXME: one thing isn't handled yet: 9 * - mandatory locks (requires lots of changes elsewhere) 10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994. 11 * 12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code. 13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994 14 * 15 * Converted file_lock_table to a linked list from an array, which eliminates 16 * the limits on how many active file locks are open. 17 * Chad Page (pageone@netcom.com), November 27, 1994 18 * 19 * Removed dependency on file descriptors. dup()'ed file descriptors now 20 * get the same locks as the original file descriptors, and a close() on 21 * any file descriptor removes ALL the locks on the file for the current 22 * process. Since locks still depend on the process id, locks are inherited 23 * after an exec() but not after a fork(). This agrees with POSIX, and both 24 * BSD and SVR4 practice. 25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995 26 * 27 * Scrapped free list which is redundant now that we allocate locks 28 * dynamically with kmalloc()/kfree(). 29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995 30 * 31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX. 32 * 33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the 34 * fcntl() system call. They have the semantics described above. 35 * 36 * FL_FLOCK locks are created with calls to flock(), through the flock() 37 * system call, which is new. Old C libraries implement flock() via fcntl() 38 * and will continue to use the old, broken implementation. 39 * 40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated 41 * with a file pointer (filp). As a result they can be shared by a parent 42 * process and its children after a fork(). They are removed when the last 43 * file descriptor referring to the file pointer is closed (unless explicitly 44 * unlocked). 45 * 46 * FL_FLOCK locks never deadlock, an existing lock is always removed before 47 * upgrading from shared to exclusive (or vice versa). When this happens 48 * any processes blocked by the current lock are woken up and allowed to 49 * run before the new lock is applied. 50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995 51 * 52 * Removed some race conditions in flock_lock_file(), marked other possible 53 * races. Just grep for FIXME to see them. 54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996. 55 * 56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive. 57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep 58 * once we've checked for blocking and deadlocking. 59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996. 60 * 61 * Initial implementation of mandatory locks. SunOS turned out to be 62 * a rotten model, so I implemented the "obvious" semantics. 63 * See 'Documentation/filesystems/mandatory-locking.txt' for details. 64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996. 65 * 66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to 67 * check if a file has mandatory locks, used by mmap(), open() and creat() to 68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference 69 * Manual, Section 2. 70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996. 71 * 72 * Tidied up block list handling. Added '/proc/locks' interface. 73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996. 74 * 75 * Fixed deadlock condition for pathological code that mixes calls to 76 * flock() and fcntl(). 77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996. 78 * 79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use 80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to 81 * guarantee sensible behaviour in the case where file system modules might 82 * be compiled with different options than the kernel itself. 83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996. 84 * 85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel 86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this. 87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996. 88 * 89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK 90 * locks. Changed process synchronisation to avoid dereferencing locks that 91 * have already been freed. 92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996. 93 * 94 * Made the block list a circular list to minimise searching in the list. 95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996. 96 * 97 * Made mandatory locking a mount option. Default is not to allow mandatory 98 * locking. 99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996. 100 * 101 * Some adaptations for NFS support. 102 * Olaf Kirch (okir@monad.swb.de), Dec 1996, 103 * 104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed. 105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997. 106 * 107 * Use slab allocator instead of kmalloc/kfree. 108 * Use generic list implementation from <linux/list.h>. 109 * Sped up posix_locks_deadlock by only considering blocked locks. 110 * Matthew Wilcox <willy@debian.org>, March, 2000. 111 * 112 * Leases and LOCK_MAND 113 * Matthew Wilcox <willy@debian.org>, June, 2000. 114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000. 115 */ 116 117 #include <linux/capability.h> 118 #include <linux/file.h> 119 #include <linux/fdtable.h> 120 #include <linux/fs.h> 121 #include <linux/init.h> 122 #include <linux/module.h> 123 #include <linux/security.h> 124 #include <linux/slab.h> 125 #include <linux/syscalls.h> 126 #include <linux/time.h> 127 #include <linux/rcupdate.h> 128 #include <linux/pid_namespace.h> 129 #include <linux/hashtable.h> 130 #include <linux/percpu.h> 131 #include <linux/lglock.h> 132 133 #include <asm/uaccess.h> 134 135 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX) 136 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK) 137 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG)) 138 #define IS_FILE_PVT(fl) (fl->fl_flags & FL_FILE_PVT) 139 140 static bool lease_breaking(struct file_lock *fl) 141 { 142 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING); 143 } 144 145 static int target_leasetype(struct file_lock *fl) 146 { 147 if (fl->fl_flags & FL_UNLOCK_PENDING) 148 return F_UNLCK; 149 if (fl->fl_flags & FL_DOWNGRADE_PENDING) 150 return F_RDLCK; 151 return fl->fl_type; 152 } 153 154 int leases_enable = 1; 155 int lease_break_time = 45; 156 157 #define for_each_lock(inode, lockp) \ 158 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next) 159 160 /* 161 * The global file_lock_list is only used for displaying /proc/locks, so we 162 * keep a list on each CPU, with each list protected by its own spinlock via 163 * the file_lock_lglock. Note that alterations to the list also require that 164 * the relevant i_lock is held. 165 */ 166 DEFINE_STATIC_LGLOCK(file_lock_lglock); 167 static DEFINE_PER_CPU(struct hlist_head, file_lock_list); 168 169 /* 170 * The blocked_hash is used to find POSIX lock loops for deadlock detection. 171 * It is protected by blocked_lock_lock. 172 * 173 * We hash locks by lockowner in order to optimize searching for the lock a 174 * particular lockowner is waiting on. 175 * 176 * FIXME: make this value scale via some heuristic? We generally will want more 177 * buckets when we have more lockowners holding locks, but that's a little 178 * difficult to determine without knowing what the workload will look like. 179 */ 180 #define BLOCKED_HASH_BITS 7 181 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS); 182 183 /* 184 * This lock protects the blocked_hash. Generally, if you're accessing it, you 185 * want to be holding this lock. 186 * 187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next 188 * pointer for file_lock structures that are acting as lock requests (in 189 * contrast to those that are acting as records of acquired locks). 190 * 191 * Note that when we acquire this lock in order to change the above fields, 192 * we often hold the i_lock as well. In certain cases, when reading the fields 193 * protected by this lock, we can skip acquiring it iff we already hold the 194 * i_lock. 195 * 196 * In particular, adding an entry to the fl_block list requires that you hold 197 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting 198 * an entry from the list however only requires the file_lock_lock. 199 */ 200 static DEFINE_SPINLOCK(blocked_lock_lock); 201 202 static struct kmem_cache *filelock_cache __read_mostly; 203 204 static void locks_init_lock_heads(struct file_lock *fl) 205 { 206 INIT_HLIST_NODE(&fl->fl_link); 207 INIT_LIST_HEAD(&fl->fl_block); 208 init_waitqueue_head(&fl->fl_wait); 209 } 210 211 /* Allocate an empty lock structure. */ 212 struct file_lock *locks_alloc_lock(void) 213 { 214 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL); 215 216 if (fl) 217 locks_init_lock_heads(fl); 218 219 return fl; 220 } 221 EXPORT_SYMBOL_GPL(locks_alloc_lock); 222 223 void locks_release_private(struct file_lock *fl) 224 { 225 if (fl->fl_ops) { 226 if (fl->fl_ops->fl_release_private) 227 fl->fl_ops->fl_release_private(fl); 228 fl->fl_ops = NULL; 229 } 230 fl->fl_lmops = NULL; 231 232 } 233 EXPORT_SYMBOL_GPL(locks_release_private); 234 235 /* Free a lock which is not in use. */ 236 void locks_free_lock(struct file_lock *fl) 237 { 238 BUG_ON(waitqueue_active(&fl->fl_wait)); 239 BUG_ON(!list_empty(&fl->fl_block)); 240 BUG_ON(!hlist_unhashed(&fl->fl_link)); 241 242 locks_release_private(fl); 243 kmem_cache_free(filelock_cache, fl); 244 } 245 EXPORT_SYMBOL(locks_free_lock); 246 247 void locks_init_lock(struct file_lock *fl) 248 { 249 memset(fl, 0, sizeof(struct file_lock)); 250 locks_init_lock_heads(fl); 251 } 252 253 EXPORT_SYMBOL(locks_init_lock); 254 255 static void locks_copy_private(struct file_lock *new, struct file_lock *fl) 256 { 257 if (fl->fl_ops) { 258 if (fl->fl_ops->fl_copy_lock) 259 fl->fl_ops->fl_copy_lock(new, fl); 260 new->fl_ops = fl->fl_ops; 261 } 262 if (fl->fl_lmops) 263 new->fl_lmops = fl->fl_lmops; 264 } 265 266 /* 267 * Initialize a new lock from an existing file_lock structure. 268 */ 269 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl) 270 { 271 new->fl_owner = fl->fl_owner; 272 new->fl_pid = fl->fl_pid; 273 new->fl_file = NULL; 274 new->fl_flags = fl->fl_flags; 275 new->fl_type = fl->fl_type; 276 new->fl_start = fl->fl_start; 277 new->fl_end = fl->fl_end; 278 new->fl_ops = NULL; 279 new->fl_lmops = NULL; 280 } 281 EXPORT_SYMBOL(__locks_copy_lock); 282 283 void locks_copy_lock(struct file_lock *new, struct file_lock *fl) 284 { 285 locks_release_private(new); 286 287 __locks_copy_lock(new, fl); 288 new->fl_file = fl->fl_file; 289 new->fl_ops = fl->fl_ops; 290 new->fl_lmops = fl->fl_lmops; 291 292 locks_copy_private(new, fl); 293 } 294 295 EXPORT_SYMBOL(locks_copy_lock); 296 297 static inline int flock_translate_cmd(int cmd) { 298 if (cmd & LOCK_MAND) 299 return cmd & (LOCK_MAND | LOCK_RW); 300 switch (cmd) { 301 case LOCK_SH: 302 return F_RDLCK; 303 case LOCK_EX: 304 return F_WRLCK; 305 case LOCK_UN: 306 return F_UNLCK; 307 } 308 return -EINVAL; 309 } 310 311 /* Fill in a file_lock structure with an appropriate FLOCK lock. */ 312 static int flock_make_lock(struct file *filp, struct file_lock **lock, 313 unsigned int cmd) 314 { 315 struct file_lock *fl; 316 int type = flock_translate_cmd(cmd); 317 if (type < 0) 318 return type; 319 320 fl = locks_alloc_lock(); 321 if (fl == NULL) 322 return -ENOMEM; 323 324 fl->fl_file = filp; 325 fl->fl_pid = current->tgid; 326 fl->fl_flags = FL_FLOCK; 327 fl->fl_type = type; 328 fl->fl_end = OFFSET_MAX; 329 330 *lock = fl; 331 return 0; 332 } 333 334 static int assign_type(struct file_lock *fl, long type) 335 { 336 switch (type) { 337 case F_RDLCK: 338 case F_WRLCK: 339 case F_UNLCK: 340 fl->fl_type = type; 341 break; 342 default: 343 return -EINVAL; 344 } 345 return 0; 346 } 347 348 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl, 349 struct flock64 *l) 350 { 351 switch (l->l_whence) { 352 case SEEK_SET: 353 fl->fl_start = 0; 354 break; 355 case SEEK_CUR: 356 fl->fl_start = filp->f_pos; 357 break; 358 case SEEK_END: 359 fl->fl_start = i_size_read(file_inode(filp)); 360 break; 361 default: 362 return -EINVAL; 363 } 364 if (l->l_start > OFFSET_MAX - fl->fl_start) 365 return -EOVERFLOW; 366 fl->fl_start += l->l_start; 367 if (fl->fl_start < 0) 368 return -EINVAL; 369 370 /* POSIX-1996 leaves the case l->l_len < 0 undefined; 371 POSIX-2001 defines it. */ 372 if (l->l_len > 0) { 373 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start) 374 return -EOVERFLOW; 375 fl->fl_end = fl->fl_start + l->l_len - 1; 376 377 } else if (l->l_len < 0) { 378 if (fl->fl_start + l->l_len < 0) 379 return -EINVAL; 380 fl->fl_end = fl->fl_start - 1; 381 fl->fl_start += l->l_len; 382 } else 383 fl->fl_end = OFFSET_MAX; 384 385 fl->fl_owner = current->files; 386 fl->fl_pid = current->tgid; 387 fl->fl_file = filp; 388 fl->fl_flags = FL_POSIX; 389 fl->fl_ops = NULL; 390 fl->fl_lmops = NULL; 391 392 /* Ensure that fl->fl_filp has compatible f_mode */ 393 switch (l->l_type) { 394 case F_RDLCK: 395 if (!(filp->f_mode & FMODE_READ)) 396 return -EBADF; 397 break; 398 case F_WRLCK: 399 if (!(filp->f_mode & FMODE_WRITE)) 400 return -EBADF; 401 break; 402 } 403 404 return assign_type(fl, l->l_type); 405 } 406 407 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX 408 * style lock. 409 */ 410 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl, 411 struct flock *l) 412 { 413 struct flock64 ll = { 414 .l_type = l->l_type, 415 .l_whence = l->l_whence, 416 .l_start = l->l_start, 417 .l_len = l->l_len, 418 }; 419 420 return flock64_to_posix_lock(filp, fl, &ll); 421 } 422 423 /* default lease lock manager operations */ 424 static void lease_break_callback(struct file_lock *fl) 425 { 426 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG); 427 } 428 429 static const struct lock_manager_operations lease_manager_ops = { 430 .lm_break = lease_break_callback, 431 .lm_change = lease_modify, 432 }; 433 434 /* 435 * Initialize a lease, use the default lock manager operations 436 */ 437 static int lease_init(struct file *filp, long type, struct file_lock *fl) 438 { 439 if (assign_type(fl, type) != 0) 440 return -EINVAL; 441 442 fl->fl_owner = current->files; 443 fl->fl_pid = current->tgid; 444 445 fl->fl_file = filp; 446 fl->fl_flags = FL_LEASE; 447 fl->fl_start = 0; 448 fl->fl_end = OFFSET_MAX; 449 fl->fl_ops = NULL; 450 fl->fl_lmops = &lease_manager_ops; 451 return 0; 452 } 453 454 /* Allocate a file_lock initialised to this type of lease */ 455 static struct file_lock *lease_alloc(struct file *filp, long type) 456 { 457 struct file_lock *fl = locks_alloc_lock(); 458 int error = -ENOMEM; 459 460 if (fl == NULL) 461 return ERR_PTR(error); 462 463 error = lease_init(filp, type, fl); 464 if (error) { 465 locks_free_lock(fl); 466 return ERR_PTR(error); 467 } 468 return fl; 469 } 470 471 /* Check if two locks overlap each other. 472 */ 473 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2) 474 { 475 return ((fl1->fl_end >= fl2->fl_start) && 476 (fl2->fl_end >= fl1->fl_start)); 477 } 478 479 /* 480 * Check whether two locks have the same owner. 481 */ 482 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2) 483 { 484 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner) 485 return fl2->fl_lmops == fl1->fl_lmops && 486 fl1->fl_lmops->lm_compare_owner(fl1, fl2); 487 return fl1->fl_owner == fl2->fl_owner; 488 } 489 490 /* Must be called with the i_lock held! */ 491 static void locks_insert_global_locks(struct file_lock *fl) 492 { 493 lg_local_lock(&file_lock_lglock); 494 fl->fl_link_cpu = smp_processor_id(); 495 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list)); 496 lg_local_unlock(&file_lock_lglock); 497 } 498 499 /* Must be called with the i_lock held! */ 500 static void locks_delete_global_locks(struct file_lock *fl) 501 { 502 /* 503 * Avoid taking lock if already unhashed. This is safe since this check 504 * is done while holding the i_lock, and new insertions into the list 505 * also require that it be held. 506 */ 507 if (hlist_unhashed(&fl->fl_link)) 508 return; 509 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu); 510 hlist_del_init(&fl->fl_link); 511 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu); 512 } 513 514 static unsigned long 515 posix_owner_key(struct file_lock *fl) 516 { 517 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key) 518 return fl->fl_lmops->lm_owner_key(fl); 519 return (unsigned long)fl->fl_owner; 520 } 521 522 static void locks_insert_global_blocked(struct file_lock *waiter) 523 { 524 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter)); 525 } 526 527 static void locks_delete_global_blocked(struct file_lock *waiter) 528 { 529 hash_del(&waiter->fl_link); 530 } 531 532 /* Remove waiter from blocker's block list. 533 * When blocker ends up pointing to itself then the list is empty. 534 * 535 * Must be called with blocked_lock_lock held. 536 */ 537 static void __locks_delete_block(struct file_lock *waiter) 538 { 539 locks_delete_global_blocked(waiter); 540 list_del_init(&waiter->fl_block); 541 waiter->fl_next = NULL; 542 } 543 544 static void locks_delete_block(struct file_lock *waiter) 545 { 546 spin_lock(&blocked_lock_lock); 547 __locks_delete_block(waiter); 548 spin_unlock(&blocked_lock_lock); 549 } 550 551 /* Insert waiter into blocker's block list. 552 * We use a circular list so that processes can be easily woken up in 553 * the order they blocked. The documentation doesn't require this but 554 * it seems like the reasonable thing to do. 555 * 556 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block 557 * list itself is protected by the blocked_lock_lock, but by ensuring that the 558 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock 559 * in some cases when we see that the fl_block list is empty. 560 */ 561 static void __locks_insert_block(struct file_lock *blocker, 562 struct file_lock *waiter) 563 { 564 BUG_ON(!list_empty(&waiter->fl_block)); 565 waiter->fl_next = blocker; 566 list_add_tail(&waiter->fl_block, &blocker->fl_block); 567 if (IS_POSIX(blocker) && !IS_FILE_PVT(blocker)) 568 locks_insert_global_blocked(waiter); 569 } 570 571 /* Must be called with i_lock held. */ 572 static void locks_insert_block(struct file_lock *blocker, 573 struct file_lock *waiter) 574 { 575 spin_lock(&blocked_lock_lock); 576 __locks_insert_block(blocker, waiter); 577 spin_unlock(&blocked_lock_lock); 578 } 579 580 /* 581 * Wake up processes blocked waiting for blocker. 582 * 583 * Must be called with the inode->i_lock held! 584 */ 585 static void locks_wake_up_blocks(struct file_lock *blocker) 586 { 587 /* 588 * Avoid taking global lock if list is empty. This is safe since new 589 * blocked requests are only added to the list under the i_lock, and 590 * the i_lock is always held here. Note that removal from the fl_block 591 * list does not require the i_lock, so we must recheck list_empty() 592 * after acquiring the blocked_lock_lock. 593 */ 594 if (list_empty(&blocker->fl_block)) 595 return; 596 597 spin_lock(&blocked_lock_lock); 598 while (!list_empty(&blocker->fl_block)) { 599 struct file_lock *waiter; 600 601 waiter = list_first_entry(&blocker->fl_block, 602 struct file_lock, fl_block); 603 __locks_delete_block(waiter); 604 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify) 605 waiter->fl_lmops->lm_notify(waiter); 606 else 607 wake_up(&waiter->fl_wait); 608 } 609 spin_unlock(&blocked_lock_lock); 610 } 611 612 /* Insert file lock fl into an inode's lock list at the position indicated 613 * by pos. At the same time add the lock to the global file lock list. 614 * 615 * Must be called with the i_lock held! 616 */ 617 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl) 618 { 619 fl->fl_nspid = get_pid(task_tgid(current)); 620 621 /* insert into file's list */ 622 fl->fl_next = *pos; 623 *pos = fl; 624 625 locks_insert_global_locks(fl); 626 } 627 628 /** 629 * locks_delete_lock - Delete a lock and then free it. 630 * @thisfl_p: pointer that points to the fl_next field of the previous 631 * inode->i_flock list entry 632 * 633 * Unlink a lock from all lists and free the namespace reference, but don't 634 * free it yet. Wake up processes that are blocked waiting for this lock and 635 * notify the FS that the lock has been cleared. 636 * 637 * Must be called with the i_lock held! 638 */ 639 static void locks_unlink_lock(struct file_lock **thisfl_p) 640 { 641 struct file_lock *fl = *thisfl_p; 642 643 locks_delete_global_locks(fl); 644 645 *thisfl_p = fl->fl_next; 646 fl->fl_next = NULL; 647 648 if (fl->fl_nspid) { 649 put_pid(fl->fl_nspid); 650 fl->fl_nspid = NULL; 651 } 652 653 locks_wake_up_blocks(fl); 654 } 655 656 /* 657 * Unlink a lock from all lists and free it. 658 * 659 * Must be called with i_lock held! 660 */ 661 static void locks_delete_lock(struct file_lock **thisfl_p) 662 { 663 struct file_lock *fl = *thisfl_p; 664 665 locks_unlink_lock(thisfl_p); 666 locks_free_lock(fl); 667 } 668 669 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality 670 * checks for shared/exclusive status of overlapping locks. 671 */ 672 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) 673 { 674 if (sys_fl->fl_type == F_WRLCK) 675 return 1; 676 if (caller_fl->fl_type == F_WRLCK) 677 return 1; 678 return 0; 679 } 680 681 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific 682 * checking before calling the locks_conflict(). 683 */ 684 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) 685 { 686 /* POSIX locks owned by the same process do not conflict with 687 * each other. 688 */ 689 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl)) 690 return (0); 691 692 /* Check whether they overlap */ 693 if (!locks_overlap(caller_fl, sys_fl)) 694 return 0; 695 696 return (locks_conflict(caller_fl, sys_fl)); 697 } 698 699 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific 700 * checking before calling the locks_conflict(). 701 */ 702 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl) 703 { 704 /* FLOCK locks referring to the same filp do not conflict with 705 * each other. 706 */ 707 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file)) 708 return (0); 709 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND)) 710 return 0; 711 712 return (locks_conflict(caller_fl, sys_fl)); 713 } 714 715 void 716 posix_test_lock(struct file *filp, struct file_lock *fl) 717 { 718 struct file_lock *cfl; 719 struct inode *inode = file_inode(filp); 720 721 spin_lock(&inode->i_lock); 722 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) { 723 if (!IS_POSIX(cfl)) 724 continue; 725 if (posix_locks_conflict(fl, cfl)) 726 break; 727 } 728 if (cfl) { 729 __locks_copy_lock(fl, cfl); 730 if (cfl->fl_nspid) 731 fl->fl_pid = pid_vnr(cfl->fl_nspid); 732 } else 733 fl->fl_type = F_UNLCK; 734 spin_unlock(&inode->i_lock); 735 return; 736 } 737 EXPORT_SYMBOL(posix_test_lock); 738 739 /* 740 * Deadlock detection: 741 * 742 * We attempt to detect deadlocks that are due purely to posix file 743 * locks. 744 * 745 * We assume that a task can be waiting for at most one lock at a time. 746 * So for any acquired lock, the process holding that lock may be 747 * waiting on at most one other lock. That lock in turns may be held by 748 * someone waiting for at most one other lock. Given a requested lock 749 * caller_fl which is about to wait for a conflicting lock block_fl, we 750 * follow this chain of waiters to ensure we are not about to create a 751 * cycle. 752 * 753 * Since we do this before we ever put a process to sleep on a lock, we 754 * are ensured that there is never a cycle; that is what guarantees that 755 * the while() loop in posix_locks_deadlock() eventually completes. 756 * 757 * Note: the above assumption may not be true when handling lock 758 * requests from a broken NFS client. It may also fail in the presence 759 * of tasks (such as posix threads) sharing the same open file table. 760 * To handle those cases, we just bail out after a few iterations. 761 * 762 * For FL_FILE_PVT locks, the owner is the filp, not the files_struct. 763 * Because the owner is not even nominally tied to a thread of 764 * execution, the deadlock detection below can't reasonably work well. Just 765 * skip it for those. 766 * 767 * In principle, we could do a more limited deadlock detection on FL_FILE_PVT 768 * locks that just checks for the case where two tasks are attempting to 769 * upgrade from read to write locks on the same inode. 770 */ 771 772 #define MAX_DEADLK_ITERATIONS 10 773 774 /* Find a lock that the owner of the given block_fl is blocking on. */ 775 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl) 776 { 777 struct file_lock *fl; 778 779 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) { 780 if (posix_same_owner(fl, block_fl)) 781 return fl->fl_next; 782 } 783 return NULL; 784 } 785 786 /* Must be called with the blocked_lock_lock held! */ 787 static int posix_locks_deadlock(struct file_lock *caller_fl, 788 struct file_lock *block_fl) 789 { 790 int i = 0; 791 792 /* 793 * This deadlock detector can't reasonably detect deadlocks with 794 * FL_FILE_PVT locks, since they aren't owned by a process, per-se. 795 */ 796 if (IS_FILE_PVT(caller_fl)) 797 return 0; 798 799 while ((block_fl = what_owner_is_waiting_for(block_fl))) { 800 if (i++ > MAX_DEADLK_ITERATIONS) 801 return 0; 802 if (posix_same_owner(caller_fl, block_fl)) 803 return 1; 804 } 805 return 0; 806 } 807 808 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks 809 * after any leases, but before any posix locks. 810 * 811 * Note that if called with an FL_EXISTS argument, the caller may determine 812 * whether or not a lock was successfully freed by testing the return 813 * value for -ENOENT. 814 */ 815 static int flock_lock_file(struct file *filp, struct file_lock *request) 816 { 817 struct file_lock *new_fl = NULL; 818 struct file_lock **before; 819 struct inode * inode = file_inode(filp); 820 int error = 0; 821 int found = 0; 822 823 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) { 824 new_fl = locks_alloc_lock(); 825 if (!new_fl) 826 return -ENOMEM; 827 } 828 829 spin_lock(&inode->i_lock); 830 if (request->fl_flags & FL_ACCESS) 831 goto find_conflict; 832 833 for_each_lock(inode, before) { 834 struct file_lock *fl = *before; 835 if (IS_POSIX(fl)) 836 break; 837 if (IS_LEASE(fl)) 838 continue; 839 if (filp != fl->fl_file) 840 continue; 841 if (request->fl_type == fl->fl_type) 842 goto out; 843 found = 1; 844 locks_delete_lock(before); 845 break; 846 } 847 848 if (request->fl_type == F_UNLCK) { 849 if ((request->fl_flags & FL_EXISTS) && !found) 850 error = -ENOENT; 851 goto out; 852 } 853 854 /* 855 * If a higher-priority process was blocked on the old file lock, 856 * give it the opportunity to lock the file. 857 */ 858 if (found) { 859 spin_unlock(&inode->i_lock); 860 cond_resched(); 861 spin_lock(&inode->i_lock); 862 } 863 864 find_conflict: 865 for_each_lock(inode, before) { 866 struct file_lock *fl = *before; 867 if (IS_POSIX(fl)) 868 break; 869 if (IS_LEASE(fl)) 870 continue; 871 if (!flock_locks_conflict(request, fl)) 872 continue; 873 error = -EAGAIN; 874 if (!(request->fl_flags & FL_SLEEP)) 875 goto out; 876 error = FILE_LOCK_DEFERRED; 877 locks_insert_block(fl, request); 878 goto out; 879 } 880 if (request->fl_flags & FL_ACCESS) 881 goto out; 882 locks_copy_lock(new_fl, request); 883 locks_insert_lock(before, new_fl); 884 new_fl = NULL; 885 error = 0; 886 887 out: 888 spin_unlock(&inode->i_lock); 889 if (new_fl) 890 locks_free_lock(new_fl); 891 return error; 892 } 893 894 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock) 895 { 896 struct file_lock *fl; 897 struct file_lock *new_fl = NULL; 898 struct file_lock *new_fl2 = NULL; 899 struct file_lock *left = NULL; 900 struct file_lock *right = NULL; 901 struct file_lock **before; 902 int error; 903 bool added = false; 904 905 /* 906 * We may need two file_lock structures for this operation, 907 * so we get them in advance to avoid races. 908 * 909 * In some cases we can be sure, that no new locks will be needed 910 */ 911 if (!(request->fl_flags & FL_ACCESS) && 912 (request->fl_type != F_UNLCK || 913 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) { 914 new_fl = locks_alloc_lock(); 915 new_fl2 = locks_alloc_lock(); 916 } 917 918 spin_lock(&inode->i_lock); 919 /* 920 * New lock request. Walk all POSIX locks and look for conflicts. If 921 * there are any, either return error or put the request on the 922 * blocker's list of waiters and the global blocked_hash. 923 */ 924 if (request->fl_type != F_UNLCK) { 925 for_each_lock(inode, before) { 926 fl = *before; 927 if (!IS_POSIX(fl)) 928 continue; 929 if (!posix_locks_conflict(request, fl)) 930 continue; 931 if (conflock) 932 __locks_copy_lock(conflock, fl); 933 error = -EAGAIN; 934 if (!(request->fl_flags & FL_SLEEP)) 935 goto out; 936 /* 937 * Deadlock detection and insertion into the blocked 938 * locks list must be done while holding the same lock! 939 */ 940 error = -EDEADLK; 941 spin_lock(&blocked_lock_lock); 942 if (likely(!posix_locks_deadlock(request, fl))) { 943 error = FILE_LOCK_DEFERRED; 944 __locks_insert_block(fl, request); 945 } 946 spin_unlock(&blocked_lock_lock); 947 goto out; 948 } 949 } 950 951 /* If we're just looking for a conflict, we're done. */ 952 error = 0; 953 if (request->fl_flags & FL_ACCESS) 954 goto out; 955 956 /* 957 * Find the first old lock with the same owner as the new lock. 958 */ 959 960 before = &inode->i_flock; 961 962 /* First skip locks owned by other processes. */ 963 while ((fl = *before) && (!IS_POSIX(fl) || 964 !posix_same_owner(request, fl))) { 965 before = &fl->fl_next; 966 } 967 968 /* Process locks with this owner. */ 969 while ((fl = *before) && posix_same_owner(request, fl)) { 970 /* Detect adjacent or overlapping regions (if same lock type) 971 */ 972 if (request->fl_type == fl->fl_type) { 973 /* In all comparisons of start vs end, use 974 * "start - 1" rather than "end + 1". If end 975 * is OFFSET_MAX, end + 1 will become negative. 976 */ 977 if (fl->fl_end < request->fl_start - 1) 978 goto next_lock; 979 /* If the next lock in the list has entirely bigger 980 * addresses than the new one, insert the lock here. 981 */ 982 if (fl->fl_start - 1 > request->fl_end) 983 break; 984 985 /* If we come here, the new and old lock are of the 986 * same type and adjacent or overlapping. Make one 987 * lock yielding from the lower start address of both 988 * locks to the higher end address. 989 */ 990 if (fl->fl_start > request->fl_start) 991 fl->fl_start = request->fl_start; 992 else 993 request->fl_start = fl->fl_start; 994 if (fl->fl_end < request->fl_end) 995 fl->fl_end = request->fl_end; 996 else 997 request->fl_end = fl->fl_end; 998 if (added) { 999 locks_delete_lock(before); 1000 continue; 1001 } 1002 request = fl; 1003 added = true; 1004 } 1005 else { 1006 /* Processing for different lock types is a bit 1007 * more complex. 1008 */ 1009 if (fl->fl_end < request->fl_start) 1010 goto next_lock; 1011 if (fl->fl_start > request->fl_end) 1012 break; 1013 if (request->fl_type == F_UNLCK) 1014 added = true; 1015 if (fl->fl_start < request->fl_start) 1016 left = fl; 1017 /* If the next lock in the list has a higher end 1018 * address than the new one, insert the new one here. 1019 */ 1020 if (fl->fl_end > request->fl_end) { 1021 right = fl; 1022 break; 1023 } 1024 if (fl->fl_start >= request->fl_start) { 1025 /* The new lock completely replaces an old 1026 * one (This may happen several times). 1027 */ 1028 if (added) { 1029 locks_delete_lock(before); 1030 continue; 1031 } 1032 /* Replace the old lock with the new one. 1033 * Wake up anybody waiting for the old one, 1034 * as the change in lock type might satisfy 1035 * their needs. 1036 */ 1037 locks_wake_up_blocks(fl); 1038 fl->fl_start = request->fl_start; 1039 fl->fl_end = request->fl_end; 1040 fl->fl_type = request->fl_type; 1041 locks_release_private(fl); 1042 locks_copy_private(fl, request); 1043 request = fl; 1044 added = true; 1045 } 1046 } 1047 /* Go on to next lock. 1048 */ 1049 next_lock: 1050 before = &fl->fl_next; 1051 } 1052 1053 /* 1054 * The above code only modifies existing locks in case of merging or 1055 * replacing. If new lock(s) need to be inserted all modifications are 1056 * done below this, so it's safe yet to bail out. 1057 */ 1058 error = -ENOLCK; /* "no luck" */ 1059 if (right && left == right && !new_fl2) 1060 goto out; 1061 1062 error = 0; 1063 if (!added) { 1064 if (request->fl_type == F_UNLCK) { 1065 if (request->fl_flags & FL_EXISTS) 1066 error = -ENOENT; 1067 goto out; 1068 } 1069 1070 if (!new_fl) { 1071 error = -ENOLCK; 1072 goto out; 1073 } 1074 locks_copy_lock(new_fl, request); 1075 locks_insert_lock(before, new_fl); 1076 new_fl = NULL; 1077 } 1078 if (right) { 1079 if (left == right) { 1080 /* The new lock breaks the old one in two pieces, 1081 * so we have to use the second new lock. 1082 */ 1083 left = new_fl2; 1084 new_fl2 = NULL; 1085 locks_copy_lock(left, right); 1086 locks_insert_lock(before, left); 1087 } 1088 right->fl_start = request->fl_end + 1; 1089 locks_wake_up_blocks(right); 1090 } 1091 if (left) { 1092 left->fl_end = request->fl_start - 1; 1093 locks_wake_up_blocks(left); 1094 } 1095 out: 1096 spin_unlock(&inode->i_lock); 1097 /* 1098 * Free any unused locks. 1099 */ 1100 if (new_fl) 1101 locks_free_lock(new_fl); 1102 if (new_fl2) 1103 locks_free_lock(new_fl2); 1104 return error; 1105 } 1106 1107 /** 1108 * posix_lock_file - Apply a POSIX-style lock to a file 1109 * @filp: The file to apply the lock to 1110 * @fl: The lock to be applied 1111 * @conflock: Place to return a copy of the conflicting lock, if found. 1112 * 1113 * Add a POSIX style lock to a file. 1114 * We merge adjacent & overlapping locks whenever possible. 1115 * POSIX locks are sorted by owner task, then by starting address 1116 * 1117 * Note that if called with an FL_EXISTS argument, the caller may determine 1118 * whether or not a lock was successfully freed by testing the return 1119 * value for -ENOENT. 1120 */ 1121 int posix_lock_file(struct file *filp, struct file_lock *fl, 1122 struct file_lock *conflock) 1123 { 1124 return __posix_lock_file(file_inode(filp), fl, conflock); 1125 } 1126 EXPORT_SYMBOL(posix_lock_file); 1127 1128 /** 1129 * posix_lock_file_wait - Apply a POSIX-style lock to a file 1130 * @filp: The file to apply the lock to 1131 * @fl: The lock to be applied 1132 * 1133 * Add a POSIX style lock to a file. 1134 * We merge adjacent & overlapping locks whenever possible. 1135 * POSIX locks are sorted by owner task, then by starting address 1136 */ 1137 int posix_lock_file_wait(struct file *filp, struct file_lock *fl) 1138 { 1139 int error; 1140 might_sleep (); 1141 for (;;) { 1142 error = posix_lock_file(filp, fl, NULL); 1143 if (error != FILE_LOCK_DEFERRED) 1144 break; 1145 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); 1146 if (!error) 1147 continue; 1148 1149 locks_delete_block(fl); 1150 break; 1151 } 1152 return error; 1153 } 1154 EXPORT_SYMBOL(posix_lock_file_wait); 1155 1156 /** 1157 * locks_mandatory_locked - Check for an active lock 1158 * @file: the file to check 1159 * 1160 * Searches the inode's list of locks to find any POSIX locks which conflict. 1161 * This function is called from locks_verify_locked() only. 1162 */ 1163 int locks_mandatory_locked(struct file *file) 1164 { 1165 struct inode *inode = file_inode(file); 1166 fl_owner_t owner = current->files; 1167 struct file_lock *fl; 1168 1169 /* 1170 * Search the lock list for this inode for any POSIX locks. 1171 */ 1172 spin_lock(&inode->i_lock); 1173 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { 1174 if (!IS_POSIX(fl)) 1175 continue; 1176 if (fl->fl_owner != owner && fl->fl_owner != (fl_owner_t)file) 1177 break; 1178 } 1179 spin_unlock(&inode->i_lock); 1180 return fl ? -EAGAIN : 0; 1181 } 1182 1183 /** 1184 * locks_mandatory_area - Check for a conflicting lock 1185 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ 1186 * for shared 1187 * @inode: the file to check 1188 * @filp: how the file was opened (if it was) 1189 * @offset: start of area to check 1190 * @count: length of area to check 1191 * 1192 * Searches the inode's list of locks to find any POSIX locks which conflict. 1193 * This function is called from rw_verify_area() and 1194 * locks_verify_truncate(). 1195 */ 1196 int locks_mandatory_area(int read_write, struct inode *inode, 1197 struct file *filp, loff_t offset, 1198 size_t count) 1199 { 1200 struct file_lock fl; 1201 int error; 1202 bool sleep = false; 1203 1204 locks_init_lock(&fl); 1205 fl.fl_pid = current->tgid; 1206 fl.fl_file = filp; 1207 fl.fl_flags = FL_POSIX | FL_ACCESS; 1208 if (filp && !(filp->f_flags & O_NONBLOCK)) 1209 sleep = true; 1210 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK; 1211 fl.fl_start = offset; 1212 fl.fl_end = offset + count - 1; 1213 1214 for (;;) { 1215 if (filp) { 1216 fl.fl_owner = (fl_owner_t)filp; 1217 fl.fl_flags &= ~FL_SLEEP; 1218 error = __posix_lock_file(inode, &fl, NULL); 1219 if (!error) 1220 break; 1221 } 1222 1223 if (sleep) 1224 fl.fl_flags |= FL_SLEEP; 1225 fl.fl_owner = current->files; 1226 error = __posix_lock_file(inode, &fl, NULL); 1227 if (error != FILE_LOCK_DEFERRED) 1228 break; 1229 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next); 1230 if (!error) { 1231 /* 1232 * If we've been sleeping someone might have 1233 * changed the permissions behind our back. 1234 */ 1235 if (__mandatory_lock(inode)) 1236 continue; 1237 } 1238 1239 locks_delete_block(&fl); 1240 break; 1241 } 1242 1243 return error; 1244 } 1245 1246 EXPORT_SYMBOL(locks_mandatory_area); 1247 1248 static void lease_clear_pending(struct file_lock *fl, int arg) 1249 { 1250 switch (arg) { 1251 case F_UNLCK: 1252 fl->fl_flags &= ~FL_UNLOCK_PENDING; 1253 /* fall through: */ 1254 case F_RDLCK: 1255 fl->fl_flags &= ~FL_DOWNGRADE_PENDING; 1256 } 1257 } 1258 1259 /* We already had a lease on this file; just change its type */ 1260 int lease_modify(struct file_lock **before, int arg) 1261 { 1262 struct file_lock *fl = *before; 1263 int error = assign_type(fl, arg); 1264 1265 if (error) 1266 return error; 1267 lease_clear_pending(fl, arg); 1268 locks_wake_up_blocks(fl); 1269 if (arg == F_UNLCK) { 1270 struct file *filp = fl->fl_file; 1271 1272 f_delown(filp); 1273 filp->f_owner.signum = 0; 1274 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync); 1275 if (fl->fl_fasync != NULL) { 1276 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync); 1277 fl->fl_fasync = NULL; 1278 } 1279 locks_delete_lock(before); 1280 } 1281 return 0; 1282 } 1283 1284 EXPORT_SYMBOL(lease_modify); 1285 1286 static bool past_time(unsigned long then) 1287 { 1288 if (!then) 1289 /* 0 is a special value meaning "this never expires": */ 1290 return false; 1291 return time_after(jiffies, then); 1292 } 1293 1294 static void time_out_leases(struct inode *inode) 1295 { 1296 struct file_lock **before; 1297 struct file_lock *fl; 1298 1299 before = &inode->i_flock; 1300 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) { 1301 if (past_time(fl->fl_downgrade_time)) 1302 lease_modify(before, F_RDLCK); 1303 if (past_time(fl->fl_break_time)) 1304 lease_modify(before, F_UNLCK); 1305 if (fl == *before) /* lease_modify may have freed fl */ 1306 before = &fl->fl_next; 1307 } 1308 } 1309 1310 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker) 1311 { 1312 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) 1313 return false; 1314 return locks_conflict(breaker, lease); 1315 } 1316 1317 /** 1318 * __break_lease - revoke all outstanding leases on file 1319 * @inode: the inode of the file to return 1320 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR: 1321 * break all leases 1322 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break 1323 * only delegations 1324 * 1325 * break_lease (inlined for speed) has checked there already is at least 1326 * some kind of lock (maybe a lease) on this file. Leases are broken on 1327 * a call to open() or truncate(). This function can sleep unless you 1328 * specified %O_NONBLOCK to your open(). 1329 */ 1330 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type) 1331 { 1332 int error = 0; 1333 struct file_lock *new_fl, *flock; 1334 struct file_lock *fl; 1335 unsigned long break_time; 1336 int i_have_this_lease = 0; 1337 bool lease_conflict = false; 1338 int want_write = (mode & O_ACCMODE) != O_RDONLY; 1339 1340 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK); 1341 if (IS_ERR(new_fl)) 1342 return PTR_ERR(new_fl); 1343 new_fl->fl_flags = type; 1344 1345 spin_lock(&inode->i_lock); 1346 1347 time_out_leases(inode); 1348 1349 flock = inode->i_flock; 1350 if ((flock == NULL) || !IS_LEASE(flock)) 1351 goto out; 1352 1353 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) { 1354 if (leases_conflict(fl, new_fl)) { 1355 lease_conflict = true; 1356 if (fl->fl_owner == current->files) 1357 i_have_this_lease = 1; 1358 } 1359 } 1360 if (!lease_conflict) 1361 goto out; 1362 1363 break_time = 0; 1364 if (lease_break_time > 0) { 1365 break_time = jiffies + lease_break_time * HZ; 1366 if (break_time == 0) 1367 break_time++; /* so that 0 means no break time */ 1368 } 1369 1370 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) { 1371 if (!leases_conflict(fl, new_fl)) 1372 continue; 1373 if (want_write) { 1374 if (fl->fl_flags & FL_UNLOCK_PENDING) 1375 continue; 1376 fl->fl_flags |= FL_UNLOCK_PENDING; 1377 fl->fl_break_time = break_time; 1378 } else { 1379 if (lease_breaking(flock)) 1380 continue; 1381 fl->fl_flags |= FL_DOWNGRADE_PENDING; 1382 fl->fl_downgrade_time = break_time; 1383 } 1384 fl->fl_lmops->lm_break(fl); 1385 } 1386 1387 if (i_have_this_lease || (mode & O_NONBLOCK)) { 1388 error = -EWOULDBLOCK; 1389 goto out; 1390 } 1391 1392 restart: 1393 break_time = flock->fl_break_time; 1394 if (break_time != 0) { 1395 break_time -= jiffies; 1396 if (break_time == 0) 1397 break_time++; 1398 } 1399 locks_insert_block(flock, new_fl); 1400 spin_unlock(&inode->i_lock); 1401 error = wait_event_interruptible_timeout(new_fl->fl_wait, 1402 !new_fl->fl_next, break_time); 1403 spin_lock(&inode->i_lock); 1404 locks_delete_block(new_fl); 1405 if (error >= 0) { 1406 if (error == 0) 1407 time_out_leases(inode); 1408 /* 1409 * Wait for the next conflicting lease that has not been 1410 * broken yet 1411 */ 1412 for (flock = inode->i_flock; flock && IS_LEASE(flock); 1413 flock = flock->fl_next) { 1414 if (leases_conflict(new_fl, flock)) 1415 goto restart; 1416 } 1417 error = 0; 1418 } 1419 1420 out: 1421 spin_unlock(&inode->i_lock); 1422 locks_free_lock(new_fl); 1423 return error; 1424 } 1425 1426 EXPORT_SYMBOL(__break_lease); 1427 1428 /** 1429 * lease_get_mtime - get the last modified time of an inode 1430 * @inode: the inode 1431 * @time: pointer to a timespec which will contain the last modified time 1432 * 1433 * This is to force NFS clients to flush their caches for files with 1434 * exclusive leases. The justification is that if someone has an 1435 * exclusive lease, then they could be modifying it. 1436 */ 1437 void lease_get_mtime(struct inode *inode, struct timespec *time) 1438 { 1439 struct file_lock *flock = inode->i_flock; 1440 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK)) 1441 *time = current_fs_time(inode->i_sb); 1442 else 1443 *time = inode->i_mtime; 1444 } 1445 1446 EXPORT_SYMBOL(lease_get_mtime); 1447 1448 /** 1449 * fcntl_getlease - Enquire what lease is currently active 1450 * @filp: the file 1451 * 1452 * The value returned by this function will be one of 1453 * (if no lease break is pending): 1454 * 1455 * %F_RDLCK to indicate a shared lease is held. 1456 * 1457 * %F_WRLCK to indicate an exclusive lease is held. 1458 * 1459 * %F_UNLCK to indicate no lease is held. 1460 * 1461 * (if a lease break is pending): 1462 * 1463 * %F_RDLCK to indicate an exclusive lease needs to be 1464 * changed to a shared lease (or removed). 1465 * 1466 * %F_UNLCK to indicate the lease needs to be removed. 1467 * 1468 * XXX: sfr & willy disagree over whether F_INPROGRESS 1469 * should be returned to userspace. 1470 */ 1471 int fcntl_getlease(struct file *filp) 1472 { 1473 struct file_lock *fl; 1474 struct inode *inode = file_inode(filp); 1475 int type = F_UNLCK; 1476 1477 spin_lock(&inode->i_lock); 1478 time_out_leases(file_inode(filp)); 1479 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl); 1480 fl = fl->fl_next) { 1481 if (fl->fl_file == filp) { 1482 type = target_leasetype(fl); 1483 break; 1484 } 1485 } 1486 spin_unlock(&inode->i_lock); 1487 return type; 1488 } 1489 1490 /** 1491 * check_conflicting_open - see if the given dentry points to a file that has 1492 * an existing open that would conflict with the 1493 * desired lease. 1494 * @dentry: dentry to check 1495 * @arg: type of lease that we're trying to acquire 1496 * 1497 * Check to see if there's an existing open fd on this file that would 1498 * conflict with the lease we're trying to set. 1499 */ 1500 static int 1501 check_conflicting_open(const struct dentry *dentry, const long arg) 1502 { 1503 int ret = 0; 1504 struct inode *inode = dentry->d_inode; 1505 1506 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0)) 1507 return -EAGAIN; 1508 1509 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) || 1510 (atomic_read(&inode->i_count) > 1))) 1511 ret = -EAGAIN; 1512 1513 return ret; 1514 } 1515 1516 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp) 1517 { 1518 struct file_lock *fl, **before, **my_before = NULL, *lease; 1519 struct dentry *dentry = filp->f_path.dentry; 1520 struct inode *inode = dentry->d_inode; 1521 bool is_deleg = (*flp)->fl_flags & FL_DELEG; 1522 int error; 1523 1524 lease = *flp; 1525 /* 1526 * In the delegation case we need mutual exclusion with 1527 * a number of operations that take the i_mutex. We trylock 1528 * because delegations are an optional optimization, and if 1529 * there's some chance of a conflict--we'd rather not 1530 * bother, maybe that's a sign this just isn't a good file to 1531 * hand out a delegation on. 1532 */ 1533 if (is_deleg && !mutex_trylock(&inode->i_mutex)) 1534 return -EAGAIN; 1535 1536 if (is_deleg && arg == F_WRLCK) { 1537 /* Write delegations are not currently supported: */ 1538 mutex_unlock(&inode->i_mutex); 1539 WARN_ON_ONCE(1); 1540 return -EINVAL; 1541 } 1542 1543 error = check_conflicting_open(dentry, arg); 1544 if (error) 1545 goto out; 1546 1547 /* 1548 * At this point, we know that if there is an exclusive 1549 * lease on this file, then we hold it on this filp 1550 * (otherwise our open of this file would have blocked). 1551 * And if we are trying to acquire an exclusive lease, 1552 * then the file is not open by anyone (including us) 1553 * except for this filp. 1554 */ 1555 error = -EAGAIN; 1556 for (before = &inode->i_flock; 1557 ((fl = *before) != NULL) && IS_LEASE(fl); 1558 before = &fl->fl_next) { 1559 if (fl->fl_file == filp) { 1560 my_before = before; 1561 continue; 1562 } 1563 /* 1564 * No exclusive leases if someone else has a lease on 1565 * this file: 1566 */ 1567 if (arg == F_WRLCK) 1568 goto out; 1569 /* 1570 * Modifying our existing lease is OK, but no getting a 1571 * new lease if someone else is opening for write: 1572 */ 1573 if (fl->fl_flags & FL_UNLOCK_PENDING) 1574 goto out; 1575 } 1576 1577 if (my_before != NULL) { 1578 error = lease->fl_lmops->lm_change(my_before, arg); 1579 if (!error) 1580 *flp = *my_before; 1581 goto out; 1582 } 1583 1584 error = -EINVAL; 1585 if (!leases_enable) 1586 goto out; 1587 1588 locks_insert_lock(before, lease); 1589 /* 1590 * The check in break_lease() is lockless. It's possible for another 1591 * open to race in after we did the earlier check for a conflicting 1592 * open but before the lease was inserted. Check again for a 1593 * conflicting open and cancel the lease if there is one. 1594 * 1595 * We also add a barrier here to ensure that the insertion of the lock 1596 * precedes these checks. 1597 */ 1598 smp_mb(); 1599 error = check_conflicting_open(dentry, arg); 1600 if (error) 1601 locks_unlink_lock(flp); 1602 out: 1603 if (is_deleg) 1604 mutex_unlock(&inode->i_mutex); 1605 return error; 1606 } 1607 1608 static int generic_delete_lease(struct file *filp, struct file_lock **flp) 1609 { 1610 struct file_lock *fl, **before; 1611 struct dentry *dentry = filp->f_path.dentry; 1612 struct inode *inode = dentry->d_inode; 1613 1614 for (before = &inode->i_flock; 1615 ((fl = *before) != NULL) && IS_LEASE(fl); 1616 before = &fl->fl_next) { 1617 if (fl->fl_file != filp) 1618 continue; 1619 return (*flp)->fl_lmops->lm_change(before, F_UNLCK); 1620 } 1621 return -EAGAIN; 1622 } 1623 1624 /** 1625 * generic_setlease - sets a lease on an open file 1626 * @filp: file pointer 1627 * @arg: type of lease to obtain 1628 * @flp: input - file_lock to use, output - file_lock inserted 1629 * 1630 * The (input) flp->fl_lmops->lm_break function is required 1631 * by break_lease(). 1632 * 1633 * Called with inode->i_lock held. 1634 */ 1635 int generic_setlease(struct file *filp, long arg, struct file_lock **flp) 1636 { 1637 struct dentry *dentry = filp->f_path.dentry; 1638 struct inode *inode = dentry->d_inode; 1639 int error; 1640 1641 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE)) 1642 return -EACCES; 1643 if (!S_ISREG(inode->i_mode)) 1644 return -EINVAL; 1645 error = security_file_lock(filp, arg); 1646 if (error) 1647 return error; 1648 1649 time_out_leases(inode); 1650 1651 BUG_ON(!(*flp)->fl_lmops->lm_break); 1652 1653 switch (arg) { 1654 case F_UNLCK: 1655 return generic_delete_lease(filp, flp); 1656 case F_RDLCK: 1657 case F_WRLCK: 1658 return generic_add_lease(filp, arg, flp); 1659 default: 1660 return -EINVAL; 1661 } 1662 } 1663 EXPORT_SYMBOL(generic_setlease); 1664 1665 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease) 1666 { 1667 if (filp->f_op->setlease) 1668 return filp->f_op->setlease(filp, arg, lease); 1669 else 1670 return generic_setlease(filp, arg, lease); 1671 } 1672 1673 /** 1674 * vfs_setlease - sets a lease on an open file 1675 * @filp: file pointer 1676 * @arg: type of lease to obtain 1677 * @lease: file_lock to use 1678 * 1679 * Call this to establish a lease on the file. 1680 * The (*lease)->fl_lmops->lm_break operation must be set; if not, 1681 * break_lease will oops! 1682 * 1683 * This will call the filesystem's setlease file method, if 1684 * defined. Note that there is no getlease method; instead, the 1685 * filesystem setlease method should call back to setlease() to 1686 * add a lease to the inode's lease list, where fcntl_getlease() can 1687 * find it. Since fcntl_getlease() only reports whether the current 1688 * task holds a lease, a cluster filesystem need only do this for 1689 * leases held by processes on this node. 1690 * 1691 * There is also no break_lease method; filesystems that 1692 * handle their own leases should break leases themselves from the 1693 * filesystem's open, create, and (on truncate) setattr methods. 1694 * 1695 * Warning: the only current setlease methods exist only to disable 1696 * leases in certain cases. More vfs changes may be required to 1697 * allow a full filesystem lease implementation. 1698 */ 1699 1700 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease) 1701 { 1702 struct inode *inode = file_inode(filp); 1703 int error; 1704 1705 spin_lock(&inode->i_lock); 1706 error = __vfs_setlease(filp, arg, lease); 1707 spin_unlock(&inode->i_lock); 1708 1709 return error; 1710 } 1711 EXPORT_SYMBOL_GPL(vfs_setlease); 1712 1713 static int do_fcntl_delete_lease(struct file *filp) 1714 { 1715 struct file_lock fl, *flp = &fl; 1716 1717 lease_init(filp, F_UNLCK, flp); 1718 1719 return vfs_setlease(filp, F_UNLCK, &flp); 1720 } 1721 1722 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg) 1723 { 1724 struct file_lock *fl, *ret; 1725 struct inode *inode = file_inode(filp); 1726 struct fasync_struct *new; 1727 int error; 1728 1729 fl = lease_alloc(filp, arg); 1730 if (IS_ERR(fl)) 1731 return PTR_ERR(fl); 1732 1733 new = fasync_alloc(); 1734 if (!new) { 1735 locks_free_lock(fl); 1736 return -ENOMEM; 1737 } 1738 ret = fl; 1739 spin_lock(&inode->i_lock); 1740 error = __vfs_setlease(filp, arg, &ret); 1741 if (error) { 1742 spin_unlock(&inode->i_lock); 1743 locks_free_lock(fl); 1744 goto out_free_fasync; 1745 } 1746 if (ret != fl) 1747 locks_free_lock(fl); 1748 1749 /* 1750 * fasync_insert_entry() returns the old entry if any. 1751 * If there was no old entry, then it used 'new' and 1752 * inserted it into the fasync list. Clear new so that 1753 * we don't release it here. 1754 */ 1755 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new)) 1756 new = NULL; 1757 1758 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0); 1759 spin_unlock(&inode->i_lock); 1760 1761 out_free_fasync: 1762 if (new) 1763 fasync_free(new); 1764 return error; 1765 } 1766 1767 /** 1768 * fcntl_setlease - sets a lease on an open file 1769 * @fd: open file descriptor 1770 * @filp: file pointer 1771 * @arg: type of lease to obtain 1772 * 1773 * Call this fcntl to establish a lease on the file. 1774 * Note that you also need to call %F_SETSIG to 1775 * receive a signal when the lease is broken. 1776 */ 1777 int fcntl_setlease(unsigned int fd, struct file *filp, long arg) 1778 { 1779 if (arg == F_UNLCK) 1780 return do_fcntl_delete_lease(filp); 1781 return do_fcntl_add_lease(fd, filp, arg); 1782 } 1783 1784 /** 1785 * flock_lock_file_wait - Apply a FLOCK-style lock to a file 1786 * @filp: The file to apply the lock to 1787 * @fl: The lock to be applied 1788 * 1789 * Add a FLOCK style lock to a file. 1790 */ 1791 int flock_lock_file_wait(struct file *filp, struct file_lock *fl) 1792 { 1793 int error; 1794 might_sleep(); 1795 for (;;) { 1796 error = flock_lock_file(filp, fl); 1797 if (error != FILE_LOCK_DEFERRED) 1798 break; 1799 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); 1800 if (!error) 1801 continue; 1802 1803 locks_delete_block(fl); 1804 break; 1805 } 1806 return error; 1807 } 1808 1809 EXPORT_SYMBOL(flock_lock_file_wait); 1810 1811 /** 1812 * sys_flock: - flock() system call. 1813 * @fd: the file descriptor to lock. 1814 * @cmd: the type of lock to apply. 1815 * 1816 * Apply a %FL_FLOCK style lock to an open file descriptor. 1817 * The @cmd can be one of 1818 * 1819 * %LOCK_SH -- a shared lock. 1820 * 1821 * %LOCK_EX -- an exclusive lock. 1822 * 1823 * %LOCK_UN -- remove an existing lock. 1824 * 1825 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes. 1826 * 1827 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other 1828 * processes read and write access respectively. 1829 */ 1830 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd) 1831 { 1832 struct fd f = fdget(fd); 1833 struct file_lock *lock; 1834 int can_sleep, unlock; 1835 int error; 1836 1837 error = -EBADF; 1838 if (!f.file) 1839 goto out; 1840 1841 can_sleep = !(cmd & LOCK_NB); 1842 cmd &= ~LOCK_NB; 1843 unlock = (cmd == LOCK_UN); 1844 1845 if (!unlock && !(cmd & LOCK_MAND) && 1846 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE))) 1847 goto out_putf; 1848 1849 error = flock_make_lock(f.file, &lock, cmd); 1850 if (error) 1851 goto out_putf; 1852 if (can_sleep) 1853 lock->fl_flags |= FL_SLEEP; 1854 1855 error = security_file_lock(f.file, lock->fl_type); 1856 if (error) 1857 goto out_free; 1858 1859 if (f.file->f_op->flock) 1860 error = f.file->f_op->flock(f.file, 1861 (can_sleep) ? F_SETLKW : F_SETLK, 1862 lock); 1863 else 1864 error = flock_lock_file_wait(f.file, lock); 1865 1866 out_free: 1867 locks_free_lock(lock); 1868 1869 out_putf: 1870 fdput(f); 1871 out: 1872 return error; 1873 } 1874 1875 /** 1876 * vfs_test_lock - test file byte range lock 1877 * @filp: The file to test lock for 1878 * @fl: The lock to test; also used to hold result 1879 * 1880 * Returns -ERRNO on failure. Indicates presence of conflicting lock by 1881 * setting conf->fl_type to something other than F_UNLCK. 1882 */ 1883 int vfs_test_lock(struct file *filp, struct file_lock *fl) 1884 { 1885 if (filp->f_op->lock) 1886 return filp->f_op->lock(filp, F_GETLK, fl); 1887 posix_test_lock(filp, fl); 1888 return 0; 1889 } 1890 EXPORT_SYMBOL_GPL(vfs_test_lock); 1891 1892 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl) 1893 { 1894 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid; 1895 #if BITS_PER_LONG == 32 1896 /* 1897 * Make sure we can represent the posix lock via 1898 * legacy 32bit flock. 1899 */ 1900 if (fl->fl_start > OFFT_OFFSET_MAX) 1901 return -EOVERFLOW; 1902 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX) 1903 return -EOVERFLOW; 1904 #endif 1905 flock->l_start = fl->fl_start; 1906 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : 1907 fl->fl_end - fl->fl_start + 1; 1908 flock->l_whence = 0; 1909 flock->l_type = fl->fl_type; 1910 return 0; 1911 } 1912 1913 #if BITS_PER_LONG == 32 1914 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl) 1915 { 1916 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid; 1917 flock->l_start = fl->fl_start; 1918 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 : 1919 fl->fl_end - fl->fl_start + 1; 1920 flock->l_whence = 0; 1921 flock->l_type = fl->fl_type; 1922 } 1923 #endif 1924 1925 /* Report the first existing lock that would conflict with l. 1926 * This implements the F_GETLK command of fcntl(). 1927 */ 1928 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l) 1929 { 1930 struct file_lock file_lock; 1931 struct flock flock; 1932 int error; 1933 1934 error = -EFAULT; 1935 if (copy_from_user(&flock, l, sizeof(flock))) 1936 goto out; 1937 error = -EINVAL; 1938 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) 1939 goto out; 1940 1941 error = flock_to_posix_lock(filp, &file_lock, &flock); 1942 if (error) 1943 goto out; 1944 1945 if (cmd == F_GETLKP) { 1946 error = -EINVAL; 1947 if (flock.l_pid != 0) 1948 goto out; 1949 1950 cmd = F_GETLK; 1951 file_lock.fl_flags |= FL_FILE_PVT; 1952 file_lock.fl_owner = (fl_owner_t)filp; 1953 } 1954 1955 error = vfs_test_lock(filp, &file_lock); 1956 if (error) 1957 goto out; 1958 1959 flock.l_type = file_lock.fl_type; 1960 if (file_lock.fl_type != F_UNLCK) { 1961 error = posix_lock_to_flock(&flock, &file_lock); 1962 if (error) 1963 goto out; 1964 } 1965 error = -EFAULT; 1966 if (!copy_to_user(l, &flock, sizeof(flock))) 1967 error = 0; 1968 out: 1969 return error; 1970 } 1971 1972 /** 1973 * vfs_lock_file - file byte range lock 1974 * @filp: The file to apply the lock to 1975 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.) 1976 * @fl: The lock to be applied 1977 * @conf: Place to return a copy of the conflicting lock, if found. 1978 * 1979 * A caller that doesn't care about the conflicting lock may pass NULL 1980 * as the final argument. 1981 * 1982 * If the filesystem defines a private ->lock() method, then @conf will 1983 * be left unchanged; so a caller that cares should initialize it to 1984 * some acceptable default. 1985 * 1986 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX 1987 * locks, the ->lock() interface may return asynchronously, before the lock has 1988 * been granted or denied by the underlying filesystem, if (and only if) 1989 * lm_grant is set. Callers expecting ->lock() to return asynchronously 1990 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if) 1991 * the request is for a blocking lock. When ->lock() does return asynchronously, 1992 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock 1993 * request completes. 1994 * If the request is for non-blocking lock the file system should return 1995 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine 1996 * with the result. If the request timed out the callback routine will return a 1997 * nonzero return code and the file system should release the lock. The file 1998 * system is also responsible to keep a corresponding posix lock when it 1999 * grants a lock so the VFS can find out which locks are locally held and do 2000 * the correct lock cleanup when required. 2001 * The underlying filesystem must not drop the kernel lock or call 2002 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED 2003 * return code. 2004 */ 2005 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf) 2006 { 2007 if (filp->f_op->lock) 2008 return filp->f_op->lock(filp, cmd, fl); 2009 else 2010 return posix_lock_file(filp, fl, conf); 2011 } 2012 EXPORT_SYMBOL_GPL(vfs_lock_file); 2013 2014 static int do_lock_file_wait(struct file *filp, unsigned int cmd, 2015 struct file_lock *fl) 2016 { 2017 int error; 2018 2019 error = security_file_lock(filp, fl->fl_type); 2020 if (error) 2021 return error; 2022 2023 for (;;) { 2024 error = vfs_lock_file(filp, cmd, fl, NULL); 2025 if (error != FILE_LOCK_DEFERRED) 2026 break; 2027 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next); 2028 if (!error) 2029 continue; 2030 2031 locks_delete_block(fl); 2032 break; 2033 } 2034 2035 return error; 2036 } 2037 2038 /* Apply the lock described by l to an open file descriptor. 2039 * This implements both the F_SETLK and F_SETLKW commands of fcntl(). 2040 */ 2041 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd, 2042 struct flock __user *l) 2043 { 2044 struct file_lock *file_lock = locks_alloc_lock(); 2045 struct flock flock; 2046 struct inode *inode; 2047 struct file *f; 2048 int error; 2049 2050 if (file_lock == NULL) 2051 return -ENOLCK; 2052 2053 /* 2054 * This might block, so we do it before checking the inode. 2055 */ 2056 error = -EFAULT; 2057 if (copy_from_user(&flock, l, sizeof(flock))) 2058 goto out; 2059 2060 inode = file_inode(filp); 2061 2062 /* Don't allow mandatory locks on files that may be memory mapped 2063 * and shared. 2064 */ 2065 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { 2066 error = -EAGAIN; 2067 goto out; 2068 } 2069 2070 again: 2071 error = flock_to_posix_lock(filp, file_lock, &flock); 2072 if (error) 2073 goto out; 2074 2075 /* 2076 * If the cmd is requesting file-private locks, then set the 2077 * FL_FILE_PVT flag and override the owner. 2078 */ 2079 switch (cmd) { 2080 case F_SETLKP: 2081 error = -EINVAL; 2082 if (flock.l_pid != 0) 2083 goto out; 2084 2085 cmd = F_SETLK; 2086 file_lock->fl_flags |= FL_FILE_PVT; 2087 file_lock->fl_owner = (fl_owner_t)filp; 2088 break; 2089 case F_SETLKPW: 2090 error = -EINVAL; 2091 if (flock.l_pid != 0) 2092 goto out; 2093 2094 cmd = F_SETLKW; 2095 file_lock->fl_flags |= FL_FILE_PVT; 2096 file_lock->fl_owner = (fl_owner_t)filp; 2097 /* Fallthrough */ 2098 case F_SETLKW: 2099 file_lock->fl_flags |= FL_SLEEP; 2100 } 2101 2102 error = do_lock_file_wait(filp, cmd, file_lock); 2103 2104 /* 2105 * Attempt to detect a close/fcntl race and recover by 2106 * releasing the lock that was just acquired. 2107 */ 2108 /* 2109 * we need that spin_lock here - it prevents reordering between 2110 * update of inode->i_flock and check for it done in close(). 2111 * rcu_read_lock() wouldn't do. 2112 */ 2113 spin_lock(¤t->files->file_lock); 2114 f = fcheck(fd); 2115 spin_unlock(¤t->files->file_lock); 2116 if (!error && f != filp && flock.l_type != F_UNLCK) { 2117 flock.l_type = F_UNLCK; 2118 goto again; 2119 } 2120 2121 out: 2122 locks_free_lock(file_lock); 2123 return error; 2124 } 2125 2126 #if BITS_PER_LONG == 32 2127 /* Report the first existing lock that would conflict with l. 2128 * This implements the F_GETLK command of fcntl(). 2129 */ 2130 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l) 2131 { 2132 struct file_lock file_lock; 2133 struct flock64 flock; 2134 int error; 2135 2136 error = -EFAULT; 2137 if (copy_from_user(&flock, l, sizeof(flock))) 2138 goto out; 2139 error = -EINVAL; 2140 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK)) 2141 goto out; 2142 2143 error = flock64_to_posix_lock(filp, &file_lock, &flock); 2144 if (error) 2145 goto out; 2146 2147 if (cmd == F_GETLKP) { 2148 error = -EINVAL; 2149 if (flock.l_pid != 0) 2150 goto out; 2151 2152 cmd = F_GETLK64; 2153 file_lock.fl_flags |= FL_FILE_PVT; 2154 file_lock.fl_owner = (fl_owner_t)filp; 2155 } 2156 2157 error = vfs_test_lock(filp, &file_lock); 2158 if (error) 2159 goto out; 2160 2161 flock.l_type = file_lock.fl_type; 2162 if (file_lock.fl_type != F_UNLCK) 2163 posix_lock_to_flock64(&flock, &file_lock); 2164 2165 error = -EFAULT; 2166 if (!copy_to_user(l, &flock, sizeof(flock))) 2167 error = 0; 2168 2169 out: 2170 return error; 2171 } 2172 2173 /* Apply the lock described by l to an open file descriptor. 2174 * This implements both the F_SETLK and F_SETLKW commands of fcntl(). 2175 */ 2176 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd, 2177 struct flock64 __user *l) 2178 { 2179 struct file_lock *file_lock = locks_alloc_lock(); 2180 struct flock64 flock; 2181 struct inode *inode; 2182 struct file *f; 2183 int error; 2184 2185 if (file_lock == NULL) 2186 return -ENOLCK; 2187 2188 /* 2189 * This might block, so we do it before checking the inode. 2190 */ 2191 error = -EFAULT; 2192 if (copy_from_user(&flock, l, sizeof(flock))) 2193 goto out; 2194 2195 inode = file_inode(filp); 2196 2197 /* Don't allow mandatory locks on files that may be memory mapped 2198 * and shared. 2199 */ 2200 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) { 2201 error = -EAGAIN; 2202 goto out; 2203 } 2204 2205 again: 2206 error = flock64_to_posix_lock(filp, file_lock, &flock); 2207 if (error) 2208 goto out; 2209 2210 /* 2211 * If the cmd is requesting file-private locks, then set the 2212 * FL_FILE_PVT flag and override the owner. 2213 */ 2214 switch (cmd) { 2215 case F_SETLKP: 2216 error = -EINVAL; 2217 if (flock.l_pid != 0) 2218 goto out; 2219 2220 cmd = F_SETLK64; 2221 file_lock->fl_flags |= FL_FILE_PVT; 2222 file_lock->fl_owner = (fl_owner_t)filp; 2223 break; 2224 case F_SETLKPW: 2225 error = -EINVAL; 2226 if (flock.l_pid != 0) 2227 goto out; 2228 2229 cmd = F_SETLKW64; 2230 file_lock->fl_flags |= FL_FILE_PVT; 2231 file_lock->fl_owner = (fl_owner_t)filp; 2232 /* Fallthrough */ 2233 case F_SETLKW64: 2234 file_lock->fl_flags |= FL_SLEEP; 2235 } 2236 2237 error = do_lock_file_wait(filp, cmd, file_lock); 2238 2239 /* 2240 * Attempt to detect a close/fcntl race and recover by 2241 * releasing the lock that was just acquired. 2242 */ 2243 spin_lock(¤t->files->file_lock); 2244 f = fcheck(fd); 2245 spin_unlock(¤t->files->file_lock); 2246 if (!error && f != filp && flock.l_type != F_UNLCK) { 2247 flock.l_type = F_UNLCK; 2248 goto again; 2249 } 2250 2251 out: 2252 locks_free_lock(file_lock); 2253 return error; 2254 } 2255 #endif /* BITS_PER_LONG == 32 */ 2256 2257 /* 2258 * This function is called when the file is being removed 2259 * from the task's fd array. POSIX locks belonging to this task 2260 * are deleted at this time. 2261 */ 2262 void locks_remove_posix(struct file *filp, fl_owner_t owner) 2263 { 2264 struct file_lock lock; 2265 2266 /* 2267 * If there are no locks held on this file, we don't need to call 2268 * posix_lock_file(). Another process could be setting a lock on this 2269 * file at the same time, but we wouldn't remove that lock anyway. 2270 */ 2271 if (!file_inode(filp)->i_flock) 2272 return; 2273 2274 lock.fl_type = F_UNLCK; 2275 lock.fl_flags = FL_POSIX | FL_CLOSE; 2276 lock.fl_start = 0; 2277 lock.fl_end = OFFSET_MAX; 2278 lock.fl_owner = owner; 2279 lock.fl_pid = current->tgid; 2280 lock.fl_file = filp; 2281 lock.fl_ops = NULL; 2282 lock.fl_lmops = NULL; 2283 2284 vfs_lock_file(filp, F_SETLK, &lock, NULL); 2285 2286 if (lock.fl_ops && lock.fl_ops->fl_release_private) 2287 lock.fl_ops->fl_release_private(&lock); 2288 } 2289 2290 EXPORT_SYMBOL(locks_remove_posix); 2291 2292 /* 2293 * This function is called on the last close of an open file. 2294 */ 2295 void locks_remove_file(struct file *filp) 2296 { 2297 struct inode * inode = file_inode(filp); 2298 struct file_lock *fl; 2299 struct file_lock **before; 2300 2301 if (!inode->i_flock) 2302 return; 2303 2304 locks_remove_posix(filp, (fl_owner_t)filp); 2305 2306 if (filp->f_op->flock) { 2307 struct file_lock fl = { 2308 .fl_pid = current->tgid, 2309 .fl_file = filp, 2310 .fl_flags = FL_FLOCK, 2311 .fl_type = F_UNLCK, 2312 .fl_end = OFFSET_MAX, 2313 }; 2314 filp->f_op->flock(filp, F_SETLKW, &fl); 2315 if (fl.fl_ops && fl.fl_ops->fl_release_private) 2316 fl.fl_ops->fl_release_private(&fl); 2317 } 2318 2319 spin_lock(&inode->i_lock); 2320 before = &inode->i_flock; 2321 2322 while ((fl = *before) != NULL) { 2323 if (fl->fl_file == filp) { 2324 if (IS_LEASE(fl)) { 2325 lease_modify(before, F_UNLCK); 2326 continue; 2327 } 2328 2329 /* 2330 * There's a leftover lock on the list of a type that 2331 * we didn't expect to see. Most likely a classic 2332 * POSIX lock that ended up not getting released 2333 * properly, or that raced onto the list somehow. Log 2334 * some info about it and then just remove it from 2335 * the list. 2336 */ 2337 WARN(!IS_FLOCK(fl), 2338 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n", 2339 MAJOR(inode->i_sb->s_dev), 2340 MINOR(inode->i_sb->s_dev), inode->i_ino, 2341 fl->fl_type, fl->fl_flags, 2342 fl->fl_start, fl->fl_end); 2343 2344 locks_delete_lock(before); 2345 continue; 2346 } 2347 before = &fl->fl_next; 2348 } 2349 spin_unlock(&inode->i_lock); 2350 } 2351 2352 /** 2353 * posix_unblock_lock - stop waiting for a file lock 2354 * @waiter: the lock which was waiting 2355 * 2356 * lockd needs to block waiting for locks. 2357 */ 2358 int 2359 posix_unblock_lock(struct file_lock *waiter) 2360 { 2361 int status = 0; 2362 2363 spin_lock(&blocked_lock_lock); 2364 if (waiter->fl_next) 2365 __locks_delete_block(waiter); 2366 else 2367 status = -ENOENT; 2368 spin_unlock(&blocked_lock_lock); 2369 return status; 2370 } 2371 EXPORT_SYMBOL(posix_unblock_lock); 2372 2373 /** 2374 * vfs_cancel_lock - file byte range unblock lock 2375 * @filp: The file to apply the unblock to 2376 * @fl: The lock to be unblocked 2377 * 2378 * Used by lock managers to cancel blocked requests 2379 */ 2380 int vfs_cancel_lock(struct file *filp, struct file_lock *fl) 2381 { 2382 if (filp->f_op->lock) 2383 return filp->f_op->lock(filp, F_CANCELLK, fl); 2384 return 0; 2385 } 2386 2387 EXPORT_SYMBOL_GPL(vfs_cancel_lock); 2388 2389 #ifdef CONFIG_PROC_FS 2390 #include <linux/proc_fs.h> 2391 #include <linux/seq_file.h> 2392 2393 struct locks_iterator { 2394 int li_cpu; 2395 loff_t li_pos; 2396 }; 2397 2398 static void lock_get_status(struct seq_file *f, struct file_lock *fl, 2399 loff_t id, char *pfx) 2400 { 2401 struct inode *inode = NULL; 2402 unsigned int fl_pid; 2403 2404 if (fl->fl_nspid) 2405 fl_pid = pid_vnr(fl->fl_nspid); 2406 else 2407 fl_pid = fl->fl_pid; 2408 2409 if (fl->fl_file != NULL) 2410 inode = file_inode(fl->fl_file); 2411 2412 seq_printf(f, "%lld:%s ", id, pfx); 2413 if (IS_POSIX(fl)) { 2414 if (fl->fl_flags & FL_ACCESS) 2415 seq_printf(f, "ACCESS"); 2416 else if (IS_FILE_PVT(fl)) 2417 seq_printf(f, "FLPVT "); 2418 else 2419 seq_printf(f, "POSIX "); 2420 2421 seq_printf(f, " %s ", 2422 (inode == NULL) ? "*NOINODE*" : 2423 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY "); 2424 } else if (IS_FLOCK(fl)) { 2425 if (fl->fl_type & LOCK_MAND) { 2426 seq_printf(f, "FLOCK MSNFS "); 2427 } else { 2428 seq_printf(f, "FLOCK ADVISORY "); 2429 } 2430 } else if (IS_LEASE(fl)) { 2431 seq_printf(f, "LEASE "); 2432 if (lease_breaking(fl)) 2433 seq_printf(f, "BREAKING "); 2434 else if (fl->fl_file) 2435 seq_printf(f, "ACTIVE "); 2436 else 2437 seq_printf(f, "BREAKER "); 2438 } else { 2439 seq_printf(f, "UNKNOWN UNKNOWN "); 2440 } 2441 if (fl->fl_type & LOCK_MAND) { 2442 seq_printf(f, "%s ", 2443 (fl->fl_type & LOCK_READ) 2444 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ " 2445 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE "); 2446 } else { 2447 seq_printf(f, "%s ", 2448 (lease_breaking(fl)) 2449 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ " 2450 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ "); 2451 } 2452 if (inode) { 2453 #ifdef WE_CAN_BREAK_LSLK_NOW 2454 seq_printf(f, "%d %s:%ld ", fl_pid, 2455 inode->i_sb->s_id, inode->i_ino); 2456 #else 2457 /* userspace relies on this representation of dev_t ;-( */ 2458 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid, 2459 MAJOR(inode->i_sb->s_dev), 2460 MINOR(inode->i_sb->s_dev), inode->i_ino); 2461 #endif 2462 } else { 2463 seq_printf(f, "%d <none>:0 ", fl_pid); 2464 } 2465 if (IS_POSIX(fl)) { 2466 if (fl->fl_end == OFFSET_MAX) 2467 seq_printf(f, "%Ld EOF\n", fl->fl_start); 2468 else 2469 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end); 2470 } else { 2471 seq_printf(f, "0 EOF\n"); 2472 } 2473 } 2474 2475 static int locks_show(struct seq_file *f, void *v) 2476 { 2477 struct locks_iterator *iter = f->private; 2478 struct file_lock *fl, *bfl; 2479 2480 fl = hlist_entry(v, struct file_lock, fl_link); 2481 2482 lock_get_status(f, fl, iter->li_pos, ""); 2483 2484 list_for_each_entry(bfl, &fl->fl_block, fl_block) 2485 lock_get_status(f, bfl, iter->li_pos, " ->"); 2486 2487 return 0; 2488 } 2489 2490 static void *locks_start(struct seq_file *f, loff_t *pos) 2491 __acquires(&blocked_lock_lock) 2492 { 2493 struct locks_iterator *iter = f->private; 2494 2495 iter->li_pos = *pos + 1; 2496 lg_global_lock(&file_lock_lglock); 2497 spin_lock(&blocked_lock_lock); 2498 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos); 2499 } 2500 2501 static void *locks_next(struct seq_file *f, void *v, loff_t *pos) 2502 { 2503 struct locks_iterator *iter = f->private; 2504 2505 ++iter->li_pos; 2506 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos); 2507 } 2508 2509 static void locks_stop(struct seq_file *f, void *v) 2510 __releases(&blocked_lock_lock) 2511 { 2512 spin_unlock(&blocked_lock_lock); 2513 lg_global_unlock(&file_lock_lglock); 2514 } 2515 2516 static const struct seq_operations locks_seq_operations = { 2517 .start = locks_start, 2518 .next = locks_next, 2519 .stop = locks_stop, 2520 .show = locks_show, 2521 }; 2522 2523 static int locks_open(struct inode *inode, struct file *filp) 2524 { 2525 return seq_open_private(filp, &locks_seq_operations, 2526 sizeof(struct locks_iterator)); 2527 } 2528 2529 static const struct file_operations proc_locks_operations = { 2530 .open = locks_open, 2531 .read = seq_read, 2532 .llseek = seq_lseek, 2533 .release = seq_release_private, 2534 }; 2535 2536 static int __init proc_locks_init(void) 2537 { 2538 proc_create("locks", 0, NULL, &proc_locks_operations); 2539 return 0; 2540 } 2541 module_init(proc_locks_init); 2542 #endif 2543 2544 /** 2545 * lock_may_read - checks that the region is free of locks 2546 * @inode: the inode that is being read 2547 * @start: the first byte to read 2548 * @len: the number of bytes to read 2549 * 2550 * Emulates Windows locking requirements. Whole-file 2551 * mandatory locks (share modes) can prohibit a read and 2552 * byte-range POSIX locks can prohibit a read if they overlap. 2553 * 2554 * N.B. this function is only ever called 2555 * from knfsd and ownership of locks is never checked. 2556 */ 2557 int lock_may_read(struct inode *inode, loff_t start, unsigned long len) 2558 { 2559 struct file_lock *fl; 2560 int result = 1; 2561 2562 spin_lock(&inode->i_lock); 2563 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { 2564 if (IS_POSIX(fl)) { 2565 if (fl->fl_type == F_RDLCK) 2566 continue; 2567 if ((fl->fl_end < start) || (fl->fl_start > (start + len))) 2568 continue; 2569 } else if (IS_FLOCK(fl)) { 2570 if (!(fl->fl_type & LOCK_MAND)) 2571 continue; 2572 if (fl->fl_type & LOCK_READ) 2573 continue; 2574 } else 2575 continue; 2576 result = 0; 2577 break; 2578 } 2579 spin_unlock(&inode->i_lock); 2580 return result; 2581 } 2582 2583 EXPORT_SYMBOL(lock_may_read); 2584 2585 /** 2586 * lock_may_write - checks that the region is free of locks 2587 * @inode: the inode that is being written 2588 * @start: the first byte to write 2589 * @len: the number of bytes to write 2590 * 2591 * Emulates Windows locking requirements. Whole-file 2592 * mandatory locks (share modes) can prohibit a write and 2593 * byte-range POSIX locks can prohibit a write if they overlap. 2594 * 2595 * N.B. this function is only ever called 2596 * from knfsd and ownership of locks is never checked. 2597 */ 2598 int lock_may_write(struct inode *inode, loff_t start, unsigned long len) 2599 { 2600 struct file_lock *fl; 2601 int result = 1; 2602 2603 spin_lock(&inode->i_lock); 2604 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) { 2605 if (IS_POSIX(fl)) { 2606 if ((fl->fl_end < start) || (fl->fl_start > (start + len))) 2607 continue; 2608 } else if (IS_FLOCK(fl)) { 2609 if (!(fl->fl_type & LOCK_MAND)) 2610 continue; 2611 if (fl->fl_type & LOCK_WRITE) 2612 continue; 2613 } else 2614 continue; 2615 result = 0; 2616 break; 2617 } 2618 spin_unlock(&inode->i_lock); 2619 return result; 2620 } 2621 2622 EXPORT_SYMBOL(lock_may_write); 2623 2624 static int __init filelock_init(void) 2625 { 2626 int i; 2627 2628 filelock_cache = kmem_cache_create("file_lock_cache", 2629 sizeof(struct file_lock), 0, SLAB_PANIC, NULL); 2630 2631 lg_lock_init(&file_lock_lglock, "file_lock_lglock"); 2632 2633 for_each_possible_cpu(i) 2634 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i)); 2635 2636 return 0; 2637 } 2638 2639 core_initcall(filelock_init); 2640