Documentation/filesystems/locking.rst

   1 =======
   2 Locking
   3 =======
   4
   5 The text below describes the locking rules for VFS-related methods.
   6 It is (believed to be) up-to-date. *Please*, if you change anything in
   7 prototypes or locking protocols - update this file. And update the relevant
   8 instances in the tree, don't leave that to maintainers of filesystems/devices/
   9 etc. At the very least, put the list of dubious cases in the end of this file.
  10 Don't turn it into log - maintainers of out-of-the-tree code are supposed to
  11 be able to use diff(1).
  12
  13 Thing currently missing here: socket operations. Alexey?
  14
  15 dentry_operations
  16 =================
  17
  18 prototypes::
  19
  20         int (*d_revalidate)(struct dentry *, unsigned int);
  21         int (*d_weak_revalidate)(struct dentry *, unsigned int);
  22         int (*d_hash)(const struct dentry *, struct qstr *);
  23         int (*d_compare)(const struct dentry *,
  24                         unsigned int, const char *, const struct qstr *);
  25         int (*d_delete)(struct dentry *);
  26         int (*d_init)(struct dentry *);
  27         void (*d_release)(struct dentry *);
  28         void (*d_iput)(struct dentry *, struct inode *);
  29         char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
  30         struct vfsmount *(*d_automount)(struct path *path);
  31         int (*d_manage)(const struct path *, bool);
  32         struct dentry *(*d_real)(struct dentry *, const struct inode *);
  33
  34 locking rules:
  35
  36 ================== ===========  ========        ==============  ========
  37 ops                rename_lock  ->d_lock        may block       rcu-walk
  38 ================== ===========  ========        ==============  ========
  39 d_revalidate:      no           no              yes (ref-walk)  maybe
  40 d_weak_revalidate: no           no              yes             no
  41 d_hash             no           no              no              maybe
  42 d_compare:         yes          no              no              maybe
  43 d_delete:          no           yes             no              no
  44 d_init:            no           no              yes             no
  45 d_release:         no           no              yes             no
  46 d_prune:           no           yes             no              no
  47 d_iput:            no           no              yes             no
  48 d_dname:           no           no              no              no
  49 d_automount:       no           no              yes             no
  50 d_manage:          no           no              yes (ref-walk)  maybe
  51 d_real             no           no              yes             no
  52 ================== ===========  ========        ==============  ========
  53
  54 inode_operations
  55 ================
  56
  57 prototypes::
  58
  59         int (*create) (struct inode *,struct dentry *,umode_t, bool);
  60         struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
  61         int (*link) (struct dentry *,struct inode *,struct dentry *);
  62         int (*unlink) (struct inode *,struct dentry *);
  63         int (*symlink) (struct inode *,struct dentry *,const char *);
  64         int (*mkdir) (struct inode *,struct dentry *,umode_t);
  65         int (*rmdir) (struct inode *,struct dentry *);
  66         int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
  67         int (*rename) (struct inode *, struct dentry *,
  68                         struct inode *, struct dentry *, unsigned int);
  69         int (*readlink) (struct dentry *, char __user *,int);
  70         const char *(*get_link) (struct dentry *, struct inode *, struct delayed_call *);
  71         void (*truncate) (struct inode *);
  72         int (*permission) (struct inode *, int, unsigned int);
  73         int (*get_acl)(struct inode *, int);
  74         int (*setattr) (struct dentry *, struct iattr *);
  75         int (*getattr) (const struct path *, struct kstat *, u32, unsigned int);
  76         ssize_t (*listxattr) (struct dentry *, char *, size_t);
  77         int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
  78         void (*update_time)(struct inode *, struct timespec *, int);
  79         int (*atomic_open)(struct inode *, struct dentry *,
  80                                 struct file *, unsigned open_flag,
  81                                 umode_t create_mode);
  82         int (*tmpfile) (struct inode *, struct dentry *, umode_t);
  83
  84 locking rules:
  85         all may block
  86
  87 ============    =============================================
  88 ops             i_rwsem(inode)
  89 ============    =============================================
  90 lookup:         shared
  91 create:         exclusive
  92 link:           exclusive (both)
  93 mknod:          exclusive
  94 symlink:        exclusive
  95 mkdir:          exclusive
  96 unlink:         exclusive (both)
  97 rmdir:          exclusive (both)(see below)
  98 rename:         exclusive (all) (see below)
  99 readlink:       no
 100 get_link:       no
 101 setattr:        exclusive
 102 permission:     no (may not block if called in rcu-walk mode)
 103 get_acl:        no
 104 getattr:        no
 105 listxattr:      no
 106 fiemap:         no
 107 update_time:    no
 108 atomic_open:    shared (exclusive if O_CREAT is set in open flags)
 109 tmpfile:        no
 110 ============    =============================================
 111
 112
 113         Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
 114         exclusive on victim.
 115         cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
 116
 117 See Documentation/filesystems/directory-locking.rst for more detailed discussion
 118 of the locking scheme for directory operations.
 119
 120 xattr_handler operations
 121 ========================
 122
 123 prototypes::
 124
 125         bool (*list)(struct dentry *dentry);
 126         int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
 127                    struct inode *inode, const char *name, void *buffer,
 128                    size_t size);
 129         int (*set)(const struct xattr_handler *handler, struct dentry *dentry,
 130                    struct inode *inode, const char *name, const void *buffer,
 131                    size_t size, int flags);
 132
 133 locking rules:
 134         all may block
 135
 136 =====           ==============
 137 ops             i_rwsem(inode)
 138 =====           ==============
 139 list:           no
 140 get:            no
 141 set:            exclusive
 142 =====           ==============
 143
 144 super_operations
 145 ================
 146
 147 prototypes::
 148
 149         struct inode *(*alloc_inode)(struct super_block *sb);
 150         void (*free_inode)(struct inode *);
 151         void (*destroy_inode)(struct inode *);
 152         void (*dirty_inode) (struct inode *, int flags);
 153         int (*write_inode) (struct inode *, struct writeback_control *wbc);
 154         int (*drop_inode) (struct inode *);
 155         void (*evict_inode) (struct inode *);
 156         void (*put_super) (struct super_block *);
 157         int (*sync_fs)(struct super_block *sb, int wait);
 158         int (*freeze_fs) (struct super_block *);
 159         int (*unfreeze_fs) (struct super_block *);
 160         int (*statfs) (struct dentry *, struct kstatfs *);
 161         int (*remount_fs) (struct super_block *, int *, char *);
 162         void (*umount_begin) (struct super_block *);
 163         int (*show_options)(struct seq_file *, struct dentry *);
 164         ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
 165         ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
 166         int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
 167
 168 locking rules:
 169         All may block [not true, see below]
 170
 171 ======================  ============    ========================
 172 ops                     s_umount        note
 173 ======================  ============    ========================
 174 alloc_inode:
 175 free_inode:                             called from RCU callback
 176 destroy_inode:
 177 dirty_inode:
 178 write_inode:
 179 drop_inode:                             !!!inode->i_lock!!!
 180 evict_inode:
 181 put_super:              write
 182 sync_fs:                read
 183 freeze_fs:              write
 184 unfreeze_fs:            write
 185 statfs:                 maybe(read)     (see below)
 186 remount_fs:             write
 187 umount_begin:           no
 188 show_options:           no              (namespace_sem)
 189 quota_read:             no              (see below)
 190 quota_write:            no              (see below)
 191 bdev_try_to_free_page:  no              (see below)
 192 ======================  ============    ========================
 193
 194 ->statfs() has s_umount (shared) when called by ustat(2) (native or
 195 compat), but that's an accident of bad API; s_umount is used to pin
 196 the superblock down when we only have dev_t given us by userland to
 197 identify the superblock.  Everything else (statfs(), fstatfs(), etc.)
 198 doesn't hold it when calling ->statfs() - superblock is pinned down
 199 by resolving the pathname passed to syscall.
 200
 201 ->quota_read() and ->quota_write() functions are both guaranteed to
 202 be the only ones operating on the quota file by the quota code (via
 203 dqio_sem) (unless an admin really wants to screw up something and
 204 writes to quota files with quotas on). For other details about locking
 205 see also dquot_operations section.
 206
 207 ->bdev_try_to_free_page is called from the ->releasepage handler of
 208 the block device inode.  See there for more details.
 209
 210 file_system_type
 211 ================
 212
 213 prototypes::
 214
 215         struct dentry *(*mount) (struct file_system_type *, int,
 216                        const char *, void *);
 217         void (*kill_sb) (struct super_block *);
 218
 219 locking rules:
 220
 221 =======         =========
 222 ops             may block
 223 =======         =========
 224 mount           yes
 225 kill_sb         yes
 226 =======         =========
 227
 228 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
 229 on return.
 230
 231 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
 232 unlocks and drops the reference.
 233
 234 address_space_operations
 235 ========================
 236 prototypes::
 237
 238         int (*writepage)(struct page *page, struct writeback_control *wbc);
 239         int (*readpage)(struct file *, struct page *);
 240         int (*writepages)(struct address_space *, struct writeback_control *);
 241         int (*set_page_dirty)(struct page *page);
 242         void (*readahead)(struct readahead_control *);
 243         int (*readpages)(struct file *filp, struct address_space *mapping,
 244                         struct list_head *pages, unsigned nr_pages);
 245         int (*write_begin)(struct file *, struct address_space *mapping,
 246                                 loff_t pos, unsigned len, unsigned flags,
 247                                 struct page **pagep, void **fsdata);
 248         int (*write_end)(struct file *, struct address_space *mapping,
 249                                 loff_t pos, unsigned len, unsigned copied,
 250                                 struct page *page, void *fsdata);
 251         sector_t (*bmap)(struct address_space *, sector_t);
 252         void (*invalidatepage) (struct page *, unsigned int, unsigned int);
 253         int (*releasepage) (struct page *, int);
 254         void (*freepage)(struct page *);
 255         int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
 256         bool (*isolate_page) (struct page *, isolate_mode_t);
 257         int (*migratepage)(struct address_space *, struct page *, struct page *);
 258         void (*putback_page) (struct page *);
 259         int (*launder_page)(struct page *);
 260         int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long);
 261         int (*error_remove_page)(struct address_space *, struct page *);
 262         int (*swap_activate)(struct file *);
 263         int (*swap_deactivate)(struct file *);
 264
 265 locking rules:
 266         All except set_page_dirty and freepage may block
 267
 268 ======================  ======================== =========
 269 ops                     PageLocked(page)         i_rwsem
 270 ======================  ======================== =========
 271 writepage:              yes, unlocks (see below)
 272 readpage:               yes, unlocks
 273 writepages:
 274 set_page_dirty          no
 275 readahead:              yes, unlocks
 276 readpages:              no
 277 write_begin:            locks the page           exclusive
 278 write_end:              yes, unlocks             exclusive
 279 bmap:
 280 invalidatepage:         yes
 281 releasepage:            yes
 282 freepage:               yes
 283 direct_IO:
 284 isolate_page:           yes
 285 migratepage:            yes (both)
 286 putback_page:           yes
 287 launder_page:           yes
 288 is_partially_uptodate:  yes
 289 error_remove_page:      yes
 290 swap_activate:          no
 291 swap_deactivate:        no
 292 ======================  ======================== =========
 293
 294 ->write_begin(), ->write_end() and ->readpage() may be called from
 295 the request handler (/dev/loop).
 296
 297 ->readpage() unlocks the page, either synchronously or via I/O
 298 completion.
 299
 300 ->readahead() unlocks the pages that I/O is attempted on like ->readpage().
 301
 302 ->readpages() populates the pagecache with the passed pages and starts
 303 I/O against them.  They come unlocked upon I/O completion.
 304
 305 ->writepage() is used for two purposes: for "memory cleansing" and for
 306 "sync".  These are quite different operations and the behaviour may differ
 307 depending upon the mode.
 308
 309 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
 310 it *must* start I/O against the page, even if that would involve
 311 blocking on in-progress I/O.
 312
 313 If writepage is called for memory cleansing (sync_mode ==
 314 WBC_SYNC_NONE) then its role is to get as much writeout underway as
 315 possible.  So writepage should try to avoid blocking against
 316 currently-in-progress I/O.
 317
 318 If the filesystem is not called for "sync" and it determines that it
 319 would need to block against in-progress I/O to be able to start new I/O
 320 against the page the filesystem should redirty the page with
 321 redirty_page_for_writepage(), then unlock the page and return zero.
 322 This may also be done to avoid internal deadlocks, but rarely.
 323
 324 If the filesystem is called for sync then it must wait on any
 325 in-progress I/O and then start new I/O.
 326
 327 The filesystem should unlock the page synchronously, before returning to the
 328 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
 329 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
 330 currently, and VM should stop calling ->writepage() on this page for some
 331 time. VM does this by moving page to the head of the active list, hence the
 332 name.
 333
 334 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
 335 and return zero, writepage *must* run set_page_writeback() against the page,
 336 followed by unlocking it.  Once set_page_writeback() has been run against the
 337 page, write I/O can be submitted and the write I/O completion handler must run
 338 end_page_writeback() once the I/O is complete.  If no I/O is submitted, the
 339 filesystem must run end_page_writeback() against the page before returning from
 340 writepage.
 341
 342 That is: after 2.5.12, pages which are under writeout are *not* locked.  Note,
 343 if the filesystem needs the page to be locked during writeout, that is ok, too,
 344 the page is allowed to be unlocked at any point in time between the calls to
 345 set_page_writeback() and end_page_writeback().
 346
 347 Note, failure to run either redirty_page_for_writepage() or the combination of
 348 set_page_writeback()/end_page_writeback() on a page submitted to writepage
 349 will leave the page itself marked clean but it will be tagged as dirty in the
 350 radix tree.  This incoherency can lead to all sorts of hard-to-debug problems
 351 in the filesystem like having dirty inodes at umount and losing written data.
 352
 353 ->writepages() is used for periodic writeback and for syscall-initiated
 354 sync operations.  The address_space should start I/O against at least
 355 ``*nr_to_write`` pages.  ``*nr_to_write`` must be decremented for each page
 356 which is written.  The address_space implementation may write more (or less)
 357 pages than ``*nr_to_write`` asks for, but it should try to be reasonably close.
 358 If nr_to_write is NULL, all dirty pages must be written.
 359
 360 writepages should _only_ write pages which are present on
 361 mapping->io_pages.
 362
 363 ->set_page_dirty() is called from various places in the kernel
 364 when the target page is marked as needing writeback.  It may be called
 365 under spinlock (it cannot block) and is sometimes called with the page
 366 not locked.
 367
 368 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
 369 filesystems and by the swapper. The latter will eventually go away.  Please,
 370 keep it that way and don't breed new callers.
 371
 372 ->invalidatepage() is called when the filesystem must attempt to drop
 373 some or all of the buffers from the page when it is being truncated. It
 374 returns zero on success. If ->invalidatepage is zero, the kernel uses
 375 block_invalidatepage() instead.
 376
 377 ->releasepage() is called when the kernel is about to try to drop the
 378 buffers from the page in preparation for freeing it.  It returns zero to
 379 indicate that the buffers are (or may be) freeable.  If ->releasepage is zero,
 380 the kernel assumes that the fs has no private interest in the buffers.
 381
 382 ->freepage() is called when the kernel is done dropping the page
 383 from the page cache.
 384
 385 ->launder_page() may be called prior to releasing a page if
 386 it is still found to be dirty. It returns zero if the page was successfully
 387 cleaned, or an error value if not. Note that in order to prevent the page
 388 getting mapped back in and redirtied, it needs to be kept locked
 389 across the entire operation.
 390
 391 ->swap_activate will be called with a non-zero argument on
 392 files backing (non block device backed) swapfiles. A return value
 393 of zero indicates success, in which case this file can be used for
 394 backing swapspace. The swapspace operations will be proxied to the
 395 address space operations.
 396
 397 ->swap_deactivate() will be called in the sys_swapoff()
 398 path after ->swap_activate() returned success.
 399
 400 file_lock_operations
 401 ====================
 402
 403 prototypes::
 404
 405         void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
 406         void (*fl_release_private)(struct file_lock *);
 407
 408
 409 locking rules:
 410
 411 ===================     =============   =========
 412 ops                     inode->i_lock   may block
 413 ===================     =============   =========
 414 fl_copy_lock:           yes             no
 415 fl_release_private:     maybe           maybe[1]_
 416 ===================     =============   =========
 417
 418 .. [1]:
 419    ->fl_release_private for flock or POSIX locks is currently allowed
 420    to block. Leases however can still be freed while the i_lock is held and
 421    so fl_release_private called on a lease should not block.
 422
 423 lock_manager_operations
 424 =======================
 425
 426 prototypes::
 427
 428         void (*lm_notify)(struct file_lock *);  /* unblock callback */
 429         int (*lm_grant)(struct file_lock *, struct file_lock *, int);
 430         void (*lm_break)(struct file_lock *); /* break_lease callback */
 431         int (*lm_change)(struct file_lock **, int);
 432         bool (*lm_breaker_owns_lease)(struct file_lock *);
 433
 434 locking rules:
 435
 436 ======================  =============   =================       =========
 437 ops                     inode->i_lock   blocked_lock_lock       may block
 438 ======================  =============   =================       =========
 439 lm_notify:              yes             yes                     no
 440 lm_grant:               no              no                      no
 441 lm_break:               yes             no                      no
 442 lm_change               yes             no                      no
 443 lm_breaker_owns_lease:  no              no                      no
 444 ======================  =============   =================       =========
 445
 446 buffer_head
 447 ===========
 448
 449 prototypes::
 450
 451         void (*b_end_io)(struct buffer_head *bh, int uptodate);
 452
 453 locking rules:
 454
 455 called from interrupts. In other words, extreme care is needed here.
 456 bh is locked, but that's all warranties we have here. Currently only RAID1,
 457 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
 458 call this method upon the IO completion.
 459
 460 block_device_operations
 461 =======================
 462 prototypes::
 463
 464         int (*open) (struct block_device *, fmode_t);
 465         int (*release) (struct gendisk *, fmode_t);
 466         int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
 467         int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
 468         int (*direct_access) (struct block_device *, sector_t, void **,
 469                                 unsigned long *);
 470         void (*unlock_native_capacity) (struct gendisk *);
 471         int (*revalidate_disk) (struct gendisk *);
 472         int (*getgeo)(struct block_device *, struct hd_geometry *);
 473         void (*swap_slot_free_notify) (struct block_device *, unsigned long);
 474
 475 locking rules:
 476
 477 ======================= ===================
 478 ops                     bd_mutex
 479 ======================= ===================
 480 open:                   yes
 481 release:                yes
 482 ioctl:                  no
 483 compat_ioctl:           no
 484 direct_access:          no
 485 unlock_native_capacity: no
 486 revalidate_disk:        no
 487 getgeo:                 no
 488 swap_slot_free_notify:  no      (see below)
 489 ======================= ===================
 490
 491 swap_slot_free_notify is called with swap_lock and sometimes the page lock
 492 held.
 493
 494
 495 file_operations
 496 ===============
 497
 498 prototypes::
 499
 500         loff_t (*llseek) (struct file *, loff_t, int);
 501         ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
 502         ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
 503         ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
 504         ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
 505         int (*iterate) (struct file *, struct dir_context *);
 506         int (*iterate_shared) (struct file *, struct dir_context *);
 507         __poll_t (*poll) (struct file *, struct poll_table_struct *);
 508         long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
 509         long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
 510         int (*mmap) (struct file *, struct vm_area_struct *);
 511         int (*open) (struct inode *, struct file *);
 512         int (*flush) (struct file *);
 513         int (*release) (struct inode *, struct file *);
 514         int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
 515         int (*fasync) (int, struct file *, int);
 516         int (*lock) (struct file *, int, struct file_lock *);
 517         ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
 518                         loff_t *);
 519         ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
 520                         loff_t *);
 521         ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
 522                         void __user *);
 523         ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
 524                         loff_t *, int);
 525         unsigned long (*get_unmapped_area)(struct file *, unsigned long,
 526                         unsigned long, unsigned long, unsigned long);
 527         int (*check_flags)(int);
 528         int (*flock) (struct file *, int, struct file_lock *);
 529         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
 530                         size_t, unsigned int);
 531         ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
 532                         size_t, unsigned int);
 533         int (*setlease)(struct file *, long, struct file_lock **, void **);
 534         long (*fallocate)(struct file *, int, loff_t, loff_t);
 535
 536 locking rules:
 537         All may block.
 538
 539 ->llseek() locking has moved from llseek to the individual llseek
 540 implementations.  If your fs is not using generic_file_llseek, you
 541 need to acquire and release the appropriate locks in your ->llseek().
 542 For many filesystems, it is probably safe to acquire the inode
 543 mutex or just to use i_size_read() instead.
 544 Note: this does not protect the file->f_pos against concurrent modifications
 545 since this is something the userspace has to take care about.
 546
 547 ->iterate() is called with i_rwsem exclusive.
 548
 549 ->iterate_shared() is called with i_rwsem at least shared.
 550
 551 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
 552 Most instances call fasync_helper(), which does that maintenance, so it's
 553 not normally something one needs to worry about.  Return values > 0 will be
 554 mapped to zero in the VFS layer.
 555
 556 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
 557 move ->readdir() to inode_operations and use a separate method for directory
 558 ->ioctl() or kill the latter completely. One of the problems is that for
 559 anything that resembles union-mount we won't have a struct file for all
 560 components. And there are other reasons why the current interface is a mess...
 561
 562 ->read on directories probably must go away - we should just enforce -EISDIR
 563 in sys_read() and friends.
 564
 565 ->setlease operations should call generic_setlease() before or after setting
 566 the lease within the individual filesystem to record the result of the
 567 operation
 568
 569 dquot_operations
 570 ================
 571
 572 prototypes::
 573
 574         int (*write_dquot) (struct dquot *);
 575         int (*acquire_dquot) (struct dquot *);
 576         int (*release_dquot) (struct dquot *);
 577         int (*mark_dirty) (struct dquot *);
 578         int (*write_info) (struct super_block *, int);
 579
 580 These operations are intended to be more or less wrapping functions that ensure
 581 a proper locking wrt the filesystem and call the generic quota operations.
 582
 583 What filesystem should expect from the generic quota functions:
 584
 585 ==============  ============    =========================
 586 ops             FS recursion    Held locks when called
 587 ==============  ============    =========================
 588 write_dquot:    yes             dqonoff_sem or dqptr_sem
 589 acquire_dquot:  yes             dqonoff_sem or dqptr_sem
 590 release_dquot:  yes             dqonoff_sem or dqptr_sem
 591 mark_dirty:     no              -
 592 write_info:     yes             dqonoff_sem
 593 ==============  ============    =========================
 594
 595 FS recursion means calling ->quota_read() and ->quota_write() from superblock
 596 operations.
 597
 598 More details about quota locking can be found in fs/dquot.c.
 599
 600 vm_operations_struct
 601 ====================
 602
 603 prototypes::
 604
 605         void (*open)(struct vm_area_struct*);
 606         void (*close)(struct vm_area_struct*);
 607         vm_fault_t (*fault)(struct vm_area_struct*, struct vm_fault *);
 608         vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
 609         vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
 610         int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
 611
 612 locking rules:
 613
 614 =============   =========       ===========================
 615 ops             mmap_lock       PageLocked(page)
 616 =============   =========       ===========================
 617 open:           yes
 618 close:          yes
 619 fault:          yes             can return with page locked
 620 map_pages:      yes
 621 page_mkwrite:   yes             can return with page locked
 622 pfn_mkwrite:    yes
 623 access:         yes
 624 =============   =========       ===========================
 625
 626 ->fault() is called when a previously not present pte is about
 627 to be faulted in. The filesystem must find and return the page associated
 628 with the passed in "pgoff" in the vm_fault structure. If it is possible that
 629 the page may be truncated and/or invalidated, then the filesystem must lock
 630 the page, then ensure it is not already truncated (the page lock will block
 631 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
 632 locked. The VM will unlock the page.
 633
 634 ->map_pages() is called when VM asks to map easy accessible pages.
 635 Filesystem should find and map pages associated with offsets from "start_pgoff"
 636 till "end_pgoff". ->map_pages() is called with page table locked and must
 637 not block.  If it's not possible to reach a page without blocking,
 638 filesystem should skip it. Filesystem should use do_set_pte() to setup
 639 page table entry. Pointer to entry associated with the page is passed in
 640 "pte" field in vm_fault structure. Pointers to entries for other offsets
 641 should be calculated relative to "pte".
 642
 643 ->page_mkwrite() is called when a previously read-only pte is
 644 about to become writeable. The filesystem again must ensure that there are
 645 no truncate/invalidate races, and then return with the page locked. If
 646 the page has been truncated, the filesystem should not look up a new page
 647 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
 648 will cause the VM to retry the fault.
 649
 650 ->pfn_mkwrite() is the same as page_mkwrite but when the pte is
 651 VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
 652 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
 653 after this call is to make the pte read-write, unless pfn_mkwrite returns
 654 an error.
 655
 656 ->access() is called when get_user_pages() fails in
 657 access_process_vm(), typically used to debug a process through
 658 /proc/pid/mem or ptrace.  This function is needed only for
 659 VM_IO | VM_PFNMAP VMAs.
 660
 661 --------------------------------------------------------------------------------
 662
 663                         Dubious stuff
 664
 665 (if you break something or notice that it is broken and do not fix it yourself
 666 - at least put it here)