1======= 2Locking 3======= 4 5The text below describes the locking rules for VFS-related methods. 6It is (believed to be) up-to-date. *Please*, if you change anything in 7prototypes or locking protocols - update this file. And update the relevant 8instances in the tree, don't leave that to maintainers of filesystems/devices/ 9etc. At the very least, put the list of dubious cases in the end of this file. 10Don't turn it into log - maintainers of out-of-the-tree code are supposed to 11be able to use diff(1). 12 13Thing currently missing here: socket operations. Alexey? 14 15dentry_operations 16================= 17 18prototypes:: 19 20 int (*d_revalidate)(struct inode *, const struct qstr *, 21 struct dentry *, unsigned int); 22 int (*d_weak_revalidate)(struct dentry *, unsigned int); 23 int (*d_hash)(const struct dentry *, struct qstr *); 24 int (*d_compare)(const struct dentry *, 25 unsigned int, const char *, const struct qstr *); 26 int (*d_delete)(struct dentry *); 27 int (*d_init)(struct dentry *); 28 void (*d_release)(struct dentry *); 29 void (*d_iput)(struct dentry *, struct inode *); 30 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen); 31 struct vfsmount *(*d_automount)(struct path *path); 32 int (*d_manage)(const struct path *, bool); 33 struct dentry *(*d_real)(struct dentry *, enum d_real_type type); 34 bool (*d_unalias_trylock)(const struct dentry *); 35 void (*d_unalias_unlock)(const struct dentry *); 36 37locking rules: 38 39================== =========== ======== ============== ======== 40ops rename_lock ->d_lock may block rcu-walk 41================== =========== ======== ============== ======== 42d_revalidate: no no yes (ref-walk) maybe 43d_weak_revalidate: no no yes no 44d_hash no no no maybe 45d_compare: yes no no maybe 46d_delete: no yes no no 47d_init: no no yes no 48d_release: no no yes no 49d_prune: no yes no no 50d_iput: no no yes no 51d_dname: no no no no 52d_automount: no no yes no 53d_manage: no no yes (ref-walk) maybe 54d_real no no yes no 55d_unalias_trylock yes no no no 56d_unalias_unlock yes no no no 57================== =========== ======== ============== ======== 58 59inode_operations 60================ 61 62prototypes:: 63 64 int (*create) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t, bool); 65 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int); 66 int (*link) (struct dentry *,struct inode *,struct dentry *); 67 int (*unlink) (struct inode *,struct dentry *); 68 int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *); 69 int (*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t); 70 int (*rmdir) (struct inode *,struct dentry *); 71 int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t); 72 int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *, 73 struct inode *, struct dentry *, unsigned int); 74 int (*readlink) (struct dentry *, char __user *,int); 75 const char *(*get_link) (struct dentry *, struct inode *, struct delayed_call *); 76 void (*truncate) (struct inode *); 77 int (*permission) (struct mnt_idmap *, struct inode *, int, unsigned int); 78 struct posix_acl * (*get_inode_acl)(struct inode *, int, bool); 79 int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *); 80 int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int); 81 ssize_t (*listxattr) (struct dentry *, char *, size_t); 82 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len); 83 void (*update_time)(struct inode *, struct timespec *, int); 84 int (*atomic_open)(struct inode *, struct dentry *, 85 struct file *, unsigned open_flag, 86 umode_t create_mode); 87 int (*tmpfile) (struct mnt_idmap *, struct inode *, 88 struct file *, umode_t); 89 int (*fileattr_set)(struct mnt_idmap *idmap, 90 struct dentry *dentry, struct fileattr *fa); 91 int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa); 92 struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int); 93 struct offset_ctx *(*get_offset_ctx)(struct inode *inode); 94 95locking rules: 96 all may block 97 98============== ================================================== 99ops i_rwsem(inode) 100============== ================================================== 101lookup: shared 102create: exclusive 103link: exclusive (both) 104mknod: exclusive 105symlink: exclusive 106mkdir: exclusive 107unlink: exclusive (both) 108rmdir: exclusive (both)(see below) 109rename: exclusive (both parents, some children) (see below) 110readlink: no 111get_link: no 112setattr: exclusive 113permission: no (may not block if called in rcu-walk mode) 114get_inode_acl: no 115get_acl: no 116getattr: no 117listxattr: no 118fiemap: no 119update_time: no 120atomic_open: shared (exclusive if O_CREAT is set in open flags) 121tmpfile: no 122fileattr_get: no or exclusive 123fileattr_set: exclusive 124get_offset_ctx no 125============== ================================================== 126 127 128 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem 129 exclusive on victim. 130 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem. 131 ->unlink() and ->rename() have ->i_rwsem exclusive on all non-directories 132 involved. 133 ->rename() has ->i_rwsem exclusive on any subdirectory that changes parent. 134 135See Documentation/filesystems/directory-locking.rst for more detailed discussion 136of the locking scheme for directory operations. 137 138xattr_handler operations 139======================== 140 141prototypes:: 142 143 bool (*list)(struct dentry *dentry); 144 int (*get)(const struct xattr_handler *handler, struct dentry *dentry, 145 struct inode *inode, const char *name, void *buffer, 146 size_t size); 147 int (*set)(const struct xattr_handler *handler, 148 struct mnt_idmap *idmap, 149 struct dentry *dentry, struct inode *inode, const char *name, 150 const void *buffer, size_t size, int flags); 151 152locking rules: 153 all may block 154 155===== ============== 156ops i_rwsem(inode) 157===== ============== 158list: no 159get: no 160set: exclusive 161===== ============== 162 163super_operations 164================ 165 166prototypes:: 167 168 struct inode *(*alloc_inode)(struct super_block *sb); 169 void (*free_inode)(struct inode *); 170 void (*destroy_inode)(struct inode *); 171 void (*dirty_inode) (struct inode *, int flags); 172 int (*write_inode) (struct inode *, struct writeback_control *wbc); 173 int (*drop_inode) (struct inode *); 174 void (*evict_inode) (struct inode *); 175 void (*put_super) (struct super_block *); 176 int (*sync_fs)(struct super_block *sb, int wait); 177 int (*freeze_fs) (struct super_block *); 178 int (*unfreeze_fs) (struct super_block *); 179 int (*statfs) (struct dentry *, struct kstatfs *); 180 int (*remount_fs) (struct super_block *, int *, char *); 181 void (*umount_begin) (struct super_block *); 182 int (*show_options)(struct seq_file *, struct dentry *); 183 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t); 184 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t); 185 186locking rules: 187 All may block [not true, see below] 188 189====================== ============ ======================== 190ops s_umount note 191====================== ============ ======================== 192alloc_inode: 193free_inode: called from RCU callback 194destroy_inode: 195dirty_inode: 196write_inode: 197drop_inode: !!!inode->i_lock!!! 198evict_inode: 199put_super: write 200sync_fs: read 201freeze_fs: write 202unfreeze_fs: write 203statfs: maybe(read) (see below) 204remount_fs: write 205umount_begin: no 206show_options: no (namespace_sem) 207quota_read: no (see below) 208quota_write: no (see below) 209====================== ============ ======================== 210 211->statfs() has s_umount (shared) when called by ustat(2) (native or 212compat), but that's an accident of bad API; s_umount is used to pin 213the superblock down when we only have dev_t given us by userland to 214identify the superblock. Everything else (statfs(), fstatfs(), etc.) 215doesn't hold it when calling ->statfs() - superblock is pinned down 216by resolving the pathname passed to syscall. 217 218->quota_read() and ->quota_write() functions are both guaranteed to 219be the only ones operating on the quota file by the quota code (via 220dqio_sem) (unless an admin really wants to screw up something and 221writes to quota files with quotas on). For other details about locking 222see also dquot_operations section. 223 224file_system_type 225================ 226 227prototypes:: 228 229 struct dentry *(*mount) (struct file_system_type *, int, 230 const char *, void *); 231 void (*kill_sb) (struct super_block *); 232 233locking rules: 234 235======= ========= 236ops may block 237======= ========= 238mount yes 239kill_sb yes 240======= ========= 241 242->mount() returns ERR_PTR or the root dentry; its superblock should be locked 243on return. 244 245->kill_sb() takes a write-locked superblock, does all shutdown work on it, 246unlocks and drops the reference. 247 248address_space_operations 249======================== 250prototypes:: 251 252 int (*writepage)(struct page *page, struct writeback_control *wbc); 253 int (*read_folio)(struct file *, struct folio *); 254 int (*writepages)(struct address_space *, struct writeback_control *); 255 bool (*dirty_folio)(struct address_space *, struct folio *folio); 256 void (*readahead)(struct readahead_control *); 257 int (*write_begin)(struct file *, struct address_space *mapping, 258 loff_t pos, unsigned len, 259 struct folio **foliop, void **fsdata); 260 int (*write_end)(struct file *, struct address_space *mapping, 261 loff_t pos, unsigned len, unsigned copied, 262 struct folio *folio, void *fsdata); 263 sector_t (*bmap)(struct address_space *, sector_t); 264 void (*invalidate_folio) (struct folio *, size_t start, size_t len); 265 bool (*release_folio)(struct folio *, gfp_t); 266 void (*free_folio)(struct folio *); 267 int (*direct_IO)(struct kiocb *, struct iov_iter *iter); 268 int (*migrate_folio)(struct address_space *, struct folio *dst, 269 struct folio *src, enum migrate_mode); 270 int (*launder_folio)(struct folio *); 271 bool (*is_partially_uptodate)(struct folio *, size_t from, size_t count); 272 int (*error_remove_folio)(struct address_space *, struct folio *); 273 int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span) 274 int (*swap_deactivate)(struct file *); 275 int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter); 276 277locking rules: 278 All except dirty_folio and free_folio may block 279 280====================== ======================== ========= =============== 281ops folio locked i_rwsem invalidate_lock 282====================== ======================== ========= =============== 283writepage: yes, unlocks (see below) 284read_folio: yes, unlocks shared 285writepages: 286dirty_folio: maybe 287readahead: yes, unlocks shared 288write_begin: locks the folio exclusive 289write_end: yes, unlocks exclusive 290bmap: 291invalidate_folio: yes exclusive 292release_folio: yes 293free_folio: yes 294direct_IO: 295migrate_folio: yes (both) 296launder_folio: yes 297is_partially_uptodate: yes 298error_remove_folio: yes 299swap_activate: no 300swap_deactivate: no 301swap_rw: yes, unlocks 302====================== ======================== ========= =============== 303 304->write_begin(), ->write_end() and ->read_folio() may be called from 305the request handler (/dev/loop). 306 307->read_folio() unlocks the folio, either synchronously or via I/O 308completion. 309 310->readahead() unlocks the folios that I/O is attempted on like ->read_folio(). 311 312->writepage() is used for two purposes: for "memory cleansing" and for 313"sync". These are quite different operations and the behaviour may differ 314depending upon the mode. 315 316If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then 317it *must* start I/O against the page, even if that would involve 318blocking on in-progress I/O. 319 320If writepage is called for memory cleansing (sync_mode == 321WBC_SYNC_NONE) then its role is to get as much writeout underway as 322possible. So writepage should try to avoid blocking against 323currently-in-progress I/O. 324 325If the filesystem is not called for "sync" and it determines that it 326would need to block against in-progress I/O to be able to start new I/O 327against the page the filesystem should redirty the page with 328redirty_page_for_writepage(), then unlock the page and return zero. 329This may also be done to avoid internal deadlocks, but rarely. 330 331If the filesystem is called for sync then it must wait on any 332in-progress I/O and then start new I/O. 333 334The filesystem should unlock the page synchronously, before returning to the 335caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE 336value. WRITEPAGE_ACTIVATE means that page cannot really be written out 337currently, and VM should stop calling ->writepage() on this page for some 338time. VM does this by moving page to the head of the active list, hence the 339name. 340 341Unless the filesystem is going to redirty_page_for_writepage(), unlock the page 342and return zero, writepage *must* run set_page_writeback() against the page, 343followed by unlocking it. Once set_page_writeback() has been run against the 344page, write I/O can be submitted and the write I/O completion handler must run 345end_page_writeback() once the I/O is complete. If no I/O is submitted, the 346filesystem must run end_page_writeback() against the page before returning from 347writepage. 348 349That is: after 2.5.12, pages which are under writeout are *not* locked. Note, 350if the filesystem needs the page to be locked during writeout, that is ok, too, 351the page is allowed to be unlocked at any point in time between the calls to 352set_page_writeback() and end_page_writeback(). 353 354Note, failure to run either redirty_page_for_writepage() or the combination of 355set_page_writeback()/end_page_writeback() on a page submitted to writepage 356will leave the page itself marked clean but it will be tagged as dirty in the 357radix tree. This incoherency can lead to all sorts of hard-to-debug problems 358in the filesystem like having dirty inodes at umount and losing written data. 359 360->writepages() is used for periodic writeback and for syscall-initiated 361sync operations. The address_space should start I/O against at least 362``*nr_to_write`` pages. ``*nr_to_write`` must be decremented for each page 363which is written. The address_space implementation may write more (or less) 364pages than ``*nr_to_write`` asks for, but it should try to be reasonably close. 365If nr_to_write is NULL, all dirty pages must be written. 366 367writepages should _only_ write pages which are present on 368mapping->io_pages. 369 370->dirty_folio() is called from various places in the kernel when 371the target folio is marked as needing writeback. The folio cannot be 372truncated because either the caller holds the folio lock, or the caller 373has found the folio while holding the page table lock which will block 374truncation. 375 376->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some 377filesystems and by the swapper. The latter will eventually go away. Please, 378keep it that way and don't breed new callers. 379 380->invalidate_folio() is called when the filesystem must attempt to drop 381some or all of the buffers from the page when it is being truncated. It 382returns zero on success. The filesystem must exclusively acquire 383invalidate_lock before invalidating page cache in truncate / hole punch 384path (and thus calling into ->invalidate_folio) to block races between page 385cache invalidation and page cache filling functions (fault, read, ...). 386 387->release_folio() is called when the MM wants to make a change to the 388folio that would invalidate the filesystem's private data. For example, 389it may be about to be removed from the address_space or split. The folio 390is locked and not under writeback. It may be dirty. The gfp parameter 391is not usually used for allocation, but rather to indicate what the 392filesystem may do to attempt to free the private data. The filesystem may 393return false to indicate that the folio's private data cannot be freed. 394If it returns true, it should have already removed the private data from 395the folio. If a filesystem does not provide a ->release_folio method, 396the pagecache will assume that private data is buffer_heads and call 397try_to_free_buffers(). 398 399->free_folio() is called when the kernel has dropped the folio 400from the page cache. 401 402->launder_folio() may be called prior to releasing a folio if 403it is still found to be dirty. It returns zero if the folio was successfully 404cleaned, or an error value if not. Note that in order to prevent the folio 405getting mapped back in and redirtied, it needs to be kept locked 406across the entire operation. 407 408->swap_activate() will be called to prepare the given file for swap. It 409should perform any validation and preparation necessary to ensure that 410writes can be performed with minimal memory allocation. It should call 411add_swap_extent(), or the helper iomap_swapfile_activate(), and return 412the number of extents added. If IO should be submitted through 413->swap_rw(), it should set SWP_FS_OPS, otherwise IO will be submitted 414directly to the block device ``sis->bdev``. 415 416->swap_deactivate() will be called in the sys_swapoff() 417path after ->swap_activate() returned success. 418 419->swap_rw will be called for swap IO if SWP_FS_OPS was set by ->swap_activate(). 420 421file_lock_operations 422==================== 423 424prototypes:: 425 426 void (*fl_copy_lock)(struct file_lock *, struct file_lock *); 427 void (*fl_release_private)(struct file_lock *); 428 429 430locking rules: 431 432=================== ============= ========= 433ops inode->i_lock may block 434=================== ============= ========= 435fl_copy_lock: yes no 436fl_release_private: maybe maybe[1]_ 437=================== ============= ========= 438 439.. [1]: 440 ->fl_release_private for flock or POSIX locks is currently allowed 441 to block. Leases however can still be freed while the i_lock is held and 442 so fl_release_private called on a lease should not block. 443 444lock_manager_operations 445======================= 446 447prototypes:: 448 449 void (*lm_notify)(struct file_lock *); /* unblock callback */ 450 int (*lm_grant)(struct file_lock *, struct file_lock *, int); 451 void (*lm_break)(struct file_lock *); /* break_lease callback */ 452 int (*lm_change)(struct file_lock **, int); 453 bool (*lm_breaker_owns_lease)(struct file_lock *); 454 bool (*lm_lock_expirable)(struct file_lock *); 455 void (*lm_expire_lock)(void); 456 457locking rules: 458 459====================== ============= ================= ========= 460ops flc_lock blocked_lock_lock may block 461====================== ============= ================= ========= 462lm_notify: no yes no 463lm_grant: no no no 464lm_break: yes no no 465lm_change yes no no 466lm_breaker_owns_lease: yes no no 467lm_lock_expirable yes no no 468lm_expire_lock no no yes 469====================== ============= ================= ========= 470 471buffer_head 472=========== 473 474prototypes:: 475 476 void (*b_end_io)(struct buffer_head *bh, int uptodate); 477 478locking rules: 479 480called from interrupts. In other words, extreme care is needed here. 481bh is locked, but that's all warranties we have here. Currently only RAID1, 482highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices 483call this method upon the IO completion. 484 485block_device_operations 486======================= 487prototypes:: 488 489 int (*open) (struct block_device *, fmode_t); 490 int (*release) (struct gendisk *, fmode_t); 491 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long); 492 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long); 493 int (*direct_access) (struct block_device *, sector_t, void **, 494 unsigned long *); 495 void (*unlock_native_capacity) (struct gendisk *); 496 int (*getgeo)(struct block_device *, struct hd_geometry *); 497 void (*swap_slot_free_notify) (struct block_device *, unsigned long); 498 499locking rules: 500 501======================= =================== 502ops open_mutex 503======================= =================== 504open: yes 505release: yes 506ioctl: no 507compat_ioctl: no 508direct_access: no 509unlock_native_capacity: no 510getgeo: no 511swap_slot_free_notify: no (see below) 512======================= =================== 513 514swap_slot_free_notify is called with swap_lock and sometimes the page lock 515held. 516 517 518file_operations 519=============== 520 521prototypes:: 522 523 loff_t (*llseek) (struct file *, loff_t, int); 524 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *); 525 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *); 526 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *); 527 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *); 528 int (*iopoll) (struct kiocb *kiocb, bool spin); 529 int (*iterate_shared) (struct file *, struct dir_context *); 530 __poll_t (*poll) (struct file *, struct poll_table_struct *); 531 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long); 532 long (*compat_ioctl) (struct file *, unsigned int, unsigned long); 533 int (*mmap) (struct file *, struct vm_area_struct *); 534 int (*open) (struct inode *, struct file *); 535 int (*flush) (struct file *); 536 int (*release) (struct inode *, struct file *); 537 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync); 538 int (*fasync) (int, struct file *, int); 539 int (*lock) (struct file *, int, struct file_lock *); 540 unsigned long (*get_unmapped_area)(struct file *, unsigned long, 541 unsigned long, unsigned long, unsigned long); 542 int (*check_flags)(int); 543 int (*flock) (struct file *, int, struct file_lock *); 544 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, 545 size_t, unsigned int); 546 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, 547 size_t, unsigned int); 548 int (*setlease)(struct file *, long, struct file_lock **, void **); 549 long (*fallocate)(struct file *, int, loff_t, loff_t); 550 void (*show_fdinfo)(struct seq_file *m, struct file *f); 551 unsigned (*mmap_capabilities)(struct file *); 552 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *, 553 loff_t, size_t, unsigned int); 554 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in, 555 struct file *file_out, loff_t pos_out, 556 loff_t len, unsigned int remap_flags); 557 int (*fadvise)(struct file *, loff_t, loff_t, int); 558 559locking rules: 560 All may block. 561 562->llseek() locking has moved from llseek to the individual llseek 563implementations. If your fs is not using generic_file_llseek, you 564need to acquire and release the appropriate locks in your ->llseek(). 565For many filesystems, it is probably safe to acquire the inode 566mutex or just to use i_size_read() instead. 567Note: this does not protect the file->f_pos against concurrent modifications 568since this is something the userspace has to take care about. 569 570->iterate_shared() is called with i_rwsem held for reading, and with the 571file f_pos_lock held exclusively 572 573->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags. 574Most instances call fasync_helper(), which does that maintenance, so it's 575not normally something one needs to worry about. Return values > 0 will be 576mapped to zero in the VFS layer. 577 578->readdir() and ->ioctl() on directories must be changed. Ideally we would 579move ->readdir() to inode_operations and use a separate method for directory 580->ioctl() or kill the latter completely. One of the problems is that for 581anything that resembles union-mount we won't have a struct file for all 582components. And there are other reasons why the current interface is a mess... 583 584->read on directories probably must go away - we should just enforce -EISDIR 585in sys_read() and friends. 586 587->setlease operations should call generic_setlease() before or after setting 588the lease within the individual filesystem to record the result of the 589operation 590 591->fallocate implementation must be really careful to maintain page cache 592consistency when punching holes or performing other operations that invalidate 593page cache contents. Usually the filesystem needs to call 594truncate_inode_pages_range() to invalidate relevant range of the page cache. 595However the filesystem usually also needs to update its internal (and on disk) 596view of file offset -> disk block mapping. Until this update is finished, the 597filesystem needs to block page faults and reads from reloading now-stale page 598cache contents from the disk. Since VFS acquires mapping->invalidate_lock in 599shared mode when loading pages from disk (filemap_fault(), filemap_read(), 600readahead paths), the fallocate implementation must take the invalidate_lock to 601prevent reloading. 602 603->copy_file_range and ->remap_file_range implementations need to serialize 604against modifications of file data while the operation is running. For 605blocking changes through write(2) and similar operations inode->i_rwsem can be 606used. To block changes to file contents via a memory mapping during the 607operation, the filesystem must take mapping->invalidate_lock to coordinate 608with ->page_mkwrite. 609 610dquot_operations 611================ 612 613prototypes:: 614 615 int (*write_dquot) (struct dquot *); 616 int (*acquire_dquot) (struct dquot *); 617 int (*release_dquot) (struct dquot *); 618 int (*mark_dirty) (struct dquot *); 619 int (*write_info) (struct super_block *, int); 620 621These operations are intended to be more or less wrapping functions that ensure 622a proper locking wrt the filesystem and call the generic quota operations. 623 624What filesystem should expect from the generic quota functions: 625 626============== ============ ========================= 627ops FS recursion Held locks when called 628============== ============ ========================= 629write_dquot: yes dqonoff_sem or dqptr_sem 630acquire_dquot: yes dqonoff_sem or dqptr_sem 631release_dquot: yes dqonoff_sem or dqptr_sem 632mark_dirty: no - 633write_info: yes dqonoff_sem 634============== ============ ========================= 635 636FS recursion means calling ->quota_read() and ->quota_write() from superblock 637operations. 638 639More details about quota locking can be found in fs/dquot.c. 640 641vm_operations_struct 642==================== 643 644prototypes:: 645 646 void (*open)(struct vm_area_struct *); 647 void (*close)(struct vm_area_struct *); 648 vm_fault_t (*fault)(struct vm_fault *); 649 vm_fault_t (*huge_fault)(struct vm_fault *, unsigned int order); 650 vm_fault_t (*map_pages)(struct vm_fault *, pgoff_t start, pgoff_t end); 651 vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *); 652 vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *); 653 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int); 654 655locking rules: 656 657============= ========== =========================== 658ops mmap_lock PageLocked(page) 659============= ========== =========================== 660open: write 661close: read/write 662fault: read can return with page locked 663huge_fault: maybe-read 664map_pages: maybe-read 665page_mkwrite: read can return with page locked 666pfn_mkwrite: read 667access: read 668============= ========== =========================== 669 670->fault() is called when a previously not present pte is about to be faulted 671in. The filesystem must find and return the page associated with the passed in 672"pgoff" in the vm_fault structure. If it is possible that the page may be 673truncated and/or invalidated, then the filesystem must lock invalidate_lock, 674then ensure the page is not already truncated (invalidate_lock will block 675subsequent truncate), and then return with VM_FAULT_LOCKED, and the page 676locked. The VM will unlock the page. 677 678->huge_fault() is called when there is no PUD or PMD entry present. This 679gives the filesystem the opportunity to install a PUD or PMD sized page. 680Filesystems can also use the ->fault method to return a PMD sized page, 681so implementing this function may not be necessary. In particular, 682filesystems should not call filemap_fault() from ->huge_fault(). 683The mmap_lock may not be held when this method is called. 684 685->map_pages() is called when VM asks to map easy accessible pages. 686Filesystem should find and map pages associated with offsets from "start_pgoff" 687till "end_pgoff". ->map_pages() is called with the RCU lock held and must 688not block. If it's not possible to reach a page without blocking, 689filesystem should skip it. Filesystem should use set_pte_range() to setup 690page table entry. Pointer to entry associated with the page is passed in 691"pte" field in vm_fault structure. Pointers to entries for other offsets 692should be calculated relative to "pte". 693 694->page_mkwrite() is called when a previously read-only pte is about to become 695writeable. The filesystem again must ensure that there are no 696truncate/invalidate races or races with operations such as ->remap_file_range 697or ->copy_file_range, and then return with the page locked. Usually 698mapping->invalidate_lock is suitable for proper serialization. If the page has 699been truncated, the filesystem should not look up a new page like the ->fault() 700handler, but simply return with VM_FAULT_NOPAGE, which will cause the VM to 701retry the fault. 702 703->pfn_mkwrite() is the same as page_mkwrite but when the pte is 704VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is 705VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior 706after this call is to make the pte read-write, unless pfn_mkwrite returns 707an error. 708 709->access() is called when get_user_pages() fails in 710access_process_vm(), typically used to debug a process through 711/proc/pid/mem or ptrace. This function is needed only for 712VM_IO | VM_PFNMAP VMAs. 713 714-------------------------------------------------------------------------------- 715 716 Dubious stuff 717 718(if you break something or notice that it is broken and do not fix it yourself 719- at least put it here) 720