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).
13 Thing currently missing here: socket operations. Alexey?
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 *);
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
42 d_compare: yes no no maybe
43 d_delete: no yes no no
45 d_release: no no yes no
49 d_automount: no no yes no
50 d_manage: no no yes (ref-walk) maybe
52 ================== =========== ======== ============== ========
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,
82 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
87 ============ =============================================
89 ============ =============================================
92 link: exclusive (both)
96 unlink: exclusive (both)
97 rmdir: exclusive (both)(see below)
98 rename: exclusive (all) (see below)
102 permission: no (may not block if called in rcu-walk mode)
108 atomic_open: shared (exclusive if O_CREAT is set in open flags)
110 ============ =============================================
113 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
115 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
117 See Documentation/filesystems/directory-locking.rst for more detailed discussion
118 of the locking scheme for directory operations.
120 xattr_handler operations
121 ========================
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,
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);
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);
169 All may block [not true, see below]
171 ====================== ============ ========================
173 ====================== ============ ========================
175 free_inode: called from RCU callback
179 drop_inode: !!!inode->i_lock!!!
185 statfs: maybe(read) (see below)
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 ====================== ============ ========================
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.
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.
207 ->bdev_try_to_free_page is called from the ->releasepage handler of
208 the block device inode. See there for more details.
215 struct dentry *(*mount) (struct file_system_type *, int,
216 const char *, void *);
217 void (*kill_sb) (struct super_block *);
228 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
231 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
232 unlocks and drops the reference.
234 address_space_operations
235 ========================
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 *);
266 All except set_page_dirty and freepage may block
268 ====================== ======================== =========
269 ops PageLocked(page) i_rwsem
270 ====================== ======================== =========
271 writepage: yes, unlocks (see below)
272 readpage: yes, unlocks
275 readahead: yes, unlocks
277 write_begin: locks the page exclusive
278 write_end: yes, unlocks exclusive
285 migratepage: yes (both)
288 is_partially_uptodate: yes
289 error_remove_page: yes
292 ====================== ======================== =========
294 ->write_begin(), ->write_end() and ->readpage() may be called from
295 the request handler (/dev/loop).
297 ->readpage() unlocks the page, either synchronously or via I/O
300 ->readahead() unlocks the pages that I/O is attempted on like ->readpage().
302 ->readpages() populates the pagecache with the passed pages and starts
303 I/O against them. They come unlocked upon I/O completion.
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.
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.
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.
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.
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.
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
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
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().
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.
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.
360 writepages should _only_ write pages which are present on
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
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.
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.
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.
382 ->freepage() is called when the kernel is done dropping the page
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.
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.
397 ->swap_deactivate() will be called in the sys_swapoff()
398 path after ->swap_activate() returned success.
405 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
406 void (*fl_release_private)(struct file_lock *);
411 =================== ============= =========
412 ops inode->i_lock may block
413 =================== ============= =========
415 fl_release_private: maybe maybe[1]_
416 =================== ============= =========
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.
423 lock_manager_operations
424 =======================
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 *);
436 ========== ============= ================= =========
437 ops inode->i_lock blocked_lock_lock may block
438 ========== ============= ================= =========
439 lm_notify: yes yes no
443 lm_breaker_owns_lease: no no no
444 ========== ============= ================= =========
451 void (*b_end_io)(struct buffer_head *bh, int uptodate);
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.
460 block_device_operations
461 =======================
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 **,
470 int (*media_changed) (struct gendisk *);
471 void (*unlock_native_capacity) (struct gendisk *);
472 int (*revalidate_disk) (struct gendisk *);
473 int (*getgeo)(struct block_device *, struct hd_geometry *);
474 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
478 ======================= ===================
480 ======================= ===================
487 unlock_native_capacity: no
490 swap_slot_free_notify: no (see below)
491 ======================= ===================
493 media_changed, unlock_native_capacity and revalidate_disk are called only from
496 swap_slot_free_notify is called with swap_lock and sometimes the page lock
505 loff_t (*llseek) (struct file *, loff_t, int);
506 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
507 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
508 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
509 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
510 int (*iterate) (struct file *, struct dir_context *);
511 int (*iterate_shared) (struct file *, struct dir_context *);
512 __poll_t (*poll) (struct file *, struct poll_table_struct *);
513 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
514 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
515 int (*mmap) (struct file *, struct vm_area_struct *);
516 int (*open) (struct inode *, struct file *);
517 int (*flush) (struct file *);
518 int (*release) (struct inode *, struct file *);
519 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
520 int (*fasync) (int, struct file *, int);
521 int (*lock) (struct file *, int, struct file_lock *);
522 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
524 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
526 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
528 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
530 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
531 unsigned long, unsigned long, unsigned long);
532 int (*check_flags)(int);
533 int (*flock) (struct file *, int, struct file_lock *);
534 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
535 size_t, unsigned int);
536 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
537 size_t, unsigned int);
538 int (*setlease)(struct file *, long, struct file_lock **, void **);
539 long (*fallocate)(struct file *, int, loff_t, loff_t);
544 ->llseek() locking has moved from llseek to the individual llseek
545 implementations. If your fs is not using generic_file_llseek, you
546 need to acquire and release the appropriate locks in your ->llseek().
547 For many filesystems, it is probably safe to acquire the inode
548 mutex or just to use i_size_read() instead.
549 Note: this does not protect the file->f_pos against concurrent modifications
550 since this is something the userspace has to take care about.
552 ->iterate() is called with i_rwsem exclusive.
554 ->iterate_shared() is called with i_rwsem at least shared.
556 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
557 Most instances call fasync_helper(), which does that maintenance, so it's
558 not normally something one needs to worry about. Return values > 0 will be
559 mapped to zero in the VFS layer.
561 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
562 move ->readdir() to inode_operations and use a separate method for directory
563 ->ioctl() or kill the latter completely. One of the problems is that for
564 anything that resembles union-mount we won't have a struct file for all
565 components. And there are other reasons why the current interface is a mess...
567 ->read on directories probably must go away - we should just enforce -EISDIR
568 in sys_read() and friends.
570 ->setlease operations should call generic_setlease() before or after setting
571 the lease within the individual filesystem to record the result of the
579 int (*write_dquot) (struct dquot *);
580 int (*acquire_dquot) (struct dquot *);
581 int (*release_dquot) (struct dquot *);
582 int (*mark_dirty) (struct dquot *);
583 int (*write_info) (struct super_block *, int);
585 These operations are intended to be more or less wrapping functions that ensure
586 a proper locking wrt the filesystem and call the generic quota operations.
588 What filesystem should expect from the generic quota functions:
590 ============== ============ =========================
591 ops FS recursion Held locks when called
592 ============== ============ =========================
593 write_dquot: yes dqonoff_sem or dqptr_sem
594 acquire_dquot: yes dqonoff_sem or dqptr_sem
595 release_dquot: yes dqonoff_sem or dqptr_sem
597 write_info: yes dqonoff_sem
598 ============== ============ =========================
600 FS recursion means calling ->quota_read() and ->quota_write() from superblock
603 More details about quota locking can be found in fs/dquot.c.
610 void (*open)(struct vm_area_struct*);
611 void (*close)(struct vm_area_struct*);
612 vm_fault_t (*fault)(struct vm_area_struct*, struct vm_fault *);
613 vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
614 vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
615 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
619 ============= ======== ===========================
620 ops mmap_lock PageLocked(page)
621 ============= ======== ===========================
624 fault: yes can return with page locked
626 page_mkwrite: yes can return with page locked
629 ============= ======== ===========================
631 ->fault() is called when a previously not present pte is about
632 to be faulted in. The filesystem must find and return the page associated
633 with the passed in "pgoff" in the vm_fault structure. If it is possible that
634 the page may be truncated and/or invalidated, then the filesystem must lock
635 the page, then ensure it is not already truncated (the page lock will block
636 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
637 locked. The VM will unlock the page.
639 ->map_pages() is called when VM asks to map easy accessible pages.
640 Filesystem should find and map pages associated with offsets from "start_pgoff"
641 till "end_pgoff". ->map_pages() is called with page table locked and must
642 not block. If it's not possible to reach a page without blocking,
643 filesystem should skip it. Filesystem should use do_set_pte() to setup
644 page table entry. Pointer to entry associated with the page is passed in
645 "pte" field in vm_fault structure. Pointers to entries for other offsets
646 should be calculated relative to "pte".
648 ->page_mkwrite() is called when a previously read-only pte is
649 about to become writeable. The filesystem again must ensure that there are
650 no truncate/invalidate races, and then return with the page locked. If
651 the page has been truncated, the filesystem should not look up a new page
652 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
653 will cause the VM to retry the fault.
655 ->pfn_mkwrite() is the same as page_mkwrite but when the pte is
656 VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
657 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
658 after this call is to make the pte read-write, unless pfn_mkwrite returns
661 ->access() is called when get_user_pages() fails in
662 access_process_vm(), typically used to debug a process through
663 /proc/pid/mem or ptrace. This function is needed only for
664 VM_IO | VM_PFNMAP VMAs.
666 --------------------------------------------------------------------------------
670 (if you break something or notice that it is broken and do not fix it yourself
671 - at least put it here)