1 The text below describes the locking rules for VFS-related methods.
2 It is (believed to be) up-to-date. *Please*, if you change anything in
3 prototypes or locking protocols - update this file. And update the relevant
4 instances in the tree, don't leave that to maintainers of filesystems/devices/
5 etc. At the very least, put the list of dubious cases in the end of this file.
6 Don't turn it into log - maintainers of out-of-the-tree code are supposed to
7 be able to use diff(1).
8 Thing currently missing here: socket operations. Alexey?
10 --------------------------- dentry_operations --------------------------
12 int (*d_revalidate)(struct dentry *, unsigned int);
13 int (*d_weak_revalidate)(struct dentry *, unsigned int);
14 int (*d_hash)(const struct dentry *, struct qstr *);
15 int (*d_compare)(const struct dentry *,
16 unsigned int, const char *, const struct qstr *);
17 int (*d_delete)(struct dentry *);
18 int (*d_init)(struct dentry *);
19 void (*d_release)(struct dentry *);
20 void (*d_iput)(struct dentry *, struct inode *);
21 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
22 struct vfsmount *(*d_automount)(struct path *path);
23 int (*d_manage)(const struct path *, bool);
24 struct dentry *(*d_real)(struct dentry *, const struct inode *);
27 rename_lock ->d_lock may block rcu-walk
28 d_revalidate: no no yes (ref-walk) maybe
29 d_weak_revalidate:no no yes no
31 d_compare: yes no no maybe
32 d_delete: no yes no no
34 d_release: no no yes no
38 d_automount: no no yes no
39 d_manage: no no yes (ref-walk) maybe
42 --------------------------- inode_operations ---------------------------
44 int (*create) (struct inode *,struct dentry *,umode_t, bool);
45 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
46 int (*link) (struct dentry *,struct inode *,struct dentry *);
47 int (*unlink) (struct inode *,struct dentry *);
48 int (*symlink) (struct inode *,struct dentry *,const char *);
49 int (*mkdir) (struct inode *,struct dentry *,umode_t);
50 int (*rmdir) (struct inode *,struct dentry *);
51 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
52 int (*rename) (struct inode *, struct dentry *,
53 struct inode *, struct dentry *, unsigned int);
54 int (*readlink) (struct dentry *, char __user *,int);
55 const char *(*get_link) (struct dentry *, struct inode *, void **);
56 void (*truncate) (struct inode *);
57 int (*permission) (struct inode *, int, unsigned int);
58 int (*get_acl)(struct inode *, int);
59 int (*setattr) (struct dentry *, struct iattr *);
60 int (*getattr) (const struct path *, struct kstat *, u32, unsigned int);
61 ssize_t (*listxattr) (struct dentry *, char *, size_t);
62 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
63 void (*update_time)(struct inode *, struct timespec *, int);
64 int (*atomic_open)(struct inode *, struct dentry *,
65 struct file *, unsigned open_flag,
67 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
74 link: exclusive (both)
78 unlink: exclusive (both)
79 rmdir: exclusive (both)(see below)
80 rename: exclusive (all) (see below)
84 permission: no (may not block if called in rcu-walk mode)
90 atomic_open: exclusive
94 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
96 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
98 See Documentation/filesystems/directory-locking for more detailed discussion
99 of the locking scheme for directory operations.
101 ----------------------- xattr_handler operations -----------------------
103 bool (*list)(struct dentry *dentry);
104 int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
105 struct inode *inode, const char *name, void *buffer,
107 int (*set)(const struct xattr_handler *handler, struct dentry *dentry,
108 struct inode *inode, const char *name, const void *buffer,
109 size_t size, int flags);
118 --------------------------- super_operations ---------------------------
120 struct inode *(*alloc_inode)(struct super_block *sb);
121 void (*destroy_inode)(struct inode *);
122 void (*dirty_inode) (struct inode *, int flags);
123 int (*write_inode) (struct inode *, struct writeback_control *wbc);
124 int (*drop_inode) (struct inode *);
125 void (*evict_inode) (struct inode *);
126 void (*put_super) (struct super_block *);
127 int (*sync_fs)(struct super_block *sb, int wait);
128 int (*freeze_fs) (struct super_block *);
129 int (*unfreeze_fs) (struct super_block *);
130 int (*statfs) (struct dentry *, struct kstatfs *);
131 int (*remount_fs) (struct super_block *, int *, char *);
132 void (*umount_begin) (struct super_block *);
133 int (*show_options)(struct seq_file *, struct dentry *);
134 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
135 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
136 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
139 All may block [not true, see below]
145 drop_inode: !!!inode->i_lock!!!
151 statfs: maybe(read) (see below)
154 show_options: no (namespace_sem)
155 quota_read: no (see below)
156 quota_write: no (see below)
157 bdev_try_to_free_page: no (see below)
159 ->statfs() has s_umount (shared) when called by ustat(2) (native or
160 compat), but that's an accident of bad API; s_umount is used to pin
161 the superblock down when we only have dev_t given us by userland to
162 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
163 doesn't hold it when calling ->statfs() - superblock is pinned down
164 by resolving the pathname passed to syscall.
165 ->quota_read() and ->quota_write() functions are both guaranteed to
166 be the only ones operating on the quota file by the quota code (via
167 dqio_sem) (unless an admin really wants to screw up something and
168 writes to quota files with quotas on). For other details about locking
169 see also dquot_operations section.
170 ->bdev_try_to_free_page is called from the ->releasepage handler of
171 the block device inode. See there for more details.
173 --------------------------- file_system_type ---------------------------
175 struct dentry *(*mount) (struct file_system_type *, int,
176 const char *, void *);
177 void (*kill_sb) (struct super_block *);
183 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
185 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
186 unlocks and drops the reference.
188 --------------------------- address_space_operations --------------------------
190 int (*writepage)(struct page *page, struct writeback_control *wbc);
191 int (*readpage)(struct file *, struct page *);
192 int (*writepages)(struct address_space *, struct writeback_control *);
193 int (*set_page_dirty)(struct page *page);
194 int (*readpages)(struct file *filp, struct address_space *mapping,
195 struct list_head *pages, unsigned nr_pages);
196 int (*write_begin)(struct file *, struct address_space *mapping,
197 loff_t pos, unsigned len, unsigned flags,
198 struct page **pagep, void **fsdata);
199 int (*write_end)(struct file *, struct address_space *mapping,
200 loff_t pos, unsigned len, unsigned copied,
201 struct page *page, void *fsdata);
202 sector_t (*bmap)(struct address_space *, sector_t);
203 void (*invalidatepage) (struct page *, unsigned int, unsigned int);
204 int (*releasepage) (struct page *, int);
205 void (*freepage)(struct page *);
206 int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
207 bool (*isolate_page) (struct page *, isolate_mode_t);
208 int (*migratepage)(struct address_space *, struct page *, struct page *);
209 void (*putback_page) (struct page *);
210 int (*launder_page)(struct page *);
211 int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long);
212 int (*error_remove_page)(struct address_space *, struct page *);
213 int (*swap_activate)(struct file *);
214 int (*swap_deactivate)(struct file *);
217 All except set_page_dirty and freepage may block
219 PageLocked(page) i_rwsem
220 writepage: yes, unlocks (see below)
221 readpage: yes, unlocks
225 write_begin: locks the page exclusive
226 write_end: yes, unlocks exclusive
233 migratepage: yes (both)
236 is_partially_uptodate: yes
237 error_remove_page: yes
241 ->write_begin(), ->write_end() and ->readpage() may be called from
242 the request handler (/dev/loop).
244 ->readpage() unlocks the page, either synchronously or via I/O
247 ->readpages() populates the pagecache with the passed pages and starts
248 I/O against them. They come unlocked upon I/O completion.
250 ->writepage() is used for two purposes: for "memory cleansing" and for
251 "sync". These are quite different operations and the behaviour may differ
252 depending upon the mode.
254 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
255 it *must* start I/O against the page, even if that would involve
256 blocking on in-progress I/O.
258 If writepage is called for memory cleansing (sync_mode ==
259 WBC_SYNC_NONE) then its role is to get as much writeout underway as
260 possible. So writepage should try to avoid blocking against
261 currently-in-progress I/O.
263 If the filesystem is not called for "sync" and it determines that it
264 would need to block against in-progress I/O to be able to start new I/O
265 against the page the filesystem should redirty the page with
266 redirty_page_for_writepage(), then unlock the page and return zero.
267 This may also be done to avoid internal deadlocks, but rarely.
269 If the filesystem is called for sync then it must wait on any
270 in-progress I/O and then start new I/O.
272 The filesystem should unlock the page synchronously, before returning to the
273 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
274 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
275 currently, and VM should stop calling ->writepage() on this page for some
276 time. VM does this by moving page to the head of the active list, hence the
279 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
280 and return zero, writepage *must* run set_page_writeback() against the page,
281 followed by unlocking it. Once set_page_writeback() has been run against the
282 page, write I/O can be submitted and the write I/O completion handler must run
283 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
284 filesystem must run end_page_writeback() against the page before returning from
287 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
288 if the filesystem needs the page to be locked during writeout, that is ok, too,
289 the page is allowed to be unlocked at any point in time between the calls to
290 set_page_writeback() and end_page_writeback().
292 Note, failure to run either redirty_page_for_writepage() or the combination of
293 set_page_writeback()/end_page_writeback() on a page submitted to writepage
294 will leave the page itself marked clean but it will be tagged as dirty in the
295 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
296 in the filesystem like having dirty inodes at umount and losing written data.
298 ->writepages() is used for periodic writeback and for syscall-initiated
299 sync operations. The address_space should start I/O against at least
300 *nr_to_write pages. *nr_to_write must be decremented for each page which is
301 written. The address_space implementation may write more (or less) pages
302 than *nr_to_write asks for, but it should try to be reasonably close. If
303 nr_to_write is NULL, all dirty pages must be written.
305 writepages should _only_ write pages which are present on
308 ->set_page_dirty() is called from various places in the kernel
309 when the target page is marked as needing writeback. It may be called
310 under spinlock (it cannot block) and is sometimes called with the page
313 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
314 filesystems and by the swapper. The latter will eventually go away. Please,
315 keep it that way and don't breed new callers.
317 ->invalidatepage() is called when the filesystem must attempt to drop
318 some or all of the buffers from the page when it is being truncated. It
319 returns zero on success. If ->invalidatepage is zero, the kernel uses
320 block_invalidatepage() instead.
322 ->releasepage() is called when the kernel is about to try to drop the
323 buffers from the page in preparation for freeing it. It returns zero to
324 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
325 the kernel assumes that the fs has no private interest in the buffers.
327 ->freepage() is called when the kernel is done dropping the page
330 ->launder_page() may be called prior to releasing a page if
331 it is still found to be dirty. It returns zero if the page was successfully
332 cleaned, or an error value if not. Note that in order to prevent the page
333 getting mapped back in and redirtied, it needs to be kept locked
334 across the entire operation.
336 ->swap_activate will be called with a non-zero argument on
337 files backing (non block device backed) swapfiles. A return value
338 of zero indicates success, in which case this file can be used for
339 backing swapspace. The swapspace operations will be proxied to the
340 address space operations.
342 ->swap_deactivate() will be called in the sys_swapoff()
343 path after ->swap_activate() returned success.
345 ----------------------- file_lock_operations ------------------------------
347 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
348 void (*fl_release_private)(struct file_lock *);
352 inode->i_lock may block
354 fl_release_private: maybe maybe[1]
356 [1]: ->fl_release_private for flock or POSIX locks is currently allowed
357 to block. Leases however can still be freed while the i_lock is held and
358 so fl_release_private called on a lease should not block.
360 ----------------------- lock_manager_operations ---------------------------
362 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
363 unsigned long (*lm_owner_key)(struct file_lock *);
364 void (*lm_notify)(struct file_lock *); /* unblock callback */
365 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
366 void (*lm_break)(struct file_lock *); /* break_lease callback */
367 int (*lm_change)(struct file_lock **, int);
371 inode->i_lock blocked_lock_lock may block
372 lm_compare_owner: yes[1] maybe no
373 lm_owner_key yes[1] yes no
374 lm_notify: yes yes no
379 [1]: ->lm_compare_owner and ->lm_owner_key are generally called with
380 *an* inode->i_lock held. It may not be the i_lock of the inode
381 associated with either file_lock argument! This is the case with deadlock
382 detection, since the code has to chase down the owners of locks that may
383 be entirely unrelated to the one on which the lock is being acquired.
384 For deadlock detection however, the blocked_lock_lock is also held. The
385 fact that these locks are held ensures that the file_locks do not
386 disappear out from under you while doing the comparison or generating an
389 --------------------------- buffer_head -----------------------------------
391 void (*b_end_io)(struct buffer_head *bh, int uptodate);
394 called from interrupts. In other words, extreme care is needed here.
395 bh is locked, but that's all warranties we have here. Currently only RAID1,
396 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
397 call this method upon the IO completion.
399 --------------------------- block_device_operations -----------------------
401 int (*open) (struct block_device *, fmode_t);
402 int (*release) (struct gendisk *, fmode_t);
403 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
404 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
405 int (*direct_access) (struct block_device *, sector_t, void **,
407 int (*media_changed) (struct gendisk *);
408 void (*unlock_native_capacity) (struct gendisk *);
409 int (*revalidate_disk) (struct gendisk *);
410 int (*getgeo)(struct block_device *, struct hd_geometry *);
411 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
421 unlock_native_capacity: no
424 swap_slot_free_notify: no (see below)
426 media_changed, unlock_native_capacity and revalidate_disk are called only from
429 swap_slot_free_notify is called with swap_lock and sometimes the page lock
433 --------------------------- file_operations -------------------------------
435 loff_t (*llseek) (struct file *, loff_t, int);
436 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
437 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
438 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
439 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
440 int (*iterate) (struct file *, struct dir_context *);
441 int (*iterate_shared) (struct file *, struct dir_context *);
442 __poll_t (*poll) (struct file *, struct poll_table_struct *);
443 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
444 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
445 int (*mmap) (struct file *, struct vm_area_struct *);
446 int (*open) (struct inode *, struct file *);
447 int (*flush) (struct file *);
448 int (*release) (struct inode *, struct file *);
449 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
450 int (*fasync) (int, struct file *, int);
451 int (*lock) (struct file *, int, struct file_lock *);
452 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
454 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
456 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
458 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
460 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
461 unsigned long, unsigned long, unsigned long);
462 int (*check_flags)(int);
463 int (*flock) (struct file *, int, struct file_lock *);
464 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
465 size_t, unsigned int);
466 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
467 size_t, unsigned int);
468 int (*setlease)(struct file *, long, struct file_lock **, void **);
469 long (*fallocate)(struct file *, int, loff_t, loff_t);
475 ->llseek() locking has moved from llseek to the individual llseek
476 implementations. If your fs is not using generic_file_llseek, you
477 need to acquire and release the appropriate locks in your ->llseek().
478 For many filesystems, it is probably safe to acquire the inode
479 mutex or just to use i_size_read() instead.
480 Note: this does not protect the file->f_pos against concurrent modifications
481 since this is something the userspace has to take care about.
483 ->iterate() is called with i_rwsem exclusive.
485 ->iterate_shared() is called with i_rwsem at least shared.
487 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
488 Most instances call fasync_helper(), which does that maintenance, so it's
489 not normally something one needs to worry about. Return values > 0 will be
490 mapped to zero in the VFS layer.
492 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
493 move ->readdir() to inode_operations and use a separate method for directory
494 ->ioctl() or kill the latter completely. One of the problems is that for
495 anything that resembles union-mount we won't have a struct file for all
496 components. And there are other reasons why the current interface is a mess...
498 ->read on directories probably must go away - we should just enforce -EISDIR
499 in sys_read() and friends.
501 ->setlease operations should call generic_setlease() before or after setting
502 the lease within the individual filesystem to record the result of the
505 --------------------------- dquot_operations -------------------------------
507 int (*write_dquot) (struct dquot *);
508 int (*acquire_dquot) (struct dquot *);
509 int (*release_dquot) (struct dquot *);
510 int (*mark_dirty) (struct dquot *);
511 int (*write_info) (struct super_block *, int);
513 These operations are intended to be more or less wrapping functions that ensure
514 a proper locking wrt the filesystem and call the generic quota operations.
516 What filesystem should expect from the generic quota functions:
518 FS recursion Held locks when called
519 write_dquot: yes dqonoff_sem or dqptr_sem
520 acquire_dquot: yes dqonoff_sem or dqptr_sem
521 release_dquot: yes dqonoff_sem or dqptr_sem
523 write_info: yes dqonoff_sem
525 FS recursion means calling ->quota_read() and ->quota_write() from superblock
528 More details about quota locking can be found in fs/dquot.c.
530 --------------------------- vm_operations_struct -----------------------------
532 void (*open)(struct vm_area_struct*);
533 void (*close)(struct vm_area_struct*);
534 vm_fault_t (*fault)(struct vm_area_struct*, struct vm_fault *);
535 vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
536 vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
537 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
540 mmap_sem PageLocked(page)
543 fault: yes can return with page locked
545 page_mkwrite: yes can return with page locked
549 ->fault() is called when a previously not present pte is about
550 to be faulted in. The filesystem must find and return the page associated
551 with the passed in "pgoff" in the vm_fault structure. If it is possible that
552 the page may be truncated and/or invalidated, then the filesystem must lock
553 the page, then ensure it is not already truncated (the page lock will block
554 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
555 locked. The VM will unlock the page.
557 ->map_pages() is called when VM asks to map easy accessible pages.
558 Filesystem should find and map pages associated with offsets from "start_pgoff"
559 till "end_pgoff". ->map_pages() is called with page table locked and must
560 not block. If it's not possible to reach a page without blocking,
561 filesystem should skip it. Filesystem should use do_set_pte() to setup
562 page table entry. Pointer to entry associated with the page is passed in
563 "pte" field in vm_fault structure. Pointers to entries for other offsets
564 should be calculated relative to "pte".
566 ->page_mkwrite() is called when a previously read-only pte is
567 about to become writeable. The filesystem again must ensure that there are
568 no truncate/invalidate races, and then return with the page locked. If
569 the page has been truncated, the filesystem should not look up a new page
570 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
571 will cause the VM to retry the fault.
573 ->pfn_mkwrite() is the same as page_mkwrite but when the pte is
574 VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
575 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
576 after this call is to make the pte read-write, unless pfn_mkwrite returns
579 ->access() is called when get_user_pages() fails in
580 access_process_vm(), typically used to debug a process through
581 /proc/pid/mem or ptrace. This function is needed only for
582 VM_IO | VM_PFNMAP VMAs.
584 ================================================================================
587 (if you break something or notice that it is broken and do not fix it yourself
588 - at least put it here)