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)(struct dentry *, bool);
24 struct dentry *(*d_real)(struct dentry *, const struct inode *,
28 rename_lock ->d_lock may block rcu-walk
29 d_revalidate: no no yes (ref-walk) maybe
30 d_weak_revalidate:no no yes no
32 d_compare: yes no no maybe
33 d_delete: no yes no no
35 d_release: no no yes no
39 d_automount: no no yes no
40 d_manage: no no yes (ref-walk) maybe
43 --------------------------- inode_operations ---------------------------
45 int (*create) (struct inode *,struct dentry *,umode_t, bool);
46 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
47 int (*link) (struct dentry *,struct inode *,struct dentry *);
48 int (*unlink) (struct inode *,struct dentry *);
49 int (*symlink) (struct inode *,struct dentry *,const char *);
50 int (*mkdir) (struct inode *,struct dentry *,umode_t);
51 int (*rmdir) (struct inode *,struct dentry *);
52 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
53 int (*rename) (struct inode *, struct dentry *,
54 struct inode *, struct dentry *);
55 int (*rename2) (struct inode *, struct dentry *,
56 struct inode *, struct dentry *, unsigned int);
57 int (*readlink) (struct dentry *, char __user *,int);
58 const char *(*get_link) (struct dentry *, struct inode *, void **);
59 void (*truncate) (struct inode *);
60 int (*permission) (struct inode *, int, unsigned int);
61 int (*get_acl)(struct inode *, int);
62 int (*setattr) (struct dentry *, struct iattr *);
63 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
64 ssize_t (*listxattr) (struct dentry *, char *, size_t);
65 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
66 void (*update_time)(struct inode *, struct timespec *, int);
67 int (*atomic_open)(struct inode *, struct dentry *,
68 struct file *, unsigned open_flag,
69 umode_t create_mode, int *opened);
70 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
82 rmdir: yes (both) (see below)
83 rename: yes (all) (see below)
84 rename2: yes (all) (see below)
88 permission: no (may not block if called in rcu-walk mode)
98 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
100 cross-directory ->rename() and rename2() has (per-superblock)
103 See Documentation/filesystems/directory-locking for more detailed discussion
104 of the locking scheme for directory operations.
106 ----------------------- xattr_handler operations -----------------------
108 bool (*list)(struct dentry *dentry);
109 int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
110 struct inode *inode, const char *name, void *buffer,
112 int (*set)(const struct xattr_handler *handler, struct dentry *dentry,
113 struct inode *inode, const char *name, const void *buffer,
114 size_t size, int flags);
123 --------------------------- super_operations ---------------------------
125 struct inode *(*alloc_inode)(struct super_block *sb);
126 void (*destroy_inode)(struct inode *);
127 void (*dirty_inode) (struct inode *, int flags);
128 int (*write_inode) (struct inode *, struct writeback_control *wbc);
129 int (*drop_inode) (struct inode *);
130 void (*evict_inode) (struct inode *);
131 void (*put_super) (struct super_block *);
132 int (*sync_fs)(struct super_block *sb, int wait);
133 int (*freeze_fs) (struct super_block *);
134 int (*unfreeze_fs) (struct super_block *);
135 int (*statfs) (struct dentry *, struct kstatfs *);
136 int (*remount_fs) (struct super_block *, int *, char *);
137 void (*umount_begin) (struct super_block *);
138 int (*show_options)(struct seq_file *, struct dentry *);
139 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
140 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
141 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
144 All may block [not true, see below]
150 drop_inode: !!!inode->i_lock!!!
156 statfs: maybe(read) (see below)
159 show_options: no (namespace_sem)
160 quota_read: no (see below)
161 quota_write: no (see below)
162 bdev_try_to_free_page: no (see below)
164 ->statfs() has s_umount (shared) when called by ustat(2) (native or
165 compat), but that's an accident of bad API; s_umount is used to pin
166 the superblock down when we only have dev_t given us by userland to
167 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
168 doesn't hold it when calling ->statfs() - superblock is pinned down
169 by resolving the pathname passed to syscall.
170 ->quota_read() and ->quota_write() functions are both guaranteed to
171 be the only ones operating on the quota file by the quota code (via
172 dqio_sem) (unless an admin really wants to screw up something and
173 writes to quota files with quotas on). For other details about locking
174 see also dquot_operations section.
175 ->bdev_try_to_free_page is called from the ->releasepage handler of
176 the block device inode. See there for more details.
178 --------------------------- file_system_type ---------------------------
180 struct dentry *(*mount) (struct file_system_type *, int,
181 const char *, void *);
182 void (*kill_sb) (struct super_block *);
188 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
190 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
191 unlocks and drops the reference.
193 --------------------------- address_space_operations --------------------------
195 int (*writepage)(struct page *page, struct writeback_control *wbc);
196 int (*readpage)(struct file *, struct page *);
197 int (*writepages)(struct address_space *, struct writeback_control *);
198 int (*set_page_dirty)(struct page *page);
199 int (*readpages)(struct file *filp, struct address_space *mapping,
200 struct list_head *pages, unsigned nr_pages);
201 int (*write_begin)(struct file *, struct address_space *mapping,
202 loff_t pos, unsigned len, unsigned flags,
203 struct page **pagep, void **fsdata);
204 int (*write_end)(struct file *, struct address_space *mapping,
205 loff_t pos, unsigned len, unsigned copied,
206 struct page *page, void *fsdata);
207 sector_t (*bmap)(struct address_space *, sector_t);
208 void (*invalidatepage) (struct page *, unsigned int, unsigned int);
209 int (*releasepage) (struct page *, int);
210 void (*freepage)(struct page *);
211 int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
212 bool (*isolate_page) (struct page *, isolate_mode_t);
213 int (*migratepage)(struct address_space *, struct page *, struct page *);
214 void (*putback_page) (struct page *);
215 int (*launder_page)(struct page *);
216 int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long);
217 int (*error_remove_page)(struct address_space *, struct page *);
218 int (*swap_activate)(struct file *);
219 int (*swap_deactivate)(struct file *);
222 All except set_page_dirty and freepage may block
224 PageLocked(page) i_mutex
225 writepage: yes, unlocks (see below)
226 readpage: yes, unlocks
230 write_begin: locks the page yes
231 write_end: yes, unlocks yes
238 migratepage: yes (both)
241 is_partially_uptodate: yes
242 error_remove_page: yes
246 ->write_begin(), ->write_end() and ->readpage() may be called from
247 the request handler (/dev/loop).
249 ->readpage() unlocks the page, either synchronously or via I/O
252 ->readpages() populates the pagecache with the passed pages and starts
253 I/O against them. They come unlocked upon I/O completion.
255 ->writepage() is used for two purposes: for "memory cleansing" and for
256 "sync". These are quite different operations and the behaviour may differ
257 depending upon the mode.
259 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
260 it *must* start I/O against the page, even if that would involve
261 blocking on in-progress I/O.
263 If writepage is called for memory cleansing (sync_mode ==
264 WBC_SYNC_NONE) then its role is to get as much writeout underway as
265 possible. So writepage should try to avoid blocking against
266 currently-in-progress I/O.
268 If the filesystem is not called for "sync" and it determines that it
269 would need to block against in-progress I/O to be able to start new I/O
270 against the page the filesystem should redirty the page with
271 redirty_page_for_writepage(), then unlock the page and return zero.
272 This may also be done to avoid internal deadlocks, but rarely.
274 If the filesystem is called for sync then it must wait on any
275 in-progress I/O and then start new I/O.
277 The filesystem should unlock the page synchronously, before returning to the
278 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
279 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
280 currently, and VM should stop calling ->writepage() on this page for some
281 time. VM does this by moving page to the head of the active list, hence the
284 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
285 and return zero, writepage *must* run set_page_writeback() against the page,
286 followed by unlocking it. Once set_page_writeback() has been run against the
287 page, write I/O can be submitted and the write I/O completion handler must run
288 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
289 filesystem must run end_page_writeback() against the page before returning from
292 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
293 if the filesystem needs the page to be locked during writeout, that is ok, too,
294 the page is allowed to be unlocked at any point in time between the calls to
295 set_page_writeback() and end_page_writeback().
297 Note, failure to run either redirty_page_for_writepage() or the combination of
298 set_page_writeback()/end_page_writeback() on a page submitted to writepage
299 will leave the page itself marked clean but it will be tagged as dirty in the
300 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
301 in the filesystem like having dirty inodes at umount and losing written data.
303 ->writepages() is used for periodic writeback and for syscall-initiated
304 sync operations. The address_space should start I/O against at least
305 *nr_to_write pages. *nr_to_write must be decremented for each page which is
306 written. The address_space implementation may write more (or less) pages
307 than *nr_to_write asks for, but it should try to be reasonably close. If
308 nr_to_write is NULL, all dirty pages must be written.
310 writepages should _only_ write pages which are present on
313 ->set_page_dirty() is called from various places in the kernel
314 when the target page is marked as needing writeback. It may be called
315 under spinlock (it cannot block) and is sometimes called with the page
318 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
319 filesystems and by the swapper. The latter will eventually go away. Please,
320 keep it that way and don't breed new callers.
322 ->invalidatepage() is called when the filesystem must attempt to drop
323 some or all of the buffers from the page when it is being truncated. It
324 returns zero on success. If ->invalidatepage is zero, the kernel uses
325 block_invalidatepage() instead.
327 ->releasepage() is called when the kernel is about to try to drop the
328 buffers from the page in preparation for freeing it. It returns zero to
329 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
330 the kernel assumes that the fs has no private interest in the buffers.
332 ->freepage() is called when the kernel is done dropping the page
335 ->launder_page() may be called prior to releasing a page if
336 it is still found to be dirty. It returns zero if the page was successfully
337 cleaned, or an error value if not. Note that in order to prevent the page
338 getting mapped back in and redirtied, it needs to be kept locked
339 across the entire operation.
341 ->swap_activate will be called with a non-zero argument on
342 files backing (non block device backed) swapfiles. A return value
343 of zero indicates success, in which case this file can be used for
344 backing swapspace. The swapspace operations will be proxied to the
345 address space operations.
347 ->swap_deactivate() will be called in the sys_swapoff()
348 path after ->swap_activate() returned success.
350 ----------------------- file_lock_operations ------------------------------
352 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
353 void (*fl_release_private)(struct file_lock *);
357 inode->i_lock may block
359 fl_release_private: maybe maybe[1]
361 [1]: ->fl_release_private for flock or POSIX locks is currently allowed
362 to block. Leases however can still be freed while the i_lock is held and
363 so fl_release_private called on a lease should not block.
365 ----------------------- lock_manager_operations ---------------------------
367 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
368 unsigned long (*lm_owner_key)(struct file_lock *);
369 void (*lm_notify)(struct file_lock *); /* unblock callback */
370 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
371 void (*lm_break)(struct file_lock *); /* break_lease callback */
372 int (*lm_change)(struct file_lock **, int);
376 inode->i_lock blocked_lock_lock may block
377 lm_compare_owner: yes[1] maybe no
378 lm_owner_key yes[1] yes no
379 lm_notify: yes yes no
384 [1]: ->lm_compare_owner and ->lm_owner_key are generally called with
385 *an* inode->i_lock held. It may not be the i_lock of the inode
386 associated with either file_lock argument! This is the case with deadlock
387 detection, since the code has to chase down the owners of locks that may
388 be entirely unrelated to the one on which the lock is being acquired.
389 For deadlock detection however, the blocked_lock_lock is also held. The
390 fact that these locks are held ensures that the file_locks do not
391 disappear out from under you while doing the comparison or generating an
394 --------------------------- buffer_head -----------------------------------
396 void (*b_end_io)(struct buffer_head *bh, int uptodate);
399 called from interrupts. In other words, extreme care is needed here.
400 bh is locked, but that's all warranties we have here. Currently only RAID1,
401 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
402 call this method upon the IO completion.
404 --------------------------- block_device_operations -----------------------
406 int (*open) (struct block_device *, fmode_t);
407 int (*release) (struct gendisk *, fmode_t);
408 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
409 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
410 int (*direct_access) (struct block_device *, sector_t, void **,
412 int (*media_changed) (struct gendisk *);
413 void (*unlock_native_capacity) (struct gendisk *);
414 int (*revalidate_disk) (struct gendisk *);
415 int (*getgeo)(struct block_device *, struct hd_geometry *);
416 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
426 unlock_native_capacity: no
429 swap_slot_free_notify: no (see below)
431 media_changed, unlock_native_capacity and revalidate_disk are called only from
434 swap_slot_free_notify is called with swap_lock and sometimes the page lock
438 --------------------------- file_operations -------------------------------
440 loff_t (*llseek) (struct file *, loff_t, int);
441 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
442 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
443 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
444 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
445 int (*iterate) (struct file *, struct dir_context *);
446 unsigned int (*poll) (struct file *, struct poll_table_struct *);
447 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
448 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
449 int (*mmap) (struct file *, struct vm_area_struct *);
450 int (*open) (struct inode *, struct file *);
451 int (*flush) (struct file *);
452 int (*release) (struct inode *, struct file *);
453 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
454 int (*aio_fsync) (struct kiocb *, int datasync);
455 int (*fasync) (int, struct file *, int);
456 int (*lock) (struct file *, int, struct file_lock *);
457 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
459 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
461 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
463 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
465 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
466 unsigned long, unsigned long, unsigned long);
467 int (*check_flags)(int);
468 int (*flock) (struct file *, int, struct file_lock *);
469 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
470 size_t, unsigned int);
471 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
472 size_t, unsigned int);
473 int (*setlease)(struct file *, long, struct file_lock **, void **);
474 long (*fallocate)(struct file *, int, loff_t, loff_t);
480 ->llseek() locking has moved from llseek to the individual llseek
481 implementations. If your fs is not using generic_file_llseek, you
482 need to acquire and release the appropriate locks in your ->llseek().
483 For many filesystems, it is probably safe to acquire the inode
484 mutex or just to use i_size_read() instead.
485 Note: this does not protect the file->f_pos against concurrent modifications
486 since this is something the userspace has to take care about.
488 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
489 Most instances call fasync_helper(), which does that maintenance, so it's
490 not normally something one needs to worry about. Return values > 0 will be
491 mapped to zero in the VFS layer.
493 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
494 move ->readdir() to inode_operations and use a separate method for directory
495 ->ioctl() or kill the latter completely. One of the problems is that for
496 anything that resembles union-mount we won't have a struct file for all
497 components. And there are other reasons why the current interface is a mess...
499 ->read on directories probably must go away - we should just enforce -EISDIR
500 in sys_read() and friends.
502 ->setlease operations should call generic_setlease() before or after setting
503 the lease within the individual filesystem to record the result of the
506 --------------------------- dquot_operations -------------------------------
508 int (*write_dquot) (struct dquot *);
509 int (*acquire_dquot) (struct dquot *);
510 int (*release_dquot) (struct dquot *);
511 int (*mark_dirty) (struct dquot *);
512 int (*write_info) (struct super_block *, int);
514 These operations are intended to be more or less wrapping functions that ensure
515 a proper locking wrt the filesystem and call the generic quota operations.
517 What filesystem should expect from the generic quota functions:
519 FS recursion Held locks when called
520 write_dquot: yes dqonoff_sem or dqptr_sem
521 acquire_dquot: yes dqonoff_sem or dqptr_sem
522 release_dquot: yes dqonoff_sem or dqptr_sem
524 write_info: yes dqonoff_sem
526 FS recursion means calling ->quota_read() and ->quota_write() from superblock
529 More details about quota locking can be found in fs/dquot.c.
531 --------------------------- vm_operations_struct -----------------------------
533 void (*open)(struct vm_area_struct*);
534 void (*close)(struct vm_area_struct*);
535 int (*fault)(struct vm_area_struct*, struct vm_fault *);
536 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
537 int (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
538 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
541 mmap_sem PageLocked(page)
544 fault: yes can return with page locked
546 page_mkwrite: yes can return with page locked
550 ->fault() is called when a previously not present pte is about
551 to be faulted in. The filesystem must find and return the page associated
552 with the passed in "pgoff" in the vm_fault structure. If it is possible that
553 the page may be truncated and/or invalidated, then the filesystem must lock
554 the page, then ensure it is not already truncated (the page lock will block
555 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
556 locked. The VM will unlock the page.
558 ->map_pages() is called when VM asks to map easy accessible pages.
559 Filesystem should find and map pages associated with offsets from "start_pgoff"
560 till "end_pgoff". ->map_pages() is called with page table locked and must
561 not block. If it's not possible to reach a page without blocking,
562 filesystem should skip it. Filesystem should use do_set_pte() to setup
563 page table entry. Pointer to entry associated with the page is passed in
564 "pte" field in fault_env structure. Pointers to entries for other offsets
565 should be calculated relative to "pte".
567 ->page_mkwrite() is called when a previously read-only pte is
568 about to become writeable. The filesystem again must ensure that there are
569 no truncate/invalidate races, and then return with the page locked. If
570 the page has been truncated, the filesystem should not look up a new page
571 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
572 will cause the VM to retry the fault.
574 ->pfn_mkwrite() is the same as page_mkwrite but when the pte is
575 VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
576 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
577 after this call is to make the pte read-write, unless pfn_mkwrite returns
580 ->access() is called when get_user_pages() fails in
581 access_process_vm(), typically used to debug a process through
582 /proc/pid/mem or ptrace. This function is needed only for
583 VM_IO | VM_PFNMAP VMAs.
585 ================================================================================
588 (if you break something or notice that it is broken and do not fix it yourself
589 - at least put it here)