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 *, const struct inode *,
16 int (*d_compare)(const struct dentry *, const struct inode *,
17 const struct dentry *, const struct inode *,
18 unsigned int, const char *, const struct qstr *);
19 int (*d_delete)(struct dentry *);
20 void (*d_release)(struct dentry *);
21 void (*d_iput)(struct dentry *, struct inode *);
22 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
23 struct vfsmount *(*d_automount)(struct path *path);
24 int (*d_manage)(struct dentry *, bool);
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
33 d_release: no no yes no
37 d_automount: no no yes no
38 d_manage: no no yes (ref-walk) maybe
40 --------------------------- inode_operations ---------------------------
42 int (*create) (struct inode *,struct dentry *,umode_t, bool);
43 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
44 int (*link) (struct dentry *,struct inode *,struct dentry *);
45 int (*unlink) (struct inode *,struct dentry *);
46 int (*symlink) (struct inode *,struct dentry *,const char *);
47 int (*mkdir) (struct inode *,struct dentry *,umode_t);
48 int (*rmdir) (struct inode *,struct dentry *);
49 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
50 int (*rename) (struct inode *, struct dentry *,
51 struct inode *, struct dentry *);
52 int (*readlink) (struct dentry *, char __user *,int);
53 void * (*follow_link) (struct dentry *, struct nameidata *);
54 void (*put_link) (struct dentry *, struct nameidata *, void *);
55 void (*truncate) (struct inode *);
56 int (*permission) (struct inode *, int, unsigned int);
57 int (*get_acl)(struct inode *, int);
58 int (*setattr) (struct dentry *, struct iattr *);
59 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
60 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
61 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
62 ssize_t (*listxattr) (struct dentry *, char *, size_t);
63 int (*removexattr) (struct dentry *, const char *);
64 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
65 void (*update_time)(struct inode *, struct timespec *, int);
66 int (*atomic_open)(struct inode *, struct dentry *,
67 struct file *, unsigned open_flag,
68 umode_t create_mode, int *opened);
80 rmdir: yes (both) (see below)
81 rename: yes (all) (see below)
86 permission: no (may not block if called in rcu-walk mode)
97 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
99 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
101 See Documentation/filesystems/directory-locking for more detailed discussion
102 of the locking scheme for directory operations.
104 --------------------------- super_operations ---------------------------
106 struct inode *(*alloc_inode)(struct super_block *sb);
107 void (*destroy_inode)(struct inode *);
108 void (*dirty_inode) (struct inode *, int flags);
109 int (*write_inode) (struct inode *, struct writeback_control *wbc);
110 int (*drop_inode) (struct inode *);
111 void (*evict_inode) (struct inode *);
112 void (*put_super) (struct super_block *);
113 int (*sync_fs)(struct super_block *sb, int wait);
114 int (*freeze_fs) (struct super_block *);
115 int (*unfreeze_fs) (struct super_block *);
116 int (*statfs) (struct dentry *, struct kstatfs *);
117 int (*remount_fs) (struct super_block *, int *, char *);
118 void (*umount_begin) (struct super_block *);
119 int (*show_options)(struct seq_file *, struct dentry *);
120 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
121 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
122 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
125 All may block [not true, see below]
131 drop_inode: !!!inode->i_lock!!!
137 statfs: maybe(read) (see below)
140 show_options: no (namespace_sem)
141 quota_read: no (see below)
142 quota_write: no (see below)
143 bdev_try_to_free_page: no (see below)
145 ->statfs() has s_umount (shared) when called by ustat(2) (native or
146 compat), but that's an accident of bad API; s_umount is used to pin
147 the superblock down when we only have dev_t given us by userland to
148 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
149 doesn't hold it when calling ->statfs() - superblock is pinned down
150 by resolving the pathname passed to syscall.
151 ->quota_read() and ->quota_write() functions are both guaranteed to
152 be the only ones operating on the quota file by the quota code (via
153 dqio_sem) (unless an admin really wants to screw up something and
154 writes to quota files with quotas on). For other details about locking
155 see also dquot_operations section.
156 ->bdev_try_to_free_page is called from the ->releasepage handler of
157 the block device inode. See there for more details.
159 --------------------------- file_system_type ---------------------------
161 int (*get_sb) (struct file_system_type *, int,
162 const char *, void *, struct vfsmount *);
163 struct dentry *(*mount) (struct file_system_type *, int,
164 const char *, void *);
165 void (*kill_sb) (struct super_block *);
171 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
173 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
174 unlocks and drops the reference.
176 --------------------------- address_space_operations --------------------------
178 int (*writepage)(struct page *page, struct writeback_control *wbc);
179 int (*readpage)(struct file *, struct page *);
180 int (*sync_page)(struct page *);
181 int (*writepages)(struct address_space *, struct writeback_control *);
182 int (*set_page_dirty)(struct page *page);
183 int (*readpages)(struct file *filp, struct address_space *mapping,
184 struct list_head *pages, unsigned nr_pages);
185 int (*write_begin)(struct file *, struct address_space *mapping,
186 loff_t pos, unsigned len, unsigned flags,
187 struct page **pagep, void **fsdata);
188 int (*write_end)(struct file *, struct address_space *mapping,
189 loff_t pos, unsigned len, unsigned copied,
190 struct page *page, void *fsdata);
191 sector_t (*bmap)(struct address_space *, sector_t);
192 int (*invalidatepage) (struct page *, unsigned long);
193 int (*releasepage) (struct page *, int);
194 void (*freepage)(struct page *);
195 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
196 loff_t offset, unsigned long nr_segs);
197 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
199 int (*migratepage)(struct address_space *, struct page *, struct page *);
200 int (*launder_page)(struct page *);
201 int (*is_partially_uptodate)(struct page *, read_descriptor_t *, unsigned long);
202 int (*error_remove_page)(struct address_space *, struct page *);
203 int (*swap_activate)(struct file *);
204 int (*swap_deactivate)(struct file *);
207 All except set_page_dirty and freepage may block
209 PageLocked(page) i_mutex
210 writepage: yes, unlocks (see below)
211 readpage: yes, unlocks
216 write_begin: locks the page yes
217 write_end: yes, unlocks yes
224 migratepage: yes (both)
226 is_partially_uptodate: yes
227 error_remove_page: yes
231 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
232 may be called from the request handler (/dev/loop).
234 ->readpage() unlocks the page, either synchronously or via I/O
237 ->readpages() populates the pagecache with the passed pages and starts
238 I/O against them. They come unlocked upon I/O completion.
240 ->writepage() is used for two purposes: for "memory cleansing" and for
241 "sync". These are quite different operations and the behaviour may differ
242 depending upon the mode.
244 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
245 it *must* start I/O against the page, even if that would involve
246 blocking on in-progress I/O.
248 If writepage is called for memory cleansing (sync_mode ==
249 WBC_SYNC_NONE) then its role is to get as much writeout underway as
250 possible. So writepage should try to avoid blocking against
251 currently-in-progress I/O.
253 If the filesystem is not called for "sync" and it determines that it
254 would need to block against in-progress I/O to be able to start new I/O
255 against the page the filesystem should redirty the page with
256 redirty_page_for_writepage(), then unlock the page and return zero.
257 This may also be done to avoid internal deadlocks, but rarely.
259 If the filesystem is called for sync then it must wait on any
260 in-progress I/O and then start new I/O.
262 The filesystem should unlock the page synchronously, before returning to the
263 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
264 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
265 currently, and VM should stop calling ->writepage() on this page for some
266 time. VM does this by moving page to the head of the active list, hence the
269 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
270 and return zero, writepage *must* run set_page_writeback() against the page,
271 followed by unlocking it. Once set_page_writeback() has been run against the
272 page, write I/O can be submitted and the write I/O completion handler must run
273 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
274 filesystem must run end_page_writeback() against the page before returning from
277 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
278 if the filesystem needs the page to be locked during writeout, that is ok, too,
279 the page is allowed to be unlocked at any point in time between the calls to
280 set_page_writeback() and end_page_writeback().
282 Note, failure to run either redirty_page_for_writepage() or the combination of
283 set_page_writeback()/end_page_writeback() on a page submitted to writepage
284 will leave the page itself marked clean but it will be tagged as dirty in the
285 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
286 in the filesystem like having dirty inodes at umount and losing written data.
288 ->sync_page() locking rules are not well-defined - usually it is called
289 with lock on page, but that is not guaranteed. Considering the currently
290 existing instances of this method ->sync_page() itself doesn't look
293 ->writepages() is used for periodic writeback and for syscall-initiated
294 sync operations. The address_space should start I/O against at least
295 *nr_to_write pages. *nr_to_write must be decremented for each page which is
296 written. The address_space implementation may write more (or less) pages
297 than *nr_to_write asks for, but it should try to be reasonably close. If
298 nr_to_write is NULL, all dirty pages must be written.
300 writepages should _only_ write pages which are present on
303 ->set_page_dirty() is called from various places in the kernel
304 when the target page is marked as needing writeback. It may be called
305 under spinlock (it cannot block) and is sometimes called with the page
308 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
309 filesystems and by the swapper. The latter will eventually go away. Please,
310 keep it that way and don't breed new callers.
312 ->invalidatepage() is called when the filesystem must attempt to drop
313 some or all of the buffers from the page when it is being truncated. It
314 returns zero on success. If ->invalidatepage is zero, the kernel uses
315 block_invalidatepage() instead.
317 ->releasepage() is called when the kernel is about to try to drop the
318 buffers from the page in preparation for freeing it. It returns zero to
319 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
320 the kernel assumes that the fs has no private interest in the buffers.
322 ->freepage() is called when the kernel is done dropping the page
325 ->launder_page() may be called prior to releasing a page if
326 it is still found to be dirty. It returns zero if the page was successfully
327 cleaned, or an error value if not. Note that in order to prevent the page
328 getting mapped back in and redirtied, it needs to be kept locked
329 across the entire operation.
331 ->swap_activate will be called with a non-zero argument on
332 files backing (non block device backed) swapfiles. A return value
333 of zero indicates success, in which case this file can be used for
334 backing swapspace. The swapspace operations will be proxied to the
335 address space operations.
337 ->swap_deactivate() will be called in the sys_swapoff()
338 path after ->swap_activate() returned success.
340 ----------------------- file_lock_operations ------------------------------
342 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
343 void (*fl_release_private)(struct file_lock *);
347 file_lock_lock may block
349 fl_release_private: maybe no
351 ----------------------- lock_manager_operations ---------------------------
353 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
354 void (*lm_notify)(struct file_lock *); /* unblock callback */
355 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
356 void (*lm_break)(struct file_lock *); /* break_lease callback */
357 int (*lm_change)(struct file_lock **, int);
360 file_lock_lock may block
361 lm_compare_owner: yes no
367 --------------------------- buffer_head -----------------------------------
369 void (*b_end_io)(struct buffer_head *bh, int uptodate);
372 called from interrupts. In other words, extreme care is needed here.
373 bh is locked, but that's all warranties we have here. Currently only RAID1,
374 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
375 call this method upon the IO completion.
377 --------------------------- block_device_operations -----------------------
379 int (*open) (struct block_device *, fmode_t);
380 int (*release) (struct gendisk *, fmode_t);
381 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
382 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
383 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
384 int (*media_changed) (struct gendisk *);
385 void (*unlock_native_capacity) (struct gendisk *);
386 int (*revalidate_disk) (struct gendisk *);
387 int (*getgeo)(struct block_device *, struct hd_geometry *);
388 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
398 unlock_native_capacity: no
401 swap_slot_free_notify: no (see below)
403 media_changed, unlock_native_capacity and revalidate_disk are called only from
406 swap_slot_free_notify is called with swap_lock and sometimes the page lock
410 --------------------------- file_operations -------------------------------
412 loff_t (*llseek) (struct file *, loff_t, int);
413 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
414 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
415 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
416 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
417 int (*readdir) (struct file *, void *, filldir_t);
418 unsigned int (*poll) (struct file *, struct poll_table_struct *);
419 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
420 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
421 int (*mmap) (struct file *, struct vm_area_struct *);
422 int (*open) (struct inode *, struct file *);
423 int (*flush) (struct file *);
424 int (*release) (struct inode *, struct file *);
425 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
426 int (*aio_fsync) (struct kiocb *, int datasync);
427 int (*fasync) (int, struct file *, int);
428 int (*lock) (struct file *, int, struct file_lock *);
429 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
431 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
433 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
435 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
437 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
438 unsigned long, unsigned long, unsigned long);
439 int (*check_flags)(int);
440 int (*flock) (struct file *, int, struct file_lock *);
441 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
442 size_t, unsigned int);
443 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
444 size_t, unsigned int);
445 int (*setlease)(struct file *, long, struct file_lock **);
446 long (*fallocate)(struct file *, int, loff_t, loff_t);
450 All may block except for ->setlease.
451 No VFS locks held on entry except for ->setlease.
453 ->setlease has the file_list_lock held and must not sleep.
455 ->llseek() locking has moved from llseek to the individual llseek
456 implementations. If your fs is not using generic_file_llseek, you
457 need to acquire and release the appropriate locks in your ->llseek().
458 For many filesystems, it is probably safe to acquire the inode
459 mutex or just to use i_size_read() instead.
460 Note: this does not protect the file->f_pos against concurrent modifications
461 since this is something the userspace has to take care about.
463 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
464 Most instances call fasync_helper(), which does that maintenance, so it's
465 not normally something one needs to worry about. Return values > 0 will be
466 mapped to zero in the VFS layer.
468 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
469 move ->readdir() to inode_operations and use a separate method for directory
470 ->ioctl() or kill the latter completely. One of the problems is that for
471 anything that resembles union-mount we won't have a struct file for all
472 components. And there are other reasons why the current interface is a mess...
474 ->read on directories probably must go away - we should just enforce -EISDIR
475 in sys_read() and friends.
477 --------------------------- dquot_operations -------------------------------
479 int (*write_dquot) (struct dquot *);
480 int (*acquire_dquot) (struct dquot *);
481 int (*release_dquot) (struct dquot *);
482 int (*mark_dirty) (struct dquot *);
483 int (*write_info) (struct super_block *, int);
485 These operations are intended to be more or less wrapping functions that ensure
486 a proper locking wrt the filesystem and call the generic quota operations.
488 What filesystem should expect from the generic quota functions:
490 FS recursion Held locks when called
491 write_dquot: yes dqonoff_sem or dqptr_sem
492 acquire_dquot: yes dqonoff_sem or dqptr_sem
493 release_dquot: yes dqonoff_sem or dqptr_sem
495 write_info: yes dqonoff_sem
497 FS recursion means calling ->quota_read() and ->quota_write() from superblock
500 More details about quota locking can be found in fs/dquot.c.
502 --------------------------- vm_operations_struct -----------------------------
504 void (*open)(struct vm_area_struct*);
505 void (*close)(struct vm_area_struct*);
506 int (*fault)(struct vm_area_struct*, struct vm_fault *);
507 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
508 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
511 mmap_sem PageLocked(page)
514 fault: yes can return with page locked
515 page_mkwrite: yes can return with page locked
518 ->fault() is called when a previously not present pte is about
519 to be faulted in. The filesystem must find and return the page associated
520 with the passed in "pgoff" in the vm_fault structure. If it is possible that
521 the page may be truncated and/or invalidated, then the filesystem must lock
522 the page, then ensure it is not already truncated (the page lock will block
523 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
524 locked. The VM will unlock the page.
526 ->page_mkwrite() is called when a previously read-only pte is
527 about to become writeable. The filesystem again must ensure that there are
528 no truncate/invalidate races, and then return with the page locked. If
529 the page has been truncated, the filesystem should not look up a new page
530 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
531 will cause the VM to retry the fault.
533 ->access() is called when get_user_pages() fails in
534 acces_process_vm(), typically used to debug a process through
535 /proc/pid/mem or ptrace. This function is needed only for
536 VM_IO | VM_PFNMAP VMAs.
538 ================================================================================
541 (if you break something or notice that it is broken and do not fix it yourself
542 - at least put it here)