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_hash)(const struct dentry *, const struct inode *,
15 int (*d_compare)(const struct dentry *, const struct inode *,
16 const struct dentry *, const struct inode *,
17 unsigned int, const char *, const struct qstr *);
18 int (*d_delete)(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);
26 rename_lock ->d_lock may block rcu-walk
27 d_revalidate: no no yes (ref-walk) maybe
29 d_compare: yes no no maybe
30 d_delete: no yes no no
31 d_release: no no yes no
35 d_automount: no no yes no
36 d_manage: no no yes (ref-walk) maybe
38 --------------------------- inode_operations ---------------------------
40 int (*create) (struct inode *,struct dentry *,umode_t, bool);
41 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
42 int (*link) (struct dentry *,struct inode *,struct dentry *);
43 int (*unlink) (struct inode *,struct dentry *);
44 int (*symlink) (struct inode *,struct dentry *,const char *);
45 int (*mkdir) (struct inode *,struct dentry *,umode_t);
46 int (*rmdir) (struct inode *,struct dentry *);
47 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
48 int (*rename) (struct inode *, struct dentry *,
49 struct inode *, struct dentry *);
50 int (*readlink) (struct dentry *, char __user *,int);
51 void * (*follow_link) (struct dentry *, struct nameidata *);
52 void (*put_link) (struct dentry *, struct nameidata *, void *);
53 void (*truncate) (struct inode *);
54 int (*permission) (struct inode *, int, unsigned int);
55 int (*get_acl)(struct inode *, int);
56 int (*setattr) (struct dentry *, struct iattr *);
57 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
58 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
59 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
60 ssize_t (*listxattr) (struct dentry *, char *, size_t);
61 int (*removexattr) (struct dentry *, const char *);
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,
66 umode_t create_mode, int *opened);
78 rmdir: yes (both) (see below)
79 rename: yes (all) (see below)
83 truncate: yes (see below)
85 permission: no (may not block if called in rcu-walk mode)
96 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
98 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
99 ->truncate() is never called directly - it's a callback, not a
100 method. It's called by vmtruncate() - deprecated library function used by
101 ->setattr(). Locking information above applies to that call (i.e. is
102 inherited from ->setattr() - vmtruncate() is used when ATTR_SIZE had been
105 See Documentation/filesystems/directory-locking for more detailed discussion
106 of the locking scheme for directory operations.
108 --------------------------- super_operations ---------------------------
110 struct inode *(*alloc_inode)(struct super_block *sb);
111 void (*destroy_inode)(struct inode *);
112 void (*dirty_inode) (struct inode *, int flags);
113 int (*write_inode) (struct inode *, struct writeback_control *wbc);
114 int (*drop_inode) (struct inode *);
115 void (*evict_inode) (struct inode *);
116 void (*put_super) (struct super_block *);
117 void (*write_super) (struct super_block *);
118 int (*sync_fs)(struct super_block *sb, int wait);
119 int (*freeze_fs) (struct super_block *);
120 int (*unfreeze_fs) (struct super_block *);
121 int (*statfs) (struct dentry *, struct kstatfs *);
122 int (*remount_fs) (struct super_block *, int *, char *);
123 void (*umount_begin) (struct super_block *);
124 int (*show_options)(struct seq_file *, struct dentry *);
125 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
126 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
127 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
130 All may block [not true, see below]
136 drop_inode: !!!inode->i_lock!!!
143 statfs: maybe(read) (see below)
146 show_options: no (namespace_sem)
147 quota_read: no (see below)
148 quota_write: no (see below)
149 bdev_try_to_free_page: no (see below)
151 ->statfs() has s_umount (shared) when called by ustat(2) (native or
152 compat), but that's an accident of bad API; s_umount is used to pin
153 the superblock down when we only have dev_t given us by userland to
154 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
155 doesn't hold it when calling ->statfs() - superblock is pinned down
156 by resolving the pathname passed to syscall.
157 ->quota_read() and ->quota_write() functions are both guaranteed to
158 be the only ones operating on the quota file by the quota code (via
159 dqio_sem) (unless an admin really wants to screw up something and
160 writes to quota files with quotas on). For other details about locking
161 see also dquot_operations section.
162 ->bdev_try_to_free_page is called from the ->releasepage handler of
163 the block device inode. See there for more details.
165 --------------------------- file_system_type ---------------------------
167 int (*get_sb) (struct file_system_type *, int,
168 const char *, void *, struct vfsmount *);
169 struct dentry *(*mount) (struct file_system_type *, int,
170 const char *, void *);
171 void (*kill_sb) (struct super_block *);
177 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
179 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
180 unlocks and drops the reference.
182 --------------------------- address_space_operations --------------------------
184 int (*writepage)(struct page *page, struct writeback_control *wbc);
185 int (*readpage)(struct file *, struct page *);
186 int (*sync_page)(struct page *);
187 int (*writepages)(struct address_space *, struct writeback_control *);
188 int (*set_page_dirty)(struct page *page);
189 int (*readpages)(struct file *filp, struct address_space *mapping,
190 struct list_head *pages, unsigned nr_pages);
191 int (*write_begin)(struct file *, struct address_space *mapping,
192 loff_t pos, unsigned len, unsigned flags,
193 struct page **pagep, void **fsdata);
194 int (*write_end)(struct file *, struct address_space *mapping,
195 loff_t pos, unsigned len, unsigned copied,
196 struct page *page, void *fsdata);
197 sector_t (*bmap)(struct address_space *, sector_t);
198 int (*invalidatepage) (struct page *, unsigned long);
199 int (*releasepage) (struct page *, int);
200 void (*freepage)(struct page *);
201 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
202 loff_t offset, unsigned long nr_segs);
203 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
205 int (*migratepage)(struct address_space *, struct page *, struct page *);
206 int (*launder_page)(struct page *);
207 int (*is_partially_uptodate)(struct page *, read_descriptor_t *, unsigned long);
208 int (*error_remove_page)(struct address_space *, struct page *);
211 All except set_page_dirty and freepage may block
213 PageLocked(page) i_mutex
214 writepage: yes, unlocks (see below)
215 readpage: yes, unlocks
220 write_begin: locks the page yes
221 write_end: yes, unlocks yes
228 migratepage: yes (both)
230 is_partially_uptodate: yes
231 error_remove_page: yes
233 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
234 may be called from the request handler (/dev/loop).
236 ->readpage() unlocks the page, either synchronously or via I/O
239 ->readpages() populates the pagecache with the passed pages and starts
240 I/O against them. They come unlocked upon I/O completion.
242 ->writepage() is used for two purposes: for "memory cleansing" and for
243 "sync". These are quite different operations and the behaviour may differ
244 depending upon the mode.
246 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
247 it *must* start I/O against the page, even if that would involve
248 blocking on in-progress I/O.
250 If writepage is called for memory cleansing (sync_mode ==
251 WBC_SYNC_NONE) then its role is to get as much writeout underway as
252 possible. So writepage should try to avoid blocking against
253 currently-in-progress I/O.
255 If the filesystem is not called for "sync" and it determines that it
256 would need to block against in-progress I/O to be able to start new I/O
257 against the page the filesystem should redirty the page with
258 redirty_page_for_writepage(), then unlock the page and return zero.
259 This may also be done to avoid internal deadlocks, but rarely.
261 If the filesystem is called for sync then it must wait on any
262 in-progress I/O and then start new I/O.
264 The filesystem should unlock the page synchronously, before returning to the
265 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
266 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
267 currently, and VM should stop calling ->writepage() on this page for some
268 time. VM does this by moving page to the head of the active list, hence the
271 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
272 and return zero, writepage *must* run set_page_writeback() against the page,
273 followed by unlocking it. Once set_page_writeback() has been run against the
274 page, write I/O can be submitted and the write I/O completion handler must run
275 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
276 filesystem must run end_page_writeback() against the page before returning from
279 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
280 if the filesystem needs the page to be locked during writeout, that is ok, too,
281 the page is allowed to be unlocked at any point in time between the calls to
282 set_page_writeback() and end_page_writeback().
284 Note, failure to run either redirty_page_for_writepage() or the combination of
285 set_page_writeback()/end_page_writeback() on a page submitted to writepage
286 will leave the page itself marked clean but it will be tagged as dirty in the
287 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
288 in the filesystem like having dirty inodes at umount and losing written data.
290 ->sync_page() locking rules are not well-defined - usually it is called
291 with lock on page, but that is not guaranteed. Considering the currently
292 existing instances of this method ->sync_page() itself doesn't look
295 ->writepages() is used for periodic writeback and for syscall-initiated
296 sync operations. The address_space should start I/O against at least
297 *nr_to_write pages. *nr_to_write must be decremented for each page which is
298 written. The address_space implementation may write more (or less) pages
299 than *nr_to_write asks for, but it should try to be reasonably close. If
300 nr_to_write is NULL, all dirty pages must be written.
302 writepages should _only_ write pages which are present on
305 ->set_page_dirty() is called from various places in the kernel
306 when the target page is marked as needing writeback. It may be called
307 under spinlock (it cannot block) and is sometimes called with the page
310 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
311 filesystems and by the swapper. The latter will eventually go away. Please,
312 keep it that way and don't breed new callers.
314 ->invalidatepage() is called when the filesystem must attempt to drop
315 some or all of the buffers from the page when it is being truncated. It
316 returns zero on success. If ->invalidatepage is zero, the kernel uses
317 block_invalidatepage() instead.
319 ->releasepage() is called when the kernel is about to try to drop the
320 buffers from the page in preparation for freeing it. It returns zero to
321 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
322 the kernel assumes that the fs has no private interest in the buffers.
324 ->freepage() is called when the kernel is done dropping the page
327 ->launder_page() may be called prior to releasing a page if
328 it is still found to be dirty. It returns zero if the page was successfully
329 cleaned, or an error value if not. Note that in order to prevent the page
330 getting mapped back in and redirtied, it needs to be kept locked
331 across the entire operation.
333 ----------------------- file_lock_operations ------------------------------
335 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
336 void (*fl_release_private)(struct file_lock *);
340 file_lock_lock may block
342 fl_release_private: maybe no
344 ----------------------- lock_manager_operations ---------------------------
346 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
347 void (*lm_notify)(struct file_lock *); /* unblock callback */
348 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
349 void (*lm_release_private)(struct file_lock *);
350 void (*lm_break)(struct file_lock *); /* break_lease callback */
351 int (*lm_change)(struct file_lock **, int);
354 file_lock_lock may block
355 lm_compare_owner: yes no
358 lm_release_private: maybe no
362 --------------------------- buffer_head -----------------------------------
364 void (*b_end_io)(struct buffer_head *bh, int uptodate);
367 called from interrupts. In other words, extreme care is needed here.
368 bh is locked, but that's all warranties we have here. Currently only RAID1,
369 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
370 call this method upon the IO completion.
372 --------------------------- block_device_operations -----------------------
374 int (*open) (struct block_device *, fmode_t);
375 int (*release) (struct gendisk *, fmode_t);
376 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
377 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
378 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
379 int (*media_changed) (struct gendisk *);
380 void (*unlock_native_capacity) (struct gendisk *);
381 int (*revalidate_disk) (struct gendisk *);
382 int (*getgeo)(struct block_device *, struct hd_geometry *);
383 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
393 unlock_native_capacity: no
396 swap_slot_free_notify: no (see below)
398 media_changed, unlock_native_capacity and revalidate_disk are called only from
401 swap_slot_free_notify is called with swap_lock and sometimes the page lock
405 --------------------------- file_operations -------------------------------
407 loff_t (*llseek) (struct file *, loff_t, int);
408 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
409 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
410 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
411 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
412 int (*readdir) (struct file *, void *, filldir_t);
413 unsigned int (*poll) (struct file *, struct poll_table_struct *);
414 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
415 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
416 int (*mmap) (struct file *, struct vm_area_struct *);
417 int (*open) (struct inode *, struct file *);
418 int (*flush) (struct file *);
419 int (*release) (struct inode *, struct file *);
420 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
421 int (*aio_fsync) (struct kiocb *, int datasync);
422 int (*fasync) (int, struct file *, int);
423 int (*lock) (struct file *, int, struct file_lock *);
424 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
426 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
428 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
430 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
432 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
433 unsigned long, unsigned long, unsigned long);
434 int (*check_flags)(int);
435 int (*flock) (struct file *, int, struct file_lock *);
436 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
437 size_t, unsigned int);
438 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
439 size_t, unsigned int);
440 int (*setlease)(struct file *, long, struct file_lock **);
441 long (*fallocate)(struct file *, int, loff_t, loff_t);
445 All may block except for ->setlease.
446 No VFS locks held on entry except for ->setlease.
448 ->setlease has the file_list_lock held and must not sleep.
450 ->llseek() locking has moved from llseek to the individual llseek
451 implementations. If your fs is not using generic_file_llseek, you
452 need to acquire and release the appropriate locks in your ->llseek().
453 For many filesystems, it is probably safe to acquire the inode
454 mutex or just to use i_size_read() instead.
455 Note: this does not protect the file->f_pos against concurrent modifications
456 since this is something the userspace has to take care about.
458 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
459 Most instances call fasync_helper(), which does that maintenance, so it's
460 not normally something one needs to worry about. Return values > 0 will be
461 mapped to zero in the VFS layer.
463 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
464 move ->readdir() to inode_operations and use a separate method for directory
465 ->ioctl() or kill the latter completely. One of the problems is that for
466 anything that resembles union-mount we won't have a struct file for all
467 components. And there are other reasons why the current interface is a mess...
469 ->read on directories probably must go away - we should just enforce -EISDIR
470 in sys_read() and friends.
472 --------------------------- dquot_operations -------------------------------
474 int (*write_dquot) (struct dquot *);
475 int (*acquire_dquot) (struct dquot *);
476 int (*release_dquot) (struct dquot *);
477 int (*mark_dirty) (struct dquot *);
478 int (*write_info) (struct super_block *, int);
480 These operations are intended to be more or less wrapping functions that ensure
481 a proper locking wrt the filesystem and call the generic quota operations.
483 What filesystem should expect from the generic quota functions:
485 FS recursion Held locks when called
486 write_dquot: yes dqonoff_sem or dqptr_sem
487 acquire_dquot: yes dqonoff_sem or dqptr_sem
488 release_dquot: yes dqonoff_sem or dqptr_sem
490 write_info: yes dqonoff_sem
492 FS recursion means calling ->quota_read() and ->quota_write() from superblock
495 More details about quota locking can be found in fs/dquot.c.
497 --------------------------- vm_operations_struct -----------------------------
499 void (*open)(struct vm_area_struct*);
500 void (*close)(struct vm_area_struct*);
501 int (*fault)(struct vm_area_struct*, struct vm_fault *);
502 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
503 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
506 mmap_sem PageLocked(page)
509 fault: yes can return with page locked
510 page_mkwrite: yes can return with page locked
513 ->fault() is called when a previously not present pte is about
514 to be faulted in. The filesystem must find and return the page associated
515 with the passed in "pgoff" in the vm_fault structure. If it is possible that
516 the page may be truncated and/or invalidated, then the filesystem must lock
517 the page, then ensure it is not already truncated (the page lock will block
518 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
519 locked. The VM will unlock the page.
521 ->page_mkwrite() is called when a previously read-only pte is
522 about to become writeable. The filesystem again must ensure that there are
523 no truncate/invalidate races, and then return with the page locked. If
524 the page has been truncated, the filesystem should not look up a new page
525 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
526 will cause the VM to retry the fault.
528 ->access() is called when get_user_pages() fails in
529 acces_process_vm(), typically used to debug a process through
530 /proc/pid/mem or ptrace. This function is needed only for
531 VM_IO | VM_PFNMAP VMAs.
533 ================================================================================
536 (if you break something or notice that it is broken and do not fix it yourself
537 - at least put it here)