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 *, struct nameidata *);
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
34 d_automount: no no yes no
35 d_manage: no no yes (ref-walk) maybe
37 --------------------------- inode_operations ---------------------------
39 int (*create) (struct inode *,struct dentry *,int, struct nameidata *);
40 struct dentry * (*lookup) (struct inode *,struct dentry *, struct nameid
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 *,int);
46 int (*rmdir) (struct inode *,struct dentry *);
47 int (*mknod) (struct inode *,struct dentry *,int,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 (*check_acl)(struct inode *, int, unsigned 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 void (*truncate_range)(struct inode *, loff_t, loff_t);
63 long (*fallocate)(struct inode *inode, int mode, loff_t offset, loff_t len);
64 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
76 rmdir: yes (both) (see below)
77 rename: yes (all) (see below)
81 truncate: yes (see below)
83 permission: no (may not block if called in rcu-walk mode)
93 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
95 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
96 ->truncate() is never called directly - it's a callback, not a
97 method. It's called by vmtruncate() - deprecated library function used by
98 ->setattr(). Locking information above applies to that call (i.e. is
99 inherited from ->setattr() - vmtruncate() is used when ATTR_SIZE had been
102 See Documentation/filesystems/directory-locking for more detailed discussion
103 of the locking scheme for directory operations.
105 --------------------------- super_operations ---------------------------
107 struct inode *(*alloc_inode)(struct super_block *sb);
108 void (*destroy_inode)(struct inode *);
109 void (*dirty_inode) (struct inode *);
110 int (*write_inode) (struct inode *, struct writeback_control *wbc);
111 int (*drop_inode) (struct inode *);
112 void (*evict_inode) (struct inode *);
113 void (*put_super) (struct super_block *);
114 void (*write_super) (struct super_block *);
115 int (*sync_fs)(struct super_block *sb, int wait);
116 int (*freeze_fs) (struct super_block *);
117 int (*unfreeze_fs) (struct super_block *);
118 int (*statfs) (struct dentry *, struct kstatfs *);
119 int (*remount_fs) (struct super_block *, int *, char *);
120 void (*umount_begin) (struct super_block *);
121 int (*show_options)(struct seq_file *, struct vfsmount *);
122 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
123 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
124 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
127 All may block [not true, see below]
131 dirty_inode: (must not sleep)
133 drop_inode: !!!inode_lock!!!
140 statfs: maybe(read) (see below)
143 show_options: no (namespace_sem)
144 quota_read: no (see below)
145 quota_write: no (see below)
146 bdev_try_to_free_page: no (see below)
148 ->statfs() has s_umount (shared) when called by ustat(2) (native or
149 compat), but that's an accident of bad API; s_umount is used to pin
150 the superblock down when we only have dev_t given us by userland to
151 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
152 doesn't hold it when calling ->statfs() - superblock is pinned down
153 by resolving the pathname passed to syscall.
154 ->quota_read() and ->quota_write() functions are both guaranteed to
155 be the only ones operating on the quota file by the quota code (via
156 dqio_sem) (unless an admin really wants to screw up something and
157 writes to quota files with quotas on). For other details about locking
158 see also dquot_operations section.
159 ->bdev_try_to_free_page is called from the ->releasepage handler of
160 the block device inode. See there for more details.
162 --------------------------- file_system_type ---------------------------
164 int (*get_sb) (struct file_system_type *, int,
165 const char *, void *, struct vfsmount *);
166 struct dentry *(*mount) (struct file_system_type *, int,
167 const char *, void *);
168 void (*kill_sb) (struct super_block *);
175 ->get_sb() returns error or 0 with locked superblock attached to the vfsmount
176 (exclusive on ->s_umount).
177 ->mount() returns ERR_PTR or the root dentry.
178 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
179 unlocks and drops the reference.
181 --------------------------- address_space_operations --------------------------
183 int (*writepage)(struct page *page, struct writeback_control *wbc);
184 int (*readpage)(struct file *, struct page *);
185 int (*sync_page)(struct page *);
186 int (*writepages)(struct address_space *, struct writeback_control *);
187 int (*set_page_dirty)(struct page *page);
188 int (*readpages)(struct file *filp, struct address_space *mapping,
189 struct list_head *pages, unsigned nr_pages);
190 int (*write_begin)(struct file *, struct address_space *mapping,
191 loff_t pos, unsigned len, unsigned flags,
192 struct page **pagep, void **fsdata);
193 int (*write_end)(struct file *, struct address_space *mapping,
194 loff_t pos, unsigned len, unsigned copied,
195 struct page *page, void *fsdata);
196 sector_t (*bmap)(struct address_space *, sector_t);
197 int (*invalidatepage) (struct page *, unsigned long);
198 int (*releasepage) (struct page *, int);
199 void (*freepage)(struct page *);
200 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
201 loff_t offset, unsigned long nr_segs);
202 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
204 int (*migratepage)(struct address_space *, struct page *, struct page *);
205 int (*launder_page)(struct page *);
206 int (*is_partially_uptodate)(struct page *, read_descriptor_t *, unsigned long);
207 int (*error_remove_page)(struct address_space *, struct page *);
210 All except set_page_dirty and freepage may block
212 PageLocked(page) i_mutex
213 writepage: yes, unlocks (see below)
214 readpage: yes, unlocks
219 write_begin: locks the page yes
220 write_end: yes, unlocks yes
227 migratepage: yes (both)
229 is_partially_uptodate: yes
230 error_remove_page: yes
232 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
233 may be called from the request handler (/dev/loop).
235 ->readpage() unlocks the page, either synchronously or via I/O
238 ->readpages() populates the pagecache with the passed pages and starts
239 I/O against them. They come unlocked upon I/O completion.
241 ->writepage() is used for two purposes: for "memory cleansing" and for
242 "sync". These are quite different operations and the behaviour may differ
243 depending upon the mode.
245 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
246 it *must* start I/O against the page, even if that would involve
247 blocking on in-progress I/O.
249 If writepage is called for memory cleansing (sync_mode ==
250 WBC_SYNC_NONE) then its role is to get as much writeout underway as
251 possible. So writepage should try to avoid blocking against
252 currently-in-progress I/O.
254 If the filesystem is not called for "sync" and it determines that it
255 would need to block against in-progress I/O to be able to start new I/O
256 against the page the filesystem should redirty the page with
257 redirty_page_for_writepage(), then unlock the page and return zero.
258 This may also be done to avoid internal deadlocks, but rarely.
260 If the filesystem is called for sync then it must wait on any
261 in-progress I/O and then start new I/O.
263 The filesystem should unlock the page synchronously, before returning to the
264 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
265 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
266 currently, and VM should stop calling ->writepage() on this page for some
267 time. VM does this by moving page to the head of the active list, hence the
270 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
271 and return zero, writepage *must* run set_page_writeback() against the page,
272 followed by unlocking it. Once set_page_writeback() has been run against the
273 page, write I/O can be submitted and the write I/O completion handler must run
274 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
275 filesystem must run end_page_writeback() against the page before returning from
278 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
279 if the filesystem needs the page to be locked during writeout, that is ok, too,
280 the page is allowed to be unlocked at any point in time between the calls to
281 set_page_writeback() and end_page_writeback().
283 Note, failure to run either redirty_page_for_writepage() or the combination of
284 set_page_writeback()/end_page_writeback() on a page submitted to writepage
285 will leave the page itself marked clean but it will be tagged as dirty in the
286 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
287 in the filesystem like having dirty inodes at umount and losing written data.
289 ->sync_page() locking rules are not well-defined - usually it is called
290 with lock on page, but that is not guaranteed. Considering the currently
291 existing instances of this method ->sync_page() itself doesn't look
294 ->writepages() is used for periodic writeback and for syscall-initiated
295 sync operations. The address_space should start I/O against at least
296 *nr_to_write pages. *nr_to_write must be decremented for each page which is
297 written. The address_space implementation may write more (or less) pages
298 than *nr_to_write asks for, but it should try to be reasonably close. If
299 nr_to_write is NULL, all dirty pages must be written.
301 writepages should _only_ write pages which are present on
304 ->set_page_dirty() is called from various places in the kernel
305 when the target page is marked as needing writeback. It may be called
306 under spinlock (it cannot block) and is sometimes called with the page
309 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
310 filesystems and by the swapper. The latter will eventually go away. Please,
311 keep it that way and don't breed new callers.
313 ->invalidatepage() is called when the filesystem must attempt to drop
314 some or all of the buffers from the page when it is being truncated. It
315 returns zero on success. If ->invalidatepage is zero, the kernel uses
316 block_invalidatepage() instead.
318 ->releasepage() is called when the kernel is about to try to drop the
319 buffers from the page in preparation for freeing it. It returns zero to
320 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
321 the kernel assumes that the fs has no private interest in the buffers.
323 ->freepage() is called when the kernel is done dropping the page
326 ->launder_page() may be called prior to releasing a page if
327 it is still found to be dirty. It returns zero if the page was successfully
328 cleaned, or an error value if not. Note that in order to prevent the page
329 getting mapped back in and redirtied, it needs to be kept locked
330 across the entire operation.
332 ----------------------- file_lock_operations ------------------------------
334 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
335 void (*fl_release_private)(struct file_lock *);
339 file_lock_lock may block
341 fl_release_private: maybe no
343 ----------------------- lock_manager_operations ---------------------------
345 int (*fl_compare_owner)(struct file_lock *, struct file_lock *);
346 void (*fl_notify)(struct file_lock *); /* unblock callback */
347 int (*fl_grant)(struct file_lock *, struct file_lock *, int);
348 void (*fl_release_private)(struct file_lock *);
349 void (*fl_break)(struct file_lock *); /* break_lease callback */
350 int (*fl_mylease)(struct file_lock *, struct file_lock *);
351 int (*fl_change)(struct file_lock **, int);
354 file_lock_lock may block
355 fl_compare_owner: yes no
358 fl_release_private: maybe no
363 --------------------------- buffer_head -----------------------------------
365 void (*b_end_io)(struct buffer_head *bh, int uptodate);
368 called from interrupts. In other words, extreme care is needed here.
369 bh is locked, but that's all warranties we have here. Currently only RAID1,
370 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
371 call this method upon the IO completion.
373 --------------------------- block_device_operations -----------------------
375 int (*open) (struct block_device *, fmode_t);
376 int (*release) (struct gendisk *, fmode_t);
377 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
378 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
379 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
380 int (*media_changed) (struct gendisk *);
381 void (*unlock_native_capacity) (struct gendisk *);
382 int (*revalidate_disk) (struct gendisk *);
383 int (*getgeo)(struct block_device *, struct hd_geometry *);
384 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
394 unlock_native_capacity: no
397 swap_slot_free_notify: no (see below)
399 media_changed, unlock_native_capacity and revalidate_disk are called only from
402 swap_slot_free_notify is called with swap_lock and sometimes the page lock
406 --------------------------- file_operations -------------------------------
408 loff_t (*llseek) (struct file *, loff_t, int);
409 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
410 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
411 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
412 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
413 int (*readdir) (struct file *, void *, filldir_t);
414 unsigned int (*poll) (struct file *, struct poll_table_struct *);
415 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
416 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
417 int (*mmap) (struct file *, struct vm_area_struct *);
418 int (*open) (struct inode *, struct file *);
419 int (*flush) (struct file *);
420 int (*release) (struct inode *, struct file *);
421 int (*fsync) (struct file *, int datasync);
422 int (*aio_fsync) (struct kiocb *, int datasync);
423 int (*fasync) (int, struct file *, int);
424 int (*lock) (struct file *, int, struct file_lock *);
425 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
427 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
429 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
431 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
433 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
434 unsigned long, unsigned long, unsigned long);
435 int (*check_flags)(int);
436 int (*flock) (struct file *, int, struct file_lock *);
437 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
438 size_t, unsigned int);
439 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
440 size_t, unsigned int);
441 int (*setlease)(struct file *, long, struct file_lock **);
445 All may block except for ->setlease.
446 No VFS locks held on entry except for ->fsync and ->setlease.
448 ->fsync() has i_mutex on inode.
450 ->setlease has the file_list_lock held and must not sleep.
452 ->llseek() locking has moved from llseek to the individual llseek
453 implementations. If your fs is not using generic_file_llseek, you
454 need to acquire and release the appropriate locks in your ->llseek().
455 For many filesystems, it is probably safe to acquire the inode
456 mutex or just to use i_size_read() instead.
457 Note: this does not protect the file->f_pos against concurrent modifications
458 since this is something the userspace has to take care about.
460 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
461 Most instances call fasync_helper(), which does that maintenance, so it's
462 not normally something one needs to worry about. Return values > 0 will be
463 mapped to zero in the VFS layer.
465 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
466 move ->readdir() to inode_operations and use a separate method for directory
467 ->ioctl() or kill the latter completely. One of the problems is that for
468 anything that resembles union-mount we won't have a struct file for all
469 components. And there are other reasons why the current interface is a mess...
471 ->read on directories probably must go away - we should just enforce -EISDIR
472 in sys_read() and friends.
474 --------------------------- dquot_operations -------------------------------
476 int (*write_dquot) (struct dquot *);
477 int (*acquire_dquot) (struct dquot *);
478 int (*release_dquot) (struct dquot *);
479 int (*mark_dirty) (struct dquot *);
480 int (*write_info) (struct super_block *, int);
482 These operations are intended to be more or less wrapping functions that ensure
483 a proper locking wrt the filesystem and call the generic quota operations.
485 What filesystem should expect from the generic quota functions:
487 FS recursion Held locks when called
488 write_dquot: yes dqonoff_sem or dqptr_sem
489 acquire_dquot: yes dqonoff_sem or dqptr_sem
490 release_dquot: yes dqonoff_sem or dqptr_sem
492 write_info: yes dqonoff_sem
494 FS recursion means calling ->quota_read() and ->quota_write() from superblock
497 More details about quota locking can be found in fs/dquot.c.
499 --------------------------- vm_operations_struct -----------------------------
501 void (*open)(struct vm_area_struct*);
502 void (*close)(struct vm_area_struct*);
503 int (*fault)(struct vm_area_struct*, struct vm_fault *);
504 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
505 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
508 mmap_sem PageLocked(page)
511 fault: yes can return with page locked
512 page_mkwrite: yes can return with page locked
515 ->fault() is called when a previously not present pte is about
516 to be faulted in. The filesystem must find and return the page associated
517 with the passed in "pgoff" in the vm_fault structure. If it is possible that
518 the page may be truncated and/or invalidated, then the filesystem must lock
519 the page, then ensure it is not already truncated (the page lock will block
520 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
521 locked. The VM will unlock the page.
523 ->page_mkwrite() is called when a previously read-only pte is
524 about to become writeable. The filesystem again must ensure that there are
525 no truncate/invalidate races, and then return with the page locked. If
526 the page has been truncated, the filesystem should not look up a new page
527 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
528 will cause the VM to retry the fault.
530 ->access() is called when get_user_pages() fails in
531 acces_process_vm(), typically used to debug a process through
532 /proc/pid/mem or ptrace. This function is needed only for
533 VM_IO | VM_PFNMAP VMAs.
535 ================================================================================
538 (if you break something or notice that it is broken and do not fix it yourself
539 - at least put it here)