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
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, struct nameidata *);
41 struct dentry * (*lookup) (struct inode *,struct dentry *, struct nameid
43 int (*link) (struct dentry *,struct inode *,struct dentry *);
44 int (*unlink) (struct inode *,struct dentry *);
45 int (*symlink) (struct inode *,struct dentry *,const char *);
46 int (*mkdir) (struct inode *,struct dentry *,umode_t);
47 int (*rmdir) (struct inode *,struct dentry *);
48 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
49 int (*rename) (struct inode *, struct dentry *,
50 struct inode *, struct dentry *);
51 int (*readlink) (struct dentry *, char __user *,int);
52 void * (*follow_link) (struct dentry *, struct nameidata *);
53 void (*put_link) (struct dentry *, struct nameidata *, void *);
54 void (*truncate) (struct inode *);
55 int (*permission) (struct inode *, int, unsigned int);
56 int (*get_acl)(struct inode *, int);
57 int (*setattr) (struct dentry *, struct iattr *);
58 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
59 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
60 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
61 ssize_t (*listxattr) (struct dentry *, char *, size_t);
62 int (*removexattr) (struct dentry *, const char *);
63 void (*truncate_range)(struct inode *, loff_t, loff_t);
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)
92 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
94 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
95 ->truncate() is never called directly - it's a callback, not a
96 method. It's called by vmtruncate() - deprecated library function used by
97 ->setattr(). Locking information above applies to that call (i.e. is
98 inherited from ->setattr() - vmtruncate() is used when ATTR_SIZE had been
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 void (*write_super) (struct super_block *);
114 int (*sync_fs)(struct super_block *sb, int wait);
115 int (*freeze_fs) (struct super_block *);
116 int (*unfreeze_fs) (struct super_block *);
117 int (*statfs) (struct dentry *, struct kstatfs *);
118 int (*remount_fs) (struct super_block *, int *, char *);
119 void (*umount_begin) (struct super_block *);
120 int (*show_options)(struct seq_file *, struct dentry *);
121 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
122 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
123 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
126 All may block [not true, see below]
132 drop_inode: !!!inode->i_lock!!!
139 statfs: maybe(read) (see below)
142 show_options: no (namespace_sem)
143 quota_read: no (see below)
144 quota_write: no (see below)
145 bdev_try_to_free_page: no (see below)
147 ->statfs() has s_umount (shared) when called by ustat(2) (native or
148 compat), but that's an accident of bad API; s_umount is used to pin
149 the superblock down when we only have dev_t given us by userland to
150 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
151 doesn't hold it when calling ->statfs() - superblock is pinned down
152 by resolving the pathname passed to syscall.
153 ->quota_read() and ->quota_write() functions are both guaranteed to
154 be the only ones operating on the quota file by the quota code (via
155 dqio_sem) (unless an admin really wants to screw up something and
156 writes to quota files with quotas on). For other details about locking
157 see also dquot_operations section.
158 ->bdev_try_to_free_page is called from the ->releasepage handler of
159 the block device inode. See there for more details.
161 --------------------------- file_system_type ---------------------------
163 int (*get_sb) (struct file_system_type *, int,
164 const char *, void *, struct vfsmount *);
165 struct dentry *(*mount) (struct file_system_type *, int,
166 const char *, void *);
167 void (*kill_sb) (struct super_block *);
173 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
175 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
176 unlocks and drops the reference.
178 --------------------------- address_space_operations --------------------------
180 int (*writepage)(struct page *page, struct writeback_control *wbc);
181 int (*readpage)(struct file *, struct page *);
182 int (*sync_page)(struct page *);
183 int (*writepages)(struct address_space *, struct writeback_control *);
184 int (*set_page_dirty)(struct page *page);
185 int (*readpages)(struct file *filp, struct address_space *mapping,
186 struct list_head *pages, unsigned nr_pages);
187 int (*write_begin)(struct file *, struct address_space *mapping,
188 loff_t pos, unsigned len, unsigned flags,
189 struct page **pagep, void **fsdata);
190 int (*write_end)(struct file *, struct address_space *mapping,
191 loff_t pos, unsigned len, unsigned copied,
192 struct page *page, void *fsdata);
193 sector_t (*bmap)(struct address_space *, sector_t);
194 int (*invalidatepage) (struct page *, unsigned long);
195 int (*releasepage) (struct page *, int);
196 void (*freepage)(struct page *);
197 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
198 loff_t offset, unsigned long nr_segs);
199 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
201 int (*migratepage)(struct address_space *, struct page *, struct page *);
202 int (*launder_page)(struct page *);
203 int (*is_partially_uptodate)(struct page *, read_descriptor_t *, unsigned long);
204 int (*error_remove_page)(struct address_space *, struct page *);
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
229 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
230 may be called from the request handler (/dev/loop).
232 ->readpage() unlocks the page, either synchronously or via I/O
235 ->readpages() populates the pagecache with the passed pages and starts
236 I/O against them. They come unlocked upon I/O completion.
238 ->writepage() is used for two purposes: for "memory cleansing" and for
239 "sync". These are quite different operations and the behaviour may differ
240 depending upon the mode.
242 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
243 it *must* start I/O against the page, even if that would involve
244 blocking on in-progress I/O.
246 If writepage is called for memory cleansing (sync_mode ==
247 WBC_SYNC_NONE) then its role is to get as much writeout underway as
248 possible. So writepage should try to avoid blocking against
249 currently-in-progress I/O.
251 If the filesystem is not called for "sync" and it determines that it
252 would need to block against in-progress I/O to be able to start new I/O
253 against the page the filesystem should redirty the page with
254 redirty_page_for_writepage(), then unlock the page and return zero.
255 This may also be done to avoid internal deadlocks, but rarely.
257 If the filesystem is called for sync then it must wait on any
258 in-progress I/O and then start new I/O.
260 The filesystem should unlock the page synchronously, before returning to the
261 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
262 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
263 currently, and VM should stop calling ->writepage() on this page for some
264 time. VM does this by moving page to the head of the active list, hence the
267 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
268 and return zero, writepage *must* run set_page_writeback() against the page,
269 followed by unlocking it. Once set_page_writeback() has been run against the
270 page, write I/O can be submitted and the write I/O completion handler must run
271 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
272 filesystem must run end_page_writeback() against the page before returning from
275 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
276 if the filesystem needs the page to be locked during writeout, that is ok, too,
277 the page is allowed to be unlocked at any point in time between the calls to
278 set_page_writeback() and end_page_writeback().
280 Note, failure to run either redirty_page_for_writepage() or the combination of
281 set_page_writeback()/end_page_writeback() on a page submitted to writepage
282 will leave the page itself marked clean but it will be tagged as dirty in the
283 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
284 in the filesystem like having dirty inodes at umount and losing written data.
286 ->sync_page() locking rules are not well-defined - usually it is called
287 with lock on page, but that is not guaranteed. Considering the currently
288 existing instances of this method ->sync_page() itself doesn't look
291 ->writepages() is used for periodic writeback and for syscall-initiated
292 sync operations. The address_space should start I/O against at least
293 *nr_to_write pages. *nr_to_write must be decremented for each page which is
294 written. The address_space implementation may write more (or less) pages
295 than *nr_to_write asks for, but it should try to be reasonably close. If
296 nr_to_write is NULL, all dirty pages must be written.
298 writepages should _only_ write pages which are present on
301 ->set_page_dirty() is called from various places in the kernel
302 when the target page is marked as needing writeback. It may be called
303 under spinlock (it cannot block) and is sometimes called with the page
306 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
307 filesystems and by the swapper. The latter will eventually go away. Please,
308 keep it that way and don't breed new callers.
310 ->invalidatepage() is called when the filesystem must attempt to drop
311 some or all of the buffers from the page when it is being truncated. It
312 returns zero on success. If ->invalidatepage is zero, the kernel uses
313 block_invalidatepage() instead.
315 ->releasepage() is called when the kernel is about to try to drop the
316 buffers from the page in preparation for freeing it. It returns zero to
317 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
318 the kernel assumes that the fs has no private interest in the buffers.
320 ->freepage() is called when the kernel is done dropping the page
323 ->launder_page() may be called prior to releasing a page if
324 it is still found to be dirty. It returns zero if the page was successfully
325 cleaned, or an error value if not. Note that in order to prevent the page
326 getting mapped back in and redirtied, it needs to be kept locked
327 across the entire operation.
329 ----------------------- file_lock_operations ------------------------------
331 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
332 void (*fl_release_private)(struct file_lock *);
336 file_lock_lock may block
338 fl_release_private: maybe no
340 ----------------------- lock_manager_operations ---------------------------
342 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
343 void (*lm_notify)(struct file_lock *); /* unblock callback */
344 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
345 void (*lm_release_private)(struct file_lock *);
346 void (*lm_break)(struct file_lock *); /* break_lease callback */
347 int (*lm_change)(struct file_lock **, int);
350 file_lock_lock may block
351 lm_compare_owner: yes no
354 lm_release_private: maybe no
358 --------------------------- buffer_head -----------------------------------
360 void (*b_end_io)(struct buffer_head *bh, int uptodate);
363 called from interrupts. In other words, extreme care is needed here.
364 bh is locked, but that's all warranties we have here. Currently only RAID1,
365 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
366 call this method upon the IO completion.
368 --------------------------- block_device_operations -----------------------
370 int (*open) (struct block_device *, fmode_t);
371 int (*release) (struct gendisk *, fmode_t);
372 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
373 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
374 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
375 int (*media_changed) (struct gendisk *);
376 void (*unlock_native_capacity) (struct gendisk *);
377 int (*revalidate_disk) (struct gendisk *);
378 int (*getgeo)(struct block_device *, struct hd_geometry *);
379 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
389 unlock_native_capacity: no
392 swap_slot_free_notify: no (see below)
394 media_changed, unlock_native_capacity and revalidate_disk are called only from
397 swap_slot_free_notify is called with swap_lock and sometimes the page lock
401 --------------------------- file_operations -------------------------------
403 loff_t (*llseek) (struct file *, loff_t, int);
404 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
405 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
406 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
407 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
408 int (*readdir) (struct file *, void *, filldir_t);
409 unsigned int (*poll) (struct file *, struct poll_table_struct *);
410 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
411 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
412 int (*mmap) (struct file *, struct vm_area_struct *);
413 int (*open) (struct inode *, struct file *);
414 int (*flush) (struct file *);
415 int (*release) (struct inode *, struct file *);
416 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
417 int (*aio_fsync) (struct kiocb *, int datasync);
418 int (*fasync) (int, struct file *, int);
419 int (*lock) (struct file *, int, struct file_lock *);
420 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
422 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
424 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
426 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
428 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
429 unsigned long, unsigned long, unsigned long);
430 int (*check_flags)(int);
431 int (*flock) (struct file *, int, struct file_lock *);
432 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
433 size_t, unsigned int);
434 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
435 size_t, unsigned int);
436 int (*setlease)(struct file *, long, struct file_lock **);
437 long (*fallocate)(struct file *, int, loff_t, loff_t);
441 All may block except for ->setlease.
442 No VFS locks held on entry except for ->setlease.
444 ->setlease has the file_list_lock held and must not sleep.
446 ->llseek() locking has moved from llseek to the individual llseek
447 implementations. If your fs is not using generic_file_llseek, you
448 need to acquire and release the appropriate locks in your ->llseek().
449 For many filesystems, it is probably safe to acquire the inode
450 mutex or just to use i_size_read() instead.
451 Note: this does not protect the file->f_pos against concurrent modifications
452 since this is something the userspace has to take care about.
454 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
455 Most instances call fasync_helper(), which does that maintenance, so it's
456 not normally something one needs to worry about. Return values > 0 will be
457 mapped to zero in the VFS layer.
459 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
460 move ->readdir() to inode_operations and use a separate method for directory
461 ->ioctl() or kill the latter completely. One of the problems is that for
462 anything that resembles union-mount we won't have a struct file for all
463 components. And there are other reasons why the current interface is a mess...
465 ->read on directories probably must go away - we should just enforce -EISDIR
466 in sys_read() and friends.
468 --------------------------- dquot_operations -------------------------------
470 int (*write_dquot) (struct dquot *);
471 int (*acquire_dquot) (struct dquot *);
472 int (*release_dquot) (struct dquot *);
473 int (*mark_dirty) (struct dquot *);
474 int (*write_info) (struct super_block *, int);
476 These operations are intended to be more or less wrapping functions that ensure
477 a proper locking wrt the filesystem and call the generic quota operations.
479 What filesystem should expect from the generic quota functions:
481 FS recursion Held locks when called
482 write_dquot: yes dqonoff_sem or dqptr_sem
483 acquire_dquot: yes dqonoff_sem or dqptr_sem
484 release_dquot: yes dqonoff_sem or dqptr_sem
486 write_info: yes dqonoff_sem
488 FS recursion means calling ->quota_read() and ->quota_write() from superblock
491 More details about quota locking can be found in fs/dquot.c.
493 --------------------------- vm_operations_struct -----------------------------
495 void (*open)(struct vm_area_struct*);
496 void (*close)(struct vm_area_struct*);
497 int (*fault)(struct vm_area_struct*, struct vm_fault *);
498 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
499 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
502 mmap_sem PageLocked(page)
505 fault: yes can return with page locked
506 page_mkwrite: yes can return with page locked
509 ->fault() is called when a previously not present pte is about
510 to be faulted in. The filesystem must find and return the page associated
511 with the passed in "pgoff" in the vm_fault structure. If it is possible that
512 the page may be truncated and/or invalidated, then the filesystem must lock
513 the page, then ensure it is not already truncated (the page lock will block
514 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
515 locked. The VM will unlock the page.
517 ->page_mkwrite() is called when a previously read-only pte is
518 about to become writeable. The filesystem again must ensure that there are
519 no truncate/invalidate races, and then return with the page locked. If
520 the page has been truncated, the filesystem should not look up a new page
521 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
522 will cause the VM to retry the fault.
524 ->access() is called when get_user_pages() fails in
525 acces_process_vm(), typically used to debug a process through
526 /proc/pid/mem or ptrace. This function is needed only for
527 VM_IO | VM_PFNMAP VMAs.
529 ================================================================================
532 (if you break something or notice that it is broken and do not fix it yourself
533 - at least put it here)