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 (*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 void (*truncate_range)(struct inode *, loff_t, loff_t);
63 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
75 rmdir: yes (both) (see below)
76 rename: yes (all) (see below)
80 truncate: yes (see below)
82 permission: no (may not block if called in rcu-walk mode)
91 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
93 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
94 ->truncate() is never called directly - it's a callback, not a
95 method. It's called by vmtruncate() - deprecated library function used by
96 ->setattr(). Locking information above applies to that call (i.e. is
97 inherited from ->setattr() - vmtruncate() is used when ATTR_SIZE had been
100 See Documentation/filesystems/directory-locking for more detailed discussion
101 of the locking scheme for directory operations.
103 --------------------------- super_operations ---------------------------
105 struct inode *(*alloc_inode)(struct super_block *sb);
106 void (*destroy_inode)(struct inode *);
107 void (*dirty_inode) (struct inode *, int flags);
108 int (*write_inode) (struct inode *, struct writeback_control *wbc);
109 int (*drop_inode) (struct inode *);
110 void (*evict_inode) (struct inode *);
111 void (*put_super) (struct super_block *);
112 void (*write_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 vfsmount *);
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!!!
138 statfs: maybe(read) (see below)
141 show_options: no (namespace_sem)
142 quota_read: no (see below)
143 quota_write: no (see below)
144 bdev_try_to_free_page: no (see below)
146 ->statfs() has s_umount (shared) when called by ustat(2) (native or
147 compat), but that's an accident of bad API; s_umount is used to pin
148 the superblock down when we only have dev_t given us by userland to
149 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
150 doesn't hold it when calling ->statfs() - superblock is pinned down
151 by resolving the pathname passed to syscall.
152 ->quota_read() and ->quota_write() functions are both guaranteed to
153 be the only ones operating on the quota file by the quota code (via
154 dqio_sem) (unless an admin really wants to screw up something and
155 writes to quota files with quotas on). For other details about locking
156 see also dquot_operations section.
157 ->bdev_try_to_free_page is called from the ->releasepage handler of
158 the block device inode. See there for more details.
160 --------------------------- file_system_type ---------------------------
162 int (*get_sb) (struct file_system_type *, int,
163 const char *, void *, struct vfsmount *);
164 struct dentry *(*mount) (struct file_system_type *, int,
165 const char *, void *);
166 void (*kill_sb) (struct super_block *);
172 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
174 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
175 unlocks and drops the reference.
177 --------------------------- address_space_operations --------------------------
179 int (*writepage)(struct page *page, struct writeback_control *wbc);
180 int (*readpage)(struct file *, struct page *);
181 int (*sync_page)(struct page *);
182 int (*writepages)(struct address_space *, struct writeback_control *);
183 int (*set_page_dirty)(struct page *page);
184 int (*readpages)(struct file *filp, struct address_space *mapping,
185 struct list_head *pages, unsigned nr_pages);
186 int (*write_begin)(struct file *, struct address_space *mapping,
187 loff_t pos, unsigned len, unsigned flags,
188 struct page **pagep, void **fsdata);
189 int (*write_end)(struct file *, struct address_space *mapping,
190 loff_t pos, unsigned len, unsigned copied,
191 struct page *page, void *fsdata);
192 sector_t (*bmap)(struct address_space *, sector_t);
193 int (*invalidatepage) (struct page *, unsigned long);
194 int (*releasepage) (struct page *, int);
195 void (*freepage)(struct page *);
196 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
197 loff_t offset, unsigned long nr_segs);
198 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
200 int (*migratepage)(struct address_space *, struct page *, struct page *);
201 int (*launder_page)(struct page *);
202 int (*is_partially_uptodate)(struct page *, read_descriptor_t *, unsigned long);
203 int (*error_remove_page)(struct address_space *, struct page *);
206 All except set_page_dirty and freepage may block
208 PageLocked(page) i_mutex
209 writepage: yes, unlocks (see below)
210 readpage: yes, unlocks
215 write_begin: locks the page yes
216 write_end: yes, unlocks yes
223 migratepage: yes (both)
225 is_partially_uptodate: yes
226 error_remove_page: yes
228 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
229 may be called from the request handler (/dev/loop).
231 ->readpage() unlocks the page, either synchronously or via I/O
234 ->readpages() populates the pagecache with the passed pages and starts
235 I/O against them. They come unlocked upon I/O completion.
237 ->writepage() is used for two purposes: for "memory cleansing" and for
238 "sync". These are quite different operations and the behaviour may differ
239 depending upon the mode.
241 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
242 it *must* start I/O against the page, even if that would involve
243 blocking on in-progress I/O.
245 If writepage is called for memory cleansing (sync_mode ==
246 WBC_SYNC_NONE) then its role is to get as much writeout underway as
247 possible. So writepage should try to avoid blocking against
248 currently-in-progress I/O.
250 If the filesystem is not called for "sync" and it determines that it
251 would need to block against in-progress I/O to be able to start new I/O
252 against the page the filesystem should redirty the page with
253 redirty_page_for_writepage(), then unlock the page and return zero.
254 This may also be done to avoid internal deadlocks, but rarely.
256 If the filesystem is called for sync then it must wait on any
257 in-progress I/O and then start new I/O.
259 The filesystem should unlock the page synchronously, before returning to the
260 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
261 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
262 currently, and VM should stop calling ->writepage() on this page for some
263 time. VM does this by moving page to the head of the active list, hence the
266 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
267 and return zero, writepage *must* run set_page_writeback() against the page,
268 followed by unlocking it. Once set_page_writeback() has been run against the
269 page, write I/O can be submitted and the write I/O completion handler must run
270 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
271 filesystem must run end_page_writeback() against the page before returning from
274 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
275 if the filesystem needs the page to be locked during writeout, that is ok, too,
276 the page is allowed to be unlocked at any point in time between the calls to
277 set_page_writeback() and end_page_writeback().
279 Note, failure to run either redirty_page_for_writepage() or the combination of
280 set_page_writeback()/end_page_writeback() on a page submitted to writepage
281 will leave the page itself marked clean but it will be tagged as dirty in the
282 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
283 in the filesystem like having dirty inodes at umount and losing written data.
285 ->sync_page() locking rules are not well-defined - usually it is called
286 with lock on page, but that is not guaranteed. Considering the currently
287 existing instances of this method ->sync_page() itself doesn't look
290 ->writepages() is used for periodic writeback and for syscall-initiated
291 sync operations. The address_space should start I/O against at least
292 *nr_to_write pages. *nr_to_write must be decremented for each page which is
293 written. The address_space implementation may write more (or less) pages
294 than *nr_to_write asks for, but it should try to be reasonably close. If
295 nr_to_write is NULL, all dirty pages must be written.
297 writepages should _only_ write pages which are present on
300 ->set_page_dirty() is called from various places in the kernel
301 when the target page is marked as needing writeback. It may be called
302 under spinlock (it cannot block) and is sometimes called with the page
305 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
306 filesystems and by the swapper. The latter will eventually go away. Please,
307 keep it that way and don't breed new callers.
309 ->invalidatepage() is called when the filesystem must attempt to drop
310 some or all of the buffers from the page when it is being truncated. It
311 returns zero on success. If ->invalidatepage is zero, the kernel uses
312 block_invalidatepage() instead.
314 ->releasepage() is called when the kernel is about to try to drop the
315 buffers from the page in preparation for freeing it. It returns zero to
316 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
317 the kernel assumes that the fs has no private interest in the buffers.
319 ->freepage() is called when the kernel is done dropping the page
322 ->launder_page() may be called prior to releasing a page if
323 it is still found to be dirty. It returns zero if the page was successfully
324 cleaned, or an error value if not. Note that in order to prevent the page
325 getting mapped back in and redirtied, it needs to be kept locked
326 across the entire operation.
328 ----------------------- file_lock_operations ------------------------------
330 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
331 void (*fl_release_private)(struct file_lock *);
335 file_lock_lock may block
337 fl_release_private: maybe no
339 ----------------------- lock_manager_operations ---------------------------
341 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
342 void (*lm_notify)(struct file_lock *); /* unblock callback */
343 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
344 void (*lm_release_private)(struct file_lock *);
345 void (*lm_break)(struct file_lock *); /* break_lease callback */
346 int (*lm_change)(struct file_lock **, int);
349 file_lock_lock may block
350 lm_compare_owner: yes no
353 lm_release_private: maybe no
357 --------------------------- buffer_head -----------------------------------
359 void (*b_end_io)(struct buffer_head *bh, int uptodate);
362 called from interrupts. In other words, extreme care is needed here.
363 bh is locked, but that's all warranties we have here. Currently only RAID1,
364 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
365 call this method upon the IO completion.
367 --------------------------- block_device_operations -----------------------
369 int (*open) (struct block_device *, fmode_t);
370 int (*release) (struct gendisk *, fmode_t);
371 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
372 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
373 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
374 int (*media_changed) (struct gendisk *);
375 void (*unlock_native_capacity) (struct gendisk *);
376 int (*revalidate_disk) (struct gendisk *);
377 int (*getgeo)(struct block_device *, struct hd_geometry *);
378 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
388 unlock_native_capacity: no
391 swap_slot_free_notify: no (see below)
393 media_changed, unlock_native_capacity and revalidate_disk are called only from
396 swap_slot_free_notify is called with swap_lock and sometimes the page lock
400 --------------------------- file_operations -------------------------------
402 loff_t (*llseek) (struct file *, loff_t, int);
403 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
404 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
405 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
406 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
407 int (*readdir) (struct file *, void *, filldir_t);
408 unsigned int (*poll) (struct file *, struct poll_table_struct *);
409 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
410 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
411 int (*mmap) (struct file *, struct vm_area_struct *);
412 int (*open) (struct inode *, struct file *);
413 int (*flush) (struct file *);
414 int (*release) (struct inode *, struct file *);
415 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
416 int (*aio_fsync) (struct kiocb *, int datasync);
417 int (*fasync) (int, struct file *, int);
418 int (*lock) (struct file *, int, struct file_lock *);
419 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
421 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
423 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
425 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
427 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
428 unsigned long, unsigned long, unsigned long);
429 int (*check_flags)(int);
430 int (*flock) (struct file *, int, struct file_lock *);
431 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
432 size_t, unsigned int);
433 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
434 size_t, unsigned int);
435 int (*setlease)(struct file *, long, struct file_lock **);
436 long (*fallocate)(struct file *, int, loff_t, loff_t);
440 All may block except for ->setlease.
441 No VFS locks held on entry except for ->setlease.
443 ->setlease has the file_list_lock held and must not sleep.
445 ->llseek() locking has moved from llseek to the individual llseek
446 implementations. If your fs is not using generic_file_llseek, you
447 need to acquire and release the appropriate locks in your ->llseek().
448 For many filesystems, it is probably safe to acquire the inode
449 mutex or just to use i_size_read() instead.
450 Note: this does not protect the file->f_pos against concurrent modifications
451 since this is something the userspace has to take care about.
453 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
454 Most instances call fasync_helper(), which does that maintenance, so it's
455 not normally something one needs to worry about. Return values > 0 will be
456 mapped to zero in the VFS layer.
458 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
459 move ->readdir() to inode_operations and use a separate method for directory
460 ->ioctl() or kill the latter completely. One of the problems is that for
461 anything that resembles union-mount we won't have a struct file for all
462 components. And there are other reasons why the current interface is a mess...
464 ->read on directories probably must go away - we should just enforce -EISDIR
465 in sys_read() and friends.
467 --------------------------- dquot_operations -------------------------------
469 int (*write_dquot) (struct dquot *);
470 int (*acquire_dquot) (struct dquot *);
471 int (*release_dquot) (struct dquot *);
472 int (*mark_dirty) (struct dquot *);
473 int (*write_info) (struct super_block *, int);
475 These operations are intended to be more or less wrapping functions that ensure
476 a proper locking wrt the filesystem and call the generic quota operations.
478 What filesystem should expect from the generic quota functions:
480 FS recursion Held locks when called
481 write_dquot: yes dqonoff_sem or dqptr_sem
482 acquire_dquot: yes dqonoff_sem or dqptr_sem
483 release_dquot: yes dqonoff_sem or dqptr_sem
485 write_info: yes dqonoff_sem
487 FS recursion means calling ->quota_read() and ->quota_write() from superblock
490 More details about quota locking can be found in fs/dquot.c.
492 --------------------------- vm_operations_struct -----------------------------
494 void (*open)(struct vm_area_struct*);
495 void (*close)(struct vm_area_struct*);
496 int (*fault)(struct vm_area_struct*, struct vm_fault *);
497 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
498 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
501 mmap_sem PageLocked(page)
504 fault: yes can return with page locked
505 page_mkwrite: yes can return with page locked
508 ->fault() is called when a previously not present pte is about
509 to be faulted in. The filesystem must find and return the page associated
510 with the passed in "pgoff" in the vm_fault structure. If it is possible that
511 the page may be truncated and/or invalidated, then the filesystem must lock
512 the page, then ensure it is not already truncated (the page lock will block
513 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
514 locked. The VM will unlock the page.
516 ->page_mkwrite() is called when a previously read-only pte is
517 about to become writeable. The filesystem again must ensure that there are
518 no truncate/invalidate races, and then return with the page locked. If
519 the page has been truncated, the filesystem should not look up a new page
520 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
521 will cause the VM to retry the fault.
523 ->access() is called when get_user_pages() fails in
524 acces_process_vm(), typically used to debug a process through
525 /proc/pid/mem or ptrace. This function is needed only for
526 VM_IO | VM_PFNMAP VMAs.
528 ================================================================================
531 (if you break something or notice that it is broken and do not fix it yourself
532 - at least put it here)