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 *, struct qstr *);
15 int (*d_compare)(const struct dentry *,
16 unsigned int, const char *, const struct qstr *);
17 int (*d_delete)(struct dentry *);
18 int (*d_init)(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)(const struct path *, bool);
24 struct dentry *(*d_real)(struct dentry *, const struct inode *,
28 rename_lock ->d_lock may block rcu-walk
29 d_revalidate: no no yes (ref-walk) maybe
30 d_weak_revalidate:no no yes no
32 d_compare: yes no no maybe
33 d_delete: no yes no no
35 d_release: no no yes no
39 d_automount: no no yes no
40 d_manage: no no yes (ref-walk) maybe
43 --------------------------- inode_operations ---------------------------
45 int (*create) (struct inode *,struct dentry *,umode_t, bool);
46 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
47 int (*link) (struct dentry *,struct inode *,struct dentry *);
48 int (*unlink) (struct inode *,struct dentry *);
49 int (*symlink) (struct inode *,struct dentry *,const char *);
50 int (*mkdir) (struct inode *,struct dentry *,umode_t);
51 int (*rmdir) (struct inode *,struct dentry *);
52 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
53 int (*rename) (struct inode *, struct dentry *,
54 struct inode *, struct dentry *, unsigned int);
55 int (*readlink) (struct dentry *, char __user *,int);
56 const char *(*get_link) (struct dentry *, struct inode *, void **);
57 void (*truncate) (struct inode *);
58 int (*permission) (struct inode *, int, unsigned int);
59 int (*get_acl)(struct inode *, int);
60 int (*setattr) (struct dentry *, struct iattr *);
61 int (*getattr) (const struct path *, struct dentry *, struct kstat *,
63 ssize_t (*listxattr) (struct dentry *, char *, size_t);
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);
69 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
81 rmdir: yes (both) (see below)
82 rename: yes (all) (see below)
86 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.
100 See Documentation/filesystems/directory-locking for more detailed discussion
101 of the locking scheme for directory operations.
103 ----------------------- xattr_handler operations -----------------------
105 bool (*list)(struct dentry *dentry);
106 int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
107 struct inode *inode, const char *name, void *buffer,
109 int (*set)(const struct xattr_handler *handler, struct dentry *dentry,
110 struct inode *inode, const char *name, const void *buffer,
111 size_t size, int flags);
120 --------------------------- super_operations ---------------------------
122 struct inode *(*alloc_inode)(struct super_block *sb);
123 void (*destroy_inode)(struct inode *);
124 void (*dirty_inode) (struct inode *, int flags);
125 int (*write_inode) (struct inode *, struct writeback_control *wbc);
126 int (*drop_inode) (struct inode *);
127 void (*evict_inode) (struct inode *);
128 void (*put_super) (struct super_block *);
129 int (*sync_fs)(struct super_block *sb, int wait);
130 int (*freeze_fs) (struct super_block *);
131 int (*unfreeze_fs) (struct super_block *);
132 int (*statfs) (struct dentry *, struct kstatfs *);
133 int (*remount_fs) (struct super_block *, int *, char *);
134 void (*umount_begin) (struct super_block *);
135 int (*show_options)(struct seq_file *, struct dentry *);
136 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
137 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
138 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
141 All may block [not true, see below]
147 drop_inode: !!!inode->i_lock!!!
153 statfs: maybe(read) (see below)
156 show_options: no (namespace_sem)
157 quota_read: no (see below)
158 quota_write: no (see below)
159 bdev_try_to_free_page: no (see below)
161 ->statfs() has s_umount (shared) when called by ustat(2) (native or
162 compat), but that's an accident of bad API; s_umount is used to pin
163 the superblock down when we only have dev_t given us by userland to
164 identify the superblock. Everything else (statfs(), fstatfs(), etc.)
165 doesn't hold it when calling ->statfs() - superblock is pinned down
166 by resolving the pathname passed to syscall.
167 ->quota_read() and ->quota_write() functions are both guaranteed to
168 be the only ones operating on the quota file by the quota code (via
169 dqio_sem) (unless an admin really wants to screw up something and
170 writes to quota files with quotas on). For other details about locking
171 see also dquot_operations section.
172 ->bdev_try_to_free_page is called from the ->releasepage handler of
173 the block device inode. See there for more details.
175 --------------------------- file_system_type ---------------------------
177 struct dentry *(*mount) (struct file_system_type *, int,
178 const char *, void *);
179 void (*kill_sb) (struct super_block *);
185 ->mount() returns ERR_PTR or the root dentry; its superblock should be locked
187 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
188 unlocks and drops the reference.
190 --------------------------- address_space_operations --------------------------
192 int (*writepage)(struct page *page, struct writeback_control *wbc);
193 int (*readpage)(struct file *, struct page *);
194 int (*writepages)(struct address_space *, struct writeback_control *);
195 int (*set_page_dirty)(struct page *page);
196 int (*readpages)(struct file *filp, struct address_space *mapping,
197 struct list_head *pages, unsigned nr_pages);
198 int (*write_begin)(struct file *, struct address_space *mapping,
199 loff_t pos, unsigned len, unsigned flags,
200 struct page **pagep, void **fsdata);
201 int (*write_end)(struct file *, struct address_space *mapping,
202 loff_t pos, unsigned len, unsigned copied,
203 struct page *page, void *fsdata);
204 sector_t (*bmap)(struct address_space *, sector_t);
205 void (*invalidatepage) (struct page *, unsigned int, unsigned int);
206 int (*releasepage) (struct page *, int);
207 void (*freepage)(struct page *);
208 int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
209 bool (*isolate_page) (struct page *, isolate_mode_t);
210 int (*migratepage)(struct address_space *, struct page *, struct page *);
211 void (*putback_page) (struct page *);
212 int (*launder_page)(struct page *);
213 int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long);
214 int (*error_remove_page)(struct address_space *, struct page *);
215 int (*swap_activate)(struct file *);
216 int (*swap_deactivate)(struct file *);
219 All except set_page_dirty and freepage may block
221 PageLocked(page) i_mutex
222 writepage: yes, unlocks (see below)
223 readpage: yes, unlocks
227 write_begin: locks the page yes
228 write_end: yes, unlocks yes
235 migratepage: yes (both)
238 is_partially_uptodate: yes
239 error_remove_page: yes
243 ->write_begin(), ->write_end() and ->readpage() may be called from
244 the request handler (/dev/loop).
246 ->readpage() unlocks the page, either synchronously or via I/O
249 ->readpages() populates the pagecache with the passed pages and starts
250 I/O against them. They come unlocked upon I/O completion.
252 ->writepage() is used for two purposes: for "memory cleansing" and for
253 "sync". These are quite different operations and the behaviour may differ
254 depending upon the mode.
256 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
257 it *must* start I/O against the page, even if that would involve
258 blocking on in-progress I/O.
260 If writepage is called for memory cleansing (sync_mode ==
261 WBC_SYNC_NONE) then its role is to get as much writeout underway as
262 possible. So writepage should try to avoid blocking against
263 currently-in-progress I/O.
265 If the filesystem is not called for "sync" and it determines that it
266 would need to block against in-progress I/O to be able to start new I/O
267 against the page the filesystem should redirty the page with
268 redirty_page_for_writepage(), then unlock the page and return zero.
269 This may also be done to avoid internal deadlocks, but rarely.
271 If the filesystem is called for sync then it must wait on any
272 in-progress I/O and then start new I/O.
274 The filesystem should unlock the page synchronously, before returning to the
275 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
276 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
277 currently, and VM should stop calling ->writepage() on this page for some
278 time. VM does this by moving page to the head of the active list, hence the
281 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
282 and return zero, writepage *must* run set_page_writeback() against the page,
283 followed by unlocking it. Once set_page_writeback() has been run against the
284 page, write I/O can be submitted and the write I/O completion handler must run
285 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
286 filesystem must run end_page_writeback() against the page before returning from
289 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
290 if the filesystem needs the page to be locked during writeout, that is ok, too,
291 the page is allowed to be unlocked at any point in time between the calls to
292 set_page_writeback() and end_page_writeback().
294 Note, failure to run either redirty_page_for_writepage() or the combination of
295 set_page_writeback()/end_page_writeback() on a page submitted to writepage
296 will leave the page itself marked clean but it will be tagged as dirty in the
297 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
298 in the filesystem like having dirty inodes at umount and losing written data.
300 ->writepages() is used for periodic writeback and for syscall-initiated
301 sync operations. The address_space should start I/O against at least
302 *nr_to_write pages. *nr_to_write must be decremented for each page which is
303 written. The address_space implementation may write more (or less) pages
304 than *nr_to_write asks for, but it should try to be reasonably close. If
305 nr_to_write is NULL, all dirty pages must be written.
307 writepages should _only_ write pages which are present on
310 ->set_page_dirty() is called from various places in the kernel
311 when the target page is marked as needing writeback. It may be called
312 under spinlock (it cannot block) and is sometimes called with the page
315 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
316 filesystems and by the swapper. The latter will eventually go away. Please,
317 keep it that way and don't breed new callers.
319 ->invalidatepage() is called when the filesystem must attempt to drop
320 some or all of the buffers from the page when it is being truncated. It
321 returns zero on success. If ->invalidatepage is zero, the kernel uses
322 block_invalidatepage() instead.
324 ->releasepage() is called when the kernel is about to try to drop the
325 buffers from the page in preparation for freeing it. It returns zero to
326 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
327 the kernel assumes that the fs has no private interest in the buffers.
329 ->freepage() is called when the kernel is done dropping the page
332 ->launder_page() may be called prior to releasing a page if
333 it is still found to be dirty. It returns zero if the page was successfully
334 cleaned, or an error value if not. Note that in order to prevent the page
335 getting mapped back in and redirtied, it needs to be kept locked
336 across the entire operation.
338 ->swap_activate will be called with a non-zero argument on
339 files backing (non block device backed) swapfiles. A return value
340 of zero indicates success, in which case this file can be used for
341 backing swapspace. The swapspace operations will be proxied to the
342 address space operations.
344 ->swap_deactivate() will be called in the sys_swapoff()
345 path after ->swap_activate() returned success.
347 ----------------------- file_lock_operations ------------------------------
349 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
350 void (*fl_release_private)(struct file_lock *);
354 inode->i_lock may block
356 fl_release_private: maybe maybe[1]
358 [1]: ->fl_release_private for flock or POSIX locks is currently allowed
359 to block. Leases however can still be freed while the i_lock is held and
360 so fl_release_private called on a lease should not block.
362 ----------------------- lock_manager_operations ---------------------------
364 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
365 unsigned long (*lm_owner_key)(struct file_lock *);
366 void (*lm_notify)(struct file_lock *); /* unblock callback */
367 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
368 void (*lm_break)(struct file_lock *); /* break_lease callback */
369 int (*lm_change)(struct file_lock **, int);
373 inode->i_lock blocked_lock_lock may block
374 lm_compare_owner: yes[1] maybe no
375 lm_owner_key yes[1] yes no
376 lm_notify: yes yes no
381 [1]: ->lm_compare_owner and ->lm_owner_key are generally called with
382 *an* inode->i_lock held. It may not be the i_lock of the inode
383 associated with either file_lock argument! This is the case with deadlock
384 detection, since the code has to chase down the owners of locks that may
385 be entirely unrelated to the one on which the lock is being acquired.
386 For deadlock detection however, the blocked_lock_lock is also held. The
387 fact that these locks are held ensures that the file_locks do not
388 disappear out from under you while doing the comparison or generating an
391 --------------------------- buffer_head -----------------------------------
393 void (*b_end_io)(struct buffer_head *bh, int uptodate);
396 called from interrupts. In other words, extreme care is needed here.
397 bh is locked, but that's all warranties we have here. Currently only RAID1,
398 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
399 call this method upon the IO completion.
401 --------------------------- block_device_operations -----------------------
403 int (*open) (struct block_device *, fmode_t);
404 int (*release) (struct gendisk *, fmode_t);
405 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
406 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
407 int (*direct_access) (struct block_device *, sector_t, void **,
409 int (*media_changed) (struct gendisk *);
410 void (*unlock_native_capacity) (struct gendisk *);
411 int (*revalidate_disk) (struct gendisk *);
412 int (*getgeo)(struct block_device *, struct hd_geometry *);
413 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
423 unlock_native_capacity: no
426 swap_slot_free_notify: no (see below)
428 media_changed, unlock_native_capacity and revalidate_disk are called only from
431 swap_slot_free_notify is called with swap_lock and sometimes the page lock
435 --------------------------- file_operations -------------------------------
437 loff_t (*llseek) (struct file *, loff_t, int);
438 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
439 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
440 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
441 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
442 int (*iterate) (struct file *, struct dir_context *);
443 unsigned int (*poll) (struct file *, struct poll_table_struct *);
444 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
445 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
446 int (*mmap) (struct file *, struct vm_area_struct *);
447 int (*open) (struct inode *, struct file *);
448 int (*flush) (struct file *);
449 int (*release) (struct inode *, struct file *);
450 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
451 int (*fasync) (int, struct file *, int);
452 int (*lock) (struct file *, int, struct file_lock *);
453 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
455 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
457 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
459 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
461 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
462 unsigned long, unsigned long, unsigned long);
463 int (*check_flags)(int);
464 int (*flock) (struct file *, int, struct file_lock *);
465 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
466 size_t, unsigned int);
467 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
468 size_t, unsigned int);
469 int (*setlease)(struct file *, long, struct file_lock **, void **);
470 long (*fallocate)(struct file *, int, loff_t, loff_t);
476 ->llseek() locking has moved from llseek to the individual llseek
477 implementations. If your fs is not using generic_file_llseek, you
478 need to acquire and release the appropriate locks in your ->llseek().
479 For many filesystems, it is probably safe to acquire the inode
480 mutex or just to use i_size_read() instead.
481 Note: this does not protect the file->f_pos against concurrent modifications
482 since this is something the userspace has to take care about.
484 ->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
485 Most instances call fasync_helper(), which does that maintenance, so it's
486 not normally something one needs to worry about. Return values > 0 will be
487 mapped to zero in the VFS layer.
489 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
490 move ->readdir() to inode_operations and use a separate method for directory
491 ->ioctl() or kill the latter completely. One of the problems is that for
492 anything that resembles union-mount we won't have a struct file for all
493 components. And there are other reasons why the current interface is a mess...
495 ->read on directories probably must go away - we should just enforce -EISDIR
496 in sys_read() and friends.
498 ->setlease operations should call generic_setlease() before or after setting
499 the lease within the individual filesystem to record the result of the
502 --------------------------- dquot_operations -------------------------------
504 int (*write_dquot) (struct dquot *);
505 int (*acquire_dquot) (struct dquot *);
506 int (*release_dquot) (struct dquot *);
507 int (*mark_dirty) (struct dquot *);
508 int (*write_info) (struct super_block *, int);
510 These operations are intended to be more or less wrapping functions that ensure
511 a proper locking wrt the filesystem and call the generic quota operations.
513 What filesystem should expect from the generic quota functions:
515 FS recursion Held locks when called
516 write_dquot: yes dqonoff_sem or dqptr_sem
517 acquire_dquot: yes dqonoff_sem or dqptr_sem
518 release_dquot: yes dqonoff_sem or dqptr_sem
520 write_info: yes dqonoff_sem
522 FS recursion means calling ->quota_read() and ->quota_write() from superblock
525 More details about quota locking can be found in fs/dquot.c.
527 --------------------------- vm_operations_struct -----------------------------
529 void (*open)(struct vm_area_struct*);
530 void (*close)(struct vm_area_struct*);
531 int (*fault)(struct vm_area_struct*, struct vm_fault *);
532 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
533 int (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
534 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
537 mmap_sem PageLocked(page)
540 fault: yes can return with page locked
542 page_mkwrite: yes can return with page locked
546 ->fault() is called when a previously not present pte is about
547 to be faulted in. The filesystem must find and return the page associated
548 with the passed in "pgoff" in the vm_fault structure. If it is possible that
549 the page may be truncated and/or invalidated, then the filesystem must lock
550 the page, then ensure it is not already truncated (the page lock will block
551 subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
552 locked. The VM will unlock the page.
554 ->map_pages() is called when VM asks to map easy accessible pages.
555 Filesystem should find and map pages associated with offsets from "start_pgoff"
556 till "end_pgoff". ->map_pages() is called with page table locked and must
557 not block. If it's not possible to reach a page without blocking,
558 filesystem should skip it. Filesystem should use do_set_pte() to setup
559 page table entry. Pointer to entry associated with the page is passed in
560 "pte" field in vm_fault structure. Pointers to entries for other offsets
561 should be calculated relative to "pte".
563 ->page_mkwrite() is called when a previously read-only pte is
564 about to become writeable. The filesystem again must ensure that there are
565 no truncate/invalidate races, and then return with the page locked. If
566 the page has been truncated, the filesystem should not look up a new page
567 like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
568 will cause the VM to retry the fault.
570 ->pfn_mkwrite() is the same as page_mkwrite but when the pte is
571 VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
572 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
573 after this call is to make the pte read-write, unless pfn_mkwrite returns
576 ->access() is called when get_user_pages() fails in
577 access_process_vm(), typically used to debug a process through
578 /proc/pid/mem or ptrace. This function is needed only for
579 VM_IO | VM_PFNMAP VMAs.
581 ================================================================================
584 (if you break something or notice that it is broken and do not fix it yourself
585 - at least put it here)