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 *, int);
13 int (*d_hash) (struct dentry *, struct qstr *);
14 int (*d_compare) (struct dentry *, struct qstr *, struct qstr *);
15 int (*d_delete)(struct dentry *);
16 void (*d_release)(struct dentry *);
17 void (*d_iput)(struct dentry *, struct inode *);
18 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
22 dcache_lock rename_lock ->d_lock may block
23 d_revalidate: no no no yes
25 d_compare: no yes no no
26 d_delete: yes no yes no
27 d_release: no no no yes
31 --------------------------- inode_operations ---------------------------
33 int (*create) (struct inode *,struct dentry *,int, struct nameidata *);
34 struct dentry * (*lookup) (struct inode *,struct dentry *, struct nameid
36 int (*link) (struct dentry *,struct inode *,struct dentry *);
37 int (*unlink) (struct inode *,struct dentry *);
38 int (*symlink) (struct inode *,struct dentry *,const char *);
39 int (*mkdir) (struct inode *,struct dentry *,int);
40 int (*rmdir) (struct inode *,struct dentry *);
41 int (*mknod) (struct inode *,struct dentry *,int,dev_t);
42 int (*rename) (struct inode *, struct dentry *,
43 struct inode *, struct dentry *);
44 int (*readlink) (struct dentry *, char __user *,int);
45 int (*follow_link) (struct dentry *, struct nameidata *);
46 void (*truncate) (struct inode *);
47 int (*permission) (struct inode *, int, struct nameidata *);
48 int (*setattr) (struct dentry *, struct iattr *);
49 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
50 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
51 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
52 ssize_t (*listxattr) (struct dentry *, char *, size_t);
53 int (*removexattr) (struct dentry *, const char *);
56 all may block, none have BKL
65 rmdir: yes (both) (see below)
66 rename: yes (all) (see below)
69 truncate: yes (see below)
77 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
79 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
80 ->truncate() is never called directly - it's a callback, not a
81 method. It's called by vmtruncate() - library function normally used by
82 ->setattr(). Locking information above applies to that call (i.e. is
83 inherited from ->setattr() - vmtruncate() is used when ATTR_SIZE had been
86 See Documentation/filesystems/directory-locking for more detailed discussion
87 of the locking scheme for directory operations.
89 --------------------------- super_operations ---------------------------
91 struct inode *(*alloc_inode)(struct super_block *sb);
92 void (*destroy_inode)(struct inode *);
93 void (*dirty_inode) (struct inode *);
94 int (*write_inode) (struct inode *, int);
95 void (*put_inode) (struct inode *);
96 void (*drop_inode) (struct inode *);
97 void (*delete_inode) (struct inode *);
98 void (*put_super) (struct super_block *);
99 void (*write_super) (struct super_block *);
100 int (*sync_fs)(struct super_block *sb, int wait);
101 void (*write_super_lockfs) (struct super_block *);
102 void (*unlockfs) (struct super_block *);
103 int (*statfs) (struct dentry *, struct kstatfs *);
104 int (*remount_fs) (struct super_block *, int *, char *);
105 void (*clear_inode) (struct inode *);
106 void (*umount_begin) (struct super_block *);
107 int (*show_options)(struct seq_file *, struct vfsmount *);
108 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
109 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
114 alloc_inode: no no no
116 dirty_inode: no (must not sleep)
119 drop_inode: no !!!inode_lock!!!
121 put_super: yes yes no
122 write_super: no yes read
124 write_super_lockfs: ?
127 remount_fs: yes yes maybe (see below)
129 umount_begin: yes no no
130 show_options: no (vfsmount->sem)
131 quota_read: no no no (see below)
132 quota_write: no no no (see below)
134 ->remount_fs() will have the s_umount lock if it's already mounted.
135 When called from get_sb_single, it does NOT have the s_umount lock.
136 ->quota_read() and ->quota_write() functions are both guaranteed to
137 be the only ones operating on the quota file by the quota code (via
138 dqio_sem) (unless an admin really wants to screw up something and
139 writes to quota files with quotas on). For other details about locking
140 see also dquot_operations section.
142 --------------------------- file_system_type ---------------------------
144 int (*get_sb) (struct file_system_type *, int,
145 const char *, void *, struct vfsmount *);
146 void (*kill_sb) (struct super_block *);
152 ->get_sb() returns error or 0 with locked superblock attached to the vfsmount
153 (exclusive on ->s_umount).
154 ->kill_sb() takes a write-locked superblock, does all shutdown work on it,
155 unlocks and drops the reference.
157 --------------------------- address_space_operations --------------------------
159 int (*writepage)(struct page *page, struct writeback_control *wbc);
160 int (*readpage)(struct file *, struct page *);
161 int (*sync_page)(struct page *);
162 int (*writepages)(struct address_space *, struct writeback_control *);
163 int (*set_page_dirty)(struct page *page);
164 int (*readpages)(struct file *filp, struct address_space *mapping,
165 struct list_head *pages, unsigned nr_pages);
166 int (*prepare_write)(struct file *, struct page *, unsigned, unsigned);
167 int (*commit_write)(struct file *, struct page *, unsigned, unsigned);
168 sector_t (*bmap)(struct address_space *, sector_t);
169 int (*invalidatepage) (struct page *, unsigned long);
170 int (*releasepage) (struct page *, int);
171 int (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
172 loff_t offset, unsigned long nr_segs);
173 int (*launder_page) (struct page *);
176 All except set_page_dirty may block
178 BKL PageLocked(page) i_sem
179 writepage: no yes, unlocks (see below)
180 readpage: no yes, unlocks
185 prepare_write: no yes yes
186 commit_write: no yes yes
187 write_begin: no locks the page yes
188 write_end: no yes, unlocks yes
189 perform_write: no n/a yes
191 invalidatepage: no yes
196 ->prepare_write(), ->commit_write(), ->sync_page() and ->readpage()
197 may be called from the request handler (/dev/loop).
199 ->readpage() unlocks the page, either synchronously or via I/O
202 ->readpages() populates the pagecache with the passed pages and starts
203 I/O against them. They come unlocked upon I/O completion.
205 ->writepage() is used for two purposes: for "memory cleansing" and for
206 "sync". These are quite different operations and the behaviour may differ
207 depending upon the mode.
209 If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
210 it *must* start I/O against the page, even if that would involve
211 blocking on in-progress I/O.
213 If writepage is called for memory cleansing (sync_mode ==
214 WBC_SYNC_NONE) then its role is to get as much writeout underway as
215 possible. So writepage should try to avoid blocking against
216 currently-in-progress I/O.
218 If the filesystem is not called for "sync" and it determines that it
219 would need to block against in-progress I/O to be able to start new I/O
220 against the page the filesystem should redirty the page with
221 redirty_page_for_writepage(), then unlock the page and return zero.
222 This may also be done to avoid internal deadlocks, but rarely.
224 If the filesystem is called for sync then it must wait on any
225 in-progress I/O and then start new I/O.
227 The filesystem should unlock the page synchronously, before returning to the
228 caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
229 value. WRITEPAGE_ACTIVATE means that page cannot really be written out
230 currently, and VM should stop calling ->writepage() on this page for some
231 time. VM does this by moving page to the head of the active list, hence the
234 Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
235 and return zero, writepage *must* run set_page_writeback() against the page,
236 followed by unlocking it. Once set_page_writeback() has been run against the
237 page, write I/O can be submitted and the write I/O completion handler must run
238 end_page_writeback() once the I/O is complete. If no I/O is submitted, the
239 filesystem must run end_page_writeback() against the page before returning from
242 That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
243 if the filesystem needs the page to be locked during writeout, that is ok, too,
244 the page is allowed to be unlocked at any point in time between the calls to
245 set_page_writeback() and end_page_writeback().
247 Note, failure to run either redirty_page_for_writepage() or the combination of
248 set_page_writeback()/end_page_writeback() on a page submitted to writepage
249 will leave the page itself marked clean but it will be tagged as dirty in the
250 radix tree. This incoherency can lead to all sorts of hard-to-debug problems
251 in the filesystem like having dirty inodes at umount and losing written data.
253 ->sync_page() locking rules are not well-defined - usually it is called
254 with lock on page, but that is not guaranteed. Considering the currently
255 existing instances of this method ->sync_page() itself doesn't look
258 ->writepages() is used for periodic writeback and for syscall-initiated
259 sync operations. The address_space should start I/O against at least
260 *nr_to_write pages. *nr_to_write must be decremented for each page which is
261 written. The address_space implementation may write more (or less) pages
262 than *nr_to_write asks for, but it should try to be reasonably close. If
263 nr_to_write is NULL, all dirty pages must be written.
265 writepages should _only_ write pages which are present on
268 ->set_page_dirty() is called from various places in the kernel
269 when the target page is marked as needing writeback. It may be called
270 under spinlock (it cannot block) and is sometimes called with the page
273 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
274 filesystems and by the swapper. The latter will eventually go away. All
275 instances do not actually need the BKL. Please, keep it that way and don't
278 ->invalidatepage() is called when the filesystem must attempt to drop
279 some or all of the buffers from the page when it is being truncated. It
280 returns zero on success. If ->invalidatepage is zero, the kernel uses
281 block_invalidatepage() instead.
283 ->releasepage() is called when the kernel is about to try to drop the
284 buffers from the page in preparation for freeing it. It returns zero to
285 indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
286 the kernel assumes that the fs has no private interest in the buffers.
288 ->launder_page() may be called prior to releasing a page if
289 it is still found to be dirty. It returns zero if the page was successfully
290 cleaned, or an error value if not. Note that in order to prevent the page
291 getting mapped back in and redirtied, it needs to be kept locked
292 across the entire operation.
294 Note: currently almost all instances of address_space methods are
295 using BKL for internal serialization and that's one of the worst sources
296 of contention. Normally they are calling library functions (in fs/buffer.c)
297 and pass foo_get_block() as a callback (on local block-based filesystems,
298 indeed). BKL is not needed for library stuff and is usually taken by
299 foo_get_block(). It's an overkill, since block bitmaps can be protected by
300 internal fs locking and real critical areas are much smaller than the areas
301 filesystems protect now.
303 ----------------------- file_lock_operations ------------------------------
305 void (*fl_insert)(struct file_lock *); /* lock insertion callback */
306 void (*fl_remove)(struct file_lock *); /* lock removal callback */
307 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
308 void (*fl_release_private)(struct file_lock *);
316 fl_release_private: yes yes
318 ----------------------- lock_manager_operations ---------------------------
320 int (*fl_compare_owner)(struct file_lock *, struct file_lock *);
321 void (*fl_notify)(struct file_lock *); /* unblock callback */
322 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
323 void (*fl_release_private)(struct file_lock *);
324 void (*fl_break)(struct file_lock *); /* break_lease callback */
328 fl_compare_owner: yes no
331 fl_release_private: yes yes
334 Currently only NFSD and NLM provide instances of this class. None of the
335 them block. If you have out-of-tree instances - please, show up. Locking
336 in that area will change.
337 --------------------------- buffer_head -----------------------------------
339 void (*b_end_io)(struct buffer_head *bh, int uptodate);
342 called from interrupts. In other words, extreme care is needed here.
343 bh is locked, but that's all warranties we have here. Currently only RAID1,
344 highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
345 call this method upon the IO completion.
347 --------------------------- block_device_operations -----------------------
349 int (*open) (struct inode *, struct file *);
350 int (*release) (struct inode *, struct file *);
351 int (*ioctl) (struct inode *, struct file *, unsigned, unsigned long);
352 int (*media_changed) (struct gendisk *);
353 int (*revalidate_disk) (struct gendisk *);
361 revalidate_disk: no no
363 The last two are called only from check_disk_change().
365 --------------------------- file_operations -------------------------------
367 loff_t (*llseek) (struct file *, loff_t, int);
368 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
369 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
370 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
371 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
372 int (*readdir) (struct file *, void *, filldir_t);
373 unsigned int (*poll) (struct file *, struct poll_table_struct *);
374 int (*ioctl) (struct inode *, struct file *, unsigned int,
376 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
377 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
378 int (*mmap) (struct file *, struct vm_area_struct *);
379 int (*open) (struct inode *, struct file *);
380 int (*flush) (struct file *);
381 int (*release) (struct inode *, struct file *);
382 int (*fsync) (struct file *, struct dentry *, int datasync);
383 int (*aio_fsync) (struct kiocb *, int datasync);
384 int (*fasync) (int, struct file *, int);
385 int (*lock) (struct file *, int, struct file_lock *);
386 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
388 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
390 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
392 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
394 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
395 unsigned long, unsigned long, unsigned long);
396 int (*check_flags)(int);
397 int (*dir_notify)(struct file *, unsigned long);
401 All except ->poll() may block.
403 llseek: no (see below)
410 ioctl: yes (see below)
411 unlocked_ioctl: no (see below)
414 open: maybe (see below)
417 fsync: no (see below)
419 fasync: yes (see below)
425 get_unmapped_area: no
429 ->llseek() locking has moved from llseek to the individual llseek
430 implementations. If your fs is not using generic_file_llseek, you
431 need to acquire and release the appropriate locks in your ->llseek().
432 For many filesystems, it is probably safe to acquire the inode
433 semaphore. Note some filesystems (i.e. remote ones) provide no
434 protection for i_size so you will need to use the BKL.
436 ->open() locking is in-transit: big lock partially moved into the methods.
437 The only exception is ->open() in the instances of file_operations that never
438 end up in ->i_fop/->proc_fops, i.e. ones that belong to character devices
439 (chrdev_open() takes lock before replacing ->f_op and calling the secondary
440 method. As soon as we fix the handling of module reference counters all
441 instances of ->open() will be called without the BKL.
443 Note: ext2_release() was *the* source of contention on fs-intensive
444 loads and dropping BKL on ->release() helps to get rid of that (we still
445 grab BKL for cases when we close a file that had been opened r/w, but that
446 can and should be done using the internal locking with smaller critical areas).
447 Current worst offender is ext2_get_block()...
449 ->fasync() is a mess. This area needs a big cleanup and that will probably
452 ->readdir() and ->ioctl() on directories must be changed. Ideally we would
453 move ->readdir() to inode_operations and use a separate method for directory
454 ->ioctl() or kill the latter completely. One of the problems is that for
455 anything that resembles union-mount we won't have a struct file for all
456 components. And there are other reasons why the current interface is a mess...
458 ->ioctl() on regular files is superceded by the ->unlocked_ioctl() that
459 doesn't take the BKL.
461 ->read on directories probably must go away - we should just enforce -EISDIR
462 in sys_read() and friends.
464 ->fsync() has i_mutex on inode.
466 --------------------------- dquot_operations -------------------------------
468 int (*initialize) (struct inode *, int);
469 int (*drop) (struct inode *);
470 int (*alloc_space) (struct inode *, qsize_t, int);
471 int (*alloc_inode) (const struct inode *, unsigned long);
472 int (*free_space) (struct inode *, qsize_t);
473 int (*free_inode) (const struct inode *, unsigned long);
474 int (*transfer) (struct inode *, struct iattr *);
475 int (*write_dquot) (struct dquot *);
476 int (*acquire_dquot) (struct dquot *);
477 int (*release_dquot) (struct dquot *);
478 int (*mark_dirty) (struct dquot *);
479 int (*write_info) (struct super_block *, int);
481 These operations are intended to be more or less wrapping functions that ensure
482 a proper locking wrt the filesystem and call the generic quota operations.
484 What filesystem should expect from the generic quota functions:
486 FS recursion Held locks when called
487 initialize: yes maybe dqonoff_sem
489 alloc_space: ->mark_dirty() -
490 alloc_inode: ->mark_dirty() -
491 free_space: ->mark_dirty() -
492 free_inode: ->mark_dirty() -
494 write_dquot: yes dqonoff_sem or dqptr_sem
495 acquire_dquot: yes dqonoff_sem or dqptr_sem
496 release_dquot: yes dqonoff_sem or dqptr_sem
498 write_info: yes dqonoff_sem
500 FS recursion means calling ->quota_read() and ->quota_write() from superblock
503 ->alloc_space(), ->alloc_inode(), ->free_space(), ->free_inode() are called
504 only directly by the filesystem and do not call any fs functions only
505 the ->mark_dirty() operation.
507 More details about quota locking can be found in fs/dquot.c.
509 --------------------------- vm_operations_struct -----------------------------
511 void (*open)(struct vm_area_struct*);
512 void (*close)(struct vm_area_struct*);
513 int (*fault)(struct vm_area_struct*, struct vm_fault *);
514 int (*page_mkwrite)(struct vm_area_struct *, struct page *);
517 BKL mmap_sem PageLocked(page)
521 page_mkwrite: no yes no
523 ->page_mkwrite() is called when a previously read-only page is
524 about to become writeable. The file system is responsible for
525 protecting against truncate races. Once appropriate action has been
526 taking to lock out truncate, the page range should be verified to be
527 within i_size. The page mapping should also be checked that it is not
530 ================================================================================
533 (if you break something or notice that it is broken and do not fix it yourself
534 - at least put it here)
536 ipc/shm.c::shm_delete() - may need BKL.
537 ->read() and ->write() in many drivers are (probably) missing BKL.