1 =======
2 Locking
3 =======
4
5 The text below describes the locking rules for
6 It is (believed to be) up-to-date. *Please*, i
7 prototypes or locking protocols - update this
8 instances in the tree, don't leave that to mai
9 etc. At the very least, put the list of dubiou
10 Don't turn it into log - maintainers of out-of
11 be able to use diff(1).
12
13 Thing currently missing here: socket operation
14
15 dentry_operations
16 =================
17
18 prototypes::
19
20 int (*d_revalidate)(struct dentry *, u
21 int (*d_weak_revalidate)(struct dentry
22 int (*d_hash)(const struct dentry *, s
23 int (*d_compare)(const struct dentry *
24 unsigned int, const ch
25 int (*d_delete)(struct dentry *);
26 int (*d_init)(struct dentry *);
27 void (*d_release)(struct dentry *);
28 void (*d_iput)(struct dentry *, struct
29 char *(*d_dname)((struct dentry *dentr
30 struct vfsmount *(*d_automount)(struct
31 int (*d_manage)(const struct path *, b
32 struct dentry *(*d_real)(struct dentry
33
34 locking rules:
35
36 ================== =========== ========
37 ops rename_lock ->d_lock
38 ================== =========== ========
39 d_revalidate: no no
40 d_weak_revalidate: no no
41 d_hash no no
42 d_compare: yes no
43 d_delete: no yes
44 d_init: no no
45 d_release: no no
46 d_prune: no yes
47 d_iput: no no
48 d_dname: no no
49 d_automount: no no
50 d_manage: no no
51 d_real no no
52 ================== =========== ========
53
54 inode_operations
55 ================
56
57 prototypes::
58
59 int (*create) (struct mnt_idmap *, str
60 struct dentry * (*lookup) (struct inod
61 int (*link) (struct dentry *,struct in
62 int (*unlink) (struct inode *,struct d
63 int (*symlink) (struct mnt_idmap *, st
64 int (*mkdir) (struct mnt_idmap *, stru
65 int (*rmdir) (struct inode *,struct de
66 int (*mknod) (struct mnt_idmap *, stru
67 int (*rename) (struct mnt_idmap *, str
68 struct inode *, struct
69 int (*readlink) (struct dentry *, char
70 const char *(*get_link) (struct dentry
71 void (*truncate) (struct inode *);
72 int (*permission) (struct mnt_idmap *,
73 struct posix_acl * (*get_inode_acl)(st
74 int (*setattr) (struct mnt_idmap *, st
75 int (*getattr) (struct mnt_idmap *, co
76 ssize_t (*listxattr) (struct dentry *,
77 int (*fiemap)(struct inode *, struct f
78 void (*update_time)(struct inode *, st
79 int (*atomic_open)(struct inode *, str
80 struct file *,
81 umode_t create
82 int (*tmpfile) (struct mnt_idmap *, st
83 struct file *, umode_t
84 int (*fileattr_set)(struct mnt_idmap *
85 struct dentry *den
86 int (*fileattr_get)(struct dentry *den
87 struct posix_acl * (*get_acl)(struct m
88 struct offset_ctx *(*get_offset_ctx)(s
89
90 locking rules:
91 all may block
92
93 ============== ==============================
94 ops i_rwsem(inode)
95 ============== ==============================
96 lookup: shared
97 create: exclusive
98 link: exclusive (both)
99 mknod: exclusive
100 symlink: exclusive
101 mkdir: exclusive
102 unlink: exclusive (both)
103 rmdir: exclusive (both)(see below)
104 rename: exclusive (both parents, some
105 readlink: no
106 get_link: no
107 setattr: exclusive
108 permission: no (may not block if called in
109 get_inode_acl: no
110 get_acl: no
111 getattr: no
112 listxattr: no
113 fiemap: no
114 update_time: no
115 atomic_open: shared (exclusive if O_CREAT i
116 tmpfile: no
117 fileattr_get: no or exclusive
118 fileattr_set: exclusive
119 get_offset_ctx no
120 ============== ==============================
121
122
123 Additionally, ->rmdir(), ->unlink() an
124 exclusive on victim.
125 cross-directory ->rename() has (per-su
126 ->unlink() and ->rename() have ->i_rws
127 involved.
128 ->rename() has ->i_rwsem exclusive on
129
130 See Documentation/filesystems/directory-lockin
131 of the locking scheme for directory operations
132
133 xattr_handler operations
134 ========================
135
136 prototypes::
137
138 bool (*list)(struct dentry *dentry);
139 int (*get)(const struct xattr_handler
140 struct inode *inode, const
141 size_t size);
142 int (*set)(const struct xattr_handler
143 struct mnt_idmap *idmap,
144 struct dentry *dentry, stru
145 const void *buffer, size_t
146
147 locking rules:
148 all may block
149
150 ===== ==============
151 ops i_rwsem(inode)
152 ===== ==============
153 list: no
154 get: no
155 set: exclusive
156 ===== ==============
157
158 super_operations
159 ================
160
161 prototypes::
162
163 struct inode *(*alloc_inode)(struct su
164 void (*free_inode)(struct inode *);
165 void (*destroy_inode)(struct inode *);
166 void (*dirty_inode) (struct inode *, i
167 int (*write_inode) (struct inode *, st
168 int (*drop_inode) (struct inode *);
169 void (*evict_inode) (struct inode *);
170 void (*put_super) (struct super_block
171 int (*sync_fs)(struct super_block *sb,
172 int (*freeze_fs) (struct super_block *
173 int (*unfreeze_fs) (struct super_block
174 int (*statfs) (struct dentry *, struct
175 int (*remount_fs) (struct super_block
176 void (*umount_begin) (struct super_blo
177 int (*show_options)(struct seq_file *,
178 ssize_t (*quota_read)(struct super_blo
179 ssize_t (*quota_write)(struct super_bl
180
181 locking rules:
182 All may block [not true, see below]
183
184 ====================== ============ ======
185 ops s_umount note
186 ====================== ============ ======
187 alloc_inode:
188 free_inode: called
189 destroy_inode:
190 dirty_inode:
191 write_inode:
192 drop_inode: !!!ino
193 evict_inode:
194 put_super: write
195 sync_fs: read
196 freeze_fs: write
197 unfreeze_fs: write
198 statfs: maybe(read) (see b
199 remount_fs: write
200 umount_begin: no
201 show_options: no (names
202 quota_read: no (see b
203 quota_write: no (see b
204 ====================== ============ ======
205
206 ->statfs() has s_umount (shared) when called b
207 compat), but that's an accident of bad API; s_
208 the superblock down when we only have dev_t gi
209 identify the superblock. Everything else (sta
210 doesn't hold it when calling ->statfs() - supe
211 by resolving the pathname passed to syscall.
212
213 ->quota_read() and ->quota_write() functions a
214 be the only ones operating on the quota file b
215 dqio_sem) (unless an admin really wants to scr
216 writes to quota files with quotas on). For oth
217 see also dquot_operations section.
218
219 file_system_type
220 ================
221
222 prototypes::
223
224 struct dentry *(*mount) (struct file_s
225 const char *, void *);
226 void (*kill_sb) (struct super_block *)
227
228 locking rules:
229
230 ======= =========
231 ops may block
232 ======= =========
233 mount yes
234 kill_sb yes
235 ======= =========
236
237 ->mount() returns ERR_PTR or the root dentry;
238 on return.
239
240 ->kill_sb() takes a write-locked superblock, d
241 unlocks and drops the reference.
242
243 address_space_operations
244 ========================
245 prototypes::
246
247 int (*writepage)(struct page *page, st
248 int (*read_folio)(struct file *, struc
249 int (*writepages)(struct address_space
250 bool (*dirty_folio)(struct address_spa
251 void (*readahead)(struct readahead_con
252 int (*write_begin)(struct file *, stru
253 loff_t pos, un
254 struct folio *
255 int (*write_end)(struct file *, struct
256 loff_t pos, un
257 struct folio *
258 sector_t (*bmap)(struct address_space
259 void (*invalidate_folio) (struct folio
260 bool (*release_folio)(struct folio *,
261 void (*free_folio)(struct folio *);
262 int (*direct_IO)(struct kiocb *, struc
263 int (*migrate_folio)(struct address_sp
264 struct folio *src, enu
265 int (*launder_folio)(struct folio *);
266 bool (*is_partially_uptodate)(struct f
267 int (*error_remove_folio)(struct addre
268 int (*swap_activate)(struct swap_info_
269 int (*swap_deactivate)(struct file *);
270 int (*swap_rw)(struct kiocb *iocb, str
271
272 locking rules:
273 All except dirty_folio and free_folio
274
275 ====================== ======================
276 ops folio locked
277 ====================== ======================
278 writepage: yes, unlocks (see belo
279 read_folio: yes, unlocks
280 writepages:
281 dirty_folio: maybe
282 readahead: yes, unlocks
283 write_begin: locks the folio
284 write_end: yes, unlocks
285 bmap:
286 invalidate_folio: yes
287 release_folio: yes
288 free_folio: yes
289 direct_IO:
290 migrate_folio: yes (both)
291 launder_folio: yes
292 is_partially_uptodate: yes
293 error_remove_folio: yes
294 swap_activate: no
295 swap_deactivate: no
296 swap_rw: yes, unlocks
297 ====================== ======================
298
299 ->write_begin(), ->write_end() and ->read_foli
300 the request handler (/dev/loop).
301
302 ->read_folio() unlocks the folio, either synch
303 completion.
304
305 ->readahead() unlocks the folios that I/O is a
306
307 ->writepage() is used for two purposes: for "m
308 "sync". These are quite different operations
309 depending upon the mode.
310
311 If writepage is called for sync (wbc->sync_mod
312 it *must* start I/O against the page, even if
313 blocking on in-progress I/O.
314
315 If writepage is called for memory cleansing (s
316 WBC_SYNC_NONE) then its role is to get as much
317 possible. So writepage should try to avoid bl
318 currently-in-progress I/O.
319
320 If the filesystem is not called for "sync" and
321 would need to block against in-progress I/O to
322 against the page the filesystem should redirty
323 redirty_page_for_writepage(), then unlock the
324 This may also be done to avoid internal deadlo
325
326 If the filesystem is called for sync then it m
327 in-progress I/O and then start new I/O.
328
329 The filesystem should unlock the page synchron
330 caller, unless ->writepage() returns special W
331 value. WRITEPAGE_ACTIVATE means that page cann
332 currently, and VM should stop calling ->writep
333 time. VM does this by moving page to the head
334 name.
335
336 Unless the filesystem is going to redirty_page
337 and return zero, writepage *must* run set_page
338 followed by unlocking it. Once set_page_write
339 page, write I/O can be submitted and the write
340 end_page_writeback() once the I/O is complete.
341 filesystem must run end_page_writeback() again
342 writepage.
343
344 That is: after 2.5.12, pages which are under w
345 if the filesystem needs the page to be locked
346 the page is allowed to be unlocked at any poin
347 set_page_writeback() and end_page_writeback().
348
349 Note, failure to run either redirty_page_for_w
350 set_page_writeback()/end_page_writeback() on a
351 will leave the page itself marked clean but it
352 radix tree. This incoherency can lead to all
353 in the filesystem like having dirty inodes at
354
355 ->writepages() is used for periodic writeback
356 sync operations. The address_space should sta
357 ``*nr_to_write`` pages. ``*nr_to_write`` must
358 which is written. The address_space implement
359 pages than ``*nr_to_write`` asks for, but it s
360 If nr_to_write is NULL, all dirty pages must b
361
362 writepages should _only_ write pages which are
363 mapping->io_pages.
364
365 ->dirty_folio() is called from various places
366 the target folio is marked as needing writebac
367 truncated because either the caller holds the
368 has found the folio while holding the page tab
369 truncation.
370
371 ->bmap() is currently used by legacy ioctl() (
372 filesystems and by the swapper. The latter wil
373 keep it that way and don't breed new callers.
374
375 ->invalidate_folio() is called when the filesy
376 some or all of the buffers from the page when
377 returns zero on success. The filesystem must
378 invalidate_lock before invalidating page cache
379 path (and thus calling into ->invalidate_folio
380 cache invalidation and page cache filling func
381
382 ->release_folio() is called when the MM wants
383 folio that would invalidate the filesystem's p
384 it may be about to be removed from the address
385 is locked and not under writeback. It may be
386 is not usually used for allocation, but rather
387 filesystem may do to attempt to free the priva
388 return false to indicate that the folio's priv
389 If it returns true, it should have already rem
390 the folio. If a filesystem does not provide a
391 the pagecache will assume that private data is
392 try_to_free_buffers().
393
394 ->free_folio() is called when the kernel has d
395 from the page cache.
396
397 ->launder_folio() may be called prior to relea
398 it is still found to be dirty. It returns zero
399 cleaned, or an error value if not. Note that i
400 getting mapped back in and redirtied, it needs
401 across the entire operation.
402
403 ->swap_activate() will be called to prepare th
404 should perform any validation and preparation
405 writes can be performed with minimal memory al
406 add_swap_extent(), or the helper iomap_swapfil
407 the number of extents added. If IO should be
408 ->swap_rw(), it should set SWP_FS_OPS, otherwi
409 directly to the block device ``sis->bdev``.
410
411 ->swap_deactivate() will be called in the sys_
412 path after ->swap_activate() returned success.
413
414 ->swap_rw will be called for swap IO if SWP_FS
415
416 file_lock_operations
417 ====================
418
419 prototypes::
420
421 void (*fl_copy_lock)(struct file_lock
422 void (*fl_release_private)(struct file
423
424
425 locking rules:
426
427 =================== ============= ======
428 ops inode->i_lock may bl
429 =================== ============= ======
430 fl_copy_lock: yes no
431 fl_release_private: maybe maybe[
432 =================== ============= ======
433
434 .. [1]:
435 ->fl_release_private for flock or POSIX loc
436 to block. Leases however can still be freed
437 so fl_release_private called on a lease sho
438
439 lock_manager_operations
440 =======================
441
442 prototypes::
443
444 void (*lm_notify)(struct file_lock *);
445 int (*lm_grant)(struct file_lock *, st
446 void (*lm_break)(struct file_lock *);
447 int (*lm_change)(struct file_lock **,
448 bool (*lm_breaker_owns_lease)(struct f
449 bool (*lm_lock_expirable)(struct file_
450 void (*lm_expire_lock)(void);
451
452 locking rules:
453
454 ====================== ============= ======
455 ops flc_lock blocke
456 ====================== ============= ======
457 lm_notify: no yes
458 lm_grant: no no
459 lm_break: yes no
460 lm_change yes no
461 lm_breaker_owns_lease: yes no
462 lm_lock_expirable yes no
463 lm_expire_lock no no
464 ====================== ============= ======
465
466 buffer_head
467 ===========
468
469 prototypes::
470
471 void (*b_end_io)(struct buffer_head *b
472
473 locking rules:
474
475 called from interrupts. In other words, extrem
476 bh is locked, but that's all warranties we hav
477 highmem, fs/buffer.c, and fs/ntfs/aops.c are p
478 call this method upon the IO completion.
479
480 block_device_operations
481 =======================
482 prototypes::
483
484 int (*open) (struct block_device *, fm
485 int (*release) (struct gendisk *, fmod
486 int (*ioctl) (struct block_device *, f
487 int (*compat_ioctl) (struct block_devi
488 int (*direct_access) (struct block_dev
489 unsigned long
490 void (*unlock_native_capacity) (struct
491 int (*getgeo)(struct block_device *, s
492 void (*swap_slot_free_notify) (struct
493
494 locking rules:
495
496 ======================= ===================
497 ops open_mutex
498 ======================= ===================
499 open: yes
500 release: yes
501 ioctl: no
502 compat_ioctl: no
503 direct_access: no
504 unlock_native_capacity: no
505 getgeo: no
506 swap_slot_free_notify: no (see below)
507 ======================= ===================
508
509 swap_slot_free_notify is called with swap_lock
510 held.
511
512
513 file_operations
514 ===============
515
516 prototypes::
517
518 loff_t (*llseek) (struct file *, loff_
519 ssize_t (*read) (struct file *, char _
520 ssize_t (*write) (struct file *, const
521 ssize_t (*read_iter) (struct kiocb *,
522 ssize_t (*write_iter) (struct kiocb *,
523 int (*iopoll) (struct kiocb *kiocb, bo
524 int (*iterate_shared) (struct file *,
525 __poll_t (*poll) (struct file *, struc
526 long (*unlocked_ioctl) (struct file *,
527 long (*compat_ioctl) (struct file *, u
528 int (*mmap) (struct file *, struct vm_
529 int (*open) (struct inode *, struct fi
530 int (*flush) (struct file *);
531 int (*release) (struct inode *, struct
532 int (*fsync) (struct file *, loff_t st
533 int (*fasync) (int, struct file *, int
534 int (*lock) (struct file *, int, struc
535 unsigned long (*get_unmapped_area)(str
536 unsigned long, unsigne
537 int (*check_flags)(int);
538 int (*flock) (struct file *, int, stru
539 ssize_t (*splice_write)(struct pipe_in
540 size_t, unsigned int);
541 ssize_t (*splice_read)(struct file *,
542 size_t, unsigned int);
543 int (*setlease)(struct file *, long, s
544 long (*fallocate)(struct file *, int,
545 void (*show_fdinfo)(struct seq_file *m
546 unsigned (*mmap_capabilities)(struct f
547 ssize_t (*copy_file_range)(struct file
548 loff_t, size_t, unsign
549 loff_t (*remap_file_range)(struct file
550 struct file *file_out,
551 loff_t len, unsigned i
552 int (*fadvise)(struct file *, loff_t,
553
554 locking rules:
555 All may block.
556
557 ->llseek() locking has moved from llseek to th
558 implementations. If your fs is not using gene
559 need to acquire and release the appropriate lo
560 For many filesystems, it is probably safe to a
561 mutex or just to use i_size_read() instead.
562 Note: this does not protect the file->f_pos ag
563 since this is something the userspace has to t
564
565 ->iterate_shared() is called with i_rwsem held
566 file f_pos_lock held exclusively
567
568 ->fasync() is responsible for maintaining the
569 Most instances call fasync_helper(), which doe
570 not normally something one needs to worry abou
571 mapped to zero in the VFS layer.
572
573 ->readdir() and ->ioctl() on directories must
574 move ->readdir() to inode_operations and use a
575 ->ioctl() or kill the latter completely. One o
576 anything that resembles union-mount we won't h
577 components. And there are other reasons why th
578
579 ->read on directories probably must go away -
580 in sys_read() and friends.
581
582 ->setlease operations should call generic_setl
583 the lease within the individual filesystem to
584 operation
585
586 ->fallocate implementation must be really care
587 consistency when punching holes or performing
588 page cache contents. Usually the filesystem ne
589 truncate_inode_pages_range() to invalidate rel
590 However the filesystem usually also needs to u
591 view of file offset -> disk block mapping. Unt
592 filesystem needs to block page faults and read
593 cache contents from the disk. Since VFS acquir
594 shared mode when loading pages from disk (file
595 readahead paths), the fallocate implementation
596 prevent reloading.
597
598 ->copy_file_range and ->remap_file_range imple
599 against modifications of file data while the o
600 blocking changes through write(2) and similar
601 used. To block changes to file contents via a
602 operation, the filesystem must take mapping->i
603 with ->page_mkwrite.
604
605 dquot_operations
606 ================
607
608 prototypes::
609
610 int (*write_dquot) (struct dquot *);
611 int (*acquire_dquot) (struct dquot *);
612 int (*release_dquot) (struct dquot *);
613 int (*mark_dirty) (struct dquot *);
614 int (*write_info) (struct super_block
615
616 These operations are intended to be more or le
617 a proper locking wrt the filesystem and call t
618
619 What filesystem should expect from the generic
620
621 ============== ============ ==============
622 ops FS recursion Held locks whe
623 ============== ============ ==============
624 write_dquot: yes dqonoff_sem or
625 acquire_dquot: yes dqonoff_sem or
626 release_dquot: yes dqonoff_sem or
627 mark_dirty: no -
628 write_info: yes dqonoff_sem
629 ============== ============ ==============
630
631 FS recursion means calling ->quota_read() and
632 operations.
633
634 More details about quota locking can be found
635
636 vm_operations_struct
637 ====================
638
639 prototypes::
640
641 void (*open)(struct vm_area_struct *);
642 void (*close)(struct vm_area_struct *)
643 vm_fault_t (*fault)(struct vm_fault *)
644 vm_fault_t (*huge_fault)(struct vm_fau
645 vm_fault_t (*map_pages)(struct vm_faul
646 vm_fault_t (*page_mkwrite)(struct vm_a
647 vm_fault_t (*pfn_mkwrite)(struct vm_ar
648 int (*access)(struct vm_area_struct *,
649
650 locking rules:
651
652 ============= ========== ==============
653 ops mmap_lock PageLocked(pag
654 ============= ========== ==============
655 open: write
656 close: read/write
657 fault: read can return wit
658 huge_fault: maybe-read
659 map_pages: maybe-read
660 page_mkwrite: read can return wit
661 pfn_mkwrite: read
662 access: read
663 ============= ========== ==============
664
665 ->fault() is called when a previously not pres
666 in. The filesystem must find and return the pa
667 "pgoff" in the vm_fault structure. If it is po
668 truncated and/or invalidated, then the filesys
669 then ensure the page is not already truncated
670 subsequent truncate), and then return with VM_
671 locked. The VM will unlock the page.
672
673 ->huge_fault() is called when there is no PUD
674 gives the filesystem the opportunity to instal
675 Filesystems can also use the ->fault method to
676 so implementing this function may not be neces
677 filesystems should not call filemap_fault() fr
678 The mmap_lock may not be held when this method
679
680 ->map_pages() is called when VM asks to map ea
681 Filesystem should find and map pages associate
682 till "end_pgoff". ->map_pages() is called with
683 not block. If it's not possible to reach a pa
684 filesystem should skip it. Filesystem should u
685 page table entry. Pointer to entry associated
686 "pte" field in vm_fault structure. Pointers to
687 should be calculated relative to "pte".
688
689 ->page_mkwrite() is called when a previously r
690 writeable. The filesystem again must ensure th
691 truncate/invalidate races or races with operat
692 or ->copy_file_range, and then return with the
693 mapping->invalidate_lock is suitable for prope
694 been truncated, the filesystem should not look
695 handler, but simply return with VM_FAULT_NOPAG
696 retry the fault.
697
698 ->pfn_mkwrite() is the same as page_mkwrite bu
699 VM_PFNMAP or VM_MIXEDMAP with a page-less entr
700 VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR
701 after this call is to make the pte read-write,
702 an error.
703
704 ->access() is called when get_user_pages() fai
705 access_process_vm(), typically used to debug a
706 /proc/pid/mem or ptrace. This function is nee
707 VM_IO | VM_PFNMAP VMAs.
708
709 ----------------------------------------------
710
711 Dubious stuff
712
713 (if you break something or notice that it is b
714 - at least put it here)
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.