Loading...
1=======
2Locking
3=======
4
5The text below describes the locking rules for VFS-related methods.
6It is (believed to be) up-to-date. *Please*, if you change anything in
7prototypes or locking protocols - update this file. And update the relevant
8instances in the tree, don't leave that to maintainers of filesystems/devices/
9etc. At the very least, put the list of dubious cases in the end of this file.
10Don't turn it into log - maintainers of out-of-the-tree code are supposed to
11be able to use diff(1).
12
13Thing currently missing here: socket operations. Alexey?
14
15dentry_operations
16=================
17
18prototypes::
19
20 int (*d_revalidate)(struct dentry *, unsigned int);
21 int (*d_weak_revalidate)(struct dentry *, unsigned int);
22 int (*d_hash)(const struct dentry *, struct qstr *);
23 int (*d_compare)(const struct dentry *,
24 unsigned int, const char *, const struct qstr *);
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 inode *);
29 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
30 struct vfsmount *(*d_automount)(struct path *path);
31 int (*d_manage)(const struct path *, bool);
32 struct dentry *(*d_real)(struct dentry *, enum d_real_type type);
33
34locking rules:
35
36================== =========== ======== ============== ========
37ops rename_lock ->d_lock may block rcu-walk
38================== =========== ======== ============== ========
39d_revalidate: no no yes (ref-walk) maybe
40d_weak_revalidate: no no yes no
41d_hash no no no maybe
42d_compare: yes no no maybe
43d_delete: no yes no no
44d_init: no no yes no
45d_release: no no yes no
46d_prune: no yes no no
47d_iput: no no yes no
48d_dname: no no no no
49d_automount: no no yes no
50d_manage: no no yes (ref-walk) maybe
51d_real no no yes no
52================== =========== ======== ============== ========
53
54inode_operations
55================
56
57prototypes::
58
59 int (*create) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t, bool);
60 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
61 int (*link) (struct dentry *,struct inode *,struct dentry *);
62 int (*unlink) (struct inode *,struct dentry *);
63 int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *);
64 int (*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t);
65 int (*rmdir) (struct inode *,struct dentry *);
66 int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t);
67 int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *,
68 struct inode *, struct dentry *, unsigned int);
69 int (*readlink) (struct dentry *, char __user *,int);
70 const char *(*get_link) (struct dentry *, struct inode *, struct delayed_call *);
71 void (*truncate) (struct inode *);
72 int (*permission) (struct mnt_idmap *, struct inode *, int, unsigned int);
73 struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
74 int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *);
75 int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int);
76 ssize_t (*listxattr) (struct dentry *, char *, size_t);
77 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
78 void (*update_time)(struct inode *, struct timespec *, int);
79 int (*atomic_open)(struct inode *, struct dentry *,
80 struct file *, unsigned open_flag,
81 umode_t create_mode);
82 int (*tmpfile) (struct mnt_idmap *, struct inode *,
83 struct file *, umode_t);
84 int (*fileattr_set)(struct mnt_idmap *idmap,
85 struct dentry *dentry, struct fileattr *fa);
86 int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa);
87 struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int);
88 struct offset_ctx *(*get_offset_ctx)(struct inode *inode);
89
90locking rules:
91 all may block
92
93============== ==================================================
94ops i_rwsem(inode)
95============== ==================================================
96lookup: shared
97create: exclusive
98link: exclusive (both)
99mknod: exclusive
100symlink: exclusive
101mkdir: exclusive
102unlink: exclusive (both)
103rmdir: exclusive (both)(see below)
104rename: exclusive (both parents, some children) (see below)
105readlink: no
106get_link: no
107setattr: exclusive
108permission: no (may not block if called in rcu-walk mode)
109get_inode_acl: no
110get_acl: no
111getattr: no
112listxattr: no
113fiemap: no
114update_time: no
115atomic_open: shared (exclusive if O_CREAT is set in open flags)
116tmpfile: no
117fileattr_get: no or exclusive
118fileattr_set: exclusive
119get_offset_ctx no
120============== ==================================================
121
122
123 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
124 exclusive on victim.
125 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
126 ->unlink() and ->rename() have ->i_rwsem exclusive on all non-directories
127 involved.
128 ->rename() has ->i_rwsem exclusive on any subdirectory that changes parent.
129
130See Documentation/filesystems/directory-locking.rst for more detailed discussion
131of the locking scheme for directory operations.
132
133xattr_handler operations
134========================
135
136prototypes::
137
138 bool (*list)(struct dentry *dentry);
139 int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
140 struct inode *inode, const char *name, void *buffer,
141 size_t size);
142 int (*set)(const struct xattr_handler *handler,
143 struct mnt_idmap *idmap,
144 struct dentry *dentry, struct inode *inode, const char *name,
145 const void *buffer, size_t size, int flags);
146
147locking rules:
148 all may block
149
150===== ==============
151ops i_rwsem(inode)
152===== ==============
153list: no
154get: no
155set: exclusive
156===== ==============
157
158super_operations
159================
160
161prototypes::
162
163 struct inode *(*alloc_inode)(struct super_block *sb);
164 void (*free_inode)(struct inode *);
165 void (*destroy_inode)(struct inode *);
166 void (*dirty_inode) (struct inode *, int flags);
167 int (*write_inode) (struct inode *, struct writeback_control *wbc);
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, int wait);
172 int (*freeze_fs) (struct super_block *);
173 int (*unfreeze_fs) (struct super_block *);
174 int (*statfs) (struct dentry *, struct kstatfs *);
175 int (*remount_fs) (struct super_block *, int *, char *);
176 void (*umount_begin) (struct super_block *);
177 int (*show_options)(struct seq_file *, struct dentry *);
178 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
179 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
180
181locking rules:
182 All may block [not true, see below]
183
184====================== ============ ========================
185ops s_umount note
186====================== ============ ========================
187alloc_inode:
188free_inode: called from RCU callback
189destroy_inode:
190dirty_inode:
191write_inode:
192drop_inode: !!!inode->i_lock!!!
193evict_inode:
194put_super: write
195sync_fs: read
196freeze_fs: write
197unfreeze_fs: write
198statfs: maybe(read) (see below)
199remount_fs: write
200umount_begin: no
201show_options: no (namespace_sem)
202quota_read: no (see below)
203quota_write: no (see below)
204====================== ============ ========================
205
206->statfs() has s_umount (shared) when called by ustat(2) (native or
207compat), but that's an accident of bad API; s_umount is used to pin
208the superblock down when we only have dev_t given us by userland to
209identify the superblock. Everything else (statfs(), fstatfs(), etc.)
210doesn't hold it when calling ->statfs() - superblock is pinned down
211by resolving the pathname passed to syscall.
212
213->quota_read() and ->quota_write() functions are both guaranteed to
214be the only ones operating on the quota file by the quota code (via
215dqio_sem) (unless an admin really wants to screw up something and
216writes to quota files with quotas on). For other details about locking
217see also dquot_operations section.
218
219file_system_type
220================
221
222prototypes::
223
224 struct dentry *(*mount) (struct file_system_type *, int,
225 const char *, void *);
226 void (*kill_sb) (struct super_block *);
227
228locking rules:
229
230======= =========
231ops may block
232======= =========
233mount yes
234kill_sb yes
235======= =========
236
237->mount() returns ERR_PTR or the root dentry; its superblock should be locked
238on return.
239
240->kill_sb() takes a write-locked superblock, does all shutdown work on it,
241unlocks and drops the reference.
242
243address_space_operations
244========================
245prototypes::
246
247 int (*writepage)(struct page *page, struct writeback_control *wbc);
248 int (*read_folio)(struct file *, struct folio *);
249 int (*writepages)(struct address_space *, struct writeback_control *);
250 bool (*dirty_folio)(struct address_space *, struct folio *folio);
251 void (*readahead)(struct readahead_control *);
252 int (*write_begin)(struct file *, struct address_space *mapping,
253 loff_t pos, unsigned len,
254 struct folio **foliop, void **fsdata);
255 int (*write_end)(struct file *, struct address_space *mapping,
256 loff_t pos, unsigned len, unsigned copied,
257 struct folio *folio, void *fsdata);
258 sector_t (*bmap)(struct address_space *, sector_t);
259 void (*invalidate_folio) (struct folio *, size_t start, size_t len);
260 bool (*release_folio)(struct folio *, gfp_t);
261 void (*free_folio)(struct folio *);
262 int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
263 int (*migrate_folio)(struct address_space *, struct folio *dst,
264 struct folio *src, enum migrate_mode);
265 int (*launder_folio)(struct folio *);
266 bool (*is_partially_uptodate)(struct folio *, size_t from, size_t count);
267 int (*error_remove_folio)(struct address_space *, struct folio *);
268 int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
269 int (*swap_deactivate)(struct file *);
270 int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
271
272locking rules:
273 All except dirty_folio and free_folio may block
274
275====================== ======================== ========= ===============
276ops folio locked i_rwsem invalidate_lock
277====================== ======================== ========= ===============
278writepage: yes, unlocks (see below)
279read_folio: yes, unlocks shared
280writepages:
281dirty_folio: maybe
282readahead: yes, unlocks shared
283write_begin: locks the folio exclusive
284write_end: yes, unlocks exclusive
285bmap:
286invalidate_folio: yes exclusive
287release_folio: yes
288free_folio: yes
289direct_IO:
290migrate_folio: yes (both)
291launder_folio: yes
292is_partially_uptodate: yes
293error_remove_folio: yes
294swap_activate: no
295swap_deactivate: no
296swap_rw: yes, unlocks
297====================== ======================== ========= ===============
298
299->write_begin(), ->write_end() and ->read_folio() may be called from
300the request handler (/dev/loop).
301
302->read_folio() unlocks the folio, either synchronously or via I/O
303completion.
304
305->readahead() unlocks the folios that I/O is attempted on like ->read_folio().
306
307->writepage() is used for two purposes: for "memory cleansing" and for
308"sync". These are quite different operations and the behaviour may differ
309depending upon the mode.
310
311If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
312it *must* start I/O against the page, even if that would involve
313blocking on in-progress I/O.
314
315If writepage is called for memory cleansing (sync_mode ==
316WBC_SYNC_NONE) then its role is to get as much writeout underway as
317possible. So writepage should try to avoid blocking against
318currently-in-progress I/O.
319
320If the filesystem is not called for "sync" and it determines that it
321would need to block against in-progress I/O to be able to start new I/O
322against the page the filesystem should redirty the page with
323redirty_page_for_writepage(), then unlock the page and return zero.
324This may also be done to avoid internal deadlocks, but rarely.
325
326If the filesystem is called for sync then it must wait on any
327in-progress I/O and then start new I/O.
328
329The filesystem should unlock the page synchronously, before returning to the
330caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
331value. WRITEPAGE_ACTIVATE means that page cannot really be written out
332currently, and VM should stop calling ->writepage() on this page for some
333time. VM does this by moving page to the head of the active list, hence the
334name.
335
336Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
337and return zero, writepage *must* run set_page_writeback() against the page,
338followed by unlocking it. Once set_page_writeback() has been run against the
339page, write I/O can be submitted and the write I/O completion handler must run
340end_page_writeback() once the I/O is complete. If no I/O is submitted, the
341filesystem must run end_page_writeback() against the page before returning from
342writepage.
343
344That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
345if the filesystem needs the page to be locked during writeout, that is ok, too,
346the page is allowed to be unlocked at any point in time between the calls to
347set_page_writeback() and end_page_writeback().
348
349Note, failure to run either redirty_page_for_writepage() or the combination of
350set_page_writeback()/end_page_writeback() on a page submitted to writepage
351will leave the page itself marked clean but it will be tagged as dirty in the
352radix tree. This incoherency can lead to all sorts of hard-to-debug problems
353in the filesystem like having dirty inodes at umount and losing written data.
354
355->writepages() is used for periodic writeback and for syscall-initiated
356sync operations. The address_space should start I/O against at least
357``*nr_to_write`` pages. ``*nr_to_write`` must be decremented for each page
358which is written. The address_space implementation may write more (or less)
359pages than ``*nr_to_write`` asks for, but it should try to be reasonably close.
360If nr_to_write is NULL, all dirty pages must be written.
361
362writepages should _only_ write pages which are present on
363mapping->io_pages.
364
365->dirty_folio() is called from various places in the kernel when
366the target folio is marked as needing writeback. The folio cannot be
367truncated because either the caller holds the folio lock, or the caller
368has found the folio while holding the page table lock which will block
369truncation.
370
371->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
372filesystems and by the swapper. The latter will eventually go away. Please,
373keep it that way and don't breed new callers.
374
375->invalidate_folio() is called when the filesystem must attempt to drop
376some or all of the buffers from the page when it is being truncated. It
377returns zero on success. The filesystem must exclusively acquire
378invalidate_lock before invalidating page cache in truncate / hole punch
379path (and thus calling into ->invalidate_folio) to block races between page
380cache invalidation and page cache filling functions (fault, read, ...).
381
382->release_folio() is called when the MM wants to make a change to the
383folio that would invalidate the filesystem's private data. For example,
384it may be about to be removed from the address_space or split. The folio
385is locked and not under writeback. It may be dirty. The gfp parameter
386is not usually used for allocation, but rather to indicate what the
387filesystem may do to attempt to free the private data. The filesystem may
388return false to indicate that the folio's private data cannot be freed.
389If it returns true, it should have already removed the private data from
390the folio. If a filesystem does not provide a ->release_folio method,
391the pagecache will assume that private data is buffer_heads and call
392try_to_free_buffers().
393
394->free_folio() is called when the kernel has dropped the folio
395from the page cache.
396
397->launder_folio() may be called prior to releasing a folio if
398it is still found to be dirty. It returns zero if the folio was successfully
399cleaned, or an error value if not. Note that in order to prevent the folio
400getting mapped back in and redirtied, it needs to be kept locked
401across the entire operation.
402
403->swap_activate() will be called to prepare the given file for swap. It
404should perform any validation and preparation necessary to ensure that
405writes can be performed with minimal memory allocation. It should call
406add_swap_extent(), or the helper iomap_swapfile_activate(), and return
407the number of extents added. If IO should be submitted through
408->swap_rw(), it should set SWP_FS_OPS, otherwise IO will be submitted
409directly to the block device ``sis->bdev``.
410
411->swap_deactivate() will be called in the sys_swapoff()
412path after ->swap_activate() returned success.
413
414->swap_rw will be called for swap IO if SWP_FS_OPS was set by ->swap_activate().
415
416file_lock_operations
417====================
418
419prototypes::
420
421 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
422 void (*fl_release_private)(struct file_lock *);
423
424
425locking rules:
426
427=================== ============= =========
428ops inode->i_lock may block
429=================== ============= =========
430fl_copy_lock: yes no
431fl_release_private: maybe maybe[1]_
432=================== ============= =========
433
434.. [1]:
435 ->fl_release_private for flock or POSIX locks is currently allowed
436 to block. Leases however can still be freed while the i_lock is held and
437 so fl_release_private called on a lease should not block.
438
439lock_manager_operations
440=======================
441
442prototypes::
443
444 void (*lm_notify)(struct file_lock *); /* unblock callback */
445 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
446 void (*lm_break)(struct file_lock *); /* break_lease callback */
447 int (*lm_change)(struct file_lock **, int);
448 bool (*lm_breaker_owns_lease)(struct file_lock *);
449 bool (*lm_lock_expirable)(struct file_lock *);
450 void (*lm_expire_lock)(void);
451
452locking rules:
453
454====================== ============= ================= =========
455ops flc_lock blocked_lock_lock may block
456====================== ============= ================= =========
457lm_notify: no yes no
458lm_grant: no no no
459lm_break: yes no no
460lm_change yes no no
461lm_breaker_owns_lease: yes no no
462lm_lock_expirable yes no no
463lm_expire_lock no no yes
464====================== ============= ================= =========
465
466buffer_head
467===========
468
469prototypes::
470
471 void (*b_end_io)(struct buffer_head *bh, int uptodate);
472
473locking rules:
474
475called from interrupts. In other words, extreme care is needed here.
476bh is locked, but that's all warranties we have here. Currently only RAID1,
477highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
478call this method upon the IO completion.
479
480block_device_operations
481=======================
482prototypes::
483
484 int (*open) (struct block_device *, fmode_t);
485 int (*release) (struct gendisk *, fmode_t);
486 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
487 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
488 int (*direct_access) (struct block_device *, sector_t, void **,
489 unsigned long *);
490 void (*unlock_native_capacity) (struct gendisk *);
491 int (*getgeo)(struct block_device *, struct hd_geometry *);
492 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
493
494locking rules:
495
496======================= ===================
497ops open_mutex
498======================= ===================
499open: yes
500release: yes
501ioctl: no
502compat_ioctl: no
503direct_access: no
504unlock_native_capacity: no
505getgeo: no
506swap_slot_free_notify: no (see below)
507======================= ===================
508
509swap_slot_free_notify is called with swap_lock and sometimes the page lock
510held.
511
512
513file_operations
514===============
515
516prototypes::
517
518 loff_t (*llseek) (struct file *, loff_t, int);
519 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
520 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
521 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
522 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
523 int (*iopoll) (struct kiocb *kiocb, bool spin);
524 int (*iterate_shared) (struct file *, struct dir_context *);
525 __poll_t (*poll) (struct file *, struct poll_table_struct *);
526 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
527 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
528 int (*mmap) (struct file *, struct vm_area_struct *);
529 int (*open) (struct inode *, struct file *);
530 int (*flush) (struct file *);
531 int (*release) (struct inode *, struct file *);
532 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
533 int (*fasync) (int, struct file *, int);
534 int (*lock) (struct file *, int, struct file_lock *);
535 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
536 unsigned long, unsigned long, unsigned long);
537 int (*check_flags)(int);
538 int (*flock) (struct file *, int, struct file_lock *);
539 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
540 size_t, unsigned int);
541 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
542 size_t, unsigned int);
543 int (*setlease)(struct file *, long, struct file_lock **, void **);
544 long (*fallocate)(struct file *, int, loff_t, loff_t);
545 void (*show_fdinfo)(struct seq_file *m, struct file *f);
546 unsigned (*mmap_capabilities)(struct file *);
547 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *,
548 loff_t, size_t, unsigned int);
549 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
550 struct file *file_out, loff_t pos_out,
551 loff_t len, unsigned int remap_flags);
552 int (*fadvise)(struct file *, loff_t, loff_t, int);
553
554locking rules:
555 All may block.
556
557->llseek() locking has moved from llseek to the individual llseek
558implementations. If your fs is not using generic_file_llseek, you
559need to acquire and release the appropriate locks in your ->llseek().
560For many filesystems, it is probably safe to acquire the inode
561mutex or just to use i_size_read() instead.
562Note: this does not protect the file->f_pos against concurrent modifications
563since this is something the userspace has to take care about.
564
565->iterate_shared() is called with i_rwsem held for reading, and with the
566file f_pos_lock held exclusively
567
568->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
569Most instances call fasync_helper(), which does that maintenance, so it's
570not normally something one needs to worry about. Return values > 0 will be
571mapped to zero in the VFS layer.
572
573->readdir() and ->ioctl() on directories must be changed. Ideally we would
574move ->readdir() to inode_operations and use a separate method for directory
575->ioctl() or kill the latter completely. One of the problems is that for
576anything that resembles union-mount we won't have a struct file for all
577components. And there are other reasons why the current interface is a mess...
578
579->read on directories probably must go away - we should just enforce -EISDIR
580in sys_read() and friends.
581
582->setlease operations should call generic_setlease() before or after setting
583the lease within the individual filesystem to record the result of the
584operation
585
586->fallocate implementation must be really careful to maintain page cache
587consistency when punching holes or performing other operations that invalidate
588page cache contents. Usually the filesystem needs to call
589truncate_inode_pages_range() to invalidate relevant range of the page cache.
590However the filesystem usually also needs to update its internal (and on disk)
591view of file offset -> disk block mapping. Until this update is finished, the
592filesystem needs to block page faults and reads from reloading now-stale page
593cache contents from the disk. Since VFS acquires mapping->invalidate_lock in
594shared mode when loading pages from disk (filemap_fault(), filemap_read(),
595readahead paths), the fallocate implementation must take the invalidate_lock to
596prevent reloading.
597
598->copy_file_range and ->remap_file_range implementations need to serialize
599against modifications of file data while the operation is running. For
600blocking changes through write(2) and similar operations inode->i_rwsem can be
601used. To block changes to file contents via a memory mapping during the
602operation, the filesystem must take mapping->invalidate_lock to coordinate
603with ->page_mkwrite.
604
605dquot_operations
606================
607
608prototypes::
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 *, int);
615
616These operations are intended to be more or less wrapping functions that ensure
617a proper locking wrt the filesystem and call the generic quota operations.
618
619What filesystem should expect from the generic quota functions:
620
621============== ============ =========================
622ops FS recursion Held locks when called
623============== ============ =========================
624write_dquot: yes dqonoff_sem or dqptr_sem
625acquire_dquot: yes dqonoff_sem or dqptr_sem
626release_dquot: yes dqonoff_sem or dqptr_sem
627mark_dirty: no -
628write_info: yes dqonoff_sem
629============== ============ =========================
630
631FS recursion means calling ->quota_read() and ->quota_write() from superblock
632operations.
633
634More details about quota locking can be found in fs/dquot.c.
635
636vm_operations_struct
637====================
638
639prototypes::
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_fault *, unsigned int order);
645 vm_fault_t (*map_pages)(struct vm_fault *, pgoff_t start, pgoff_t end);
646 vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
647 vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
648 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
649
650locking rules:
651
652============= ========== ===========================
653ops mmap_lock PageLocked(page)
654============= ========== ===========================
655open: write
656close: read/write
657fault: read can return with page locked
658huge_fault: maybe-read
659map_pages: maybe-read
660page_mkwrite: read can return with page locked
661pfn_mkwrite: read
662access: read
663============= ========== ===========================
664
665->fault() is called when a previously not present pte is about to be faulted
666in. The filesystem must find and return the page associated with the passed in
667"pgoff" in the vm_fault structure. If it is possible that the page may be
668truncated and/or invalidated, then the filesystem must lock invalidate_lock,
669then ensure the page is not already truncated (invalidate_lock will block
670subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
671locked. The VM will unlock the page.
672
673->huge_fault() is called when there is no PUD or PMD entry present. This
674gives the filesystem the opportunity to install a PUD or PMD sized page.
675Filesystems can also use the ->fault method to return a PMD sized page,
676so implementing this function may not be necessary. In particular,
677filesystems should not call filemap_fault() from ->huge_fault().
678The mmap_lock may not be held when this method is called.
679
680->map_pages() is called when VM asks to map easy accessible pages.
681Filesystem should find and map pages associated with offsets from "start_pgoff"
682till "end_pgoff". ->map_pages() is called with the RCU lock held and must
683not block. If it's not possible to reach a page without blocking,
684filesystem should skip it. Filesystem should use set_pte_range() to setup
685page table entry. Pointer to entry associated with the page is passed in
686"pte" field in vm_fault structure. Pointers to entries for other offsets
687should be calculated relative to "pte".
688
689->page_mkwrite() is called when a previously read-only pte is about to become
690writeable. The filesystem again must ensure that there are no
691truncate/invalidate races or races with operations such as ->remap_file_range
692or ->copy_file_range, and then return with the page locked. Usually
693mapping->invalidate_lock is suitable for proper serialization. If the page has
694been truncated, the filesystem should not look up a new page like the ->fault()
695handler, but simply return with VM_FAULT_NOPAGE, which will cause the VM to
696retry the fault.
697
698->pfn_mkwrite() is the same as page_mkwrite but when the pte is
699VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
700VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
701after this call is to make the pte read-write, unless pfn_mkwrite returns
702an error.
703
704->access() is called when get_user_pages() fails in
705access_process_vm(), typically used to debug a process through
706/proc/pid/mem or ptrace. This function is needed only for
707VM_IO | VM_PFNMAP VMAs.
708
709--------------------------------------------------------------------------------
710
711 Dubious stuff
712
713(if you break something or notice that it is broken and do not fix it yourself
714- at least put it here)
1=======
2Locking
3=======
4
5The text below describes the locking rules for VFS-related methods.
6It is (believed to be) up-to-date. *Please*, if you change anything in
7prototypes or locking protocols - update this file. And update the relevant
8instances in the tree, don't leave that to maintainers of filesystems/devices/
9etc. At the very least, put the list of dubious cases in the end of this file.
10Don't turn it into log - maintainers of out-of-the-tree code are supposed to
11be able to use diff(1).
12
13Thing currently missing here: socket operations. Alexey?
14
15dentry_operations
16=================
17
18prototypes::
19
20 int (*d_revalidate)(struct dentry *, unsigned int);
21 int (*d_weak_revalidate)(struct dentry *, unsigned int);
22 int (*d_hash)(const struct dentry *, struct qstr *);
23 int (*d_compare)(const struct dentry *,
24 unsigned int, const char *, const struct qstr *);
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 inode *);
29 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
30 struct vfsmount *(*d_automount)(struct path *path);
31 int (*d_manage)(const struct path *, bool);
32 struct dentry *(*d_real)(struct dentry *, const struct inode *);
33
34locking rules:
35
36================== =========== ======== ============== ========
37ops rename_lock ->d_lock may block rcu-walk
38================== =========== ======== ============== ========
39d_revalidate: no no yes (ref-walk) maybe
40d_weak_revalidate: no no yes no
41d_hash no no no maybe
42d_compare: yes no no maybe
43d_delete: no yes no no
44d_init: no no yes no
45d_release: no no yes no
46d_prune: no yes no no
47d_iput: no no yes no
48d_dname: no no no no
49d_automount: no no yes no
50d_manage: no no yes (ref-walk) maybe
51d_real no no yes no
52================== =========== ======== ============== ========
53
54inode_operations
55================
56
57prototypes::
58
59 int (*create) (struct inode *,struct dentry *,umode_t, bool);
60 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
61 int (*link) (struct dentry *,struct inode *,struct dentry *);
62 int (*unlink) (struct inode *,struct dentry *);
63 int (*symlink) (struct inode *,struct dentry *,const char *);
64 int (*mkdir) (struct inode *,struct dentry *,umode_t);
65 int (*rmdir) (struct inode *,struct dentry *);
66 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
67 int (*rename) (struct inode *, struct dentry *,
68 struct inode *, struct dentry *, unsigned int);
69 int (*readlink) (struct dentry *, char __user *,int);
70 const char *(*get_link) (struct dentry *, struct inode *, struct delayed_call *);
71 void (*truncate) (struct inode *);
72 int (*permission) (struct inode *, int, unsigned int);
73 struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
74 int (*setattr) (struct dentry *, struct iattr *);
75 int (*getattr) (const struct path *, struct kstat *, u32, unsigned int);
76 ssize_t (*listxattr) (struct dentry *, char *, size_t);
77 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
78 void (*update_time)(struct inode *, struct timespec *, int);
79 int (*atomic_open)(struct inode *, struct dentry *,
80 struct file *, unsigned open_flag,
81 umode_t create_mode);
82 int (*tmpfile) (struct user_namespace *, struct inode *,
83 struct file *, umode_t);
84 int (*fileattr_set)(struct user_namespace *mnt_userns,
85 struct dentry *dentry, struct fileattr *fa);
86 int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa);
87 struct posix_acl * (*get_acl)(struct user_namespace *, struct dentry *, int);
88
89locking rules:
90 all may block
91
92============== =============================================
93ops i_rwsem(inode)
94============== =============================================
95lookup: shared
96create: exclusive
97link: exclusive (both)
98mknod: exclusive
99symlink: exclusive
100mkdir: exclusive
101unlink: exclusive (both)
102rmdir: exclusive (both)(see below)
103rename: exclusive (all) (see below)
104readlink: no
105get_link: no
106setattr: exclusive
107permission: no (may not block if called in rcu-walk mode)
108get_inode_acl: no
109get_acl: no
110getattr: no
111listxattr: no
112fiemap: no
113update_time: no
114atomic_open: shared (exclusive if O_CREAT is set in open flags)
115tmpfile: no
116fileattr_get: no or exclusive
117fileattr_set: exclusive
118============== =============================================
119
120
121 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
122 exclusive on victim.
123 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
124
125See Documentation/filesystems/directory-locking.rst for more detailed discussion
126of the locking scheme for directory operations.
127
128xattr_handler operations
129========================
130
131prototypes::
132
133 bool (*list)(struct dentry *dentry);
134 int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
135 struct inode *inode, const char *name, void *buffer,
136 size_t size);
137 int (*set)(const struct xattr_handler *handler,
138 struct user_namespace *mnt_userns,
139 struct dentry *dentry, struct inode *inode, const char *name,
140 const void *buffer, size_t size, int flags);
141
142locking rules:
143 all may block
144
145===== ==============
146ops i_rwsem(inode)
147===== ==============
148list: no
149get: no
150set: exclusive
151===== ==============
152
153super_operations
154================
155
156prototypes::
157
158 struct inode *(*alloc_inode)(struct super_block *sb);
159 void (*free_inode)(struct inode *);
160 void (*destroy_inode)(struct inode *);
161 void (*dirty_inode) (struct inode *, int flags);
162 int (*write_inode) (struct inode *, struct writeback_control *wbc);
163 int (*drop_inode) (struct inode *);
164 void (*evict_inode) (struct inode *);
165 void (*put_super) (struct super_block *);
166 int (*sync_fs)(struct super_block *sb, int wait);
167 int (*freeze_fs) (struct super_block *);
168 int (*unfreeze_fs) (struct super_block *);
169 int (*statfs) (struct dentry *, struct kstatfs *);
170 int (*remount_fs) (struct super_block *, int *, char *);
171 void (*umount_begin) (struct super_block *);
172 int (*show_options)(struct seq_file *, struct dentry *);
173 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
174 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
175
176locking rules:
177 All may block [not true, see below]
178
179====================== ============ ========================
180ops s_umount note
181====================== ============ ========================
182alloc_inode:
183free_inode: called from RCU callback
184destroy_inode:
185dirty_inode:
186write_inode:
187drop_inode: !!!inode->i_lock!!!
188evict_inode:
189put_super: write
190sync_fs: read
191freeze_fs: write
192unfreeze_fs: write
193statfs: maybe(read) (see below)
194remount_fs: write
195umount_begin: no
196show_options: no (namespace_sem)
197quota_read: no (see below)
198quota_write: no (see below)
199====================== ============ ========================
200
201->statfs() has s_umount (shared) when called by ustat(2) (native or
202compat), but that's an accident of bad API; s_umount is used to pin
203the superblock down when we only have dev_t given us by userland to
204identify the superblock. Everything else (statfs(), fstatfs(), etc.)
205doesn't hold it when calling ->statfs() - superblock is pinned down
206by resolving the pathname passed to syscall.
207
208->quota_read() and ->quota_write() functions are both guaranteed to
209be the only ones operating on the quota file by the quota code (via
210dqio_sem) (unless an admin really wants to screw up something and
211writes to quota files with quotas on). For other details about locking
212see also dquot_operations section.
213
214file_system_type
215================
216
217prototypes::
218
219 struct dentry *(*mount) (struct file_system_type *, int,
220 const char *, void *);
221 void (*kill_sb) (struct super_block *);
222
223locking rules:
224
225======= =========
226ops may block
227======= =========
228mount yes
229kill_sb yes
230======= =========
231
232->mount() returns ERR_PTR or the root dentry; its superblock should be locked
233on return.
234
235->kill_sb() takes a write-locked superblock, does all shutdown work on it,
236unlocks and drops the reference.
237
238address_space_operations
239========================
240prototypes::
241
242 int (*writepage)(struct page *page, struct writeback_control *wbc);
243 int (*read_folio)(struct file *, struct folio *);
244 int (*writepages)(struct address_space *, struct writeback_control *);
245 bool (*dirty_folio)(struct address_space *, struct folio *folio);
246 void (*readahead)(struct readahead_control *);
247 int (*write_begin)(struct file *, struct address_space *mapping,
248 loff_t pos, unsigned len,
249 struct page **pagep, void **fsdata);
250 int (*write_end)(struct file *, struct address_space *mapping,
251 loff_t pos, unsigned len, unsigned copied,
252 struct page *page, void *fsdata);
253 sector_t (*bmap)(struct address_space *, sector_t);
254 void (*invalidate_folio) (struct folio *, size_t start, size_t len);
255 bool (*release_folio)(struct folio *, gfp_t);
256 void (*free_folio)(struct folio *);
257 int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
258 int (*migrate_folio)(struct address_space *, struct folio *dst,
259 struct folio *src, enum migrate_mode);
260 int (*launder_folio)(struct folio *);
261 bool (*is_partially_uptodate)(struct folio *, size_t from, size_t count);
262 int (*error_remove_page)(struct address_space *, struct page *);
263 int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
264 int (*swap_deactivate)(struct file *);
265 int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
266
267locking rules:
268 All except dirty_folio and free_folio may block
269
270====================== ======================== ========= ===============
271ops folio locked i_rwsem invalidate_lock
272====================== ======================== ========= ===============
273writepage: yes, unlocks (see below)
274read_folio: yes, unlocks shared
275writepages:
276dirty_folio: maybe
277readahead: yes, unlocks shared
278write_begin: locks the page exclusive
279write_end: yes, unlocks exclusive
280bmap:
281invalidate_folio: yes exclusive
282release_folio: yes
283free_folio: yes
284direct_IO:
285migrate_folio: yes (both)
286launder_folio: yes
287is_partially_uptodate: yes
288error_remove_page: yes
289swap_activate: no
290swap_deactivate: no
291swap_rw: yes, unlocks
292====================== ======================== ========= ===============
293
294->write_begin(), ->write_end() and ->read_folio() may be called from
295the request handler (/dev/loop).
296
297->read_folio() unlocks the folio, either synchronously or via I/O
298completion.
299
300->readahead() unlocks the folios that I/O is attempted on like ->read_folio().
301
302->writepage() is used for two purposes: for "memory cleansing" and for
303"sync". These are quite different operations and the behaviour may differ
304depending upon the mode.
305
306If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
307it *must* start I/O against the page, even if that would involve
308blocking on in-progress I/O.
309
310If writepage is called for memory cleansing (sync_mode ==
311WBC_SYNC_NONE) then its role is to get as much writeout underway as
312possible. So writepage should try to avoid blocking against
313currently-in-progress I/O.
314
315If the filesystem is not called for "sync" and it determines that it
316would need to block against in-progress I/O to be able to start new I/O
317against the page the filesystem should redirty the page with
318redirty_page_for_writepage(), then unlock the page and return zero.
319This may also be done to avoid internal deadlocks, but rarely.
320
321If the filesystem is called for sync then it must wait on any
322in-progress I/O and then start new I/O.
323
324The filesystem should unlock the page synchronously, before returning to the
325caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
326value. WRITEPAGE_ACTIVATE means that page cannot really be written out
327currently, and VM should stop calling ->writepage() on this page for some
328time. VM does this by moving page to the head of the active list, hence the
329name.
330
331Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
332and return zero, writepage *must* run set_page_writeback() against the page,
333followed by unlocking it. Once set_page_writeback() has been run against the
334page, write I/O can be submitted and the write I/O completion handler must run
335end_page_writeback() once the I/O is complete. If no I/O is submitted, the
336filesystem must run end_page_writeback() against the page before returning from
337writepage.
338
339That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
340if the filesystem needs the page to be locked during writeout, that is ok, too,
341the page is allowed to be unlocked at any point in time between the calls to
342set_page_writeback() and end_page_writeback().
343
344Note, failure to run either redirty_page_for_writepage() or the combination of
345set_page_writeback()/end_page_writeback() on a page submitted to writepage
346will leave the page itself marked clean but it will be tagged as dirty in the
347radix tree. This incoherency can lead to all sorts of hard-to-debug problems
348in the filesystem like having dirty inodes at umount and losing written data.
349
350->writepages() is used for periodic writeback and for syscall-initiated
351sync operations. The address_space should start I/O against at least
352``*nr_to_write`` pages. ``*nr_to_write`` must be decremented for each page
353which is written. The address_space implementation may write more (or less)
354pages than ``*nr_to_write`` asks for, but it should try to be reasonably close.
355If nr_to_write is NULL, all dirty pages must be written.
356
357writepages should _only_ write pages which are present on
358mapping->io_pages.
359
360->dirty_folio() is called from various places in the kernel when
361the target folio is marked as needing writeback. The folio cannot be
362truncated because either the caller holds the folio lock, or the caller
363has found the folio while holding the page table lock which will block
364truncation.
365
366->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
367filesystems and by the swapper. The latter will eventually go away. Please,
368keep it that way and don't breed new callers.
369
370->invalidate_folio() is called when the filesystem must attempt to drop
371some or all of the buffers from the page when it is being truncated. It
372returns zero on success. The filesystem must exclusively acquire
373invalidate_lock before invalidating page cache in truncate / hole punch
374path (and thus calling into ->invalidate_folio) to block races between page
375cache invalidation and page cache filling functions (fault, read, ...).
376
377->release_folio() is called when the kernel is about to try to drop the
378buffers from the folio in preparation for freeing it. It returns false to
379indicate that the buffers are (or may be) freeable. If ->release_folio is
380NULL, the kernel assumes that the fs has no private interest in the buffers.
381
382->free_folio() is called when the kernel has dropped the folio
383from the page cache.
384
385->launder_folio() may be called prior to releasing a folio if
386it is still found to be dirty. It returns zero if the folio was successfully
387cleaned, or an error value if not. Note that in order to prevent the folio
388getting mapped back in and redirtied, it needs to be kept locked
389across the entire operation.
390
391->swap_activate() will be called to prepare the given file for swap. It
392should perform any validation and preparation necessary to ensure that
393writes can be performed with minimal memory allocation. It should call
394add_swap_extent(), or the helper iomap_swapfile_activate(), and return
395the number of extents added. If IO should be submitted through
396->swap_rw(), it should set SWP_FS_OPS, otherwise IO will be submitted
397directly to the block device ``sis->bdev``.
398
399->swap_deactivate() will be called in the sys_swapoff()
400path after ->swap_activate() returned success.
401
402->swap_rw will be called for swap IO if SWP_FS_OPS was set by ->swap_activate().
403
404file_lock_operations
405====================
406
407prototypes::
408
409 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
410 void (*fl_release_private)(struct file_lock *);
411
412
413locking rules:
414
415=================== ============= =========
416ops inode->i_lock may block
417=================== ============= =========
418fl_copy_lock: yes no
419fl_release_private: maybe maybe[1]_
420=================== ============= =========
421
422.. [1]:
423 ->fl_release_private for flock or POSIX locks is currently allowed
424 to block. Leases however can still be freed while the i_lock is held and
425 so fl_release_private called on a lease should not block.
426
427lock_manager_operations
428=======================
429
430prototypes::
431
432 void (*lm_notify)(struct file_lock *); /* unblock callback */
433 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
434 void (*lm_break)(struct file_lock *); /* break_lease callback */
435 int (*lm_change)(struct file_lock **, int);
436 bool (*lm_breaker_owns_lease)(struct file_lock *);
437 bool (*lm_lock_expirable)(struct file_lock *);
438 void (*lm_expire_lock)(void);
439
440locking rules:
441
442====================== ============= ================= =========
443ops flc_lock blocked_lock_lock may block
444====================== ============= ================= =========
445lm_notify: no yes no
446lm_grant: no no no
447lm_break: yes no no
448lm_change yes no no
449lm_breaker_owns_lease: yes no no
450lm_lock_expirable yes no no
451lm_expire_lock no no yes
452====================== ============= ================= =========
453
454buffer_head
455===========
456
457prototypes::
458
459 void (*b_end_io)(struct buffer_head *bh, int uptodate);
460
461locking rules:
462
463called from interrupts. In other words, extreme care is needed here.
464bh is locked, but that's all warranties we have here. Currently only RAID1,
465highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
466call this method upon the IO completion.
467
468block_device_operations
469=======================
470prototypes::
471
472 int (*open) (struct block_device *, fmode_t);
473 int (*release) (struct gendisk *, fmode_t);
474 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
475 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
476 int (*direct_access) (struct block_device *, sector_t, void **,
477 unsigned long *);
478 void (*unlock_native_capacity) (struct gendisk *);
479 int (*getgeo)(struct block_device *, struct hd_geometry *);
480 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
481
482locking rules:
483
484======================= ===================
485ops open_mutex
486======================= ===================
487open: yes
488release: yes
489ioctl: no
490compat_ioctl: no
491direct_access: no
492unlock_native_capacity: no
493getgeo: no
494swap_slot_free_notify: no (see below)
495======================= ===================
496
497swap_slot_free_notify is called with swap_lock and sometimes the page lock
498held.
499
500
501file_operations
502===============
503
504prototypes::
505
506 loff_t (*llseek) (struct file *, loff_t, int);
507 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
508 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
509 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
510 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
511 int (*iopoll) (struct kiocb *kiocb, bool spin);
512 int (*iterate) (struct file *, struct dir_context *);
513 int (*iterate_shared) (struct file *, struct dir_context *);
514 __poll_t (*poll) (struct file *, struct poll_table_struct *);
515 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
516 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
517 int (*mmap) (struct file *, struct vm_area_struct *);
518 int (*open) (struct inode *, struct file *);
519 int (*flush) (struct file *);
520 int (*release) (struct inode *, struct file *);
521 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
522 int (*fasync) (int, struct file *, int);
523 int (*lock) (struct file *, int, struct file_lock *);
524 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
525 loff_t *, int);
526 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
527 unsigned long, unsigned long, unsigned long);
528 int (*check_flags)(int);
529 int (*flock) (struct file *, int, struct file_lock *);
530 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
531 size_t, unsigned int);
532 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
533 size_t, unsigned int);
534 int (*setlease)(struct file *, long, struct file_lock **, void **);
535 long (*fallocate)(struct file *, int, loff_t, loff_t);
536 void (*show_fdinfo)(struct seq_file *m, struct file *f);
537 unsigned (*mmap_capabilities)(struct file *);
538 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *,
539 loff_t, size_t, unsigned int);
540 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
541 struct file *file_out, loff_t pos_out,
542 loff_t len, unsigned int remap_flags);
543 int (*fadvise)(struct file *, loff_t, loff_t, int);
544
545locking rules:
546 All may block.
547
548->llseek() locking has moved from llseek to the individual llseek
549implementations. If your fs is not using generic_file_llseek, you
550need to acquire and release the appropriate locks in your ->llseek().
551For many filesystems, it is probably safe to acquire the inode
552mutex or just to use i_size_read() instead.
553Note: this does not protect the file->f_pos against concurrent modifications
554since this is something the userspace has to take care about.
555
556->iterate() is called with i_rwsem exclusive.
557
558->iterate_shared() is called with i_rwsem at least shared.
559
560->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
561Most instances call fasync_helper(), which does that maintenance, so it's
562not normally something one needs to worry about. Return values > 0 will be
563mapped to zero in the VFS layer.
564
565->readdir() and ->ioctl() on directories must be changed. Ideally we would
566move ->readdir() to inode_operations and use a separate method for directory
567->ioctl() or kill the latter completely. One of the problems is that for
568anything that resembles union-mount we won't have a struct file for all
569components. And there are other reasons why the current interface is a mess...
570
571->read on directories probably must go away - we should just enforce -EISDIR
572in sys_read() and friends.
573
574->setlease operations should call generic_setlease() before or after setting
575the lease within the individual filesystem to record the result of the
576operation
577
578->fallocate implementation must be really careful to maintain page cache
579consistency when punching holes or performing other operations that invalidate
580page cache contents. Usually the filesystem needs to call
581truncate_inode_pages_range() to invalidate relevant range of the page cache.
582However the filesystem usually also needs to update its internal (and on disk)
583view of file offset -> disk block mapping. Until this update is finished, the
584filesystem needs to block page faults and reads from reloading now-stale page
585cache contents from the disk. Since VFS acquires mapping->invalidate_lock in
586shared mode when loading pages from disk (filemap_fault(), filemap_read(),
587readahead paths), the fallocate implementation must take the invalidate_lock to
588prevent reloading.
589
590->copy_file_range and ->remap_file_range implementations need to serialize
591against modifications of file data while the operation is running. For
592blocking changes through write(2) and similar operations inode->i_rwsem can be
593used. To block changes to file contents via a memory mapping during the
594operation, the filesystem must take mapping->invalidate_lock to coordinate
595with ->page_mkwrite.
596
597dquot_operations
598================
599
600prototypes::
601
602 int (*write_dquot) (struct dquot *);
603 int (*acquire_dquot) (struct dquot *);
604 int (*release_dquot) (struct dquot *);
605 int (*mark_dirty) (struct dquot *);
606 int (*write_info) (struct super_block *, int);
607
608These operations are intended to be more or less wrapping functions that ensure
609a proper locking wrt the filesystem and call the generic quota operations.
610
611What filesystem should expect from the generic quota functions:
612
613============== ============ =========================
614ops FS recursion Held locks when called
615============== ============ =========================
616write_dquot: yes dqonoff_sem or dqptr_sem
617acquire_dquot: yes dqonoff_sem or dqptr_sem
618release_dquot: yes dqonoff_sem or dqptr_sem
619mark_dirty: no -
620write_info: yes dqonoff_sem
621============== ============ =========================
622
623FS recursion means calling ->quota_read() and ->quota_write() from superblock
624operations.
625
626More details about quota locking can be found in fs/dquot.c.
627
628vm_operations_struct
629====================
630
631prototypes::
632
633 void (*open)(struct vm_area_struct*);
634 void (*close)(struct vm_area_struct*);
635 vm_fault_t (*fault)(struct vm_area_struct*, struct vm_fault *);
636 vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
637 vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
638 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
639
640locking rules:
641
642============= ========= ===========================
643ops mmap_lock PageLocked(page)
644============= ========= ===========================
645open: yes
646close: yes
647fault: yes can return with page locked
648map_pages: yes
649page_mkwrite: yes can return with page locked
650pfn_mkwrite: yes
651access: yes
652============= ========= ===========================
653
654->fault() is called when a previously not present pte is about to be faulted
655in. The filesystem must find and return the page associated with the passed in
656"pgoff" in the vm_fault structure. If it is possible that the page may be
657truncated and/or invalidated, then the filesystem must lock invalidate_lock,
658then ensure the page is not already truncated (invalidate_lock will block
659subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
660locked. The VM will unlock the page.
661
662->map_pages() is called when VM asks to map easy accessible pages.
663Filesystem should find and map pages associated with offsets from "start_pgoff"
664till "end_pgoff". ->map_pages() is called with page table locked and must
665not block. If it's not possible to reach a page without blocking,
666filesystem should skip it. Filesystem should use do_set_pte() to setup
667page table entry. Pointer to entry associated with the page is passed in
668"pte" field in vm_fault structure. Pointers to entries for other offsets
669should be calculated relative to "pte".
670
671->page_mkwrite() is called when a previously read-only pte is about to become
672writeable. The filesystem again must ensure that there are no
673truncate/invalidate races or races with operations such as ->remap_file_range
674or ->copy_file_range, and then return with the page locked. Usually
675mapping->invalidate_lock is suitable for proper serialization. If the page has
676been truncated, the filesystem should not look up a new page like the ->fault()
677handler, but simply return with VM_FAULT_NOPAGE, which will cause the VM to
678retry the fault.
679
680->pfn_mkwrite() is the same as page_mkwrite but when the pte is
681VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
682VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
683after this call is to make the pte read-write, unless pfn_mkwrite returns
684an error.
685
686->access() is called when get_user_pages() fails in
687access_process_vm(), typically used to debug a process through
688/proc/pid/mem or ptrace. This function is needed only for
689VM_IO | VM_PFNMAP VMAs.
690
691--------------------------------------------------------------------------------
692
693 Dubious stuff
694
695(if you break something or notice that it is broken and do not fix it yourself
696- at least put it here)