blob: b30753cbf4311641ebd0ffe62735de05b452e77a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001 The text below describes the locking rules for VFS-related methods.
2It is (believed to be) up-to-date. *Please*, if you change anything in
3prototypes or locking protocols - update this file. And update the relevant
4instances in the tree, don't leave that to maintainers of filesystems/devices/
5etc. At the very least, put the list of dubious cases in the end of this file.
6Don't turn it into log - maintainers of out-of-the-tree code are supposed to
7be able to use diff(1).
8 Thing currently missing here: socket operations. Alexey?
9
10--------------------------- dentry_operations --------------------------
11prototypes:
Al Viro0b728e12012-06-10 16:03:43 -040012 int (*d_revalidate)(struct dentry *, unsigned int);
Jeff Laytonecf3d1f2013-02-20 11:19:05 -050013 int (*d_weak_revalidate)(struct dentry *, unsigned int);
Linus Torvaldsda53be12013-05-21 15:22:44 -070014 int (*d_hash)(const struct dentry *, struct qstr *);
15 int (*d_compare)(const struct dentry *, const struct dentry *,
Nick Piggin621e1552011-01-07 17:49:27 +110016 unsigned int, const char *, const struct qstr *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 int (*d_delete)(struct dentry *);
18 void (*d_release)(struct dentry *);
19 void (*d_iput)(struct dentry *, struct inode *);
Eric Dumazetc23fbb62007-05-08 00:26:18 -070020 char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
David Howells9875cf82011-01-14 18:45:21 +000021 struct vfsmount *(*d_automount)(struct path *path);
David Howellscc53ce52011-01-14 18:45:26 +000022 int (*d_manage)(struct dentry *, bool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24locking rules:
Nick Piggin34286d62011-01-07 17:49:57 +110025 rename_lock ->d_lock may block rcu-walk
26d_revalidate: no no yes (ref-walk) maybe
Jeff Laytonecf3d1f2013-02-20 11:19:05 -050027d_weak_revalidate:no no yes no
Nick Piggin34286d62011-01-07 17:49:57 +110028d_hash no no no maybe
29d_compare: yes no no maybe
30d_delete: no yes no no
31d_release: no no yes no
Sage Weilf0023bc2011-10-28 10:02:42 -070032d_prune: no yes no no
Nick Piggin34286d62011-01-07 17:49:57 +110033d_iput: no no yes no
34d_dname: no no no no
David Howells9875cf82011-01-14 18:45:21 +000035d_automount: no no yes no
David Howellsab909112011-01-14 18:46:51 +000036d_manage: no no yes (ref-walk) maybe
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38--------------------------- inode_operations ---------------------------
39prototypes:
Al Viroebfc3b42012-06-10 18:05:36 -040040 int (*create) (struct inode *,struct dentry *,umode_t, bool);
Al Viro00cd8dd2012-06-10 17:13:09 -040041 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 int (*link) (struct dentry *,struct inode *,struct dentry *);
43 int (*unlink) (struct inode *,struct dentry *);
44 int (*symlink) (struct inode *,struct dentry *,const char *);
Al Viro18bb1db2011-07-26 01:41:39 -040045 int (*mkdir) (struct inode *,struct dentry *,umode_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 int (*rmdir) (struct inode *,struct dentry *);
Al Viro1a67aaf2011-07-26 01:52:52 -040047 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 int (*rename) (struct inode *, struct dentry *,
49 struct inode *, struct dentry *);
Miklos Szeredi520c8b12014-04-01 17:08:42 +020050 int (*rename2) (struct inode *, struct dentry *,
51 struct inode *, struct dentry *, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 int (*readlink) (struct dentry *, char __user *,int);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +010053 void * (*follow_link) (struct dentry *, struct nameidata *);
54 void (*put_link) (struct dentry *, struct nameidata *, void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 void (*truncate) (struct inode *);
Nick Pigginb74c79e2011-01-07 17:49:58 +110056 int (*permission) (struct inode *, int, unsigned int);
Christoph Hellwig4e34e712011-07-23 17:37:31 +020057 int (*get_acl)(struct inode *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 int (*setattr) (struct dentry *, struct iattr *);
59 int (*getattr) (struct vfsmount *, struct dentry *, struct kstat *);
60 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
61 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
62 ssize_t (*listxattr) (struct dentry *, char *, size_t);
63 int (*removexattr) (struct dentry *, const char *);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +010064 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
Josef Bacikc3b2da32012-03-26 09:59:21 -040065 void (*update_time)(struct inode *, struct timespec *, int);
Al Virod9585272012-06-22 12:39:14 +040066 int (*atomic_open)(struct inode *, struct dentry *,
Al Viro30d90492012-06-22 12:40:19 +040067 struct file *, unsigned open_flag,
Al Viro47237682012-06-10 05:01:45 -040068 umode_t create_mode, int *opened);
Al Viro48bde8d2013-07-03 16:19:23 +040069 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
Miklos Szeredi4aa7c632014-10-24 00:14:35 +020070 int (*dentry_open)(struct dentry *, struct file *, const struct cred *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72locking rules:
Christoph Hellwigb83be6f2010-12-16 12:04:54 +010073 all may block
Artem Bityutskiya7bc02f2007-05-09 07:53:16 +020074 i_mutex(inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075lookup: yes
76create: yes
77link: yes (both)
78mknod: yes
79symlink: yes
80mkdir: yes
81unlink: yes (both)
82rmdir: yes (both) (see below)
83rename: yes (all) (see below)
Miklos Szeredi520c8b12014-04-01 17:08:42 +020084rename2: yes (all) (see below)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085readlink: no
86follow_link: no
Christoph Hellwigb83be6f2010-12-16 12:04:54 +010087put_link: no
Linus Torvalds1da177e2005-04-16 15:20:36 -070088setattr: yes
Nick Pigginb74c79e2011-01-07 17:49:58 +110089permission: no (may not block if called in rcu-walk mode)
Christoph Hellwig4e34e712011-07-23 17:37:31 +020090get_acl: no
Linus Torvalds1da177e2005-04-16 15:20:36 -070091getattr: no
92setxattr: yes
93getxattr: no
94listxattr: no
95removexattr: yes
Christoph Hellwigb83be6f2010-12-16 12:04:54 +010096fiemap: no
Josef Bacikc3b2da32012-03-26 09:59:21 -040097update_time: no
Miklos Szeredid18e9002012-06-05 15:10:17 +020098atomic_open: yes
Al Viro48bde8d2013-07-03 16:19:23 +040099tmpfile: no
Miklos Szeredi4aa7c632014-10-24 00:14:35 +0200100dentry_open: no
Josef Bacikc3b2da32012-03-26 09:59:21 -0400101
Artem Bityutskiya7bc02f2007-05-09 07:53:16 +0200102 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103victim.
Miklos Szeredi520c8b12014-04-01 17:08:42 +0200104 cross-directory ->rename() and rename2() has (per-superblock)
105->s_vfs_rename_sem.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106
107See Documentation/filesystems/directory-locking for more detailed discussion
108of the locking scheme for directory operations.
109
110--------------------------- super_operations ---------------------------
111prototypes:
112 struct inode *(*alloc_inode)(struct super_block *sb);
113 void (*destroy_inode)(struct inode *);
Christoph Hellwigaa385722011-05-27 06:53:02 -0400114 void (*dirty_inode) (struct inode *, int flags);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100115 int (*write_inode) (struct inode *, struct writeback_control *wbc);
Al Viro336fb3b2010-06-08 00:37:12 -0400116 int (*drop_inode) (struct inode *);
117 void (*evict_inode) (struct inode *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 void (*put_super) (struct super_block *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 int (*sync_fs)(struct super_block *sb, int wait);
Takashi Satoc4be0c12009-01-09 16:40:58 -0800120 int (*freeze_fs) (struct super_block *);
121 int (*unfreeze_fs) (struct super_block *);
David Howells726c3342006-06-23 02:02:58 -0700122 int (*statfs) (struct dentry *, struct kstatfs *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 int (*remount_fs) (struct super_block *, int *, char *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 void (*umount_begin) (struct super_block *);
Al Viro34c80b12011-12-08 21:32:45 -0500125 int (*show_options)(struct seq_file *, struct dentry *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
127 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100128 int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130locking rules:
Al Viro336fb3b2010-06-08 00:37:12 -0400131 All may block [not true, see below]
Christoph Hellwig7e325d32009-06-19 20:22:37 +0200132 s_umount
133alloc_inode:
134destroy_inode:
Christoph Hellwigaa385722011-05-27 06:53:02 -0400135dirty_inode:
Christoph Hellwig7e325d32009-06-19 20:22:37 +0200136write_inode:
Dave Chinnerf283c862011-03-22 22:23:39 +1100137drop_inode: !!!inode->i_lock!!!
Al Viro336fb3b2010-06-08 00:37:12 -0400138evict_inode:
Christoph Hellwig7e325d32009-06-19 20:22:37 +0200139put_super: write
Christoph Hellwig7e325d32009-06-19 20:22:37 +0200140sync_fs: read
Valerie Aurora06fd5162012-06-12 16:20:48 +0200141freeze_fs: write
142unfreeze_fs: write
Al Viro336fb3b2010-06-08 00:37:12 -0400143statfs: maybe(read) (see below)
144remount_fs: write
Christoph Hellwig7e325d32009-06-19 20:22:37 +0200145umount_begin: no
146show_options: no (namespace_sem)
147quota_read: no (see below)
148quota_write: no (see below)
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100149bdev_try_to_free_page: no (see below)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
Al Viro336fb3b2010-06-08 00:37:12 -0400151->statfs() has s_umount (shared) when called by ustat(2) (native or
152compat), but that's an accident of bad API; s_umount is used to pin
153the superblock down when we only have dev_t given us by userland to
154identify the superblock. Everything else (statfs(), fstatfs(), etc.)
155doesn't hold it when calling ->statfs() - superblock is pinned down
156by resolving the pathname passed to syscall.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157->quota_read() and ->quota_write() functions are both guaranteed to
158be the only ones operating on the quota file by the quota code (via
159dqio_sem) (unless an admin really wants to screw up something and
160writes to quota files with quotas on). For other details about locking
161see also dquot_operations section.
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100162->bdev_try_to_free_page is called from the ->releasepage handler of
163the block device inode. See there for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165--------------------------- file_system_type ---------------------------
166prototypes:
Jonathan Corbet5d8b2eb2006-07-10 04:44:07 -0700167 int (*get_sb) (struct file_system_type *, int,
168 const char *, void *, struct vfsmount *);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100169 struct dentry *(*mount) (struct file_system_type *, int,
170 const char *, void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 void (*kill_sb) (struct super_block *);
172locking rules:
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100173 may block
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100174mount yes
175kill_sb yes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Al Viro1a102ff2011-03-16 09:07:58 -0400177->mount() returns ERR_PTR or the root dentry; its superblock should be locked
178on return.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179->kill_sb() takes a write-locked superblock, does all shutdown work on it,
180unlocks and drops the reference.
181
182--------------------------- address_space_operations --------------------------
183prototypes:
184 int (*writepage)(struct page *page, struct writeback_control *wbc);
185 int (*readpage)(struct file *, struct page *);
186 int (*sync_page)(struct page *);
187 int (*writepages)(struct address_space *, struct writeback_control *);
188 int (*set_page_dirty)(struct page *page);
189 int (*readpages)(struct file *filp, struct address_space *mapping,
190 struct list_head *pages, unsigned nr_pages);
Nick Piggin4e02ed42008-10-29 14:00:55 -0700191 int (*write_begin)(struct file *, struct address_space *mapping,
192 loff_t pos, unsigned len, unsigned flags,
193 struct page **pagep, void **fsdata);
194 int (*write_end)(struct file *, struct address_space *mapping,
195 loff_t pos, unsigned len, unsigned copied,
196 struct page *page, void *fsdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 sector_t (*bmap)(struct address_space *, sector_t);
Lukas Czernerd47992f2013-05-21 23:17:23 -0400198 void (*invalidatepage) (struct page *, unsigned int, unsigned int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 int (*releasepage) (struct page *, int);
Linus Torvalds6072d132010-12-01 13:35:19 -0500200 void (*freepage)(struct page *);
Al Virod8d3d942014-03-04 21:27:34 -0500201 int (*direct_IO)(int, struct kiocb *, struct iov_iter *iter, loff_t offset);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100202 int (*get_xip_mem)(struct address_space *, pgoff_t, int, void **,
203 unsigned long *);
204 int (*migratepage)(struct address_space *, struct page *, struct page *);
205 int (*launder_page)(struct page *);
Al Viroc186afb42014-02-02 21:16:54 -0500206 int (*is_partially_uptodate)(struct page *, unsigned long, unsigned long);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100207 int (*error_remove_page)(struct address_space *, struct page *);
Mel Gorman62c230b2012-07-31 16:44:55 -0700208 int (*swap_activate)(struct file *);
209 int (*swap_deactivate)(struct file *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211locking rules:
Linus Torvalds6072d132010-12-01 13:35:19 -0500212 All except set_page_dirty and freepage may block
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100214 PageLocked(page) i_mutex
215writepage: yes, unlocks (see below)
216readpage: yes, unlocks
217sync_page: maybe
218writepages:
219set_page_dirty no
220readpages:
221write_begin: locks the page yes
222write_end: yes, unlocks yes
223bmap:
224invalidatepage: yes
225releasepage: yes
226freepage: yes
227direct_IO:
228get_xip_mem: maybe
229migratepage: yes (both)
230launder_page: yes
231is_partially_uptodate: yes
232error_remove_page: yes
Mel Gorman62c230b2012-07-31 16:44:55 -0700233swap_activate: no
234swap_deactivate: no
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Nick Piggin4e02ed42008-10-29 14:00:55 -0700236 ->write_begin(), ->write_end(), ->sync_page() and ->readpage()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237may be called from the request handler (/dev/loop).
238
239 ->readpage() unlocks the page, either synchronously or via I/O
240completion.
241
242 ->readpages() populates the pagecache with the passed pages and starts
243I/O against them. They come unlocked upon I/O completion.
244
245 ->writepage() is used for two purposes: for "memory cleansing" and for
246"sync". These are quite different operations and the behaviour may differ
247depending upon the mode.
248
249If writepage is called for sync (wbc->sync_mode != WBC_SYNC_NONE) then
250it *must* start I/O against the page, even if that would involve
251blocking on in-progress I/O.
252
253If writepage is called for memory cleansing (sync_mode ==
254WBC_SYNC_NONE) then its role is to get as much writeout underway as
255possible. So writepage should try to avoid blocking against
256currently-in-progress I/O.
257
258If the filesystem is not called for "sync" and it determines that it
259would need to block against in-progress I/O to be able to start new I/O
260against the page the filesystem should redirty the page with
261redirty_page_for_writepage(), then unlock the page and return zero.
262This may also be done to avoid internal deadlocks, but rarely.
263
Robert P. J. Day3a4fa0a2007-10-19 23:10:43 +0200264If the filesystem is called for sync then it must wait on any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265in-progress I/O and then start new I/O.
266
Nikita Danilov20546062005-05-01 08:58:37 -0700267The filesystem should unlock the page synchronously, before returning to the
268caller, unless ->writepage() returns special WRITEPAGE_ACTIVATE
269value. WRITEPAGE_ACTIVATE means that page cannot really be written out
270currently, and VM should stop calling ->writepage() on this page for some
271time. VM does this by moving page to the head of the active list, hence the
272name.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274Unless the filesystem is going to redirty_page_for_writepage(), unlock the page
275and return zero, writepage *must* run set_page_writeback() against the page,
276followed by unlocking it. Once set_page_writeback() has been run against the
277page, write I/O can be submitted and the write I/O completion handler must run
278end_page_writeback() once the I/O is complete. If no I/O is submitted, the
279filesystem must run end_page_writeback() against the page before returning from
280writepage.
281
282That is: after 2.5.12, pages which are under writeout are *not* locked. Note,
283if the filesystem needs the page to be locked during writeout, that is ok, too,
284the page is allowed to be unlocked at any point in time between the calls to
285set_page_writeback() and end_page_writeback().
286
287Note, failure to run either redirty_page_for_writepage() or the combination of
288set_page_writeback()/end_page_writeback() on a page submitted to writepage
289will leave the page itself marked clean but it will be tagged as dirty in the
290radix tree. This incoherency can lead to all sorts of hard-to-debug problems
291in the filesystem like having dirty inodes at umount and losing written data.
292
293 ->sync_page() locking rules are not well-defined - usually it is called
294with lock on page, but that is not guaranteed. Considering the currently
295existing instances of this method ->sync_page() itself doesn't look
296well-defined...
297
298 ->writepages() is used for periodic writeback and for syscall-initiated
299sync operations. The address_space should start I/O against at least
300*nr_to_write pages. *nr_to_write must be decremented for each page which is
301written. The address_space implementation may write more (or less) pages
302than *nr_to_write asks for, but it should try to be reasonably close. If
303nr_to_write is NULL, all dirty pages must be written.
304
305writepages should _only_ write pages which are present on
306mapping->io_pages.
307
308 ->set_page_dirty() is called from various places in the kernel
309when the target page is marked as needing writeback. It may be called
310under spinlock (it cannot block) and is sometimes called with the page
311not locked.
312
313 ->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100314filesystems and by the swapper. The latter will eventually go away. Please,
315keep it that way and don't breed new callers.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 ->invalidatepage() is called when the filesystem must attempt to drop
Lukas Czernerd47992f2013-05-21 23:17:23 -0400318some or all of the buffers from the page when it is being truncated. It
319returns zero on success. If ->invalidatepage is zero, the kernel uses
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320block_invalidatepage() instead.
321
322 ->releasepage() is called when the kernel is about to try to drop the
323buffers from the page in preparation for freeing it. It returns zero to
324indicate that the buffers are (or may be) freeable. If ->releasepage is zero,
325the kernel assumes that the fs has no private interest in the buffers.
326
Linus Torvalds6072d132010-12-01 13:35:19 -0500327 ->freepage() is called when the kernel is done dropping the page
328from the page cache.
329
Trond Myklebuste3db7692007-01-10 23:15:39 -0800330 ->launder_page() may be called prior to releasing a page if
331it is still found to be dirty. It returns zero if the page was successfully
332cleaned, or an error value if not. Note that in order to prevent the page
333getting mapped back in and redirtied, it needs to be kept locked
334across the entire operation.
335
Mel Gorman62c230b2012-07-31 16:44:55 -0700336 ->swap_activate will be called with a non-zero argument on
337files backing (non block device backed) swapfiles. A return value
338of zero indicates success, in which case this file can be used for
339backing swapspace. The swapspace operations will be proxied to the
340address space operations.
341
342 ->swap_deactivate() will be called in the sys_swapoff()
343path after ->swap_activate() returned success.
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345----------------------- file_lock_operations ------------------------------
346prototypes:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
348 void (*fl_release_private)(struct file_lock *);
349
350
351locking rules:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400352 inode->i_lock may block
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100353fl_copy_lock: yes no
Jeff Layton2ece1732014-08-12 10:38:07 -0400354fl_release_private: maybe maybe[1]
355
356[1]: ->fl_release_private for flock or POSIX locks is currently allowed
357to block. Leases however can still be freed while the i_lock is held and
358so fl_release_private called on a lease should not block.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360----------------------- lock_manager_operations ---------------------------
361prototypes:
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400362 int (*lm_compare_owner)(struct file_lock *, struct file_lock *);
Jeff Layton3999e492013-06-21 08:58:19 -0400363 unsigned long (*lm_owner_key)(struct file_lock *);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400364 void (*lm_notify)(struct file_lock *); /* unblock callback */
365 int (*lm_grant)(struct file_lock *, struct file_lock *, int);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400366 void (*lm_break)(struct file_lock *); /* break_lease callback */
367 int (*lm_change)(struct file_lock **, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369locking rules:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400370
Jeff Layton7b2296a2013-06-21 08:58:20 -0400371 inode->i_lock blocked_lock_lock may block
372lm_compare_owner: yes[1] maybe no
373lm_owner_key yes[1] yes no
374lm_notify: yes yes no
375lm_grant: no no no
376lm_break: yes no no
377lm_change yes no no
Jeff Layton1c8c6012013-06-21 08:58:15 -0400378
Jeff Layton3999e492013-06-21 08:58:19 -0400379[1]: ->lm_compare_owner and ->lm_owner_key are generally called with
380*an* inode->i_lock held. It may not be the i_lock of the inode
381associated with either file_lock argument! This is the case with deadlock
382detection, since the code has to chase down the owners of locks that may
383be entirely unrelated to the one on which the lock is being acquired.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400384For deadlock detection however, the blocked_lock_lock is also held. The
Jeff Layton3999e492013-06-21 08:58:19 -0400385fact that these locks are held ensures that the file_locks do not
386disappear out from under you while doing the comparison or generating an
387owner key.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389--------------------------- buffer_head -----------------------------------
390prototypes:
391 void (*b_end_io)(struct buffer_head *bh, int uptodate);
392
393locking rules:
394 called from interrupts. In other words, extreme care is needed here.
395bh is locked, but that's all warranties we have here. Currently only RAID1,
396highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
397call this method upon the IO completion.
398
399--------------------------- block_device_operations -----------------------
400prototypes:
Christoph Hellwige1455d12010-10-06 10:46:53 +0200401 int (*open) (struct block_device *, fmode_t);
402 int (*release) (struct gendisk *, fmode_t);
403 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
404 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
405 int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 int (*media_changed) (struct gendisk *);
Christoph Hellwige1455d12010-10-06 10:46:53 +0200407 void (*unlock_native_capacity) (struct gendisk *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 int (*revalidate_disk) (struct gendisk *);
Christoph Hellwige1455d12010-10-06 10:46:53 +0200409 int (*getgeo)(struct block_device *, struct hd_geometry *);
410 void (*swap_slot_free_notify) (struct block_device *, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412locking rules:
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100413 bd_mutex
414open: yes
415release: yes
416ioctl: no
417compat_ioctl: no
418direct_access: no
419media_changed: no
420unlock_native_capacity: no
421revalidate_disk: no
422getgeo: no
423swap_slot_free_notify: no (see below)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
Christoph Hellwige1455d12010-10-06 10:46:53 +0200425media_changed, unlock_native_capacity and revalidate_disk are called only from
426check_disk_change().
427
428swap_slot_free_notify is called with swap_lock and sometimes the page lock
429held.
430
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432--------------------------- file_operations -------------------------------
433prototypes:
434 loff_t (*llseek) (struct file *, loff_t, int);
435 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
Badari Pulavarty027445c2006-09-30 23:28:46 -0700437 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
438 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
Al Viro293bc982014-02-11 18:37:41 -0500439 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
440 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
Al Viro2233f312013-05-22 21:44:23 -0400441 int (*iterate) (struct file *, struct dir_context *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 unsigned int (*poll) (struct file *, struct poll_table_struct *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
444 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
445 int (*mmap) (struct file *, struct vm_area_struct *);
446 int (*open) (struct inode *, struct file *);
447 int (*flush) (struct file *);
448 int (*release) (struct inode *, struct file *);
Josef Bacik02c24a82011-07-16 20:44:56 -0400449 int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 int (*aio_fsync) (struct kiocb *, int datasync);
451 int (*fasync) (int, struct file *, int);
452 int (*lock) (struct file *, int, struct file_lock *);
453 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long,
454 loff_t *);
455 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long,
456 loff_t *);
457 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t,
458 void __user *);
459 ssize_t (*sendpage) (struct file *, struct page *, int, size_t,
460 loff_t *, int);
461 unsigned long (*get_unmapped_area)(struct file *, unsigned long,
462 unsigned long, unsigned long, unsigned long);
463 int (*check_flags)(int);
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100464 int (*flock) (struct file *, int, struct file_lock *);
465 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
466 size_t, unsigned int);
467 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
468 size_t, unsigned int);
Jeff Laytone6f5c782014-08-22 10:40:25 -0400469 int (*setlease)(struct file *, long, struct file_lock **, void **);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100470 long (*fallocate)(struct file *, int, loff_t, loff_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471};
472
473locking rules:
Jeff Laytonc45198e2014-09-01 07:12:07 -0400474 All may block.
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476->llseek() locking has moved from llseek to the individual llseek
477implementations. If your fs is not using generic_file_llseek, you
478need to acquire and release the appropriate locks in your ->llseek().
479For many filesystems, it is probably safe to acquire the inode
Jan Blunck866707f2010-05-26 14:44:54 -0700480mutex or just to use i_size_read() instead.
481Note: this does not protect the file->f_pos against concurrent modifications
482since this is something the userspace has to take care about.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100484->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
485Most instances call fasync_helper(), which does that maintenance, so it's
486not normally something one needs to worry about. Return values > 0 will be
487mapped to zero in the VFS layer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489->readdir() and ->ioctl() on directories must be changed. Ideally we would
490move ->readdir() to inode_operations and use a separate method for directory
491->ioctl() or kill the latter completely. One of the problems is that for
492anything that resembles union-mount we won't have a struct file for all
493components. And there are other reasons why the current interface is a mess...
494
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495->read on directories probably must go away - we should just enforce -EISDIR
496in sys_read() and friends.
497
Jeff Laytonf82b4b62014-08-22 18:50:48 -0400498->setlease operations should call generic_setlease() before or after setting
499the lease within the individual filesystem to record the result of the
500operation
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502--------------------------- dquot_operations -------------------------------
503prototypes:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 int (*write_dquot) (struct dquot *);
505 int (*acquire_dquot) (struct dquot *);
506 int (*release_dquot) (struct dquot *);
507 int (*mark_dirty) (struct dquot *);
508 int (*write_info) (struct super_block *, int);
509
510These operations are intended to be more or less wrapping functions that ensure
511a proper locking wrt the filesystem and call the generic quota operations.
512
513What filesystem should expect from the generic quota functions:
514
515 FS recursion Held locks when called
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516write_dquot: yes dqonoff_sem or dqptr_sem
517acquire_dquot: yes dqonoff_sem or dqptr_sem
518release_dquot: yes dqonoff_sem or dqptr_sem
519mark_dirty: no -
520write_info: yes dqonoff_sem
521
522FS recursion means calling ->quota_read() and ->quota_write() from superblock
523operations.
524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525More details about quota locking can be found in fs/dquot.c.
526
527--------------------------- vm_operations_struct -----------------------------
528prototypes:
529 void (*open)(struct vm_area_struct*);
530 void (*close)(struct vm_area_struct*);
Nick Piggind0217ac2007-07-19 01:47:03 -0700531 int (*fault)(struct vm_area_struct*, struct vm_fault *);
Nick Pigginc2ec1752009-03-31 15:23:21 -0700532 int (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
Rik van Riel28b2ee22008-07-23 21:27:05 -0700533 int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534
535locking rules:
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100536 mmap_sem PageLocked(page)
537open: yes
538close: yes
539fault: yes can return with page locked
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -0700540map_pages: yes
Christoph Hellwigb83be6f2010-12-16 12:04:54 +0100541page_mkwrite: yes can return with page locked
542access: yes
Mark Fashehed2f2f92007-07-19 01:47:01 -0700543
Nick Pigginb827e492009-04-30 15:08:16 -0700544 ->fault() is called when a previously not present pte is about
545to be faulted in. The filesystem must find and return the page associated
546with the passed in "pgoff" in the vm_fault structure. If it is possible that
547the page may be truncated and/or invalidated, then the filesystem must lock
548the page, then ensure it is not already truncated (the page lock will block
549subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
550locked. The VM will unlock the page.
551
Kirill A. Shutemov8c6e50b2014-04-07 15:37:18 -0700552 ->map_pages() is called when VM asks to map easy accessible pages.
553Filesystem should find and map pages associated with offsets from "pgoff"
554till "max_pgoff". ->map_pages() is called with page table locked and must
555not block. If it's not possible to reach a page without blocking,
556filesystem should skip it. Filesystem should use do_set_pte() to setup
557page table entry. Pointer to entry associated with offset "pgoff" is
558passed in "pte" field in vm_fault structure. Pointers to entries for other
559offsets should be calculated relative to "pte".
560
Nick Pigginb827e492009-04-30 15:08:16 -0700561 ->page_mkwrite() is called when a previously read-only pte is
562about to become writeable. The filesystem again must ensure that there are
563no truncate/invalidate races, and then return with the page locked. If
564the page has been truncated, the filesystem should not look up a new page
565like the ->fault() handler, but simply return with VM_FAULT_NOPAGE, which
566will cause the VM to retry the fault.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Rik van Riel28b2ee22008-07-23 21:27:05 -0700568 ->access() is called when get_user_pages() fails in
Stefan Weil507da6a2013-12-05 20:34:05 +0100569access_process_vm(), typically used to debug a process through
Rik van Riel28b2ee22008-07-23 21:27:05 -0700570/proc/pid/mem or ptrace. This function is needed only for
571VM_IO | VM_PFNMAP VMAs.
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573================================================================================
574 Dubious stuff
575
576(if you break something or notice that it is broken and do not fix it yourself
577- at least put it here)