xref: /linux/Documentation/filesystems/locking.rst (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
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 inode *, const struct qstr *,
21			    struct dentry *, unsigned int);
22	int (*d_weak_revalidate)(struct dentry *, unsigned int);
23	int (*d_hash)(const struct dentry *, struct qstr *);
24	int (*d_compare)(const struct dentry *,
25			unsigned int, const char *, const struct qstr *);
26	int (*d_delete)(struct dentry *);
27	int (*d_init)(struct dentry *);
28	void (*d_release)(struct dentry *);
29	void (*d_iput)(struct dentry *, struct inode *);
30	char *(*d_dname)((struct dentry *dentry, char *buffer, int buflen);
31	struct vfsmount *(*d_automount)(struct path *path);
32	int (*d_manage)(const struct path *, bool);
33	struct dentry *(*d_real)(struct dentry *, enum d_real_type type);
34	bool (*d_unalias_trylock)(const struct dentry *);
35	void (*d_unalias_unlock)(const struct dentry *);
36
37locking rules:
38
39================== ===========	========	==============	========
40ops		   rename_lock	->d_lock	may block	rcu-walk
41================== ===========	========	==============	========
42d_revalidate:	   no		no		yes (ref-walk)	maybe
43d_weak_revalidate: no		no		yes	 	no
44d_hash		   no		no		no		maybe
45d_compare:	   yes		no		no		maybe
46d_delete:	   no		yes		no		no
47d_init:		   no		no		yes		no
48d_release:	   no		no		yes		no
49d_prune:           no		yes		no		no
50d_iput:		   no		no		yes		no
51d_dname:	   no		no		no		no
52d_automount:	   no		no		yes		no
53d_manage:	   no		no		yes (ref-walk)	maybe
54d_real		   no		no		yes 		no
55d_unalias_trylock  yes		no		no 		no
56d_unalias_unlock   yes		no		no 		no
57================== ===========	========	==============	========
58
59inode_operations
60================
61
62prototypes::
63
64	int (*create) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t, bool);
65	struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
66	int (*link) (struct dentry *,struct inode *,struct dentry *);
67	int (*unlink) (struct inode *,struct dentry *);
68	int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *);
69	struct dentry *(*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t);
70	int (*rmdir) (struct inode *,struct dentry *);
71	int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t);
72	int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *,
73			struct inode *, struct dentry *, unsigned int);
74	int (*readlink) (struct dentry *, char __user *,int);
75	const char *(*get_link) (struct dentry *, struct inode *, struct delayed_call *);
76	void (*truncate) (struct inode *);
77	int (*permission) (struct mnt_idmap *, struct inode *, int, unsigned int);
78	struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
79	int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *);
80	int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int);
81	ssize_t (*listxattr) (struct dentry *, char *, size_t);
82	int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
83	void (*update_time)(struct inode *inode, enum fs_update_time type,
84			    int flags);
85	void (*sync_lazytime)(struct inode *inode);
86	int (*atomic_open)(struct inode *, struct dentry *,
87				struct file *, unsigned open_flag,
88				umode_t create_mode);
89	int (*tmpfile) (struct mnt_idmap *, struct inode *,
90			struct file *, umode_t);
91	int (*fileattr_set)(struct mnt_idmap *idmap,
92			    struct dentry *dentry, struct file_kattr *fa);
93	int (*fileattr_get)(struct dentry *dentry, struct file_kattr *fa);
94	struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int);
95	struct offset_ctx *(*get_offset_ctx)(struct inode *inode);
96
97locking rules:
98	all may block
99
100==============	==================================================
101ops		i_rwsem(inode)
102==============	==================================================
103lookup:		shared
104create:		exclusive
105link:		exclusive (both)
106mknod:		exclusive
107symlink:	exclusive
108mkdir:		exclusive
109unlink:		exclusive (both)
110rmdir:		exclusive (both)(see below)
111rename:		exclusive (both parents, some children)	(see below)
112readlink:	no
113get_link:	no
114setattr:	exclusive
115permission:	no (may not block if called in rcu-walk mode)
116get_inode_acl:	no
117get_acl:	no
118getattr:	no
119listxattr:	no
120fiemap:		no
121update_time:	no
122sync_lazytime:	no
123atomic_open:	shared (exclusive if O_CREAT is set in open flags)
124tmpfile:	no
125fileattr_get:	no or exclusive
126fileattr_set:	exclusive
127get_offset_ctx  no
128==============	==================================================
129
130
131	Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_rwsem
132	exclusive on victim.
133	cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
134	->unlink() and ->rename() have ->i_rwsem exclusive on all non-directories
135	involved.
136	->rename() has ->i_rwsem exclusive on any subdirectory that changes parent.
137
138See Documentation/filesystems/directory-locking.rst for more detailed discussion
139of the locking scheme for directory operations.
140
141xattr_handler operations
142========================
143
144prototypes::
145
146	bool (*list)(struct dentry *dentry);
147	int (*get)(const struct xattr_handler *handler, struct dentry *dentry,
148		   struct inode *inode, const char *name, void *buffer,
149		   size_t size);
150	int (*set)(const struct xattr_handler *handler,
151                   struct mnt_idmap *idmap,
152                   struct dentry *dentry, struct inode *inode, const char *name,
153                   const void *buffer, size_t size, int flags);
154
155locking rules:
156	all may block
157
158=====		==============
159ops		i_rwsem(inode)
160=====		==============
161list:		no
162get:		no
163set:		exclusive
164=====		==============
165
166super_operations
167================
168
169prototypes::
170
171	struct inode *(*alloc_inode)(struct super_block *sb);
172	void (*free_inode)(struct inode *);
173	void (*destroy_inode)(struct inode *);
174	void (*dirty_inode) (struct inode *, int flags);
175	int (*write_inode) (struct inode *, struct writeback_control *wbc);
176	int (*drop_inode) (struct inode *);
177	void (*evict_inode) (struct inode *);
178	void (*put_super) (struct super_block *);
179	int (*sync_fs)(struct super_block *sb, int wait);
180	int (*freeze_fs) (struct super_block *);
181	int (*unfreeze_fs) (struct super_block *);
182	int (*statfs) (struct dentry *, struct kstatfs *);
183	void (*umount_begin) (struct super_block *);
184	int (*show_options)(struct seq_file *, struct dentry *);
185	ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
186	ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
187
188locking rules:
189	All may block [not true, see below]
190
191======================	============	========================
192ops			s_umount	note
193======================	============	========================
194alloc_inode:
195free_inode:				called from RCU callback
196destroy_inode:
197dirty_inode:
198write_inode:
199drop_inode:				!!!inode->i_lock!!!
200evict_inode:
201put_super:		write
202sync_fs:		read
203freeze_fs:		write
204unfreeze_fs:		write
205statfs:			maybe(read)	(see below)
206umount_begin:		no
207show_options:		no		(namespace_sem)
208quota_read:		no		(see below)
209quota_write:		no		(see below)
210======================	============	========================
211
212->statfs() has s_umount (shared) when called by ustat(2) (native or
213compat), but that's an accident of bad API; s_umount is used to pin
214the superblock down when we only have dev_t given us by userland to
215identify the superblock.  Everything else (statfs(), fstatfs(), etc.)
216doesn't hold it when calling ->statfs() - superblock is pinned down
217by resolving the pathname passed to syscall.
218
219->quota_read() and ->quota_write() functions are both guaranteed to
220be the only ones operating on the quota file by the quota code (via
221dqio_sem) (unless an admin really wants to screw up something and
222writes to quota files with quotas on). For other details about locking
223see also dquot_operations section.
224
225file_system_type
226================
227
228prototypes::
229
230	void (*kill_sb) (struct super_block *);
231
232locking rules:
233
234=======		=========
235ops		may block
236=======		=========
237kill_sb		yes
238=======		=========
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 (*read_folio)(struct file *, struct folio *);
248	int (*writepages)(struct address_space *, struct writeback_control *);
249	bool (*dirty_folio)(struct address_space *, struct folio *folio);
250	void (*readahead)(struct readahead_control *);
251	int (*write_begin)(const struct kiocb *, struct address_space *mapping,
252				loff_t pos, unsigned len,
253				struct folio **foliop, void **fsdata);
254	int (*write_end)(const struct kiocb *, struct address_space *mapping,
255				loff_t pos, unsigned len, unsigned copied,
256				struct folio *folio, void *fsdata);
257	sector_t (*bmap)(struct address_space *, sector_t);
258	void (*invalidate_folio) (struct folio *, size_t start, size_t len);
259	bool (*release_folio)(struct folio *, gfp_t);
260	void (*free_folio)(struct folio *);
261	int (*direct_IO)(struct kiocb *, struct iov_iter *iter);
262	int (*migrate_folio)(struct address_space *, struct folio *dst,
263			struct folio *src, enum migrate_mode);
264	int (*launder_folio)(struct folio *);
265	bool (*is_partially_uptodate)(struct folio *, size_t from, size_t count);
266	int (*error_remove_folio)(struct address_space *, struct folio *);
267	int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
268	int (*swap_deactivate)(struct file *);
269	int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
270
271locking rules:
272	All except dirty_folio and free_folio may block
273
274======================	======================== =========	===============
275ops			folio locked		 i_rwsem	invalidate_lock
276======================	======================== =========	===============
277read_folio:		yes, unlocks				shared
278writepages:
279dirty_folio:		maybe
280readahead:		yes, unlocks				shared
281write_begin:		locks the folio		 exclusive
282write_end:		yes, unlocks		 exclusive
283bmap:
284invalidate_folio:	yes					exclusive
285release_folio:		yes
286free_folio:		yes
287direct_IO:
288migrate_folio:		yes (both)
289launder_folio:		yes
290is_partially_uptodate:	yes
291error_remove_folio:	yes
292swap_activate:		no
293swap_deactivate:	no
294swap_rw:		yes, unlocks
295======================	======================== =========	===============
296
297->write_begin(), ->write_end() and ->read_folio() may be called from
298the request handler (/dev/loop).
299
300->read_folio() unlocks the folio, either synchronously or via I/O
301completion.
302
303->readahead() unlocks the folios that I/O is attempted on like ->read_folio().
304
305->writepages() is used for periodic writeback and for syscall-initiated
306sync operations.  The address_space should start I/O against at least
307``*nr_to_write`` pages.  ``*nr_to_write`` must be decremented for each page
308which is written.  The address_space implementation may write more (or less)
309pages than ``*nr_to_write`` asks for, but it should try to be reasonably close.
310If nr_to_write is NULL, all dirty pages must be written.
311
312writepages should _only_ write pages which are present in
313mapping->i_pages.
314
315->dirty_folio() is called from various places in the kernel when
316the target folio is marked as needing writeback.  The folio cannot be
317truncated because either the caller holds the folio lock, or the caller
318has found the folio while holding the page table lock which will block
319truncation.
320
321->bmap() is currently used by legacy ioctl() (FIBMAP) provided by some
322filesystems and by the swapper. The latter will eventually go away.  Please,
323keep it that way and don't breed new callers.
324
325->invalidate_folio() is called when the filesystem must attempt to drop
326some or all of the buffers from the page when it is being truncated. It
327returns zero on success.  The filesystem must exclusively acquire
328invalidate_lock before invalidating page cache in truncate / hole punch
329path (and thus calling into ->invalidate_folio) to block races between page
330cache invalidation and page cache filling functions (fault, read, ...).
331
332->release_folio() is called when the MM wants to make a change to the
333folio that would invalidate the filesystem's private data.  For example,
334it may be about to be removed from the address_space or split.  The folio
335is locked and not under writeback.  It may be dirty.  The gfp parameter
336is not usually used for allocation, but rather to indicate what the
337filesystem may do to attempt to free the private data.  The filesystem may
338return false to indicate that the folio's private data cannot be freed.
339If it returns true, it should have already removed the private data from
340the folio.  If a filesystem does not provide a ->release_folio method,
341the pagecache will assume that private data is buffer_heads and call
342try_to_free_buffers().
343
344->free_folio() is called when the kernel has dropped the folio
345from the page cache.
346
347->launder_folio() may be called prior to releasing a folio if
348it is still found to be dirty. It returns zero if the folio was successfully
349cleaned, or an error value if not. Note that in order to prevent the folio
350getting mapped back in and redirtied, it needs to be kept locked
351across the entire operation.
352
353->swap_activate() will be called to prepare the given file for swap.  It
354should perform any validation and preparation necessary to ensure that
355writes can be performed with minimal memory allocation.  It should call
356add_swap_extent(), or the helper iomap_swapfile_activate(), and return
357the number of extents added.  If IO should be submitted through
358->swap_rw(), it should set SWP_FS_OPS, otherwise IO will be submitted
359directly to the block device ``sis->bdev``.
360
361->swap_deactivate() will be called in the sys_swapoff()
362path after ->swap_activate() returned success.
363
364->swap_rw will be called for swap IO if SWP_FS_OPS was set by ->swap_activate().
365
366file_lock_operations
367====================
368
369prototypes::
370
371	void (*fl_copy_lock)(struct file_lock *, struct file_lock *);
372	void (*fl_release_private)(struct file_lock *);
373
374
375locking rules:
376
377===================	=============	=========
378ops			inode->i_lock	may block
379===================	=============	=========
380fl_copy_lock:		yes		no
381fl_release_private:	maybe		maybe[1]_
382===================	=============	=========
383
384.. [1]:
385   ->fl_release_private for flock or POSIX locks is currently allowed
386   to block. Leases however can still be freed while the i_lock is held and
387   so fl_release_private called on a lease should not block.
388
389lock_manager_operations
390=======================
391
392prototypes::
393
394	void (*lm_notify)(struct file_lock *);  /* unblock callback */
395	int (*lm_grant)(struct file_lock *, struct file_lock *, int);
396	void (*lm_break)(struct file_lock *); /* break_lease callback */
397	int (*lm_change)(struct file_lock **, int);
398	bool (*lm_breaker_owns_lease)(struct file_lock *);
399        bool (*lm_lock_expirable)(struct file_lock *);
400        void (*lm_expire_lock)(void);
401
402locking rules:
403
404======================	=============	=================	=========
405ops			   flc_lock  	blocked_lock_lock	may block
406======================	=============	=================	=========
407lm_notify:		no      	yes			no
408lm_grant:		no		no			no
409lm_break:		yes		no			no
410lm_change		yes		no			no
411lm_breaker_owns_lease:	yes     	no			no
412lm_lock_expirable	yes		no			no
413lm_expire_lock		no		no			yes
414lm_open_conflict	yes		no			no
415======================	=============	=================	=========
416
417buffer_head
418===========
419
420prototypes::
421
422	void (*b_end_io)(struct buffer_head *bh, int uptodate);
423
424locking rules:
425
426called from interrupts. In other words, extreme care is needed here.
427bh is locked, but that's all warranties we have here. Currently only RAID1,
428highmem, fs/buffer.c, and fs/ntfs/aops.c are providing these. Block devices
429call this method upon the IO completion.
430
431block_device_operations
432=======================
433prototypes::
434
435	int (*open) (struct block_device *, fmode_t);
436	int (*release) (struct gendisk *, fmode_t);
437	int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
438	int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
439	int (*direct_access) (struct block_device *, sector_t, void **,
440				unsigned long *);
441	void (*unlock_native_capacity) (struct gendisk *);
442	int (*getgeo)(struct gendisk *, struct hd_geometry *);
443	void (*swap_slot_free_notify) (struct block_device *, unsigned long);
444
445locking rules:
446
447======================= ===================
448ops			open_mutex
449======================= ===================
450open:			yes
451release:		yes
452ioctl:			no
453compat_ioctl:		no
454direct_access:		no
455unlock_native_capacity:	no
456getgeo:			no
457swap_slot_free_notify:	no	(see below)
458======================= ===================
459
460swap_slot_free_notify is called with swap_lock and sometimes the page lock
461held.
462
463
464file_operations
465===============
466
467prototypes::
468
469	loff_t (*llseek) (struct file *, loff_t, int);
470	ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
471	ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
472	ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
473	ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
474	int (*iopoll) (struct kiocb *kiocb, bool spin);
475	int (*iterate_shared) (struct file *, struct dir_context *);
476	__poll_t (*poll) (struct file *, struct poll_table_struct *);
477	long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
478	long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
479	int (*mmap) (struct file *, struct vm_area_struct *);
480	int (*open) (struct inode *, struct file *);
481	int (*flush) (struct file *);
482	int (*release) (struct inode *, struct file *);
483	int (*fsync) (struct file *, loff_t start, loff_t end, int datasync);
484	int (*fasync) (int, struct file *, int);
485	int (*lock) (struct file *, int, struct file_lock *);
486	unsigned long (*get_unmapped_area)(struct file *, unsigned long,
487			unsigned long, unsigned long, unsigned long);
488	int (*check_flags)(int);
489	int (*flock) (struct file *, int, struct file_lock *);
490	ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *,
491			size_t, unsigned int);
492	ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *,
493			size_t, unsigned int);
494	int (*setlease)(struct file *, long, struct file_lock **, void **);
495	long (*fallocate)(struct file *, int, loff_t, loff_t);
496	void (*show_fdinfo)(struct seq_file *m, struct file *f);
497	unsigned (*mmap_capabilities)(struct file *);
498	ssize_t (*copy_file_range)(struct file *, loff_t, struct file *,
499			loff_t, size_t, unsigned int);
500	loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
501			struct file *file_out, loff_t pos_out,
502			loff_t len, unsigned int remap_flags);
503	int (*fadvise)(struct file *, loff_t, loff_t, int);
504
505locking rules:
506	All may block.
507
508->llseek() locking has moved from llseek to the individual llseek
509implementations.  If your fs is not using generic_file_llseek, you
510need to acquire and release the appropriate locks in your ->llseek().
511For many filesystems, it is probably safe to acquire the inode
512mutex or just to use i_size_read() instead.
513Note: this does not protect the file->f_pos against concurrent modifications
514since this is something the userspace has to take care about.
515
516->iterate_shared() is called with i_rwsem held for reading, and with the
517file f_pos_lock held exclusively
518
519->fasync() is responsible for maintaining the FASYNC bit in filp->f_flags.
520Most instances call fasync_helper(), which does that maintenance, so it's
521not normally something one needs to worry about.  Return values > 0 will be
522mapped to zero in the VFS layer.
523
524->readdir() and ->ioctl() on directories must be changed. Ideally we would
525move ->readdir() to inode_operations and use a separate method for directory
526->ioctl() or kill the latter completely. One of the problems is that for
527anything that resembles union-mount we won't have a struct file for all
528components. And there are other reasons why the current interface is a mess...
529
530->read on directories probably must go away - we should just enforce -EISDIR
531in sys_read() and friends.
532
533->setlease operations should call generic_setlease() before or after setting
534the lease within the individual filesystem to record the result of the
535operation
536
537->fallocate implementation must be really careful to maintain page cache
538consistency when punching holes or performing other operations that invalidate
539page cache contents. Usually the filesystem needs to call
540truncate_inode_pages_range() to invalidate relevant range of the page cache.
541However the filesystem usually also needs to update its internal (and on disk)
542view of file offset -> disk block mapping. Until this update is finished, the
543filesystem needs to block page faults and reads from reloading now-stale page
544cache contents from the disk. Since VFS acquires mapping->invalidate_lock in
545shared mode when loading pages from disk (filemap_fault(), filemap_read(),
546readahead paths), the fallocate implementation must take the invalidate_lock to
547prevent reloading.
548
549->copy_file_range and ->remap_file_range implementations need to serialize
550against modifications of file data while the operation is running. For
551blocking changes through write(2) and similar operations inode->i_rwsem can be
552used. To block changes to file contents via a memory mapping during the
553operation, the filesystem must take mapping->invalidate_lock to coordinate
554with ->page_mkwrite.
555
556dquot_operations
557================
558
559prototypes::
560
561	int (*write_dquot) (struct dquot *);
562	int (*acquire_dquot) (struct dquot *);
563	int (*release_dquot) (struct dquot *);
564	int (*mark_dirty) (struct dquot *);
565	int (*write_info) (struct super_block *, int);
566
567These operations are intended to be more or less wrapping functions that ensure
568a proper locking wrt the filesystem and call the generic quota operations.
569
570What filesystem should expect from the generic quota functions:
571
572==============	============	=========================
573ops		FS recursion	Held locks when called
574==============	============	=========================
575write_dquot:	yes		dqonoff_sem or dqptr_sem
576acquire_dquot:	yes		dqonoff_sem or dqptr_sem
577release_dquot:	yes		dqonoff_sem or dqptr_sem
578mark_dirty:	no		-
579write_info:	yes		dqonoff_sem
580==============	============	=========================
581
582FS recursion means calling ->quota_read() and ->quota_write() from superblock
583operations.
584
585More details about quota locking can be found in fs/dquot.c.
586
587vm_operations_struct
588====================
589
590prototypes::
591
592	void (*open)(struct vm_area_struct *);
593	void (*close)(struct vm_area_struct *);
594	vm_fault_t (*fault)(struct vm_fault *);
595	vm_fault_t (*huge_fault)(struct vm_fault *, unsigned int order);
596	vm_fault_t (*map_pages)(struct vm_fault *, pgoff_t start, pgoff_t end);
597	vm_fault_t (*page_mkwrite)(struct vm_area_struct *, struct vm_fault *);
598	vm_fault_t (*pfn_mkwrite)(struct vm_area_struct *, struct vm_fault *);
599	int (*access)(struct vm_area_struct *, unsigned long, void*, int, int);
600
601locking rules:
602
603=============	==========	===========================
604ops		mmap_lock	PageLocked(page)
605=============	==========	===========================
606open:		write
607close:		read/write
608fault:		read		can return with page locked
609huge_fault:	maybe-read
610map_pages:	maybe-read
611page_mkwrite:	read		can return with page locked
612pfn_mkwrite:	read
613access:		read
614=============	==========	===========================
615
616->fault() is called when a previously not present pte is about to be faulted
617in. The filesystem must find and return the page associated with the passed in
618"pgoff" in the vm_fault structure. If it is possible that the page may be
619truncated and/or invalidated, then the filesystem must lock invalidate_lock,
620then ensure the page is not already truncated (invalidate_lock will block
621subsequent truncate), and then return with VM_FAULT_LOCKED, and the page
622locked. The VM will unlock the page.
623
624->huge_fault() is called when there is no PUD or PMD entry present.  This
625gives the filesystem the opportunity to install a PUD or PMD sized page.
626Filesystems can also use the ->fault method to return a PMD sized page,
627so implementing this function may not be necessary.  In particular,
628filesystems should not call filemap_fault() from ->huge_fault().
629The mmap_lock may not be held when this method is called.
630
631->map_pages() is called when VM asks to map easy accessible pages.
632Filesystem should find and map pages associated with offsets from "start_pgoff"
633till "end_pgoff". ->map_pages() is called with the RCU lock held and must
634not block.  If it's not possible to reach a page without blocking,
635filesystem should skip it. Filesystem should use set_pte_range() to setup
636page table entry. Pointer to entry associated with the page is passed in
637"pte" field in vm_fault structure. Pointers to entries for other offsets
638should be calculated relative to "pte".
639
640->page_mkwrite() is called when a previously read-only pte is about to become
641writeable. The filesystem again must ensure that there are no
642truncate/invalidate races or races with operations such as ->remap_file_range
643or ->copy_file_range, and then return with the page locked. Usually
644mapping->invalidate_lock is suitable for proper serialization. If the page has
645been truncated, the filesystem should not look up a new page like the ->fault()
646handler, but simply return with VM_FAULT_NOPAGE, which will cause the VM to
647retry the fault.
648
649->pfn_mkwrite() is the same as page_mkwrite but when the pte is
650VM_PFNMAP or VM_MIXEDMAP with a page-less entry. Expected return is
651VM_FAULT_NOPAGE. Or one of the VM_FAULT_ERROR types. The default behavior
652after this call is to make the pte read-write, unless pfn_mkwrite returns
653an error.
654
655->access() is called when get_user_pages() fails in
656access_process_vm(), typically used to debug a process through
657/proc/pid/mem or ptrace.  This function is needed only for
658VM_IO | VM_PFNMAP VMAs.
659
660--------------------------------------------------------------------------------
661
662			Dubious stuff
663
664(if you break something or notice that it is broken and do not fix it yourself
665- at least put it here)
666