1 .. SPDX-License-Identifier: GPL-2.0
2
3 =========================================
4 Overview of the Linux Virtual File System
5 =========================================
6
7 Original author: Richard Gooch <rgooch@atnf.csiro.au>
8
9 - Copyright (C) 1999 Richard Gooch
10 - Copyright (C) 2005 Pekka Enberg
11
12
13 Introduction
14 ============
15
16 The Virtual File System (also known as the Virtual Filesystem Switch) is
17 the software layer in the kernel that provides the filesystem interface
18 to userspace programs. It also provides an abstraction within the
19 kernel which allows different filesystem implementations to coexist.
20
21 VFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so on
22 are called from a process context. Filesystem locking is described in
23 the document Documentation/filesystems/locking.rst.
24
25
26 Directory Entry Cache (dcache)
27 ------------------------------
28
29 The VFS implements the open(2), stat(2), chmod(2), and similar system
30 calls. The pathname argument that is passed to them is used by the VFS
31 to search through the directory entry cache (also known as the dentry
32 cache or dcache). This provides a very fast look-up mechanism to
33 translate a pathname (filename) into a specific dentry. Dentries live
34 in RAM and are never saved to disc: they exist only for performance.
35
36 The dentry cache is meant to be a view into your entire filespace. As
37 most computers cannot fit all dentries in the RAM at the same time, some
38 bits of the cache are missing. In order to resolve your pathname into a
39 dentry, the VFS may have to resort to creating dentries along the way,
40 and then loading the inode. This is done by looking up the inode.
41
42
43 The Inode Object
44 ----------------
45
46 An individual dentry usually has a pointer to an inode. Inodes are
47 filesystem objects such as regular files, directories, FIFOs and other
48 beasts. They live either on the disc (for block device filesystems) or
49 in the memory (for pseudo filesystems). Inodes that live on the disc
50 are copied into the memory when required and changes to the inode are
51 written back to disc. A single inode can be pointed to by multiple
52 dentries (hard links, for example, do this).
53
54 To look up an inode requires that the VFS calls the lookup() method of
55 the parent directory inode. This method is installed by the specific
56 filesystem implementation that the inode lives in. Once the VFS has the
57 required dentry (and hence the inode), we can do all those boring things
58 like open(2) the file, or stat(2) it to peek at the inode data. The
59 stat(2) operation is fairly simple: once the VFS has the dentry, it
60 peeks at the inode data and passes some of it back to userspace.
61
62
63 The File Object
64 ---------------
65
66 Opening a file requires another operation: allocation of a file
67 structure (this is the kernel-side implementation of file descriptors).
68 The freshly allocated file structure is initialized with a pointer to
69 the dentry and a set of file operation member functions. These are
70 taken from the inode data. The open() file method is then called so the
71 specific filesystem implementation can do its work. You can see that
72 this is another switch performed by the VFS. The file structure is
73 placed into the file descriptor table for the process.
74
75 Reading, writing and closing files (and other assorted VFS operations)
76 is done by using the userspace file descriptor to grab the appropriate
77 file structure, and then calling the required file structure method to
78 do whatever is required. For as long as the file is open, it keeps the
79 dentry in use, which in turn means that the VFS inode is still in use.
80
81
82 Registering and Mounting a Filesystem
83 =====================================
84
85 To register and unregister a filesystem, use the following API
86 functions:
87
88 .. code-block:: c
89
90 #include <linux/fs.h>
91
92 extern int register_filesystem(struct file_system_type *);
93 extern int unregister_filesystem(struct file_system_type *);
94
95 The passed struct file_system_type describes your filesystem. When a
96 request is made to mount a filesystem onto a directory in your
97 namespace, the VFS will call the appropriate mount() method for the
98 specific filesystem. New vfsmount referring to the tree returned by
99 ->mount() will be attached to the mountpoint, so that when pathname
100 resolution reaches the mountpoint it will jump into the root of that
101 vfsmount.
102
103 You can see all filesystems that are registered to the kernel in the
104 file /proc/filesystems.
105
106
107 struct file_system_type
108 -----------------------
109
110 This describes the filesystem. The following
111 members are defined:
112
113 .. code-block:: c
114
115 struct file_system_type {
116 const char *name;
117 int fs_flags;
118 int (*init_fs_context)(struct fs_context *);
119 const struct fs_parameter_spec *parameters;
120 struct dentry *(*mount) (struct file_system_type *, int,
121 const char *, void *);
122 void (*kill_sb) (struct super_block *);
123 struct module *owner;
124 struct file_system_type * next;
125 struct hlist_head fs_supers;
126
127 struct lock_class_key s_lock_key;
128 struct lock_class_key s_umount_key;
129 struct lock_class_key s_vfs_rename_key;
130 struct lock_class_key s_writers_key[SB_FREEZE_LEVELS];
131
132 struct lock_class_key i_lock_key;
133 struct lock_class_key i_mutex_key;
134 struct lock_class_key invalidate_lock_key;
135 struct lock_class_key i_mutex_dir_key;
136 };
137
138 ``name``
139 the name of the filesystem type, such as "ext2", "iso9660",
140 "msdos" and so on
141
142 ``fs_flags``
143 various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
144
145 ``init_fs_context``
146 Initializes 'struct fs_context' ->ops and ->fs_private fields with
147 filesystem-specific data.
148
149 ``parameters``
150 Pointer to the array of filesystem parameters descriptors
151 'struct fs_parameter_spec'.
152 More info in Documentation/filesystems/mount_api.rst.
153
154 ``mount``
155 the method to call when a new instance of this filesystem should
156 be mounted
157
158 ``kill_sb``
159 the method to call when an instance of this filesystem should be
160 shut down
161
162
163 ``owner``
164 for internal VFS use: you should initialize this to THIS_MODULE
165 in most cases.
166
167 ``next``
168 for internal VFS use: you should initialize this to NULL
169
170 ``fs_supers``
171 for internal VFS use: hlist of filesystem instances (superblocks)
172
173 s_lock_key, s_umount_key, s_vfs_rename_key, s_writers_key,
174 i_lock_key, i_mutex_key, invalidate_lock_key, i_mutex_dir_key: lockdep-specific
175
176 The mount() method has the following arguments:
177
178 ``struct file_system_type *fs_type``
179 describes the filesystem, partly initialized by the specific
180 filesystem code
181
182 ``int flags``
183 mount flags
184
185 ``const char *dev_name``
186 the device name we are mounting.
187
188 ``void *data``
189 arbitrary mount options, usually comes as an ASCII string (see
190 "Mount Options" section)
191
192 The mount() method must return the root dentry of the tree requested by
193 caller. An active reference to its superblock must be grabbed and the
194 superblock must be locked. On failure it should return ERR_PTR(error).
195
196 The arguments match those of mount(2) and their interpretation depends
197 on filesystem type. E.g. for block filesystems, dev_name is interpreted
198 as block device name, that device is opened and if it contains a
199 suitable filesystem image the method creates and initializes struct
200 super_block accordingly, returning its root dentry to caller.
201
202 ->mount() may choose to return a subtree of existing filesystem - it
203 doesn't have to create a new one. The main result from the caller's
204 point of view is a reference to dentry at the root of (sub)tree to be
205 attached; creation of new superblock is a common side effect.
206
207 The most interesting member of the superblock structure that the mount()
208 method fills in is the "s_op" field. This is a pointer to a "struct
209 super_operations" which describes the next level of the filesystem
210 implementation.
211
212 Usually, a filesystem uses one of the generic mount() implementations
213 and provides a fill_super() callback instead. The generic variants are:
214
215 ``mount_bdev``
216 mount a filesystem residing on a block device
217
218 ``mount_nodev``
219 mount a filesystem that is not backed by a device
220
221 ``mount_single``
222 mount a filesystem which shares the instance between all mounts
223
224 A fill_super() callback implementation has the following arguments:
225
226 ``struct super_block *sb``
227 the superblock structure. The callback must initialize this
228 properly.
229
230 ``void *data``
231 arbitrary mount options, usually comes as an ASCII string (see
232 "Mount Options" section)
233
234 ``int silent``
235 whether or not to be silent on error
236
237
238 The Superblock Object
239 =====================
240
241 A superblock object represents a mounted filesystem.
242
243
244 struct super_operations
245 -----------------------
246
247 This describes how the VFS can manipulate the superblock of your
248 filesystem. The following members are defined:
249
250 .. code-block:: c
251
252 struct super_operations {
253 struct inode *(*alloc_inode)(struct super_block *sb);
254 void (*destroy_inode)(struct inode *);
255 void (*free_inode)(struct inode *);
256
257 void (*dirty_inode) (struct inode *, int flags);
258 int (*write_inode) (struct inode *, struct writeback_control *wbc);
259 int (*drop_inode) (struct inode *);
260 void (*evict_inode) (struct inode *);
261 void (*put_super) (struct super_block *);
262 int (*sync_fs)(struct super_block *sb, int wait);
263 int (*freeze_super) (struct super_block *sb,
264 enum freeze_holder who);
265 int (*freeze_fs) (struct super_block *);
266 int (*thaw_super) (struct super_block *sb,
267 enum freeze_wholder who);
268 int (*unfreeze_fs) (struct super_block *);
269 int (*statfs) (struct dentry *, struct kstatfs *);
270 int (*remount_fs) (struct super_block *, int *, char *);
271 void (*umount_begin) (struct super_block *);
272
273 int (*show_options)(struct seq_file *, struct dentry *);
274 int (*show_devname)(struct seq_file *, struct dentry *);
275 int (*show_path)(struct seq_file *, struct dentry *);
276 int (*show_stats)(struct seq_file *, struct dentry *);
277
278 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
279 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
280 struct dquot **(*get_dquots)(struct inode *);
281
282 long (*nr_cached_objects)(struct super_block *,
283 struct shrink_control *);
284 long (*free_cached_objects)(struct super_block *,
285 struct shrink_control *);
286 };
287
288 All methods are called without any locks being held, unless otherwise
289 noted. This means that most methods can block safely. All methods are
290 only called from a process context (i.e. not from an interrupt handler
291 or bottom half).
292
293 ``alloc_inode``
294 this method is called by alloc_inode() to allocate memory for
295 struct inode and initialize it. If this function is not
296 defined, a simple 'struct inode' is allocated. Normally
297 alloc_inode will be used to allocate a larger structure which
298 contains a 'struct inode' embedded within it.
299
300 ``destroy_inode``
301 this method is called by destroy_inode() to release resources
302 allocated for struct inode. It is only required if
303 ->alloc_inode was defined and simply undoes anything done by
304 ->alloc_inode.
305
306 ``free_inode``
307 this method is called from RCU callback. If you use call_rcu()
308 in ->destroy_inode to free 'struct inode' memory, then it's
309 better to release memory in this method.
310
311 ``dirty_inode``
312 this method is called by the VFS when an inode is marked dirty.
313 This is specifically for the inode itself being marked dirty,
314 not its data. If the update needs to be persisted by fdatasync(),
315 then I_DIRTY_DATASYNC will be set in the flags argument.
316 I_DIRTY_TIME will be set in the flags in case lazytime is enabled
317 and struct inode has times updated since the last ->dirty_inode
318 call.
319
320 ``write_inode``
321 this method is called when the VFS needs to write an inode to
322 disc. The second parameter indicates whether the write should
323 be synchronous or not, not all filesystems check this flag.
324
325 ``drop_inode``
326 called when the last access to the inode is dropped, with the
327 inode->i_lock spinlock held.
328
329 This method should be either NULL (normal UNIX filesystem
330 semantics) or "generic_delete_inode" (for filesystems that do
331 not want to cache inodes - causing "delete_inode" to always be
332 called regardless of the value of i_nlink)
333
334 The "generic_delete_inode()" behavior is equivalent to the old
335 practice of using "force_delete" in the put_inode() case, but
336 does not have the races that the "force_delete()" approach had.
337
338 ``evict_inode``
339 called when the VFS wants to evict an inode. Caller does
340 *not* evict the pagecache or inode-associated metadata buffers;
341 the method has to use truncate_inode_pages_final() to get rid
342 of those. Caller makes sure async writeback cannot be running for
343 the inode while (or after) ->evict_inode() is called. Optional.
344
345 ``put_super``
346 called when the VFS wishes to free the superblock
347 (i.e. unmount). This is called with the superblock lock held
348
349 ``sync_fs``
350 called when VFS is writing out all dirty data associated with a
351 superblock. The second parameter indicates whether the method
352 should wait until the write out has been completed. Optional.
353
354 ``freeze_super``
355 Called instead of ->freeze_fs callback if provided.
356 Main difference is that ->freeze_super is called without taking
357 down_write(&sb->s_umount). If filesystem implements it and wants
358 ->freeze_fs to be called too, then it has to call ->freeze_fs
359 explicitly from this callback. Optional.
360
361 ``freeze_fs``
362 called when VFS is locking a filesystem and forcing it into a
363 consistent state. This method is currently used by the Logical
364 Volume Manager (LVM) and ioctl(FIFREEZE). Optional.
365
366 ``thaw_super``
367 called when VFS is unlocking a filesystem and making it writable
368 again after ->freeze_super. Optional.
369
370 ``unfreeze_fs``
371 called when VFS is unlocking a filesystem and making it writable
372 again after ->freeze_fs. Optional.
373
374 ``statfs``
375 called when the VFS needs to get filesystem statistics.
376
377 ``remount_fs``
378 called when the filesystem is remounted. This is called with
379 the kernel lock held
380
381 ``umount_begin``
382 called when the VFS is unmounting a filesystem.
383
384 ``show_options``
385 called by the VFS to show mount options for /proc/<pid>/mounts
386 and /proc/<pid>/mountinfo.
387 (see "Mount Options" section)
388
389 ``show_devname``
390 Optional. Called by the VFS to show device name for
391 /proc/<pid>/{mounts,mountinfo,mountstats}. If not provided then
392 '(struct mount).mnt_devname' will be used.
393
394 ``show_path``
395 Optional. Called by the VFS (for /proc/<pid>/mountinfo) to show
396 the mount root dentry path relative to the filesystem root.
397
398 ``show_stats``
399 Optional. Called by the VFS (for /proc/<pid>/mountstats) to show
400 filesystem-specific mount statistics.
401
402 ``quota_read``
403 called by the VFS to read from filesystem quota file.
404
405 ``quota_write``
406 called by the VFS to write to filesystem quota file.
407
408 ``get_dquots``
409 called by quota to get 'struct dquot' array for a particular inode.
410 Optional.
411
412 ``nr_cached_objects``
413 called by the sb cache shrinking function for the filesystem to
414 return the number of freeable cached objects it contains.
415 Optional.
416
417 ``free_cache_objects``
418 called by the sb cache shrinking function for the filesystem to
419 scan the number of objects indicated to try to free them.
420 Optional, but any filesystem implementing this method needs to
421 also implement ->nr_cached_objects for it to be called
422 correctly.
423
424 We can't do anything with any errors that the filesystem might
425 encountered, hence the void return type. This will never be
426 called if the VM is trying to reclaim under GFP_NOFS conditions,
427 hence this method does not need to handle that situation itself.
428
429 Implementations must include conditional reschedule calls inside
430 any scanning loop that is done. This allows the VFS to
431 determine appropriate scan batch sizes without having to worry
432 about whether implementations will cause holdoff problems due to
433 large scan batch sizes.
434
435 Whoever sets up the inode is responsible for filling in the "i_op"
436 field. This is a pointer to a "struct inode_operations" which describes
437 the methods that can be performed on individual inodes.
438
439
440 struct xattr_handler
441 ---------------------
442
443 On filesystems that support extended attributes (xattrs), the s_xattr
444 superblock field points to a NULL-terminated array of xattr handlers.
445 Extended attributes are name:value pairs.
446
447 ``name``
448 Indicates that the handler matches attributes with the specified
449 name (such as "system.posix_acl_access"); the prefix field must
450 be NULL.
451
452 ``prefix``
453 Indicates that the handler matches all attributes with the
454 specified name prefix (such as "user."); the name field must be
455 NULL.
456
457 ``list``
458 Determine if attributes matching this xattr handler should be
459 listed for a particular dentry. Used by some listxattr
460 implementations like generic_listxattr.
461
462 ``get``
463 Called by the VFS to get the value of a particular extended
464 attribute. This method is called by the getxattr(2) system
465 call.
466
467 ``set``
468 Called by the VFS to set the value of a particular extended
469 attribute. When the new value is NULL, called to remove a
470 particular extended attribute. This method is called by the
471 setxattr(2) and removexattr(2) system calls.
472
473 When none of the xattr handlers of a filesystem match the specified
474 attribute name or when a filesystem doesn't support extended attributes,
475 the various ``*xattr(2)`` system calls return -EOPNOTSUPP.
476
477
478 The Inode Object
479 ================
480
481 An inode object represents an object within the filesystem.
482
483
484 struct inode_operations
485 -----------------------
486
487 This describes how the VFS can manipulate an inode in your filesystem.
488 As of kernel 2.6.22, the following members are defined:
489
490 .. code-block:: c
491
492 struct inode_operations {
493 int (*create) (struct mnt_idmap *, struct inode *,struct dentry *, umode_t, bool);
494 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
495 int (*link) (struct dentry *,struct inode *,struct dentry *);
496 int (*unlink) (struct inode *,struct dentry *);
497 int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *);
498 int (*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t);
499 int (*rmdir) (struct inode *,struct dentry *);
500 int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t);
501 int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *,
502 struct inode *, struct dentry *, unsigned int);
503 int (*readlink) (struct dentry *, char __user *,int);
504 const char *(*get_link) (struct dentry *, struct inode *,
505 struct delayed_call *);
506 int (*permission) (struct mnt_idmap *, struct inode *, int);
507 struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
508 int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *);
509 int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int);
510 ssize_t (*listxattr) (struct dentry *, char *, size_t);
511 void (*update_time)(struct inode *, struct timespec *, int);
512 int (*atomic_open)(struct inode *, struct dentry *, struct file *,
513 unsigned open_flag, umode_t create_mode);
514 int (*tmpfile) (struct mnt_idmap *, struct inode *, struct file *, umode_t);
515 struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int);
516 int (*set_acl)(struct mnt_idmap *, struct dentry *, struct posix_acl *, int);
517 int (*fileattr_set)(struct mnt_idmap *idmap,
518 struct dentry *dentry, struct fileattr *fa);
519 int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa);
520 struct offset_ctx *(*get_offset_ctx)(struct inode *inode);
521 };
522
523 Again, all methods are called without any locks being held, unless
524 otherwise noted.
525
526 ``create``
527 called by the open(2) and creat(2) system calls. Only required
528 if you want to support regular files. The dentry you get should
529 not have an inode (i.e. it should be a negative dentry). Here
530 you will probably call d_instantiate() with the dentry and the
531 newly created inode
532
533 ``lookup``
534 called when the VFS needs to look up an inode in a parent
535 directory. The name to look for is found in the dentry. This
536 method must call d_add() to insert the found inode into the
537 dentry. The "i_count" field in the inode structure should be
538 incremented. If the named inode does not exist a NULL inode
539 should be inserted into the dentry (this is called a negative
540 dentry). Returning an error code from this routine must only be
541 done on a real error, otherwise creating inodes with system
542 calls like create(2), mknod(2), mkdir(2) and so on will fail.
543 If you wish to overload the dentry methods then you should
544 initialise the "d_dop" field in the dentry; this is a pointer to
545 a struct "dentry_operations". This method is called with the
546 directory inode semaphore held
547
548 ``link``
549 called by the link(2) system call. Only required if you want to
550 support hard links. You will probably need to call
551 d_instantiate() just as you would in the create() method
552
553 ``unlink``
554 called by the unlink(2) system call. Only required if you want
555 to support deleting inodes
556
557 ``symlink``
558 called by the symlink(2) system call. Only required if you want
559 to support symlinks. You will probably need to call
560 d_instantiate() just as you would in the create() method
561
562 ``mkdir``
563 called by the mkdir(2) system call. Only required if you want
564 to support creating subdirectories. You will probably need to
565 call d_instantiate() just as you would in the create() method
566
567 ``rmdir``
568 called by the rmdir(2) system call. Only required if you want
569 to support deleting subdirectories
570
571 ``mknod``
572 called by the mknod(2) system call to create a device (char,
573 block) inode or a named pipe (FIFO) or socket. Only required if
574 you want to support creating these types of inodes. You will
575 probably need to call d_instantiate() just as you would in the
576 create() method
577
578 ``rename``
579 called by the rename(2) system call to rename the object to have
580 the parent and name given by the second inode and dentry.
581
582 The filesystem must return -EINVAL for any unsupported or
583 unknown flags. Currently the following flags are implemented:
584 (1) RENAME_NOREPLACE: this flag indicates that if the target of
585 the rename exists the rename should fail with -EEXIST instead of
586 replacing the target. The VFS already checks for existence, so
587 for local filesystems the RENAME_NOREPLACE implementation is
588 equivalent to plain rename.
589 (2) RENAME_EXCHANGE: exchange source and target. Both must
590 exist; this is checked by the VFS. Unlike plain rename, source
591 and target may be of different type.
592
593 ``get_link``
594 called by the VFS to follow a symbolic link to the inode it
595 points to. Only required if you want to support symbolic links.
596 This method returns the symlink body to traverse (and possibly
597 resets the current position with nd_jump_link()). If the body
598 won't go away until the inode is gone, nothing else is needed;
599 if it needs to be otherwise pinned, arrange for its release by
600 having get_link(..., ..., done) do set_delayed_call(done,
601 destructor, argument). In that case destructor(argument) will
602 be called once VFS is done with the body you've returned. May
603 be called in RCU mode; that is indicated by NULL dentry
604 argument. If request can't be handled without leaving RCU mode,
605 have it return ERR_PTR(-ECHILD).
606
607 If the filesystem stores the symlink target in ->i_link, the
608 VFS may use it directly without calling ->get_link(); however,
609 ->get_link() must still be provided. ->i_link must not be
610 freed until after an RCU grace period. Writing to ->i_link
611 post-iget() time requires a 'release' memory barrier.
612
613 ``readlink``
614 this is now just an override for use by readlink(2) for the
615 cases when ->get_link uses nd_jump_link() or object is not in
616 fact a symlink. Normally filesystems should only implement
617 ->get_link for symlinks and readlink(2) will automatically use
618 that.
619
620 ``permission``
621 called by the VFS to check for access rights on a POSIX-like
622 filesystem.
623
624 May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in
625 rcu-walk mode, the filesystem must check the permission without
626 blocking or storing to the inode.
627
628 If a situation is encountered that rcu-walk cannot handle,
629 return
630 -ECHILD and it will be called again in ref-walk mode.
631
632 ``setattr``
633 called by the VFS to set attributes for a file. This method is
634 called by chmod(2) and related system calls.
635
636 ``getattr``
637 called by the VFS to get attributes of a file. This method is
638 called by stat(2) and related system calls.
639
640 ``listxattr``
641 called by the VFS to list all extended attributes for a given
642 file. This method is called by the listxattr(2) system call.
643
644 ``update_time``
645 called by the VFS to update a specific time or the i_version of
646 an inode. If this is not defined the VFS will update the inode
647 itself and call mark_inode_dirty_sync.
648
649 ``atomic_open``
650 called on the last component of an open. Using this optional
651 method the filesystem can look up, possibly create and open the
652 file in one atomic operation. If it wants to leave actual
653 opening to the caller (e.g. if the file turned out to be a
654 symlink, device, or just something filesystem won't do atomic
655 open for), it may signal this by returning finish_no_open(file,
656 dentry). This method is only called if the last component is
657 negative or needs lookup. Cached positive dentries are still
658 handled by f_op->open(). If the file was created, FMODE_CREATED
659 flag should be set in file->f_mode. In case of O_EXCL the
660 method must only succeed if the file didn't exist and hence
661 FMODE_CREATED shall always be set on success.
662
663 ``tmpfile``
664 called in the end of O_TMPFILE open(). Optional, equivalent to
665 atomically creating, opening and unlinking a file in given
666 directory. On success needs to return with the file already
667 open; this can be done by calling finish_open_simple() right at
668 the end.
669
670 ``fileattr_get``
671 called on ioctl(FS_IOC_GETFLAGS) and ioctl(FS_IOC_FSGETXATTR) to
672 retrieve miscellaneous file flags and attributes. Also called
673 before the relevant SET operation to check what is being changed
674 (in this case with i_rwsem locked exclusive). If unset, then
675 fall back to f_op->ioctl().
676
677 ``fileattr_set``
678 called on ioctl(FS_IOC_SETFLAGS) and ioctl(FS_IOC_FSSETXATTR) to
679 change miscellaneous file flags and attributes. Callers hold
680 i_rwsem exclusive. If unset, then fall back to f_op->ioctl().
681 ``get_offset_ctx``
682 called to get the offset context for a directory inode. A
683 filesystem must define this operation to use
684 simple_offset_dir_operations.
685
686 The Address Space Object
687 ========================
688
689 The address space object is used to group and manage pages in the page
690 cache. It can be used to keep track of the pages in a file (or anything
691 else) and also track the mapping of sections of the file into process
692 address spaces.
693
694 There are a number of distinct yet related services that an
695 address-space can provide. These include communicating memory pressure,
696 page lookup by address, and keeping track of pages tagged as Dirty or
697 Writeback.
698
699 The first can be used independently to the others. The VM can try to
700 either write dirty pages in order to clean them, or release clean pages
701 in order to reuse them. To do this it can call the ->writepage method
702 on dirty pages, and ->release_folio on clean folios with the private
703 flag set. Clean pages without PagePrivate and with no external references
704 will be released without notice being given to the address_space.
705
706 To achieve this functionality, pages need to be placed on an LRU with
707 lru_cache_add and mark_page_active needs to be called whenever the page
708 is used.
709
710 Pages are normally kept in a radix tree index by ->index. This tree
711 maintains information about the PG_Dirty and PG_Writeback status of each
712 page, so that pages with either of these flags can be found quickly.
713
714 The Dirty tag is primarily used by mpage_writepages - the default
715 ->writepages method. It uses the tag to find dirty pages to call
716 ->writepage on. If mpage_writepages is not used (i.e. the address
717 provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is almost
718 unused. write_inode_now and sync_inode do use it (through
719 __sync_single_inode) to check if ->writepages has been successful in
720 writing out the whole address_space.
721
722 The Writeback tag is used by filemap*wait* and sync_page* functions, via
723 filemap_fdatawait_range, to wait for all writeback to complete.
724
725 An address_space handler may attach extra information to a page,
726 typically using the 'private' field in the 'struct page'. If such
727 information is attached, the PG_Private flag should be set. This will
728 cause various VM routines to make extra calls into the address_space
729 handler to deal with that data.
730
731 An address space acts as an intermediate between storage and
732 application. Data is read into the address space a whole page at a
733 time, and provided to the application either by copying of the page, or
734 by memory-mapping the page. Data is written into the address space by
735 the application, and then written-back to storage typically in whole
736 pages, however the address_space has finer control of write sizes.
737
738 The read process essentially only requires 'read_folio'. The write
739 process is more complicated and uses write_begin/write_end or
740 dirty_folio to write data into the address_space, and writepage and
741 writepages to writeback data to storage.
742
743 Adding and removing pages to/from an address_space is protected by the
744 inode's i_mutex.
745
746 When data is written to a page, the PG_Dirty flag should be set. It
747 typically remains set until writepage asks for it to be written. This
748 should clear PG_Dirty and set PG_Writeback. It can be actually written
749 at any point after PG_Dirty is clear. Once it is known to be safe,
750 PG_Writeback is cleared.
751
752 Writeback makes use of a writeback_control structure to direct the
753 operations. This gives the writepage and writepages operations some
754 information about the nature of and reason for the writeback request,
755 and the constraints under which it is being done. It is also used to
756 return information back to the caller about the result of a writepage or
757 writepages request.
758
759
760 Handling errors during writeback
761 --------------------------------
762
763 Most applications that do buffered I/O will periodically call a file
764 synchronization call (fsync, fdatasync, msync or sync_file_range) to
765 ensure that data written has made it to the backing store. When there
766 is an error during writeback, they expect that error to be reported when
767 a file sync request is made. After an error has been reported on one
768 request, subsequent requests on the same file descriptor should return
769 0, unless further writeback errors have occurred since the previous file
770 synchronization.
771
772 Ideally, the kernel would report errors only on file descriptions on
773 which writes were done that subsequently failed to be written back. The
774 generic pagecache infrastructure does not track the file descriptions
775 that have dirtied each individual page however, so determining which
776 file descriptors should get back an error is not possible.
777
778 Instead, the generic writeback error tracking infrastructure in the
779 kernel settles for reporting errors to fsync on all file descriptions
780 that were open at the time that the error occurred. In a situation with
781 multiple writers, all of them will get back an error on a subsequent
782 fsync, even if all of the writes done through that particular file
783 descriptor succeeded (or even if there were no writes on that file
784 descriptor at all).
785
786 Filesystems that wish to use this infrastructure should call
787 mapping_set_error to record the error in the address_space when it
788 occurs. Then, after writing back data from the pagecache in their
789 file->fsync operation, they should call file_check_and_advance_wb_err to
790 ensure that the struct file's error cursor has advanced to the correct
791 point in the stream of errors emitted by the backing device(s).
792
793
794 struct address_space_operations
795 -------------------------------
796
797 This describes how the VFS can manipulate mapping of a file to page
798 cache in your filesystem. The following members are defined:
799
800 .. code-block:: c
801
802 struct address_space_operations {
803 int (*writepage)(struct page *page, struct writeback_control *wbc);
804 int (*read_folio)(struct file *, struct folio *);
805 int (*writepages)(struct address_space *, struct writeback_control *);
806 bool (*dirty_folio)(struct address_space *, struct folio *);
807 void (*readahead)(struct readahead_control *);
808 int (*write_begin)(struct file *, struct address_space *mapping,
809 loff_t pos, unsigned len,
810 struct page **pagep, void **fsdata);
811 int (*write_end)(struct file *, struct address_space *mapping,
812 loff_t pos, unsigned len, unsigned copied,
813 struct folio *folio, void *fsdata);
814 sector_t (*bmap)(struct address_space *, sector_t);
815 void (*invalidate_folio) (struct folio *, size_t start, size_t len);
816 bool (*release_folio)(struct folio *, gfp_t);
817 void (*free_folio)(struct folio *);
818 ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter);
819 int (*migrate_folio)(struct mapping *, struct folio *dst,
820 struct folio *src, enum migrate_mode);
821 int (*launder_folio) (struct folio *);
822
823 bool (*is_partially_uptodate) (struct folio *, size_t from,
824 size_t count);
825 void (*is_dirty_writeback)(struct folio *, bool *, bool *);
826 int (*error_remove_folio)(struct mapping *mapping, struct folio *);
827 int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
828 int (*swap_deactivate)(struct file *);
829 int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
830 };
831
832 ``writepage``
833 called by the VM to write a dirty page to backing store. This
834 may happen for data integrity reasons (i.e. 'sync'), or to free
835 up memory (flush). The difference can be seen in
836 wbc->sync_mode. The PG_Dirty flag has been cleared and
837 PageLocked is true. writepage should start writeout, should set
838 PG_Writeback, and should make sure the page is unlocked, either
839 synchronously or asynchronously when the write operation
840 completes.
841
842 If wbc->sync_mode is WB_SYNC_NONE, ->writepage doesn't have to
843 try too hard if there are problems, and may choose to write out
844 other pages from the mapping if that is easier (e.g. due to
845 internal dependencies). If it chooses not to start writeout, it
846 should return AOP_WRITEPAGE_ACTIVATE so that the VM will not
847 keep calling ->writepage on that page.
848
849 See the file "Locking" for more details.
850
851 ``read_folio``
852 Called by the page cache to read a folio from the backing store.
853 The 'file' argument supplies authentication information to network
854 filesystems, and is generally not used by block based filesystems.
855 It may be NULL if the caller does not have an open file (eg if
856 the kernel is performing a read for itself rather than on behalf
857 of a userspace process with an open file).
858
859 If the mapping does not support large folios, the folio will
860 contain a single page. The folio will be locked when read_folio
861 is called. If the read completes successfully, the folio should
862 be marked uptodate. The filesystem should unlock the folio
863 once the read has completed, whether it was successful or not.
864 The filesystem does not need to modify the refcount on the folio;
865 the page cache holds a reference count and that will not be
866 released until the folio is unlocked.
867
868 Filesystems may implement ->read_folio() synchronously.
869 In normal operation, folios are read through the ->readahead()
870 method. Only if this fails, or if the caller needs to wait for
871 the read to complete will the page cache call ->read_folio().
872 Filesystems should not attempt to perform their own readahead
873 in the ->read_folio() operation.
874
875 If the filesystem cannot perform the read at this time, it can
876 unlock the folio, do whatever action it needs to ensure that the
877 read will succeed in the future and return AOP_TRUNCATED_PAGE.
878 In this case, the caller should look up the folio, lock it,
879 and call ->read_folio again.
880
881 Callers may invoke the ->read_folio() method directly, but using
882 read_mapping_folio() will take care of locking, waiting for the
883 read to complete and handle cases such as AOP_TRUNCATED_PAGE.
884
885 ``writepages``
886 called by the VM to write out pages associated with the
887 address_space object. If wbc->sync_mode is WB_SYNC_ALL, then
888 the writeback_control will specify a range of pages that must be
889 written out. If it is WB_SYNC_NONE, then a nr_to_write is
890 given and that many pages should be written if possible. If no
891 ->writepages is given, then mpage_writepages is used instead.
892 This will choose pages from the address space that are tagged as
893 DIRTY and will pass them to ->writepage.
894
895 ``dirty_folio``
896 called by the VM to mark a folio as dirty. This is particularly
897 needed if an address space attaches private data to a folio, and
898 that data needs to be updated when a folio is dirtied. This is
899 called, for example, when a memory mapped page gets modified.
900 If defined, it should set the folio dirty flag, and the
901 PAGECACHE_TAG_DIRTY search mark in i_pages.
902
903 ``readahead``
904 Called by the VM to read pages associated with the address_space
905 object. The pages are consecutive in the page cache and are
906 locked. The implementation should decrement the page refcount
907 after starting I/O on each page. Usually the page will be
908 unlocked by the I/O completion handler. The set of pages are
909 divided into some sync pages followed by some async pages,
910 rac->ra->async_size gives the number of async pages. The
911 filesystem should attempt to read all sync pages but may decide
912 to stop once it reaches the async pages. If it does decide to
913 stop attempting I/O, it can simply return. The caller will
914 remove the remaining pages from the address space, unlock them
915 and decrement the page refcount. Set PageUptodate if the I/O
916 completes successfully.
917
918 ``write_begin``
919 Called by the generic buffered write code to ask the filesystem
920 to prepare to write len bytes at the given offset in the file.
921 The address_space should check that the write will be able to
922 complete, by allocating space if necessary and doing any other
923 internal housekeeping. If the write will update parts of any
924 basic-blocks on storage, then those blocks should be pre-read
925 (if they haven't been read already) so that the updated blocks
926 can be written out properly.
927
928 The filesystem must return the locked pagecache folio for the
929 specified offset, in ``*foliop``, for the caller to write into.
930
931 It must be able to cope with short writes (where the length
932 passed to write_begin is greater than the number of bytes copied
933 into the folio).
934
935 A void * may be returned in fsdata, which then gets passed into
936 write_end.
937
938 Returns 0 on success; < 0 on failure (which is the error code),
939 in which case write_end is not called.
940
941 ``write_end``
942 After a successful write_begin, and data copy, write_end must be
943 called. len is the original len passed to write_begin, and
944 copied is the amount that was able to be copied.
945
946 The filesystem must take care of unlocking the folio,
947 decrementing its refcount, and updating i_size.
948
949 Returns < 0 on failure, otherwise the number of bytes (<=
950 'copied') that were able to be copied into pagecache.
951
952 ``bmap``
953 called by the VFS to map a logical block offset within object to
954 physical block number. This method is used by the FIBMAP ioctl
955 and for working with swap-files. To be able to swap to a file,
956 the file must have a stable mapping to a block device. The swap
957 system does not go through the filesystem but instead uses bmap
958 to find out where the blocks in the file are and uses those
959 addresses directly.
960
961 ``invalidate_folio``
962 If a folio has private data, then invalidate_folio will be
963 called when part or all of the folio is to be removed from the
964 address space. This generally corresponds to either a
965 truncation, punch hole or a complete invalidation of the address
966 space (in the latter case 'offset' will always be 0 and 'length'
967 will be folio_size()). Any private data associated with the folio
968 should be updated to reflect this truncation. If offset is 0
969 and length is folio_size(), then the private data should be
970 released, because the folio must be able to be completely
971 discarded. This may be done by calling the ->release_folio
972 function, but in this case the release MUST succeed.
973
974 ``release_folio``
975 release_folio is called on folios with private data to tell the
976 filesystem that the folio is about to be freed. ->release_folio
977 should remove any private data from the folio and clear the
978 private flag. If release_folio() fails, it should return false.
979 release_folio() is used in two distinct though related cases.
980 The first is when the VM wants to free a clean folio with no
981 active users. If ->release_folio succeeds, the folio will be
982 removed from the address_space and be freed.
983
984 The second case is when a request has been made to invalidate
985 some or all folios in an address_space. This can happen
986 through the fadvise(POSIX_FADV_DONTNEED) system call or by the
987 filesystem explicitly requesting it as nfs and 9p do (when they
988 believe the cache may be out of date with storage) by calling
989 invalidate_inode_pages2(). If the filesystem makes such a call,
990 and needs to be certain that all folios are invalidated, then
991 its release_folio will need to ensure this. Possibly it can
992 clear the uptodate flag if it cannot free private data yet.
993
994 ``free_folio``
995 free_folio is called once the folio is no longer visible in the
996 page cache in order to allow the cleanup of any private data.
997 Since it may be called by the memory reclaimer, it should not
998 assume that the original address_space mapping still exists, and
999 it should not block.
1000
1001 ``direct_IO``
1002 called by the generic read/write routines to perform direct_IO -
1003 that is IO requests which bypass the page cache and transfer
1004 data directly between the storage and the application's address
1005 space.
1006
1007 ``migrate_folio``
1008 This is used to compact the physical memory usage. If the VM
1009 wants to relocate a folio (maybe from a memory device that is
1010 signalling imminent failure) it will pass a new folio and an old
1011 folio to this function. migrate_folio should transfer any private
1012 data across and update any references that it has to the folio.
1013
1014 ``launder_folio``
1015 Called before freeing a folio - it writes back the dirty folio.
1016 To prevent redirtying the folio, it is kept locked during the
1017 whole operation.
1018
1019 ``is_partially_uptodate``
1020 Called by the VM when reading a file through the pagecache when
1021 the underlying blocksize is smaller than the size of the folio.
1022 If the required block is up to date then the read can complete
1023 without needing I/O to bring the whole page up to date.
1024
1025 ``is_dirty_writeback``
1026 Called by the VM when attempting to reclaim a folio. The VM uses
1027 dirty and writeback information to determine if it needs to
1028 stall to allow flushers a chance to complete some IO.
1029 Ordinarily it can use folio_test_dirty and folio_test_writeback but
1030 some filesystems have more complex state (unstable folios in NFS
1031 prevent reclaim) or do not set those flags due to locking
1032 problems. This callback allows a filesystem to indicate to the
1033 VM if a folio should be treated as dirty or writeback for the
1034 purposes of stalling.
1035
1036 ``error_remove_folio``
1037 normally set to generic_error_remove_folio if truncation is ok
1038 for this address space. Used for memory failure handling.
1039 Setting this implies you deal with pages going away under you,
1040 unless you have them locked or reference counts increased.
1041
1042 ``swap_activate``
1043
1044 Called to prepare the given file for swap. It should perform
1045 any validation and preparation necessary to ensure that writes
1046 can be performed with minimal memory allocation. It should call
1047 add_swap_extent(), or the helper iomap_swapfile_activate(), and
1048 return the number of extents added. If IO should be submitted
1049 through ->swap_rw(), it should set SWP_FS_OPS, otherwise IO will
1050 be submitted directly to the block device ``sis->bdev``.
1051
1052 ``swap_deactivate``
1053 Called during swapoff on files where swap_activate was
1054 successful.
1055
1056 ``swap_rw``
1057 Called to read or write swap pages when SWP_FS_OPS is set.
1058
1059 The File Object
1060 ===============
1061
1062 A file object represents a file opened by a process. This is also known
1063 as an "open file description" in POSIX parlance.
1064
1065
1066 struct file_operations
1067 ----------------------
1068
1069 This describes how the VFS can manipulate an open file. As of kernel
1070 4.18, the following members are defined:
1071
1072 .. code-block:: c
1073
1074 struct file_operations {
1075 struct module *owner;
1076 loff_t (*llseek) (struct file *, loff_t, int);
1077 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
1078 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
1079 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
1080 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
1081 int (*iopoll)(struct kiocb *kiocb, bool spin);
1082 int (*iterate_shared) (struct file *, struct dir_context *);
1083 __poll_t (*poll) (struct file *, struct poll_table_struct *);
1084 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
1085 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
1086 int (*mmap) (struct file *, struct vm_area_struct *);
1087 int (*open) (struct inode *, struct file *);
1088 int (*flush) (struct file *, fl_owner_t id);
1089 int (*release) (struct inode *, struct file *);
1090 int (*fsync) (struct file *, loff_t, loff_t, int datasync);
1091 int (*fasync) (int, struct file *, int);
1092 int (*lock) (struct file *, int, struct file_lock *);
1093 unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1094 int (*check_flags)(int);
1095 int (*flock) (struct file *, int, struct file_lock *);
1096 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
1097 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
1098 int (*setlease)(struct file *, long, struct file_lock **, void **);
1099 long (*fallocate)(struct file *file, int mode, loff_t offset,
1100 loff_t len);
1101 void (*show_fdinfo)(struct seq_file *m, struct file *f);
1102 #ifndef CONFIG_MMU
1103 unsigned (*mmap_capabilities)(struct file *);
1104 #endif
1105 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *, loff_t, size_t, unsigned int);
1106 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
1107 struct file *file_out, loff_t pos_out,
1108 loff_t len, unsigned int remap_flags);
1109 int (*fadvise)(struct file *, loff_t, loff_t, int);
1110 };
1111
1112 Again, all methods are called without any locks being held, unless
1113 otherwise noted.
1114
1115 ``llseek``
1116 called when the VFS needs to move the file position index
1117
1118 ``read``
1119 called by read(2) and related system calls
1120
1121 ``read_iter``
1122 possibly asynchronous read with iov_iter as destination
1123
1124 ``write``
1125 called by write(2) and related system calls
1126
1127 ``write_iter``
1128 possibly asynchronous write with iov_iter as source
1129
1130 ``iopoll``
1131 called when aio wants to poll for completions on HIPRI iocbs
1132
1133 ``iterate_shared``
1134 called when the VFS needs to read the directory contents
1135
1136 ``poll``
1137 called by the VFS when a process wants to check if there is
1138 activity on this file and (optionally) go to sleep until there
1139 is activity. Called by the select(2) and poll(2) system calls
1140
1141 ``unlocked_ioctl``
1142 called by the ioctl(2) system call.
1143
1144 ``compat_ioctl``
1145 called by the ioctl(2) system call when 32 bit system calls are
1146 used on 64 bit kernels.
1147
1148 ``mmap``
1149 called by the mmap(2) system call
1150
1151 ``open``
1152 called by the VFS when an inode should be opened. When the VFS
1153 opens a file, it creates a new "struct file". It then calls the
1154 open method for the newly allocated file structure. You might
1155 think that the open method really belongs in "struct
1156 inode_operations", and you may be right. I think it's done the
1157 way it is because it makes filesystems simpler to implement.
1158 The open() method is a good place to initialize the
1159 "private_data" member in the file structure if you want to point
1160 to a device structure
1161
1162 ``flush``
1163 called by the close(2) system call to flush a file
1164
1165 ``release``
1166 called when the last reference to an open file is closed
1167
1168 ``fsync``
1169 called by the fsync(2) system call. Also see the section above
1170 entitled "Handling errors during writeback".
1171
1172 ``fasync``
1173 called by the fcntl(2) system call when asynchronous
1174 (non-blocking) mode is enabled for a file
1175
1176 ``lock``
1177 called by the fcntl(2) system call for F_GETLK, F_SETLK, and
1178 F_SETLKW commands
1179
1180 ``get_unmapped_area``
1181 called by the mmap(2) system call
1182
1183 ``check_flags``
1184 called by the fcntl(2) system call for F_SETFL command
1185
1186 ``flock``
1187 called by the flock(2) system call
1188
1189 ``splice_write``
1190 called by the VFS to splice data from a pipe to a file. This
1191 method is used by the splice(2) system call
1192
1193 ``splice_read``
1194 called by the VFS to splice data from file to a pipe. This
1195 method is used by the splice(2) system call
1196
1197 ``setlease``
1198 called by the VFS to set or release a file lock lease. setlease
1199 implementations should call generic_setlease to record or remove
1200 the lease in the inode after setting it.
1201
1202 ``fallocate``
1203 called by the VFS to preallocate blocks or punch a hole.
1204
1205 ``copy_file_range``
1206 called by the copy_file_range(2) system call.
1207
1208 ``remap_file_range``
1209 called by the ioctl(2) system call for FICLONERANGE and FICLONE
1210 and FIDEDUPERANGE commands to remap file ranges. An
1211 implementation should remap len bytes at pos_in of the source
1212 file into the dest file at pos_out. Implementations must handle
1213 callers passing in len == 0; this means "remap to the end of the
1214 source file". The return value should the number of bytes
1215 remapped, or the usual negative error code if errors occurred
1216 before any bytes were remapped. The remap_flags parameter
1217 accepts REMAP_FILE_* flags. If REMAP_FILE_DEDUP is set then the
1218 implementation must only remap if the requested file ranges have
1219 identical contents. If REMAP_FILE_CAN_SHORTEN is set, the caller is
1220 ok with the implementation shortening the request length to
1221 satisfy alignment or EOF requirements (or any other reason).
1222
1223 ``fadvise``
1224 possibly called by the fadvise64() system call.
1225
1226 Note that the file operations are implemented by the specific
1227 filesystem in which the inode resides. When opening a device node
1228 (character or block special) most filesystems will call special
1229 support routines in the VFS which will locate the required device
1230 driver information. These support routines replace the filesystem file
1231 operations with those for the device driver, and then proceed to call
1232 the new open() method for the file. This is how opening a device file
1233 in the filesystem eventually ends up calling the device driver open()
1234 method.
1235
1236
1237 Directory Entry Cache (dcache)
1238 ==============================
1239
1240
1241 struct dentry_operations
1242 ------------------------
1243
1244 This describes how a filesystem can overload the standard dentry
1245 operations. Dentries and the dcache are the domain of the VFS and the
1246 individual filesystem implementations. Device drivers have no business
1247 here. These methods may be set to NULL, as they are either optional or
1248 the VFS uses a default. As of kernel 2.6.22, the following members are
1249 defined:
1250
1251 .. code-block:: c
1252
1253 struct dentry_operations {
1254 int (*d_revalidate)(struct dentry *, unsigned int);
1255 int (*d_weak_revalidate)(struct dentry *, unsigned int);
1256 int (*d_hash)(const struct dentry *, struct qstr *);
1257 int (*d_compare)(const struct dentry *,
1258 unsigned int, const char *, const struct qstr *);
1259 int (*d_delete)(const struct dentry *);
1260 int (*d_init)(struct dentry *);
1261 void (*d_release)(struct dentry *);
1262 void (*d_iput)(struct dentry *, struct inode *);
1263 char *(*d_dname)(struct dentry *, char *, int);
1264 struct vfsmount *(*d_automount)(struct path *);
1265 int (*d_manage)(const struct path *, bool);
1266 struct dentry *(*d_real)(struct dentry *, enum d_real_type type);
1267 };
1268
1269 ``d_revalidate``
1270 called when the VFS needs to revalidate a dentry. This is
1271 called whenever a name look-up finds a dentry in the dcache.
1272 Most local filesystems leave this as NULL, because all their
1273 dentries in the dcache are valid. Network filesystems are
1274 different since things can change on the server without the
1275 client necessarily being aware of it.
1276
1277 This function should return a positive value if the dentry is
1278 still valid, and zero or a negative error code if it isn't.
1279
1280 d_revalidate may be called in rcu-walk mode (flags &
1281 LOOKUP_RCU). If in rcu-walk mode, the filesystem must
1282 revalidate the dentry without blocking or storing to the dentry,
1283 d_parent and d_inode should not be used without care (because
1284 they can change and, in d_inode case, even become NULL under
1285 us).
1286
1287 If a situation is encountered that rcu-walk cannot handle,
1288 return
1289 -ECHILD and it will be called again in ref-walk mode.
1290
1291 ``d_weak_revalidate``
1292 called when the VFS needs to revalidate a "jumped" dentry. This
1293 is called when a path-walk ends at dentry that was not acquired
1294 by doing a lookup in the parent directory. This includes "/",
1295 "." and "..", as well as procfs-style symlinks and mountpoint
1296 traversal.
1297
1298 In this case, we are less concerned with whether the dentry is
1299 still fully correct, but rather that the inode is still valid.
1300 As with d_revalidate, most local filesystems will set this to
1301 NULL since their dcache entries are always valid.
1302
1303 This function has the same return code semantics as
1304 d_revalidate.
1305
1306 d_weak_revalidate is only called after leaving rcu-walk mode.
1307
1308 ``d_hash``
1309 called when the VFS adds a dentry to the hash table. The first
1310 dentry passed to d_hash is the parent directory that the name is
1311 to be hashed into.
1312
1313 Same locking and synchronisation rules as d_compare regarding
1314 what is safe to dereference etc.
1315
1316 ``d_compare``
1317 called to compare a dentry name with a given name. The first
1318 dentry is the parent of the dentry to be compared, the second is
1319 the child dentry. len and name string are properties of the
1320 dentry to be compared. qstr is the name to compare it with.
1321
1322 Must be constant and idempotent, and should not take locks if
1323 possible, and should not or store into the dentry. Should not
1324 dereference pointers outside the dentry without lots of care
1325 (eg. d_parent, d_inode, d_name should not be used).
1326
1327 However, our vfsmount is pinned, and RCU held, so the dentries
1328 and inodes won't disappear, neither will our sb or filesystem
1329 module. ->d_sb may be used.
1330
1331 It is a tricky calling convention because it needs to be called
1332 under "rcu-walk", ie. without any locks or references on things.
1333
1334 ``d_delete``
1335 called when the last reference to a dentry is dropped and the
1336 dcache is deciding whether or not to cache it. Return 1 to
1337 delete immediately, or 0 to cache the dentry. Default is NULL
1338 which means to always cache a reachable dentry. d_delete must
1339 be constant and idempotent.
1340
1341 ``d_init``
1342 called when a dentry is allocated
1343
1344 ``d_release``
1345 called when a dentry is really deallocated
1346
1347 ``d_iput``
1348 called when a dentry loses its inode (just prior to its being
1349 deallocated). The default when this is NULL is that the VFS
1350 calls iput(). If you define this method, you must call iput()
1351 yourself
1352
1353 ``d_dname``
1354 called when the pathname of a dentry should be generated.
1355 Useful for some pseudo filesystems (sockfs, pipefs, ...) to
1356 delay pathname generation. (Instead of doing it when dentry is
1357 created, it's done only when the path is needed.). Real
1358 filesystems probably dont want to use it, because their dentries
1359 are present in global dcache hash, so their hash should be an
1360 invariant. As no lock is held, d_dname() should not try to
1361 modify the dentry itself, unless appropriate SMP safety is used.
1362 CAUTION : d_path() logic is quite tricky. The correct way to
1363 return for example "Hello" is to put it at the end of the
1364 buffer, and returns a pointer to the first char.
1365 dynamic_dname() helper function is provided to take care of
1366 this.
1367
1368 Example :
1369
1370 .. code-block:: c
1371
1372 static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1373 {
1374 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1375 dentry->d_inode->i_ino);
1376 }
1377
1378 ``d_automount``
1379 called when an automount dentry is to be traversed (optional).
1380 This should create a new VFS mount record and return the record
1381 to the caller. The caller is supplied with a path parameter
1382 giving the automount directory to describe the automount target
1383 and the parent VFS mount record to provide inheritable mount
1384 parameters. NULL should be returned if someone else managed to
1385 make the automount first. If the vfsmount creation failed, then
1386 an error code should be returned. If -EISDIR is returned, then
1387 the directory will be treated as an ordinary directory and
1388 returned to pathwalk to continue walking.
1389
1390 If a vfsmount is returned, the caller will attempt to mount it
1391 on the mountpoint and will remove the vfsmount from its
1392 expiration list in the case of failure. The vfsmount should be
1393 returned with 2 refs on it to prevent automatic expiration - the
1394 caller will clean up the additional ref.
1395
1396 This function is only used if DCACHE_NEED_AUTOMOUNT is set on
1397 the dentry. This is set by __d_instantiate() if S_AUTOMOUNT is
1398 set on the inode being added.
1399
1400 ``d_manage``
1401 called to allow the filesystem to manage the transition from a
1402 dentry (optional). This allows autofs, for example, to hold up
1403 clients waiting to explore behind a 'mountpoint' while letting
1404 the daemon go past and construct the subtree there. 0 should be
1405 returned to let the calling process continue. -EISDIR can be
1406 returned to tell pathwalk to use this directory as an ordinary
1407 directory and to ignore anything mounted on it and not to check
1408 the automount flag. Any other error code will abort pathwalk
1409 completely.
1410
1411 If the 'rcu_walk' parameter is true, then the caller is doing a
1412 pathwalk in RCU-walk mode. Sleeping is not permitted in this
1413 mode, and the caller can be asked to leave it and call again by
1414 returning -ECHILD. -EISDIR may also be returned to tell
1415 pathwalk to ignore d_automount or any mounts.
1416
1417 This function is only used if DCACHE_MANAGE_TRANSIT is set on
1418 the dentry being transited from.
1419
1420 ``d_real``
1421 overlay/union type filesystems implement this method to return one
1422 of the underlying dentries of a regular file hidden by the overlay.
1423
1424 The 'type' argument takes the values D_REAL_DATA or D_REAL_METADATA
1425 for returning the real underlying dentry that refers to the inode
1426 hosting the file's data or metadata respectively.
1427
1428 For non-regular files, the 'dentry' argument is returned.
1429
1430 Each dentry has a pointer to its parent dentry, as well as a hash list
1431 of child dentries. Child dentries are basically like files in a
1432 directory.
1433
1434
1435 Directory Entry Cache API
1436 --------------------------
1437
1438 There are a number of functions defined which permit a filesystem to
1439 manipulate dentries:
1440
1441 ``dget``
1442 open a new handle for an existing dentry (this just increments
1443 the usage count)
1444
1445 ``dput``
1446 close a handle for a dentry (decrements the usage count). If
1447 the usage count drops to 0, and the dentry is still in its
1448 parent's hash, the "d_delete" method is called to check whether
1449 it should be cached. If it should not be cached, or if the
1450 dentry is not hashed, it is deleted. Otherwise cached dentries
1451 are put into an LRU list to be reclaimed on memory shortage.
1452
1453 ``d_drop``
1454 this unhashes a dentry from its parents hash list. A subsequent
1455 call to dput() will deallocate the dentry if its usage count
1456 drops to 0
1457
1458 ``d_delete``
1459 delete a dentry. If there are no other open references to the
1460 dentry then the dentry is turned into a negative dentry (the
1461 d_iput() method is called). If there are other references, then
1462 d_drop() is called instead
1463
1464 ``d_add``
1465 add a dentry to its parents hash list and then calls
1466 d_instantiate()
1467
1468 ``d_instantiate``
1469 add a dentry to the alias hash list for the inode and updates
1470 the "d_inode" member. The "i_count" member in the inode
1471 structure should be set/incremented. If the inode pointer is
1472 NULL, the dentry is called a "negative dentry". This function
1473 is commonly called when an inode is created for an existing
1474 negative dentry
1475
1476 ``d_lookup``
1477 look up a dentry given its parent and path name component It
1478 looks up the child of that given name from the dcache hash
1479 table. If it is found, the reference count is incremented and
1480 the dentry is returned. The caller must use dput() to free the
1481 dentry when it finishes using it.
1482
1483
1484 Mount Options
1485 =============
1486
1487
1488 Parsing options
1489 ---------------
1490
1491 On mount and remount the filesystem is passed a string containing a
1492 comma separated list of mount options. The options can have either of
1493 these forms:
1494
1495 option
1496 option=value
1497
1498 The <linux/parser.h> header defines an API that helps parse these
1499 options. There are plenty of examples on how to use it in existing
1500 filesystems.
1501
1502
1503 Showing options
1504 ---------------
1505
1506 If a filesystem accepts mount options, it must define show_options() to
1507 show all the currently active options. The rules are:
1508
1509 - options MUST be shown which are not default or their values differ
1510 from the default
1511
1512 - options MAY be shown which are enabled by default or have their
1513 default value
1514
1515 Options used only internally between a mount helper and the kernel (such
1516 as file descriptors), or which only have an effect during the mounting
1517 (such as ones controlling the creation of a journal) are exempt from the
1518 above rules.
1519
1520 The underlying reason for the above rules is to make sure, that a mount
1521 can be accurately replicated (e.g. umounting and mounting again) based
1522 on the information found in /proc/mounts.
1523
1524
1525 Resources
1526 =========
1527
1528 (Note some of these resources are not up-to-date with the latest kernel
1529 version.)
1530
1531 Creating Linux virtual filesystems. 2002
1532 <https://lwn.net/Articles/13325/>
1533
1534 The Linux Virtual File-system Layer by Neil Brown. 1999
1535 <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1536
1537 A tour of the Linux VFS by Michael K. Johnson. 1996
1538 <https://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1539
1540 A small trail through the Linux kernel by Andries Brouwer. 2001
1541 <https://www.win.tue.nl/~aeb/linux/vfs/trail.html>
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.