Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1.. SPDX-License-Identifier: GPL-2.0
2
3=========================================
4Overview of the Linux Virtual File System
5=========================================
6
7Original author: Richard Gooch <rgooch@atnf.csiro.au>
8
9- Copyright (C) 1999 Richard Gooch
10- Copyright (C) 2005 Pekka Enberg
11
12
13Introduction
14============
15
16The Virtual File System (also known as the Virtual Filesystem Switch) is
17the software layer in the kernel that provides the filesystem interface
18to userspace programs. It also provides an abstraction within the
19kernel which allows different filesystem implementations to coexist.
20
21VFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so on
22are called from a process context. Filesystem locking is described in
23the document Documentation/filesystems/locking.rst.
24
25
26Directory Entry Cache (dcache)
27------------------------------
28
29The VFS implements the open(2), stat(2), chmod(2), and similar system
30calls. The pathname argument that is passed to them is used by the VFS
31to search through the directory entry cache (also known as the dentry
32cache or dcache). This provides a very fast look-up mechanism to
33translate a pathname (filename) into a specific dentry. Dentries live
34in RAM and are never saved to disc: they exist only for performance.
35
36The dentry cache is meant to be a view into your entire filespace. As
37most computers cannot fit all dentries in the RAM at the same time, some
38bits of the cache are missing. In order to resolve your pathname into a
39dentry, the VFS may have to resort to creating dentries along the way,
40and then loading the inode. This is done by looking up the inode.
41
42
43The Inode Object
44----------------
45
46An individual dentry usually has a pointer to an inode. Inodes are
47filesystem objects such as regular files, directories, FIFOs and other
48beasts. They live either on the disc (for block device filesystems) or
49in the memory (for pseudo filesystems). Inodes that live on the disc
50are copied into the memory when required and changes to the inode are
51written back to disc. A single inode can be pointed to by multiple
52dentries (hard links, for example, do this).
53
54To look up an inode requires that the VFS calls the lookup() method of
55the parent directory inode. This method is installed by the specific
56filesystem implementation that the inode lives in. Once the VFS has the
57required dentry (and hence the inode), we can do all those boring things
58like open(2) the file, or stat(2) it to peek at the inode data. The
59stat(2) operation is fairly simple: once the VFS has the dentry, it
60peeks at the inode data and passes some of it back to userspace.
61
62
63The File Object
64---------------
65
66Opening a file requires another operation: allocation of a file
67structure (this is the kernel-side implementation of file descriptors).
68The freshly allocated file structure is initialized with a pointer to
69the dentry and a set of file operation member functions. These are
70taken from the inode data. The open() file method is then called so the
71specific filesystem implementation can do its work. You can see that
72this is another switch performed by the VFS. The file structure is
73placed into the file descriptor table for the process.
74
75Reading, writing and closing files (and other assorted VFS operations)
76is done by using the userspace file descriptor to grab the appropriate
77file structure, and then calling the required file structure method to
78do whatever is required. For as long as the file is open, it keeps the
79dentry in use, which in turn means that the VFS inode is still in use.
80
81
82Registering and Mounting a Filesystem
83=====================================
84
85To register and unregister a filesystem, use the following API
86functions:
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
95The passed struct file_system_type describes your filesystem. When a
96request is made to mount a filesystem onto a directory in your
97namespace, the VFS will call the appropriate get_tree() method for the
98specific filesystem. See Documentation/filesystems/mount_api.rst
99for more details.
100
101You can see all filesystems that are registered to the kernel in the
102file /proc/filesystems.
103
104
105struct file_system_type
106-----------------------
107
108This describes the filesystem. The following
109members are defined:
110
111.. code-block:: c
112
113 struct file_system_type {
114 const char *name;
115 int fs_flags;
116 int (*init_fs_context)(struct fs_context *);
117 const struct fs_parameter_spec *parameters;
118 void (*kill_sb) (struct super_block *);
119 struct module *owner;
120 struct file_system_type * next;
121 struct hlist_head fs_supers;
122
123 struct lock_class_key s_lock_key;
124 struct lock_class_key s_umount_key;
125 struct lock_class_key s_vfs_rename_key;
126 struct lock_class_key s_writers_key[SB_FREEZE_LEVELS];
127
128 struct lock_class_key i_lock_key;
129 struct lock_class_key i_mutex_key;
130 struct lock_class_key invalidate_lock_key;
131 struct lock_class_key i_mutex_dir_key;
132 };
133
134``name``
135 the name of the filesystem type, such as "ext2", "iso9660",
136 "msdos" and so on
137
138``fs_flags``
139 various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
140
141``init_fs_context``
142 Initializes 'struct fs_context' ->ops and ->fs_private fields with
143 filesystem-specific data.
144
145``parameters``
146 Pointer to the array of filesystem parameters descriptors
147 'struct fs_parameter_spec'.
148 More info in Documentation/filesystems/mount_api.rst.
149
150``kill_sb``
151 the method to call when an instance of this filesystem should be
152 shut down
153
154
155``owner``
156 for internal VFS use: you should initialize this to THIS_MODULE
157 in most cases.
158
159``next``
160 for internal VFS use: you should initialize this to NULL
161
162``fs_supers``
163 for internal VFS use: hlist of filesystem instances (superblocks)
164
165 s_lock_key, s_umount_key, s_vfs_rename_key, s_writers_key,
166 i_lock_key, i_mutex_key, invalidate_lock_key, i_mutex_dir_key: lockdep-specific
167
168The Superblock Object
169=====================
170
171A superblock object represents a mounted filesystem.
172
173
174struct super_operations
175-----------------------
176
177This describes how the VFS can manipulate the superblock of your
178filesystem. The following members are defined:
179
180.. code-block:: c
181
182 struct super_operations {
183 struct inode *(*alloc_inode)(struct super_block *sb);
184 void (*destroy_inode)(struct inode *);
185 void (*free_inode)(struct inode *);
186
187 void (*dirty_inode) (struct inode *, int flags);
188 int (*write_inode) (struct inode *, struct writeback_control *wbc);
189 int (*drop_inode) (struct inode *);
190 void (*evict_inode) (struct inode *);
191 void (*put_super) (struct super_block *);
192 int (*sync_fs)(struct super_block *sb, int wait);
193 int (*freeze_super) (struct super_block *sb,
194 enum freeze_holder who);
195 int (*freeze_fs) (struct super_block *);
196 int (*thaw_super) (struct super_block *sb,
197 enum freeze_wholder who);
198 int (*unfreeze_fs) (struct super_block *);
199 int (*statfs) (struct dentry *, struct kstatfs *);
200 void (*umount_begin) (struct super_block *);
201
202 int (*show_options)(struct seq_file *, struct dentry *);
203 int (*show_devname)(struct seq_file *, struct dentry *);
204 int (*show_path)(struct seq_file *, struct dentry *);
205 int (*show_stats)(struct seq_file *, struct dentry *);
206
207 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
208 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
209 struct dquot **(*get_dquots)(struct inode *);
210
211 long (*nr_cached_objects)(struct super_block *,
212 struct shrink_control *);
213 long (*free_cached_objects)(struct super_block *,
214 struct shrink_control *);
215 };
216
217All methods are called without any locks being held, unless otherwise
218noted. This means that most methods can block safely. All methods are
219only called from a process context (i.e. not from an interrupt handler
220or bottom half).
221
222``alloc_inode``
223 this method is called by alloc_inode() to allocate memory for
224 struct inode and initialize it. If this function is not
225 defined, a simple 'struct inode' is allocated. Normally
226 alloc_inode will be used to allocate a larger structure which
227 contains a 'struct inode' embedded within it.
228
229``destroy_inode``
230 this method is called by destroy_inode() to release resources
231 allocated for struct inode. It is only required if
232 ->alloc_inode was defined and simply undoes anything done by
233 ->alloc_inode.
234
235``free_inode``
236 this method is called from RCU callback. If you use call_rcu()
237 in ->destroy_inode to free 'struct inode' memory, then it's
238 better to release memory in this method.
239
240``dirty_inode``
241 this method is called by the VFS when an inode is marked dirty.
242 This is specifically for the inode itself being marked dirty,
243 not its data. If the update needs to be persisted by fdatasync(),
244 then I_DIRTY_DATASYNC will be set in the flags argument.
245 I_DIRTY_TIME will be set in the flags in case lazytime is enabled
246 and struct inode has times updated since the last ->dirty_inode
247 call.
248
249``write_inode``
250 this method is called when the VFS needs to write an inode to
251 disc. The second parameter indicates whether the write should
252 be synchronous or not, not all filesystems check this flag.
253
254``drop_inode``
255 called when the last access to the inode is dropped, with the
256 inode->i_lock spinlock held.
257
258 This method should be either NULL (normal UNIX filesystem
259 semantics) or "inode_just_drop" (for filesystems that do
260 not want to cache inodes - causing "delete_inode" to always be
261 called regardless of the value of i_nlink)
262
263 The "inode_just_drop()" behavior is equivalent to the old
264 practice of using "force_delete" in the put_inode() case, but
265 does not have the races that the "force_delete()" approach had.
266
267``evict_inode``
268 called when the VFS wants to evict an inode. Caller does
269 *not* evict the pagecache or inode-associated metadata buffers;
270 the method has to use truncate_inode_pages_final() to get rid
271 of those. Caller makes sure async writeback cannot be running for
272 the inode while (or after) ->evict_inode() is called. Optional.
273
274``put_super``
275 called when the VFS wishes to free the superblock
276 (i.e. unmount). This is called with the superblock lock held
277
278``sync_fs``
279 called when VFS is writing out all dirty data associated with a
280 superblock. The second parameter indicates whether the method
281 should wait until the write out has been completed. Optional.
282
283``freeze_super``
284 Called instead of ->freeze_fs callback if provided.
285 Main difference is that ->freeze_super is called without taking
286 down_write(&sb->s_umount). If filesystem implements it and wants
287 ->freeze_fs to be called too, then it has to call ->freeze_fs
288 explicitly from this callback. Optional.
289
290``freeze_fs``
291 called when VFS is locking a filesystem and forcing it into a
292 consistent state. This method is currently used by the Logical
293 Volume Manager (LVM) and ioctl(FIFREEZE). Optional.
294
295``thaw_super``
296 called when VFS is unlocking a filesystem and making it writable
297 again after ->freeze_super. Optional.
298
299``unfreeze_fs``
300 called when VFS is unlocking a filesystem and making it writable
301 again after ->freeze_fs. Optional.
302
303``statfs``
304 called when the VFS needs to get filesystem statistics.
305
306``umount_begin``
307 called when the VFS is unmounting a filesystem.
308
309``show_options``
310 called by the VFS to show mount options for /proc/<pid>/mounts
311 and /proc/<pid>/mountinfo.
312 (see "Mount Options" section)
313
314``show_devname``
315 Optional. Called by the VFS to show device name for
316 /proc/<pid>/{mounts,mountinfo,mountstats}. If not provided then
317 '(struct mount).mnt_devname' will be used.
318
319``show_path``
320 Optional. Called by the VFS (for /proc/<pid>/mountinfo) to show
321 the mount root dentry path relative to the filesystem root.
322
323``show_stats``
324 Optional. Called by the VFS (for /proc/<pid>/mountstats) to show
325 filesystem-specific mount statistics.
326
327``quota_read``
328 called by the VFS to read from filesystem quota file.
329
330``quota_write``
331 called by the VFS to write to filesystem quota file.
332
333``get_dquots``
334 called by quota to get 'struct dquot' array for a particular inode.
335 Optional.
336
337``nr_cached_objects``
338 called by the sb cache shrinking function for the filesystem to
339 return the number of freeable cached objects it contains.
340 Optional.
341
342``free_cache_objects``
343 called by the sb cache shrinking function for the filesystem to
344 scan the number of objects indicated to try to free them.
345 Optional, but any filesystem implementing this method needs to
346 also implement ->nr_cached_objects for it to be called
347 correctly.
348
349 We can't do anything with any errors that the filesystem might
350 encountered, hence the void return type. This will never be
351 called if the VM is trying to reclaim under GFP_NOFS conditions,
352 hence this method does not need to handle that situation itself.
353
354 Implementations must include conditional reschedule calls inside
355 any scanning loop that is done. This allows the VFS to
356 determine appropriate scan batch sizes without having to worry
357 about whether implementations will cause holdoff problems due to
358 large scan batch sizes.
359
360Whoever sets up the inode is responsible for filling in the "i_op"
361field. This is a pointer to a "struct inode_operations" which describes
362the methods that can be performed on individual inodes.
363
364
365struct xattr_handler
366---------------------
367
368On filesystems that support extended attributes (xattrs), the s_xattr
369superblock field points to a NULL-terminated array of xattr handlers.
370Extended attributes are name:value pairs.
371
372``name``
373 Indicates that the handler matches attributes with the specified
374 name (such as "system.posix_acl_access"); the prefix field must
375 be NULL.
376
377``prefix``
378 Indicates that the handler matches all attributes with the
379 specified name prefix (such as "user."); the name field must be
380 NULL.
381
382``list``
383 Determine if attributes matching this xattr handler should be
384 listed for a particular dentry. Used by some listxattr
385 implementations like generic_listxattr.
386
387``get``
388 Called by the VFS to get the value of a particular extended
389 attribute. This method is called by the getxattr(2) system
390 call.
391
392``set``
393 Called by the VFS to set the value of a particular extended
394 attribute. When the new value is NULL, called to remove a
395 particular extended attribute. This method is called by the
396 setxattr(2) and removexattr(2) system calls.
397
398When none of the xattr handlers of a filesystem match the specified
399attribute name or when a filesystem doesn't support extended attributes,
400the various ``*xattr(2)`` system calls return -EOPNOTSUPP.
401
402
403The Inode Object
404================
405
406An inode object represents an object within the filesystem.
407
408
409struct inode_operations
410-----------------------
411
412This describes how the VFS can manipulate an inode in your filesystem.
413As of kernel 2.6.22, the following members are defined:
414
415.. code-block:: c
416
417 struct inode_operations {
418 int (*create) (struct mnt_idmap *, struct inode *,struct dentry *, umode_t, bool);
419 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
420 int (*link) (struct dentry *,struct inode *,struct dentry *);
421 int (*unlink) (struct inode *,struct dentry *);
422 int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *);
423 struct dentry *(*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t);
424 int (*rmdir) (struct inode *,struct dentry *);
425 int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t);
426 int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *,
427 struct inode *, struct dentry *, unsigned int);
428 int (*readlink) (struct dentry *, char __user *,int);
429 const char *(*get_link) (struct dentry *, struct inode *,
430 struct delayed_call *);
431 int (*permission) (struct mnt_idmap *, struct inode *, int);
432 struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
433 int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *);
434 int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int);
435 ssize_t (*listxattr) (struct dentry *, char *, size_t);
436 void (*update_time)(struct inode *inode, enum fs_update_time type,
437 int flags);
438 void (*sync_lazytime)(struct inode *inode);
439 int (*atomic_open)(struct inode *, struct dentry *, struct file *,
440 unsigned open_flag, umode_t create_mode);
441 int (*tmpfile) (struct mnt_idmap *, struct inode *, struct file *, umode_t);
442 struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int);
443 int (*set_acl)(struct mnt_idmap *, struct dentry *, struct posix_acl *, int);
444 int (*fileattr_set)(struct mnt_idmap *idmap,
445 struct dentry *dentry, struct file_kattr *fa);
446 int (*fileattr_get)(struct dentry *dentry, struct file_kattr *fa);
447 struct offset_ctx *(*get_offset_ctx)(struct inode *inode);
448 };
449
450Again, all methods are called without any locks being held, unless
451otherwise noted.
452
453``create``
454 called by the open(2) and creat(2) system calls. Only required
455 if you want to support regular files. The dentry you get should
456 not have an inode (i.e. it should be a negative dentry). Here
457 you will probably call d_instantiate() with the dentry and the
458 newly created inode
459
460``lookup``
461 called when the VFS needs to look up an inode in a parent
462 directory. The name to look for is found in the dentry. This
463 method must call d_add() to insert the found inode into the
464 dentry. The "i_count" field in the inode structure should be
465 incremented. If the named inode does not exist a NULL inode
466 should be inserted into the dentry (this is called a negative
467 dentry). Returning an error code from this routine must only be
468 done on a real error, otherwise creating inodes with system
469 calls like create(2), mknod(2), mkdir(2) and so on will fail.
470 If you wish to overload the dentry methods then you should
471 initialise the "d_dop" field in the dentry; this is a pointer to
472 a struct "dentry_operations". This method is called with the
473 directory inode semaphore held
474
475``link``
476 called by the link(2) system call. Only required if you want to
477 support hard links. You will probably need to call
478 d_instantiate() just as you would in the create() method
479
480``unlink``
481 called by the unlink(2) system call. Only required if you want
482 to support deleting inodes
483
484``symlink``
485 called by the symlink(2) system call. Only required if you want
486 to support symlinks. You will probably need to call
487 d_instantiate() just as you would in the create() method
488
489``mkdir``
490 called by the mkdir(2) system call. Only required if you want
491 to support creating subdirectories. You will probably need to
492 call d_instantiate_new() just as you would in the create() method.
493
494 If d_instantiate_new() is not used and if the fh_to_dentry()
495 export operation is provided, or if the storage might be
496 accessible by another path (e.g. with a network filesystem)
497 then more care may be needed. Importantly d_instantate()
498 should not be used with an inode that is no longer I_NEW if there
499 any chance that the inode could already be attached to a dentry.
500 This is because of a hard rule in the VFS that a directory must
501 only ever have one dentry.
502
503 For example, if an NFS filesystem is mounted twice the new directory
504 could be visible on the other mount before it is on the original
505 mount, and a pair of name_to_handle_at(), open_by_handle_at()
506 calls could instantiate the directory inode with an IS_ROOT()
507 dentry before the first mkdir returns.
508
509 If there is any chance this could happen, then the new inode
510 should be d_drop()ed and attached with d_splice_alias(). The
511 returned dentry (if any) should be returned by ->mkdir().
512
513``rmdir``
514 called by the rmdir(2) system call. Only required if you want
515 to support deleting subdirectories
516
517``mknod``
518 called by the mknod(2) system call to create a device (char,
519 block) inode or a named pipe (FIFO) or socket. Only required if
520 you want to support creating these types of inodes. You will
521 probably need to call d_instantiate() just as you would in the
522 create() method
523
524``rename``
525 called by the rename(2) system call to rename the object to have
526 the parent and name given by the second inode and dentry.
527
528 The filesystem must return -EINVAL for any unsupported or
529 unknown flags. Currently the following flags are implemented:
530 (1) RENAME_NOREPLACE: this flag indicates that if the target of
531 the rename exists the rename should fail with -EEXIST instead of
532 replacing the target. The VFS already checks for existence, so
533 for local filesystems the RENAME_NOREPLACE implementation is
534 equivalent to plain rename.
535 (2) RENAME_EXCHANGE: exchange source and target. Both must
536 exist; this is checked by the VFS. Unlike plain rename, source
537 and target may be of different type.
538
539``get_link``
540 called by the VFS to follow a symbolic link to the inode it
541 points to. Only required if you want to support symbolic links.
542 This method returns the symlink body to traverse (and possibly
543 resets the current position with nd_jump_link()). If the body
544 won't go away until the inode is gone, nothing else is needed;
545 if it needs to be otherwise pinned, arrange for its release by
546 having get_link(..., ..., done) do set_delayed_call(done,
547 destructor, argument). In that case destructor(argument) will
548 be called once VFS is done with the body you've returned. May
549 be called in RCU mode; that is indicated by NULL dentry
550 argument. If request can't be handled without leaving RCU mode,
551 have it return ERR_PTR(-ECHILD).
552
553 If the filesystem stores the symlink target in ->i_link, the
554 VFS may use it directly without calling ->get_link(); however,
555 ->get_link() must still be provided. ->i_link must not be
556 freed until after an RCU grace period. Writing to ->i_link
557 post-iget() time requires a 'release' memory barrier.
558
559``readlink``
560 this is now just an override for use by readlink(2) for the
561 cases when ->get_link uses nd_jump_link() or object is not in
562 fact a symlink. Normally filesystems should only implement
563 ->get_link for symlinks and readlink(2) will automatically use
564 that.
565
566``permission``
567 called by the VFS to check for access rights on a POSIX-like
568 filesystem.
569
570 May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in
571 rcu-walk mode, the filesystem must check the permission without
572 blocking or storing to the inode.
573
574 If a situation is encountered that rcu-walk cannot handle,
575 return
576 -ECHILD and it will be called again in ref-walk mode.
577
578``setattr``
579 called by the VFS to set attributes for a file. This method is
580 called by chmod(2) and related system calls.
581
582``getattr``
583 called by the VFS to get attributes of a file. This method is
584 called by stat(2) and related system calls.
585
586``listxattr``
587 called by the VFS to list all extended attributes for a given
588 file. This method is called by the listxattr(2) system call.
589
590``update_time``
591 called by the VFS to update a specific time or the i_version of
592 an inode. If this is not defined the VFS will update the inode
593 itself and call mark_inode_dirty_sync.
594
595``sync_lazytime``:
596 called by the writeback code to update the lazy time stamps to
597 regular time stamp updates that get syncing into the on-disk
598 inode.
599
600``atomic_open``
601 called on the last component of an open. Using this optional
602 method the filesystem can look up, possibly create and open the
603 file in one atomic operation. If it wants to leave actual
604 opening to the caller (e.g. if the file turned out to be a
605 symlink, device, or just something filesystem won't do atomic
606 open for), it may signal this by returning finish_no_open(file,
607 dentry). This method is only called if the last component is
608 negative or needs lookup. Cached positive dentries are still
609 handled by f_op->open(). If the file was created, FMODE_CREATED
610 flag should be set in file->f_mode. In case of O_EXCL the
611 method must only succeed if the file didn't exist and hence
612 FMODE_CREATED shall always be set on success.
613
614``tmpfile``
615 called in the end of O_TMPFILE open(). Optional, equivalent to
616 atomically creating, opening and unlinking a file in given
617 directory. On success needs to return with the file already
618 open; this can be done by calling finish_open_simple() right at
619 the end.
620
621``fileattr_get``
622 called on ioctl(FS_IOC_GETFLAGS) and ioctl(FS_IOC_FSGETXATTR) to
623 retrieve miscellaneous file flags and attributes. Also called
624 before the relevant SET operation to check what is being changed
625 (in this case with i_rwsem locked exclusive). If unset, then
626 fall back to f_op->ioctl().
627
628``fileattr_set``
629 called on ioctl(FS_IOC_SETFLAGS) and ioctl(FS_IOC_FSSETXATTR) to
630 change miscellaneous file flags and attributes. Callers hold
631 i_rwsem exclusive. If unset, then fall back to f_op->ioctl().
632``get_offset_ctx``
633 called to get the offset context for a directory inode. A
634 filesystem must define this operation to use
635 simple_offset_dir_operations.
636
637The Address Space Object
638========================
639
640The address space object is used to group and manage pages in the page
641cache. It can be used to keep track of the pages in a file (or anything
642else) and also track the mapping of sections of the file into process
643address spaces.
644
645There are a number of distinct yet related services that an
646address-space can provide. These include communicating memory pressure,
647page lookup by address, and keeping track of pages tagged as Dirty or
648Writeback.
649
650The first can be used independently to the others. The VM can try to
651release clean pages in order to reuse them. To do this it can call
652->release_folio on clean folios with the private
653flag set. Clean pages without PagePrivate and with no external references
654will be released without notice being given to the address_space.
655
656To achieve this functionality, pages need to be placed on an LRU with
657lru_cache_add and mark_page_active needs to be called whenever the page
658is used.
659
660Pages are normally kept in a radix tree index by ->index. This tree
661maintains information about the PG_Dirty and PG_Writeback status of each
662page, so that pages with either of these flags can be found quickly.
663
664The Dirty tag is primarily used by mpage_writepages - the default
665->writepages method. It uses the tag to find dirty pages to
666write back. If mpage_writepages is not used (i.e. the address
667provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is almost
668unused. write_inode_now and sync_inode do use it (through
669__sync_single_inode) to check if ->writepages has been successful in
670writing out the whole address_space.
671
672The Writeback tag is used by filemap*wait* and sync_page* functions, via
673filemap_fdatawait_range, to wait for all writeback to complete.
674
675An address_space handler may attach extra information to a page,
676typically using the 'private' field in the 'struct page'. If such
677information is attached, the PG_Private flag should be set. This will
678cause various VM routines to make extra calls into the address_space
679handler to deal with that data.
680
681An address space acts as an intermediate between storage and
682application. Data is read into the address space a whole page at a
683time, and provided to the application either by copying of the page, or
684by memory-mapping the page. Data is written into the address space by
685the application, and then written-back to storage typically in whole
686pages, however the address_space has finer control of write sizes.
687
688The read process essentially only requires 'read_folio'. The write
689process is more complicated and uses write_begin/write_end or
690dirty_folio to write data into the address_space, and
691writepages to writeback data to storage.
692
693Removing pages from an address_space requires holding the inode's i_rwsem
694exclusively, while adding pages to the address_space requires holding the
695inode's i_mapping->invalidate_lock exclusively.
696
697When data is written to a page, the PG_Dirty flag should be set. It
698typically remains set until writepages asks for it to be written. This
699should clear PG_Dirty and set PG_Writeback. It can be actually written
700at any point after PG_Dirty is clear. Once it is known to be safe,
701PG_Writeback is cleared.
702
703Writeback makes use of a writeback_control structure to direct the
704operations. This gives the writepages operation some
705information about the nature of and reason for the writeback request,
706and the constraints under which it is being done. It is also used to
707return information back to the caller about the result of a
708writepages request.
709
710
711Handling errors during writeback
712--------------------------------
713
714Most applications that do buffered I/O will periodically call a file
715synchronization call (fsync, fdatasync, msync or sync_file_range) to
716ensure that data written has made it to the backing store. When there
717is an error during writeback, they expect that error to be reported when
718a file sync request is made. After an error has been reported on one
719request, subsequent requests on the same file descriptor should return
7200, unless further writeback errors have occurred since the previous file
721synchronization.
722
723Ideally, the kernel would report errors only on file descriptions on
724which writes were done that subsequently failed to be written back. The
725generic pagecache infrastructure does not track the file descriptions
726that have dirtied each individual page however, so determining which
727file descriptors should get back an error is not possible.
728
729Instead, the generic writeback error tracking infrastructure in the
730kernel settles for reporting errors to fsync on all file descriptions
731that were open at the time that the error occurred. In a situation with
732multiple writers, all of them will get back an error on a subsequent
733fsync, even if all of the writes done through that particular file
734descriptor succeeded (or even if there were no writes on that file
735descriptor at all).
736
737Filesystems that wish to use this infrastructure should call
738mapping_set_error to record the error in the address_space when it
739occurs. Then, after writing back data from the pagecache in their
740file->fsync operation, they should call file_check_and_advance_wb_err to
741ensure that the struct file's error cursor has advanced to the correct
742point in the stream of errors emitted by the backing device(s).
743
744
745struct address_space_operations
746-------------------------------
747
748This describes how the VFS can manipulate mapping of a file to page
749cache in your filesystem. The following members are defined:
750
751.. code-block:: c
752
753 struct address_space_operations {
754 int (*read_folio)(struct file *, struct folio *);
755 int (*writepages)(struct address_space *, struct writeback_control *);
756 bool (*dirty_folio)(struct address_space *, struct folio *);
757 void (*readahead)(struct readahead_control *);
758 int (*write_begin)(const struct kiocb *, struct address_space *mapping,
759 loff_t pos, unsigned len,
760 struct page **pagep, void **fsdata);
761 int (*write_end)(const struct kiocb *, struct address_space *mapping,
762 loff_t pos, unsigned len, unsigned copied,
763 struct folio *folio, void *fsdata);
764 sector_t (*bmap)(struct address_space *, sector_t);
765 void (*invalidate_folio) (struct folio *, size_t start, size_t len);
766 bool (*release_folio)(struct folio *, gfp_t);
767 void (*free_folio)(struct folio *);
768 ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter);
769 int (*migrate_folio)(struct mapping *, struct folio *dst,
770 struct folio *src, enum migrate_mode);
771 int (*launder_folio) (struct folio *);
772
773 bool (*is_partially_uptodate) (struct folio *, size_t from,
774 size_t count);
775 void (*is_dirty_writeback)(struct folio *, bool *, bool *);
776 int (*error_remove_folio)(struct mapping *mapping, struct folio *);
777 int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
778 int (*swap_deactivate)(struct file *);
779 int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
780 };
781
782``read_folio``
783 Called by the page cache to read a folio from the backing store.
784 The 'file' argument supplies authentication information to network
785 filesystems, and is generally not used by block based filesystems.
786 It may be NULL if the caller does not have an open file (eg if
787 the kernel is performing a read for itself rather than on behalf
788 of a userspace process with an open file).
789
790 If the mapping does not support large folios, the folio will
791 contain a single page. The folio will be locked when read_folio
792 is called. If the read completes successfully, the folio should
793 be marked uptodate. The filesystem should unlock the folio
794 once the read has completed, whether it was successful or not.
795 The filesystem does not need to modify the refcount on the folio;
796 the page cache holds a reference count and that will not be
797 released until the folio is unlocked.
798
799 Filesystems may implement ->read_folio() synchronously.
800 In normal operation, folios are read through the ->readahead()
801 method. Only if this fails, or if the caller needs to wait for
802 the read to complete will the page cache call ->read_folio().
803 Filesystems should not attempt to perform their own readahead
804 in the ->read_folio() operation.
805
806 If the filesystem cannot perform the read at this time, it can
807 unlock the folio, do whatever action it needs to ensure that the
808 read will succeed in the future and return AOP_TRUNCATED_PAGE.
809 In this case, the caller should look up the folio, lock it,
810 and call ->read_folio again.
811
812 Callers may invoke the ->read_folio() method directly, but using
813 read_mapping_folio() will take care of locking, waiting for the
814 read to complete and handle cases such as AOP_TRUNCATED_PAGE.
815
816``writepages``
817 called by the VM to write out pages associated with the
818 address_space object. If wbc->sync_mode is WB_SYNC_ALL, then
819 the writeback_control will specify a range of pages that must be
820 written out. If it is WB_SYNC_NONE, then a nr_to_write is
821 given and that many pages should be written if possible. If no
822 ->writepages is given, then mpage_writepages is used instead.
823 This will choose pages from the address space that are tagged as
824 DIRTY and will write them back.
825
826``dirty_folio``
827 called by the VM to mark a folio as dirty. This is particularly
828 needed if an address space attaches private data to a folio, and
829 that data needs to be updated when a folio is dirtied. This is
830 called, for example, when a memory mapped page gets modified.
831 If defined, it should set the folio dirty flag, and the
832 PAGECACHE_TAG_DIRTY search mark in i_pages.
833
834``readahead``
835 Called by the VM to read pages associated with the address_space
836 object. The pages are consecutive in the page cache and are
837 locked. The implementation should decrement the page refcount
838 after starting I/O on each page. Usually the page will be
839 unlocked by the I/O completion handler. The set of pages are
840 divided into some sync pages followed by some async pages,
841 rac->ra->async_size gives the number of async pages. The
842 filesystem should attempt to read all sync pages but may decide
843 to stop once it reaches the async pages. If it does decide to
844 stop attempting I/O, it can simply return. The caller will
845 remove the remaining pages from the address space, unlock them
846 and decrement the page refcount. Set PageUptodate if the I/O
847 completes successfully.
848
849``write_begin``
850 Called by the generic buffered write code to ask the filesystem
851 to prepare to write len bytes at the given offset in the file.
852 The address_space should check that the write will be able to
853 complete, by allocating space if necessary and doing any other
854 internal housekeeping. If the write will update parts of any
855 basic-blocks on storage, then those blocks should be pre-read
856 (if they haven't been read already) so that the updated blocks
857 can be written out properly.
858
859 The filesystem must return the locked pagecache folio for the
860 specified offset, in ``*foliop``, for the caller to write into.
861
862 It must be able to cope with short writes (where the length
863 passed to write_begin is greater than the number of bytes copied
864 into the folio).
865
866 A void * may be returned in fsdata, which then gets passed into
867 write_end.
868
869 Returns 0 on success; < 0 on failure (which is the error code),
870 in which case write_end is not called.
871
872``write_end``
873 After a successful write_begin, and data copy, write_end must be
874 called. len is the original len passed to write_begin, and
875 copied is the amount that was able to be copied.
876
877 The filesystem must take care of unlocking the folio,
878 decrementing its refcount, and updating i_size.
879
880 Returns < 0 on failure, otherwise the number of bytes (<=
881 'copied') that were able to be copied into pagecache.
882
883``bmap``
884 called by the VFS to map a logical block offset within object to
885 physical block number. This method is used by the FIBMAP ioctl
886 and for working with swap-files. To be able to swap to a file,
887 the file must have a stable mapping to a block device. The swap
888 system does not go through the filesystem but instead uses bmap
889 to find out where the blocks in the file are and uses those
890 addresses directly.
891
892``invalidate_folio``
893 If a folio has private data, then invalidate_folio will be
894 called when part or all of the folio is to be removed from the
895 address space. This generally corresponds to either a
896 truncation, punch hole or a complete invalidation of the address
897 space (in the latter case 'offset' will always be 0 and 'length'
898 will be folio_size()). Any private data associated with the folio
899 should be updated to reflect this truncation. If offset is 0
900 and length is folio_size(), then the private data should be
901 released, because the folio must be able to be completely
902 discarded. This may be done by calling the ->release_folio
903 function, but in this case the release MUST succeed.
904
905``release_folio``
906 release_folio is called on folios with private data to tell the
907 filesystem that the folio is about to be freed. ->release_folio
908 should remove any private data from the folio and clear the
909 private flag. If release_folio() fails, it should return false.
910 release_folio() is used in two distinct though related cases.
911 The first is when the VM wants to free a clean folio with no
912 active users. If ->release_folio succeeds, the folio will be
913 removed from the address_space and be freed.
914
915 The second case is when a request has been made to invalidate
916 some or all folios in an address_space. This can happen
917 through the fadvise(POSIX_FADV_DONTNEED) system call or by the
918 filesystem explicitly requesting it as nfs and 9p do (when they
919 believe the cache may be out of date with storage) by calling
920 invalidate_inode_pages2(). If the filesystem makes such a call,
921 and needs to be certain that all folios are invalidated, then
922 its release_folio will need to ensure this. Possibly it can
923 clear the uptodate flag if it cannot free private data yet.
924
925``free_folio``
926 free_folio is called once the folio is no longer visible in the
927 page cache in order to allow the cleanup of any private data.
928 Since it may be called by the memory reclaimer, it should not
929 assume that the original address_space mapping still exists, and
930 it should not block.
931
932``direct_IO``
933 called by the generic read/write routines to perform direct_IO -
934 that is IO requests which bypass the page cache and transfer
935 data directly between the storage and the application's address
936 space.
937
938``migrate_folio``
939 This is used to compact the physical memory usage. If the VM
940 wants to relocate a folio (maybe from a memory device that is
941 signalling imminent failure) it will pass a new folio and an old
942 folio to this function. migrate_folio should transfer any private
943 data across and update any references that it has to the folio.
944
945``launder_folio``
946 Called before freeing a folio - it writes back the dirty folio.
947 To prevent redirtying the folio, it is kept locked during the
948 whole operation.
949
950``is_partially_uptodate``
951 Called by the VM when reading a file through the pagecache when
952 the underlying blocksize is smaller than the size of the folio.
953 If the required block is up to date then the read can complete
954 without needing I/O to bring the whole page up to date.
955
956``is_dirty_writeback``
957 Called by the VM when attempting to reclaim a folio. The VM uses
958 dirty and writeback information to determine if it needs to
959 stall to allow flushers a chance to complete some IO.
960 Ordinarily it can use folio_test_dirty and folio_test_writeback but
961 some filesystems have more complex state (unstable folios in NFS
962 prevent reclaim) or do not set those flags due to locking
963 problems. This callback allows a filesystem to indicate to the
964 VM if a folio should be treated as dirty or writeback for the
965 purposes of stalling.
966
967``error_remove_folio``
968 normally set to generic_error_remove_folio if truncation is ok
969 for this address space. Used for memory failure handling.
970 Setting this implies you deal with pages going away under you,
971 unless you have them locked or reference counts increased.
972
973``swap_activate``
974
975 Called to prepare the given file for swap. It should perform
976 any validation and preparation necessary to ensure that writes
977 can be performed with minimal memory allocation. It should call
978 add_swap_extent(), or the helper iomap_swapfile_activate(), and
979 return the number of extents added. If IO should be submitted
980 through ->swap_rw(), it should set SWP_FS_OPS, otherwise IO will
981 be submitted directly to the block device ``sis->bdev``.
982
983``swap_deactivate``
984 Called during swapoff on files where swap_activate was
985 successful.
986
987``swap_rw``
988 Called to read or write swap pages when SWP_FS_OPS is set.
989
990The File Object
991===============
992
993A file object represents a file opened by a process. This is also known
994as an "open file description" in POSIX parlance.
995
996
997struct file_operations
998----------------------
999
1000This describes how the VFS can manipulate an open file. As of kernel
10014.18, the following members are defined:
1002
1003.. code-block:: c
1004
1005 struct file_operations {
1006 struct module *owner;
1007 fop_flags_t fop_flags;
1008 loff_t (*llseek) (struct file *, loff_t, int);
1009 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
1010 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
1011 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
1012 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
1013 int (*iopoll)(struct kiocb *kiocb, struct io_comp_batch *,
1014 unsigned int flags);
1015 int (*iterate_shared) (struct file *, struct dir_context *);
1016 __poll_t (*poll) (struct file *, struct poll_table_struct *);
1017 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
1018 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
1019 int (*mmap) (struct file *, struct vm_area_struct *);
1020 int (*open) (struct inode *, struct file *);
1021 int (*flush) (struct file *, fl_owner_t id);
1022 int (*release) (struct inode *, struct file *);
1023 int (*fsync) (struct file *, loff_t, loff_t, int datasync);
1024 int (*fasync) (int, struct file *, int);
1025 int (*lock) (struct file *, int, struct file_lock *);
1026 unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1027 int (*check_flags)(int);
1028 int (*flock) (struct file *, int, struct file_lock *);
1029 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
1030 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
1031 void (*splice_eof)(struct file *file);
1032 int (*setlease)(struct file *, int, struct file_lease **, void **);
1033 long (*fallocate)(struct file *file, int mode, loff_t offset,
1034 loff_t len);
1035 void (*show_fdinfo)(struct seq_file *m, struct file *f);
1036 #ifndef CONFIG_MMU
1037 unsigned (*mmap_capabilities)(struct file *);
1038 #endif
1039 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *,
1040 loff_t, size_t, unsigned int);
1041 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
1042 struct file *file_out, loff_t pos_out,
1043 loff_t len, unsigned int remap_flags);
1044 int (*fadvise)(struct file *, loff_t, loff_t, int);
1045 int (*uring_cmd)(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
1046 int (*uring_cmd_iopoll)(struct io_uring_cmd *, struct io_comp_batch *,
1047 unsigned int poll_flags);
1048 int (*mmap_prepare)(struct vm_area_desc *);
1049 };
1050
1051Again, all methods are called without any locks being held, unless
1052otherwise noted.
1053
1054``llseek``
1055 called when the VFS needs to move the file position index
1056
1057``read``
1058 called by read(2) and related system calls
1059
1060``read_iter``
1061 possibly asynchronous read with iov_iter as destination
1062
1063``write``
1064 called by write(2) and related system calls
1065
1066``write_iter``
1067 possibly asynchronous write with iov_iter as source
1068
1069``iopoll``
1070 called when aio wants to poll for completions on HIPRI iocbs
1071
1072``iterate_shared``
1073 called when the VFS needs to read the directory contents
1074
1075``poll``
1076 called by the VFS when a process wants to check if there is
1077 activity on this file and (optionally) go to sleep until there
1078 is activity. Called by the select(2) and poll(2) system calls
1079
1080``unlocked_ioctl``
1081 called by the ioctl(2) system call.
1082
1083``compat_ioctl``
1084 called by the ioctl(2) system call when 32 bit system calls are
1085 used on 64 bit kernels.
1086
1087``mmap``
1088 called by the mmap(2) system call. Deprecated in favour of
1089 ``mmap_prepare``.
1090
1091``open``
1092 called by the VFS when an inode should be opened. When the VFS
1093 opens a file, it creates a new "struct file". It then calls the
1094 open method for the newly allocated file structure. You might
1095 think that the open method really belongs in "struct
1096 inode_operations", and you may be right. I think it's done the
1097 way it is because it makes filesystems simpler to implement.
1098 The open() method is a good place to initialize the
1099 "private_data" member in the file structure if you want to point
1100 to a device structure
1101
1102``flush``
1103 called by the close(2) system call to flush a file
1104
1105``release``
1106 called when the last reference to an open file is closed
1107
1108``fsync``
1109 called by the fsync(2) system call. Also see the section above
1110 entitled "Handling errors during writeback".
1111
1112``fasync``
1113 called by the fcntl(2) system call when asynchronous
1114 (non-blocking) mode is enabled for a file
1115
1116``lock``
1117 called by the fcntl(2) system call for F_GETLK, F_SETLK, and
1118 F_SETLKW commands
1119
1120``get_unmapped_area``
1121 called by the mmap(2) system call
1122
1123``check_flags``
1124 called by the fcntl(2) system call for F_SETFL command
1125
1126``flock``
1127 called by the flock(2) system call
1128
1129``splice_write``
1130 called by the VFS to splice data from a pipe to a file. This
1131 method is used by the splice(2) system call
1132
1133``splice_read``
1134 called by the VFS to splice data from file to a pipe. This
1135 method is used by the splice(2) system call
1136
1137``setlease``
1138 called by the VFS to set or release a file lock lease. Local
1139 filesystems that wish to use the kernel-internal lease implementation
1140 should set this to generic_setlease(). Other setlease implementations
1141 should call generic_setlease() to record or remove the lease in the inode
1142 after setting it. When set to NULL, attempts to set or remove a lease will
1143 return -EINVAL.
1144
1145``fallocate``
1146 called by the VFS to preallocate blocks or punch a hole.
1147
1148``copy_file_range``
1149 called by the copy_file_range(2) system call.
1150
1151``remap_file_range``
1152 called by the ioctl(2) system call for FICLONERANGE and FICLONE
1153 and FIDEDUPERANGE commands to remap file ranges. An
1154 implementation should remap len bytes at pos_in of the source
1155 file into the dest file at pos_out. Implementations must handle
1156 callers passing in len == 0; this means "remap to the end of the
1157 source file". The return value should the number of bytes
1158 remapped, or the usual negative error code if errors occurred
1159 before any bytes were remapped. The remap_flags parameter
1160 accepts REMAP_FILE_* flags. If REMAP_FILE_DEDUP is set then the
1161 implementation must only remap if the requested file ranges have
1162 identical contents. If REMAP_FILE_CAN_SHORTEN is set, the caller is
1163 ok with the implementation shortening the request length to
1164 satisfy alignment or EOF requirements (or any other reason).
1165
1166``fadvise``
1167 possibly called by the fadvise64() system call.
1168
1169``mmap_prepare``
1170 Called by the mmap(2) system call. Allows a VFS to set up a
1171 file-backed memory mapping, most notably establishing relevant
1172 private state and VMA callbacks.
1173
1174 If further action such as pre-population of page tables is required,
1175 this can be specified by the vm_area_desc->action field and related
1176 parameters.
1177
1178Note that the file operations are implemented by the specific
1179filesystem in which the inode resides. When opening a device node
1180(character or block special) most filesystems will call special
1181support routines in the VFS which will locate the required device
1182driver information. These support routines replace the filesystem file
1183operations with those for the device driver, and then proceed to call
1184the new open() method for the file. This is how opening a device file
1185in the filesystem eventually ends up calling the device driver open()
1186method.
1187
1188
1189Directory Entry Cache (dcache)
1190==============================
1191
1192
1193struct dentry_operations
1194------------------------
1195
1196This describes how a filesystem can overload the standard dentry
1197operations. Dentries and the dcache are the domain of the VFS and the
1198individual filesystem implementations. Device drivers have no business
1199here. These methods may be set to NULL, as they are either optional or
1200the VFS uses a default. As of kernel 2.6.22, the following members are
1201defined:
1202
1203.. code-block:: c
1204
1205 struct dentry_operations {
1206 int (*d_revalidate)(struct inode *, const struct qstr *,
1207 struct dentry *, unsigned int);
1208 int (*d_weak_revalidate)(struct dentry *, unsigned int);
1209 int (*d_hash)(const struct dentry *, struct qstr *);
1210 int (*d_compare)(const struct dentry *,
1211 unsigned int, const char *, const struct qstr *);
1212 int (*d_delete)(const struct dentry *);
1213 int (*d_init)(struct dentry *);
1214 void (*d_release)(struct dentry *);
1215 void (*d_iput)(struct dentry *, struct inode *);
1216 char *(*d_dname)(struct dentry *, char *, int);
1217 struct vfsmount *(*d_automount)(struct path *);
1218 int (*d_manage)(const struct path *, bool);
1219 struct dentry *(*d_real)(struct dentry *, enum d_real_type type);
1220 bool (*d_unalias_trylock)(const struct dentry *);
1221 void (*d_unalias_unlock)(const struct dentry *);
1222 };
1223
1224``d_revalidate``
1225 called when the VFS needs to revalidate a dentry. This is
1226 called whenever a name look-up finds a dentry in the dcache.
1227 Most local filesystems leave this as NULL, because all their
1228 dentries in the dcache are valid. Network filesystems are
1229 different since things can change on the server without the
1230 client necessarily being aware of it.
1231
1232 This function should return a positive value if the dentry is
1233 still valid, and zero or a negative error code if it isn't.
1234
1235 d_revalidate may be called in rcu-walk mode (flags &
1236 LOOKUP_RCU). If in rcu-walk mode, the filesystem must
1237 revalidate the dentry without blocking or storing to the dentry,
1238 d_parent and d_inode should not be used without care (because
1239 they can change and, in d_inode case, even become NULL under
1240 us).
1241
1242 If a situation is encountered that rcu-walk cannot handle,
1243 return
1244 -ECHILD and it will be called again in ref-walk mode.
1245
1246``d_weak_revalidate``
1247 called when the VFS needs to revalidate a "jumped" dentry. This
1248 is called when a path-walk ends at dentry that was not acquired
1249 by doing a lookup in the parent directory. This includes "/",
1250 "." and "..", as well as procfs-style symlinks and mountpoint
1251 traversal.
1252
1253 In this case, we are less concerned with whether the dentry is
1254 still fully correct, but rather that the inode is still valid.
1255 As with d_revalidate, most local filesystems will set this to
1256 NULL since their dcache entries are always valid.
1257
1258 This function has the same return code semantics as
1259 d_revalidate.
1260
1261 d_weak_revalidate is only called after leaving rcu-walk mode.
1262
1263``d_hash``
1264 called when the VFS adds a dentry to the hash table. The first
1265 dentry passed to d_hash is the parent directory that the name is
1266 to be hashed into.
1267
1268 Same locking and synchronisation rules as d_compare regarding
1269 what is safe to dereference etc.
1270
1271``d_compare``
1272 called to compare a dentry name with a given name. The first
1273 dentry is the parent of the dentry to be compared, the second is
1274 the child dentry. len and name string are properties of the
1275 dentry to be compared. qstr is the name to compare it with.
1276
1277 Must be constant and idempotent, and should not take locks if
1278 possible, and should not or store into the dentry. Should not
1279 dereference pointers outside the dentry without lots of care
1280 (eg. d_parent, d_inode, d_name should not be used).
1281
1282 However, our vfsmount is pinned, and RCU held, so the dentries
1283 and inodes won't disappear, neither will our sb or filesystem
1284 module. ->d_sb may be used.
1285
1286 It is a tricky calling convention because it needs to be called
1287 under "rcu-walk", ie. without any locks or references on things.
1288
1289``d_delete``
1290 called when the last reference to a dentry is dropped and the
1291 dcache is deciding whether or not to cache it. Return 1 to
1292 delete immediately, or 0 to cache the dentry. Default is NULL
1293 which means to always cache a reachable dentry. d_delete must
1294 be constant and idempotent.
1295
1296``d_init``
1297 called when a dentry is allocated
1298
1299``d_release``
1300 called when a dentry is really deallocated
1301
1302``d_iput``
1303 called when a dentry loses its inode (just prior to its being
1304 deallocated). The default when this is NULL is that the VFS
1305 calls iput(). If you define this method, you must call iput()
1306 yourself
1307
1308``d_dname``
1309 called when the pathname of a dentry should be generated.
1310 Useful for some pseudo filesystems (sockfs, pipefs, ...) to
1311 delay pathname generation. (Instead of doing it when dentry is
1312 created, it's done only when the path is needed.). Real
1313 filesystems probably dont want to use it, because their dentries
1314 are present in global dcache hash, so their hash should be an
1315 invariant. As no lock is held, d_dname() should not try to
1316 modify the dentry itself, unless appropriate SMP safety is used.
1317 CAUTION : d_path() logic is quite tricky. The correct way to
1318 return for example "Hello" is to put it at the end of the
1319 buffer, and returns a pointer to the first char.
1320 dynamic_dname() helper function is provided to take care of
1321 this.
1322
1323 Example :
1324
1325.. code-block:: c
1326
1327 static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1328 {
1329 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1330 dentry->d_inode->i_ino);
1331 }
1332
1333``d_automount``
1334 called when an automount dentry is to be traversed (optional).
1335 This should create a new VFS mount record and return the record
1336 to the caller. The caller is supplied with a path parameter
1337 giving the automount directory to describe the automount target
1338 and the parent VFS mount record to provide inheritable mount
1339 parameters. NULL should be returned if someone else managed to
1340 make the automount first. If the vfsmount creation failed, then
1341 an error code should be returned. If -EISDIR is returned, then
1342 the directory will be treated as an ordinary directory and
1343 returned to pathwalk to continue walking.
1344
1345 If a vfsmount is returned, the caller will attempt to mount it
1346 on the mountpoint and will remove the vfsmount from its
1347 expiration list in the case of failure.
1348
1349 This function is only used if DCACHE_NEED_AUTOMOUNT is set on
1350 the dentry. This is set by __d_instantiate() if S_AUTOMOUNT is
1351 set on the inode being added.
1352
1353``d_manage``
1354 called to allow the filesystem to manage the transition from a
1355 dentry (optional). This allows autofs, for example, to hold up
1356 clients waiting to explore behind a 'mountpoint' while letting
1357 the daemon go past and construct the subtree there. 0 should be
1358 returned to let the calling process continue. -EISDIR can be
1359 returned to tell pathwalk to use this directory as an ordinary
1360 directory and to ignore anything mounted on it and not to check
1361 the automount flag. Any other error code will abort pathwalk
1362 completely.
1363
1364 If the 'rcu_walk' parameter is true, then the caller is doing a
1365 pathwalk in RCU-walk mode. Sleeping is not permitted in this
1366 mode, and the caller can be asked to leave it and call again by
1367 returning -ECHILD. -EISDIR may also be returned to tell
1368 pathwalk to ignore d_automount or any mounts.
1369
1370 This function is only used if DCACHE_MANAGE_TRANSIT is set on
1371 the dentry being transited from.
1372
1373``d_real``
1374 overlay/union type filesystems implement this method to return one
1375 of the underlying dentries of a regular file hidden by the overlay.
1376
1377 The 'type' argument takes the values D_REAL_DATA or D_REAL_METADATA
1378 for returning the real underlying dentry that refers to the inode
1379 hosting the file's data or metadata respectively.
1380
1381 For non-regular files, the 'dentry' argument is returned.
1382
1383``d_unalias_trylock``
1384 if present, will be called by d_splice_alias() before moving a
1385 preexisting attached alias. Returning false prevents __d_move(),
1386 making d_splice_alias() fail with -ESTALE.
1387
1388 Rationale: setting FS_RENAME_DOES_D_MOVE will prevent d_move()
1389 and d_exchange() calls from the outside of filesystem methods;
1390 however, it does not guarantee that attached dentries won't
1391 be renamed or moved by d_splice_alias() finding a preexisting
1392 alias for a directory inode. Normally we would not care;
1393 however, something that wants to stabilize the entire path to
1394 root over a blocking operation might need that. See 9p for one
1395 (and hopefully only) example.
1396
1397``d_unalias_unlock``
1398 should be paired with ``d_unalias_trylock``; that one is called after
1399 __d_move() call in __d_unalias().
1400
1401
1402Each dentry has a pointer to its parent dentry, as well as a hash list
1403of child dentries. Child dentries are basically like files in a
1404directory.
1405
1406
1407Directory Entry Cache API
1408--------------------------
1409
1410There are a number of functions defined which permit a filesystem to
1411manipulate dentries:
1412
1413``dget``
1414 open a new handle for an existing dentry (this just increments
1415 the usage count)
1416
1417``dput``
1418 close a handle for a dentry (decrements the usage count). If
1419 the usage count drops to 0, and the dentry is still in its
1420 parent's hash, the "d_delete" method is called to check whether
1421 it should be cached. If it should not be cached, or if the
1422 dentry is not hashed, it is deleted. Otherwise cached dentries
1423 are put into an LRU list to be reclaimed on memory shortage.
1424
1425``d_drop``
1426 this unhashes a dentry from its parents hash list. A subsequent
1427 call to dput() will deallocate the dentry if its usage count
1428 drops to 0
1429
1430``d_delete``
1431 delete a dentry. If there are no other open references to the
1432 dentry then the dentry is turned into a negative dentry (the
1433 d_iput() method is called). If there are other references, then
1434 d_drop() is called instead
1435
1436``d_add``
1437 add a dentry to its parents hash list and then calls
1438 d_instantiate()
1439
1440``d_instantiate``
1441 add a dentry to the alias hash list for the inode and updates
1442 the "d_inode" member. The "i_count" member in the inode
1443 structure should be set/incremented. If the inode pointer is
1444 NULL, the dentry is called a "negative dentry". This function
1445 is commonly called when an inode is created for an existing
1446 negative dentry
1447
1448``d_lookup``
1449 look up a dentry given its parent and path name component It
1450 looks up the child of that given name from the dcache hash
1451 table. If it is found, the reference count is incremented and
1452 the dentry is returned. The caller must use dput() to free the
1453 dentry when it finishes using it.
1454
1455
1456Mount Options
1457=============
1458
1459
1460Parsing options
1461---------------
1462
1463On mount and remount the filesystem is passed a string containing a
1464comma separated list of mount options. The options can have either of
1465these forms:
1466
1467 option
1468 option=value
1469
1470The <linux/parser.h> header defines an API that helps parse these
1471options. There are plenty of examples on how to use it in existing
1472filesystems.
1473
1474
1475Showing options
1476---------------
1477
1478If a filesystem accepts mount options, it must define show_options() to
1479show all the currently active options. The rules are:
1480
1481 - options MUST be shown which are not default or their values differ
1482 from the default
1483
1484 - options MAY be shown which are enabled by default or have their
1485 default value
1486
1487Options used only internally between a mount helper and the kernel (such
1488as file descriptors), or which only have an effect during the mounting
1489(such as ones controlling the creation of a journal) are exempt from the
1490above rules.
1491
1492The underlying reason for the above rules is to make sure, that a mount
1493can be accurately replicated (e.g. umounting and mounting again) based
1494on the information found in /proc/mounts.
1495
1496
1497Resources
1498=========
1499
1500(Note some of these resources are not up-to-date with the latest kernel
1501 version.)
1502
1503Creating Linux virtual filesystems. 2002
1504 <https://lwn.net/Articles/13325/>
1505
1506The Linux Virtual File-system Layer by Neil Brown. 1999
1507 <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1508
1509A tour of the Linux VFS by Michael K. Johnson. 1996
1510 <https://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1511
1512A small trail through the Linux kernel by Andries Brouwer. 2001
1513 <https://www.win.tue.nl/~aeb/linux/vfs/trail.html>