1 .. SPDX-License-Identifier: GPL-2.0
3 =========================================
4 Overview of the Linux Virtual File System
5 =========================================
7 Original author: Richard Gooch <rgooch@atnf.csiro.au>
9 - Copyright (C) 1999 Richard Gooch
10 - Copyright (C) 2005 Pekka Enberg
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.
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.
26 Directory Entry Cache (dcache)
27 ------------------------------
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.
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.
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).
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.
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.
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.
82 Registering and Mounting a Filesystem
83 =====================================
85 To register and unregister a filesystem, use the following API
92 extern int register_filesystem(struct file_system_type *);
93 extern int unregister_filesystem(struct file_system_type *);
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
103 You can see all filesystems that are registered to the kernel in the
104 file /proc/filesystems.
107 struct file_system_type
108 -----------------------
110 This describes the filesystem. As of kernel 2.6.39, the following
115 struct file_system_type {
118 struct dentry *(*mount) (struct file_system_type *, int,
119 const char *, void *);
120 void (*kill_sb) (struct super_block *);
121 struct module *owner;
122 struct file_system_type * next;
123 struct list_head fs_supers;
124 struct lock_class_key s_lock_key;
125 struct lock_class_key s_umount_key;
129 the name of the filesystem type, such as "ext2", "iso9660",
133 various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
136 the method to call when a new instance of this filesystem should
140 the method to call when an instance of this filesystem should be
145 for internal VFS use: you should initialize this to THIS_MODULE
149 for internal VFS use: you should initialize this to NULL
151 s_lock_key, s_umount_key: lockdep-specific
153 The mount() method has the following arguments:
155 ``struct file_system_type *fs_type``
156 describes the filesystem, partly initialized by the specific
162 ``const char *dev_name``
163 the device name we are mounting.
166 arbitrary mount options, usually comes as an ASCII string (see
167 "Mount Options" section)
169 The mount() method must return the root dentry of the tree requested by
170 caller. An active reference to its superblock must be grabbed and the
171 superblock must be locked. On failure it should return ERR_PTR(error).
173 The arguments match those of mount(2) and their interpretation depends
174 on filesystem type. E.g. for block filesystems, dev_name is interpreted
175 as block device name, that device is opened and if it contains a
176 suitable filesystem image the method creates and initializes struct
177 super_block accordingly, returning its root dentry to caller.
179 ->mount() may choose to return a subtree of existing filesystem - it
180 doesn't have to create a new one. The main result from the caller's
181 point of view is a reference to dentry at the root of (sub)tree to be
182 attached; creation of new superblock is a common side effect.
184 The most interesting member of the superblock structure that the mount()
185 method fills in is the "s_op" field. This is a pointer to a "struct
186 super_operations" which describes the next level of the filesystem
189 Usually, a filesystem uses one of the generic mount() implementations
190 and provides a fill_super() callback instead. The generic variants are:
193 mount a filesystem residing on a block device
196 mount a filesystem that is not backed by a device
199 mount a filesystem which shares the instance between all mounts
201 A fill_super() callback implementation has the following arguments:
203 ``struct super_block *sb``
204 the superblock structure. The callback must initialize this
208 arbitrary mount options, usually comes as an ASCII string (see
209 "Mount Options" section)
212 whether or not to be silent on error
215 The Superblock Object
216 =====================
218 A superblock object represents a mounted filesystem.
221 struct super_operations
222 -----------------------
224 This describes how the VFS can manipulate the superblock of your
225 filesystem. As of kernel 2.6.22, the following members are defined:
229 struct super_operations {
230 struct inode *(*alloc_inode)(struct super_block *sb);
231 void (*destroy_inode)(struct inode *);
233 void (*dirty_inode) (struct inode *, int flags);
234 int (*write_inode) (struct inode *, int);
235 void (*drop_inode) (struct inode *);
236 void (*delete_inode) (struct inode *);
237 void (*put_super) (struct super_block *);
238 int (*sync_fs)(struct super_block *sb, int wait);
239 int (*freeze_fs) (struct super_block *);
240 int (*unfreeze_fs) (struct super_block *);
241 int (*statfs) (struct dentry *, struct kstatfs *);
242 int (*remount_fs) (struct super_block *, int *, char *);
243 void (*clear_inode) (struct inode *);
244 void (*umount_begin) (struct super_block *);
246 int (*show_options)(struct seq_file *, struct dentry *);
248 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
249 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
250 int (*nr_cached_objects)(struct super_block *);
251 void (*free_cached_objects)(struct super_block *, int);
254 All methods are called without any locks being held, unless otherwise
255 noted. This means that most methods can block safely. All methods are
256 only called from a process context (i.e. not from an interrupt handler
260 this method is called by alloc_inode() to allocate memory for
261 struct inode and initialize it. If this function is not
262 defined, a simple 'struct inode' is allocated. Normally
263 alloc_inode will be used to allocate a larger structure which
264 contains a 'struct inode' embedded within it.
267 this method is called by destroy_inode() to release resources
268 allocated for struct inode. It is only required if
269 ->alloc_inode was defined and simply undoes anything done by
273 this method is called by the VFS when an inode is marked dirty.
274 This is specifically for the inode itself being marked dirty,
275 not its data. If the update needs to be persisted by fdatasync(),
276 then I_DIRTY_DATASYNC will be set in the flags argument.
279 this method is called when the VFS needs to write an inode to
280 disc. The second parameter indicates whether the write should
281 be synchronous or not, not all filesystems check this flag.
284 called when the last access to the inode is dropped, with the
285 inode->i_lock spinlock held.
287 This method should be either NULL (normal UNIX filesystem
288 semantics) or "generic_delete_inode" (for filesystems that do
289 not want to cache inodes - causing "delete_inode" to always be
290 called regardless of the value of i_nlink)
292 The "generic_delete_inode()" behavior is equivalent to the old
293 practice of using "force_delete" in the put_inode() case, but
294 does not have the races that the "force_delete()" approach had.
297 called when the VFS wants to delete an inode
300 called when the VFS wishes to free the superblock
301 (i.e. unmount). This is called with the superblock lock held
304 called when VFS is writing out all dirty data associated with a
305 superblock. The second parameter indicates whether the method
306 should wait until the write out has been completed. Optional.
309 called when VFS is locking a filesystem and forcing it into a
310 consistent state. This method is currently used by the Logical
311 Volume Manager (LVM).
314 called when VFS is unlocking a filesystem and making it writable
318 called when the VFS needs to get filesystem statistics.
321 called when the filesystem is remounted. This is called with
325 called then the VFS clears the inode. Optional
328 called when the VFS is unmounting a filesystem.
331 called by the VFS to show mount options for /proc/<pid>/mounts.
332 (see "Mount Options" section)
335 called by the VFS to read from filesystem quota file.
338 called by the VFS to write to filesystem quota file.
340 ``nr_cached_objects``
341 called by the sb cache shrinking function for the filesystem to
342 return the number of freeable cached objects it contains.
345 ``free_cache_objects``
346 called by the sb cache shrinking function for the filesystem to
347 scan the number of objects indicated to try to free them.
348 Optional, but any filesystem implementing this method needs to
349 also implement ->nr_cached_objects for it to be called
352 We can't do anything with any errors that the filesystem might
353 encountered, hence the void return type. This will never be
354 called if the VM is trying to reclaim under GFP_NOFS conditions,
355 hence this method does not need to handle that situation itself.
357 Implementations must include conditional reschedule calls inside
358 any scanning loop that is done. This allows the VFS to
359 determine appropriate scan batch sizes without having to worry
360 about whether implementations will cause holdoff problems due to
361 large scan batch sizes.
363 Whoever sets up the inode is responsible for filling in the "i_op"
364 field. This is a pointer to a "struct inode_operations" which describes
365 the methods that can be performed on individual inodes.
368 struct xattr_handlers
369 ---------------------
371 On filesystems that support extended attributes (xattrs), the s_xattr
372 superblock field points to a NULL-terminated array of xattr handlers.
373 Extended attributes are name:value pairs.
376 Indicates that the handler matches attributes with the specified
377 name (such as "system.posix_acl_access"); the prefix field must
381 Indicates that the handler matches all attributes with the
382 specified name prefix (such as "user."); the name field must be
386 Determine if attributes matching this xattr handler should be
387 listed for a particular dentry. Used by some listxattr
388 implementations like generic_listxattr.
391 Called by the VFS to get the value of a particular extended
392 attribute. This method is called by the getxattr(2) system
396 Called by the VFS to set the value of a particular extended
397 attribute. When the new value is NULL, called to remove a
398 particular extended attribute. This method is called by the
399 setxattr(2) and removexattr(2) system calls.
401 When none of the xattr handlers of a filesystem match the specified
402 attribute name or when a filesystem doesn't support extended attributes,
403 the various ``*xattr(2)`` system calls return -EOPNOTSUPP.
409 An inode object represents an object within the filesystem.
412 struct inode_operations
413 -----------------------
415 This describes how the VFS can manipulate an inode in your filesystem.
416 As of kernel 2.6.22, the following members are defined:
420 struct inode_operations {
421 int (*create) (struct user_namespace *, struct inode *,struct dentry *, umode_t, bool);
422 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
423 int (*link) (struct dentry *,struct inode *,struct dentry *);
424 int (*unlink) (struct inode *,struct dentry *);
425 int (*symlink) (struct user_namespace *, struct inode *,struct dentry *,const char *);
426 int (*mkdir) (struct user_namespace *, struct inode *,struct dentry *,umode_t);
427 int (*rmdir) (struct inode *,struct dentry *);
428 int (*mknod) (struct user_namespace *, struct inode *,struct dentry *,umode_t,dev_t);
429 int (*rename) (struct user_namespace *, struct inode *, struct dentry *,
430 struct inode *, struct dentry *, unsigned int);
431 int (*readlink) (struct dentry *, char __user *,int);
432 const char *(*get_link) (struct dentry *, struct inode *,
433 struct delayed_call *);
434 int (*permission) (struct user_namespace *, struct inode *, int);
435 struct posix_acl * (*get_acl)(struct inode *, int, bool);
436 int (*setattr) (struct user_namespace *, struct dentry *, struct iattr *);
437 int (*getattr) (struct user_namespace *, const struct path *, struct kstat *, u32, unsigned int);
438 ssize_t (*listxattr) (struct dentry *, char *, size_t);
439 void (*update_time)(struct inode *, struct timespec *, int);
440 int (*atomic_open)(struct inode *, struct dentry *, struct file *,
441 unsigned open_flag, umode_t create_mode);
442 int (*tmpfile) (struct user_namespace *, struct inode *, struct dentry *, umode_t);
443 int (*set_acl)(struct user_namespace *, struct inode *, struct posix_acl *, int);
444 int (*fileattr_set)(struct user_namespace *mnt_userns,
445 struct dentry *dentry, struct fileattr *fa);
446 int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa);
449 Again, all methods are called without any locks being held, unless
453 called by the open(2) and creat(2) system calls. Only required
454 if you want to support regular files. The dentry you get should
455 not have an inode (i.e. it should be a negative dentry). Here
456 you will probably call d_instantiate() with the dentry and the
460 called when the VFS needs to look up an inode in a parent
461 directory. The name to look for is found in the dentry. This
462 method must call d_add() to insert the found inode into the
463 dentry. The "i_count" field in the inode structure should be
464 incremented. If the named inode does not exist a NULL inode
465 should be inserted into the dentry (this is called a negative
466 dentry). Returning an error code from this routine must only be
467 done on a real error, otherwise creating inodes with system
468 calls like create(2), mknod(2), mkdir(2) and so on will fail.
469 If you wish to overload the dentry methods then you should
470 initialise the "d_dop" field in the dentry; this is a pointer to
471 a struct "dentry_operations". This method is called with the
472 directory inode semaphore held
475 called by the link(2) system call. Only required if you want to
476 support hard links. You will probably need to call
477 d_instantiate() just as you would in the create() method
480 called by the unlink(2) system call. Only required if you want
481 to support deleting inodes
484 called by the symlink(2) system call. Only required if you want
485 to support symlinks. You will probably need to call
486 d_instantiate() just as you would in the create() method
489 called by the mkdir(2) system call. Only required if you want
490 to support creating subdirectories. You will probably need to
491 call d_instantiate() just as you would in the create() method
494 called by the rmdir(2) system call. Only required if you want
495 to support deleting subdirectories
498 called by the mknod(2) system call to create a device (char,
499 block) inode or a named pipe (FIFO) or socket. Only required if
500 you want to support creating these types of inodes. You will
501 probably need to call d_instantiate() just as you would in the
505 called by the rename(2) system call to rename the object to have
506 the parent and name given by the second inode and dentry.
508 The filesystem must return -EINVAL for any unsupported or
509 unknown flags. Currently the following flags are implemented:
510 (1) RENAME_NOREPLACE: this flag indicates that if the target of
511 the rename exists the rename should fail with -EEXIST instead of
512 replacing the target. The VFS already checks for existence, so
513 for local filesystems the RENAME_NOREPLACE implementation is
514 equivalent to plain rename.
515 (2) RENAME_EXCHANGE: exchange source and target. Both must
516 exist; this is checked by the VFS. Unlike plain rename, source
517 and target may be of different type.
520 called by the VFS to follow a symbolic link to the inode it
521 points to. Only required if you want to support symbolic links.
522 This method returns the symlink body to traverse (and possibly
523 resets the current position with nd_jump_link()). If the body
524 won't go away until the inode is gone, nothing else is needed;
525 if it needs to be otherwise pinned, arrange for its release by
526 having get_link(..., ..., done) do set_delayed_call(done,
527 destructor, argument). In that case destructor(argument) will
528 be called once VFS is done with the body you've returned. May
529 be called in RCU mode; that is indicated by NULL dentry
530 argument. If request can't be handled without leaving RCU mode,
531 have it return ERR_PTR(-ECHILD).
533 If the filesystem stores the symlink target in ->i_link, the
534 VFS may use it directly without calling ->get_link(); however,
535 ->get_link() must still be provided. ->i_link must not be
536 freed until after an RCU grace period. Writing to ->i_link
537 post-iget() time requires a 'release' memory barrier.
540 this is now just an override for use by readlink(2) for the
541 cases when ->get_link uses nd_jump_link() or object is not in
542 fact a symlink. Normally filesystems should only implement
543 ->get_link for symlinks and readlink(2) will automatically use
547 called by the VFS to check for access rights on a POSIX-like
550 May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in
551 rcu-walk mode, the filesystem must check the permission without
552 blocking or storing to the inode.
554 If a situation is encountered that rcu-walk cannot handle,
556 -ECHILD and it will be called again in ref-walk mode.
559 called by the VFS to set attributes for a file. This method is
560 called by chmod(2) and related system calls.
563 called by the VFS to get attributes of a file. This method is
564 called by stat(2) and related system calls.
567 called by the VFS to list all extended attributes for a given
568 file. This method is called by the listxattr(2) system call.
571 called by the VFS to update a specific time or the i_version of
572 an inode. If this is not defined the VFS will update the inode
573 itself and call mark_inode_dirty_sync.
576 called on the last component of an open. Using this optional
577 method the filesystem can look up, possibly create and open the
578 file in one atomic operation. If it wants to leave actual
579 opening to the caller (e.g. if the file turned out to be a
580 symlink, device, or just something filesystem won't do atomic
581 open for), it may signal this by returning finish_no_open(file,
582 dentry). This method is only called if the last component is
583 negative or needs lookup. Cached positive dentries are still
584 handled by f_op->open(). If the file was created, FMODE_CREATED
585 flag should be set in file->f_mode. In case of O_EXCL the
586 method must only succeed if the file didn't exist and hence
587 FMODE_CREATED shall always be set on success.
590 called in the end of O_TMPFILE open(). Optional, equivalent to
591 atomically creating, opening and unlinking a file in given
595 called on ioctl(FS_IOC_GETFLAGS) and ioctl(FS_IOC_FSGETXATTR) to
596 retrieve miscellaneous file flags and attributes. Also called
597 before the relevant SET operation to check what is being changed
598 (in this case with i_rwsem locked exclusive). If unset, then
599 fall back to f_op->ioctl().
602 called on ioctl(FS_IOC_SETFLAGS) and ioctl(FS_IOC_FSSETXATTR) to
603 change miscellaneous file flags and attributes. Callers hold
604 i_rwsem exclusive. If unset, then fall back to f_op->ioctl().
607 The Address Space Object
608 ========================
610 The address space object is used to group and manage pages in the page
611 cache. It can be used to keep track of the pages in a file (or anything
612 else) and also track the mapping of sections of the file into process
615 There are a number of distinct yet related services that an
616 address-space can provide. These include communicating memory pressure,
617 page lookup by address, and keeping track of pages tagged as Dirty or
620 The first can be used independently to the others. The VM can try to
621 either write dirty pages in order to clean them, or release clean pages
622 in order to reuse them. To do this it can call the ->writepage method
623 on dirty pages, and ->releasepage on clean pages with PagePrivate set.
624 Clean pages without PagePrivate and with no external references will be
625 released without notice being given to the address_space.
627 To achieve this functionality, pages need to be placed on an LRU with
628 lru_cache_add and mark_page_active needs to be called whenever the page
631 Pages are normally kept in a radix tree index by ->index. This tree
632 maintains information about the PG_Dirty and PG_Writeback status of each
633 page, so that pages with either of these flags can be found quickly.
635 The Dirty tag is primarily used by mpage_writepages - the default
636 ->writepages method. It uses the tag to find dirty pages to call
637 ->writepage on. If mpage_writepages is not used (i.e. the address
638 provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is almost
639 unused. write_inode_now and sync_inode do use it (through
640 __sync_single_inode) to check if ->writepages has been successful in
641 writing out the whole address_space.
643 The Writeback tag is used by filemap*wait* and sync_page* functions, via
644 filemap_fdatawait_range, to wait for all writeback to complete.
646 An address_space handler may attach extra information to a page,
647 typically using the 'private' field in the 'struct page'. If such
648 information is attached, the PG_Private flag should be set. This will
649 cause various VM routines to make extra calls into the address_space
650 handler to deal with that data.
652 An address space acts as an intermediate between storage and
653 application. Data is read into the address space a whole page at a
654 time, and provided to the application either by copying of the page, or
655 by memory-mapping the page. Data is written into the address space by
656 the application, and then written-back to storage typically in whole
657 pages, however the address_space has finer control of write sizes.
659 The read process essentially only requires 'readpage'. The write
660 process is more complicated and uses write_begin/write_end or
661 dirty_folio to write data into the address_space, and writepage and
662 writepages to writeback data to storage.
664 Adding and removing pages to/from an address_space is protected by the
667 When data is written to a page, the PG_Dirty flag should be set. It
668 typically remains set until writepage asks for it to be written. This
669 should clear PG_Dirty and set PG_Writeback. It can be actually written
670 at any point after PG_Dirty is clear. Once it is known to be safe,
671 PG_Writeback is cleared.
673 Writeback makes use of a writeback_control structure to direct the
674 operations. This gives the writepage and writepages operations some
675 information about the nature of and reason for the writeback request,
676 and the constraints under which it is being done. It is also used to
677 return information back to the caller about the result of a writepage or
681 Handling errors during writeback
682 --------------------------------
684 Most applications that do buffered I/O will periodically call a file
685 synchronization call (fsync, fdatasync, msync or sync_file_range) to
686 ensure that data written has made it to the backing store. When there
687 is an error during writeback, they expect that error to be reported when
688 a file sync request is made. After an error has been reported on one
689 request, subsequent requests on the same file descriptor should return
690 0, unless further writeback errors have occurred since the previous file
693 Ideally, the kernel would report errors only on file descriptions on
694 which writes were done that subsequently failed to be written back. The
695 generic pagecache infrastructure does not track the file descriptions
696 that have dirtied each individual page however, so determining which
697 file descriptors should get back an error is not possible.
699 Instead, the generic writeback error tracking infrastructure in the
700 kernel settles for reporting errors to fsync on all file descriptions
701 that were open at the time that the error occurred. In a situation with
702 multiple writers, all of them will get back an error on a subsequent
703 fsync, even if all of the writes done through that particular file
704 descriptor succeeded (or even if there were no writes on that file
707 Filesystems that wish to use this infrastructure should call
708 mapping_set_error to record the error in the address_space when it
709 occurs. Then, after writing back data from the pagecache in their
710 file->fsync operation, they should call file_check_and_advance_wb_err to
711 ensure that the struct file's error cursor has advanced to the correct
712 point in the stream of errors emitted by the backing device(s).
715 struct address_space_operations
716 -------------------------------
718 This describes how the VFS can manipulate mapping of a file to page
719 cache in your filesystem. The following members are defined:
723 struct address_space_operations {
724 int (*writepage)(struct page *page, struct writeback_control *wbc);
725 int (*readpage)(struct file *, struct page *);
726 int (*writepages)(struct address_space *, struct writeback_control *);
727 bool (*dirty_folio)(struct address_space *, struct folio *);
728 void (*readahead)(struct readahead_control *);
729 int (*write_begin)(struct file *, struct address_space *mapping,
730 loff_t pos, unsigned len, unsigned flags,
731 struct page **pagep, void **fsdata);
732 int (*write_end)(struct file *, struct address_space *mapping,
733 loff_t pos, unsigned len, unsigned copied,
734 struct page *page, void *fsdata);
735 sector_t (*bmap)(struct address_space *, sector_t);
736 void (*invalidate_folio) (struct folio *, size_t start, size_t len);
737 int (*releasepage) (struct page *, int);
738 void (*freepage)(struct page *);
739 ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter);
740 /* isolate a page for migration */
741 bool (*isolate_page) (struct page *, isolate_mode_t);
742 /* migrate the contents of a page to the specified target */
743 int (*migratepage) (struct page *, struct page *);
744 /* put migration-failed page back to right list */
745 void (*putback_page) (struct page *);
746 int (*launder_folio) (struct folio *);
748 bool (*is_partially_uptodate) (struct folio *, size_t from,
750 void (*is_dirty_writeback) (struct page *, bool *, bool *);
751 int (*error_remove_page) (struct mapping *mapping, struct page *page);
752 int (*swap_activate)(struct file *);
753 int (*swap_deactivate)(struct file *);
757 called by the VM to write a dirty page to backing store. This
758 may happen for data integrity reasons (i.e. 'sync'), or to free
759 up memory (flush). The difference can be seen in
760 wbc->sync_mode. The PG_Dirty flag has been cleared and
761 PageLocked is true. writepage should start writeout, should set
762 PG_Writeback, and should make sure the page is unlocked, either
763 synchronously or asynchronously when the write operation
766 If wbc->sync_mode is WB_SYNC_NONE, ->writepage doesn't have to
767 try too hard if there are problems, and may choose to write out
768 other pages from the mapping if that is easier (e.g. due to
769 internal dependencies). If it chooses not to start writeout, it
770 should return AOP_WRITEPAGE_ACTIVATE so that the VM will not
771 keep calling ->writepage on that page.
773 See the file "Locking" for more details.
776 called by the VM to read a page from backing store. The page
777 will be Locked when readpage is called, and should be unlocked
778 and marked uptodate once the read completes. If ->readpage
779 discovers that it needs to unlock the page for some reason, it
780 can do so, and then return AOP_TRUNCATED_PAGE. In this case,
781 the page will be relocated, relocked and if that all succeeds,
782 ->readpage will be called again.
785 called by the VM to write out pages associated with the
786 address_space object. If wbc->sync_mode is WB_SYNC_ALL, then
787 the writeback_control will specify a range of pages that must be
788 written out. If it is WB_SYNC_NONE, then a nr_to_write is
789 given and that many pages should be written if possible. If no
790 ->writepages is given, then mpage_writepages is used instead.
791 This will choose pages from the address space that are tagged as
792 DIRTY and will pass them to ->writepage.
795 called by the VM to mark a folio as dirty. This is particularly
796 needed if an address space attaches private data to a folio, and
797 that data needs to be updated when a folio is dirtied. This is
798 called, for example, when a memory mapped page gets modified.
799 If defined, it should set the folio dirty flag, and the
800 PAGECACHE_TAG_DIRTY search mark in i_pages.
803 Called by the VM to read pages associated with the address_space
804 object. The pages are consecutive in the page cache and are
805 locked. The implementation should decrement the page refcount
806 after starting I/O on each page. Usually the page will be
807 unlocked by the I/O completion handler. The set of pages are
808 divided into some sync pages followed by some async pages,
809 rac->ra->async_size gives the number of async pages. The
810 filesystem should attempt to read all sync pages but may decide
811 to stop once it reaches the async pages. If it does decide to
812 stop attempting I/O, it can simply return. The caller will
813 remove the remaining pages from the address space, unlock them
814 and decrement the page refcount. Set PageUptodate if the I/O
815 completes successfully. Setting PageError on any page will be
816 ignored; simply unlock the page if an I/O error occurs.
819 Called by the generic buffered write code to ask the filesystem
820 to prepare to write len bytes at the given offset in the file.
821 The address_space should check that the write will be able to
822 complete, by allocating space if necessary and doing any other
823 internal housekeeping. If the write will update parts of any
824 basic-blocks on storage, then those blocks should be pre-read
825 (if they haven't been read already) so that the updated blocks
826 can be written out properly.
828 The filesystem must return the locked pagecache page for the
829 specified offset, in ``*pagep``, for the caller to write into.
831 It must be able to cope with short writes (where the length
832 passed to write_begin is greater than the number of bytes copied
835 flags is a field for AOP_FLAG_xxx flags, described in
838 A void * may be returned in fsdata, which then gets passed into
841 Returns 0 on success; < 0 on failure (which is the error code),
842 in which case write_end is not called.
845 After a successful write_begin, and data copy, write_end must be
846 called. len is the original len passed to write_begin, and
847 copied is the amount that was able to be copied.
849 The filesystem must take care of unlocking the page and
850 releasing it refcount, and updating i_size.
852 Returns < 0 on failure, otherwise the number of bytes (<=
853 'copied') that were able to be copied into pagecache.
856 called by the VFS to map a logical block offset within object to
857 physical block number. This method is used by the FIBMAP ioctl
858 and for working with swap-files. To be able to swap to a file,
859 the file must have a stable mapping to a block device. The swap
860 system does not go through the filesystem but instead uses bmap
861 to find out where the blocks in the file are and uses those
865 If a folio has private data, then invalidate_folio will be
866 called when part or all of the folio is to be removed from the
867 address space. This generally corresponds to either a
868 truncation, punch hole or a complete invalidation of the address
869 space (in the latter case 'offset' will always be 0 and 'length'
870 will be folio_size()). Any private data associated with the page
871 should be updated to reflect this truncation. If offset is 0
872 and length is folio_size(), then the private data should be
873 released, because the page must be able to be completely
874 discarded. This may be done by calling the ->releasepage
875 function, but in this case the release MUST succeed.
878 releasepage is called on PagePrivate pages to indicate that the
879 page should be freed if possible. ->releasepage should remove
880 any private data from the page and clear the PagePrivate flag.
881 If releasepage() fails for some reason, it must indicate failure
882 with a 0 return value. releasepage() is used in two distinct
883 though related cases. The first is when the VM finds a clean
884 page with no active users and wants to make it a free page. If
885 ->releasepage succeeds, the page will be removed from the
886 address_space and become free.
888 The second case is when a request has been made to invalidate
889 some or all pages in an address_space. This can happen through
890 the fadvise(POSIX_FADV_DONTNEED) system call or by the
891 filesystem explicitly requesting it as nfs and 9fs do (when they
892 believe the cache may be out of date with storage) by calling
893 invalidate_inode_pages2(). If the filesystem makes such a call,
894 and needs to be certain that all pages are invalidated, then its
895 releasepage will need to ensure this. Possibly it can clear the
896 PageUptodate bit if it cannot free private data yet.
899 freepage is called once the page is no longer visible in the
900 page cache in order to allow the cleanup of any private data.
901 Since it may be called by the memory reclaimer, it should not
902 assume that the original address_space mapping still exists, and
906 called by the generic read/write routines to perform direct_IO -
907 that is IO requests which bypass the page cache and transfer
908 data directly between the storage and the application's address
912 Called by the VM when isolating a movable non-lru page. If page
913 is successfully isolated, VM marks the page as PG_isolated via
917 This is used to compact the physical memory usage. If the VM
918 wants to relocate a page (maybe off a memory card that is
919 signalling imminent failure) it will pass a new page and an old
920 page to this function. migrate_page should transfer any private
921 data across and update any references that it has to the page.
924 Called by the VM when isolated page's migration fails.
927 Called before freeing a folio - it writes back the dirty folio.
928 To prevent redirtying the folio, it is kept locked during the
931 ``is_partially_uptodate``
932 Called by the VM when reading a file through the pagecache when
933 the underlying blocksize is smaller than the size of the folio.
934 If the required block is up to date then the read can complete
935 without needing I/O to bring the whole page up to date.
937 ``is_dirty_writeback``
938 Called by the VM when attempting to reclaim a page. The VM uses
939 dirty and writeback information to determine if it needs to
940 stall to allow flushers a chance to complete some IO.
941 Ordinarily it can use PageDirty and PageWriteback but some
942 filesystems have more complex state (unstable pages in NFS
943 prevent reclaim) or do not set those flags due to locking
944 problems. This callback allows a filesystem to indicate to the
945 VM if a page should be treated as dirty or writeback for the
946 purposes of stalling.
948 ``error_remove_page``
949 normally set to generic_error_remove_page if truncation is ok
950 for this address space. Used for memory failure handling.
951 Setting this implies you deal with pages going away under you,
952 unless you have them locked or reference counts increased.
955 Called when swapon is used on a file to allocate space if
956 necessary and pin the block lookup information in memory. A
957 return value of zero indicates success, in which case this file
958 can be used to back swapspace.
961 Called during swapoff on files where swap_activate was
968 A file object represents a file opened by a process. This is also known
969 as an "open file description" in POSIX parlance.
972 struct file_operations
973 ----------------------
975 This describes how the VFS can manipulate an open file. As of kernel
976 4.18, the following members are defined:
980 struct file_operations {
981 struct module *owner;
982 loff_t (*llseek) (struct file *, loff_t, int);
983 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
984 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
985 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
986 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
987 int (*iopoll)(struct kiocb *kiocb, bool spin);
988 int (*iterate) (struct file *, struct dir_context *);
989 int (*iterate_shared) (struct file *, struct dir_context *);
990 __poll_t (*poll) (struct file *, struct poll_table_struct *);
991 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
992 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
993 int (*mmap) (struct file *, struct vm_area_struct *);
994 int (*open) (struct inode *, struct file *);
995 int (*flush) (struct file *, fl_owner_t id);
996 int (*release) (struct inode *, struct file *);
997 int (*fsync) (struct file *, loff_t, loff_t, int datasync);
998 int (*fasync) (int, struct file *, int);
999 int (*lock) (struct file *, int, struct file_lock *);
1000 ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
1001 unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1002 int (*check_flags)(int);
1003 int (*flock) (struct file *, int, struct file_lock *);
1004 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
1005 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
1006 int (*setlease)(struct file *, long, struct file_lock **, void **);
1007 long (*fallocate)(struct file *file, int mode, loff_t offset,
1009 void (*show_fdinfo)(struct seq_file *m, struct file *f);
1011 unsigned (*mmap_capabilities)(struct file *);
1013 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *, loff_t, size_t, unsigned int);
1014 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
1015 struct file *file_out, loff_t pos_out,
1016 loff_t len, unsigned int remap_flags);
1017 int (*fadvise)(struct file *, loff_t, loff_t, int);
1020 Again, all methods are called without any locks being held, unless
1024 called when the VFS needs to move the file position index
1027 called by read(2) and related system calls
1030 possibly asynchronous read with iov_iter as destination
1033 called by write(2) and related system calls
1036 possibly asynchronous write with iov_iter as source
1039 called when aio wants to poll for completions on HIPRI iocbs
1042 called when the VFS needs to read the directory contents
1045 called when the VFS needs to read the directory contents when
1046 filesystem supports concurrent dir iterators
1049 called by the VFS when a process wants to check if there is
1050 activity on this file and (optionally) go to sleep until there
1051 is activity. Called by the select(2) and poll(2) system calls
1054 called by the ioctl(2) system call.
1057 called by the ioctl(2) system call when 32 bit system calls are
1058 used on 64 bit kernels.
1061 called by the mmap(2) system call
1064 called by the VFS when an inode should be opened. When the VFS
1065 opens a file, it creates a new "struct file". It then calls the
1066 open method for the newly allocated file structure. You might
1067 think that the open method really belongs in "struct
1068 inode_operations", and you may be right. I think it's done the
1069 way it is because it makes filesystems simpler to implement.
1070 The open() method is a good place to initialize the
1071 "private_data" member in the file structure if you want to point
1072 to a device structure
1075 called by the close(2) system call to flush a file
1078 called when the last reference to an open file is closed
1081 called by the fsync(2) system call. Also see the section above
1082 entitled "Handling errors during writeback".
1085 called by the fcntl(2) system call when asynchronous
1086 (non-blocking) mode is enabled for a file
1089 called by the fcntl(2) system call for F_GETLK, F_SETLK, and
1092 ``get_unmapped_area``
1093 called by the mmap(2) system call
1096 called by the fcntl(2) system call for F_SETFL command
1099 called by the flock(2) system call
1102 called by the VFS to splice data from a pipe to a file. This
1103 method is used by the splice(2) system call
1106 called by the VFS to splice data from file to a pipe. This
1107 method is used by the splice(2) system call
1110 called by the VFS to set or release a file lock lease. setlease
1111 implementations should call generic_setlease to record or remove
1112 the lease in the inode after setting it.
1115 called by the VFS to preallocate blocks or punch a hole.
1118 called by the copy_file_range(2) system call.
1120 ``remap_file_range``
1121 called by the ioctl(2) system call for FICLONERANGE and FICLONE
1122 and FIDEDUPERANGE commands to remap file ranges. An
1123 implementation should remap len bytes at pos_in of the source
1124 file into the dest file at pos_out. Implementations must handle
1125 callers passing in len == 0; this means "remap to the end of the
1126 source file". The return value should the number of bytes
1127 remapped, or the usual negative error code if errors occurred
1128 before any bytes were remapped. The remap_flags parameter
1129 accepts REMAP_FILE_* flags. If REMAP_FILE_DEDUP is set then the
1130 implementation must only remap if the requested file ranges have
1131 identical contents. If REMAP_FILE_CAN_SHORTEN is set, the caller is
1132 ok with the implementation shortening the request length to
1133 satisfy alignment or EOF requirements (or any other reason).
1136 possibly called by the fadvise64() system call.
1138 Note that the file operations are implemented by the specific
1139 filesystem in which the inode resides. When opening a device node
1140 (character or block special) most filesystems will call special
1141 support routines in the VFS which will locate the required device
1142 driver information. These support routines replace the filesystem file
1143 operations with those for the device driver, and then proceed to call
1144 the new open() method for the file. This is how opening a device file
1145 in the filesystem eventually ends up calling the device driver open()
1149 Directory Entry Cache (dcache)
1150 ==============================
1153 struct dentry_operations
1154 ------------------------
1156 This describes how a filesystem can overload the standard dentry
1157 operations. Dentries and the dcache are the domain of the VFS and the
1158 individual filesystem implementations. Device drivers have no business
1159 here. These methods may be set to NULL, as they are either optional or
1160 the VFS uses a default. As of kernel 2.6.22, the following members are
1165 struct dentry_operations {
1166 int (*d_revalidate)(struct dentry *, unsigned int);
1167 int (*d_weak_revalidate)(struct dentry *, unsigned int);
1168 int (*d_hash)(const struct dentry *, struct qstr *);
1169 int (*d_compare)(const struct dentry *,
1170 unsigned int, const char *, const struct qstr *);
1171 int (*d_delete)(const struct dentry *);
1172 int (*d_init)(struct dentry *);
1173 void (*d_release)(struct dentry *);
1174 void (*d_iput)(struct dentry *, struct inode *);
1175 char *(*d_dname)(struct dentry *, char *, int);
1176 struct vfsmount *(*d_automount)(struct path *);
1177 int (*d_manage)(const struct path *, bool);
1178 struct dentry *(*d_real)(struct dentry *, const struct inode *);
1182 called when the VFS needs to revalidate a dentry. This is
1183 called whenever a name look-up finds a dentry in the dcache.
1184 Most local filesystems leave this as NULL, because all their
1185 dentries in the dcache are valid. Network filesystems are
1186 different since things can change on the server without the
1187 client necessarily being aware of it.
1189 This function should return a positive value if the dentry is
1190 still valid, and zero or a negative error code if it isn't.
1192 d_revalidate may be called in rcu-walk mode (flags &
1193 LOOKUP_RCU). If in rcu-walk mode, the filesystem must
1194 revalidate the dentry without blocking or storing to the dentry,
1195 d_parent and d_inode should not be used without care (because
1196 they can change and, in d_inode case, even become NULL under
1199 If a situation is encountered that rcu-walk cannot handle,
1201 -ECHILD and it will be called again in ref-walk mode.
1203 ``_weak_revalidate``
1204 called when the VFS needs to revalidate a "jumped" dentry. This
1205 is called when a path-walk ends at dentry that was not acquired
1206 by doing a lookup in the parent directory. This includes "/",
1207 "." and "..", as well as procfs-style symlinks and mountpoint
1210 In this case, we are less concerned with whether the dentry is
1211 still fully correct, but rather that the inode is still valid.
1212 As with d_revalidate, most local filesystems will set this to
1213 NULL since their dcache entries are always valid.
1215 This function has the same return code semantics as
1218 d_weak_revalidate is only called after leaving rcu-walk mode.
1221 called when the VFS adds a dentry to the hash table. The first
1222 dentry passed to d_hash is the parent directory that the name is
1225 Same locking and synchronisation rules as d_compare regarding
1226 what is safe to dereference etc.
1229 called to compare a dentry name with a given name. The first
1230 dentry is the parent of the dentry to be compared, the second is
1231 the child dentry. len and name string are properties of the
1232 dentry to be compared. qstr is the name to compare it with.
1234 Must be constant and idempotent, and should not take locks if
1235 possible, and should not or store into the dentry. Should not
1236 dereference pointers outside the dentry without lots of care
1237 (eg. d_parent, d_inode, d_name should not be used).
1239 However, our vfsmount is pinned, and RCU held, so the dentries
1240 and inodes won't disappear, neither will our sb or filesystem
1241 module. ->d_sb may be used.
1243 It is a tricky calling convention because it needs to be called
1244 under "rcu-walk", ie. without any locks or references on things.
1247 called when the last reference to a dentry is dropped and the
1248 dcache is deciding whether or not to cache it. Return 1 to
1249 delete immediately, or 0 to cache the dentry. Default is NULL
1250 which means to always cache a reachable dentry. d_delete must
1251 be constant and idempotent.
1254 called when a dentry is allocated
1257 called when a dentry is really deallocated
1260 called when a dentry loses its inode (just prior to its being
1261 deallocated). The default when this is NULL is that the VFS
1262 calls iput(). If you define this method, you must call iput()
1266 called when the pathname of a dentry should be generated.
1267 Useful for some pseudo filesystems (sockfs, pipefs, ...) to
1268 delay pathname generation. (Instead of doing it when dentry is
1269 created, it's done only when the path is needed.). Real
1270 filesystems probably dont want to use it, because their dentries
1271 are present in global dcache hash, so their hash should be an
1272 invariant. As no lock is held, d_dname() should not try to
1273 modify the dentry itself, unless appropriate SMP safety is used.
1274 CAUTION : d_path() logic is quite tricky. The correct way to
1275 return for example "Hello" is to put it at the end of the
1276 buffer, and returns a pointer to the first char.
1277 dynamic_dname() helper function is provided to take care of
1284 static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1286 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1287 dentry->d_inode->i_ino);
1291 called when an automount dentry is to be traversed (optional).
1292 This should create a new VFS mount record and return the record
1293 to the caller. The caller is supplied with a path parameter
1294 giving the automount directory to describe the automount target
1295 and the parent VFS mount record to provide inheritable mount
1296 parameters. NULL should be returned if someone else managed to
1297 make the automount first. If the vfsmount creation failed, then
1298 an error code should be returned. If -EISDIR is returned, then
1299 the directory will be treated as an ordinary directory and
1300 returned to pathwalk to continue walking.
1302 If a vfsmount is returned, the caller will attempt to mount it
1303 on the mountpoint and will remove the vfsmount from its
1304 expiration list in the case of failure. The vfsmount should be
1305 returned with 2 refs on it to prevent automatic expiration - the
1306 caller will clean up the additional ref.
1308 This function is only used if DCACHE_NEED_AUTOMOUNT is set on
1309 the dentry. This is set by __d_instantiate() if S_AUTOMOUNT is
1310 set on the inode being added.
1313 called to allow the filesystem to manage the transition from a
1314 dentry (optional). This allows autofs, for example, to hold up
1315 clients waiting to explore behind a 'mountpoint' while letting
1316 the daemon go past and construct the subtree there. 0 should be
1317 returned to let the calling process continue. -EISDIR can be
1318 returned to tell pathwalk to use this directory as an ordinary
1319 directory and to ignore anything mounted on it and not to check
1320 the automount flag. Any other error code will abort pathwalk
1323 If the 'rcu_walk' parameter is true, then the caller is doing a
1324 pathwalk in RCU-walk mode. Sleeping is not permitted in this
1325 mode, and the caller can be asked to leave it and call again by
1326 returning -ECHILD. -EISDIR may also be returned to tell
1327 pathwalk to ignore d_automount or any mounts.
1329 This function is only used if DCACHE_MANAGE_TRANSIT is set on
1330 the dentry being transited from.
1333 overlay/union type filesystems implement this method to return
1334 one of the underlying dentries hidden by the overlay. It is
1335 used in two different modes:
1337 Called from file_dentry() it returns the real dentry matching
1338 the inode argument. The real dentry may be from a lower layer
1339 already copied up, but still referenced from the file. This
1340 mode is selected with a non-NULL inode argument.
1342 With NULL inode the topmost real underlying dentry is returned.
1344 Each dentry has a pointer to its parent dentry, as well as a hash list
1345 of child dentries. Child dentries are basically like files in a
1349 Directory Entry Cache API
1350 --------------------------
1352 There are a number of functions defined which permit a filesystem to
1353 manipulate dentries:
1356 open a new handle for an existing dentry (this just increments
1360 close a handle for a dentry (decrements the usage count). If
1361 the usage count drops to 0, and the dentry is still in its
1362 parent's hash, the "d_delete" method is called to check whether
1363 it should be cached. If it should not be cached, or if the
1364 dentry is not hashed, it is deleted. Otherwise cached dentries
1365 are put into an LRU list to be reclaimed on memory shortage.
1368 this unhashes a dentry from its parents hash list. A subsequent
1369 call to dput() will deallocate the dentry if its usage count
1373 delete a dentry. If there are no other open references to the
1374 dentry then the dentry is turned into a negative dentry (the
1375 d_iput() method is called). If there are other references, then
1376 d_drop() is called instead
1379 add a dentry to its parents hash list and then calls
1383 add a dentry to the alias hash list for the inode and updates
1384 the "d_inode" member. The "i_count" member in the inode
1385 structure should be set/incremented. If the inode pointer is
1386 NULL, the dentry is called a "negative dentry". This function
1387 is commonly called when an inode is created for an existing
1391 look up a dentry given its parent and path name component It
1392 looks up the child of that given name from the dcache hash
1393 table. If it is found, the reference count is incremented and
1394 the dentry is returned. The caller must use dput() to free the
1395 dentry when it finishes using it.
1405 On mount and remount the filesystem is passed a string containing a
1406 comma separated list of mount options. The options can have either of
1412 The <linux/parser.h> header defines an API that helps parse these
1413 options. There are plenty of examples on how to use it in existing
1420 If a filesystem accepts mount options, it must define show_options() to
1421 show all the currently active options. The rules are:
1423 - options MUST be shown which are not default or their values differ
1426 - options MAY be shown which are enabled by default or have their
1429 Options used only internally between a mount helper and the kernel (such
1430 as file descriptors), or which only have an effect during the mounting
1431 (such as ones controlling the creation of a journal) are exempt from the
1434 The underlying reason for the above rules is to make sure, that a mount
1435 can be accurately replicated (e.g. umounting and mounting again) based
1436 on the information found in /proc/mounts.
1442 (Note some of these resources are not up-to-date with the latest kernel
1445 Creating Linux virtual filesystems. 2002
1446 <https://lwn.net/Articles/13325/>
1448 The Linux Virtual File-system Layer by Neil Brown. 1999
1449 <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1451 A tour of the Linux VFS by Michael K. Johnson. 1996
1452 <https://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1454 A small trail through the Linux kernel by Andries Brouwer. 2001
1455 <https://www.win.tue.nl/~aeb/linux/vfs/trail.html>