2 Overview of the Linux Virtual File System
4 Original author: Richard Gooch <rgooch@atnf.csiro.au>
6 Copyright (C) 1999 Richard Gooch
7 Copyright (C) 2005 Pekka Enberg
9 This file is released under the GPLv2.
15 The Virtual File System (also known as the Virtual Filesystem Switch)
16 is the software layer in the kernel that provides the filesystem
17 interface to userspace programs. It also provides an abstraction
18 within the kernel which allows different filesystem implementations to
21 VFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so
22 on are called from a process context. Filesystem locking is described
23 in the document Documentation/filesystems/Locking.
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,
38 some bits of the cache are missing. In order to resolve your pathname
39 into a dentry, the VFS may have to resort to creating dentries along
40 the way, and then loading the inode. This is done by looking up the
47 An individual dentry usually has a pointer to an inode. Inodes are
48 filesystem objects such as regular files, directories, FIFOs and other
49 beasts. They live either on the disc (for block device filesystems)
50 or in the memory (for pseudo filesystems). Inodes that live on the
51 disc are copied into the memory when required and changes to the inode
52 are written back to disc. A single inode can be pointed to by multiple
53 dentries (hard links, for example, do this).
55 To look up an inode requires that the VFS calls the lookup() method of
56 the parent directory inode. This method is installed by the specific
57 filesystem implementation that the inode lives in. Once the VFS has
58 the required dentry (and hence the inode), we can do all those boring
59 things like open(2) the file, or stat(2) it to peek at the inode
60 data. The stat(2) operation is fairly simple: once the VFS has the
61 dentry, it peeks at the inode data and passes some of it back to
68 Opening a file requires another operation: allocation of a file
69 structure (this is the kernel-side implementation of file
70 descriptors). The freshly allocated file structure is initialized with
71 a pointer to the dentry and a set of file operation member functions.
72 These are taken from the inode data. The open() file method is then
73 called so the specific filesystem implementation can do its work. You
74 can see that this is another switch performed by the VFS. The file
75 structure is placed into the file descriptor table for the process.
77 Reading, writing and closing files (and other assorted VFS operations)
78 is done by using the userspace file descriptor to grab the appropriate
79 file structure, and then calling the required file structure method to
80 do whatever is required. For as long as the file is open, it keeps the
81 dentry in use, which in turn means that the VFS inode is still in use.
84 Registering and Mounting a Filesystem
85 =====================================
87 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 namespace,
97 the VFS will call the appropriate mount() method for the specific
98 filesystem. New vfsmount referring to the tree returned by ->mount()
99 will be attached to the mountpoint, so that when pathname resolution
100 reaches the mountpoint it will jump into the root of that vfsmount.
102 You can see all filesystems that are registered to the kernel in the
103 file /proc/filesystems.
106 struct file_system_type
107 -----------------------
109 This describes the filesystem. As of kernel 2.6.39, the following
112 struct file_system_type {
115 struct dentry *(*mount) (struct file_system_type *, int,
116 const char *, void *);
117 void (*kill_sb) (struct super_block *);
118 struct module *owner;
119 struct file_system_type * next;
120 struct list_head fs_supers;
121 struct lock_class_key s_lock_key;
122 struct lock_class_key s_umount_key;
125 name: the name of the filesystem type, such as "ext2", "iso9660",
128 fs_flags: various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
130 mount: the method to call when a new instance of this
131 filesystem should be mounted
133 kill_sb: the method to call when an instance of this filesystem
136 owner: for internal VFS use: you should initialize this to THIS_MODULE in
139 next: for internal VFS use: you should initialize this to NULL
141 s_lock_key, s_umount_key: lockdep-specific
143 The mount() method has the following arguments:
145 struct file_system_type *fs_type: describes the filesystem, partly initialized
146 by the specific filesystem code
148 int flags: mount flags
150 const char *dev_name: the device name we are mounting.
152 void *data: arbitrary mount options, usually comes as an ASCII
153 string (see "Mount Options" section)
155 The mount() method must return the root dentry of the tree requested by
156 caller. An active reference to its superblock must be grabbed and the
157 superblock must be locked. On failure it should return ERR_PTR(error).
159 The arguments match those of mount(2) and their interpretation
160 depends on filesystem type. E.g. for block filesystems, dev_name is
161 interpreted as block device name, that device is opened and if it
162 contains a suitable filesystem image the method creates and initializes
163 struct super_block accordingly, returning its root dentry to caller.
165 ->mount() may choose to return a subtree of existing filesystem - it
166 doesn't have to create a new one. The main result from the caller's
167 point of view is a reference to dentry at the root of (sub)tree to
168 be attached; creation of new superblock is a common side effect.
170 The most interesting member of the superblock structure that the
171 mount() method fills in is the "s_op" field. This is a pointer to
172 a "struct super_operations" which describes the next level of the
173 filesystem implementation.
175 Usually, a filesystem uses one of the generic mount() implementations
176 and provides a fill_super() callback instead. The generic variants are:
178 mount_bdev: mount a filesystem residing on a block device
180 mount_nodev: mount a filesystem that is not backed by a device
182 mount_single: mount a filesystem which shares the instance between
185 A fill_super() callback implementation has the following arguments:
187 struct super_block *sb: the superblock structure. The callback
188 must initialize this properly.
190 void *data: arbitrary mount options, usually comes as an ASCII
191 string (see "Mount Options" section)
193 int silent: whether or not to be silent on error
196 The Superblock Object
197 =====================
199 A superblock object represents a mounted filesystem.
202 struct super_operations
203 -----------------------
205 This describes how the VFS can manipulate the superblock of your
206 filesystem. As of kernel 2.6.22, the following members are defined:
208 struct super_operations {
209 struct inode *(*alloc_inode)(struct super_block *sb);
210 void (*destroy_inode)(struct inode *);
212 void (*dirty_inode) (struct inode *, int flags);
213 int (*write_inode) (struct inode *, int);
214 void (*drop_inode) (struct inode *);
215 void (*delete_inode) (struct inode *);
216 void (*put_super) (struct super_block *);
217 int (*sync_fs)(struct super_block *sb, int wait);
218 int (*freeze_fs) (struct super_block *);
219 int (*unfreeze_fs) (struct super_block *);
220 int (*statfs) (struct dentry *, struct kstatfs *);
221 int (*remount_fs) (struct super_block *, int *, char *);
222 void (*clear_inode) (struct inode *);
223 void (*umount_begin) (struct super_block *);
225 int (*show_options)(struct seq_file *, struct dentry *);
227 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
228 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
229 int (*nr_cached_objects)(struct super_block *);
230 void (*free_cached_objects)(struct super_block *, int);
233 All methods are called without any locks being held, unless otherwise
234 noted. This means that most methods can block safely. All methods are
235 only called from a process context (i.e. not from an interrupt handler
238 alloc_inode: this method is called by alloc_inode() to allocate memory
239 for struct inode and initialize it. If this function is not
240 defined, a simple 'struct inode' is allocated. Normally
241 alloc_inode will be used to allocate a larger structure which
242 contains a 'struct inode' embedded within it.
244 destroy_inode: this method is called by destroy_inode() to release
245 resources allocated for struct inode. It is only required if
246 ->alloc_inode was defined and simply undoes anything done by
249 dirty_inode: this method is called by the VFS to mark an inode dirty.
251 write_inode: this method is called when the VFS needs to write an
252 inode to disc. The second parameter indicates whether the write
253 should be synchronous or not, not all filesystems check this flag.
255 drop_inode: called when the last access to the inode is dropped,
256 with the inode->i_lock spinlock held.
258 This method should be either NULL (normal UNIX filesystem
259 semantics) or "generic_delete_inode" (for filesystems that do not
260 want to cache inodes - causing "delete_inode" to always be
261 called regardless of the value of i_nlink)
263 The "generic_delete_inode()" behavior is equivalent to the
264 old practice of using "force_delete" in the put_inode() case,
265 but does not have the races that the "force_delete()" approach
268 delete_inode: called when the VFS wants to delete an inode
270 put_super: called when the VFS wishes to free the superblock
271 (i.e. unmount). This is called with the superblock lock held
273 sync_fs: called when VFS is writing out all dirty data associated with
274 a superblock. The second parameter indicates whether the method
275 should wait until the write out has been completed. Optional.
277 freeze_fs: called when VFS is locking a filesystem and
278 forcing it into a consistent state. This method is currently
279 used by the Logical Volume Manager (LVM).
281 unfreeze_fs: called when VFS is unlocking a filesystem and making it writable
284 statfs: called when the VFS needs to get filesystem statistics.
286 remount_fs: called when the filesystem is remounted. This is called
287 with the kernel lock held
289 clear_inode: called then the VFS clears the inode. Optional
291 umount_begin: called when the VFS is unmounting a filesystem.
293 show_options: called by the VFS to show mount options for
294 /proc/<pid>/mounts. (see "Mount Options" section)
296 quota_read: called by the VFS to read from filesystem quota file.
298 quota_write: called by the VFS to write to filesystem quota file.
300 nr_cached_objects: called by the sb cache shrinking function for the
301 filesystem to return the number of freeable cached objects it contains.
304 free_cache_objects: called by the sb cache shrinking function for the
305 filesystem to scan the number of objects indicated to try to free them.
306 Optional, but any filesystem implementing this method needs to also
307 implement ->nr_cached_objects for it to be called correctly.
309 We can't do anything with any errors that the filesystem might
310 encountered, hence the void return type. This will never be called if
311 the VM is trying to reclaim under GFP_NOFS conditions, hence this
312 method does not need to handle that situation itself.
314 Implementations must include conditional reschedule calls inside any
315 scanning loop that is done. This allows the VFS to determine
316 appropriate scan batch sizes without having to worry about whether
317 implementations will cause holdoff problems due to large scan batch
320 Whoever sets up the inode is responsible for filling in the "i_op" field. This
321 is a pointer to a "struct inode_operations" which describes the methods that
322 can be performed on individual inodes.
324 struct xattr_handlers
325 ---------------------
327 On filesystems that support extended attributes (xattrs), the s_xattr
328 superblock field points to a NULL-terminated array of xattr handlers. Extended
329 attributes are name:value pairs.
331 name: Indicates that the handler matches attributes with the specified name
332 (such as "system.posix_acl_access"); the prefix field must be NULL.
334 prefix: Indicates that the handler matches all attributes with the specified
335 name prefix (such as "user."); the name field must be NULL.
337 list: Determine if attributes matching this xattr handler should be listed
338 for a particular dentry. Used by some listxattr implementations like
341 get: Called by the VFS to get the value of a particular extended attribute.
342 This method is called by the getxattr(2) system call.
344 set: Called by the VFS to set the value of a particular extended attribute.
345 When the new value is NULL, called to remove a particular extended
346 attribute. This method is called by the the setxattr(2) and
347 removexattr(2) system calls.
349 When none of the xattr handlers of a filesystem match the specified attribute
350 name or when a filesystem doesn't support extended attributes, the various
351 *xattr(2) system calls return -EOPNOTSUPP.
357 An inode object represents an object within the filesystem.
360 struct inode_operations
361 -----------------------
363 This describes how the VFS can manipulate an inode in your
364 filesystem. As of kernel 2.6.22, the following members are defined:
366 struct inode_operations {
367 int (*create) (struct inode *,struct dentry *, umode_t, bool);
368 struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
369 int (*link) (struct dentry *,struct inode *,struct dentry *);
370 int (*unlink) (struct inode *,struct dentry *);
371 int (*symlink) (struct inode *,struct dentry *,const char *);
372 int (*mkdir) (struct inode *,struct dentry *,umode_t);
373 int (*rmdir) (struct inode *,struct dentry *);
374 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
375 int (*rename) (struct inode *, struct dentry *,
376 struct inode *, struct dentry *, unsigned int);
377 int (*readlink) (struct dentry *, char __user *,int);
378 const char *(*get_link) (struct dentry *, struct inode *,
379 struct delayed_call *);
380 int (*permission) (struct inode *, int);
381 int (*get_acl)(struct inode *, int);
382 int (*setattr) (struct dentry *, struct iattr *);
383 int (*getattr) (const struct path *, struct kstat *, u32, unsigned int);
384 ssize_t (*listxattr) (struct dentry *, char *, size_t);
385 void (*update_time)(struct inode *, struct timespec *, int);
386 int (*atomic_open)(struct inode *, struct dentry *, struct file *,
387 unsigned open_flag, umode_t create_mode);
388 int (*tmpfile) (struct inode *, struct dentry *, umode_t);
391 Again, all methods are called without any locks being held, unless
394 create: called by the open(2) and creat(2) system calls. Only
395 required if you want to support regular files. The dentry you
396 get should not have an inode (i.e. it should be a negative
397 dentry). Here you will probably call d_instantiate() with the
398 dentry and the newly created inode
400 lookup: called when the VFS needs to look up an inode in a parent
401 directory. The name to look for is found in the dentry. This
402 method must call d_add() to insert the found inode into the
403 dentry. The "i_count" field in the inode structure should be
404 incremented. If the named inode does not exist a NULL inode
405 should be inserted into the dentry (this is called a negative
406 dentry). Returning an error code from this routine must only
407 be done on a real error, otherwise creating inodes with system
408 calls like create(2), mknod(2), mkdir(2) and so on will fail.
409 If you wish to overload the dentry methods then you should
410 initialise the "d_dop" field in the dentry; this is a pointer
411 to a struct "dentry_operations".
412 This method is called with the directory inode semaphore held
414 link: called by the link(2) system call. Only required if you want
415 to support hard links. You will probably need to call
416 d_instantiate() just as you would in the create() method
418 unlink: called by the unlink(2) system call. Only required if you
419 want to support deleting inodes
421 symlink: called by the symlink(2) system call. Only required if you
422 want to support symlinks. You will probably need to call
423 d_instantiate() just as you would in the create() method
425 mkdir: called by the mkdir(2) system call. Only required if you want
426 to support creating subdirectories. You will probably need to
427 call d_instantiate() just as you would in the create() method
429 rmdir: called by the rmdir(2) system call. Only required if you want
430 to support deleting subdirectories
432 mknod: called by the mknod(2) system call to create a device (char,
433 block) inode or a named pipe (FIFO) or socket. Only required
434 if you want to support creating these types of inodes. You
435 will probably need to call d_instantiate() just as you would
436 in the create() method
438 rename: called by the rename(2) system call to rename the object to
439 have the parent and name given by the second inode and dentry.
441 The filesystem must return -EINVAL for any unsupported or
442 unknown flags. Currently the following flags are implemented:
443 (1) RENAME_NOREPLACE: this flag indicates that if the target
444 of the rename exists the rename should fail with -EEXIST
445 instead of replacing the target. The VFS already checks for
446 existence, so for local filesystems the RENAME_NOREPLACE
447 implementation is equivalent to plain rename.
448 (2) RENAME_EXCHANGE: exchange source and target. Both must
449 exist; this is checked by the VFS. Unlike plain rename,
450 source and target may be of different type.
452 get_link: called by the VFS to follow a symbolic link to the
453 inode it points to. Only required if you want to support
454 symbolic links. This method returns the symlink body
455 to traverse (and possibly resets the current position with
456 nd_jump_link()). If the body won't go away until the inode
457 is gone, nothing else is needed; if it needs to be otherwise
458 pinned, arrange for its release by having get_link(..., ..., done)
459 do set_delayed_call(done, destructor, argument).
460 In that case destructor(argument) will be called once VFS is
461 done with the body you've returned.
462 May be called in RCU mode; that is indicated by NULL dentry
463 argument. If request can't be handled without leaving RCU mode,
464 have it return ERR_PTR(-ECHILD).
466 If the filesystem stores the symlink target in ->i_link, the
467 VFS may use it directly without calling ->get_link(); however,
468 ->get_link() must still be provided. ->i_link must not be
469 freed until after an RCU grace period. Writing to ->i_link
470 post-iget() time requires a 'release' memory barrier.
472 readlink: this is now just an override for use by readlink(2) for the
473 cases when ->get_link uses nd_jump_link() or object is not in
474 fact a symlink. Normally filesystems should only implement
475 ->get_link for symlinks and readlink(2) will automatically use
478 permission: called by the VFS to check for access rights on a POSIX-like
481 May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in rcu-walk
482 mode, the filesystem must check the permission without blocking or
483 storing to the inode.
485 If a situation is encountered that rcu-walk cannot handle, return
486 -ECHILD and it will be called again in ref-walk mode.
488 setattr: called by the VFS to set attributes for a file. This method
489 is called by chmod(2) and related system calls.
491 getattr: called by the VFS to get attributes of a file. This method
492 is called by stat(2) and related system calls.
494 listxattr: called by the VFS to list all extended attributes for a
495 given file. This method is called by the listxattr(2) system call.
497 update_time: called by the VFS to update a specific time or the i_version of
498 an inode. If this is not defined the VFS will update the inode itself
499 and call mark_inode_dirty_sync.
501 atomic_open: called on the last component of an open. Using this optional
502 method the filesystem can look up, possibly create and open the file in
503 one atomic operation. If it wants to leave actual opening to the
504 caller (e.g. if the file turned out to be a symlink, device, or just
505 something filesystem won't do atomic open for), it may signal this by
506 returning finish_no_open(file, dentry). This method is only called if
507 the last component is negative or needs lookup. Cached positive dentries
508 are still handled by f_op->open(). If the file was created,
509 FMODE_CREATED flag should be set in file->f_mode. In case of O_EXCL
510 the method must only succeed if the file didn't exist and hence FMODE_CREATED
511 shall always be set on success.
513 tmpfile: called in the end of O_TMPFILE open(). Optional, equivalent to
514 atomically creating, opening and unlinking a file in given directory.
516 The Address Space Object
517 ========================
519 The address space object is used to group and manage pages in the page
520 cache. It can be used to keep track of the pages in a file (or
521 anything else) and also track the mapping of sections of the file into
522 process address spaces.
524 There are a number of distinct yet related services that an
525 address-space can provide. These include communicating memory
526 pressure, page lookup by address, and keeping track of pages tagged as
529 The first can be used independently to the others. The VM can try to
530 either write dirty pages in order to clean them, or release clean
531 pages in order to reuse them. To do this it can call the ->writepage
532 method on dirty pages, and ->releasepage on clean pages with
533 PagePrivate set. Clean pages without PagePrivate and with no external
534 references will be released without notice being given to the
537 To achieve this functionality, pages need to be placed on an LRU with
538 lru_cache_add and mark_page_active needs to be called whenever the
541 Pages are normally kept in a radix tree index by ->index. This tree
542 maintains information about the PG_Dirty and PG_Writeback status of
543 each page, so that pages with either of these flags can be found
546 The Dirty tag is primarily used by mpage_writepages - the default
547 ->writepages method. It uses the tag to find dirty pages to call
548 ->writepage on. If mpage_writepages is not used (i.e. the address
549 provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is
550 almost unused. write_inode_now and sync_inode do use it (through
551 __sync_single_inode) to check if ->writepages has been successful in
552 writing out the whole address_space.
554 The Writeback tag is used by filemap*wait* and sync_page* functions,
555 via filemap_fdatawait_range, to wait for all writeback to complete.
557 An address_space handler may attach extra information to a page,
558 typically using the 'private' field in the 'struct page'. If such
559 information is attached, the PG_Private flag should be set. This will
560 cause various VM routines to make extra calls into the address_space
561 handler to deal with that data.
563 An address space acts as an intermediate between storage and
564 application. Data is read into the address space a whole page at a
565 time, and provided to the application either by copying of the page,
566 or by memory-mapping the page.
567 Data is written into the address space by the application, and then
568 written-back to storage typically in whole pages, however the
569 address_space has finer control of write sizes.
571 The read process essentially only requires 'readpage'. The write
572 process is more complicated and uses write_begin/write_end or
573 set_page_dirty to write data into the address_space, and writepage
574 and writepages to writeback data to storage.
576 Adding and removing pages to/from an address_space is protected by the
579 When data is written to a page, the PG_Dirty flag should be set. It
580 typically remains set until writepage asks for it to be written. This
581 should clear PG_Dirty and set PG_Writeback. It can be actually
582 written at any point after PG_Dirty is clear. Once it is known to be
583 safe, PG_Writeback is cleared.
585 Writeback makes use of a writeback_control structure to direct the
586 operations. This gives the the writepage and writepages operations some
587 information about the nature of and reason for the writeback request,
588 and the constraints under which it is being done. It is also used to
589 return information back to the caller about the result of a writepage or
592 Handling errors during writeback
593 --------------------------------
594 Most applications that do buffered I/O will periodically call a file
595 synchronization call (fsync, fdatasync, msync or sync_file_range) to
596 ensure that data written has made it to the backing store. When there
597 is an error during writeback, they expect that error to be reported when
598 a file sync request is made. After an error has been reported on one
599 request, subsequent requests on the same file descriptor should return
600 0, unless further writeback errors have occurred since the previous file
603 Ideally, the kernel would report errors only on file descriptions on
604 which writes were done that subsequently failed to be written back. The
605 generic pagecache infrastructure does not track the file descriptions
606 that have dirtied each individual page however, so determining which
607 file descriptors should get back an error is not possible.
609 Instead, the generic writeback error tracking infrastructure in the
610 kernel settles for reporting errors to fsync on all file descriptions
611 that were open at the time that the error occurred. In a situation with
612 multiple writers, all of them will get back an error on a subsequent fsync,
613 even if all of the writes done through that particular file descriptor
614 succeeded (or even if there were no writes on that file descriptor at all).
616 Filesystems that wish to use this infrastructure should call
617 mapping_set_error to record the error in the address_space when it
618 occurs. Then, after writing back data from the pagecache in their
619 file->fsync operation, they should call file_check_and_advance_wb_err to
620 ensure that the struct file's error cursor has advanced to the correct
621 point in the stream of errors emitted by the backing device(s).
623 struct address_space_operations
624 -------------------------------
626 This describes how the VFS can manipulate mapping of a file to page cache in
627 your filesystem. The following members are defined:
629 struct address_space_operations {
630 int (*writepage)(struct page *page, struct writeback_control *wbc);
631 int (*readpage)(struct file *, struct page *);
632 int (*writepages)(struct address_space *, struct writeback_control *);
633 int (*set_page_dirty)(struct page *page);
634 int (*readpages)(struct file *filp, struct address_space *mapping,
635 struct list_head *pages, unsigned nr_pages);
636 int (*write_begin)(struct file *, struct address_space *mapping,
637 loff_t pos, unsigned len, unsigned flags,
638 struct page **pagep, void **fsdata);
639 int (*write_end)(struct file *, struct address_space *mapping,
640 loff_t pos, unsigned len, unsigned copied,
641 struct page *page, void *fsdata);
642 sector_t (*bmap)(struct address_space *, sector_t);
643 void (*invalidatepage) (struct page *, unsigned int, unsigned int);
644 int (*releasepage) (struct page *, int);
645 void (*freepage)(struct page *);
646 ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter);
647 /* isolate a page for migration */
648 bool (*isolate_page) (struct page *, isolate_mode_t);
649 /* migrate the contents of a page to the specified target */
650 int (*migratepage) (struct page *, struct page *);
651 /* put migration-failed page back to right list */
652 void (*putback_page) (struct page *);
653 int (*launder_page) (struct page *);
655 int (*is_partially_uptodate) (struct page *, unsigned long,
657 void (*is_dirty_writeback) (struct page *, bool *, bool *);
658 int (*error_remove_page) (struct mapping *mapping, struct page *page);
659 int (*swap_activate)(struct file *);
660 int (*swap_deactivate)(struct file *);
663 writepage: called by the VM to write a dirty page to backing store.
664 This may happen for data integrity reasons (i.e. 'sync'), or
665 to free up memory (flush). The difference can be seen in
667 The PG_Dirty flag has been cleared and PageLocked is true.
668 writepage should start writeout, should set PG_Writeback,
669 and should make sure the page is unlocked, either synchronously
670 or asynchronously when the write operation completes.
672 If wbc->sync_mode is WB_SYNC_NONE, ->writepage doesn't have to
673 try too hard if there are problems, and may choose to write out
674 other pages from the mapping if that is easier (e.g. due to
675 internal dependencies). If it chooses not to start writeout, it
676 should return AOP_WRITEPAGE_ACTIVATE so that the VM will not keep
677 calling ->writepage on that page.
679 See the file "Locking" for more details.
681 readpage: called by the VM to read a page from backing store.
682 The page will be Locked when readpage is called, and should be
683 unlocked and marked uptodate once the read completes.
684 If ->readpage discovers that it needs to unlock the page for
685 some reason, it can do so, and then return AOP_TRUNCATED_PAGE.
686 In this case, the page will be relocated, relocked and if
687 that all succeeds, ->readpage will be called again.
689 writepages: called by the VM to write out pages associated with the
690 address_space object. If wbc->sync_mode is WBC_SYNC_ALL, then
691 the writeback_control will specify a range of pages that must be
692 written out. If it is WBC_SYNC_NONE, then a nr_to_write is given
693 and that many pages should be written if possible.
694 If no ->writepages is given, then mpage_writepages is used
695 instead. This will choose pages from the address space that are
696 tagged as DIRTY and will pass them to ->writepage.
698 set_page_dirty: called by the VM to set a page dirty.
699 This is particularly needed if an address space attaches
700 private data to a page, and that data needs to be updated when
701 a page is dirtied. This is called, for example, when a memory
702 mapped page gets modified.
703 If defined, it should set the PageDirty flag, and the
704 PAGECACHE_TAG_DIRTY tag in the radix tree.
706 readpages: called by the VM to read pages associated with the address_space
707 object. This is essentially just a vector version of
708 readpage. Instead of just one page, several pages are
710 readpages is only used for read-ahead, so read errors are
711 ignored. If anything goes wrong, feel free to give up.
714 Called by the generic buffered write code to ask the filesystem to
715 prepare to write len bytes at the given offset in the file. The
716 address_space should check that the write will be able to complete,
717 by allocating space if necessary and doing any other internal
718 housekeeping. If the write will update parts of any basic-blocks on
719 storage, then those blocks should be pre-read (if they haven't been
720 read already) so that the updated blocks can be written out properly.
722 The filesystem must return the locked pagecache page for the specified
723 offset, in *pagep, for the caller to write into.
725 It must be able to cope with short writes (where the length passed to
726 write_begin is greater than the number of bytes copied into the page).
728 flags is a field for AOP_FLAG_xxx flags, described in
731 A void * may be returned in fsdata, which then gets passed into
734 Returns 0 on success; < 0 on failure (which is the error code), in
735 which case write_end is not called.
737 write_end: After a successful write_begin, and data copy, write_end must
738 be called. len is the original len passed to write_begin, and copied
739 is the amount that was able to be copied.
741 The filesystem must take care of unlocking the page and releasing it
742 refcount, and updating i_size.
744 Returns < 0 on failure, otherwise the number of bytes (<= 'copied')
745 that were able to be copied into pagecache.
747 bmap: called by the VFS to map a logical block offset within object to
748 physical block number. This method is used by the FIBMAP
749 ioctl and for working with swap-files. To be able to swap to
750 a file, the file must have a stable mapping to a block
751 device. The swap system does not go through the filesystem
752 but instead uses bmap to find out where the blocks in the file
753 are and uses those addresses directly.
755 invalidatepage: If a page has PagePrivate set, then invalidatepage
756 will be called when part or all of the page is to be removed
757 from the address space. This generally corresponds to either a
758 truncation, punch hole or a complete invalidation of the address
759 space (in the latter case 'offset' will always be 0 and 'length'
760 will be PAGE_SIZE). Any private data associated with the page
761 should be updated to reflect this truncation. If offset is 0 and
762 length is PAGE_SIZE, then the private data should be released,
763 because the page must be able to be completely discarded. This may
764 be done by calling the ->releasepage function, but in this case the
765 release MUST succeed.
767 releasepage: releasepage is called on PagePrivate pages to indicate
768 that the page should be freed if possible. ->releasepage
769 should remove any private data from the page and clear the
770 PagePrivate flag. If releasepage() fails for some reason, it must
771 indicate failure with a 0 return value.
772 releasepage() is used in two distinct though related cases. The
773 first is when the VM finds a clean page with no active users and
774 wants to make it a free page. If ->releasepage succeeds, the
775 page will be removed from the address_space and become free.
777 The second case is when a request has been made to invalidate
778 some or all pages in an address_space. This can happen
779 through the fadvise(POSIX_FADV_DONTNEED) system call or by the
780 filesystem explicitly requesting it as nfs and 9fs do (when
781 they believe the cache may be out of date with storage) by
782 calling invalidate_inode_pages2().
783 If the filesystem makes such a call, and needs to be certain
784 that all pages are invalidated, then its releasepage will
785 need to ensure this. Possibly it can clear the PageUptodate
786 bit if it cannot free private data yet.
788 freepage: freepage is called once the page is no longer visible in
789 the page cache in order to allow the cleanup of any private
790 data. Since it may be called by the memory reclaimer, it
791 should not assume that the original address_space mapping still
792 exists, and it should not block.
794 direct_IO: called by the generic read/write routines to perform
795 direct_IO - that is IO requests which bypass the page cache
796 and transfer data directly between the storage and the
797 application's address space.
799 isolate_page: Called by the VM when isolating a movable non-lru page.
800 If page is successfully isolated, VM marks the page as PG_isolated
801 via __SetPageIsolated.
803 migrate_page: This is used to compact the physical memory usage.
804 If the VM wants to relocate a page (maybe off a memory card
805 that is signalling imminent failure) it will pass a new page
806 and an old page to this function. migrate_page should
807 transfer any private data across and update any references
808 that it has to the page.
810 putback_page: Called by the VM when isolated page's migration fails.
812 launder_page: Called before freeing a page - it writes back the dirty page. To
813 prevent redirtying the page, it is kept locked during the whole
816 is_partially_uptodate: Called by the VM when reading a file through the
817 pagecache when the underlying blocksize != pagesize. If the required
818 block is up to date then the read can complete without needing the IO
819 to bring the whole page up to date.
821 is_dirty_writeback: Called by the VM when attempting to reclaim a page.
822 The VM uses dirty and writeback information to determine if it needs
823 to stall to allow flushers a chance to complete some IO. Ordinarily
824 it can use PageDirty and PageWriteback but some filesystems have
825 more complex state (unstable pages in NFS prevent reclaim) or
826 do not set those flags due to locking problems. This callback
827 allows a filesystem to indicate to the VM if a page should be
828 treated as dirty or writeback for the purposes of stalling.
830 error_remove_page: normally set to generic_error_remove_page if truncation
831 is ok for this address space. Used for memory failure handling.
832 Setting this implies you deal with pages going away under you,
833 unless you have them locked or reference counts increased.
835 swap_activate: Called when swapon is used on a file to allocate
836 space if necessary and pin the block lookup information in
837 memory. A return value of zero indicates success,
838 in which case this file can be used to back swapspace.
840 swap_deactivate: Called during swapoff on files where swap_activate
847 A file object represents a file opened by a process. This is also known
848 as an "open file description" in POSIX parlance.
851 struct file_operations
852 ----------------------
854 This describes how the VFS can manipulate an open file. As of kernel
855 4.18, the following members are defined:
857 struct file_operations {
858 struct module *owner;
859 loff_t (*llseek) (struct file *, loff_t, int);
860 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
861 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
862 ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
863 ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
864 int (*iopoll)(struct kiocb *kiocb, bool spin);
865 int (*iterate) (struct file *, struct dir_context *);
866 int (*iterate_shared) (struct file *, struct dir_context *);
867 __poll_t (*poll) (struct file *, struct poll_table_struct *);
868 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
869 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
870 int (*mmap) (struct file *, struct vm_area_struct *);
871 int (*open) (struct inode *, struct file *);
872 int (*flush) (struct file *, fl_owner_t id);
873 int (*release) (struct inode *, struct file *);
874 int (*fsync) (struct file *, loff_t, loff_t, int datasync);
875 int (*fasync) (int, struct file *, int);
876 int (*lock) (struct file *, int, struct file_lock *);
877 ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
878 unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
879 int (*check_flags)(int);
880 int (*flock) (struct file *, int, struct file_lock *);
881 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
882 ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
883 int (*setlease)(struct file *, long, struct file_lock **, void **);
884 long (*fallocate)(struct file *file, int mode, loff_t offset,
886 void (*show_fdinfo)(struct seq_file *m, struct file *f);
888 unsigned (*mmap_capabilities)(struct file *);
890 ssize_t (*copy_file_range)(struct file *, loff_t, struct file *, loff_t, size_t, unsigned int);
891 loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
892 struct file *file_out, loff_t pos_out,
893 loff_t len, unsigned int remap_flags);
894 int (*fadvise)(struct file *, loff_t, loff_t, int);
897 Again, all methods are called without any locks being held, unless
900 llseek: called when the VFS needs to move the file position index
902 read: called by read(2) and related system calls
904 read_iter: possibly asynchronous read with iov_iter as destination
906 write: called by write(2) and related system calls
908 write_iter: possibly asynchronous write with iov_iter as source
910 iopoll: called when aio wants to poll for completions on HIPRI iocbs
912 iterate: called when the VFS needs to read the directory contents
914 iterate_shared: called when the VFS needs to read the directory contents
915 when filesystem supports concurrent dir iterators
917 poll: called by the VFS when a process wants to check if there is
918 activity on this file and (optionally) go to sleep until there
919 is activity. Called by the select(2) and poll(2) system calls
921 unlocked_ioctl: called by the ioctl(2) system call.
923 compat_ioctl: called by the ioctl(2) system call when 32 bit system calls
924 are used on 64 bit kernels.
926 mmap: called by the mmap(2) system call
928 open: called by the VFS when an inode should be opened. When the VFS
929 opens a file, it creates a new "struct file". It then calls the
930 open method for the newly allocated file structure. You might
931 think that the open method really belongs in
932 "struct inode_operations", and you may be right. I think it's
933 done the way it is because it makes filesystems simpler to
934 implement. The open() method is a good place to initialize the
935 "private_data" member in the file structure if you want to point
936 to a device structure
938 flush: called by the close(2) system call to flush a file
940 release: called when the last reference to an open file is closed
942 fsync: called by the fsync(2) system call. Also see the section above
943 entitled "Handling errors during writeback".
945 fasync: called by the fcntl(2) system call when asynchronous
946 (non-blocking) mode is enabled for a file
948 lock: called by the fcntl(2) system call for F_GETLK, F_SETLK, and F_SETLKW
951 get_unmapped_area: called by the mmap(2) system call
953 check_flags: called by the fcntl(2) system call for F_SETFL command
955 flock: called by the flock(2) system call
957 splice_write: called by the VFS to splice data from a pipe to a file. This
958 method is used by the splice(2) system call
960 splice_read: called by the VFS to splice data from file to a pipe. This
961 method is used by the splice(2) system call
963 setlease: called by the VFS to set or release a file lock lease. setlease
964 implementations should call generic_setlease to record or remove
965 the lease in the inode after setting it.
967 fallocate: called by the VFS to preallocate blocks or punch a hole.
969 copy_file_range: called by the copy_file_range(2) system call.
971 remap_file_range: called by the ioctl(2) system call for FICLONERANGE and
972 FICLONE and FIDEDUPERANGE commands to remap file ranges. An
973 implementation should remap len bytes at pos_in of the source file into
974 the dest file at pos_out. Implementations must handle callers passing
975 in len == 0; this means "remap to the end of the source file". The
976 return value should the number of bytes remapped, or the usual
977 negative error code if errors occurred before any bytes were remapped.
978 The remap_flags parameter accepts REMAP_FILE_* flags. If
979 REMAP_FILE_DEDUP is set then the implementation must only remap if the
980 requested file ranges have identical contents. If REMAP_CAN_SHORTEN is
981 set, the caller is ok with the implementation shortening the request
982 length to satisfy alignment or EOF requirements (or any other reason).
984 fadvise: possibly called by the fadvise64() system call.
986 Note that the file operations are implemented by the specific
987 filesystem in which the inode resides. When opening a device node
988 (character or block special) most filesystems will call special
989 support routines in the VFS which will locate the required device
990 driver information. These support routines replace the filesystem file
991 operations with those for the device driver, and then proceed to call
992 the new open() method for the file. This is how opening a device file
993 in the filesystem eventually ends up calling the device driver open()
997 Directory Entry Cache (dcache)
998 ==============================
1001 struct dentry_operations
1002 ------------------------
1004 This describes how a filesystem can overload the standard dentry
1005 operations. Dentries and the dcache are the domain of the VFS and the
1006 individual filesystem implementations. Device drivers have no business
1007 here. These methods may be set to NULL, as they are either optional or
1008 the VFS uses a default. As of kernel 2.6.22, the following members are
1011 struct dentry_operations {
1012 int (*d_revalidate)(struct dentry *, unsigned int);
1013 int (*d_weak_revalidate)(struct dentry *, unsigned int);
1014 int (*d_hash)(const struct dentry *, struct qstr *);
1015 int (*d_compare)(const struct dentry *,
1016 unsigned int, const char *, const struct qstr *);
1017 int (*d_delete)(const struct dentry *);
1018 int (*d_init)(struct dentry *);
1019 void (*d_release)(struct dentry *);
1020 void (*d_iput)(struct dentry *, struct inode *);
1021 char *(*d_dname)(struct dentry *, char *, int);
1022 struct vfsmount *(*d_automount)(struct path *);
1023 int (*d_manage)(const struct path *, bool);
1024 struct dentry *(*d_real)(struct dentry *, const struct inode *);
1027 d_revalidate: called when the VFS needs to revalidate a dentry. This
1028 is called whenever a name look-up finds a dentry in the
1029 dcache. Most local filesystems leave this as NULL, because all their
1030 dentries in the dcache are valid. Network filesystems are different
1031 since things can change on the server without the client necessarily
1034 This function should return a positive value if the dentry is still
1035 valid, and zero or a negative error code if it isn't.
1037 d_revalidate may be called in rcu-walk mode (flags & LOOKUP_RCU).
1038 If in rcu-walk mode, the filesystem must revalidate the dentry without
1039 blocking or storing to the dentry, d_parent and d_inode should not be
1040 used without care (because they can change and, in d_inode case, even
1041 become NULL under us).
1043 If a situation is encountered that rcu-walk cannot handle, return
1044 -ECHILD and it will be called again in ref-walk mode.
1046 d_weak_revalidate: called when the VFS needs to revalidate a "jumped" dentry.
1047 This is called when a path-walk ends at dentry that was not acquired by
1048 doing a lookup in the parent directory. This includes "/", "." and "..",
1049 as well as procfs-style symlinks and mountpoint traversal.
1051 In this case, we are less concerned with whether the dentry is still
1052 fully correct, but rather that the inode is still valid. As with
1053 d_revalidate, most local filesystems will set this to NULL since their
1054 dcache entries are always valid.
1056 This function has the same return code semantics as d_revalidate.
1058 d_weak_revalidate is only called after leaving rcu-walk mode.
1060 d_hash: called when the VFS adds a dentry to the hash table. The first
1061 dentry passed to d_hash is the parent directory that the name is
1064 Same locking and synchronisation rules as d_compare regarding
1065 what is safe to dereference etc.
1067 d_compare: called to compare a dentry name with a given name. The first
1068 dentry is the parent of the dentry to be compared, the second is
1069 the child dentry. len and name string are properties of the dentry
1070 to be compared. qstr is the name to compare it with.
1072 Must be constant and idempotent, and should not take locks if
1073 possible, and should not or store into the dentry.
1074 Should not dereference pointers outside the dentry without
1075 lots of care (eg. d_parent, d_inode, d_name should not be used).
1077 However, our vfsmount is pinned, and RCU held, so the dentries and
1078 inodes won't disappear, neither will our sb or filesystem module.
1081 It is a tricky calling convention because it needs to be called under
1082 "rcu-walk", ie. without any locks or references on things.
1084 d_delete: called when the last reference to a dentry is dropped and the
1085 dcache is deciding whether or not to cache it. Return 1 to delete
1086 immediately, or 0 to cache the dentry. Default is NULL which means to
1087 always cache a reachable dentry. d_delete must be constant and
1090 d_init: called when a dentry is allocated
1092 d_release: called when a dentry is really deallocated
1094 d_iput: called when a dentry loses its inode (just prior to its
1095 being deallocated). The default when this is NULL is that the
1096 VFS calls iput(). If you define this method, you must call
1099 d_dname: called when the pathname of a dentry should be generated.
1100 Useful for some pseudo filesystems (sockfs, pipefs, ...) to delay
1101 pathname generation. (Instead of doing it when dentry is created,
1102 it's done only when the path is needed.). Real filesystems probably
1103 dont want to use it, because their dentries are present in global
1104 dcache hash, so their hash should be an invariant. As no lock is
1105 held, d_dname() should not try to modify the dentry itself, unless
1106 appropriate SMP safety is used. CAUTION : d_path() logic is quite
1107 tricky. The correct way to return for example "Hello" is to put it
1108 at the end of the buffer, and returns a pointer to the first char.
1109 dynamic_dname() helper function is provided to take care of this.
1113 static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1115 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1116 dentry->d_inode->i_ino);
1119 d_automount: called when an automount dentry is to be traversed (optional).
1120 This should create a new VFS mount record and return the record to the
1121 caller. The caller is supplied with a path parameter giving the
1122 automount directory to describe the automount target and the parent
1123 VFS mount record to provide inheritable mount parameters. NULL should
1124 be returned if someone else managed to make the automount first. If
1125 the vfsmount creation failed, then an error code should be returned.
1126 If -EISDIR is returned, then the directory will be treated as an
1127 ordinary directory and returned to pathwalk to continue walking.
1129 If a vfsmount is returned, the caller will attempt to mount it on the
1130 mountpoint and will remove the vfsmount from its expiration list in
1131 the case of failure. The vfsmount should be returned with 2 refs on
1132 it to prevent automatic expiration - the caller will clean up the
1135 This function is only used if DCACHE_NEED_AUTOMOUNT is set on the
1136 dentry. This is set by __d_instantiate() if S_AUTOMOUNT is set on the
1139 d_manage: called to allow the filesystem to manage the transition from a
1140 dentry (optional). This allows autofs, for example, to hold up clients
1141 waiting to explore behind a 'mountpoint' while letting the daemon go
1142 past and construct the subtree there. 0 should be returned to let the
1143 calling process continue. -EISDIR can be returned to tell pathwalk to
1144 use this directory as an ordinary directory and to ignore anything
1145 mounted on it and not to check the automount flag. Any other error
1146 code will abort pathwalk completely.
1148 If the 'rcu_walk' parameter is true, then the caller is doing a
1149 pathwalk in RCU-walk mode. Sleeping is not permitted in this mode,
1150 and the caller can be asked to leave it and call again by returning
1151 -ECHILD. -EISDIR may also be returned to tell pathwalk to
1152 ignore d_automount or any mounts.
1154 This function is only used if DCACHE_MANAGE_TRANSIT is set on the
1155 dentry being transited from.
1157 d_real: overlay/union type filesystems implement this method to return one of
1158 the underlying dentries hidden by the overlay. It is used in two
1161 Called from file_dentry() it returns the real dentry matching the inode
1162 argument. The real dentry may be from a lower layer already copied up,
1163 but still referenced from the file. This mode is selected with a
1164 non-NULL inode argument.
1166 With NULL inode the topmost real underlying dentry is returned.
1168 Each dentry has a pointer to its parent dentry, as well as a hash list
1169 of child dentries. Child dentries are basically like files in a
1173 Directory Entry Cache API
1174 --------------------------
1176 There are a number of functions defined which permit a filesystem to
1177 manipulate dentries:
1179 dget: open a new handle for an existing dentry (this just increments
1182 dput: close a handle for a dentry (decrements the usage count). If
1183 the usage count drops to 0, and the dentry is still in its
1184 parent's hash, the "d_delete" method is called to check whether
1185 it should be cached. If it should not be cached, or if the dentry
1186 is not hashed, it is deleted. Otherwise cached dentries are put
1187 into an LRU list to be reclaimed on memory shortage.
1189 d_drop: this unhashes a dentry from its parents hash list. A
1190 subsequent call to dput() will deallocate the dentry if its
1191 usage count drops to 0
1193 d_delete: delete a dentry. If there are no other open references to
1194 the dentry then the dentry is turned into a negative dentry
1195 (the d_iput() method is called). If there are other
1196 references, then d_drop() is called instead
1198 d_add: add a dentry to its parents hash list and then calls
1201 d_instantiate: add a dentry to the alias hash list for the inode and
1202 updates the "d_inode" member. The "i_count" member in the
1203 inode structure should be set/incremented. If the inode
1204 pointer is NULL, the dentry is called a "negative
1205 dentry". This function is commonly called when an inode is
1206 created for an existing negative dentry
1208 d_lookup: look up a dentry given its parent and path name component
1209 It looks up the child of that given name from the dcache
1210 hash table. If it is found, the reference count is incremented
1211 and the dentry is returned. The caller must use dput()
1212 to free the dentry when it finishes using it.
1220 On mount and remount the filesystem is passed a string containing a
1221 comma separated list of mount options. The options can have either of
1227 The <linux/parser.h> header defines an API that helps parse these
1228 options. There are plenty of examples on how to use it in existing
1234 If a filesystem accepts mount options, it must define show_options()
1235 to show all the currently active options. The rules are:
1237 - options MUST be shown which are not default or their values differ
1240 - options MAY be shown which are enabled by default or have their
1243 Options used only internally between a mount helper and the kernel
1244 (such as file descriptors), or which only have an effect during the
1245 mounting (such as ones controlling the creation of a journal) are exempt
1246 from the above rules.
1248 The underlying reason for the above rules is to make sure, that a
1249 mount can be accurately replicated (e.g. umounting and mounting again)
1250 based on the information found in /proc/mounts.
1255 (Note some of these resources are not up-to-date with the latest kernel
1258 Creating Linux virtual filesystems. 2002
1259 <http://lwn.net/Articles/13325/>
1261 The Linux Virtual File-system Layer by Neil Brown. 1999
1262 <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1264 A tour of the Linux VFS by Michael K. Johnson. 1996
1265 <http://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1267 A small trail through the Linux kernel by Andries Brouwer. 2001
1268 <http://www.win.tue.nl/~aeb/linux/vfs/trail.html>