2 Overview of the Linux Virtual File System
4 Original author: Richard Gooch <rgooch@atnf.csiro.au>
6 Last updated on June 24, 2007.
8 Copyright (C) 1999 Richard Gooch
9 Copyright (C) 2005 Pekka Enberg
11 This file is released under the GPLv2.
17 The Virtual File System (also known as the Virtual Filesystem Switch)
18 is the software layer in the kernel that provides the filesystem
19 interface to userspace programs. It also provides an abstraction
20 within the kernel which allows different filesystem implementations to
23 VFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so
24 on are called from a process context. Filesystem locking is described
25 in the document Documentation/filesystems/Locking.
28 Directory Entry Cache (dcache)
29 ------------------------------
31 The VFS implements the open(2), stat(2), chmod(2), and similar system
32 calls. The pathname argument that is passed to them is used by the VFS
33 to search through the directory entry cache (also known as the dentry
34 cache or dcache). This provides a very fast look-up mechanism to
35 translate a pathname (filename) into a specific dentry. Dentries live
36 in RAM and are never saved to disc: they exist only for performance.
38 The dentry cache is meant to be a view into your entire filespace. As
39 most computers cannot fit all dentries in the RAM at the same time,
40 some bits of the cache are missing. In order to resolve your pathname
41 into a dentry, the VFS may have to resort to creating dentries along
42 the way, and then loading the inode. This is done by looking up the
49 An individual dentry usually has a pointer to an inode. Inodes are
50 filesystem objects such as regular files, directories, FIFOs and other
51 beasts. They live either on the disc (for block device filesystems)
52 or in the memory (for pseudo filesystems). Inodes that live on the
53 disc are copied into the memory when required and changes to the inode
54 are written back to disc. A single inode can be pointed to by multiple
55 dentries (hard links, for example, do this).
57 To look up an inode requires that the VFS calls the lookup() method of
58 the parent directory inode. This method is installed by the specific
59 filesystem implementation that the inode lives in. Once the VFS has
60 the required dentry (and hence the inode), we can do all those boring
61 things like open(2) the file, or stat(2) it to peek at the inode
62 data. The stat(2) operation is fairly simple: once the VFS has the
63 dentry, it peeks at the inode data and passes some of it back to
70 Opening a file requires another operation: allocation of a file
71 structure (this is the kernel-side implementation of file
72 descriptors). The freshly allocated file structure is initialized with
73 a pointer to the dentry and a set of file operation member functions.
74 These are taken from the inode data. The open() file method is then
75 called so the specific filesystem implementation can do its work. You
76 can see that this is another switch performed by the VFS. The file
77 structure is placed into the file descriptor table for the process.
79 Reading, writing and closing files (and other assorted VFS operations)
80 is done by using the userspace file descriptor to grab the appropriate
81 file structure, and then calling the required file structure method to
82 do whatever is required. For as long as the file is open, it keeps the
83 dentry in use, which in turn means that the VFS inode is still in use.
86 Registering and Mounting a Filesystem
87 =====================================
89 To register and unregister a filesystem, use the following API
94 extern int register_filesystem(struct file_system_type *);
95 extern int unregister_filesystem(struct file_system_type *);
97 The passed struct file_system_type describes your filesystem. When a
98 request is made to mount a filesystem onto a directory in your namespace,
99 the VFS will call the appropriate mount() method for the specific
100 filesystem. New vfsmount referring to the tree returned by ->mount()
101 will be attached to the mountpoint, so that when pathname resolution
102 reaches the mountpoint it will jump into the root of that vfsmount.
104 You can see all filesystems that are registered to the kernel in the
105 file /proc/filesystems.
108 struct file_system_type
109 -----------------------
111 This describes the filesystem. As of kernel 2.6.39, the following
114 struct file_system_type {
117 struct dentry *(*mount) (struct file_system_type *, int,
118 const char *, void *);
119 void (*kill_sb) (struct super_block *);
120 struct module *owner;
121 struct file_system_type * next;
122 struct list_head fs_supers;
123 struct lock_class_key s_lock_key;
124 struct lock_class_key s_umount_key;
127 name: the name of the filesystem type, such as "ext2", "iso9660",
130 fs_flags: various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
132 mount: the method to call when a new instance of this
133 filesystem should be mounted
135 kill_sb: the method to call when an instance of this filesystem
138 owner: for internal VFS use: you should initialize this to THIS_MODULE in
141 next: for internal VFS use: you should initialize this to NULL
143 s_lock_key, s_umount_key: lockdep-specific
145 The mount() method has the following arguments:
147 struct file_system_type *fs_type: describes the filesystem, partly initialized
148 by the specific filesystem code
150 int flags: mount flags
152 const char *dev_name: the device name we are mounting.
154 void *data: arbitrary mount options, usually comes as an ASCII
155 string (see "Mount Options" section)
157 The mount() method must return the root dentry of the tree requested by
158 caller. An active reference to its superblock must be grabbed and the
159 superblock must be locked. On failure it should return ERR_PTR(error).
161 The arguments match those of mount(2) and their interpretation
162 depends on filesystem type. E.g. for block filesystems, dev_name is
163 interpreted as block device name, that device is opened and if it
164 contains a suitable filesystem image the method creates and initializes
165 struct super_block accordingly, returning its root dentry to caller.
167 ->mount() may choose to return a subtree of existing filesystem - it
168 doesn't have to create a new one. The main result from the caller's
169 point of view is a reference to dentry at the root of (sub)tree to
170 be attached; creation of new superblock is a common side effect.
172 The most interesting member of the superblock structure that the
173 mount() method fills in is the "s_op" field. This is a pointer to
174 a "struct super_operations" which describes the next level of the
175 filesystem implementation.
177 Usually, a filesystem uses one of the generic mount() implementations
178 and provides a fill_super() callback instead. The generic variants are:
180 mount_bdev: mount a filesystem residing on a block device
182 mount_nodev: mount a filesystem that is not backed by a device
184 mount_single: mount a filesystem which shares the instance between
187 A fill_super() callback implementation has the following arguments:
189 struct super_block *sb: the superblock structure. The callback
190 must initialize this properly.
192 void *data: arbitrary mount options, usually comes as an ASCII
193 string (see "Mount Options" section)
195 int silent: whether or not to be silent on error
198 The Superblock Object
199 =====================
201 A superblock object represents a mounted filesystem.
204 struct super_operations
205 -----------------------
207 This describes how the VFS can manipulate the superblock of your
208 filesystem. As of kernel 2.6.22, the following members are defined:
210 struct super_operations {
211 struct inode *(*alloc_inode)(struct super_block *sb);
212 void (*destroy_inode)(struct inode *);
214 void (*dirty_inode) (struct inode *, int flags);
215 int (*write_inode) (struct inode *, int);
216 void (*drop_inode) (struct inode *);
217 void (*delete_inode) (struct inode *);
218 void (*put_super) (struct super_block *);
219 void (*write_super) (struct super_block *);
220 int (*sync_fs)(struct super_block *sb, int wait);
221 int (*freeze_fs) (struct super_block *);
222 int (*unfreeze_fs) (struct super_block *);
223 int (*statfs) (struct dentry *, struct kstatfs *);
224 int (*remount_fs) (struct super_block *, int *, char *);
225 void (*clear_inode) (struct inode *);
226 void (*umount_begin) (struct super_block *);
228 int (*show_options)(struct seq_file *, struct dentry *);
230 ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
231 ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
232 int (*nr_cached_objects)(struct super_block *);
233 void (*free_cached_objects)(struct super_block *, int);
236 All methods are called without any locks being held, unless otherwise
237 noted. This means that most methods can block safely. All methods are
238 only called from a process context (i.e. not from an interrupt handler
241 alloc_inode: this method is called by inode_alloc() to allocate memory
242 for struct inode and initialize it. If this function is not
243 defined, a simple 'struct inode' is allocated. Normally
244 alloc_inode will be used to allocate a larger structure which
245 contains a 'struct inode' embedded within it.
247 destroy_inode: this method is called by destroy_inode() to release
248 resources allocated for struct inode. It is only required if
249 ->alloc_inode was defined and simply undoes anything done by
252 dirty_inode: this method is called by the VFS to mark an inode dirty.
254 write_inode: this method is called when the VFS needs to write an
255 inode to disc. The second parameter indicates whether the write
256 should be synchronous or not, not all filesystems check this flag.
258 drop_inode: called when the last access to the inode is dropped,
259 with the inode->i_lock spinlock held.
261 This method should be either NULL (normal UNIX filesystem
262 semantics) or "generic_delete_inode" (for filesystems that do not
263 want to cache inodes - causing "delete_inode" to always be
264 called regardless of the value of i_nlink)
266 The "generic_delete_inode()" behavior is equivalent to the
267 old practice of using "force_delete" in the put_inode() case,
268 but does not have the races that the "force_delete()" approach
271 delete_inode: called when the VFS wants to delete an inode
273 put_super: called when the VFS wishes to free the superblock
274 (i.e. unmount). This is called with the superblock lock held
276 write_super: called when the VFS superblock needs to be written to
277 disc. This method is optional
279 sync_fs: called when VFS is writing out all dirty data associated with
280 a superblock. The second parameter indicates whether the method
281 should wait until the write out has been completed. Optional.
283 freeze_fs: called when VFS is locking a filesystem and
284 forcing it into a consistent state. This method is currently
285 used by the Logical Volume Manager (LVM).
287 unfreeze_fs: called when VFS is unlocking a filesystem and making it writable
290 statfs: called when the VFS needs to get filesystem statistics.
292 remount_fs: called when the filesystem is remounted. This is called
293 with the kernel lock held
295 clear_inode: called then the VFS clears the inode. Optional
297 umount_begin: called when the VFS is unmounting a filesystem.
299 show_options: called by the VFS to show mount options for
300 /proc/<pid>/mounts. (see "Mount Options" section)
302 quota_read: called by the VFS to read from filesystem quota file.
304 quota_write: called by the VFS to write to filesystem quota file.
306 nr_cached_objects: called by the sb cache shrinking function for the
307 filesystem to return the number of freeable cached objects it contains.
310 free_cache_objects: called by the sb cache shrinking function for the
311 filesystem to scan the number of objects indicated to try to free them.
312 Optional, but any filesystem implementing this method needs to also
313 implement ->nr_cached_objects for it to be called correctly.
315 We can't do anything with any errors that the filesystem might
316 encountered, hence the void return type. This will never be called if
317 the VM is trying to reclaim under GFP_NOFS conditions, hence this
318 method does not need to handle that situation itself.
320 Implementations must include conditional reschedule calls inside any
321 scanning loop that is done. This allows the VFS to determine
322 appropriate scan batch sizes without having to worry about whether
323 implementations will cause holdoff problems due to large scan batch
326 Whoever sets up the inode is responsible for filling in the "i_op" field. This
327 is a pointer to a "struct inode_operations" which describes the methods that
328 can be performed on individual inodes.
334 An inode object represents an object within the filesystem.
337 struct inode_operations
338 -----------------------
340 This describes how the VFS can manipulate an inode in your
341 filesystem. As of kernel 2.6.22, the following members are defined:
343 struct inode_operations {
344 int (*create) (struct inode *,struct dentry *, umode_t, struct nameidata *);
345 struct dentry * (*lookup) (struct inode *,struct dentry *, struct nameidata *);
346 int (*link) (struct dentry *,struct inode *,struct dentry *);
347 int (*unlink) (struct inode *,struct dentry *);
348 int (*symlink) (struct inode *,struct dentry *,const char *);
349 int (*mkdir) (struct inode *,struct dentry *,umode_t);
350 int (*rmdir) (struct inode *,struct dentry *);
351 int (*mknod) (struct inode *,struct dentry *,umode_t,dev_t);
352 int (*rename) (struct inode *, struct dentry *,
353 struct inode *, struct dentry *);
354 int (*readlink) (struct dentry *, char __user *,int);
355 void * (*follow_link) (struct dentry *, struct nameidata *);
356 void (*put_link) (struct dentry *, struct nameidata *, void *);
357 void (*truncate) (struct inode *);
358 int (*permission) (struct inode *, int);
359 int (*get_acl)(struct inode *, int);
360 int (*setattr) (struct dentry *, struct iattr *);
361 int (*getattr) (struct vfsmount *mnt, struct dentry *, struct kstat *);
362 int (*setxattr) (struct dentry *, const char *,const void *,size_t,int);
363 ssize_t (*getxattr) (struct dentry *, const char *, void *, size_t);
364 ssize_t (*listxattr) (struct dentry *, char *, size_t);
365 int (*removexattr) (struct dentry *, const char *);
366 void (*update_time)(struct inode *, struct timespec *, int);
367 struct file * (*atomic_open)(struct inode *, struct dentry *,
368 struct opendata *, unsigned open_flag,
369 umode_t create_mode, bool *created);
372 Again, all methods are called without any locks being held, unless
375 create: called by the open(2) and creat(2) system calls. Only
376 required if you want to support regular files. The dentry you
377 get should not have an inode (i.e. it should be a negative
378 dentry). Here you will probably call d_instantiate() with the
379 dentry and the newly created inode
381 lookup: called when the VFS needs to look up an inode in a parent
382 directory. The name to look for is found in the dentry. This
383 method must call d_add() to insert the found inode into the
384 dentry. The "i_count" field in the inode structure should be
385 incremented. If the named inode does not exist a NULL inode
386 should be inserted into the dentry (this is called a negative
387 dentry). Returning an error code from this routine must only
388 be done on a real error, otherwise creating inodes with system
389 calls like create(2), mknod(2), mkdir(2) and so on will fail.
390 If you wish to overload the dentry methods then you should
391 initialise the "d_dop" field in the dentry; this is a pointer
392 to a struct "dentry_operations".
393 This method is called with the directory inode semaphore held
395 link: called by the link(2) system call. Only required if you want
396 to support hard links. You will probably need to call
397 d_instantiate() just as you would in the create() method
399 unlink: called by the unlink(2) system call. Only required if you
400 want to support deleting inodes
402 symlink: called by the symlink(2) system call. Only required if you
403 want to support symlinks. You will probably need to call
404 d_instantiate() just as you would in the create() method
406 mkdir: called by the mkdir(2) system call. Only required if you want
407 to support creating subdirectories. You will probably need to
408 call d_instantiate() just as you would in the create() method
410 rmdir: called by the rmdir(2) system call. Only required if you want
411 to support deleting subdirectories
413 mknod: called by the mknod(2) system call to create a device (char,
414 block) inode or a named pipe (FIFO) or socket. Only required
415 if you want to support creating these types of inodes. You
416 will probably need to call d_instantiate() just as you would
417 in the create() method
419 rename: called by the rename(2) system call to rename the object to
420 have the parent and name given by the second inode and dentry.
422 readlink: called by the readlink(2) system call. Only required if
423 you want to support reading symbolic links
425 follow_link: called by the VFS to follow a symbolic link to the
426 inode it points to. Only required if you want to support
427 symbolic links. This method returns a void pointer cookie
428 that is passed to put_link().
430 put_link: called by the VFS to release resources allocated by
431 follow_link(). The cookie returned by follow_link() is passed
432 to this method as the last parameter. It is used by
433 filesystems such as NFS where page cache is not stable
434 (i.e. page that was installed when the symbolic link walk
435 started might not be in the page cache at the end of the
438 truncate: Deprecated. This will not be called if ->setsize is defined.
439 Called by the VFS to change the size of a file. The
440 i_size field of the inode is set to the desired size by the
441 VFS before this method is called. This method is called by
442 the truncate(2) system call and related functionality.
444 Note: ->truncate and vmtruncate are deprecated. Do not add new
445 instances/calls of these. Filesystems should be converted to do their
446 truncate sequence via ->setattr().
448 permission: called by the VFS to check for access rights on a POSIX-like
451 May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in rcu-walk
452 mode, the filesystem must check the permission without blocking or
453 storing to the inode.
455 If a situation is encountered that rcu-walk cannot handle, return
456 -ECHILD and it will be called again in ref-walk mode.
458 setattr: called by the VFS to set attributes for a file. This method
459 is called by chmod(2) and related system calls.
461 getattr: called by the VFS to get attributes of a file. This method
462 is called by stat(2) and related system calls.
464 setxattr: called by the VFS to set an extended attribute for a file.
465 Extended attribute is a name:value pair associated with an
466 inode. This method is called by setxattr(2) system call.
468 getxattr: called by the VFS to retrieve the value of an extended
469 attribute name. This method is called by getxattr(2) function
472 listxattr: called by the VFS to list all extended attributes for a
473 given file. This method is called by listxattr(2) system call.
475 removexattr: called by the VFS to remove an extended attribute from
476 a file. This method is called by removexattr(2) system call.
478 update_time: called by the VFS to update a specific time or the i_version of
479 an inode. If this is not defined the VFS will update the inode itself
480 and call mark_inode_dirty_sync.
482 atomic_open: called on the last component of an open. Using this optional
483 method the filesystem can look up, possibly create and open the file in
484 one atomic operation. If it cannot perform this (e.g. the file type
485 turned out to be wrong) it may signal this by returning NULL instead of
486 an open struct file pointer. This method is only called if the last
487 component is negative or needs lookup. Cached positive dentries are
488 still handled by f_op->open().
490 The Address Space Object
491 ========================
493 The address space object is used to group and manage pages in the page
494 cache. It can be used to keep track of the pages in a file (or
495 anything else) and also track the mapping of sections of the file into
496 process address spaces.
498 There are a number of distinct yet related services that an
499 address-space can provide. These include communicating memory
500 pressure, page lookup by address, and keeping track of pages tagged as
503 The first can be used independently to the others. The VM can try to
504 either write dirty pages in order to clean them, or release clean
505 pages in order to reuse them. To do this it can call the ->writepage
506 method on dirty pages, and ->releasepage on clean pages with
507 PagePrivate set. Clean pages without PagePrivate and with no external
508 references will be released without notice being given to the
511 To achieve this functionality, pages need to be placed on an LRU with
512 lru_cache_add and mark_page_active needs to be called whenever the
515 Pages are normally kept in a radix tree index by ->index. This tree
516 maintains information about the PG_Dirty and PG_Writeback status of
517 each page, so that pages with either of these flags can be found
520 The Dirty tag is primarily used by mpage_writepages - the default
521 ->writepages method. It uses the tag to find dirty pages to call
522 ->writepage on. If mpage_writepages is not used (i.e. the address
523 provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is
524 almost unused. write_inode_now and sync_inode do use it (through
525 __sync_single_inode) to check if ->writepages has been successful in
526 writing out the whole address_space.
528 The Writeback tag is used by filemap*wait* and sync_page* functions,
529 via filemap_fdatawait_range, to wait for all writeback to
530 complete. While waiting ->sync_page (if defined) will be called on
531 each page that is found to require writeback.
533 An address_space handler may attach extra information to a page,
534 typically using the 'private' field in the 'struct page'. If such
535 information is attached, the PG_Private flag should be set. This will
536 cause various VM routines to make extra calls into the address_space
537 handler to deal with that data.
539 An address space acts as an intermediate between storage and
540 application. Data is read into the address space a whole page at a
541 time, and provided to the application either by copying of the page,
542 or by memory-mapping the page.
543 Data is written into the address space by the application, and then
544 written-back to storage typically in whole pages, however the
545 address_space has finer control of write sizes.
547 The read process essentially only requires 'readpage'. The write
548 process is more complicated and uses write_begin/write_end or
549 set_page_dirty to write data into the address_space, and writepage,
550 sync_page, and writepages to writeback data to storage.
552 Adding and removing pages to/from an address_space is protected by the
555 When data is written to a page, the PG_Dirty flag should be set. It
556 typically remains set until writepage asks for it to be written. This
557 should clear PG_Dirty and set PG_Writeback. It can be actually
558 written at any point after PG_Dirty is clear. Once it is known to be
559 safe, PG_Writeback is cleared.
561 Writeback makes use of a writeback_control structure...
563 struct address_space_operations
564 -------------------------------
566 This describes how the VFS can manipulate mapping of a file to page cache in
567 your filesystem. As of kernel 2.6.22, the following members are defined:
569 struct address_space_operations {
570 int (*writepage)(struct page *page, struct writeback_control *wbc);
571 int (*readpage)(struct file *, struct page *);
572 int (*sync_page)(struct page *);
573 int (*writepages)(struct address_space *, struct writeback_control *);
574 int (*set_page_dirty)(struct page *page);
575 int (*readpages)(struct file *filp, struct address_space *mapping,
576 struct list_head *pages, unsigned nr_pages);
577 int (*write_begin)(struct file *, struct address_space *mapping,
578 loff_t pos, unsigned len, unsigned flags,
579 struct page **pagep, void **fsdata);
580 int (*write_end)(struct file *, struct address_space *mapping,
581 loff_t pos, unsigned len, unsigned copied,
582 struct page *page, void *fsdata);
583 sector_t (*bmap)(struct address_space *, sector_t);
584 int (*invalidatepage) (struct page *, unsigned long);
585 int (*releasepage) (struct page *, int);
586 void (*freepage)(struct page *);
587 ssize_t (*direct_IO)(int, struct kiocb *, const struct iovec *iov,
588 loff_t offset, unsigned long nr_segs);
589 struct page* (*get_xip_page)(struct address_space *, sector_t,
591 /* migrate the contents of a page to the specified target */
592 int (*migratepage) (struct page *, struct page *);
593 int (*launder_page) (struct page *);
594 int (*error_remove_page) (struct mapping *mapping, struct page *page);
597 writepage: called by the VM to write a dirty page to backing store.
598 This may happen for data integrity reasons (i.e. 'sync'), or
599 to free up memory (flush). The difference can be seen in
601 The PG_Dirty flag has been cleared and PageLocked is true.
602 writepage should start writeout, should set PG_Writeback,
603 and should make sure the page is unlocked, either synchronously
604 or asynchronously when the write operation completes.
606 If wbc->sync_mode is WB_SYNC_NONE, ->writepage doesn't have to
607 try too hard if there are problems, and may choose to write out
608 other pages from the mapping if that is easier (e.g. due to
609 internal dependencies). If it chooses not to start writeout, it
610 should return AOP_WRITEPAGE_ACTIVATE so that the VM will not keep
611 calling ->writepage on that page.
613 See the file "Locking" for more details.
615 readpage: called by the VM to read a page from backing store.
616 The page will be Locked when readpage is called, and should be
617 unlocked and marked uptodate once the read completes.
618 If ->readpage discovers that it needs to unlock the page for
619 some reason, it can do so, and then return AOP_TRUNCATED_PAGE.
620 In this case, the page will be relocated, relocked and if
621 that all succeeds, ->readpage will be called again.
623 sync_page: called by the VM to notify the backing store to perform all
624 queued I/O operations for a page. I/O operations for other pages
625 associated with this address_space object may also be performed.
627 This function is optional and is called only for pages with
628 PG_Writeback set while waiting for the writeback to complete.
630 writepages: called by the VM to write out pages associated with the
631 address_space object. If wbc->sync_mode is WBC_SYNC_ALL, then
632 the writeback_control will specify a range of pages that must be
633 written out. If it is WBC_SYNC_NONE, then a nr_to_write is given
634 and that many pages should be written if possible.
635 If no ->writepages is given, then mpage_writepages is used
636 instead. This will choose pages from the address space that are
637 tagged as DIRTY and will pass them to ->writepage.
639 set_page_dirty: called by the VM to set a page dirty.
640 This is particularly needed if an address space attaches
641 private data to a page, and that data needs to be updated when
642 a page is dirtied. This is called, for example, when a memory
643 mapped page gets modified.
644 If defined, it should set the PageDirty flag, and the
645 PAGECACHE_TAG_DIRTY tag in the radix tree.
647 readpages: called by the VM to read pages associated with the address_space
648 object. This is essentially just a vector version of
649 readpage. Instead of just one page, several pages are
651 readpages is only used for read-ahead, so read errors are
652 ignored. If anything goes wrong, feel free to give up.
655 Called by the generic buffered write code to ask the filesystem to
656 prepare to write len bytes at the given offset in the file. The
657 address_space should check that the write will be able to complete,
658 by allocating space if necessary and doing any other internal
659 housekeeping. If the write will update parts of any basic-blocks on
660 storage, then those blocks should be pre-read (if they haven't been
661 read already) so that the updated blocks can be written out properly.
663 The filesystem must return the locked pagecache page for the specified
664 offset, in *pagep, for the caller to write into.
666 It must be able to cope with short writes (where the length passed to
667 write_begin is greater than the number of bytes copied into the page).
669 flags is a field for AOP_FLAG_xxx flags, described in
672 A void * may be returned in fsdata, which then gets passed into
675 Returns 0 on success; < 0 on failure (which is the error code), in
676 which case write_end is not called.
678 write_end: After a successful write_begin, and data copy, write_end must
679 be called. len is the original len passed to write_begin, and copied
680 is the amount that was able to be copied (copied == len is always true
681 if write_begin was called with the AOP_FLAG_UNINTERRUPTIBLE flag).
683 The filesystem must take care of unlocking the page and releasing it
684 refcount, and updating i_size.
686 Returns < 0 on failure, otherwise the number of bytes (<= 'copied')
687 that were able to be copied into pagecache.
689 bmap: called by the VFS to map a logical block offset within object to
690 physical block number. This method is used by the FIBMAP
691 ioctl and for working with swap-files. To be able to swap to
692 a file, the file must have a stable mapping to a block
693 device. The swap system does not go through the filesystem
694 but instead uses bmap to find out where the blocks in the file
695 are and uses those addresses directly.
698 invalidatepage: If a page has PagePrivate set, then invalidatepage
699 will be called when part or all of the page is to be removed
700 from the address space. This generally corresponds to either a
701 truncation or a complete invalidation of the address space
702 (in the latter case 'offset' will always be 0).
703 Any private data associated with the page should be updated
704 to reflect this truncation. If offset is 0, then
705 the private data should be released, because the page
706 must be able to be completely discarded. This may be done by
707 calling the ->releasepage function, but in this case the
708 release MUST succeed.
710 releasepage: releasepage is called on PagePrivate pages to indicate
711 that the page should be freed if possible. ->releasepage
712 should remove any private data from the page and clear the
713 PagePrivate flag. If releasepage() fails for some reason, it must
714 indicate failure with a 0 return value.
715 releasepage() is used in two distinct though related cases. The
716 first is when the VM finds a clean page with no active users and
717 wants to make it a free page. If ->releasepage succeeds, the
718 page will be removed from the address_space and become free.
720 The second case is when a request has been made to invalidate
721 some or all pages in an address_space. This can happen
722 through the fadvice(POSIX_FADV_DONTNEED) system call or by the
723 filesystem explicitly requesting it as nfs and 9fs do (when
724 they believe the cache may be out of date with storage) by
725 calling invalidate_inode_pages2().
726 If the filesystem makes such a call, and needs to be certain
727 that all pages are invalidated, then its releasepage will
728 need to ensure this. Possibly it can clear the PageUptodate
729 bit if it cannot free private data yet.
731 freepage: freepage is called once the page is no longer visible in
732 the page cache in order to allow the cleanup of any private
733 data. Since it may be called by the memory reclaimer, it
734 should not assume that the original address_space mapping still
735 exists, and it should not block.
737 direct_IO: called by the generic read/write routines to perform
738 direct_IO - that is IO requests which bypass the page cache
739 and transfer data directly between the storage and the
740 application's address space.
742 get_xip_page: called by the VM to translate a block number to a page.
743 The page is valid until the corresponding filesystem is unmounted.
744 Filesystems that want to use execute-in-place (XIP) need to implement
745 it. An example implementation can be found in fs/ext2/xip.c.
747 migrate_page: This is used to compact the physical memory usage.
748 If the VM wants to relocate a page (maybe off a memory card
749 that is signalling imminent failure) it will pass a new page
750 and an old page to this function. migrate_page should
751 transfer any private data across and update any references
752 that it has to the page.
754 launder_page: Called before freeing a page - it writes back the dirty page. To
755 prevent redirtying the page, it is kept locked during the whole
758 error_remove_page: normally set to generic_error_remove_page if truncation
759 is ok for this address space. Used for memory failure handling.
760 Setting this implies you deal with pages going away under you,
761 unless you have them locked or reference counts increased.
767 A file object represents a file opened by a process.
770 struct file_operations
771 ----------------------
773 This describes how the VFS can manipulate an open file. As of kernel
774 3.5, the following members are defined:
776 struct file_operations {
777 struct module *owner;
778 loff_t (*llseek) (struct file *, loff_t, int);
779 ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
780 ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
781 ssize_t (*aio_read) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
782 ssize_t (*aio_write) (struct kiocb *, const struct iovec *, unsigned long, loff_t);
783 int (*readdir) (struct file *, void *, filldir_t);
784 unsigned int (*poll) (struct file *, struct poll_table_struct *);
785 long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
786 long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
787 int (*mmap) (struct file *, struct vm_area_struct *);
788 int (*open) (struct inode *, struct file *);
789 int (*flush) (struct file *);
790 int (*release) (struct inode *, struct file *);
791 int (*fsync) (struct file *, loff_t, loff_t, int datasync);
792 int (*aio_fsync) (struct kiocb *, int datasync);
793 int (*fasync) (int, struct file *, int);
794 int (*lock) (struct file *, int, struct file_lock *);
795 ssize_t (*readv) (struct file *, const struct iovec *, unsigned long, loff_t *);
796 ssize_t (*writev) (struct file *, const struct iovec *, unsigned long, loff_t *);
797 ssize_t (*sendfile) (struct file *, loff_t *, size_t, read_actor_t, void *);
798 ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
799 unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
800 int (*check_flags)(int);
801 int (*flock) (struct file *, int, struct file_lock *);
802 ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, size_t, unsigned int);
803 ssize_t (*splice_read)(struct file *, struct pipe_inode_info *, size_t, unsigned int);
804 int (*setlease)(struct file *, long arg, struct file_lock **);
805 long (*fallocate)(struct file *, int mode, loff_t offset, loff_t len);
808 Again, all methods are called without any locks being held, unless
811 llseek: called when the VFS needs to move the file position index
813 read: called by read(2) and related system calls
815 aio_read: called by io_submit(2) and other asynchronous I/O operations
817 write: called by write(2) and related system calls
819 aio_write: called by io_submit(2) and other asynchronous I/O operations
821 readdir: called when the VFS needs to read the directory contents
823 poll: called by the VFS when a process wants to check if there is
824 activity on this file and (optionally) go to sleep until there
825 is activity. Called by the select(2) and poll(2) system calls
827 unlocked_ioctl: called by the ioctl(2) system call.
829 compat_ioctl: called by the ioctl(2) system call when 32 bit system calls
830 are used on 64 bit kernels.
832 mmap: called by the mmap(2) system call
834 open: called by the VFS when an inode should be opened. When the VFS
835 opens a file, it creates a new "struct file". It then calls the
836 open method for the newly allocated file structure. You might
837 think that the open method really belongs in
838 "struct inode_operations", and you may be right. I think it's
839 done the way it is because it makes filesystems simpler to
840 implement. The open() method is a good place to initialize the
841 "private_data" member in the file structure if you want to point
842 to a device structure
844 flush: called by the close(2) system call to flush a file
846 release: called when the last reference to an open file is closed
848 fsync: called by the fsync(2) system call
850 fasync: called by the fcntl(2) system call when asynchronous
851 (non-blocking) mode is enabled for a file
853 lock: called by the fcntl(2) system call for F_GETLK, F_SETLK, and F_SETLKW
856 readv: called by the readv(2) system call
858 writev: called by the writev(2) system call
860 sendfile: called by the sendfile(2) system call
862 get_unmapped_area: called by the mmap(2) system call
864 check_flags: called by the fcntl(2) system call for F_SETFL command
866 flock: called by the flock(2) system call
868 splice_write: called by the VFS to splice data from a pipe to a file. This
869 method is used by the splice(2) system call
871 splice_read: called by the VFS to splice data from file to a pipe. This
872 method is used by the splice(2) system call
874 setlease: called by the VFS to set or release a file lock lease.
875 setlease has the file_lock_lock held and must not sleep.
877 fallocate: called by the VFS to preallocate blocks or punch a hole.
879 Note that the file operations are implemented by the specific
880 filesystem in which the inode resides. When opening a device node
881 (character or block special) most filesystems will call special
882 support routines in the VFS which will locate the required device
883 driver information. These support routines replace the filesystem file
884 operations with those for the device driver, and then proceed to call
885 the new open() method for the file. This is how opening a device file
886 in the filesystem eventually ends up calling the device driver open()
890 Directory Entry Cache (dcache)
891 ==============================
894 struct dentry_operations
895 ------------------------
897 This describes how a filesystem can overload the standard dentry
898 operations. Dentries and the dcache are the domain of the VFS and the
899 individual filesystem implementations. Device drivers have no business
900 here. These methods may be set to NULL, as they are either optional or
901 the VFS uses a default. As of kernel 2.6.22, the following members are
904 struct dentry_operations {
905 int (*d_revalidate)(struct dentry *, struct nameidata *);
906 int (*d_hash)(const struct dentry *, const struct inode *,
908 int (*d_compare)(const struct dentry *, const struct inode *,
909 const struct dentry *, const struct inode *,
910 unsigned int, const char *, const struct qstr *);
911 int (*d_delete)(const struct dentry *);
912 void (*d_release)(struct dentry *);
913 void (*d_iput)(struct dentry *, struct inode *);
914 char *(*d_dname)(struct dentry *, char *, int);
915 struct vfsmount *(*d_automount)(struct path *);
916 int (*d_manage)(struct dentry *, bool);
919 d_revalidate: called when the VFS needs to revalidate a dentry. This
920 is called whenever a name look-up finds a dentry in the
921 dcache. Most filesystems leave this as NULL, because all their
922 dentries in the dcache are valid
924 d_revalidate may be called in rcu-walk mode (nd->flags & LOOKUP_RCU).
925 If in rcu-walk mode, the filesystem must revalidate the dentry without
926 blocking or storing to the dentry, d_parent and d_inode should not be
927 used without care (because they can go NULL), instead nd->inode should
930 If a situation is encountered that rcu-walk cannot handle, return
931 -ECHILD and it will be called again in ref-walk mode.
933 d_hash: called when the VFS adds a dentry to the hash table. The first
934 dentry passed to d_hash is the parent directory that the name is
935 to be hashed into. The inode is the dentry's inode.
937 Same locking and synchronisation rules as d_compare regarding
938 what is safe to dereference etc.
940 d_compare: called to compare a dentry name with a given name. The first
941 dentry is the parent of the dentry to be compared, the second is
942 the parent's inode, then the dentry and inode (may be NULL) of the
943 child dentry. len and name string are properties of the dentry to be
944 compared. qstr is the name to compare it with.
946 Must be constant and idempotent, and should not take locks if
947 possible, and should not or store into the dentry or inodes.
948 Should not dereference pointers outside the dentry or inodes without
949 lots of care (eg. d_parent, d_inode, d_name should not be used).
951 However, our vfsmount is pinned, and RCU held, so the dentries and
952 inodes won't disappear, neither will our sb or filesystem module.
953 ->i_sb and ->d_sb may be used.
955 It is a tricky calling convention because it needs to be called under
956 "rcu-walk", ie. without any locks or references on things.
958 d_delete: called when the last reference to a dentry is dropped and the
959 dcache is deciding whether or not to cache it. Return 1 to delete
960 immediately, or 0 to cache the dentry. Default is NULL which means to
961 always cache a reachable dentry. d_delete must be constant and
964 d_release: called when a dentry is really deallocated
966 d_iput: called when a dentry loses its inode (just prior to its
967 being deallocated). The default when this is NULL is that the
968 VFS calls iput(). If you define this method, you must call
971 d_dname: called when the pathname of a dentry should be generated.
972 Useful for some pseudo filesystems (sockfs, pipefs, ...) to delay
973 pathname generation. (Instead of doing it when dentry is created,
974 it's done only when the path is needed.). Real filesystems probably
975 dont want to use it, because their dentries are present in global
976 dcache hash, so their hash should be an invariant. As no lock is
977 held, d_dname() should not try to modify the dentry itself, unless
978 appropriate SMP safety is used. CAUTION : d_path() logic is quite
979 tricky. The correct way to return for example "Hello" is to put it
980 at the end of the buffer, and returns a pointer to the first char.
981 dynamic_dname() helper function is provided to take care of this.
983 d_automount: called when an automount dentry is to be traversed (optional).
984 This should create a new VFS mount record and return the record to the
985 caller. The caller is supplied with a path parameter giving the
986 automount directory to describe the automount target and the parent
987 VFS mount record to provide inheritable mount parameters. NULL should
988 be returned if someone else managed to make the automount first. If
989 the vfsmount creation failed, then an error code should be returned.
990 If -EISDIR is returned, then the directory will be treated as an
991 ordinary directory and returned to pathwalk to continue walking.
993 If a vfsmount is returned, the caller will attempt to mount it on the
994 mountpoint and will remove the vfsmount from its expiration list in
995 the case of failure. The vfsmount should be returned with 2 refs on
996 it to prevent automatic expiration - the caller will clean up the
999 This function is only used if DCACHE_NEED_AUTOMOUNT is set on the
1000 dentry. This is set by __d_instantiate() if S_AUTOMOUNT is set on the
1003 d_manage: called to allow the filesystem to manage the transition from a
1004 dentry (optional). This allows autofs, for example, to hold up clients
1005 waiting to explore behind a 'mountpoint' whilst letting the daemon go
1006 past and construct the subtree there. 0 should be returned to let the
1007 calling process continue. -EISDIR can be returned to tell pathwalk to
1008 use this directory as an ordinary directory and to ignore anything
1009 mounted on it and not to check the automount flag. Any other error
1010 code will abort pathwalk completely.
1012 If the 'rcu_walk' parameter is true, then the caller is doing a
1013 pathwalk in RCU-walk mode. Sleeping is not permitted in this mode,
1014 and the caller can be asked to leave it and call again by returning
1017 This function is only used if DCACHE_MANAGE_TRANSIT is set on the
1018 dentry being transited from.
1022 static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1024 return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1025 dentry->d_inode->i_ino);
1028 Each dentry has a pointer to its parent dentry, as well as a hash list
1029 of child dentries. Child dentries are basically like files in a
1033 Directory Entry Cache API
1034 --------------------------
1036 There are a number of functions defined which permit a filesystem to
1037 manipulate dentries:
1039 dget: open a new handle for an existing dentry (this just increments
1042 dput: close a handle for a dentry (decrements the usage count). If
1043 the usage count drops to 0, and the dentry is still in its
1044 parent's hash, the "d_delete" method is called to check whether
1045 it should be cached. If it should not be cached, or if the dentry
1046 is not hashed, it is deleted. Otherwise cached dentries are put
1047 into an LRU list to be reclaimed on memory shortage.
1049 d_drop: this unhashes a dentry from its parents hash list. A
1050 subsequent call to dput() will deallocate the dentry if its
1051 usage count drops to 0
1053 d_delete: delete a dentry. If there are no other open references to
1054 the dentry then the dentry is turned into a negative dentry
1055 (the d_iput() method is called). If there are other
1056 references, then d_drop() is called instead
1058 d_add: add a dentry to its parents hash list and then calls
1061 d_instantiate: add a dentry to the alias hash list for the inode and
1062 updates the "d_inode" member. The "i_count" member in the
1063 inode structure should be set/incremented. If the inode
1064 pointer is NULL, the dentry is called a "negative
1065 dentry". This function is commonly called when an inode is
1066 created for an existing negative dentry
1068 d_lookup: look up a dentry given its parent and path name component
1069 It looks up the child of that given name from the dcache
1070 hash table. If it is found, the reference count is incremented
1071 and the dentry is returned. The caller must use dput()
1072 to free the dentry when it finishes using it.
1080 On mount and remount the filesystem is passed a string containing a
1081 comma separated list of mount options. The options can have either of
1087 The <linux/parser.h> header defines an API that helps parse these
1088 options. There are plenty of examples on how to use it in existing
1094 If a filesystem accepts mount options, it must define show_options()
1095 to show all the currently active options. The rules are:
1097 - options MUST be shown which are not default or their values differ
1100 - options MAY be shown which are enabled by default or have their
1103 Options used only internally between a mount helper and the kernel
1104 (such as file descriptors), or which only have an effect during the
1105 mounting (such as ones controlling the creation of a journal) are exempt
1106 from the above rules.
1108 The underlying reason for the above rules is to make sure, that a
1109 mount can be accurately replicated (e.g. umounting and mounting again)
1110 based on the information found in /proc/mounts.
1112 A simple method of saving options at mount/remount time and showing
1113 them is provided with the save_mount_options() and
1114 generic_show_options() helper functions. Please note, that using
1115 these may have drawbacks. For more info see header comments for these
1116 functions in fs/namespace.c.
1121 (Note some of these resources are not up-to-date with the latest kernel
1124 Creating Linux virtual filesystems. 2002
1125 <http://lwn.net/Articles/13325/>
1127 The Linux Virtual File-system Layer by Neil Brown. 1999
1128 <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1130 A tour of the Linux VFS by Michael K. Johnson. 1996
1131 <http://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1133 A small trail through the Linux kernel by Andries Brouwer. 2001
1134 <http://www.win.tue.nl/~aeb/linux/vfs/trail.html>