vfs.rst

   1.. SPDX-License-Identifier: GPL-2.0
   2
   3=========================================
   4Overview of the Linux Virtual File System
   5=========================================
   6
   7Original author: Richard Gooch <rgooch@atnf.csiro.au>
   8
   9- Copyright (C) 1999 Richard Gooch
  10- Copyright (C) 2005 Pekka Enberg
  11
  12
  13Introduction
  14============
  15
  16The Virtual File System (also known as the Virtual Filesystem Switch) is
  17the software layer in the kernel that provides the filesystem interface
  18to userspace programs.  It also provides an abstraction within the
  19kernel which allows different filesystem implementations to coexist.
  20
  21VFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so on
  22are called from a process context.  Filesystem locking is described in
  23the document Documentation/filesystems/locking.rst.
  24
  25
  26Directory Entry Cache (dcache)
  27------------------------------
  28
  29The VFS implements the open(2), stat(2), chmod(2), and similar system
  30calls.  The pathname argument that is passed to them is used by the VFS
  31to search through the directory entry cache (also known as the dentry
  32cache or dcache).  This provides a very fast look-up mechanism to
  33translate a pathname (filename) into a specific dentry.  Dentries live
  34in RAM and are never saved to disc: they exist only for performance.
  35
  36The dentry cache is meant to be a view into your entire filespace.  As
  37most computers cannot fit all dentries in the RAM at the same time, some
  38bits of the cache are missing.  In order to resolve your pathname into a
  39dentry, the VFS may have to resort to creating dentries along the way,
  40and then loading the inode.  This is done by looking up the inode.
  41
  42
  43The Inode Object
  44----------------
  45
  46An individual dentry usually has a pointer to an inode.  Inodes are
  47filesystem objects such as regular files, directories, FIFOs and other
  48beasts.  They live either on the disc (for block device filesystems) or
  49in the memory (for pseudo filesystems).  Inodes that live on the disc
  50are copied into the memory when required and changes to the inode are
  51written back to disc.  A single inode can be pointed to by multiple
  52dentries (hard links, for example, do this).
  53
  54To look up an inode requires that the VFS calls the lookup() method of
  55the parent directory inode.  This method is installed by the specific
  56filesystem implementation that the inode lives in.  Once the VFS has the
  57required dentry (and hence the inode), we can do all those boring things
  58like open(2) the file, or stat(2) it to peek at the inode data.  The
  59stat(2) operation is fairly simple: once the VFS has the dentry, it
  60peeks at the inode data and passes some of it back to userspace.
  61
  62
  63The File Object
  64---------------
  65
  66Opening a file requires another operation: allocation of a file
  67structure (this is the kernel-side implementation of file descriptors).
  68The freshly allocated file structure is initialized with a pointer to
  69the dentry and a set of file operation member functions.  These are
  70taken from the inode data.  The open() file method is then called so the
  71specific filesystem implementation can do its work.  You can see that
  72this is another switch performed by the VFS.  The file structure is
  73placed into the file descriptor table for the process.
  74
  75Reading, writing and closing files (and other assorted VFS operations)
  76is done by using the userspace file descriptor to grab the appropriate
  77file structure, and then calling the required file structure method to
  78do whatever is required.  For as long as the file is open, it keeps the
  79dentry in use, which in turn means that the VFS inode is still in use.
  80
  81
  82Registering and Mounting a Filesystem
  83=====================================
  84
  85To register and unregister a filesystem, use the following API
  86functions:
  87
  88.. code-block:: c
  89
  90	#include <linux/fs.h>
  91
  92	extern int register_filesystem(struct file_system_type *);
  93	extern int unregister_filesystem(struct file_system_type *);
  94
  95The passed struct file_system_type describes your filesystem.  When a
  96request is made to mount a filesystem onto a directory in your
  97namespace, the VFS will call the appropriate mount() method for the
  98specific filesystem.  New vfsmount referring to the tree returned by
  99->mount() will be attached to the mountpoint, so that when pathname
 100resolution reaches the mountpoint it will jump into the root of that
 101vfsmount.
 102
 103You can see all filesystems that are registered to the kernel in the
 104file /proc/filesystems.
 105
 106
 107struct file_system_type
 108-----------------------
 109
 110This describes the filesystem.  The following
 111members are defined:
 112
 113.. code-block:: c
 114
 115	struct file_system_type {
 116		const char *name;
 117		int fs_flags;
 118		int (*init_fs_context)(struct fs_context *);
 119		const struct fs_parameter_spec *parameters;
 120		struct dentry *(*mount) (struct file_system_type *, int,
 121			const char *, void *);
 122		void (*kill_sb) (struct super_block *);
 123		struct module *owner;
 124		struct file_system_type * next;
 125		struct hlist_head fs_supers;
 126
 127		struct lock_class_key s_lock_key;
 128		struct lock_class_key s_umount_key;
 129		struct lock_class_key s_vfs_rename_key;
 130		struct lock_class_key s_writers_key[SB_FREEZE_LEVELS];
 131
 132		struct lock_class_key i_lock_key;
 133		struct lock_class_key i_mutex_key;
 134		struct lock_class_key invalidate_lock_key;
 135		struct lock_class_key i_mutex_dir_key;
 136	};
 137
 138``name``
 139	the name of the filesystem type, such as "ext2", "iso9660",
 140	"msdos" and so on
 141
 142``fs_flags``
 143	various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.)
 144
 145``init_fs_context``
 146	Initializes 'struct fs_context' ->ops and ->fs_private fields with
 147	filesystem-specific data.
 148
 149``parameters``
 150	Pointer to the array of filesystem parameters descriptors
 151	'struct fs_parameter_spec'.
 152	More info in Documentation/filesystems/mount_api.rst.
 153
 154``mount``
 155	the method to call when a new instance of this filesystem should
 156	be mounted
 157
 158``kill_sb``
 159	the method to call when an instance of this filesystem should be
 160	shut down
 161
 162
 163``owner``
 164	for internal VFS use: you should initialize this to THIS_MODULE
 165	in most cases.
 166
 167``next``
 168	for internal VFS use: you should initialize this to NULL
 169
 170``fs_supers``
 171	for internal VFS use: hlist of filesystem instances (superblocks)
 172
 173  s_lock_key, s_umount_key, s_vfs_rename_key, s_writers_key,
 174  i_lock_key, i_mutex_key, invalidate_lock_key, i_mutex_dir_key: lockdep-specific
 175
 176The mount() method has the following arguments:
 177
 178``struct file_system_type *fs_type``
 179	describes the filesystem, partly initialized by the specific
 180	filesystem code
 181
 182``int flags``
 183	mount flags
 184
 185``const char *dev_name``
 186	the device name we are mounting.
 187
 188``void *data``
 189	arbitrary mount options, usually comes as an ASCII string (see
 190	"Mount Options" section)
 191
 192The mount() method must return the root dentry of the tree requested by
 193caller.  An active reference to its superblock must be grabbed and the
 194superblock must be locked.  On failure it should return ERR_PTR(error).
 195
 196The arguments match those of mount(2) and their interpretation depends
 197on filesystem type.  E.g. for block filesystems, dev_name is interpreted
 198as block device name, that device is opened and if it contains a
 199suitable filesystem image the method creates and initializes struct
 200super_block accordingly, returning its root dentry to caller.
 201
 202->mount() may choose to return a subtree of existing filesystem - it
 203doesn't have to create a new one.  The main result from the caller's
 204point of view is a reference to dentry at the root of (sub)tree to be
 205attached; creation of new superblock is a common side effect.
 206
 207The most interesting member of the superblock structure that the mount()
 208method fills in is the "s_op" field.  This is a pointer to a "struct
 209super_operations" which describes the next level of the filesystem
 210implementation.
 211
 212For more information on mounting (and the new mount API), see
 213Documentation/filesystems/mount_api.rst.
 214
 215The Superblock Object
 216=====================
 217
 218A superblock object represents a mounted filesystem.
 219
 220
 221struct super_operations
 222-----------------------
 223
 224This describes how the VFS can manipulate the superblock of your
 225filesystem.  The following members are defined:
 226
 227.. code-block:: c
 228
 229	struct super_operations {
 230		struct inode *(*alloc_inode)(struct super_block *sb);
 231		void (*destroy_inode)(struct inode *);
 232		void (*free_inode)(struct inode *);
 233
 234		void (*dirty_inode) (struct inode *, int flags);
 235		int (*write_inode) (struct inode *, struct writeback_control *wbc);
 236		int (*drop_inode) (struct inode *);
 237		void (*evict_inode) (struct inode *);
 238		void (*put_super) (struct super_block *);
 239		int (*sync_fs)(struct super_block *sb, int wait);
 240		int (*freeze_super) (struct super_block *sb,
 241					enum freeze_holder who);
 242		int (*freeze_fs) (struct super_block *);
 243		int (*thaw_super) (struct super_block *sb,
 244					enum freeze_wholder who);
 245		int (*unfreeze_fs) (struct super_block *);
 246		int (*statfs) (struct dentry *, struct kstatfs *);
 247		int (*remount_fs) (struct super_block *, int *, char *);
 248		void (*umount_begin) (struct super_block *);
 249
 250		int (*show_options)(struct seq_file *, struct dentry *);
 251		int (*show_devname)(struct seq_file *, struct dentry *);
 252		int (*show_path)(struct seq_file *, struct dentry *);
 253		int (*show_stats)(struct seq_file *, struct dentry *);
 254
 255		ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t);
 256		ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t);
 257		struct dquot **(*get_dquots)(struct inode *);
 258
 259		long (*nr_cached_objects)(struct super_block *,
 260					struct shrink_control *);
 261		long (*free_cached_objects)(struct super_block *,
 262					struct shrink_control *);
 263	};
 264
 265All methods are called without any locks being held, unless otherwise
 266noted.  This means that most methods can block safely.  All methods are
 267only called from a process context (i.e. not from an interrupt handler
 268or bottom half).
 269
 270``alloc_inode``
 271	this method is called by alloc_inode() to allocate memory for
 272	struct inode and initialize it.  If this function is not
 273	defined, a simple 'struct inode' is allocated.  Normally
 274	alloc_inode will be used to allocate a larger structure which
 275	contains a 'struct inode' embedded within it.
 276
 277``destroy_inode``
 278	this method is called by destroy_inode() to release resources
 279	allocated for struct inode.  It is only required if
 280	->alloc_inode was defined and simply undoes anything done by
 281	->alloc_inode.
 282
 283``free_inode``
 284	this method is called from RCU callback. If you use call_rcu()
 285	in ->destroy_inode to free 'struct inode' memory, then it's
 286	better to release memory in this method.
 287
 288``dirty_inode``
 289	this method is called by the VFS when an inode is marked dirty.
 290	This is specifically for the inode itself being marked dirty,
 291	not its data.  If the update needs to be persisted by fdatasync(),
 292	then I_DIRTY_DATASYNC will be set in the flags argument.
 293	I_DIRTY_TIME will be set in the flags in case lazytime is enabled
 294	and struct inode has times updated since the last ->dirty_inode
 295	call.
 296
 297``write_inode``
 298	this method is called when the VFS needs to write an inode to
 299	disc.  The second parameter indicates whether the write should
 300	be synchronous or not, not all filesystems check this flag.
 301
 302``drop_inode``
 303	called when the last access to the inode is dropped, with the
 304	inode->i_lock spinlock held.
 305
 306	This method should be either NULL (normal UNIX filesystem
 307	semantics) or "inode_just_drop" (for filesystems that do
 308	not want to cache inodes - causing "delete_inode" to always be
 309	called regardless of the value of i_nlink)
 310
 311	The "inode_just_drop()" behavior is equivalent to the old
 312	practice of using "force_delete" in the put_inode() case, but
 313	does not have the races that the "force_delete()" approach had.
 314
 315``evict_inode``
 316	called when the VFS wants to evict an inode. Caller does
 317	*not* evict the pagecache or inode-associated metadata buffers;
 318	the method has to use truncate_inode_pages_final() to get rid
 319	of those. Caller makes sure async writeback cannot be running for
 320	the inode while (or after) ->evict_inode() is called. Optional.
 321
 322``put_super``
 323	called when the VFS wishes to free the superblock
 324	(i.e. unmount).  This is called with the superblock lock held
 325
 326``sync_fs``
 327	called when VFS is writing out all dirty data associated with a
 328	superblock.  The second parameter indicates whether the method
 329	should wait until the write out has been completed.  Optional.
 330
 331``freeze_super``
 332	Called instead of ->freeze_fs callback if provided.
 333	Main difference is that ->freeze_super is called without taking
 334	down_write(&sb->s_umount). If filesystem implements it and wants
 335	->freeze_fs to be called too, then it has to call ->freeze_fs
 336	explicitly from this callback. Optional.
 337
 338``freeze_fs``
 339	called when VFS is locking a filesystem and forcing it into a
 340	consistent state.  This method is currently used by the Logical
 341	Volume Manager (LVM) and ioctl(FIFREEZE). Optional.
 342
 343``thaw_super``
 344	called when VFS is unlocking a filesystem and making it writable
 345	again after ->freeze_super. Optional.
 346
 347``unfreeze_fs``
 348	called when VFS is unlocking a filesystem and making it writable
 349	again after ->freeze_fs. Optional.
 350
 351``statfs``
 352	called when the VFS needs to get filesystem statistics.
 353
 354``remount_fs``
 355	called when the filesystem is remounted.  This is called with
 356	the kernel lock held
 357
 358``umount_begin``
 359	called when the VFS is unmounting a filesystem.
 360
 361``show_options``
 362	called by the VFS to show mount options for /proc/<pid>/mounts
 363	and /proc/<pid>/mountinfo.
 364	(see "Mount Options" section)
 365
 366``show_devname``
 367	Optional. Called by the VFS to show device name for
 368	/proc/<pid>/{mounts,mountinfo,mountstats}. If not provided then
 369	'(struct mount).mnt_devname' will be used.
 370
 371``show_path``
 372	Optional. Called by the VFS (for /proc/<pid>/mountinfo) to show
 373	the mount root dentry path relative to the filesystem root.
 374
 375``show_stats``
 376	Optional. Called by the VFS (for /proc/<pid>/mountstats) to show
 377	filesystem-specific mount statistics.
 378
 379``quota_read``
 380	called by the VFS to read from filesystem quota file.
 381
 382``quota_write``
 383	called by the VFS to write to filesystem quota file.
 384
 385``get_dquots``
 386	called by quota to get 'struct dquot' array for a particular inode.
 387	Optional.
 388
 389``nr_cached_objects``
 390	called by the sb cache shrinking function for the filesystem to
 391	return the number of freeable cached objects it contains.
 392	Optional.
 393
 394``free_cache_objects``
 395	called by the sb cache shrinking function for the filesystem to
 396	scan the number of objects indicated to try to free them.
 397	Optional, but any filesystem implementing this method needs to
 398	also implement ->nr_cached_objects for it to be called
 399	correctly.
 400
 401	We can't do anything with any errors that the filesystem might
 402	encountered, hence the void return type.  This will never be
 403	called if the VM is trying to reclaim under GFP_NOFS conditions,
 404	hence this method does not need to handle that situation itself.
 405
 406	Implementations must include conditional reschedule calls inside
 407	any scanning loop that is done.  This allows the VFS to
 408	determine appropriate scan batch sizes without having to worry
 409	about whether implementations will cause holdoff problems due to
 410	large scan batch sizes.
 411
 412Whoever sets up the inode is responsible for filling in the "i_op"
 413field.  This is a pointer to a "struct inode_operations" which describes
 414the methods that can be performed on individual inodes.
 415
 416
 417struct xattr_handler
 418---------------------
 419
 420On filesystems that support extended attributes (xattrs), the s_xattr
 421superblock field points to a NULL-terminated array of xattr handlers.
 422Extended attributes are name:value pairs.
 423
 424``name``
 425	Indicates that the handler matches attributes with the specified
 426	name (such as "system.posix_acl_access"); the prefix field must
 427	be NULL.
 428
 429``prefix``
 430	Indicates that the handler matches all attributes with the
 431	specified name prefix (such as "user."); the name field must be
 432	NULL.
 433
 434``list``
 435	Determine if attributes matching this xattr handler should be
 436	listed for a particular dentry.  Used by some listxattr
 437	implementations like generic_listxattr.
 438
 439``get``
 440	Called by the VFS to get the value of a particular extended
 441	attribute.  This method is called by the getxattr(2) system
 442	call.
 443
 444``set``
 445	Called by the VFS to set the value of a particular extended
 446	attribute.  When the new value is NULL, called to remove a
 447	particular extended attribute.  This method is called by the
 448	setxattr(2) and removexattr(2) system calls.
 449
 450When none of the xattr handlers of a filesystem match the specified
 451attribute name or when a filesystem doesn't support extended attributes,
 452the various ``*xattr(2)`` system calls return -EOPNOTSUPP.
 453
 454
 455The Inode Object
 456================
 457
 458An inode object represents an object within the filesystem.
 459
 460
 461struct inode_operations
 462-----------------------
 463
 464This describes how the VFS can manipulate an inode in your filesystem.
 465As of kernel 2.6.22, the following members are defined:
 466
 467.. code-block:: c
 468
 469	struct inode_operations {
 470		int (*create) (struct mnt_idmap *, struct inode *,struct dentry *, umode_t, bool);
 471		struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int);
 472		int (*link) (struct dentry *,struct inode *,struct dentry *);
 473		int (*unlink) (struct inode *,struct dentry *);
 474		int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *);
 475		struct dentry *(*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t);
 476		int (*rmdir) (struct inode *,struct dentry *);
 477		int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t);
 478		int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *,
 479			       struct inode *, struct dentry *, unsigned int);
 480		int (*readlink) (struct dentry *, char __user *,int);
 481		const char *(*get_link) (struct dentry *, struct inode *,
 482					 struct delayed_call *);
 483		int (*permission) (struct mnt_idmap *, struct inode *, int);
 484		struct posix_acl * (*get_inode_acl)(struct inode *, int, bool);
 485		int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *);
 486		int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int);
 487		ssize_t (*listxattr) (struct dentry *, char *, size_t);
 488		void (*update_time)(struct inode *, struct timespec *, int);
 489		int (*atomic_open)(struct inode *, struct dentry *, struct file *,
 490				   unsigned open_flag, umode_t create_mode);
 491		int (*tmpfile) (struct mnt_idmap *, struct inode *, struct file *, umode_t);
 492		struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int);
 493	        int (*set_acl)(struct mnt_idmap *, struct dentry *, struct posix_acl *, int);
 494		int (*fileattr_set)(struct mnt_idmap *idmap,
 495				    struct dentry *dentry, struct file_kattr *fa);
 496		int (*fileattr_get)(struct dentry *dentry, struct file_kattr *fa);
 497	        struct offset_ctx *(*get_offset_ctx)(struct inode *inode);
 498	};
 499
 500Again, all methods are called without any locks being held, unless
 501otherwise noted.
 502
 503``create``
 504	called by the open(2) and creat(2) system calls.  Only required
 505	if you want to support regular files.  The dentry you get should
 506	not have an inode (i.e. it should be a negative dentry).  Here
 507	you will probably call d_instantiate() with the dentry and the
 508	newly created inode
 509
 510``lookup``
 511	called when the VFS needs to look up an inode in a parent
 512	directory.  The name to look for is found in the dentry.  This
 513	method must call d_add() to insert the found inode into the
 514	dentry.  The "i_count" field in the inode structure should be
 515	incremented.  If the named inode does not exist a NULL inode
 516	should be inserted into the dentry (this is called a negative
 517	dentry).  Returning an error code from this routine must only be
 518	done on a real error, otherwise creating inodes with system
 519	calls like create(2), mknod(2), mkdir(2) and so on will fail.
 520	If you wish to overload the dentry methods then you should
 521	initialise the "d_dop" field in the dentry; this is a pointer to
 522	a struct "dentry_operations".  This method is called with the
 523	directory inode semaphore held
 524
 525``link``
 526	called by the link(2) system call.  Only required if you want to
 527	support hard links.  You will probably need to call
 528	d_instantiate() just as you would in the create() method
 529
 530``unlink``
 531	called by the unlink(2) system call.  Only required if you want
 532	to support deleting inodes
 533
 534``symlink``
 535	called by the symlink(2) system call.  Only required if you want
 536	to support symlinks.  You will probably need to call
 537	d_instantiate() just as you would in the create() method
 538
 539``mkdir``
 540	called by the mkdir(2) system call.  Only required if you want
 541	to support creating subdirectories.  You will probably need to
 542	call d_instantiate_new() just as you would in the create() method.
 543
 544	If d_instantiate_new() is not used and if the fh_to_dentry()
 545	export operation is provided, or if the storage might be
 546	accessible by another path (e.g. with a network filesystem)
 547	then more care may be needed.  Importantly d_instantate()
 548	should not be used with an inode that is no longer I_NEW if there
 549	any chance that the inode could already be attached to a dentry.
 550	This is because of a hard rule in the VFS that a directory must
 551	only ever have one dentry.
 552
 553	For example, if an NFS filesystem is mounted twice the new directory
 554	could be visible on the other mount before it is on the original
 555	mount, and a pair of name_to_handle_at(), open_by_handle_at()
 556	calls could instantiate the directory inode with an IS_ROOT()
 557	dentry before the first mkdir returns.
 558
 559	If there is any chance this could happen, then the new inode
 560	should be d_drop()ed and attached with d_splice_alias().  The
 561	returned dentry (if any) should be returned by ->mkdir().
 562
 563``rmdir``
 564	called by the rmdir(2) system call.  Only required if you want
 565	to support deleting subdirectories
 566
 567``mknod``
 568	called by the mknod(2) system call to create a device (char,
 569	block) inode or a named pipe (FIFO) or socket.  Only required if
 570	you want to support creating these types of inodes.  You will
 571	probably need to call d_instantiate() just as you would in the
 572	create() method
 573
 574``rename``
 575	called by the rename(2) system call to rename the object to have
 576	the parent and name given by the second inode and dentry.
 577
 578	The filesystem must return -EINVAL for any unsupported or
 579	unknown flags.  Currently the following flags are implemented:
 580	(1) RENAME_NOREPLACE: this flag indicates that if the target of
 581	the rename exists the rename should fail with -EEXIST instead of
 582	replacing the target.  The VFS already checks for existence, so
 583	for local filesystems the RENAME_NOREPLACE implementation is
 584	equivalent to plain rename.
 585	(2) RENAME_EXCHANGE: exchange source and target.  Both must
 586	exist; this is checked by the VFS.  Unlike plain rename, source
 587	and target may be of different type.
 588
 589``get_link``
 590	called by the VFS to follow a symbolic link to the inode it
 591	points to.  Only required if you want to support symbolic links.
 592	This method returns the symlink body to traverse (and possibly
 593	resets the current position with nd_jump_link()).  If the body
 594	won't go away until the inode is gone, nothing else is needed;
 595	if it needs to be otherwise pinned, arrange for its release by
 596	having get_link(..., ..., done) do set_delayed_call(done,
 597	destructor, argument).  In that case destructor(argument) will
 598	be called once VFS is done with the body you've returned.  May
 599	be called in RCU mode; that is indicated by NULL dentry
 600	argument.  If request can't be handled without leaving RCU mode,
 601	have it return ERR_PTR(-ECHILD).
 602
 603	If the filesystem stores the symlink target in ->i_link, the
 604	VFS may use it directly without calling ->get_link(); however,
 605	->get_link() must still be provided.  ->i_link must not be
 606	freed until after an RCU grace period.  Writing to ->i_link
 607	post-iget() time requires a 'release' memory barrier.
 608
 609``readlink``
 610	this is now just an override for use by readlink(2) for the
 611	cases when ->get_link uses nd_jump_link() or object is not in
 612	fact a symlink.  Normally filesystems should only implement
 613	->get_link for symlinks and readlink(2) will automatically use
 614	that.
 615
 616``permission``
 617	called by the VFS to check for access rights on a POSIX-like
 618	filesystem.
 619
 620	May be called in rcu-walk mode (mask & MAY_NOT_BLOCK).  If in
 621	rcu-walk mode, the filesystem must check the permission without
 622	blocking or storing to the inode.
 623
 624	If a situation is encountered that rcu-walk cannot handle,
 625	return
 626	-ECHILD and it will be called again in ref-walk mode.
 627
 628``setattr``
 629	called by the VFS to set attributes for a file.  This method is
 630	called by chmod(2) and related system calls.
 631
 632``getattr``
 633	called by the VFS to get attributes of a file.  This method is
 634	called by stat(2) and related system calls.
 635
 636``listxattr``
 637	called by the VFS to list all extended attributes for a given
 638	file.  This method is called by the listxattr(2) system call.
 639
 640``update_time``
 641	called by the VFS to update a specific time or the i_version of
 642	an inode.  If this is not defined the VFS will update the inode
 643	itself and call mark_inode_dirty_sync.
 644
 645``atomic_open``
 646	called on the last component of an open.  Using this optional
 647	method the filesystem can look up, possibly create and open the
 648	file in one atomic operation.  If it wants to leave actual
 649	opening to the caller (e.g. if the file turned out to be a
 650	symlink, device, or just something filesystem won't do atomic
 651	open for), it may signal this by returning finish_no_open(file,
 652	dentry).  This method is only called if the last component is
 653	negative or needs lookup.  Cached positive dentries are still
 654	handled by f_op->open().  If the file was created, FMODE_CREATED
 655	flag should be set in file->f_mode.  In case of O_EXCL the
 656	method must only succeed if the file didn't exist and hence
 657	FMODE_CREATED shall always be set on success.
 658
 659``tmpfile``
 660	called in the end of O_TMPFILE open().  Optional, equivalent to
 661	atomically creating, opening and unlinking a file in given
 662	directory.  On success needs to return with the file already
 663	open; this can be done by calling finish_open_simple() right at
 664	the end.
 665
 666``fileattr_get``
 667	called on ioctl(FS_IOC_GETFLAGS) and ioctl(FS_IOC_FSGETXATTR) to
 668	retrieve miscellaneous file flags and attributes.  Also called
 669	before the relevant SET operation to check what is being changed
 670	(in this case with i_rwsem locked exclusive).  If unset, then
 671	fall back to f_op->ioctl().
 672
 673``fileattr_set``
 674	called on ioctl(FS_IOC_SETFLAGS) and ioctl(FS_IOC_FSSETXATTR) to
 675	change miscellaneous file flags and attributes.  Callers hold
 676	i_rwsem exclusive.  If unset, then fall back to f_op->ioctl().
 677``get_offset_ctx``
 678	called to get the offset context for a directory inode. A
 679        filesystem must define this operation to use
 680        simple_offset_dir_operations.
 681
 682The Address Space Object
 683========================
 684
 685The address space object is used to group and manage pages in the page
 686cache.  It can be used to keep track of the pages in a file (or anything
 687else) and also track the mapping of sections of the file into process
 688address spaces.
 689
 690There are a number of distinct yet related services that an
 691address-space can provide.  These include communicating memory pressure,
 692page lookup by address, and keeping track of pages tagged as Dirty or
 693Writeback.
 694
 695The first can be used independently to the others.  The VM can try to
 696release clean pages in order to reuse them.  To do this it can call
 697->release_folio on clean folios with the private
 698flag set.  Clean pages without PagePrivate and with no external references
 699will be released without notice being given to the address_space.
 700
 701To achieve this functionality, pages need to be placed on an LRU with
 702lru_cache_add and mark_page_active needs to be called whenever the page
 703is used.
 704
 705Pages are normally kept in a radix tree index by ->index.  This tree
 706maintains information about the PG_Dirty and PG_Writeback status of each
 707page, so that pages with either of these flags can be found quickly.
 708
 709The Dirty tag is primarily used by mpage_writepages - the default
 710->writepages method.  It uses the tag to find dirty pages to
 711write back.  If mpage_writepages is not used (i.e. the address
 712provides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is almost
 713unused.  write_inode_now and sync_inode do use it (through
 714__sync_single_inode) to check if ->writepages has been successful in
 715writing out the whole address_space.
 716
 717The Writeback tag is used by filemap*wait* and sync_page* functions, via
 718filemap_fdatawait_range, to wait for all writeback to complete.
 719
 720An address_space handler may attach extra information to a page,
 721typically using the 'private' field in the 'struct page'.  If such
 722information is attached, the PG_Private flag should be set.  This will
 723cause various VM routines to make extra calls into the address_space
 724handler to deal with that data.
 725
 726An address space acts as an intermediate between storage and
 727application.  Data is read into the address space a whole page at a
 728time, and provided to the application either by copying of the page, or
 729by memory-mapping the page.  Data is written into the address space by
 730the application, and then written-back to storage typically in whole
 731pages, however the address_space has finer control of write sizes.
 732
 733The read process essentially only requires 'read_folio'.  The write
 734process is more complicated and uses write_begin/write_end or
 735dirty_folio to write data into the address_space, and
 736writepages to writeback data to storage.
 737
 738Removing pages from an address_space requires holding the inode's i_rwsem
 739exclusively, while adding pages to the address_space requires holding the
 740inode's i_mapping->invalidate_lock exclusively.
 741
 742When data is written to a page, the PG_Dirty flag should be set.  It
 743typically remains set until writepages asks for it to be written.  This
 744should clear PG_Dirty and set PG_Writeback.  It can be actually written
 745at any point after PG_Dirty is clear.  Once it is known to be safe,
 746PG_Writeback is cleared.
 747
 748Writeback makes use of a writeback_control structure to direct the
 749operations.  This gives the writepages operation some
 750information about the nature of and reason for the writeback request,
 751and the constraints under which it is being done.  It is also used to
 752return information back to the caller about the result of a
 753writepages request.
 754
 755
 756Handling errors during writeback
 757--------------------------------
 758
 759Most applications that do buffered I/O will periodically call a file
 760synchronization call (fsync, fdatasync, msync or sync_file_range) to
 761ensure that data written has made it to the backing store.  When there
 762is an error during writeback, they expect that error to be reported when
 763a file sync request is made.  After an error has been reported on one
 764request, subsequent requests on the same file descriptor should return
 7650, unless further writeback errors have occurred since the previous file
 766synchronization.
 767
 768Ideally, the kernel would report errors only on file descriptions on
 769which writes were done that subsequently failed to be written back.  The
 770generic pagecache infrastructure does not track the file descriptions
 771that have dirtied each individual page however, so determining which
 772file descriptors should get back an error is not possible.
 773
 774Instead, the generic writeback error tracking infrastructure in the
 775kernel settles for reporting errors to fsync on all file descriptions
 776that were open at the time that the error occurred.  In a situation with
 777multiple writers, all of them will get back an error on a subsequent
 778fsync, even if all of the writes done through that particular file
 779descriptor succeeded (or even if there were no writes on that file
 780descriptor at all).
 781
 782Filesystems that wish to use this infrastructure should call
 783mapping_set_error to record the error in the address_space when it
 784occurs.  Then, after writing back data from the pagecache in their
 785file->fsync operation, they should call file_check_and_advance_wb_err to
 786ensure that the struct file's error cursor has advanced to the correct
 787point in the stream of errors emitted by the backing device(s).
 788
 789
 790struct address_space_operations
 791-------------------------------
 792
 793This describes how the VFS can manipulate mapping of a file to page
 794cache in your filesystem.  The following members are defined:
 795
 796.. code-block:: c
 797
 798	struct address_space_operations {
 799		int (*read_folio)(struct file *, struct folio *);
 800		int (*writepages)(struct address_space *, struct writeback_control *);
 801		bool (*dirty_folio)(struct address_space *, struct folio *);
 802		void (*readahead)(struct readahead_control *);
 803		int (*write_begin)(const struct kiocb *, struct address_space *mapping,
 804				   loff_t pos, unsigned len,
 805				   struct page **pagep, void **fsdata);
 806		int (*write_end)(const struct kiocb *, struct address_space *mapping,
 807				 loff_t pos, unsigned len, unsigned copied,
 808				 struct folio *folio, void *fsdata);
 809		sector_t (*bmap)(struct address_space *, sector_t);
 810		void (*invalidate_folio) (struct folio *, size_t start, size_t len);
 811		bool (*release_folio)(struct folio *, gfp_t);
 812		void (*free_folio)(struct folio *);
 813		ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter);
 814		int (*migrate_folio)(struct mapping *, struct folio *dst,
 815				struct folio *src, enum migrate_mode);
 816		int (*launder_folio) (struct folio *);
 817
 818		bool (*is_partially_uptodate) (struct folio *, size_t from,
 819					       size_t count);
 820		void (*is_dirty_writeback)(struct folio *, bool *, bool *);
 821		int (*error_remove_folio)(struct mapping *mapping, struct folio *);
 822		int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span)
 823		int (*swap_deactivate)(struct file *);
 824		int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter);
 825	};
 826
 827``read_folio``
 828	Called by the page cache to read a folio from the backing store.
 829	The 'file' argument supplies authentication information to network
 830	filesystems, and is generally not used by block based filesystems.
 831	It may be NULL if the caller does not have an open file (eg if
 832	the kernel is performing a read for itself rather than on behalf
 833	of a userspace process with an open file).
 834
 835	If the mapping does not support large folios, the folio will
 836	contain a single page.	The folio will be locked when read_folio
 837	is called.  If the read completes successfully, the folio should
 838	be marked uptodate.  The filesystem should unlock the folio
 839	once the read has completed, whether it was successful or not.
 840	The filesystem does not need to modify the refcount on the folio;
 841	the page cache holds a reference count and that will not be
 842	released until the folio is unlocked.
 843
 844	Filesystems may implement ->read_folio() synchronously.
 845	In normal operation, folios are read through the ->readahead()
 846	method.  Only if this fails, or if the caller needs to wait for
 847	the read to complete will the page cache call ->read_folio().
 848	Filesystems should not attempt to perform their own readahead
 849	in the ->read_folio() operation.
 850
 851	If the filesystem cannot perform the read at this time, it can
 852	unlock the folio, do whatever action it needs to ensure that the
 853	read will succeed in the future and return AOP_TRUNCATED_PAGE.
 854	In this case, the caller should look up the folio, lock it,
 855	and call ->read_folio again.
 856
 857	Callers may invoke the ->read_folio() method directly, but using
 858	read_mapping_folio() will take care of locking, waiting for the
 859	read to complete and handle cases such as AOP_TRUNCATED_PAGE.
 860
 861``writepages``
 862	called by the VM to write out pages associated with the
 863	address_space object.  If wbc->sync_mode is WB_SYNC_ALL, then
 864	the writeback_control will specify a range of pages that must be
 865	written out.  If it is WB_SYNC_NONE, then a nr_to_write is
 866	given and that many pages should be written if possible.  If no
 867	->writepages is given, then mpage_writepages is used instead.
 868	This will choose pages from the address space that are tagged as
 869	DIRTY and will write them back.
 870
 871``dirty_folio``
 872	called by the VM to mark a folio as dirty.  This is particularly
 873	needed if an address space attaches private data to a folio, and
 874	that data needs to be updated when a folio is dirtied.  This is
 875	called, for example, when a memory mapped page gets modified.
 876	If defined, it should set the folio dirty flag, and the
 877	PAGECACHE_TAG_DIRTY search mark in i_pages.
 878
 879``readahead``
 880	Called by the VM to read pages associated with the address_space
 881	object.  The pages are consecutive in the page cache and are
 882	locked.  The implementation should decrement the page refcount
 883	after starting I/O on each page.  Usually the page will be
 884	unlocked by the I/O completion handler.  The set of pages are
 885	divided into some sync pages followed by some async pages,
 886	rac->ra->async_size gives the number of async pages.  The
 887	filesystem should attempt to read all sync pages but may decide
 888	to stop once it reaches the async pages.  If it does decide to
 889	stop attempting I/O, it can simply return.  The caller will
 890	remove the remaining pages from the address space, unlock them
 891	and decrement the page refcount.  Set PageUptodate if the I/O
 892	completes successfully.
 893
 894``write_begin``
 895	Called by the generic buffered write code to ask the filesystem
 896	to prepare to write len bytes at the given offset in the file.
 897	The address_space should check that the write will be able to
 898	complete, by allocating space if necessary and doing any other
 899	internal housekeeping.  If the write will update parts of any
 900	basic-blocks on storage, then those blocks should be pre-read
 901	(if they haven't been read already) so that the updated blocks
 902	can be written out properly.
 903
 904	The filesystem must return the locked pagecache folio for the
 905	specified offset, in ``*foliop``, for the caller to write into.
 906
 907	It must be able to cope with short writes (where the length
 908	passed to write_begin is greater than the number of bytes copied
 909	into the folio).
 910
 911	A void * may be returned in fsdata, which then gets passed into
 912	write_end.
 913
 914	Returns 0 on success; < 0 on failure (which is the error code),
 915	in which case write_end is not called.
 916
 917``write_end``
 918	After a successful write_begin, and data copy, write_end must be
 919	called.  len is the original len passed to write_begin, and
 920	copied is the amount that was able to be copied.
 921
 922	The filesystem must take care of unlocking the folio,
 923	decrementing its refcount, and updating i_size.
 924
 925	Returns < 0 on failure, otherwise the number of bytes (<=
 926	'copied') that were able to be copied into pagecache.
 927
 928``bmap``
 929	called by the VFS to map a logical block offset within object to
 930	physical block number.  This method is used by the FIBMAP ioctl
 931	and for working with swap-files.  To be able to swap to a file,
 932	the file must have a stable mapping to a block device.  The swap
 933	system does not go through the filesystem but instead uses bmap
 934	to find out where the blocks in the file are and uses those
 935	addresses directly.
 936
 937``invalidate_folio``
 938	If a folio has private data, then invalidate_folio will be
 939	called when part or all of the folio is to be removed from the
 940	address space.  This generally corresponds to either a
 941	truncation, punch hole or a complete invalidation of the address
 942	space (in the latter case 'offset' will always be 0 and 'length'
 943	will be folio_size()).  Any private data associated with the folio
 944	should be updated to reflect this truncation.  If offset is 0
 945	and length is folio_size(), then the private data should be
 946	released, because the folio must be able to be completely
 947	discarded.  This may be done by calling the ->release_folio
 948	function, but in this case the release MUST succeed.
 949
 950``release_folio``
 951	release_folio is called on folios with private data to tell the
 952	filesystem that the folio is about to be freed.  ->release_folio
 953	should remove any private data from the folio and clear the
 954	private flag.  If release_folio() fails, it should return false.
 955	release_folio() is used in two distinct though related cases.
 956	The first is when the VM wants to free a clean folio with no
 957	active users.  If ->release_folio succeeds, the folio will be
 958	removed from the address_space and be freed.
 959
 960	The second case is when a request has been made to invalidate
 961	some or all folios in an address_space.  This can happen
 962	through the fadvise(POSIX_FADV_DONTNEED) system call or by the
 963	filesystem explicitly requesting it as nfs and 9p do (when they
 964	believe the cache may be out of date with storage) by calling
 965	invalidate_inode_pages2().  If the filesystem makes such a call,
 966	and needs to be certain that all folios are invalidated, then
 967	its release_folio will need to ensure this.  Possibly it can
 968	clear the uptodate flag if it cannot free private data yet.
 969
 970``free_folio``
 971	free_folio is called once the folio is no longer visible in the
 972	page cache in order to allow the cleanup of any private data.
 973	Since it may be called by the memory reclaimer, it should not
 974	assume that the original address_space mapping still exists, and
 975	it should not block.
 976
 977``direct_IO``
 978	called by the generic read/write routines to perform direct_IO -
 979	that is IO requests which bypass the page cache and transfer
 980	data directly between the storage and the application's address
 981	space.
 982
 983``migrate_folio``
 984	This is used to compact the physical memory usage.  If the VM
 985	wants to relocate a folio (maybe from a memory device that is
 986	signalling imminent failure) it will pass a new folio and an old
 987	folio to this function.  migrate_folio should transfer any private
 988	data across and update any references that it has to the folio.
 989
 990``launder_folio``
 991	Called before freeing a folio - it writes back the dirty folio.
 992	To prevent redirtying the folio, it is kept locked during the
 993	whole operation.
 994
 995``is_partially_uptodate``
 996	Called by the VM when reading a file through the pagecache when
 997	the underlying blocksize is smaller than the size of the folio.
 998	If the required block is up to date then the read can complete
 999	without needing I/O to bring the whole page up to date.
1000
1001``is_dirty_writeback``
1002	Called by the VM when attempting to reclaim a folio.  The VM uses
1003	dirty and writeback information to determine if it needs to
1004	stall to allow flushers a chance to complete some IO.
1005	Ordinarily it can use folio_test_dirty and folio_test_writeback but
1006	some filesystems have more complex state (unstable folios in NFS
1007	prevent reclaim) or do not set those flags due to locking
1008	problems.  This callback allows a filesystem to indicate to the
1009	VM if a folio should be treated as dirty or writeback for the
1010	purposes of stalling.
1011
1012``error_remove_folio``
1013	normally set to generic_error_remove_folio if truncation is ok
1014	for this address space.  Used for memory failure handling.
1015	Setting this implies you deal with pages going away under you,
1016	unless you have them locked or reference counts increased.
1017
1018``swap_activate``
1019
1020	Called to prepare the given file for swap.  It should perform
1021	any validation and preparation necessary to ensure that writes
1022	can be performed with minimal memory allocation.  It should call
1023	add_swap_extent(), or the helper iomap_swapfile_activate(), and
1024	return the number of extents added.  If IO should be submitted
1025	through ->swap_rw(), it should set SWP_FS_OPS, otherwise IO will
1026	be submitted directly to the block device ``sis->bdev``.
1027
1028``swap_deactivate``
1029	Called during swapoff on files where swap_activate was
1030	successful.
1031
1032``swap_rw``
1033	Called to read or write swap pages when SWP_FS_OPS is set.
1034
1035The File Object
1036===============
1037
1038A file object represents a file opened by a process.  This is also known
1039as an "open file description" in POSIX parlance.
1040
1041
1042struct file_operations
1043----------------------
1044
1045This describes how the VFS can manipulate an open file.  As of kernel
10464.18, the following members are defined:
1047
1048.. code-block:: c
1049
1050	struct file_operations {
1051		struct module *owner;
1052		fop_flags_t fop_flags;
1053		loff_t (*llseek) (struct file *, loff_t, int);
1054		ssize_t (*read) (struct file *, char __user *, size_t, loff_t *);
1055		ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *);
1056		ssize_t (*read_iter) (struct kiocb *, struct iov_iter *);
1057		ssize_t (*write_iter) (struct kiocb *, struct iov_iter *);
1058		int (*iopoll)(struct kiocb *kiocb, struct io_comp_batch *,
1059				unsigned int flags);
1060		int (*iterate_shared) (struct file *, struct dir_context *);
1061		__poll_t (*poll) (struct file *, struct poll_table_struct *);
1062		long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
1063		long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
1064		int (*mmap) (struct file *, struct vm_area_struct *);
1065		int (*open) (struct inode *, struct file *);
1066		int (*flush) (struct file *, fl_owner_t id);
1067		int (*release) (struct inode *, struct file *);
1068		int (*fsync) (struct file *, loff_t, loff_t, int datasync);
1069		int (*fasync) (int, struct file *, int);
1070		int (*lock) (struct file *, int, struct file_lock *);
1071		unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
1072		int (*check_flags)(int);
1073		int (*flock) (struct file *, int, struct file_lock *);
1074		ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
1075		ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
1076		void (*splice_eof)(struct file *file);
1077		int (*setlease)(struct file *, int, struct file_lease **, void **);
1078		long (*fallocate)(struct file *file, int mode, loff_t offset,
1079				  loff_t len);
1080		void (*show_fdinfo)(struct seq_file *m, struct file *f);
1081	#ifndef CONFIG_MMU
1082		unsigned (*mmap_capabilities)(struct file *);
1083	#endif
1084		ssize_t (*copy_file_range)(struct file *, loff_t, struct file *,
1085				loff_t, size_t, unsigned int);
1086		loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in,
1087					   struct file *file_out, loff_t pos_out,
1088					   loff_t len, unsigned int remap_flags);
1089		int (*fadvise)(struct file *, loff_t, loff_t, int);
1090		int (*uring_cmd)(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
1091		int (*uring_cmd_iopoll)(struct io_uring_cmd *, struct io_comp_batch *,
1092					unsigned int poll_flags);
1093		int (*mmap_prepare)(struct vm_area_desc *);
1094	};
1095
1096Again, all methods are called without any locks being held, unless
1097otherwise noted.
1098
1099``llseek``
1100	called when the VFS needs to move the file position index
1101
1102``read``
1103	called by read(2) and related system calls
1104
1105``read_iter``
1106	possibly asynchronous read with iov_iter as destination
1107
1108``write``
1109	called by write(2) and related system calls
1110
1111``write_iter``
1112	possibly asynchronous write with iov_iter as source
1113
1114``iopoll``
1115	called when aio wants to poll for completions on HIPRI iocbs
1116
1117``iterate_shared``
1118	called when the VFS needs to read the directory contents
1119
1120``poll``
1121	called by the VFS when a process wants to check if there is
1122	activity on this file and (optionally) go to sleep until there
1123	is activity.  Called by the select(2) and poll(2) system calls
1124
1125``unlocked_ioctl``
1126	called by the ioctl(2) system call.
1127
1128``compat_ioctl``
1129	called by the ioctl(2) system call when 32 bit system calls are
1130	 used on 64 bit kernels.
1131
1132``mmap``
1133	called by the mmap(2) system call. Deprecated in favour of
1134	``mmap_prepare``.
1135
1136``open``
1137	called by the VFS when an inode should be opened.  When the VFS
1138	opens a file, it creates a new "struct file".  It then calls the
1139	open method for the newly allocated file structure.  You might
1140	think that the open method really belongs in "struct
1141	inode_operations", and you may be right.  I think it's done the
1142	way it is because it makes filesystems simpler to implement.
1143	The open() method is a good place to initialize the
1144	"private_data" member in the file structure if you want to point
1145	to a device structure
1146
1147``flush``
1148	called by the close(2) system call to flush a file
1149
1150``release``
1151	called when the last reference to an open file is closed
1152
1153``fsync``
1154	called by the fsync(2) system call.  Also see the section above
1155	entitled "Handling errors during writeback".
1156
1157``fasync``
1158	called by the fcntl(2) system call when asynchronous
1159	(non-blocking) mode is enabled for a file
1160
1161``lock``
1162	called by the fcntl(2) system call for F_GETLK, F_SETLK, and
1163	F_SETLKW commands
1164
1165``get_unmapped_area``
1166	called by the mmap(2) system call
1167
1168``check_flags``
1169	called by the fcntl(2) system call for F_SETFL command
1170
1171``flock``
1172	called by the flock(2) system call
1173
1174``splice_write``
1175	called by the VFS to splice data from a pipe to a file.  This
1176	method is used by the splice(2) system call
1177
1178``splice_read``
1179	called by the VFS to splice data from file to a pipe.  This
1180	method is used by the splice(2) system call
1181
1182``setlease``
1183	called by the VFS to set or release a file lock lease.  setlease
1184	implementations should call generic_setlease to record or remove
1185	the lease in the inode after setting it.
1186
1187``fallocate``
1188	called by the VFS to preallocate blocks or punch a hole.
1189
1190``copy_file_range``
1191	called by the copy_file_range(2) system call.
1192
1193``remap_file_range``
1194	called by the ioctl(2) system call for FICLONERANGE and FICLONE
1195	and FIDEDUPERANGE commands to remap file ranges.  An
1196	implementation should remap len bytes at pos_in of the source
1197	file into the dest file at pos_out.  Implementations must handle
1198	callers passing in len == 0; this means "remap to the end of the
1199	source file".  The return value should the number of bytes
1200	remapped, or the usual negative error code if errors occurred
1201	before any bytes were remapped.  The remap_flags parameter
1202	accepts REMAP_FILE_* flags.  If REMAP_FILE_DEDUP is set then the
1203	implementation must only remap if the requested file ranges have
1204	identical contents.  If REMAP_FILE_CAN_SHORTEN is set, the caller is
1205	ok with the implementation shortening the request length to
1206	satisfy alignment or EOF requirements (or any other reason).
1207
1208``fadvise``
1209	possibly called by the fadvise64() system call.
1210
1211``mmap_prepare``
1212	Called by the mmap(2) system call. Allows a VFS to set up a
1213	file-backed memory mapping, most notably establishing relevant
1214	private state and VMA callbacks.
1215
1216	If further action such as pre-population of page tables is required,
1217	this can be specified by the vm_area_desc->action field and related
1218	parameters.
1219
1220Note that the file operations are implemented by the specific
1221filesystem in which the inode resides.  When opening a device node
1222(character or block special) most filesystems will call special
1223support routines in the VFS which will locate the required device
1224driver information.  These support routines replace the filesystem file
1225operations with those for the device driver, and then proceed to call
1226the new open() method for the file.  This is how opening a device file
1227in the filesystem eventually ends up calling the device driver open()
1228method.
1229
1230
1231Directory Entry Cache (dcache)
1232==============================
1233
1234
1235struct dentry_operations
1236------------------------
1237
1238This describes how a filesystem can overload the standard dentry
1239operations.  Dentries and the dcache are the domain of the VFS and the
1240individual filesystem implementations.  Device drivers have no business
1241here.  These methods may be set to NULL, as they are either optional or
1242the VFS uses a default.  As of kernel 2.6.22, the following members are
1243defined:
1244
1245.. code-block:: c
1246
1247	struct dentry_operations {
1248		int (*d_revalidate)(struct inode *, const struct qstr *,
1249				    struct dentry *, unsigned int);
1250		int (*d_weak_revalidate)(struct dentry *, unsigned int);
1251		int (*d_hash)(const struct dentry *, struct qstr *);
1252		int (*d_compare)(const struct dentry *,
1253				 unsigned int, const char *, const struct qstr *);
1254		int (*d_delete)(const struct dentry *);
1255		int (*d_init)(struct dentry *);
1256		void (*d_release)(struct dentry *);
1257		void (*d_iput)(struct dentry *, struct inode *);
1258		char *(*d_dname)(struct dentry *, char *, int);
1259		struct vfsmount *(*d_automount)(struct path *);
1260		int (*d_manage)(const struct path *, bool);
1261		struct dentry *(*d_real)(struct dentry *, enum d_real_type type);
1262		bool (*d_unalias_trylock)(const struct dentry *);
1263		void (*d_unalias_unlock)(const struct dentry *);
1264	};
1265
1266``d_revalidate``
1267	called when the VFS needs to revalidate a dentry.  This is
1268	called whenever a name look-up finds a dentry in the dcache.
1269	Most local filesystems leave this as NULL, because all their
1270	dentries in the dcache are valid.  Network filesystems are
1271	different since things can change on the server without the
1272	client necessarily being aware of it.
1273
1274	This function should return a positive value if the dentry is
1275	still valid, and zero or a negative error code if it isn't.
1276
1277	d_revalidate may be called in rcu-walk mode (flags &
1278	LOOKUP_RCU).  If in rcu-walk mode, the filesystem must
1279	revalidate the dentry without blocking or storing to the dentry,
1280	d_parent and d_inode should not be used without care (because
1281	they can change and, in d_inode case, even become NULL under
1282	us).
1283
1284	If a situation is encountered that rcu-walk cannot handle,
1285	return
1286	-ECHILD and it will be called again in ref-walk mode.
1287
1288``d_weak_revalidate``
1289	called when the VFS needs to revalidate a "jumped" dentry.  This
1290	is called when a path-walk ends at dentry that was not acquired
1291	by doing a lookup in the parent directory.  This includes "/",
1292	"." and "..", as well as procfs-style symlinks and mountpoint
1293	traversal.
1294
1295	In this case, we are less concerned with whether the dentry is
1296	still fully correct, but rather that the inode is still valid.
1297	As with d_revalidate, most local filesystems will set this to
1298	NULL since their dcache entries are always valid.
1299
1300	This function has the same return code semantics as
1301	d_revalidate.
1302
1303	d_weak_revalidate is only called after leaving rcu-walk mode.
1304
1305``d_hash``
1306	called when the VFS adds a dentry to the hash table.  The first
1307	dentry passed to d_hash is the parent directory that the name is
1308	to be hashed into.
1309
1310	Same locking and synchronisation rules as d_compare regarding
1311	what is safe to dereference etc.
1312
1313``d_compare``
1314	called to compare a dentry name with a given name.  The first
1315	dentry is the parent of the dentry to be compared, the second is
1316	the child dentry.  len and name string are properties of the
1317	dentry to be compared.  qstr is the name to compare it with.
1318
1319	Must be constant and idempotent, and should not take locks if
1320	possible, and should not or store into the dentry.  Should not
1321	dereference pointers outside the dentry without lots of care
1322	(eg.  d_parent, d_inode, d_name should not be used).
1323
1324	However, our vfsmount is pinned, and RCU held, so the dentries
1325	and inodes won't disappear, neither will our sb or filesystem
1326	module.  ->d_sb may be used.
1327
1328	It is a tricky calling convention because it needs to be called
1329	under "rcu-walk", ie. without any locks or references on things.
1330
1331``d_delete``
1332	called when the last reference to a dentry is dropped and the
1333	dcache is deciding whether or not to cache it.  Return 1 to
1334	delete immediately, or 0 to cache the dentry.  Default is NULL
1335	which means to always cache a reachable dentry.  d_delete must
1336	be constant and idempotent.
1337
1338``d_init``
1339	called when a dentry is allocated
1340
1341``d_release``
1342	called when a dentry is really deallocated
1343
1344``d_iput``
1345	called when a dentry loses its inode (just prior to its being
1346	deallocated).  The default when this is NULL is that the VFS
1347	calls iput().  If you define this method, you must call iput()
1348	yourself
1349
1350``d_dname``
1351	called when the pathname of a dentry should be generated.
1352	Useful for some pseudo filesystems (sockfs, pipefs, ...) to
1353	delay pathname generation.  (Instead of doing it when dentry is
1354	created, it's done only when the path is needed.).  Real
1355	filesystems probably dont want to use it, because their dentries
1356	are present in global dcache hash, so their hash should be an
1357	invariant.  As no lock is held, d_dname() should not try to
1358	modify the dentry itself, unless appropriate SMP safety is used.
1359	CAUTION : d_path() logic is quite tricky.  The correct way to
1360	return for example "Hello" is to put it at the end of the
1361	buffer, and returns a pointer to the first char.
1362	dynamic_dname() helper function is provided to take care of
1363	this.
1364
1365	Example :
1366
1367.. code-block:: c
1368
1369	static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen)
1370	{
1371		return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
1372				dentry->d_inode->i_ino);
1373	}
1374
1375``d_automount``
1376	called when an automount dentry is to be traversed (optional).
1377	This should create a new VFS mount record and return the record
1378	to the caller.  The caller is supplied with a path parameter
1379	giving the automount directory to describe the automount target
1380	and the parent VFS mount record to provide inheritable mount
1381	parameters.  NULL should be returned if someone else managed to
1382	make the automount first.  If the vfsmount creation failed, then
1383	an error code should be returned.  If -EISDIR is returned, then
1384	the directory will be treated as an ordinary directory and
1385	returned to pathwalk to continue walking.
1386
1387	If a vfsmount is returned, the caller will attempt to mount it
1388	on the mountpoint and will remove the vfsmount from its
1389	expiration list in the case of failure.
1390
1391	This function is only used if DCACHE_NEED_AUTOMOUNT is set on
1392	the dentry.  This is set by __d_instantiate() if S_AUTOMOUNT is
1393	set on the inode being added.
1394
1395``d_manage``
1396	called to allow the filesystem to manage the transition from a
1397	dentry (optional).  This allows autofs, for example, to hold up
1398	clients waiting to explore behind a 'mountpoint' while letting
1399	the daemon go past and construct the subtree there.  0 should be
1400	returned to let the calling process continue.  -EISDIR can be
1401	returned to tell pathwalk to use this directory as an ordinary
1402	directory and to ignore anything mounted on it and not to check
1403	the automount flag.  Any other error code will abort pathwalk
1404	completely.
1405
1406	If the 'rcu_walk' parameter is true, then the caller is doing a
1407	pathwalk in RCU-walk mode.  Sleeping is not permitted in this
1408	mode, and the caller can be asked to leave it and call again by
1409	returning -ECHILD.  -EISDIR may also be returned to tell
1410	pathwalk to ignore d_automount or any mounts.
1411
1412	This function is only used if DCACHE_MANAGE_TRANSIT is set on
1413	the dentry being transited from.
1414
1415``d_real``
1416	overlay/union type filesystems implement this method to return one
1417	of the underlying dentries of a regular file hidden by the overlay.
1418
1419	The 'type' argument takes the values D_REAL_DATA or D_REAL_METADATA
1420	for returning the real underlying dentry that refers to the inode
1421	hosting the file's data or metadata respectively.
1422
1423	For non-regular files, the 'dentry' argument is returned.
1424
1425``d_unalias_trylock``
1426	if present, will be called by d_splice_alias() before moving a
1427	preexisting attached alias.  Returning false prevents __d_move(),
1428	making d_splice_alias() fail with -ESTALE.
1429
1430	Rationale: setting FS_RENAME_DOES_D_MOVE will prevent d_move()
1431	and d_exchange() calls from the outside of filesystem methods;
1432	however, it does not guarantee that attached dentries won't
1433	be renamed or moved by d_splice_alias() finding a preexisting
1434	alias for a directory inode.  Normally we would not care;
1435	however, something that wants to stabilize the entire path to
1436	root over a blocking operation might need that.  See 9p for one
1437	(and hopefully only) example.
1438
1439``d_unalias_unlock``
1440	should be paired with ``d_unalias_trylock``; that one is called after
1441	__d_move() call in __d_unalias().
1442
1443
1444Each dentry has a pointer to its parent dentry, as well as a hash list
1445of child dentries.  Child dentries are basically like files in a
1446directory.
1447
1448
1449Directory Entry Cache API
1450--------------------------
1451
1452There are a number of functions defined which permit a filesystem to
1453manipulate dentries:
1454
1455``dget``
1456	open a new handle for an existing dentry (this just increments
1457	the usage count)
1458
1459``dput``
1460	close a handle for a dentry (decrements the usage count).  If
1461	the usage count drops to 0, and the dentry is still in its
1462	parent's hash, the "d_delete" method is called to check whether
1463	it should be cached.  If it should not be cached, or if the
1464	dentry is not hashed, it is deleted.  Otherwise cached dentries
1465	are put into an LRU list to be reclaimed on memory shortage.
1466
1467``d_drop``
1468	this unhashes a dentry from its parents hash list.  A subsequent
1469	call to dput() will deallocate the dentry if its usage count
1470	drops to 0
1471
1472``d_delete``
1473	delete a dentry.  If there are no other open references to the
1474	dentry then the dentry is turned into a negative dentry (the
1475	d_iput() method is called).  If there are other references, then
1476	d_drop() is called instead
1477
1478``d_add``
1479	add a dentry to its parents hash list and then calls
1480	d_instantiate()
1481
1482``d_instantiate``
1483	add a dentry to the alias hash list for the inode and updates
1484	the "d_inode" member.  The "i_count" member in the inode
1485	structure should be set/incremented.  If the inode pointer is
1486	NULL, the dentry is called a "negative dentry".  This function
1487	is commonly called when an inode is created for an existing
1488	negative dentry
1489
1490``d_lookup``
1491	look up a dentry given its parent and path name component It
1492	looks up the child of that given name from the dcache hash
1493	table.  If it is found, the reference count is incremented and
1494	the dentry is returned.  The caller must use dput() to free the
1495	dentry when it finishes using it.
1496
1497
1498Mount Options
1499=============
1500
1501
1502Parsing options
1503---------------
1504
1505On mount and remount the filesystem is passed a string containing a
1506comma separated list of mount options.  The options can have either of
1507these forms:
1508
1509  option
1510  option=value
1511
1512The <linux/parser.h> header defines an API that helps parse these
1513options.  There are plenty of examples on how to use it in existing
1514filesystems.
1515
1516
1517Showing options
1518---------------
1519
1520If a filesystem accepts mount options, it must define show_options() to
1521show all the currently active options.  The rules are:
1522
1523  - options MUST be shown which are not default or their values differ
1524    from the default
1525
1526  - options MAY be shown which are enabled by default or have their
1527    default value
1528
1529Options used only internally between a mount helper and the kernel (such
1530as file descriptors), or which only have an effect during the mounting
1531(such as ones controlling the creation of a journal) are exempt from the
1532above rules.
1533
1534The underlying reason for the above rules is to make sure, that a mount
1535can be accurately replicated (e.g. umounting and mounting again) based
1536on the information found in /proc/mounts.
1537
1538
1539Resources
1540=========
1541
1542(Note some of these resources are not up-to-date with the latest kernel
1543 version.)
1544
1545Creating Linux virtual filesystems. 2002
1546    <https://lwn.net/Articles/13325/>
1547
1548The Linux Virtual File-system Layer by Neil Brown. 1999
1549    <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html>
1550
1551A tour of the Linux VFS by Michael K. Johnson. 1996
1552    <https://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html>
1553
1554A small trail through the Linux kernel by Andries Brouwer. 2001
1555    <https://www.win.tue.nl/~aeb/linux/vfs/trail.html>