1af96c1e3STobin C. Harding.. SPDX-License-Identifier: GPL-2.0 2af96c1e3STobin C. Harding 3af96c1e3STobin C. Harding========================================= 4af96c1e3STobin C. HardingOverview of the Linux Virtual File System 5af96c1e3STobin C. Harding========================================= 6af96c1e3STobin C. Harding 7af96c1e3STobin C. HardingOriginal author: Richard Gooch <rgooch@atnf.csiro.au> 8af96c1e3STobin C. Harding 9af96c1e3STobin C. Harding- Copyright (C) 1999 Richard Gooch 10af96c1e3STobin C. Harding- Copyright (C) 2005 Pekka Enberg 11af96c1e3STobin C. Harding 12af96c1e3STobin C. Harding 13af96c1e3STobin C. HardingIntroduction 14af96c1e3STobin C. Harding============ 15af96c1e3STobin C. Harding 16af96c1e3STobin C. HardingThe Virtual File System (also known as the Virtual Filesystem Switch) is 17af96c1e3STobin C. Hardingthe software layer in the kernel that provides the filesystem interface 18af96c1e3STobin C. Hardingto userspace programs. It also provides an abstraction within the 19af96c1e3STobin C. Hardingkernel which allows different filesystem implementations to coexist. 20af96c1e3STobin C. Harding 21af96c1e3STobin C. HardingVFS system calls open(2), stat(2), read(2), write(2), chmod(2) and so on 22af96c1e3STobin C. Hardingare called from a process context. Filesystem locking is described in 23ec23eb54SMauro Carvalho Chehabthe document Documentation/filesystems/locking.rst. 24af96c1e3STobin C. Harding 25af96c1e3STobin C. Harding 26af96c1e3STobin C. HardingDirectory Entry Cache (dcache) 27af96c1e3STobin C. Harding------------------------------ 28af96c1e3STobin C. Harding 29af96c1e3STobin C. HardingThe VFS implements the open(2), stat(2), chmod(2), and similar system 30af96c1e3STobin C. Hardingcalls. The pathname argument that is passed to them is used by the VFS 31af96c1e3STobin C. Hardingto search through the directory entry cache (also known as the dentry 32af96c1e3STobin C. Hardingcache or dcache). This provides a very fast look-up mechanism to 33af96c1e3STobin C. Hardingtranslate a pathname (filename) into a specific dentry. Dentries live 34af96c1e3STobin C. Hardingin RAM and are never saved to disc: they exist only for performance. 35af96c1e3STobin C. Harding 36af96c1e3STobin C. HardingThe dentry cache is meant to be a view into your entire filespace. As 37af96c1e3STobin C. Hardingmost computers cannot fit all dentries in the RAM at the same time, some 38af96c1e3STobin C. Hardingbits of the cache are missing. In order to resolve your pathname into a 39af96c1e3STobin C. Hardingdentry, the VFS may have to resort to creating dentries along the way, 40af96c1e3STobin C. Hardingand then loading the inode. This is done by looking up the inode. 41af96c1e3STobin C. Harding 42af96c1e3STobin C. Harding 43af96c1e3STobin C. HardingThe Inode Object 44af96c1e3STobin C. Harding---------------- 45af96c1e3STobin C. Harding 46af96c1e3STobin C. HardingAn individual dentry usually has a pointer to an inode. Inodes are 47af96c1e3STobin C. Hardingfilesystem objects such as regular files, directories, FIFOs and other 48af96c1e3STobin C. Hardingbeasts. They live either on the disc (for block device filesystems) or 49af96c1e3STobin C. Hardingin the memory (for pseudo filesystems). Inodes that live on the disc 50af96c1e3STobin C. Hardingare copied into the memory when required and changes to the inode are 51af96c1e3STobin C. Hardingwritten back to disc. A single inode can be pointed to by multiple 52af96c1e3STobin C. Hardingdentries (hard links, for example, do this). 53af96c1e3STobin C. Harding 54af96c1e3STobin C. HardingTo look up an inode requires that the VFS calls the lookup() method of 55af96c1e3STobin C. Hardingthe parent directory inode. This method is installed by the specific 56af96c1e3STobin C. Hardingfilesystem implementation that the inode lives in. Once the VFS has the 57af96c1e3STobin C. Hardingrequired dentry (and hence the inode), we can do all those boring things 58af96c1e3STobin C. Hardinglike open(2) the file, or stat(2) it to peek at the inode data. The 59af96c1e3STobin C. Hardingstat(2) operation is fairly simple: once the VFS has the dentry, it 60af96c1e3STobin C. Hardingpeeks at the inode data and passes some of it back to userspace. 61af96c1e3STobin C. Harding 62af96c1e3STobin C. Harding 63af96c1e3STobin C. HardingThe File Object 64af96c1e3STobin C. Harding--------------- 65af96c1e3STobin C. Harding 66af96c1e3STobin C. HardingOpening a file requires another operation: allocation of a file 67af96c1e3STobin C. Hardingstructure (this is the kernel-side implementation of file descriptors). 68af96c1e3STobin C. HardingThe freshly allocated file structure is initialized with a pointer to 69af96c1e3STobin C. Hardingthe dentry and a set of file operation member functions. These are 70af96c1e3STobin C. Hardingtaken from the inode data. The open() file method is then called so the 71af96c1e3STobin C. Hardingspecific filesystem implementation can do its work. You can see that 72af96c1e3STobin C. Hardingthis is another switch performed by the VFS. The file structure is 73af96c1e3STobin C. Hardingplaced into the file descriptor table for the process. 74af96c1e3STobin C. Harding 75af96c1e3STobin C. HardingReading, writing and closing files (and other assorted VFS operations) 76af96c1e3STobin C. Hardingis done by using the userspace file descriptor to grab the appropriate 77af96c1e3STobin C. Hardingfile structure, and then calling the required file structure method to 78af96c1e3STobin C. Hardingdo whatever is required. For as long as the file is open, it keeps the 79af96c1e3STobin C. Hardingdentry in use, which in turn means that the VFS inode is still in use. 80af96c1e3STobin C. Harding 81af96c1e3STobin C. Harding 82af96c1e3STobin C. HardingRegistering and Mounting a Filesystem 83af96c1e3STobin C. Harding===================================== 84af96c1e3STobin C. Harding 85af96c1e3STobin C. HardingTo register and unregister a filesystem, use the following API 86af96c1e3STobin C. Hardingfunctions: 87af96c1e3STobin C. Harding 88af96c1e3STobin C. Harding.. code-block:: c 89af96c1e3STobin C. Harding 90af96c1e3STobin C. Harding #include <linux/fs.h> 91af96c1e3STobin C. Harding 92af96c1e3STobin C. Harding extern int register_filesystem(struct file_system_type *); 93af96c1e3STobin C. Harding extern int unregister_filesystem(struct file_system_type *); 94af96c1e3STobin C. Harding 95af96c1e3STobin C. HardingThe passed struct file_system_type describes your filesystem. When a 96af96c1e3STobin C. Hardingrequest is made to mount a filesystem onto a directory in your 97af96c1e3STobin C. Hardingnamespace, the VFS will call the appropriate mount() method for the 98af96c1e3STobin C. Hardingspecific filesystem. New vfsmount referring to the tree returned by 99af96c1e3STobin C. Harding->mount() will be attached to the mountpoint, so that when pathname 100af96c1e3STobin C. Hardingresolution reaches the mountpoint it will jump into the root of that 101af96c1e3STobin C. Hardingvfsmount. 102af96c1e3STobin C. Harding 103af96c1e3STobin C. HardingYou can see all filesystems that are registered to the kernel in the 104af96c1e3STobin C. Hardingfile /proc/filesystems. 105af96c1e3STobin C. Harding 106af96c1e3STobin C. Harding 107af96c1e3STobin C. Hardingstruct file_system_type 108af96c1e3STobin C. Harding----------------------- 109af96c1e3STobin C. Harding 11085bf9a0eSAlexander MikhalitsynThis describes the filesystem. The following 111af96c1e3STobin C. Hardingmembers are defined: 112af96c1e3STobin C. Harding 113af96c1e3STobin C. Harding.. code-block:: c 114af96c1e3STobin C. Harding 1156a2195a1SLiao Pingfang struct file_system_type { 116af96c1e3STobin C. Harding const char *name; 117af96c1e3STobin C. Harding int fs_flags; 11885bf9a0eSAlexander Mikhalitsyn int (*init_fs_context)(struct fs_context *); 11985bf9a0eSAlexander Mikhalitsyn const struct fs_parameter_spec *parameters; 120af96c1e3STobin C. Harding struct dentry *(*mount) (struct file_system_type *, int, 121af96c1e3STobin C. Harding const char *, void *); 122af96c1e3STobin C. Harding void (*kill_sb) (struct super_block *); 123af96c1e3STobin C. Harding struct module *owner; 124af96c1e3STobin C. Harding struct file_system_type * next; 12585bf9a0eSAlexander Mikhalitsyn struct hlist_head fs_supers; 12685bf9a0eSAlexander Mikhalitsyn 127af96c1e3STobin C. Harding struct lock_class_key s_lock_key; 128af96c1e3STobin C. Harding struct lock_class_key s_umount_key; 12985bf9a0eSAlexander Mikhalitsyn struct lock_class_key s_vfs_rename_key; 13085bf9a0eSAlexander Mikhalitsyn struct lock_class_key s_writers_key[SB_FREEZE_LEVELS]; 13185bf9a0eSAlexander Mikhalitsyn 13285bf9a0eSAlexander Mikhalitsyn struct lock_class_key i_lock_key; 13385bf9a0eSAlexander Mikhalitsyn struct lock_class_key i_mutex_key; 13485bf9a0eSAlexander Mikhalitsyn struct lock_class_key invalidate_lock_key; 13585bf9a0eSAlexander Mikhalitsyn struct lock_class_key i_mutex_dir_key; 136af96c1e3STobin C. Harding }; 137af96c1e3STobin C. Harding 138ee5dc049STobin C. Harding``name`` 139ee5dc049STobin C. Harding the name of the filesystem type, such as "ext2", "iso9660", 140af96c1e3STobin C. Harding "msdos" and so on 141af96c1e3STobin C. Harding 142ee5dc049STobin C. Harding``fs_flags`` 143ee5dc049STobin C. Harding various flags (i.e. FS_REQUIRES_DEV, FS_NO_DCACHE, etc.) 144af96c1e3STobin C. Harding 14585bf9a0eSAlexander Mikhalitsyn``init_fs_context`` 14685bf9a0eSAlexander Mikhalitsyn Initializes 'struct fs_context' ->ops and ->fs_private fields with 14785bf9a0eSAlexander Mikhalitsyn filesystem-specific data. 14885bf9a0eSAlexander Mikhalitsyn 14985bf9a0eSAlexander Mikhalitsyn``parameters`` 15085bf9a0eSAlexander Mikhalitsyn Pointer to the array of filesystem parameters descriptors 15185bf9a0eSAlexander Mikhalitsyn 'struct fs_parameter_spec'. 15285bf9a0eSAlexander Mikhalitsyn More info in Documentation/filesystems/mount_api.rst. 15385bf9a0eSAlexander Mikhalitsyn 154ee5dc049STobin C. Harding``mount`` 155ee5dc049STobin C. Harding the method to call when a new instance of this filesystem should 156af96c1e3STobin C. Harding be mounted 157af96c1e3STobin C. Harding 158ee5dc049STobin C. Harding``kill_sb`` 159ee5dc049STobin C. Harding the method to call when an instance of this filesystem should be 160ee5dc049STobin C. Harding shut down 161af96c1e3STobin C. Harding 162af96c1e3STobin C. Harding 163ee5dc049STobin C. Harding``owner`` 164ee5dc049STobin C. Harding for internal VFS use: you should initialize this to THIS_MODULE 165ee5dc049STobin C. Harding in most cases. 166ee5dc049STobin C. Harding 167ee5dc049STobin C. Harding``next`` 168ee5dc049STobin C. Harding for internal VFS use: you should initialize this to NULL 169af96c1e3STobin C. Harding 17085bf9a0eSAlexander Mikhalitsyn``fs_supers`` 17185bf9a0eSAlexander Mikhalitsyn for internal VFS use: hlist of filesystem instances (superblocks) 17285bf9a0eSAlexander Mikhalitsyn 17385bf9a0eSAlexander Mikhalitsyn s_lock_key, s_umount_key, s_vfs_rename_key, s_writers_key, 17485bf9a0eSAlexander Mikhalitsyn i_lock_key, i_mutex_key, invalidate_lock_key, i_mutex_dir_key: lockdep-specific 175af96c1e3STobin C. Harding 176af96c1e3STobin C. HardingThe mount() method has the following arguments: 177af96c1e3STobin C. Harding 178ee5dc049STobin C. Harding``struct file_system_type *fs_type`` 179ee5dc049STobin C. Harding describes the filesystem, partly initialized by the specific 180ee5dc049STobin C. Harding filesystem code 181af96c1e3STobin C. Harding 182ee5dc049STobin C. Harding``int flags`` 183ee5dc049STobin C. Harding mount flags 184af96c1e3STobin C. Harding 185ee5dc049STobin C. Harding``const char *dev_name`` 186ee5dc049STobin C. Harding the device name we are mounting. 187af96c1e3STobin C. Harding 188ee5dc049STobin C. Harding``void *data`` 189ee5dc049STobin C. Harding arbitrary mount options, usually comes as an ASCII string (see 190ee5dc049STobin C. Harding "Mount Options" section) 191af96c1e3STobin C. Harding 192af96c1e3STobin C. HardingThe mount() method must return the root dentry of the tree requested by 193af96c1e3STobin C. Hardingcaller. An active reference to its superblock must be grabbed and the 194af96c1e3STobin C. Hardingsuperblock must be locked. On failure it should return ERR_PTR(error). 195af96c1e3STobin C. Harding 196af96c1e3STobin C. HardingThe arguments match those of mount(2) and their interpretation depends 197af96c1e3STobin C. Hardingon filesystem type. E.g. for block filesystems, dev_name is interpreted 198af96c1e3STobin C. Hardingas block device name, that device is opened and if it contains a 199af96c1e3STobin C. Hardingsuitable filesystem image the method creates and initializes struct 200af96c1e3STobin C. Hardingsuper_block accordingly, returning its root dentry to caller. 201af96c1e3STobin C. Harding 202af96c1e3STobin C. Harding->mount() may choose to return a subtree of existing filesystem - it 203af96c1e3STobin C. Hardingdoesn't have to create a new one. The main result from the caller's 204af96c1e3STobin C. Hardingpoint of view is a reference to dentry at the root of (sub)tree to be 205af96c1e3STobin C. Hardingattached; creation of new superblock is a common side effect. 206af96c1e3STobin C. Harding 207af96c1e3STobin C. HardingThe most interesting member of the superblock structure that the mount() 208af96c1e3STobin C. Hardingmethod fills in is the "s_op" field. This is a pointer to a "struct 209af96c1e3STobin C. Hardingsuper_operations" which describes the next level of the filesystem 210af96c1e3STobin C. Hardingimplementation. 211af96c1e3STobin C. Harding 212af96c1e3STobin C. HardingUsually, a filesystem uses one of the generic mount() implementations 213af96c1e3STobin C. Hardingand provides a fill_super() callback instead. The generic variants are: 214af96c1e3STobin C. Harding 215ee5dc049STobin C. Harding``mount_bdev`` 216ee5dc049STobin C. Harding mount a filesystem residing on a block device 217af96c1e3STobin C. Harding 218ee5dc049STobin C. Harding``mount_nodev`` 219ee5dc049STobin C. Harding mount a filesystem that is not backed by a device 220af96c1e3STobin C. Harding 221ee5dc049STobin C. Harding``mount_single`` 222ee5dc049STobin C. Harding mount a filesystem which shares the instance between all mounts 223af96c1e3STobin C. Harding 224af96c1e3STobin C. HardingA fill_super() callback implementation has the following arguments: 225af96c1e3STobin C. Harding 226ee5dc049STobin C. Harding``struct super_block *sb`` 227ee5dc049STobin C. Harding the superblock structure. The callback must initialize this 228ee5dc049STobin C. Harding properly. 229af96c1e3STobin C. Harding 230ee5dc049STobin C. Harding``void *data`` 231ee5dc049STobin C. Harding arbitrary mount options, usually comes as an ASCII string (see 232ee5dc049STobin C. Harding "Mount Options" section) 233af96c1e3STobin C. Harding 234ee5dc049STobin C. Harding``int silent`` 235ee5dc049STobin C. Harding whether or not to be silent on error 236af96c1e3STobin C. Harding 237af96c1e3STobin C. Harding 238af96c1e3STobin C. HardingThe Superblock Object 239af96c1e3STobin C. Harding===================== 240af96c1e3STobin C. Harding 241af96c1e3STobin C. HardingA superblock object represents a mounted filesystem. 242af96c1e3STobin C. Harding 243af96c1e3STobin C. Harding 244af96c1e3STobin C. Hardingstruct super_operations 245af96c1e3STobin C. Harding----------------------- 246af96c1e3STobin C. Harding 247af96c1e3STobin C. HardingThis describes how the VFS can manipulate the superblock of your 248592d8072SAlexander Mikhalitsynfilesystem. The following members are defined: 249af96c1e3STobin C. Harding 250af96c1e3STobin C. Harding.. code-block:: c 251af96c1e3STobin C. Harding 252af96c1e3STobin C. Harding struct super_operations { 253af96c1e3STobin C. Harding struct inode *(*alloc_inode)(struct super_block *sb); 254af96c1e3STobin C. Harding void (*destroy_inode)(struct inode *); 255592d8072SAlexander Mikhalitsyn void (*free_inode)(struct inode *); 256af96c1e3STobin C. Harding 257af96c1e3STobin C. Harding void (*dirty_inode) (struct inode *, int flags); 258592d8072SAlexander Mikhalitsyn int (*write_inode) (struct inode *, struct writeback_control *wbc); 259592d8072SAlexander Mikhalitsyn int (*drop_inode) (struct inode *); 260592d8072SAlexander Mikhalitsyn void (*evict_inode) (struct inode *); 261af96c1e3STobin C. Harding void (*put_super) (struct super_block *); 262af96c1e3STobin C. Harding int (*sync_fs)(struct super_block *sb, int wait); 263592d8072SAlexander Mikhalitsyn int (*freeze_super) (struct super_block *sb, 264af96c1e3STobin C. Harding enum freeze_holder who); 265592d8072SAlexander Mikhalitsyn int (*freeze_fs) (struct super_block *); 266af96c1e3STobin C. Harding int (*thaw_super) (struct super_block *sb, 267af96c1e3STobin C. Harding enum freeze_wholder who); 268af96c1e3STobin C. Harding int (*unfreeze_fs) (struct super_block *); 269af96c1e3STobin C. Harding int (*statfs) (struct dentry *, struct kstatfs *); 270af96c1e3STobin C. Harding int (*remount_fs) (struct super_block *, int *, char *); 271af96c1e3STobin C. Harding void (*umount_begin) (struct super_block *); 272592d8072SAlexander Mikhalitsyn 273592d8072SAlexander Mikhalitsyn int (*show_options)(struct seq_file *, struct dentry *); 274592d8072SAlexander Mikhalitsyn int (*show_devname)(struct seq_file *, struct dentry *); 275af96c1e3STobin C. Harding int (*show_path)(struct seq_file *, struct dentry *); 276af96c1e3STobin C. Harding int (*show_stats)(struct seq_file *, struct dentry *); 277af96c1e3STobin C. Harding 278592d8072SAlexander Mikhalitsyn ssize_t (*quota_read)(struct super_block *, int, char *, size_t, loff_t); 279592d8072SAlexander Mikhalitsyn ssize_t (*quota_write)(struct super_block *, int, const char *, size_t, loff_t); 280592d8072SAlexander Mikhalitsyn struct dquot **(*get_dquots)(struct inode *); 281592d8072SAlexander Mikhalitsyn 282592d8072SAlexander Mikhalitsyn long (*nr_cached_objects)(struct super_block *, 283592d8072SAlexander Mikhalitsyn struct shrink_control *); 284af96c1e3STobin C. Harding long (*free_cached_objects)(struct super_block *, 285af96c1e3STobin C. Harding struct shrink_control *); 286af96c1e3STobin C. Harding }; 287af96c1e3STobin C. Harding 288af96c1e3STobin C. HardingAll methods are called without any locks being held, unless otherwise 289af96c1e3STobin C. Hardingnoted. This means that most methods can block safely. All methods are 290af96c1e3STobin C. Hardingonly called from a process context (i.e. not from an interrupt handler 291ee5dc049STobin C. Hardingor bottom half). 292ee5dc049STobin C. Harding 293ee5dc049STobin C. Harding``alloc_inode`` 294af96c1e3STobin C. Harding this method is called by alloc_inode() to allocate memory for 295af96c1e3STobin C. Harding struct inode and initialize it. If this function is not 296af96c1e3STobin C. Harding defined, a simple 'struct inode' is allocated. Normally 297af96c1e3STobin C. Harding alloc_inode will be used to allocate a larger structure which 298ee5dc049STobin C. Harding contains a 'struct inode' embedded within it. 299ee5dc049STobin C. Harding 300ee5dc049STobin C. Harding``destroy_inode`` 301af96c1e3STobin C. Harding this method is called by destroy_inode() to release resources 302af96c1e3STobin C. Harding allocated for struct inode. It is only required if 303af96c1e3STobin C. Harding ->alloc_inode was defined and simply undoes anything done by 304592d8072SAlexander Mikhalitsyn ->alloc_inode. 305592d8072SAlexander Mikhalitsyn 306592d8072SAlexander Mikhalitsyn``free_inode`` 307592d8072SAlexander Mikhalitsyn this method is called from RCU callback. If you use call_rcu() 308592d8072SAlexander Mikhalitsyn in ->destroy_inode to free 'struct inode' memory, then it's 309ee5dc049STobin C. Harding better to release memory in this method. 310a38ed483SEric Biggers 311a38ed483SEric Biggers``dirty_inode`` 312a38ed483SEric Biggers this method is called by the VFS when an inode is marked dirty. 313a38ed483SEric Biggers This is specifically for the inode itself being marked dirty, 314cbfecb92SLukas Czerner not its data. If the update needs to be persisted by fdatasync(), 315cbfecb92SLukas Czerner then I_DIRTY_DATASYNC will be set in the flags argument. 316cbfecb92SLukas Czerner I_DIRTY_TIME will be set in the flags in case lazytime is enabled 317af96c1e3STobin C. Harding and struct inode has times updated since the last ->dirty_inode 318ee5dc049STobin C. Harding call. 319ee5dc049STobin C. Harding 320ee5dc049STobin C. Harding``write_inode`` 321ee5dc049STobin C. Harding this method is called when the VFS needs to write an inode to 322af96c1e3STobin C. Harding disc. The second parameter indicates whether the write should 323ee5dc049STobin C. Harding be synchronous or not, not all filesystems check this flag. 324ee5dc049STobin C. Harding 325ee5dc049STobin C. Harding``drop_inode`` 326af96c1e3STobin C. Harding called when the last access to the inode is dropped, with the 327af96c1e3STobin C. Harding inode->i_lock spinlock held. 328ee5dc049STobin C. Harding 329ee5dc049STobin C. Harding This method should be either NULL (normal UNIX filesystem 330af96c1e3STobin C. Harding semantics) or "generic_delete_inode" (for filesystems that do 331af96c1e3STobin C. Harding not want to cache inodes - causing "delete_inode" to always be 332ee5dc049STobin C. Harding called regardless of the value of i_nlink) 333ee5dc049STobin C. Harding 334ee5dc049STobin C. Harding The "generic_delete_inode()" behavior is equivalent to the old 335af96c1e3STobin C. Harding practice of using "force_delete" in the put_inode() case, but 336592d8072SAlexander Mikhalitsyn does not have the races that the "force_delete()" approach had. 337592d8072SAlexander Mikhalitsyn 338592d8072SAlexander Mikhalitsyn``evict_inode`` 339592d8072SAlexander Mikhalitsyn called when the VFS wants to evict an inode. Caller does 340592d8072SAlexander Mikhalitsyn *not* evict the pagecache or inode-associated metadata buffers; 341592d8072SAlexander Mikhalitsyn the method has to use truncate_inode_pages_final() to get rid 342af96c1e3STobin C. Harding of those. Caller makes sure async writeback cannot be running for 343ee5dc049STobin C. Harding the inode while (or after) ->evict_inode() is called. Optional. 344ee5dc049STobin C. Harding 345af96c1e3STobin C. Harding``put_super`` 346af96c1e3STobin C. Harding called when the VFS wishes to free the superblock 347ee5dc049STobin C. Harding (i.e. unmount). This is called with the superblock lock held 348ee5dc049STobin C. Harding 349ee5dc049STobin C. Harding``sync_fs`` 350af96c1e3STobin C. Harding called when VFS is writing out all dirty data associated with a 351af96c1e3STobin C. Harding superblock. The second parameter indicates whether the method 352592d8072SAlexander Mikhalitsyn should wait until the write out has been completed. Optional. 353592d8072SAlexander Mikhalitsyn 354592d8072SAlexander Mikhalitsyn``freeze_super`` 355592d8072SAlexander Mikhalitsyn Called instead of ->freeze_fs callback if provided. 356592d8072SAlexander Mikhalitsyn Main difference is that ->freeze_super is called without taking 357592d8072SAlexander Mikhalitsyn down_write(&sb->s_umount). If filesystem implements it and wants 358592d8072SAlexander Mikhalitsyn ->freeze_fs to be called too, then it has to call ->freeze_fs 359ee5dc049STobin C. Harding explicitly from this callback. Optional. 360ee5dc049STobin C. Harding 361ee5dc049STobin C. Harding``freeze_fs`` 362592d8072SAlexander Mikhalitsyn called when VFS is locking a filesystem and forcing it into a 363592d8072SAlexander Mikhalitsyn consistent state. This method is currently used by the Logical 364592d8072SAlexander Mikhalitsyn Volume Manager (LVM) and ioctl(FIFREEZE). Optional. 365592d8072SAlexander Mikhalitsyn 366592d8072SAlexander Mikhalitsyn``thaw_super`` 367af96c1e3STobin C. Harding called when VFS is unlocking a filesystem and making it writable 368ee5dc049STobin C. Harding again after ->freeze_super. Optional. 369ee5dc049STobin C. Harding 370592d8072SAlexander Mikhalitsyn``unfreeze_fs`` 371af96c1e3STobin C. Harding called when VFS is unlocking a filesystem and making it writable 372ee5dc049STobin C. Harding again after ->freeze_fs. Optional. 373ee5dc049STobin C. Harding 374af96c1e3STobin C. Harding``statfs`` 375ee5dc049STobin C. Harding called when the VFS needs to get filesystem statistics. 376ee5dc049STobin C. Harding 377ee5dc049STobin C. Harding``remount_fs`` 378af96c1e3STobin C. Harding called when the filesystem is remounted. This is called with 379ee5dc049STobin C. Harding the kernel lock held 380ee5dc049STobin C. Harding 381af96c1e3STobin C. Harding``umount_begin`` 382ee5dc049STobin C. Harding called when the VFS is unmounting a filesystem. 383592d8072SAlexander Mikhalitsyn 384592d8072SAlexander Mikhalitsyn``show_options`` 385ee5dc049STobin C. Harding called by the VFS to show mount options for /proc/<pid>/mounts 386af96c1e3STobin C. Harding and /proc/<pid>/mountinfo. 387592d8072SAlexander Mikhalitsyn (see "Mount Options" section) 388592d8072SAlexander Mikhalitsyn 389592d8072SAlexander Mikhalitsyn``show_devname`` 390592d8072SAlexander Mikhalitsyn Optional. Called by the VFS to show device name for 391592d8072SAlexander Mikhalitsyn /proc/<pid>/{mounts,mountinfo,mountstats}. If not provided then 392592d8072SAlexander Mikhalitsyn '(struct mount).mnt_devname' will be used. 393592d8072SAlexander Mikhalitsyn 394592d8072SAlexander Mikhalitsyn``show_path`` 395592d8072SAlexander Mikhalitsyn Optional. Called by the VFS (for /proc/<pid>/mountinfo) to show 396592d8072SAlexander Mikhalitsyn the mount root dentry path relative to the filesystem root. 397592d8072SAlexander Mikhalitsyn 398592d8072SAlexander Mikhalitsyn``show_stats`` 399592d8072SAlexander Mikhalitsyn Optional. Called by the VFS (for /proc/<pid>/mountstats) to show 400ee5dc049STobin C. Harding filesystem-specific mount statistics. 401ee5dc049STobin C. Harding 402af96c1e3STobin C. Harding``quota_read`` 403ee5dc049STobin C. Harding called by the VFS to read from filesystem quota file. 404ee5dc049STobin C. Harding 405af96c1e3STobin C. Harding``quota_write`` 406592d8072SAlexander Mikhalitsyn called by the VFS to write to filesystem quota file. 407592d8072SAlexander Mikhalitsyn 408592d8072SAlexander Mikhalitsyn``get_dquots`` 409592d8072SAlexander Mikhalitsyn called by quota to get 'struct dquot' array for a particular inode. 410ee5dc049STobin C. Harding Optional. 411ee5dc049STobin C. Harding 412ee5dc049STobin C. Harding``nr_cached_objects`` 413af96c1e3STobin C. Harding called by the sb cache shrinking function for the filesystem to 414af96c1e3STobin C. Harding return the number of freeable cached objects it contains. 415ee5dc049STobin C. Harding Optional. 416ee5dc049STobin C. Harding 417ee5dc049STobin C. Harding``free_cache_objects`` 418ee5dc049STobin C. Harding called by the sb cache shrinking function for the filesystem to 419ee5dc049STobin C. Harding scan the number of objects indicated to try to free them. 420ee5dc049STobin C. Harding Optional, but any filesystem implementing this method needs to 421af96c1e3STobin C. Harding also implement ->nr_cached_objects for it to be called 422af96c1e3STobin C. Harding correctly. 423ee5dc049STobin C. Harding 424ee5dc049STobin C. Harding We can't do anything with any errors that the filesystem might 425ee5dc049STobin C. Harding encountered, hence the void return type. This will never be 426af96c1e3STobin C. Harding called if the VM is trying to reclaim under GFP_NOFS conditions, 427ee5dc049STobin C. Harding hence this method does not need to handle that situation itself. 428ee5dc049STobin C. Harding 429ee5dc049STobin C. Harding Implementations must include conditional reschedule calls inside 430ee5dc049STobin C. Harding any scanning loop that is done. This allows the VFS to 431ee5dc049STobin C. Harding determine appropriate scan batch sizes without having to worry 432af96c1e3STobin C. Harding about whether implementations will cause holdoff problems due to 433af96c1e3STobin C. Harding large scan batch sizes. 434af96c1e3STobin C. Harding 435af96c1e3STobin C. HardingWhoever sets up the inode is responsible for filling in the "i_op" 436af96c1e3STobin C. Hardingfield. This is a pointer to a "struct inode_operations" which describes 437af96c1e3STobin C. Hardingthe methods that can be performed on individual inodes. 438af96c1e3STobin C. Harding 439af96c1e3STobin C. Harding 440af96c1e3STobin C. Hardingstruct xattr_handlers 441af96c1e3STobin C. Harding--------------------- 442af96c1e3STobin C. Harding 443af96c1e3STobin C. HardingOn filesystems that support extended attributes (xattrs), the s_xattr 444af96c1e3STobin C. Hardingsuperblock field points to a NULL-terminated array of xattr handlers. 445ee5dc049STobin C. HardingExtended attributes are name:value pairs. 446ee5dc049STobin C. Harding 447ee5dc049STobin C. Harding``name`` 448ee5dc049STobin C. Harding Indicates that the handler matches attributes with the specified 449af96c1e3STobin C. Harding name (such as "system.posix_acl_access"); the prefix field must 450ee5dc049STobin C. Harding be NULL. 451ee5dc049STobin C. Harding 452ee5dc049STobin C. Harding``prefix`` 453ee5dc049STobin C. Harding Indicates that the handler matches all attributes with the 454af96c1e3STobin C. Harding specified name prefix (such as "user."); the name field must be 455ee5dc049STobin C. Harding NULL. 456ee5dc049STobin C. Harding 457ee5dc049STobin C. Harding``list`` 458ee5dc049STobin C. Harding Determine if attributes matching this xattr handler should be 459af96c1e3STobin C. Harding listed for a particular dentry. Used by some listxattr 460ee5dc049STobin C. Harding implementations like generic_listxattr. 461ee5dc049STobin C. Harding 462ee5dc049STobin C. Harding``get`` 463ee5dc049STobin C. Harding Called by the VFS to get the value of a particular extended 464af96c1e3STobin C. Harding attribute. This method is called by the getxattr(2) system 465ee5dc049STobin C. Harding call. 466ee5dc049STobin C. Harding 467ee5dc049STobin C. Harding``set`` 4688286de7cSRandy Dunlap Called by the VFS to set the value of a particular extended 469ee5dc049STobin C. Harding attribute. When the new value is NULL, called to remove a 470af96c1e3STobin C. Harding particular extended attribute. This method is called by the 471af96c1e3STobin C. Harding setxattr(2) and removexattr(2) system calls. 472af96c1e3STobin C. Harding 473af96c1e3STobin C. HardingWhen none of the xattr handlers of a filesystem match the specified 474af96c1e3STobin C. Hardingattribute name or when a filesystem doesn't support extended attributes, 475af96c1e3STobin C. Hardingthe various ``*xattr(2)`` system calls return -EOPNOTSUPP. 476af96c1e3STobin C. Harding 477af96c1e3STobin C. Harding 478af96c1e3STobin C. HardingThe Inode Object 479af96c1e3STobin C. Harding================ 480af96c1e3STobin C. Harding 481af96c1e3STobin C. HardingAn inode object represents an object within the filesystem. 482af96c1e3STobin C. Harding 483af96c1e3STobin C. Harding 484af96c1e3STobin C. Hardingstruct inode_operations 485af96c1e3STobin C. Harding----------------------- 486af96c1e3STobin C. Harding 487af96c1e3STobin C. HardingThis describes how the VFS can manipulate an inode in your filesystem. 488af96c1e3STobin C. HardingAs of kernel 2.6.22, the following members are defined: 489af96c1e3STobin C. Harding 490af96c1e3STobin C. Harding.. code-block:: c 4916c960e68SChristian Brauner 492af96c1e3STobin C. Harding struct inode_operations { 493af96c1e3STobin C. Harding int (*create) (struct mnt_idmap *, struct inode *,struct dentry *, umode_t, bool); 494af96c1e3STobin C. Harding struct dentry * (*lookup) (struct inode *,struct dentry *, unsigned int); 4957a77db95SChristian Brauner int (*link) (struct dentry *,struct inode *,struct dentry *); 496c54bd91eSChristian Brauner int (*unlink) (struct inode *,struct dentry *); 497af96c1e3STobin C. Harding int (*symlink) (struct mnt_idmap *, struct inode *,struct dentry *,const char *); 4985ebb29beSChristian Brauner int (*mkdir) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t); 499e18275aeSChristian Brauner int (*rmdir) (struct inode *,struct dentry *); 500af96c1e3STobin C. Harding int (*mknod) (struct mnt_idmap *, struct inode *,struct dentry *,umode_t,dev_t); 501af96c1e3STobin C. Harding int (*rename) (struct mnt_idmap *, struct inode *, struct dentry *, 502af96c1e3STobin C. Harding struct inode *, struct dentry *, unsigned int); 503af96c1e3STobin C. Harding int (*readlink) (struct dentry *, char __user *,int); 5044609e1f1SChristian Brauner const char *(*get_link) (struct dentry *, struct inode *, 505cac2f8b8SChristian Brauner struct delayed_call *); 506c1632a0fSChristian Brauner int (*permission) (struct mnt_idmap *, struct inode *, int); 507b74d24f7SChristian Brauner struct posix_acl * (*get_inode_acl)(struct inode *, int, bool); 508af96c1e3STobin C. Harding int (*setattr) (struct mnt_idmap *, struct dentry *, struct iattr *); 509af96c1e3STobin C. Harding int (*getattr) (struct mnt_idmap *, const struct path *, struct kstat *, u32, unsigned int); 510af96c1e3STobin C. Harding ssize_t (*listxattr) (struct dentry *, char *, size_t); 511af96c1e3STobin C. Harding void (*update_time)(struct inode *, struct timespec *, int); 512011e2b71SChristian Brauner int (*atomic_open)(struct inode *, struct dentry *, struct file *, 51377435322SChristian Brauner unsigned open_flag, umode_t create_mode); 51413e83a49SChristian Brauner int (*tmpfile) (struct mnt_idmap *, struct inode *, struct file *, umode_t); 5158782a9aeSChristian Brauner struct posix_acl * (*get_acl)(struct mnt_idmap *, struct dentry *, int); 5164c5b4799SMiklos Szeredi int (*set_acl)(struct mnt_idmap *, struct dentry *, struct posix_acl *, int); 5174c5b4799SMiklos Szeredi int (*fileattr_set)(struct mnt_idmap *idmap, 518af96c1e3STobin C. Harding struct dentry *dentry, struct fileattr *fa); 519af96c1e3STobin C. Harding int (*fileattr_get)(struct dentry *dentry, struct fileattr *fa); 520af96c1e3STobin C. Harding struct offset_ctx *(*get_offset_ctx)(struct inode *inode); 521af96c1e3STobin C. Harding }; 522af96c1e3STobin C. Harding 523ee5dc049STobin C. HardingAgain, all methods are called without any locks being held, unless 524ee5dc049STobin C. Hardingotherwise noted. 525ee5dc049STobin C. Harding 526ee5dc049STobin C. Harding``create`` 527ee5dc049STobin C. Harding called by the open(2) and creat(2) system calls. Only required 528ee5dc049STobin C. Harding if you want to support regular files. The dentry you get should 529af96c1e3STobin C. Harding not have an inode (i.e. it should be a negative dentry). Here 530ee5dc049STobin C. Harding you will probably call d_instantiate() with the dentry and the 531ee5dc049STobin C. Harding newly created inode 532af96c1e3STobin C. Harding 533af96c1e3STobin C. Harding``lookup`` 534af96c1e3STobin C. Harding called when the VFS needs to look up an inode in a parent 535af96c1e3STobin C. Harding directory. The name to look for is found in the dentry. This 536af96c1e3STobin C. Harding method must call d_add() to insert the found inode into the 537ee5dc049STobin C. Harding dentry. The "i_count" field in the inode structure should be 538ee5dc049STobin C. Harding incremented. If the named inode does not exist a NULL inode 539af96c1e3STobin C. Harding should be inserted into the dentry (this is called a negative 540af96c1e3STobin C. Harding dentry). Returning an error code from this routine must only be 541ee5dc049STobin C. Harding done on a real error, otherwise creating inodes with system 542ee5dc049STobin C. Harding calls like create(2), mknod(2), mkdir(2) and so on will fail. 543ee5dc049STobin C. Harding If you wish to overload the dentry methods then you should 544af96c1e3STobin C. Harding initialise the "d_dop" field in the dentry; this is a pointer to 545ee5dc049STobin C. Harding a struct "dentry_operations". This method is called with the 546ee5dc049STobin C. Harding directory inode semaphore held 547ee5dc049STobin C. Harding 548af96c1e3STobin C. Harding``link`` 549af96c1e3STobin C. Harding called by the link(2) system call. Only required if you want to 550ee5dc049STobin C. Harding support hard links. You will probably need to call 551ee5dc049STobin C. Harding d_instantiate() just as you would in the create() method 552ee5dc049STobin C. Harding 553af96c1e3STobin C. Harding``unlink`` 554ee5dc049STobin C. Harding called by the unlink(2) system call. Only required if you want 555ee5dc049STobin C. Harding to support deleting inodes 556ee5dc049STobin C. Harding 557af96c1e3STobin C. Harding``symlink`` 558af96c1e3STobin C. Harding called by the symlink(2) system call. Only required if you want 559ee5dc049STobin C. Harding to support symlinks. You will probably need to call 560ee5dc049STobin C. Harding d_instantiate() just as you would in the create() method 561af96c1e3STobin C. Harding 562af96c1e3STobin C. Harding``mkdir`` 563af96c1e3STobin C. Harding called by the mkdir(2) system call. Only required if you want 564ee5dc049STobin C. Harding to support creating subdirectories. You will probably need to 565ee5dc049STobin C. Harding call d_instantiate() just as you would in the create() method 566af96c1e3STobin C. Harding 567af96c1e3STobin C. Harding``rmdir`` 568ee5dc049STobin C. Harding called by the rmdir(2) system call. Only required if you want 569ee5dc049STobin C. Harding to support deleting subdirectories 570ee5dc049STobin C. Harding 571ee5dc049STobin C. Harding``mknod`` 572ee5dc049STobin C. Harding called by the mknod(2) system call to create a device (char, 573ee5dc049STobin C. Harding block) inode or a named pipe (FIFO) or socket. Only required if 574af96c1e3STobin C. Harding you want to support creating these types of inodes. You will 575ee5dc049STobin C. Harding probably need to call d_instantiate() just as you would in the 576ee5dc049STobin C. Harding create() method 577ee5dc049STobin C. Harding 578af96c1e3STobin C. Harding``rename`` 579af96c1e3STobin C. Harding called by the rename(2) system call to rename the object to have 580af96c1e3STobin C. Harding the parent and name given by the second inode and dentry. 581ee5dc049STobin C. Harding 582ee5dc049STobin C. Harding The filesystem must return -EINVAL for any unsupported or 583ee5dc049STobin C. Harding unknown flags. Currently the following flags are implemented: 584ee5dc049STobin C. Harding (1) RENAME_NOREPLACE: this flag indicates that if the target of 585ee5dc049STobin C. Harding the rename exists the rename should fail with -EEXIST instead of 586af96c1e3STobin C. Harding replacing the target. The VFS already checks for existence, so 587ee5dc049STobin C. Harding for local filesystems the RENAME_NOREPLACE implementation is 588ee5dc049STobin C. Harding equivalent to plain rename. 589af96c1e3STobin C. Harding (2) RENAME_EXCHANGE: exchange source and target. Both must 590ee5dc049STobin C. Harding exist; this is checked by the VFS. Unlike plain rename, source 591ee5dc049STobin C. Harding and target may be of different type. 592ee5dc049STobin C. Harding 593ee5dc049STobin C. Harding``get_link`` 594ee5dc049STobin C. Harding called by the VFS to follow a symbolic link to the inode it 595ee5dc049STobin C. Harding points to. Only required if you want to support symbolic links. 596ee5dc049STobin C. Harding This method returns the symlink body to traverse (and possibly 597ee5dc049STobin C. Harding resets the current position with nd_jump_link()). If the body 598ee5dc049STobin C. Harding won't go away until the inode is gone, nothing else is needed; 599ee5dc049STobin C. Harding if it needs to be otherwise pinned, arrange for its release by 600ee5dc049STobin C. Harding having get_link(..., ..., done) do set_delayed_call(done, 601af96c1e3STobin C. Harding destructor, argument). In that case destructor(argument) will 602af96c1e3STobin C. Harding be called once VFS is done with the body you've returned. May 603af96c1e3STobin C. Harding be called in RCU mode; that is indicated by NULL dentry 604af96c1e3STobin C. Harding argument. If request can't be handled without leaving RCU mode, 605af96c1e3STobin C. Harding have it return ERR_PTR(-ECHILD). 606af96c1e3STobin C. Harding 607af96c1e3STobin C. Harding If the filesystem stores the symlink target in ->i_link, the 608af96c1e3STobin C. Harding VFS may use it directly without calling ->get_link(); however, 609af96c1e3STobin C. Harding ->get_link() must still be provided. ->i_link must not be 610ee5dc049STobin C. Harding freed until after an RCU grace period. Writing to ->i_link 611ee5dc049STobin C. Harding post-iget() time requires a 'release' memory barrier. 612af96c1e3STobin C. Harding 613af96c1e3STobin C. Harding``readlink`` 614af96c1e3STobin C. Harding this is now just an override for use by readlink(2) for the 615af96c1e3STobin C. Harding cases when ->get_link uses nd_jump_link() or object is not in 616af96c1e3STobin C. Harding fact a symlink. Normally filesystems should only implement 617ee5dc049STobin C. Harding ->get_link for symlinks and readlink(2) will automatically use 618ee5dc049STobin C. Harding that. 619af96c1e3STobin C. Harding 620af96c1e3STobin C. Harding``permission`` 621ee5dc049STobin C. Harding called by the VFS to check for access rights on a POSIX-like 622ee5dc049STobin C. Harding filesystem. 623ee5dc049STobin C. Harding 624af96c1e3STobin C. Harding May be called in rcu-walk mode (mask & MAY_NOT_BLOCK). If in 625ee5dc049STobin C. Harding rcu-walk mode, the filesystem must check the permission without 626ee5dc049STobin C. Harding blocking or storing to the inode. 627af96c1e3STobin C. Harding 628af96c1e3STobin C. Harding If a situation is encountered that rcu-walk cannot handle, 629ee5dc049STobin C. Harding return 630ee5dc049STobin C. Harding -ECHILD and it will be called again in ref-walk mode. 631ee5dc049STobin C. Harding 632af96c1e3STobin C. Harding``setattr`` 633ee5dc049STobin C. Harding called by the VFS to set attributes for a file. This method is 634ee5dc049STobin C. Harding called by chmod(2) and related system calls. 635ee5dc049STobin C. Harding 636af96c1e3STobin C. Harding``getattr`` 637ee5dc049STobin C. Harding called by the VFS to get attributes of a file. This method is 638ee5dc049STobin C. Harding called by stat(2) and related system calls. 639ee5dc049STobin C. Harding 640af96c1e3STobin C. Harding``listxattr`` 641ee5dc049STobin C. Harding called by the VFS to list all extended attributes for a given 642ee5dc049STobin C. Harding file. This method is called by the listxattr(2) system call. 643ee5dc049STobin C. Harding 644ee5dc049STobin C. Harding``update_time`` 645af96c1e3STobin C. Harding called by the VFS to update a specific time or the i_version of 646ee5dc049STobin C. Harding an inode. If this is not defined the VFS will update the inode 647ee5dc049STobin C. Harding itself and call mark_inode_dirty_sync. 648ee5dc049STobin C. Harding 649ee5dc049STobin C. Harding``atomic_open`` 650ee5dc049STobin C. Harding called on the last component of an open. Using this optional 651ee5dc049STobin C. Harding method the filesystem can look up, possibly create and open the 652ee5dc049STobin C. Harding file in one atomic operation. If it wants to leave actual 653ee5dc049STobin C. Harding opening to the caller (e.g. if the file turned out to be a 654ee5dc049STobin C. Harding symlink, device, or just something filesystem won't do atomic 655ee5dc049STobin C. Harding open for), it may signal this by returning finish_no_open(file, 656ee5dc049STobin C. Harding dentry). This method is only called if the last component is 657ee5dc049STobin C. Harding negative or needs lookup. Cached positive dentries are still 658ee5dc049STobin C. Harding handled by f_op->open(). If the file was created, FMODE_CREATED 659af96c1e3STobin C. Harding flag should be set in file->f_mode. In case of O_EXCL the 660ee5dc049STobin C. Harding method must only succeed if the file didn't exist and hence 661ee5dc049STobin C. Harding FMODE_CREATED shall always be set on success. 662ee5dc049STobin C. Harding 663863f144fSMiklos Szeredi``tmpfile`` 664863f144fSMiklos Szeredi called in the end of O_TMPFILE open(). Optional, equivalent to 665863f144fSMiklos Szeredi atomically creating, opening and unlinking a file in given 666af96c1e3STobin C. Harding directory. On success needs to return with the file already 6674c5b4799SMiklos Szeredi open; this can be done by calling finish_open_simple() right at 6684c5b4799SMiklos Szeredi the end. 6694c5b4799SMiklos Szeredi 6704c5b4799SMiklos Szeredi``fileattr_get`` 6714c5b4799SMiklos Szeredi called on ioctl(FS_IOC_GETFLAGS) and ioctl(FS_IOC_FSGETXATTR) to 6724c5b4799SMiklos Szeredi retrieve miscellaneous file flags and attributes. Also called 6734c5b4799SMiklos Szeredi before the relevant SET operation to check what is being changed 6744c5b4799SMiklos Szeredi (in this case with i_rwsem locked exclusive). If unset, then 6754c5b4799SMiklos Szeredi fall back to f_op->ioctl(). 6764c5b4799SMiklos Szeredi 6774c5b4799SMiklos Szeredi``fileattr_set`` 6784c5b4799SMiklos Szeredi called on ioctl(FS_IOC_SETFLAGS) and ioctl(FS_IOC_FSSETXATTR) to 679af96c1e3STobin C. Harding change miscellaneous file flags and attributes. Callers hold 680af96c1e3STobin C. Harding i_rwsem exclusive. If unset, then fall back to f_op->ioctl(). 681af96c1e3STobin C. Harding``get_offset_ctx`` 682af96c1e3STobin C. Harding called to get the offset context for a directory inode. A 683af96c1e3STobin C. Harding filesystem must define this operation to use 684af96c1e3STobin C. Harding simple_offset_dir_operations. 685af96c1e3STobin C. Harding 686af96c1e3STobin C. HardingThe Address Space Object 687af96c1e3STobin C. Harding======================== 688af96c1e3STobin C. Harding 689af96c1e3STobin C. HardingThe address space object is used to group and manage pages in the page 690af96c1e3STobin C. Hardingcache. It can be used to keep track of the pages in a file (or anything 691af96c1e3STobin C. Hardingelse) and also track the mapping of sections of the file into process 692af96c1e3STobin C. Hardingaddress spaces. 693af96c1e3STobin C. Harding 694af96c1e3STobin C. HardingThere are a number of distinct yet related services that an 695af96c1e3STobin C. Hardingaddress-space can provide. These include communicating memory pressure, 696fa29000bSMatthew Wilcox (Oracle)page lookup by address, and keeping track of pages tagged as Dirty or 697fa29000bSMatthew Wilcox (Oracle)Writeback. 698fa29000bSMatthew Wilcox (Oracle) 699af96c1e3STobin C. HardingThe first can be used independently to the others. The VM can try to 700af96c1e3STobin C. Hardingeither write dirty pages in order to clean them, or release clean pages 701af96c1e3STobin C. Hardingin order to reuse them. To do this it can call the ->writepage method 702af96c1e3STobin C. Hardingon dirty pages, and ->release_folio on clean folios with the private 703af96c1e3STobin C. Hardingflag set. Clean pages without PagePrivate and with no external references 704af96c1e3STobin C. Hardingwill be released without notice being given to the address_space. 705af96c1e3STobin C. Harding 706af96c1e3STobin C. HardingTo achieve this functionality, pages need to be placed on an LRU with 707af96c1e3STobin C. Hardinglru_cache_add and mark_page_active needs to be called whenever the page 708af96c1e3STobin C. Hardingis used. 709af96c1e3STobin C. Harding 710af96c1e3STobin C. HardingPages are normally kept in a radix tree index by ->index. This tree 711af96c1e3STobin C. Hardingmaintains information about the PG_Dirty and PG_Writeback status of each 712af96c1e3STobin C. Hardingpage, so that pages with either of these flags can be found quickly. 713af96c1e3STobin C. Harding 714af96c1e3STobin C. HardingThe Dirty tag is primarily used by mpage_writepages - the default 715af96c1e3STobin C. Harding->writepages method. It uses the tag to find dirty pages to call 716af96c1e3STobin C. Harding->writepage on. If mpage_writepages is not used (i.e. the address 717af96c1e3STobin C. Hardingprovides its own ->writepages) , the PAGECACHE_TAG_DIRTY tag is almost 718af96c1e3STobin C. Hardingunused. write_inode_now and sync_inode do use it (through 719af96c1e3STobin C. Harding__sync_single_inode) to check if ->writepages has been successful in 720af96c1e3STobin C. Hardingwriting out the whole address_space. 721af96c1e3STobin C. Harding 722af96c1e3STobin C. HardingThe Writeback tag is used by filemap*wait* and sync_page* functions, via 723af96c1e3STobin C. Hardingfilemap_fdatawait_range, to wait for all writeback to complete. 724af96c1e3STobin C. Harding 725af96c1e3STobin C. HardingAn address_space handler may attach extra information to a page, 726af96c1e3STobin C. Hardingtypically using the 'private' field in the 'struct page'. If such 727af96c1e3STobin C. Hardinginformation is attached, the PG_Private flag should be set. This will 728af96c1e3STobin C. Hardingcause various VM routines to make extra calls into the address_space 729af96c1e3STobin C. Hardinghandler to deal with that data. 730af96c1e3STobin C. Harding 731af96c1e3STobin C. HardingAn address space acts as an intermediate between storage and 73208830c8bSMatthew Wilcox (Oracle)application. Data is read into the address space a whole page at a 733af96c1e3STobin C. Hardingtime, and provided to the application either by copying of the page, or 7346f31a5a2SMatthew Wilcox (Oracle)by memory-mapping the page. Data is written into the address space by 735af96c1e3STobin C. Hardingthe application, and then written-back to storage typically in whole 736af96c1e3STobin C. Hardingpages, however the address_space has finer control of write sizes. 737af96c1e3STobin C. Harding 738af96c1e3STobin C. HardingThe read process essentially only requires 'read_folio'. The write 739af96c1e3STobin C. Hardingprocess is more complicated and uses write_begin/write_end or 740af96c1e3STobin C. Hardingdirty_folio to write data into the address_space, and writepage and 741af96c1e3STobin C. Hardingwritepages to writeback data to storage. 742af96c1e3STobin C. Harding 743af96c1e3STobin C. HardingAdding and removing pages to/from an address_space is protected by the 744af96c1e3STobin C. Hardinginode's i_mutex. 745af96c1e3STobin C. Harding 746af96c1e3STobin C. HardingWhen data is written to a page, the PG_Dirty flag should be set. It 7478286de7cSRandy Dunlaptypically remains set until writepage asks for it to be written. This 748af96c1e3STobin C. Hardingshould clear PG_Dirty and set PG_Writeback. It can be actually written 749af96c1e3STobin C. Hardingat any point after PG_Dirty is clear. Once it is known to be safe, 750af96c1e3STobin C. HardingPG_Writeback is cleared. 751af96c1e3STobin C. Harding 752af96c1e3STobin C. HardingWriteback makes use of a writeback_control structure to direct the 753af96c1e3STobin C. Hardingoperations. This gives the writepage and writepages operations some 754af96c1e3STobin C. Hardinginformation about the nature of and reason for the writeback request, 755af96c1e3STobin C. Hardingand the constraints under which it is being done. It is also used to 756af96c1e3STobin C. Hardingreturn information back to the caller about the result of a writepage or 757af96c1e3STobin C. Hardingwritepages request. 758af96c1e3STobin C. Harding 759af96c1e3STobin C. Harding 760af96c1e3STobin C. HardingHandling errors during writeback 761af96c1e3STobin C. Harding-------------------------------- 762af96c1e3STobin C. Harding 763af96c1e3STobin C. HardingMost applications that do buffered I/O will periodically call a file 764*d56b699dSBjorn Helgaassynchronization call (fsync, fdatasync, msync or sync_file_range) to 765af96c1e3STobin C. Hardingensure that data written has made it to the backing store. When there 766af96c1e3STobin C. Hardingis an error during writeback, they expect that error to be reported when 767af96c1e3STobin C. Hardinga file sync request is made. After an error has been reported on one 768af96c1e3STobin C. Hardingrequest, subsequent requests on the same file descriptor should return 769af96c1e3STobin C. Harding0, unless further writeback errors have occurred since the previous file 770af96c1e3STobin C. Hardingsynchronization. 771af96c1e3STobin C. Harding 772af96c1e3STobin C. HardingIdeally, the kernel would report errors only on file descriptions on 773af96c1e3STobin C. Hardingwhich writes were done that subsequently failed to be written back. The 774af96c1e3STobin C. Hardinggeneric pagecache infrastructure does not track the file descriptions 775af96c1e3STobin C. Hardingthat have dirtied each individual page however, so determining which 776af96c1e3STobin C. Hardingfile descriptors should get back an error is not possible. 777af96c1e3STobin C. Harding 778af96c1e3STobin C. HardingInstead, the generic writeback error tracking infrastructure in the 779af96c1e3STobin C. Hardingkernel settles for reporting errors to fsync on all file descriptions 780af96c1e3STobin C. Hardingthat were open at the time that the error occurred. In a situation with 781af96c1e3STobin C. Hardingmultiple writers, all of them will get back an error on a subsequent 782af96c1e3STobin C. Hardingfsync, even if all of the writes done through that particular file 783af96c1e3STobin C. Hardingdescriptor succeeded (or even if there were no writes on that file 784af96c1e3STobin C. Hardingdescriptor at all). 785af96c1e3STobin C. Harding 786af96c1e3STobin C. HardingFilesystems that wish to use this infrastructure should call 787af96c1e3STobin C. Hardingmapping_set_error to record the error in the address_space when it 788af96c1e3STobin C. Hardingoccurs. Then, after writing back data from the pagecache in their 789af96c1e3STobin C. Hardingfile->fsync operation, they should call file_check_and_advance_wb_err to 790af96c1e3STobin C. Hardingensure that the struct file's error cursor has advanced to the correct 791af96c1e3STobin C. Hardingpoint in the stream of errors emitted by the backing device(s). 792af96c1e3STobin C. Harding 793af96c1e3STobin C. Harding 794af96c1e3STobin C. Hardingstruct address_space_operations 795af96c1e3STobin C. Harding------------------------------- 796af96c1e3STobin C. Harding 797af96c1e3STobin C. HardingThis describes how the VFS can manipulate mapping of a file to page 79808830c8bSMatthew Wilcox (Oracle)cache in your filesystem. The following members are defined: 799af96c1e3STobin C. Harding 8006f31a5a2SMatthew Wilcox (Oracle).. code-block:: c 8018151b4c8SMatthew Wilcox (Oracle) 802af96c1e3STobin C. Harding struct address_space_operations { 8039d6b0cd7SMatthew Wilcox (Oracle) int (*writepage)(struct page *page, struct writeback_control *wbc); 804af96c1e3STobin C. Harding int (*read_folio)(struct file *, struct folio *); 805af96c1e3STobin C. Harding int (*writepages)(struct address_space *, struct writeback_control *); 806af96c1e3STobin C. Harding bool (*dirty_folio)(struct address_space *, struct folio *); 807af96c1e3STobin C. Harding void (*readahead)(struct readahead_control *); 808af96c1e3STobin C. Harding int (*write_begin)(struct file *, struct address_space *mapping, 809128d1f82SMatthew Wilcox (Oracle) loff_t pos, unsigned len, 810fa29000bSMatthew Wilcox (Oracle) struct page **pagep, void **fsdata); 811d2329aa0SMatthew Wilcox (Oracle) int (*write_end)(struct file *, struct address_space *mapping, 812af96c1e3STobin C. Harding loff_t pos, unsigned len, unsigned copied, 8135490da4fSMatthew Wilcox (Oracle) struct page *page, void *fsdata); 8145490da4fSMatthew Wilcox (Oracle) sector_t (*bmap)(struct address_space *, sector_t); 815affa80e8SMatthew Wilcox (Oracle) void (*invalidate_folio) (struct folio *, size_t start, size_t len); 816af96c1e3STobin C. Harding bool (*release_folio)(struct folio *, gfp_t); 8172e7e80f7SMatthew Wilcox (Oracle) void (*free_folio)(struct folio *); 8182e7e80f7SMatthew Wilcox (Oracle) ssize_t (*direct_IO)(struct kiocb *, struct iov_iter *iter); 819520f301cSMatthew Wilcox (Oracle) int (*migrate_folio)(struct mapping *, struct folio *dst, 820af96c1e3STobin C. Harding struct folio *src, enum migrate_mode); 821cba738f6SNeilBrown int (*launder_folio) (struct folio *); 822af96c1e3STobin C. Harding 823cba738f6SNeilBrown bool (*is_partially_uptodate) (struct folio *, size_t from, 824af96c1e3STobin C. Harding size_t count); 825af96c1e3STobin C. Harding void (*is_dirty_writeback)(struct folio *, bool *, bool *); 826ee5dc049STobin C. Harding int (*error_remove_page) (struct mapping *mapping, struct page *page); 827ee5dc049STobin C. Harding int (*swap_activate)(struct swap_info_struct *sis, struct file *f, sector_t *span) 828ee5dc049STobin C. Harding int (*swap_deactivate)(struct file *); 829ee5dc049STobin C. Harding int (*swap_rw)(struct kiocb *iocb, struct iov_iter *iter); 830ee5dc049STobin C. Harding }; 831ee5dc049STobin C. Harding 832ee5dc049STobin C. Harding``writepage`` 833ee5dc049STobin C. Harding called by the VM to write a dirty page to backing store. This 834ee5dc049STobin C. Harding may happen for data integrity reasons (i.e. 'sync'), or to free 835af96c1e3STobin C. Harding up memory (flush). The difference can be seen in 836af96c1e3STobin C. Harding wbc->sync_mode. The PG_Dirty flag has been cleared and 837af96c1e3STobin C. Harding PageLocked is true. writepage should start writeout, should set 838af96c1e3STobin C. Harding PG_Writeback, and should make sure the page is unlocked, either 839af96c1e3STobin C. Harding synchronously or asynchronously when the write operation 840ee5dc049STobin C. Harding completes. 841ee5dc049STobin C. Harding 842af96c1e3STobin C. Harding If wbc->sync_mode is WB_SYNC_NONE, ->writepage doesn't have to 843af96c1e3STobin C. Harding try too hard if there are problems, and may choose to write out 844af96c1e3STobin C. Harding other pages from the mapping if that is easier (e.g. due to 84508830c8bSMatthew Wilcox (Oracle) internal dependencies). If it chooses not to start writeout, it 84690c02eb9SMatthew Wilcox (Oracle) should return AOP_WRITEPAGE_ACTIVATE so that the VM will not 84790c02eb9SMatthew Wilcox (Oracle) keep calling ->writepage on that page. 84890c02eb9SMatthew Wilcox (Oracle) 84990c02eb9SMatthew Wilcox (Oracle) See the file "Locking" for more details. 85090c02eb9SMatthew Wilcox (Oracle) 85190c02eb9SMatthew Wilcox (Oracle)``read_folio`` 85290c02eb9SMatthew Wilcox (Oracle) Called by the page cache to read a folio from the backing store. 85390c02eb9SMatthew Wilcox (Oracle) The 'file' argument supplies authentication information to network 85490c02eb9SMatthew Wilcox (Oracle) filesystems, and is generally not used by block based filesystems. 85590c02eb9SMatthew Wilcox (Oracle) It may be NULL if the caller does not have an open file (eg if 85690c02eb9SMatthew Wilcox (Oracle) the kernel is performing a read for itself rather than on behalf 85790c02eb9SMatthew Wilcox (Oracle) of a userspace process with an open file). 85890c02eb9SMatthew Wilcox (Oracle) 85990c02eb9SMatthew Wilcox (Oracle) If the mapping does not support large folios, the folio will 86090c02eb9SMatthew Wilcox (Oracle) contain a single page. The folio will be locked when read_folio 86190c02eb9SMatthew Wilcox (Oracle) is called. If the read completes successfully, the folio should 86290c02eb9SMatthew Wilcox (Oracle) be marked uptodate. The filesystem should unlock the folio 86390c02eb9SMatthew Wilcox (Oracle) once the read has completed, whether it was successful or not. 86490c02eb9SMatthew Wilcox (Oracle) The filesystem does not need to modify the refcount on the folio; 86590c02eb9SMatthew Wilcox (Oracle) the page cache holds a reference count and that will not be 86690c02eb9SMatthew Wilcox (Oracle) released until the folio is unlocked. 86790c02eb9SMatthew Wilcox (Oracle) 86890c02eb9SMatthew Wilcox (Oracle) Filesystems may implement ->read_folio() synchronously. 86990c02eb9SMatthew Wilcox (Oracle) In normal operation, folios are read through the ->readahead() 87090c02eb9SMatthew Wilcox (Oracle) method. Only if this fails, or if the caller needs to wait for 87190c02eb9SMatthew Wilcox (Oracle) the read to complete will the page cache call ->read_folio(). 87290c02eb9SMatthew Wilcox (Oracle) Filesystems should not attempt to perform their own readahead 87390c02eb9SMatthew Wilcox (Oracle) in the ->read_folio() operation. 87490c02eb9SMatthew Wilcox (Oracle) 87590c02eb9SMatthew Wilcox (Oracle) If the filesystem cannot perform the read at this time, it can 87690c02eb9SMatthew Wilcox (Oracle) unlock the folio, do whatever action it needs to ensure that the 87790c02eb9SMatthew Wilcox (Oracle) read will succeed in the future and return AOP_TRUNCATED_PAGE. 878af96c1e3STobin C. Harding In this case, the caller should look up the folio, lock it, 879ee5dc049STobin C. Harding and call ->read_folio again. 880ee5dc049STobin C. Harding 881e9b2f15bSJulia Lawall Callers may invoke the ->read_folio() method directly, but using 882af96c1e3STobin C. Harding read_mapping_folio() will take care of locking, waiting for the 883e9b2f15bSJulia Lawall read to complete and handle cases such as AOP_TRUNCATED_PAGE. 884ee5dc049STobin C. Harding 885ee5dc049STobin C. Harding``writepages`` 886ee5dc049STobin C. Harding called by the VM to write out pages associated with the 887ee5dc049STobin C. Harding address_space object. If wbc->sync_mode is WB_SYNC_ALL, then 888af96c1e3STobin C. Harding the writeback_control will specify a range of pages that must be 8896f31a5a2SMatthew Wilcox (Oracle) written out. If it is WB_SYNC_NONE, then a nr_to_write is 8906f31a5a2SMatthew Wilcox (Oracle) given and that many pages should be written if possible. If no 8916f31a5a2SMatthew Wilcox (Oracle) ->writepages is given, then mpage_writepages is used instead. 8926f31a5a2SMatthew Wilcox (Oracle) This will choose pages from the address space that are tagged as 893ee5dc049STobin C. Harding DIRTY and will pass them to ->writepage. 8946f31a5a2SMatthew Wilcox (Oracle) 8956f31a5a2SMatthew Wilcox (Oracle)``dirty_folio`` 896af96c1e3STobin C. Harding called by the VM to mark a folio as dirty. This is particularly 8978151b4c8SMatthew Wilcox (Oracle) needed if an address space attaches private data to a folio, and 8988151b4c8SMatthew Wilcox (Oracle) that data needs to be updated when a folio is dirtied. This is 8998151b4c8SMatthew Wilcox (Oracle) called, for example, when a memory mapped page gets modified. 9008151b4c8SMatthew Wilcox (Oracle) If defined, it should set the folio dirty flag, and the 9018151b4c8SMatthew Wilcox (Oracle) PAGECACHE_TAG_DIRTY search mark in i_pages. 90284dacdbdSNeilBrown 90384dacdbdSNeilBrown``readahead`` 90484dacdbdSNeilBrown Called by the VM to read pages associated with the address_space 90584dacdbdSNeilBrown object. The pages are consecutive in the page cache and are 90684dacdbdSNeilBrown locked. The implementation should decrement the page refcount 90784dacdbdSNeilBrown after starting I/O on each page. Usually the page will be 90884dacdbdSNeilBrown unlocked by the I/O completion handler. The set of pages are 90984dacdbdSNeilBrown divided into some sync pages followed by some async pages, 91084dacdbdSNeilBrown rac->ra->async_size gives the number of async pages. The 91184dacdbdSNeilBrown filesystem should attempt to read all sync pages but may decide 9128151b4c8SMatthew Wilcox (Oracle) to stop once it reaches the async pages. If it does decide to 913ee5dc049STobin C. Harding stop attempting I/O, it can simply return. The caller will 914ee5dc049STobin C. Harding remove the remaining pages from the address space, unlock them 915ee5dc049STobin C. Harding and decrement the page refcount. Set PageUptodate if the I/O 916ee5dc049STobin C. Harding completes successfully. Setting PageError on any page will be 917ee5dc049STobin C. Harding ignored; simply unlock the page if an I/O error occurs. 918ee5dc049STobin C. Harding 919ee5dc049STobin C. Harding``write_begin`` 920ee5dc049STobin C. Harding Called by the generic buffered write code to ask the filesystem 921ee5dc049STobin C. Harding to prepare to write len bytes at the given offset in the file. 922af96c1e3STobin C. Harding The address_space should check that the write will be able to 923ee5dc049STobin C. Harding complete, by allocating space if necessary and doing any other 924ee5dc049STobin C. Harding internal housekeeping. If the write will update parts of any 925af96c1e3STobin C. Harding basic-blocks on storage, then those blocks should be pre-read 926ee5dc049STobin C. Harding (if they haven't been read already) so that the updated blocks 927ee5dc049STobin C. Harding can be written out properly. 928ee5dc049STobin C. Harding 929af96c1e3STobin C. Harding The filesystem must return the locked pagecache page for the 930af96c1e3STobin C. Harding specified offset, in ``*pagep``, for the caller to write into. 931af96c1e3STobin C. Harding 932af96c1e3STobin C. Harding It must be able to cope with short writes (where the length 933ee5dc049STobin C. Harding passed to write_begin is greater than the number of bytes copied 934ee5dc049STobin C. Harding into the page). 935af96c1e3STobin C. Harding 936ee5dc049STobin C. Harding A void * may be returned in fsdata, which then gets passed into 937ee5dc049STobin C. Harding write_end. 938ee5dc049STobin C. Harding 939ee5dc049STobin C. Harding Returns 0 on success; < 0 on failure (which is the error code), 940af96c1e3STobin C. Harding in which case write_end is not called. 941ee5dc049STobin C. Harding 942ee5dc049STobin C. Harding``write_end`` 943af96c1e3STobin C. Harding After a successful write_begin, and data copy, write_end must be 944ee5dc049STobin C. Harding called. len is the original len passed to write_begin, and 945ee5dc049STobin C. Harding copied is the amount that was able to be copied. 946af96c1e3STobin C. Harding 947ee5dc049STobin C. Harding The filesystem must take care of unlocking the page and 948ee5dc049STobin C. Harding releasing it refcount, and updating i_size. 949ee5dc049STobin C. Harding 950ee5dc049STobin C. Harding Returns < 0 on failure, otherwise the number of bytes (<= 951ee5dc049STobin C. Harding 'copied') that were able to be copied into pagecache. 952ee5dc049STobin C. Harding 953ee5dc049STobin C. Harding``bmap`` 954ee5dc049STobin C. Harding called by the VFS to map a logical block offset within object to 955af96c1e3STobin C. Harding physical block number. This method is used by the FIBMAP ioctl 956128d1f82SMatthew Wilcox (Oracle) and for working with swap-files. To be able to swap to a file, 957128d1f82SMatthew Wilcox (Oracle) the file must have a stable mapping to a block device. The swap 958128d1f82SMatthew Wilcox (Oracle) system does not go through the filesystem but instead uses bmap 959ee5dc049STobin C. Harding to find out where the blocks in the file are and uses those 960af96c1e3STobin C. Harding addresses directly. 961af96c1e3STobin C. Harding 962fa29000bSMatthew Wilcox (Oracle)``invalidate_folio`` 963ee5dc049STobin C. Harding If a folio has private data, then invalidate_folio will be 964128d1f82SMatthew Wilcox (Oracle) called when part or all of the folio is to be removed from the 965fa29000bSMatthew Wilcox (Oracle) address space. This generally corresponds to either a 966fa29000bSMatthew Wilcox (Oracle) truncation, punch hole or a complete invalidation of the address 967ee5dc049STobin C. Harding space (in the latter case 'offset' will always be 0 and 'length' 968af96c1e3STobin C. Harding will be folio_size()). Any private data associated with the folio 969fa29000bSMatthew Wilcox (Oracle) should be updated to reflect this truncation. If offset is 0 970fa29000bSMatthew Wilcox (Oracle) and length is folio_size(), then the private data should be 971fa29000bSMatthew Wilcox (Oracle) released, because the folio must be able to be completely 972fa29000bSMatthew Wilcox (Oracle) discarded. This may be done by calling the ->release_folio 973fa29000bSMatthew Wilcox (Oracle) function, but in this case the release MUST succeed. 974fa29000bSMatthew Wilcox (Oracle) 975fa29000bSMatthew Wilcox (Oracle)``release_folio`` 976fa29000bSMatthew Wilcox (Oracle) release_folio is called on folios with private data to tell the 977fa29000bSMatthew Wilcox (Oracle) filesystem that the folio is about to be freed. ->release_folio 978af96c1e3STobin C. Harding should remove any private data from the folio and clear the 979af96c1e3STobin C. Harding private flag. If release_folio() fails, it should return false. 980fa29000bSMatthew Wilcox (Oracle) release_folio() is used in two distinct though related cases. 981fa29000bSMatthew Wilcox (Oracle) The first is when the VM wants to free a clean folio with no 982fa29000bSMatthew Wilcox (Oracle) active users. If ->release_folio succeeds, the folio will be 983ee5dc049STobin C. Harding removed from the address_space and be freed. 984ee5dc049STobin C. Harding 985fa29000bSMatthew Wilcox (Oracle) The second case is when a request has been made to invalidate 986fa29000bSMatthew Wilcox (Oracle) some or all folios in an address_space. This can happen 987fa29000bSMatthew Wilcox (Oracle) through the fadvise(POSIX_FADV_DONTNEED) system call or by the 988af96c1e3STobin C. Harding filesystem explicitly requesting it as nfs and 9p do (when they 989d2329aa0SMatthew Wilcox (Oracle) believe the cache may be out of date with storage) by calling 990d2329aa0SMatthew Wilcox (Oracle) invalidate_inode_pages2(). If the filesystem makes such a call, 991ee5dc049STobin C. Harding and needs to be certain that all folios are invalidated, then 992ee5dc049STobin C. Harding its release_folio will need to ensure this. Possibly it can 993ee5dc049STobin C. Harding clear the uptodate flag if it cannot free private data yet. 994ee5dc049STobin C. Harding 995af96c1e3STobin C. Harding``free_folio`` 996ee5dc049STobin C. Harding free_folio is called once the folio is no longer visible in the 997ee5dc049STobin C. Harding page cache in order to allow the cleanup of any private data. 998ee5dc049STobin C. Harding Since it may be called by the memory reclaimer, it should not 999ee5dc049STobin C. Harding assume that the original address_space mapping still exists, and 1000ee5dc049STobin C. Harding it should not block. 1001af96c1e3STobin C. Harding 10025490da4fSMatthew Wilcox (Oracle)``direct_IO`` 1003ee5dc049STobin C. Harding called by the generic read/write routines to perform direct_IO - 10045490da4fSMatthew Wilcox (Oracle) that is IO requests which bypass the page cache and transfer 10055490da4fSMatthew Wilcox (Oracle) data directly between the storage and the application's address 10065490da4fSMatthew Wilcox (Oracle) space. 10075490da4fSMatthew Wilcox (Oracle) 1008af96c1e3STobin C. Harding``migrate_folio`` 1009affa80e8SMatthew Wilcox (Oracle) This is used to compact the physical memory usage. If the VM 1010affa80e8SMatthew Wilcox (Oracle) wants to relocate a folio (maybe from a memory device that is 1011affa80e8SMatthew Wilcox (Oracle) signalling imminent failure) it will pass a new folio and an old 1012ee5dc049STobin C. Harding folio to this function. migrate_folio should transfer any private 1013af96c1e3STobin C. Harding data across and update any references that it has to the folio. 1014ee5dc049STobin C. Harding 1015ee5dc049STobin C. Harding``launder_folio`` 10162e7e80f7SMatthew Wilcox (Oracle) Called before freeing a folio - it writes back the dirty folio. 10172e7e80f7SMatthew Wilcox (Oracle) To prevent redirtying the folio, it is kept locked during the 10182e7e80f7SMatthew Wilcox (Oracle) whole operation. 1019af96c1e3STobin C. Harding 1020ee5dc049STobin C. Harding``is_partially_uptodate`` 1021520f301cSMatthew Wilcox (Oracle) Called by the VM when reading a file through the pagecache when 1022ee5dc049STobin C. Harding the underlying blocksize is smaller than the size of the folio. 1023ee5dc049STobin C. Harding If the required block is up to date then the read can complete 1024520f301cSMatthew Wilcox (Oracle) without needing I/O to bring the whole page up to date. 1025520f301cSMatthew Wilcox (Oracle) 1026ee5dc049STobin C. Harding``is_dirty_writeback`` 1027ee5dc049STobin C. Harding Called by the VM when attempting to reclaim a folio. The VM uses 1028520f301cSMatthew Wilcox (Oracle) dirty and writeback information to determine if it needs to 1029ee5dc049STobin C. Harding stall to allow flushers a chance to complete some IO. 1030af96c1e3STobin C. Harding Ordinarily it can use folio_test_dirty and folio_test_writeback but 1031ee5dc049STobin C. Harding some filesystems have more complex state (unstable folios in NFS 1032ee5dc049STobin C. Harding prevent reclaim) or do not set those flags due to locking 1033ee5dc049STobin C. Harding problems. This callback allows a filesystem to indicate to the 1034af96c1e3STobin C. Harding VM if a folio should be treated as dirty or writeback for the 1035af96c1e3STobin C. Harding purposes of stalling. 1036af96c1e3STobin C. Harding 1037ee5dc049STobin C. Harding``error_remove_page`` 1038cba738f6SNeilBrown normally set to generic_error_remove_page if truncation is ok 1039cba738f6SNeilBrown for this address space. Used for memory failure handling. 1040cba738f6SNeilBrown Setting this implies you deal with pages going away under you, 1041cba738f6SNeilBrown unless you have them locked or reference counts increased. 1042cba738f6SNeilBrown 1043cba738f6SNeilBrown``swap_activate`` 1044cba738f6SNeilBrown 1045cba738f6SNeilBrown Called to prepare the given file for swap. It should perform 1046af96c1e3STobin C. Harding any validation and preparation necessary to ensure that writes 1047ee5dc049STobin C. Harding can be performed with minimal memory allocation. It should call 1048ee5dc049STobin C. Harding add_swap_extent(), or the helper iomap_swapfile_activate(), and 1049ee5dc049STobin C. Harding return the number of extents added. If IO should be submitted 1050af96c1e3STobin C. Harding through ->swap_rw(), it should set SWP_FS_OPS, otherwise IO will 1051cba738f6SNeilBrown be submitted directly to the block device ``sis->bdev``. 1052cba738f6SNeilBrown 1053af96c1e3STobin C. Harding``swap_deactivate`` 1054af96c1e3STobin C. Harding Called during swapoff on files where swap_activate was 1055af96c1e3STobin C. Harding successful. 1056af96c1e3STobin C. Harding 1057af96c1e3STobin C. Harding``swap_rw`` 1058af96c1e3STobin C. Harding Called to read or write swap pages when SWP_FS_OPS is set. 1059af96c1e3STobin C. Harding 1060af96c1e3STobin C. HardingThe File Object 1061af96c1e3STobin C. Harding=============== 1062af96c1e3STobin C. Harding 1063af96c1e3STobin C. HardingA file object represents a file opened by a process. This is also known 1064af96c1e3STobin C. Hardingas an "open file description" in POSIX parlance. 1065af96c1e3STobin C. Harding 1066af96c1e3STobin C. Harding 1067af96c1e3STobin C. Hardingstruct file_operations 1068af96c1e3STobin C. Harding---------------------- 1069af96c1e3STobin C. Harding 1070af96c1e3STobin C. HardingThis describes how the VFS can manipulate an open file. As of kernel 1071af96c1e3STobin C. Harding4.18, the following members are defined: 1072af96c1e3STobin C. Harding 1073af96c1e3STobin C. Harding.. code-block:: c 1074af96c1e3STobin C. Harding 1075af96c1e3STobin C. Harding struct file_operations { 1076af96c1e3STobin C. Harding struct module *owner; 1077af96c1e3STobin C. Harding loff_t (*llseek) (struct file *, loff_t, int); 1078af96c1e3STobin C. Harding ssize_t (*read) (struct file *, char __user *, size_t, loff_t *); 1079af96c1e3STobin C. Harding ssize_t (*write) (struct file *, const char __user *, size_t, loff_t *); 1080af96c1e3STobin C. Harding ssize_t (*read_iter) (struct kiocb *, struct iov_iter *); 1081af96c1e3STobin C. Harding ssize_t (*write_iter) (struct kiocb *, struct iov_iter *); 1082af96c1e3STobin C. Harding int (*iopoll)(struct kiocb *kiocb, bool spin); 1083af96c1e3STobin C. Harding int (*iterate_shared) (struct file *, struct dir_context *); 1084af96c1e3STobin C. Harding __poll_t (*poll) (struct file *, struct poll_table_struct *); 1085af96c1e3STobin C. Harding long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long); 1086af96c1e3STobin C. Harding long (*compat_ioctl) (struct file *, unsigned int, unsigned long); 1087af96c1e3STobin C. Harding int (*mmap) (struct file *, struct vm_area_struct *); 1088af96c1e3STobin C. Harding int (*open) (struct inode *, struct file *); 1089af96c1e3STobin C. Harding int (*flush) (struct file *, fl_owner_t id); 1090af96c1e3STobin C. Harding int (*release) (struct inode *, struct file *); 1091af96c1e3STobin C. Harding int (*fsync) (struct file *, loff_t, loff_t, int datasync); 1092af96c1e3STobin C. Harding int (*fasync) (int, struct file *, int); 1093af96c1e3STobin C. Harding int (*lock) (struct file *, int, struct file_lock *); 1094af96c1e3STobin C. Harding unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long); 1095af96c1e3STobin C. Harding int (*check_flags)(int); 1096af96c1e3STobin C. Harding int (*flock) (struct file *, int, struct file_lock *); 1097af96c1e3STobin C. Harding ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int); 1098af96c1e3STobin C. Harding ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int); 1099af96c1e3STobin C. Harding int (*setlease)(struct file *, long, struct file_lock **, void **); 1100af96c1e3STobin C. Harding long (*fallocate)(struct file *file, int mode, loff_t offset, 1101af96c1e3STobin C. Harding loff_t len); 1102af96c1e3STobin C. Harding void (*show_fdinfo)(struct seq_file *m, struct file *f); 1103af96c1e3STobin C. Harding #ifndef CONFIG_MMU 1104af96c1e3STobin C. Harding unsigned (*mmap_capabilities)(struct file *); 1105af96c1e3STobin C. Harding #endif 1106af96c1e3STobin C. Harding ssize_t (*copy_file_range)(struct file *, loff_t, struct file *, loff_t, size_t, unsigned int); 1107af96c1e3STobin C. Harding loff_t (*remap_file_range)(struct file *file_in, loff_t pos_in, 1108af96c1e3STobin C. Harding struct file *file_out, loff_t pos_out, 1109af96c1e3STobin C. Harding loff_t len, unsigned int remap_flags); 1110ee5dc049STobin C. Harding int (*fadvise)(struct file *, loff_t, loff_t, int); 1111ee5dc049STobin C. Harding }; 1112af96c1e3STobin C. Harding 1113ee5dc049STobin C. HardingAgain, all methods are called without any locks being held, unless 1114ee5dc049STobin C. Hardingotherwise noted. 1115af96c1e3STobin C. Harding 1116ee5dc049STobin C. Harding``llseek`` 1117ee5dc049STobin C. Harding called when the VFS needs to move the file position index 1118af96c1e3STobin C. Harding 1119ee5dc049STobin C. Harding``read`` 1120ee5dc049STobin C. Harding called by read(2) and related system calls 1121af96c1e3STobin C. Harding 1122ee5dc049STobin C. Harding``read_iter`` 1123ee5dc049STobin C. Harding possibly asynchronous read with iov_iter as destination 1124af96c1e3STobin C. Harding 1125ee5dc049STobin C. Harding``write`` 1126ee5dc049STobin C. Harding called by write(2) and related system calls 1127af96c1e3STobin C. Harding 1128ee5dc049STobin C. Harding``write_iter`` 112999b319d3SJonathan Corbet possibly asynchronous write with iov_iter as source 1130af96c1e3STobin C. Harding 1131ee5dc049STobin C. Harding``iopoll`` 1132ee5dc049STobin C. Harding called when aio wants to poll for completions on HIPRI iocbs 1133af96c1e3STobin C. Harding 1134af96c1e3STobin C. Harding``iterate_shared`` 1135af96c1e3STobin C. Harding called when the VFS needs to read the directory contents 1136ee5dc049STobin C. Harding 1137ee5dc049STobin C. Harding``poll`` 1138af96c1e3STobin C. Harding called by the VFS when a process wants to check if there is 1139ee5dc049STobin C. Harding activity on this file and (optionally) go to sleep until there 1140ee5dc049STobin C. Harding is activity. Called by the select(2) and poll(2) system calls 1141ee5dc049STobin C. Harding 1142af96c1e3STobin C. Harding``unlocked_ioctl`` 1143ee5dc049STobin C. Harding called by the ioctl(2) system call. 1144ee5dc049STobin C. Harding 1145af96c1e3STobin C. Harding``compat_ioctl`` 1146ee5dc049STobin C. Harding called by the ioctl(2) system call when 32 bit system calls are 1147ee5dc049STobin C. Harding used on 64 bit kernels. 1148af96c1e3STobin C. Harding 1149af96c1e3STobin C. Harding``mmap`` 1150ee5dc049STobin C. Harding called by the mmap(2) system call 1151ee5dc049STobin C. Harding 1152ee5dc049STobin C. Harding``open`` 1153ee5dc049STobin C. Harding called by the VFS when an inode should be opened. When the VFS 1154af96c1e3STobin C. Harding opens a file, it creates a new "struct file". It then calls the 1155af96c1e3STobin C. Harding open method for the newly allocated file structure. You might 1156af96c1e3STobin C. Harding think that the open method really belongs in "struct 1157ee5dc049STobin C. Harding inode_operations", and you may be right. I think it's done the 1158ee5dc049STobin C. Harding way it is because it makes filesystems simpler to implement. 1159af96c1e3STobin C. Harding The open() method is a good place to initialize the 1160ee5dc049STobin C. Harding "private_data" member in the file structure if you want to point 1161ee5dc049STobin C. Harding to a device structure 1162af96c1e3STobin C. Harding 1163ee5dc049STobin C. Harding``flush`` 1164ee5dc049STobin C. Harding called by the close(2) system call to flush a file 1165af96c1e3STobin C. Harding 1166af96c1e3STobin C. Harding``release`` 1167ee5dc049STobin C. Harding called when the last reference to an open file is closed 1168ee5dc049STobin C. Harding 1169af96c1e3STobin C. Harding``fsync`` 1170af96c1e3STobin C. Harding called by the fsync(2) system call. Also see the section above 1171ee5dc049STobin C. Harding entitled "Handling errors during writeback". 1172ee5dc049STobin C. Harding 1173ee5dc049STobin C. Harding``fasync`` 1174af96c1e3STobin C. Harding called by the fcntl(2) system call when asynchronous 1175ee5dc049STobin C. Harding (non-blocking) mode is enabled for a file 1176ee5dc049STobin C. Harding 1177af96c1e3STobin C. Harding``lock`` 1178ee5dc049STobin C. Harding called by the fcntl(2) system call for F_GETLK, F_SETLK, and 1179ee5dc049STobin C. Harding F_SETLKW commands 1180af96c1e3STobin C. Harding 1181ee5dc049STobin C. Harding``get_unmapped_area`` 1182ee5dc049STobin C. Harding called by the mmap(2) system call 1183af96c1e3STobin C. Harding 1184ee5dc049STobin C. Harding``check_flags`` 1185ee5dc049STobin C. Harding called by the fcntl(2) system call for F_SETFL command 1186af96c1e3STobin C. Harding 1187af96c1e3STobin C. Harding``flock`` 1188ee5dc049STobin C. Harding called by the flock(2) system call 1189ee5dc049STobin C. Harding 1190af96c1e3STobin C. Harding``splice_write`` 1191af96c1e3STobin C. Harding called by the VFS to splice data from a pipe to a file. This 1192ee5dc049STobin C. Harding method is used by the splice(2) system call 1193ee5dc049STobin C. Harding 1194af96c1e3STobin C. Harding``splice_read`` 1195af96c1e3STobin C. Harding called by the VFS to splice data from file to a pipe. This 1196af96c1e3STobin C. Harding method is used by the splice(2) system call 1197ee5dc049STobin C. Harding 1198ee5dc049STobin C. Harding``setlease`` 1199af96c1e3STobin C. Harding called by the VFS to set or release a file lock lease. setlease 1200ee5dc049STobin C. Harding implementations should call generic_setlease to record or remove 1201ee5dc049STobin C. Harding the lease in the inode after setting it. 1202af96c1e3STobin C. Harding 1203ee5dc049STobin C. Harding``fallocate`` 1204ee5dc049STobin C. Harding called by the VFS to preallocate blocks or punch a hole. 1205ee5dc049STobin C. Harding 1206ee5dc049STobin C. Harding``copy_file_range`` 1207ee5dc049STobin C. Harding called by the copy_file_range(2) system call. 1208ee5dc049STobin C. Harding 1209ee5dc049STobin C. Harding``remap_file_range`` 1210ee5dc049STobin C. Harding called by the ioctl(2) system call for FICLONERANGE and FICLONE 1211ee5dc049STobin C. Harding and FIDEDUPERANGE commands to remap file ranges. An 1212ee5dc049STobin C. Harding implementation should remap len bytes at pos_in of the source 1213ee5dc049STobin C. Harding file into the dest file at pos_out. Implementations must handle 1214cb56ecaeSJulia Lawall callers passing in len == 0; this means "remap to the end of the 1215ee5dc049STobin C. Harding source file". The return value should the number of bytes 1216ee5dc049STobin C. Harding remapped, or the usual negative error code if errors occurred 1217af96c1e3STobin C. Harding before any bytes were remapped. The remap_flags parameter 1218ee5dc049STobin C. Harding accepts REMAP_FILE_* flags. If REMAP_FILE_DEDUP is set then the 1219ee5dc049STobin C. Harding implementation must only remap if the requested file ranges have 1220af96c1e3STobin C. Harding identical contents. If REMAP_FILE_CAN_SHORTEN is set, the caller is 1221af96c1e3STobin C. Harding ok with the implementation shortening the request length to 1222af96c1e3STobin C. Harding satisfy alignment or EOF requirements (or any other reason). 1223af96c1e3STobin C. Harding 1224af96c1e3STobin C. Harding``fadvise`` 1225af96c1e3STobin C. Harding possibly called by the fadvise64() system call. 1226af96c1e3STobin C. Harding 1227af96c1e3STobin C. HardingNote that the file operations are implemented by the specific 1228af96c1e3STobin C. Hardingfilesystem in which the inode resides. When opening a device node 1229af96c1e3STobin C. Harding(character or block special) most filesystems will call special 1230af96c1e3STobin C. Hardingsupport routines in the VFS which will locate the required device 1231af96c1e3STobin C. Hardingdriver information. These support routines replace the filesystem file 1232af96c1e3STobin C. Hardingoperations with those for the device driver, and then proceed to call 1233af96c1e3STobin C. Hardingthe new open() method for the file. This is how opening a device file 1234af96c1e3STobin C. Hardingin the filesystem eventually ends up calling the device driver open() 1235af96c1e3STobin C. Hardingmethod. 1236af96c1e3STobin C. Harding 1237af96c1e3STobin C. Harding 1238af96c1e3STobin C. HardingDirectory Entry Cache (dcache) 1239af96c1e3STobin C. Harding============================== 1240af96c1e3STobin C. Harding 1241af96c1e3STobin C. Harding 1242af96c1e3STobin C. Hardingstruct dentry_operations 1243af96c1e3STobin C. Harding------------------------ 1244af96c1e3STobin C. Harding 1245af96c1e3STobin C. HardingThis describes how a filesystem can overload the standard dentry 1246af96c1e3STobin C. Hardingoperations. Dentries and the dcache are the domain of the VFS and the 1247af96c1e3STobin C. Hardingindividual filesystem implementations. Device drivers have no business 1248af96c1e3STobin C. Hardinghere. These methods may be set to NULL, as they are either optional or 1249af96c1e3STobin C. Hardingthe VFS uses a default. As of kernel 2.6.22, the following members are 1250af96c1e3STobin C. Hardingdefined: 1251af96c1e3STobin C. Harding 1252af96c1e3STobin C. Harding.. code-block:: c 1253af96c1e3STobin C. Harding 1254af96c1e3STobin C. Harding struct dentry_operations { 1255af96c1e3STobin C. Harding int (*d_revalidate)(struct dentry *, unsigned int); 1256af96c1e3STobin C. Harding int (*d_weak_revalidate)(struct dentry *, unsigned int); 1257af96c1e3STobin C. Harding int (*d_hash)(const struct dentry *, struct qstr *); 1258af96c1e3STobin C. Harding int (*d_compare)(const struct dentry *, 1259af96c1e3STobin C. Harding unsigned int, const char *, const struct qstr *); 1260af96c1e3STobin C. Harding int (*d_delete)(const struct dentry *); 1261af96c1e3STobin C. Harding int (*d_init)(struct dentry *); 1262af96c1e3STobin C. Harding void (*d_release)(struct dentry *); 1263af96c1e3STobin C. Harding void (*d_iput)(struct dentry *, struct inode *); 1264ee5dc049STobin C. Harding char *(*d_dname)(struct dentry *, char *, int); 1265ee5dc049STobin C. Harding struct vfsmount *(*d_automount)(struct path *); 1266ee5dc049STobin C. Harding int (*d_manage)(const struct path *, bool); 1267ee5dc049STobin C. Harding struct dentry *(*d_real)(struct dentry *, const struct inode *); 1268ee5dc049STobin C. Harding }; 1269ee5dc049STobin C. Harding 1270ee5dc049STobin C. Harding``d_revalidate`` 1271af96c1e3STobin C. Harding called when the VFS needs to revalidate a dentry. This is 1272ee5dc049STobin C. Harding called whenever a name look-up finds a dentry in the dcache. 1273ee5dc049STobin C. Harding Most local filesystems leave this as NULL, because all their 1274af96c1e3STobin C. Harding dentries in the dcache are valid. Network filesystems are 1275ee5dc049STobin C. Harding different since things can change on the server without the 1276ee5dc049STobin C. Harding client necessarily being aware of it. 1277ee5dc049STobin C. Harding 1278ee5dc049STobin C. Harding This function should return a positive value if the dentry is 1279ee5dc049STobin C. Harding still valid, and zero or a negative error code if it isn't. 1280ee5dc049STobin C. Harding 1281af96c1e3STobin C. Harding d_revalidate may be called in rcu-walk mode (flags & 1282ee5dc049STobin C. Harding LOOKUP_RCU). If in rcu-walk mode, the filesystem must 1283ee5dc049STobin C. Harding revalidate the dentry without blocking or storing to the dentry, 1284af96c1e3STobin C. Harding d_parent and d_inode should not be used without care (because 1285af96c1e3STobin C. Harding they can change and, in d_inode case, even become NULL under 128674596085SGlenn Washburn us). 1287ee5dc049STobin C. Harding 1288ee5dc049STobin C. Harding If a situation is encountered that rcu-walk cannot handle, 1289ee5dc049STobin C. Harding return 1290ee5dc049STobin C. Harding -ECHILD and it will be called again in ref-walk mode. 1291ee5dc049STobin C. Harding 1292af96c1e3STobin C. Harding``d_weak_revalidate`` 1293ee5dc049STobin C. Harding called when the VFS needs to revalidate a "jumped" dentry. This 1294ee5dc049STobin C. Harding is called when a path-walk ends at dentry that was not acquired 1295ee5dc049STobin C. Harding by doing a lookup in the parent directory. This includes "/", 1296ee5dc049STobin C. Harding "." and "..", as well as procfs-style symlinks and mountpoint 1297af96c1e3STobin C. Harding traversal. 1298ee5dc049STobin C. Harding 1299ee5dc049STobin C. Harding In this case, we are less concerned with whether the dentry is 1300af96c1e3STobin C. Harding still fully correct, but rather that the inode is still valid. 1301af96c1e3STobin C. Harding As with d_revalidate, most local filesystems will set this to 1302af96c1e3STobin C. Harding NULL since their dcache entries are always valid. 1303ee5dc049STobin C. Harding 1304ee5dc049STobin C. Harding This function has the same return code semantics as 1305af96c1e3STobin C. Harding d_revalidate. 1306af96c1e3STobin C. Harding 1307af96c1e3STobin C. Harding d_weak_revalidate is only called after leaving rcu-walk mode. 1308af96c1e3STobin C. Harding 1309af96c1e3STobin C. Harding``d_hash`` 1310af96c1e3STobin C. Harding called when the VFS adds a dentry to the hash table. The first 1311ee5dc049STobin C. Harding dentry passed to d_hash is the parent directory that the name is 1312ee5dc049STobin C. Harding to be hashed into. 1313af96c1e3STobin C. Harding 1314ee5dc049STobin C. Harding Same locking and synchronisation rules as d_compare regarding 1315ee5dc049STobin C. Harding what is safe to dereference etc. 1316af96c1e3STobin C. Harding 1317af96c1e3STobin C. Harding``d_compare`` 1318ee5dc049STobin C. Harding called to compare a dentry name with a given name. The first 1319ee5dc049STobin C. Harding dentry is the parent of the dentry to be compared, the second is 1320ee5dc049STobin C. Harding the child dentry. len and name string are properties of the 1321af96c1e3STobin C. Harding dentry to be compared. qstr is the name to compare it with. 1322ee5dc049STobin C. Harding 1323ee5dc049STobin C. Harding Must be constant and idempotent, and should not take locks if 1324ee5dc049STobin C. Harding possible, and should not or store into the dentry. Should not 1325af96c1e3STobin C. Harding dereference pointers outside the dentry without lots of care 1326ee5dc049STobin C. Harding (eg. d_parent, d_inode, d_name should not be used). 1327ee5dc049STobin C. Harding 1328af96c1e3STobin C. Harding However, our vfsmount is pinned, and RCU held, so the dentries 1329ee5dc049STobin C. Harding and inodes won't disappear, neither will our sb or filesystem 1330ee5dc049STobin C. Harding module. ->d_sb may be used. 1331ee5dc049STobin C. Harding 1332ee5dc049STobin C. Harding It is a tricky calling convention because it needs to be called 1333ee5dc049STobin C. Harding under "rcu-walk", ie. without any locks or references on things. 1334ee5dc049STobin C. Harding 1335af96c1e3STobin C. Harding``d_delete`` 1336ee5dc049STobin C. Harding called when the last reference to a dentry is dropped and the 1337ee5dc049STobin C. Harding dcache is deciding whether or not to cache it. Return 1 to 1338af96c1e3STobin C. Harding delete immediately, or 0 to cache the dentry. Default is NULL 1339ee5dc049STobin C. Harding which means to always cache a reachable dentry. d_delete must 1340ee5dc049STobin C. Harding be constant and idempotent. 1341af96c1e3STobin C. Harding 1342ee5dc049STobin C. Harding``d_init`` 1343ee5dc049STobin C. Harding called when a dentry is allocated 1344ee5dc049STobin C. Harding 1345ee5dc049STobin C. Harding``d_release`` 1346ee5dc049STobin C. Harding called when a dentry is really deallocated 1347af96c1e3STobin C. Harding 1348ee5dc049STobin C. Harding``d_iput`` 1349ee5dc049STobin C. Harding called when a dentry loses its inode (just prior to its being 1350ee5dc049STobin C. Harding deallocated). The default when this is NULL is that the VFS 1351ee5dc049STobin C. Harding calls iput(). If you define this method, you must call iput() 1352ee5dc049STobin C. Harding yourself 1353ee5dc049STobin C. Harding 1354ee5dc049STobin C. Harding``d_dname`` 1355ee5dc049STobin C. Harding called when the pathname of a dentry should be generated. 1356ee5dc049STobin C. Harding Useful for some pseudo filesystems (sockfs, pipefs, ...) to 1357ee5dc049STobin C. Harding delay pathname generation. (Instead of doing it when dentry is 1358ee5dc049STobin C. Harding created, it's done only when the path is needed.). Real 1359ee5dc049STobin C. Harding filesystems probably dont want to use it, because their dentries 1360ee5dc049STobin C. Harding are present in global dcache hash, so their hash should be an 1361ee5dc049STobin C. Harding invariant. As no lock is held, d_dname() should not try to 1362af96c1e3STobin C. Harding modify the dentry itself, unless appropriate SMP safety is used. 1363af96c1e3STobin C. Harding CAUTION : d_path() logic is quite tricky. The correct way to 1364af96c1e3STobin C. Harding return for example "Hello" is to put it at the end of the 1365af96c1e3STobin C. Harding buffer, and returns a pointer to the first char. 1366af96c1e3STobin C. Harding dynamic_dname() helper function is provided to take care of 1367af96c1e3STobin C. Harding this. 1368af96c1e3STobin C. Harding 1369af96c1e3STobin C. Harding Example : 1370af96c1e3STobin C. Harding 1371af96c1e3STobin C. Harding.. code-block:: c 1372af96c1e3STobin C. Harding 1373ee5dc049STobin C. Harding static char *pipefs_dname(struct dentry *dent, char *buffer, int buflen) 1374ee5dc049STobin C. Harding { 1375ee5dc049STobin C. Harding return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]", 1376ee5dc049STobin C. Harding dentry->d_inode->i_ino); 1377ee5dc049STobin C. Harding } 1378ee5dc049STobin C. Harding 1379ee5dc049STobin C. Harding``d_automount`` 1380ee5dc049STobin C. Harding called when an automount dentry is to be traversed (optional). 1381ee5dc049STobin C. Harding This should create a new VFS mount record and return the record 1382ee5dc049STobin C. Harding to the caller. The caller is supplied with a path parameter 1383ee5dc049STobin C. Harding giving the automount directory to describe the automount target 1384af96c1e3STobin C. Harding and the parent VFS mount record to provide inheritable mount 1385ee5dc049STobin C. Harding parameters. NULL should be returned if someone else managed to 1386ee5dc049STobin C. Harding make the automount first. If the vfsmount creation failed, then 1387ee5dc049STobin C. Harding an error code should be returned. If -EISDIR is returned, then 1388ee5dc049STobin C. Harding the directory will be treated as an ordinary directory and 1389ee5dc049STobin C. Harding returned to pathwalk to continue walking. 1390af96c1e3STobin C. Harding 1391ee5dc049STobin C. Harding If a vfsmount is returned, the caller will attempt to mount it 1392ee5dc049STobin C. Harding on the mountpoint and will remove the vfsmount from its 1393ee5dc049STobin C. Harding expiration list in the case of failure. The vfsmount should be 1394af96c1e3STobin C. Harding returned with 2 refs on it to prevent automatic expiration - the 1395ee5dc049STobin C. Harding caller will clean up the additional ref. 1396ee5dc049STobin C. Harding 1397ee5dc049STobin C. Harding This function is only used if DCACHE_NEED_AUTOMOUNT is set on 1398ee5dc049STobin C. Harding the dentry. This is set by __d_instantiate() if S_AUTOMOUNT is 1399ee5dc049STobin C. Harding set on the inode being added. 1400ee5dc049STobin C. Harding 1401ee5dc049STobin C. Harding``d_manage`` 1402ee5dc049STobin C. Harding called to allow the filesystem to manage the transition from a 1403ee5dc049STobin C. Harding dentry (optional). This allows autofs, for example, to hold up 1404ee5dc049STobin C. Harding clients waiting to explore behind a 'mountpoint' while letting 1405af96c1e3STobin C. Harding the daemon go past and construct the subtree there. 0 should be 1406af96c1e3STobin C. Harding returned to let the calling process continue. -EISDIR can be 1407ee5dc049STobin C. Harding returned to tell pathwalk to use this directory as an ordinary 1408ee5dc049STobin C. Harding directory and to ignore anything mounted on it and not to check 1409ee5dc049STobin C. Harding the automount flag. Any other error code will abort pathwalk 1410ee5dc049STobin C. Harding completely. 1411af96c1e3STobin C. Harding 1412ee5dc049STobin C. Harding If the 'rcu_walk' parameter is true, then the caller is doing a 1413ee5dc049STobin C. Harding pathwalk in RCU-walk mode. Sleeping is not permitted in this 1414af96c1e3STobin C. Harding mode, and the caller can be asked to leave it and call again by 1415ee5dc049STobin C. Harding returning -ECHILD. -EISDIR may also be returned to tell 1416ee5dc049STobin C. Harding pathwalk to ignore d_automount or any mounts. 1417ee5dc049STobin C. Harding 1418ee5dc049STobin C. Harding This function is only used if DCACHE_MANAGE_TRANSIT is set on 1419af96c1e3STobin C. Harding the dentry being transited from. 1420ee5dc049STobin C. Harding 1421ee5dc049STobin C. Harding``d_real`` 1422ee5dc049STobin C. Harding overlay/union type filesystems implement this method to return 1423ee5dc049STobin C. Harding one of the underlying dentries hidden by the overlay. It is 1424af96c1e3STobin C. Harding used in two different modes: 1425af96c1e3STobin C. Harding 1426af96c1e3STobin C. Harding Called from file_dentry() it returns the real dentry matching 1427af96c1e3STobin C. Harding the inode argument. The real dentry may be from a lower layer 1428af96c1e3STobin C. Harding already copied up, but still referenced from the file. This 1429af96c1e3STobin C. Harding mode is selected with a non-NULL inode argument. 1430af96c1e3STobin C. Harding 1431af96c1e3STobin C. Harding With NULL inode the topmost real underlying dentry is returned. 1432af96c1e3STobin C. Harding 1433af96c1e3STobin C. HardingEach dentry has a pointer to its parent dentry, as well as a hash list 1434af96c1e3STobin C. Hardingof child dentries. Child dentries are basically like files in a 1435af96c1e3STobin C. Hardingdirectory. 1436af96c1e3STobin C. Harding 1437af96c1e3STobin C. Harding 1438ee5dc049STobin C. HardingDirectory Entry Cache API 1439ee5dc049STobin C. Harding-------------------------- 1440af96c1e3STobin C. Harding 1441af96c1e3STobin C. HardingThere are a number of functions defined which permit a filesystem to 1442ee5dc049STobin C. Hardingmanipulate dentries: 1443ee5dc049STobin C. Harding 1444af96c1e3STobin C. Harding``dget`` 1445af96c1e3STobin C. Harding open a new handle for an existing dentry (this just increments 1446ee5dc049STobin C. Harding the usage count) 1447ee5dc049STobin C. Harding 1448ee5dc049STobin C. Harding``dput`` 1449af96c1e3STobin C. Harding close a handle for a dentry (decrements the usage count). If 1450ee5dc049STobin C. Harding the usage count drops to 0, and the dentry is still in its 1451ee5dc049STobin C. Harding parent's hash, the "d_delete" method is called to check whether 1452ee5dc049STobin C. Harding it should be cached. If it should not be cached, or if the 1453ee5dc049STobin C. Harding dentry is not hashed, it is deleted. Otherwise cached dentries 1454af96c1e3STobin C. Harding are put into an LRU list to be reclaimed on memory shortage. 1455ee5dc049STobin C. Harding 1456ee5dc049STobin C. Harding``d_drop`` 1457ee5dc049STobin C. Harding this unhashes a dentry from its parents hash list. A subsequent 1458ee5dc049STobin C. Harding call to dput() will deallocate the dentry if its usage count 1459ee5dc049STobin C. Harding drops to 0 1460af96c1e3STobin C. Harding 1461ee5dc049STobin C. Harding``d_delete`` 1462ee5dc049STobin C. Harding delete a dentry. If there are no other open references to the 1463af96c1e3STobin C. Harding dentry then the dentry is turned into a negative dentry (the 1464af96c1e3STobin C. Harding d_iput() method is called). If there are other references, then 1465ee5dc049STobin C. Harding d_drop() is called instead 1466ee5dc049STobin C. Harding 1467ee5dc049STobin C. Harding``d_add`` 1468ee5dc049STobin C. Harding add a dentry to its parents hash list and then calls 1469ee5dc049STobin C. Harding d_instantiate() 1470ee5dc049STobin C. Harding 1471ee5dc049STobin C. Harding``d_instantiate`` 1472af96c1e3STobin C. Harding add a dentry to the alias hash list for the inode and updates 1473ee5dc049STobin C. Harding the "d_inode" member. The "i_count" member in the inode 1474ee5dc049STobin C. Harding structure should be set/incremented. If the inode pointer is 1475ee5dc049STobin C. Harding NULL, the dentry is called a "negative dentry". This function 1476ee5dc049STobin C. Harding is commonly called when an inode is created for an existing 1477ee5dc049STobin C. Harding negative dentry 1478ee5dc049STobin C. Harding 1479af96c1e3STobin C. Harding``d_lookup`` 1480af96c1e3STobin C. Harding look up a dentry given its parent and path name component It 1481af96c1e3STobin C. Harding looks up the child of that given name from the dcache hash 1482af96c1e3STobin C. Harding table. If it is found, the reference count is incremented and 1483af96c1e3STobin C. Harding the dentry is returned. The caller must use dput() to free the 1484af96c1e3STobin C. Harding dentry when it finishes using it. 1485af96c1e3STobin C. Harding 1486af96c1e3STobin C. Harding 1487af96c1e3STobin C. HardingMount Options 1488af96c1e3STobin C. Harding============= 1489af96c1e3STobin C. Harding 1490af96c1e3STobin C. Harding 1491af96c1e3STobin C. HardingParsing options 1492af96c1e3STobin C. Harding--------------- 1493af96c1e3STobin C. Harding 1494af96c1e3STobin C. HardingOn mount and remount the filesystem is passed a string containing a 1495af96c1e3STobin C. Hardingcomma separated list of mount options. The options can have either of 1496af96c1e3STobin C. Hardingthese forms: 1497af96c1e3STobin C. Harding 1498af96c1e3STobin C. Harding option 1499af96c1e3STobin C. Harding option=value 1500af96c1e3STobin C. Harding 1501af96c1e3STobin C. HardingThe <linux/parser.h> header defines an API that helps parse these 1502af96c1e3STobin C. Hardingoptions. There are plenty of examples on how to use it in existing 1503af96c1e3STobin C. Hardingfilesystems. 1504af96c1e3STobin C. Harding 1505af96c1e3STobin C. Harding 1506af96c1e3STobin C. HardingShowing options 1507af96c1e3STobin C. Harding--------------- 1508af96c1e3STobin C. Harding 1509af96c1e3STobin C. HardingIf a filesystem accepts mount options, it must define show_options() to 1510af96c1e3STobin C. Hardingshow all the currently active options. The rules are: 1511af96c1e3STobin C. Harding 1512af96c1e3STobin C. Harding - options MUST be shown which are not default or their values differ 1513af96c1e3STobin C. Harding from the default 1514af96c1e3STobin C. Harding 1515af96c1e3STobin C. Harding - options MAY be shown which are enabled by default or have their 1516af96c1e3STobin C. Harding default value 1517af96c1e3STobin C. Harding 1518af96c1e3STobin C. HardingOptions used only internally between a mount helper and the kernel (such 1519af96c1e3STobin C. Hardingas file descriptors), or which only have an effect during the mounting 1520af96c1e3STobin C. Harding(such as ones controlling the creation of a journal) are exempt from the 1521af96c1e3STobin C. Hardingabove rules. 1522af96c1e3STobin C. Harding 1523af96c1e3STobin C. HardingThe underlying reason for the above rules is to make sure, that a mount 1524af96c1e3STobin C. Hardingcan be accurately replicated (e.g. umounting and mounting again) based 1525af96c1e3STobin C. Hardingon the information found in /proc/mounts. 1526af96c1e3STobin C. Harding 1527af96c1e3STobin C. Harding 1528af96c1e3STobin C. HardingResources 1529c69f22f2SAlexander A. Klimov========= 1530af96c1e3STobin C. Harding 1531af96c1e3STobin C. Harding(Note some of these resources are not up-to-date with the latest kernel 1532af96c1e3STobin C. Harding version.) 1533af96c1e3STobin C. Harding 1534af96c1e3STobin C. HardingCreating Linux virtual filesystems. 2002 1535c69f22f2SAlexander A. Klimov <https://lwn.net/Articles/13325/> 1536af96c1e3STobin C. Harding 1537af96c1e3STobin C. HardingThe Linux Virtual File-system Layer by Neil Brown. 1999 1538c69f22f2SAlexander A. Klimov <http://www.cse.unsw.edu.au/~neilb/oss/linux-commentary/vfs.html> 1539 1540A tour of the Linux VFS by Michael K. Johnson. 1996 1541 <https://www.tldp.org/LDP/khg/HyperNews/get/fs/vfstour.html> 1542 1543A small trail through the Linux kernel by Andries Brouwer. 2001 1544 <https://www.win.tue.nl/~aeb/linux/vfs/trail.html> 1545