About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Documentation / filesystems / porting

Custom Search

Based on kernel version 4.16.1. Page generated on 2018-04-09 11:53 EST.

1	Changes since 2.5.0:
3	---
4	[recommended]
6	New helpers: sb_bread(), sb_getblk(), sb_find_get_block(), set_bh(),
7		sb_set_blocksize() and sb_min_blocksize().
9	Use them.
11	(sb_find_get_block() replaces 2.4's get_hash_table())
13	---
14	[recommended]
16	New methods: ->alloc_inode() and ->destroy_inode().
18	Remove inode->u.foo_inode_i
19	Declare
20		struct foo_inode_info {
21			/* fs-private stuff */
22			struct inode vfs_inode;
23		};
24		static inline struct foo_inode_info *FOO_I(struct inode *inode)
25		{
26			return list_entry(inode, struct foo_inode_info, vfs_inode);
27		}
29	Use FOO_I(inode) instead of &inode->u.foo_inode_i;
31	Add foo_alloc_inode() and foo_destroy_inode() - the former should allocate
32	foo_inode_info and return the address of ->vfs_inode, the latter should free
33	FOO_I(inode) (see in-tree filesystems for examples).
35	Make them ->alloc_inode and ->destroy_inode in your super_operations.
37	Keep in mind that now you need explicit initialization of private data
38	typically between calling iget_locked() and unlocking the inode.
40	At some point that will become mandatory.
42	---
43	[mandatory]
45	Change of file_system_type method (->read_super to ->get_sb)
47	->read_super() is no more.  Ditto for DECLARE_FSTYPE and DECLARE_FSTYPE_DEV.
49	Turn your foo_read_super() into a function that would return 0 in case of
50	success and negative number in case of error (-EINVAL unless you have more
51	informative error value to report).  Call it foo_fill_super().  Now declare
53	int foo_get_sb(struct file_system_type *fs_type,
54		int flags, const char *dev_name, void *data, struct vfsmount *mnt)
55	{
56		return get_sb_bdev(fs_type, flags, dev_name, data, foo_fill_super,
57				   mnt);
58	}
60	(or similar with s/bdev/nodev/ or s/bdev/single/, depending on the kind of
61	filesystem).
63	Replace DECLARE_FSTYPE... with explicit initializer and have ->get_sb set as
64	foo_get_sb.
66	---
67	[mandatory]
69	Locking change: ->s_vfs_rename_sem is taken only by cross-directory renames.
70	Most likely there is no need to change anything, but if you relied on
71	global exclusion between renames for some internal purpose - you need to
72	change your internal locking.  Otherwise exclusion warranties remain the
73	same (i.e. parents and victim are locked, etc.).
75	---
76	[informational]
78	Now we have the exclusion between ->lookup() and directory removal (by
79	->rmdir() and ->rename()).  If you used to need that exclusion and do
80	it by internal locking (most of filesystems couldn't care less) - you
81	can relax your locking.
83	---
84	[mandatory]
86	->lookup(), ->truncate(), ->create(), ->unlink(), ->mknod(), ->mkdir(),
87	->rmdir(), ->link(), ->lseek(), ->symlink(), ->rename()
88	and ->readdir() are called without BKL now.  Grab it on entry, drop upon return
89	- that will guarantee the same locking you used to have.  If your method or its
90	parts do not need BKL - better yet, now you can shift lock_kernel() and
91	unlock_kernel() so that they would protect exactly what needs to be
92	protected.
94	---
95	[mandatory]
97	BKL is also moved from around sb operations. BKL should have been shifted into
98	individual fs sb_op functions.  If you don't need it, remove it.
100	---
101	[informational]
103	check for ->link() target not being a directory is done by callers.  Feel
104	free to drop it...
106	---
107	[informational]
109	->link() callers hold ->i_mutex on the object we are linking to.  Some of your
110	problems might be over...
112	---
113	[mandatory]
115	new file_system_type method - kill_sb(superblock).  If you are converting
116	an existing filesystem, set it according to ->fs_flags:
117		FS_REQUIRES_DEV		-	kill_block_super
118		FS_LITTER		-	kill_litter_super
119		neither			-	kill_anon_super
120	FS_LITTER is gone - just remove it from fs_flags.
122	---
123	[mandatory]
125		FS_SINGLE is gone (actually, that had happened back when ->get_sb()
126	went in - and hadn't been documented ;-/).  Just remove it from fs_flags
127	(and see ->get_sb() entry for other actions).
129	---
130	[mandatory]
132	->setattr() is called without BKL now.  Caller _always_ holds ->i_mutex, so
133	watch for ->i_mutex-grabbing code that might be used by your ->setattr().
134	Callers of notify_change() need ->i_mutex now.
136	---
137	[recommended]
139	New super_block field "struct export_operations *s_export_op" for
140	explicit support for exporting, e.g. via NFS.  The structure is fully
141	documented at its declaration in include/linux/fs.h, and in
142	Documentation/filesystems/nfs/Exporting.
144	Briefly it allows for the definition of decode_fh and encode_fh operations
145	to encode and decode filehandles, and allows the filesystem to use
146	a standard helper function for decode_fh, and provide file-system specific
147	support for this helper, particularly get_parent.
149	It is planned that this will be required for exporting once the code
150	settles down a bit.
152	[mandatory]
154	s_export_op is now required for exporting a filesystem.
155	isofs, ext2, ext3, resierfs, fat
156	can be used as examples of very different filesystems.
158	---
159	[mandatory]
161	iget4() and the read_inode2 callback have been superseded by iget5_locked()
162	which has the following prototype,
164	    struct inode *iget5_locked(struct super_block *sb, unsigned long ino,
165					int (*test)(struct inode *, void *),
166					int (*set)(struct inode *, void *),
167					void *data);
169	'test' is an additional function that can be used when the inode
170	number is not sufficient to identify the actual file object. 'set'
171	should be a non-blocking function that initializes those parts of a
172	newly created inode to allow the test function to succeed. 'data' is
173	passed as an opaque value to both test and set functions.
175	When the inode has been created by iget5_locked(), it will be returned with the
176	I_NEW flag set and will still be locked.  The filesystem then needs to finalize
177	the initialization. Once the inode is initialized it must be unlocked by
178	calling unlock_new_inode().
180	The filesystem is responsible for setting (and possibly testing) i_ino
181	when appropriate. There is also a simpler iget_locked function that
182	just takes the superblock and inode number as arguments and does the
183	test and set for you.
185	e.g.
186		inode = iget_locked(sb, ino);
187		if (inode->i_state & I_NEW) {
188			err = read_inode_from_disk(inode);
189			if (err < 0) {
190				iget_failed(inode);
191				return err;
192			}
193			unlock_new_inode(inode);
194		}
196	Note that if the process of setting up a new inode fails, then iget_failed()
197	should be called on the inode to render it dead, and an appropriate error
198	should be passed back to the caller.
200	---
201	[recommended]
203	->getattr() finally getting used.  See instances in nfs, minix, etc.
205	---
206	[mandatory]
208	->revalidate() is gone.  If your filesystem had it - provide ->getattr()
209	and let it call whatever you had as ->revlidate() + (for symlinks that
210	had ->revalidate()) add calls in ->follow_link()/->readlink().
212	---
213	[mandatory]
215	->d_parent changes are not protected by BKL anymore.  Read access is safe
216	if at least one of the following is true:
217		* filesystem has no cross-directory rename()
218		* we know that parent had been locked (e.g. we are looking at
219	->d_parent of ->lookup() argument).
220		* we are called from ->rename().
221		* the child's ->d_lock is held
222	Audit your code and add locking if needed.  Notice that any place that is
223	not protected by the conditions above is risky even in the old tree - you
224	had been relying on BKL and that's prone to screwups.  Old tree had quite
225	a few holes of that kind - unprotected access to ->d_parent leading to
226	anything from oops to silent memory corruption.
228	---
229	[mandatory]
231		FS_NOMOUNT is gone.  If you use it - just set SB_NOUSER in flags
232	(see rootfs for one kind of solution and bdev/socket/pipe for another).
234	---
235	[recommended]
237		Use bdev_read_only(bdev) instead of is_read_only(kdev).  The latter
238	is still alive, but only because of the mess in drivers/s390/block/dasd.c.
239	As soon as it gets fixed is_read_only() will die.
241	---
242	[mandatory]
244	->permission() is called without BKL now. Grab it on entry, drop upon
245	return - that will guarantee the same locking you used to have.  If
246	your method or its parts do not need BKL - better yet, now you can
247	shift lock_kernel() and unlock_kernel() so that they would protect
248	exactly what needs to be protected.
250	---
251	[mandatory]
253	->statfs() is now called without BKL held.  BKL should have been
254	shifted into individual fs sb_op functions where it's not clear that
255	it's safe to remove it.  If you don't need it, remove it.
257	---
258	[mandatory]
260		is_read_only() is gone; use bdev_read_only() instead.
262	---
263	[mandatory]
265		destroy_buffers() is gone; use invalidate_bdev().
267	---
268	[mandatory]
270		fsync_dev() is gone; use fsync_bdev().  NOTE: lvm breakage is
271	deliberate; as soon as struct block_device * is propagated in a reasonable
272	way by that code fixing will become trivial; until then nothing can be
273	done.
275	[mandatory]
277		block truncatation on error exit from ->write_begin, and ->direct_IO
278	moved from generic methods (block_write_begin, cont_write_begin,
279	nobh_write_begin, blockdev_direct_IO*) to callers.  Take a look at
280	ext2_write_failed and callers for an example.
282	[mandatory]
284		->truncate is gone.  The whole truncate sequence needs to be
285	implemented in ->setattr, which is now mandatory for filesystems
286	implementing on-disk size changes.  Start with a copy of the old inode_setattr
287	and vmtruncate, and the reorder the vmtruncate + foofs_vmtruncate sequence to
288	be in order of zeroing blocks using block_truncate_page or similar helpers,
289	size update and on finally on-disk truncation which should not fail.
290	setattr_prepare (which used to be inode_change_ok) now includes the size checks
291	for ATTR_SIZE and must be called in the beginning of ->setattr unconditionally.
293	[mandatory]
295		->clear_inode() and ->delete_inode() are gone; ->evict_inode() should
296	be used instead.  It gets called whenever the inode is evicted, whether it has
297	remaining links or not.  Caller does *not* evict the pagecache or inode-associated
298	metadata buffers; the method has to use truncate_inode_pages_final() to get rid
299	of those. Caller makes sure async writeback cannot be running for the inode while
300	(or after) ->evict_inode() is called.
302		->drop_inode() returns int now; it's called on final iput() with
303	inode->i_lock held and it returns true if filesystems wants the inode to be
304	dropped.  As before, generic_drop_inode() is still the default and it's been
305	updated appropriately.  generic_delete_inode() is also alive and it consists
306	simply of return 1.  Note that all actual eviction work is done by caller after
307	->drop_inode() returns.
309		As before, clear_inode() must be called exactly once on each call of
310	->evict_inode() (as it used to be for each call of ->delete_inode()).  Unlike
311	before, if you are using inode-associated metadata buffers (i.e.
312	mark_buffer_dirty_inode()), it's your responsibility to call
313	invalidate_inode_buffers() before clear_inode().
315		NOTE: checking i_nlink in the beginning of ->write_inode() and bailing out
316	if it's zero is not *and* *never* *had* *been* enough.  Final unlink() and iput()
317	may happen while the inode is in the middle of ->write_inode(); e.g. if you blindly
318	free the on-disk inode, you may end up doing that while ->write_inode() is writing
319	to it.
321	---
322	[mandatory]
324		.d_delete() now only advises the dcache as to whether or not to cache
325	unreferenced dentries, and is now only called when the dentry refcount goes to
326	0. Even on 0 refcount transition, it must be able to tolerate being called 0,
327	1, or more times (eg. constant, idempotent).
329	---
330	[mandatory]
332		.d_compare() calling convention and locking rules are significantly
333	changed. Read updated documentation in Documentation/filesystems/vfs.txt (and
334	look at examples of other filesystems) for guidance.
336	---
337	[mandatory]
339		.d_hash() calling convention and locking rules are significantly
340	changed. Read updated documentation in Documentation/filesystems/vfs.txt (and
341	look at examples of other filesystems) for guidance.
343	---
344	[mandatory]
345		dcache_lock is gone, replaced by fine grained locks. See fs/dcache.c
346	for details of what locks to replace dcache_lock with in order to protect
347	particular things. Most of the time, a filesystem only needs ->d_lock, which
348	protects *all* the dcache state of a given dentry.
350	--
351	[mandatory]
353		Filesystems must RCU-free their inodes, if they can have been accessed
354	via rcu-walk path walk (basically, if the file can have had a path name in the
355	vfs namespace).
357		Even though i_dentry and i_rcu share storage in a union, we will
358	initialize the former in inode_init_always(), so just leave it alone in
359	the callback.  It used to be necessary to clean it there, but not anymore
360	(starting at 3.2).
362	--
363	[recommended]
364		vfs now tries to do path walking in "rcu-walk mode", which avoids
365	atomic operations and scalability hazards on dentries and inodes (see
366	Documentation/filesystems/path-lookup.txt). d_hash and d_compare changes
367	(above) are examples of the changes required to support this. For more complex
368	filesystem callbacks, the vfs drops out of rcu-walk mode before the fs call, so
369	no changes are required to the filesystem. However, this is costly and loses
370	the benefits of rcu-walk mode. We will begin to add filesystem callbacks that
371	are rcu-walk aware, shown below. Filesystems should take advantage of this
372	where possible.
374	--
375	[mandatory]
376		d_revalidate is a callback that is made on every path element (if
377	the filesystem provides it), which requires dropping out of rcu-walk mode. This
378	may now be called in rcu-walk mode (nd->flags & LOOKUP_RCU). -ECHILD should be
379	returned if the filesystem cannot handle rcu-walk. See
380	Documentation/filesystems/vfs.txt for more details.
382		permission is an inode permission check that is called on many or all
383	directory inodes on the way down a path walk (to check for exec permission). It
384	must now be rcu-walk aware (mask & MAY_NOT_BLOCK).  See
385	Documentation/filesystems/vfs.txt for more details.
387	--
388	[mandatory]
389		In ->fallocate() you must check the mode option passed in.  If your
390	filesystem does not support hole punching (deallocating space in the middle of a
391	file) you must return -EOPNOTSUPP if FALLOC_FL_PUNCH_HOLE is set in mode.
392	Currently you can only have FALLOC_FL_PUNCH_HOLE with FALLOC_FL_KEEP_SIZE set,
393	so the i_size should not change when hole punching, even when puching the end of
394	a file off.
396	--
397	[mandatory]
398		->get_sb() is gone.  Switch to use of ->mount().  Typically it's just
399	a matter of switching from calling get_sb_... to mount_... and changing the
400	function type.  If you were doing it manually, just switch from setting ->mnt_root
401	to some pointer to returning that pointer.  On errors return ERR_PTR(...).
403	--
404	[mandatory]
405		->permission() and generic_permission()have lost flags
406	argument; instead of passing IPERM_FLAG_RCU we add MAY_NOT_BLOCK into mask.
407		generic_permission() has also lost the check_acl argument; ACL checking
408	has been taken to VFS and filesystems need to provide a non-NULL ->i_op->get_acl
409	to read an ACL from disk.
411	--
412	[mandatory]
413		If you implement your own ->llseek() you must handle SEEK_HOLE and
414	SEEK_DATA.  You can hanle this by returning -EINVAL, but it would be nicer to
415	support it in some way.  The generic handler assumes that the entire file is
416	data and there is a virtual hole at the end of the file.  So if the provided
417	offset is less than i_size and SEEK_DATA is specified, return the same offset.
418	If the above is true for the offset and you are given SEEK_HOLE, return the end
419	of the file.  If the offset is i_size or greater return -ENXIO in either case.
421	[mandatory]
422		If you have your own ->fsync() you must make sure to call
423	filemap_write_and_wait_range() so that all dirty pages are synced out properly.
424	You must also keep in mind that ->fsync() is not called with i_mutex held
425	anymore, so if you require i_mutex locking you must make sure to take it and
426	release it yourself.
428	--
429	[mandatory]
430		d_alloc_root() is gone, along with a lot of bugs caused by code
431	misusing it.  Replacement: d_make_root(inode).  The difference is,
432	d_make_root() drops the reference to inode if dentry allocation fails.  
434	--
435	[mandatory]
436		The witch is dead!  Well, 2/3 of it, anyway.  ->d_revalidate() and
437	->lookup() do *not* take struct nameidata anymore; just the flags.
438	--
439	[mandatory]
440		->create() doesn't take struct nameidata *; unlike the previous
441	two, it gets "is it an O_EXCL or equivalent?" boolean argument.  Note that
442	local filesystems can ignore tha argument - they are guaranteed that the
443	object doesn't exist.  It's remote/distributed ones that might care...
444	--
445	[mandatory]
446		FS_REVAL_DOT is gone; if you used to have it, add ->d_weak_revalidate()
447	in your dentry operations instead.
448	--
449	[mandatory]
450		vfs_readdir() is gone; switch to iterate_dir() instead
451	--
452	[mandatory]
453		->readdir() is gone now; switch to ->iterate()
454	[mandatory]
455		vfs_follow_link has been removed.  Filesystems must use nd_set_link
456		from ->follow_link for normal symlinks, or nd_jump_link for magic
457		/proc/<pid> style links.
458	--
459	[mandatory]
460		iget5_locked()/ilookup5()/ilookup5_nowait() test() callback used to be
461		called with both ->i_lock and inode_hash_lock held; the former is *not*
462		taken anymore, so verify that your callbacks do not rely on it (none
463		of the in-tree instances did).  inode_hash_lock is still held,
464		of course, so they are still serialized wrt removal from inode hash,
465		as well as wrt set() callback of iget5_locked().
466	--
467	[mandatory]
468		d_materialise_unique() is gone; d_splice_alias() does everything you
469		need now.  Remember that they have opposite orders of arguments ;-/
470	--
471	[mandatory]
472		f_dentry is gone; use f_path.dentry, or, better yet, see if you can avoid
473		it entirely.
474	--
475	[mandatory]
476		never call ->read() and ->write() directly; use __vfs_{read,write} or
477		wrappers; instead of checking for ->write or ->read being NULL, look for
478		FMODE_CAN_{WRITE,READ} in file->f_mode.
479	--
480	[mandatory]
481		do _not_ use new_sync_{read,write} for ->read/->write; leave it NULL
482		instead.
483	--
484	[mandatory]
485		->aio_read/->aio_write are gone.  Use ->read_iter/->write_iter.
486	---
487	[recommended]
488		for embedded ("fast") symlinks just set inode->i_link to wherever the
489		symlink body is and use simple_follow_link() as ->follow_link().
490	--
491	[mandatory]
492		calling conventions for ->follow_link() have changed.  Instead of returning
493		cookie and using nd_set_link() to store the body to traverse, we return
494		the body to traverse and store the cookie using explicit void ** argument.
495		nameidata isn't passed at all - nd_jump_link() doesn't need it and
496		nd_[gs]et_link() is gone.
497	--
498	[mandatory]
499		calling conventions for ->put_link() have changed.  It gets inode instead of
500		dentry,  it does not get nameidata at all and it gets called only when cookie
501		is non-NULL.  Note that link body isn't available anymore, so if you need it,
502		store it as cookie.
503	--
504	[mandatory]
505		any symlink that might use page_follow_link_light/page_put_link() must
506		have inode_nohighmem(inode) called before anything might start playing with
507		its pagecache.  No highmem pages should end up in the pagecache of such
508		symlinks.  That includes any preseeding that might be done during symlink
509		creation.  __page_symlink() will honour the mapping gfp flags, so once
510		you've done inode_nohighmem() it's safe to use, but if you allocate and
511		insert the page manually, make sure to use the right gfp flags.
512	--
513	[mandatory]
514		->follow_link() is replaced with ->get_link(); same API, except that
515			* ->get_link() gets inode as a separate argument
516			* ->get_link() may be called in RCU mode - in that case NULL
517			  dentry is passed
518	--
519	[mandatory]
520		->get_link() gets struct delayed_call *done now, and should do
521		set_delayed_call() where it used to set *cookie.
522		->put_link() is gone - just give the destructor to set_delayed_call()
523		in ->get_link().
524	--
525	[mandatory]
526		->getxattr() and xattr_handler.get() get dentry and inode passed separately.
527		dentry might be yet to be attached to inode, so do _not_ use its ->d_inode
528		in the instances.  Rationale: !@#!@# security_d_instantiate() needs to be
529		called before we attach dentry to inode.
530	--
531	[mandatory]
532		symlinks are no longer the only inodes that do *not* have i_bdev/i_cdev/
533		i_pipe/i_link union zeroed out at inode eviction.  As the result, you can't
534		assume that non-NULL value in ->i_nlink at ->destroy_inode() implies that
535		it's a symlink.  Checking ->i_mode is really needed now.  In-tree we had
536		to fix shmem_destroy_callback() that used to take that kind of shortcut;
537		watch out, since that shortcut is no longer valid.
538	--
539	[mandatory]
540		->i_mutex is replaced with ->i_rwsem now.  inode_lock() et.al. work as
541		they used to - they just take it exclusive.  However, ->lookup() may be
542		called with parent locked shared.  Its instances must not
543			* use d_instantiate) and d_rehash() separately - use d_add() or
544			  d_splice_alias() instead.
545			* use d_rehash() alone - call d_add(new_dentry, NULL) instead.
546			* in the unlikely case when (read-only) access to filesystem
547			  data structures needs exclusion for some reason, arrange it
548			  yourself.  None of the in-tree filesystems needed that.
549			* rely on ->d_parent and ->d_name not changing after dentry has
550			  been fed to d_add() or d_splice_alias().  Again, none of the
551			  in-tree instances relied upon that.
552		We are guaranteed that lookups of the same name in the same directory
553		will not happen in parallel ("same" in the sense of your ->d_compare()).
554		Lookups on different names in the same directory can and do happen in
555		parallel now.
556	--
557	[recommended]
558		->iterate_shared() is added; it's a parallel variant of ->iterate().
559		Exclusion on struct file level is still provided (as well as that
560		between it and lseek on the same struct file), but if your directory
561		has been opened several times, you can get these called in parallel.
562		Exclusion between that method and all directory-modifying ones is
563		still provided, of course.
565		Often enough ->iterate() can serve as ->iterate_shared() without any
566		changes - it is a read-only operation, after all.  If you have any
567		per-inode or per-dentry in-core data structures modified by ->iterate(),
568		you might need something to serialize the access to them.  If you
569		do dcache pre-seeding, you'll need to switch to d_alloc_parallel() for
570		that; look for in-tree examples.
572		Old method is only used if the new one is absent; eventually it will
573		be removed.  Switch while you still can; the old one won't stay.
574	--
575	[mandatory]
576		->atomic_open() calls without O_CREAT may happen in parallel.
577	--
578	[mandatory]
579		->setxattr() and xattr_handler.set() get dentry and inode passed separately.
580		dentry might be yet to be attached to inode, so do _not_ use its ->d_inode
581		in the instances.  Rationale: !@#!@# security_d_instantiate() needs to be
582		called before we attach dentry to inode and !@#!@##!@$!$#!@#$!@$!@$ smack
583		->d_instantiate() uses not just ->getxattr() but ->setxattr() as well.
584	--
585	[mandatory]
586		->d_compare() doesn't get parent as a separate argument anymore.  If you
587		used it for finding the struct super_block involved, dentry->d_sb will
588		work just as well; if it's something more complicated, use dentry->d_parent.
589		Just be careful not to assume that fetching it more than once will yield
590		the same value - in RCU mode it could change under you.
591	--
592	[mandatory]
593		->rename() has an added flags argument.  Any flags not handled by the
594	        filesystem should result in EINVAL being returned.
595	--
596	[recommended]
597		->readlink is optional for symlinks.  Don't set, unless filesystem needs
598		to fake something for readlink(2).
599	--
600	[mandatory]
601		->getattr() is now passed a struct path rather than a vfsmount and
602		dentry separately, and it now has request_mask and query_flags arguments
603		to specify the fields and sync type requested by statx.  Filesystems not
604		supporting any statx-specific features may ignore the new arguments.
Hide Line Numbers
About Kernel Documentation Linux Kernel Contact Linux Resources Linux Blog

Information is copyright its respective author. All material is available from the Linux Kernel Source distributed under a GPL License. This page is provided as a free service by mjmwired.net.