1 ====================
2 Changes since 2.5.0:
3 ====================
4
5 ---
6
7 **recommended**
8
9 New helpers: sb_bread(), sb_getblk(), sb_find_get_block(), set_bh(),
10 sb_set_blocksize() and sb_min_blocksize().
11
12 Use them.
13
14 (sb_find_get_block() replaces 2.4's get_hash_table())
15
16 ---
17
18 **recommended**
19
20 New methods: ->alloc_inode() and ->destroy_inode().
21
22 Remove inode->u.foo_inode_i
23
24 Declare::
25
26 struct foo_inode_info {
27 /* fs-private stuff */
28 struct inode vfs_inode;
29 };
30 static inline struct foo_inode_info *FOO_I(struct inode *inode)
31 {
32 return list_entry(inode, struct foo_inode_info, vfs_inode);
33 }
34
35 Use FOO_I(inode) instead of &inode->u.foo_inode_i;
36
37 Add foo_alloc_inode() and foo_destroy_inode() - the former should allocate
38 foo_inode_info and return the address of ->vfs_inode, the latter should free
39 FOO_I(inode) (see in-tree filesystems for examples).
40
41 Make them ->alloc_inode and ->destroy_inode in your super_operations.
42
43 Keep in mind that now you need explicit initialization of private data
44 typically between calling iget_locked() and unlocking the inode.
45
46 At some point that will become mandatory.
47
48 **mandatory**
49
50 The foo_inode_info should always be allocated through alloc_inode_sb() rather
51 than kmem_cache_alloc() or kmalloc() related to set up the inode reclaim context
52 correctly.
53
54 ---
55
56 **mandatory**
57
58 Change of file_system_type method (->read_super to ->get_sb)
59
60 ->read_super() is no more. Ditto for DECLARE_FSTYPE and DECLARE_FSTYPE_DEV.
61
62 Turn your foo_read_super() into a function that would return 0 in case of
63 success and negative number in case of error (-EINVAL unless you have more
64 informative error value to report). Call it foo_fill_super(). Now declare::
65
66 int foo_get_sb(struct file_system_type *fs_type,
67 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
68 {
69 return get_sb_bdev(fs_type, flags, dev_name, data, foo_fill_super,
70 mnt);
71 }
72
73 (or similar with s/bdev/nodev/ or s/bdev/single/, depending on the kind of
74 filesystem).
75
76 Replace DECLARE_FSTYPE... with explicit initializer and have ->get_sb set as
77 foo_get_sb.
78
79 ---
80
81 **mandatory**
82
83 Locking change: ->s_vfs_rename_sem is taken only by cross-directory renames.
84 Most likely there is no need to change anything, but if you relied on
85 global exclusion between renames for some internal purpose - you need to
86 change your internal locking. Otherwise exclusion warranties remain the
87 same (i.e. parents and victim are locked, etc.).
88
89 ---
90
91 **informational**
92
93 Now we have the exclusion between ->lookup() and directory removal (by
94 ->rmdir() and ->rename()). If you used to need that exclusion and do
95 it by internal locking (most of filesystems couldn't care less) - you
96 can relax your locking.
97
98 ---
99
100 **mandatory**
101
102 ->lookup(), ->truncate(), ->create(), ->unlink(), ->mknod(), ->mkdir(),
103 ->rmdir(), ->link(), ->lseek(), ->symlink(), ->rename()
104 and ->readdir() are called without BKL now. Grab it on entry, drop upon return
105 - that will guarantee the same locking you used to have. If your method or its
106 parts do not need BKL - better yet, now you can shift lock_kernel() and
107 unlock_kernel() so that they would protect exactly what needs to be
108 protected.
109
110 ---
111
112 **mandatory**
113
114 BKL is also moved from around sb operations. BKL should have been shifted into
115 individual fs sb_op functions. If you don't need it, remove it.
116
117 ---
118
119 **informational**
120
121 check for ->link() target not being a directory is done by callers. Feel
122 free to drop it...
123
124 ---
125
126 **informational**
127
128 ->link() callers hold ->i_mutex on the object we are linking to. Some of your
129 problems might be over...
130
131 ---
132
133 **mandatory**
134
135 new file_system_type method - kill_sb(superblock). If you are converting
136 an existing filesystem, set it according to ->fs_flags::
137
138 FS_REQUIRES_DEV - kill_block_super
139 FS_LITTER - kill_litter_super
140 neither - kill_anon_super
141
142 FS_LITTER is gone - just remove it from fs_flags.
143
144 ---
145
146 **mandatory**
147
148 FS_SINGLE is gone (actually, that had happened back when ->get_sb()
149 went in - and hadn't been documented ;-/). Just remove it from fs_flags
150 (and see ->get_sb() entry for other actions).
151
152 ---
153
154 **mandatory**
155
156 ->setattr() is called without BKL now. Caller _always_ holds ->i_mutex, so
157 watch for ->i_mutex-grabbing code that might be used by your ->setattr().
158 Callers of notify_change() need ->i_mutex now.
159
160 ---
161
162 **recommended**
163
164 New super_block field ``struct export_operations *s_export_op`` for
165 explicit support for exporting, e.g. via NFS. The structure is fully
166 documented at its declaration in include/linux/fs.h, and in
167 Documentation/filesystems/nfs/exporting.rst.
168
169 Briefly it allows for the definition of decode_fh and encode_fh operations
170 to encode and decode filehandles, and allows the filesystem to use
171 a standard helper function for decode_fh, and provide file-system specific
172 support for this helper, particularly get_parent.
173
174 It is planned that this will be required for exporting once the code
175 settles down a bit.
176
177 **mandatory**
178
179 s_export_op is now required for exporting a filesystem.
180 isofs, ext2, ext3, reiserfs, fat
181 can be used as examples of very different filesystems.
182
183 ---
184
185 **mandatory**
186
187 iget4() and the read_inode2 callback have been superseded by iget5_locked()
188 which has the following prototype::
189
190 struct inode *iget5_locked(struct super_block *sb, unsigned long ino,
191 int (*test)(struct inode *, void *),
192 int (*set)(struct inode *, void *),
193 void *data);
194
195 'test' is an additional function that can be used when the inode
196 number is not sufficient to identify the actual file object. 'set'
197 should be a non-blocking function that initializes those parts of a
198 newly created inode to allow the test function to succeed. 'data' is
199 passed as an opaque value to both test and set functions.
200
201 When the inode has been created by iget5_locked(), it will be returned with the
202 I_NEW flag set and will still be locked. The filesystem then needs to finalize
203 the initialization. Once the inode is initialized it must be unlocked by
204 calling unlock_new_inode().
205
206 The filesystem is responsible for setting (and possibly testing) i_ino
207 when appropriate. There is also a simpler iget_locked function that
208 just takes the superblock and inode number as arguments and does the
209 test and set for you.
210
211 e.g.::
212
213 inode = iget_locked(sb, ino);
214 if (inode->i_state & I_NEW) {
215 err = read_inode_from_disk(inode);
216 if (err < 0) {
217 iget_failed(inode);
218 return err;
219 }
220 unlock_new_inode(inode);
221 }
222
223 Note that if the process of setting up a new inode fails, then iget_failed()
224 should be called on the inode to render it dead, and an appropriate error
225 should be passed back to the caller.
226
227 ---
228
229 **recommended**
230
231 ->getattr() finally getting used. See instances in nfs, minix, etc.
232
233 ---
234
235 **mandatory**
236
237 ->revalidate() is gone. If your filesystem had it - provide ->getattr()
238 and let it call whatever you had as ->revlidate() + (for symlinks that
239 had ->revalidate()) add calls in ->follow_link()/->readlink().
240
241 ---
242
243 **mandatory**
244
245 ->d_parent changes are not protected by BKL anymore. Read access is safe
246 if at least one of the following is true:
247
248 * filesystem has no cross-directory rename()
249 * we know that parent had been locked (e.g. we are looking at
250 ->d_parent of ->lookup() argument).
251 * we are called from ->rename().
252 * the child's ->d_lock is held
253
254 Audit your code and add locking if needed. Notice that any place that is
255 not protected by the conditions above is risky even in the old tree - you
256 had been relying on BKL and that's prone to screwups. Old tree had quite
257 a few holes of that kind - unprotected access to ->d_parent leading to
258 anything from oops to silent memory corruption.
259
260 ---
261
262 **mandatory**
263
264 FS_NOMOUNT is gone. If you use it - just set SB_NOUSER in flags
265 (see rootfs for one kind of solution and bdev/socket/pipe for another).
266
267 ---
268
269 **recommended**
270
271 Use bdev_read_only(bdev) instead of is_read_only(kdev). The latter
272 is still alive, but only because of the mess in drivers/s390/block/dasd.c.
273 As soon as it gets fixed is_read_only() will die.
274
275 ---
276
277 **mandatory**
278
279 ->permission() is called without BKL now. Grab it on entry, drop upon
280 return - that will guarantee the same locking you used to have. If
281 your method or its parts do not need BKL - better yet, now you can
282 shift lock_kernel() and unlock_kernel() so that they would protect
283 exactly what needs to be protected.
284
285 ---
286
287 **mandatory**
288
289 ->statfs() is now called without BKL held. BKL should have been
290 shifted into individual fs sb_op functions where it's not clear that
291 it's safe to remove it. If you don't need it, remove it.
292
293 ---
294
295 **mandatory**
296
297 is_read_only() is gone; use bdev_read_only() instead.
298
299 ---
300
301 **mandatory**
302
303 destroy_buffers() is gone; use invalidate_bdev().
304
305 ---
306
307 **mandatory**
308
309 fsync_dev() is gone; use fsync_bdev(). NOTE: lvm breakage is
310 deliberate; as soon as struct block_device * is propagated in a reasonable
311 way by that code fixing will become trivial; until then nothing can be
312 done.
313
314 **mandatory**
315
316 block truncatation on error exit from ->write_begin, and ->direct_IO
317 moved from generic methods (block_write_begin, cont_write_begin,
318 nobh_write_begin, blockdev_direct_IO*) to callers. Take a look at
319 ext2_write_failed and callers for an example.
320
321 **mandatory**
322
323 ->truncate is gone. The whole truncate sequence needs to be
324 implemented in ->setattr, which is now mandatory for filesystems
325 implementing on-disk size changes. Start with a copy of the old inode_setattr
326 and vmtruncate, and the reorder the vmtruncate + foofs_vmtruncate sequence to
327 be in order of zeroing blocks using block_truncate_page or similar helpers,
328 size update and on finally on-disk truncation which should not fail.
329 setattr_prepare (which used to be inode_change_ok) now includes the size checks
330 for ATTR_SIZE and must be called in the beginning of ->setattr unconditionally.
331
332 **mandatory**
333
334 ->clear_inode() and ->delete_inode() are gone; ->evict_inode() should
335 be used instead. It gets called whenever the inode is evicted, whether it has
336 remaining links or not. Caller does *not* evict the pagecache or inode-associated
337 metadata buffers; the method has to use truncate_inode_pages_final() to get rid
338 of those. Caller makes sure async writeback cannot be running for the inode while
339 (or after) ->evict_inode() is called.
340
341 ->drop_inode() returns int now; it's called on final iput() with
342 inode->i_lock held and it returns true if filesystems wants the inode to be
343 dropped. As before, generic_drop_inode() is still the default and it's been
344 updated appropriately. generic_delete_inode() is also alive and it consists
345 simply of return 1. Note that all actual eviction work is done by caller after
346 ->drop_inode() returns.
347
348 As before, clear_inode() must be called exactly once on each call of
349 ->evict_inode() (as it used to be for each call of ->delete_inode()). Unlike
350 before, if you are using inode-associated metadata buffers (i.e.
351 mark_buffer_dirty_inode()), it's your responsibility to call
352 invalidate_inode_buffers() before clear_inode().
353
354 NOTE: checking i_nlink in the beginning of ->write_inode() and bailing out
355 if it's zero is not *and* *never* *had* *been* enough. Final unlink() and iput()
356 may happen while the inode is in the middle of ->write_inode(); e.g. if you blindly
357 free the on-disk inode, you may end up doing that while ->write_inode() is writing
358 to it.
359
360 ---
361
362 **mandatory**
363
364 .d_delete() now only advises the dcache as to whether or not to cache
365 unreferenced dentries, and is now only called when the dentry refcount goes to
366 0. Even on 0 refcount transition, it must be able to tolerate being called 0,
367 1, or more times (eg. constant, idempotent).
368
369 ---
370
371 **mandatory**
372
373 .d_compare() calling convention and locking rules are significantly
374 changed. Read updated documentation in Documentation/filesystems/vfs.rst (and
375 look at examples of other filesystems) for guidance.
376
377 ---
378
379 **mandatory**
380
381 .d_hash() calling convention and locking rules are significantly
382 changed. Read updated documentation in Documentation/filesystems/vfs.rst (and
383 look at examples of other filesystems) for guidance.
384
385 ---
386
387 **mandatory**
388
389 dcache_lock is gone, replaced by fine grained locks. See fs/dcache.c
390 for details of what locks to replace dcache_lock with in order to protect
391 particular things. Most of the time, a filesystem only needs ->d_lock, which
392 protects *all* the dcache state of a given dentry.
393
394 ---
395
396 **mandatory**
397
398 Filesystems must RCU-free their inodes, if they can have been accessed
399 via rcu-walk path walk (basically, if the file can have had a path name in the
400 vfs namespace).
401
402 Even though i_dentry and i_rcu share storage in a union, we will
403 initialize the former in inode_init_always(), so just leave it alone in
404 the callback. It used to be necessary to clean it there, but not anymore
405 (starting at 3.2).
406
407 ---
408
409 **recommended**
410
411 vfs now tries to do path walking in "rcu-walk mode", which avoids
412 atomic operations and scalability hazards on dentries and inodes (see
413 Documentation/filesystems/path-lookup.txt). d_hash and d_compare changes
414 (above) are examples of the changes required to support this. For more complex
415 filesystem callbacks, the vfs drops out of rcu-walk mode before the fs call, so
416 no changes are required to the filesystem. However, this is costly and loses
417 the benefits of rcu-walk mode. We will begin to add filesystem callbacks that
418 are rcu-walk aware, shown below. Filesystems should take advantage of this
419 where possible.
420
421 ---
422
423 **mandatory**
424
425 d_revalidate is a callback that is made on every path element (if
426 the filesystem provides it), which requires dropping out of rcu-walk mode. This
427 may now be called in rcu-walk mode (nd->flags & LOOKUP_RCU). -ECHILD should be
428 returned if the filesystem cannot handle rcu-walk. See
429 Documentation/filesystems/vfs.rst for more details.
430
431 permission is an inode permission check that is called on many or all
432 directory inodes on the way down a path walk (to check for exec permission). It
433 must now be rcu-walk aware (mask & MAY_NOT_BLOCK). See
434 Documentation/filesystems/vfs.rst for more details.
435
436 ---
437
438 **mandatory**
439
440 In ->fallocate() you must check the mode option passed in. If your
441 filesystem does not support hole punching (deallocating space in the middle of a
442 file) you must return -EOPNOTSUPP if FALLOC_FL_PUNCH_HOLE is set in mode.
443 Currently you can only have FALLOC_FL_PUNCH_HOLE with FALLOC_FL_KEEP_SIZE set,
444 so the i_size should not change when hole punching, even when puching the end of
445 a file off.
446
447 ---
448
449 **mandatory**
450
451 ->get_sb() is gone. Switch to use of ->mount(). Typically it's just
452 a matter of switching from calling ``get_sb_``... to ``mount_``... and changing
453 the function type. If you were doing it manually, just switch from setting
454 ->mnt_root to some pointer to returning that pointer. On errors return
455 ERR_PTR(...).
456
457 ---
458
459 **mandatory**
460
461 ->permission() and generic_permission()have lost flags
462 argument; instead of passing IPERM_FLAG_RCU we add MAY_NOT_BLOCK into mask.
463
464 generic_permission() has also lost the check_acl argument; ACL checking
465 has been taken to VFS and filesystems need to provide a non-NULL
466 ->i_op->get_inode_acl to read an ACL from disk.
467
468 ---
469
470 **mandatory**
471
472 If you implement your own ->llseek() you must handle SEEK_HOLE and
473 SEEK_DATA. You can handle this by returning -EINVAL, but it would be nicer to
474 support it in some way. The generic handler assumes that the entire file is
475 data and there is a virtual hole at the end of the file. So if the provided
476 offset is less than i_size and SEEK_DATA is specified, return the same offset.
477 If the above is true for the offset and you are given SEEK_HOLE, return the end
478 of the file. If the offset is i_size or greater return -ENXIO in either case.
479
480 **mandatory**
481
482 If you have your own ->fsync() you must make sure to call
483 filemap_write_and_wait_range() so that all dirty pages are synced out properly.
484 You must also keep in mind that ->fsync() is not called with i_mutex held
485 anymore, so if you require i_mutex locking you must make sure to take it and
486 release it yourself.
487
488 ---
489
490 **mandatory**
491
492 d_alloc_root() is gone, along with a lot of bugs caused by code
493 misusing it. Replacement: d_make_root(inode). On success d_make_root(inode)
494 allocates and returns a new dentry instantiated with the passed in inode.
495 On failure NULL is returned and the passed in inode is dropped so the reference
496 to inode is consumed in all cases and failure handling need not do any cleanup
497 for the inode. If d_make_root(inode) is passed a NULL inode it returns NULL
498 and also requires no further error handling. Typical usage is::
499
500 inode = foofs_new_inode(....);
501 s->s_root = d_make_root(inode);
502 if (!s->s_root)
503 /* Nothing needed for the inode cleanup */
504 return -ENOMEM;
505 ...
506
507 ---
508
509 **mandatory**
510
511 The witch is dead! Well, 2/3 of it, anyway. ->d_revalidate() and
512 ->lookup() do *not* take struct nameidata anymore; just the flags.
513
514 ---
515
516 **mandatory**
517
518 ->create() doesn't take ``struct nameidata *``; unlike the previous
519 two, it gets "is it an O_EXCL or equivalent?" boolean argument. Note that
520 local filesystems can ignore this argument - they are guaranteed that the
521 object doesn't exist. It's remote/distributed ones that might care...
522
523 ---
524
525 **mandatory**
526
527 FS_REVAL_DOT is gone; if you used to have it, add ->d_weak_revalidate()
528 in your dentry operations instead.
529
530 ---
531
532 **mandatory**
533
534 vfs_readdir() is gone; switch to iterate_dir() instead
535
536 ---
537
538 **mandatory**
539
540 ->readdir() is gone now; switch to ->iterate_shared()
541
542 **mandatory**
543
544 vfs_follow_link has been removed. Filesystems must use nd_set_link
545 from ->follow_link for normal symlinks, or nd_jump_link for magic
546 /proc/<pid> style links.
547
548 ---
549
550 **mandatory**
551
552 iget5_locked()/ilookup5()/ilookup5_nowait() test() callback used to be
553 called with both ->i_lock and inode_hash_lock held; the former is *not*
554 taken anymore, so verify that your callbacks do not rely on it (none
555 of the in-tree instances did). inode_hash_lock is still held,
556 of course, so they are still serialized wrt removal from inode hash,
557 as well as wrt set() callback of iget5_locked().
558
559 ---
560
561 **mandatory**
562
563 d_materialise_unique() is gone; d_splice_alias() does everything you
564 need now. Remember that they have opposite orders of arguments ;-/
565
566 ---
567
568 **mandatory**
569
570 f_dentry is gone; use f_path.dentry, or, better yet, see if you can avoid
571 it entirely.
572
573 ---
574
575 **mandatory**
576
577 never call ->read() and ->write() directly; use __vfs_{read,write} or
578 wrappers; instead of checking for ->write or ->read being NULL, look for
579 FMODE_CAN_{WRITE,READ} in file->f_mode.
580
581 ---
582
583 **mandatory**
584
585 do _not_ use new_sync_{read,write} for ->read/->write; leave it NULL
586 instead.
587
588 ---
589
590 **mandatory**
591 ->aio_read/->aio_write are gone. Use ->read_iter/->write_iter.
592
593 ---
594
595 **recommended**
596
597 for embedded ("fast") symlinks just set inode->i_link to wherever the
598 symlink body is and use simple_follow_link() as ->follow_link().
599
600 ---
601
602 **mandatory**
603
604 calling conventions for ->follow_link() have changed. Instead of returning
605 cookie and using nd_set_link() to store the body to traverse, we return
606 the body to traverse and store the cookie using explicit void ** argument.
607 nameidata isn't passed at all - nd_jump_link() doesn't need it and
608 nd_[gs]et_link() is gone.
609
610 ---
611
612 **mandatory**
613
614 calling conventions for ->put_link() have changed. It gets inode instead of
615 dentry, it does not get nameidata at all and it gets called only when cookie
616 is non-NULL. Note that link body isn't available anymore, so if you need it,
617 store it as cookie.
618
619 ---
620
621 **mandatory**
622
623 any symlink that might use page_follow_link_light/page_put_link() must
624 have inode_nohighmem(inode) called before anything might start playing with
625 its pagecache. No highmem pages should end up in the pagecache of such
626 symlinks. That includes any preseeding that might be done during symlink
627 creation. page_symlink() will honour the mapping gfp flags, so once
628 you've done inode_nohighmem() it's safe to use, but if you allocate and
629 insert the page manually, make sure to use the right gfp flags.
630
631 ---
632
633 **mandatory**
634
635 ->follow_link() is replaced with ->get_link(); same API, except that
636
637 * ->get_link() gets inode as a separate argument
638 * ->get_link() may be called in RCU mode - in that case NULL
639 dentry is passed
640
641 ---
642
643 **mandatory**
644
645 ->get_link() gets struct delayed_call ``*done`` now, and should do
646 set_delayed_call() where it used to set ``*cookie``.
647
648 ->put_link() is gone - just give the destructor to set_delayed_call()
649 in ->get_link().
650
651 ---
652
653 **mandatory**
654
655 ->getxattr() and xattr_handler.get() get dentry and inode passed separately.
656 dentry might be yet to be attached to inode, so do _not_ use its ->d_inode
657 in the instances. Rationale: !@#!@# security_d_instantiate() needs to be
658 called before we attach dentry to inode.
659
660 ---
661
662 **mandatory**
663
664 symlinks are no longer the only inodes that do *not* have i_bdev/i_cdev/
665 i_pipe/i_link union zeroed out at inode eviction. As the result, you can't
666 assume that non-NULL value in ->i_nlink at ->destroy_inode() implies that
667 it's a symlink. Checking ->i_mode is really needed now. In-tree we had
668 to fix shmem_destroy_callback() that used to take that kind of shortcut;
669 watch out, since that shortcut is no longer valid.
670
671 ---
672
673 **mandatory**
674
675 ->i_mutex is replaced with ->i_rwsem now. inode_lock() et.al. work as
676 they used to - they just take it exclusive. However, ->lookup() may be
677 called with parent locked shared. Its instances must not
678
679 * use d_instantiate) and d_rehash() separately - use d_add() or
680 d_splice_alias() instead.
681 * use d_rehash() alone - call d_add(new_dentry, NULL) instead.
682 * in the unlikely case when (read-only) access to filesystem
683 data structures needs exclusion for some reason, arrange it
684 yourself. None of the in-tree filesystems needed that.
685 * rely on ->d_parent and ->d_name not changing after dentry has
686 been fed to d_add() or d_splice_alias(). Again, none of the
687 in-tree instances relied upon that.
688
689 We are guaranteed that lookups of the same name in the same directory
690 will not happen in parallel ("same" in the sense of your ->d_compare()).
691 Lookups on different names in the same directory can and do happen in
692 parallel now.
693
694 ---
695
696 **mandatory**
697
698 ->iterate_shared() is added.
699 Exclusion on struct file level is still provided (as well as that
700 between it and lseek on the same struct file), but if your directory
701 has been opened several times, you can get these called in parallel.
702 Exclusion between that method and all directory-modifying ones is
703 still provided, of course.
704
705 If you have any per-inode or per-dentry in-core data structures modified
706 by ->iterate_shared(), you might need something to serialize the access
707 to them. If you do dcache pre-seeding, you'll need to switch to
708 d_alloc_parallel() for that; look for in-tree examples.
709
710 ---
711
712 **mandatory**
713
714 ->atomic_open() calls without O_CREAT may happen in parallel.
715
716 ---
717
718 **mandatory**
719
720 ->setxattr() and xattr_handler.set() get dentry and inode passed separately.
721 The xattr_handler.set() gets passed the user namespace of the mount the inode
722 is seen from so filesystems can idmap the i_uid and i_gid accordingly.
723 dentry might be yet to be attached to inode, so do _not_ use its ->d_inode
724 in the instances. Rationale: !@#!@# security_d_instantiate() needs to be
725 called before we attach dentry to inode and !@#!@##!@$!$#!@#$!@$!@$ smack
726 ->d_instantiate() uses not just ->getxattr() but ->setxattr() as well.
727
728 ---
729
730 **mandatory**
731
732 ->d_compare() doesn't get parent as a separate argument anymore. If you
733 used it for finding the struct super_block involved, dentry->d_sb will
734 work just as well; if it's something more complicated, use dentry->d_parent.
735 Just be careful not to assume that fetching it more than once will yield
736 the same value - in RCU mode it could change under you.
737
738 ---
739
740 **mandatory**
741
742 ->rename() has an added flags argument. Any flags not handled by the
743 filesystem should result in EINVAL being returned.
744
745 ---
746
747
748 **recommended**
749
750 ->readlink is optional for symlinks. Don't set, unless filesystem needs
751 to fake something for readlink(2).
752
753 ---
754
755 **mandatory**
756
757 ->getattr() is now passed a struct path rather than a vfsmount and
758 dentry separately, and it now has request_mask and query_flags arguments
759 to specify the fields and sync type requested by statx. Filesystems not
760 supporting any statx-specific features may ignore the new arguments.
761
762 ---
763
764 **mandatory**
765
766 ->atomic_open() calling conventions have changed. Gone is ``int *opened``,
767 along with FILE_OPENED/FILE_CREATED. In place of those we have
768 FMODE_OPENED/FMODE_CREATED, set in file->f_mode. Additionally, return
769 value for 'called finish_no_open(), open it yourself' case has become
770 0, not 1. Since finish_no_open() itself is returning 0 now, that part
771 does not need any changes in ->atomic_open() instances.
772
773 ---
774
775 **mandatory**
776
777 alloc_file() has become static now; two wrappers are to be used instead.
778 alloc_file_pseudo(inode, vfsmount, name, flags, ops) is for the cases
779 when dentry needs to be created; that's the majority of old alloc_file()
780 users. Calling conventions: on success a reference to new struct file
781 is returned and callers reference to inode is subsumed by that. On
782 failure, ERR_PTR() is returned and no caller's references are affected,
783 so the caller needs to drop the inode reference it held.
784 alloc_file_clone(file, flags, ops) does not affect any caller's references.
785 On success you get a new struct file sharing the mount/dentry with the
786 original, on failure - ERR_PTR().
787
788 ---
789
790 **mandatory**
791
792 ->clone_file_range() and ->dedupe_file_range have been replaced with
793 ->remap_file_range(). See Documentation/filesystems/vfs.rst for more
794 information.
795
796 ---
797
798 **recommended**
799
800 ->lookup() instances doing an equivalent of::
801
802 if (IS_ERR(inode))
803 return ERR_CAST(inode);
804 return d_splice_alias(inode, dentry);
805
806 don't need to bother with the check - d_splice_alias() will do the
807 right thing when given ERR_PTR(...) as inode. Moreover, passing NULL
808 inode to d_splice_alias() will also do the right thing (equivalent of
809 d_add(dentry, NULL); return NULL;), so that kind of special cases
810 also doesn't need a separate treatment.
811
812 ---
813
814 **strongly recommended**
815
816 take the RCU-delayed parts of ->destroy_inode() into a new method -
817 ->free_inode(). If ->destroy_inode() becomes empty - all the better,
818 just get rid of it. Synchronous work (e.g. the stuff that can't
819 be done from an RCU callback, or any WARN_ON() where we want the
820 stack trace) *might* be movable to ->evict_inode(); however,
821 that goes only for the things that are not needed to balance something
822 done by ->alloc_inode(). IOW, if it's cleaning up the stuff that
823 might have accumulated over the life of in-core inode, ->evict_inode()
824 might be a fit.
825
826 Rules for inode destruction:
827
828 * if ->destroy_inode() is non-NULL, it gets called
829 * if ->free_inode() is non-NULL, it gets scheduled by call_rcu()
830 * combination of NULL ->destroy_inode and NULL ->free_inode is
831 treated as NULL/free_inode_nonrcu, to preserve the compatibility.
832
833 Note that the callback (be it via ->free_inode() or explicit call_rcu()
834 in ->destroy_inode()) is *NOT* ordered wrt superblock destruction;
835 as the matter of fact, the superblock and all associated structures
836 might be already gone. The filesystem driver is guaranteed to be still
837 there, but that's it. Freeing memory in the callback is fine; doing
838 more than that is possible, but requires a lot of care and is best
839 avoided.
840
841 ---
842
843 **mandatory**
844
845 DCACHE_RCUACCESS is gone; having an RCU delay on dentry freeing is the
846 default. DCACHE_NORCU opts out, and only d_alloc_pseudo() has any
847 business doing so.
848
849 ---
850
851 **mandatory**
852
853 d_alloc_pseudo() is internal-only; uses outside of alloc_file_pseudo() are
854 very suspect (and won't work in modules). Such uses are very likely to
855 be misspelled d_alloc_anon().
856
857 ---
858
859 **mandatory**
860
861 [should've been added in 2016] stale comment in finish_open() notwithstanding,
862 failure exits in ->atomic_open() instances should *NOT* fput() the file,
863 no matter what. Everything is handled by the caller.
864
865 ---
866
867 **mandatory**
868
869 clone_private_mount() returns a longterm mount now, so the proper destructor of
870 its result is kern_unmount() or kern_unmount_array().
871
872 ---
873
874 **mandatory**
875
876 zero-length bvec segments are disallowed, they must be filtered out before
877 passed on to an iterator.
878
879 ---
880
881 **mandatory**
882
883 For bvec based itererators bio_iov_iter_get_pages() now doesn't copy bvecs but
884 uses the one provided. Anyone issuing kiocb-I/O should ensure that the bvec and
885 page references stay until I/O has completed, i.e. until ->ki_complete() has
886 been called or returned with non -EIOCBQUEUED code.
887
888 ---
889
890 **mandatory**
891
892 mnt_want_write_file() can now only be paired with mnt_drop_write_file(),
893 whereas previously it could be paired with mnt_drop_write() as well.
894
895 ---
896
897 **mandatory**
898
899 iov_iter_copy_from_user_atomic() is gone; use copy_page_from_iter_atomic().
900 The difference is copy_page_from_iter_atomic() advances the iterator and
901 you don't need iov_iter_advance() after it. However, if you decide to use
902 only a part of obtained data, you should do iov_iter_revert().
903
904 ---
905
906 **mandatory**
907
908 Calling conventions for file_open_root() changed; now it takes struct path *
909 instead of passing mount and dentry separately. For callers that used to
910 pass <mnt, mnt->mnt_root> pair (i.e. the root of given mount), a new helper
911 is provided - file_open_root_mnt(). In-tree users adjusted.
912
913 ---
914
915 **mandatory**
916
917 no_llseek is gone; don't set .llseek to that - just leave it NULL instead.
918 Checks for "does that file have llseek(2), or should it fail with ESPIPE"
919 should be done by looking at FMODE_LSEEK in file->f_mode.
920
921 ---
922
923 *mandatory*
924
925 filldir_t (readdir callbacks) calling conventions have changed. Instead of
926 returning 0 or -E... it returns bool now. false means "no more" (as -E... used
927 to) and true - "keep going" (as 0 in old calling conventions). Rationale:
928 callers never looked at specific -E... values anyway. -> iterate_shared()
929 instances require no changes at all, all filldir_t ones in the tree
930 converted.
931
932 ---
933
934 **mandatory**
935
936 Calling conventions for ->tmpfile() have changed. It now takes a struct
937 file pointer instead of struct dentry pointer. d_tmpfile() is similarly
938 changed to simplify callers. The passed file is in a non-open state and on
939 success must be opened before returning (e.g. by calling
940 finish_open_simple()).
941
942 ---
943
944 **mandatory**
945
946 Calling convention for ->huge_fault has changed. It now takes a page
947 order instead of an enum page_entry_size, and it may be called without the
948 mmap_lock held. All in-tree users have been audited and do not seem to
949 depend on the mmap_lock being held, but out of tree users should verify
950 for themselves. If they do need it, they can return VM_FAULT_RETRY to
951 be called with the mmap_lock held.
952
953 ---
954
955 **mandatory**
956
957 The order of opening block devices and matching or creating superblocks has
958 changed.
959
960 The old logic opened block devices first and then tried to find a
961 suitable superblock to reuse based on the block device pointer.
962
963 The new logic tries to find a suitable superblock first based on the device
964 number, and opening the block device afterwards.
965
966 Since opening block devices cannot happen under s_umount because of lock
967 ordering requirements s_umount is now dropped while opening block devices and
968 reacquired before calling fill_super().
969
970 In the old logic concurrent mounters would find the superblock on the list of
971 superblocks for the filesystem type. Since the first opener of the block device
972 would hold s_umount they would wait until the superblock became either born or
973 was discarded due to initialization failure.
974
975 Since the new logic drops s_umount concurrent mounters could grab s_umount and
976 would spin. Instead they are now made to wait using an explicit wait-wake
977 mechanism without having to hold s_umount.
978
979 ---
980
981 **mandatory**
982
983 The holder of a block device is now the superblock.
984
985 The holder of a block device used to be the file_system_type which wasn't
986 particularly useful. It wasn't possible to go from block device to owning
987 superblock without matching on the device pointer stored in the superblock.
988 This mechanism would only work for a single device so the block layer couldn't
989 find the owning superblock of any additional devices.
990
991 In the old mechanism reusing or creating a superblock for a racing mount(2) and
992 umount(2) relied on the file_system_type as the holder. This was severely
993 underdocumented however:
994
995 (1) Any concurrent mounter that managed to grab an active reference on an
996 existing superblock was made to wait until the superblock either became
997 ready or until the superblock was removed from the list of superblocks of
998 the filesystem type. If the superblock is ready the caller would simple
999 reuse it.
1000
1001 (2) If the mounter came after deactivate_locked_super() but before
1002 the superblock had been removed from the list of superblocks of the
1003 filesystem type the mounter would wait until the superblock was shutdown,
1004 reuse the block device and allocate a new superblock.
1005
1006 (3) If the mounter came after deactivate_locked_super() and after
1007 the superblock had been removed from the list of superblocks of the
1008 filesystem type the mounter would reuse the block device and allocate a new
1009 superblock (the bd_holder point may still be set to the filesystem type).
1010
1011 Because the holder of the block device was the file_system_type any concurrent
1012 mounter could open the block devices of any superblock of the same
1013 file_system_type without risking seeing EBUSY because the block device was
1014 still in use by another superblock.
1015
1016 Making the superblock the owner of the block device changes this as the holder
1017 is now a unique superblock and thus block devices associated with it cannot be
1018 reused by concurrent mounters. So a concurrent mounter in (2) could suddenly
1019 see EBUSY when trying to open a block device whose holder was a different
1020 superblock.
1021
1022 The new logic thus waits until the superblock and the devices are shutdown in
1023 ->kill_sb(). Removal of the superblock from the list of superblocks of the
1024 filesystem type is now moved to a later point when the devices are closed:
1025
1026 (1) Any concurrent mounter managing to grab an active reference on an existing
1027 superblock is made to wait until the superblock is either ready or until
1028 the superblock and all devices are shutdown in ->kill_sb(). If the
1029 superblock is ready the caller will simply reuse it.
1030
1031 (2) If the mounter comes after deactivate_locked_super() but before
1032 the superblock has been removed from the list of superblocks of the
1033 filesystem type the mounter is made to wait until the superblock and the
1034 devices are shut down in ->kill_sb() and the superblock is removed from the
1035 list of superblocks of the filesystem type. The mounter will allocate a new
1036 superblock and grab ownership of the block device (the bd_holder pointer of
1037 the block device will be set to the newly allocated superblock).
1038
1039 (3) This case is now collapsed into (2) as the superblock is left on the list
1040 of superblocks of the filesystem type until all devices are shutdown in
1041 ->kill_sb(). In other words, if the superblock isn't on the list of
1042 superblock of the filesystem type anymore then it has given up ownership of
1043 all associated block devices (the bd_holder pointer is NULL).
1044
1045 As this is a VFS level change it has no practical consequences for filesystems
1046 other than that all of them must use one of the provided kill_litter_super(),
1047 kill_anon_super(), or kill_block_super() helpers.
1048
1049 ---
1050
1051 **mandatory**
1052
1053 Lock ordering has been changed so that s_umount ranks above open_mutex again.
1054 All places where s_umount was taken under open_mutex have been fixed up.
1055
1056 ---
1057
1058 **mandatory**
1059
1060 export_operations ->encode_fh() no longer has a default implementation to
1061 encode FILEID_INO32_GEN* file handles.
1062 Filesystems that used the default implementation may use the generic helper
1063 generic_encode_ino32_fh() explicitly.
1064
1065 ---
1066
1067 **mandatory**
1068
1069 If ->rename() update of .. on cross-directory move needs an exclusion with
1070 directory modifications, do *not* lock the subdirectory in question in your
1071 ->rename() - it's done by the caller now [that item should've been added in
1072 28eceeda130f "fs: Lock moved directories"].
1073
1074 ---
1075
1076 **mandatory**
1077
1078 On same-directory ->rename() the (tautological) update of .. is not protected
1079 by any locks; just don't do it if the old parent is the same as the new one.
1080 We really can't lock two subdirectories in same-directory rename - not without
1081 deadlocks.
1082
1083 ---
1084
1085 **mandatory**
1086
1087 lock_rename() and lock_rename_child() may fail in cross-directory case, if
1088 their arguments do not have a common ancestor. In that case ERR_PTR(-EXDEV)
1089 is returned, with no locks taken. In-tree users updated; out-of-tree ones
1090 would need to do so.
1091
1092 ---
1093
1094 **mandatory**
1095
1096 The list of children anchored in parent dentry got turned into hlist now.
1097 Field names got changed (->d_children/->d_sib instead of ->d_subdirs/->d_child
1098 for anchor/entries resp.), so any affected places will be immediately caught
1099 by compiler.
1100
1101 ---
1102
1103 **mandatory**
1104
1105 ->d_delete() instances are now called for dentries with ->d_lock held
1106 and refcount equal to 0. They are not permitted to drop/regain ->d_lock.
1107 None of in-tree instances did anything of that sort. Make sure yours do not...
1108
1109 ---
1110
1111 **mandatory**
1112
1113 ->d_prune() instances are now called without ->d_lock held on the parent.
1114 ->d_lock on dentry itself is still held; if you need per-parent exclusions (none
1115 of the in-tree instances did), use your own spinlock.
1116
1117 ->d_iput() and ->d_release() are called with victim dentry still in the
1118 list of parent's children. It is still unhashed, marked killed, etc., just not
1119 removed from parent's ->d_children yet.
1120
1121 Anyone iterating through the list of children needs to be aware of the
1122 half-killed dentries that might be seen there; taking ->d_lock on those will
1123 see them negative, unhashed and with negative refcount, which means that most
1124 of the in-kernel users would've done the right thing anyway without any adjustment.
1125
1126 ---
1127
1128 **recommended**
1129
1130 Block device freezing and thawing have been moved to holder operations.
1131
1132 Before this change, get_active_super() would only be able to find the
1133 superblock of the main block device, i.e., the one stored in sb->s_bdev. Block
1134 device freezing now works for any block device owned by a given superblock, not
1135 just the main block device. The get_active_super() helper and bd_fsfreeze_sb
1136 pointer are gone.
1137
1138 ---
1139
1140 **mandatory**
1141
1142 set_blocksize() takes opened struct file instead of struct block_device now
1143 and it *must* be opened exclusive.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.