Root/Documentation/filesystems/porting

1Changes since 2.5.0:
2
3---
4[recommended]
5
6New helpers: sb_bread(), sb_getblk(), sb_find_get_block(), set_bh(),
7    sb_set_blocksize() and sb_min_blocksize().
8
9Use them.
10
11(sb_find_get_block() replaces 2.4's get_hash_table())
12
13---
14[recommended]
15
16New methods: ->alloc_inode() and ->destroy_inode().
17
18Remove inode->u.foo_inode_i
19Declare
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    }
28
29Use FOO_I(inode) instead of &inode->u.foo_inode_i;
30
31Add foo_alloc_inode() and foo_destroy_inode() - the former should allocate
32foo_inode_info and return the address of ->vfs_inode, the latter should free
33FOO_I(inode) (see in-tree filesystems for examples).
34
35Make them ->alloc_inode and ->destroy_inode in your super_operations.
36
37Keep in mind that now you need explicit initialization of private data
38typically between calling iget_locked() and unlocking the inode.
39
40At some point that will become mandatory.
41
42---
43[mandatory]
44
45Change of file_system_type method (->read_super to ->get_sb)
46
47->read_super() is no more. Ditto for DECLARE_FSTYPE and DECLARE_FSTYPE_DEV.
48
49Turn your foo_read_super() into a function that would return 0 in case of
50success and negative number in case of error (-EINVAL unless you have more
51informative error value to report). Call it foo_fill_super(). Now declare
52
53int 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}
59
60(or similar with s/bdev/nodev/ or s/bdev/single/, depending on the kind of
61filesystem).
62
63Replace DECLARE_FSTYPE... with explicit initializer and have ->get_sb set as
64foo_get_sb.
65
66---
67[mandatory]
68
69Locking change: ->s_vfs_rename_sem is taken only by cross-directory renames.
70Most likely there is no need to change anything, but if you relied on
71global exclusion between renames for some internal purpose - you need to
72change your internal locking. Otherwise exclusion warranties remain the
73same (i.e. parents and victim are locked, etc.).
74
75---
76[informational]
77
78Now we have the exclusion between ->lookup() and directory removal (by
79->rmdir() and ->rename()). If you used to need that exclusion and do
80it by internal locking (most of filesystems couldn't care less) - you
81can relax your locking.
82
83---
84[mandatory]
85
86->lookup(), ->truncate(), ->create(), ->unlink(), ->mknod(), ->mkdir(),
87->rmdir(), ->link(), ->lseek(), ->symlink(), ->rename()
88and ->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
90parts do not need BKL - better yet, now you can shift lock_kernel() and
91unlock_kernel() so that they would protect exactly what needs to be
92protected.
93
94---
95[mandatory]
96
97BKL is also moved from around sb operations. ->write_super() Is now called
98without BKL held. BKL should have been shifted into individual fs sb_op
99functions. If you don't need it, remove it.
100
101---
102[informational]
103
104check for ->link() target not being a directory is done by callers. Feel
105free to drop it...
106
107---
108[informational]
109
110->link() callers hold ->i_mutex on the object we are linking to. Some of your
111problems might be over...
112
113---
114[mandatory]
115
116new file_system_type method - kill_sb(superblock). If you are converting
117an existing filesystem, set it according to ->fs_flags:
118    FS_REQUIRES_DEV - kill_block_super
119    FS_LITTER - kill_litter_super
120    neither - kill_anon_super
121FS_LITTER is gone - just remove it from fs_flags.
122
123---
124[mandatory]
125
126    FS_SINGLE is gone (actually, that had happened back when ->get_sb()
127went in - and hadn't been documented ;-/). Just remove it from fs_flags
128(and see ->get_sb() entry for other actions).
129
130---
131[mandatory]
132
133->setattr() is called without BKL now. Caller _always_ holds ->i_mutex, so
134watch for ->i_mutex-grabbing code that might be used by your ->setattr().
135Callers of notify_change() need ->i_mutex now.
136
137---
138[recommended]
139
140New super_block field "struct export_operations *s_export_op" for
141explicit support for exporting, e.g. via NFS. The structure is fully
142documented at its declaration in include/linux/fs.h, and in
143Documentation/filesystems/nfs/Exporting.
144
145Briefly it allows for the definition of decode_fh and encode_fh operations
146to encode and decode filehandles, and allows the filesystem to use
147a standard helper function for decode_fh, and provide file-system specific
148support for this helper, particularly get_parent.
149
150It is planned that this will be required for exporting once the code
151settles down a bit.
152
153[mandatory]
154
155s_export_op is now required for exporting a filesystem.
156isofs, ext2, ext3, resierfs, fat
157can be used as examples of very different filesystems.
158
159---
160[mandatory]
161
162iget4() and the read_inode2 callback have been superseded by iget5_locked()
163which has the following prototype,
164
165    struct inode *iget5_locked(struct super_block *sb, unsigned long ino,
166                int (*test)(struct inode *, void *),
167                int (*set)(struct inode *, void *),
168                void *data);
169
170'test' is an additional function that can be used when the inode
171number is not sufficient to identify the actual file object. 'set'
172should be a non-blocking function that initializes those parts of a
173newly created inode to allow the test function to succeed. 'data' is
174passed as an opaque value to both test and set functions.
175
176When the inode has been created by iget5_locked(), it will be returned with the
177I_NEW flag set and will still be locked. The filesystem then needs to finalize
178the initialization. Once the inode is initialized it must be unlocked by
179calling unlock_new_inode().
180
181The filesystem is responsible for setting (and possibly testing) i_ino
182when appropriate. There is also a simpler iget_locked function that
183just takes the superblock and inode number as arguments and does the
184test and set for you.
185
186e.g.
187    inode = iget_locked(sb, ino);
188    if (inode->i_state & I_NEW) {
189        err = read_inode_from_disk(inode);
190        if (err < 0) {
191            iget_failed(inode);
192            return err;
193        }
194        unlock_new_inode(inode);
195    }
196
197Note that if the process of setting up a new inode fails, then iget_failed()
198should be called on the inode to render it dead, and an appropriate error
199should be passed back to the caller.
200
201---
202[recommended]
203
204->getattr() finally getting used. See instances in nfs, minix, etc.
205
206---
207[mandatory]
208
209->revalidate() is gone. If your filesystem had it - provide ->getattr()
210and let it call whatever you had as ->revlidate() + (for symlinks that
211had ->revalidate()) add calls in ->follow_link()/->readlink().
212
213---
214[mandatory]
215
216->d_parent changes are not protected by BKL anymore. Read access is safe
217if at least one of the following is true:
218    * filesystem has no cross-directory rename()
219    * we know that parent had been locked (e.g. we are looking at
220->d_parent of ->lookup() argument).
221    * we are called from ->rename().
222    * the child's ->d_lock is held
223Audit your code and add locking if needed. Notice that any place that is
224not protected by the conditions above is risky even in the old tree - you
225had been relying on BKL and that's prone to screwups. Old tree had quite
226a few holes of that kind - unprotected access to ->d_parent leading to
227anything from oops to silent memory corruption.
228
229---
230[mandatory]
231
232    FS_NOMOUNT is gone. If you use it - just set MS_NOUSER in flags
233(see rootfs for one kind of solution and bdev/socket/pipe for another).
234
235---
236[recommended]
237
238    Use bdev_read_only(bdev) instead of is_read_only(kdev). The latter
239is still alive, but only because of the mess in drivers/s390/block/dasd.c.
240As soon as it gets fixed is_read_only() will die.
241
242---
243[mandatory]
244
245->permission() is called without BKL now. Grab it on entry, drop upon
246return - that will guarantee the same locking you used to have. If
247your method or its parts do not need BKL - better yet, now you can
248shift lock_kernel() and unlock_kernel() so that they would protect
249exactly what needs to be protected.
250
251---
252[mandatory]
253
254->statfs() is now called without BKL held. BKL should have been
255shifted into individual fs sb_op functions where it's not clear that
256it's safe to remove it. If you don't need it, remove it.
257
258---
259[mandatory]
260
261    is_read_only() is gone; use bdev_read_only() instead.
262
263---
264[mandatory]
265
266    destroy_buffers() is gone; use invalidate_bdev().
267
268---
269[mandatory]
270
271    fsync_dev() is gone; use fsync_bdev(). NOTE: lvm breakage is
272deliberate; as soon as struct block_device * is propagated in a reasonable
273way by that code fixing will become trivial; until then nothing can be
274done.
275
276[mandatory]
277
278    block truncatation on error exit from ->write_begin, and ->direct_IO
279moved from generic methods (block_write_begin, cont_write_begin,
280nobh_write_begin, blockdev_direct_IO*) to callers. Take a look at
281ext2_write_failed and callers for an example.
282
283[mandatory]
284
285    ->truncate is going away. The whole truncate sequence needs to be
286implemented in ->setattr, which is now mandatory for filesystems
287implementing on-disk size changes. Start with a copy of the old inode_setattr
288and vmtruncate, and the reorder the vmtruncate + foofs_vmtruncate sequence to
289be in order of zeroing blocks using block_truncate_page or similar helpers,
290size update and on finally on-disk truncation which should not fail.
291inode_change_ok now includes the size checks for ATTR_SIZE and must be called
292in the beginning of ->setattr unconditionally.
293
294[mandatory]
295
296    ->clear_inode() and ->delete_inode() are gone; ->evict_inode() should
297be used instead. It gets called whenever the inode is evicted, whether it has
298remaining links or not. Caller does *not* evict the pagecache or inode-associated
299metadata buffers; getting rid of those is responsibility of method, as it had
300been for ->delete_inode().
301
302    ->drop_inode() returns int now; it's called on final iput() with
303inode->i_lock held and it returns true if filesystems wants the inode to be
304dropped. As before, generic_drop_inode() is still the default and it's been
305updated appropriately. generic_delete_inode() is also alive and it consists
306simply of return 1. Note that all actual eviction work is done by caller after
307->drop_inode() returns.
308
309    clear_inode() is gone; use end_writeback() instead. As before, it must
310be called exactly once on each call of ->evict_inode() (as it used to be for
311each call of ->delete_inode()). Unlike before, if you are using inode-associated
312metadata buffers (i.e. mark_buffer_dirty_inode()), it's your responsibility to
313call invalidate_inode_buffers() before end_writeback().
314    No async writeback (and thus no calls of ->write_inode()) will happen
315after end_writeback() returns, so actions that should not overlap with ->write_inode()
316(e.g. freeing on-disk inode if i_nlink is 0) ought to be done after that call.
317
318    NOTE: checking i_nlink in the beginning of ->write_inode() and bailing out
319if it's zero is not *and* *never* *had* *been* enough. Final unlink() and iput()
320may happen while the inode is in the middle of ->write_inode(); e.g. if you blindly
321free the on-disk inode, you may end up doing that while ->write_inode() is writing
322to it.
323
324---
325[mandatory]
326
327    .d_delete() now only advises the dcache as to whether or not to cache
328unreferenced dentries, and is now only called when the dentry refcount goes to
3290. Even on 0 refcount transition, it must be able to tolerate being called 0,
3301, or more times (eg. constant, idempotent).
331
332---
333[mandatory]
334
335    .d_compare() calling convention and locking rules are significantly
336changed. Read updated documentation in Documentation/filesystems/vfs.txt (and
337look at examples of other filesystems) for guidance.
338
339---
340[mandatory]
341
342    .d_hash() calling convention and locking rules are significantly
343changed. Read updated documentation in Documentation/filesystems/vfs.txt (and
344look at examples of other filesystems) for guidance.
345
346---
347[mandatory]
348    dcache_lock is gone, replaced by fine grained locks. See fs/dcache.c
349for details of what locks to replace dcache_lock with in order to protect
350particular things. Most of the time, a filesystem only needs ->d_lock, which
351protects *all* the dcache state of a given dentry.
352
353--
354[mandatory]
355
356    Filesystems must RCU-free their inodes, if they can have been accessed
357via rcu-walk path walk (basically, if the file can have had a path name in the
358vfs namespace).
359
360    i_dentry and i_rcu share storage in a union, and the vfs expects
361i_dentry to be reinitialized before it is freed, so an:
362
363  INIT_LIST_HEAD(&inode->i_dentry);
364
365must be done in the RCU callback.
366
367--
368[recommended]
369    vfs now tries to do path walking in "rcu-walk mode", which avoids
370atomic operations and scalability hazards on dentries and inodes (see
371Documentation/filesystems/path-lookup.txt). d_hash and d_compare changes
372(above) are examples of the changes required to support this. For more complex
373filesystem callbacks, the vfs drops out of rcu-walk mode before the fs call, so
374no changes are required to the filesystem. However, this is costly and loses
375the benefits of rcu-walk mode. We will begin to add filesystem callbacks that
376are rcu-walk aware, shown below. Filesystems should take advantage of this
377where possible.
378
379--
380[mandatory]
381    d_revalidate is a callback that is made on every path element (if
382the filesystem provides it), which requires dropping out of rcu-walk mode. This
383may now be called in rcu-walk mode (nd->flags & LOOKUP_RCU). -ECHILD should be
384returned if the filesystem cannot handle rcu-walk. See
385Documentation/filesystems/vfs.txt for more details.
386
387    permission and check_acl are inode permission checks that are called
388on many or all directory inodes on the way down a path walk (to check for
389exec permission). These must now be rcu-walk aware (flags & IPERM_FLAG_RCU).
390See Documentation/filesystems/vfs.txt for more details.
391 
392--
393[mandatory]
394    In ->fallocate() you must check the mode option passed in. If your
395filesystem does not support hole punching (deallocating space in the middle of a
396file) you must return -EOPNOTSUPP if FALLOC_FL_PUNCH_HOLE is set in mode.
397Currently you can only have FALLOC_FL_PUNCH_HOLE with FALLOC_FL_KEEP_SIZE set,
398so the i_size should not change when hole punching, even when puching the end of
399a file off.
400
401--
402[mandatory]
403    ->get_sb() is gone. Switch to use of ->mount(). Typically it's just
404a matter of switching from calling get_sb_... to mount_... and changing the
405function type. If you were doing it manually, just switch from setting ->mnt_root
406to some pointer to returning that pointer. On errors return ERR_PTR(...).
407
408--
409[mandatory]
410    ->permission() and generic_permission()have lost flags
411argument; instead of passing IPERM_FLAG_RCU we add MAY_NOT_BLOCK into mask.
412    generic_permission() has also lost the check_acl argument; ACL checking
413has been taken to VFS and filesystems need to provide a non-NULL ->i_op->get_acl
414to read an ACL from disk.
415
416--
417[mandatory]
418    If you implement your own ->llseek() you must handle SEEK_HOLE and
419SEEK_DATA. You can hanle this by returning -EINVAL, but it would be nicer to
420support it in some way. The generic handler assumes that the entire file is
421data and there is a virtual hole at the end of the file. So if the provided
422offset is less than i_size and SEEK_DATA is specified, return the same offset.
423If the above is true for the offset and you are given SEEK_HOLE, return the end
424of the file. If the offset is i_size or greater return -ENXIO in either case.
425
426[mandatory]
427    If you have your own ->fsync() you must make sure to call
428filemap_write_and_wait_range() so that all dirty pages are synced out properly.
429You must also keep in mind that ->fsync() is not called with i_mutex held
430anymore, so if you require i_mutex locking you must make sure to take it and
431release it yourself.
432

Archive Download this file



interactive