1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
10 #include "mds_client.h"
13 * Directory operations: readdir, lookup, create, link, unlink,
18 * Ceph MDS operations are specified in terms of a base ino and
19 * relative path. Thus, the client can specify an operation on a
20 * specific inode (e.g., a getattr due to fstat(2)), or as a path
21 * relative to, say, the root directory.
23 * Normally, we limit ourselves to strict inode ops (no path component)
24 * or dentry operations (a single path component relative to an ino). The
25 * exception to this is open_root_dentry(), which will open the mount
29 const struct inode_operations ceph_dir_iops
;
30 const struct file_operations ceph_dir_fops
;
31 const struct dentry_operations ceph_dentry_ops
;
34 * Initialize ceph dentry state.
36 int ceph_init_dentry(struct dentry
*dentry
)
38 struct ceph_dentry_info
*di
;
43 di
= kmem_cache_alloc(ceph_dentry_cachep
, GFP_NOFS
| __GFP_ZERO
);
45 return -ENOMEM
; /* oh well */
47 spin_lock(&dentry
->d_lock
);
48 if (dentry
->d_fsdata
) {
50 kmem_cache_free(ceph_dentry_cachep
, di
);
54 if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_NOSNAP
)
55 d_set_d_op(dentry
, &ceph_dentry_ops
);
56 else if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_SNAPDIR
)
57 d_set_d_op(dentry
, &ceph_snapdir_dentry_ops
);
59 d_set_d_op(dentry
, &ceph_snap_dentry_ops
);
62 di
->lease_session
= NULL
;
63 dentry
->d_time
= jiffies
;
64 /* avoid reordering d_fsdata setup so that the check above is safe */
66 dentry
->d_fsdata
= di
;
67 ceph_dentry_lru_add(dentry
);
69 spin_unlock(&dentry
->d_lock
);
73 struct inode
*ceph_get_dentry_parent_inode(struct dentry
*dentry
)
75 struct inode
*inode
= NULL
;
80 spin_lock(&dentry
->d_lock
);
81 if (!IS_ROOT(dentry
)) {
82 inode
= dentry
->d_parent
->d_inode
;
85 spin_unlock(&dentry
->d_lock
);
91 * for readdir, we encode the directory frag and offset within that
94 static unsigned fpos_frag(loff_t p
)
98 static unsigned fpos_off(loff_t p
)
100 return p
& 0xffffffff;
103 static int fpos_cmp(loff_t l
, loff_t r
)
105 int v
= ceph_frag_compare(fpos_frag(l
), fpos_frag(r
));
108 return (int)(fpos_off(l
) - fpos_off(r
));
112 * When possible, we try to satisfy a readdir by peeking at the
113 * dcache. We make this work by carefully ordering dentries on
114 * d_u.d_child when we initially get results back from the MDS, and
115 * falling back to a "normal" sync readdir if any dentries in the dir
118 * Complete dir indicates that we have all dentries in the dir. It is
119 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
120 * the MDS if/when the directory is modified).
122 static int __dcache_readdir(struct file
*file
, struct dir_context
*ctx
)
124 struct ceph_file_info
*fi
= file
->private_data
;
125 struct dentry
*parent
= file
->f_dentry
;
126 struct inode
*dir
= parent
->d_inode
;
128 struct dentry
*dentry
, *last
;
129 struct ceph_dentry_info
*di
;
132 /* claim ref on last dentry we returned */
136 dout("__dcache_readdir %p at %llu (last %p)\n", dir
, ctx
->pos
,
139 spin_lock(&parent
->d_lock
);
141 /* start at beginning? */
142 if (ctx
->pos
== 2 || last
== NULL
||
143 ctx
->pos
< ceph_dentry(last
)->offset
) {
144 if (list_empty(&parent
->d_subdirs
))
146 p
= parent
->d_subdirs
.prev
;
147 dout(" initial p %p/%p\n", p
->prev
, p
->next
);
149 p
= last
->d_u
.d_child
.prev
;
153 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
154 di
= ceph_dentry(dentry
);
156 dout(" p %p/%p %s d_subdirs %p/%p\n", p
->prev
, p
->next
,
157 d_unhashed(dentry
) ? "!hashed" : "hashed",
158 parent
->d_subdirs
.prev
, parent
->d_subdirs
.next
);
159 if (p
== &parent
->d_subdirs
) {
160 fi
->flags
|= CEPH_F_ATEND
;
163 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
164 if (!d_unhashed(dentry
) && dentry
->d_inode
&&
165 ceph_snap(dentry
->d_inode
) != CEPH_SNAPDIR
&&
166 ceph_ino(dentry
->d_inode
) != CEPH_INO_CEPH
&&
167 fpos_cmp(ctx
->pos
, di
->offset
) <= 0)
169 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry
,
170 dentry
->d_name
.len
, dentry
->d_name
.name
, di
->offset
,
171 ctx
->pos
, d_unhashed(dentry
) ? " unhashed" : "",
172 !dentry
->d_inode
? " null" : "");
173 spin_unlock(&dentry
->d_lock
);
175 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
176 di
= ceph_dentry(dentry
);
180 spin_unlock(&dentry
->d_lock
);
181 spin_unlock(&parent
->d_lock
);
183 dout(" %llu (%llu) dentry %p %.*s %p\n", di
->offset
, ctx
->pos
,
184 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
185 ctx
->pos
= di
->offset
;
186 if (!dir_emit(ctx
, dentry
->d_name
.name
,
188 ceph_translate_ino(dentry
->d_sb
, dentry
->d_inode
->i_ino
),
189 dentry
->d_inode
->i_mode
>> 12)) {
191 /* remember our position */
193 fi
->next_offset
= di
->offset
;
205 /* make sure a dentry wasn't dropped while we didn't have parent lock */
206 if (!ceph_dir_is_complete(dir
)) {
207 dout(" lost dir complete on %p; falling back to mds\n", dir
);
212 spin_lock(&parent
->d_lock
);
213 p
= p
->prev
; /* advance to next dentry */
217 spin_unlock(&parent
->d_lock
);
225 * make note of the last dentry we read, so we can
226 * continue at the same lexicographical point,
227 * regardless of what dir changes take place on the
230 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
233 kfree(fi
->last_name
);
234 fi
->last_name
= kmalloc(len
+1, GFP_NOFS
);
237 memcpy(fi
->last_name
, name
, len
);
238 fi
->last_name
[len
] = 0;
239 dout("note_last_dentry '%s'\n", fi
->last_name
);
243 static int ceph_readdir(struct file
*file
, struct dir_context
*ctx
)
245 struct ceph_file_info
*fi
= file
->private_data
;
246 struct inode
*inode
= file_inode(file
);
247 struct ceph_inode_info
*ci
= ceph_inode(inode
);
248 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
249 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
250 unsigned frag
= fpos_frag(ctx
->pos
);
251 int off
= fpos_off(ctx
->pos
);
254 struct ceph_mds_reply_info_parsed
*rinfo
;
255 const int max_entries
= fsc
->mount_options
->max_readdir
;
256 const int max_bytes
= fsc
->mount_options
->max_readdir_bytes
;
258 dout("readdir %p file %p frag %u off %u\n", inode
, file
, frag
, off
);
259 if (fi
->flags
& CEPH_F_ATEND
)
262 /* always start with . and .. */
264 /* note dir version at start of readdir so we can tell
265 * if any dentries get dropped */
266 fi
->dir_release_count
= atomic_read(&ci
->i_release_count
);
268 dout("readdir off 0 -> '.'\n");
269 if (!dir_emit(ctx
, ".", 1,
270 ceph_translate_ino(inode
->i_sb
, inode
->i_ino
),
271 inode
->i_mode
>> 12))
277 ino_t ino
= parent_ino(file
->f_dentry
);
278 dout("readdir off 1 -> '..'\n");
279 if (!dir_emit(ctx
, "..", 2,
280 ceph_translate_ino(inode
->i_sb
, ino
),
281 inode
->i_mode
>> 12))
287 /* can we use the dcache? */
288 spin_lock(&ci
->i_ceph_lock
);
289 if ((ctx
->pos
== 2 || fi
->dentry
) &&
290 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
291 ceph_snap(inode
) != CEPH_SNAPDIR
&&
292 __ceph_dir_is_complete(ci
) &&
293 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
294 spin_unlock(&ci
->i_ceph_lock
);
295 err
= __dcache_readdir(file
, ctx
);
299 spin_unlock(&ci
->i_ceph_lock
);
302 err
= note_last_dentry(fi
, fi
->dentry
->d_name
.name
,
303 fi
->dentry
->d_name
.len
);
310 /* proceed with a normal readdir */
313 /* do we have the correct frag content buffered? */
314 if (fi
->frag
!= frag
|| fi
->last_readdir
== NULL
) {
315 struct ceph_mds_request
*req
;
316 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
317 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
319 /* discard old result, if any */
320 if (fi
->last_readdir
) {
321 ceph_mdsc_put_request(fi
->last_readdir
);
322 fi
->last_readdir
= NULL
;
325 /* requery frag tree, as the frag topology may have changed */
326 frag
= ceph_choose_frag(ceph_inode(inode
), frag
, NULL
, NULL
);
328 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
329 ceph_vinop(inode
), frag
, fi
->last_name
);
330 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
333 req
->r_inode
= inode
;
335 req
->r_dentry
= dget(file
->f_dentry
);
336 /* hints to request -> mds selection code */
337 req
->r_direct_mode
= USE_AUTH_MDS
;
338 req
->r_direct_hash
= ceph_frag_value(frag
);
339 req
->r_direct_is_hash
= true;
340 req
->r_path2
= kstrdup(fi
->last_name
, GFP_NOFS
);
341 req
->r_readdir_offset
= fi
->next_offset
;
342 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
343 req
->r_args
.readdir
.max_entries
= cpu_to_le32(max_entries
);
344 req
->r_args
.readdir
.max_bytes
= cpu_to_le32(max_bytes
);
345 req
->r_num_caps
= max_entries
+ 1;
346 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
348 ceph_mdsc_put_request(req
);
351 dout("readdir got and parsed readdir result=%d"
352 " on frag %x, end=%d, complete=%d\n", err
, frag
,
353 (int)req
->r_reply_info
.dir_end
,
354 (int)req
->r_reply_info
.dir_complete
);
356 if (!req
->r_did_prepopulate
) {
357 dout("readdir !did_prepopulate");
358 /* preclude from marking dir complete */
359 fi
->dir_release_count
--;
362 /* note next offset and last dentry name */
363 rinfo
= &req
->r_reply_info
;
364 if (le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
365 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
366 if (ceph_frag_is_leftmost(frag
))
370 off
= fi
->next_offset
;
372 fi
->offset
= fi
->next_offset
;
373 fi
->last_readdir
= req
;
376 if (req
->r_reply_info
.dir_end
) {
377 kfree(fi
->last_name
);
378 fi
->last_name
= NULL
;
379 if (ceph_frag_is_rightmost(frag
))
384 err
= note_last_dentry(fi
,
385 rinfo
->dir_dname
[rinfo
->dir_nr
-1],
386 rinfo
->dir_dname_len
[rinfo
->dir_nr
-1]);
389 fi
->next_offset
+= rinfo
->dir_nr
;
393 rinfo
= &fi
->last_readdir
->r_reply_info
;
394 dout("readdir frag %x num %d off %d chunkoff %d\n", frag
,
395 rinfo
->dir_nr
, off
, fi
->offset
);
397 ctx
->pos
= ceph_make_fpos(frag
, off
);
398 while (off
>= fi
->offset
&& off
- fi
->offset
< rinfo
->dir_nr
) {
399 struct ceph_mds_reply_inode
*in
=
400 rinfo
->dir_in
[off
- fi
->offset
].in
;
401 struct ceph_vino vino
;
404 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
405 off
, off
- fi
->offset
, rinfo
->dir_nr
, ctx
->pos
,
406 rinfo
->dir_dname_len
[off
- fi
->offset
],
407 rinfo
->dir_dname
[off
- fi
->offset
], in
);
409 ftype
= le32_to_cpu(in
->mode
) >> 12;
410 vino
.ino
= le64_to_cpu(in
->ino
);
411 vino
.snap
= le64_to_cpu(in
->snapid
);
412 ino
= ceph_vino_to_ino(vino
);
414 rinfo
->dir_dname
[off
- fi
->offset
],
415 rinfo
->dir_dname_len
[off
- fi
->offset
],
416 ceph_translate_ino(inode
->i_sb
, ino
), ftype
)) {
417 dout("filldir stopping us...\n");
425 ceph_mdsc_put_request(fi
->last_readdir
);
426 fi
->last_readdir
= NULL
;
431 if (!ceph_frag_is_rightmost(frag
)) {
432 frag
= ceph_frag_next(frag
);
434 ctx
->pos
= ceph_make_fpos(frag
, off
);
435 dout("readdir next frag is %x\n", frag
);
438 fi
->flags
|= CEPH_F_ATEND
;
441 * if dir_release_count still matches the dir, no dentries
442 * were released during the whole readdir, and we should have
443 * the complete dir contents in our cache.
445 spin_lock(&ci
->i_ceph_lock
);
446 if (atomic_read(&ci
->i_release_count
) == fi
->dir_release_count
) {
447 dout(" marking %p complete\n", inode
);
448 __ceph_dir_set_complete(ci
, fi
->dir_release_count
);
449 ci
->i_max_offset
= ctx
->pos
;
451 spin_unlock(&ci
->i_ceph_lock
);
453 dout("readdir %p file %p done.\n", inode
, file
);
457 static void reset_readdir(struct ceph_file_info
*fi
)
459 if (fi
->last_readdir
) {
460 ceph_mdsc_put_request(fi
->last_readdir
);
461 fi
->last_readdir
= NULL
;
463 kfree(fi
->last_name
);
464 fi
->last_name
= NULL
;
465 fi
->next_offset
= 2; /* compensate for . and .. */
470 fi
->flags
&= ~CEPH_F_ATEND
;
473 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
475 struct ceph_file_info
*fi
= file
->private_data
;
476 struct inode
*inode
= file
->f_mapping
->host
;
477 loff_t old_offset
= offset
;
480 mutex_lock(&inode
->i_mutex
);
484 offset
+= inode
->i_size
+ 2; /* FIXME */
487 offset
+= file
->f_pos
;
494 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
495 if (offset
!= file
->f_pos
) {
496 file
->f_pos
= offset
;
498 fi
->flags
&= ~CEPH_F_ATEND
;
503 * discard buffered readdir content on seekdir(0), or
504 * seek to new frag, or seek prior to current chunk.
507 fpos_frag(offset
) != fpos_frag(old_offset
) ||
508 fpos_off(offset
) < fi
->offset
) {
509 dout("dir_llseek dropping %p content\n", file
);
513 /* bump dir_release_count if we did a forward seek */
514 if (offset
> old_offset
)
515 fi
->dir_release_count
--;
518 mutex_unlock(&inode
->i_mutex
);
523 * Handle lookups for the hidden .snap directory.
525 int ceph_handle_snapdir(struct ceph_mds_request
*req
,
526 struct dentry
*dentry
, int err
)
528 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
529 struct inode
*parent
= dentry
->d_parent
->d_inode
; /* we hold i_mutex */
532 if (err
== -ENOENT
&&
533 ceph_snap(parent
) == CEPH_NOSNAP
&&
534 strcmp(dentry
->d_name
.name
,
535 fsc
->mount_options
->snapdir_name
) == 0) {
536 struct inode
*inode
= ceph_get_snapdir(parent
);
537 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
538 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, inode
);
539 BUG_ON(!d_unhashed(dentry
));
540 d_add(dentry
, inode
);
547 * Figure out final result of a lookup/open request.
549 * Mainly, make sure we return the final req->r_dentry (if it already
550 * existed) in place of the original VFS-provided dentry when they
553 * Gracefully handle the case where the MDS replies with -ENOENT and
554 * no trace (which it may do, at its discretion, e.g., if it doesn't
555 * care to issue a lease on the negative dentry).
557 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
558 struct dentry
*dentry
, int err
)
560 if (err
== -ENOENT
) {
563 if (!req
->r_reply_info
.head
->is_dentry
) {
564 dout("ENOENT and no trace, dentry %p inode %p\n",
565 dentry
, dentry
->d_inode
);
566 if (dentry
->d_inode
) {
575 dentry
= ERR_PTR(err
);
576 else if (dentry
!= req
->r_dentry
)
577 dentry
= dget(req
->r_dentry
); /* we got spliced */
583 static int is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
585 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
586 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
590 * Look up a single dir entry. If there is a lookup intent, inform
591 * the MDS so that it gets our 'caps wanted' value in a single op.
593 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
596 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
597 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
598 struct ceph_mds_request
*req
;
602 dout("lookup %p dentry %p '%.*s'\n",
603 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
605 if (dentry
->d_name
.len
> NAME_MAX
)
606 return ERR_PTR(-ENAMETOOLONG
);
608 err
= ceph_init_dentry(dentry
);
612 /* can we conclude ENOENT locally? */
613 if (dentry
->d_inode
== NULL
) {
614 struct ceph_inode_info
*ci
= ceph_inode(dir
);
615 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
617 spin_lock(&ci
->i_ceph_lock
);
618 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
619 if (strncmp(dentry
->d_name
.name
,
620 fsc
->mount_options
->snapdir_name
,
621 dentry
->d_name
.len
) &&
622 !is_root_ceph_dentry(dir
, dentry
) &&
623 __ceph_dir_is_complete(ci
) &&
624 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
625 spin_unlock(&ci
->i_ceph_lock
);
626 dout(" dir %p complete, -ENOENT\n", dir
);
628 di
->lease_shared_gen
= ci
->i_shared_gen
;
631 spin_unlock(&ci
->i_ceph_lock
);
634 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
635 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
636 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
638 return ERR_CAST(req
);
639 req
->r_dentry
= dget(dentry
);
641 /* we only need inode linkage */
642 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
643 req
->r_locked_dir
= dir
;
644 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
645 err
= ceph_handle_snapdir(req
, dentry
, err
);
646 dentry
= ceph_finish_lookup(req
, dentry
, err
);
647 ceph_mdsc_put_request(req
); /* will dput(dentry) */
648 dout("lookup result=%p\n", dentry
);
653 * If we do a create but get no trace back from the MDS, follow up with
654 * a lookup (the VFS expects us to link up the provided dentry).
656 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
658 struct dentry
*result
= ceph_lookup(dir
, dentry
, 0);
660 if (result
&& !IS_ERR(result
)) {
662 * We created the item, then did a lookup, and found
663 * it was already linked to another inode we already
664 * had in our cache (and thus got spliced). Link our
665 * dentry to that inode, but don't hash it, just in
666 * case the VFS wants to dereference it.
668 BUG_ON(!result
->d_inode
);
669 d_instantiate(dentry
, result
->d_inode
);
672 return PTR_ERR(result
);
675 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
676 umode_t mode
, dev_t rdev
)
678 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
679 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
680 struct ceph_mds_request
*req
;
683 if (ceph_snap(dir
) != CEPH_NOSNAP
)
686 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
687 dir
, dentry
, mode
, rdev
);
688 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
693 req
->r_dentry
= dget(dentry
);
695 req
->r_locked_dir
= dir
;
696 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
697 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
698 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
699 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
700 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
701 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
702 err
= ceph_handle_notrace_create(dir
, dentry
);
703 ceph_mdsc_put_request(req
);
706 ceph_init_acl(dentry
, dentry
->d_inode
, dir
);
712 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
715 return ceph_mknod(dir
, dentry
, mode
, 0);
718 static int ceph_symlink(struct inode
*dir
, struct dentry
*dentry
,
721 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
722 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
723 struct ceph_mds_request
*req
;
726 if (ceph_snap(dir
) != CEPH_NOSNAP
)
729 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
730 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
735 req
->r_dentry
= dget(dentry
);
737 req
->r_path2
= kstrdup(dest
, GFP_NOFS
);
738 req
->r_locked_dir
= dir
;
739 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
740 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
741 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
742 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
743 err
= ceph_handle_notrace_create(dir
, dentry
);
744 ceph_mdsc_put_request(req
);
746 ceph_init_acl(dentry
, dentry
->d_inode
, dir
);
752 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
754 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
755 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
756 struct ceph_mds_request
*req
;
760 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
761 /* mkdir .snap/foo is a MKSNAP */
762 op
= CEPH_MDS_OP_MKSNAP
;
763 dout("mksnap dir %p snap '%.*s' dn %p\n", dir
,
764 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
);
765 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
766 dout("mkdir dir %p dn %p mode 0%ho\n", dir
, dentry
, mode
);
767 op
= CEPH_MDS_OP_MKDIR
;
771 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
777 req
->r_dentry
= dget(dentry
);
779 req
->r_locked_dir
= dir
;
780 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
781 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
782 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
783 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
784 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
785 err
= ceph_handle_notrace_create(dir
, dentry
);
786 ceph_mdsc_put_request(req
);
789 ceph_init_acl(dentry
, dentry
->d_inode
, dir
);
795 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
796 struct dentry
*dentry
)
798 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
799 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
800 struct ceph_mds_request
*req
;
803 if (ceph_snap(dir
) != CEPH_NOSNAP
)
806 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
808 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
813 req
->r_dentry
= dget(dentry
);
815 req
->r_old_dentry
= dget(old_dentry
); /* or inode? hrm. */
816 req
->r_old_dentry_dir
= ceph_get_dentry_parent_inode(old_dentry
);
817 req
->r_locked_dir
= dir
;
818 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
819 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
820 /* release LINK_SHARED on source inode (mds will lock it) */
821 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
822 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
825 } else if (!req
->r_reply_info
.head
->is_dentry
) {
826 ihold(old_dentry
->d_inode
);
827 d_instantiate(dentry
, old_dentry
->d_inode
);
829 ceph_mdsc_put_request(req
);
834 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
835 * looks like the link count will hit 0, drop any other caps (other
836 * than PIN) we don't specifically want (due to the file still being
839 static int drop_caps_for_unlink(struct inode
*inode
)
841 struct ceph_inode_info
*ci
= ceph_inode(inode
);
842 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
844 spin_lock(&ci
->i_ceph_lock
);
845 if (inode
->i_nlink
== 1) {
846 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
847 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
849 spin_unlock(&ci
->i_ceph_lock
);
854 * rmdir and unlink are differ only by the metadata op code
856 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
858 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
859 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
860 struct inode
*inode
= dentry
->d_inode
;
861 struct ceph_mds_request
*req
;
865 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
866 /* rmdir .snap/foo is RMSNAP */
867 dout("rmsnap dir %p '%.*s' dn %p\n", dir
, dentry
->d_name
.len
,
868 dentry
->d_name
.name
, dentry
);
869 op
= CEPH_MDS_OP_RMSNAP
;
870 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
871 dout("unlink/rmdir dir %p dn %p inode %p\n",
873 op
= S_ISDIR(dentry
->d_inode
->i_mode
) ?
874 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
877 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
882 req
->r_dentry
= dget(dentry
);
884 req
->r_locked_dir
= dir
;
885 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
886 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
887 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
888 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
889 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
891 ceph_mdsc_put_request(req
);
896 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
897 struct inode
*new_dir
, struct dentry
*new_dentry
)
899 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
900 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
901 struct ceph_mds_request
*req
;
904 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
906 if (ceph_snap(old_dir
) != CEPH_NOSNAP
||
907 ceph_snap(new_dir
) != CEPH_NOSNAP
)
909 dout("rename dir %p dentry %p to dir %p dentry %p\n",
910 old_dir
, old_dentry
, new_dir
, new_dentry
);
911 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_RENAME
, USE_AUTH_MDS
);
914 req
->r_dentry
= dget(new_dentry
);
916 req
->r_old_dentry
= dget(old_dentry
);
917 req
->r_old_dentry_dir
= ceph_get_dentry_parent_inode(old_dentry
);
918 req
->r_locked_dir
= new_dir
;
919 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
920 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
921 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
922 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
923 /* release LINK_RDCACHE on source inode (mds will lock it) */
924 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
925 if (new_dentry
->d_inode
)
926 req
->r_inode_drop
= drop_caps_for_unlink(new_dentry
->d_inode
);
927 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
928 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
930 * Normally d_move() is done by fill_trace (called by
931 * do_request, above). If there is no trace, we need
935 /* d_move screws up d_subdirs order */
936 ceph_dir_clear_complete(new_dir
);
938 d_move(old_dentry
, new_dentry
);
940 /* ensure target dentry is invalidated, despite
941 rehashing bug in vfs_rename_dir */
942 ceph_invalidate_dentry_lease(new_dentry
);
944 ceph_mdsc_put_request(req
);
949 * Ensure a dentry lease will no longer revalidate.
951 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
953 spin_lock(&dentry
->d_lock
);
954 dentry
->d_time
= jiffies
;
955 ceph_dentry(dentry
)->lease_shared_gen
= 0;
956 spin_unlock(&dentry
->d_lock
);
960 * Check if dentry lease is valid. If not, delete the lease. Try to
961 * renew if the least is more than half up.
963 static int dentry_lease_is_valid(struct dentry
*dentry
)
965 struct ceph_dentry_info
*di
;
966 struct ceph_mds_session
*s
;
970 struct ceph_mds_session
*session
= NULL
;
971 struct inode
*dir
= NULL
;
974 spin_lock(&dentry
->d_lock
);
975 di
= ceph_dentry(dentry
);
976 if (di
->lease_session
) {
977 s
= di
->lease_session
;
978 spin_lock(&s
->s_gen_ttl_lock
);
981 spin_unlock(&s
->s_gen_ttl_lock
);
983 if (di
->lease_gen
== gen
&&
984 time_before(jiffies
, dentry
->d_time
) &&
985 time_before(jiffies
, ttl
)) {
987 if (di
->lease_renew_after
&&
988 time_after(jiffies
, di
->lease_renew_after
)) {
989 /* we should renew */
990 dir
= dentry
->d_parent
->d_inode
;
991 session
= ceph_get_mds_session(s
);
993 di
->lease_renew_after
= 0;
994 di
->lease_renew_from
= jiffies
;
998 spin_unlock(&dentry
->d_lock
);
1001 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
1002 CEPH_MDS_LEASE_RENEW
, seq
);
1003 ceph_put_mds_session(session
);
1005 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
1010 * Check if directory-wide content lease/cap is valid.
1012 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
1014 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1015 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1018 spin_lock(&ci
->i_ceph_lock
);
1019 if (ci
->i_shared_gen
== di
->lease_shared_gen
)
1020 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
1021 spin_unlock(&ci
->i_ceph_lock
);
1022 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1023 dir
, (unsigned)ci
->i_shared_gen
, dentry
,
1024 (unsigned)di
->lease_shared_gen
, valid
);
1029 * Check if cached dentry can be trusted.
1031 static int ceph_d_revalidate(struct dentry
*dentry
, unsigned int flags
)
1036 if (flags
& LOOKUP_RCU
)
1039 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry
,
1040 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
,
1041 ceph_dentry(dentry
)->offset
);
1043 dir
= ceph_get_dentry_parent_inode(dentry
);
1045 /* always trust cached snapped dentries, snapdir dentry */
1046 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
1047 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry
,
1048 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
1050 } else if (dentry
->d_inode
&&
1051 ceph_snap(dentry
->d_inode
) == CEPH_SNAPDIR
) {
1053 } else if (dentry_lease_is_valid(dentry
) ||
1054 dir_lease_is_valid(dir
, dentry
)) {
1055 if (dentry
->d_inode
)
1056 valid
= ceph_is_any_caps(dentry
->d_inode
);
1061 dout("d_revalidate %p %s\n", dentry
, valid
? "valid" : "invalid");
1063 ceph_dentry_lru_touch(dentry
);
1065 ceph_dir_clear_complete(dir
);
1073 * Release our ceph_dentry_info.
1075 static void ceph_d_release(struct dentry
*dentry
)
1077 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1079 dout("d_release %p\n", dentry
);
1080 ceph_dentry_lru_del(dentry
);
1081 if (di
->lease_session
)
1082 ceph_put_mds_session(di
->lease_session
);
1083 kmem_cache_free(ceph_dentry_cachep
, di
);
1084 dentry
->d_fsdata
= NULL
;
1087 static int ceph_snapdir_d_revalidate(struct dentry
*dentry
,
1091 * Eventually, we'll want to revalidate snapped metadata
1092 * too... probably...
1098 * When the VFS prunes a dentry from the cache, we need to clear the
1099 * complete flag on the parent directory.
1101 * Called under dentry->d_lock.
1103 static void ceph_d_prune(struct dentry
*dentry
)
1105 dout("ceph_d_prune %p\n", dentry
);
1107 /* do we have a valid parent? */
1108 if (IS_ROOT(dentry
))
1111 /* if we are not hashed, we don't affect dir's completeness */
1112 if (d_unhashed(dentry
))
1116 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1117 * cleared until d_release
1119 ceph_dir_clear_complete(dentry
->d_parent
->d_inode
);
1123 * read() on a dir. This weird interface hack only works if mounted
1124 * with '-o dirstat'.
1126 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1129 struct ceph_file_info
*cf
= file
->private_data
;
1130 struct inode
*inode
= file_inode(file
);
1131 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1133 const int bufsize
= 1024;
1135 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1138 if (!cf
->dir_info
) {
1139 cf
->dir_info
= kmalloc(bufsize
, GFP_NOFS
);
1143 snprintf(cf
->dir_info
, bufsize
,
1146 " subdirs: %20lld\n"
1147 "rentries: %20lld\n"
1149 " rsubdirs: %20lld\n"
1151 "rctime: %10ld.%09ld\n",
1152 ci
->i_files
+ ci
->i_subdirs
,
1155 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1159 (long)ci
->i_rctime
.tv_sec
,
1160 (long)ci
->i_rctime
.tv_nsec
);
1163 if (*ppos
>= cf
->dir_info_len
)
1165 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1166 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1169 *ppos
+= (size
- left
);
1174 * an fsync() on a dir will wait for any uncommitted directory
1175 * operations to commit.
1177 static int ceph_dir_fsync(struct file
*file
, loff_t start
, loff_t end
,
1180 struct inode
*inode
= file_inode(file
);
1181 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1182 struct list_head
*head
= &ci
->i_unsafe_dirops
;
1183 struct ceph_mds_request
*req
;
1187 dout("dir_fsync %p\n", inode
);
1188 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1191 mutex_lock(&inode
->i_mutex
);
1193 spin_lock(&ci
->i_unsafe_lock
);
1194 if (list_empty(head
))
1197 req
= list_entry(head
->prev
,
1198 struct ceph_mds_request
, r_unsafe_dir_item
);
1199 last_tid
= req
->r_tid
;
1202 ceph_mdsc_get_request(req
);
1203 spin_unlock(&ci
->i_unsafe_lock
);
1205 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1206 inode
, req
->r_tid
, last_tid
);
1207 if (req
->r_timeout
) {
1208 ret
= wait_for_completion_timeout(
1209 &req
->r_safe_completion
, req
->r_timeout
);
1213 ret
= -EIO
; /* timed out */
1215 wait_for_completion(&req
->r_safe_completion
);
1217 ceph_mdsc_put_request(req
);
1219 spin_lock(&ci
->i_unsafe_lock
);
1220 if (ret
|| list_empty(head
))
1222 req
= list_entry(head
->next
,
1223 struct ceph_mds_request
, r_unsafe_dir_item
);
1224 } while (req
->r_tid
< last_tid
);
1226 spin_unlock(&ci
->i_unsafe_lock
);
1227 mutex_unlock(&inode
->i_mutex
);
1233 * We maintain a private dentry LRU.
1235 * FIXME: this needs to be changed to a per-mds lru to be useful.
1237 void ceph_dentry_lru_add(struct dentry
*dn
)
1239 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1240 struct ceph_mds_client
*mdsc
;
1242 dout("dentry_lru_add %p %p '%.*s'\n", di
, dn
,
1243 dn
->d_name
.len
, dn
->d_name
.name
);
1244 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1245 spin_lock(&mdsc
->dentry_lru_lock
);
1246 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1248 spin_unlock(&mdsc
->dentry_lru_lock
);
1251 void ceph_dentry_lru_touch(struct dentry
*dn
)
1253 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1254 struct ceph_mds_client
*mdsc
;
1256 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di
, dn
,
1257 dn
->d_name
.len
, dn
->d_name
.name
, di
->offset
);
1258 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1259 spin_lock(&mdsc
->dentry_lru_lock
);
1260 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1261 spin_unlock(&mdsc
->dentry_lru_lock
);
1264 void ceph_dentry_lru_del(struct dentry
*dn
)
1266 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1267 struct ceph_mds_client
*mdsc
;
1269 dout("dentry_lru_del %p %p '%.*s'\n", di
, dn
,
1270 dn
->d_name
.len
, dn
->d_name
.name
);
1271 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1272 spin_lock(&mdsc
->dentry_lru_lock
);
1273 list_del_init(&di
->lru
);
1275 spin_unlock(&mdsc
->dentry_lru_lock
);
1279 * Return name hash for a given dentry. This is dependent on
1280 * the parent directory's hash function.
1282 unsigned ceph_dentry_hash(struct inode
*dir
, struct dentry
*dn
)
1284 struct ceph_inode_info
*dci
= ceph_inode(dir
);
1286 switch (dci
->i_dir_layout
.dl_dir_hash
) {
1287 case 0: /* for backward compat */
1288 case CEPH_STR_HASH_LINUX
:
1289 return dn
->d_name
.hash
;
1292 return ceph_str_hash(dci
->i_dir_layout
.dl_dir_hash
,
1293 dn
->d_name
.name
, dn
->d_name
.len
);
1297 const struct file_operations ceph_dir_fops
= {
1298 .read
= ceph_read_dir
,
1299 .iterate
= ceph_readdir
,
1300 .llseek
= ceph_dir_llseek
,
1302 .release
= ceph_release
,
1303 .unlocked_ioctl
= ceph_ioctl
,
1304 .fsync
= ceph_dir_fsync
,
1307 const struct inode_operations ceph_dir_iops
= {
1308 .lookup
= ceph_lookup
,
1309 .permission
= ceph_permission
,
1310 .getattr
= ceph_getattr
,
1311 .setattr
= ceph_setattr
,
1312 .setxattr
= ceph_setxattr
,
1313 .getxattr
= ceph_getxattr
,
1314 .listxattr
= ceph_listxattr
,
1315 .removexattr
= ceph_removexattr
,
1316 .get_acl
= ceph_get_acl
,
1317 .set_acl
= ceph_set_acl
,
1318 .mknod
= ceph_mknod
,
1319 .symlink
= ceph_symlink
,
1320 .mkdir
= ceph_mkdir
,
1322 .unlink
= ceph_unlink
,
1323 .rmdir
= ceph_unlink
,
1324 .rename
= ceph_rename
,
1325 .create
= ceph_create
,
1326 .atomic_open
= ceph_atomic_open
,
1329 const struct dentry_operations ceph_dentry_ops
= {
1330 .d_revalidate
= ceph_d_revalidate
,
1331 .d_release
= ceph_d_release
,
1332 .d_prune
= ceph_d_prune
,
1335 const struct dentry_operations ceph_snapdir_dentry_ops
= {
1336 .d_revalidate
= ceph_snapdir_d_revalidate
,
1337 .d_release
= ceph_d_release
,
1340 const struct dentry_operations ceph_snap_dentry_ops
= {
1341 .d_release
= ceph_d_release
,
1342 .d_prune
= ceph_d_prune
,