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 (dentry
->d_parent
== NULL
|| /* nfs fh_to_dentry */
55 ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_NOSNAP
)
56 d_set_d_op(dentry
, &ceph_dentry_ops
);
57 else if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_SNAPDIR
)
58 d_set_d_op(dentry
, &ceph_snapdir_dentry_ops
);
60 d_set_d_op(dentry
, &ceph_snap_dentry_ops
);
63 di
->lease_session
= NULL
;
64 dentry
->d_time
= jiffies
;
65 /* avoid reordering d_fsdata setup so that the check above is safe */
67 dentry
->d_fsdata
= di
;
68 ceph_dentry_lru_add(dentry
);
70 spin_unlock(&dentry
->d_lock
);
74 struct inode
*ceph_get_dentry_parent_inode(struct dentry
*dentry
)
76 struct inode
*inode
= NULL
;
81 spin_lock(&dentry
->d_lock
);
82 if (dentry
->d_parent
) {
83 inode
= dentry
->d_parent
->d_inode
;
86 spin_unlock(&dentry
->d_lock
);
92 * for readdir, we encode the directory frag and offset within that
95 static unsigned fpos_frag(loff_t p
)
99 static unsigned fpos_off(loff_t p
)
101 return p
& 0xffffffff;
105 * When possible, we try to satisfy a readdir by peeking at the
106 * dcache. We make this work by carefully ordering dentries on
107 * d_u.d_child when we initially get results back from the MDS, and
108 * falling back to a "normal" sync readdir if any dentries in the dir
111 * D_COMPLETE tells indicates we have all dentries in the dir. It is
112 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
113 * the MDS if/when the directory is modified).
115 static int __dcache_readdir(struct file
*filp
,
116 void *dirent
, filldir_t filldir
)
118 struct ceph_file_info
*fi
= filp
->private_data
;
119 struct dentry
*parent
= filp
->f_dentry
;
120 struct inode
*dir
= parent
->d_inode
;
122 struct dentry
*dentry
, *last
;
123 struct ceph_dentry_info
*di
;
126 /* claim ref on last dentry we returned */
130 dout("__dcache_readdir %p at %llu (last %p)\n", dir
, filp
->f_pos
,
133 spin_lock(&parent
->d_lock
);
135 /* start at beginning? */
136 if (filp
->f_pos
== 2 || last
== NULL
||
137 filp
->f_pos
< ceph_dentry(last
)->offset
) {
138 if (list_empty(&parent
->d_subdirs
))
140 p
= parent
->d_subdirs
.prev
;
141 dout(" initial p %p/%p\n", p
->prev
, p
->next
);
143 p
= last
->d_u
.d_child
.prev
;
147 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
148 di
= ceph_dentry(dentry
);
150 dout(" p %p/%p %s d_subdirs %p/%p\n", p
->prev
, p
->next
,
151 d_unhashed(dentry
) ? "!hashed" : "hashed",
152 parent
->d_subdirs
.prev
, parent
->d_subdirs
.next
);
153 if (p
== &parent
->d_subdirs
) {
154 fi
->flags
|= CEPH_F_ATEND
;
157 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
158 if (!d_unhashed(dentry
) && dentry
->d_inode
&&
159 ceph_snap(dentry
->d_inode
) != CEPH_SNAPDIR
&&
160 ceph_ino(dentry
->d_inode
) != CEPH_INO_CEPH
&&
161 filp
->f_pos
<= di
->offset
)
163 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry
,
164 dentry
->d_name
.len
, dentry
->d_name
.name
, di
->offset
,
165 filp
->f_pos
, d_unhashed(dentry
) ? " unhashed" : "",
166 !dentry
->d_inode
? " null" : "");
167 spin_unlock(&dentry
->d_lock
);
169 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
170 di
= ceph_dentry(dentry
);
174 spin_unlock(&dentry
->d_lock
);
175 spin_unlock(&parent
->d_lock
);
177 dout(" %llu (%llu) dentry %p %.*s %p\n", di
->offset
, filp
->f_pos
,
178 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
179 filp
->f_pos
= di
->offset
;
180 err
= filldir(dirent
, dentry
->d_name
.name
,
181 dentry
->d_name
.len
, di
->offset
,
182 ceph_translate_ino(dentry
->d_sb
, dentry
->d_inode
->i_ino
),
183 dentry
->d_inode
->i_mode
>> 12);
187 /* remember our position */
189 fi
->next_offset
= di
->offset
;
201 /* make sure a dentry wasn't dropped while we didn't have parent lock */
202 if (!ceph_dir_test_complete(dir
)) {
203 dout(" lost D_COMPLETE on %p; falling back to mds\n", dir
);
208 spin_lock(&parent
->d_lock
);
209 p
= p
->prev
; /* advance to next dentry */
213 spin_unlock(&parent
->d_lock
);
221 * make note of the last dentry we read, so we can
222 * continue at the same lexicographical point,
223 * regardless of what dir changes take place on the
226 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
229 kfree(fi
->last_name
);
230 fi
->last_name
= kmalloc(len
+1, GFP_NOFS
);
233 memcpy(fi
->last_name
, name
, len
);
234 fi
->last_name
[len
] = 0;
235 dout("note_last_dentry '%s'\n", fi
->last_name
);
239 static int ceph_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
241 struct ceph_file_info
*fi
= filp
->private_data
;
242 struct inode
*inode
= filp
->f_dentry
->d_inode
;
243 struct ceph_inode_info
*ci
= ceph_inode(inode
);
244 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
245 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
246 unsigned frag
= fpos_frag(filp
->f_pos
);
247 int off
= fpos_off(filp
->f_pos
);
250 struct ceph_mds_reply_info_parsed
*rinfo
;
251 const int max_entries
= fsc
->mount_options
->max_readdir
;
252 const int max_bytes
= fsc
->mount_options
->max_readdir_bytes
;
254 dout("readdir %p filp %p frag %u off %u\n", inode
, filp
, frag
, off
);
255 if (fi
->flags
& CEPH_F_ATEND
)
258 /* always start with . and .. */
259 if (filp
->f_pos
== 0) {
260 /* note dir version at start of readdir so we can tell
261 * if any dentries get dropped */
262 fi
->dir_release_count
= ci
->i_release_count
;
264 dout("readdir off 0 -> '.'\n");
265 if (filldir(dirent
, ".", 1, ceph_make_fpos(0, 0),
266 ceph_translate_ino(inode
->i_sb
, inode
->i_ino
),
267 inode
->i_mode
>> 12) < 0)
272 if (filp
->f_pos
== 1) {
273 ino_t ino
= parent_ino(filp
->f_dentry
);
274 dout("readdir off 1 -> '..'\n");
275 if (filldir(dirent
, "..", 2, ceph_make_fpos(0, 1),
276 ceph_translate_ino(inode
->i_sb
, ino
),
277 inode
->i_mode
>> 12) < 0)
283 /* can we use the dcache? */
284 spin_lock(&ci
->i_ceph_lock
);
285 if ((filp
->f_pos
== 2 || fi
->dentry
) &&
286 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
287 ceph_snap(inode
) != CEPH_SNAPDIR
&&
288 ceph_dir_test_complete(inode
) &&
289 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
290 spin_unlock(&ci
->i_ceph_lock
);
291 err
= __dcache_readdir(filp
, dirent
, filldir
);
295 spin_unlock(&ci
->i_ceph_lock
);
298 err
= note_last_dentry(fi
, fi
->dentry
->d_name
.name
,
299 fi
->dentry
->d_name
.len
);
306 /* proceed with a normal readdir */
309 /* do we have the correct frag content buffered? */
310 if (fi
->frag
!= frag
|| fi
->last_readdir
== NULL
) {
311 struct ceph_mds_request
*req
;
312 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
313 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
315 /* discard old result, if any */
316 if (fi
->last_readdir
) {
317 ceph_mdsc_put_request(fi
->last_readdir
);
318 fi
->last_readdir
= NULL
;
321 /* requery frag tree, as the frag topology may have changed */
322 frag
= ceph_choose_frag(ceph_inode(inode
), frag
, NULL
, NULL
);
324 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
325 ceph_vinop(inode
), frag
, fi
->last_name
);
326 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
329 req
->r_inode
= inode
;
331 req
->r_dentry
= dget(filp
->f_dentry
);
332 /* hints to request -> mds selection code */
333 req
->r_direct_mode
= USE_AUTH_MDS
;
334 req
->r_direct_hash
= ceph_frag_value(frag
);
335 req
->r_direct_is_hash
= true;
336 req
->r_path2
= kstrdup(fi
->last_name
, GFP_NOFS
);
337 req
->r_readdir_offset
= fi
->next_offset
;
338 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
339 req
->r_args
.readdir
.max_entries
= cpu_to_le32(max_entries
);
340 req
->r_args
.readdir
.max_bytes
= cpu_to_le32(max_bytes
);
341 req
->r_num_caps
= max_entries
+ 1;
342 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
344 ceph_mdsc_put_request(req
);
347 dout("readdir got and parsed readdir result=%d"
348 " on frag %x, end=%d, complete=%d\n", err
, frag
,
349 (int)req
->r_reply_info
.dir_end
,
350 (int)req
->r_reply_info
.dir_complete
);
352 if (!req
->r_did_prepopulate
) {
353 dout("readdir !did_prepopulate");
354 fi
->dir_release_count
--; /* preclude D_COMPLETE */
357 /* note next offset and last dentry name */
358 fi
->offset
= fi
->next_offset
;
359 fi
->last_readdir
= req
;
361 if (req
->r_reply_info
.dir_end
) {
362 kfree(fi
->last_name
);
363 fi
->last_name
= NULL
;
364 if (ceph_frag_is_rightmost(frag
))
369 rinfo
= &req
->r_reply_info
;
370 err
= note_last_dentry(fi
,
371 rinfo
->dir_dname
[rinfo
->dir_nr
-1],
372 rinfo
->dir_dname_len
[rinfo
->dir_nr
-1]);
375 fi
->next_offset
+= rinfo
->dir_nr
;
379 rinfo
= &fi
->last_readdir
->r_reply_info
;
380 dout("readdir frag %x num %d off %d chunkoff %d\n", frag
,
381 rinfo
->dir_nr
, off
, fi
->offset
);
382 while (off
>= fi
->offset
&& off
- fi
->offset
< rinfo
->dir_nr
) {
383 u64 pos
= ceph_make_fpos(frag
, off
);
384 struct ceph_mds_reply_inode
*in
=
385 rinfo
->dir_in
[off
- fi
->offset
].in
;
386 struct ceph_vino vino
;
389 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
390 off
, off
- fi
->offset
, rinfo
->dir_nr
, pos
,
391 rinfo
->dir_dname_len
[off
- fi
->offset
],
392 rinfo
->dir_dname
[off
- fi
->offset
], in
);
394 ftype
= le32_to_cpu(in
->mode
) >> 12;
395 vino
.ino
= le64_to_cpu(in
->ino
);
396 vino
.snap
= le64_to_cpu(in
->snapid
);
397 ino
= ceph_vino_to_ino(vino
);
399 rinfo
->dir_dname
[off
- fi
->offset
],
400 rinfo
->dir_dname_len
[off
- fi
->offset
],
402 ceph_translate_ino(inode
->i_sb
, ino
), ftype
) < 0) {
403 dout("filldir stopping us...\n");
407 filp
->f_pos
= pos
+ 1;
411 ceph_mdsc_put_request(fi
->last_readdir
);
412 fi
->last_readdir
= NULL
;
417 if (!ceph_frag_is_rightmost(frag
)) {
418 frag
= ceph_frag_next(frag
);
420 filp
->f_pos
= ceph_make_fpos(frag
, off
);
421 dout("readdir next frag is %x\n", frag
);
424 fi
->flags
|= CEPH_F_ATEND
;
427 * if dir_release_count still matches the dir, no dentries
428 * were released during the whole readdir, and we should have
429 * the complete dir contents in our cache.
431 spin_lock(&ci
->i_ceph_lock
);
432 if (ci
->i_release_count
== fi
->dir_release_count
) {
433 ceph_dir_set_complete(inode
);
434 ci
->i_max_offset
= filp
->f_pos
;
436 spin_unlock(&ci
->i_ceph_lock
);
438 dout("readdir %p filp %p done.\n", inode
, filp
);
442 static void reset_readdir(struct ceph_file_info
*fi
)
444 if (fi
->last_readdir
) {
445 ceph_mdsc_put_request(fi
->last_readdir
);
446 fi
->last_readdir
= NULL
;
448 kfree(fi
->last_name
);
449 fi
->last_name
= NULL
;
450 fi
->next_offset
= 2; /* compensate for . and .. */
455 fi
->flags
&= ~CEPH_F_ATEND
;
458 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int origin
)
460 struct ceph_file_info
*fi
= file
->private_data
;
461 struct inode
*inode
= file
->f_mapping
->host
;
462 loff_t old_offset
= offset
;
465 mutex_lock(&inode
->i_mutex
);
469 offset
+= inode
->i_size
+ 2; /* FIXME */
472 offset
+= file
->f_pos
;
479 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
480 if (offset
!= file
->f_pos
) {
481 file
->f_pos
= offset
;
483 fi
->flags
&= ~CEPH_F_ATEND
;
488 * discard buffered readdir content on seekdir(0), or
489 * seek to new frag, or seek prior to current chunk.
492 fpos_frag(offset
) != fpos_frag(old_offset
) ||
493 fpos_off(offset
) < fi
->offset
) {
494 dout("dir_llseek dropping %p content\n", file
);
498 /* bump dir_release_count if we did a forward seek */
499 if (offset
> old_offset
)
500 fi
->dir_release_count
--;
503 mutex_unlock(&inode
->i_mutex
);
508 * Handle lookups for the hidden .snap directory.
510 int ceph_handle_snapdir(struct ceph_mds_request
*req
,
511 struct dentry
*dentry
, int err
)
513 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
514 struct inode
*parent
= dentry
->d_parent
->d_inode
; /* we hold i_mutex */
517 if (err
== -ENOENT
&&
518 ceph_snap(parent
) == CEPH_NOSNAP
&&
519 strcmp(dentry
->d_name
.name
,
520 fsc
->mount_options
->snapdir_name
) == 0) {
521 struct inode
*inode
= ceph_get_snapdir(parent
);
522 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
523 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, inode
);
524 BUG_ON(!d_unhashed(dentry
));
525 d_add(dentry
, inode
);
532 * Figure out final result of a lookup/open request.
534 * Mainly, make sure we return the final req->r_dentry (if it already
535 * existed) in place of the original VFS-provided dentry when they
538 * Gracefully handle the case where the MDS replies with -ENOENT and
539 * no trace (which it may do, at its discretion, e.g., if it doesn't
540 * care to issue a lease on the negative dentry).
542 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
543 struct dentry
*dentry
, int err
)
545 if (err
== -ENOENT
) {
548 if (!req
->r_reply_info
.head
->is_dentry
) {
549 dout("ENOENT and no trace, dentry %p inode %p\n",
550 dentry
, dentry
->d_inode
);
551 if (dentry
->d_inode
) {
560 dentry
= ERR_PTR(err
);
561 else if (dentry
!= req
->r_dentry
)
562 dentry
= dget(req
->r_dentry
); /* we got spliced */
568 static int is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
570 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
571 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
575 * Look up a single dir entry. If there is a lookup intent, inform
576 * the MDS so that it gets our 'caps wanted' value in a single op.
578 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
579 struct nameidata
*nd
)
581 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
582 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
583 struct ceph_mds_request
*req
;
587 dout("lookup %p dentry %p '%.*s'\n",
588 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
590 if (dentry
->d_name
.len
> NAME_MAX
)
591 return ERR_PTR(-ENAMETOOLONG
);
593 err
= ceph_init_dentry(dentry
);
597 /* can we conclude ENOENT locally? */
598 if (dentry
->d_inode
== NULL
) {
599 struct ceph_inode_info
*ci
= ceph_inode(dir
);
600 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
602 spin_lock(&ci
->i_ceph_lock
);
603 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
604 if (strncmp(dentry
->d_name
.name
,
605 fsc
->mount_options
->snapdir_name
,
606 dentry
->d_name
.len
) &&
607 !is_root_ceph_dentry(dir
, dentry
) &&
608 ceph_dir_test_complete(dir
) &&
609 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
610 spin_unlock(&ci
->i_ceph_lock
);
611 dout(" dir %p complete, -ENOENT\n", dir
);
613 di
->lease_shared_gen
= ci
->i_shared_gen
;
616 spin_unlock(&ci
->i_ceph_lock
);
619 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
620 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
621 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
623 return ERR_CAST(req
);
624 req
->r_dentry
= dget(dentry
);
626 /* we only need inode linkage */
627 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
628 req
->r_locked_dir
= dir
;
629 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
630 err
= ceph_handle_snapdir(req
, dentry
, err
);
631 dentry
= ceph_finish_lookup(req
, dentry
, err
);
632 ceph_mdsc_put_request(req
); /* will dput(dentry) */
633 dout("lookup result=%p\n", dentry
);
637 int ceph_atomic_open(struct inode
*dir
, struct dentry
*dentry
,
638 struct file
*file
, unsigned flags
, umode_t mode
,
642 struct dentry
*res
= NULL
;
644 if (!(flags
& O_CREAT
)) {
645 if (dentry
->d_name
.len
> NAME_MAX
)
646 return -ENAMETOOLONG
;
648 err
= ceph_init_dentry(dentry
);
652 return ceph_lookup_open(dir
, dentry
, file
, flags
, mode
, opened
);
655 if (d_unhashed(dentry
)) {
656 res
= ceph_lookup(dir
, dentry
, NULL
);
664 /* We don't deal with positive dentries here */
666 return finish_no_open(file
, res
);
668 *opened
|= FILE_CREATED
;
669 err
= ceph_lookup_open(dir
, dentry
, file
, flags
, mode
, opened
);
676 * If we do a create but get no trace back from the MDS, follow up with
677 * a lookup (the VFS expects us to link up the provided dentry).
679 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
681 struct dentry
*result
= ceph_lookup(dir
, dentry
, NULL
);
683 if (result
&& !IS_ERR(result
)) {
685 * We created the item, then did a lookup, and found
686 * it was already linked to another inode we already
687 * had in our cache (and thus got spliced). Link our
688 * dentry to that inode, but don't hash it, just in
689 * case the VFS wants to dereference it.
691 BUG_ON(!result
->d_inode
);
692 d_instantiate(dentry
, result
->d_inode
);
695 return PTR_ERR(result
);
698 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
699 umode_t mode
, dev_t rdev
)
701 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
702 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
703 struct ceph_mds_request
*req
;
706 if (ceph_snap(dir
) != CEPH_NOSNAP
)
709 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
710 dir
, dentry
, mode
, rdev
);
711 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
716 req
->r_dentry
= dget(dentry
);
718 req
->r_locked_dir
= dir
;
719 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
720 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
721 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
722 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
723 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
724 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
725 err
= ceph_handle_notrace_create(dir
, dentry
);
726 ceph_mdsc_put_request(req
);
732 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
733 struct nameidata
*nd
)
735 return ceph_mknod(dir
, dentry
, mode
, 0);
738 static int ceph_symlink(struct inode
*dir
, struct dentry
*dentry
,
741 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
742 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
743 struct ceph_mds_request
*req
;
746 if (ceph_snap(dir
) != CEPH_NOSNAP
)
749 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
750 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
755 req
->r_dentry
= dget(dentry
);
757 req
->r_path2
= kstrdup(dest
, GFP_NOFS
);
758 req
->r_locked_dir
= dir
;
759 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
760 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
761 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
762 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
763 err
= ceph_handle_notrace_create(dir
, dentry
);
764 ceph_mdsc_put_request(req
);
770 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
772 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
773 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
774 struct ceph_mds_request
*req
;
778 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
779 /* mkdir .snap/foo is a MKSNAP */
780 op
= CEPH_MDS_OP_MKSNAP
;
781 dout("mksnap dir %p snap '%.*s' dn %p\n", dir
,
782 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
);
783 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
784 dout("mkdir dir %p dn %p mode 0%ho\n", dir
, dentry
, mode
);
785 op
= CEPH_MDS_OP_MKDIR
;
789 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
795 req
->r_dentry
= dget(dentry
);
797 req
->r_locked_dir
= dir
;
798 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
799 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
800 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
801 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
802 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
803 err
= ceph_handle_notrace_create(dir
, dentry
);
804 ceph_mdsc_put_request(req
);
811 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
812 struct dentry
*dentry
)
814 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
815 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
816 struct ceph_mds_request
*req
;
819 if (ceph_snap(dir
) != CEPH_NOSNAP
)
822 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
824 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
829 req
->r_dentry
= dget(dentry
);
831 req
->r_old_dentry
= dget(old_dentry
); /* or inode? hrm. */
832 req
->r_old_dentry_dir
= ceph_get_dentry_parent_inode(old_dentry
);
833 req
->r_locked_dir
= dir
;
834 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
835 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
836 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
839 } else if (!req
->r_reply_info
.head
->is_dentry
) {
840 ihold(old_dentry
->d_inode
);
841 d_instantiate(dentry
, old_dentry
->d_inode
);
843 ceph_mdsc_put_request(req
);
848 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
849 * looks like the link count will hit 0, drop any other caps (other
850 * than PIN) we don't specifically want (due to the file still being
853 static int drop_caps_for_unlink(struct inode
*inode
)
855 struct ceph_inode_info
*ci
= ceph_inode(inode
);
856 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
858 spin_lock(&ci
->i_ceph_lock
);
859 if (inode
->i_nlink
== 1) {
860 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
861 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
863 spin_unlock(&ci
->i_ceph_lock
);
868 * rmdir and unlink are differ only by the metadata op code
870 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
872 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
873 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
874 struct inode
*inode
= dentry
->d_inode
;
875 struct ceph_mds_request
*req
;
879 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
880 /* rmdir .snap/foo is RMSNAP */
881 dout("rmsnap dir %p '%.*s' dn %p\n", dir
, dentry
->d_name
.len
,
882 dentry
->d_name
.name
, dentry
);
883 op
= CEPH_MDS_OP_RMSNAP
;
884 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
885 dout("unlink/rmdir dir %p dn %p inode %p\n",
887 op
= S_ISDIR(dentry
->d_inode
->i_mode
) ?
888 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
891 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
896 req
->r_dentry
= dget(dentry
);
898 req
->r_locked_dir
= dir
;
899 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
900 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
901 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
902 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
903 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
905 ceph_mdsc_put_request(req
);
910 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
911 struct inode
*new_dir
, struct dentry
*new_dentry
)
913 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
914 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
915 struct ceph_mds_request
*req
;
918 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
920 if (ceph_snap(old_dir
) != CEPH_NOSNAP
||
921 ceph_snap(new_dir
) != CEPH_NOSNAP
)
923 dout("rename dir %p dentry %p to dir %p dentry %p\n",
924 old_dir
, old_dentry
, new_dir
, new_dentry
);
925 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_RENAME
, USE_AUTH_MDS
);
928 req
->r_dentry
= dget(new_dentry
);
930 req
->r_old_dentry
= dget(old_dentry
);
931 req
->r_old_dentry_dir
= ceph_get_dentry_parent_inode(old_dentry
);
932 req
->r_locked_dir
= new_dir
;
933 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
934 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
935 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
936 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
937 /* release LINK_RDCACHE on source inode (mds will lock it) */
938 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
939 if (new_dentry
->d_inode
)
940 req
->r_inode_drop
= drop_caps_for_unlink(new_dentry
->d_inode
);
941 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
942 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
944 * Normally d_move() is done by fill_trace (called by
945 * do_request, above). If there is no trace, we need
949 /* d_move screws up d_subdirs order */
950 ceph_dir_clear_complete(new_dir
);
952 d_move(old_dentry
, new_dentry
);
954 /* ensure target dentry is invalidated, despite
955 rehashing bug in vfs_rename_dir */
956 ceph_invalidate_dentry_lease(new_dentry
);
958 ceph_mdsc_put_request(req
);
963 * Ensure a dentry lease will no longer revalidate.
965 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
967 spin_lock(&dentry
->d_lock
);
968 dentry
->d_time
= jiffies
;
969 ceph_dentry(dentry
)->lease_shared_gen
= 0;
970 spin_unlock(&dentry
->d_lock
);
974 * Check if dentry lease is valid. If not, delete the lease. Try to
975 * renew if the least is more than half up.
977 static int dentry_lease_is_valid(struct dentry
*dentry
)
979 struct ceph_dentry_info
*di
;
980 struct ceph_mds_session
*s
;
984 struct ceph_mds_session
*session
= NULL
;
985 struct inode
*dir
= NULL
;
988 spin_lock(&dentry
->d_lock
);
989 di
= ceph_dentry(dentry
);
990 if (di
->lease_session
) {
991 s
= di
->lease_session
;
992 spin_lock(&s
->s_gen_ttl_lock
);
995 spin_unlock(&s
->s_gen_ttl_lock
);
997 if (di
->lease_gen
== gen
&&
998 time_before(jiffies
, dentry
->d_time
) &&
999 time_before(jiffies
, ttl
)) {
1001 if (di
->lease_renew_after
&&
1002 time_after(jiffies
, di
->lease_renew_after
)) {
1003 /* we should renew */
1004 dir
= dentry
->d_parent
->d_inode
;
1005 session
= ceph_get_mds_session(s
);
1006 seq
= di
->lease_seq
;
1007 di
->lease_renew_after
= 0;
1008 di
->lease_renew_from
= jiffies
;
1012 spin_unlock(&dentry
->d_lock
);
1015 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
1016 CEPH_MDS_LEASE_RENEW
, seq
);
1017 ceph_put_mds_session(session
);
1019 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
1024 * Check if directory-wide content lease/cap is valid.
1026 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
1028 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1029 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1032 spin_lock(&ci
->i_ceph_lock
);
1033 if (ci
->i_shared_gen
== di
->lease_shared_gen
)
1034 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
1035 spin_unlock(&ci
->i_ceph_lock
);
1036 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1037 dir
, (unsigned)ci
->i_shared_gen
, dentry
,
1038 (unsigned)di
->lease_shared_gen
, valid
);
1043 * Check if cached dentry can be trusted.
1045 static int ceph_d_revalidate(struct dentry
*dentry
, unsigned int flags
)
1050 if (flags
& LOOKUP_RCU
)
1053 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry
,
1054 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
,
1055 ceph_dentry(dentry
)->offset
);
1057 dir
= ceph_get_dentry_parent_inode(dentry
);
1059 /* always trust cached snapped dentries, snapdir dentry */
1060 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
1061 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry
,
1062 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
1064 } else if (dentry
->d_inode
&&
1065 ceph_snap(dentry
->d_inode
) == CEPH_SNAPDIR
) {
1067 } else if (dentry_lease_is_valid(dentry
) ||
1068 dir_lease_is_valid(dir
, dentry
)) {
1072 dout("d_revalidate %p %s\n", dentry
, valid
? "valid" : "invalid");
1074 ceph_dentry_lru_touch(dentry
);
1082 * Release our ceph_dentry_info.
1084 static void ceph_d_release(struct dentry
*dentry
)
1086 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1088 dout("d_release %p\n", dentry
);
1089 ceph_dentry_lru_del(dentry
);
1090 if (di
->lease_session
)
1091 ceph_put_mds_session(di
->lease_session
);
1092 kmem_cache_free(ceph_dentry_cachep
, di
);
1093 dentry
->d_fsdata
= NULL
;
1096 static int ceph_snapdir_d_revalidate(struct dentry
*dentry
,
1100 * Eventually, we'll want to revalidate snapped metadata
1101 * too... probably...
1107 * Set/clear/test dir complete flag on the dir's dentry.
1109 void ceph_dir_set_complete(struct inode
*inode
)
1111 struct dentry
*dentry
= d_find_any_alias(inode
);
1113 if (dentry
&& ceph_dentry(dentry
) &&
1114 ceph_test_mount_opt(ceph_sb_to_client(dentry
->d_sb
), DCACHE
)) {
1115 dout(" marking %p (%p) complete\n", inode
, dentry
);
1116 set_bit(CEPH_D_COMPLETE
, &ceph_dentry(dentry
)->flags
);
1121 void ceph_dir_clear_complete(struct inode
*inode
)
1123 struct dentry
*dentry
= d_find_any_alias(inode
);
1125 if (dentry
&& ceph_dentry(dentry
)) {
1126 dout(" marking %p (%p) complete\n", inode
, dentry
);
1127 set_bit(CEPH_D_COMPLETE
, &ceph_dentry(dentry
)->flags
);
1132 bool ceph_dir_test_complete(struct inode
*inode
)
1134 struct dentry
*dentry
= d_find_any_alias(inode
);
1136 if (dentry
&& ceph_dentry(dentry
)) {
1137 dout(" marking %p (%p) NOT complete\n", inode
, dentry
);
1138 clear_bit(CEPH_D_COMPLETE
, &ceph_dentry(dentry
)->flags
);
1145 * When the VFS prunes a dentry from the cache, we need to clear the
1146 * complete flag on the parent directory.
1148 * Called under dentry->d_lock.
1150 static void ceph_d_prune(struct dentry
*dentry
)
1152 struct ceph_dentry_info
*di
;
1154 dout("ceph_d_prune %p\n", dentry
);
1156 /* do we have a valid parent? */
1157 if (!dentry
->d_parent
|| IS_ROOT(dentry
))
1160 /* if we are not hashed, we don't affect D_COMPLETE */
1161 if (d_unhashed(dentry
))
1165 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1166 * cleared until d_release
1168 di
= ceph_dentry(dentry
->d_parent
);
1169 clear_bit(CEPH_D_COMPLETE
, &di
->flags
);
1173 * read() on a dir. This weird interface hack only works if mounted
1174 * with '-o dirstat'.
1176 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1179 struct ceph_file_info
*cf
= file
->private_data
;
1180 struct inode
*inode
= file
->f_dentry
->d_inode
;
1181 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1183 const int bufsize
= 1024;
1185 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1188 if (!cf
->dir_info
) {
1189 cf
->dir_info
= kmalloc(bufsize
, GFP_NOFS
);
1193 snprintf(cf
->dir_info
, bufsize
,
1196 " subdirs: %20lld\n"
1197 "rentries: %20lld\n"
1199 " rsubdirs: %20lld\n"
1201 "rctime: %10ld.%09ld\n",
1202 ci
->i_files
+ ci
->i_subdirs
,
1205 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1209 (long)ci
->i_rctime
.tv_sec
,
1210 (long)ci
->i_rctime
.tv_nsec
);
1213 if (*ppos
>= cf
->dir_info_len
)
1215 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1216 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1219 *ppos
+= (size
- left
);
1224 * an fsync() on a dir will wait for any uncommitted directory
1225 * operations to commit.
1227 static int ceph_dir_fsync(struct file
*file
, loff_t start
, loff_t end
,
1230 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1231 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1232 struct list_head
*head
= &ci
->i_unsafe_dirops
;
1233 struct ceph_mds_request
*req
;
1237 dout("dir_fsync %p\n", inode
);
1238 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1241 mutex_lock(&inode
->i_mutex
);
1243 spin_lock(&ci
->i_unsafe_lock
);
1244 if (list_empty(head
))
1247 req
= list_entry(head
->prev
,
1248 struct ceph_mds_request
, r_unsafe_dir_item
);
1249 last_tid
= req
->r_tid
;
1252 ceph_mdsc_get_request(req
);
1253 spin_unlock(&ci
->i_unsafe_lock
);
1255 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1256 inode
, req
->r_tid
, last_tid
);
1257 if (req
->r_timeout
) {
1258 ret
= wait_for_completion_timeout(
1259 &req
->r_safe_completion
, req
->r_timeout
);
1263 ret
= -EIO
; /* timed out */
1265 wait_for_completion(&req
->r_safe_completion
);
1267 ceph_mdsc_put_request(req
);
1269 spin_lock(&ci
->i_unsafe_lock
);
1270 if (ret
|| list_empty(head
))
1272 req
= list_entry(head
->next
,
1273 struct ceph_mds_request
, r_unsafe_dir_item
);
1274 } while (req
->r_tid
< last_tid
);
1276 spin_unlock(&ci
->i_unsafe_lock
);
1277 mutex_unlock(&inode
->i_mutex
);
1283 * We maintain a private dentry LRU.
1285 * FIXME: this needs to be changed to a per-mds lru to be useful.
1287 void ceph_dentry_lru_add(struct dentry
*dn
)
1289 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1290 struct ceph_mds_client
*mdsc
;
1292 dout("dentry_lru_add %p %p '%.*s'\n", di
, dn
,
1293 dn
->d_name
.len
, dn
->d_name
.name
);
1294 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1295 spin_lock(&mdsc
->dentry_lru_lock
);
1296 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1298 spin_unlock(&mdsc
->dentry_lru_lock
);
1301 void ceph_dentry_lru_touch(struct dentry
*dn
)
1303 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1304 struct ceph_mds_client
*mdsc
;
1306 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di
, dn
,
1307 dn
->d_name
.len
, dn
->d_name
.name
, di
->offset
);
1308 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1309 spin_lock(&mdsc
->dentry_lru_lock
);
1310 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1311 spin_unlock(&mdsc
->dentry_lru_lock
);
1314 void ceph_dentry_lru_del(struct dentry
*dn
)
1316 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1317 struct ceph_mds_client
*mdsc
;
1319 dout("dentry_lru_del %p %p '%.*s'\n", di
, dn
,
1320 dn
->d_name
.len
, dn
->d_name
.name
);
1321 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1322 spin_lock(&mdsc
->dentry_lru_lock
);
1323 list_del_init(&di
->lru
);
1325 spin_unlock(&mdsc
->dentry_lru_lock
);
1329 * Return name hash for a given dentry. This is dependent on
1330 * the parent directory's hash function.
1332 unsigned ceph_dentry_hash(struct inode
*dir
, struct dentry
*dn
)
1334 struct ceph_inode_info
*dci
= ceph_inode(dir
);
1336 switch (dci
->i_dir_layout
.dl_dir_hash
) {
1337 case 0: /* for backward compat */
1338 case CEPH_STR_HASH_LINUX
:
1339 return dn
->d_name
.hash
;
1342 return ceph_str_hash(dci
->i_dir_layout
.dl_dir_hash
,
1343 dn
->d_name
.name
, dn
->d_name
.len
);
1347 const struct file_operations ceph_dir_fops
= {
1348 .read
= ceph_read_dir
,
1349 .readdir
= ceph_readdir
,
1350 .llseek
= ceph_dir_llseek
,
1352 .release
= ceph_release
,
1353 .unlocked_ioctl
= ceph_ioctl
,
1354 .fsync
= ceph_dir_fsync
,
1357 const struct inode_operations ceph_dir_iops
= {
1358 .lookup
= ceph_lookup
,
1359 .permission
= ceph_permission
,
1360 .getattr
= ceph_getattr
,
1361 .setattr
= ceph_setattr
,
1362 .setxattr
= ceph_setxattr
,
1363 .getxattr
= ceph_getxattr
,
1364 .listxattr
= ceph_listxattr
,
1365 .removexattr
= ceph_removexattr
,
1366 .mknod
= ceph_mknod
,
1367 .symlink
= ceph_symlink
,
1368 .mkdir
= ceph_mkdir
,
1370 .unlink
= ceph_unlink
,
1371 .rmdir
= ceph_unlink
,
1372 .rename
= ceph_rename
,
1373 .create
= ceph_create
,
1374 .atomic_open
= ceph_atomic_open
,
1377 const struct dentry_operations ceph_dentry_ops
= {
1378 .d_revalidate
= ceph_d_revalidate
,
1379 .d_release
= ceph_d_release
,
1380 .d_prune
= ceph_d_prune
,
1383 const struct dentry_operations ceph_snapdir_dentry_ops
= {
1384 .d_revalidate
= ceph_snapdir_d_revalidate
,
1385 .d_release
= ceph_d_release
,
1388 const struct dentry_operations ceph_snap_dentry_ops
= {
1389 .d_release
= ceph_d_release
,
1390 .d_prune
= ceph_d_prune
,