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 dentry_operations ceph_dentry_ops
;
32 * Initialize ceph dentry state.
34 int ceph_init_dentry(struct dentry
*dentry
)
36 struct ceph_dentry_info
*di
;
41 di
= kmem_cache_alloc(ceph_dentry_cachep
, GFP_KERNEL
| __GFP_ZERO
);
43 return -ENOMEM
; /* oh well */
45 spin_lock(&dentry
->d_lock
);
46 if (dentry
->d_fsdata
) {
48 kmem_cache_free(ceph_dentry_cachep
, di
);
52 if (ceph_snap(d_inode(dentry
->d_parent
)) == CEPH_NOSNAP
)
53 d_set_d_op(dentry
, &ceph_dentry_ops
);
54 else if (ceph_snap(d_inode(dentry
->d_parent
)) == CEPH_SNAPDIR
)
55 d_set_d_op(dentry
, &ceph_snapdir_dentry_ops
);
57 d_set_d_op(dentry
, &ceph_snap_dentry_ops
);
60 di
->lease_session
= NULL
;
61 dentry
->d_time
= jiffies
;
62 /* avoid reordering d_fsdata setup so that the check above is safe */
64 dentry
->d_fsdata
= di
;
65 ceph_dentry_lru_add(dentry
);
67 spin_unlock(&dentry
->d_lock
);
71 struct inode
*ceph_get_dentry_parent_inode(struct dentry
*dentry
)
73 struct inode
*inode
= NULL
;
78 spin_lock(&dentry
->d_lock
);
79 if (!IS_ROOT(dentry
)) {
80 inode
= d_inode(dentry
->d_parent
);
83 spin_unlock(&dentry
->d_lock
);
89 * for readdir, we encode the directory frag and offset within that
92 static unsigned fpos_frag(loff_t p
)
96 static unsigned fpos_off(loff_t p
)
98 return p
& 0xffffffff;
101 static int fpos_cmp(loff_t l
, loff_t r
)
103 int v
= ceph_frag_compare(fpos_frag(l
), fpos_frag(r
));
106 return (int)(fpos_off(l
) - fpos_off(r
));
110 * make note of the last dentry we read, so we can
111 * continue at the same lexicographical point,
112 * regardless of what dir changes take place on the
115 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
116 int len
, unsigned next_offset
)
118 char *buf
= kmalloc(len
+1, GFP_KERNEL
);
121 kfree(fi
->last_name
);
123 memcpy(fi
->last_name
, name
, len
);
124 fi
->last_name
[len
] = 0;
125 fi
->next_offset
= next_offset
;
126 dout("note_last_dentry '%s'\n", fi
->last_name
);
131 * When possible, we try to satisfy a readdir by peeking at the
132 * dcache. We make this work by carefully ordering dentries on
133 * d_child when we initially get results back from the MDS, and
134 * falling back to a "normal" sync readdir if any dentries in the dir
137 * Complete dir indicates that we have all dentries in the dir. It is
138 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
139 * the MDS if/when the directory is modified).
141 static int __dcache_readdir(struct file
*file
, struct dir_context
*ctx
,
144 struct ceph_file_info
*fi
= file
->private_data
;
145 struct dentry
*parent
= file
->f_path
.dentry
;
146 struct inode
*dir
= d_inode(parent
);
147 struct dentry
*dentry
, *last
= NULL
;
148 struct ceph_dentry_info
*di
;
149 unsigned nsize
= PAGE_CACHE_SIZE
/ sizeof(struct dentry
*);
152 struct ceph_readdir_cache_control cache_ctl
= {};
154 dout("__dcache_readdir %p v%u at %llu\n", dir
, shared_gen
, ctx
->pos
);
156 /* we can calculate cache index for the first dirfrag */
157 if (ceph_frag_is_leftmost(fpos_frag(ctx
->pos
))) {
158 cache_ctl
.index
= fpos_off(ctx
->pos
) - 2;
159 BUG_ON(cache_ctl
.index
< 0);
160 ptr_pos
= cache_ctl
.index
* sizeof(struct dentry
*);
167 if (ptr_pos
>= i_size_read(dir
)) {
168 fi
->flags
|= CEPH_F_ATEND
;
174 pgoff
= ptr_pos
>> PAGE_CACHE_SHIFT
;
175 if (!cache_ctl
.page
|| pgoff
!= page_index(cache_ctl
.page
)) {
176 ceph_readdir_cache_release(&cache_ctl
);
177 cache_ctl
.page
= find_lock_page(&dir
->i_data
, pgoff
);
178 if (!cache_ctl
.page
) {
179 dout(" page %lu not found\n", pgoff
);
182 /* reading/filling the cache are serialized by
183 * i_mutex, no need to use page lock */
184 unlock_page(cache_ctl
.page
);
185 cache_ctl
.dentries
= kmap(cache_ctl
.page
);
189 spin_lock(&parent
->d_lock
);
190 /* check i_size again here, because empty directory can be
191 * marked as complete while not holding the i_mutex. */
192 if (ceph_dir_is_complete_ordered(dir
) &&
193 ptr_pos
< i_size_read(dir
))
194 dentry
= cache_ctl
.dentries
[cache_ctl
.index
% nsize
];
197 spin_unlock(&parent
->d_lock
);
198 if (dentry
&& !lockref_get_not_dead(&dentry
->d_lockref
))
205 di
= ceph_dentry(dentry
);
206 spin_lock(&dentry
->d_lock
);
207 if (di
->lease_shared_gen
== shared_gen
&&
208 d_really_is_positive(dentry
) &&
209 ceph_snap(d_inode(dentry
)) != CEPH_SNAPDIR
&&
210 ceph_ino(d_inode(dentry
)) != CEPH_INO_CEPH
&&
211 fpos_cmp(ctx
->pos
, di
->offset
) <= 0) {
214 spin_unlock(&dentry
->d_lock
);
217 dout(" %llu (%llu) dentry %p %pd %p\n", di
->offset
, ctx
->pos
,
218 dentry
, dentry
, d_inode(dentry
));
219 ctx
->pos
= di
->offset
;
220 if (!dir_emit(ctx
, dentry
->d_name
.name
,
222 ceph_translate_ino(dentry
->d_sb
,
223 d_inode(dentry
)->i_ino
),
224 d_inode(dentry
)->i_mode
>> 12)) {
239 ptr_pos
+= sizeof(struct dentry
*);
241 ceph_readdir_cache_release(&cache_ctl
);
244 di
= ceph_dentry(last
);
245 ret
= note_last_dentry(fi
, last
->d_name
.name
, last
->d_name
.len
,
246 fpos_off(di
->offset
) + 1);
250 /* last_name no longer match cache index */
251 if (fi
->readdir_cache_idx
>= 0) {
252 fi
->readdir_cache_idx
= -1;
253 fi
->dir_release_count
= 0;
259 static int ceph_readdir(struct file
*file
, struct dir_context
*ctx
)
261 struct ceph_file_info
*fi
= file
->private_data
;
262 struct inode
*inode
= file_inode(file
);
263 struct ceph_inode_info
*ci
= ceph_inode(inode
);
264 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
265 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
266 unsigned frag
= fpos_frag(ctx
->pos
);
267 int off
= fpos_off(ctx
->pos
);
270 struct ceph_mds_reply_info_parsed
*rinfo
;
272 dout("readdir %p file %p frag %u off %u\n", inode
, file
, frag
, off
);
273 if (fi
->flags
& CEPH_F_ATEND
)
276 /* always start with . and .. */
278 dout("readdir off 0 -> '.'\n");
279 if (!dir_emit(ctx
, ".", 1,
280 ceph_translate_ino(inode
->i_sb
, inode
->i_ino
),
281 inode
->i_mode
>> 12))
287 ino_t ino
= parent_ino(file
->f_path
.dentry
);
288 dout("readdir off 1 -> '..'\n");
289 if (!dir_emit(ctx
, "..", 2,
290 ceph_translate_ino(inode
->i_sb
, ino
),
291 inode
->i_mode
>> 12))
297 /* can we use the dcache? */
298 spin_lock(&ci
->i_ceph_lock
);
299 if (ceph_test_mount_opt(fsc
, DCACHE
) &&
300 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
301 ceph_snap(inode
) != CEPH_SNAPDIR
&&
302 __ceph_dir_is_complete_ordered(ci
) &&
303 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
304 u32 shared_gen
= ci
->i_shared_gen
;
305 spin_unlock(&ci
->i_ceph_lock
);
306 err
= __dcache_readdir(file
, ctx
, shared_gen
);
309 frag
= fpos_frag(ctx
->pos
);
310 off
= fpos_off(ctx
->pos
);
312 spin_unlock(&ci
->i_ceph_lock
);
315 /* proceed with a normal readdir */
317 /* do we have the correct frag content buffered? */
318 if (fi
->frag
!= frag
|| fi
->last_readdir
== NULL
) {
319 struct ceph_mds_request
*req
;
320 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
321 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
323 /* discard old result, if any */
324 if (fi
->last_readdir
) {
325 ceph_mdsc_put_request(fi
->last_readdir
);
326 fi
->last_readdir
= NULL
;
329 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
330 ceph_vinop(inode
), frag
, fi
->last_name
);
331 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
334 err
= ceph_alloc_readdir_reply_buffer(req
, inode
);
336 ceph_mdsc_put_request(req
);
339 /* hints to request -> mds selection code */
340 req
->r_direct_mode
= USE_AUTH_MDS
;
341 req
->r_direct_hash
= ceph_frag_value(frag
);
342 req
->r_direct_is_hash
= true;
344 req
->r_path2
= kstrdup(fi
->last_name
, GFP_KERNEL
);
346 ceph_mdsc_put_request(req
);
350 req
->r_dir_release_cnt
= fi
->dir_release_count
;
351 req
->r_dir_ordered_cnt
= fi
->dir_ordered_count
;
352 req
->r_readdir_cache_idx
= fi
->readdir_cache_idx
;
353 req
->r_readdir_offset
= fi
->next_offset
;
354 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
356 req
->r_inode
= inode
;
358 req
->r_dentry
= dget(file
->f_path
.dentry
);
359 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
361 ceph_mdsc_put_request(req
);
364 dout("readdir got and parsed readdir result=%d"
365 " on frag %x, end=%d, complete=%d\n", err
, frag
,
366 (int)req
->r_reply_info
.dir_end
,
367 (int)req
->r_reply_info
.dir_complete
);
370 /* note next offset and last dentry name */
371 rinfo
= &req
->r_reply_info
;
372 if (le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
373 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
374 off
= req
->r_readdir_offset
;
375 fi
->next_offset
= off
;
379 fi
->offset
= fi
->next_offset
;
380 fi
->last_readdir
= req
;
382 if (req
->r_did_prepopulate
) {
383 fi
->readdir_cache_idx
= req
->r_readdir_cache_idx
;
384 if (fi
->readdir_cache_idx
< 0) {
385 /* preclude from marking dir ordered */
386 fi
->dir_ordered_count
= 0;
387 } else if (ceph_frag_is_leftmost(frag
) && off
== 2) {
388 /* note dir version at start of readdir so
389 * we can tell if any dentries get dropped */
390 fi
->dir_release_count
= req
->r_dir_release_cnt
;
391 fi
->dir_ordered_count
= req
->r_dir_ordered_cnt
;
394 dout("readdir !did_prepopulate");
395 /* disable readdir cache */
396 fi
->readdir_cache_idx
= -1;
397 /* preclude from marking dir complete */
398 fi
->dir_release_count
= 0;
401 if (req
->r_reply_info
.dir_end
) {
402 kfree(fi
->last_name
);
403 fi
->last_name
= NULL
;
404 if (ceph_frag_is_rightmost(frag
))
409 err
= note_last_dentry(fi
,
410 rinfo
->dir_dname
[rinfo
->dir_nr
-1],
411 rinfo
->dir_dname_len
[rinfo
->dir_nr
-1],
412 fi
->next_offset
+ rinfo
->dir_nr
);
418 rinfo
= &fi
->last_readdir
->r_reply_info
;
419 dout("readdir frag %x num %d off %d chunkoff %d\n", frag
,
420 rinfo
->dir_nr
, off
, fi
->offset
);
422 ctx
->pos
= ceph_make_fpos(frag
, off
);
423 while (off
>= fi
->offset
&& off
- fi
->offset
< rinfo
->dir_nr
) {
424 struct ceph_mds_reply_inode
*in
=
425 rinfo
->dir_in
[off
- fi
->offset
].in
;
426 struct ceph_vino vino
;
429 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
430 off
, off
- fi
->offset
, rinfo
->dir_nr
, ctx
->pos
,
431 rinfo
->dir_dname_len
[off
- fi
->offset
],
432 rinfo
->dir_dname
[off
- fi
->offset
], in
);
434 ftype
= le32_to_cpu(in
->mode
) >> 12;
435 vino
.ino
= le64_to_cpu(in
->ino
);
436 vino
.snap
= le64_to_cpu(in
->snapid
);
437 ino
= ceph_vino_to_ino(vino
);
439 rinfo
->dir_dname
[off
- fi
->offset
],
440 rinfo
->dir_dname_len
[off
- fi
->offset
],
441 ceph_translate_ino(inode
->i_sb
, ino
), ftype
)) {
442 dout("filldir stopping us...\n");
450 ceph_mdsc_put_request(fi
->last_readdir
);
451 fi
->last_readdir
= NULL
;
456 if (!ceph_frag_is_rightmost(frag
)) {
457 frag
= ceph_frag_next(frag
);
459 ctx
->pos
= ceph_make_fpos(frag
, off
);
460 dout("readdir next frag is %x\n", frag
);
463 fi
->flags
|= CEPH_F_ATEND
;
466 * if dir_release_count still matches the dir, no dentries
467 * were released during the whole readdir, and we should have
468 * the complete dir contents in our cache.
470 if (atomic64_read(&ci
->i_release_count
) == fi
->dir_release_count
) {
471 spin_lock(&ci
->i_ceph_lock
);
472 if (fi
->dir_ordered_count
== atomic64_read(&ci
->i_ordered_count
)) {
473 dout(" marking %p complete and ordered\n", inode
);
474 /* use i_size to track number of entries in
476 BUG_ON(fi
->readdir_cache_idx
< 0);
477 i_size_write(inode
, fi
->readdir_cache_idx
*
478 sizeof(struct dentry
*));
480 dout(" marking %p complete\n", inode
);
482 __ceph_dir_set_complete(ci
, fi
->dir_release_count
,
483 fi
->dir_ordered_count
);
484 spin_unlock(&ci
->i_ceph_lock
);
487 dout("readdir %p file %p done.\n", inode
, file
);
491 static void reset_readdir(struct ceph_file_info
*fi
, unsigned frag
)
493 if (fi
->last_readdir
) {
494 ceph_mdsc_put_request(fi
->last_readdir
);
495 fi
->last_readdir
= NULL
;
497 kfree(fi
->last_name
);
498 fi
->last_name
= NULL
;
499 fi
->dir_release_count
= 0;
500 fi
->readdir_cache_idx
= -1;
501 if (ceph_frag_is_leftmost(frag
))
502 fi
->next_offset
= 2; /* compensate for . and .. */
505 fi
->flags
&= ~CEPH_F_ATEND
;
508 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
510 struct ceph_file_info
*fi
= file
->private_data
;
511 struct inode
*inode
= file
->f_mapping
->host
;
512 loff_t old_offset
= ceph_make_fpos(fi
->frag
, fi
->next_offset
);
515 mutex_lock(&inode
->i_mutex
);
519 offset
+= file
->f_pos
;
523 retval
= -EOPNOTSUPP
;
529 if (offset
!= file
->f_pos
) {
530 file
->f_pos
= offset
;
532 fi
->flags
&= ~CEPH_F_ATEND
;
537 fpos_frag(offset
) != fi
->frag
||
538 fpos_off(offset
) < fi
->offset
) {
539 /* discard buffered readdir content on seekdir(0), or
540 * seek to new frag, or seek prior to current chunk */
541 dout("dir_llseek dropping %p content\n", file
);
542 reset_readdir(fi
, fpos_frag(offset
));
543 } else if (fpos_cmp(offset
, old_offset
) > 0) {
544 /* reset dir_release_count if we did a forward seek */
545 fi
->dir_release_count
= 0;
546 fi
->readdir_cache_idx
= -1;
550 mutex_unlock(&inode
->i_mutex
);
555 * Handle lookups for the hidden .snap directory.
557 int ceph_handle_snapdir(struct ceph_mds_request
*req
,
558 struct dentry
*dentry
, int err
)
560 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
561 struct inode
*parent
= d_inode(dentry
->d_parent
); /* we hold i_mutex */
564 if (err
== -ENOENT
&&
565 ceph_snap(parent
) == CEPH_NOSNAP
&&
566 strcmp(dentry
->d_name
.name
,
567 fsc
->mount_options
->snapdir_name
) == 0) {
568 struct inode
*inode
= ceph_get_snapdir(parent
);
569 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
570 dentry
, dentry
, inode
);
571 BUG_ON(!d_unhashed(dentry
));
572 d_add(dentry
, inode
);
579 * Figure out final result of a lookup/open request.
581 * Mainly, make sure we return the final req->r_dentry (if it already
582 * existed) in place of the original VFS-provided dentry when they
585 * Gracefully handle the case where the MDS replies with -ENOENT and
586 * no trace (which it may do, at its discretion, e.g., if it doesn't
587 * care to issue a lease on the negative dentry).
589 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
590 struct dentry
*dentry
, int err
)
592 if (err
== -ENOENT
) {
595 if (!req
->r_reply_info
.head
->is_dentry
) {
596 dout("ENOENT and no trace, dentry %p inode %p\n",
597 dentry
, d_inode(dentry
));
598 if (d_really_is_positive(dentry
)) {
607 dentry
= ERR_PTR(err
);
608 else if (dentry
!= req
->r_dentry
)
609 dentry
= dget(req
->r_dentry
); /* we got spliced */
615 static int is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
617 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
618 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
622 * Look up a single dir entry. If there is a lookup intent, inform
623 * the MDS so that it gets our 'caps wanted' value in a single op.
625 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
628 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
629 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
630 struct ceph_mds_request
*req
;
634 dout("lookup %p dentry %p '%pd'\n",
635 dir
, dentry
, dentry
);
637 if (dentry
->d_name
.len
> NAME_MAX
)
638 return ERR_PTR(-ENAMETOOLONG
);
640 err
= ceph_init_dentry(dentry
);
644 /* can we conclude ENOENT locally? */
645 if (d_really_is_negative(dentry
)) {
646 struct ceph_inode_info
*ci
= ceph_inode(dir
);
647 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
649 spin_lock(&ci
->i_ceph_lock
);
650 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
651 if (strncmp(dentry
->d_name
.name
,
652 fsc
->mount_options
->snapdir_name
,
653 dentry
->d_name
.len
) &&
654 !is_root_ceph_dentry(dir
, dentry
) &&
655 ceph_test_mount_opt(fsc
, DCACHE
) &&
656 __ceph_dir_is_complete(ci
) &&
657 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
658 spin_unlock(&ci
->i_ceph_lock
);
659 dout(" dir %p complete, -ENOENT\n", dir
);
661 di
->lease_shared_gen
= ci
->i_shared_gen
;
664 spin_unlock(&ci
->i_ceph_lock
);
667 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
668 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
669 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
671 return ERR_CAST(req
);
672 req
->r_dentry
= dget(dentry
);
674 /* we only need inode linkage */
675 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
676 req
->r_locked_dir
= dir
;
677 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
678 err
= ceph_handle_snapdir(req
, dentry
, err
);
679 dentry
= ceph_finish_lookup(req
, dentry
, err
);
680 ceph_mdsc_put_request(req
); /* will dput(dentry) */
681 dout("lookup result=%p\n", dentry
);
686 * If we do a create but get no trace back from the MDS, follow up with
687 * a lookup (the VFS expects us to link up the provided dentry).
689 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
691 struct dentry
*result
= ceph_lookup(dir
, dentry
, 0);
693 if (result
&& !IS_ERR(result
)) {
695 * We created the item, then did a lookup, and found
696 * it was already linked to another inode we already
697 * had in our cache (and thus got spliced). To not
698 * confuse VFS (especially when inode is a directory),
699 * we don't link our dentry to that inode, return an
702 * This event should be rare and it happens only when
703 * we talk to old MDS. Recent MDS does not send traceless
704 * reply for request that creates new inode.
709 return PTR_ERR(result
);
712 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
713 umode_t mode
, dev_t rdev
)
715 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
716 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
717 struct ceph_mds_request
*req
;
718 struct ceph_acls_info acls
= {};
721 if (ceph_snap(dir
) != CEPH_NOSNAP
)
724 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
728 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
729 dir
, dentry
, mode
, rdev
);
730 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
735 req
->r_dentry
= dget(dentry
);
737 req
->r_locked_dir
= dir
;
738 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
739 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
740 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
741 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
743 req
->r_pagelist
= acls
.pagelist
;
744 acls
.pagelist
= NULL
;
746 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
747 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
748 err
= ceph_handle_notrace_create(dir
, dentry
);
749 ceph_mdsc_put_request(req
);
752 ceph_init_inode_acls(d_inode(dentry
), &acls
);
755 ceph_release_acls_info(&acls
);
759 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
762 return ceph_mknod(dir
, dentry
, mode
, 0);
765 static int ceph_symlink(struct inode
*dir
, struct dentry
*dentry
,
768 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
769 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
770 struct ceph_mds_request
*req
;
773 if (ceph_snap(dir
) != CEPH_NOSNAP
)
776 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
777 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
782 req
->r_path2
= kstrdup(dest
, GFP_KERNEL
);
785 ceph_mdsc_put_request(req
);
788 req
->r_locked_dir
= dir
;
789 req
->r_dentry
= dget(dentry
);
791 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
792 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
793 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
794 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
795 err
= ceph_handle_notrace_create(dir
, dentry
);
796 ceph_mdsc_put_request(req
);
803 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
805 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
806 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
807 struct ceph_mds_request
*req
;
808 struct ceph_acls_info acls
= {};
812 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
813 /* mkdir .snap/foo is a MKSNAP */
814 op
= CEPH_MDS_OP_MKSNAP
;
815 dout("mksnap dir %p snap '%pd' dn %p\n", dir
,
817 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
818 dout("mkdir dir %p dn %p mode 0%ho\n", dir
, dentry
, mode
);
819 op
= CEPH_MDS_OP_MKDIR
;
825 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
829 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
835 req
->r_dentry
= dget(dentry
);
837 req
->r_locked_dir
= dir
;
838 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
839 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
840 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
842 req
->r_pagelist
= acls
.pagelist
;
843 acls
.pagelist
= NULL
;
845 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
847 !req
->r_reply_info
.head
->is_target
&&
848 !req
->r_reply_info
.head
->is_dentry
)
849 err
= ceph_handle_notrace_create(dir
, dentry
);
850 ceph_mdsc_put_request(req
);
853 ceph_init_inode_acls(d_inode(dentry
), &acls
);
856 ceph_release_acls_info(&acls
);
860 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
861 struct dentry
*dentry
)
863 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
864 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
865 struct ceph_mds_request
*req
;
868 if (ceph_snap(dir
) != CEPH_NOSNAP
)
871 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
873 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
878 req
->r_dentry
= dget(dentry
);
880 req
->r_old_dentry
= dget(old_dentry
);
881 req
->r_locked_dir
= dir
;
882 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
883 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
884 /* release LINK_SHARED on source inode (mds will lock it) */
885 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
886 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
889 } else if (!req
->r_reply_info
.head
->is_dentry
) {
890 ihold(d_inode(old_dentry
));
891 d_instantiate(dentry
, d_inode(old_dentry
));
893 ceph_mdsc_put_request(req
);
898 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
899 * looks like the link count will hit 0, drop any other caps (other
900 * than PIN) we don't specifically want (due to the file still being
903 static int drop_caps_for_unlink(struct inode
*inode
)
905 struct ceph_inode_info
*ci
= ceph_inode(inode
);
906 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
908 spin_lock(&ci
->i_ceph_lock
);
909 if (inode
->i_nlink
== 1) {
910 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
911 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
913 spin_unlock(&ci
->i_ceph_lock
);
918 * rmdir and unlink are differ only by the metadata op code
920 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
922 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
923 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
924 struct inode
*inode
= d_inode(dentry
);
925 struct ceph_mds_request
*req
;
929 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
930 /* rmdir .snap/foo is RMSNAP */
931 dout("rmsnap dir %p '%pd' dn %p\n", dir
, dentry
, dentry
);
932 op
= CEPH_MDS_OP_RMSNAP
;
933 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
934 dout("unlink/rmdir dir %p dn %p inode %p\n",
936 op
= d_is_dir(dentry
) ?
937 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
940 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
945 req
->r_dentry
= dget(dentry
);
947 req
->r_locked_dir
= dir
;
948 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
949 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
950 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
951 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
952 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
954 ceph_mdsc_put_request(req
);
959 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
960 struct inode
*new_dir
, struct dentry
*new_dentry
)
962 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
963 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
964 struct ceph_mds_request
*req
;
965 int op
= CEPH_MDS_OP_RENAME
;
968 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
970 if (ceph_snap(old_dir
) != CEPH_NOSNAP
) {
971 if (old_dir
== new_dir
&& ceph_snap(old_dir
) == CEPH_SNAPDIR
)
972 op
= CEPH_MDS_OP_RENAMESNAP
;
976 dout("rename dir %p dentry %p to dir %p dentry %p\n",
977 old_dir
, old_dentry
, new_dir
, new_dentry
);
978 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
982 req
->r_dentry
= dget(new_dentry
);
984 req
->r_old_dentry
= dget(old_dentry
);
985 req
->r_old_dentry_dir
= old_dir
;
986 req
->r_locked_dir
= new_dir
;
987 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
988 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
989 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
990 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
991 /* release LINK_RDCACHE on source inode (mds will lock it) */
992 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
993 if (d_really_is_positive(new_dentry
))
994 req
->r_inode_drop
= drop_caps_for_unlink(d_inode(new_dentry
));
995 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
996 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
998 * Normally d_move() is done by fill_trace (called by
999 * do_request, above). If there is no trace, we need
1003 /* d_move screws up sibling dentries' offsets */
1004 ceph_dir_clear_complete(old_dir
);
1005 ceph_dir_clear_complete(new_dir
);
1007 d_move(old_dentry
, new_dentry
);
1009 /* ensure target dentry is invalidated, despite
1010 rehashing bug in vfs_rename_dir */
1011 ceph_invalidate_dentry_lease(new_dentry
);
1013 ceph_mdsc_put_request(req
);
1018 * Ensure a dentry lease will no longer revalidate.
1020 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
1022 spin_lock(&dentry
->d_lock
);
1023 dentry
->d_time
= jiffies
;
1024 ceph_dentry(dentry
)->lease_shared_gen
= 0;
1025 spin_unlock(&dentry
->d_lock
);
1029 * Check if dentry lease is valid. If not, delete the lease. Try to
1030 * renew if the least is more than half up.
1032 static int dentry_lease_is_valid(struct dentry
*dentry
)
1034 struct ceph_dentry_info
*di
;
1035 struct ceph_mds_session
*s
;
1039 struct ceph_mds_session
*session
= NULL
;
1040 struct inode
*dir
= NULL
;
1043 spin_lock(&dentry
->d_lock
);
1044 di
= ceph_dentry(dentry
);
1045 if (di
->lease_session
) {
1046 s
= di
->lease_session
;
1047 spin_lock(&s
->s_gen_ttl_lock
);
1050 spin_unlock(&s
->s_gen_ttl_lock
);
1052 if (di
->lease_gen
== gen
&&
1053 time_before(jiffies
, dentry
->d_time
) &&
1054 time_before(jiffies
, ttl
)) {
1056 if (di
->lease_renew_after
&&
1057 time_after(jiffies
, di
->lease_renew_after
)) {
1058 /* we should renew */
1059 dir
= d_inode(dentry
->d_parent
);
1060 session
= ceph_get_mds_session(s
);
1061 seq
= di
->lease_seq
;
1062 di
->lease_renew_after
= 0;
1063 di
->lease_renew_from
= jiffies
;
1067 spin_unlock(&dentry
->d_lock
);
1070 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
1071 CEPH_MDS_LEASE_RENEW
, seq
);
1072 ceph_put_mds_session(session
);
1074 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
1079 * Check if directory-wide content lease/cap is valid.
1081 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
1083 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1084 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1087 spin_lock(&ci
->i_ceph_lock
);
1088 if (ci
->i_shared_gen
== di
->lease_shared_gen
)
1089 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
1090 spin_unlock(&ci
->i_ceph_lock
);
1091 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1092 dir
, (unsigned)ci
->i_shared_gen
, dentry
,
1093 (unsigned)di
->lease_shared_gen
, valid
);
1098 * Check if cached dentry can be trusted.
1100 static int ceph_d_revalidate(struct dentry
*dentry
, unsigned int flags
)
1105 if (flags
& LOOKUP_RCU
)
1108 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry
,
1109 dentry
, d_inode(dentry
), ceph_dentry(dentry
)->offset
);
1111 dir
= ceph_get_dentry_parent_inode(dentry
);
1113 /* always trust cached snapped dentries, snapdir dentry */
1114 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
1115 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry
,
1116 dentry
, d_inode(dentry
));
1118 } else if (d_really_is_positive(dentry
) &&
1119 ceph_snap(d_inode(dentry
)) == CEPH_SNAPDIR
) {
1121 } else if (dentry_lease_is_valid(dentry
) ||
1122 dir_lease_is_valid(dir
, dentry
)) {
1123 if (d_really_is_positive(dentry
))
1124 valid
= ceph_is_any_caps(d_inode(dentry
));
1129 dout("d_revalidate %p %s\n", dentry
, valid
? "valid" : "invalid");
1131 ceph_dentry_lru_touch(dentry
);
1133 ceph_dir_clear_complete(dir
);
1140 * Release our ceph_dentry_info.
1142 static void ceph_d_release(struct dentry
*dentry
)
1144 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1146 dout("d_release %p\n", dentry
);
1147 ceph_dentry_lru_del(dentry
);
1148 if (di
->lease_session
)
1149 ceph_put_mds_session(di
->lease_session
);
1150 kmem_cache_free(ceph_dentry_cachep
, di
);
1151 dentry
->d_fsdata
= NULL
;
1154 static int ceph_snapdir_d_revalidate(struct dentry
*dentry
,
1158 * Eventually, we'll want to revalidate snapped metadata
1159 * too... probably...
1165 * When the VFS prunes a dentry from the cache, we need to clear the
1166 * complete flag on the parent directory.
1168 * Called under dentry->d_lock.
1170 static void ceph_d_prune(struct dentry
*dentry
)
1172 dout("ceph_d_prune %p\n", dentry
);
1174 /* do we have a valid parent? */
1175 if (IS_ROOT(dentry
))
1178 /* if we are not hashed, we don't affect dir's completeness */
1179 if (d_unhashed(dentry
))
1183 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1184 * cleared until d_release
1186 ceph_dir_clear_complete(d_inode(dentry
->d_parent
));
1190 * read() on a dir. This weird interface hack only works if mounted
1191 * with '-o dirstat'.
1193 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1196 struct ceph_file_info
*cf
= file
->private_data
;
1197 struct inode
*inode
= file_inode(file
);
1198 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1200 const int bufsize
= 1024;
1202 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1205 if (!cf
->dir_info
) {
1206 cf
->dir_info
= kmalloc(bufsize
, GFP_KERNEL
);
1210 snprintf(cf
->dir_info
, bufsize
,
1213 " subdirs: %20lld\n"
1214 "rentries: %20lld\n"
1216 " rsubdirs: %20lld\n"
1218 "rctime: %10ld.%09ld\n",
1219 ci
->i_files
+ ci
->i_subdirs
,
1222 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1226 (long)ci
->i_rctime
.tv_sec
,
1227 (long)ci
->i_rctime
.tv_nsec
);
1230 if (*ppos
>= cf
->dir_info_len
)
1232 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1233 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1236 *ppos
+= (size
- left
);
1241 * We maintain a private dentry LRU.
1243 * FIXME: this needs to be changed to a per-mds lru to be useful.
1245 void ceph_dentry_lru_add(struct dentry
*dn
)
1247 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1248 struct ceph_mds_client
*mdsc
;
1250 dout("dentry_lru_add %p %p '%pd'\n", di
, dn
, dn
);
1251 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1252 spin_lock(&mdsc
->dentry_lru_lock
);
1253 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1255 spin_unlock(&mdsc
->dentry_lru_lock
);
1258 void ceph_dentry_lru_touch(struct dentry
*dn
)
1260 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1261 struct ceph_mds_client
*mdsc
;
1263 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di
, dn
, dn
,
1265 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1266 spin_lock(&mdsc
->dentry_lru_lock
);
1267 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1268 spin_unlock(&mdsc
->dentry_lru_lock
);
1271 void ceph_dentry_lru_del(struct dentry
*dn
)
1273 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1274 struct ceph_mds_client
*mdsc
;
1276 dout("dentry_lru_del %p %p '%pd'\n", di
, dn
, dn
);
1277 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1278 spin_lock(&mdsc
->dentry_lru_lock
);
1279 list_del_init(&di
->lru
);
1281 spin_unlock(&mdsc
->dentry_lru_lock
);
1285 * Return name hash for a given dentry. This is dependent on
1286 * the parent directory's hash function.
1288 unsigned ceph_dentry_hash(struct inode
*dir
, struct dentry
*dn
)
1290 struct ceph_inode_info
*dci
= ceph_inode(dir
);
1293 switch (dci
->i_dir_layout
.dl_dir_hash
) {
1294 case 0: /* for backward compat */
1295 case CEPH_STR_HASH_LINUX
:
1296 return dn
->d_name
.hash
;
1299 spin_lock(&dn
->d_lock
);
1300 hash
= ceph_str_hash(dci
->i_dir_layout
.dl_dir_hash
,
1301 dn
->d_name
.name
, dn
->d_name
.len
);
1302 spin_unlock(&dn
->d_lock
);
1307 const struct file_operations ceph_dir_fops
= {
1308 .read
= ceph_read_dir
,
1309 .iterate
= ceph_readdir
,
1310 .llseek
= ceph_dir_llseek
,
1312 .release
= ceph_release
,
1313 .unlocked_ioctl
= ceph_ioctl
,
1314 .fsync
= ceph_fsync
,
1317 const struct file_operations ceph_snapdir_fops
= {
1318 .iterate
= ceph_readdir
,
1319 .llseek
= ceph_dir_llseek
,
1321 .release
= ceph_release
,
1324 const struct inode_operations ceph_dir_iops
= {
1325 .lookup
= ceph_lookup
,
1326 .permission
= ceph_permission
,
1327 .getattr
= ceph_getattr
,
1328 .setattr
= ceph_setattr
,
1329 .setxattr
= ceph_setxattr
,
1330 .getxattr
= ceph_getxattr
,
1331 .listxattr
= ceph_listxattr
,
1332 .removexattr
= ceph_removexattr
,
1333 .get_acl
= ceph_get_acl
,
1334 .set_acl
= ceph_set_acl
,
1335 .mknod
= ceph_mknod
,
1336 .symlink
= ceph_symlink
,
1337 .mkdir
= ceph_mkdir
,
1339 .unlink
= ceph_unlink
,
1340 .rmdir
= ceph_unlink
,
1341 .rename
= ceph_rename
,
1342 .create
= ceph_create
,
1343 .atomic_open
= ceph_atomic_open
,
1346 const struct inode_operations ceph_snapdir_iops
= {
1347 .lookup
= ceph_lookup
,
1348 .permission
= ceph_permission
,
1349 .getattr
= ceph_getattr
,
1350 .mkdir
= ceph_mkdir
,
1351 .rmdir
= ceph_unlink
,
1352 .rename
= ceph_rename
,
1355 const struct dentry_operations ceph_dentry_ops
= {
1356 .d_revalidate
= ceph_d_revalidate
,
1357 .d_release
= ceph_d_release
,
1358 .d_prune
= ceph_d_prune
,
1361 const struct dentry_operations ceph_snapdir_dentry_ops
= {
1362 .d_revalidate
= ceph_snapdir_d_revalidate
,
1363 .d_release
= ceph_d_release
,
1366 const struct dentry_operations ceph_snap_dentry_ops
= {
1367 .d_release
= ceph_d_release
,
1368 .d_prune
= ceph_d_prune
,