1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/spinlock.h>
5 #include <linux/fs_struct.h>
6 #include <linux/namei.h>
7 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/xattr.h>
12 #include "mds_client.h"
15 * Directory operations: readdir, lookup, create, link, unlink,
20 * Ceph MDS operations are specified in terms of a base ino and
21 * relative path. Thus, the client can specify an operation on a
22 * specific inode (e.g., a getattr due to fstat(2)), or as a path
23 * relative to, say, the root directory.
25 * Normally, we limit ourselves to strict inode ops (no path component)
26 * or dentry operations (a single path component relative to an ino). The
27 * exception to this is open_root_dentry(), which will open the mount
31 const struct dentry_operations ceph_dentry_ops
;
34 * Initialize ceph dentry state.
36 static int ceph_d_init(struct dentry
*dentry
)
38 struct ceph_dentry_info
*di
;
40 di
= kmem_cache_zalloc(ceph_dentry_cachep
, GFP_KERNEL
);
42 return -ENOMEM
; /* oh well */
45 di
->lease_session
= NULL
;
47 dentry
->d_fsdata
= di
;
48 ceph_dentry_lru_add(dentry
);
53 * for f_pos for readdir:
55 * (0xff << 52) | ((24 bits hash) << 28) |
56 * (the nth entry has hash collision);
58 * ((frag value) << 28) | (the nth entry in frag);
60 #define OFFSET_BITS 28
61 #define OFFSET_MASK ((1 << OFFSET_BITS) - 1)
62 #define HASH_ORDER (0xffull << (OFFSET_BITS + 24))
63 loff_t
ceph_make_fpos(unsigned high
, unsigned off
, bool hash_order
)
65 loff_t fpos
= ((loff_t
)high
<< 28) | (loff_t
)off
;
71 static bool is_hash_order(loff_t p
)
73 return (p
& HASH_ORDER
) == HASH_ORDER
;
76 static unsigned fpos_frag(loff_t p
)
78 return p
>> OFFSET_BITS
;
81 static unsigned fpos_hash(loff_t p
)
83 return ceph_frag_value(fpos_frag(p
));
86 static unsigned fpos_off(loff_t p
)
88 return p
& OFFSET_MASK
;
91 static int fpos_cmp(loff_t l
, loff_t r
)
93 int v
= ceph_frag_compare(fpos_frag(l
), fpos_frag(r
));
96 return (int)(fpos_off(l
) - fpos_off(r
));
100 * make note of the last dentry we read, so we can
101 * continue at the same lexicographical point,
102 * regardless of what dir changes take place on the
105 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
106 int len
, unsigned next_offset
)
108 char *buf
= kmalloc(len
+1, GFP_KERNEL
);
111 kfree(fi
->last_name
);
113 memcpy(fi
->last_name
, name
, len
);
114 fi
->last_name
[len
] = 0;
115 fi
->next_offset
= next_offset
;
116 dout("note_last_dentry '%s'\n", fi
->last_name
);
121 static struct dentry
*
122 __dcache_find_get_entry(struct dentry
*parent
, u64 idx
,
123 struct ceph_readdir_cache_control
*cache_ctl
)
125 struct inode
*dir
= d_inode(parent
);
126 struct dentry
*dentry
;
127 unsigned idx_mask
= (PAGE_SIZE
/ sizeof(struct dentry
*)) - 1;
128 loff_t ptr_pos
= idx
* sizeof(struct dentry
*);
129 pgoff_t ptr_pgoff
= ptr_pos
>> PAGE_SHIFT
;
131 if (ptr_pos
>= i_size_read(dir
))
134 if (!cache_ctl
->page
|| ptr_pgoff
!= page_index(cache_ctl
->page
)) {
135 ceph_readdir_cache_release(cache_ctl
);
136 cache_ctl
->page
= find_lock_page(&dir
->i_data
, ptr_pgoff
);
137 if (!cache_ctl
->page
) {
138 dout(" page %lu not found\n", ptr_pgoff
);
139 return ERR_PTR(-EAGAIN
);
141 /* reading/filling the cache are serialized by
142 i_mutex, no need to use page lock */
143 unlock_page(cache_ctl
->page
);
144 cache_ctl
->dentries
= kmap(cache_ctl
->page
);
147 cache_ctl
->index
= idx
& idx_mask
;
150 spin_lock(&parent
->d_lock
);
151 /* check i_size again here, because empty directory can be
152 * marked as complete while not holding the i_mutex. */
153 if (ceph_dir_is_complete_ordered(dir
) && ptr_pos
< i_size_read(dir
))
154 dentry
= cache_ctl
->dentries
[cache_ctl
->index
];
157 spin_unlock(&parent
->d_lock
);
158 if (dentry
&& !lockref_get_not_dead(&dentry
->d_lockref
))
161 return dentry
? : ERR_PTR(-EAGAIN
);
165 * When possible, we try to satisfy a readdir by peeking at the
166 * dcache. We make this work by carefully ordering dentries on
167 * d_child when we initially get results back from the MDS, and
168 * falling back to a "normal" sync readdir if any dentries in the dir
171 * Complete dir indicates that we have all dentries in the dir. It is
172 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
173 * the MDS if/when the directory is modified).
175 static int __dcache_readdir(struct file
*file
, struct dir_context
*ctx
,
178 struct ceph_file_info
*fi
= file
->private_data
;
179 struct dentry
*parent
= file
->f_path
.dentry
;
180 struct inode
*dir
= d_inode(parent
);
181 struct dentry
*dentry
, *last
= NULL
;
182 struct ceph_dentry_info
*di
;
183 struct ceph_readdir_cache_control cache_ctl
= {};
187 dout("__dcache_readdir %p v%u at %llx\n", dir
, (unsigned)shared_gen
, ctx
->pos
);
189 /* search start position */
191 u64 count
= div_u64(i_size_read(dir
), sizeof(struct dentry
*));
193 u64 step
= count
>> 1;
194 dentry
= __dcache_find_get_entry(parent
, idx
+ step
,
197 /* use linar search */
201 if (IS_ERR(dentry
)) {
202 err
= PTR_ERR(dentry
);
205 di
= ceph_dentry(dentry
);
206 spin_lock(&dentry
->d_lock
);
207 if (fpos_cmp(di
->offset
, ctx
->pos
) < 0) {
213 spin_unlock(&dentry
->d_lock
);
217 dout("__dcache_readdir %p cache idx %llu\n", dir
, idx
);
222 bool emit_dentry
= false;
223 dentry
= __dcache_find_get_entry(parent
, idx
++, &cache_ctl
);
225 fi
->flags
|= CEPH_F_ATEND
;
229 if (IS_ERR(dentry
)) {
230 err
= PTR_ERR(dentry
);
234 spin_lock(&dentry
->d_lock
);
235 di
= ceph_dentry(dentry
);
236 if (d_unhashed(dentry
) ||
237 d_really_is_negative(dentry
) ||
238 di
->lease_shared_gen
!= shared_gen
) {
239 spin_unlock(&dentry
->d_lock
);
244 if (fpos_cmp(ctx
->pos
, di
->offset
) <= 0) {
247 spin_unlock(&dentry
->d_lock
);
250 dout(" %llx dentry %p %pd %p\n", di
->offset
,
251 dentry
, dentry
, d_inode(dentry
));
252 ctx
->pos
= di
->offset
;
253 if (!dir_emit(ctx
, dentry
->d_name
.name
,
255 ceph_translate_ino(dentry
->d_sb
,
256 d_inode(dentry
)->i_ino
),
257 d_inode(dentry
)->i_mode
>> 12)) {
272 ceph_readdir_cache_release(&cache_ctl
);
275 di
= ceph_dentry(last
);
276 ret
= note_last_dentry(fi
, last
->d_name
.name
, last
->d_name
.len
,
277 fpos_off(di
->offset
) + 1);
281 /* last_name no longer match cache index */
282 if (fi
->readdir_cache_idx
>= 0) {
283 fi
->readdir_cache_idx
= -1;
284 fi
->dir_release_count
= 0;
290 static bool need_send_readdir(struct ceph_file_info
*fi
, loff_t pos
)
292 if (!fi
->last_readdir
)
294 if (is_hash_order(pos
))
295 return !ceph_frag_contains_value(fi
->frag
, fpos_hash(pos
));
297 return fi
->frag
!= fpos_frag(pos
);
300 static int ceph_readdir(struct file
*file
, struct dir_context
*ctx
)
302 struct ceph_file_info
*fi
= file
->private_data
;
303 struct inode
*inode
= file_inode(file
);
304 struct ceph_inode_info
*ci
= ceph_inode(inode
);
305 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
306 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
310 struct ceph_mds_reply_info_parsed
*rinfo
;
312 dout("readdir %p file %p pos %llx\n", inode
, file
, ctx
->pos
);
313 if (fi
->flags
& CEPH_F_ATEND
)
316 /* always start with . and .. */
318 dout("readdir off 0 -> '.'\n");
319 if (!dir_emit(ctx
, ".", 1,
320 ceph_translate_ino(inode
->i_sb
, inode
->i_ino
),
321 inode
->i_mode
>> 12))
326 ino_t ino
= parent_ino(file
->f_path
.dentry
);
327 dout("readdir off 1 -> '..'\n");
328 if (!dir_emit(ctx
, "..", 2,
329 ceph_translate_ino(inode
->i_sb
, ino
),
330 inode
->i_mode
>> 12))
335 /* can we use the dcache? */
336 spin_lock(&ci
->i_ceph_lock
);
337 if (ceph_test_mount_opt(fsc
, DCACHE
) &&
338 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
339 ceph_snap(inode
) != CEPH_SNAPDIR
&&
340 __ceph_dir_is_complete_ordered(ci
) &&
341 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
342 int shared_gen
= atomic_read(&ci
->i_shared_gen
);
343 spin_unlock(&ci
->i_ceph_lock
);
344 err
= __dcache_readdir(file
, ctx
, shared_gen
);
348 spin_unlock(&ci
->i_ceph_lock
);
351 /* proceed with a normal readdir */
353 /* do we have the correct frag content buffered? */
354 if (need_send_readdir(fi
, ctx
->pos
)) {
355 struct ceph_mds_request
*req
;
356 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
357 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
359 /* discard old result, if any */
360 if (fi
->last_readdir
) {
361 ceph_mdsc_put_request(fi
->last_readdir
);
362 fi
->last_readdir
= NULL
;
365 if (is_hash_order(ctx
->pos
)) {
366 /* fragtree isn't always accurate. choose frag
367 * based on previous reply when possible. */
368 if (frag
== (unsigned)-1)
369 frag
= ceph_choose_frag(ci
, fpos_hash(ctx
->pos
),
372 frag
= fpos_frag(ctx
->pos
);
375 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
376 ceph_vinop(inode
), frag
, fi
->last_name
);
377 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
380 err
= ceph_alloc_readdir_reply_buffer(req
, inode
);
382 ceph_mdsc_put_request(req
);
385 /* hints to request -> mds selection code */
386 req
->r_direct_mode
= USE_AUTH_MDS
;
387 if (op
== CEPH_MDS_OP_READDIR
) {
388 req
->r_direct_hash
= ceph_frag_value(frag
);
389 __set_bit(CEPH_MDS_R_DIRECT_IS_HASH
, &req
->r_req_flags
);
392 req
->r_path2
= kstrdup(fi
->last_name
, GFP_KERNEL
);
394 ceph_mdsc_put_request(req
);
397 } else if (is_hash_order(ctx
->pos
)) {
398 req
->r_args
.readdir
.offset_hash
=
399 cpu_to_le32(fpos_hash(ctx
->pos
));
402 req
->r_dir_release_cnt
= fi
->dir_release_count
;
403 req
->r_dir_ordered_cnt
= fi
->dir_ordered_count
;
404 req
->r_readdir_cache_idx
= fi
->readdir_cache_idx
;
405 req
->r_readdir_offset
= fi
->next_offset
;
406 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
407 req
->r_args
.readdir
.flags
=
408 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS
);
410 req
->r_inode
= inode
;
412 req
->r_dentry
= dget(file
->f_path
.dentry
);
413 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
415 ceph_mdsc_put_request(req
);
418 dout("readdir got and parsed readdir result=%d on "
419 "frag %x, end=%d, complete=%d, hash_order=%d\n",
421 (int)req
->r_reply_info
.dir_end
,
422 (int)req
->r_reply_info
.dir_complete
,
423 (int)req
->r_reply_info
.hash_order
);
425 rinfo
= &req
->r_reply_info
;
426 if (le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
427 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
428 if (!rinfo
->hash_order
) {
429 fi
->next_offset
= req
->r_readdir_offset
;
430 /* adjust ctx->pos to beginning of frag */
431 ctx
->pos
= ceph_make_fpos(frag
,
438 fi
->last_readdir
= req
;
440 if (test_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
)) {
441 fi
->readdir_cache_idx
= req
->r_readdir_cache_idx
;
442 if (fi
->readdir_cache_idx
< 0) {
443 /* preclude from marking dir ordered */
444 fi
->dir_ordered_count
= 0;
445 } else if (ceph_frag_is_leftmost(frag
) &&
446 fi
->next_offset
== 2) {
447 /* note dir version at start of readdir so
448 * we can tell if any dentries get dropped */
449 fi
->dir_release_count
= req
->r_dir_release_cnt
;
450 fi
->dir_ordered_count
= req
->r_dir_ordered_cnt
;
453 dout("readdir !did_prepopulate");
454 /* disable readdir cache */
455 fi
->readdir_cache_idx
= -1;
456 /* preclude from marking dir complete */
457 fi
->dir_release_count
= 0;
460 /* note next offset and last dentry name */
461 if (rinfo
->dir_nr
> 0) {
462 struct ceph_mds_reply_dir_entry
*rde
=
463 rinfo
->dir_entries
+ (rinfo
->dir_nr
-1);
464 unsigned next_offset
= req
->r_reply_info
.dir_end
?
465 2 : (fpos_off(rde
->offset
) + 1);
466 err
= note_last_dentry(fi
, rde
->name
, rde
->name_len
,
470 } else if (req
->r_reply_info
.dir_end
) {
476 rinfo
= &fi
->last_readdir
->r_reply_info
;
477 dout("readdir frag %x num %d pos %llx chunk first %llx\n",
478 fi
->frag
, rinfo
->dir_nr
, ctx
->pos
,
479 rinfo
->dir_nr
? rinfo
->dir_entries
[0].offset
: 0LL);
482 /* search start position */
483 if (rinfo
->dir_nr
> 0) {
484 int step
, nr
= rinfo
->dir_nr
;
487 if (rinfo
->dir_entries
[i
+ step
].offset
< ctx
->pos
) {
495 for (; i
< rinfo
->dir_nr
; i
++) {
496 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
497 struct ceph_vino vino
;
501 BUG_ON(rde
->offset
< ctx
->pos
);
503 ctx
->pos
= rde
->offset
;
504 dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
505 i
, rinfo
->dir_nr
, ctx
->pos
,
506 rde
->name_len
, rde
->name
, &rde
->inode
.in
);
508 BUG_ON(!rde
->inode
.in
);
509 ftype
= le32_to_cpu(rde
->inode
.in
->mode
) >> 12;
510 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
511 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
512 ino
= ceph_vino_to_ino(vino
);
514 if (!dir_emit(ctx
, rde
->name
, rde
->name_len
,
515 ceph_translate_ino(inode
->i_sb
, ino
), ftype
)) {
516 dout("filldir stopping us...\n");
522 ceph_mdsc_put_request(fi
->last_readdir
);
523 fi
->last_readdir
= NULL
;
525 if (fi
->next_offset
> 2) {
531 if (!ceph_frag_is_rightmost(fi
->frag
)) {
532 frag
= ceph_frag_next(fi
->frag
);
533 if (is_hash_order(ctx
->pos
)) {
534 loff_t new_pos
= ceph_make_fpos(ceph_frag_value(frag
),
535 fi
->next_offset
, true);
536 if (new_pos
> ctx
->pos
)
540 ctx
->pos
= ceph_make_fpos(frag
, fi
->next_offset
, false);
541 kfree(fi
->last_name
);
542 fi
->last_name
= NULL
;
544 dout("readdir next frag is %x\n", frag
);
547 fi
->flags
|= CEPH_F_ATEND
;
550 * if dir_release_count still matches the dir, no dentries
551 * were released during the whole readdir, and we should have
552 * the complete dir contents in our cache.
554 if (atomic64_read(&ci
->i_release_count
) == fi
->dir_release_count
) {
555 spin_lock(&ci
->i_ceph_lock
);
556 if (fi
->dir_ordered_count
== atomic64_read(&ci
->i_ordered_count
)) {
557 dout(" marking %p complete and ordered\n", inode
);
558 /* use i_size to track number of entries in
560 BUG_ON(fi
->readdir_cache_idx
< 0);
561 i_size_write(inode
, fi
->readdir_cache_idx
*
562 sizeof(struct dentry
*));
564 dout(" marking %p complete\n", inode
);
566 __ceph_dir_set_complete(ci
, fi
->dir_release_count
,
567 fi
->dir_ordered_count
);
568 spin_unlock(&ci
->i_ceph_lock
);
571 dout("readdir %p file %p done.\n", inode
, file
);
575 static void reset_readdir(struct ceph_file_info
*fi
)
577 if (fi
->last_readdir
) {
578 ceph_mdsc_put_request(fi
->last_readdir
);
579 fi
->last_readdir
= NULL
;
581 kfree(fi
->last_name
);
582 fi
->last_name
= NULL
;
583 fi
->dir_release_count
= 0;
584 fi
->readdir_cache_idx
= -1;
585 fi
->next_offset
= 2; /* compensate for . and .. */
586 fi
->flags
&= ~CEPH_F_ATEND
;
590 * discard buffered readdir content on seekdir(0), or seek to new frag,
591 * or seek prior to current chunk
593 static bool need_reset_readdir(struct ceph_file_info
*fi
, loff_t new_pos
)
595 struct ceph_mds_reply_info_parsed
*rinfo
;
599 if (is_hash_order(new_pos
)) {
600 /* no need to reset last_name for a forward seek when
601 * dentries are sotred in hash order */
602 } else if (fi
->frag
!= fpos_frag(new_pos
)) {
605 rinfo
= fi
->last_readdir
? &fi
->last_readdir
->r_reply_info
: NULL
;
606 if (!rinfo
|| !rinfo
->dir_nr
)
608 chunk_offset
= rinfo
->dir_entries
[0].offset
;
609 return new_pos
< chunk_offset
||
610 is_hash_order(new_pos
) != is_hash_order(chunk_offset
);
613 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
615 struct ceph_file_info
*fi
= file
->private_data
;
616 struct inode
*inode
= file
->f_mapping
->host
;
623 offset
+= file
->f_pos
;
627 retval
= -EOPNOTSUPP
;
633 if (need_reset_readdir(fi
, offset
)) {
634 dout("dir_llseek dropping %p content\n", file
);
636 } else if (is_hash_order(offset
) && offset
> file
->f_pos
) {
637 /* for hash offset, we don't know if a forward seek
638 * is within same frag */
639 fi
->dir_release_count
= 0;
640 fi
->readdir_cache_idx
= -1;
643 if (offset
!= file
->f_pos
) {
644 file
->f_pos
= offset
;
646 fi
->flags
&= ~CEPH_F_ATEND
;
656 * Handle lookups for the hidden .snap directory.
658 int ceph_handle_snapdir(struct ceph_mds_request
*req
,
659 struct dentry
*dentry
, int err
)
661 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
662 struct inode
*parent
= d_inode(dentry
->d_parent
); /* we hold i_mutex */
665 if (err
== -ENOENT
&&
666 ceph_snap(parent
) == CEPH_NOSNAP
&&
667 strcmp(dentry
->d_name
.name
,
668 fsc
->mount_options
->snapdir_name
) == 0) {
669 struct inode
*inode
= ceph_get_snapdir(parent
);
670 dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
671 dentry
, dentry
, inode
);
672 BUG_ON(!d_unhashed(dentry
));
673 d_add(dentry
, inode
);
680 * Figure out final result of a lookup/open request.
682 * Mainly, make sure we return the final req->r_dentry (if it already
683 * existed) in place of the original VFS-provided dentry when they
686 * Gracefully handle the case where the MDS replies with -ENOENT and
687 * no trace (which it may do, at its discretion, e.g., if it doesn't
688 * care to issue a lease on the negative dentry).
690 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
691 struct dentry
*dentry
, int err
)
693 if (err
== -ENOENT
) {
696 if (!req
->r_reply_info
.head
->is_dentry
) {
697 dout("ENOENT and no trace, dentry %p inode %p\n",
698 dentry
, d_inode(dentry
));
699 if (d_really_is_positive(dentry
)) {
708 dentry
= ERR_PTR(err
);
709 else if (dentry
!= req
->r_dentry
)
710 dentry
= dget(req
->r_dentry
); /* we got spliced */
716 static bool is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
718 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
719 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
723 * Look up a single dir entry. If there is a lookup intent, inform
724 * the MDS so that it gets our 'caps wanted' value in a single op.
726 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
729 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
730 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
731 struct ceph_mds_request
*req
;
736 dout("lookup %p dentry %p '%pd'\n",
737 dir
, dentry
, dentry
);
739 if (dentry
->d_name
.len
> NAME_MAX
)
740 return ERR_PTR(-ENAMETOOLONG
);
742 /* can we conclude ENOENT locally? */
743 if (d_really_is_negative(dentry
)) {
744 struct ceph_inode_info
*ci
= ceph_inode(dir
);
745 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
747 spin_lock(&ci
->i_ceph_lock
);
748 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
749 if (strncmp(dentry
->d_name
.name
,
750 fsc
->mount_options
->snapdir_name
,
751 dentry
->d_name
.len
) &&
752 !is_root_ceph_dentry(dir
, dentry
) &&
753 ceph_test_mount_opt(fsc
, DCACHE
) &&
754 __ceph_dir_is_complete(ci
) &&
755 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
756 spin_unlock(&ci
->i_ceph_lock
);
757 dout(" dir %p complete, -ENOENT\n", dir
);
759 di
->lease_shared_gen
= atomic_read(&ci
->i_shared_gen
);
762 spin_unlock(&ci
->i_ceph_lock
);
765 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
766 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
767 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
769 return ERR_CAST(req
);
770 req
->r_dentry
= dget(dentry
);
773 mask
= CEPH_STAT_CAP_INODE
| CEPH_CAP_AUTH_SHARED
;
774 if (ceph_security_xattr_wanted(dir
))
775 mask
|= CEPH_CAP_XATTR_SHARED
;
776 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
779 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
780 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
781 err
= ceph_handle_snapdir(req
, dentry
, err
);
782 dentry
= ceph_finish_lookup(req
, dentry
, err
);
783 ceph_mdsc_put_request(req
); /* will dput(dentry) */
784 dout("lookup result=%p\n", dentry
);
789 * If we do a create but get no trace back from the MDS, follow up with
790 * a lookup (the VFS expects us to link up the provided dentry).
792 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
794 struct dentry
*result
= ceph_lookup(dir
, dentry
, 0);
796 if (result
&& !IS_ERR(result
)) {
798 * We created the item, then did a lookup, and found
799 * it was already linked to another inode we already
800 * had in our cache (and thus got spliced). To not
801 * confuse VFS (especially when inode is a directory),
802 * we don't link our dentry to that inode, return an
805 * This event should be rare and it happens only when
806 * we talk to old MDS. Recent MDS does not send traceless
807 * reply for request that creates new inode.
812 return PTR_ERR(result
);
815 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
816 umode_t mode
, dev_t rdev
)
818 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
819 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
820 struct ceph_mds_request
*req
;
821 struct ceph_acls_info acls
= {};
824 if (ceph_snap(dir
) != CEPH_NOSNAP
)
827 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
831 dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
832 dir
, dentry
, mode
, rdev
);
833 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
838 req
->r_dentry
= dget(dentry
);
841 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
842 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
843 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
844 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
845 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
847 req
->r_pagelist
= acls
.pagelist
;
848 acls
.pagelist
= NULL
;
850 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
851 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
852 err
= ceph_handle_notrace_create(dir
, dentry
);
853 ceph_mdsc_put_request(req
);
856 ceph_init_inode_acls(d_inode(dentry
), &acls
);
859 ceph_release_acls_info(&acls
);
863 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
866 return ceph_mknod(dir
, dentry
, mode
, 0);
869 static int ceph_symlink(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 ceph_mds_request
*req
;
877 if (ceph_snap(dir
) != CEPH_NOSNAP
)
880 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
881 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
886 req
->r_path2
= kstrdup(dest
, GFP_KERNEL
);
889 ceph_mdsc_put_request(req
);
893 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
894 req
->r_dentry
= dget(dentry
);
896 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
897 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
898 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
899 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
900 err
= ceph_handle_notrace_create(dir
, dentry
);
901 ceph_mdsc_put_request(req
);
908 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
910 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
911 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
912 struct ceph_mds_request
*req
;
913 struct ceph_acls_info acls
= {};
917 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
918 /* mkdir .snap/foo is a MKSNAP */
919 op
= CEPH_MDS_OP_MKSNAP
;
920 dout("mksnap dir %p snap '%pd' dn %p\n", dir
,
922 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
923 dout("mkdir dir %p dn %p mode 0%ho\n", dir
, dentry
, mode
);
924 op
= CEPH_MDS_OP_MKDIR
;
930 err
= ceph_pre_init_acls(dir
, &mode
, &acls
);
934 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
940 req
->r_dentry
= dget(dentry
);
943 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
944 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
945 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
946 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
948 req
->r_pagelist
= acls
.pagelist
;
949 acls
.pagelist
= NULL
;
951 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
953 !req
->r_reply_info
.head
->is_target
&&
954 !req
->r_reply_info
.head
->is_dentry
)
955 err
= ceph_handle_notrace_create(dir
, dentry
);
956 ceph_mdsc_put_request(req
);
959 ceph_init_inode_acls(d_inode(dentry
), &acls
);
962 ceph_release_acls_info(&acls
);
966 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
967 struct dentry
*dentry
)
969 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
970 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
971 struct ceph_mds_request
*req
;
974 if (ceph_snap(dir
) != CEPH_NOSNAP
)
977 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
979 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
984 req
->r_dentry
= dget(dentry
);
986 req
->r_old_dentry
= dget(old_dentry
);
988 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
989 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
990 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
991 /* release LINK_SHARED on source inode (mds will lock it) */
992 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
993 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
996 } else if (!req
->r_reply_info
.head
->is_dentry
) {
997 ihold(d_inode(old_dentry
));
998 d_instantiate(dentry
, d_inode(old_dentry
));
1000 ceph_mdsc_put_request(req
);
1005 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
1006 * looks like the link count will hit 0, drop any other caps (other
1007 * than PIN) we don't specifically want (due to the file still being
1010 static int drop_caps_for_unlink(struct inode
*inode
)
1012 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1013 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
1015 spin_lock(&ci
->i_ceph_lock
);
1016 if (inode
->i_nlink
== 1) {
1017 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
1018 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
1020 spin_unlock(&ci
->i_ceph_lock
);
1025 * rmdir and unlink are differ only by the metadata op code
1027 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
1029 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
1030 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1031 struct inode
*inode
= d_inode(dentry
);
1032 struct ceph_mds_request
*req
;
1036 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
1037 /* rmdir .snap/foo is RMSNAP */
1038 dout("rmsnap dir %p '%pd' dn %p\n", dir
, dentry
, dentry
);
1039 op
= CEPH_MDS_OP_RMSNAP
;
1040 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
1041 dout("unlink/rmdir dir %p dn %p inode %p\n",
1042 dir
, dentry
, inode
);
1043 op
= d_is_dir(dentry
) ?
1044 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
1047 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
1052 req
->r_dentry
= dget(dentry
);
1053 req
->r_num_caps
= 2;
1054 req
->r_parent
= dir
;
1055 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
1056 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
1057 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
1058 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
1059 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
1060 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
1062 ceph_mdsc_put_request(req
);
1067 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
1068 struct inode
*new_dir
, struct dentry
*new_dentry
,
1071 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
1072 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1073 struct ceph_mds_request
*req
;
1074 int op
= CEPH_MDS_OP_RENAME
;
1080 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
1082 if (ceph_snap(old_dir
) != CEPH_NOSNAP
) {
1083 if (old_dir
== new_dir
&& ceph_snap(old_dir
) == CEPH_SNAPDIR
)
1084 op
= CEPH_MDS_OP_RENAMESNAP
;
1088 dout("rename dir %p dentry %p to dir %p dentry %p\n",
1089 old_dir
, old_dentry
, new_dir
, new_dentry
);
1090 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
1092 return PTR_ERR(req
);
1094 req
->r_dentry
= dget(new_dentry
);
1095 req
->r_num_caps
= 2;
1096 req
->r_old_dentry
= dget(old_dentry
);
1097 req
->r_old_dentry_dir
= old_dir
;
1098 req
->r_parent
= new_dir
;
1099 set_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
);
1100 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
1101 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
1102 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
1103 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
1104 /* release LINK_RDCACHE on source inode (mds will lock it) */
1105 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
1106 if (d_really_is_positive(new_dentry
))
1107 req
->r_inode_drop
= drop_caps_for_unlink(d_inode(new_dentry
));
1108 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
1109 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
1111 * Normally d_move() is done by fill_trace (called by
1112 * do_request, above). If there is no trace, we need
1116 /* d_move screws up sibling dentries' offsets */
1117 ceph_dir_clear_complete(old_dir
);
1118 ceph_dir_clear_complete(new_dir
);
1120 d_move(old_dentry
, new_dentry
);
1122 /* ensure target dentry is invalidated, despite
1123 rehashing bug in vfs_rename_dir */
1124 ceph_invalidate_dentry_lease(new_dentry
);
1126 ceph_mdsc_put_request(req
);
1131 * Ensure a dentry lease will no longer revalidate.
1133 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
1135 spin_lock(&dentry
->d_lock
);
1136 ceph_dentry(dentry
)->time
= jiffies
;
1137 ceph_dentry(dentry
)->lease_shared_gen
= 0;
1138 spin_unlock(&dentry
->d_lock
);
1142 * Check if dentry lease is valid. If not, delete the lease. Try to
1143 * renew if the least is more than half up.
1145 static int dentry_lease_is_valid(struct dentry
*dentry
, unsigned int flags
,
1148 struct ceph_dentry_info
*di
;
1149 struct ceph_mds_session
*s
;
1153 struct ceph_mds_session
*session
= NULL
;
1156 spin_lock(&dentry
->d_lock
);
1157 di
= ceph_dentry(dentry
);
1158 if (di
&& di
->lease_session
) {
1159 s
= di
->lease_session
;
1160 spin_lock(&s
->s_gen_ttl_lock
);
1163 spin_unlock(&s
->s_gen_ttl_lock
);
1165 if (di
->lease_gen
== gen
&&
1166 time_before(jiffies
, di
->time
) &&
1167 time_before(jiffies
, ttl
)) {
1169 if (di
->lease_renew_after
&&
1170 time_after(jiffies
, di
->lease_renew_after
)) {
1172 * We should renew. If we're in RCU walk mode
1173 * though, we can't do that so just return
1176 if (flags
& LOOKUP_RCU
) {
1179 session
= ceph_get_mds_session(s
);
1180 seq
= di
->lease_seq
;
1181 di
->lease_renew_after
= 0;
1182 di
->lease_renew_from
= jiffies
;
1187 spin_unlock(&dentry
->d_lock
);
1190 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
1191 CEPH_MDS_LEASE_RENEW
, seq
);
1192 ceph_put_mds_session(session
);
1194 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
1199 * Check if directory-wide content lease/cap is valid.
1201 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
1203 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1204 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1207 spin_lock(&ci
->i_ceph_lock
);
1208 if (atomic_read(&ci
->i_shared_gen
) == di
->lease_shared_gen
)
1209 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
1210 spin_unlock(&ci
->i_ceph_lock
);
1211 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1212 dir
, (unsigned)atomic_read(&ci
->i_shared_gen
),
1213 dentry
, (unsigned)di
->lease_shared_gen
, valid
);
1218 * Check if cached dentry can be trusted.
1220 static int ceph_d_revalidate(struct dentry
*dentry
, unsigned int flags
)
1223 struct dentry
*parent
;
1226 if (flags
& LOOKUP_RCU
) {
1227 parent
= READ_ONCE(dentry
->d_parent
);
1228 dir
= d_inode_rcu(parent
);
1232 parent
= dget_parent(dentry
);
1233 dir
= d_inode(parent
);
1236 dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry
,
1237 dentry
, d_inode(dentry
), ceph_dentry(dentry
)->offset
);
1239 /* always trust cached snapped dentries, snapdir dentry */
1240 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
1241 dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry
,
1242 dentry
, d_inode(dentry
));
1244 } else if (d_really_is_positive(dentry
) &&
1245 ceph_snap(d_inode(dentry
)) == CEPH_SNAPDIR
) {
1248 valid
= dentry_lease_is_valid(dentry
, flags
, dir
);
1249 if (valid
== -ECHILD
)
1251 if (valid
|| dir_lease_is_valid(dir
, dentry
)) {
1252 if (d_really_is_positive(dentry
))
1253 valid
= ceph_is_any_caps(d_inode(dentry
));
1260 struct ceph_mds_client
*mdsc
=
1261 ceph_sb_to_client(dir
->i_sb
)->mdsc
;
1262 struct ceph_mds_request
*req
;
1266 if (flags
& LOOKUP_RCU
)
1269 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
1270 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
1271 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
1273 req
->r_dentry
= dget(dentry
);
1274 req
->r_num_caps
= 2;
1275 req
->r_parent
= dir
;
1277 mask
= CEPH_STAT_CAP_INODE
| CEPH_CAP_AUTH_SHARED
;
1278 if (ceph_security_xattr_wanted(dir
))
1279 mask
|= CEPH_CAP_XATTR_SHARED
;
1280 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1282 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1285 if (d_really_is_positive(dentry
) &&
1286 d_inode(dentry
) == req
->r_target_inode
)
1290 if (d_really_is_negative(dentry
))
1296 ceph_mdsc_put_request(req
);
1297 dout("d_revalidate %p lookup result=%d\n",
1302 dout("d_revalidate %p %s\n", dentry
, valid
? "valid" : "invalid");
1304 ceph_dentry_lru_touch(dentry
);
1306 ceph_dir_clear_complete(dir
);
1309 if (!(flags
& LOOKUP_RCU
))
1315 * Release our ceph_dentry_info.
1317 static void ceph_d_release(struct dentry
*dentry
)
1319 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1321 dout("d_release %p\n", dentry
);
1322 ceph_dentry_lru_del(dentry
);
1324 spin_lock(&dentry
->d_lock
);
1325 dentry
->d_fsdata
= NULL
;
1326 spin_unlock(&dentry
->d_lock
);
1328 if (di
->lease_session
)
1329 ceph_put_mds_session(di
->lease_session
);
1330 kmem_cache_free(ceph_dentry_cachep
, di
);
1334 * When the VFS prunes a dentry from the cache, we need to clear the
1335 * complete flag on the parent directory.
1337 * Called under dentry->d_lock.
1339 static void ceph_d_prune(struct dentry
*dentry
)
1341 struct ceph_inode_info
*dir_ci
;
1342 struct ceph_dentry_info
*di
;
1344 dout("ceph_d_prune %pd %p\n", dentry
, dentry
);
1346 /* do we have a valid parent? */
1347 if (IS_ROOT(dentry
))
1350 /* we hold d_lock, so d_parent is stable */
1351 dir_ci
= ceph_inode(d_inode(dentry
->d_parent
));
1352 if (dir_ci
->i_vino
.snap
== CEPH_SNAPDIR
)
1355 /* who calls d_delete() should also disable dcache readdir */
1356 if (d_really_is_negative(dentry
))
1359 /* d_fsdata does not get cleared until d_release */
1360 if (!d_unhashed(dentry
)) {
1361 __ceph_dir_clear_complete(dir_ci
);
1365 /* Disable dcache readdir just in case that someone called d_drop()
1366 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1367 * properly (dcache readdir is still enabled) */
1368 di
= ceph_dentry(dentry
);
1369 if (di
->offset
> 0 &&
1370 di
->lease_shared_gen
== atomic_read(&dir_ci
->i_shared_gen
))
1371 __ceph_dir_clear_ordered(dir_ci
);
1375 * read() on a dir. This weird interface hack only works if mounted
1376 * with '-o dirstat'.
1378 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1381 struct ceph_file_info
*cf
= file
->private_data
;
1382 struct inode
*inode
= file_inode(file
);
1383 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1385 const int bufsize
= 1024;
1387 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1390 if (!cf
->dir_info
) {
1391 cf
->dir_info
= kmalloc(bufsize
, GFP_KERNEL
);
1395 snprintf(cf
->dir_info
, bufsize
,
1398 " subdirs: %20lld\n"
1399 "rentries: %20lld\n"
1401 " rsubdirs: %20lld\n"
1403 "rctime: %10ld.%09ld\n",
1404 ci
->i_files
+ ci
->i_subdirs
,
1407 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1411 (long)ci
->i_rctime
.tv_sec
,
1412 (long)ci
->i_rctime
.tv_nsec
);
1415 if (*ppos
>= cf
->dir_info_len
)
1417 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1418 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1421 *ppos
+= (size
- left
);
1426 * We maintain a private dentry LRU.
1428 * FIXME: this needs to be changed to a per-mds lru to be useful.
1430 void ceph_dentry_lru_add(struct dentry
*dn
)
1432 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1433 struct ceph_mds_client
*mdsc
;
1435 dout("dentry_lru_add %p %p '%pd'\n", di
, dn
, dn
);
1436 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1437 spin_lock(&mdsc
->dentry_lru_lock
);
1438 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1440 spin_unlock(&mdsc
->dentry_lru_lock
);
1443 void ceph_dentry_lru_touch(struct dentry
*dn
)
1445 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1446 struct ceph_mds_client
*mdsc
;
1448 dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di
, dn
, dn
,
1450 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1451 spin_lock(&mdsc
->dentry_lru_lock
);
1452 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1453 spin_unlock(&mdsc
->dentry_lru_lock
);
1456 void ceph_dentry_lru_del(struct dentry
*dn
)
1458 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1459 struct ceph_mds_client
*mdsc
;
1461 dout("dentry_lru_del %p %p '%pd'\n", di
, dn
, dn
);
1462 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1463 spin_lock(&mdsc
->dentry_lru_lock
);
1464 list_del_init(&di
->lru
);
1466 spin_unlock(&mdsc
->dentry_lru_lock
);
1470 * Return name hash for a given dentry. This is dependent on
1471 * the parent directory's hash function.
1473 unsigned ceph_dentry_hash(struct inode
*dir
, struct dentry
*dn
)
1475 struct ceph_inode_info
*dci
= ceph_inode(dir
);
1477 switch (dci
->i_dir_layout
.dl_dir_hash
) {
1478 case 0: /* for backward compat */
1479 case CEPH_STR_HASH_LINUX
:
1480 return dn
->d_name
.hash
;
1483 return ceph_str_hash(dci
->i_dir_layout
.dl_dir_hash
,
1484 dn
->d_name
.name
, dn
->d_name
.len
);
1488 const struct file_operations ceph_dir_fops
= {
1489 .read
= ceph_read_dir
,
1490 .iterate
= ceph_readdir
,
1491 .llseek
= ceph_dir_llseek
,
1493 .release
= ceph_release
,
1494 .unlocked_ioctl
= ceph_ioctl
,
1495 .fsync
= ceph_fsync
,
1498 const struct file_operations ceph_snapdir_fops
= {
1499 .iterate
= ceph_readdir
,
1500 .llseek
= ceph_dir_llseek
,
1502 .release
= ceph_release
,
1505 const struct inode_operations ceph_dir_iops
= {
1506 .lookup
= ceph_lookup
,
1507 .permission
= ceph_permission
,
1508 .getattr
= ceph_getattr
,
1509 .setattr
= ceph_setattr
,
1510 .listxattr
= ceph_listxattr
,
1511 .get_acl
= ceph_get_acl
,
1512 .set_acl
= ceph_set_acl
,
1513 .mknod
= ceph_mknod
,
1514 .symlink
= ceph_symlink
,
1515 .mkdir
= ceph_mkdir
,
1517 .unlink
= ceph_unlink
,
1518 .rmdir
= ceph_unlink
,
1519 .rename
= ceph_rename
,
1520 .create
= ceph_create
,
1521 .atomic_open
= ceph_atomic_open
,
1524 const struct inode_operations ceph_snapdir_iops
= {
1525 .lookup
= ceph_lookup
,
1526 .permission
= ceph_permission
,
1527 .getattr
= ceph_getattr
,
1528 .mkdir
= ceph_mkdir
,
1529 .rmdir
= ceph_unlink
,
1530 .rename
= ceph_rename
,
1533 const struct dentry_operations ceph_dentry_ops
= {
1534 .d_revalidate
= ceph_d_revalidate
,
1535 .d_release
= ceph_d_release
,
1536 .d_prune
= ceph_d_prune
,
1537 .d_init
= ceph_d_init
,