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 if (dentry
->d_parent
== NULL
|| /* nfs fh_to_dentry */
44 ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_NOSNAP
)
45 d_set_d_op(dentry
, &ceph_dentry_ops
);
46 else if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_SNAPDIR
)
47 d_set_d_op(dentry
, &ceph_snapdir_dentry_ops
);
49 d_set_d_op(dentry
, &ceph_snap_dentry_ops
);
51 di
= kmem_cache_alloc(ceph_dentry_cachep
, GFP_NOFS
| __GFP_ZERO
);
53 return -ENOMEM
; /* oh well */
55 spin_lock(&dentry
->d_lock
);
56 if (dentry
->d_fsdata
) {
58 kmem_cache_free(ceph_dentry_cachep
, di
);
62 di
->lease_session
= NULL
;
63 di
->parent_inode
= igrab(dentry
->d_parent
->d_inode
);
64 dentry
->d_fsdata
= di
;
65 dentry
->d_time
= jiffies
;
66 ceph_dentry_lru_add(dentry
);
68 spin_unlock(&dentry
->d_lock
);
75 * for readdir, we encode the directory frag and offset within that
78 static unsigned fpos_frag(loff_t p
)
82 static unsigned fpos_off(loff_t p
)
84 return p
& 0xffffffff;
88 * When possible, we try to satisfy a readdir by peeking at the
89 * dcache. We make this work by carefully ordering dentries on
90 * d_u.d_child when we initially get results back from the MDS, and
91 * falling back to a "normal" sync readdir if any dentries in the dir
94 * I_COMPLETE tells indicates we have all dentries in the dir. It is
95 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
96 * the MDS if/when the directory is modified).
98 static int __dcache_readdir(struct file
*filp
,
99 void *dirent
, filldir_t filldir
)
101 struct ceph_file_info
*fi
= filp
->private_data
;
102 struct dentry
*parent
= filp
->f_dentry
;
103 struct inode
*dir
= parent
->d_inode
;
105 struct dentry
*dentry
, *last
;
106 struct ceph_dentry_info
*di
;
109 /* claim ref on last dentry we returned */
113 dout("__dcache_readdir %p at %llu (last %p)\n", dir
, filp
->f_pos
,
116 spin_lock(&parent
->d_lock
);
118 /* start at beginning? */
119 if (filp
->f_pos
== 2 || last
== NULL
||
120 filp
->f_pos
< ceph_dentry(last
)->offset
) {
121 if (list_empty(&parent
->d_subdirs
))
123 p
= parent
->d_subdirs
.prev
;
124 dout(" initial p %p/%p\n", p
->prev
, p
->next
);
126 p
= last
->d_u
.d_child
.prev
;
130 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
131 di
= ceph_dentry(dentry
);
133 dout(" p %p/%p %s d_subdirs %p/%p\n", p
->prev
, p
->next
,
134 d_unhashed(dentry
) ? "!hashed" : "hashed",
135 parent
->d_subdirs
.prev
, parent
->d_subdirs
.next
);
136 if (p
== &parent
->d_subdirs
) {
140 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
141 if (!d_unhashed(dentry
) && dentry
->d_inode
&&
142 ceph_snap(dentry
->d_inode
) != CEPH_SNAPDIR
&&
143 ceph_ino(dentry
->d_inode
) != CEPH_INO_CEPH
&&
144 filp
->f_pos
<= di
->offset
)
146 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry
,
147 dentry
->d_name
.len
, dentry
->d_name
.name
, di
->offset
,
148 filp
->f_pos
, d_unhashed(dentry
) ? " unhashed" : "",
149 !dentry
->d_inode
? " null" : "");
150 spin_unlock(&dentry
->d_lock
);
152 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
153 di
= ceph_dentry(dentry
);
157 spin_unlock(&dentry
->d_lock
);
158 spin_unlock(&parent
->d_lock
);
160 dout(" %llu (%llu) dentry %p %.*s %p\n", di
->offset
, filp
->f_pos
,
161 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
162 filp
->f_pos
= di
->offset
;
163 err
= filldir(dirent
, dentry
->d_name
.name
,
164 dentry
->d_name
.len
, di
->offset
,
165 dentry
->d_inode
->i_ino
,
166 dentry
->d_inode
->i_mode
>> 12);
170 /* remember our position */
172 fi
->next_offset
= di
->offset
;
184 /* make sure a dentry wasn't dropped while we didn't have parent lock */
185 if (!ceph_i_test(dir
, CEPH_I_COMPLETE
)) {
186 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir
);
191 spin_lock(&parent
->d_lock
);
192 p
= p
->prev
; /* advance to next dentry */
196 spin_unlock(&parent
->d_lock
);
204 * make note of the last dentry we read, so we can
205 * continue at the same lexicographical point,
206 * regardless of what dir changes take place on the
209 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
212 kfree(fi
->last_name
);
213 fi
->last_name
= kmalloc(len
+1, GFP_NOFS
);
216 memcpy(fi
->last_name
, name
, len
);
217 fi
->last_name
[len
] = 0;
218 dout("note_last_dentry '%s'\n", fi
->last_name
);
222 static int ceph_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
224 struct ceph_file_info
*fi
= filp
->private_data
;
225 struct inode
*inode
= filp
->f_dentry
->d_inode
;
226 struct ceph_inode_info
*ci
= ceph_inode(inode
);
227 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
228 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
229 unsigned frag
= fpos_frag(filp
->f_pos
);
230 int off
= fpos_off(filp
->f_pos
);
233 struct ceph_mds_reply_info_parsed
*rinfo
;
234 const int max_entries
= fsc
->mount_options
->max_readdir
;
235 const int max_bytes
= fsc
->mount_options
->max_readdir_bytes
;
237 dout("readdir %p filp %p frag %u off %u\n", inode
, filp
, frag
, off
);
241 /* always start with . and .. */
242 if (filp
->f_pos
== 0) {
243 /* note dir version at start of readdir so we can tell
244 * if any dentries get dropped */
245 fi
->dir_release_count
= ci
->i_release_count
;
247 dout("readdir off 0 -> '.'\n");
248 if (filldir(dirent
, ".", 1, ceph_make_fpos(0, 0),
249 inode
->i_ino
, inode
->i_mode
>> 12) < 0)
254 if (filp
->f_pos
== 1) {
255 dout("readdir off 1 -> '..'\n");
256 if (filldir(dirent
, "..", 2, ceph_make_fpos(0, 1),
257 filp
->f_dentry
->d_parent
->d_inode
->i_ino
,
258 inode
->i_mode
>> 12) < 0)
264 /* can we use the dcache? */
265 spin_lock(&inode
->i_lock
);
266 if ((filp
->f_pos
== 2 || fi
->dentry
) &&
267 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
268 ceph_snap(inode
) != CEPH_SNAPDIR
&&
269 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) &&
270 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
271 spin_unlock(&inode
->i_lock
);
272 err
= __dcache_readdir(filp
, dirent
, filldir
);
276 spin_unlock(&inode
->i_lock
);
279 err
= note_last_dentry(fi
, fi
->dentry
->d_name
.name
,
280 fi
->dentry
->d_name
.len
);
287 /* proceed with a normal readdir */
290 /* do we have the correct frag content buffered? */
291 if (fi
->frag
!= frag
|| fi
->last_readdir
== NULL
) {
292 struct ceph_mds_request
*req
;
293 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
294 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
296 /* discard old result, if any */
297 if (fi
->last_readdir
) {
298 ceph_mdsc_put_request(fi
->last_readdir
);
299 fi
->last_readdir
= NULL
;
302 /* requery frag tree, as the frag topology may have changed */
303 frag
= ceph_choose_frag(ceph_inode(inode
), frag
, NULL
, NULL
);
305 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
306 ceph_vinop(inode
), frag
, fi
->last_name
);
307 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
310 req
->r_inode
= igrab(inode
);
311 req
->r_dentry
= dget(filp
->f_dentry
);
312 /* hints to request -> mds selection code */
313 req
->r_direct_mode
= USE_AUTH_MDS
;
314 req
->r_direct_hash
= ceph_frag_value(frag
);
315 req
->r_direct_is_hash
= true;
316 req
->r_path2
= kstrdup(fi
->last_name
, GFP_NOFS
);
317 req
->r_readdir_offset
= fi
->next_offset
;
318 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
319 req
->r_args
.readdir
.max_entries
= cpu_to_le32(max_entries
);
320 req
->r_args
.readdir
.max_bytes
= cpu_to_le32(max_bytes
);
321 req
->r_num_caps
= max_entries
+ 1;
322 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
324 ceph_mdsc_put_request(req
);
327 dout("readdir got and parsed readdir result=%d"
328 " on frag %x, end=%d, complete=%d\n", err
, frag
,
329 (int)req
->r_reply_info
.dir_end
,
330 (int)req
->r_reply_info
.dir_complete
);
332 if (!req
->r_did_prepopulate
) {
333 dout("readdir !did_prepopulate");
334 fi
->dir_release_count
--; /* preclude I_COMPLETE */
337 /* note next offset and last dentry name */
338 fi
->offset
= fi
->next_offset
;
339 fi
->last_readdir
= req
;
341 if (req
->r_reply_info
.dir_end
) {
342 kfree(fi
->last_name
);
343 fi
->last_name
= NULL
;
344 if (ceph_frag_is_rightmost(frag
))
349 rinfo
= &req
->r_reply_info
;
350 err
= note_last_dentry(fi
,
351 rinfo
->dir_dname
[rinfo
->dir_nr
-1],
352 rinfo
->dir_dname_len
[rinfo
->dir_nr
-1]);
355 fi
->next_offset
+= rinfo
->dir_nr
;
359 rinfo
= &fi
->last_readdir
->r_reply_info
;
360 dout("readdir frag %x num %d off %d chunkoff %d\n", frag
,
361 rinfo
->dir_nr
, off
, fi
->offset
);
362 while (off
- fi
->offset
>= 0 && off
- fi
->offset
< rinfo
->dir_nr
) {
363 u64 pos
= ceph_make_fpos(frag
, off
);
364 struct ceph_mds_reply_inode
*in
=
365 rinfo
->dir_in
[off
- fi
->offset
].in
;
366 struct ceph_vino vino
;
369 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
370 off
, off
- fi
->offset
, rinfo
->dir_nr
, pos
,
371 rinfo
->dir_dname_len
[off
- fi
->offset
],
372 rinfo
->dir_dname
[off
- fi
->offset
], in
);
374 ftype
= le32_to_cpu(in
->mode
) >> 12;
375 vino
.ino
= le64_to_cpu(in
->ino
);
376 vino
.snap
= le64_to_cpu(in
->snapid
);
377 ino
= ceph_vino_to_ino(vino
);
379 rinfo
->dir_dname
[off
- fi
->offset
],
380 rinfo
->dir_dname_len
[off
- fi
->offset
],
381 pos
, ino
, ftype
) < 0) {
382 dout("filldir stopping us...\n");
386 filp
->f_pos
= pos
+ 1;
390 ceph_mdsc_put_request(fi
->last_readdir
);
391 fi
->last_readdir
= NULL
;
396 if (!ceph_frag_is_rightmost(frag
)) {
397 frag
= ceph_frag_next(frag
);
399 filp
->f_pos
= ceph_make_fpos(frag
, off
);
400 dout("readdir next frag is %x\n", frag
);
406 * if dir_release_count still matches the dir, no dentries
407 * were released during the whole readdir, and we should have
408 * the complete dir contents in our cache.
410 spin_lock(&inode
->i_lock
);
411 if (ci
->i_release_count
== fi
->dir_release_count
) {
412 dout(" marking %p complete\n", inode
);
413 ci
->i_ceph_flags
|= CEPH_I_COMPLETE
;
414 ci
->i_max_offset
= filp
->f_pos
;
416 spin_unlock(&inode
->i_lock
);
418 dout("readdir %p filp %p done.\n", inode
, filp
);
422 static void reset_readdir(struct ceph_file_info
*fi
)
424 if (fi
->last_readdir
) {
425 ceph_mdsc_put_request(fi
->last_readdir
);
426 fi
->last_readdir
= NULL
;
428 kfree(fi
->last_name
);
429 fi
->last_name
= NULL
;
430 fi
->next_offset
= 2; /* compensate for . and .. */
438 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int origin
)
440 struct ceph_file_info
*fi
= file
->private_data
;
441 struct inode
*inode
= file
->f_mapping
->host
;
442 loff_t old_offset
= offset
;
445 mutex_lock(&inode
->i_mutex
);
448 offset
+= inode
->i_size
+ 2; /* FIXME */
451 offset
+= file
->f_pos
;
454 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
455 if (offset
!= file
->f_pos
) {
456 file
->f_pos
= offset
;
463 * discard buffered readdir content on seekdir(0), or
464 * seek to new frag, or seek prior to current chunk.
467 fpos_frag(offset
) != fpos_frag(old_offset
) ||
468 fpos_off(offset
) < fi
->offset
) {
469 dout("dir_llseek dropping %p content\n", file
);
473 /* bump dir_release_count if we did a forward seek */
474 if (offset
> old_offset
)
475 fi
->dir_release_count
--;
477 mutex_unlock(&inode
->i_mutex
);
482 * Process result of a lookup/open request.
484 * Mainly, make sure we return the final req->r_dentry (if it already
485 * existed) in place of the original VFS-provided dentry when they
488 * Gracefully handle the case where the MDS replies with -ENOENT and
489 * no trace (which it may do, at its discretion, e.g., if it doesn't
490 * care to issue a lease on the negative dentry).
492 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
493 struct dentry
*dentry
, int err
)
495 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
496 struct inode
*parent
= dentry
->d_parent
->d_inode
;
499 if (err
== -ENOENT
&&
500 strcmp(dentry
->d_name
.name
,
501 fsc
->mount_options
->snapdir_name
) == 0) {
502 struct inode
*inode
= ceph_get_snapdir(parent
);
503 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
504 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, inode
);
505 BUG_ON(!d_unhashed(dentry
));
506 d_add(dentry
, inode
);
510 if (err
== -ENOENT
) {
513 if (!req
->r_reply_info
.head
->is_dentry
) {
514 dout("ENOENT and no trace, dentry %p inode %p\n",
515 dentry
, dentry
->d_inode
);
516 if (dentry
->d_inode
) {
525 dentry
= ERR_PTR(err
);
526 else if (dentry
!= req
->r_dentry
)
527 dentry
= dget(req
->r_dentry
); /* we got spliced */
533 static int is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
535 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
536 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
540 * Look up a single dir entry. If there is a lookup intent, inform
541 * the MDS so that it gets our 'caps wanted' value in a single op.
543 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
544 struct nameidata
*nd
)
546 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
547 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
548 struct ceph_mds_request
*req
;
552 dout("lookup %p dentry %p '%.*s'\n",
553 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
555 if (dentry
->d_name
.len
> NAME_MAX
)
556 return ERR_PTR(-ENAMETOOLONG
);
558 err
= ceph_init_dentry(dentry
);
562 /* open (but not create!) intent? */
564 (nd
->flags
& LOOKUP_OPEN
) &&
565 (nd
->flags
& LOOKUP_CONTINUE
) == 0 && /* only open last component */
566 !(nd
->intent
.open
.flags
& O_CREAT
)) {
567 int mode
= nd
->intent
.open
.create_mode
& ~current
->fs
->umask
;
568 return ceph_lookup_open(dir
, dentry
, nd
, mode
, 1);
571 /* can we conclude ENOENT locally? */
572 if (dentry
->d_inode
== NULL
) {
573 struct ceph_inode_info
*ci
= ceph_inode(dir
);
574 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
576 spin_lock(&dir
->i_lock
);
577 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
578 if (strncmp(dentry
->d_name
.name
,
579 fsc
->mount_options
->snapdir_name
,
580 dentry
->d_name
.len
) &&
581 !is_root_ceph_dentry(dir
, dentry
) &&
582 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) &&
583 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
584 spin_unlock(&dir
->i_lock
);
585 dout(" dir %p complete, -ENOENT\n", dir
);
587 di
->lease_shared_gen
= ci
->i_shared_gen
;
590 spin_unlock(&dir
->i_lock
);
593 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
594 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
595 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
597 return ERR_CAST(req
);
598 req
->r_dentry
= dget(dentry
);
600 /* we only need inode linkage */
601 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
602 req
->r_locked_dir
= dir
;
603 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
604 dentry
= ceph_finish_lookup(req
, dentry
, err
);
605 ceph_mdsc_put_request(req
); /* will dput(dentry) */
606 dout("lookup result=%p\n", dentry
);
611 * If we do a create but get no trace back from the MDS, follow up with
612 * a lookup (the VFS expects us to link up the provided dentry).
614 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
616 struct dentry
*result
= ceph_lookup(dir
, dentry
, NULL
);
618 if (result
&& !IS_ERR(result
)) {
620 * We created the item, then did a lookup, and found
621 * it was already linked to another inode we already
622 * had in our cache (and thus got spliced). Link our
623 * dentry to that inode, but don't hash it, just in
624 * case the VFS wants to dereference it.
626 BUG_ON(!result
->d_inode
);
627 d_instantiate(dentry
, result
->d_inode
);
630 return PTR_ERR(result
);
633 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
634 int mode
, dev_t rdev
)
636 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
637 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
638 struct ceph_mds_request
*req
;
641 if (ceph_snap(dir
) != CEPH_NOSNAP
)
644 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
645 dir
, dentry
, mode
, rdev
);
646 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
651 req
->r_dentry
= dget(dentry
);
653 req
->r_locked_dir
= dir
;
654 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
655 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
656 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
657 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
658 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
659 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
660 err
= ceph_handle_notrace_create(dir
, dentry
);
661 ceph_mdsc_put_request(req
);
667 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
668 struct nameidata
*nd
)
670 dout("create in dir %p dentry %p name '%.*s'\n",
671 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
673 if (ceph_snap(dir
) != CEPH_NOSNAP
)
677 BUG_ON((nd
->flags
& LOOKUP_OPEN
) == 0);
678 dentry
= ceph_lookup_open(dir
, dentry
, nd
, mode
, 0);
679 /* hrm, what should i do here if we get aliased? */
681 return PTR_ERR(dentry
);
685 /* fall back to mknod */
686 return ceph_mknod(dir
, dentry
, (mode
& ~S_IFMT
) | S_IFREG
, 0);
689 static int ceph_symlink(struct inode
*dir
, struct dentry
*dentry
,
692 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
693 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
694 struct ceph_mds_request
*req
;
697 if (ceph_snap(dir
) != CEPH_NOSNAP
)
700 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
701 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
706 req
->r_dentry
= dget(dentry
);
708 req
->r_path2
= kstrdup(dest
, GFP_NOFS
);
709 req
->r_locked_dir
= dir
;
710 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
711 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
712 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
713 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
714 err
= ceph_handle_notrace_create(dir
, dentry
);
715 ceph_mdsc_put_request(req
);
721 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
723 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
724 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
725 struct ceph_mds_request
*req
;
729 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
730 /* mkdir .snap/foo is a MKSNAP */
731 op
= CEPH_MDS_OP_MKSNAP
;
732 dout("mksnap dir %p snap '%.*s' dn %p\n", dir
,
733 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
);
734 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
735 dout("mkdir dir %p dn %p mode 0%o\n", dir
, dentry
, mode
);
736 op
= CEPH_MDS_OP_MKDIR
;
740 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
746 req
->r_dentry
= dget(dentry
);
748 req
->r_locked_dir
= dir
;
749 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
750 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
751 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
752 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
753 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
754 err
= ceph_handle_notrace_create(dir
, dentry
);
755 ceph_mdsc_put_request(req
);
762 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
763 struct dentry
*dentry
)
765 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
766 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
767 struct ceph_mds_request
*req
;
770 if (ceph_snap(dir
) != CEPH_NOSNAP
)
773 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
775 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
780 req
->r_dentry
= dget(dentry
);
782 req
->r_old_dentry
= dget(old_dentry
); /* or inode? hrm. */
783 req
->r_locked_dir
= dir
;
784 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
785 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
786 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
789 else if (!req
->r_reply_info
.head
->is_dentry
)
790 d_instantiate(dentry
, igrab(old_dentry
->d_inode
));
791 ceph_mdsc_put_request(req
);
796 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
797 * looks like the link count will hit 0, drop any other caps (other
798 * than PIN) we don't specifically want (due to the file still being
801 static int drop_caps_for_unlink(struct inode
*inode
)
803 struct ceph_inode_info
*ci
= ceph_inode(inode
);
804 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
806 spin_lock(&inode
->i_lock
);
807 if (inode
->i_nlink
== 1) {
808 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
809 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
811 spin_unlock(&inode
->i_lock
);
816 * rmdir and unlink are differ only by the metadata op code
818 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
820 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
821 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
822 struct inode
*inode
= dentry
->d_inode
;
823 struct ceph_mds_request
*req
;
827 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
828 /* rmdir .snap/foo is RMSNAP */
829 dout("rmsnap dir %p '%.*s' dn %p\n", dir
, dentry
->d_name
.len
,
830 dentry
->d_name
.name
, dentry
);
831 op
= CEPH_MDS_OP_RMSNAP
;
832 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
833 dout("unlink/rmdir dir %p dn %p inode %p\n",
835 op
= ((dentry
->d_inode
->i_mode
& S_IFMT
) == S_IFDIR
) ?
836 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
839 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
844 req
->r_dentry
= dget(dentry
);
846 req
->r_locked_dir
= dir
;
847 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
848 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
849 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
850 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
851 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
853 ceph_mdsc_put_request(req
);
858 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
859 struct inode
*new_dir
, struct dentry
*new_dentry
)
861 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
862 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
863 struct ceph_mds_request
*req
;
866 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
868 if (ceph_snap(old_dir
) != CEPH_NOSNAP
||
869 ceph_snap(new_dir
) != CEPH_NOSNAP
)
871 dout("rename dir %p dentry %p to dir %p dentry %p\n",
872 old_dir
, old_dentry
, new_dir
, new_dentry
);
873 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_RENAME
, USE_AUTH_MDS
);
876 req
->r_dentry
= dget(new_dentry
);
878 req
->r_old_dentry
= dget(old_dentry
);
879 req
->r_locked_dir
= new_dir
;
880 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
881 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
882 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
883 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
884 /* release LINK_RDCACHE on source inode (mds will lock it) */
885 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
886 if (new_dentry
->d_inode
)
887 req
->r_inode_drop
= drop_caps_for_unlink(new_dentry
->d_inode
);
888 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
889 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
891 * Normally d_move() is done by fill_trace (called by
892 * do_request, above). If there is no trace, we need
896 /* d_move screws up d_subdirs order */
897 ceph_i_clear(new_dir
, CEPH_I_COMPLETE
);
899 d_move(old_dentry
, new_dentry
);
901 /* ensure target dentry is invalidated, despite
902 rehashing bug in vfs_rename_dir */
903 ceph_invalidate_dentry_lease(new_dentry
);
905 ceph_mdsc_put_request(req
);
910 * Ensure a dentry lease will no longer revalidate.
912 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
914 spin_lock(&dentry
->d_lock
);
915 dentry
->d_time
= jiffies
;
916 ceph_dentry(dentry
)->lease_shared_gen
= 0;
917 spin_unlock(&dentry
->d_lock
);
921 * Check if dentry lease is valid. If not, delete the lease. Try to
922 * renew if the least is more than half up.
924 static int dentry_lease_is_valid(struct dentry
*dentry
)
926 struct ceph_dentry_info
*di
;
927 struct ceph_mds_session
*s
;
931 struct ceph_mds_session
*session
= NULL
;
932 struct inode
*dir
= NULL
;
935 spin_lock(&dentry
->d_lock
);
936 di
= ceph_dentry(dentry
);
937 if (di
&& di
->lease_session
) {
938 s
= di
->lease_session
;
939 spin_lock(&s
->s_cap_lock
);
942 spin_unlock(&s
->s_cap_lock
);
944 if (di
->lease_gen
== gen
&&
945 time_before(jiffies
, dentry
->d_time
) &&
946 time_before(jiffies
, ttl
)) {
948 if (di
->lease_renew_after
&&
949 time_after(jiffies
, di
->lease_renew_after
)) {
950 /* we should renew */
951 dir
= dentry
->d_parent
->d_inode
;
952 session
= ceph_get_mds_session(s
);
954 di
->lease_renew_after
= 0;
955 di
->lease_renew_from
= jiffies
;
959 spin_unlock(&dentry
->d_lock
);
962 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
963 CEPH_MDS_LEASE_RENEW
, seq
);
964 ceph_put_mds_session(session
);
966 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
971 * Check if directory-wide content lease/cap is valid.
973 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
975 struct ceph_inode_info
*ci
= ceph_inode(dir
);
976 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
979 spin_lock(&dir
->i_lock
);
980 if (ci
->i_shared_gen
== di
->lease_shared_gen
)
981 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
982 spin_unlock(&dir
->i_lock
);
983 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
984 dir
, (unsigned)ci
->i_shared_gen
, dentry
,
985 (unsigned)di
->lease_shared_gen
, valid
);
990 * Check if cached dentry can be trusted.
992 static int ceph_d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
996 if (nd
->flags
& LOOKUP_RCU
)
999 dir
= dentry
->d_parent
->d_inode
;
1001 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry
,
1002 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
,
1003 ceph_dentry(dentry
)->offset
);
1005 /* always trust cached snapped dentries, snapdir dentry */
1006 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
1007 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry
,
1008 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
1011 if (dentry
->d_inode
&& ceph_snap(dentry
->d_inode
) == CEPH_SNAPDIR
)
1014 if (dentry_lease_is_valid(dentry
) ||
1015 dir_lease_is_valid(dir
, dentry
))
1018 dout("d_revalidate %p invalid\n", dentry
);
1022 ceph_dentry_lru_touch(dentry
);
1027 * When a dentry is released, clear the dir I_COMPLETE if it was part
1028 * of the current dir gen or if this is in the snapshot namespace.
1030 static void ceph_dentry_release(struct dentry
*dentry
)
1032 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1033 struct inode
*parent_inode
= NULL
;
1034 u64 snapid
= CEPH_NOSNAP
;
1036 if (!IS_ROOT(dentry
)) {
1037 parent_inode
= di
->parent_inode
;
1039 snapid
= ceph_snap(parent_inode
);
1041 dout("dentry_release %p parent %p\n", dentry
, parent_inode
);
1042 if (parent_inode
&& snapid
!= CEPH_SNAPDIR
) {
1043 struct ceph_inode_info
*ci
= ceph_inode(parent_inode
);
1045 spin_lock(&parent_inode
->i_lock
);
1046 if (ci
->i_shared_gen
== di
->lease_shared_gen
||
1047 snapid
<= CEPH_MAXSNAP
) {
1048 dout(" clearing %p complete (d_release)\n",
1050 ci
->i_ceph_flags
&= ~CEPH_I_COMPLETE
;
1051 ci
->i_release_count
++;
1053 spin_unlock(&parent_inode
->i_lock
);
1056 ceph_dentry_lru_del(dentry
);
1057 if (di
->lease_session
)
1058 ceph_put_mds_session(di
->lease_session
);
1059 kmem_cache_free(ceph_dentry_cachep
, di
);
1060 dentry
->d_fsdata
= NULL
;
1066 static int ceph_snapdir_d_revalidate(struct dentry
*dentry
,
1067 struct nameidata
*nd
)
1070 * Eventually, we'll want to revalidate snapped metadata
1071 * too... probably...
1079 * read() on a dir. This weird interface hack only works if mounted
1080 * with '-o dirstat'.
1082 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1085 struct ceph_file_info
*cf
= file
->private_data
;
1086 struct inode
*inode
= file
->f_dentry
->d_inode
;
1087 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1090 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1093 if (!cf
->dir_info
) {
1094 cf
->dir_info
= kmalloc(1024, GFP_NOFS
);
1098 sprintf(cf
->dir_info
,
1101 " subdirs: %20lld\n"
1102 "rentries: %20lld\n"
1104 " rsubdirs: %20lld\n"
1106 "rctime: %10ld.%09ld\n",
1107 ci
->i_files
+ ci
->i_subdirs
,
1110 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1114 (long)ci
->i_rctime
.tv_sec
,
1115 (long)ci
->i_rctime
.tv_nsec
);
1118 if (*ppos
>= cf
->dir_info_len
)
1120 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1121 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1124 *ppos
+= (size
- left
);
1129 * an fsync() on a dir will wait for any uncommitted directory
1130 * operations to commit.
1132 static int ceph_dir_fsync(struct file
*file
, int datasync
)
1134 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1135 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1136 struct list_head
*head
= &ci
->i_unsafe_dirops
;
1137 struct ceph_mds_request
*req
;
1141 dout("dir_fsync %p\n", inode
);
1142 spin_lock(&ci
->i_unsafe_lock
);
1143 if (list_empty(head
))
1146 req
= list_entry(head
->prev
,
1147 struct ceph_mds_request
, r_unsafe_dir_item
);
1148 last_tid
= req
->r_tid
;
1151 ceph_mdsc_get_request(req
);
1152 spin_unlock(&ci
->i_unsafe_lock
);
1153 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1154 inode
, req
->r_tid
, last_tid
);
1155 if (req
->r_timeout
) {
1156 ret
= wait_for_completion_timeout(
1157 &req
->r_safe_completion
, req
->r_timeout
);
1161 ret
= -EIO
; /* timed out */
1163 wait_for_completion(&req
->r_safe_completion
);
1165 spin_lock(&ci
->i_unsafe_lock
);
1166 ceph_mdsc_put_request(req
);
1168 if (ret
|| list_empty(head
))
1170 req
= list_entry(head
->next
,
1171 struct ceph_mds_request
, r_unsafe_dir_item
);
1172 } while (req
->r_tid
< last_tid
);
1174 spin_unlock(&ci
->i_unsafe_lock
);
1179 * We maintain a private dentry LRU.
1181 * FIXME: this needs to be changed to a per-mds lru to be useful.
1183 void ceph_dentry_lru_add(struct dentry
*dn
)
1185 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1186 struct ceph_mds_client
*mdsc
;
1188 dout("dentry_lru_add %p %p '%.*s'\n", di
, dn
,
1189 dn
->d_name
.len
, dn
->d_name
.name
);
1191 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1192 spin_lock(&mdsc
->dentry_lru_lock
);
1193 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1195 spin_unlock(&mdsc
->dentry_lru_lock
);
1199 void ceph_dentry_lru_touch(struct dentry
*dn
)
1201 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1202 struct ceph_mds_client
*mdsc
;
1204 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di
, dn
,
1205 dn
->d_name
.len
, dn
->d_name
.name
, di
->offset
);
1207 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1208 spin_lock(&mdsc
->dentry_lru_lock
);
1209 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1210 spin_unlock(&mdsc
->dentry_lru_lock
);
1214 void ceph_dentry_lru_del(struct dentry
*dn
)
1216 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1217 struct ceph_mds_client
*mdsc
;
1219 dout("dentry_lru_del %p %p '%.*s'\n", di
, dn
,
1220 dn
->d_name
.len
, dn
->d_name
.name
);
1222 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1223 spin_lock(&mdsc
->dentry_lru_lock
);
1224 list_del_init(&di
->lru
);
1226 spin_unlock(&mdsc
->dentry_lru_lock
);
1231 * Return name hash for a given dentry. This is dependent on
1232 * the parent directory's hash function.
1234 unsigned ceph_dentry_hash(struct dentry
*dn
)
1236 struct inode
*dir
= dn
->d_parent
->d_inode
;
1237 struct ceph_inode_info
*dci
= ceph_inode(dir
);
1239 switch (dci
->i_dir_layout
.dl_dir_hash
) {
1240 case 0: /* for backward compat */
1241 case CEPH_STR_HASH_LINUX
:
1242 return dn
->d_name
.hash
;
1245 return ceph_str_hash(dci
->i_dir_layout
.dl_dir_hash
,
1246 dn
->d_name
.name
, dn
->d_name
.len
);
1250 const struct file_operations ceph_dir_fops
= {
1251 .read
= ceph_read_dir
,
1252 .readdir
= ceph_readdir
,
1253 .llseek
= ceph_dir_llseek
,
1255 .release
= ceph_release
,
1256 .unlocked_ioctl
= ceph_ioctl
,
1257 .fsync
= ceph_dir_fsync
,
1260 const struct inode_operations ceph_dir_iops
= {
1261 .lookup
= ceph_lookup
,
1262 .permission
= ceph_permission
,
1263 .getattr
= ceph_getattr
,
1264 .setattr
= ceph_setattr
,
1265 .setxattr
= ceph_setxattr
,
1266 .getxattr
= ceph_getxattr
,
1267 .listxattr
= ceph_listxattr
,
1268 .removexattr
= ceph_removexattr
,
1269 .mknod
= ceph_mknod
,
1270 .symlink
= ceph_symlink
,
1271 .mkdir
= ceph_mkdir
,
1273 .unlink
= ceph_unlink
,
1274 .rmdir
= ceph_unlink
,
1275 .rename
= ceph_rename
,
1276 .create
= ceph_create
,
1279 const struct dentry_operations ceph_dentry_ops
= {
1280 .d_revalidate
= ceph_d_revalidate
,
1281 .d_release
= ceph_dentry_release
,
1284 const struct dentry_operations ceph_snapdir_dentry_ops
= {
1285 .d_revalidate
= ceph_snapdir_d_revalidate
,
1286 .d_release
= ceph_dentry_release
,
1289 const struct dentry_operations ceph_snap_dentry_ops
= {
1290 .d_release
= ceph_dentry_release
,