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 (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_NOSNAP
)
44 dentry
->d_op
= &ceph_dentry_ops
;
45 else if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_SNAPDIR
)
46 dentry
->d_op
= &ceph_snapdir_dentry_ops
;
48 dentry
->d_op
= &ceph_snap_dentry_ops
;
50 di
= kmem_cache_alloc(ceph_dentry_cachep
, GFP_NOFS
| __GFP_ZERO
);
52 return -ENOMEM
; /* oh well */
54 spin_lock(&dentry
->d_lock
);
55 if (dentry
->d_fsdata
) {
57 kmem_cache_free(ceph_dentry_cachep
, di
);
61 di
->lease_session
= NULL
;
62 dentry
->d_fsdata
= di
;
63 dentry
->d_time
= jiffies
;
64 ceph_dentry_lru_add(dentry
);
66 spin_unlock(&dentry
->d_lock
);
73 * for readdir, we encode the directory frag and offset within that
76 static unsigned fpos_frag(loff_t p
)
80 static unsigned fpos_off(loff_t p
)
82 return p
& 0xffffffff;
86 * When possible, we try to satisfy a readdir by peeking at the
87 * dcache. We make this work by carefully ordering dentries on
88 * d_u.d_child when we initially get results back from the MDS, and
89 * falling back to a "normal" sync readdir if any dentries in the dir
92 * I_COMPLETE tells indicates we have all dentries in the dir. It is
93 * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
94 * the MDS if/when the directory is modified).
96 static int __dcache_readdir(struct file
*filp
,
97 void *dirent
, filldir_t filldir
)
99 struct ceph_file_info
*fi
= filp
->private_data
;
100 struct dentry
*parent
= filp
->f_dentry
;
101 struct inode
*dir
= parent
->d_inode
;
103 struct dentry
*dentry
, *last
;
104 struct ceph_dentry_info
*di
;
107 /* claim ref on last dentry we returned */
111 dout("__dcache_readdir %p at %llu (last %p)\n", dir
, filp
->f_pos
,
114 spin_lock(&dcache_lock
);
116 /* start at beginning? */
117 if (filp
->f_pos
== 2 || (last
&&
118 filp
->f_pos
< ceph_dentry(last
)->offset
)) {
119 if (list_empty(&parent
->d_subdirs
))
121 p
= parent
->d_subdirs
.prev
;
122 dout(" initial p %p/%p\n", p
->prev
, p
->next
);
124 p
= last
->d_u
.d_child
.prev
;
128 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
129 di
= ceph_dentry(dentry
);
131 dout(" p %p/%p %s d_subdirs %p/%p\n", p
->prev
, p
->next
,
132 d_unhashed(dentry
) ? "!hashed" : "hashed",
133 parent
->d_subdirs
.prev
, parent
->d_subdirs
.next
);
134 if (p
== &parent
->d_subdirs
) {
138 if (!d_unhashed(dentry
) && dentry
->d_inode
&&
139 ceph_snap(dentry
->d_inode
) != CEPH_SNAPDIR
&&
140 ceph_ino(dentry
->d_inode
) != CEPH_INO_CEPH
&&
141 filp
->f_pos
<= di
->offset
)
143 dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry
,
144 dentry
->d_name
.len
, dentry
->d_name
.name
, di
->offset
,
145 filp
->f_pos
, d_unhashed(dentry
) ? " unhashed" : "",
146 !dentry
->d_inode
? " null" : "");
148 dentry
= list_entry(p
, struct dentry
, d_u
.d_child
);
149 di
= ceph_dentry(dentry
);
152 atomic_inc(&dentry
->d_count
);
153 spin_unlock(&dcache_lock
);
155 dout(" %llu (%llu) dentry %p %.*s %p\n", di
->offset
, filp
->f_pos
,
156 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
157 filp
->f_pos
= di
->offset
;
158 err
= filldir(dirent
, dentry
->d_name
.name
,
159 dentry
->d_name
.len
, di
->offset
,
160 dentry
->d_inode
->i_ino
,
161 dentry
->d_inode
->i_mode
>> 12);
165 /* remember our position */
167 fi
->next_offset
= di
->offset
;
179 /* make sure a dentry wasn't dropped while we didn't have dcache_lock */
180 if (!ceph_i_test(dir
, CEPH_I_COMPLETE
)) {
181 dout(" lost I_COMPLETE on %p; falling back to mds\n", dir
);
186 spin_lock(&dcache_lock
);
187 p
= p
->prev
; /* advance to next dentry */
191 spin_unlock(&dcache_lock
);
199 * make note of the last dentry we read, so we can
200 * continue at the same lexicographical point,
201 * regardless of what dir changes take place on the
204 static int note_last_dentry(struct ceph_file_info
*fi
, const char *name
,
207 kfree(fi
->last_name
);
208 fi
->last_name
= kmalloc(len
+1, GFP_NOFS
);
211 memcpy(fi
->last_name
, name
, len
);
212 fi
->last_name
[len
] = 0;
213 dout("note_last_dentry '%s'\n", fi
->last_name
);
217 static int ceph_readdir(struct file
*filp
, void *dirent
, filldir_t filldir
)
219 struct ceph_file_info
*fi
= filp
->private_data
;
220 struct inode
*inode
= filp
->f_dentry
->d_inode
;
221 struct ceph_inode_info
*ci
= ceph_inode(inode
);
222 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
223 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
224 unsigned frag
= fpos_frag(filp
->f_pos
);
225 int off
= fpos_off(filp
->f_pos
);
228 struct ceph_mds_reply_info_parsed
*rinfo
;
229 const int max_entries
= fsc
->mount_options
->max_readdir
;
230 const int max_bytes
= fsc
->mount_options
->max_readdir_bytes
;
232 dout("readdir %p filp %p frag %u off %u\n", inode
, filp
, frag
, off
);
236 /* always start with . and .. */
237 if (filp
->f_pos
== 0) {
238 /* note dir version at start of readdir so we can tell
239 * if any dentries get dropped */
240 fi
->dir_release_count
= ci
->i_release_count
;
242 dout("readdir off 0 -> '.'\n");
243 if (filldir(dirent
, ".", 1, ceph_make_fpos(0, 0),
244 inode
->i_ino
, inode
->i_mode
>> 12) < 0)
249 if (filp
->f_pos
== 1) {
250 dout("readdir off 1 -> '..'\n");
251 if (filldir(dirent
, "..", 2, ceph_make_fpos(0, 1),
252 filp
->f_dentry
->d_parent
->d_inode
->i_ino
,
253 inode
->i_mode
>> 12) < 0)
259 /* can we use the dcache? */
260 spin_lock(&inode
->i_lock
);
261 if ((filp
->f_pos
== 2 || fi
->dentry
) &&
262 !ceph_test_mount_opt(fsc
, NOASYNCREADDIR
) &&
263 ceph_snap(inode
) != CEPH_SNAPDIR
&&
264 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) &&
265 __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1)) {
266 spin_unlock(&inode
->i_lock
);
267 err
= __dcache_readdir(filp
, dirent
, filldir
);
271 spin_unlock(&inode
->i_lock
);
274 err
= note_last_dentry(fi
, fi
->dentry
->d_name
.name
,
275 fi
->dentry
->d_name
.len
);
282 /* proceed with a normal readdir */
285 /* do we have the correct frag content buffered? */
286 if (fi
->frag
!= frag
|| fi
->last_readdir
== NULL
) {
287 struct ceph_mds_request
*req
;
288 int op
= ceph_snap(inode
) == CEPH_SNAPDIR
?
289 CEPH_MDS_OP_LSSNAP
: CEPH_MDS_OP_READDIR
;
291 /* discard old result, if any */
292 if (fi
->last_readdir
) {
293 ceph_mdsc_put_request(fi
->last_readdir
);
294 fi
->last_readdir
= NULL
;
297 /* requery frag tree, as the frag topology may have changed */
298 frag
= ceph_choose_frag(ceph_inode(inode
), frag
, NULL
, NULL
);
300 dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
301 ceph_vinop(inode
), frag
, fi
->last_name
);
302 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
305 req
->r_inode
= igrab(inode
);
306 req
->r_dentry
= dget(filp
->f_dentry
);
307 /* hints to request -> mds selection code */
308 req
->r_direct_mode
= USE_AUTH_MDS
;
309 req
->r_direct_hash
= ceph_frag_value(frag
);
310 req
->r_direct_is_hash
= true;
311 req
->r_path2
= kstrdup(fi
->last_name
, GFP_NOFS
);
312 req
->r_readdir_offset
= fi
->next_offset
;
313 req
->r_args
.readdir
.frag
= cpu_to_le32(frag
);
314 req
->r_args
.readdir
.max_entries
= cpu_to_le32(max_entries
);
315 req
->r_args
.readdir
.max_bytes
= cpu_to_le32(max_bytes
);
316 req
->r_num_caps
= max_entries
+ 1;
317 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
319 ceph_mdsc_put_request(req
);
322 dout("readdir got and parsed readdir result=%d"
323 " on frag %x, end=%d, complete=%d\n", err
, frag
,
324 (int)req
->r_reply_info
.dir_end
,
325 (int)req
->r_reply_info
.dir_complete
);
327 if (!req
->r_did_prepopulate
) {
328 dout("readdir !did_prepopulate");
329 fi
->dir_release_count
--; /* preclude I_COMPLETE */
332 /* note next offset and last dentry name */
333 fi
->offset
= fi
->next_offset
;
334 fi
->last_readdir
= req
;
336 if (req
->r_reply_info
.dir_end
) {
337 kfree(fi
->last_name
);
338 fi
->last_name
= NULL
;
339 if (ceph_frag_is_rightmost(frag
))
344 rinfo
= &req
->r_reply_info
;
345 err
= note_last_dentry(fi
,
346 rinfo
->dir_dname
[rinfo
->dir_nr
-1],
347 rinfo
->dir_dname_len
[rinfo
->dir_nr
-1]);
350 fi
->next_offset
+= rinfo
->dir_nr
;
354 rinfo
= &fi
->last_readdir
->r_reply_info
;
355 dout("readdir frag %x num %d off %d chunkoff %d\n", frag
,
356 rinfo
->dir_nr
, off
, fi
->offset
);
357 while (off
- fi
->offset
>= 0 && off
- fi
->offset
< rinfo
->dir_nr
) {
358 u64 pos
= ceph_make_fpos(frag
, off
);
359 struct ceph_mds_reply_inode
*in
=
360 rinfo
->dir_in
[off
- fi
->offset
].in
;
361 dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
362 off
, off
- fi
->offset
, rinfo
->dir_nr
, pos
,
363 rinfo
->dir_dname_len
[off
- fi
->offset
],
364 rinfo
->dir_dname
[off
- fi
->offset
], in
);
366 ftype
= le32_to_cpu(in
->mode
) >> 12;
368 rinfo
->dir_dname
[off
- fi
->offset
],
369 rinfo
->dir_dname_len
[off
- fi
->offset
],
371 le64_to_cpu(in
->ino
),
373 dout("filldir stopping us...\n");
377 filp
->f_pos
= pos
+ 1;
381 ceph_mdsc_put_request(fi
->last_readdir
);
382 fi
->last_readdir
= NULL
;
387 if (!ceph_frag_is_rightmost(frag
)) {
388 frag
= ceph_frag_next(frag
);
390 filp
->f_pos
= ceph_make_fpos(frag
, off
);
391 dout("readdir next frag is %x\n", frag
);
397 * if dir_release_count still matches the dir, no dentries
398 * were released during the whole readdir, and we should have
399 * the complete dir contents in our cache.
401 spin_lock(&inode
->i_lock
);
402 if (ci
->i_release_count
== fi
->dir_release_count
) {
403 dout(" marking %p complete\n", inode
);
404 ci
->i_ceph_flags
|= CEPH_I_COMPLETE
;
405 ci
->i_max_offset
= filp
->f_pos
;
407 spin_unlock(&inode
->i_lock
);
409 dout("readdir %p filp %p done.\n", inode
, filp
);
413 static void reset_readdir(struct ceph_file_info
*fi
)
415 if (fi
->last_readdir
) {
416 ceph_mdsc_put_request(fi
->last_readdir
);
417 fi
->last_readdir
= NULL
;
419 kfree(fi
->last_name
);
420 fi
->last_name
= NULL
;
421 fi
->next_offset
= 2; /* compensate for . and .. */
429 static loff_t
ceph_dir_llseek(struct file
*file
, loff_t offset
, int origin
)
431 struct ceph_file_info
*fi
= file
->private_data
;
432 struct inode
*inode
= file
->f_mapping
->host
;
433 loff_t old_offset
= offset
;
436 mutex_lock(&inode
->i_mutex
);
439 offset
+= inode
->i_size
+ 2; /* FIXME */
442 offset
+= file
->f_pos
;
445 if (offset
>= 0 && offset
<= inode
->i_sb
->s_maxbytes
) {
446 if (offset
!= file
->f_pos
) {
447 file
->f_pos
= offset
;
454 * discard buffered readdir content on seekdir(0), or
455 * seek to new frag, or seek prior to current chunk.
458 fpos_frag(offset
) != fpos_frag(old_offset
) ||
459 fpos_off(offset
) < fi
->offset
) {
460 dout("dir_llseek dropping %p content\n", file
);
464 /* bump dir_release_count if we did a forward seek */
465 if (offset
> old_offset
)
466 fi
->dir_release_count
--;
468 mutex_unlock(&inode
->i_mutex
);
473 * Process result of a lookup/open request.
475 * Mainly, make sure we return the final req->r_dentry (if it already
476 * existed) in place of the original VFS-provided dentry when they
479 * Gracefully handle the case where the MDS replies with -ENOENT and
480 * no trace (which it may do, at its discretion, e.g., if it doesn't
481 * care to issue a lease on the negative dentry).
483 struct dentry
*ceph_finish_lookup(struct ceph_mds_request
*req
,
484 struct dentry
*dentry
, int err
)
486 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dentry
->d_sb
);
487 struct inode
*parent
= dentry
->d_parent
->d_inode
;
490 if (err
== -ENOENT
&&
491 strcmp(dentry
->d_name
.name
,
492 fsc
->mount_options
->snapdir_name
) == 0) {
493 struct inode
*inode
= ceph_get_snapdir(parent
);
494 dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
495 dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
, inode
);
496 BUG_ON(!d_unhashed(dentry
));
497 d_add(dentry
, inode
);
501 if (err
== -ENOENT
) {
504 if (!req
->r_reply_info
.head
->is_dentry
) {
505 dout("ENOENT and no trace, dentry %p inode %p\n",
506 dentry
, dentry
->d_inode
);
507 if (dentry
->d_inode
) {
516 dentry
= ERR_PTR(err
);
517 else if (dentry
!= req
->r_dentry
)
518 dentry
= dget(req
->r_dentry
); /* we got spliced */
524 static int is_root_ceph_dentry(struct inode
*inode
, struct dentry
*dentry
)
526 return ceph_ino(inode
) == CEPH_INO_ROOT
&&
527 strncmp(dentry
->d_name
.name
, ".ceph", 5) == 0;
531 * Look up a single dir entry. If there is a lookup intent, inform
532 * the MDS so that it gets our 'caps wanted' value in a single op.
534 static struct dentry
*ceph_lookup(struct inode
*dir
, struct dentry
*dentry
,
535 struct nameidata
*nd
)
537 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
538 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
539 struct ceph_mds_request
*req
;
543 dout("lookup %p dentry %p '%.*s'\n",
544 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
546 if (dentry
->d_name
.len
> NAME_MAX
)
547 return ERR_PTR(-ENAMETOOLONG
);
549 err
= ceph_init_dentry(dentry
);
553 /* open (but not create!) intent? */
555 (nd
->flags
& LOOKUP_OPEN
) &&
556 (nd
->flags
& LOOKUP_CONTINUE
) == 0 && /* only open last component */
557 !(nd
->intent
.open
.flags
& O_CREAT
)) {
558 int mode
= nd
->intent
.open
.create_mode
& ~current
->fs
->umask
;
559 return ceph_lookup_open(dir
, dentry
, nd
, mode
, 1);
562 /* can we conclude ENOENT locally? */
563 if (dentry
->d_inode
== NULL
) {
564 struct ceph_inode_info
*ci
= ceph_inode(dir
);
565 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
567 spin_lock(&dir
->i_lock
);
568 dout(" dir %p flags are %d\n", dir
, ci
->i_ceph_flags
);
569 if (strncmp(dentry
->d_name
.name
,
570 fsc
->mount_options
->snapdir_name
,
571 dentry
->d_name
.len
) &&
572 !is_root_ceph_dentry(dir
, dentry
) &&
573 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) &&
574 (__ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1))) {
575 spin_unlock(&dir
->i_lock
);
576 dout(" dir %p complete, -ENOENT\n", dir
);
578 di
->lease_shared_gen
= ci
->i_shared_gen
;
581 spin_unlock(&dir
->i_lock
);
584 op
= ceph_snap(dir
) == CEPH_SNAPDIR
?
585 CEPH_MDS_OP_LOOKUPSNAP
: CEPH_MDS_OP_LOOKUP
;
586 req
= ceph_mdsc_create_request(mdsc
, op
, USE_ANY_MDS
);
588 return ERR_CAST(req
);
589 req
->r_dentry
= dget(dentry
);
591 /* we only need inode linkage */
592 req
->r_args
.getattr
.mask
= cpu_to_le32(CEPH_STAT_CAP_INODE
);
593 req
->r_locked_dir
= dir
;
594 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
595 dentry
= ceph_finish_lookup(req
, dentry
, err
);
596 ceph_mdsc_put_request(req
); /* will dput(dentry) */
597 dout("lookup result=%p\n", dentry
);
602 * If we do a create but get no trace back from the MDS, follow up with
603 * a lookup (the VFS expects us to link up the provided dentry).
605 int ceph_handle_notrace_create(struct inode
*dir
, struct dentry
*dentry
)
607 struct dentry
*result
= ceph_lookup(dir
, dentry
, NULL
);
609 if (result
&& !IS_ERR(result
)) {
611 * We created the item, then did a lookup, and found
612 * it was already linked to another inode we already
613 * had in our cache (and thus got spliced). Link our
614 * dentry to that inode, but don't hash it, just in
615 * case the VFS wants to dereference it.
617 BUG_ON(!result
->d_inode
);
618 d_instantiate(dentry
, result
->d_inode
);
621 return PTR_ERR(result
);
624 static int ceph_mknod(struct inode
*dir
, struct dentry
*dentry
,
625 int mode
, dev_t rdev
)
627 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
628 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
629 struct ceph_mds_request
*req
;
632 if (ceph_snap(dir
) != CEPH_NOSNAP
)
635 dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
636 dir
, dentry
, mode
, rdev
);
637 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_MKNOD
, USE_AUTH_MDS
);
642 req
->r_dentry
= dget(dentry
);
644 req
->r_locked_dir
= dir
;
645 req
->r_args
.mknod
.mode
= cpu_to_le32(mode
);
646 req
->r_args
.mknod
.rdev
= cpu_to_le32(rdev
);
647 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
648 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
649 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
650 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
651 err
= ceph_handle_notrace_create(dir
, dentry
);
652 ceph_mdsc_put_request(req
);
658 static int ceph_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
659 struct nameidata
*nd
)
661 dout("create in dir %p dentry %p name '%.*s'\n",
662 dir
, dentry
, dentry
->d_name
.len
, dentry
->d_name
.name
);
664 if (ceph_snap(dir
) != CEPH_NOSNAP
)
668 BUG_ON((nd
->flags
& LOOKUP_OPEN
) == 0);
669 dentry
= ceph_lookup_open(dir
, dentry
, nd
, mode
, 0);
670 /* hrm, what should i do here if we get aliased? */
672 return PTR_ERR(dentry
);
676 /* fall back to mknod */
677 return ceph_mknod(dir
, dentry
, (mode
& ~S_IFMT
) | S_IFREG
, 0);
680 static int ceph_symlink(struct inode
*dir
, struct dentry
*dentry
,
683 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
684 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
685 struct ceph_mds_request
*req
;
688 if (ceph_snap(dir
) != CEPH_NOSNAP
)
691 dout("symlink in dir %p dentry %p to '%s'\n", dir
, dentry
, dest
);
692 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SYMLINK
, USE_AUTH_MDS
);
697 req
->r_dentry
= dget(dentry
);
699 req
->r_path2
= kstrdup(dest
, GFP_NOFS
);
700 req
->r_locked_dir
= dir
;
701 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
702 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
703 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
704 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
705 err
= ceph_handle_notrace_create(dir
, dentry
);
706 ceph_mdsc_put_request(req
);
712 static int ceph_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
714 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
715 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
716 struct ceph_mds_request
*req
;
720 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
721 /* mkdir .snap/foo is a MKSNAP */
722 op
= CEPH_MDS_OP_MKSNAP
;
723 dout("mksnap dir %p snap '%.*s' dn %p\n", dir
,
724 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
);
725 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
726 dout("mkdir dir %p dn %p mode 0%o\n", dir
, dentry
, mode
);
727 op
= CEPH_MDS_OP_MKDIR
;
731 req
= ceph_mdsc_create_request(mdsc
, op
, USE_AUTH_MDS
);
737 req
->r_dentry
= dget(dentry
);
739 req
->r_locked_dir
= dir
;
740 req
->r_args
.mkdir
.mode
= cpu_to_le32(mode
);
741 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
742 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
743 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
744 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
745 err
= ceph_handle_notrace_create(dir
, dentry
);
746 ceph_mdsc_put_request(req
);
753 static int ceph_link(struct dentry
*old_dentry
, struct inode
*dir
,
754 struct dentry
*dentry
)
756 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
757 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
758 struct ceph_mds_request
*req
;
761 if (ceph_snap(dir
) != CEPH_NOSNAP
)
764 dout("link in dir %p old_dentry %p dentry %p\n", dir
,
766 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_LINK
, USE_AUTH_MDS
);
771 req
->r_dentry
= dget(dentry
);
773 req
->r_old_dentry
= dget(old_dentry
); /* or inode? hrm. */
774 req
->r_locked_dir
= dir
;
775 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
776 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
777 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
780 else if (!req
->r_reply_info
.head
->is_dentry
)
781 d_instantiate(dentry
, igrab(old_dentry
->d_inode
));
782 ceph_mdsc_put_request(req
);
787 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
788 * looks like the link count will hit 0, drop any other caps (other
789 * than PIN) we don't specifically want (due to the file still being
792 static int drop_caps_for_unlink(struct inode
*inode
)
794 struct ceph_inode_info
*ci
= ceph_inode(inode
);
795 int drop
= CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
;
797 spin_lock(&inode
->i_lock
);
798 if (inode
->i_nlink
== 1) {
799 drop
|= ~(__ceph_caps_wanted(ci
) | CEPH_CAP_PIN
);
800 ci
->i_ceph_flags
|= CEPH_I_NODELAY
;
802 spin_unlock(&inode
->i_lock
);
807 * rmdir and unlink are differ only by the metadata op code
809 static int ceph_unlink(struct inode
*dir
, struct dentry
*dentry
)
811 struct ceph_fs_client
*fsc
= ceph_sb_to_client(dir
->i_sb
);
812 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
813 struct inode
*inode
= dentry
->d_inode
;
814 struct ceph_mds_request
*req
;
818 if (ceph_snap(dir
) == CEPH_SNAPDIR
) {
819 /* rmdir .snap/foo is RMSNAP */
820 dout("rmsnap dir %p '%.*s' dn %p\n", dir
, dentry
->d_name
.len
,
821 dentry
->d_name
.name
, dentry
);
822 op
= CEPH_MDS_OP_RMSNAP
;
823 } else if (ceph_snap(dir
) == CEPH_NOSNAP
) {
824 dout("unlink/rmdir dir %p dn %p inode %p\n",
826 op
= ((dentry
->d_inode
->i_mode
& S_IFMT
) == S_IFDIR
) ?
827 CEPH_MDS_OP_RMDIR
: CEPH_MDS_OP_UNLINK
;
830 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_dentry_drop
= CEPH_CAP_FILE_SHARED
;
839 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
840 req
->r_inode_drop
= drop_caps_for_unlink(inode
);
841 err
= ceph_mdsc_do_request(mdsc
, dir
, req
);
842 if (!err
&& !req
->r_reply_info
.head
->is_dentry
)
844 ceph_mdsc_put_request(req
);
849 static int ceph_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
850 struct inode
*new_dir
, struct dentry
*new_dentry
)
852 struct ceph_fs_client
*fsc
= ceph_sb_to_client(old_dir
->i_sb
);
853 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
854 struct ceph_mds_request
*req
;
857 if (ceph_snap(old_dir
) != ceph_snap(new_dir
))
859 if (ceph_snap(old_dir
) != CEPH_NOSNAP
||
860 ceph_snap(new_dir
) != CEPH_NOSNAP
)
862 dout("rename dir %p dentry %p to dir %p dentry %p\n",
863 old_dir
, old_dentry
, new_dir
, new_dentry
);
864 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_RENAME
, USE_AUTH_MDS
);
867 req
->r_dentry
= dget(new_dentry
);
869 req
->r_old_dentry
= dget(old_dentry
);
870 req
->r_locked_dir
= new_dir
;
871 req
->r_old_dentry_drop
= CEPH_CAP_FILE_SHARED
;
872 req
->r_old_dentry_unless
= CEPH_CAP_FILE_EXCL
;
873 req
->r_dentry_drop
= CEPH_CAP_FILE_SHARED
;
874 req
->r_dentry_unless
= CEPH_CAP_FILE_EXCL
;
875 /* release LINK_RDCACHE on source inode (mds will lock it) */
876 req
->r_old_inode_drop
= CEPH_CAP_LINK_SHARED
;
877 if (new_dentry
->d_inode
)
878 req
->r_inode_drop
= drop_caps_for_unlink(new_dentry
->d_inode
);
879 err
= ceph_mdsc_do_request(mdsc
, old_dir
, req
);
880 if (!err
&& !req
->r_reply_info
.head
->is_dentry
) {
882 * Normally d_move() is done by fill_trace (called by
883 * do_request, above). If there is no trace, we need
887 /* d_move screws up d_subdirs order */
888 ceph_i_clear(new_dir
, CEPH_I_COMPLETE
);
890 d_move(old_dentry
, new_dentry
);
892 /* ensure target dentry is invalidated, despite
893 rehashing bug in vfs_rename_dir */
894 ceph_invalidate_dentry_lease(new_dentry
);
896 ceph_mdsc_put_request(req
);
901 * Ensure a dentry lease will no longer revalidate.
903 void ceph_invalidate_dentry_lease(struct dentry
*dentry
)
905 spin_lock(&dentry
->d_lock
);
906 dentry
->d_time
= jiffies
;
907 ceph_dentry(dentry
)->lease_shared_gen
= 0;
908 spin_unlock(&dentry
->d_lock
);
912 * Check if dentry lease is valid. If not, delete the lease. Try to
913 * renew if the least is more than half up.
915 static int dentry_lease_is_valid(struct dentry
*dentry
)
917 struct ceph_dentry_info
*di
;
918 struct ceph_mds_session
*s
;
922 struct ceph_mds_session
*session
= NULL
;
923 struct inode
*dir
= NULL
;
926 spin_lock(&dentry
->d_lock
);
927 di
= ceph_dentry(dentry
);
928 if (di
&& di
->lease_session
) {
929 s
= di
->lease_session
;
930 spin_lock(&s
->s_cap_lock
);
933 spin_unlock(&s
->s_cap_lock
);
935 if (di
->lease_gen
== gen
&&
936 time_before(jiffies
, dentry
->d_time
) &&
937 time_before(jiffies
, ttl
)) {
939 if (di
->lease_renew_after
&&
940 time_after(jiffies
, di
->lease_renew_after
)) {
941 /* we should renew */
942 dir
= dentry
->d_parent
->d_inode
;
943 session
= ceph_get_mds_session(s
);
945 di
->lease_renew_after
= 0;
946 di
->lease_renew_from
= jiffies
;
950 spin_unlock(&dentry
->d_lock
);
953 ceph_mdsc_lease_send_msg(session
, dir
, dentry
,
954 CEPH_MDS_LEASE_RENEW
, seq
);
955 ceph_put_mds_session(session
);
957 dout("dentry_lease_is_valid - dentry %p = %d\n", dentry
, valid
);
962 * Check if directory-wide content lease/cap is valid.
964 static int dir_lease_is_valid(struct inode
*dir
, struct dentry
*dentry
)
966 struct ceph_inode_info
*ci
= ceph_inode(dir
);
967 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
970 spin_lock(&dir
->i_lock
);
971 if (ci
->i_shared_gen
== di
->lease_shared_gen
)
972 valid
= __ceph_caps_issued_mask(ci
, CEPH_CAP_FILE_SHARED
, 1);
973 spin_unlock(&dir
->i_lock
);
974 dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
975 dir
, (unsigned)ci
->i_shared_gen
, dentry
,
976 (unsigned)di
->lease_shared_gen
, valid
);
981 * Check if cached dentry can be trusted.
983 static int ceph_d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
985 struct inode
*dir
= dentry
->d_parent
->d_inode
;
987 dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry
,
988 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
,
989 ceph_dentry(dentry
)->offset
);
991 /* always trust cached snapped dentries, snapdir dentry */
992 if (ceph_snap(dir
) != CEPH_NOSNAP
) {
993 dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry
,
994 dentry
->d_name
.len
, dentry
->d_name
.name
, dentry
->d_inode
);
997 if (dentry
->d_inode
&& ceph_snap(dentry
->d_inode
) == CEPH_SNAPDIR
)
1000 if (dentry_lease_is_valid(dentry
) ||
1001 dir_lease_is_valid(dir
, dentry
))
1004 dout("d_revalidate %p invalid\n", dentry
);
1008 ceph_dentry_lru_touch(dentry
);
1013 * When a dentry is released, clear the dir I_COMPLETE if it was part
1014 * of the current dir gen or if this is in the snapshot namespace.
1016 static void ceph_dentry_release(struct dentry
*dentry
)
1018 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1019 struct inode
*parent_inode
= NULL
;
1020 u64 snapid
= CEPH_NOSNAP
;
1022 if (!IS_ROOT(dentry
)) {
1023 parent_inode
= dentry
->d_parent
->d_inode
;
1025 snapid
= ceph_snap(parent_inode
);
1027 dout("dentry_release %p parent %p\n", dentry
, parent_inode
);
1028 if (parent_inode
&& snapid
!= CEPH_SNAPDIR
) {
1029 struct ceph_inode_info
*ci
= ceph_inode(parent_inode
);
1031 spin_lock(&parent_inode
->i_lock
);
1032 if (ci
->i_shared_gen
== di
->lease_shared_gen
||
1033 snapid
<= CEPH_MAXSNAP
) {
1034 dout(" clearing %p complete (d_release)\n",
1036 ci
->i_ceph_flags
&= ~CEPH_I_COMPLETE
;
1037 ci
->i_release_count
++;
1039 spin_unlock(&parent_inode
->i_lock
);
1042 ceph_dentry_lru_del(dentry
);
1043 if (di
->lease_session
)
1044 ceph_put_mds_session(di
->lease_session
);
1045 kmem_cache_free(ceph_dentry_cachep
, di
);
1046 dentry
->d_fsdata
= NULL
;
1050 static int ceph_snapdir_d_revalidate(struct dentry
*dentry
,
1051 struct nameidata
*nd
)
1054 * Eventually, we'll want to revalidate snapped metadata
1055 * too... probably...
1063 * read() on a dir. This weird interface hack only works if mounted
1064 * with '-o dirstat'.
1066 static ssize_t
ceph_read_dir(struct file
*file
, char __user
*buf
, size_t size
,
1069 struct ceph_file_info
*cf
= file
->private_data
;
1070 struct inode
*inode
= file
->f_dentry
->d_inode
;
1071 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1074 if (!ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), DIRSTAT
))
1077 if (!cf
->dir_info
) {
1078 cf
->dir_info
= kmalloc(1024, GFP_NOFS
);
1082 sprintf(cf
->dir_info
,
1085 " subdirs: %20lld\n"
1086 "rentries: %20lld\n"
1088 " rsubdirs: %20lld\n"
1090 "rctime: %10ld.%09ld\n",
1091 ci
->i_files
+ ci
->i_subdirs
,
1094 ci
->i_rfiles
+ ci
->i_rsubdirs
,
1098 (long)ci
->i_rctime
.tv_sec
,
1099 (long)ci
->i_rctime
.tv_nsec
);
1102 if (*ppos
>= cf
->dir_info_len
)
1104 size
= min_t(unsigned, size
, cf
->dir_info_len
-*ppos
);
1105 left
= copy_to_user(buf
, cf
->dir_info
+ *ppos
, size
);
1108 *ppos
+= (size
- left
);
1113 * an fsync() on a dir will wait for any uncommitted directory
1114 * operations to commit.
1116 static int ceph_dir_fsync(struct file
*file
, int datasync
)
1118 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
1119 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1120 struct list_head
*head
= &ci
->i_unsafe_dirops
;
1121 struct ceph_mds_request
*req
;
1125 dout("dir_fsync %p\n", inode
);
1126 spin_lock(&ci
->i_unsafe_lock
);
1127 if (list_empty(head
))
1130 req
= list_entry(head
->prev
,
1131 struct ceph_mds_request
, r_unsafe_dir_item
);
1132 last_tid
= req
->r_tid
;
1135 ceph_mdsc_get_request(req
);
1136 spin_unlock(&ci
->i_unsafe_lock
);
1137 dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1138 inode
, req
->r_tid
, last_tid
);
1139 if (req
->r_timeout
) {
1140 ret
= wait_for_completion_timeout(
1141 &req
->r_safe_completion
, req
->r_timeout
);
1145 ret
= -EIO
; /* timed out */
1147 wait_for_completion(&req
->r_safe_completion
);
1149 spin_lock(&ci
->i_unsafe_lock
);
1150 ceph_mdsc_put_request(req
);
1152 if (ret
|| list_empty(head
))
1154 req
= list_entry(head
->next
,
1155 struct ceph_mds_request
, r_unsafe_dir_item
);
1156 } while (req
->r_tid
< last_tid
);
1158 spin_unlock(&ci
->i_unsafe_lock
);
1163 * We maintain a private dentry LRU.
1165 * FIXME: this needs to be changed to a per-mds lru to be useful.
1167 void ceph_dentry_lru_add(struct dentry
*dn
)
1169 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1170 struct ceph_mds_client
*mdsc
;
1172 dout("dentry_lru_add %p %p '%.*s'\n", di
, dn
,
1173 dn
->d_name
.len
, dn
->d_name
.name
);
1175 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1176 spin_lock(&mdsc
->dentry_lru_lock
);
1177 list_add_tail(&di
->lru
, &mdsc
->dentry_lru
);
1179 spin_unlock(&mdsc
->dentry_lru_lock
);
1183 void ceph_dentry_lru_touch(struct dentry
*dn
)
1185 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1186 struct ceph_mds_client
*mdsc
;
1188 dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di
, dn
,
1189 dn
->d_name
.len
, dn
->d_name
.name
, di
->offset
);
1191 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1192 spin_lock(&mdsc
->dentry_lru_lock
);
1193 list_move_tail(&di
->lru
, &mdsc
->dentry_lru
);
1194 spin_unlock(&mdsc
->dentry_lru_lock
);
1198 void ceph_dentry_lru_del(struct dentry
*dn
)
1200 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1201 struct ceph_mds_client
*mdsc
;
1203 dout("dentry_lru_del %p %p '%.*s'\n", di
, dn
,
1204 dn
->d_name
.len
, dn
->d_name
.name
);
1206 mdsc
= ceph_sb_to_client(dn
->d_sb
)->mdsc
;
1207 spin_lock(&mdsc
->dentry_lru_lock
);
1208 list_del_init(&di
->lru
);
1210 spin_unlock(&mdsc
->dentry_lru_lock
);
1214 const struct file_operations ceph_dir_fops
= {
1215 .read
= ceph_read_dir
,
1216 .readdir
= ceph_readdir
,
1217 .llseek
= ceph_dir_llseek
,
1219 .release
= ceph_release
,
1220 .unlocked_ioctl
= ceph_ioctl
,
1221 .fsync
= ceph_dir_fsync
,
1224 const struct inode_operations ceph_dir_iops
= {
1225 .lookup
= ceph_lookup
,
1226 .permission
= ceph_permission
,
1227 .getattr
= ceph_getattr
,
1228 .setattr
= ceph_setattr
,
1229 .setxattr
= ceph_setxattr
,
1230 .getxattr
= ceph_getxattr
,
1231 .listxattr
= ceph_listxattr
,
1232 .removexattr
= ceph_removexattr
,
1233 .mknod
= ceph_mknod
,
1234 .symlink
= ceph_symlink
,
1235 .mkdir
= ceph_mkdir
,
1237 .unlink
= ceph_unlink
,
1238 .rmdir
= ceph_unlink
,
1239 .rename
= ceph_rename
,
1240 .create
= ceph_create
,
1243 const struct dentry_operations ceph_dentry_ops
= {
1244 .d_revalidate
= ceph_d_revalidate
,
1245 .d_release
= ceph_dentry_release
,
1248 const struct dentry_operations ceph_snapdir_dentry_ops
= {
1249 .d_revalidate
= ceph_snapdir_d_revalidate
,
1250 .d_release
= ceph_dentry_release
,
1253 const struct dentry_operations ceph_snap_dentry_ops
= {
1254 .d_release
= ceph_dentry_release
,