1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/writeback.h>
10 #include <linux/vmalloc.h>
11 #include <linux/posix_acl.h>
12 #include <linux/random.h>
15 #include "mds_client.h"
17 #include <linux/ceph/decode.h>
20 * Ceph inode operations
22 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23 * setattr, etc.), xattr helpers, and helpers for assimilating
24 * metadata returned by the MDS into our cache.
26 * Also define helpers for doing asynchronous writeback, invalidation,
27 * and truncation for the benefit of those who can't afford to block
28 * (typically because they are in the message handler path).
31 static const struct inode_operations ceph_symlink_iops
;
33 static void ceph_invalidate_work(struct work_struct
*work
);
34 static void ceph_writeback_work(struct work_struct
*work
);
35 static void ceph_vmtruncate_work(struct work_struct
*work
);
38 * find or create an inode, given the ceph ino number
40 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
42 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
43 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
47 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
50 ino_t t
= ceph_vino_to_ino(vino
);
52 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
54 return ERR_PTR(-ENOMEM
);
55 if (inode
->i_state
& I_NEW
) {
56 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
57 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
58 unlock_new_inode(inode
);
61 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
67 * get/constuct snapdir inode for a given directory
69 struct inode
*ceph_get_snapdir(struct inode
*parent
)
71 struct ceph_vino vino
= {
72 .ino
= ceph_ino(parent
),
75 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
76 struct ceph_inode_info
*ci
= ceph_inode(inode
);
78 BUG_ON(!S_ISDIR(parent
->i_mode
));
81 inode
->i_mode
= parent
->i_mode
;
82 inode
->i_uid
= parent
->i_uid
;
83 inode
->i_gid
= parent
->i_gid
;
84 inode
->i_op
= &ceph_snapdir_iops
;
85 inode
->i_fop
= &ceph_snapdir_fops
;
86 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
91 const struct inode_operations ceph_file_iops
= {
92 .permission
= ceph_permission
,
93 .setattr
= ceph_setattr
,
94 .getattr
= ceph_getattr
,
95 .setxattr
= ceph_setxattr
,
96 .getxattr
= ceph_getxattr
,
97 .listxattr
= ceph_listxattr
,
98 .removexattr
= ceph_removexattr
,
99 .get_acl
= ceph_get_acl
,
100 .set_acl
= ceph_set_acl
,
105 * We use a 'frag tree' to keep track of the MDS's directory fragments
106 * for a given inode (usually there is just a single fragment). We
107 * need to know when a child frag is delegated to a new MDS, or when
108 * it is flagged as replicated, so we can direct our requests
113 * find/create a frag in the tree
115 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
119 struct rb_node
*parent
= NULL
;
120 struct ceph_inode_frag
*frag
;
123 p
= &ci
->i_fragtree
.rb_node
;
126 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
127 c
= ceph_frag_compare(f
, frag
->frag
);
136 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
138 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
139 "frag %x\n", &ci
->vfs_inode
,
140 ceph_vinop(&ci
->vfs_inode
), f
);
141 return ERR_PTR(-ENOMEM
);
148 rb_link_node(&frag
->node
, parent
, p
);
149 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
151 dout("get_or_create_frag added %llx.%llx frag %x\n",
152 ceph_vinop(&ci
->vfs_inode
), f
);
157 * find a specific frag @f
159 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
161 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
164 struct ceph_inode_frag
*frag
=
165 rb_entry(n
, struct ceph_inode_frag
, node
);
166 int c
= ceph_frag_compare(f
, frag
->frag
);
178 * Choose frag containing the given value @v. If @pfrag is
179 * specified, copy the frag delegation info to the caller if
182 static u32
__ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
183 struct ceph_inode_frag
*pfrag
, int *found
)
185 u32 t
= ceph_frag_make(0, 0);
186 struct ceph_inode_frag
*frag
;
194 WARN_ON(!ceph_frag_contains_value(t
, v
));
195 frag
= __ceph_find_frag(ci
, t
);
197 break; /* t is a leaf */
198 if (frag
->split_by
== 0) {
200 memcpy(pfrag
, frag
, sizeof(*pfrag
));
207 nway
= 1 << frag
->split_by
;
208 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
209 frag
->split_by
, nway
);
210 for (i
= 0; i
< nway
; i
++) {
211 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
212 if (ceph_frag_contains_value(n
, v
)) {
219 dout("choose_frag(%x) = %x\n", v
, t
);
224 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
225 struct ceph_inode_frag
*pfrag
, int *found
)
228 mutex_lock(&ci
->i_fragtree_mutex
);
229 ret
= __ceph_choose_frag(ci
, v
, pfrag
, found
);
230 mutex_unlock(&ci
->i_fragtree_mutex
);
235 * Process dirfrag (delegation) info from the mds. Include leaf
236 * fragment in tree ONLY if ndist > 0. Otherwise, only
237 * branches/splits are included in i_fragtree)
239 static int ceph_fill_dirfrag(struct inode
*inode
,
240 struct ceph_mds_reply_dirfrag
*dirinfo
)
242 struct ceph_inode_info
*ci
= ceph_inode(inode
);
243 struct ceph_inode_frag
*frag
;
244 u32 id
= le32_to_cpu(dirinfo
->frag
);
245 int mds
= le32_to_cpu(dirinfo
->auth
);
246 int ndist
= le32_to_cpu(dirinfo
->ndist
);
251 spin_lock(&ci
->i_ceph_lock
);
253 diri_auth
= ci
->i_auth_cap
->mds
;
254 spin_unlock(&ci
->i_ceph_lock
);
256 mutex_lock(&ci
->i_fragtree_mutex
);
257 if (ndist
== 0 && mds
== diri_auth
) {
258 /* no delegation info needed. */
259 frag
= __ceph_find_frag(ci
, id
);
262 if (frag
->split_by
== 0) {
263 /* tree leaf, remove */
264 dout("fill_dirfrag removed %llx.%llx frag %x"
265 " (no ref)\n", ceph_vinop(inode
), id
);
266 rb_erase(&frag
->node
, &ci
->i_fragtree
);
269 /* tree branch, keep and clear */
270 dout("fill_dirfrag cleared %llx.%llx frag %x"
271 " referral\n", ceph_vinop(inode
), id
);
279 /* find/add this frag to store mds delegation info */
280 frag
= __get_or_create_frag(ci
, id
);
282 /* this is not the end of the world; we can continue
283 with bad/inaccurate delegation info */
284 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
285 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
291 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
292 for (i
= 0; i
< frag
->ndist
; i
++)
293 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
294 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
295 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
298 mutex_unlock(&ci
->i_fragtree_mutex
);
302 static int ceph_fill_fragtree(struct inode
*inode
,
303 struct ceph_frag_tree_head
*fragtree
,
304 struct ceph_mds_reply_dirfrag
*dirinfo
)
306 struct ceph_inode_info
*ci
= ceph_inode(inode
);
307 struct ceph_inode_frag
*frag
;
308 struct rb_node
*rb_node
;
313 mutex_lock(&ci
->i_fragtree_mutex
);
314 nsplits
= le32_to_cpu(fragtree
->nsplits
);
316 i
= prandom_u32() % nsplits
;
317 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
318 if (!__ceph_find_frag(ci
, id
))
320 } else if (!RB_EMPTY_ROOT(&ci
->i_fragtree
)) {
321 rb_node
= rb_first(&ci
->i_fragtree
);
322 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
323 if (frag
->frag
!= ceph_frag_make(0, 0) || rb_next(rb_node
))
326 if (!update
&& dirinfo
) {
327 id
= le32_to_cpu(dirinfo
->frag
);
328 if (id
!= __ceph_choose_frag(ci
, id
, NULL
, NULL
))
334 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode
));
335 rb_node
= rb_first(&ci
->i_fragtree
);
336 for (i
= 0; i
< nsplits
; i
++) {
337 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
340 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
341 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
342 if (frag
->frag
!= id
)
345 rb_node
= rb_next(rb_node
);
348 rb_node
= rb_next(rb_node
);
349 rb_erase(&frag
->node
, &ci
->i_fragtree
);
354 frag
= __get_or_create_frag(ci
, id
);
358 frag
->split_by
= le32_to_cpu(fragtree
->splits
[i
].by
);
359 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
362 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
363 rb_node
= rb_next(rb_node
);
364 rb_erase(&frag
->node
, &ci
->i_fragtree
);
368 mutex_unlock(&ci
->i_fragtree_mutex
);
373 * initialize a newly allocated inode.
375 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
377 struct ceph_inode_info
*ci
;
380 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
384 dout("alloc_inode %p\n", &ci
->vfs_inode
);
386 spin_lock_init(&ci
->i_ceph_lock
);
389 ci
->i_inline_version
= 0;
390 ci
->i_time_warp_seq
= 0;
391 ci
->i_ceph_flags
= 0;
392 atomic64_set(&ci
->i_ordered_count
, 1);
393 atomic64_set(&ci
->i_release_count
, 1);
394 atomic64_set(&ci
->i_complete_seq
[0], 0);
395 atomic64_set(&ci
->i_complete_seq
[1], 0);
396 ci
->i_symlink
= NULL
;
398 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
400 ci
->i_fragtree
= RB_ROOT
;
401 mutex_init(&ci
->i_fragtree_mutex
);
403 ci
->i_xattrs
.blob
= NULL
;
404 ci
->i_xattrs
.prealloc_blob
= NULL
;
405 ci
->i_xattrs
.dirty
= false;
406 ci
->i_xattrs
.index
= RB_ROOT
;
407 ci
->i_xattrs
.count
= 0;
408 ci
->i_xattrs
.names_size
= 0;
409 ci
->i_xattrs
.vals_size
= 0;
410 ci
->i_xattrs
.version
= 0;
411 ci
->i_xattrs
.index_version
= 0;
413 ci
->i_caps
= RB_ROOT
;
414 ci
->i_auth_cap
= NULL
;
415 ci
->i_dirty_caps
= 0;
416 ci
->i_flushing_caps
= 0;
417 INIT_LIST_HEAD(&ci
->i_dirty_item
);
418 INIT_LIST_HEAD(&ci
->i_flushing_item
);
419 ci
->i_prealloc_cap_flush
= NULL
;
420 ci
->i_cap_flush_tree
= RB_ROOT
;
421 init_waitqueue_head(&ci
->i_cap_wq
);
422 ci
->i_hold_caps_min
= 0;
423 ci
->i_hold_caps_max
= 0;
424 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
425 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
426 ci
->i_head_snapc
= NULL
;
429 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
430 ci
->i_nr_by_mode
[i
] = 0;
432 mutex_init(&ci
->i_truncate_mutex
);
433 ci
->i_truncate_seq
= 0;
434 ci
->i_truncate_size
= 0;
435 ci
->i_truncate_pending
= 0;
438 ci
->i_reported_size
= 0;
439 ci
->i_wanted_max_size
= 0;
440 ci
->i_requested_max_size
= 0;
444 ci
->i_rdcache_ref
= 0;
447 ci
->i_wrbuffer_ref
= 0;
448 ci
->i_wrbuffer_ref_head
= 0;
449 ci
->i_shared_gen
= 0;
450 ci
->i_rdcache_gen
= 0;
451 ci
->i_rdcache_revoking
= 0;
453 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
454 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
455 spin_lock_init(&ci
->i_unsafe_lock
);
457 ci
->i_snap_realm
= NULL
;
458 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
459 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
461 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
462 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
464 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
466 ceph_fscache_inode_init(ci
);
468 return &ci
->vfs_inode
;
471 static void ceph_i_callback(struct rcu_head
*head
)
473 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
474 struct ceph_inode_info
*ci
= ceph_inode(inode
);
476 kmem_cache_free(ceph_inode_cachep
, ci
);
479 void ceph_destroy_inode(struct inode
*inode
)
481 struct ceph_inode_info
*ci
= ceph_inode(inode
);
482 struct ceph_inode_frag
*frag
;
485 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
487 ceph_fscache_unregister_inode_cookie(ci
);
489 ceph_queue_caps_release(inode
);
492 * we may still have a snap_realm reference if there are stray
493 * caps in i_snap_caps.
495 if (ci
->i_snap_realm
) {
496 struct ceph_mds_client
*mdsc
=
497 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
498 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
500 dout(" dropping residual ref to snap realm %p\n", realm
);
501 spin_lock(&realm
->inodes_with_caps_lock
);
502 list_del_init(&ci
->i_snap_realm_item
);
503 spin_unlock(&realm
->inodes_with_caps_lock
);
504 ceph_put_snap_realm(mdsc
, realm
);
507 kfree(ci
->i_symlink
);
508 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
509 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
510 rb_erase(n
, &ci
->i_fragtree
);
514 __ceph_destroy_xattrs(ci
);
515 if (ci
->i_xattrs
.blob
)
516 ceph_buffer_put(ci
->i_xattrs
.blob
);
517 if (ci
->i_xattrs
.prealloc_blob
)
518 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
520 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
523 int ceph_drop_inode(struct inode
*inode
)
526 * Positve dentry and corresponding inode are always accompanied
527 * in MDS reply. So no need to keep inode in the cache after
528 * dropping all its aliases.
534 * Helpers to fill in size, ctime, mtime, and atime. We have to be
535 * careful because either the client or MDS may have more up to date
536 * info, depending on which capabilities are held, and whether
537 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
538 * and size are monotonically increasing, except when utimes() or
539 * truncate() increments the corresponding _seq values.)
541 int ceph_fill_file_size(struct inode
*inode
, int issued
,
542 u32 truncate_seq
, u64 truncate_size
, u64 size
)
544 struct ceph_inode_info
*ci
= ceph_inode(inode
);
547 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
548 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
549 dout("size %lld -> %llu\n", inode
->i_size
, size
);
550 inode
->i_size
= size
;
551 inode
->i_blocks
= (size
+ (1<<9) - 1) >> 9;
552 ci
->i_reported_size
= size
;
553 if (truncate_seq
!= ci
->i_truncate_seq
) {
554 dout("truncate_seq %u -> %u\n",
555 ci
->i_truncate_seq
, truncate_seq
);
556 ci
->i_truncate_seq
= truncate_seq
;
558 /* the MDS should have revoked these caps */
559 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
562 CEPH_CAP_FILE_LAZYIO
));
564 * If we hold relevant caps, or in the case where we're
565 * not the only client referencing this file and we
566 * don't hold those caps, then we need to check whether
567 * the file is either opened or mmaped
569 if ((issued
& (CEPH_CAP_FILE_CACHE
|
570 CEPH_CAP_FILE_BUFFER
)) ||
571 mapping_mapped(inode
->i_mapping
) ||
572 __ceph_caps_file_wanted(ci
)) {
573 ci
->i_truncate_pending
++;
578 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
579 ci
->i_truncate_size
!= truncate_size
) {
580 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
582 ci
->i_truncate_size
= truncate_size
;
586 ceph_fscache_invalidate(inode
);
591 void ceph_fill_file_time(struct inode
*inode
, int issued
,
592 u64 time_warp_seq
, struct timespec
*ctime
,
593 struct timespec
*mtime
, struct timespec
*atime
)
595 struct ceph_inode_info
*ci
= ceph_inode(inode
);
598 if (issued
& (CEPH_CAP_FILE_EXCL
|
600 CEPH_CAP_FILE_BUFFER
|
602 CEPH_CAP_XATTR_EXCL
)) {
603 if (timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
604 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
605 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
606 ctime
->tv_sec
, ctime
->tv_nsec
);
607 inode
->i_ctime
= *ctime
;
609 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
610 /* the MDS did a utimes() */
611 dout("mtime %ld.%09ld -> %ld.%09ld "
613 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
614 mtime
->tv_sec
, mtime
->tv_nsec
,
615 ci
->i_time_warp_seq
, (int)time_warp_seq
);
617 inode
->i_mtime
= *mtime
;
618 inode
->i_atime
= *atime
;
619 ci
->i_time_warp_seq
= time_warp_seq
;
620 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
621 /* nobody did utimes(); take the max */
622 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
623 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
624 inode
->i_mtime
.tv_sec
,
625 inode
->i_mtime
.tv_nsec
,
626 mtime
->tv_sec
, mtime
->tv_nsec
);
627 inode
->i_mtime
= *mtime
;
629 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
630 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
631 inode
->i_atime
.tv_sec
,
632 inode
->i_atime
.tv_nsec
,
633 atime
->tv_sec
, atime
->tv_nsec
);
634 inode
->i_atime
= *atime
;
636 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
637 /* we did a utimes(); ignore mds values */
642 /* we have no write|excl caps; whatever the MDS says is true */
643 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
644 inode
->i_ctime
= *ctime
;
645 inode
->i_mtime
= *mtime
;
646 inode
->i_atime
= *atime
;
647 ci
->i_time_warp_seq
= time_warp_seq
;
652 if (warn
) /* time_warp_seq shouldn't go backwards */
653 dout("%p mds time_warp_seq %llu < %u\n",
654 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
658 * Populate an inode based on info from mds. May be called on new or
661 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
662 struct ceph_mds_reply_info_in
*iinfo
,
663 struct ceph_mds_reply_dirfrag
*dirinfo
,
664 struct ceph_mds_session
*session
,
665 unsigned long ttl_from
, int cap_fmode
,
666 struct ceph_cap_reservation
*caps_reservation
)
668 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
669 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
670 struct ceph_inode_info
*ci
= ceph_inode(inode
);
671 int issued
= 0, implemented
, new_issued
;
672 struct timespec mtime
, atime
, ctime
;
673 struct ceph_buffer
*xattr_blob
= NULL
;
674 struct ceph_cap
*new_cap
= NULL
;
677 bool queue_trunc
= false;
678 bool new_version
= false;
679 bool fill_inline
= false;
681 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
682 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
685 /* prealloc new cap struct */
686 if (info
->cap
.caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
687 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
690 * prealloc xattr data, if it looks like we'll need it. only
691 * if len > 4 (meaning there are actually xattrs; the first 4
692 * bytes are the xattr count).
694 if (iinfo
->xattr_len
> 4) {
695 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
697 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
701 spin_lock(&ci
->i_ceph_lock
);
704 * provided version will be odd if inode value is projected,
705 * even if stable. skip the update if we have newer stable
706 * info (ours>=theirs, e.g. due to racing mds replies), unless
707 * we are getting projected (unstable) info (in which case the
708 * version is odd, and we want ours>theirs).
714 if (ci
->i_version
== 0 ||
715 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
716 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
719 issued
= __ceph_caps_issued(ci
, &implemented
);
720 issued
|= implemented
| __ceph_caps_dirty(ci
);
721 new_issued
= ~issued
& le32_to_cpu(info
->cap
.caps
);
724 ci
->i_version
= le64_to_cpu(info
->version
);
726 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
727 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
729 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
730 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
731 inode
->i_mode
= le32_to_cpu(info
->mode
);
732 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
733 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
734 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
735 from_kuid(&init_user_ns
, inode
->i_uid
),
736 from_kgid(&init_user_ns
, inode
->i_gid
));
739 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
740 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
741 set_nlink(inode
, le32_to_cpu(info
->nlink
));
743 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
744 /* be careful with mtime, atime, size */
745 ceph_decode_timespec(&atime
, &info
->atime
);
746 ceph_decode_timespec(&mtime
, &info
->mtime
);
747 ceph_decode_timespec(&ctime
, &info
->ctime
);
748 ceph_fill_file_time(inode
, issued
,
749 le32_to_cpu(info
->time_warp_seq
),
750 &ctime
, &mtime
, &atime
);
754 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
755 if (ci
->i_layout
.fl_pg_pool
!= info
->layout
.fl_pg_pool
)
756 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
757 ci
->i_layout
= info
->layout
;
759 queue_trunc
= ceph_fill_file_size(inode
, issued
,
760 le32_to_cpu(info
->truncate_seq
),
761 le64_to_cpu(info
->truncate_size
),
762 le64_to_cpu(info
->size
));
763 /* only update max_size on auth cap */
764 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
765 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
766 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
767 le64_to_cpu(info
->max_size
));
768 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
773 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
774 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
775 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
776 if (ci
->i_xattrs
.blob
)
777 ceph_buffer_put(ci
->i_xattrs
.blob
);
778 ci
->i_xattrs
.blob
= xattr_blob
;
780 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
781 iinfo
->xattr_data
, iinfo
->xattr_len
);
782 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
783 ceph_forget_all_cached_acls(inode
);
787 inode
->i_mapping
->a_ops
= &ceph_aops
;
789 switch (inode
->i_mode
& S_IFMT
) {
794 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
795 inode
->i_op
= &ceph_file_iops
;
798 inode
->i_op
= &ceph_file_iops
;
799 inode
->i_fop
= &ceph_file_fops
;
802 inode
->i_op
= &ceph_symlink_iops
;
803 if (!ci
->i_symlink
) {
804 u32 symlen
= iinfo
->symlink_len
;
807 spin_unlock(&ci
->i_ceph_lock
);
810 if (WARN_ON(symlen
!= inode
->i_size
))
814 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
818 spin_lock(&ci
->i_ceph_lock
);
822 kfree(sym
); /* lost a race */
824 inode
->i_link
= ci
->i_symlink
;
827 inode
->i_op
= &ceph_dir_iops
;
828 inode
->i_fop
= &ceph_dir_fops
;
830 ci
->i_dir_layout
= iinfo
->dir_layout
;
832 ci
->i_files
= le64_to_cpu(info
->files
);
833 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
834 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
835 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
836 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
837 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
840 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
841 ceph_vinop(inode
), inode
->i_mode
);
844 /* were we issued a capability? */
845 if (info
->cap
.caps
) {
846 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
847 unsigned caps
= le32_to_cpu(info
->cap
.caps
);
848 ceph_add_cap(inode
, session
,
849 le64_to_cpu(info
->cap
.cap_id
),
851 le32_to_cpu(info
->cap
.wanted
),
852 le32_to_cpu(info
->cap
.seq
),
853 le32_to_cpu(info
->cap
.mseq
),
854 le64_to_cpu(info
->cap
.realm
),
855 info
->cap
.flags
, &new_cap
);
857 /* set dir completion flag? */
858 if (S_ISDIR(inode
->i_mode
) &&
859 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
860 (caps
& CEPH_CAP_FILE_SHARED
) &&
861 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
862 !__ceph_dir_is_complete(ci
)) {
863 dout(" marking %p complete (empty)\n", inode
);
864 i_size_write(inode
, 0);
865 __ceph_dir_set_complete(ci
,
866 atomic64_read(&ci
->i_release_count
),
867 atomic64_read(&ci
->i_ordered_count
));
872 dout(" %p got snap_caps %s\n", inode
,
873 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
874 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
876 __ceph_get_fmode(ci
, cap_fmode
);
878 } else if (cap_fmode
>= 0) {
879 pr_warn("mds issued no caps on %llx.%llx\n",
881 __ceph_get_fmode(ci
, cap_fmode
);
884 if (iinfo
->inline_version
> 0 &&
885 iinfo
->inline_version
>= ci
->i_inline_version
) {
886 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
887 ci
->i_inline_version
= iinfo
->inline_version
;
888 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
890 (le32_to_cpu(info
->cap
.caps
) & cache_caps
)))
894 spin_unlock(&ci
->i_ceph_lock
);
897 ceph_fill_inline_data(inode
, locked_page
,
898 iinfo
->inline_data
, iinfo
->inline_len
);
901 wake_up_all(&ci
->i_cap_wq
);
903 /* queue truncate if we saw i_size decrease */
905 ceph_queue_vmtruncate(inode
);
907 /* populate frag tree */
908 if (S_ISDIR(inode
->i_mode
))
909 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
911 /* update delegation info? */
913 ceph_fill_dirfrag(inode
, dirinfo
);
918 ceph_put_cap(mdsc
, new_cap
);
920 ceph_buffer_put(xattr_blob
);
925 * caller should hold session s_mutex.
927 static void update_dentry_lease(struct dentry
*dentry
,
928 struct ceph_mds_reply_lease
*lease
,
929 struct ceph_mds_session
*session
,
930 unsigned long from_time
)
932 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
933 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
934 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
935 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
938 /* only track leases on regular dentries */
939 if (dentry
->d_op
!= &ceph_dentry_ops
)
942 spin_lock(&dentry
->d_lock
);
943 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
944 dentry
, duration
, ttl
);
946 /* make lease_rdcache_gen match directory */
947 dir
= d_inode(dentry
->d_parent
);
948 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
953 if (di
->lease_gen
== session
->s_cap_gen
&&
954 time_before(ttl
, dentry
->d_time
))
955 goto out_unlock
; /* we already have a newer lease. */
957 if (di
->lease_session
&& di
->lease_session
!= session
)
960 ceph_dentry_lru_touch(dentry
);
962 if (!di
->lease_session
)
963 di
->lease_session
= ceph_get_mds_session(session
);
964 di
->lease_gen
= session
->s_cap_gen
;
965 di
->lease_seq
= le32_to_cpu(lease
->seq
);
966 di
->lease_renew_after
= half_ttl
;
967 di
->lease_renew_from
= 0;
968 dentry
->d_time
= ttl
;
970 spin_unlock(&dentry
->d_lock
);
975 * splice a dentry to an inode.
976 * caller must hold directory i_mutex for this to be safe.
978 * we will only rehash the resulting dentry if @prehash is
979 * true; @prehash will be set to false (for the benefit of
980 * the caller) if we fail.
982 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
985 struct dentry
*realdn
;
989 /* dn must be unhashed */
992 realdn
= d_splice_alias(in
, dn
);
993 if (IS_ERR(realdn
)) {
994 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
995 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
997 *prehash
= false; /* don't rehash on error */
998 dn
= realdn
; /* note realdn contains the error */
1000 } else if (realdn
) {
1001 dout("dn %p (%d) spliced with %p (%d) "
1002 "inode %p ino %llx.%llx\n",
1004 realdn
, d_count(realdn
),
1005 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1009 BUG_ON(!ceph_dentry(dn
));
1010 dout("dn %p attached to %p ino %llx.%llx\n",
1011 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1013 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
1020 * Incorporate results into the local cache. This is either just
1021 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1024 * A reply may contain
1025 * a directory inode along with a dentry.
1026 * and/or a target inode
1028 * Called with snap_rwsem (read).
1030 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
1031 struct ceph_mds_session
*session
)
1033 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1034 struct inode
*in
= NULL
;
1035 struct ceph_vino vino
;
1036 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1039 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1040 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1046 * If we resend completed ops to a recovering mds, we get no
1047 * trace. Since that is very rare, pretend this is the case
1048 * to ensure the 'no trace' handlers in the callers behave.
1050 * Fill in inodes unconditionally to avoid breaking cap
1053 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
1054 pr_info("fill_trace faking empty trace on %lld %s\n",
1055 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
1056 if (rinfo
->head
->is_dentry
) {
1057 rinfo
->head
->is_dentry
= 0;
1058 err
= fill_inode(req
->r_locked_dir
,
1059 &rinfo
->diri
, rinfo
->dirfrag
,
1060 session
, req
->r_request_started
, -1);
1062 if (rinfo
->head
->is_target
) {
1063 rinfo
->head
->is_target
= 0;
1064 ininfo
= rinfo
->targeti
.in
;
1065 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1066 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1067 in
= ceph_get_inode(sb
, vino
);
1068 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1069 session
, req
->r_request_started
,
1076 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1077 dout("fill_trace reply is empty!\n");
1078 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
1079 ceph_invalidate_dir_request(req
);
1083 if (rinfo
->head
->is_dentry
) {
1084 struct inode
*dir
= req
->r_locked_dir
;
1087 err
= fill_inode(dir
, NULL
,
1088 &rinfo
->diri
, rinfo
->dirfrag
,
1089 session
, req
->r_request_started
, -1,
1090 &req
->r_caps_reservation
);
1097 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1099 struct dentry
*dn
, *parent
;
1101 BUG_ON(!rinfo
->head
->is_target
);
1102 BUG_ON(req
->r_dentry
);
1104 parent
= d_find_any_alias(dir
);
1107 dname
.name
= rinfo
->dname
;
1108 dname
.len
= rinfo
->dname_len
;
1109 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1110 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1111 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1113 dn
= d_lookup(parent
, &dname
);
1114 dout("d_lookup on parent=%p name=%.*s got %p\n",
1115 parent
, dname
.len
, dname
.name
, dn
);
1118 dn
= d_alloc(parent
, &dname
);
1119 dout("d_alloc %p '%.*s' = %p\n", parent
,
1120 dname
.len
, dname
.name
, dn
);
1126 err
= ceph_init_dentry(dn
);
1132 } else if (d_really_is_positive(dn
) &&
1133 (ceph_ino(d_inode(dn
)) != vino
.ino
||
1134 ceph_snap(d_inode(dn
)) != vino
.snap
)) {
1135 dout(" dn %p points to wrong inode %p\n",
1147 if (rinfo
->head
->is_target
) {
1148 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1149 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1151 in
= ceph_get_inode(sb
, vino
);
1156 req
->r_target_inode
= in
;
1158 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1159 session
, req
->r_request_started
,
1160 (!req
->r_aborted
&& rinfo
->head
->result
== 0) ?
1162 &req
->r_caps_reservation
);
1164 pr_err("fill_inode badness %p %llx.%llx\n",
1165 in
, ceph_vinop(in
));
1171 * ignore null lease/binding on snapdir ENOENT, or else we
1172 * will have trouble splicing in the virtual snapdir later
1174 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1175 req
->r_locked_dir
&&
1176 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1177 fsc
->mount_options
->snapdir_name
,
1178 req
->r_dentry
->d_name
.len
))) {
1180 * lookup link rename : null -> possibly existing inode
1181 * mknod symlink mkdir : null -> new inode
1182 * unlink : linked -> null
1184 struct inode
*dir
= req
->r_locked_dir
;
1185 struct dentry
*dn
= req
->r_dentry
;
1186 bool have_dir_cap
, have_lease
;
1190 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1191 BUG_ON(ceph_ino(dir
) !=
1192 le64_to_cpu(rinfo
->diri
.in
->ino
));
1193 BUG_ON(ceph_snap(dir
) !=
1194 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1196 /* do we have a lease on the whole dir? */
1198 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1199 CEPH_CAP_FILE_SHARED
);
1201 /* do we have a dn lease? */
1202 have_lease
= have_dir_cap
||
1203 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1205 dout("fill_trace no dentry lease or dir cap\n");
1208 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1209 struct inode
*olddir
= req
->r_old_dentry_dir
;
1212 dout(" src %p '%pd' dst %p '%pd'\n",
1216 dout("fill_trace doing d_move %p -> %p\n",
1217 req
->r_old_dentry
, dn
);
1219 /* d_move screws up sibling dentries' offsets */
1220 ceph_dir_clear_ordered(dir
);
1221 ceph_dir_clear_ordered(olddir
);
1223 d_move(req
->r_old_dentry
, dn
);
1224 dout(" src %p '%pd' dst %p '%pd'\n",
1229 /* ensure target dentry is invalidated, despite
1230 rehashing bug in vfs_rename_dir */
1231 ceph_invalidate_dentry_lease(dn
);
1233 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1234 ceph_dentry(req
->r_old_dentry
)->offset
);
1236 dn
= req
->r_old_dentry
; /* use old_dentry */
1240 if (!rinfo
->head
->is_target
) {
1241 dout("fill_trace null dentry\n");
1242 if (d_really_is_positive(dn
)) {
1243 ceph_dir_clear_ordered(dir
);
1244 dout("d_delete %p\n", dn
);
1247 dout("d_instantiate %p NULL\n", dn
);
1248 d_instantiate(dn
, NULL
);
1249 if (have_lease
&& d_unhashed(dn
))
1251 update_dentry_lease(dn
, rinfo
->dlease
,
1253 req
->r_request_started
);
1258 /* attach proper inode */
1259 if (d_really_is_negative(dn
)) {
1260 ceph_dir_clear_ordered(dir
);
1262 dn
= splice_dentry(dn
, in
, &have_lease
);
1267 req
->r_dentry
= dn
; /* may have spliced */
1268 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1269 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1270 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1276 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1277 req
->r_request_started
);
1278 dout(" final dn %p\n", dn
);
1279 } else if (!req
->r_aborted
&&
1280 (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1281 req
->r_op
== CEPH_MDS_OP_MKSNAP
)) {
1282 struct dentry
*dn
= req
->r_dentry
;
1283 struct inode
*dir
= req
->r_locked_dir
;
1285 /* fill out a snapdir LOOKUPSNAP dentry */
1288 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1289 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1290 ceph_dir_clear_ordered(dir
);
1292 dn
= splice_dentry(dn
, in
, NULL
);
1297 req
->r_dentry
= dn
; /* may have spliced */
1300 dout("fill_trace done err=%d\n", err
);
1305 * Prepopulate our cache with readdir results, leases, etc.
1307 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1308 struct ceph_mds_session
*session
)
1310 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1313 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1314 struct ceph_vino vino
;
1318 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1319 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1321 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1324 dout("new_inode badness got %d\n", err
);
1327 rc
= fill_inode(in
, NULL
, &rinfo
->dir_in
[i
], NULL
, session
,
1328 req
->r_request_started
, -1,
1329 &req
->r_caps_reservation
);
1331 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1340 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1344 page_cache_release(ctl
->page
);
1349 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1350 struct ceph_readdir_cache_control
*ctl
,
1351 struct ceph_mds_request
*req
)
1353 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1354 unsigned nsize
= PAGE_CACHE_SIZE
/ sizeof(struct dentry
*);
1355 unsigned idx
= ctl
->index
% nsize
;
1356 pgoff_t pgoff
= ctl
->index
/ nsize
;
1358 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1359 ceph_readdir_cache_release(ctl
);
1360 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1365 /* reading/filling the cache are serialized by
1366 * i_mutex, no need to use page lock */
1367 unlock_page(ctl
->page
);
1368 ctl
->dentries
= kmap(ctl
->page
);
1371 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1372 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1373 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1374 ctl
->dentries
[idx
] = dn
;
1377 dout("disable readdir cache\n");
1383 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1384 struct ceph_mds_session
*session
)
1386 struct dentry
*parent
= req
->r_dentry
;
1387 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1391 int err
= 0, ret
, i
;
1392 struct inode
*snapdir
= NULL
;
1393 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1394 struct ceph_dentry_info
*di
;
1395 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1396 struct ceph_readdir_cache_control cache_ctl
= {};
1399 return readdir_prepopulate_inodes_only(req
, session
);
1401 if (rinfo
->dir_dir
&&
1402 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1403 dout("readdir_prepopulate got new frag %x -> %x\n",
1404 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1405 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1406 if (ceph_frag_is_leftmost(frag
))
1407 req
->r_readdir_offset
= 2;
1409 req
->r_readdir_offset
= 0;
1412 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1413 snapdir
= ceph_get_snapdir(d_inode(parent
));
1414 parent
= d_find_alias(snapdir
);
1415 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1416 rinfo
->dir_nr
, parent
);
1418 dout("readdir_prepopulate %d items under dn %p\n",
1419 rinfo
->dir_nr
, parent
);
1421 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1424 if (ceph_frag_is_leftmost(frag
) && req
->r_readdir_offset
== 2) {
1425 /* note dir version at start of readdir so we can tell
1426 * if any dentries get dropped */
1427 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1428 req
->r_dir_release_cnt
= atomic64_read(&ci
->i_release_count
);
1429 req
->r_dir_ordered_cnt
= atomic64_read(&ci
->i_ordered_count
);
1430 req
->r_readdir_cache_idx
= 0;
1433 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1435 /* FIXME: release caps/leases if error occurs */
1436 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1437 struct ceph_vino vino
;
1439 dname
.name
= rinfo
->dir_dname
[i
];
1440 dname
.len
= rinfo
->dir_dname_len
[i
];
1441 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1443 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1444 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1447 dn
= d_lookup(parent
, &dname
);
1448 dout("d_lookup on parent=%p name=%.*s got %p\n",
1449 parent
, dname
.len
, dname
.name
, dn
);
1452 dn
= d_alloc(parent
, &dname
);
1453 dout("d_alloc %p '%.*s' = %p\n", parent
,
1454 dname
.len
, dname
.name
, dn
);
1456 dout("d_alloc badness\n");
1460 ret
= ceph_init_dentry(dn
);
1466 } else if (d_really_is_positive(dn
) &&
1467 (ceph_ino(d_inode(dn
)) != vino
.ino
||
1468 ceph_snap(d_inode(dn
)) != vino
.snap
)) {
1469 dout(" dn %p points to wrong inode %p\n",
1477 if (d_really_is_positive(dn
)) {
1480 in
= ceph_get_inode(parent
->d_sb
, vino
);
1482 dout("new_inode badness\n");
1490 ret
= fill_inode(in
, NULL
, &rinfo
->dir_in
[i
], NULL
, session
,
1491 req
->r_request_started
, -1,
1492 &req
->r_caps_reservation
);
1494 pr_err("fill_inode badness on %p\n", in
);
1495 if (d_really_is_negative(dn
))
1502 if (d_really_is_negative(dn
)) {
1503 struct dentry
*realdn
= splice_dentry(dn
, in
, NULL
);
1504 if (IS_ERR(realdn
)) {
1505 err
= PTR_ERR(realdn
);
1514 di
->offset
= ceph_make_fpos(frag
, i
+ req
->r_readdir_offset
);
1516 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1518 req
->r_request_started
);
1520 if (err
== 0 && cache_ctl
.index
>= 0) {
1521 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1532 req
->r_did_prepopulate
= true;
1533 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1535 ceph_readdir_cache_release(&cache_ctl
);
1540 dout("readdir_prepopulate done\n");
1544 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1546 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1549 spin_lock(&ci
->i_ceph_lock
);
1550 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1551 inode
->i_size
= size
;
1552 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1554 /* tell the MDS if we are approaching max_size */
1555 if ((size
<< 1) >= ci
->i_max_size
&&
1556 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1559 spin_unlock(&ci
->i_ceph_lock
);
1564 * Write back inode data in a worker thread. (This can't be done
1565 * in the message handler context.)
1567 void ceph_queue_writeback(struct inode
*inode
)
1570 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1571 &ceph_inode(inode
)->i_wb_work
)) {
1572 dout("ceph_queue_writeback %p\n", inode
);
1574 dout("ceph_queue_writeback %p failed\n", inode
);
1579 static void ceph_writeback_work(struct work_struct
*work
)
1581 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1583 struct inode
*inode
= &ci
->vfs_inode
;
1585 dout("writeback %p\n", inode
);
1586 filemap_fdatawrite(&inode
->i_data
);
1591 * queue an async invalidation
1593 void ceph_queue_invalidate(struct inode
*inode
)
1596 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1597 &ceph_inode(inode
)->i_pg_inv_work
)) {
1598 dout("ceph_queue_invalidate %p\n", inode
);
1600 dout("ceph_queue_invalidate %p failed\n", inode
);
1606 * Invalidate inode pages in a worker thread. (This can't be done
1607 * in the message handler context.)
1609 static void ceph_invalidate_work(struct work_struct
*work
)
1611 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1613 struct inode
*inode
= &ci
->vfs_inode
;
1617 mutex_lock(&ci
->i_truncate_mutex
);
1618 spin_lock(&ci
->i_ceph_lock
);
1619 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1620 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1621 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1622 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1624 spin_unlock(&ci
->i_ceph_lock
);
1625 mutex_unlock(&ci
->i_truncate_mutex
);
1628 orig_gen
= ci
->i_rdcache_gen
;
1629 spin_unlock(&ci
->i_ceph_lock
);
1631 truncate_pagecache(inode
, 0);
1633 spin_lock(&ci
->i_ceph_lock
);
1634 if (orig_gen
== ci
->i_rdcache_gen
&&
1635 orig_gen
== ci
->i_rdcache_revoking
) {
1636 dout("invalidate_pages %p gen %d successful\n", inode
,
1638 ci
->i_rdcache_revoking
--;
1641 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1642 inode
, orig_gen
, ci
->i_rdcache_gen
,
1643 ci
->i_rdcache_revoking
);
1644 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1647 spin_unlock(&ci
->i_ceph_lock
);
1648 mutex_unlock(&ci
->i_truncate_mutex
);
1651 ceph_check_caps(ci
, 0, NULL
);
1657 * called by trunc_wq;
1659 * We also truncate in a separate thread as well.
1661 static void ceph_vmtruncate_work(struct work_struct
*work
)
1663 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1665 struct inode
*inode
= &ci
->vfs_inode
;
1667 dout("vmtruncate_work %p\n", inode
);
1668 __ceph_do_pending_vmtruncate(inode
);
1673 * Queue an async vmtruncate. If we fail to queue work, we will handle
1674 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1676 void ceph_queue_vmtruncate(struct inode
*inode
)
1678 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1682 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1683 &ci
->i_vmtruncate_work
)) {
1684 dout("ceph_queue_vmtruncate %p\n", inode
);
1686 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1687 inode
, ci
->i_truncate_pending
);
1693 * Make sure any pending truncation is applied before doing anything
1694 * that may depend on it.
1696 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1698 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1700 int wrbuffer_refs
, finish
= 0;
1702 mutex_lock(&ci
->i_truncate_mutex
);
1704 spin_lock(&ci
->i_ceph_lock
);
1705 if (ci
->i_truncate_pending
== 0) {
1706 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1707 spin_unlock(&ci
->i_ceph_lock
);
1708 mutex_unlock(&ci
->i_truncate_mutex
);
1713 * make sure any dirty snapped pages are flushed before we
1714 * possibly truncate them.. so write AND block!
1716 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1717 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1719 spin_unlock(&ci
->i_ceph_lock
);
1720 filemap_write_and_wait_range(&inode
->i_data
, 0,
1721 inode
->i_sb
->s_maxbytes
);
1725 /* there should be no reader or writer */
1726 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1728 to
= ci
->i_truncate_size
;
1729 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1730 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1731 ci
->i_truncate_pending
, to
);
1732 spin_unlock(&ci
->i_ceph_lock
);
1734 truncate_pagecache(inode
, to
);
1736 spin_lock(&ci
->i_ceph_lock
);
1737 if (to
== ci
->i_truncate_size
) {
1738 ci
->i_truncate_pending
= 0;
1741 spin_unlock(&ci
->i_ceph_lock
);
1745 mutex_unlock(&ci
->i_truncate_mutex
);
1747 if (wrbuffer_refs
== 0)
1748 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1750 wake_up_all(&ci
->i_cap_wq
);
1756 static const struct inode_operations ceph_symlink_iops
= {
1757 .readlink
= generic_readlink
,
1758 .follow_link
= simple_follow_link
,
1759 .setattr
= ceph_setattr
,
1760 .getattr
= ceph_getattr
,
1761 .setxattr
= ceph_setxattr
,
1762 .getxattr
= ceph_getxattr
,
1763 .listxattr
= ceph_listxattr
,
1764 .removexattr
= ceph_removexattr
,
1770 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1772 struct inode
*inode
= d_inode(dentry
);
1773 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1774 const unsigned int ia_valid
= attr
->ia_valid
;
1775 struct ceph_mds_request
*req
;
1776 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1777 struct ceph_cap_flush
*prealloc_cf
;
1779 int release
= 0, dirtied
= 0;
1782 int inode_dirty_flags
= 0;
1783 bool lock_snap_rwsem
= false;
1785 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1788 err
= inode_change_ok(inode
, attr
);
1792 prealloc_cf
= ceph_alloc_cap_flush();
1796 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1799 ceph_free_cap_flush(prealloc_cf
);
1800 return PTR_ERR(req
);
1803 spin_lock(&ci
->i_ceph_lock
);
1804 issued
= __ceph_caps_issued(ci
, NULL
);
1806 if (!ci
->i_head_snapc
&&
1807 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1808 lock_snap_rwsem
= true;
1809 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1810 spin_unlock(&ci
->i_ceph_lock
);
1811 down_read(&mdsc
->snap_rwsem
);
1812 spin_lock(&ci
->i_ceph_lock
);
1813 issued
= __ceph_caps_issued(ci
, NULL
);
1817 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1819 if (ia_valid
& ATTR_UID
) {
1820 dout("setattr %p uid %d -> %d\n", inode
,
1821 from_kuid(&init_user_ns
, inode
->i_uid
),
1822 from_kuid(&init_user_ns
, attr
->ia_uid
));
1823 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1824 inode
->i_uid
= attr
->ia_uid
;
1825 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1826 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1827 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1828 req
->r_args
.setattr
.uid
= cpu_to_le32(
1829 from_kuid(&init_user_ns
, attr
->ia_uid
));
1830 mask
|= CEPH_SETATTR_UID
;
1831 release
|= CEPH_CAP_AUTH_SHARED
;
1834 if (ia_valid
& ATTR_GID
) {
1835 dout("setattr %p gid %d -> %d\n", inode
,
1836 from_kgid(&init_user_ns
, inode
->i_gid
),
1837 from_kgid(&init_user_ns
, attr
->ia_gid
));
1838 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1839 inode
->i_gid
= attr
->ia_gid
;
1840 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1841 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1842 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1843 req
->r_args
.setattr
.gid
= cpu_to_le32(
1844 from_kgid(&init_user_ns
, attr
->ia_gid
));
1845 mask
|= CEPH_SETATTR_GID
;
1846 release
|= CEPH_CAP_AUTH_SHARED
;
1849 if (ia_valid
& ATTR_MODE
) {
1850 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1852 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1853 inode
->i_mode
= attr
->ia_mode
;
1854 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1855 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1856 attr
->ia_mode
!= inode
->i_mode
) {
1857 inode
->i_mode
= attr
->ia_mode
;
1858 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1859 mask
|= CEPH_SETATTR_MODE
;
1860 release
|= CEPH_CAP_AUTH_SHARED
;
1864 if (ia_valid
& ATTR_ATIME
) {
1865 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1866 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1867 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1868 if (issued
& CEPH_CAP_FILE_EXCL
) {
1869 ci
->i_time_warp_seq
++;
1870 inode
->i_atime
= attr
->ia_atime
;
1871 dirtied
|= CEPH_CAP_FILE_EXCL
;
1872 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1873 timespec_compare(&inode
->i_atime
,
1874 &attr
->ia_atime
) < 0) {
1875 inode
->i_atime
= attr
->ia_atime
;
1876 dirtied
|= CEPH_CAP_FILE_WR
;
1877 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1878 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1879 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1881 mask
|= CEPH_SETATTR_ATIME
;
1882 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1886 if (ia_valid
& ATTR_MTIME
) {
1887 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1888 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1889 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1890 if (issued
& CEPH_CAP_FILE_EXCL
) {
1891 ci
->i_time_warp_seq
++;
1892 inode
->i_mtime
= attr
->ia_mtime
;
1893 dirtied
|= CEPH_CAP_FILE_EXCL
;
1894 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1895 timespec_compare(&inode
->i_mtime
,
1896 &attr
->ia_mtime
) < 0) {
1897 inode
->i_mtime
= attr
->ia_mtime
;
1898 dirtied
|= CEPH_CAP_FILE_WR
;
1899 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1900 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1901 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1903 mask
|= CEPH_SETATTR_MTIME
;
1904 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1908 if (ia_valid
& ATTR_SIZE
) {
1909 dout("setattr %p size %lld -> %lld\n", inode
,
1910 inode
->i_size
, attr
->ia_size
);
1911 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1912 attr
->ia_size
> inode
->i_size
) {
1913 inode
->i_size
= attr
->ia_size
;
1915 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1916 inode
->i_ctime
= attr
->ia_ctime
;
1917 ci
->i_reported_size
= attr
->ia_size
;
1918 dirtied
|= CEPH_CAP_FILE_EXCL
;
1919 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1920 attr
->ia_size
!= inode
->i_size
) {
1921 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1922 req
->r_args
.setattr
.old_size
=
1923 cpu_to_le64(inode
->i_size
);
1924 mask
|= CEPH_SETATTR_SIZE
;
1925 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1930 /* these do nothing */
1931 if (ia_valid
& ATTR_CTIME
) {
1932 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1933 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1934 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1935 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1936 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1937 only
? "ctime only" : "ignored");
1938 inode
->i_ctime
= attr
->ia_ctime
;
1941 * if kernel wants to dirty ctime but nothing else,
1942 * we need to choose a cap to dirty under, or do
1943 * a almost-no-op setattr
1945 if (issued
& CEPH_CAP_AUTH_EXCL
)
1946 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1947 else if (issued
& CEPH_CAP_FILE_EXCL
)
1948 dirtied
|= CEPH_CAP_FILE_EXCL
;
1949 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1950 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1952 mask
|= CEPH_SETATTR_CTIME
;
1955 if (ia_valid
& ATTR_FILE
)
1956 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1959 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
1961 inode
->i_ctime
= CURRENT_TIME
;
1965 spin_unlock(&ci
->i_ceph_lock
);
1966 if (lock_snap_rwsem
)
1967 up_read(&mdsc
->snap_rwsem
);
1969 if (inode_dirty_flags
)
1970 __mark_inode_dirty(inode
, inode_dirty_flags
);
1972 if (ia_valid
& ATTR_MODE
) {
1973 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
1979 req
->r_inode
= inode
;
1981 req
->r_inode_drop
= release
;
1982 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1983 req
->r_num_caps
= 1;
1984 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1986 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1987 ceph_cap_string(dirtied
), mask
);
1989 ceph_mdsc_put_request(req
);
1990 if (mask
& CEPH_SETATTR_SIZE
)
1991 __ceph_do_pending_vmtruncate(inode
);
1992 ceph_free_cap_flush(prealloc_cf
);
1995 ceph_mdsc_put_request(req
);
1996 ceph_free_cap_flush(prealloc_cf
);
2001 * Verify that we have a lease on the given mask. If not,
2002 * do a getattr against an mds.
2004 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2005 int mask
, bool force
)
2007 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2008 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2009 struct ceph_mds_request
*req
;
2012 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2013 dout("do_getattr inode %p SNAPDIR\n", inode
);
2017 dout("do_getattr inode %p mask %s mode 0%o\n",
2018 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2019 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2022 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
2024 return PTR_ERR(req
);
2025 req
->r_inode
= inode
;
2027 req
->r_num_caps
= 1;
2028 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2029 req
->r_locked_page
= locked_page
;
2030 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2031 if (locked_page
&& err
== 0) {
2032 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2033 if (inline_version
== 0) {
2034 /* the reply is supposed to contain inline data */
2036 } else if (inline_version
== CEPH_INLINE_NONE
) {
2039 err
= req
->r_reply_info
.targeti
.inline_len
;
2042 ceph_mdsc_put_request(req
);
2043 dout("do_getattr result=%d\n", err
);
2049 * Check inode permissions. We verify we have a valid value for
2050 * the AUTH cap, then call the generic handler.
2052 int ceph_permission(struct inode
*inode
, int mask
)
2056 if (mask
& MAY_NOT_BLOCK
)
2059 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2062 err
= generic_permission(inode
, mask
);
2067 * Get all attributes. Hopefully somedata we'll have a statlite()
2068 * and can limit the fields we require to be accurate.
2070 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
2073 struct inode
*inode
= d_inode(dentry
);
2074 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2077 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2079 generic_fillattr(inode
, stat
);
2080 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2081 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2082 stat
->dev
= ceph_snap(inode
);
2085 if (S_ISDIR(inode
->i_mode
)) {
2086 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2088 stat
->size
= ci
->i_rbytes
;
2090 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2092 stat
->blksize
= 65536;