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
));
399 ci
->i_pool_ns_len
= 0;
401 ci
->i_fragtree
= RB_ROOT
;
402 mutex_init(&ci
->i_fragtree_mutex
);
404 ci
->i_xattrs
.blob
= NULL
;
405 ci
->i_xattrs
.prealloc_blob
= NULL
;
406 ci
->i_xattrs
.dirty
= false;
407 ci
->i_xattrs
.index
= RB_ROOT
;
408 ci
->i_xattrs
.count
= 0;
409 ci
->i_xattrs
.names_size
= 0;
410 ci
->i_xattrs
.vals_size
= 0;
411 ci
->i_xattrs
.version
= 0;
412 ci
->i_xattrs
.index_version
= 0;
414 ci
->i_caps
= RB_ROOT
;
415 ci
->i_auth_cap
= NULL
;
416 ci
->i_dirty_caps
= 0;
417 ci
->i_flushing_caps
= 0;
418 INIT_LIST_HEAD(&ci
->i_dirty_item
);
419 INIT_LIST_HEAD(&ci
->i_flushing_item
);
420 ci
->i_prealloc_cap_flush
= NULL
;
421 ci
->i_cap_flush_tree
= RB_ROOT
;
422 init_waitqueue_head(&ci
->i_cap_wq
);
423 ci
->i_hold_caps_min
= 0;
424 ci
->i_hold_caps_max
= 0;
425 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
426 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
427 ci
->i_head_snapc
= NULL
;
430 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
431 ci
->i_nr_by_mode
[i
] = 0;
433 mutex_init(&ci
->i_truncate_mutex
);
434 ci
->i_truncate_seq
= 0;
435 ci
->i_truncate_size
= 0;
436 ci
->i_truncate_pending
= 0;
439 ci
->i_reported_size
= 0;
440 ci
->i_wanted_max_size
= 0;
441 ci
->i_requested_max_size
= 0;
445 ci
->i_rdcache_ref
= 0;
448 ci
->i_wrbuffer_ref
= 0;
449 ci
->i_wrbuffer_ref_head
= 0;
450 ci
->i_shared_gen
= 0;
451 ci
->i_rdcache_gen
= 0;
452 ci
->i_rdcache_revoking
= 0;
454 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
455 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
456 INIT_LIST_HEAD(&ci
->i_unsafe_iops
);
457 spin_lock_init(&ci
->i_unsafe_lock
);
459 ci
->i_snap_realm
= NULL
;
460 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
461 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
463 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
464 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
466 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
468 ceph_fscache_inode_init(ci
);
470 return &ci
->vfs_inode
;
473 static void ceph_i_callback(struct rcu_head
*head
)
475 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
476 struct ceph_inode_info
*ci
= ceph_inode(inode
);
478 kmem_cache_free(ceph_inode_cachep
, ci
);
481 void ceph_destroy_inode(struct inode
*inode
)
483 struct ceph_inode_info
*ci
= ceph_inode(inode
);
484 struct ceph_inode_frag
*frag
;
487 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
489 ceph_fscache_unregister_inode_cookie(ci
);
491 ceph_queue_caps_release(inode
);
494 * we may still have a snap_realm reference if there are stray
495 * caps in i_snap_caps.
497 if (ci
->i_snap_realm
) {
498 struct ceph_mds_client
*mdsc
=
499 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
500 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
502 dout(" dropping residual ref to snap realm %p\n", realm
);
503 spin_lock(&realm
->inodes_with_caps_lock
);
504 list_del_init(&ci
->i_snap_realm_item
);
505 spin_unlock(&realm
->inodes_with_caps_lock
);
506 ceph_put_snap_realm(mdsc
, realm
);
509 kfree(ci
->i_symlink
);
510 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
511 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
512 rb_erase(n
, &ci
->i_fragtree
);
516 __ceph_destroy_xattrs(ci
);
517 if (ci
->i_xattrs
.blob
)
518 ceph_buffer_put(ci
->i_xattrs
.blob
);
519 if (ci
->i_xattrs
.prealloc_blob
)
520 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
522 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
525 int ceph_drop_inode(struct inode
*inode
)
528 * Positve dentry and corresponding inode are always accompanied
529 * in MDS reply. So no need to keep inode in the cache after
530 * dropping all its aliases.
536 * Helpers to fill in size, ctime, mtime, and atime. We have to be
537 * careful because either the client or MDS may have more up to date
538 * info, depending on which capabilities are held, and whether
539 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
540 * and size are monotonically increasing, except when utimes() or
541 * truncate() increments the corresponding _seq values.)
543 int ceph_fill_file_size(struct inode
*inode
, int issued
,
544 u32 truncate_seq
, u64 truncate_size
, u64 size
)
546 struct ceph_inode_info
*ci
= ceph_inode(inode
);
549 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
550 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
551 dout("size %lld -> %llu\n", inode
->i_size
, size
);
552 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
553 pr_err("fill_file_size non-zero size for directory\n");
556 i_size_write(inode
, size
);
557 inode
->i_blocks
= (size
+ (1<<9) - 1) >> 9;
558 ci
->i_reported_size
= size
;
559 if (truncate_seq
!= ci
->i_truncate_seq
) {
560 dout("truncate_seq %u -> %u\n",
561 ci
->i_truncate_seq
, truncate_seq
);
562 ci
->i_truncate_seq
= truncate_seq
;
564 /* the MDS should have revoked these caps */
565 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
568 CEPH_CAP_FILE_LAZYIO
));
570 * If we hold relevant caps, or in the case where we're
571 * not the only client referencing this file and we
572 * don't hold those caps, then we need to check whether
573 * the file is either opened or mmaped
575 if ((issued
& (CEPH_CAP_FILE_CACHE
|
576 CEPH_CAP_FILE_BUFFER
)) ||
577 mapping_mapped(inode
->i_mapping
) ||
578 __ceph_caps_file_wanted(ci
)) {
579 ci
->i_truncate_pending
++;
584 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
585 ci
->i_truncate_size
!= truncate_size
) {
586 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
588 ci
->i_truncate_size
= truncate_size
;
592 ceph_fscache_invalidate(inode
);
597 void ceph_fill_file_time(struct inode
*inode
, int issued
,
598 u64 time_warp_seq
, struct timespec
*ctime
,
599 struct timespec
*mtime
, struct timespec
*atime
)
601 struct ceph_inode_info
*ci
= ceph_inode(inode
);
604 if (issued
& (CEPH_CAP_FILE_EXCL
|
606 CEPH_CAP_FILE_BUFFER
|
608 CEPH_CAP_XATTR_EXCL
)) {
609 if (timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
610 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
611 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
612 ctime
->tv_sec
, ctime
->tv_nsec
);
613 inode
->i_ctime
= *ctime
;
615 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
616 /* the MDS did a utimes() */
617 dout("mtime %ld.%09ld -> %ld.%09ld "
619 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
620 mtime
->tv_sec
, mtime
->tv_nsec
,
621 ci
->i_time_warp_seq
, (int)time_warp_seq
);
623 inode
->i_mtime
= *mtime
;
624 inode
->i_atime
= *atime
;
625 ci
->i_time_warp_seq
= time_warp_seq
;
626 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
627 /* nobody did utimes(); take the max */
628 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
629 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
630 inode
->i_mtime
.tv_sec
,
631 inode
->i_mtime
.tv_nsec
,
632 mtime
->tv_sec
, mtime
->tv_nsec
);
633 inode
->i_mtime
= *mtime
;
635 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
636 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
637 inode
->i_atime
.tv_sec
,
638 inode
->i_atime
.tv_nsec
,
639 atime
->tv_sec
, atime
->tv_nsec
);
640 inode
->i_atime
= *atime
;
642 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
643 /* we did a utimes(); ignore mds values */
648 /* we have no write|excl caps; whatever the MDS says is true */
649 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
650 inode
->i_ctime
= *ctime
;
651 inode
->i_mtime
= *mtime
;
652 inode
->i_atime
= *atime
;
653 ci
->i_time_warp_seq
= time_warp_seq
;
658 if (warn
) /* time_warp_seq shouldn't go backwards */
659 dout("%p mds time_warp_seq %llu < %u\n",
660 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
664 * Populate an inode based on info from mds. May be called on new or
667 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
668 struct ceph_mds_reply_info_in
*iinfo
,
669 struct ceph_mds_reply_dirfrag
*dirinfo
,
670 struct ceph_mds_session
*session
,
671 unsigned long ttl_from
, int cap_fmode
,
672 struct ceph_cap_reservation
*caps_reservation
)
674 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
675 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
676 struct ceph_inode_info
*ci
= ceph_inode(inode
);
677 int issued
= 0, implemented
, new_issued
;
678 struct timespec mtime
, atime
, ctime
;
679 struct ceph_buffer
*xattr_blob
= NULL
;
680 struct ceph_cap
*new_cap
= NULL
;
683 bool queue_trunc
= false;
684 bool new_version
= false;
685 bool fill_inline
= false;
687 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
688 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
691 /* prealloc new cap struct */
692 if (info
->cap
.caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
693 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
696 * prealloc xattr data, if it looks like we'll need it. only
697 * if len > 4 (meaning there are actually xattrs; the first 4
698 * bytes are the xattr count).
700 if (iinfo
->xattr_len
> 4) {
701 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
703 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
707 spin_lock(&ci
->i_ceph_lock
);
710 * provided version will be odd if inode value is projected,
711 * even if stable. skip the update if we have newer stable
712 * info (ours>=theirs, e.g. due to racing mds replies), unless
713 * we are getting projected (unstable) info (in which case the
714 * version is odd, and we want ours>theirs).
720 if (ci
->i_version
== 0 ||
721 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
722 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
725 issued
= __ceph_caps_issued(ci
, &implemented
);
726 issued
|= implemented
| __ceph_caps_dirty(ci
);
727 new_issued
= ~issued
& le32_to_cpu(info
->cap
.caps
);
730 ci
->i_version
= le64_to_cpu(info
->version
);
732 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
733 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
735 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
736 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
737 inode
->i_mode
= le32_to_cpu(info
->mode
);
738 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
739 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
740 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
741 from_kuid(&init_user_ns
, inode
->i_uid
),
742 from_kgid(&init_user_ns
, inode
->i_gid
));
745 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
746 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
747 set_nlink(inode
, le32_to_cpu(info
->nlink
));
749 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
750 /* be careful with mtime, atime, size */
751 ceph_decode_timespec(&atime
, &info
->atime
);
752 ceph_decode_timespec(&mtime
, &info
->mtime
);
753 ceph_decode_timespec(&ctime
, &info
->ctime
);
754 ceph_fill_file_time(inode
, issued
,
755 le32_to_cpu(info
->time_warp_seq
),
756 &ctime
, &mtime
, &atime
);
760 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
761 if (ci
->i_layout
.fl_pg_pool
!= info
->layout
.fl_pg_pool
)
762 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
763 ci
->i_layout
= info
->layout
;
764 ci
->i_pool_ns_len
= iinfo
->pool_ns_len
;
766 queue_trunc
= ceph_fill_file_size(inode
, issued
,
767 le32_to_cpu(info
->truncate_seq
),
768 le64_to_cpu(info
->truncate_size
),
769 le64_to_cpu(info
->size
));
770 /* only update max_size on auth cap */
771 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
772 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
773 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
774 le64_to_cpu(info
->max_size
));
775 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
780 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
781 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
782 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
783 if (ci
->i_xattrs
.blob
)
784 ceph_buffer_put(ci
->i_xattrs
.blob
);
785 ci
->i_xattrs
.blob
= xattr_blob
;
787 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
788 iinfo
->xattr_data
, iinfo
->xattr_len
);
789 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
790 ceph_forget_all_cached_acls(inode
);
794 inode
->i_mapping
->a_ops
= &ceph_aops
;
796 switch (inode
->i_mode
& S_IFMT
) {
801 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
802 inode
->i_op
= &ceph_file_iops
;
805 inode
->i_op
= &ceph_file_iops
;
806 inode
->i_fop
= &ceph_file_fops
;
809 inode
->i_op
= &ceph_symlink_iops
;
810 if (!ci
->i_symlink
) {
811 u32 symlen
= iinfo
->symlink_len
;
814 spin_unlock(&ci
->i_ceph_lock
);
817 if (WARN_ON(symlen
!= i_size_read(inode
)))
821 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
825 spin_lock(&ci
->i_ceph_lock
);
829 kfree(sym
); /* lost a race */
831 inode
->i_link
= ci
->i_symlink
;
834 inode
->i_op
= &ceph_dir_iops
;
835 inode
->i_fop
= &ceph_dir_fops
;
837 ci
->i_dir_layout
= iinfo
->dir_layout
;
839 ci
->i_files
= le64_to_cpu(info
->files
);
840 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
841 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
842 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
843 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
844 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
847 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
848 ceph_vinop(inode
), inode
->i_mode
);
851 /* were we issued a capability? */
852 if (info
->cap
.caps
) {
853 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
854 unsigned caps
= le32_to_cpu(info
->cap
.caps
);
855 ceph_add_cap(inode
, session
,
856 le64_to_cpu(info
->cap
.cap_id
),
858 le32_to_cpu(info
->cap
.wanted
),
859 le32_to_cpu(info
->cap
.seq
),
860 le32_to_cpu(info
->cap
.mseq
),
861 le64_to_cpu(info
->cap
.realm
),
862 info
->cap
.flags
, &new_cap
);
864 /* set dir completion flag? */
865 if (S_ISDIR(inode
->i_mode
) &&
866 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
867 (caps
& CEPH_CAP_FILE_SHARED
) &&
868 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
869 !__ceph_dir_is_complete(ci
)) {
870 dout(" marking %p complete (empty)\n", inode
);
871 i_size_write(inode
, 0);
872 __ceph_dir_set_complete(ci
,
873 atomic64_read(&ci
->i_release_count
),
874 atomic64_read(&ci
->i_ordered_count
));
879 dout(" %p got snap_caps %s\n", inode
,
880 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
881 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
883 __ceph_get_fmode(ci
, cap_fmode
);
885 } else if (cap_fmode
>= 0) {
886 pr_warn("mds issued no caps on %llx.%llx\n",
888 __ceph_get_fmode(ci
, cap_fmode
);
891 if (iinfo
->inline_version
> 0 &&
892 iinfo
->inline_version
>= ci
->i_inline_version
) {
893 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
894 ci
->i_inline_version
= iinfo
->inline_version
;
895 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
897 (le32_to_cpu(info
->cap
.caps
) & cache_caps
)))
901 spin_unlock(&ci
->i_ceph_lock
);
904 ceph_fill_inline_data(inode
, locked_page
,
905 iinfo
->inline_data
, iinfo
->inline_len
);
908 wake_up_all(&ci
->i_cap_wq
);
910 /* queue truncate if we saw i_size decrease */
912 ceph_queue_vmtruncate(inode
);
914 /* populate frag tree */
915 if (S_ISDIR(inode
->i_mode
))
916 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
918 /* update delegation info? */
920 ceph_fill_dirfrag(inode
, dirinfo
);
925 ceph_put_cap(mdsc
, new_cap
);
927 ceph_buffer_put(xattr_blob
);
932 * caller should hold session s_mutex.
934 static void update_dentry_lease(struct dentry
*dentry
,
935 struct ceph_mds_reply_lease
*lease
,
936 struct ceph_mds_session
*session
,
937 unsigned long from_time
)
939 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
940 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
941 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
942 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
945 /* only track leases on regular dentries */
946 if (dentry
->d_op
!= &ceph_dentry_ops
)
949 spin_lock(&dentry
->d_lock
);
950 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
951 dentry
, duration
, ttl
);
953 /* make lease_rdcache_gen match directory */
954 dir
= d_inode(dentry
->d_parent
);
955 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
960 if (di
->lease_gen
== session
->s_cap_gen
&&
961 time_before(ttl
, dentry
->d_time
))
962 goto out_unlock
; /* we already have a newer lease. */
964 if (di
->lease_session
&& di
->lease_session
!= session
)
967 ceph_dentry_lru_touch(dentry
);
969 if (!di
->lease_session
)
970 di
->lease_session
= ceph_get_mds_session(session
);
971 di
->lease_gen
= session
->s_cap_gen
;
972 di
->lease_seq
= le32_to_cpu(lease
->seq
);
973 di
->lease_renew_after
= half_ttl
;
974 di
->lease_renew_from
= 0;
975 dentry
->d_time
= ttl
;
977 spin_unlock(&dentry
->d_lock
);
982 * splice a dentry to an inode.
983 * caller must hold directory i_mutex for this to be safe.
985 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
)
987 struct dentry
*realdn
;
991 /* dn must be unhashed */
994 realdn
= d_splice_alias(in
, dn
);
995 if (IS_ERR(realdn
)) {
996 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
997 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
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
)));
1018 * Incorporate results into the local cache. This is either just
1019 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1022 * A reply may contain
1023 * a directory inode along with a dentry.
1024 * and/or a target inode
1026 * Called with snap_rwsem (read).
1028 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
1029 struct ceph_mds_session
*session
)
1031 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1032 struct inode
*in
= NULL
;
1033 struct ceph_vino vino
;
1034 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1037 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1038 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1044 * If we resend completed ops to a recovering mds, we get no
1045 * trace. Since that is very rare, pretend this is the case
1046 * to ensure the 'no trace' handlers in the callers behave.
1048 * Fill in inodes unconditionally to avoid breaking cap
1051 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
1052 pr_info("fill_trace faking empty trace on %lld %s\n",
1053 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
1054 if (rinfo
->head
->is_dentry
) {
1055 rinfo
->head
->is_dentry
= 0;
1056 err
= fill_inode(req
->r_locked_dir
,
1057 &rinfo
->diri
, rinfo
->dirfrag
,
1058 session
, req
->r_request_started
, -1);
1060 if (rinfo
->head
->is_target
) {
1061 rinfo
->head
->is_target
= 0;
1062 ininfo
= rinfo
->targeti
.in
;
1063 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1064 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1065 in
= ceph_get_inode(sb
, vino
);
1066 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1067 session
, req
->r_request_started
,
1074 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1075 dout("fill_trace reply is empty!\n");
1076 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
1077 ceph_invalidate_dir_request(req
);
1081 if (rinfo
->head
->is_dentry
) {
1082 struct inode
*dir
= req
->r_locked_dir
;
1085 err
= fill_inode(dir
, NULL
,
1086 &rinfo
->diri
, rinfo
->dirfrag
,
1087 session
, req
->r_request_started
, -1,
1088 &req
->r_caps_reservation
);
1095 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1097 struct dentry
*dn
, *parent
;
1099 BUG_ON(!rinfo
->head
->is_target
);
1100 BUG_ON(req
->r_dentry
);
1102 parent
= d_find_any_alias(dir
);
1105 dname
.name
= rinfo
->dname
;
1106 dname
.len
= rinfo
->dname_len
;
1107 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1108 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1109 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1111 dn
= d_lookup(parent
, &dname
);
1112 dout("d_lookup on parent=%p name=%.*s got %p\n",
1113 parent
, dname
.len
, dname
.name
, dn
);
1116 dn
= d_alloc(parent
, &dname
);
1117 dout("d_alloc %p '%.*s' = %p\n", parent
,
1118 dname
.len
, dname
.name
, dn
);
1124 err
= ceph_init_dentry(dn
);
1130 } else if (d_really_is_positive(dn
) &&
1131 (ceph_ino(d_inode(dn
)) != vino
.ino
||
1132 ceph_snap(d_inode(dn
)) != vino
.snap
)) {
1133 dout(" dn %p points to wrong inode %p\n",
1145 if (rinfo
->head
->is_target
) {
1146 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1147 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1149 in
= ceph_get_inode(sb
, vino
);
1154 req
->r_target_inode
= in
;
1156 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1157 session
, req
->r_request_started
,
1158 (!req
->r_aborted
&& rinfo
->head
->result
== 0) ?
1160 &req
->r_caps_reservation
);
1162 pr_err("fill_inode badness %p %llx.%llx\n",
1163 in
, ceph_vinop(in
));
1169 * ignore null lease/binding on snapdir ENOENT, or else we
1170 * will have trouble splicing in the virtual snapdir later
1172 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1173 req
->r_locked_dir
&&
1174 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1175 fsc
->mount_options
->snapdir_name
,
1176 req
->r_dentry
->d_name
.len
))) {
1178 * lookup link rename : null -> possibly existing inode
1179 * mknod symlink mkdir : null -> new inode
1180 * unlink : linked -> null
1182 struct inode
*dir
= req
->r_locked_dir
;
1183 struct dentry
*dn
= req
->r_dentry
;
1184 bool have_dir_cap
, have_lease
;
1188 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1189 BUG_ON(ceph_ino(dir
) !=
1190 le64_to_cpu(rinfo
->diri
.in
->ino
));
1191 BUG_ON(ceph_snap(dir
) !=
1192 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1194 /* do we have a lease on the whole dir? */
1196 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1197 CEPH_CAP_FILE_SHARED
);
1199 /* do we have a dn lease? */
1200 have_lease
= have_dir_cap
||
1201 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1203 dout("fill_trace no dentry lease or dir cap\n");
1206 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1207 struct inode
*olddir
= req
->r_old_dentry_dir
;
1210 dout(" src %p '%pd' dst %p '%pd'\n",
1214 dout("fill_trace doing d_move %p -> %p\n",
1215 req
->r_old_dentry
, dn
);
1217 /* d_move screws up sibling dentries' offsets */
1218 ceph_dir_clear_ordered(dir
);
1219 ceph_dir_clear_ordered(olddir
);
1221 d_move(req
->r_old_dentry
, dn
);
1222 dout(" src %p '%pd' dst %p '%pd'\n",
1227 /* ensure target dentry is invalidated, despite
1228 rehashing bug in vfs_rename_dir */
1229 ceph_invalidate_dentry_lease(dn
);
1231 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1232 ceph_dentry(req
->r_old_dentry
)->offset
);
1234 dn
= req
->r_old_dentry
; /* use old_dentry */
1238 if (!rinfo
->head
->is_target
) {
1239 dout("fill_trace null dentry\n");
1240 if (d_really_is_positive(dn
)) {
1241 ceph_dir_clear_ordered(dir
);
1242 dout("d_delete %p\n", dn
);
1245 if (have_lease
&& d_unhashed(dn
))
1247 update_dentry_lease(dn
, rinfo
->dlease
,
1249 req
->r_request_started
);
1254 /* attach proper inode */
1255 if (d_really_is_negative(dn
)) {
1256 ceph_dir_clear_ordered(dir
);
1258 dn
= splice_dentry(dn
, in
);
1263 req
->r_dentry
= dn
; /* may have spliced */
1264 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1265 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1266 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1273 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1274 req
->r_request_started
);
1275 dout(" final dn %p\n", dn
);
1276 } else if (!req
->r_aborted
&&
1277 (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1278 req
->r_op
== CEPH_MDS_OP_MKSNAP
)) {
1279 struct dentry
*dn
= req
->r_dentry
;
1280 struct inode
*dir
= req
->r_locked_dir
;
1282 /* fill out a snapdir LOOKUPSNAP dentry */
1285 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1286 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1287 ceph_dir_clear_ordered(dir
);
1289 dn
= splice_dentry(dn
, in
);
1294 req
->r_dentry
= dn
; /* may have spliced */
1297 dout("fill_trace done err=%d\n", err
);
1302 * Prepopulate our cache with readdir results, leases, etc.
1304 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1305 struct ceph_mds_session
*session
)
1307 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1310 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1311 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1312 struct ceph_vino vino
;
1316 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1317 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1319 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1322 dout("new_inode badness got %d\n", err
);
1325 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1326 req
->r_request_started
, -1,
1327 &req
->r_caps_reservation
);
1329 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1338 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1342 put_page(ctl
->page
);
1347 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1348 struct ceph_readdir_cache_control
*ctl
,
1349 struct ceph_mds_request
*req
)
1351 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1352 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1353 unsigned idx
= ctl
->index
% nsize
;
1354 pgoff_t pgoff
= ctl
->index
/ nsize
;
1356 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1357 ceph_readdir_cache_release(ctl
);
1359 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1361 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1364 return idx
== 0 ? -ENOMEM
: 0;
1366 /* reading/filling the cache are serialized by
1367 * i_mutex, no need to use page lock */
1368 unlock_page(ctl
->page
);
1369 ctl
->dentries
= kmap(ctl
->page
);
1371 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1374 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1375 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1376 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1377 ctl
->dentries
[idx
] = dn
;
1380 dout("disable readdir cache\n");
1386 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1387 struct ceph_mds_session
*session
)
1389 struct dentry
*parent
= req
->r_dentry
;
1390 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1394 int err
= 0, skipped
= 0, ret
, i
;
1395 struct inode
*snapdir
= NULL
;
1396 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1397 struct ceph_dentry_info
*di
;
1398 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1399 struct ceph_readdir_cache_control cache_ctl
= {};
1402 return readdir_prepopulate_inodes_only(req
, session
);
1404 if (rinfo
->dir_dir
&&
1405 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1406 dout("readdir_prepopulate got new frag %x -> %x\n",
1407 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1408 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1409 req
->r_readdir_offset
= 2;
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_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1438 struct ceph_vino vino
;
1440 dname
.name
= rde
->name
;
1441 dname
.len
= rde
->name_len
;
1442 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1444 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1445 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1448 dn
= d_lookup(parent
, &dname
);
1449 dout("d_lookup on parent=%p name=%.*s got %p\n",
1450 parent
, dname
.len
, dname
.name
, dn
);
1453 dn
= d_alloc(parent
, &dname
);
1454 dout("d_alloc %p '%.*s' = %p\n", parent
,
1455 dname
.len
, dname
.name
, dn
);
1457 dout("d_alloc badness\n");
1461 ret
= ceph_init_dentry(dn
);
1467 } else if (d_really_is_positive(dn
) &&
1468 (ceph_ino(d_inode(dn
)) != vino
.ino
||
1469 ceph_snap(d_inode(dn
)) != vino
.snap
)) {
1470 dout(" dn %p points to wrong inode %p\n",
1478 if (d_really_is_positive(dn
)) {
1481 in
= ceph_get_inode(parent
->d_sb
, vino
);
1483 dout("new_inode badness\n");
1491 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1492 req
->r_request_started
, -1,
1493 &req
->r_caps_reservation
);
1495 pr_err("fill_inode badness on %p\n", in
);
1496 if (d_really_is_negative(dn
))
1503 if (d_really_is_negative(dn
)) {
1504 struct dentry
*realdn
;
1506 if (ceph_security_xattr_deadlock(in
)) {
1507 dout(" skip splicing dn %p to inode %p"
1508 " (security xattr deadlock)\n", dn
, in
);
1514 realdn
= splice_dentry(dn
, in
);
1515 if (IS_ERR(realdn
)) {
1516 err
= PTR_ERR(realdn
);
1525 di
->offset
= ceph_make_fpos(frag
, i
+ req
->r_readdir_offset
);
1526 rde
->offset
= di
->offset
;
1528 update_dentry_lease(dn
, rde
->lease
, req
->r_session
,
1529 req
->r_request_started
);
1531 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1532 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1542 if (err
== 0 && skipped
== 0) {
1543 req
->r_did_prepopulate
= true;
1544 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1546 ceph_readdir_cache_release(&cache_ctl
);
1551 dout("readdir_prepopulate done\n");
1555 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1557 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1560 spin_lock(&ci
->i_ceph_lock
);
1561 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1562 i_size_write(inode
, size
);
1563 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1565 /* tell the MDS if we are approaching max_size */
1566 if ((size
<< 1) >= ci
->i_max_size
&&
1567 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1570 spin_unlock(&ci
->i_ceph_lock
);
1575 * Write back inode data in a worker thread. (This can't be done
1576 * in the message handler context.)
1578 void ceph_queue_writeback(struct inode
*inode
)
1581 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1582 &ceph_inode(inode
)->i_wb_work
)) {
1583 dout("ceph_queue_writeback %p\n", inode
);
1585 dout("ceph_queue_writeback %p failed\n", inode
);
1590 static void ceph_writeback_work(struct work_struct
*work
)
1592 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1594 struct inode
*inode
= &ci
->vfs_inode
;
1596 dout("writeback %p\n", inode
);
1597 filemap_fdatawrite(&inode
->i_data
);
1602 * queue an async invalidation
1604 void ceph_queue_invalidate(struct inode
*inode
)
1607 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1608 &ceph_inode(inode
)->i_pg_inv_work
)) {
1609 dout("ceph_queue_invalidate %p\n", inode
);
1611 dout("ceph_queue_invalidate %p failed\n", inode
);
1617 * Invalidate inode pages in a worker thread. (This can't be done
1618 * in the message handler context.)
1620 static void ceph_invalidate_work(struct work_struct
*work
)
1622 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1624 struct inode
*inode
= &ci
->vfs_inode
;
1625 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1629 mutex_lock(&ci
->i_truncate_mutex
);
1631 if (ACCESS_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1632 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1633 inode
, ceph_ino(inode
));
1634 mapping_set_error(inode
->i_mapping
, -EIO
);
1635 truncate_pagecache(inode
, 0);
1636 mutex_unlock(&ci
->i_truncate_mutex
);
1640 spin_lock(&ci
->i_ceph_lock
);
1641 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1642 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1643 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1644 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1646 spin_unlock(&ci
->i_ceph_lock
);
1647 mutex_unlock(&ci
->i_truncate_mutex
);
1650 orig_gen
= ci
->i_rdcache_gen
;
1651 spin_unlock(&ci
->i_ceph_lock
);
1653 truncate_pagecache(inode
, 0);
1655 spin_lock(&ci
->i_ceph_lock
);
1656 if (orig_gen
== ci
->i_rdcache_gen
&&
1657 orig_gen
== ci
->i_rdcache_revoking
) {
1658 dout("invalidate_pages %p gen %d successful\n", inode
,
1660 ci
->i_rdcache_revoking
--;
1663 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1664 inode
, orig_gen
, ci
->i_rdcache_gen
,
1665 ci
->i_rdcache_revoking
);
1666 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1669 spin_unlock(&ci
->i_ceph_lock
);
1670 mutex_unlock(&ci
->i_truncate_mutex
);
1673 ceph_check_caps(ci
, 0, NULL
);
1679 * called by trunc_wq;
1681 * We also truncate in a separate thread as well.
1683 static void ceph_vmtruncate_work(struct work_struct
*work
)
1685 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1687 struct inode
*inode
= &ci
->vfs_inode
;
1689 dout("vmtruncate_work %p\n", inode
);
1690 __ceph_do_pending_vmtruncate(inode
);
1695 * Queue an async vmtruncate. If we fail to queue work, we will handle
1696 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1698 void ceph_queue_vmtruncate(struct inode
*inode
)
1700 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1704 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1705 &ci
->i_vmtruncate_work
)) {
1706 dout("ceph_queue_vmtruncate %p\n", inode
);
1708 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1709 inode
, ci
->i_truncate_pending
);
1715 * Make sure any pending truncation is applied before doing anything
1716 * that may depend on it.
1718 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1720 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1722 int wrbuffer_refs
, finish
= 0;
1724 mutex_lock(&ci
->i_truncate_mutex
);
1726 spin_lock(&ci
->i_ceph_lock
);
1727 if (ci
->i_truncate_pending
== 0) {
1728 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1729 spin_unlock(&ci
->i_ceph_lock
);
1730 mutex_unlock(&ci
->i_truncate_mutex
);
1735 * make sure any dirty snapped pages are flushed before we
1736 * possibly truncate them.. so write AND block!
1738 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1739 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1741 spin_unlock(&ci
->i_ceph_lock
);
1742 filemap_write_and_wait_range(&inode
->i_data
, 0,
1743 inode
->i_sb
->s_maxbytes
);
1747 /* there should be no reader or writer */
1748 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1750 to
= ci
->i_truncate_size
;
1751 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1752 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1753 ci
->i_truncate_pending
, to
);
1754 spin_unlock(&ci
->i_ceph_lock
);
1756 truncate_pagecache(inode
, to
);
1758 spin_lock(&ci
->i_ceph_lock
);
1759 if (to
== ci
->i_truncate_size
) {
1760 ci
->i_truncate_pending
= 0;
1763 spin_unlock(&ci
->i_ceph_lock
);
1767 mutex_unlock(&ci
->i_truncate_mutex
);
1769 if (wrbuffer_refs
== 0)
1770 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1772 wake_up_all(&ci
->i_cap_wq
);
1778 static const struct inode_operations ceph_symlink_iops
= {
1779 .readlink
= generic_readlink
,
1780 .get_link
= simple_get_link
,
1781 .setattr
= ceph_setattr
,
1782 .getattr
= ceph_getattr
,
1783 .setxattr
= ceph_setxattr
,
1784 .getxattr
= ceph_getxattr
,
1785 .listxattr
= ceph_listxattr
,
1786 .removexattr
= ceph_removexattr
,
1792 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1794 struct inode
*inode
= d_inode(dentry
);
1795 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1796 const unsigned int ia_valid
= attr
->ia_valid
;
1797 struct ceph_mds_request
*req
;
1798 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1799 struct ceph_cap_flush
*prealloc_cf
;
1801 int release
= 0, dirtied
= 0;
1804 int inode_dirty_flags
= 0;
1805 bool lock_snap_rwsem
= false;
1807 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1810 err
= inode_change_ok(inode
, attr
);
1814 prealloc_cf
= ceph_alloc_cap_flush();
1818 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1821 ceph_free_cap_flush(prealloc_cf
);
1822 return PTR_ERR(req
);
1825 spin_lock(&ci
->i_ceph_lock
);
1826 issued
= __ceph_caps_issued(ci
, NULL
);
1828 if (!ci
->i_head_snapc
&&
1829 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1830 lock_snap_rwsem
= true;
1831 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1832 spin_unlock(&ci
->i_ceph_lock
);
1833 down_read(&mdsc
->snap_rwsem
);
1834 spin_lock(&ci
->i_ceph_lock
);
1835 issued
= __ceph_caps_issued(ci
, NULL
);
1839 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1841 if (ia_valid
& ATTR_UID
) {
1842 dout("setattr %p uid %d -> %d\n", inode
,
1843 from_kuid(&init_user_ns
, inode
->i_uid
),
1844 from_kuid(&init_user_ns
, attr
->ia_uid
));
1845 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1846 inode
->i_uid
= attr
->ia_uid
;
1847 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1848 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1849 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1850 req
->r_args
.setattr
.uid
= cpu_to_le32(
1851 from_kuid(&init_user_ns
, attr
->ia_uid
));
1852 mask
|= CEPH_SETATTR_UID
;
1853 release
|= CEPH_CAP_AUTH_SHARED
;
1856 if (ia_valid
& ATTR_GID
) {
1857 dout("setattr %p gid %d -> %d\n", inode
,
1858 from_kgid(&init_user_ns
, inode
->i_gid
),
1859 from_kgid(&init_user_ns
, attr
->ia_gid
));
1860 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1861 inode
->i_gid
= attr
->ia_gid
;
1862 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1863 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1864 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1865 req
->r_args
.setattr
.gid
= cpu_to_le32(
1866 from_kgid(&init_user_ns
, attr
->ia_gid
));
1867 mask
|= CEPH_SETATTR_GID
;
1868 release
|= CEPH_CAP_AUTH_SHARED
;
1871 if (ia_valid
& ATTR_MODE
) {
1872 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1874 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1875 inode
->i_mode
= attr
->ia_mode
;
1876 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1877 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1878 attr
->ia_mode
!= inode
->i_mode
) {
1879 inode
->i_mode
= attr
->ia_mode
;
1880 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1881 mask
|= CEPH_SETATTR_MODE
;
1882 release
|= CEPH_CAP_AUTH_SHARED
;
1886 if (ia_valid
& ATTR_ATIME
) {
1887 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1888 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1889 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1890 if (issued
& CEPH_CAP_FILE_EXCL
) {
1891 ci
->i_time_warp_seq
++;
1892 inode
->i_atime
= attr
->ia_atime
;
1893 dirtied
|= CEPH_CAP_FILE_EXCL
;
1894 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1895 timespec_compare(&inode
->i_atime
,
1896 &attr
->ia_atime
) < 0) {
1897 inode
->i_atime
= attr
->ia_atime
;
1898 dirtied
|= CEPH_CAP_FILE_WR
;
1899 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1900 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1901 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1903 mask
|= CEPH_SETATTR_ATIME
;
1904 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1908 if (ia_valid
& ATTR_MTIME
) {
1909 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1910 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1911 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1912 if (issued
& CEPH_CAP_FILE_EXCL
) {
1913 ci
->i_time_warp_seq
++;
1914 inode
->i_mtime
= attr
->ia_mtime
;
1915 dirtied
|= CEPH_CAP_FILE_EXCL
;
1916 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1917 timespec_compare(&inode
->i_mtime
,
1918 &attr
->ia_mtime
) < 0) {
1919 inode
->i_mtime
= attr
->ia_mtime
;
1920 dirtied
|= CEPH_CAP_FILE_WR
;
1921 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1922 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1923 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1925 mask
|= CEPH_SETATTR_MTIME
;
1926 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1930 if (ia_valid
& ATTR_SIZE
) {
1931 dout("setattr %p size %lld -> %lld\n", inode
,
1932 inode
->i_size
, attr
->ia_size
);
1933 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1934 attr
->ia_size
> inode
->i_size
) {
1935 i_size_write(inode
, attr
->ia_size
);
1937 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1938 inode
->i_ctime
= attr
->ia_ctime
;
1939 ci
->i_reported_size
= attr
->ia_size
;
1940 dirtied
|= CEPH_CAP_FILE_EXCL
;
1941 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1942 attr
->ia_size
!= inode
->i_size
) {
1943 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1944 req
->r_args
.setattr
.old_size
=
1945 cpu_to_le64(inode
->i_size
);
1946 mask
|= CEPH_SETATTR_SIZE
;
1947 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1952 /* these do nothing */
1953 if (ia_valid
& ATTR_CTIME
) {
1954 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1955 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1956 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1957 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1958 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1959 only
? "ctime only" : "ignored");
1960 inode
->i_ctime
= attr
->ia_ctime
;
1963 * if kernel wants to dirty ctime but nothing else,
1964 * we need to choose a cap to dirty under, or do
1965 * a almost-no-op setattr
1967 if (issued
& CEPH_CAP_AUTH_EXCL
)
1968 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1969 else if (issued
& CEPH_CAP_FILE_EXCL
)
1970 dirtied
|= CEPH_CAP_FILE_EXCL
;
1971 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1972 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1974 mask
|= CEPH_SETATTR_CTIME
;
1977 if (ia_valid
& ATTR_FILE
)
1978 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1981 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
1983 inode
->i_ctime
= current_fs_time(inode
->i_sb
);
1987 spin_unlock(&ci
->i_ceph_lock
);
1988 if (lock_snap_rwsem
)
1989 up_read(&mdsc
->snap_rwsem
);
1991 if (inode_dirty_flags
)
1992 __mark_inode_dirty(inode
, inode_dirty_flags
);
1994 if (ia_valid
& ATTR_MODE
) {
1995 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2001 req
->r_inode
= inode
;
2003 req
->r_inode_drop
= release
;
2004 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2005 req
->r_num_caps
= 1;
2006 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2008 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2009 ceph_cap_string(dirtied
), mask
);
2011 ceph_mdsc_put_request(req
);
2012 if (mask
& CEPH_SETATTR_SIZE
)
2013 __ceph_do_pending_vmtruncate(inode
);
2014 ceph_free_cap_flush(prealloc_cf
);
2017 ceph_mdsc_put_request(req
);
2018 ceph_free_cap_flush(prealloc_cf
);
2023 * Verify that we have a lease on the given mask. If not,
2024 * do a getattr against an mds.
2026 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2027 int mask
, bool force
)
2029 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2030 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2031 struct ceph_mds_request
*req
;
2034 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2035 dout("do_getattr inode %p SNAPDIR\n", inode
);
2039 dout("do_getattr inode %p mask %s mode 0%o\n",
2040 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2041 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2044 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
2046 return PTR_ERR(req
);
2047 req
->r_inode
= inode
;
2049 req
->r_num_caps
= 1;
2050 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2051 req
->r_locked_page
= locked_page
;
2052 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2053 if (locked_page
&& err
== 0) {
2054 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2055 if (inline_version
== 0) {
2056 /* the reply is supposed to contain inline data */
2058 } else if (inline_version
== CEPH_INLINE_NONE
) {
2061 err
= req
->r_reply_info
.targeti
.inline_len
;
2064 ceph_mdsc_put_request(req
);
2065 dout("do_getattr result=%d\n", err
);
2071 * Check inode permissions. We verify we have a valid value for
2072 * the AUTH cap, then call the generic handler.
2074 int ceph_permission(struct inode
*inode
, int mask
)
2078 if (mask
& MAY_NOT_BLOCK
)
2081 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2084 err
= generic_permission(inode
, mask
);
2089 * Get all attributes. Hopefully somedata we'll have a statlite()
2090 * and can limit the fields we require to be accurate.
2092 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
2095 struct inode
*inode
= d_inode(dentry
);
2096 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2099 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2101 generic_fillattr(inode
, stat
);
2102 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2103 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2104 stat
->dev
= ceph_snap(inode
);
2107 if (S_ISDIR(inode
->i_mode
)) {
2108 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2110 stat
->size
= ci
->i_rbytes
;
2112 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2114 stat
->blksize
= 65536;