1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/xattr.h>
13 #include <linux/posix_acl.h>
14 #include <linux/random.h>
15 #include <linux/sort.h>
18 #include "mds_client.h"
20 #include <linux/ceph/decode.h>
23 * Ceph inode operations
25 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
26 * setattr, etc.), xattr helpers, and helpers for assimilating
27 * metadata returned by the MDS into our cache.
29 * Also define helpers for doing asynchronous writeback, invalidation,
30 * and truncation for the benefit of those who can't afford to block
31 * (typically because they are in the message handler path).
34 static const struct inode_operations ceph_symlink_iops
;
36 static void ceph_inode_work(struct work_struct
*work
);
39 * find or create an inode, given the ceph ino number
41 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
43 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
44 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
48 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
51 ino_t t
= ceph_vino_to_ino(vino
);
53 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
55 return ERR_PTR(-ENOMEM
);
56 if (inode
->i_state
& I_NEW
) {
57 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
59 unlock_new_inode(inode
);
62 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
68 * get/constuct snapdir inode for a given directory
70 struct inode
*ceph_get_snapdir(struct inode
*parent
)
72 struct ceph_vino vino
= {
73 .ino
= ceph_ino(parent
),
76 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
77 struct ceph_inode_info
*ci
= ceph_inode(inode
);
79 BUG_ON(!S_ISDIR(parent
->i_mode
));
82 inode
->i_mode
= parent
->i_mode
;
83 inode
->i_uid
= parent
->i_uid
;
84 inode
->i_gid
= parent
->i_gid
;
85 inode
->i_op
= &ceph_snapdir_iops
;
86 inode
->i_fop
= &ceph_snapdir_fops
;
87 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
92 const struct inode_operations ceph_file_iops
= {
93 .permission
= ceph_permission
,
94 .setattr
= ceph_setattr
,
95 .getattr
= ceph_getattr
,
96 .listxattr
= ceph_listxattr
,
97 .get_acl
= ceph_get_acl
,
98 .set_acl
= ceph_set_acl
,
103 * We use a 'frag tree' to keep track of the MDS's directory fragments
104 * for a given inode (usually there is just a single fragment). We
105 * need to know when a child frag is delegated to a new MDS, or when
106 * it is flagged as replicated, so we can direct our requests
111 * find/create a frag in the tree
113 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
117 struct rb_node
*parent
= NULL
;
118 struct ceph_inode_frag
*frag
;
121 p
= &ci
->i_fragtree
.rb_node
;
124 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
125 c
= ceph_frag_compare(f
, frag
->frag
);
134 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
136 return ERR_PTR(-ENOMEM
);
143 rb_link_node(&frag
->node
, parent
, p
);
144 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
146 dout("get_or_create_frag added %llx.%llx frag %x\n",
147 ceph_vinop(&ci
->vfs_inode
), f
);
152 * find a specific frag @f
154 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
156 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
159 struct ceph_inode_frag
*frag
=
160 rb_entry(n
, struct ceph_inode_frag
, node
);
161 int c
= ceph_frag_compare(f
, frag
->frag
);
173 * Choose frag containing the given value @v. If @pfrag is
174 * specified, copy the frag delegation info to the caller if
177 static u32
__ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
178 struct ceph_inode_frag
*pfrag
, int *found
)
180 u32 t
= ceph_frag_make(0, 0);
181 struct ceph_inode_frag
*frag
;
189 WARN_ON(!ceph_frag_contains_value(t
, v
));
190 frag
= __ceph_find_frag(ci
, t
);
192 break; /* t is a leaf */
193 if (frag
->split_by
== 0) {
195 memcpy(pfrag
, frag
, sizeof(*pfrag
));
202 nway
= 1 << frag
->split_by
;
203 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
204 frag
->split_by
, nway
);
205 for (i
= 0; i
< nway
; i
++) {
206 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
207 if (ceph_frag_contains_value(n
, v
)) {
214 dout("choose_frag(%x) = %x\n", v
, t
);
219 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
220 struct ceph_inode_frag
*pfrag
, int *found
)
223 mutex_lock(&ci
->i_fragtree_mutex
);
224 ret
= __ceph_choose_frag(ci
, v
, pfrag
, found
);
225 mutex_unlock(&ci
->i_fragtree_mutex
);
230 * Process dirfrag (delegation) info from the mds. Include leaf
231 * fragment in tree ONLY if ndist > 0. Otherwise, only
232 * branches/splits are included in i_fragtree)
234 static int ceph_fill_dirfrag(struct inode
*inode
,
235 struct ceph_mds_reply_dirfrag
*dirinfo
)
237 struct ceph_inode_info
*ci
= ceph_inode(inode
);
238 struct ceph_inode_frag
*frag
;
239 u32 id
= le32_to_cpu(dirinfo
->frag
);
240 int mds
= le32_to_cpu(dirinfo
->auth
);
241 int ndist
= le32_to_cpu(dirinfo
->ndist
);
246 spin_lock(&ci
->i_ceph_lock
);
248 diri_auth
= ci
->i_auth_cap
->mds
;
249 spin_unlock(&ci
->i_ceph_lock
);
251 if (mds
== -1) /* CDIR_AUTH_PARENT */
254 mutex_lock(&ci
->i_fragtree_mutex
);
255 if (ndist
== 0 && mds
== diri_auth
) {
256 /* no delegation info needed. */
257 frag
= __ceph_find_frag(ci
, id
);
260 if (frag
->split_by
== 0) {
261 /* tree leaf, remove */
262 dout("fill_dirfrag removed %llx.%llx frag %x"
263 " (no ref)\n", ceph_vinop(inode
), id
);
264 rb_erase(&frag
->node
, &ci
->i_fragtree
);
267 /* tree branch, keep and clear */
268 dout("fill_dirfrag cleared %llx.%llx frag %x"
269 " referral\n", ceph_vinop(inode
), id
);
277 /* find/add this frag to store mds delegation info */
278 frag
= __get_or_create_frag(ci
, id
);
280 /* this is not the end of the world; we can continue
281 with bad/inaccurate delegation info */
282 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
283 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
289 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
290 for (i
= 0; i
< frag
->ndist
; i
++)
291 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
292 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
293 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
296 mutex_unlock(&ci
->i_fragtree_mutex
);
300 static int frag_tree_split_cmp(const void *l
, const void *r
)
302 struct ceph_frag_tree_split
*ls
= (struct ceph_frag_tree_split
*)l
;
303 struct ceph_frag_tree_split
*rs
= (struct ceph_frag_tree_split
*)r
;
304 return ceph_frag_compare(le32_to_cpu(ls
->frag
),
305 le32_to_cpu(rs
->frag
));
308 static bool is_frag_child(u32 f
, struct ceph_inode_frag
*frag
)
311 return f
== ceph_frag_make(0, 0);
312 if (ceph_frag_bits(f
) != ceph_frag_bits(frag
->frag
) + frag
->split_by
)
314 return ceph_frag_contains_value(frag
->frag
, ceph_frag_value(f
));
317 static int ceph_fill_fragtree(struct inode
*inode
,
318 struct ceph_frag_tree_head
*fragtree
,
319 struct ceph_mds_reply_dirfrag
*dirinfo
)
321 struct ceph_inode_info
*ci
= ceph_inode(inode
);
322 struct ceph_inode_frag
*frag
, *prev_frag
= NULL
;
323 struct rb_node
*rb_node
;
324 unsigned i
, split_by
, nsplits
;
328 mutex_lock(&ci
->i_fragtree_mutex
);
329 nsplits
= le32_to_cpu(fragtree
->nsplits
);
330 if (nsplits
!= ci
->i_fragtree_nsplits
) {
332 } else if (nsplits
) {
333 i
= prandom_u32() % nsplits
;
334 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
335 if (!__ceph_find_frag(ci
, id
))
337 } else if (!RB_EMPTY_ROOT(&ci
->i_fragtree
)) {
338 rb_node
= rb_first(&ci
->i_fragtree
);
339 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
340 if (frag
->frag
!= ceph_frag_make(0, 0) || rb_next(rb_node
))
343 if (!update
&& dirinfo
) {
344 id
= le32_to_cpu(dirinfo
->frag
);
345 if (id
!= __ceph_choose_frag(ci
, id
, NULL
, NULL
))
352 sort(fragtree
->splits
, nsplits
, sizeof(fragtree
->splits
[0]),
353 frag_tree_split_cmp
, NULL
);
356 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode
));
357 rb_node
= rb_first(&ci
->i_fragtree
);
358 for (i
= 0; i
< nsplits
; i
++) {
359 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
360 split_by
= le32_to_cpu(fragtree
->splits
[i
].by
);
361 if (split_by
== 0 || ceph_frag_bits(id
) + split_by
> 24) {
362 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
363 "frag %x split by %d\n", ceph_vinop(inode
),
364 i
, nsplits
, id
, split_by
);
369 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
370 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
371 if (frag
->frag
!= id
)
374 rb_node
= rb_next(rb_node
);
377 rb_node
= rb_next(rb_node
);
378 /* delete stale split/leaf node */
379 if (frag
->split_by
> 0 ||
380 !is_frag_child(frag
->frag
, prev_frag
)) {
381 rb_erase(&frag
->node
, &ci
->i_fragtree
);
382 if (frag
->split_by
> 0)
383 ci
->i_fragtree_nsplits
--;
389 frag
= __get_or_create_frag(ci
, id
);
393 if (frag
->split_by
== 0)
394 ci
->i_fragtree_nsplits
++;
395 frag
->split_by
= split_by
;
396 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
400 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
401 rb_node
= rb_next(rb_node
);
402 /* delete stale split/leaf node */
403 if (frag
->split_by
> 0 ||
404 !is_frag_child(frag
->frag
, prev_frag
)) {
405 rb_erase(&frag
->node
, &ci
->i_fragtree
);
406 if (frag
->split_by
> 0)
407 ci
->i_fragtree_nsplits
--;
412 mutex_unlock(&ci
->i_fragtree_mutex
);
417 * initialize a newly allocated inode.
419 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
421 struct ceph_inode_info
*ci
;
424 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
428 dout("alloc_inode %p\n", &ci
->vfs_inode
);
430 spin_lock_init(&ci
->i_ceph_lock
);
433 ci
->i_inline_version
= 0;
434 ci
->i_time_warp_seq
= 0;
435 ci
->i_ceph_flags
= 0;
436 atomic64_set(&ci
->i_ordered_count
, 1);
437 atomic64_set(&ci
->i_release_count
, 1);
438 atomic64_set(&ci
->i_complete_seq
[0], 0);
439 atomic64_set(&ci
->i_complete_seq
[1], 0);
440 ci
->i_symlink
= NULL
;
445 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
446 RCU_INIT_POINTER(ci
->i_layout
.pool_ns
, NULL
);
448 ci
->i_fragtree
= RB_ROOT
;
449 mutex_init(&ci
->i_fragtree_mutex
);
451 ci
->i_xattrs
.blob
= NULL
;
452 ci
->i_xattrs
.prealloc_blob
= NULL
;
453 ci
->i_xattrs
.dirty
= false;
454 ci
->i_xattrs
.index
= RB_ROOT
;
455 ci
->i_xattrs
.count
= 0;
456 ci
->i_xattrs
.names_size
= 0;
457 ci
->i_xattrs
.vals_size
= 0;
458 ci
->i_xattrs
.version
= 0;
459 ci
->i_xattrs
.index_version
= 0;
461 ci
->i_caps
= RB_ROOT
;
462 ci
->i_auth_cap
= NULL
;
463 ci
->i_dirty_caps
= 0;
464 ci
->i_flushing_caps
= 0;
465 INIT_LIST_HEAD(&ci
->i_dirty_item
);
466 INIT_LIST_HEAD(&ci
->i_flushing_item
);
467 ci
->i_prealloc_cap_flush
= NULL
;
468 INIT_LIST_HEAD(&ci
->i_cap_flush_list
);
469 init_waitqueue_head(&ci
->i_cap_wq
);
470 ci
->i_hold_caps_min
= 0;
471 ci
->i_hold_caps_max
= 0;
472 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
473 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
474 ci
->i_head_snapc
= NULL
;
477 for (i
= 0; i
< CEPH_FILE_MODE_BITS
; i
++)
478 ci
->i_nr_by_mode
[i
] = 0;
480 mutex_init(&ci
->i_truncate_mutex
);
481 ci
->i_truncate_seq
= 0;
482 ci
->i_truncate_size
= 0;
483 ci
->i_truncate_pending
= 0;
486 ci
->i_reported_size
= 0;
487 ci
->i_wanted_max_size
= 0;
488 ci
->i_requested_max_size
= 0;
492 ci
->i_rdcache_ref
= 0;
495 ci
->i_wrbuffer_ref
= 0;
496 ci
->i_wrbuffer_ref_head
= 0;
497 atomic_set(&ci
->i_filelock_ref
, 0);
498 atomic_set(&ci
->i_shared_gen
, 0);
499 ci
->i_rdcache_gen
= 0;
500 ci
->i_rdcache_revoking
= 0;
502 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
503 INIT_LIST_HEAD(&ci
->i_unsafe_iops
);
504 spin_lock_init(&ci
->i_unsafe_lock
);
506 ci
->i_snap_realm
= NULL
;
507 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
508 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
510 INIT_WORK(&ci
->i_work
, ceph_inode_work
);
513 ceph_fscache_inode_init(ci
);
515 return &ci
->vfs_inode
;
518 static void ceph_i_callback(struct rcu_head
*head
)
520 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
521 struct ceph_inode_info
*ci
= ceph_inode(inode
);
523 kfree(ci
->i_symlink
);
524 kmem_cache_free(ceph_inode_cachep
, ci
);
527 void ceph_evict_inode(struct inode
*inode
)
529 struct ceph_inode_info
*ci
= ceph_inode(inode
);
530 struct ceph_inode_frag
*frag
;
533 dout("evict_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
535 truncate_inode_pages_final(&inode
->i_data
);
538 ceph_fscache_unregister_inode_cookie(ci
);
540 __ceph_remove_caps(inode
);
542 if (__ceph_has_any_quota(ci
))
543 ceph_adjust_quota_realms_count(inode
, false);
546 * we may still have a snap_realm reference if there are stray
547 * caps in i_snap_caps.
549 if (ci
->i_snap_realm
) {
550 struct ceph_mds_client
*mdsc
=
551 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
552 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
554 dout(" dropping residual ref to snap realm %p\n", realm
);
555 spin_lock(&realm
->inodes_with_caps_lock
);
556 list_del_init(&ci
->i_snap_realm_item
);
557 spin_unlock(&realm
->inodes_with_caps_lock
);
558 ceph_put_snap_realm(mdsc
, realm
);
561 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
562 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
563 rb_erase(n
, &ci
->i_fragtree
);
566 ci
->i_fragtree_nsplits
= 0;
568 __ceph_destroy_xattrs(ci
);
569 if (ci
->i_xattrs
.blob
)
570 ceph_buffer_put(ci
->i_xattrs
.blob
);
571 if (ci
->i_xattrs
.prealloc_blob
)
572 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
574 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
577 void ceph_destroy_inode(struct inode
*inode
)
579 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
582 int ceph_drop_inode(struct inode
*inode
)
585 * Positve dentry and corresponding inode are always accompanied
586 * in MDS reply. So no need to keep inode in the cache after
587 * dropping all its aliases.
592 static inline blkcnt_t
calc_inode_blocks(u64 size
)
594 return (size
+ (1<<9) - 1) >> 9;
598 * Helpers to fill in size, ctime, mtime, and atime. We have to be
599 * careful because either the client or MDS may have more up to date
600 * info, depending on which capabilities are held, and whether
601 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
602 * and size are monotonically increasing, except when utimes() or
603 * truncate() increments the corresponding _seq values.)
605 int ceph_fill_file_size(struct inode
*inode
, int issued
,
606 u32 truncate_seq
, u64 truncate_size
, u64 size
)
608 struct ceph_inode_info
*ci
= ceph_inode(inode
);
611 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
612 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
613 dout("size %lld -> %llu\n", inode
->i_size
, size
);
614 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
615 pr_err("fill_file_size non-zero size for directory\n");
618 i_size_write(inode
, size
);
619 inode
->i_blocks
= calc_inode_blocks(size
);
620 ci
->i_reported_size
= size
;
621 if (truncate_seq
!= ci
->i_truncate_seq
) {
622 dout("truncate_seq %u -> %u\n",
623 ci
->i_truncate_seq
, truncate_seq
);
624 ci
->i_truncate_seq
= truncate_seq
;
626 /* the MDS should have revoked these caps */
627 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
630 CEPH_CAP_FILE_LAZYIO
));
632 * If we hold relevant caps, or in the case where we're
633 * not the only client referencing this file and we
634 * don't hold those caps, then we need to check whether
635 * the file is either opened or mmaped
637 if ((issued
& (CEPH_CAP_FILE_CACHE
|
638 CEPH_CAP_FILE_BUFFER
)) ||
639 mapping_mapped(inode
->i_mapping
) ||
640 __ceph_caps_file_wanted(ci
)) {
641 ci
->i_truncate_pending
++;
646 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
647 ci
->i_truncate_size
!= truncate_size
) {
648 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
650 ci
->i_truncate_size
= truncate_size
;
654 ceph_fscache_invalidate(inode
);
659 void ceph_fill_file_time(struct inode
*inode
, int issued
,
660 u64 time_warp_seq
, struct timespec
*ctime
,
661 struct timespec
*mtime
, struct timespec
*atime
)
663 struct ceph_inode_info
*ci
= ceph_inode(inode
);
666 if (issued
& (CEPH_CAP_FILE_EXCL
|
668 CEPH_CAP_FILE_BUFFER
|
670 CEPH_CAP_XATTR_EXCL
)) {
671 if (ci
->i_version
== 0 ||
672 timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
673 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
674 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
675 ctime
->tv_sec
, ctime
->tv_nsec
);
676 inode
->i_ctime
= *ctime
;
678 if (ci
->i_version
== 0 ||
679 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
680 /* the MDS did a utimes() */
681 dout("mtime %ld.%09ld -> %ld.%09ld "
683 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
684 mtime
->tv_sec
, mtime
->tv_nsec
,
685 ci
->i_time_warp_seq
, (int)time_warp_seq
);
687 inode
->i_mtime
= *mtime
;
688 inode
->i_atime
= *atime
;
689 ci
->i_time_warp_seq
= time_warp_seq
;
690 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
691 /* nobody did utimes(); take the max */
692 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
693 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
694 inode
->i_mtime
.tv_sec
,
695 inode
->i_mtime
.tv_nsec
,
696 mtime
->tv_sec
, mtime
->tv_nsec
);
697 inode
->i_mtime
= *mtime
;
699 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
700 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
701 inode
->i_atime
.tv_sec
,
702 inode
->i_atime
.tv_nsec
,
703 atime
->tv_sec
, atime
->tv_nsec
);
704 inode
->i_atime
= *atime
;
706 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
707 /* we did a utimes(); ignore mds values */
712 /* we have no write|excl caps; whatever the MDS says is true */
713 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
714 inode
->i_ctime
= *ctime
;
715 inode
->i_mtime
= *mtime
;
716 inode
->i_atime
= *atime
;
717 ci
->i_time_warp_seq
= time_warp_seq
;
722 if (warn
) /* time_warp_seq shouldn't go backwards */
723 dout("%p mds time_warp_seq %llu < %u\n",
724 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
728 * Populate an inode based on info from mds. May be called on new or
731 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
732 struct ceph_mds_reply_info_in
*iinfo
,
733 struct ceph_mds_reply_dirfrag
*dirinfo
,
734 struct ceph_mds_session
*session
,
735 unsigned long ttl_from
, int cap_fmode
,
736 struct ceph_cap_reservation
*caps_reservation
)
738 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
739 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
740 struct ceph_inode_info
*ci
= ceph_inode(inode
);
741 int issued
= 0, implemented
, new_issued
;
742 struct timespec mtime
, atime
, ctime
;
743 struct ceph_buffer
*xattr_blob
= NULL
;
744 struct ceph_buffer
*old_blob
= NULL
;
745 struct ceph_string
*pool_ns
= NULL
;
746 struct ceph_cap
*new_cap
= NULL
;
749 bool queue_trunc
= false;
750 bool new_version
= false;
751 bool fill_inline
= false;
753 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
754 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
757 /* prealloc new cap struct */
758 if (info
->cap
.caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
759 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
762 * prealloc xattr data, if it looks like we'll need it. only
763 * if len > 4 (meaning there are actually xattrs; the first 4
764 * bytes are the xattr count).
766 if (iinfo
->xattr_len
> 4) {
767 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
769 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
773 if (iinfo
->pool_ns_len
> 0)
774 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
777 spin_lock(&ci
->i_ceph_lock
);
780 * provided version will be odd if inode value is projected,
781 * even if stable. skip the update if we have newer stable
782 * info (ours>=theirs, e.g. due to racing mds replies), unless
783 * we are getting projected (unstable) info (in which case the
784 * version is odd, and we want ours>theirs).
790 if (ci
->i_version
== 0 ||
791 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
792 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
795 issued
= __ceph_caps_issued(ci
, &implemented
);
796 issued
|= implemented
| __ceph_caps_dirty(ci
);
797 new_issued
= ~issued
& le32_to_cpu(info
->cap
.caps
);
800 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
801 /* directories have fl_stripe_unit set to zero */
802 if (le32_to_cpu(info
->layout
.fl_stripe_unit
))
804 fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
806 inode
->i_blkbits
= CEPH_BLOCK_SHIFT
;
808 __ceph_update_quota(ci
, iinfo
->max_bytes
, iinfo
->max_files
);
810 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
811 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
812 inode
->i_mode
= le32_to_cpu(info
->mode
);
813 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
814 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
815 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
816 from_kuid(&init_user_ns
, inode
->i_uid
),
817 from_kgid(&init_user_ns
, inode
->i_gid
));
820 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
821 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
822 set_nlink(inode
, le32_to_cpu(info
->nlink
));
824 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
825 /* be careful with mtime, atime, size */
826 ceph_decode_timespec(&atime
, &info
->atime
);
827 ceph_decode_timespec(&mtime
, &info
->mtime
);
828 ceph_decode_timespec(&ctime
, &info
->ctime
);
829 ceph_fill_file_time(inode
, issued
,
830 le32_to_cpu(info
->time_warp_seq
),
831 &ctime
, &mtime
, &atime
);
835 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
836 s64 old_pool
= ci
->i_layout
.pool_id
;
837 struct ceph_string
*old_ns
;
839 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
840 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
841 lockdep_is_held(&ci
->i_ceph_lock
));
842 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
844 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
845 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
849 queue_trunc
= ceph_fill_file_size(inode
, issued
,
850 le32_to_cpu(info
->truncate_seq
),
851 le64_to_cpu(info
->truncate_size
),
852 le64_to_cpu(info
->size
));
853 /* only update max_size on auth cap */
854 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
855 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
856 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
857 le64_to_cpu(info
->max_size
));
858 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
863 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
864 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
865 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
866 if (ci
->i_xattrs
.blob
)
867 old_blob
= ci
->i_xattrs
.blob
;
868 ci
->i_xattrs
.blob
= xattr_blob
;
870 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
871 iinfo
->xattr_data
, iinfo
->xattr_len
);
872 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
873 ceph_forget_all_cached_acls(inode
);
877 /* finally update i_version */
878 ci
->i_version
= le64_to_cpu(info
->version
);
880 inode
->i_mapping
->a_ops
= &ceph_aops
;
882 switch (inode
->i_mode
& S_IFMT
) {
887 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
888 inode
->i_op
= &ceph_file_iops
;
891 inode
->i_op
= &ceph_file_iops
;
892 inode
->i_fop
= &ceph_file_fops
;
895 inode
->i_op
= &ceph_symlink_iops
;
896 if (!ci
->i_symlink
) {
897 u32 symlen
= iinfo
->symlink_len
;
900 spin_unlock(&ci
->i_ceph_lock
);
902 if (symlen
!= i_size_read(inode
)) {
903 pr_err("fill_inode %llx.%llx BAD symlink "
904 "size %lld\n", ceph_vinop(inode
),
906 i_size_write(inode
, symlen
);
907 inode
->i_blocks
= calc_inode_blocks(symlen
);
911 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
915 spin_lock(&ci
->i_ceph_lock
);
919 kfree(sym
); /* lost a race */
921 inode
->i_link
= ci
->i_symlink
;
924 inode
->i_op
= &ceph_dir_iops
;
925 inode
->i_fop
= &ceph_dir_fops
;
927 ci
->i_dir_layout
= iinfo
->dir_layout
;
929 ci
->i_files
= le64_to_cpu(info
->files
);
930 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
931 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
932 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
933 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
934 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
937 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
938 ceph_vinop(inode
), inode
->i_mode
);
941 /* were we issued a capability? */
942 if (info
->cap
.caps
) {
943 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
944 unsigned caps
= le32_to_cpu(info
->cap
.caps
);
945 ceph_add_cap(inode
, session
,
946 le64_to_cpu(info
->cap
.cap_id
),
948 le32_to_cpu(info
->cap
.wanted
),
949 le32_to_cpu(info
->cap
.seq
),
950 le32_to_cpu(info
->cap
.mseq
),
951 le64_to_cpu(info
->cap
.realm
),
952 info
->cap
.flags
, &new_cap
);
954 /* set dir completion flag? */
955 if (S_ISDIR(inode
->i_mode
) &&
956 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
957 (caps
& CEPH_CAP_FILE_SHARED
) &&
958 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
959 !__ceph_dir_is_complete(ci
)) {
960 dout(" marking %p complete (empty)\n", inode
);
961 i_size_write(inode
, 0);
962 __ceph_dir_set_complete(ci
,
963 atomic64_read(&ci
->i_release_count
),
964 atomic64_read(&ci
->i_ordered_count
));
969 dout(" %p got snap_caps %s\n", inode
,
970 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
971 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
973 __ceph_get_fmode(ci
, cap_fmode
);
975 } else if (cap_fmode
>= 0) {
976 pr_warn("mds issued no caps on %llx.%llx\n",
978 __ceph_get_fmode(ci
, cap_fmode
);
981 if (iinfo
->inline_version
> 0 &&
982 iinfo
->inline_version
>= ci
->i_inline_version
) {
983 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
984 ci
->i_inline_version
= iinfo
->inline_version
;
985 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
987 (le32_to_cpu(info
->cap
.caps
) & cache_caps
)))
991 spin_unlock(&ci
->i_ceph_lock
);
994 ceph_fill_inline_data(inode
, locked_page
,
995 iinfo
->inline_data
, iinfo
->inline_len
);
998 wake_up_all(&ci
->i_cap_wq
);
1000 /* queue truncate if we saw i_size decrease */
1002 ceph_queue_vmtruncate(inode
);
1004 /* populate frag tree */
1005 if (S_ISDIR(inode
->i_mode
))
1006 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
1008 /* update delegation info? */
1010 ceph_fill_dirfrag(inode
, dirinfo
);
1015 ceph_put_cap(mdsc
, new_cap
);
1016 ceph_buffer_put(old_blob
);
1017 ceph_buffer_put(xattr_blob
);
1018 ceph_put_string(pool_ns
);
1023 * caller should hold session s_mutex.
1025 static void update_dentry_lease(struct dentry
*dentry
,
1026 struct ceph_mds_reply_lease
*lease
,
1027 struct ceph_mds_session
*session
,
1028 unsigned long from_time
,
1029 struct ceph_vino
*tgt_vino
,
1030 struct ceph_vino
*dir_vino
)
1032 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1033 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1034 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1035 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1037 struct ceph_mds_session
*old_lease_session
= NULL
;
1040 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1041 * we expect a negative dentry.
1043 if (!tgt_vino
&& d_really_is_positive(dentry
))
1046 if (tgt_vino
&& (d_really_is_negative(dentry
) ||
1047 !ceph_ino_compare(d_inode(dentry
), tgt_vino
)))
1050 spin_lock(&dentry
->d_lock
);
1051 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1052 dentry
, duration
, ttl
);
1054 dir
= d_inode(dentry
->d_parent
);
1056 /* make sure parent matches dir_vino */
1057 if (!ceph_ino_compare(dir
, dir_vino
))
1060 /* only track leases on regular dentries */
1061 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1064 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1069 if (di
->lease_gen
== session
->s_cap_gen
&&
1070 time_before(ttl
, di
->time
))
1071 goto out_unlock
; /* we already have a newer lease. */
1073 if (di
->lease_session
&& di
->lease_session
!= session
) {
1074 old_lease_session
= di
->lease_session
;
1075 di
->lease_session
= NULL
;
1078 ceph_dentry_lru_touch(dentry
);
1080 if (!di
->lease_session
)
1081 di
->lease_session
= ceph_get_mds_session(session
);
1082 di
->lease_gen
= session
->s_cap_gen
;
1083 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1084 di
->lease_renew_after
= half_ttl
;
1085 di
->lease_renew_from
= 0;
1088 spin_unlock(&dentry
->d_lock
);
1089 if (old_lease_session
)
1090 ceph_put_mds_session(old_lease_session
);
1094 * splice a dentry to an inode.
1095 * caller must hold directory i_mutex for this to be safe.
1097 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
)
1099 struct dentry
*realdn
;
1101 BUG_ON(d_inode(dn
));
1103 if (S_ISDIR(in
->i_mode
)) {
1104 /* If inode is directory, d_splice_alias() below will remove
1105 * 'realdn' from its origin parent. We need to ensure that
1106 * origin parent's readdir cache will not reference 'realdn'
1108 realdn
= d_find_any_alias(in
);
1110 struct ceph_dentry_info
*di
= ceph_dentry(realdn
);
1111 spin_lock(&realdn
->d_lock
);
1113 realdn
->d_op
->d_prune(realdn
);
1116 di
->lease_shared_gen
= 0;
1119 spin_unlock(&realdn
->d_lock
);
1124 /* dn must be unhashed */
1125 if (!d_unhashed(dn
))
1127 realdn
= d_splice_alias(in
, dn
);
1128 if (IS_ERR(realdn
)) {
1129 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1130 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1131 dn
= realdn
; /* note realdn contains the error */
1133 } else if (realdn
) {
1134 dout("dn %p (%d) spliced with %p (%d) "
1135 "inode %p ino %llx.%llx\n",
1137 realdn
, d_count(realdn
),
1138 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1142 BUG_ON(!ceph_dentry(dn
));
1143 dout("dn %p attached to %p ino %llx.%llx\n",
1144 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1151 * Incorporate results into the local cache. This is either just
1152 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1155 * A reply may contain
1156 * a directory inode along with a dentry.
1157 * and/or a target inode
1159 * Called with snap_rwsem (read).
1161 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1163 struct ceph_mds_session
*session
= req
->r_session
;
1164 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1165 struct inode
*in
= NULL
;
1166 struct ceph_vino tvino
, dvino
;
1167 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1170 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1171 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1173 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1174 dout("fill_trace reply is empty!\n");
1175 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1176 ceph_invalidate_dir_request(req
);
1180 if (rinfo
->head
->is_dentry
) {
1181 struct inode
*dir
= req
->r_parent
;
1184 err
= fill_inode(dir
, NULL
,
1185 &rinfo
->diri
, rinfo
->dirfrag
,
1186 session
, req
->r_request_started
, -1,
1187 &req
->r_caps_reservation
);
1194 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1196 struct dentry
*dn
, *parent
;
1198 BUG_ON(!rinfo
->head
->is_target
);
1199 BUG_ON(req
->r_dentry
);
1201 parent
= d_find_any_alias(dir
);
1204 dname
.name
= rinfo
->dname
;
1205 dname
.len
= rinfo
->dname_len
;
1206 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1207 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1208 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1210 dn
= d_lookup(parent
, &dname
);
1211 dout("d_lookup on parent=%p name=%.*s got %p\n",
1212 parent
, dname
.len
, dname
.name
, dn
);
1215 dn
= d_alloc(parent
, &dname
);
1216 dout("d_alloc %p '%.*s' = %p\n", parent
,
1217 dname
.len
, dname
.name
, dn
);
1224 } else if (d_really_is_positive(dn
) &&
1225 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1226 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1227 dout(" dn %p points to wrong inode %p\n",
1229 ceph_dir_clear_ordered(dir
);
1240 if (rinfo
->head
->is_target
) {
1241 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1242 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1244 in
= ceph_get_inode(sb
, tvino
);
1249 req
->r_target_inode
= in
;
1251 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1252 session
, req
->r_request_started
,
1253 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1254 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1255 &req
->r_caps_reservation
);
1257 pr_err("fill_inode badness %p %llx.%llx\n",
1258 in
, ceph_vinop(in
));
1264 * ignore null lease/binding on snapdir ENOENT, or else we
1265 * will have trouble splicing in the virtual snapdir later
1267 if (rinfo
->head
->is_dentry
&&
1268 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1269 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1270 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1271 fsc
->mount_options
->snapdir_name
,
1272 req
->r_dentry
->d_name
.len
))) {
1274 * lookup link rename : null -> possibly existing inode
1275 * mknod symlink mkdir : null -> new inode
1276 * unlink : linked -> null
1278 struct inode
*dir
= req
->r_parent
;
1279 struct dentry
*dn
= req
->r_dentry
;
1280 bool have_dir_cap
, have_lease
;
1284 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1286 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1287 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1289 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1290 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1292 /* do we have a lease on the whole dir? */
1294 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1295 CEPH_CAP_FILE_SHARED
);
1297 /* do we have a dn lease? */
1298 have_lease
= have_dir_cap
||
1299 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1301 dout("fill_trace no dentry lease or dir cap\n");
1304 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1305 struct inode
*olddir
= req
->r_old_dentry_dir
;
1308 dout(" src %p '%pd' dst %p '%pd'\n",
1312 dout("fill_trace doing d_move %p -> %p\n",
1313 req
->r_old_dentry
, dn
);
1315 /* d_move screws up sibling dentries' offsets */
1316 ceph_dir_clear_ordered(dir
);
1317 ceph_dir_clear_ordered(olddir
);
1319 d_move(req
->r_old_dentry
, dn
);
1320 dout(" src %p '%pd' dst %p '%pd'\n",
1325 /* ensure target dentry is invalidated, despite
1326 rehashing bug in vfs_rename_dir */
1327 ceph_invalidate_dentry_lease(dn
);
1329 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1330 ceph_dentry(req
->r_old_dentry
)->offset
);
1332 dn
= req
->r_old_dentry
; /* use old_dentry */
1336 if (!rinfo
->head
->is_target
) {
1337 dout("fill_trace null dentry\n");
1338 if (d_really_is_positive(dn
)) {
1339 dout("d_delete %p\n", dn
);
1340 ceph_dir_clear_ordered(dir
);
1342 } else if (have_lease
) {
1345 update_dentry_lease(dn
, rinfo
->dlease
,
1347 req
->r_request_started
,
1353 /* attach proper inode */
1354 if (d_really_is_negative(dn
)) {
1355 ceph_dir_clear_ordered(dir
);
1357 dn
= splice_dentry(dn
, in
);
1362 req
->r_dentry
= dn
; /* may have spliced */
1363 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1364 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1365 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1372 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1373 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1374 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1375 req
->r_request_started
,
1378 dout(" final dn %p\n", dn
);
1379 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1380 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1381 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1382 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1383 struct dentry
*dn
= req
->r_dentry
;
1384 struct inode
*dir
= req
->r_parent
;
1386 /* fill out a snapdir LOOKUPSNAP dentry */
1389 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1390 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1391 ceph_dir_clear_ordered(dir
);
1393 dn
= splice_dentry(dn
, in
);
1398 req
->r_dentry
= dn
; /* may have spliced */
1399 } else if (rinfo
->head
->is_dentry
) {
1400 struct ceph_vino
*ptvino
= NULL
;
1402 if ((le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) ||
1403 le32_to_cpu(rinfo
->dlease
->duration_ms
)) {
1404 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1405 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1407 if (rinfo
->head
->is_target
) {
1408 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1409 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1413 update_dentry_lease(req
->r_dentry
, rinfo
->dlease
,
1414 session
, req
->r_request_started
, ptvino
,
1417 dout("%s: no dentry lease or dir cap\n", __func__
);
1421 dout("fill_trace done err=%d\n", err
);
1426 * Prepopulate our cache with readdir results, leases, etc.
1428 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1429 struct ceph_mds_session
*session
)
1431 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1434 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1435 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1436 struct ceph_vino vino
;
1440 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1441 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1443 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1446 dout("new_inode badness got %d\n", err
);
1449 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1450 req
->r_request_started
, -1,
1451 &req
->r_caps_reservation
);
1453 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1456 /* avoid calling iput_final() in mds dispatch threads */
1457 ceph_async_iput(in
);
1463 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1467 put_page(ctl
->page
);
1472 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1473 struct ceph_readdir_cache_control
*ctl
,
1474 struct ceph_mds_request
*req
)
1476 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1477 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1478 unsigned idx
= ctl
->index
% nsize
;
1479 pgoff_t pgoff
= ctl
->index
/ nsize
;
1481 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1482 ceph_readdir_cache_release(ctl
);
1484 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1486 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1489 return idx
== 0 ? -ENOMEM
: 0;
1491 /* reading/filling the cache are serialized by
1492 * i_mutex, no need to use page lock */
1493 unlock_page(ctl
->page
);
1494 ctl
->dentries
= kmap(ctl
->page
);
1496 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1499 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1500 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1501 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1502 ctl
->dentries
[idx
] = dn
;
1505 dout("disable readdir cache\n");
1511 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1512 struct ceph_mds_session
*session
)
1514 struct dentry
*parent
= req
->r_dentry
;
1515 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1516 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1520 int err
= 0, skipped
= 0, ret
, i
;
1521 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1522 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1525 struct ceph_readdir_cache_control cache_ctl
= {};
1527 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1528 return readdir_prepopulate_inodes_only(req
, session
);
1530 if (rinfo
->hash_order
) {
1532 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1534 strlen(req
->r_path2
));
1535 last_hash
= ceph_frag_value(last_hash
);
1536 } else if (rinfo
->offset_hash
) {
1537 /* mds understands offset_hash */
1538 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1539 last_hash
= le32_to_cpu(rhead
->args
.readdir
.offset_hash
);
1543 if (rinfo
->dir_dir
&&
1544 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1545 dout("readdir_prepopulate got new frag %x -> %x\n",
1546 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1547 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1548 if (!rinfo
->hash_order
)
1549 req
->r_readdir_offset
= 2;
1552 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1553 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1554 rinfo
->dir_nr
, parent
);
1556 dout("readdir_prepopulate %d items under dn %p\n",
1557 rinfo
->dir_nr
, parent
);
1559 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1561 if (ceph_frag_is_leftmost(frag
) &&
1562 req
->r_readdir_offset
== 2 &&
1563 !(rinfo
->hash_order
&& last_hash
)) {
1564 /* note dir version at start of readdir so we can
1565 * tell if any dentries get dropped */
1566 req
->r_dir_release_cnt
=
1567 atomic64_read(&ci
->i_release_count
);
1568 req
->r_dir_ordered_cnt
=
1569 atomic64_read(&ci
->i_ordered_count
);
1570 req
->r_readdir_cache_idx
= 0;
1574 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1575 fpos_offset
= req
->r_readdir_offset
;
1577 /* FIXME: release caps/leases if error occurs */
1578 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1579 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1580 struct ceph_vino tvino
, dvino
;
1582 dname
.name
= rde
->name
;
1583 dname
.len
= rde
->name_len
;
1584 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1586 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1587 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1589 if (rinfo
->hash_order
) {
1590 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1591 rde
->name
, rde
->name_len
);
1592 hash
= ceph_frag_value(hash
);
1593 if (hash
!= last_hash
)
1596 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1598 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1602 dn
= d_lookup(parent
, &dname
);
1603 dout("d_lookup on parent=%p name=%.*s got %p\n",
1604 parent
, dname
.len
, dname
.name
, dn
);
1607 dn
= d_alloc(parent
, &dname
);
1608 dout("d_alloc %p '%.*s' = %p\n", parent
,
1609 dname
.len
, dname
.name
, dn
);
1611 dout("d_alloc badness\n");
1615 } else if (d_really_is_positive(dn
) &&
1616 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1617 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1618 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1619 dout(" dn %p points to wrong inode %p\n",
1622 spin_lock(&dn
->d_lock
);
1623 if (di
->offset
> 0 &&
1624 di
->lease_shared_gen
==
1625 atomic_read(&ci
->i_shared_gen
)) {
1626 __ceph_dir_clear_ordered(ci
);
1629 spin_unlock(&dn
->d_lock
);
1637 if (d_really_is_positive(dn
)) {
1640 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1642 dout("new_inode badness\n");
1650 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1651 req
->r_request_started
, -1,
1652 &req
->r_caps_reservation
);
1654 pr_err("fill_inode badness on %p\n", in
);
1655 if (d_really_is_negative(dn
)) {
1656 /* avoid calling iput_final() in mds
1657 * dispatch threads */
1658 ceph_async_iput(in
);
1665 if (d_really_is_negative(dn
)) {
1666 struct dentry
*realdn
;
1668 if (ceph_security_xattr_deadlock(in
)) {
1669 dout(" skip splicing dn %p to inode %p"
1670 " (security xattr deadlock)\n", dn
, in
);
1671 ceph_async_iput(in
);
1676 realdn
= splice_dentry(dn
, in
);
1677 if (IS_ERR(realdn
)) {
1678 err
= PTR_ERR(realdn
);
1685 ceph_dentry(dn
)->offset
= rde
->offset
;
1687 dvino
= ceph_vino(d_inode(parent
));
1688 update_dentry_lease(dn
, rde
->lease
, req
->r_session
,
1689 req
->r_request_started
, &tvino
, &dvino
);
1691 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1692 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1702 if (err
== 0 && skipped
== 0) {
1703 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1704 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1706 ceph_readdir_cache_release(&cache_ctl
);
1707 dout("readdir_prepopulate done\n");
1711 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1713 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1716 spin_lock(&ci
->i_ceph_lock
);
1717 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1718 i_size_write(inode
, size
);
1719 inode
->i_blocks
= calc_inode_blocks(size
);
1721 ret
= __ceph_should_report_size(ci
);
1723 spin_unlock(&ci
->i_ceph_lock
);
1728 * Put reference to inode, but avoid calling iput_final() in current thread.
1729 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1732 void ceph_async_iput(struct inode
*inode
)
1737 if (atomic_add_unless(&inode
->i_count
, -1, 1))
1739 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1740 &ceph_inode(inode
)->i_work
))
1742 /* queue work failed, i_count must be at least 2 */
1747 * Write back inode data in a worker thread. (This can't be done
1748 * in the message handler context.)
1750 void ceph_queue_writeback(struct inode
*inode
)
1752 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1753 set_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
);
1756 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1758 dout("ceph_queue_writeback %p\n", inode
);
1760 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1761 inode
, ci
->i_work_mask
);
1767 * queue an async invalidation
1769 void ceph_queue_invalidate(struct inode
*inode
)
1771 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1772 set_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
);
1775 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1776 &ceph_inode(inode
)->i_work
)) {
1777 dout("ceph_queue_invalidate %p\n", inode
);
1779 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1780 inode
, ci
->i_work_mask
);
1786 * Queue an async vmtruncate. If we fail to queue work, we will handle
1787 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1789 void ceph_queue_vmtruncate(struct inode
*inode
)
1791 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1792 set_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
);
1795 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1797 dout("ceph_queue_vmtruncate %p\n", inode
);
1799 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1800 inode
, ci
->i_work_mask
);
1805 static void ceph_do_invalidate_pages(struct inode
*inode
)
1807 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1808 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1812 mutex_lock(&ci
->i_truncate_mutex
);
1814 if (READ_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1815 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1816 inode
, ceph_ino(inode
));
1817 mapping_set_error(inode
->i_mapping
, -EIO
);
1818 truncate_pagecache(inode
, 0);
1819 mutex_unlock(&ci
->i_truncate_mutex
);
1823 spin_lock(&ci
->i_ceph_lock
);
1824 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1825 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1826 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1827 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1829 spin_unlock(&ci
->i_ceph_lock
);
1830 mutex_unlock(&ci
->i_truncate_mutex
);
1833 orig_gen
= ci
->i_rdcache_gen
;
1834 spin_unlock(&ci
->i_ceph_lock
);
1836 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1837 pr_err("invalidate_pages %p fails\n", inode
);
1840 spin_lock(&ci
->i_ceph_lock
);
1841 if (orig_gen
== ci
->i_rdcache_gen
&&
1842 orig_gen
== ci
->i_rdcache_revoking
) {
1843 dout("invalidate_pages %p gen %d successful\n", inode
,
1845 ci
->i_rdcache_revoking
--;
1848 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1849 inode
, orig_gen
, ci
->i_rdcache_gen
,
1850 ci
->i_rdcache_revoking
);
1851 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1854 spin_unlock(&ci
->i_ceph_lock
);
1855 mutex_unlock(&ci
->i_truncate_mutex
);
1858 ceph_check_caps(ci
, 0, NULL
);
1862 * Make sure any pending truncation is applied before doing anything
1863 * that may depend on it.
1865 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1867 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1869 int wrbuffer_refs
, finish
= 0;
1871 mutex_lock(&ci
->i_truncate_mutex
);
1873 spin_lock(&ci
->i_ceph_lock
);
1874 if (ci
->i_truncate_pending
== 0) {
1875 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1876 spin_unlock(&ci
->i_ceph_lock
);
1877 mutex_unlock(&ci
->i_truncate_mutex
);
1882 * make sure any dirty snapped pages are flushed before we
1883 * possibly truncate them.. so write AND block!
1885 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1886 struct ceph_cap_snap
*capsnap
;
1887 to
= ci
->i_truncate_size
;
1888 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1889 // MDS should have revoked Frw caps
1890 WARN_ON_ONCE(capsnap
->writing
);
1891 if (capsnap
->dirty_pages
&& capsnap
->size
> to
)
1894 spin_unlock(&ci
->i_ceph_lock
);
1895 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1898 truncate_pagecache(inode
, to
);
1900 filemap_write_and_wait_range(&inode
->i_data
, 0,
1901 inode
->i_sb
->s_maxbytes
);
1905 /* there should be no reader or writer */
1906 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1908 to
= ci
->i_truncate_size
;
1909 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1910 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1911 ci
->i_truncate_pending
, to
);
1912 spin_unlock(&ci
->i_ceph_lock
);
1914 truncate_pagecache(inode
, to
);
1916 spin_lock(&ci
->i_ceph_lock
);
1917 if (to
== ci
->i_truncate_size
) {
1918 ci
->i_truncate_pending
= 0;
1921 spin_unlock(&ci
->i_ceph_lock
);
1925 mutex_unlock(&ci
->i_truncate_mutex
);
1927 if (wrbuffer_refs
== 0)
1928 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1930 wake_up_all(&ci
->i_cap_wq
);
1933 static void ceph_inode_work(struct work_struct
*work
)
1935 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1937 struct inode
*inode
= &ci
->vfs_inode
;
1939 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
)) {
1940 dout("writeback %p\n", inode
);
1941 filemap_fdatawrite(&inode
->i_data
);
1943 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
))
1944 ceph_do_invalidate_pages(inode
);
1946 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
))
1947 __ceph_do_pending_vmtruncate(inode
);
1955 static const struct inode_operations ceph_symlink_iops
= {
1956 .get_link
= simple_get_link
,
1957 .setattr
= ceph_setattr
,
1958 .getattr
= ceph_getattr
,
1959 .listxattr
= ceph_listxattr
,
1962 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
1964 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1965 const unsigned int ia_valid
= attr
->ia_valid
;
1966 struct ceph_mds_request
*req
;
1967 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1968 struct ceph_cap_flush
*prealloc_cf
;
1970 int release
= 0, dirtied
= 0;
1973 int inode_dirty_flags
= 0;
1974 bool lock_snap_rwsem
= false;
1976 prealloc_cf
= ceph_alloc_cap_flush();
1980 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1983 ceph_free_cap_flush(prealloc_cf
);
1984 return PTR_ERR(req
);
1987 spin_lock(&ci
->i_ceph_lock
);
1988 issued
= __ceph_caps_issued(ci
, NULL
);
1990 if (!ci
->i_head_snapc
&&
1991 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1992 lock_snap_rwsem
= true;
1993 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1994 spin_unlock(&ci
->i_ceph_lock
);
1995 down_read(&mdsc
->snap_rwsem
);
1996 spin_lock(&ci
->i_ceph_lock
);
1997 issued
= __ceph_caps_issued(ci
, NULL
);
2001 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
2003 if (ia_valid
& ATTR_UID
) {
2004 dout("setattr %p uid %d -> %d\n", inode
,
2005 from_kuid(&init_user_ns
, inode
->i_uid
),
2006 from_kuid(&init_user_ns
, attr
->ia_uid
));
2007 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2008 inode
->i_uid
= attr
->ia_uid
;
2009 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2010 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2011 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
2012 req
->r_args
.setattr
.uid
= cpu_to_le32(
2013 from_kuid(&init_user_ns
, attr
->ia_uid
));
2014 mask
|= CEPH_SETATTR_UID
;
2015 release
|= CEPH_CAP_AUTH_SHARED
;
2018 if (ia_valid
& ATTR_GID
) {
2019 dout("setattr %p gid %d -> %d\n", inode
,
2020 from_kgid(&init_user_ns
, inode
->i_gid
),
2021 from_kgid(&init_user_ns
, attr
->ia_gid
));
2022 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2023 inode
->i_gid
= attr
->ia_gid
;
2024 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2025 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2026 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
2027 req
->r_args
.setattr
.gid
= cpu_to_le32(
2028 from_kgid(&init_user_ns
, attr
->ia_gid
));
2029 mask
|= CEPH_SETATTR_GID
;
2030 release
|= CEPH_CAP_AUTH_SHARED
;
2033 if (ia_valid
& ATTR_MODE
) {
2034 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
2036 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2037 inode
->i_mode
= attr
->ia_mode
;
2038 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2039 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2040 attr
->ia_mode
!= inode
->i_mode
) {
2041 inode
->i_mode
= attr
->ia_mode
;
2042 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
2043 mask
|= CEPH_SETATTR_MODE
;
2044 release
|= CEPH_CAP_AUTH_SHARED
;
2048 if (ia_valid
& ATTR_ATIME
) {
2049 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
2050 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
2051 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
2052 if (issued
& CEPH_CAP_FILE_EXCL
) {
2053 ci
->i_time_warp_seq
++;
2054 inode
->i_atime
= attr
->ia_atime
;
2055 dirtied
|= CEPH_CAP_FILE_EXCL
;
2056 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2057 timespec_compare(&inode
->i_atime
,
2058 &attr
->ia_atime
) < 0) {
2059 inode
->i_atime
= attr
->ia_atime
;
2060 dirtied
|= CEPH_CAP_FILE_WR
;
2061 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2062 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2063 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
2065 mask
|= CEPH_SETATTR_ATIME
;
2066 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
2070 if (ia_valid
& ATTR_MTIME
) {
2071 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
2072 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2073 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2074 if (issued
& CEPH_CAP_FILE_EXCL
) {
2075 ci
->i_time_warp_seq
++;
2076 inode
->i_mtime
= attr
->ia_mtime
;
2077 dirtied
|= CEPH_CAP_FILE_EXCL
;
2078 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2079 timespec_compare(&inode
->i_mtime
,
2080 &attr
->ia_mtime
) < 0) {
2081 inode
->i_mtime
= attr
->ia_mtime
;
2082 dirtied
|= CEPH_CAP_FILE_WR
;
2083 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2084 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2085 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
2087 mask
|= CEPH_SETATTR_MTIME
;
2088 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2092 if (ia_valid
& ATTR_SIZE
) {
2093 dout("setattr %p size %lld -> %lld\n", inode
,
2094 inode
->i_size
, attr
->ia_size
);
2095 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
2096 attr
->ia_size
> inode
->i_size
) {
2097 i_size_write(inode
, attr
->ia_size
);
2098 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2099 ci
->i_reported_size
= attr
->ia_size
;
2100 dirtied
|= CEPH_CAP_FILE_EXCL
;
2101 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2102 attr
->ia_size
!= inode
->i_size
) {
2103 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2104 req
->r_args
.setattr
.old_size
=
2105 cpu_to_le64(inode
->i_size
);
2106 mask
|= CEPH_SETATTR_SIZE
;
2107 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2112 /* these do nothing */
2113 if (ia_valid
& ATTR_CTIME
) {
2114 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2115 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2116 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
2117 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2118 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2119 only
? "ctime only" : "ignored");
2122 * if kernel wants to dirty ctime but nothing else,
2123 * we need to choose a cap to dirty under, or do
2124 * a almost-no-op setattr
2126 if (issued
& CEPH_CAP_AUTH_EXCL
)
2127 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2128 else if (issued
& CEPH_CAP_FILE_EXCL
)
2129 dirtied
|= CEPH_CAP_FILE_EXCL
;
2130 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2131 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2133 mask
|= CEPH_SETATTR_CTIME
;
2136 if (ia_valid
& ATTR_FILE
)
2137 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2140 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2142 inode
->i_ctime
= attr
->ia_ctime
;
2146 spin_unlock(&ci
->i_ceph_lock
);
2147 if (lock_snap_rwsem
)
2148 up_read(&mdsc
->snap_rwsem
);
2150 if (inode_dirty_flags
)
2151 __mark_inode_dirty(inode
, inode_dirty_flags
);
2155 req
->r_inode
= inode
;
2157 req
->r_inode_drop
= release
;
2158 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2159 req
->r_num_caps
= 1;
2160 req
->r_stamp
= attr
->ia_ctime
;
2161 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2163 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2164 ceph_cap_string(dirtied
), mask
);
2166 ceph_mdsc_put_request(req
);
2167 ceph_free_cap_flush(prealloc_cf
);
2169 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2170 __ceph_do_pending_vmtruncate(inode
);
2178 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2180 struct inode
*inode
= d_inode(dentry
);
2183 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2186 err
= setattr_prepare(dentry
, attr
);
2190 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2191 ceph_quota_is_max_bytes_exceeded(inode
, attr
->ia_size
))
2194 err
= __ceph_setattr(inode
, attr
);
2196 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2197 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2203 * Verify that we have a lease on the given mask. If not,
2204 * do a getattr against an mds.
2206 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2207 int mask
, bool force
)
2209 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2210 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2211 struct ceph_mds_request
*req
;
2214 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2215 dout("do_getattr inode %p SNAPDIR\n", inode
);
2219 dout("do_getattr inode %p mask %s mode 0%o\n",
2220 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2221 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2224 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
2226 return PTR_ERR(req
);
2227 req
->r_inode
= inode
;
2229 req
->r_num_caps
= 1;
2230 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2231 req
->r_locked_page
= locked_page
;
2232 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2233 if (locked_page
&& err
== 0) {
2234 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2235 if (inline_version
== 0) {
2236 /* the reply is supposed to contain inline data */
2238 } else if (inline_version
== CEPH_INLINE_NONE
) {
2241 err
= req
->r_reply_info
.targeti
.inline_len
;
2244 ceph_mdsc_put_request(req
);
2245 dout("do_getattr result=%d\n", err
);
2251 * Check inode permissions. We verify we have a valid value for
2252 * the AUTH cap, then call the generic handler.
2254 int ceph_permission(struct inode
*inode
, int mask
)
2258 if (mask
& MAY_NOT_BLOCK
)
2261 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2264 err
= generic_permission(inode
, mask
);
2269 * Get all attributes. Hopefully somedata we'll have a statlite()
2270 * and can limit the fields we require to be accurate.
2272 int ceph_getattr(const struct path
*path
, struct kstat
*stat
,
2273 u32 request_mask
, unsigned int flags
)
2275 struct inode
*inode
= d_inode(path
->dentry
);
2276 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2279 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2281 generic_fillattr(inode
, stat
);
2282 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2283 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2284 stat
->dev
= ceph_snap(inode
);
2287 if (S_ISDIR(inode
->i_mode
)) {
2288 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2290 stat
->size
= ci
->i_rbytes
;
2292 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2294 stat
->blksize
= 65536;