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 kmem_cache_free(ceph_inode_cachep
, ci
);
526 void ceph_destroy_inode(struct inode
*inode
)
528 struct ceph_inode_info
*ci
= ceph_inode(inode
);
529 struct ceph_inode_frag
*frag
;
532 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
534 ceph_fscache_unregister_inode_cookie(ci
);
536 __ceph_remove_caps(inode
);
539 * we may still have a snap_realm reference if there are stray
540 * caps in i_snap_caps.
542 if (ci
->i_snap_realm
) {
543 struct ceph_mds_client
*mdsc
=
544 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
545 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
547 dout(" dropping residual ref to snap realm %p\n", realm
);
548 spin_lock(&realm
->inodes_with_caps_lock
);
549 list_del_init(&ci
->i_snap_realm_item
);
550 spin_unlock(&realm
->inodes_with_caps_lock
);
551 ceph_put_snap_realm(mdsc
, realm
);
554 kfree(ci
->i_symlink
);
555 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
556 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
557 rb_erase(n
, &ci
->i_fragtree
);
560 ci
->i_fragtree_nsplits
= 0;
562 __ceph_destroy_xattrs(ci
);
563 if (ci
->i_xattrs
.blob
)
564 ceph_buffer_put(ci
->i_xattrs
.blob
);
565 if (ci
->i_xattrs
.prealloc_blob
)
566 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
568 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
570 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
573 int ceph_drop_inode(struct inode
*inode
)
576 * Positve dentry and corresponding inode are always accompanied
577 * in MDS reply. So no need to keep inode in the cache after
578 * dropping all its aliases.
583 static inline blkcnt_t
calc_inode_blocks(u64 size
)
585 return (size
+ (1<<9) - 1) >> 9;
589 * Helpers to fill in size, ctime, mtime, and atime. We have to be
590 * careful because either the client or MDS may have more up to date
591 * info, depending on which capabilities are held, and whether
592 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
593 * and size are monotonically increasing, except when utimes() or
594 * truncate() increments the corresponding _seq values.)
596 int ceph_fill_file_size(struct inode
*inode
, int issued
,
597 u32 truncate_seq
, u64 truncate_size
, u64 size
)
599 struct ceph_inode_info
*ci
= ceph_inode(inode
);
602 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
603 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
604 dout("size %lld -> %llu\n", inode
->i_size
, size
);
605 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
606 pr_err("fill_file_size non-zero size for directory\n");
609 i_size_write(inode
, size
);
610 inode
->i_blocks
= calc_inode_blocks(size
);
611 ci
->i_reported_size
= size
;
612 if (truncate_seq
!= ci
->i_truncate_seq
) {
613 dout("truncate_seq %u -> %u\n",
614 ci
->i_truncate_seq
, truncate_seq
);
615 ci
->i_truncate_seq
= truncate_seq
;
617 /* the MDS should have revoked these caps */
618 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
621 CEPH_CAP_FILE_LAZYIO
));
623 * If we hold relevant caps, or in the case where we're
624 * not the only client referencing this file and we
625 * don't hold those caps, then we need to check whether
626 * the file is either opened or mmaped
628 if ((issued
& (CEPH_CAP_FILE_CACHE
|
629 CEPH_CAP_FILE_BUFFER
)) ||
630 mapping_mapped(inode
->i_mapping
) ||
631 __ceph_caps_file_wanted(ci
)) {
632 ci
->i_truncate_pending
++;
637 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
638 ci
->i_truncate_size
!= truncate_size
) {
639 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
641 ci
->i_truncate_size
= truncate_size
;
645 ceph_fscache_invalidate(inode
);
650 void ceph_fill_file_time(struct inode
*inode
, int issued
,
651 u64 time_warp_seq
, struct timespec
*ctime
,
652 struct timespec
*mtime
, struct timespec
*atime
)
654 struct ceph_inode_info
*ci
= ceph_inode(inode
);
657 if (issued
& (CEPH_CAP_FILE_EXCL
|
659 CEPH_CAP_FILE_BUFFER
|
661 CEPH_CAP_XATTR_EXCL
)) {
662 if (ci
->i_version
== 0 ||
663 timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
664 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
665 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
666 ctime
->tv_sec
, ctime
->tv_nsec
);
667 inode
->i_ctime
= *ctime
;
669 if (ci
->i_version
== 0 ||
670 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
671 /* the MDS did a utimes() */
672 dout("mtime %ld.%09ld -> %ld.%09ld "
674 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
675 mtime
->tv_sec
, mtime
->tv_nsec
,
676 ci
->i_time_warp_seq
, (int)time_warp_seq
);
678 inode
->i_mtime
= *mtime
;
679 inode
->i_atime
= *atime
;
680 ci
->i_time_warp_seq
= time_warp_seq
;
681 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
682 /* nobody did utimes(); take the max */
683 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
684 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
685 inode
->i_mtime
.tv_sec
,
686 inode
->i_mtime
.tv_nsec
,
687 mtime
->tv_sec
, mtime
->tv_nsec
);
688 inode
->i_mtime
= *mtime
;
690 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
691 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
692 inode
->i_atime
.tv_sec
,
693 inode
->i_atime
.tv_nsec
,
694 atime
->tv_sec
, atime
->tv_nsec
);
695 inode
->i_atime
= *atime
;
697 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
698 /* we did a utimes(); ignore mds values */
703 /* we have no write|excl caps; whatever the MDS says is true */
704 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
705 inode
->i_ctime
= *ctime
;
706 inode
->i_mtime
= *mtime
;
707 inode
->i_atime
= *atime
;
708 ci
->i_time_warp_seq
= time_warp_seq
;
713 if (warn
) /* time_warp_seq shouldn't go backwards */
714 dout("%p mds time_warp_seq %llu < %u\n",
715 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
719 * Populate an inode based on info from mds. May be called on new or
722 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
723 struct ceph_mds_reply_info_in
*iinfo
,
724 struct ceph_mds_reply_dirfrag
*dirinfo
,
725 struct ceph_mds_session
*session
,
726 unsigned long ttl_from
, int cap_fmode
,
727 struct ceph_cap_reservation
*caps_reservation
)
729 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
730 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
731 struct ceph_inode_info
*ci
= ceph_inode(inode
);
732 int issued
= 0, implemented
, new_issued
;
733 struct timespec mtime
, atime
, ctime
;
734 struct ceph_buffer
*xattr_blob
= NULL
;
735 struct ceph_string
*pool_ns
= NULL
;
736 struct ceph_cap
*new_cap
= NULL
;
739 bool queue_trunc
= false;
740 bool new_version
= false;
741 bool fill_inline
= false;
743 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
744 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
747 /* prealloc new cap struct */
748 if (info
->cap
.caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
749 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
752 * prealloc xattr data, if it looks like we'll need it. only
753 * if len > 4 (meaning there are actually xattrs; the first 4
754 * bytes are the xattr count).
756 if (iinfo
->xattr_len
> 4) {
757 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
759 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
763 if (iinfo
->pool_ns_len
> 0)
764 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
767 spin_lock(&ci
->i_ceph_lock
);
770 * provided version will be odd if inode value is projected,
771 * even if stable. skip the update if we have newer stable
772 * info (ours>=theirs, e.g. due to racing mds replies), unless
773 * we are getting projected (unstable) info (in which case the
774 * version is odd, and we want ours>theirs).
780 if (ci
->i_version
== 0 ||
781 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
782 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
785 issued
= __ceph_caps_issued(ci
, &implemented
);
786 issued
|= implemented
| __ceph_caps_dirty(ci
);
787 new_issued
= ~issued
& le32_to_cpu(info
->cap
.caps
);
790 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
791 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
793 ci
->i_max_bytes
= iinfo
->max_bytes
;
794 ci
->i_max_files
= iinfo
->max_files
;
796 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
797 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
798 inode
->i_mode
= le32_to_cpu(info
->mode
);
799 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
800 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
801 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
802 from_kuid(&init_user_ns
, inode
->i_uid
),
803 from_kgid(&init_user_ns
, inode
->i_gid
));
806 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
807 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
808 set_nlink(inode
, le32_to_cpu(info
->nlink
));
810 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
811 /* be careful with mtime, atime, size */
812 ceph_decode_timespec(&atime
, &info
->atime
);
813 ceph_decode_timespec(&mtime
, &info
->mtime
);
814 ceph_decode_timespec(&ctime
, &info
->ctime
);
815 ceph_fill_file_time(inode
, issued
,
816 le32_to_cpu(info
->time_warp_seq
),
817 &ctime
, &mtime
, &atime
);
821 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
822 s64 old_pool
= ci
->i_layout
.pool_id
;
823 struct ceph_string
*old_ns
;
825 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
826 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
827 lockdep_is_held(&ci
->i_ceph_lock
));
828 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
830 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
831 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
835 queue_trunc
= ceph_fill_file_size(inode
, issued
,
836 le32_to_cpu(info
->truncate_seq
),
837 le64_to_cpu(info
->truncate_size
),
838 le64_to_cpu(info
->size
));
839 /* only update max_size on auth cap */
840 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
841 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
842 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
843 le64_to_cpu(info
->max_size
));
844 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
849 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
850 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
851 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
852 if (ci
->i_xattrs
.blob
)
853 ceph_buffer_put(ci
->i_xattrs
.blob
);
854 ci
->i_xattrs
.blob
= xattr_blob
;
856 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
857 iinfo
->xattr_data
, iinfo
->xattr_len
);
858 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
859 ceph_forget_all_cached_acls(inode
);
863 /* finally update i_version */
864 ci
->i_version
= le64_to_cpu(info
->version
);
866 inode
->i_mapping
->a_ops
= &ceph_aops
;
868 switch (inode
->i_mode
& S_IFMT
) {
873 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
874 inode
->i_op
= &ceph_file_iops
;
877 inode
->i_op
= &ceph_file_iops
;
878 inode
->i_fop
= &ceph_file_fops
;
881 inode
->i_op
= &ceph_symlink_iops
;
882 if (!ci
->i_symlink
) {
883 u32 symlen
= iinfo
->symlink_len
;
886 spin_unlock(&ci
->i_ceph_lock
);
888 if (symlen
!= i_size_read(inode
)) {
889 pr_err("fill_inode %llx.%llx BAD symlink "
890 "size %lld\n", ceph_vinop(inode
),
892 i_size_write(inode
, symlen
);
893 inode
->i_blocks
= calc_inode_blocks(symlen
);
897 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
901 spin_lock(&ci
->i_ceph_lock
);
905 kfree(sym
); /* lost a race */
907 inode
->i_link
= ci
->i_symlink
;
910 inode
->i_op
= &ceph_dir_iops
;
911 inode
->i_fop
= &ceph_dir_fops
;
913 ci
->i_dir_layout
= iinfo
->dir_layout
;
915 ci
->i_files
= le64_to_cpu(info
->files
);
916 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
917 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
918 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
919 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
920 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
923 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
924 ceph_vinop(inode
), inode
->i_mode
);
927 /* were we issued a capability? */
928 if (info
->cap
.caps
) {
929 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
930 unsigned caps
= le32_to_cpu(info
->cap
.caps
);
931 ceph_add_cap(inode
, session
,
932 le64_to_cpu(info
->cap
.cap_id
),
934 le32_to_cpu(info
->cap
.wanted
),
935 le32_to_cpu(info
->cap
.seq
),
936 le32_to_cpu(info
->cap
.mseq
),
937 le64_to_cpu(info
->cap
.realm
),
938 info
->cap
.flags
, &new_cap
);
940 /* set dir completion flag? */
941 if (S_ISDIR(inode
->i_mode
) &&
942 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
943 (caps
& CEPH_CAP_FILE_SHARED
) &&
944 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
945 !__ceph_dir_is_complete(ci
)) {
946 dout(" marking %p complete (empty)\n", inode
);
947 i_size_write(inode
, 0);
948 __ceph_dir_set_complete(ci
,
949 atomic64_read(&ci
->i_release_count
),
950 atomic64_read(&ci
->i_ordered_count
));
955 dout(" %p got snap_caps %s\n", inode
,
956 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
957 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
959 __ceph_get_fmode(ci
, cap_fmode
);
961 } else if (cap_fmode
>= 0) {
962 pr_warn("mds issued no caps on %llx.%llx\n",
964 __ceph_get_fmode(ci
, cap_fmode
);
967 if (iinfo
->inline_version
> 0 &&
968 iinfo
->inline_version
>= ci
->i_inline_version
) {
969 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
970 ci
->i_inline_version
= iinfo
->inline_version
;
971 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
973 (le32_to_cpu(info
->cap
.caps
) & cache_caps
)))
977 spin_unlock(&ci
->i_ceph_lock
);
980 ceph_fill_inline_data(inode
, locked_page
,
981 iinfo
->inline_data
, iinfo
->inline_len
);
984 wake_up_all(&ci
->i_cap_wq
);
986 /* queue truncate if we saw i_size decrease */
988 ceph_queue_vmtruncate(inode
);
990 /* populate frag tree */
991 if (S_ISDIR(inode
->i_mode
))
992 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
994 /* update delegation info? */
996 ceph_fill_dirfrag(inode
, dirinfo
);
1001 ceph_put_cap(mdsc
, new_cap
);
1003 ceph_buffer_put(xattr_blob
);
1004 ceph_put_string(pool_ns
);
1009 * caller should hold session s_mutex.
1011 static void update_dentry_lease(struct dentry
*dentry
,
1012 struct ceph_mds_reply_lease
*lease
,
1013 struct ceph_mds_session
*session
,
1014 unsigned long from_time
,
1015 struct ceph_vino
*tgt_vino
,
1016 struct ceph_vino
*dir_vino
)
1018 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1019 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1020 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1021 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1023 struct ceph_mds_session
*old_lease_session
= NULL
;
1026 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1027 * we expect a negative dentry.
1029 if (!tgt_vino
&& d_really_is_positive(dentry
))
1032 if (tgt_vino
&& (d_really_is_negative(dentry
) ||
1033 !ceph_ino_compare(d_inode(dentry
), tgt_vino
)))
1036 spin_lock(&dentry
->d_lock
);
1037 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1038 dentry
, duration
, ttl
);
1040 dir
= d_inode(dentry
->d_parent
);
1042 /* make sure parent matches dir_vino */
1043 if (!ceph_ino_compare(dir
, dir_vino
))
1046 /* only track leases on regular dentries */
1047 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1050 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1055 if (di
->lease_gen
== session
->s_cap_gen
&&
1056 time_before(ttl
, di
->time
))
1057 goto out_unlock
; /* we already have a newer lease. */
1059 if (di
->lease_session
&& di
->lease_session
!= session
) {
1060 old_lease_session
= di
->lease_session
;
1061 di
->lease_session
= NULL
;
1064 ceph_dentry_lru_touch(dentry
);
1066 if (!di
->lease_session
)
1067 di
->lease_session
= ceph_get_mds_session(session
);
1068 di
->lease_gen
= session
->s_cap_gen
;
1069 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1070 di
->lease_renew_after
= half_ttl
;
1071 di
->lease_renew_from
= 0;
1074 spin_unlock(&dentry
->d_lock
);
1075 if (old_lease_session
)
1076 ceph_put_mds_session(old_lease_session
);
1080 * splice a dentry to an inode.
1081 * caller must hold directory i_mutex for this to be safe.
1083 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
)
1085 struct dentry
*realdn
;
1087 BUG_ON(d_inode(dn
));
1089 if (S_ISDIR(in
->i_mode
)) {
1090 /* If inode is directory, d_splice_alias() below will remove
1091 * 'realdn' from its origin parent. We need to ensure that
1092 * origin parent's readdir cache will not reference 'realdn'
1094 realdn
= d_find_any_alias(in
);
1096 struct ceph_dentry_info
*di
= ceph_dentry(realdn
);
1097 spin_lock(&realdn
->d_lock
);
1099 realdn
->d_op
->d_prune(realdn
);
1102 di
->lease_shared_gen
= 0;
1105 spin_unlock(&realdn
->d_lock
);
1110 /* dn must be unhashed */
1111 if (!d_unhashed(dn
))
1113 realdn
= d_splice_alias(in
, dn
);
1114 if (IS_ERR(realdn
)) {
1115 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1116 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1117 dn
= realdn
; /* note realdn contains the error */
1119 } else if (realdn
) {
1120 dout("dn %p (%d) spliced with %p (%d) "
1121 "inode %p ino %llx.%llx\n",
1123 realdn
, d_count(realdn
),
1124 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1128 BUG_ON(!ceph_dentry(dn
));
1129 dout("dn %p attached to %p ino %llx.%llx\n",
1130 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1137 * Incorporate results into the local cache. This is either just
1138 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1141 * A reply may contain
1142 * a directory inode along with a dentry.
1143 * and/or a target inode
1145 * Called with snap_rwsem (read).
1147 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1149 struct ceph_mds_session
*session
= req
->r_session
;
1150 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1151 struct inode
*in
= NULL
;
1152 struct ceph_vino tvino
, dvino
;
1153 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1156 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1157 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1159 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1160 dout("fill_trace reply is empty!\n");
1161 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1162 ceph_invalidate_dir_request(req
);
1166 if (rinfo
->head
->is_dentry
) {
1167 struct inode
*dir
= req
->r_parent
;
1170 err
= fill_inode(dir
, NULL
,
1171 &rinfo
->diri
, rinfo
->dirfrag
,
1172 session
, req
->r_request_started
, -1,
1173 &req
->r_caps_reservation
);
1180 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1182 struct dentry
*dn
, *parent
;
1184 BUG_ON(!rinfo
->head
->is_target
);
1185 BUG_ON(req
->r_dentry
);
1187 parent
= d_find_any_alias(dir
);
1190 dname
.name
= rinfo
->dname
;
1191 dname
.len
= rinfo
->dname_len
;
1192 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1193 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1194 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1196 dn
= d_lookup(parent
, &dname
);
1197 dout("d_lookup on parent=%p name=%.*s got %p\n",
1198 parent
, dname
.len
, dname
.name
, dn
);
1201 dn
= d_alloc(parent
, &dname
);
1202 dout("d_alloc %p '%.*s' = %p\n", parent
,
1203 dname
.len
, dname
.name
, dn
);
1210 } else if (d_really_is_positive(dn
) &&
1211 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1212 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1213 dout(" dn %p points to wrong inode %p\n",
1215 ceph_dir_clear_ordered(dir
);
1226 if (rinfo
->head
->is_target
) {
1227 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1228 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1230 in
= ceph_get_inode(sb
, tvino
);
1235 req
->r_target_inode
= in
;
1237 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1238 session
, req
->r_request_started
,
1239 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1240 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1241 &req
->r_caps_reservation
);
1243 pr_err("fill_inode badness %p %llx.%llx\n",
1244 in
, ceph_vinop(in
));
1250 * ignore null lease/binding on snapdir ENOENT, or else we
1251 * will have trouble splicing in the virtual snapdir later
1253 if (rinfo
->head
->is_dentry
&&
1254 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1255 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1256 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1257 fsc
->mount_options
->snapdir_name
,
1258 req
->r_dentry
->d_name
.len
))) {
1260 * lookup link rename : null -> possibly existing inode
1261 * mknod symlink mkdir : null -> new inode
1262 * unlink : linked -> null
1264 struct inode
*dir
= req
->r_parent
;
1265 struct dentry
*dn
= req
->r_dentry
;
1266 bool have_dir_cap
, have_lease
;
1270 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1272 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1273 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1275 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1276 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1278 /* do we have a lease on the whole dir? */
1280 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1281 CEPH_CAP_FILE_SHARED
);
1283 /* do we have a dn lease? */
1284 have_lease
= have_dir_cap
||
1285 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1287 dout("fill_trace no dentry lease or dir cap\n");
1290 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1291 struct inode
*olddir
= req
->r_old_dentry_dir
;
1294 dout(" src %p '%pd' dst %p '%pd'\n",
1298 dout("fill_trace doing d_move %p -> %p\n",
1299 req
->r_old_dentry
, dn
);
1301 /* d_move screws up sibling dentries' offsets */
1302 ceph_dir_clear_ordered(dir
);
1303 ceph_dir_clear_ordered(olddir
);
1305 d_move(req
->r_old_dentry
, dn
);
1306 dout(" src %p '%pd' dst %p '%pd'\n",
1311 /* ensure target dentry is invalidated, despite
1312 rehashing bug in vfs_rename_dir */
1313 ceph_invalidate_dentry_lease(dn
);
1315 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1316 ceph_dentry(req
->r_old_dentry
)->offset
);
1318 dn
= req
->r_old_dentry
; /* use old_dentry */
1322 if (!rinfo
->head
->is_target
) {
1323 dout("fill_trace null dentry\n");
1324 if (d_really_is_positive(dn
)) {
1325 dout("d_delete %p\n", dn
);
1326 ceph_dir_clear_ordered(dir
);
1328 } else if (have_lease
) {
1331 update_dentry_lease(dn
, rinfo
->dlease
,
1333 req
->r_request_started
,
1339 /* attach proper inode */
1340 if (d_really_is_negative(dn
)) {
1341 ceph_dir_clear_ordered(dir
);
1343 dn
= splice_dentry(dn
, in
);
1348 req
->r_dentry
= dn
; /* may have spliced */
1349 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1350 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1351 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1358 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1359 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1360 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1361 req
->r_request_started
,
1364 dout(" final dn %p\n", dn
);
1365 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1366 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1367 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1368 struct dentry
*dn
= req
->r_dentry
;
1369 struct inode
*dir
= req
->r_parent
;
1371 /* fill out a snapdir LOOKUPSNAP dentry */
1374 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1375 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1376 ceph_dir_clear_ordered(dir
);
1378 dn
= splice_dentry(dn
, in
);
1383 req
->r_dentry
= dn
; /* may have spliced */
1384 } else if (rinfo
->head
->is_dentry
) {
1385 struct ceph_vino
*ptvino
= NULL
;
1387 if ((le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) ||
1388 le32_to_cpu(rinfo
->dlease
->duration_ms
)) {
1389 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1390 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1392 if (rinfo
->head
->is_target
) {
1393 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1394 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1398 update_dentry_lease(req
->r_dentry
, rinfo
->dlease
,
1399 session
, req
->r_request_started
, ptvino
,
1402 dout("%s: no dentry lease or dir cap\n", __func__
);
1406 dout("fill_trace done err=%d\n", err
);
1411 * Prepopulate our cache with readdir results, leases, etc.
1413 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1414 struct ceph_mds_session
*session
)
1416 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1419 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1420 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1421 struct ceph_vino vino
;
1425 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1426 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1428 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1431 dout("new_inode badness got %d\n", err
);
1434 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1435 req
->r_request_started
, -1,
1436 &req
->r_caps_reservation
);
1438 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1447 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1451 put_page(ctl
->page
);
1456 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1457 struct ceph_readdir_cache_control
*ctl
,
1458 struct ceph_mds_request
*req
)
1460 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1461 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1462 unsigned idx
= ctl
->index
% nsize
;
1463 pgoff_t pgoff
= ctl
->index
/ nsize
;
1465 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1466 ceph_readdir_cache_release(ctl
);
1468 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1470 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1473 return idx
== 0 ? -ENOMEM
: 0;
1475 /* reading/filling the cache are serialized by
1476 * i_mutex, no need to use page lock */
1477 unlock_page(ctl
->page
);
1478 ctl
->dentries
= kmap(ctl
->page
);
1480 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1483 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1484 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1485 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1486 ctl
->dentries
[idx
] = dn
;
1489 dout("disable readdir cache\n");
1495 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1496 struct ceph_mds_session
*session
)
1498 struct dentry
*parent
= req
->r_dentry
;
1499 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1500 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1504 int err
= 0, skipped
= 0, ret
, i
;
1505 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1506 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1509 struct ceph_readdir_cache_control cache_ctl
= {};
1511 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1512 return readdir_prepopulate_inodes_only(req
, session
);
1514 if (rinfo
->hash_order
) {
1516 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1518 strlen(req
->r_path2
));
1519 last_hash
= ceph_frag_value(last_hash
);
1520 } else if (rinfo
->offset_hash
) {
1521 /* mds understands offset_hash */
1522 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1523 last_hash
= le32_to_cpu(rhead
->args
.readdir
.offset_hash
);
1527 if (rinfo
->dir_dir
&&
1528 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1529 dout("readdir_prepopulate got new frag %x -> %x\n",
1530 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1531 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1532 if (!rinfo
->hash_order
)
1533 req
->r_readdir_offset
= 2;
1536 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1537 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1538 rinfo
->dir_nr
, parent
);
1540 dout("readdir_prepopulate %d items under dn %p\n",
1541 rinfo
->dir_nr
, parent
);
1543 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1545 if (ceph_frag_is_leftmost(frag
) &&
1546 req
->r_readdir_offset
== 2 &&
1547 !(rinfo
->hash_order
&& last_hash
)) {
1548 /* note dir version at start of readdir so we can
1549 * tell if any dentries get dropped */
1550 req
->r_dir_release_cnt
=
1551 atomic64_read(&ci
->i_release_count
);
1552 req
->r_dir_ordered_cnt
=
1553 atomic64_read(&ci
->i_ordered_count
);
1554 req
->r_readdir_cache_idx
= 0;
1558 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1559 fpos_offset
= req
->r_readdir_offset
;
1561 /* FIXME: release caps/leases if error occurs */
1562 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1563 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1564 struct ceph_vino tvino
, dvino
;
1566 dname
.name
= rde
->name
;
1567 dname
.len
= rde
->name_len
;
1568 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1570 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1571 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1573 if (rinfo
->hash_order
) {
1574 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1575 rde
->name
, rde
->name_len
);
1576 hash
= ceph_frag_value(hash
);
1577 if (hash
!= last_hash
)
1580 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1582 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1586 dn
= d_lookup(parent
, &dname
);
1587 dout("d_lookup on parent=%p name=%.*s got %p\n",
1588 parent
, dname
.len
, dname
.name
, dn
);
1591 dn
= d_alloc(parent
, &dname
);
1592 dout("d_alloc %p '%.*s' = %p\n", parent
,
1593 dname
.len
, dname
.name
, dn
);
1595 dout("d_alloc badness\n");
1599 } else if (d_really_is_positive(dn
) &&
1600 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1601 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1602 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1603 dout(" dn %p points to wrong inode %p\n",
1606 spin_lock(&dn
->d_lock
);
1607 if (di
->offset
> 0 &&
1608 di
->lease_shared_gen
==
1609 atomic_read(&ci
->i_shared_gen
)) {
1610 __ceph_dir_clear_ordered(ci
);
1613 spin_unlock(&dn
->d_lock
);
1621 if (d_really_is_positive(dn
)) {
1624 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1626 dout("new_inode badness\n");
1634 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1635 req
->r_request_started
, -1,
1636 &req
->r_caps_reservation
);
1638 pr_err("fill_inode badness on %p\n", in
);
1639 if (d_really_is_negative(dn
))
1646 if (d_really_is_negative(dn
)) {
1647 struct dentry
*realdn
;
1649 if (ceph_security_xattr_deadlock(in
)) {
1650 dout(" skip splicing dn %p to inode %p"
1651 " (security xattr deadlock)\n", dn
, in
);
1657 realdn
= splice_dentry(dn
, in
);
1658 if (IS_ERR(realdn
)) {
1659 err
= PTR_ERR(realdn
);
1667 ceph_dentry(dn
)->offset
= rde
->offset
;
1669 dvino
= ceph_vino(d_inode(parent
));
1670 update_dentry_lease(dn
, rde
->lease
, req
->r_session
,
1671 req
->r_request_started
, &tvino
, &dvino
);
1673 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1674 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1684 if (err
== 0 && skipped
== 0) {
1685 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1686 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1688 ceph_readdir_cache_release(&cache_ctl
);
1689 dout("readdir_prepopulate done\n");
1693 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1695 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1698 spin_lock(&ci
->i_ceph_lock
);
1699 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1700 i_size_write(inode
, size
);
1701 inode
->i_blocks
= calc_inode_blocks(size
);
1703 ret
= __ceph_should_report_size(ci
);
1705 spin_unlock(&ci
->i_ceph_lock
);
1710 * Write back inode data in a worker thread. (This can't be done
1711 * in the message handler context.)
1713 void ceph_queue_writeback(struct inode
*inode
)
1715 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1716 set_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
);
1719 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1721 dout("ceph_queue_writeback %p\n", inode
);
1723 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1724 inode
, ci
->i_work_mask
);
1730 * queue an async invalidation
1732 void ceph_queue_invalidate(struct inode
*inode
)
1734 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1735 set_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
);
1738 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1739 &ceph_inode(inode
)->i_work
)) {
1740 dout("ceph_queue_invalidate %p\n", inode
);
1742 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1743 inode
, ci
->i_work_mask
);
1749 * Queue an async vmtruncate. If we fail to queue work, we will handle
1750 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1752 void ceph_queue_vmtruncate(struct inode
*inode
)
1754 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1755 set_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
);
1758 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1760 dout("ceph_queue_vmtruncate %p\n", inode
);
1762 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1763 inode
, ci
->i_work_mask
);
1768 static void ceph_do_invalidate_pages(struct inode
*inode
)
1770 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1771 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1775 mutex_lock(&ci
->i_truncate_mutex
);
1777 if (READ_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1778 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1779 inode
, ceph_ino(inode
));
1780 mapping_set_error(inode
->i_mapping
, -EIO
);
1781 truncate_pagecache(inode
, 0);
1782 mutex_unlock(&ci
->i_truncate_mutex
);
1786 spin_lock(&ci
->i_ceph_lock
);
1787 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1788 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1789 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1790 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1792 spin_unlock(&ci
->i_ceph_lock
);
1793 mutex_unlock(&ci
->i_truncate_mutex
);
1796 orig_gen
= ci
->i_rdcache_gen
;
1797 spin_unlock(&ci
->i_ceph_lock
);
1799 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1800 pr_err("invalidate_pages %p fails\n", inode
);
1803 spin_lock(&ci
->i_ceph_lock
);
1804 if (orig_gen
== ci
->i_rdcache_gen
&&
1805 orig_gen
== ci
->i_rdcache_revoking
) {
1806 dout("invalidate_pages %p gen %d successful\n", inode
,
1808 ci
->i_rdcache_revoking
--;
1811 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1812 inode
, orig_gen
, ci
->i_rdcache_gen
,
1813 ci
->i_rdcache_revoking
);
1814 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1817 spin_unlock(&ci
->i_ceph_lock
);
1818 mutex_unlock(&ci
->i_truncate_mutex
);
1821 ceph_check_caps(ci
, 0, NULL
);
1825 * Make sure any pending truncation is applied before doing anything
1826 * that may depend on it.
1828 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1830 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1832 int wrbuffer_refs
, finish
= 0;
1834 mutex_lock(&ci
->i_truncate_mutex
);
1836 spin_lock(&ci
->i_ceph_lock
);
1837 if (ci
->i_truncate_pending
== 0) {
1838 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1839 spin_unlock(&ci
->i_ceph_lock
);
1840 mutex_unlock(&ci
->i_truncate_mutex
);
1845 * make sure any dirty snapped pages are flushed before we
1846 * possibly truncate them.. so write AND block!
1848 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1849 struct ceph_cap_snap
*capsnap
;
1850 to
= ci
->i_truncate_size
;
1851 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1852 // MDS should have revoked Frw caps
1853 WARN_ON_ONCE(capsnap
->writing
);
1854 if (capsnap
->dirty_pages
&& capsnap
->size
> to
)
1857 spin_unlock(&ci
->i_ceph_lock
);
1858 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1861 truncate_pagecache(inode
, to
);
1863 filemap_write_and_wait_range(&inode
->i_data
, 0,
1864 inode
->i_sb
->s_maxbytes
);
1868 /* there should be no reader or writer */
1869 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1871 to
= ci
->i_truncate_size
;
1872 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1873 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1874 ci
->i_truncate_pending
, to
);
1875 spin_unlock(&ci
->i_ceph_lock
);
1877 truncate_pagecache(inode
, to
);
1879 spin_lock(&ci
->i_ceph_lock
);
1880 if (to
== ci
->i_truncate_size
) {
1881 ci
->i_truncate_pending
= 0;
1884 spin_unlock(&ci
->i_ceph_lock
);
1888 mutex_unlock(&ci
->i_truncate_mutex
);
1890 if (wrbuffer_refs
== 0)
1891 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1893 wake_up_all(&ci
->i_cap_wq
);
1896 static void ceph_inode_work(struct work_struct
*work
)
1898 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1900 struct inode
*inode
= &ci
->vfs_inode
;
1902 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
)) {
1903 dout("writeback %p\n", inode
);
1904 filemap_fdatawrite(&inode
->i_data
);
1906 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
))
1907 ceph_do_invalidate_pages(inode
);
1909 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
))
1910 __ceph_do_pending_vmtruncate(inode
);
1918 static const struct inode_operations ceph_symlink_iops
= {
1919 .get_link
= simple_get_link
,
1920 .setattr
= ceph_setattr
,
1921 .getattr
= ceph_getattr
,
1922 .listxattr
= ceph_listxattr
,
1925 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
1927 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1928 const unsigned int ia_valid
= attr
->ia_valid
;
1929 struct ceph_mds_request
*req
;
1930 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1931 struct ceph_cap_flush
*prealloc_cf
;
1933 int release
= 0, dirtied
= 0;
1936 int inode_dirty_flags
= 0;
1937 bool lock_snap_rwsem
= false;
1939 prealloc_cf
= ceph_alloc_cap_flush();
1943 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1946 ceph_free_cap_flush(prealloc_cf
);
1947 return PTR_ERR(req
);
1950 spin_lock(&ci
->i_ceph_lock
);
1951 issued
= __ceph_caps_issued(ci
, NULL
);
1953 if (!ci
->i_head_snapc
&&
1954 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1955 lock_snap_rwsem
= true;
1956 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1957 spin_unlock(&ci
->i_ceph_lock
);
1958 down_read(&mdsc
->snap_rwsem
);
1959 spin_lock(&ci
->i_ceph_lock
);
1960 issued
= __ceph_caps_issued(ci
, NULL
);
1964 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1966 if (ia_valid
& ATTR_UID
) {
1967 dout("setattr %p uid %d -> %d\n", inode
,
1968 from_kuid(&init_user_ns
, inode
->i_uid
),
1969 from_kuid(&init_user_ns
, attr
->ia_uid
));
1970 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1971 inode
->i_uid
= attr
->ia_uid
;
1972 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1973 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1974 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1975 req
->r_args
.setattr
.uid
= cpu_to_le32(
1976 from_kuid(&init_user_ns
, attr
->ia_uid
));
1977 mask
|= CEPH_SETATTR_UID
;
1978 release
|= CEPH_CAP_AUTH_SHARED
;
1981 if (ia_valid
& ATTR_GID
) {
1982 dout("setattr %p gid %d -> %d\n", inode
,
1983 from_kgid(&init_user_ns
, inode
->i_gid
),
1984 from_kgid(&init_user_ns
, attr
->ia_gid
));
1985 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1986 inode
->i_gid
= attr
->ia_gid
;
1987 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1988 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1989 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1990 req
->r_args
.setattr
.gid
= cpu_to_le32(
1991 from_kgid(&init_user_ns
, attr
->ia_gid
));
1992 mask
|= CEPH_SETATTR_GID
;
1993 release
|= CEPH_CAP_AUTH_SHARED
;
1996 if (ia_valid
& ATTR_MODE
) {
1997 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1999 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2000 inode
->i_mode
= attr
->ia_mode
;
2001 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2002 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2003 attr
->ia_mode
!= inode
->i_mode
) {
2004 inode
->i_mode
= attr
->ia_mode
;
2005 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
2006 mask
|= CEPH_SETATTR_MODE
;
2007 release
|= CEPH_CAP_AUTH_SHARED
;
2011 if (ia_valid
& ATTR_ATIME
) {
2012 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
2013 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
2014 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
2015 if (issued
& CEPH_CAP_FILE_EXCL
) {
2016 ci
->i_time_warp_seq
++;
2017 inode
->i_atime
= attr
->ia_atime
;
2018 dirtied
|= CEPH_CAP_FILE_EXCL
;
2019 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2020 timespec_compare(&inode
->i_atime
,
2021 &attr
->ia_atime
) < 0) {
2022 inode
->i_atime
= attr
->ia_atime
;
2023 dirtied
|= CEPH_CAP_FILE_WR
;
2024 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2025 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2026 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
2028 mask
|= CEPH_SETATTR_ATIME
;
2029 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
2033 if (ia_valid
& ATTR_MTIME
) {
2034 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
2035 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2036 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2037 if (issued
& CEPH_CAP_FILE_EXCL
) {
2038 ci
->i_time_warp_seq
++;
2039 inode
->i_mtime
= attr
->ia_mtime
;
2040 dirtied
|= CEPH_CAP_FILE_EXCL
;
2041 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2042 timespec_compare(&inode
->i_mtime
,
2043 &attr
->ia_mtime
) < 0) {
2044 inode
->i_mtime
= attr
->ia_mtime
;
2045 dirtied
|= CEPH_CAP_FILE_WR
;
2046 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2047 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2048 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
2050 mask
|= CEPH_SETATTR_MTIME
;
2051 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2055 if (ia_valid
& ATTR_SIZE
) {
2056 dout("setattr %p size %lld -> %lld\n", inode
,
2057 inode
->i_size
, attr
->ia_size
);
2058 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
2059 attr
->ia_size
> inode
->i_size
) {
2060 i_size_write(inode
, attr
->ia_size
);
2061 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2062 ci
->i_reported_size
= attr
->ia_size
;
2063 dirtied
|= CEPH_CAP_FILE_EXCL
;
2064 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2065 attr
->ia_size
!= inode
->i_size
) {
2066 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2067 req
->r_args
.setattr
.old_size
=
2068 cpu_to_le64(inode
->i_size
);
2069 mask
|= CEPH_SETATTR_SIZE
;
2070 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2075 /* these do nothing */
2076 if (ia_valid
& ATTR_CTIME
) {
2077 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2078 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2079 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
2080 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2081 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2082 only
? "ctime only" : "ignored");
2085 * if kernel wants to dirty ctime but nothing else,
2086 * we need to choose a cap to dirty under, or do
2087 * a almost-no-op setattr
2089 if (issued
& CEPH_CAP_AUTH_EXCL
)
2090 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2091 else if (issued
& CEPH_CAP_FILE_EXCL
)
2092 dirtied
|= CEPH_CAP_FILE_EXCL
;
2093 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2094 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2096 mask
|= CEPH_SETATTR_CTIME
;
2099 if (ia_valid
& ATTR_FILE
)
2100 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2103 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2105 inode
->i_ctime
= attr
->ia_ctime
;
2109 spin_unlock(&ci
->i_ceph_lock
);
2110 if (lock_snap_rwsem
)
2111 up_read(&mdsc
->snap_rwsem
);
2113 if (inode_dirty_flags
)
2114 __mark_inode_dirty(inode
, inode_dirty_flags
);
2118 req
->r_inode
= inode
;
2120 req
->r_inode_drop
= release
;
2121 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2122 req
->r_num_caps
= 1;
2123 req
->r_stamp
= attr
->ia_ctime
;
2124 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2126 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2127 ceph_cap_string(dirtied
), mask
);
2129 ceph_mdsc_put_request(req
);
2130 ceph_free_cap_flush(prealloc_cf
);
2132 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2133 __ceph_do_pending_vmtruncate(inode
);
2141 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2143 struct inode
*inode
= d_inode(dentry
);
2146 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2149 err
= setattr_prepare(dentry
, attr
);
2153 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2154 ceph_quota_is_max_bytes_exceeded(inode
, attr
->ia_size
))
2157 err
= __ceph_setattr(inode
, attr
);
2159 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2160 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2166 * Verify that we have a lease on the given mask. If not,
2167 * do a getattr against an mds.
2169 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2170 int mask
, bool force
)
2172 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2173 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2174 struct ceph_mds_request
*req
;
2177 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2178 dout("do_getattr inode %p SNAPDIR\n", inode
);
2182 dout("do_getattr inode %p mask %s mode 0%o\n",
2183 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2184 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2187 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
2189 return PTR_ERR(req
);
2190 req
->r_inode
= inode
;
2192 req
->r_num_caps
= 1;
2193 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2194 req
->r_locked_page
= locked_page
;
2195 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2196 if (locked_page
&& err
== 0) {
2197 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2198 if (inline_version
== 0) {
2199 /* the reply is supposed to contain inline data */
2201 } else if (inline_version
== CEPH_INLINE_NONE
) {
2204 err
= req
->r_reply_info
.targeti
.inline_len
;
2207 ceph_mdsc_put_request(req
);
2208 dout("do_getattr result=%d\n", err
);
2214 * Check inode permissions. We verify we have a valid value for
2215 * the AUTH cap, then call the generic handler.
2217 int ceph_permission(struct inode
*inode
, int mask
)
2221 if (mask
& MAY_NOT_BLOCK
)
2224 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2227 err
= generic_permission(inode
, mask
);
2232 * Get all attributes. Hopefully somedata we'll have a statlite()
2233 * and can limit the fields we require to be accurate.
2235 int ceph_getattr(const struct path
*path
, struct kstat
*stat
,
2236 u32 request_mask
, unsigned int flags
)
2238 struct inode
*inode
= d_inode(path
->dentry
);
2239 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2242 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2244 generic_fillattr(inode
, stat
);
2245 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2246 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2247 stat
->dev
= ceph_snap(inode
);
2250 if (S_ISDIR(inode
->i_mode
)) {
2251 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2253 stat
->size
= ci
->i_rbytes
;
2255 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2257 stat
->blksize
= 65536;