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_invalidate_work(struct work_struct
*work
);
37 static void ceph_writeback_work(struct work_struct
*work
);
38 static void ceph_vmtruncate_work(struct work_struct
*work
);
41 * find or create an inode, given the ceph ino number
43 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
45 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
46 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
50 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
53 ino_t t
= ceph_vino_to_ino(vino
);
55 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
57 return ERR_PTR(-ENOMEM
);
58 if (inode
->i_state
& I_NEW
) {
59 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
60 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
61 unlock_new_inode(inode
);
64 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
70 * get/constuct snapdir inode for a given directory
72 struct inode
*ceph_get_snapdir(struct inode
*parent
)
74 struct ceph_vino vino
= {
75 .ino
= ceph_ino(parent
),
78 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
79 struct ceph_inode_info
*ci
= ceph_inode(inode
);
81 BUG_ON(!S_ISDIR(parent
->i_mode
));
84 inode
->i_mode
= parent
->i_mode
;
85 inode
->i_uid
= parent
->i_uid
;
86 inode
->i_gid
= parent
->i_gid
;
87 inode
->i_op
= &ceph_snapdir_iops
;
88 inode
->i_fop
= &ceph_snapdir_fops
;
89 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
94 const struct inode_operations ceph_file_iops
= {
95 .permission
= ceph_permission
,
96 .setattr
= ceph_setattr
,
97 .getattr
= ceph_getattr
,
98 .listxattr
= ceph_listxattr
,
99 .get_acl
= ceph_get_acl
,
100 .set_acl
= ceph_set_acl
,
105 * We use a 'frag tree' to keep track of the MDS's directory fragments
106 * for a given inode (usually there is just a single fragment). We
107 * need to know when a child frag is delegated to a new MDS, or when
108 * it is flagged as replicated, so we can direct our requests
113 * find/create a frag in the tree
115 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
119 struct rb_node
*parent
= NULL
;
120 struct ceph_inode_frag
*frag
;
123 p
= &ci
->i_fragtree
.rb_node
;
126 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
127 c
= ceph_frag_compare(f
, frag
->frag
);
136 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
138 return ERR_PTR(-ENOMEM
);
145 rb_link_node(&frag
->node
, parent
, p
);
146 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
148 dout("get_or_create_frag added %llx.%llx frag %x\n",
149 ceph_vinop(&ci
->vfs_inode
), f
);
154 * find a specific frag @f
156 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
158 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
161 struct ceph_inode_frag
*frag
=
162 rb_entry(n
, struct ceph_inode_frag
, node
);
163 int c
= ceph_frag_compare(f
, frag
->frag
);
175 * Choose frag containing the given value @v. If @pfrag is
176 * specified, copy the frag delegation info to the caller if
179 static u32
__ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
180 struct ceph_inode_frag
*pfrag
, int *found
)
182 u32 t
= ceph_frag_make(0, 0);
183 struct ceph_inode_frag
*frag
;
191 WARN_ON(!ceph_frag_contains_value(t
, v
));
192 frag
= __ceph_find_frag(ci
, t
);
194 break; /* t is a leaf */
195 if (frag
->split_by
== 0) {
197 memcpy(pfrag
, frag
, sizeof(*pfrag
));
204 nway
= 1 << frag
->split_by
;
205 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
206 frag
->split_by
, nway
);
207 for (i
= 0; i
< nway
; i
++) {
208 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
209 if (ceph_frag_contains_value(n
, v
)) {
216 dout("choose_frag(%x) = %x\n", v
, t
);
221 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
222 struct ceph_inode_frag
*pfrag
, int *found
)
225 mutex_lock(&ci
->i_fragtree_mutex
);
226 ret
= __ceph_choose_frag(ci
, v
, pfrag
, found
);
227 mutex_unlock(&ci
->i_fragtree_mutex
);
232 * Process dirfrag (delegation) info from the mds. Include leaf
233 * fragment in tree ONLY if ndist > 0. Otherwise, only
234 * branches/splits are included in i_fragtree)
236 static int ceph_fill_dirfrag(struct inode
*inode
,
237 struct ceph_mds_reply_dirfrag
*dirinfo
)
239 struct ceph_inode_info
*ci
= ceph_inode(inode
);
240 struct ceph_inode_frag
*frag
;
241 u32 id
= le32_to_cpu(dirinfo
->frag
);
242 int mds
= le32_to_cpu(dirinfo
->auth
);
243 int ndist
= le32_to_cpu(dirinfo
->ndist
);
248 spin_lock(&ci
->i_ceph_lock
);
250 diri_auth
= ci
->i_auth_cap
->mds
;
251 spin_unlock(&ci
->i_ceph_lock
);
253 if (mds
== -1) /* CDIR_AUTH_PARENT */
256 mutex_lock(&ci
->i_fragtree_mutex
);
257 if (ndist
== 0 && mds
== diri_auth
) {
258 /* no delegation info needed. */
259 frag
= __ceph_find_frag(ci
, id
);
262 if (frag
->split_by
== 0) {
263 /* tree leaf, remove */
264 dout("fill_dirfrag removed %llx.%llx frag %x"
265 " (no ref)\n", ceph_vinop(inode
), id
);
266 rb_erase(&frag
->node
, &ci
->i_fragtree
);
269 /* tree branch, keep and clear */
270 dout("fill_dirfrag cleared %llx.%llx frag %x"
271 " referral\n", ceph_vinop(inode
), id
);
279 /* find/add this frag to store mds delegation info */
280 frag
= __get_or_create_frag(ci
, id
);
282 /* this is not the end of the world; we can continue
283 with bad/inaccurate delegation info */
284 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
285 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
291 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
292 for (i
= 0; i
< frag
->ndist
; i
++)
293 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
294 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
295 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
298 mutex_unlock(&ci
->i_fragtree_mutex
);
302 static int frag_tree_split_cmp(const void *l
, const void *r
)
304 struct ceph_frag_tree_split
*ls
= (struct ceph_frag_tree_split
*)l
;
305 struct ceph_frag_tree_split
*rs
= (struct ceph_frag_tree_split
*)r
;
306 return ceph_frag_compare(le32_to_cpu(ls
->frag
),
307 le32_to_cpu(rs
->frag
));
310 static bool is_frag_child(u32 f
, struct ceph_inode_frag
*frag
)
313 return f
== ceph_frag_make(0, 0);
314 if (ceph_frag_bits(f
) != ceph_frag_bits(frag
->frag
) + frag
->split_by
)
316 return ceph_frag_contains_value(frag
->frag
, ceph_frag_value(f
));
319 static int ceph_fill_fragtree(struct inode
*inode
,
320 struct ceph_frag_tree_head
*fragtree
,
321 struct ceph_mds_reply_dirfrag
*dirinfo
)
323 struct ceph_inode_info
*ci
= ceph_inode(inode
);
324 struct ceph_inode_frag
*frag
, *prev_frag
= NULL
;
325 struct rb_node
*rb_node
;
326 unsigned i
, split_by
, nsplits
;
330 mutex_lock(&ci
->i_fragtree_mutex
);
331 nsplits
= le32_to_cpu(fragtree
->nsplits
);
332 if (nsplits
!= ci
->i_fragtree_nsplits
) {
334 } else if (nsplits
) {
335 i
= prandom_u32() % nsplits
;
336 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
337 if (!__ceph_find_frag(ci
, id
))
339 } else if (!RB_EMPTY_ROOT(&ci
->i_fragtree
)) {
340 rb_node
= rb_first(&ci
->i_fragtree
);
341 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
342 if (frag
->frag
!= ceph_frag_make(0, 0) || rb_next(rb_node
))
345 if (!update
&& dirinfo
) {
346 id
= le32_to_cpu(dirinfo
->frag
);
347 if (id
!= __ceph_choose_frag(ci
, id
, NULL
, NULL
))
354 sort(fragtree
->splits
, nsplits
, sizeof(fragtree
->splits
[0]),
355 frag_tree_split_cmp
, NULL
);
358 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode
));
359 rb_node
= rb_first(&ci
->i_fragtree
);
360 for (i
= 0; i
< nsplits
; i
++) {
361 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
362 split_by
= le32_to_cpu(fragtree
->splits
[i
].by
);
363 if (split_by
== 0 || ceph_frag_bits(id
) + split_by
> 24) {
364 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
365 "frag %x split by %d\n", ceph_vinop(inode
),
366 i
, nsplits
, id
, split_by
);
371 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
372 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
373 if (frag
->frag
!= id
)
376 rb_node
= rb_next(rb_node
);
379 rb_node
= rb_next(rb_node
);
380 /* delete stale split/leaf node */
381 if (frag
->split_by
> 0 ||
382 !is_frag_child(frag
->frag
, prev_frag
)) {
383 rb_erase(&frag
->node
, &ci
->i_fragtree
);
384 if (frag
->split_by
> 0)
385 ci
->i_fragtree_nsplits
--;
391 frag
= __get_or_create_frag(ci
, id
);
395 if (frag
->split_by
== 0)
396 ci
->i_fragtree_nsplits
++;
397 frag
->split_by
= split_by
;
398 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
402 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
403 rb_node
= rb_next(rb_node
);
404 /* delete stale split/leaf node */
405 if (frag
->split_by
> 0 ||
406 !is_frag_child(frag
->frag
, prev_frag
)) {
407 rb_erase(&frag
->node
, &ci
->i_fragtree
);
408 if (frag
->split_by
> 0)
409 ci
->i_fragtree_nsplits
--;
414 mutex_unlock(&ci
->i_fragtree_mutex
);
419 * initialize a newly allocated inode.
421 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
423 struct ceph_inode_info
*ci
;
426 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
430 dout("alloc_inode %p\n", &ci
->vfs_inode
);
432 spin_lock_init(&ci
->i_ceph_lock
);
435 ci
->i_inline_version
= 0;
436 ci
->i_time_warp_seq
= 0;
437 ci
->i_ceph_flags
= 0;
438 atomic64_set(&ci
->i_ordered_count
, 1);
439 atomic64_set(&ci
->i_release_count
, 1);
440 atomic64_set(&ci
->i_complete_seq
[0], 0);
441 atomic64_set(&ci
->i_complete_seq
[1], 0);
442 ci
->i_symlink
= NULL
;
444 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
445 RCU_INIT_POINTER(ci
->i_layout
.pool_ns
, NULL
);
447 ci
->i_fragtree
= RB_ROOT
;
448 mutex_init(&ci
->i_fragtree_mutex
);
450 ci
->i_xattrs
.blob
= NULL
;
451 ci
->i_xattrs
.prealloc_blob
= NULL
;
452 ci
->i_xattrs
.dirty
= false;
453 ci
->i_xattrs
.index
= RB_ROOT
;
454 ci
->i_xattrs
.count
= 0;
455 ci
->i_xattrs
.names_size
= 0;
456 ci
->i_xattrs
.vals_size
= 0;
457 ci
->i_xattrs
.version
= 0;
458 ci
->i_xattrs
.index_version
= 0;
460 ci
->i_caps
= RB_ROOT
;
461 ci
->i_auth_cap
= NULL
;
462 ci
->i_dirty_caps
= 0;
463 ci
->i_flushing_caps
= 0;
464 INIT_LIST_HEAD(&ci
->i_dirty_item
);
465 INIT_LIST_HEAD(&ci
->i_flushing_item
);
466 ci
->i_prealloc_cap_flush
= NULL
;
467 INIT_LIST_HEAD(&ci
->i_cap_flush_list
);
468 init_waitqueue_head(&ci
->i_cap_wq
);
469 ci
->i_hold_caps_min
= 0;
470 ci
->i_hold_caps_max
= 0;
471 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
472 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
473 ci
->i_head_snapc
= NULL
;
476 for (i
= 0; i
< CEPH_FILE_MODE_BITS
; i
++)
477 ci
->i_nr_by_mode
[i
] = 0;
479 mutex_init(&ci
->i_truncate_mutex
);
480 ci
->i_truncate_seq
= 0;
481 ci
->i_truncate_size
= 0;
482 ci
->i_truncate_pending
= 0;
485 ci
->i_reported_size
= 0;
486 ci
->i_wanted_max_size
= 0;
487 ci
->i_requested_max_size
= 0;
491 ci
->i_rdcache_ref
= 0;
494 ci
->i_wrbuffer_ref
= 0;
495 ci
->i_wrbuffer_ref_head
= 0;
496 atomic_set(&ci
->i_filelock_ref
, 0);
497 atomic_set(&ci
->i_shared_gen
, 0);
498 ci
->i_rdcache_gen
= 0;
499 ci
->i_rdcache_revoking
= 0;
501 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
502 INIT_LIST_HEAD(&ci
->i_unsafe_iops
);
503 spin_lock_init(&ci
->i_unsafe_lock
);
505 ci
->i_snap_realm
= NULL
;
506 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
507 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
509 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
510 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
512 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
514 ceph_fscache_inode_init(ci
);
516 return &ci
->vfs_inode
;
519 static void ceph_i_callback(struct rcu_head
*head
)
521 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
522 struct ceph_inode_info
*ci
= ceph_inode(inode
);
524 kmem_cache_free(ceph_inode_cachep
, ci
);
527 void ceph_destroy_inode(struct inode
*inode
)
529 struct ceph_inode_info
*ci
= ceph_inode(inode
);
530 struct ceph_inode_frag
*frag
;
533 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
535 ceph_fscache_unregister_inode_cookie(ci
);
537 __ceph_remove_caps(inode
);
540 * we may still have a snap_realm reference if there are stray
541 * caps in i_snap_caps.
543 if (ci
->i_snap_realm
) {
544 struct ceph_mds_client
*mdsc
=
545 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
546 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
548 dout(" dropping residual ref to snap realm %p\n", realm
);
549 spin_lock(&realm
->inodes_with_caps_lock
);
550 list_del_init(&ci
->i_snap_realm_item
);
551 spin_unlock(&realm
->inodes_with_caps_lock
);
552 ceph_put_snap_realm(mdsc
, realm
);
555 kfree(ci
->i_symlink
);
556 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
557 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
558 rb_erase(n
, &ci
->i_fragtree
);
561 ci
->i_fragtree_nsplits
= 0;
563 __ceph_destroy_xattrs(ci
);
564 if (ci
->i_xattrs
.blob
)
565 ceph_buffer_put(ci
->i_xattrs
.blob
);
566 if (ci
->i_xattrs
.prealloc_blob
)
567 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
569 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
571 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
574 int ceph_drop_inode(struct inode
*inode
)
577 * Positve dentry and corresponding inode are always accompanied
578 * in MDS reply. So no need to keep inode in the cache after
579 * dropping all its aliases.
584 static inline blkcnt_t
calc_inode_blocks(u64 size
)
586 return (size
+ (1<<9) - 1) >> 9;
590 * Helpers to fill in size, ctime, mtime, and atime. We have to be
591 * careful because either the client or MDS may have more up to date
592 * info, depending on which capabilities are held, and whether
593 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
594 * and size are monotonically increasing, except when utimes() or
595 * truncate() increments the corresponding _seq values.)
597 int ceph_fill_file_size(struct inode
*inode
, int issued
,
598 u32 truncate_seq
, u64 truncate_size
, u64 size
)
600 struct ceph_inode_info
*ci
= ceph_inode(inode
);
603 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
604 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
605 dout("size %lld -> %llu\n", inode
->i_size
, size
);
606 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
607 pr_err("fill_file_size non-zero size for directory\n");
610 i_size_write(inode
, size
);
611 inode
->i_blocks
= calc_inode_blocks(size
);
612 ci
->i_reported_size
= size
;
613 if (truncate_seq
!= ci
->i_truncate_seq
) {
614 dout("truncate_seq %u -> %u\n",
615 ci
->i_truncate_seq
, truncate_seq
);
616 ci
->i_truncate_seq
= truncate_seq
;
618 /* the MDS should have revoked these caps */
619 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
622 CEPH_CAP_FILE_LAZYIO
));
624 * If we hold relevant caps, or in the case where we're
625 * not the only client referencing this file and we
626 * don't hold those caps, then we need to check whether
627 * the file is either opened or mmaped
629 if ((issued
& (CEPH_CAP_FILE_CACHE
|
630 CEPH_CAP_FILE_BUFFER
)) ||
631 mapping_mapped(inode
->i_mapping
) ||
632 __ceph_caps_file_wanted(ci
)) {
633 ci
->i_truncate_pending
++;
638 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
639 ci
->i_truncate_size
!= truncate_size
) {
640 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
642 ci
->i_truncate_size
= truncate_size
;
646 ceph_fscache_invalidate(inode
);
651 void ceph_fill_file_time(struct inode
*inode
, int issued
,
652 u64 time_warp_seq
, struct timespec
*ctime
,
653 struct timespec
*mtime
, struct timespec
*atime
)
655 struct ceph_inode_info
*ci
= ceph_inode(inode
);
658 if (issued
& (CEPH_CAP_FILE_EXCL
|
660 CEPH_CAP_FILE_BUFFER
|
662 CEPH_CAP_XATTR_EXCL
)) {
663 if (ci
->i_version
== 0 ||
664 timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
665 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
666 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
667 ctime
->tv_sec
, ctime
->tv_nsec
);
668 inode
->i_ctime
= *ctime
;
670 if (ci
->i_version
== 0 ||
671 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
672 /* the MDS did a utimes() */
673 dout("mtime %ld.%09ld -> %ld.%09ld "
675 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
676 mtime
->tv_sec
, mtime
->tv_nsec
,
677 ci
->i_time_warp_seq
, (int)time_warp_seq
);
679 inode
->i_mtime
= *mtime
;
680 inode
->i_atime
= *atime
;
681 ci
->i_time_warp_seq
= time_warp_seq
;
682 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
683 /* nobody did utimes(); take the max */
684 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
685 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
686 inode
->i_mtime
.tv_sec
,
687 inode
->i_mtime
.tv_nsec
,
688 mtime
->tv_sec
, mtime
->tv_nsec
);
689 inode
->i_mtime
= *mtime
;
691 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
692 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
693 inode
->i_atime
.tv_sec
,
694 inode
->i_atime
.tv_nsec
,
695 atime
->tv_sec
, atime
->tv_nsec
);
696 inode
->i_atime
= *atime
;
698 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
699 /* we did a utimes(); ignore mds values */
704 /* we have no write|excl caps; whatever the MDS says is true */
705 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
706 inode
->i_ctime
= *ctime
;
707 inode
->i_mtime
= *mtime
;
708 inode
->i_atime
= *atime
;
709 ci
->i_time_warp_seq
= time_warp_seq
;
714 if (warn
) /* time_warp_seq shouldn't go backwards */
715 dout("%p mds time_warp_seq %llu < %u\n",
716 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
720 * Populate an inode based on info from mds. May be called on new or
723 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
724 struct ceph_mds_reply_info_in
*iinfo
,
725 struct ceph_mds_reply_dirfrag
*dirinfo
,
726 struct ceph_mds_session
*session
,
727 unsigned long ttl_from
, int cap_fmode
,
728 struct ceph_cap_reservation
*caps_reservation
)
730 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
731 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
732 struct ceph_inode_info
*ci
= ceph_inode(inode
);
733 int issued
= 0, implemented
, new_issued
;
734 struct timespec mtime
, atime
, ctime
;
735 struct ceph_buffer
*xattr_blob
= NULL
;
736 struct ceph_string
*pool_ns
= NULL
;
737 struct ceph_cap
*new_cap
= NULL
;
740 bool queue_trunc
= false;
741 bool new_version
= false;
742 bool fill_inline
= false;
744 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
745 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
748 /* prealloc new cap struct */
749 if (info
->cap
.caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
750 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
753 * prealloc xattr data, if it looks like we'll need it. only
754 * if len > 4 (meaning there are actually xattrs; the first 4
755 * bytes are the xattr count).
757 if (iinfo
->xattr_len
> 4) {
758 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
760 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
764 if (iinfo
->pool_ns_len
> 0)
765 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
768 spin_lock(&ci
->i_ceph_lock
);
771 * provided version will be odd if inode value is projected,
772 * even if stable. skip the update if we have newer stable
773 * info (ours>=theirs, e.g. due to racing mds replies), unless
774 * we are getting projected (unstable) info (in which case the
775 * version is odd, and we want ours>theirs).
781 if (ci
->i_version
== 0 ||
782 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
783 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
786 issued
= __ceph_caps_issued(ci
, &implemented
);
787 issued
|= implemented
| __ceph_caps_dirty(ci
);
788 new_issued
= ~issued
& le32_to_cpu(info
->cap
.caps
);
791 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
792 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
794 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
795 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
796 inode
->i_mode
= le32_to_cpu(info
->mode
);
797 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
798 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
799 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
800 from_kuid(&init_user_ns
, inode
->i_uid
),
801 from_kgid(&init_user_ns
, inode
->i_gid
));
804 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
805 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
806 set_nlink(inode
, le32_to_cpu(info
->nlink
));
808 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
809 /* be careful with mtime, atime, size */
810 ceph_decode_timespec(&atime
, &info
->atime
);
811 ceph_decode_timespec(&mtime
, &info
->mtime
);
812 ceph_decode_timespec(&ctime
, &info
->ctime
);
813 ceph_fill_file_time(inode
, issued
,
814 le32_to_cpu(info
->time_warp_seq
),
815 &ctime
, &mtime
, &atime
);
819 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
820 s64 old_pool
= ci
->i_layout
.pool_id
;
821 struct ceph_string
*old_ns
;
823 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
824 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
825 lockdep_is_held(&ci
->i_ceph_lock
));
826 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
828 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
829 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
833 queue_trunc
= ceph_fill_file_size(inode
, issued
,
834 le32_to_cpu(info
->truncate_seq
),
835 le64_to_cpu(info
->truncate_size
),
836 le64_to_cpu(info
->size
));
837 /* only update max_size on auth cap */
838 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
839 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
840 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
841 le64_to_cpu(info
->max_size
));
842 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
847 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
848 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
849 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
850 if (ci
->i_xattrs
.blob
)
851 ceph_buffer_put(ci
->i_xattrs
.blob
);
852 ci
->i_xattrs
.blob
= xattr_blob
;
854 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
855 iinfo
->xattr_data
, iinfo
->xattr_len
);
856 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
857 ceph_forget_all_cached_acls(inode
);
861 /* finally update i_version */
862 ci
->i_version
= le64_to_cpu(info
->version
);
864 inode
->i_mapping
->a_ops
= &ceph_aops
;
866 switch (inode
->i_mode
& S_IFMT
) {
871 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
872 inode
->i_op
= &ceph_file_iops
;
875 inode
->i_op
= &ceph_file_iops
;
876 inode
->i_fop
= &ceph_file_fops
;
879 inode
->i_op
= &ceph_symlink_iops
;
880 if (!ci
->i_symlink
) {
881 u32 symlen
= iinfo
->symlink_len
;
884 spin_unlock(&ci
->i_ceph_lock
);
886 if (symlen
!= i_size_read(inode
)) {
887 pr_err("fill_inode %llx.%llx BAD symlink "
888 "size %lld\n", ceph_vinop(inode
),
890 i_size_write(inode
, symlen
);
891 inode
->i_blocks
= calc_inode_blocks(symlen
);
895 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
899 spin_lock(&ci
->i_ceph_lock
);
903 kfree(sym
); /* lost a race */
905 inode
->i_link
= ci
->i_symlink
;
908 inode
->i_op
= &ceph_dir_iops
;
909 inode
->i_fop
= &ceph_dir_fops
;
911 ci
->i_dir_layout
= iinfo
->dir_layout
;
913 ci
->i_files
= le64_to_cpu(info
->files
);
914 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
915 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
916 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
917 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
918 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
921 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
922 ceph_vinop(inode
), inode
->i_mode
);
925 /* were we issued a capability? */
926 if (info
->cap
.caps
) {
927 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
928 unsigned caps
= le32_to_cpu(info
->cap
.caps
);
929 ceph_add_cap(inode
, session
,
930 le64_to_cpu(info
->cap
.cap_id
),
932 le32_to_cpu(info
->cap
.wanted
),
933 le32_to_cpu(info
->cap
.seq
),
934 le32_to_cpu(info
->cap
.mseq
),
935 le64_to_cpu(info
->cap
.realm
),
936 info
->cap
.flags
, &new_cap
);
938 /* set dir completion flag? */
939 if (S_ISDIR(inode
->i_mode
) &&
940 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
941 (caps
& CEPH_CAP_FILE_SHARED
) &&
942 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
943 !__ceph_dir_is_complete(ci
)) {
944 dout(" marking %p complete (empty)\n", inode
);
945 i_size_write(inode
, 0);
946 __ceph_dir_set_complete(ci
,
947 atomic64_read(&ci
->i_release_count
),
948 atomic64_read(&ci
->i_ordered_count
));
953 dout(" %p got snap_caps %s\n", inode
,
954 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
955 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
957 __ceph_get_fmode(ci
, cap_fmode
);
959 } else if (cap_fmode
>= 0) {
960 pr_warn("mds issued no caps on %llx.%llx\n",
962 __ceph_get_fmode(ci
, cap_fmode
);
965 if (iinfo
->inline_version
> 0 &&
966 iinfo
->inline_version
>= ci
->i_inline_version
) {
967 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
968 ci
->i_inline_version
= iinfo
->inline_version
;
969 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
971 (le32_to_cpu(info
->cap
.caps
) & cache_caps
)))
975 spin_unlock(&ci
->i_ceph_lock
);
978 ceph_fill_inline_data(inode
, locked_page
,
979 iinfo
->inline_data
, iinfo
->inline_len
);
982 wake_up_all(&ci
->i_cap_wq
);
984 /* queue truncate if we saw i_size decrease */
986 ceph_queue_vmtruncate(inode
);
988 /* populate frag tree */
989 if (S_ISDIR(inode
->i_mode
))
990 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
992 /* update delegation info? */
994 ceph_fill_dirfrag(inode
, dirinfo
);
999 ceph_put_cap(mdsc
, new_cap
);
1001 ceph_buffer_put(xattr_blob
);
1002 ceph_put_string(pool_ns
);
1007 * caller should hold session s_mutex.
1009 static void update_dentry_lease(struct dentry
*dentry
,
1010 struct ceph_mds_reply_lease
*lease
,
1011 struct ceph_mds_session
*session
,
1012 unsigned long from_time
,
1013 struct ceph_vino
*tgt_vino
,
1014 struct ceph_vino
*dir_vino
)
1016 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1017 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1018 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1019 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1021 struct ceph_mds_session
*old_lease_session
= NULL
;
1024 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1025 * we expect a negative dentry.
1027 if (!tgt_vino
&& d_really_is_positive(dentry
))
1030 if (tgt_vino
&& (d_really_is_negative(dentry
) ||
1031 !ceph_ino_compare(d_inode(dentry
), tgt_vino
)))
1034 spin_lock(&dentry
->d_lock
);
1035 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1036 dentry
, duration
, ttl
);
1038 dir
= d_inode(dentry
->d_parent
);
1040 /* make sure parent matches dir_vino */
1041 if (!ceph_ino_compare(dir
, dir_vino
))
1044 /* only track leases on regular dentries */
1045 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1048 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1053 if (di
->lease_gen
== session
->s_cap_gen
&&
1054 time_before(ttl
, di
->time
))
1055 goto out_unlock
; /* we already have a newer lease. */
1057 if (di
->lease_session
&& di
->lease_session
!= session
) {
1058 old_lease_session
= di
->lease_session
;
1059 di
->lease_session
= NULL
;
1062 ceph_dentry_lru_touch(dentry
);
1064 if (!di
->lease_session
)
1065 di
->lease_session
= ceph_get_mds_session(session
);
1066 di
->lease_gen
= session
->s_cap_gen
;
1067 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1068 di
->lease_renew_after
= half_ttl
;
1069 di
->lease_renew_from
= 0;
1072 spin_unlock(&dentry
->d_lock
);
1073 if (old_lease_session
)
1074 ceph_put_mds_session(old_lease_session
);
1078 * splice a dentry to an inode.
1079 * caller must hold directory i_mutex for this to be safe.
1081 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
)
1083 struct dentry
*realdn
;
1085 BUG_ON(d_inode(dn
));
1087 /* dn must be unhashed */
1088 if (!d_unhashed(dn
))
1090 realdn
= d_splice_alias(in
, dn
);
1091 if (IS_ERR(realdn
)) {
1092 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1093 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1094 dn
= realdn
; /* note realdn contains the error */
1096 } else if (realdn
) {
1097 dout("dn %p (%d) spliced with %p (%d) "
1098 "inode %p ino %llx.%llx\n",
1100 realdn
, d_count(realdn
),
1101 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1105 BUG_ON(!ceph_dentry(dn
));
1106 dout("dn %p attached to %p ino %llx.%llx\n",
1107 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1114 * Incorporate results into the local cache. This is either just
1115 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1118 * A reply may contain
1119 * a directory inode along with a dentry.
1120 * and/or a target inode
1122 * Called with snap_rwsem (read).
1124 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1126 struct ceph_mds_session
*session
= req
->r_session
;
1127 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1128 struct inode
*in
= NULL
;
1129 struct ceph_vino tvino
, dvino
;
1130 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1133 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1134 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1136 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1137 dout("fill_trace reply is empty!\n");
1138 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1139 ceph_invalidate_dir_request(req
);
1143 if (rinfo
->head
->is_dentry
) {
1144 struct inode
*dir
= req
->r_parent
;
1147 err
= fill_inode(dir
, NULL
,
1148 &rinfo
->diri
, rinfo
->dirfrag
,
1149 session
, req
->r_request_started
, -1,
1150 &req
->r_caps_reservation
);
1157 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1159 struct dentry
*dn
, *parent
;
1161 BUG_ON(!rinfo
->head
->is_target
);
1162 BUG_ON(req
->r_dentry
);
1164 parent
= d_find_any_alias(dir
);
1167 dname
.name
= rinfo
->dname
;
1168 dname
.len
= rinfo
->dname_len
;
1169 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1170 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1171 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1173 dn
= d_lookup(parent
, &dname
);
1174 dout("d_lookup on parent=%p name=%.*s got %p\n",
1175 parent
, dname
.len
, dname
.name
, dn
);
1178 dn
= d_alloc(parent
, &dname
);
1179 dout("d_alloc %p '%.*s' = %p\n", parent
,
1180 dname
.len
, dname
.name
, dn
);
1187 } else if (d_really_is_positive(dn
) &&
1188 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1189 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1190 dout(" dn %p points to wrong inode %p\n",
1192 ceph_dir_clear_ordered(dir
);
1203 if (rinfo
->head
->is_target
) {
1204 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1205 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1207 in
= ceph_get_inode(sb
, tvino
);
1212 req
->r_target_inode
= in
;
1214 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1215 session
, req
->r_request_started
,
1216 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1217 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1218 &req
->r_caps_reservation
);
1220 pr_err("fill_inode badness %p %llx.%llx\n",
1221 in
, ceph_vinop(in
));
1227 * ignore null lease/binding on snapdir ENOENT, or else we
1228 * will have trouble splicing in the virtual snapdir later
1230 if (rinfo
->head
->is_dentry
&&
1231 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1232 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1233 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1234 fsc
->mount_options
->snapdir_name
,
1235 req
->r_dentry
->d_name
.len
))) {
1237 * lookup link rename : null -> possibly existing inode
1238 * mknod symlink mkdir : null -> new inode
1239 * unlink : linked -> null
1241 struct inode
*dir
= req
->r_parent
;
1242 struct dentry
*dn
= req
->r_dentry
;
1243 bool have_dir_cap
, have_lease
;
1247 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1249 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1250 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1252 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1253 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1255 /* do we have a lease on the whole dir? */
1257 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1258 CEPH_CAP_FILE_SHARED
);
1260 /* do we have a dn lease? */
1261 have_lease
= have_dir_cap
||
1262 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1264 dout("fill_trace no dentry lease or dir cap\n");
1267 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1268 struct inode
*olddir
= req
->r_old_dentry_dir
;
1271 dout(" src %p '%pd' dst %p '%pd'\n",
1275 dout("fill_trace doing d_move %p -> %p\n",
1276 req
->r_old_dentry
, dn
);
1278 /* d_move screws up sibling dentries' offsets */
1279 ceph_dir_clear_ordered(dir
);
1280 ceph_dir_clear_ordered(olddir
);
1282 d_move(req
->r_old_dentry
, dn
);
1283 dout(" src %p '%pd' dst %p '%pd'\n",
1288 /* ensure target dentry is invalidated, despite
1289 rehashing bug in vfs_rename_dir */
1290 ceph_invalidate_dentry_lease(dn
);
1292 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1293 ceph_dentry(req
->r_old_dentry
)->offset
);
1295 dn
= req
->r_old_dentry
; /* use old_dentry */
1299 if (!rinfo
->head
->is_target
) {
1300 dout("fill_trace null dentry\n");
1301 if (d_really_is_positive(dn
)) {
1302 ceph_dir_clear_ordered(dir
);
1303 dout("d_delete %p\n", dn
);
1305 } else if (have_lease
) {
1308 update_dentry_lease(dn
, rinfo
->dlease
,
1310 req
->r_request_started
,
1316 /* attach proper inode */
1317 if (d_really_is_negative(dn
)) {
1318 ceph_dir_clear_ordered(dir
);
1320 dn
= splice_dentry(dn
, in
);
1325 req
->r_dentry
= dn
; /* may have spliced */
1326 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1327 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1328 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1330 ceph_dir_clear_ordered(dir
);
1336 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1337 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1338 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1339 req
->r_request_started
,
1342 dout(" final dn %p\n", dn
);
1343 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1344 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1345 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1346 struct dentry
*dn
= req
->r_dentry
;
1347 struct inode
*dir
= req
->r_parent
;
1349 /* fill out a snapdir LOOKUPSNAP dentry */
1352 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1353 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1354 ceph_dir_clear_ordered(dir
);
1356 dn
= splice_dentry(dn
, in
);
1361 req
->r_dentry
= dn
; /* may have spliced */
1362 } else if (rinfo
->head
->is_dentry
) {
1363 struct ceph_vino
*ptvino
= NULL
;
1365 if ((le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) ||
1366 le32_to_cpu(rinfo
->dlease
->duration_ms
)) {
1367 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1368 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1370 if (rinfo
->head
->is_target
) {
1371 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1372 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1376 update_dentry_lease(req
->r_dentry
, rinfo
->dlease
,
1377 session
, req
->r_request_started
, ptvino
,
1380 dout("%s: no dentry lease or dir cap\n", __func__
);
1384 dout("fill_trace done err=%d\n", err
);
1389 * Prepopulate our cache with readdir results, leases, etc.
1391 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1392 struct ceph_mds_session
*session
)
1394 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1397 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1398 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1399 struct ceph_vino vino
;
1403 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1404 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1406 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1409 dout("new_inode badness got %d\n", err
);
1412 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1413 req
->r_request_started
, -1,
1414 &req
->r_caps_reservation
);
1416 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1425 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1429 put_page(ctl
->page
);
1434 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1435 struct ceph_readdir_cache_control
*ctl
,
1436 struct ceph_mds_request
*req
)
1438 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1439 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1440 unsigned idx
= ctl
->index
% nsize
;
1441 pgoff_t pgoff
= ctl
->index
/ nsize
;
1443 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1444 ceph_readdir_cache_release(ctl
);
1446 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1448 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1451 return idx
== 0 ? -ENOMEM
: 0;
1453 /* reading/filling the cache are serialized by
1454 * i_mutex, no need to use page lock */
1455 unlock_page(ctl
->page
);
1456 ctl
->dentries
= kmap(ctl
->page
);
1458 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1461 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1462 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1463 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1464 ctl
->dentries
[idx
] = dn
;
1467 dout("disable readdir cache\n");
1473 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1474 struct ceph_mds_session
*session
)
1476 struct dentry
*parent
= req
->r_dentry
;
1477 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1478 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1482 int err
= 0, skipped
= 0, ret
, i
;
1483 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1484 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1487 struct ceph_readdir_cache_control cache_ctl
= {};
1489 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1490 return readdir_prepopulate_inodes_only(req
, session
);
1492 if (rinfo
->hash_order
) {
1494 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1496 strlen(req
->r_path2
));
1497 last_hash
= ceph_frag_value(last_hash
);
1498 } else if (rinfo
->offset_hash
) {
1499 /* mds understands offset_hash */
1500 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1501 last_hash
= le32_to_cpu(rhead
->args
.readdir
.offset_hash
);
1505 if (rinfo
->dir_dir
&&
1506 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1507 dout("readdir_prepopulate got new frag %x -> %x\n",
1508 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1509 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1510 if (!rinfo
->hash_order
)
1511 req
->r_readdir_offset
= 2;
1514 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1515 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1516 rinfo
->dir_nr
, parent
);
1518 dout("readdir_prepopulate %d items under dn %p\n",
1519 rinfo
->dir_nr
, parent
);
1521 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1523 if (ceph_frag_is_leftmost(frag
) &&
1524 req
->r_readdir_offset
== 2 &&
1525 !(rinfo
->hash_order
&& last_hash
)) {
1526 /* note dir version at start of readdir so we can
1527 * tell if any dentries get dropped */
1528 req
->r_dir_release_cnt
=
1529 atomic64_read(&ci
->i_release_count
);
1530 req
->r_dir_ordered_cnt
=
1531 atomic64_read(&ci
->i_ordered_count
);
1532 req
->r_readdir_cache_idx
= 0;
1536 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1537 fpos_offset
= req
->r_readdir_offset
;
1539 /* FIXME: release caps/leases if error occurs */
1540 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1541 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1542 struct ceph_vino tvino
, dvino
;
1544 dname
.name
= rde
->name
;
1545 dname
.len
= rde
->name_len
;
1546 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1548 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1549 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1551 if (rinfo
->hash_order
) {
1552 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1553 rde
->name
, rde
->name_len
);
1554 hash
= ceph_frag_value(hash
);
1555 if (hash
!= last_hash
)
1558 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1560 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1564 dn
= d_lookup(parent
, &dname
);
1565 dout("d_lookup on parent=%p name=%.*s got %p\n",
1566 parent
, dname
.len
, dname
.name
, dn
);
1569 dn
= d_alloc(parent
, &dname
);
1570 dout("d_alloc %p '%.*s' = %p\n", parent
,
1571 dname
.len
, dname
.name
, dn
);
1573 dout("d_alloc badness\n");
1577 } else if (d_really_is_positive(dn
) &&
1578 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1579 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1580 dout(" dn %p points to wrong inode %p\n",
1582 __ceph_dir_clear_ordered(ci
);
1589 if (d_really_is_positive(dn
)) {
1592 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1594 dout("new_inode badness\n");
1602 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1603 req
->r_request_started
, -1,
1604 &req
->r_caps_reservation
);
1606 pr_err("fill_inode badness on %p\n", in
);
1607 if (d_really_is_positive(dn
))
1608 __ceph_dir_clear_ordered(ci
);
1616 if (d_really_is_negative(dn
)) {
1617 struct dentry
*realdn
;
1619 if (ceph_security_xattr_deadlock(in
)) {
1620 dout(" skip splicing dn %p to inode %p"
1621 " (security xattr deadlock)\n", dn
, in
);
1627 realdn
= splice_dentry(dn
, in
);
1628 if (IS_ERR(realdn
)) {
1629 err
= PTR_ERR(realdn
);
1637 ceph_dentry(dn
)->offset
= rde
->offset
;
1639 dvino
= ceph_vino(d_inode(parent
));
1640 update_dentry_lease(dn
, rde
->lease
, req
->r_session
,
1641 req
->r_request_started
, &tvino
, &dvino
);
1643 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1644 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1654 if (err
== 0 && skipped
== 0) {
1655 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1656 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1658 ceph_readdir_cache_release(&cache_ctl
);
1659 dout("readdir_prepopulate done\n");
1663 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1665 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1668 spin_lock(&ci
->i_ceph_lock
);
1669 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1670 i_size_write(inode
, size
);
1671 inode
->i_blocks
= calc_inode_blocks(size
);
1673 ret
= __ceph_should_report_size(ci
);
1675 spin_unlock(&ci
->i_ceph_lock
);
1680 * Write back inode data in a worker thread. (This can't be done
1681 * in the message handler context.)
1683 void ceph_queue_writeback(struct inode
*inode
)
1686 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1687 &ceph_inode(inode
)->i_wb_work
)) {
1688 dout("ceph_queue_writeback %p\n", inode
);
1690 dout("ceph_queue_writeback %p failed\n", inode
);
1695 static void ceph_writeback_work(struct work_struct
*work
)
1697 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1699 struct inode
*inode
= &ci
->vfs_inode
;
1701 dout("writeback %p\n", inode
);
1702 filemap_fdatawrite(&inode
->i_data
);
1707 * queue an async invalidation
1709 void ceph_queue_invalidate(struct inode
*inode
)
1712 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1713 &ceph_inode(inode
)->i_pg_inv_work
)) {
1714 dout("ceph_queue_invalidate %p\n", inode
);
1716 dout("ceph_queue_invalidate %p failed\n", inode
);
1722 * Invalidate inode pages in a worker thread. (This can't be done
1723 * in the message handler context.)
1725 static void ceph_invalidate_work(struct work_struct
*work
)
1727 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1729 struct inode
*inode
= &ci
->vfs_inode
;
1730 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1734 mutex_lock(&ci
->i_truncate_mutex
);
1736 if (READ_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1737 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1738 inode
, ceph_ino(inode
));
1739 mapping_set_error(inode
->i_mapping
, -EIO
);
1740 truncate_pagecache(inode
, 0);
1741 mutex_unlock(&ci
->i_truncate_mutex
);
1745 spin_lock(&ci
->i_ceph_lock
);
1746 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1747 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1748 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1749 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1751 spin_unlock(&ci
->i_ceph_lock
);
1752 mutex_unlock(&ci
->i_truncate_mutex
);
1755 orig_gen
= ci
->i_rdcache_gen
;
1756 spin_unlock(&ci
->i_ceph_lock
);
1758 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1759 pr_err("invalidate_pages %p fails\n", inode
);
1762 spin_lock(&ci
->i_ceph_lock
);
1763 if (orig_gen
== ci
->i_rdcache_gen
&&
1764 orig_gen
== ci
->i_rdcache_revoking
) {
1765 dout("invalidate_pages %p gen %d successful\n", inode
,
1767 ci
->i_rdcache_revoking
--;
1770 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1771 inode
, orig_gen
, ci
->i_rdcache_gen
,
1772 ci
->i_rdcache_revoking
);
1773 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1776 spin_unlock(&ci
->i_ceph_lock
);
1777 mutex_unlock(&ci
->i_truncate_mutex
);
1780 ceph_check_caps(ci
, 0, NULL
);
1786 * called by trunc_wq;
1788 * We also truncate in a separate thread as well.
1790 static void ceph_vmtruncate_work(struct work_struct
*work
)
1792 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1794 struct inode
*inode
= &ci
->vfs_inode
;
1796 dout("vmtruncate_work %p\n", inode
);
1797 __ceph_do_pending_vmtruncate(inode
);
1802 * Queue an async vmtruncate. If we fail to queue work, we will handle
1803 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1805 void ceph_queue_vmtruncate(struct inode
*inode
)
1807 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1811 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1812 &ci
->i_vmtruncate_work
)) {
1813 dout("ceph_queue_vmtruncate %p\n", inode
);
1815 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1816 inode
, ci
->i_truncate_pending
);
1822 * Make sure any pending truncation is applied before doing anything
1823 * that may depend on it.
1825 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1827 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1829 int wrbuffer_refs
, finish
= 0;
1831 mutex_lock(&ci
->i_truncate_mutex
);
1833 spin_lock(&ci
->i_ceph_lock
);
1834 if (ci
->i_truncate_pending
== 0) {
1835 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1836 spin_unlock(&ci
->i_ceph_lock
);
1837 mutex_unlock(&ci
->i_truncate_mutex
);
1842 * make sure any dirty snapped pages are flushed before we
1843 * possibly truncate them.. so write AND block!
1845 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1846 struct ceph_cap_snap
*capsnap
;
1847 to
= ci
->i_truncate_size
;
1848 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1849 // MDS should have revoked Frw caps
1850 WARN_ON_ONCE(capsnap
->writing
);
1851 if (capsnap
->dirty_pages
&& capsnap
->size
> to
)
1854 spin_unlock(&ci
->i_ceph_lock
);
1855 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1858 truncate_pagecache(inode
, to
);
1860 filemap_write_and_wait_range(&inode
->i_data
, 0,
1861 inode
->i_sb
->s_maxbytes
);
1865 /* there should be no reader or writer */
1866 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1868 to
= ci
->i_truncate_size
;
1869 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1870 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1871 ci
->i_truncate_pending
, to
);
1872 spin_unlock(&ci
->i_ceph_lock
);
1874 truncate_pagecache(inode
, to
);
1876 spin_lock(&ci
->i_ceph_lock
);
1877 if (to
== ci
->i_truncate_size
) {
1878 ci
->i_truncate_pending
= 0;
1881 spin_unlock(&ci
->i_ceph_lock
);
1885 mutex_unlock(&ci
->i_truncate_mutex
);
1887 if (wrbuffer_refs
== 0)
1888 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1890 wake_up_all(&ci
->i_cap_wq
);
1896 static const struct inode_operations ceph_symlink_iops
= {
1897 .get_link
= simple_get_link
,
1898 .setattr
= ceph_setattr
,
1899 .getattr
= ceph_getattr
,
1900 .listxattr
= ceph_listxattr
,
1903 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
1905 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1906 const unsigned int ia_valid
= attr
->ia_valid
;
1907 struct ceph_mds_request
*req
;
1908 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1909 struct ceph_cap_flush
*prealloc_cf
;
1911 int release
= 0, dirtied
= 0;
1914 int inode_dirty_flags
= 0;
1915 bool lock_snap_rwsem
= false;
1917 prealloc_cf
= ceph_alloc_cap_flush();
1921 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1924 ceph_free_cap_flush(prealloc_cf
);
1925 return PTR_ERR(req
);
1928 spin_lock(&ci
->i_ceph_lock
);
1929 issued
= __ceph_caps_issued(ci
, NULL
);
1931 if (!ci
->i_head_snapc
&&
1932 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1933 lock_snap_rwsem
= true;
1934 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1935 spin_unlock(&ci
->i_ceph_lock
);
1936 down_read(&mdsc
->snap_rwsem
);
1937 spin_lock(&ci
->i_ceph_lock
);
1938 issued
= __ceph_caps_issued(ci
, NULL
);
1942 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1944 if (ia_valid
& ATTR_UID
) {
1945 dout("setattr %p uid %d -> %d\n", inode
,
1946 from_kuid(&init_user_ns
, inode
->i_uid
),
1947 from_kuid(&init_user_ns
, attr
->ia_uid
));
1948 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1949 inode
->i_uid
= attr
->ia_uid
;
1950 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1951 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1952 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1953 req
->r_args
.setattr
.uid
= cpu_to_le32(
1954 from_kuid(&init_user_ns
, attr
->ia_uid
));
1955 mask
|= CEPH_SETATTR_UID
;
1956 release
|= CEPH_CAP_AUTH_SHARED
;
1959 if (ia_valid
& ATTR_GID
) {
1960 dout("setattr %p gid %d -> %d\n", inode
,
1961 from_kgid(&init_user_ns
, inode
->i_gid
),
1962 from_kgid(&init_user_ns
, attr
->ia_gid
));
1963 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1964 inode
->i_gid
= attr
->ia_gid
;
1965 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1966 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1967 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1968 req
->r_args
.setattr
.gid
= cpu_to_le32(
1969 from_kgid(&init_user_ns
, attr
->ia_gid
));
1970 mask
|= CEPH_SETATTR_GID
;
1971 release
|= CEPH_CAP_AUTH_SHARED
;
1974 if (ia_valid
& ATTR_MODE
) {
1975 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1977 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1978 inode
->i_mode
= attr
->ia_mode
;
1979 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1980 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1981 attr
->ia_mode
!= inode
->i_mode
) {
1982 inode
->i_mode
= attr
->ia_mode
;
1983 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1984 mask
|= CEPH_SETATTR_MODE
;
1985 release
|= CEPH_CAP_AUTH_SHARED
;
1989 if (ia_valid
& ATTR_ATIME
) {
1990 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1991 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1992 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1993 if (issued
& CEPH_CAP_FILE_EXCL
) {
1994 ci
->i_time_warp_seq
++;
1995 inode
->i_atime
= attr
->ia_atime
;
1996 dirtied
|= CEPH_CAP_FILE_EXCL
;
1997 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1998 timespec_compare(&inode
->i_atime
,
1999 &attr
->ia_atime
) < 0) {
2000 inode
->i_atime
= attr
->ia_atime
;
2001 dirtied
|= CEPH_CAP_FILE_WR
;
2002 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2003 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2004 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
2006 mask
|= CEPH_SETATTR_ATIME
;
2007 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
2011 if (ia_valid
& ATTR_MTIME
) {
2012 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
2013 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2014 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2015 if (issued
& CEPH_CAP_FILE_EXCL
) {
2016 ci
->i_time_warp_seq
++;
2017 inode
->i_mtime
= attr
->ia_mtime
;
2018 dirtied
|= CEPH_CAP_FILE_EXCL
;
2019 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2020 timespec_compare(&inode
->i_mtime
,
2021 &attr
->ia_mtime
) < 0) {
2022 inode
->i_mtime
= attr
->ia_mtime
;
2023 dirtied
|= CEPH_CAP_FILE_WR
;
2024 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2025 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2026 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
2028 mask
|= CEPH_SETATTR_MTIME
;
2029 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2033 if (ia_valid
& ATTR_SIZE
) {
2034 dout("setattr %p size %lld -> %lld\n", inode
,
2035 inode
->i_size
, attr
->ia_size
);
2036 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
2037 attr
->ia_size
> inode
->i_size
) {
2038 i_size_write(inode
, attr
->ia_size
);
2039 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2040 ci
->i_reported_size
= attr
->ia_size
;
2041 dirtied
|= CEPH_CAP_FILE_EXCL
;
2042 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2043 attr
->ia_size
!= inode
->i_size
) {
2044 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2045 req
->r_args
.setattr
.old_size
=
2046 cpu_to_le64(inode
->i_size
);
2047 mask
|= CEPH_SETATTR_SIZE
;
2048 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
2053 /* these do nothing */
2054 if (ia_valid
& ATTR_CTIME
) {
2055 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2056 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2057 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
2058 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2059 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2060 only
? "ctime only" : "ignored");
2063 * if kernel wants to dirty ctime but nothing else,
2064 * we need to choose a cap to dirty under, or do
2065 * a almost-no-op setattr
2067 if (issued
& CEPH_CAP_AUTH_EXCL
)
2068 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2069 else if (issued
& CEPH_CAP_FILE_EXCL
)
2070 dirtied
|= CEPH_CAP_FILE_EXCL
;
2071 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2072 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2074 mask
|= CEPH_SETATTR_CTIME
;
2077 if (ia_valid
& ATTR_FILE
)
2078 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2081 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2083 inode
->i_ctime
= attr
->ia_ctime
;
2087 spin_unlock(&ci
->i_ceph_lock
);
2088 if (lock_snap_rwsem
)
2089 up_read(&mdsc
->snap_rwsem
);
2091 if (inode_dirty_flags
)
2092 __mark_inode_dirty(inode
, inode_dirty_flags
);
2096 req
->r_inode
= inode
;
2098 req
->r_inode_drop
= release
;
2099 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2100 req
->r_num_caps
= 1;
2101 req
->r_stamp
= attr
->ia_ctime
;
2102 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2104 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2105 ceph_cap_string(dirtied
), mask
);
2107 ceph_mdsc_put_request(req
);
2108 ceph_free_cap_flush(prealloc_cf
);
2110 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2111 __ceph_do_pending_vmtruncate(inode
);
2119 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2121 struct inode
*inode
= d_inode(dentry
);
2124 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2127 err
= setattr_prepare(dentry
, attr
);
2131 err
= __ceph_setattr(inode
, attr
);
2133 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2134 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2140 * Verify that we have a lease on the given mask. If not,
2141 * do a getattr against an mds.
2143 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2144 int mask
, bool force
)
2146 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2147 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2148 struct ceph_mds_request
*req
;
2151 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2152 dout("do_getattr inode %p SNAPDIR\n", inode
);
2156 dout("do_getattr inode %p mask %s mode 0%o\n",
2157 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2158 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2161 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
2163 return PTR_ERR(req
);
2164 req
->r_inode
= inode
;
2166 req
->r_num_caps
= 1;
2167 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2168 req
->r_locked_page
= locked_page
;
2169 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2170 if (locked_page
&& err
== 0) {
2171 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2172 if (inline_version
== 0) {
2173 /* the reply is supposed to contain inline data */
2175 } else if (inline_version
== CEPH_INLINE_NONE
) {
2178 err
= req
->r_reply_info
.targeti
.inline_len
;
2181 ceph_mdsc_put_request(req
);
2182 dout("do_getattr result=%d\n", err
);
2188 * Check inode permissions. We verify we have a valid value for
2189 * the AUTH cap, then call the generic handler.
2191 int ceph_permission(struct inode
*inode
, int mask
)
2195 if (mask
& MAY_NOT_BLOCK
)
2198 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2201 err
= generic_permission(inode
, mask
);
2206 * Get all attributes. Hopefully somedata we'll have a statlite()
2207 * and can limit the fields we require to be accurate.
2209 int ceph_getattr(const struct path
*path
, struct kstat
*stat
,
2210 u32 request_mask
, unsigned int flags
)
2212 struct inode
*inode
= d_inode(path
->dentry
);
2213 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2216 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2218 generic_fillattr(inode
, stat
);
2219 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2220 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2221 stat
->dev
= ceph_snap(inode
);
2224 if (S_ISDIR(inode
->i_mode
)) {
2225 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2227 stat
->size
= ci
->i_rbytes
;
2229 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2231 stat
->blksize
= 65536;