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>
16 #include <linux/iversion.h>
19 #include "mds_client.h"
21 #include <linux/ceph/decode.h>
24 * Ceph inode operations
26 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
27 * setattr, etc.), xattr helpers, and helpers for assimilating
28 * metadata returned by the MDS into our cache.
30 * Also define helpers for doing asynchronous writeback, invalidation,
31 * and truncation for the benefit of those who can't afford to block
32 * (typically because they are in the message handler path).
35 static const struct inode_operations ceph_symlink_iops
;
37 static void ceph_inode_work(struct work_struct
*work
);
40 * find or create an inode, given the ceph ino number
42 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
44 struct ceph_inode_info
*ci
= ceph_inode(inode
);
45 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(inode
->i_sb
);
47 ci
->i_vino
= *(struct ceph_vino
*)data
;
48 inode
->i_ino
= ceph_vino_to_ino_t(ci
->i_vino
);
49 inode_set_iversion_raw(inode
, 0);
50 percpu_counter_inc(&mdsc
->metric
.total_inodes
);
55 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
59 if (ceph_vino_is_reserved(vino
))
60 return ERR_PTR(-EREMOTEIO
);
62 inode
= iget5_locked(sb
, (unsigned long)vino
.ino
, ceph_ino_compare
,
63 ceph_set_ino_cb
, &vino
);
65 return ERR_PTR(-ENOMEM
);
67 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode
),
68 ceph_vinop(inode
), inode
, !!(inode
->i_state
& I_NEW
));
73 * get/constuct snapdir inode for a given directory
75 struct inode
*ceph_get_snapdir(struct inode
*parent
)
77 struct ceph_vino vino
= {
78 .ino
= ceph_ino(parent
),
81 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
82 struct ceph_inode_info
*ci
= ceph_inode(inode
);
87 if (!S_ISDIR(parent
->i_mode
)) {
88 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
93 if (!(inode
->i_state
& I_NEW
) && !S_ISDIR(inode
->i_mode
)) {
94 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
99 inode
->i_mode
= parent
->i_mode
;
100 inode
->i_uid
= parent
->i_uid
;
101 inode
->i_gid
= parent
->i_gid
;
102 inode
->i_mtime
= parent
->i_mtime
;
103 inode
->i_ctime
= parent
->i_ctime
;
104 inode
->i_atime
= parent
->i_atime
;
106 ci
->i_btime
= ceph_inode(parent
)->i_btime
;
108 if (inode
->i_state
& I_NEW
) {
109 inode
->i_op
= &ceph_snapdir_iops
;
110 inode
->i_fop
= &ceph_snapdir_fops
;
111 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
112 unlock_new_inode(inode
);
117 if ((inode
->i_state
& I_NEW
))
118 discard_new_inode(inode
);
121 return ERR_PTR(-ENOTDIR
);
124 const struct inode_operations ceph_file_iops
= {
125 .permission
= ceph_permission
,
126 .setattr
= ceph_setattr
,
127 .getattr
= ceph_getattr
,
128 .listxattr
= ceph_listxattr
,
129 .get_inode_acl
= ceph_get_acl
,
130 .set_acl
= ceph_set_acl
,
135 * We use a 'frag tree' to keep track of the MDS's directory fragments
136 * for a given inode (usually there is just a single fragment). We
137 * need to know when a child frag is delegated to a new MDS, or when
138 * it is flagged as replicated, so we can direct our requests
143 * find/create a frag in the tree
145 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
149 struct rb_node
*parent
= NULL
;
150 struct ceph_inode_frag
*frag
;
153 p
= &ci
->i_fragtree
.rb_node
;
156 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
157 c
= ceph_frag_compare(f
, frag
->frag
);
166 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
168 return ERR_PTR(-ENOMEM
);
175 rb_link_node(&frag
->node
, parent
, p
);
176 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
178 dout("get_or_create_frag added %llx.%llx frag %x\n",
179 ceph_vinop(&ci
->netfs
.inode
), f
);
184 * find a specific frag @f
186 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
188 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
191 struct ceph_inode_frag
*frag
=
192 rb_entry(n
, struct ceph_inode_frag
, node
);
193 int c
= ceph_frag_compare(f
, frag
->frag
);
205 * Choose frag containing the given value @v. If @pfrag is
206 * specified, copy the frag delegation info to the caller if
209 static u32
__ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
210 struct ceph_inode_frag
*pfrag
, int *found
)
212 u32 t
= ceph_frag_make(0, 0);
213 struct ceph_inode_frag
*frag
;
221 WARN_ON(!ceph_frag_contains_value(t
, v
));
222 frag
= __ceph_find_frag(ci
, t
);
224 break; /* t is a leaf */
225 if (frag
->split_by
== 0) {
227 memcpy(pfrag
, frag
, sizeof(*pfrag
));
234 nway
= 1 << frag
->split_by
;
235 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
236 frag
->split_by
, nway
);
237 for (i
= 0; i
< nway
; i
++) {
238 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
239 if (ceph_frag_contains_value(n
, v
)) {
246 dout("choose_frag(%x) = %x\n", v
, t
);
251 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
252 struct ceph_inode_frag
*pfrag
, int *found
)
255 mutex_lock(&ci
->i_fragtree_mutex
);
256 ret
= __ceph_choose_frag(ci
, v
, pfrag
, found
);
257 mutex_unlock(&ci
->i_fragtree_mutex
);
262 * Process dirfrag (delegation) info from the mds. Include leaf
263 * fragment in tree ONLY if ndist > 0. Otherwise, only
264 * branches/splits are included in i_fragtree)
266 static int ceph_fill_dirfrag(struct inode
*inode
,
267 struct ceph_mds_reply_dirfrag
*dirinfo
)
269 struct ceph_inode_info
*ci
= ceph_inode(inode
);
270 struct ceph_inode_frag
*frag
;
271 u32 id
= le32_to_cpu(dirinfo
->frag
);
272 int mds
= le32_to_cpu(dirinfo
->auth
);
273 int ndist
= le32_to_cpu(dirinfo
->ndist
);
278 spin_lock(&ci
->i_ceph_lock
);
280 diri_auth
= ci
->i_auth_cap
->mds
;
281 spin_unlock(&ci
->i_ceph_lock
);
283 if (mds
== -1) /* CDIR_AUTH_PARENT */
286 mutex_lock(&ci
->i_fragtree_mutex
);
287 if (ndist
== 0 && mds
== diri_auth
) {
288 /* no delegation info needed. */
289 frag
= __ceph_find_frag(ci
, id
);
292 if (frag
->split_by
== 0) {
293 /* tree leaf, remove */
294 dout("fill_dirfrag removed %llx.%llx frag %x"
295 " (no ref)\n", ceph_vinop(inode
), id
);
296 rb_erase(&frag
->node
, &ci
->i_fragtree
);
299 /* tree branch, keep and clear */
300 dout("fill_dirfrag cleared %llx.%llx frag %x"
301 " referral\n", ceph_vinop(inode
), id
);
309 /* find/add this frag to store mds delegation info */
310 frag
= __get_or_create_frag(ci
, id
);
312 /* this is not the end of the world; we can continue
313 with bad/inaccurate delegation info */
314 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
315 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
321 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
322 for (i
= 0; i
< frag
->ndist
; i
++)
323 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
324 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
325 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
328 mutex_unlock(&ci
->i_fragtree_mutex
);
332 static int frag_tree_split_cmp(const void *l
, const void *r
)
334 struct ceph_frag_tree_split
*ls
= (struct ceph_frag_tree_split
*)l
;
335 struct ceph_frag_tree_split
*rs
= (struct ceph_frag_tree_split
*)r
;
336 return ceph_frag_compare(le32_to_cpu(ls
->frag
),
337 le32_to_cpu(rs
->frag
));
340 static bool is_frag_child(u32 f
, struct ceph_inode_frag
*frag
)
343 return f
== ceph_frag_make(0, 0);
344 if (ceph_frag_bits(f
) != ceph_frag_bits(frag
->frag
) + frag
->split_by
)
346 return ceph_frag_contains_value(frag
->frag
, ceph_frag_value(f
));
349 static int ceph_fill_fragtree(struct inode
*inode
,
350 struct ceph_frag_tree_head
*fragtree
,
351 struct ceph_mds_reply_dirfrag
*dirinfo
)
353 struct ceph_inode_info
*ci
= ceph_inode(inode
);
354 struct ceph_inode_frag
*frag
, *prev_frag
= NULL
;
355 struct rb_node
*rb_node
;
356 unsigned i
, split_by
, nsplits
;
360 mutex_lock(&ci
->i_fragtree_mutex
);
361 nsplits
= le32_to_cpu(fragtree
->nsplits
);
362 if (nsplits
!= ci
->i_fragtree_nsplits
) {
364 } else if (nsplits
) {
365 i
= get_random_u32_below(nsplits
);
366 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
367 if (!__ceph_find_frag(ci
, id
))
369 } else if (!RB_EMPTY_ROOT(&ci
->i_fragtree
)) {
370 rb_node
= rb_first(&ci
->i_fragtree
);
371 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
372 if (frag
->frag
!= ceph_frag_make(0, 0) || rb_next(rb_node
))
375 if (!update
&& dirinfo
) {
376 id
= le32_to_cpu(dirinfo
->frag
);
377 if (id
!= __ceph_choose_frag(ci
, id
, NULL
, NULL
))
384 sort(fragtree
->splits
, nsplits
, sizeof(fragtree
->splits
[0]),
385 frag_tree_split_cmp
, NULL
);
388 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode
));
389 rb_node
= rb_first(&ci
->i_fragtree
);
390 for (i
= 0; i
< nsplits
; i
++) {
391 id
= le32_to_cpu(fragtree
->splits
[i
].frag
);
392 split_by
= le32_to_cpu(fragtree
->splits
[i
].by
);
393 if (split_by
== 0 || ceph_frag_bits(id
) + split_by
> 24) {
394 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
395 "frag %x split by %d\n", ceph_vinop(inode
),
396 i
, nsplits
, id
, split_by
);
401 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
402 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
403 if (frag
->frag
!= id
)
406 rb_node
= rb_next(rb_node
);
409 rb_node
= rb_next(rb_node
);
410 /* delete stale split/leaf node */
411 if (frag
->split_by
> 0 ||
412 !is_frag_child(frag
->frag
, prev_frag
)) {
413 rb_erase(&frag
->node
, &ci
->i_fragtree
);
414 if (frag
->split_by
> 0)
415 ci
->i_fragtree_nsplits
--;
421 frag
= __get_or_create_frag(ci
, id
);
425 if (frag
->split_by
== 0)
426 ci
->i_fragtree_nsplits
++;
427 frag
->split_by
= split_by
;
428 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
432 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
433 rb_node
= rb_next(rb_node
);
434 /* delete stale split/leaf node */
435 if (frag
->split_by
> 0 ||
436 !is_frag_child(frag
->frag
, prev_frag
)) {
437 rb_erase(&frag
->node
, &ci
->i_fragtree
);
438 if (frag
->split_by
> 0)
439 ci
->i_fragtree_nsplits
--;
444 mutex_unlock(&ci
->i_fragtree_mutex
);
449 * initialize a newly allocated inode.
451 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
453 struct ceph_inode_info
*ci
;
456 ci
= alloc_inode_sb(sb
, ceph_inode_cachep
, GFP_NOFS
);
460 dout("alloc_inode %p\n", &ci
->netfs
.inode
);
462 /* Set parameters for the netfs library */
463 netfs_inode_init(&ci
->netfs
, &ceph_netfs_ops
);
465 spin_lock_init(&ci
->i_ceph_lock
);
468 ci
->i_inline_version
= 0;
469 ci
->i_time_warp_seq
= 0;
470 ci
->i_ceph_flags
= 0;
471 atomic64_set(&ci
->i_ordered_count
, 1);
472 atomic64_set(&ci
->i_release_count
, 1);
473 atomic64_set(&ci
->i_complete_seq
[0], 0);
474 atomic64_set(&ci
->i_complete_seq
[1], 0);
475 ci
->i_symlink
= NULL
;
480 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
481 memset(&ci
->i_cached_layout
, 0, sizeof(ci
->i_cached_layout
));
482 RCU_INIT_POINTER(ci
->i_layout
.pool_ns
, NULL
);
484 ci
->i_fragtree
= RB_ROOT
;
485 mutex_init(&ci
->i_fragtree_mutex
);
487 ci
->i_xattrs
.blob
= NULL
;
488 ci
->i_xattrs
.prealloc_blob
= NULL
;
489 ci
->i_xattrs
.dirty
= false;
490 ci
->i_xattrs
.index
= RB_ROOT
;
491 ci
->i_xattrs
.count
= 0;
492 ci
->i_xattrs
.names_size
= 0;
493 ci
->i_xattrs
.vals_size
= 0;
494 ci
->i_xattrs
.version
= 0;
495 ci
->i_xattrs
.index_version
= 0;
497 ci
->i_caps
= RB_ROOT
;
498 ci
->i_auth_cap
= NULL
;
499 ci
->i_dirty_caps
= 0;
500 ci
->i_flushing_caps
= 0;
501 INIT_LIST_HEAD(&ci
->i_dirty_item
);
502 INIT_LIST_HEAD(&ci
->i_flushing_item
);
503 ci
->i_prealloc_cap_flush
= NULL
;
504 INIT_LIST_HEAD(&ci
->i_cap_flush_list
);
505 init_waitqueue_head(&ci
->i_cap_wq
);
506 ci
->i_hold_caps_max
= 0;
507 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
508 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
509 ci
->i_head_snapc
= NULL
;
512 ci
->i_last_rd
= ci
->i_last_wr
= jiffies
- 3600 * HZ
;
513 for (i
= 0; i
< CEPH_FILE_MODE_BITS
; i
++)
514 ci
->i_nr_by_mode
[i
] = 0;
516 mutex_init(&ci
->i_truncate_mutex
);
517 ci
->i_truncate_seq
= 0;
518 ci
->i_truncate_size
= 0;
519 ci
->i_truncate_pending
= 0;
522 ci
->i_reported_size
= 0;
523 ci
->i_wanted_max_size
= 0;
524 ci
->i_requested_max_size
= 0;
528 ci
->i_rdcache_ref
= 0;
532 ci
->i_wrbuffer_ref
= 0;
533 ci
->i_wrbuffer_ref_head
= 0;
534 atomic_set(&ci
->i_filelock_ref
, 0);
535 atomic_set(&ci
->i_shared_gen
, 1);
536 ci
->i_rdcache_gen
= 0;
537 ci
->i_rdcache_revoking
= 0;
539 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
540 INIT_LIST_HEAD(&ci
->i_unsafe_iops
);
541 spin_lock_init(&ci
->i_unsafe_lock
);
543 ci
->i_snap_realm
= NULL
;
544 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
545 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
547 INIT_WORK(&ci
->i_work
, ceph_inode_work
);
549 memset(&ci
->i_btime
, '\0', sizeof(ci
->i_btime
));
550 return &ci
->netfs
.inode
;
553 void ceph_free_inode(struct inode
*inode
)
555 struct ceph_inode_info
*ci
= ceph_inode(inode
);
557 kfree(ci
->i_symlink
);
558 kmem_cache_free(ceph_inode_cachep
, ci
);
561 void ceph_evict_inode(struct inode
*inode
)
563 struct ceph_inode_info
*ci
= ceph_inode(inode
);
564 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(inode
->i_sb
);
565 struct ceph_inode_frag
*frag
;
568 dout("evict_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
570 percpu_counter_dec(&mdsc
->metric
.total_inodes
);
572 truncate_inode_pages_final(&inode
->i_data
);
573 if (inode
->i_state
& I_PINNING_FSCACHE_WB
)
574 ceph_fscache_unuse_cookie(inode
, true);
577 ceph_fscache_unregister_inode_cookie(ci
);
579 __ceph_remove_caps(ci
);
581 if (__ceph_has_quota(ci
, QUOTA_GET_ANY
))
582 ceph_adjust_quota_realms_count(inode
, false);
585 * we may still have a snap_realm reference if there are stray
586 * caps in i_snap_caps.
588 if (ci
->i_snap_realm
) {
589 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
590 dout(" dropping residual ref to snap realm %p\n",
592 ceph_change_snap_realm(inode
, NULL
);
594 ceph_put_snapid_map(mdsc
, ci
->i_snapid_map
);
595 ci
->i_snap_realm
= NULL
;
599 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
600 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
601 rb_erase(n
, &ci
->i_fragtree
);
604 ci
->i_fragtree_nsplits
= 0;
606 __ceph_destroy_xattrs(ci
);
607 if (ci
->i_xattrs
.blob
)
608 ceph_buffer_put(ci
->i_xattrs
.blob
);
609 if (ci
->i_xattrs
.prealloc_blob
)
610 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
612 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
613 ceph_put_string(rcu_dereference_raw(ci
->i_cached_layout
.pool_ns
));
616 static inline blkcnt_t
calc_inode_blocks(u64 size
)
618 return (size
+ (1<<9) - 1) >> 9;
622 * Helpers to fill in size, ctime, mtime, and atime. We have to be
623 * careful because either the client or MDS may have more up to date
624 * info, depending on which capabilities are held, and whether
625 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
626 * and size are monotonically increasing, except when utimes() or
627 * truncate() increments the corresponding _seq values.)
629 int ceph_fill_file_size(struct inode
*inode
, int issued
,
630 u32 truncate_seq
, u64 truncate_size
, u64 size
)
632 struct ceph_inode_info
*ci
= ceph_inode(inode
);
634 loff_t isize
= i_size_read(inode
);
636 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
637 (truncate_seq
== ci
->i_truncate_seq
&& size
> isize
)) {
638 dout("size %lld -> %llu\n", isize
, size
);
639 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
640 pr_err("fill_file_size non-zero size for directory\n");
643 i_size_write(inode
, size
);
644 inode
->i_blocks
= calc_inode_blocks(size
);
646 * If we're expanding, then we should be able to just update
647 * the existing cookie.
650 ceph_fscache_update(inode
);
651 ci
->i_reported_size
= size
;
652 if (truncate_seq
!= ci
->i_truncate_seq
) {
653 dout("truncate_seq %u -> %u\n",
654 ci
->i_truncate_seq
, truncate_seq
);
655 ci
->i_truncate_seq
= truncate_seq
;
657 /* the MDS should have revoked these caps */
658 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
661 CEPH_CAP_FILE_LAZYIO
));
663 * If we hold relevant caps, or in the case where we're
664 * not the only client referencing this file and we
665 * don't hold those caps, then we need to check whether
666 * the file is either opened or mmaped
668 if ((issued
& (CEPH_CAP_FILE_CACHE
|
669 CEPH_CAP_FILE_BUFFER
)) ||
670 mapping_mapped(inode
->i_mapping
) ||
671 __ceph_is_file_opened(ci
)) {
672 ci
->i_truncate_pending
++;
677 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
678 ci
->i_truncate_size
!= truncate_size
) {
679 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
681 ci
->i_truncate_size
= truncate_size
;
686 void ceph_fill_file_time(struct inode
*inode
, int issued
,
687 u64 time_warp_seq
, struct timespec64
*ctime
,
688 struct timespec64
*mtime
, struct timespec64
*atime
)
690 struct ceph_inode_info
*ci
= ceph_inode(inode
);
693 if (issued
& (CEPH_CAP_FILE_EXCL
|
695 CEPH_CAP_FILE_BUFFER
|
697 CEPH_CAP_XATTR_EXCL
)) {
698 if (ci
->i_version
== 0 ||
699 timespec64_compare(ctime
, &inode
->i_ctime
) > 0) {
700 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
701 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
702 ctime
->tv_sec
, ctime
->tv_nsec
);
703 inode
->i_ctime
= *ctime
;
705 if (ci
->i_version
== 0 ||
706 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
707 /* the MDS did a utimes() */
708 dout("mtime %lld.%09ld -> %lld.%09ld "
710 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
711 mtime
->tv_sec
, mtime
->tv_nsec
,
712 ci
->i_time_warp_seq
, (int)time_warp_seq
);
714 inode
->i_mtime
= *mtime
;
715 inode
->i_atime
= *atime
;
716 ci
->i_time_warp_seq
= time_warp_seq
;
717 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
718 /* nobody did utimes(); take the max */
719 if (timespec64_compare(mtime
, &inode
->i_mtime
) > 0) {
720 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
721 inode
->i_mtime
.tv_sec
,
722 inode
->i_mtime
.tv_nsec
,
723 mtime
->tv_sec
, mtime
->tv_nsec
);
724 inode
->i_mtime
= *mtime
;
726 if (timespec64_compare(atime
, &inode
->i_atime
) > 0) {
727 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
728 inode
->i_atime
.tv_sec
,
729 inode
->i_atime
.tv_nsec
,
730 atime
->tv_sec
, atime
->tv_nsec
);
731 inode
->i_atime
= *atime
;
733 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
734 /* we did a utimes(); ignore mds values */
739 /* we have no write|excl caps; whatever the MDS says is true */
740 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
741 inode
->i_ctime
= *ctime
;
742 inode
->i_mtime
= *mtime
;
743 inode
->i_atime
= *atime
;
744 ci
->i_time_warp_seq
= time_warp_seq
;
749 if (warn
) /* time_warp_seq shouldn't go backwards */
750 dout("%p mds time_warp_seq %llu < %u\n",
751 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
755 * Populate an inode based on info from mds. May be called on new or
758 int ceph_fill_inode(struct inode
*inode
, struct page
*locked_page
,
759 struct ceph_mds_reply_info_in
*iinfo
,
760 struct ceph_mds_reply_dirfrag
*dirinfo
,
761 struct ceph_mds_session
*session
, int cap_fmode
,
762 struct ceph_cap_reservation
*caps_reservation
)
764 struct ceph_mds_client
*mdsc
= ceph_sb_to_mdsc(inode
->i_sb
);
765 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
766 struct ceph_inode_info
*ci
= ceph_inode(inode
);
767 int issued
, new_issued
, info_caps
;
768 struct timespec64 mtime
, atime
, ctime
;
769 struct ceph_buffer
*xattr_blob
= NULL
;
770 struct ceph_buffer
*old_blob
= NULL
;
771 struct ceph_string
*pool_ns
= NULL
;
772 struct ceph_cap
*new_cap
= NULL
;
775 bool queue_trunc
= false;
776 bool new_version
= false;
777 bool fill_inline
= false;
778 umode_t mode
= le32_to_cpu(info
->mode
);
779 dev_t rdev
= le32_to_cpu(info
->rdev
);
781 lockdep_assert_held(&mdsc
->snap_rwsem
);
783 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__
,
784 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
787 /* Once I_NEW is cleared, we can't change type or dev numbers */
788 if (inode
->i_state
& I_NEW
) {
789 inode
->i_mode
= mode
;
791 if (inode_wrong_type(inode
, mode
)) {
792 pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
793 ceph_vinop(inode
), inode
->i_mode
, mode
);
797 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && inode
->i_rdev
!= rdev
) {
798 pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
799 ceph_vinop(inode
), MAJOR(inode
->i_rdev
),
800 MINOR(inode
->i_rdev
), MAJOR(rdev
),
806 info_caps
= le32_to_cpu(info
->cap
.caps
);
808 /* prealloc new cap struct */
809 if (info_caps
&& ceph_snap(inode
) == CEPH_NOSNAP
) {
810 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
816 * prealloc xattr data, if it looks like we'll need it. only
817 * if len > 4 (meaning there are actually xattrs; the first 4
818 * bytes are the xattr count).
820 if (iinfo
->xattr_len
> 4) {
821 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
823 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__
,
827 if (iinfo
->pool_ns_len
> 0)
828 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
831 if (ceph_snap(inode
) != CEPH_NOSNAP
&& !ci
->i_snapid_map
)
832 ci
->i_snapid_map
= ceph_get_snapid_map(mdsc
, ceph_snap(inode
));
834 spin_lock(&ci
->i_ceph_lock
);
837 * provided version will be odd if inode value is projected,
838 * even if stable. skip the update if we have newer stable
839 * info (ours>=theirs, e.g. due to racing mds replies), unless
840 * we are getting projected (unstable) info (in which case the
841 * version is odd, and we want ours>theirs).
847 if (ci
->i_version
== 0 ||
848 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
849 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
852 /* Update change_attribute */
853 inode_set_max_iversion_raw(inode
, iinfo
->change_attr
);
855 __ceph_caps_issued(ci
, &issued
);
856 issued
|= __ceph_caps_dirty(ci
);
857 new_issued
= ~issued
& info_caps
;
859 /* directories have fl_stripe_unit set to zero */
860 if (le32_to_cpu(info
->layout
.fl_stripe_unit
))
862 fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
864 inode
->i_blkbits
= CEPH_BLOCK_SHIFT
;
866 __ceph_update_quota(ci
, iinfo
->max_bytes
, iinfo
->max_files
);
868 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
869 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
870 inode
->i_mode
= mode
;
871 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
872 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
873 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
874 from_kuid(&init_user_ns
, inode
->i_uid
),
875 from_kgid(&init_user_ns
, inode
->i_gid
));
876 ceph_decode_timespec64(&ci
->i_btime
, &iinfo
->btime
);
877 ceph_decode_timespec64(&ci
->i_snap_btime
, &iinfo
->snap_btime
);
880 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
881 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
882 set_nlink(inode
, le32_to_cpu(info
->nlink
));
884 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
885 /* be careful with mtime, atime, size */
886 ceph_decode_timespec64(&atime
, &info
->atime
);
887 ceph_decode_timespec64(&mtime
, &info
->mtime
);
888 ceph_decode_timespec64(&ctime
, &info
->ctime
);
889 ceph_fill_file_time(inode
, issued
,
890 le32_to_cpu(info
->time_warp_seq
),
891 &ctime
, &mtime
, &atime
);
894 if (new_version
|| (info_caps
& CEPH_CAP_FILE_SHARED
)) {
895 ci
->i_files
= le64_to_cpu(info
->files
);
896 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
900 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
901 s64 old_pool
= ci
->i_layout
.pool_id
;
902 struct ceph_string
*old_ns
;
904 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
905 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
906 lockdep_is_held(&ci
->i_ceph_lock
));
907 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
909 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
910 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
914 queue_trunc
= ceph_fill_file_size(inode
, issued
,
915 le32_to_cpu(info
->truncate_seq
),
916 le64_to_cpu(info
->truncate_size
),
917 le64_to_cpu(info
->size
));
918 /* only update max_size on auth cap */
919 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
920 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
921 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
922 le64_to_cpu(info
->max_size
));
923 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
927 /* layout and rstat are not tracked by capability, update them if
928 * the inode info is from auth mds */
929 if (new_version
|| (info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
)) {
930 if (S_ISDIR(inode
->i_mode
)) {
931 ci
->i_dir_layout
= iinfo
->dir_layout
;
932 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
933 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
934 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
935 ci
->i_dir_pin
= iinfo
->dir_pin
;
936 ci
->i_rsnaps
= iinfo
->rsnaps
;
937 ceph_decode_timespec64(&ci
->i_rctime
, &info
->rctime
);
942 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
943 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
944 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
945 if (ci
->i_xattrs
.blob
)
946 old_blob
= ci
->i_xattrs
.blob
;
947 ci
->i_xattrs
.blob
= xattr_blob
;
949 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
950 iinfo
->xattr_data
, iinfo
->xattr_len
);
951 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
952 ceph_forget_all_cached_acls(inode
);
953 ceph_security_invalidate_secctx(inode
);
957 /* finally update i_version */
958 if (le64_to_cpu(info
->version
) > ci
->i_version
)
959 ci
->i_version
= le64_to_cpu(info
->version
);
961 inode
->i_mapping
->a_ops
= &ceph_aops
;
963 switch (inode
->i_mode
& S_IFMT
) {
968 inode
->i_blkbits
= PAGE_SHIFT
;
969 init_special_inode(inode
, inode
->i_mode
, rdev
);
970 inode
->i_op
= &ceph_file_iops
;
973 inode
->i_op
= &ceph_file_iops
;
974 inode
->i_fop
= &ceph_file_fops
;
977 inode
->i_op
= &ceph_symlink_iops
;
978 if (!ci
->i_symlink
) {
979 u32 symlen
= iinfo
->symlink_len
;
982 spin_unlock(&ci
->i_ceph_lock
);
984 if (symlen
!= i_size_read(inode
)) {
985 pr_err("%s %llx.%llx BAD symlink "
986 "size %lld\n", __func__
,
989 i_size_write(inode
, symlen
);
990 inode
->i_blocks
= calc_inode_blocks(symlen
);
994 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
998 spin_lock(&ci
->i_ceph_lock
);
1000 ci
->i_symlink
= sym
;
1002 kfree(sym
); /* lost a race */
1004 inode
->i_link
= ci
->i_symlink
;
1007 inode
->i_op
= &ceph_dir_iops
;
1008 inode
->i_fop
= &ceph_dir_fops
;
1011 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__
,
1012 ceph_vinop(inode
), inode
->i_mode
);
1015 /* were we issued a capability? */
1017 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
1018 ceph_add_cap(inode
, session
,
1019 le64_to_cpu(info
->cap
.cap_id
),
1021 le32_to_cpu(info
->cap
.wanted
),
1022 le32_to_cpu(info
->cap
.seq
),
1023 le32_to_cpu(info
->cap
.mseq
),
1024 le64_to_cpu(info
->cap
.realm
),
1025 info
->cap
.flags
, &new_cap
);
1027 /* set dir completion flag? */
1028 if (S_ISDIR(inode
->i_mode
) &&
1029 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
1030 (info_caps
& CEPH_CAP_FILE_SHARED
) &&
1031 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
1032 !__ceph_dir_is_complete(ci
)) {
1033 dout(" marking %p complete (empty)\n", inode
);
1034 i_size_write(inode
, 0);
1035 __ceph_dir_set_complete(ci
,
1036 atomic64_read(&ci
->i_release_count
),
1037 atomic64_read(&ci
->i_ordered_count
));
1042 dout(" %p got snap_caps %s\n", inode
,
1043 ceph_cap_string(info_caps
));
1044 ci
->i_snap_caps
|= info_caps
;
1048 if (iinfo
->inline_version
> 0 &&
1049 iinfo
->inline_version
>= ci
->i_inline_version
) {
1050 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
1051 ci
->i_inline_version
= iinfo
->inline_version
;
1052 if (ceph_has_inline_data(ci
) &&
1053 (locked_page
|| (info_caps
& cache_caps
)))
1057 if (cap_fmode
>= 0) {
1059 pr_warn("mds issued no caps on %llx.%llx\n",
1061 __ceph_touch_fmode(ci
, mdsc
, cap_fmode
);
1064 spin_unlock(&ci
->i_ceph_lock
);
1066 ceph_fscache_register_inode_cookie(inode
);
1069 ceph_fill_inline_data(inode
, locked_page
,
1070 iinfo
->inline_data
, iinfo
->inline_len
);
1073 wake_up_all(&ci
->i_cap_wq
);
1075 /* queue truncate if we saw i_size decrease */
1077 ceph_queue_vmtruncate(inode
);
1079 /* populate frag tree */
1080 if (S_ISDIR(inode
->i_mode
))
1081 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
1083 /* update delegation info? */
1085 ceph_fill_dirfrag(inode
, dirinfo
);
1090 ceph_put_cap(mdsc
, new_cap
);
1091 ceph_buffer_put(old_blob
);
1092 ceph_buffer_put(xattr_blob
);
1093 ceph_put_string(pool_ns
);
1098 * caller should hold session s_mutex and dentry->d_lock.
1100 static void __update_dentry_lease(struct inode
*dir
, struct dentry
*dentry
,
1101 struct ceph_mds_reply_lease
*lease
,
1102 struct ceph_mds_session
*session
,
1103 unsigned long from_time
,
1104 struct ceph_mds_session
**old_lease_session
)
1106 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1107 unsigned mask
= le16_to_cpu(lease
->mask
);
1108 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1109 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1110 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1112 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1113 dentry
, duration
, ttl
);
1115 /* only track leases on regular dentries */
1116 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1119 if (mask
& CEPH_LEASE_PRIMARY_LINK
)
1120 di
->flags
|= CEPH_DENTRY_PRIMARY_LINK
;
1122 di
->flags
&= ~CEPH_DENTRY_PRIMARY_LINK
;
1124 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1125 if (!(mask
& CEPH_LEASE_VALID
)) {
1126 __ceph_dentry_dir_lease_touch(di
);
1130 if (di
->lease_gen
== atomic_read(&session
->s_cap_gen
) &&
1131 time_before(ttl
, di
->time
))
1132 return; /* we already have a newer lease. */
1134 if (di
->lease_session
&& di
->lease_session
!= session
) {
1135 *old_lease_session
= di
->lease_session
;
1136 di
->lease_session
= NULL
;
1139 if (!di
->lease_session
)
1140 di
->lease_session
= ceph_get_mds_session(session
);
1141 di
->lease_gen
= atomic_read(&session
->s_cap_gen
);
1142 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1143 di
->lease_renew_after
= half_ttl
;
1144 di
->lease_renew_from
= 0;
1147 __ceph_dentry_lease_touch(di
);
1150 static inline void update_dentry_lease(struct inode
*dir
, struct dentry
*dentry
,
1151 struct ceph_mds_reply_lease
*lease
,
1152 struct ceph_mds_session
*session
,
1153 unsigned long from_time
)
1155 struct ceph_mds_session
*old_lease_session
= NULL
;
1156 spin_lock(&dentry
->d_lock
);
1157 __update_dentry_lease(dir
, dentry
, lease
, session
, from_time
,
1158 &old_lease_session
);
1159 spin_unlock(&dentry
->d_lock
);
1160 ceph_put_mds_session(old_lease_session
);
1164 * update dentry lease without having parent inode locked
1166 static void update_dentry_lease_careful(struct dentry
*dentry
,
1167 struct ceph_mds_reply_lease
*lease
,
1168 struct ceph_mds_session
*session
,
1169 unsigned long from_time
,
1170 char *dname
, u32 dname_len
,
1171 struct ceph_vino
*pdvino
,
1172 struct ceph_vino
*ptvino
)
1176 struct ceph_mds_session
*old_lease_session
= NULL
;
1178 spin_lock(&dentry
->d_lock
);
1179 /* make sure dentry's name matches target */
1180 if (dentry
->d_name
.len
!= dname_len
||
1181 memcmp(dentry
->d_name
.name
, dname
, dname_len
))
1184 dir
= d_inode(dentry
->d_parent
);
1185 /* make sure parent matches dvino */
1186 if (!ceph_ino_compare(dir
, pdvino
))
1189 /* make sure dentry's inode matches target. NULL ptvino means that
1190 * we expect a negative dentry */
1192 if (d_really_is_negative(dentry
))
1194 if (!ceph_ino_compare(d_inode(dentry
), ptvino
))
1197 if (d_really_is_positive(dentry
))
1201 __update_dentry_lease(dir
, dentry
, lease
, session
,
1202 from_time
, &old_lease_session
);
1204 spin_unlock(&dentry
->d_lock
);
1205 ceph_put_mds_session(old_lease_session
);
1209 * splice a dentry to an inode.
1210 * caller must hold directory i_rwsem for this to be safe.
1212 static int splice_dentry(struct dentry
**pdn
, struct inode
*in
)
1214 struct dentry
*dn
= *pdn
;
1215 struct dentry
*realdn
;
1217 BUG_ON(d_inode(dn
));
1219 if (S_ISDIR(in
->i_mode
)) {
1220 /* If inode is directory, d_splice_alias() below will remove
1221 * 'realdn' from its origin parent. We need to ensure that
1222 * origin parent's readdir cache will not reference 'realdn'
1224 realdn
= d_find_any_alias(in
);
1226 struct ceph_dentry_info
*di
= ceph_dentry(realdn
);
1227 spin_lock(&realdn
->d_lock
);
1229 realdn
->d_op
->d_prune(realdn
);
1232 di
->lease_shared_gen
= 0;
1235 spin_unlock(&realdn
->d_lock
);
1240 /* dn must be unhashed */
1241 if (!d_unhashed(dn
))
1243 realdn
= d_splice_alias(in
, dn
);
1244 if (IS_ERR(realdn
)) {
1245 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1246 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1247 return PTR_ERR(realdn
);
1251 dout("dn %p (%d) spliced with %p (%d) "
1252 "inode %p ino %llx.%llx\n",
1254 realdn
, d_count(realdn
),
1255 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1259 BUG_ON(!ceph_dentry(dn
));
1260 dout("dn %p attached to %p ino %llx.%llx\n",
1261 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1267 * Incorporate results into the local cache. This is either just
1268 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1271 * A reply may contain
1272 * a directory inode along with a dentry.
1273 * and/or a target inode
1275 * Called with snap_rwsem (read).
1277 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1279 struct ceph_mds_session
*session
= req
->r_session
;
1280 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1281 struct inode
*in
= NULL
;
1282 struct ceph_vino tvino
, dvino
;
1283 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1286 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1287 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1289 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1290 dout("fill_trace reply is empty!\n");
1291 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1292 ceph_invalidate_dir_request(req
);
1296 if (rinfo
->head
->is_dentry
) {
1297 struct inode
*dir
= req
->r_parent
;
1300 err
= ceph_fill_inode(dir
, NULL
, &rinfo
->diri
,
1301 rinfo
->dirfrag
, session
, -1,
1302 &req
->r_caps_reservation
);
1309 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
&&
1310 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1311 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1313 struct dentry
*dn
, *parent
;
1315 BUG_ON(!rinfo
->head
->is_target
);
1316 BUG_ON(req
->r_dentry
);
1318 parent
= d_find_any_alias(dir
);
1321 dname
.name
= rinfo
->dname
;
1322 dname
.len
= rinfo
->dname_len
;
1323 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1324 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1325 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1327 dn
= d_lookup(parent
, &dname
);
1328 dout("d_lookup on parent=%p name=%.*s got %p\n",
1329 parent
, dname
.len
, dname
.name
, dn
);
1332 dn
= d_alloc(parent
, &dname
);
1333 dout("d_alloc %p '%.*s' = %p\n", parent
,
1334 dname
.len
, dname
.name
, dn
);
1341 } else if (d_really_is_positive(dn
) &&
1342 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1343 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1344 dout(" dn %p points to wrong inode %p\n",
1346 ceph_dir_clear_ordered(dir
);
1357 if (rinfo
->head
->is_target
) {
1358 /* Should be filled in by handle_reply */
1359 BUG_ON(!req
->r_target_inode
);
1361 in
= req
->r_target_inode
;
1362 err
= ceph_fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
,
1364 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1365 !test_bit(CEPH_MDS_R_ASYNC
, &req
->r_req_flags
) &&
1366 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1367 &req
->r_caps_reservation
);
1369 pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1370 in
, ceph_vinop(in
));
1371 req
->r_target_inode
= NULL
;
1372 if (in
->i_state
& I_NEW
)
1373 discard_new_inode(in
);
1378 if (in
->i_state
& I_NEW
)
1379 unlock_new_inode(in
);
1383 * ignore null lease/binding on snapdir ENOENT, or else we
1384 * will have trouble splicing in the virtual snapdir later
1386 if (rinfo
->head
->is_dentry
&&
1387 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1388 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1389 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1390 fsc
->mount_options
->snapdir_name
,
1391 req
->r_dentry
->d_name
.len
))) {
1393 * lookup link rename : null -> possibly existing inode
1394 * mknod symlink mkdir : null -> new inode
1395 * unlink : linked -> null
1397 struct inode
*dir
= req
->r_parent
;
1398 struct dentry
*dn
= req
->r_dentry
;
1399 bool have_dir_cap
, have_lease
;
1403 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1405 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1406 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1408 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1409 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1411 /* do we have a lease on the whole dir? */
1413 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1414 CEPH_CAP_FILE_SHARED
);
1416 /* do we have a dn lease? */
1417 have_lease
= have_dir_cap
||
1418 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1420 dout("fill_trace no dentry lease or dir cap\n");
1423 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1424 struct inode
*olddir
= req
->r_old_dentry_dir
;
1427 dout(" src %p '%pd' dst %p '%pd'\n",
1431 dout("fill_trace doing d_move %p -> %p\n",
1432 req
->r_old_dentry
, dn
);
1434 /* d_move screws up sibling dentries' offsets */
1435 ceph_dir_clear_ordered(dir
);
1436 ceph_dir_clear_ordered(olddir
);
1438 d_move(req
->r_old_dentry
, dn
);
1439 dout(" src %p '%pd' dst %p '%pd'\n",
1444 /* ensure target dentry is invalidated, despite
1445 rehashing bug in vfs_rename_dir */
1446 ceph_invalidate_dentry_lease(dn
);
1448 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1449 ceph_dentry(req
->r_old_dentry
)->offset
);
1451 /* swap r_dentry and r_old_dentry in case that
1452 * splice_dentry() gets called later. This is safe
1453 * because no other place will use them */
1454 req
->r_dentry
= req
->r_old_dentry
;
1455 req
->r_old_dentry
= dn
;
1460 if (!rinfo
->head
->is_target
) {
1461 dout("fill_trace null dentry\n");
1462 if (d_really_is_positive(dn
)) {
1463 dout("d_delete %p\n", dn
);
1464 ceph_dir_clear_ordered(dir
);
1466 } else if (have_lease
) {
1471 if (!d_unhashed(dn
) && have_lease
)
1472 update_dentry_lease(dir
, dn
,
1473 rinfo
->dlease
, session
,
1474 req
->r_request_started
);
1478 /* attach proper inode */
1479 if (d_really_is_negative(dn
)) {
1480 ceph_dir_clear_ordered(dir
);
1482 err
= splice_dentry(&req
->r_dentry
, in
);
1485 dn
= req
->r_dentry
; /* may have spliced */
1486 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1487 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1488 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1495 update_dentry_lease(dir
, dn
,
1496 rinfo
->dlease
, session
,
1497 req
->r_request_started
);
1499 dout(" final dn %p\n", dn
);
1500 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1501 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1502 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1503 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1504 struct inode
*dir
= req
->r_parent
;
1506 /* fill out a snapdir LOOKUPSNAP dentry */
1508 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1509 BUG_ON(!req
->r_dentry
);
1510 dout(" linking snapped dir %p to dn %p\n", in
, req
->r_dentry
);
1511 ceph_dir_clear_ordered(dir
);
1513 err
= splice_dentry(&req
->r_dentry
, in
);
1516 } else if (rinfo
->head
->is_dentry
&& req
->r_dentry
) {
1517 /* parent inode is not locked, be carefull */
1518 struct ceph_vino
*ptvino
= NULL
;
1519 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1520 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1521 if (rinfo
->head
->is_target
) {
1522 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1523 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1526 update_dentry_lease_careful(req
->r_dentry
, rinfo
->dlease
,
1527 session
, req
->r_request_started
,
1528 rinfo
->dname
, rinfo
->dname_len
,
1532 dout("fill_trace done err=%d\n", err
);
1537 * Prepopulate our cache with readdir results, leases, etc.
1539 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1540 struct ceph_mds_session
*session
)
1542 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1545 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1546 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1547 struct ceph_vino vino
;
1551 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1552 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1554 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1557 dout("new_inode badness got %d\n", err
);
1560 rc
= ceph_fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1561 -1, &req
->r_caps_reservation
);
1563 pr_err("ceph_fill_inode badness on %p got %d\n",
1566 if (in
->i_state
& I_NEW
) {
1568 discard_new_inode(in
);
1570 } else if (in
->i_state
& I_NEW
) {
1571 unlock_new_inode(in
);
1580 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1584 put_page(ctl
->page
);
1589 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1590 struct ceph_readdir_cache_control
*ctl
,
1591 struct ceph_mds_request
*req
)
1593 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1594 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1595 unsigned idx
= ctl
->index
% nsize
;
1596 pgoff_t pgoff
= ctl
->index
/ nsize
;
1598 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1599 ceph_readdir_cache_release(ctl
);
1601 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1603 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1606 return idx
== 0 ? -ENOMEM
: 0;
1608 /* reading/filling the cache are serialized by
1609 * i_rwsem, no need to use page lock */
1610 unlock_page(ctl
->page
);
1611 ctl
->dentries
= kmap(ctl
->page
);
1613 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1616 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1617 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1618 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1619 ctl
->dentries
[idx
] = dn
;
1622 dout("disable readdir cache\n");
1628 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1629 struct ceph_mds_session
*session
)
1631 struct dentry
*parent
= req
->r_dentry
;
1632 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1633 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1637 int err
= 0, skipped
= 0, ret
, i
;
1638 u32 frag
= le32_to_cpu(req
->r_args
.readdir
.frag
);
1641 struct ceph_readdir_cache_control cache_ctl
= {};
1643 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1644 return readdir_prepopulate_inodes_only(req
, session
);
1646 if (rinfo
->hash_order
) {
1648 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1650 strlen(req
->r_path2
));
1651 last_hash
= ceph_frag_value(last_hash
);
1652 } else if (rinfo
->offset_hash
) {
1653 /* mds understands offset_hash */
1654 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1655 last_hash
= le32_to_cpu(req
->r_args
.readdir
.offset_hash
);
1659 if (rinfo
->dir_dir
&&
1660 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1661 dout("readdir_prepopulate got new frag %x -> %x\n",
1662 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1663 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1664 if (!rinfo
->hash_order
)
1665 req
->r_readdir_offset
= 2;
1668 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1669 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1670 rinfo
->dir_nr
, parent
);
1672 dout("readdir_prepopulate %d items under dn %p\n",
1673 rinfo
->dir_nr
, parent
);
1675 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1677 if (ceph_frag_is_leftmost(frag
) &&
1678 req
->r_readdir_offset
== 2 &&
1679 !(rinfo
->hash_order
&& last_hash
)) {
1680 /* note dir version at start of readdir so we can
1681 * tell if any dentries get dropped */
1682 req
->r_dir_release_cnt
=
1683 atomic64_read(&ci
->i_release_count
);
1684 req
->r_dir_ordered_cnt
=
1685 atomic64_read(&ci
->i_ordered_count
);
1686 req
->r_readdir_cache_idx
= 0;
1690 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1691 fpos_offset
= req
->r_readdir_offset
;
1693 /* FIXME: release caps/leases if error occurs */
1694 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1695 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1696 struct ceph_vino tvino
;
1698 dname
.name
= rde
->name
;
1699 dname
.len
= rde
->name_len
;
1700 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1702 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1703 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1705 if (rinfo
->hash_order
) {
1706 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1707 rde
->name
, rde
->name_len
);
1708 hash
= ceph_frag_value(hash
);
1709 if (hash
!= last_hash
)
1712 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1714 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1718 dn
= d_lookup(parent
, &dname
);
1719 dout("d_lookup on parent=%p name=%.*s got %p\n",
1720 parent
, dname
.len
, dname
.name
, dn
);
1723 dn
= d_alloc(parent
, &dname
);
1724 dout("d_alloc %p '%.*s' = %p\n", parent
,
1725 dname
.len
, dname
.name
, dn
);
1727 dout("d_alloc badness\n");
1731 } else if (d_really_is_positive(dn
) &&
1732 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1733 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1734 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1735 dout(" dn %p points to wrong inode %p\n",
1738 spin_lock(&dn
->d_lock
);
1739 if (di
->offset
> 0 &&
1740 di
->lease_shared_gen
==
1741 atomic_read(&ci
->i_shared_gen
)) {
1742 __ceph_dir_clear_ordered(ci
);
1745 spin_unlock(&dn
->d_lock
);
1753 if (d_really_is_positive(dn
)) {
1756 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1758 dout("new_inode badness\n");
1766 ret
= ceph_fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1767 -1, &req
->r_caps_reservation
);
1769 pr_err("ceph_fill_inode badness on %p\n", in
);
1770 if (d_really_is_negative(dn
)) {
1771 if (in
->i_state
& I_NEW
) {
1773 discard_new_inode(in
);
1781 if (in
->i_state
& I_NEW
)
1782 unlock_new_inode(in
);
1784 if (d_really_is_negative(dn
)) {
1785 if (ceph_security_xattr_deadlock(in
)) {
1786 dout(" skip splicing dn %p to inode %p"
1787 " (security xattr deadlock)\n", dn
, in
);
1793 err
= splice_dentry(&dn
, in
);
1798 ceph_dentry(dn
)->offset
= rde
->offset
;
1800 update_dentry_lease(d_inode(parent
), dn
,
1801 rde
->lease
, req
->r_session
,
1802 req
->r_request_started
);
1804 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1805 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1814 if (err
== 0 && skipped
== 0) {
1815 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1816 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1818 ceph_readdir_cache_release(&cache_ctl
);
1819 dout("readdir_prepopulate done\n");
1823 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1825 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1828 spin_lock(&ci
->i_ceph_lock
);
1829 dout("set_size %p %llu -> %llu\n", inode
, i_size_read(inode
), size
);
1830 i_size_write(inode
, size
);
1831 ceph_fscache_update(inode
);
1832 inode
->i_blocks
= calc_inode_blocks(size
);
1834 ret
= __ceph_should_report_size(ci
);
1836 spin_unlock(&ci
->i_ceph_lock
);
1841 void ceph_queue_inode_work(struct inode
*inode
, int work_bit
)
1843 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1844 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1845 set_bit(work_bit
, &ci
->i_work_mask
);
1848 if (queue_work(fsc
->inode_wq
, &ci
->i_work
)) {
1849 dout("queue_inode_work %p, mask=%lx\n", inode
, ci
->i_work_mask
);
1851 dout("queue_inode_work %p already queued, mask=%lx\n",
1852 inode
, ci
->i_work_mask
);
1857 static void ceph_do_invalidate_pages(struct inode
*inode
)
1859 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1863 ceph_fscache_invalidate(inode
, false);
1865 mutex_lock(&ci
->i_truncate_mutex
);
1867 if (ceph_inode_is_shutdown(inode
)) {
1868 pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n",
1869 __func__
, ceph_vinop(inode
));
1870 mapping_set_error(inode
->i_mapping
, -EIO
);
1871 truncate_pagecache(inode
, 0);
1872 mutex_unlock(&ci
->i_truncate_mutex
);
1876 spin_lock(&ci
->i_ceph_lock
);
1877 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1878 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1879 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1880 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1882 spin_unlock(&ci
->i_ceph_lock
);
1883 mutex_unlock(&ci
->i_truncate_mutex
);
1886 orig_gen
= ci
->i_rdcache_gen
;
1887 spin_unlock(&ci
->i_ceph_lock
);
1889 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1890 pr_err("invalidate_inode_pages2 %llx.%llx failed\n",
1894 spin_lock(&ci
->i_ceph_lock
);
1895 if (orig_gen
== ci
->i_rdcache_gen
&&
1896 orig_gen
== ci
->i_rdcache_revoking
) {
1897 dout("invalidate_pages %p gen %d successful\n", inode
,
1899 ci
->i_rdcache_revoking
--;
1902 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1903 inode
, orig_gen
, ci
->i_rdcache_gen
,
1904 ci
->i_rdcache_revoking
);
1905 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1908 spin_unlock(&ci
->i_ceph_lock
);
1909 mutex_unlock(&ci
->i_truncate_mutex
);
1912 ceph_check_caps(ci
, 0);
1916 * Make sure any pending truncation is applied before doing anything
1917 * that may depend on it.
1919 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1921 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1923 int wrbuffer_refs
, finish
= 0;
1925 mutex_lock(&ci
->i_truncate_mutex
);
1927 spin_lock(&ci
->i_ceph_lock
);
1928 if (ci
->i_truncate_pending
== 0) {
1929 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1930 spin_unlock(&ci
->i_ceph_lock
);
1931 mutex_unlock(&ci
->i_truncate_mutex
);
1936 * make sure any dirty snapped pages are flushed before we
1937 * possibly truncate them.. so write AND block!
1939 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1940 spin_unlock(&ci
->i_ceph_lock
);
1941 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1943 filemap_write_and_wait_range(&inode
->i_data
, 0,
1944 inode
->i_sb
->s_maxbytes
);
1948 /* there should be no reader or writer */
1949 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1951 to
= ci
->i_truncate_size
;
1952 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1953 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1954 ci
->i_truncate_pending
, to
);
1955 spin_unlock(&ci
->i_ceph_lock
);
1957 ceph_fscache_resize(inode
, to
);
1958 truncate_pagecache(inode
, to
);
1960 spin_lock(&ci
->i_ceph_lock
);
1961 if (to
== ci
->i_truncate_size
) {
1962 ci
->i_truncate_pending
= 0;
1965 spin_unlock(&ci
->i_ceph_lock
);
1969 mutex_unlock(&ci
->i_truncate_mutex
);
1971 if (wrbuffer_refs
== 0)
1972 ceph_check_caps(ci
, 0);
1974 wake_up_all(&ci
->i_cap_wq
);
1977 static void ceph_inode_work(struct work_struct
*work
)
1979 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1981 struct inode
*inode
= &ci
->netfs
.inode
;
1983 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
)) {
1984 dout("writeback %p\n", inode
);
1985 filemap_fdatawrite(&inode
->i_data
);
1987 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
))
1988 ceph_do_invalidate_pages(inode
);
1990 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
))
1991 __ceph_do_pending_vmtruncate(inode
);
1993 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS
, &ci
->i_work_mask
))
1994 ceph_check_caps(ci
, 0);
1996 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS
, &ci
->i_work_mask
))
1997 ceph_flush_snaps(ci
, NULL
);
2005 static const struct inode_operations ceph_symlink_iops
= {
2006 .get_link
= simple_get_link
,
2007 .setattr
= ceph_setattr
,
2008 .getattr
= ceph_getattr
,
2009 .listxattr
= ceph_listxattr
,
2012 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
2014 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2015 unsigned int ia_valid
= attr
->ia_valid
;
2016 struct ceph_mds_request
*req
;
2017 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2018 struct ceph_cap_flush
*prealloc_cf
;
2020 int release
= 0, dirtied
= 0;
2023 int inode_dirty_flags
= 0;
2024 bool lock_snap_rwsem
= false;
2026 prealloc_cf
= ceph_alloc_cap_flush();
2030 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
2033 ceph_free_cap_flush(prealloc_cf
);
2034 return PTR_ERR(req
);
2037 spin_lock(&ci
->i_ceph_lock
);
2038 issued
= __ceph_caps_issued(ci
, NULL
);
2040 if (!ci
->i_head_snapc
&&
2041 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
2042 lock_snap_rwsem
= true;
2043 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
2044 spin_unlock(&ci
->i_ceph_lock
);
2045 down_read(&mdsc
->snap_rwsem
);
2046 spin_lock(&ci
->i_ceph_lock
);
2047 issued
= __ceph_caps_issued(ci
, NULL
);
2051 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
2053 if (ia_valid
& ATTR_UID
) {
2054 dout("setattr %p uid %d -> %d\n", inode
,
2055 from_kuid(&init_user_ns
, inode
->i_uid
),
2056 from_kuid(&init_user_ns
, attr
->ia_uid
));
2057 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2058 inode
->i_uid
= attr
->ia_uid
;
2059 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2060 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2061 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
2062 req
->r_args
.setattr
.uid
= cpu_to_le32(
2063 from_kuid(&init_user_ns
, attr
->ia_uid
));
2064 mask
|= CEPH_SETATTR_UID
;
2065 release
|= CEPH_CAP_AUTH_SHARED
;
2068 if (ia_valid
& ATTR_GID
) {
2069 dout("setattr %p gid %d -> %d\n", inode
,
2070 from_kgid(&init_user_ns
, inode
->i_gid
),
2071 from_kgid(&init_user_ns
, attr
->ia_gid
));
2072 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2073 inode
->i_gid
= attr
->ia_gid
;
2074 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2075 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2076 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
2077 req
->r_args
.setattr
.gid
= cpu_to_le32(
2078 from_kgid(&init_user_ns
, attr
->ia_gid
));
2079 mask
|= CEPH_SETATTR_GID
;
2080 release
|= CEPH_CAP_AUTH_SHARED
;
2083 if (ia_valid
& ATTR_MODE
) {
2084 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
2086 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2087 inode
->i_mode
= attr
->ia_mode
;
2088 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2089 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2090 attr
->ia_mode
!= inode
->i_mode
) {
2091 inode
->i_mode
= attr
->ia_mode
;
2092 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
2093 mask
|= CEPH_SETATTR_MODE
;
2094 release
|= CEPH_CAP_AUTH_SHARED
;
2098 if (ia_valid
& ATTR_ATIME
) {
2099 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode
,
2100 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
2101 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
2102 if (issued
& CEPH_CAP_FILE_EXCL
) {
2103 ci
->i_time_warp_seq
++;
2104 inode
->i_atime
= attr
->ia_atime
;
2105 dirtied
|= CEPH_CAP_FILE_EXCL
;
2106 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2107 timespec64_compare(&inode
->i_atime
,
2108 &attr
->ia_atime
) < 0) {
2109 inode
->i_atime
= attr
->ia_atime
;
2110 dirtied
|= CEPH_CAP_FILE_WR
;
2111 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2112 !timespec64_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2113 ceph_encode_timespec64(&req
->r_args
.setattr
.atime
,
2115 mask
|= CEPH_SETATTR_ATIME
;
2116 release
|= CEPH_CAP_FILE_SHARED
|
2117 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2120 if (ia_valid
& ATTR_SIZE
) {
2121 loff_t isize
= i_size_read(inode
);
2123 dout("setattr %p size %lld -> %lld\n", inode
, isize
, attr
->ia_size
);
2124 if ((issued
& CEPH_CAP_FILE_EXCL
) && attr
->ia_size
>= isize
) {
2125 if (attr
->ia_size
> isize
) {
2126 i_size_write(inode
, attr
->ia_size
);
2127 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2128 ci
->i_reported_size
= attr
->ia_size
;
2129 dirtied
|= CEPH_CAP_FILE_EXCL
;
2130 ia_valid
|= ATTR_MTIME
;
2132 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2133 attr
->ia_size
!= isize
) {
2134 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2135 req
->r_args
.setattr
.old_size
= cpu_to_le64(isize
);
2136 mask
|= CEPH_SETATTR_SIZE
;
2137 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_EXCL
|
2138 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2141 if (ia_valid
& ATTR_MTIME
) {
2142 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode
,
2143 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2144 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2145 if (issued
& CEPH_CAP_FILE_EXCL
) {
2146 ci
->i_time_warp_seq
++;
2147 inode
->i_mtime
= attr
->ia_mtime
;
2148 dirtied
|= CEPH_CAP_FILE_EXCL
;
2149 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2150 timespec64_compare(&inode
->i_mtime
,
2151 &attr
->ia_mtime
) < 0) {
2152 inode
->i_mtime
= attr
->ia_mtime
;
2153 dirtied
|= CEPH_CAP_FILE_WR
;
2154 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2155 !timespec64_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2156 ceph_encode_timespec64(&req
->r_args
.setattr
.mtime
,
2158 mask
|= CEPH_SETATTR_MTIME
;
2159 release
|= CEPH_CAP_FILE_SHARED
|
2160 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2164 /* these do nothing */
2165 if (ia_valid
& ATTR_CTIME
) {
2166 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2167 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2168 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode
,
2169 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2170 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2171 only
? "ctime only" : "ignored");
2174 * if kernel wants to dirty ctime but nothing else,
2175 * we need to choose a cap to dirty under, or do
2176 * a almost-no-op setattr
2178 if (issued
& CEPH_CAP_AUTH_EXCL
)
2179 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2180 else if (issued
& CEPH_CAP_FILE_EXCL
)
2181 dirtied
|= CEPH_CAP_FILE_EXCL
;
2182 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2183 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2185 mask
|= CEPH_SETATTR_CTIME
;
2188 if (ia_valid
& ATTR_FILE
)
2189 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2192 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2194 inode
->i_ctime
= attr
->ia_ctime
;
2195 inode_inc_iversion_raw(inode
);
2199 spin_unlock(&ci
->i_ceph_lock
);
2200 if (lock_snap_rwsem
)
2201 up_read(&mdsc
->snap_rwsem
);
2203 if (inode_dirty_flags
)
2204 __mark_inode_dirty(inode
, inode_dirty_flags
);
2207 req
->r_inode
= inode
;
2209 req
->r_inode_drop
= release
;
2210 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2211 req
->r_num_caps
= 1;
2212 req
->r_stamp
= attr
->ia_ctime
;
2213 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2215 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2216 ceph_cap_string(dirtied
), mask
);
2218 ceph_mdsc_put_request(req
);
2219 ceph_free_cap_flush(prealloc_cf
);
2221 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2222 __ceph_do_pending_vmtruncate(inode
);
2230 int ceph_setattr(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
2233 struct inode
*inode
= d_inode(dentry
);
2234 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
2237 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2240 if (ceph_inode_is_shutdown(inode
))
2243 err
= setattr_prepare(&nop_mnt_idmap
, dentry
, attr
);
2247 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2248 attr
->ia_size
> max(i_size_read(inode
), fsc
->max_file_size
))
2251 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2252 ceph_quota_is_max_bytes_exceeded(inode
, attr
->ia_size
))
2255 err
= __ceph_setattr(inode
, attr
);
2257 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2258 err
= posix_acl_chmod(&nop_mnt_idmap
, dentry
, attr
->ia_mode
);
2263 int ceph_try_to_choose_auth_mds(struct inode
*inode
, int mask
)
2265 int issued
= ceph_caps_issued(ceph_inode(inode
));
2268 * If any 'x' caps is issued we can just choose the auth MDS
2269 * instead of the random replica MDSes. Because only when the
2270 * Locker is in LOCK_EXEC state will the loner client could
2271 * get the 'x' caps. And if we send the getattr requests to
2272 * any replica MDS it must auth pin and tries to rdlock from
2273 * the auth MDS, and then the auth MDS need to do the Locker
2274 * state transition to LOCK_SYNC. And after that the lock state
2277 * This cost much when doing the Locker state transition and
2278 * usually will need to revoke caps from clients.
2280 * And for the 'Xs' caps for getxattr we will also choose the
2281 * auth MDS, because the MDS side code is buggy due to setxattr
2282 * won't notify the replica MDSes when the values changed and
2283 * the replica MDS will return the old values. Though we will
2284 * fix it in MDS code, but this still makes sense for old ceph.
2286 if (((mask
& CEPH_CAP_ANY_SHARED
) && (issued
& CEPH_CAP_ANY_EXCL
))
2287 || (mask
& (CEPH_STAT_RSTAT
| CEPH_STAT_CAP_XATTR
)))
2288 return USE_AUTH_MDS
;
2294 * Verify that we have a lease on the given mask. If not,
2295 * do a getattr against an mds.
2297 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2298 int mask
, bool force
)
2300 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2301 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2302 struct ceph_mds_request
*req
;
2306 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2307 dout("do_getattr inode %p SNAPDIR\n", inode
);
2311 dout("do_getattr inode %p mask %s mode 0%o\n",
2312 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2313 if (!force
&& ceph_caps_issued_mask_metric(ceph_inode(inode
), mask
, 1))
2316 mode
= ceph_try_to_choose_auth_mds(inode
, mask
);
2317 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, mode
);
2319 return PTR_ERR(req
);
2320 req
->r_inode
= inode
;
2322 req
->r_num_caps
= 1;
2323 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2324 req
->r_locked_page
= locked_page
;
2325 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2326 if (locked_page
&& err
== 0) {
2327 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2328 if (inline_version
== 0) {
2329 /* the reply is supposed to contain inline data */
2331 } else if (inline_version
== CEPH_INLINE_NONE
||
2332 inline_version
== 1) {
2335 err
= req
->r_reply_info
.targeti
.inline_len
;
2338 ceph_mdsc_put_request(req
);
2339 dout("do_getattr result=%d\n", err
);
2343 int ceph_do_getvxattr(struct inode
*inode
, const char *name
, void *value
,
2346 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2347 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2348 struct ceph_mds_request
*req
;
2349 int mode
= USE_AUTH_MDS
;
2352 size_t xattr_value_len
;
2354 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETVXATTR
, mode
);
2360 req
->r_feature_needed
= CEPHFS_FEATURE_OP_GETVXATTR
;
2361 req
->r_path2
= kstrdup(name
, GFP_NOFS
);
2362 if (!req
->r_path2
) {
2368 req
->r_inode
= inode
;
2369 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2373 xattr_value
= req
->r_reply_info
.xattr_info
.xattr_value
;
2374 xattr_value_len
= req
->r_reply_info
.xattr_info
.xattr_value_len
;
2376 dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len
, size
);
2378 err
= (int)xattr_value_len
;
2382 if (xattr_value_len
> size
) {
2387 memcpy(value
, xattr_value
, xattr_value_len
);
2389 ceph_mdsc_put_request(req
);
2391 dout("do_getvxattr result=%d\n", err
);
2397 * Check inode permissions. We verify we have a valid value for
2398 * the AUTH cap, then call the generic handler.
2400 int ceph_permission(struct user_namespace
*mnt_userns
, struct inode
*inode
,
2405 if (mask
& MAY_NOT_BLOCK
)
2408 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2411 err
= generic_permission(&init_user_ns
, inode
, mask
);
2415 /* Craft a mask of needed caps given a set of requested statx attrs. */
2416 static int statx_to_caps(u32 want
, umode_t mode
)
2420 if (want
& (STATX_MODE
|STATX_UID
|STATX_GID
|STATX_CTIME
|STATX_BTIME
))
2421 mask
|= CEPH_CAP_AUTH_SHARED
;
2423 if (want
& (STATX_NLINK
|STATX_CTIME
)) {
2425 * The link count for directories depends on inode->i_subdirs,
2426 * and that is only updated when Fs caps are held.
2429 mask
|= CEPH_CAP_FILE_SHARED
;
2431 mask
|= CEPH_CAP_LINK_SHARED
;
2434 if (want
& (STATX_ATIME
|STATX_MTIME
|STATX_CTIME
|STATX_SIZE
|
2436 mask
|= CEPH_CAP_FILE_SHARED
;
2438 if (want
& (STATX_CTIME
))
2439 mask
|= CEPH_CAP_XATTR_SHARED
;
2445 * Get all the attributes. If we have sufficient caps for the requested attrs,
2446 * then we can avoid talking to the MDS at all.
2448 int ceph_getattr(struct mnt_idmap
*idmap
, const struct path
*path
,
2449 struct kstat
*stat
, u32 request_mask
, unsigned int flags
)
2451 struct inode
*inode
= d_inode(path
->dentry
);
2452 struct super_block
*sb
= inode
->i_sb
;
2453 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2454 u32 valid_mask
= STATX_BASIC_STATS
;
2457 if (ceph_inode_is_shutdown(inode
))
2460 /* Skip the getattr altogether if we're asked not to sync */
2461 if ((flags
& AT_STATX_SYNC_TYPE
) != AT_STATX_DONT_SYNC
) {
2462 err
= ceph_do_getattr(inode
,
2463 statx_to_caps(request_mask
, inode
->i_mode
),
2464 flags
& AT_STATX_FORCE_SYNC
);
2469 generic_fillattr(&nop_mnt_idmap
, inode
, stat
);
2470 stat
->ino
= ceph_present_inode(inode
);
2473 * btime on newly-allocated inodes is 0, so if this is still set to
2474 * that, then assume that it's not valid.
2476 if (ci
->i_btime
.tv_sec
|| ci
->i_btime
.tv_nsec
) {
2477 stat
->btime
= ci
->i_btime
;
2478 valid_mask
|= STATX_BTIME
;
2481 if (ceph_snap(inode
) == CEPH_NOSNAP
)
2482 stat
->dev
= sb
->s_dev
;
2484 stat
->dev
= ci
->i_snapid_map
? ci
->i_snapid_map
->dev
: 0;
2486 if (S_ISDIR(inode
->i_mode
)) {
2487 if (ceph_test_mount_opt(ceph_sb_to_client(sb
), RBYTES
)) {
2488 stat
->size
= ci
->i_rbytes
;
2489 } else if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2490 struct ceph_inode_info
*pci
;
2491 struct ceph_snap_realm
*realm
;
2492 struct inode
*parent
;
2494 parent
= ceph_lookup_inode(sb
, ceph_ino(inode
));
2496 return PTR_ERR(parent
);
2498 pci
= ceph_inode(parent
);
2499 spin_lock(&pci
->i_ceph_lock
);
2500 realm
= pci
->i_snap_realm
;
2502 stat
->size
= realm
->num_snaps
;
2505 spin_unlock(&pci
->i_ceph_lock
);
2508 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2511 stat
->blksize
= 65536;
2513 * Some applications rely on the number of st_nlink
2514 * value on directories to be either 0 (if unlinked)
2515 * or 2 + number of subdirectories.
2517 if (stat
->nlink
== 1)
2518 /* '.' + '..' + subdirs */
2519 stat
->nlink
= 1 + 1 + ci
->i_subdirs
;
2522 stat
->result_mask
= request_mask
& valid_mask
;
2526 void ceph_inode_shutdown(struct inode
*inode
)
2528 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2531 bool invalidate
= false;
2533 spin_lock(&ci
->i_ceph_lock
);
2534 ci
->i_ceph_flags
|= CEPH_I_SHUTDOWN
;
2535 p
= rb_first(&ci
->i_caps
);
2537 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
2540 iputs
+= ceph_purge_inode_cap(inode
, cap
, &invalidate
);
2542 spin_unlock(&ci
->i_ceph_lock
);
2545 ceph_queue_invalidate(inode
);