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
;
447 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
448 RCU_INIT_POINTER(ci
->i_layout
.pool_ns
, NULL
);
450 ci
->i_fragtree
= RB_ROOT
;
451 mutex_init(&ci
->i_fragtree_mutex
);
453 ci
->i_xattrs
.blob
= NULL
;
454 ci
->i_xattrs
.prealloc_blob
= NULL
;
455 ci
->i_xattrs
.dirty
= false;
456 ci
->i_xattrs
.index
= RB_ROOT
;
457 ci
->i_xattrs
.count
= 0;
458 ci
->i_xattrs
.names_size
= 0;
459 ci
->i_xattrs
.vals_size
= 0;
460 ci
->i_xattrs
.version
= 0;
461 ci
->i_xattrs
.index_version
= 0;
463 ci
->i_caps
= RB_ROOT
;
464 ci
->i_auth_cap
= NULL
;
465 ci
->i_dirty_caps
= 0;
466 ci
->i_flushing_caps
= 0;
467 INIT_LIST_HEAD(&ci
->i_dirty_item
);
468 INIT_LIST_HEAD(&ci
->i_flushing_item
);
469 ci
->i_prealloc_cap_flush
= NULL
;
470 INIT_LIST_HEAD(&ci
->i_cap_flush_list
);
471 init_waitqueue_head(&ci
->i_cap_wq
);
472 ci
->i_hold_caps_min
= 0;
473 ci
->i_hold_caps_max
= 0;
474 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
475 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
476 ci
->i_head_snapc
= NULL
;
479 for (i
= 0; i
< CEPH_FILE_MODE_BITS
; i
++)
480 ci
->i_nr_by_mode
[i
] = 0;
482 mutex_init(&ci
->i_truncate_mutex
);
483 ci
->i_truncate_seq
= 0;
484 ci
->i_truncate_size
= 0;
485 ci
->i_truncate_pending
= 0;
488 ci
->i_reported_size
= 0;
489 ci
->i_wanted_max_size
= 0;
490 ci
->i_requested_max_size
= 0;
494 ci
->i_rdcache_ref
= 0;
497 ci
->i_wrbuffer_ref
= 0;
498 ci
->i_wrbuffer_ref_head
= 0;
499 atomic_set(&ci
->i_filelock_ref
, 0);
500 atomic_set(&ci
->i_shared_gen
, 0);
501 ci
->i_rdcache_gen
= 0;
502 ci
->i_rdcache_revoking
= 0;
504 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
505 INIT_LIST_HEAD(&ci
->i_unsafe_iops
);
506 spin_lock_init(&ci
->i_unsafe_lock
);
508 ci
->i_snap_realm
= NULL
;
509 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
510 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
512 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
513 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
515 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
517 ceph_fscache_inode_init(ci
);
519 return &ci
->vfs_inode
;
522 static void ceph_i_callback(struct rcu_head
*head
)
524 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
525 struct ceph_inode_info
*ci
= ceph_inode(inode
);
527 kfree(ci
->i_symlink
);
528 kmem_cache_free(ceph_inode_cachep
, ci
);
531 void ceph_evict_inode(struct inode
*inode
)
533 struct ceph_inode_info
*ci
= ceph_inode(inode
);
534 struct ceph_inode_frag
*frag
;
537 dout("evict_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
539 truncate_inode_pages_final(&inode
->i_data
);
542 ceph_fscache_unregister_inode_cookie(ci
);
544 ceph_queue_caps_release(inode
);
546 if (__ceph_has_any_quota(ci
))
547 ceph_adjust_quota_realms_count(inode
, false);
550 * we may still have a snap_realm reference if there are stray
551 * caps in i_snap_caps.
553 if (ci
->i_snap_realm
) {
554 struct ceph_mds_client
*mdsc
=
555 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
556 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
558 dout(" dropping residual ref to snap realm %p\n", realm
);
559 spin_lock(&realm
->inodes_with_caps_lock
);
560 list_del_init(&ci
->i_snap_realm_item
);
561 ci
->i_snap_realm
= NULL
;
562 if (realm
->ino
== ci
->i_vino
.ino
)
564 spin_unlock(&realm
->inodes_with_caps_lock
);
565 ceph_put_snap_realm(mdsc
, realm
);
568 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
569 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
570 rb_erase(n
, &ci
->i_fragtree
);
573 ci
->i_fragtree_nsplits
= 0;
575 __ceph_destroy_xattrs(ci
);
576 if (ci
->i_xattrs
.blob
)
577 ceph_buffer_put(ci
->i_xattrs
.blob
);
578 if (ci
->i_xattrs
.prealloc_blob
)
579 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
581 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
584 void ceph_destroy_inode(struct inode
*inode
)
586 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
589 int ceph_drop_inode(struct inode
*inode
)
592 * Positve dentry and corresponding inode are always accompanied
593 * in MDS reply. So no need to keep inode in the cache after
594 * dropping all its aliases.
599 static inline blkcnt_t
calc_inode_blocks(u64 size
)
601 return (size
+ (1<<9) - 1) >> 9;
605 * Helpers to fill in size, ctime, mtime, and atime. We have to be
606 * careful because either the client or MDS may have more up to date
607 * info, depending on which capabilities are held, and whether
608 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
609 * and size are monotonically increasing, except when utimes() or
610 * truncate() increments the corresponding _seq values.)
612 int ceph_fill_file_size(struct inode
*inode
, int issued
,
613 u32 truncate_seq
, u64 truncate_size
, u64 size
)
615 struct ceph_inode_info
*ci
= ceph_inode(inode
);
618 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
619 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
620 dout("size %lld -> %llu\n", inode
->i_size
, size
);
621 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
622 pr_err("fill_file_size non-zero size for directory\n");
625 i_size_write(inode
, size
);
626 inode
->i_blocks
= calc_inode_blocks(size
);
627 ci
->i_reported_size
= size
;
628 if (truncate_seq
!= ci
->i_truncate_seq
) {
629 dout("truncate_seq %u -> %u\n",
630 ci
->i_truncate_seq
, truncate_seq
);
631 ci
->i_truncate_seq
= truncate_seq
;
633 /* the MDS should have revoked these caps */
634 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
637 CEPH_CAP_FILE_LAZYIO
));
639 * If we hold relevant caps, or in the case where we're
640 * not the only client referencing this file and we
641 * don't hold those caps, then we need to check whether
642 * the file is either opened or mmaped
644 if ((issued
& (CEPH_CAP_FILE_CACHE
|
645 CEPH_CAP_FILE_BUFFER
)) ||
646 mapping_mapped(inode
->i_mapping
) ||
647 __ceph_caps_file_wanted(ci
)) {
648 ci
->i_truncate_pending
++;
653 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
654 ci
->i_truncate_size
!= truncate_size
) {
655 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
657 ci
->i_truncate_size
= truncate_size
;
661 ceph_fscache_invalidate(inode
);
666 void ceph_fill_file_time(struct inode
*inode
, int issued
,
667 u64 time_warp_seq
, struct timespec64
*ctime
,
668 struct timespec64
*mtime
, struct timespec64
*atime
)
670 struct ceph_inode_info
*ci
= ceph_inode(inode
);
673 if (issued
& (CEPH_CAP_FILE_EXCL
|
675 CEPH_CAP_FILE_BUFFER
|
677 CEPH_CAP_XATTR_EXCL
)) {
678 if (ci
->i_version
== 0 ||
679 timespec64_compare(ctime
, &inode
->i_ctime
) > 0) {
680 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
681 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
682 ctime
->tv_sec
, ctime
->tv_nsec
);
683 inode
->i_ctime
= *ctime
;
685 if (ci
->i_version
== 0 ||
686 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
687 /* the MDS did a utimes() */
688 dout("mtime %lld.%09ld -> %lld.%09ld "
690 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
691 mtime
->tv_sec
, mtime
->tv_nsec
,
692 ci
->i_time_warp_seq
, (int)time_warp_seq
);
694 inode
->i_mtime
= *mtime
;
695 inode
->i_atime
= *atime
;
696 ci
->i_time_warp_seq
= time_warp_seq
;
697 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
698 /* nobody did utimes(); take the max */
699 if (timespec64_compare(mtime
, &inode
->i_mtime
) > 0) {
700 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
701 inode
->i_mtime
.tv_sec
,
702 inode
->i_mtime
.tv_nsec
,
703 mtime
->tv_sec
, mtime
->tv_nsec
);
704 inode
->i_mtime
= *mtime
;
706 if (timespec64_compare(atime
, &inode
->i_atime
) > 0) {
707 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
708 inode
->i_atime
.tv_sec
,
709 inode
->i_atime
.tv_nsec
,
710 atime
->tv_sec
, atime
->tv_nsec
);
711 inode
->i_atime
= *atime
;
713 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
714 /* we did a utimes(); ignore mds values */
719 /* we have no write|excl caps; whatever the MDS says is true */
720 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
721 inode
->i_ctime
= *ctime
;
722 inode
->i_mtime
= *mtime
;
723 inode
->i_atime
= *atime
;
724 ci
->i_time_warp_seq
= time_warp_seq
;
729 if (warn
) /* time_warp_seq shouldn't go backwards */
730 dout("%p mds time_warp_seq %llu < %u\n",
731 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
735 * Populate an inode based on info from mds. May be called on new or
738 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
739 struct ceph_mds_reply_info_in
*iinfo
,
740 struct ceph_mds_reply_dirfrag
*dirinfo
,
741 struct ceph_mds_session
*session
,
742 unsigned long ttl_from
, int cap_fmode
,
743 struct ceph_cap_reservation
*caps_reservation
)
745 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
746 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
747 struct ceph_inode_info
*ci
= ceph_inode(inode
);
748 int issued
, new_issued
, info_caps
;
749 struct timespec64 mtime
, atime
, ctime
;
750 struct ceph_buffer
*xattr_blob
= NULL
;
751 struct ceph_buffer
*old_blob
= NULL
;
752 struct ceph_string
*pool_ns
= NULL
;
753 struct ceph_cap
*new_cap
= NULL
;
756 bool queue_trunc
= false;
757 bool new_version
= false;
758 bool fill_inline
= false;
760 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
761 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
764 info_caps
= le32_to_cpu(info
->cap
.caps
);
766 /* prealloc new cap struct */
767 if (info_caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
768 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
771 * prealloc xattr data, if it looks like we'll need it. only
772 * if len > 4 (meaning there are actually xattrs; the first 4
773 * bytes are the xattr count).
775 if (iinfo
->xattr_len
> 4) {
776 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
778 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
782 if (iinfo
->pool_ns_len
> 0)
783 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
786 spin_lock(&ci
->i_ceph_lock
);
789 * provided version will be odd if inode value is projected,
790 * even if stable. skip the update if we have newer stable
791 * info (ours>=theirs, e.g. due to racing mds replies), unless
792 * we are getting projected (unstable) info (in which case the
793 * version is odd, and we want ours>theirs).
799 if (ci
->i_version
== 0 ||
800 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
801 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
804 __ceph_caps_issued(ci
, &issued
);
805 issued
|= __ceph_caps_dirty(ci
);
806 new_issued
= ~issued
& info_caps
;
809 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
810 /* directories have fl_stripe_unit set to zero */
811 if (le32_to_cpu(info
->layout
.fl_stripe_unit
))
813 fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
815 inode
->i_blkbits
= CEPH_BLOCK_SHIFT
;
817 __ceph_update_quota(ci
, iinfo
->max_bytes
, iinfo
->max_files
);
819 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
820 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
821 inode
->i_mode
= le32_to_cpu(info
->mode
);
822 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
823 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
824 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
825 from_kuid(&init_user_ns
, inode
->i_uid
),
826 from_kgid(&init_user_ns
, inode
->i_gid
));
829 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
830 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
831 set_nlink(inode
, le32_to_cpu(info
->nlink
));
833 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
834 /* be careful with mtime, atime, size */
835 ceph_decode_timespec64(&atime
, &info
->atime
);
836 ceph_decode_timespec64(&mtime
, &info
->mtime
);
837 ceph_decode_timespec64(&ctime
, &info
->ctime
);
838 ceph_fill_file_time(inode
, issued
,
839 le32_to_cpu(info
->time_warp_seq
),
840 &ctime
, &mtime
, &atime
);
843 if (new_version
|| (info_caps
& CEPH_CAP_FILE_SHARED
)) {
844 ci
->i_files
= le64_to_cpu(info
->files
);
845 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
849 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
850 s64 old_pool
= ci
->i_layout
.pool_id
;
851 struct ceph_string
*old_ns
;
853 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
854 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
855 lockdep_is_held(&ci
->i_ceph_lock
));
856 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
858 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
859 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
863 queue_trunc
= ceph_fill_file_size(inode
, issued
,
864 le32_to_cpu(info
->truncate_seq
),
865 le64_to_cpu(info
->truncate_size
),
866 le64_to_cpu(info
->size
));
867 /* only update max_size on auth cap */
868 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
869 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
870 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
871 le64_to_cpu(info
->max_size
));
872 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
876 /* layout and rstat are not tracked by capability, update them if
877 * the inode info is from auth mds */
878 if (new_version
|| (info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
)) {
879 if (S_ISDIR(inode
->i_mode
)) {
880 ci
->i_dir_layout
= iinfo
->dir_layout
;
881 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
882 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
883 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
884 ceph_decode_timespec64(&ci
->i_rctime
, &info
->rctime
);
889 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
890 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
891 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
892 if (ci
->i_xattrs
.blob
)
893 old_blob
= ci
->i_xattrs
.blob
;
894 ci
->i_xattrs
.blob
= xattr_blob
;
896 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
897 iinfo
->xattr_data
, iinfo
->xattr_len
);
898 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
899 ceph_forget_all_cached_acls(inode
);
903 /* finally update i_version */
904 if (le64_to_cpu(info
->version
) > ci
->i_version
)
905 ci
->i_version
= le64_to_cpu(info
->version
);
907 inode
->i_mapping
->a_ops
= &ceph_aops
;
909 switch (inode
->i_mode
& S_IFMT
) {
914 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
915 inode
->i_op
= &ceph_file_iops
;
918 inode
->i_op
= &ceph_file_iops
;
919 inode
->i_fop
= &ceph_file_fops
;
922 inode
->i_op
= &ceph_symlink_iops
;
923 if (!ci
->i_symlink
) {
924 u32 symlen
= iinfo
->symlink_len
;
927 spin_unlock(&ci
->i_ceph_lock
);
929 if (symlen
!= i_size_read(inode
)) {
930 pr_err("fill_inode %llx.%llx BAD symlink "
931 "size %lld\n", ceph_vinop(inode
),
933 i_size_write(inode
, symlen
);
934 inode
->i_blocks
= calc_inode_blocks(symlen
);
938 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
942 spin_lock(&ci
->i_ceph_lock
);
946 kfree(sym
); /* lost a race */
948 inode
->i_link
= ci
->i_symlink
;
951 inode
->i_op
= &ceph_dir_iops
;
952 inode
->i_fop
= &ceph_dir_fops
;
955 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
956 ceph_vinop(inode
), inode
->i_mode
);
959 /* were we issued a capability? */
961 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
962 ceph_add_cap(inode
, session
,
963 le64_to_cpu(info
->cap
.cap_id
),
964 cap_fmode
, info_caps
,
965 le32_to_cpu(info
->cap
.wanted
),
966 le32_to_cpu(info
->cap
.seq
),
967 le32_to_cpu(info
->cap
.mseq
),
968 le64_to_cpu(info
->cap
.realm
),
969 info
->cap
.flags
, &new_cap
);
971 /* set dir completion flag? */
972 if (S_ISDIR(inode
->i_mode
) &&
973 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
974 (info_caps
& CEPH_CAP_FILE_SHARED
) &&
975 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
976 !__ceph_dir_is_complete(ci
)) {
977 dout(" marking %p complete (empty)\n", inode
);
978 i_size_write(inode
, 0);
979 __ceph_dir_set_complete(ci
,
980 atomic64_read(&ci
->i_release_count
),
981 atomic64_read(&ci
->i_ordered_count
));
986 dout(" %p got snap_caps %s\n", inode
,
987 ceph_cap_string(info_caps
));
988 ci
->i_snap_caps
|= info_caps
;
990 __ceph_get_fmode(ci
, cap_fmode
);
992 } else if (cap_fmode
>= 0) {
993 pr_warn("mds issued no caps on %llx.%llx\n",
995 __ceph_get_fmode(ci
, cap_fmode
);
998 if (iinfo
->inline_version
> 0 &&
999 iinfo
->inline_version
>= ci
->i_inline_version
) {
1000 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
1001 ci
->i_inline_version
= iinfo
->inline_version
;
1002 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
1003 (locked_page
|| (info_caps
& cache_caps
)))
1007 spin_unlock(&ci
->i_ceph_lock
);
1010 ceph_fill_inline_data(inode
, locked_page
,
1011 iinfo
->inline_data
, iinfo
->inline_len
);
1014 wake_up_all(&ci
->i_cap_wq
);
1016 /* queue truncate if we saw i_size decrease */
1018 ceph_queue_vmtruncate(inode
);
1020 /* populate frag tree */
1021 if (S_ISDIR(inode
->i_mode
))
1022 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
1024 /* update delegation info? */
1026 ceph_fill_dirfrag(inode
, dirinfo
);
1031 ceph_put_cap(mdsc
, new_cap
);
1032 ceph_buffer_put(old_blob
);
1033 ceph_buffer_put(xattr_blob
);
1034 ceph_put_string(pool_ns
);
1039 * caller should hold session s_mutex.
1041 static void update_dentry_lease(struct dentry
*dentry
,
1042 struct ceph_mds_reply_lease
*lease
,
1043 struct ceph_mds_session
*session
,
1044 unsigned long from_time
,
1045 struct ceph_vino
*tgt_vino
,
1046 struct ceph_vino
*dir_vino
)
1048 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1049 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1050 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1051 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1053 struct ceph_mds_session
*old_lease_session
= NULL
;
1056 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1057 * we expect a negative dentry.
1059 if (!tgt_vino
&& d_really_is_positive(dentry
))
1062 if (tgt_vino
&& (d_really_is_negative(dentry
) ||
1063 !ceph_ino_compare(d_inode(dentry
), tgt_vino
)))
1066 spin_lock(&dentry
->d_lock
);
1067 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1068 dentry
, duration
, ttl
);
1070 dir
= d_inode(dentry
->d_parent
);
1072 /* make sure parent matches dir_vino */
1073 if (!ceph_ino_compare(dir
, dir_vino
))
1076 /* only track leases on regular dentries */
1077 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1080 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1085 if (di
->lease_gen
== session
->s_cap_gen
&&
1086 time_before(ttl
, di
->time
))
1087 goto out_unlock
; /* we already have a newer lease. */
1089 if (di
->lease_session
&& di
->lease_session
!= session
) {
1090 old_lease_session
= di
->lease_session
;
1091 di
->lease_session
= NULL
;
1094 ceph_dentry_lru_touch(dentry
);
1096 if (!di
->lease_session
)
1097 di
->lease_session
= ceph_get_mds_session(session
);
1098 di
->lease_gen
= session
->s_cap_gen
;
1099 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1100 di
->lease_renew_after
= half_ttl
;
1101 di
->lease_renew_from
= 0;
1104 spin_unlock(&dentry
->d_lock
);
1105 if (old_lease_session
)
1106 ceph_put_mds_session(old_lease_session
);
1110 * splice a dentry to an inode.
1111 * caller must hold directory i_mutex for this to be safe.
1113 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
)
1115 struct dentry
*realdn
;
1117 BUG_ON(d_inode(dn
));
1119 if (S_ISDIR(in
->i_mode
)) {
1120 /* If inode is directory, d_splice_alias() below will remove
1121 * 'realdn' from its origin parent. We need to ensure that
1122 * origin parent's readdir cache will not reference 'realdn'
1124 realdn
= d_find_any_alias(in
);
1126 struct ceph_dentry_info
*di
= ceph_dentry(realdn
);
1127 spin_lock(&realdn
->d_lock
);
1129 realdn
->d_op
->d_prune(realdn
);
1132 di
->lease_shared_gen
= 0;
1135 spin_unlock(&realdn
->d_lock
);
1140 /* dn must be unhashed */
1141 if (!d_unhashed(dn
))
1143 realdn
= d_splice_alias(in
, dn
);
1144 if (IS_ERR(realdn
)) {
1145 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1146 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1149 * Caller should release 'dn' in the case of error.
1150 * If 'req->r_dentry' is passed to this function,
1151 * caller should leave 'req->r_dentry' untouched.
1154 } else if (realdn
) {
1155 dout("dn %p (%d) spliced with %p (%d) "
1156 "inode %p ino %llx.%llx\n",
1158 realdn
, d_count(realdn
),
1159 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1163 BUG_ON(!ceph_dentry(dn
));
1164 dout("dn %p attached to %p ino %llx.%llx\n",
1165 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1172 * Incorporate results into the local cache. This is either just
1173 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1176 * A reply may contain
1177 * a directory inode along with a dentry.
1178 * and/or a target inode
1180 * Called with snap_rwsem (read).
1182 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1184 struct ceph_mds_session
*session
= req
->r_session
;
1185 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1186 struct inode
*in
= NULL
;
1187 struct ceph_vino tvino
, dvino
;
1188 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1191 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1192 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1194 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1195 dout("fill_trace reply is empty!\n");
1196 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1197 ceph_invalidate_dir_request(req
);
1201 if (rinfo
->head
->is_dentry
) {
1202 struct inode
*dir
= req
->r_parent
;
1205 err
= fill_inode(dir
, NULL
,
1206 &rinfo
->diri
, rinfo
->dirfrag
,
1207 session
, req
->r_request_started
, -1,
1208 &req
->r_caps_reservation
);
1215 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1217 struct dentry
*dn
, *parent
;
1219 BUG_ON(!rinfo
->head
->is_target
);
1220 BUG_ON(req
->r_dentry
);
1222 parent
= d_find_any_alias(dir
);
1225 dname
.name
= rinfo
->dname
;
1226 dname
.len
= rinfo
->dname_len
;
1227 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1228 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1229 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1231 dn
= d_lookup(parent
, &dname
);
1232 dout("d_lookup on parent=%p name=%.*s got %p\n",
1233 parent
, dname
.len
, dname
.name
, dn
);
1236 dn
= d_alloc(parent
, &dname
);
1237 dout("d_alloc %p '%.*s' = %p\n", parent
,
1238 dname
.len
, dname
.name
, dn
);
1245 } else if (d_really_is_positive(dn
) &&
1246 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1247 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1248 dout(" dn %p points to wrong inode %p\n",
1250 ceph_dir_clear_ordered(dir
);
1261 if (rinfo
->head
->is_target
) {
1262 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1263 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1265 in
= ceph_get_inode(sb
, tvino
);
1270 req
->r_target_inode
= in
;
1272 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1273 session
, req
->r_request_started
,
1274 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1275 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1276 &req
->r_caps_reservation
);
1278 pr_err("fill_inode badness %p %llx.%llx\n",
1279 in
, ceph_vinop(in
));
1285 * ignore null lease/binding on snapdir ENOENT, or else we
1286 * will have trouble splicing in the virtual snapdir later
1288 if (rinfo
->head
->is_dentry
&&
1289 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1290 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1291 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1292 fsc
->mount_options
->snapdir_name
,
1293 req
->r_dentry
->d_name
.len
))) {
1295 * lookup link rename : null -> possibly existing inode
1296 * mknod symlink mkdir : null -> new inode
1297 * unlink : linked -> null
1299 struct inode
*dir
= req
->r_parent
;
1300 struct dentry
*dn
= req
->r_dentry
;
1301 bool have_dir_cap
, have_lease
;
1305 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1307 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1308 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1310 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1311 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1313 /* do we have a lease on the whole dir? */
1315 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1316 CEPH_CAP_FILE_SHARED
);
1318 /* do we have a dn lease? */
1319 have_lease
= have_dir_cap
||
1320 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1322 dout("fill_trace no dentry lease or dir cap\n");
1325 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1326 struct inode
*olddir
= req
->r_old_dentry_dir
;
1329 dout(" src %p '%pd' dst %p '%pd'\n",
1333 dout("fill_trace doing d_move %p -> %p\n",
1334 req
->r_old_dentry
, dn
);
1336 /* d_move screws up sibling dentries' offsets */
1337 ceph_dir_clear_ordered(dir
);
1338 ceph_dir_clear_ordered(olddir
);
1340 d_move(req
->r_old_dentry
, dn
);
1341 dout(" src %p '%pd' dst %p '%pd'\n",
1346 /* ensure target dentry is invalidated, despite
1347 rehashing bug in vfs_rename_dir */
1348 ceph_invalidate_dentry_lease(dn
);
1350 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1351 ceph_dentry(req
->r_old_dentry
)->offset
);
1353 dn
= req
->r_old_dentry
; /* use old_dentry */
1357 if (!rinfo
->head
->is_target
) {
1358 dout("fill_trace null dentry\n");
1359 if (d_really_is_positive(dn
)) {
1360 dout("d_delete %p\n", dn
);
1361 ceph_dir_clear_ordered(dir
);
1363 } else if (have_lease
) {
1366 update_dentry_lease(dn
, rinfo
->dlease
,
1368 req
->r_request_started
,
1374 /* attach proper inode */
1375 if (d_really_is_negative(dn
)) {
1376 ceph_dir_clear_ordered(dir
);
1378 dn
= splice_dentry(dn
, in
);
1383 req
->r_dentry
= dn
; /* may have spliced */
1384 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1385 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1386 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1393 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1394 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1395 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1396 req
->r_request_started
,
1399 dout(" final dn %p\n", dn
);
1400 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1401 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1402 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1403 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1404 struct dentry
*dn
= req
->r_dentry
;
1405 struct inode
*dir
= req
->r_parent
;
1407 /* fill out a snapdir LOOKUPSNAP dentry */
1410 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1411 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1412 ceph_dir_clear_ordered(dir
);
1414 dn
= splice_dentry(dn
, in
);
1419 req
->r_dentry
= dn
; /* may have spliced */
1420 } else if (rinfo
->head
->is_dentry
) {
1421 struct ceph_vino
*ptvino
= NULL
;
1423 if ((le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) ||
1424 le32_to_cpu(rinfo
->dlease
->duration_ms
)) {
1425 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1426 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1428 if (rinfo
->head
->is_target
) {
1429 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1430 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1434 update_dentry_lease(req
->r_dentry
, rinfo
->dlease
,
1435 session
, req
->r_request_started
, ptvino
,
1438 dout("%s: no dentry lease or dir cap\n", __func__
);
1442 dout("fill_trace done err=%d\n", err
);
1447 * Prepopulate our cache with readdir results, leases, etc.
1449 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1450 struct ceph_mds_session
*session
)
1452 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1455 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1456 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1457 struct ceph_vino vino
;
1461 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1462 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1464 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1467 dout("new_inode badness got %d\n", err
);
1470 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1471 req
->r_request_started
, -1,
1472 &req
->r_caps_reservation
);
1474 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1483 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1487 put_page(ctl
->page
);
1492 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1493 struct ceph_readdir_cache_control
*ctl
,
1494 struct ceph_mds_request
*req
)
1496 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1497 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1498 unsigned idx
= ctl
->index
% nsize
;
1499 pgoff_t pgoff
= ctl
->index
/ nsize
;
1501 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1502 ceph_readdir_cache_release(ctl
);
1504 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1506 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1509 return idx
== 0 ? -ENOMEM
: 0;
1511 /* reading/filling the cache are serialized by
1512 * i_mutex, no need to use page lock */
1513 unlock_page(ctl
->page
);
1514 ctl
->dentries
= kmap(ctl
->page
);
1516 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1519 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1520 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1521 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1522 ctl
->dentries
[idx
] = dn
;
1525 dout("disable readdir cache\n");
1531 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1532 struct ceph_mds_session
*session
)
1534 struct dentry
*parent
= req
->r_dentry
;
1535 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1536 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1540 int err
= 0, skipped
= 0, ret
, i
;
1541 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1542 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1545 struct ceph_readdir_cache_control cache_ctl
= {};
1547 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1548 return readdir_prepopulate_inodes_only(req
, session
);
1550 if (rinfo
->hash_order
) {
1552 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1554 strlen(req
->r_path2
));
1555 last_hash
= ceph_frag_value(last_hash
);
1556 } else if (rinfo
->offset_hash
) {
1557 /* mds understands offset_hash */
1558 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1559 last_hash
= le32_to_cpu(rhead
->args
.readdir
.offset_hash
);
1563 if (rinfo
->dir_dir
&&
1564 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1565 dout("readdir_prepopulate got new frag %x -> %x\n",
1566 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1567 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1568 if (!rinfo
->hash_order
)
1569 req
->r_readdir_offset
= 2;
1572 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1573 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1574 rinfo
->dir_nr
, parent
);
1576 dout("readdir_prepopulate %d items under dn %p\n",
1577 rinfo
->dir_nr
, parent
);
1579 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1581 if (ceph_frag_is_leftmost(frag
) &&
1582 req
->r_readdir_offset
== 2 &&
1583 !(rinfo
->hash_order
&& last_hash
)) {
1584 /* note dir version at start of readdir so we can
1585 * tell if any dentries get dropped */
1586 req
->r_dir_release_cnt
=
1587 atomic64_read(&ci
->i_release_count
);
1588 req
->r_dir_ordered_cnt
=
1589 atomic64_read(&ci
->i_ordered_count
);
1590 req
->r_readdir_cache_idx
= 0;
1594 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1595 fpos_offset
= req
->r_readdir_offset
;
1597 /* FIXME: release caps/leases if error occurs */
1598 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1599 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1600 struct ceph_vino tvino
, dvino
;
1602 dname
.name
= rde
->name
;
1603 dname
.len
= rde
->name_len
;
1604 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1606 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1607 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1609 if (rinfo
->hash_order
) {
1610 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1611 rde
->name
, rde
->name_len
);
1612 hash
= ceph_frag_value(hash
);
1613 if (hash
!= last_hash
)
1616 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1618 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1622 dn
= d_lookup(parent
, &dname
);
1623 dout("d_lookup on parent=%p name=%.*s got %p\n",
1624 parent
, dname
.len
, dname
.name
, dn
);
1627 dn
= d_alloc(parent
, &dname
);
1628 dout("d_alloc %p '%.*s' = %p\n", parent
,
1629 dname
.len
, dname
.name
, dn
);
1631 dout("d_alloc badness\n");
1635 } else if (d_really_is_positive(dn
) &&
1636 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1637 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1638 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1639 dout(" dn %p points to wrong inode %p\n",
1642 spin_lock(&dn
->d_lock
);
1643 if (di
->offset
> 0 &&
1644 di
->lease_shared_gen
==
1645 atomic_read(&ci
->i_shared_gen
)) {
1646 __ceph_dir_clear_ordered(ci
);
1649 spin_unlock(&dn
->d_lock
);
1657 if (d_really_is_positive(dn
)) {
1660 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1662 dout("new_inode badness\n");
1670 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1671 req
->r_request_started
, -1,
1672 &req
->r_caps_reservation
);
1674 pr_err("fill_inode badness on %p\n", in
);
1675 if (d_really_is_negative(dn
))
1682 if (d_really_is_negative(dn
)) {
1683 struct dentry
*realdn
;
1685 if (ceph_security_xattr_deadlock(in
)) {
1686 dout(" skip splicing dn %p to inode %p"
1687 " (security xattr deadlock)\n", dn
, in
);
1693 realdn
= splice_dentry(dn
, in
);
1694 if (IS_ERR(realdn
)) {
1695 err
= PTR_ERR(realdn
);
1702 ceph_dentry(dn
)->offset
= rde
->offset
;
1704 dvino
= ceph_vino(d_inode(parent
));
1705 update_dentry_lease(dn
, rde
->lease
, req
->r_session
,
1706 req
->r_request_started
, &tvino
, &dvino
);
1708 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1709 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1719 if (err
== 0 && skipped
== 0) {
1720 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1721 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1723 ceph_readdir_cache_release(&cache_ctl
);
1724 dout("readdir_prepopulate done\n");
1728 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1730 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1733 spin_lock(&ci
->i_ceph_lock
);
1734 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1735 i_size_write(inode
, size
);
1736 inode
->i_blocks
= calc_inode_blocks(size
);
1738 ret
= __ceph_should_report_size(ci
);
1740 spin_unlock(&ci
->i_ceph_lock
);
1745 * Write back inode data in a worker thread. (This can't be done
1746 * in the message handler context.)
1748 void ceph_queue_writeback(struct inode
*inode
)
1751 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1752 &ceph_inode(inode
)->i_wb_work
)) {
1753 dout("ceph_queue_writeback %p\n", inode
);
1755 dout("ceph_queue_writeback %p failed\n", inode
);
1760 static void ceph_writeback_work(struct work_struct
*work
)
1762 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1764 struct inode
*inode
= &ci
->vfs_inode
;
1766 dout("writeback %p\n", inode
);
1767 filemap_fdatawrite(&inode
->i_data
);
1772 * queue an async invalidation
1774 void ceph_queue_invalidate(struct inode
*inode
)
1777 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1778 &ceph_inode(inode
)->i_pg_inv_work
)) {
1779 dout("ceph_queue_invalidate %p\n", inode
);
1781 dout("ceph_queue_invalidate %p failed\n", inode
);
1787 * Invalidate inode pages in a worker thread. (This can't be done
1788 * in the message handler context.)
1790 static void ceph_invalidate_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
;
1795 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1799 mutex_lock(&ci
->i_truncate_mutex
);
1801 if (READ_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1802 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1803 inode
, ceph_ino(inode
));
1804 mapping_set_error(inode
->i_mapping
, -EIO
);
1805 truncate_pagecache(inode
, 0);
1806 mutex_unlock(&ci
->i_truncate_mutex
);
1810 spin_lock(&ci
->i_ceph_lock
);
1811 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1812 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1813 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1814 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1816 spin_unlock(&ci
->i_ceph_lock
);
1817 mutex_unlock(&ci
->i_truncate_mutex
);
1820 orig_gen
= ci
->i_rdcache_gen
;
1821 spin_unlock(&ci
->i_ceph_lock
);
1823 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1824 pr_err("invalidate_pages %p fails\n", inode
);
1827 spin_lock(&ci
->i_ceph_lock
);
1828 if (orig_gen
== ci
->i_rdcache_gen
&&
1829 orig_gen
== ci
->i_rdcache_revoking
) {
1830 dout("invalidate_pages %p gen %d successful\n", inode
,
1832 ci
->i_rdcache_revoking
--;
1835 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1836 inode
, orig_gen
, ci
->i_rdcache_gen
,
1837 ci
->i_rdcache_revoking
);
1838 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1841 spin_unlock(&ci
->i_ceph_lock
);
1842 mutex_unlock(&ci
->i_truncate_mutex
);
1845 ceph_check_caps(ci
, 0, NULL
);
1851 * called by trunc_wq;
1853 * We also truncate in a separate thread as well.
1855 static void ceph_vmtruncate_work(struct work_struct
*work
)
1857 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1859 struct inode
*inode
= &ci
->vfs_inode
;
1861 dout("vmtruncate_work %p\n", inode
);
1862 __ceph_do_pending_vmtruncate(inode
);
1867 * Queue an async vmtruncate. If we fail to queue work, we will handle
1868 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1870 void ceph_queue_vmtruncate(struct inode
*inode
)
1872 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1876 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1877 &ci
->i_vmtruncate_work
)) {
1878 dout("ceph_queue_vmtruncate %p\n", inode
);
1880 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1881 inode
, ci
->i_truncate_pending
);
1887 * Make sure any pending truncation is applied before doing anything
1888 * that may depend on it.
1890 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1892 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1894 int wrbuffer_refs
, finish
= 0;
1896 mutex_lock(&ci
->i_truncate_mutex
);
1898 spin_lock(&ci
->i_ceph_lock
);
1899 if (ci
->i_truncate_pending
== 0) {
1900 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1901 spin_unlock(&ci
->i_ceph_lock
);
1902 mutex_unlock(&ci
->i_truncate_mutex
);
1907 * make sure any dirty snapped pages are flushed before we
1908 * possibly truncate them.. so write AND block!
1910 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1911 spin_unlock(&ci
->i_ceph_lock
);
1912 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1914 filemap_write_and_wait_range(&inode
->i_data
, 0,
1915 inode
->i_sb
->s_maxbytes
);
1919 /* there should be no reader or writer */
1920 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1922 to
= ci
->i_truncate_size
;
1923 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1924 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1925 ci
->i_truncate_pending
, to
);
1926 spin_unlock(&ci
->i_ceph_lock
);
1928 truncate_pagecache(inode
, to
);
1930 spin_lock(&ci
->i_ceph_lock
);
1931 if (to
== ci
->i_truncate_size
) {
1932 ci
->i_truncate_pending
= 0;
1935 spin_unlock(&ci
->i_ceph_lock
);
1939 mutex_unlock(&ci
->i_truncate_mutex
);
1941 if (wrbuffer_refs
== 0)
1942 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1944 wake_up_all(&ci
->i_cap_wq
);
1950 static const struct inode_operations ceph_symlink_iops
= {
1951 .get_link
= simple_get_link
,
1952 .setattr
= ceph_setattr
,
1953 .getattr
= ceph_getattr
,
1954 .listxattr
= ceph_listxattr
,
1957 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
1959 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1960 const unsigned int ia_valid
= attr
->ia_valid
;
1961 struct ceph_mds_request
*req
;
1962 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1963 struct ceph_cap_flush
*prealloc_cf
;
1965 int release
= 0, dirtied
= 0;
1968 int inode_dirty_flags
= 0;
1969 bool lock_snap_rwsem
= false;
1971 prealloc_cf
= ceph_alloc_cap_flush();
1975 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1978 ceph_free_cap_flush(prealloc_cf
);
1979 return PTR_ERR(req
);
1982 spin_lock(&ci
->i_ceph_lock
);
1983 issued
= __ceph_caps_issued(ci
, NULL
);
1985 if (!ci
->i_head_snapc
&&
1986 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
1987 lock_snap_rwsem
= true;
1988 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
1989 spin_unlock(&ci
->i_ceph_lock
);
1990 down_read(&mdsc
->snap_rwsem
);
1991 spin_lock(&ci
->i_ceph_lock
);
1992 issued
= __ceph_caps_issued(ci
, NULL
);
1996 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1998 if (ia_valid
& ATTR_UID
) {
1999 dout("setattr %p uid %d -> %d\n", inode
,
2000 from_kuid(&init_user_ns
, inode
->i_uid
),
2001 from_kuid(&init_user_ns
, attr
->ia_uid
));
2002 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2003 inode
->i_uid
= attr
->ia_uid
;
2004 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2005 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2006 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
2007 req
->r_args
.setattr
.uid
= cpu_to_le32(
2008 from_kuid(&init_user_ns
, attr
->ia_uid
));
2009 mask
|= CEPH_SETATTR_UID
;
2010 release
|= CEPH_CAP_AUTH_SHARED
;
2013 if (ia_valid
& ATTR_GID
) {
2014 dout("setattr %p gid %d -> %d\n", inode
,
2015 from_kgid(&init_user_ns
, inode
->i_gid
),
2016 from_kgid(&init_user_ns
, attr
->ia_gid
));
2017 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2018 inode
->i_gid
= attr
->ia_gid
;
2019 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2020 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2021 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
2022 req
->r_args
.setattr
.gid
= cpu_to_le32(
2023 from_kgid(&init_user_ns
, attr
->ia_gid
));
2024 mask
|= CEPH_SETATTR_GID
;
2025 release
|= CEPH_CAP_AUTH_SHARED
;
2028 if (ia_valid
& ATTR_MODE
) {
2029 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
2031 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2032 inode
->i_mode
= attr
->ia_mode
;
2033 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2034 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2035 attr
->ia_mode
!= inode
->i_mode
) {
2036 inode
->i_mode
= attr
->ia_mode
;
2037 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
2038 mask
|= CEPH_SETATTR_MODE
;
2039 release
|= CEPH_CAP_AUTH_SHARED
;
2043 if (ia_valid
& ATTR_ATIME
) {
2044 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode
,
2045 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
2046 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
2047 if (issued
& CEPH_CAP_FILE_EXCL
) {
2048 ci
->i_time_warp_seq
++;
2049 inode
->i_atime
= attr
->ia_atime
;
2050 dirtied
|= CEPH_CAP_FILE_EXCL
;
2051 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2052 timespec64_compare(&inode
->i_atime
,
2053 &attr
->ia_atime
) < 0) {
2054 inode
->i_atime
= attr
->ia_atime
;
2055 dirtied
|= CEPH_CAP_FILE_WR
;
2056 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2057 !timespec64_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2058 ceph_encode_timespec64(&req
->r_args
.setattr
.atime
,
2060 mask
|= CEPH_SETATTR_ATIME
;
2061 release
|= CEPH_CAP_FILE_SHARED
|
2062 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2065 if (ia_valid
& ATTR_MTIME
) {
2066 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode
,
2067 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2068 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2069 if (issued
& CEPH_CAP_FILE_EXCL
) {
2070 ci
->i_time_warp_seq
++;
2071 inode
->i_mtime
= attr
->ia_mtime
;
2072 dirtied
|= CEPH_CAP_FILE_EXCL
;
2073 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2074 timespec64_compare(&inode
->i_mtime
,
2075 &attr
->ia_mtime
) < 0) {
2076 inode
->i_mtime
= attr
->ia_mtime
;
2077 dirtied
|= CEPH_CAP_FILE_WR
;
2078 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2079 !timespec64_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2080 ceph_encode_timespec64(&req
->r_args
.setattr
.mtime
,
2082 mask
|= CEPH_SETATTR_MTIME
;
2083 release
|= CEPH_CAP_FILE_SHARED
|
2084 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2087 if (ia_valid
& ATTR_SIZE
) {
2088 dout("setattr %p size %lld -> %lld\n", inode
,
2089 inode
->i_size
, attr
->ia_size
);
2090 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
2091 attr
->ia_size
> inode
->i_size
) {
2092 i_size_write(inode
, attr
->ia_size
);
2093 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2094 ci
->i_reported_size
= attr
->ia_size
;
2095 dirtied
|= CEPH_CAP_FILE_EXCL
;
2096 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2097 attr
->ia_size
!= inode
->i_size
) {
2098 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2099 req
->r_args
.setattr
.old_size
=
2100 cpu_to_le64(inode
->i_size
);
2101 mask
|= CEPH_SETATTR_SIZE
;
2102 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_EXCL
|
2103 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2107 /* these do nothing */
2108 if (ia_valid
& ATTR_CTIME
) {
2109 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2110 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2111 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode
,
2112 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2113 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2114 only
? "ctime only" : "ignored");
2117 * if kernel wants to dirty ctime but nothing else,
2118 * we need to choose a cap to dirty under, or do
2119 * a almost-no-op setattr
2121 if (issued
& CEPH_CAP_AUTH_EXCL
)
2122 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2123 else if (issued
& CEPH_CAP_FILE_EXCL
)
2124 dirtied
|= CEPH_CAP_FILE_EXCL
;
2125 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2126 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2128 mask
|= CEPH_SETATTR_CTIME
;
2131 if (ia_valid
& ATTR_FILE
)
2132 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2135 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2137 inode
->i_ctime
= attr
->ia_ctime
;
2141 spin_unlock(&ci
->i_ceph_lock
);
2142 if (lock_snap_rwsem
)
2143 up_read(&mdsc
->snap_rwsem
);
2145 if (inode_dirty_flags
)
2146 __mark_inode_dirty(inode
, inode_dirty_flags
);
2150 req
->r_inode
= inode
;
2152 req
->r_inode_drop
= release
;
2153 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2154 req
->r_num_caps
= 1;
2155 req
->r_stamp
= attr
->ia_ctime
;
2156 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2158 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2159 ceph_cap_string(dirtied
), mask
);
2161 ceph_mdsc_put_request(req
);
2162 ceph_free_cap_flush(prealloc_cf
);
2164 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2165 __ceph_do_pending_vmtruncate(inode
);
2173 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2175 struct inode
*inode
= d_inode(dentry
);
2176 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
2179 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2182 err
= setattr_prepare(dentry
, attr
);
2186 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2187 attr
->ia_size
> max(inode
->i_size
, fsc
->max_file_size
))
2190 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2191 ceph_quota_is_max_bytes_exceeded(inode
, attr
->ia_size
))
2194 err
= __ceph_setattr(inode
, attr
);
2196 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2197 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2203 * Verify that we have a lease on the given mask. If not,
2204 * do a getattr against an mds.
2206 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2207 int mask
, bool force
)
2209 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2210 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2211 struct ceph_mds_request
*req
;
2215 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2216 dout("do_getattr inode %p SNAPDIR\n", inode
);
2220 dout("do_getattr inode %p mask %s mode 0%o\n",
2221 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2222 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2225 mode
= (mask
& CEPH_STAT_RSTAT
) ? USE_AUTH_MDS
: USE_ANY_MDS
;
2226 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, mode
);
2228 return PTR_ERR(req
);
2229 req
->r_inode
= inode
;
2231 req
->r_num_caps
= 1;
2232 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2233 req
->r_locked_page
= locked_page
;
2234 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2235 if (locked_page
&& err
== 0) {
2236 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2237 if (inline_version
== 0) {
2238 /* the reply is supposed to contain inline data */
2240 } else if (inline_version
== CEPH_INLINE_NONE
) {
2243 err
= req
->r_reply_info
.targeti
.inline_len
;
2246 ceph_mdsc_put_request(req
);
2247 dout("do_getattr result=%d\n", err
);
2253 * Check inode permissions. We verify we have a valid value for
2254 * the AUTH cap, then call the generic handler.
2256 int ceph_permission(struct inode
*inode
, int mask
)
2260 if (mask
& MAY_NOT_BLOCK
)
2263 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2266 err
= generic_permission(inode
, mask
);
2271 * Get all attributes. Hopefully somedata we'll have a statlite()
2272 * and can limit the fields we require to be accurate.
2274 int ceph_getattr(const struct path
*path
, struct kstat
*stat
,
2275 u32 request_mask
, unsigned int flags
)
2277 struct inode
*inode
= d_inode(path
->dentry
);
2278 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2281 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
, false);
2283 generic_fillattr(inode
, stat
);
2284 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2285 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2286 stat
->dev
= ceph_snap(inode
);
2289 if (S_ISDIR(inode
->i_mode
)) {
2290 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2292 stat
->size
= ci
->i_rbytes
;
2294 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2296 stat
->blksize
= 65536;
2298 * Some applications rely on the number of st_nlink
2299 * value on directories to be either 0 (if unlinked)
2300 * or 2 + number of subdirectories.
2302 if (stat
->nlink
== 1)
2303 /* '.' + '..' + subdirs */
2304 stat
->nlink
= 1 + 1 + ci
->i_subdirs
;