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 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
45 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
46 inode_set_iversion_raw(inode
, 0);
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
, 1);
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_work
, ceph_inode_work
);
514 memset(&ci
->i_btime
, '\0', sizeof(ci
->i_btime
));
516 ceph_fscache_inode_init(ci
);
518 return &ci
->vfs_inode
;
521 void ceph_free_inode(struct inode
*inode
)
523 struct ceph_inode_info
*ci
= ceph_inode(inode
);
525 kfree(ci
->i_symlink
);
526 kmem_cache_free(ceph_inode_cachep
, ci
);
529 void ceph_evict_inode(struct inode
*inode
)
531 struct ceph_inode_info
*ci
= ceph_inode(inode
);
532 struct ceph_inode_frag
*frag
;
535 dout("evict_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
537 truncate_inode_pages_final(&inode
->i_data
);
540 ceph_fscache_unregister_inode_cookie(ci
);
542 __ceph_remove_caps(ci
);
544 if (__ceph_has_any_quota(ci
))
545 ceph_adjust_quota_realms_count(inode
, false);
548 * we may still have a snap_realm reference if there are stray
549 * caps in i_snap_caps.
551 if (ci
->i_snap_realm
) {
552 struct ceph_mds_client
*mdsc
=
553 ceph_inode_to_client(inode
)->mdsc
;
554 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
555 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
556 dout(" dropping residual ref to snap realm %p\n",
558 spin_lock(&realm
->inodes_with_caps_lock
);
559 list_del_init(&ci
->i_snap_realm_item
);
560 ci
->i_snap_realm
= NULL
;
561 if (realm
->ino
== ci
->i_vino
.ino
)
563 spin_unlock(&realm
->inodes_with_caps_lock
);
564 ceph_put_snap_realm(mdsc
, realm
);
566 ceph_put_snapid_map(mdsc
, ci
->i_snapid_map
);
567 ci
->i_snap_realm
= NULL
;
571 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
572 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
573 rb_erase(n
, &ci
->i_fragtree
);
576 ci
->i_fragtree_nsplits
= 0;
578 __ceph_destroy_xattrs(ci
);
579 if (ci
->i_xattrs
.blob
)
580 ceph_buffer_put(ci
->i_xattrs
.blob
);
581 if (ci
->i_xattrs
.prealloc_blob
)
582 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
584 ceph_put_string(rcu_dereference_raw(ci
->i_layout
.pool_ns
));
587 static inline blkcnt_t
calc_inode_blocks(u64 size
)
589 return (size
+ (1<<9) - 1) >> 9;
593 * Helpers to fill in size, ctime, mtime, and atime. We have to be
594 * careful because either the client or MDS may have more up to date
595 * info, depending on which capabilities are held, and whether
596 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
597 * and size are monotonically increasing, except when utimes() or
598 * truncate() increments the corresponding _seq values.)
600 int ceph_fill_file_size(struct inode
*inode
, int issued
,
601 u32 truncate_seq
, u64 truncate_size
, u64 size
)
603 struct ceph_inode_info
*ci
= ceph_inode(inode
);
606 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
607 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
608 dout("size %lld -> %llu\n", inode
->i_size
, size
);
609 if (size
> 0 && S_ISDIR(inode
->i_mode
)) {
610 pr_err("fill_file_size non-zero size for directory\n");
613 i_size_write(inode
, size
);
614 inode
->i_blocks
= calc_inode_blocks(size
);
615 ci
->i_reported_size
= size
;
616 if (truncate_seq
!= ci
->i_truncate_seq
) {
617 dout("truncate_seq %u -> %u\n",
618 ci
->i_truncate_seq
, truncate_seq
);
619 ci
->i_truncate_seq
= truncate_seq
;
621 /* the MDS should have revoked these caps */
622 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
625 CEPH_CAP_FILE_LAZYIO
));
627 * If we hold relevant caps, or in the case where we're
628 * not the only client referencing this file and we
629 * don't hold those caps, then we need to check whether
630 * the file is either opened or mmaped
632 if ((issued
& (CEPH_CAP_FILE_CACHE
|
633 CEPH_CAP_FILE_BUFFER
)) ||
634 mapping_mapped(inode
->i_mapping
) ||
635 __ceph_caps_file_wanted(ci
)) {
636 ci
->i_truncate_pending
++;
641 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
642 ci
->i_truncate_size
!= truncate_size
) {
643 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
645 ci
->i_truncate_size
= truncate_size
;
649 ceph_fscache_invalidate(inode
);
654 void ceph_fill_file_time(struct inode
*inode
, int issued
,
655 u64 time_warp_seq
, struct timespec64
*ctime
,
656 struct timespec64
*mtime
, struct timespec64
*atime
)
658 struct ceph_inode_info
*ci
= ceph_inode(inode
);
661 if (issued
& (CEPH_CAP_FILE_EXCL
|
663 CEPH_CAP_FILE_BUFFER
|
665 CEPH_CAP_XATTR_EXCL
)) {
666 if (ci
->i_version
== 0 ||
667 timespec64_compare(ctime
, &inode
->i_ctime
) > 0) {
668 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
669 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
670 ctime
->tv_sec
, ctime
->tv_nsec
);
671 inode
->i_ctime
= *ctime
;
673 if (ci
->i_version
== 0 ||
674 ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
675 /* the MDS did a utimes() */
676 dout("mtime %lld.%09ld -> %lld.%09ld "
678 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
679 mtime
->tv_sec
, mtime
->tv_nsec
,
680 ci
->i_time_warp_seq
, (int)time_warp_seq
);
682 inode
->i_mtime
= *mtime
;
683 inode
->i_atime
= *atime
;
684 ci
->i_time_warp_seq
= time_warp_seq
;
685 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
686 /* nobody did utimes(); take the max */
687 if (timespec64_compare(mtime
, &inode
->i_mtime
) > 0) {
688 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
689 inode
->i_mtime
.tv_sec
,
690 inode
->i_mtime
.tv_nsec
,
691 mtime
->tv_sec
, mtime
->tv_nsec
);
692 inode
->i_mtime
= *mtime
;
694 if (timespec64_compare(atime
, &inode
->i_atime
) > 0) {
695 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
696 inode
->i_atime
.tv_sec
,
697 inode
->i_atime
.tv_nsec
,
698 atime
->tv_sec
, atime
->tv_nsec
);
699 inode
->i_atime
= *atime
;
701 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
702 /* we did a utimes(); ignore mds values */
707 /* we have no write|excl caps; whatever the MDS says is true */
708 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
709 inode
->i_ctime
= *ctime
;
710 inode
->i_mtime
= *mtime
;
711 inode
->i_atime
= *atime
;
712 ci
->i_time_warp_seq
= time_warp_seq
;
717 if (warn
) /* time_warp_seq shouldn't go backwards */
718 dout("%p mds time_warp_seq %llu < %u\n",
719 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
723 * Populate an inode based on info from mds. May be called on new or
726 static int fill_inode(struct inode
*inode
, struct page
*locked_page
,
727 struct ceph_mds_reply_info_in
*iinfo
,
728 struct ceph_mds_reply_dirfrag
*dirinfo
,
729 struct ceph_mds_session
*session
,
730 unsigned long ttl_from
, int cap_fmode
,
731 struct ceph_cap_reservation
*caps_reservation
)
733 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
734 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
735 struct ceph_inode_info
*ci
= ceph_inode(inode
);
736 int issued
, new_issued
, info_caps
;
737 struct timespec64 mtime
, atime
, ctime
;
738 struct ceph_buffer
*xattr_blob
= NULL
;
739 struct ceph_buffer
*old_blob
= NULL
;
740 struct ceph_string
*pool_ns
= NULL
;
741 struct ceph_cap
*new_cap
= NULL
;
744 bool queue_trunc
= false;
745 bool new_version
= false;
746 bool fill_inline
= false;
748 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
749 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
752 info_caps
= le32_to_cpu(info
->cap
.caps
);
754 /* prealloc new cap struct */
755 if (info_caps
&& ceph_snap(inode
) == CEPH_NOSNAP
)
756 new_cap
= ceph_get_cap(mdsc
, caps_reservation
);
759 * prealloc xattr data, if it looks like we'll need it. only
760 * if len > 4 (meaning there are actually xattrs; the first 4
761 * bytes are the xattr count).
763 if (iinfo
->xattr_len
> 4) {
764 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
766 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
770 if (iinfo
->pool_ns_len
> 0)
771 pool_ns
= ceph_find_or_create_string(iinfo
->pool_ns_data
,
774 if (ceph_snap(inode
) != CEPH_NOSNAP
&& !ci
->i_snapid_map
)
775 ci
->i_snapid_map
= ceph_get_snapid_map(mdsc
, ceph_snap(inode
));
777 spin_lock(&ci
->i_ceph_lock
);
780 * provided version will be odd if inode value is projected,
781 * even if stable. skip the update if we have newer stable
782 * info (ours>=theirs, e.g. due to racing mds replies), unless
783 * we are getting projected (unstable) info (in which case the
784 * version is odd, and we want ours>theirs).
790 if (ci
->i_version
== 0 ||
791 ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
792 le64_to_cpu(info
->version
) > (ci
->i_version
& ~1)))
795 /* Update change_attribute */
796 inode_set_max_iversion_raw(inode
, iinfo
->change_attr
);
798 __ceph_caps_issued(ci
, &issued
);
799 issued
|= __ceph_caps_dirty(ci
);
800 new_issued
= ~issued
& info_caps
;
803 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
804 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
806 __ceph_update_quota(ci
, iinfo
->max_bytes
, iinfo
->max_files
);
808 if ((new_version
|| (new_issued
& CEPH_CAP_AUTH_SHARED
)) &&
809 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
810 inode
->i_mode
= le32_to_cpu(info
->mode
);
811 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
812 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
813 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
814 from_kuid(&init_user_ns
, inode
->i_uid
),
815 from_kgid(&init_user_ns
, inode
->i_gid
));
816 ceph_decode_timespec64(&ci
->i_btime
, &iinfo
->btime
);
817 ceph_decode_timespec64(&ci
->i_snap_btime
, &iinfo
->snap_btime
);
820 if ((new_version
|| (new_issued
& CEPH_CAP_LINK_SHARED
)) &&
821 (issued
& CEPH_CAP_LINK_EXCL
) == 0)
822 set_nlink(inode
, le32_to_cpu(info
->nlink
));
824 if (new_version
|| (new_issued
& CEPH_CAP_ANY_RD
)) {
825 /* be careful with mtime, atime, size */
826 ceph_decode_timespec64(&atime
, &info
->atime
);
827 ceph_decode_timespec64(&mtime
, &info
->mtime
);
828 ceph_decode_timespec64(&ctime
, &info
->ctime
);
829 ceph_fill_file_time(inode
, issued
,
830 le32_to_cpu(info
->time_warp_seq
),
831 &ctime
, &mtime
, &atime
);
834 if (new_version
|| (info_caps
& CEPH_CAP_FILE_SHARED
)) {
835 ci
->i_files
= le64_to_cpu(info
->files
);
836 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
840 (new_issued
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
))) {
841 s64 old_pool
= ci
->i_layout
.pool_id
;
842 struct ceph_string
*old_ns
;
844 ceph_file_layout_from_legacy(&ci
->i_layout
, &info
->layout
);
845 old_ns
= rcu_dereference_protected(ci
->i_layout
.pool_ns
,
846 lockdep_is_held(&ci
->i_ceph_lock
));
847 rcu_assign_pointer(ci
->i_layout
.pool_ns
, pool_ns
);
849 if (ci
->i_layout
.pool_id
!= old_pool
|| pool_ns
!= old_ns
)
850 ci
->i_ceph_flags
&= ~CEPH_I_POOL_PERM
;
854 queue_trunc
= ceph_fill_file_size(inode
, issued
,
855 le32_to_cpu(info
->truncate_seq
),
856 le64_to_cpu(info
->truncate_size
),
857 le64_to_cpu(info
->size
));
858 /* only update max_size on auth cap */
859 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
860 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
861 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
862 le64_to_cpu(info
->max_size
));
863 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
867 /* layout and rstat are not tracked by capability, update them if
868 * the inode info is from auth mds */
869 if (new_version
|| (info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
)) {
870 if (S_ISDIR(inode
->i_mode
)) {
871 ci
->i_dir_layout
= iinfo
->dir_layout
;
872 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
873 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
874 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
875 ci
->i_dir_pin
= iinfo
->dir_pin
;
876 ceph_decode_timespec64(&ci
->i_rctime
, &info
->rctime
);
881 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
882 if ((ci
->i_xattrs
.version
== 0 || !(issued
& CEPH_CAP_XATTR_EXCL
)) &&
883 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
884 if (ci
->i_xattrs
.blob
)
885 old_blob
= ci
->i_xattrs
.blob
;
886 ci
->i_xattrs
.blob
= xattr_blob
;
888 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
889 iinfo
->xattr_data
, iinfo
->xattr_len
);
890 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
891 ceph_forget_all_cached_acls(inode
);
892 ceph_security_invalidate_secctx(inode
);
896 /* finally update i_version */
897 if (le64_to_cpu(info
->version
) > ci
->i_version
)
898 ci
->i_version
= le64_to_cpu(info
->version
);
900 inode
->i_mapping
->a_ops
= &ceph_aops
;
902 switch (inode
->i_mode
& S_IFMT
) {
907 inode
->i_blkbits
= PAGE_SHIFT
;
908 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
909 inode
->i_op
= &ceph_file_iops
;
912 inode
->i_op
= &ceph_file_iops
;
913 inode
->i_fop
= &ceph_file_fops
;
916 inode
->i_op
= &ceph_symlink_iops
;
917 if (!ci
->i_symlink
) {
918 u32 symlen
= iinfo
->symlink_len
;
921 spin_unlock(&ci
->i_ceph_lock
);
923 if (symlen
!= i_size_read(inode
)) {
924 pr_err("fill_inode %llx.%llx BAD symlink "
925 "size %lld\n", ceph_vinop(inode
),
927 i_size_write(inode
, symlen
);
928 inode
->i_blocks
= calc_inode_blocks(symlen
);
932 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
936 spin_lock(&ci
->i_ceph_lock
);
940 kfree(sym
); /* lost a race */
942 inode
->i_link
= ci
->i_symlink
;
945 inode
->i_op
= &ceph_dir_iops
;
946 inode
->i_fop
= &ceph_dir_fops
;
949 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
950 ceph_vinop(inode
), inode
->i_mode
);
953 /* were we issued a capability? */
955 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
956 ceph_add_cap(inode
, session
,
957 le64_to_cpu(info
->cap
.cap_id
),
958 cap_fmode
, info_caps
,
959 le32_to_cpu(info
->cap
.wanted
),
960 le32_to_cpu(info
->cap
.seq
),
961 le32_to_cpu(info
->cap
.mseq
),
962 le64_to_cpu(info
->cap
.realm
),
963 info
->cap
.flags
, &new_cap
);
965 /* set dir completion flag? */
966 if (S_ISDIR(inode
->i_mode
) &&
967 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
968 (info_caps
& CEPH_CAP_FILE_SHARED
) &&
969 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
970 !__ceph_dir_is_complete(ci
)) {
971 dout(" marking %p complete (empty)\n", inode
);
972 i_size_write(inode
, 0);
973 __ceph_dir_set_complete(ci
,
974 atomic64_read(&ci
->i_release_count
),
975 atomic64_read(&ci
->i_ordered_count
));
980 dout(" %p got snap_caps %s\n", inode
,
981 ceph_cap_string(info_caps
));
982 ci
->i_snap_caps
|= info_caps
;
984 __ceph_get_fmode(ci
, cap_fmode
);
986 } else if (cap_fmode
>= 0) {
987 pr_warn("mds issued no caps on %llx.%llx\n",
989 __ceph_get_fmode(ci
, cap_fmode
);
992 if (iinfo
->inline_version
> 0 &&
993 iinfo
->inline_version
>= ci
->i_inline_version
) {
994 int cache_caps
= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_LAZYIO
;
995 ci
->i_inline_version
= iinfo
->inline_version
;
996 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
997 (locked_page
|| (info_caps
& cache_caps
)))
1001 spin_unlock(&ci
->i_ceph_lock
);
1004 ceph_fill_inline_data(inode
, locked_page
,
1005 iinfo
->inline_data
, iinfo
->inline_len
);
1008 wake_up_all(&ci
->i_cap_wq
);
1010 /* queue truncate if we saw i_size decrease */
1012 ceph_queue_vmtruncate(inode
);
1014 /* populate frag tree */
1015 if (S_ISDIR(inode
->i_mode
))
1016 ceph_fill_fragtree(inode
, &info
->fragtree
, dirinfo
);
1018 /* update delegation info? */
1020 ceph_fill_dirfrag(inode
, dirinfo
);
1025 ceph_put_cap(mdsc
, new_cap
);
1026 ceph_buffer_put(old_blob
);
1027 ceph_buffer_put(xattr_blob
);
1028 ceph_put_string(pool_ns
);
1033 * caller should hold session s_mutex and dentry->d_lock.
1035 static void __update_dentry_lease(struct inode
*dir
, struct dentry
*dentry
,
1036 struct ceph_mds_reply_lease
*lease
,
1037 struct ceph_mds_session
*session
,
1038 unsigned long from_time
,
1039 struct ceph_mds_session
**old_lease_session
)
1041 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
1042 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
1043 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
1044 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
1046 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1047 dentry
, duration
, ttl
);
1049 /* only track leases on regular dentries */
1050 if (ceph_snap(dir
) != CEPH_NOSNAP
)
1053 di
->lease_shared_gen
= atomic_read(&ceph_inode(dir
)->i_shared_gen
);
1054 if (duration
== 0) {
1055 __ceph_dentry_dir_lease_touch(di
);
1059 if (di
->lease_gen
== session
->s_cap_gen
&&
1060 time_before(ttl
, di
->time
))
1061 return; /* we already have a newer lease. */
1063 if (di
->lease_session
&& di
->lease_session
!= session
) {
1064 *old_lease_session
= di
->lease_session
;
1065 di
->lease_session
= NULL
;
1068 if (!di
->lease_session
)
1069 di
->lease_session
= ceph_get_mds_session(session
);
1070 di
->lease_gen
= session
->s_cap_gen
;
1071 di
->lease_seq
= le32_to_cpu(lease
->seq
);
1072 di
->lease_renew_after
= half_ttl
;
1073 di
->lease_renew_from
= 0;
1076 __ceph_dentry_lease_touch(di
);
1079 static inline void update_dentry_lease(struct inode
*dir
, struct dentry
*dentry
,
1080 struct ceph_mds_reply_lease
*lease
,
1081 struct ceph_mds_session
*session
,
1082 unsigned long from_time
)
1084 struct ceph_mds_session
*old_lease_session
= NULL
;
1085 spin_lock(&dentry
->d_lock
);
1086 __update_dentry_lease(dir
, dentry
, lease
, session
, from_time
,
1087 &old_lease_session
);
1088 spin_unlock(&dentry
->d_lock
);
1089 if (old_lease_session
)
1090 ceph_put_mds_session(old_lease_session
);
1094 * update dentry lease without having parent inode locked
1096 static void update_dentry_lease_careful(struct dentry
*dentry
,
1097 struct ceph_mds_reply_lease
*lease
,
1098 struct ceph_mds_session
*session
,
1099 unsigned long from_time
,
1100 char *dname
, u32 dname_len
,
1101 struct ceph_vino
*pdvino
,
1102 struct ceph_vino
*ptvino
)
1106 struct ceph_mds_session
*old_lease_session
= NULL
;
1108 spin_lock(&dentry
->d_lock
);
1109 /* make sure dentry's name matches target */
1110 if (dentry
->d_name
.len
!= dname_len
||
1111 memcmp(dentry
->d_name
.name
, dname
, dname_len
))
1114 dir
= d_inode(dentry
->d_parent
);
1115 /* make sure parent matches dvino */
1116 if (!ceph_ino_compare(dir
, pdvino
))
1119 /* make sure dentry's inode matches target. NULL ptvino means that
1120 * we expect a negative dentry */
1122 if (d_really_is_negative(dentry
))
1124 if (!ceph_ino_compare(d_inode(dentry
), ptvino
))
1127 if (d_really_is_positive(dentry
))
1131 __update_dentry_lease(dir
, dentry
, lease
, session
,
1132 from_time
, &old_lease_session
);
1134 spin_unlock(&dentry
->d_lock
);
1135 if (old_lease_session
)
1136 ceph_put_mds_session(old_lease_session
);
1140 * splice a dentry to an inode.
1141 * caller must hold directory i_mutex for this to be safe.
1143 static int splice_dentry(struct dentry
**pdn
, struct inode
*in
)
1145 struct dentry
*dn
= *pdn
;
1146 struct dentry
*realdn
;
1148 BUG_ON(d_inode(dn
));
1150 if (S_ISDIR(in
->i_mode
)) {
1151 /* If inode is directory, d_splice_alias() below will remove
1152 * 'realdn' from its origin parent. We need to ensure that
1153 * origin parent's readdir cache will not reference 'realdn'
1155 realdn
= d_find_any_alias(in
);
1157 struct ceph_dentry_info
*di
= ceph_dentry(realdn
);
1158 spin_lock(&realdn
->d_lock
);
1160 realdn
->d_op
->d_prune(realdn
);
1163 di
->lease_shared_gen
= 0;
1166 spin_unlock(&realdn
->d_lock
);
1171 /* dn must be unhashed */
1172 if (!d_unhashed(dn
))
1174 realdn
= d_splice_alias(in
, dn
);
1175 if (IS_ERR(realdn
)) {
1176 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1177 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
1178 return PTR_ERR(realdn
);
1182 dout("dn %p (%d) spliced with %p (%d) "
1183 "inode %p ino %llx.%llx\n",
1185 realdn
, d_count(realdn
),
1186 d_inode(realdn
), ceph_vinop(d_inode(realdn
)));
1190 BUG_ON(!ceph_dentry(dn
));
1191 dout("dn %p attached to %p ino %llx.%llx\n",
1192 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)));
1198 * Incorporate results into the local cache. This is either just
1199 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1202 * A reply may contain
1203 * a directory inode along with a dentry.
1204 * and/or a target inode
1206 * Called with snap_rwsem (read).
1208 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
)
1210 struct ceph_mds_session
*session
= req
->r_session
;
1211 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1212 struct inode
*in
= NULL
;
1213 struct ceph_vino tvino
, dvino
;
1214 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
1217 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
1218 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1220 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1221 dout("fill_trace reply is empty!\n");
1222 if (rinfo
->head
->result
== 0 && req
->r_parent
)
1223 ceph_invalidate_dir_request(req
);
1227 if (rinfo
->head
->is_dentry
) {
1228 struct inode
*dir
= req
->r_parent
;
1231 err
= fill_inode(dir
, NULL
,
1232 &rinfo
->diri
, rinfo
->dirfrag
,
1233 session
, req
->r_request_started
, -1,
1234 &req
->r_caps_reservation
);
1241 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
&&
1242 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1243 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1245 struct dentry
*dn
, *parent
;
1247 BUG_ON(!rinfo
->head
->is_target
);
1248 BUG_ON(req
->r_dentry
);
1250 parent
= d_find_any_alias(dir
);
1253 dname
.name
= rinfo
->dname
;
1254 dname
.len
= rinfo
->dname_len
;
1255 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1256 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1257 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1259 dn
= d_lookup(parent
, &dname
);
1260 dout("d_lookup on parent=%p name=%.*s got %p\n",
1261 parent
, dname
.len
, dname
.name
, dn
);
1264 dn
= d_alloc(parent
, &dname
);
1265 dout("d_alloc %p '%.*s' = %p\n", parent
,
1266 dname
.len
, dname
.name
, dn
);
1273 } else if (d_really_is_positive(dn
) &&
1274 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1275 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1276 dout(" dn %p points to wrong inode %p\n",
1278 ceph_dir_clear_ordered(dir
);
1289 if (rinfo
->head
->is_target
) {
1290 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1291 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1293 in
= ceph_get_inode(sb
, tvino
);
1298 req
->r_target_inode
= in
;
1300 err
= fill_inode(in
, req
->r_locked_page
, &rinfo
->targeti
, NULL
,
1301 session
, req
->r_request_started
,
1302 (!test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1303 rinfo
->head
->result
== 0) ? req
->r_fmode
: -1,
1304 &req
->r_caps_reservation
);
1306 pr_err("fill_inode badness %p %llx.%llx\n",
1307 in
, ceph_vinop(in
));
1313 * ignore null lease/binding on snapdir ENOENT, or else we
1314 * will have trouble splicing in the virtual snapdir later
1316 if (rinfo
->head
->is_dentry
&&
1317 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
) &&
1318 test_bit(CEPH_MDS_R_PARENT_LOCKED
, &req
->r_req_flags
) &&
1319 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1320 fsc
->mount_options
->snapdir_name
,
1321 req
->r_dentry
->d_name
.len
))) {
1323 * lookup link rename : null -> possibly existing inode
1324 * mknod symlink mkdir : null -> new inode
1325 * unlink : linked -> null
1327 struct inode
*dir
= req
->r_parent
;
1328 struct dentry
*dn
= req
->r_dentry
;
1329 bool have_dir_cap
, have_lease
;
1333 BUG_ON(d_inode(dn
->d_parent
) != dir
);
1335 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1336 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1338 BUG_ON(ceph_ino(dir
) != dvino
.ino
);
1339 BUG_ON(ceph_snap(dir
) != dvino
.snap
);
1341 /* do we have a lease on the whole dir? */
1343 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1344 CEPH_CAP_FILE_SHARED
);
1346 /* do we have a dn lease? */
1347 have_lease
= have_dir_cap
||
1348 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1350 dout("fill_trace no dentry lease or dir cap\n");
1353 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1354 struct inode
*olddir
= req
->r_old_dentry_dir
;
1357 dout(" src %p '%pd' dst %p '%pd'\n",
1361 dout("fill_trace doing d_move %p -> %p\n",
1362 req
->r_old_dentry
, dn
);
1364 /* d_move screws up sibling dentries' offsets */
1365 ceph_dir_clear_ordered(dir
);
1366 ceph_dir_clear_ordered(olddir
);
1368 d_move(req
->r_old_dentry
, dn
);
1369 dout(" src %p '%pd' dst %p '%pd'\n",
1374 /* ensure target dentry is invalidated, despite
1375 rehashing bug in vfs_rename_dir */
1376 ceph_invalidate_dentry_lease(dn
);
1378 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1379 ceph_dentry(req
->r_old_dentry
)->offset
);
1381 /* swap r_dentry and r_old_dentry in case that
1382 * splice_dentry() gets called later. This is safe
1383 * because no other place will use them */
1384 req
->r_dentry
= req
->r_old_dentry
;
1385 req
->r_old_dentry
= dn
;
1390 if (!rinfo
->head
->is_target
) {
1391 dout("fill_trace null dentry\n");
1392 if (d_really_is_positive(dn
)) {
1393 dout("d_delete %p\n", dn
);
1394 ceph_dir_clear_ordered(dir
);
1396 } else if (have_lease
) {
1399 update_dentry_lease(dir
, dn
,
1400 rinfo
->dlease
, session
,
1401 req
->r_request_started
);
1406 /* attach proper inode */
1407 if (d_really_is_negative(dn
)) {
1408 ceph_dir_clear_ordered(dir
);
1410 err
= splice_dentry(&req
->r_dentry
, in
);
1413 dn
= req
->r_dentry
; /* may have spliced */
1414 } else if (d_really_is_positive(dn
) && d_inode(dn
) != in
) {
1415 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1416 dn
, d_inode(dn
), ceph_vinop(d_inode(dn
)),
1423 update_dentry_lease(dir
, dn
,
1424 rinfo
->dlease
, session
,
1425 req
->r_request_started
);
1427 dout(" final dn %p\n", dn
);
1428 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1429 req
->r_op
== CEPH_MDS_OP_MKSNAP
) &&
1430 !test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
)) {
1431 struct inode
*dir
= req
->r_parent
;
1433 /* fill out a snapdir LOOKUPSNAP dentry */
1435 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1436 BUG_ON(!req
->r_dentry
);
1437 dout(" linking snapped dir %p to dn %p\n", in
, req
->r_dentry
);
1438 ceph_dir_clear_ordered(dir
);
1440 err
= splice_dentry(&req
->r_dentry
, in
);
1443 } else if (rinfo
->head
->is_dentry
&& req
->r_dentry
) {
1444 /* parent inode is not locked, be carefull */
1445 struct ceph_vino
*ptvino
= NULL
;
1446 dvino
.ino
= le64_to_cpu(rinfo
->diri
.in
->ino
);
1447 dvino
.snap
= le64_to_cpu(rinfo
->diri
.in
->snapid
);
1448 if (rinfo
->head
->is_target
) {
1449 tvino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1450 tvino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1453 update_dentry_lease_careful(req
->r_dentry
, rinfo
->dlease
,
1454 session
, req
->r_request_started
,
1455 rinfo
->dname
, rinfo
->dname_len
,
1459 dout("fill_trace done err=%d\n", err
);
1464 * Prepopulate our cache with readdir results, leases, etc.
1466 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1467 struct ceph_mds_session
*session
)
1469 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1472 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1473 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1474 struct ceph_vino vino
;
1478 vino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1479 vino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1481 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1484 dout("new_inode badness got %d\n", err
);
1487 rc
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1488 req
->r_request_started
, -1,
1489 &req
->r_caps_reservation
);
1491 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1494 /* avoid calling iput_final() in mds dispatch threads */
1495 ceph_async_iput(in
);
1501 void ceph_readdir_cache_release(struct ceph_readdir_cache_control
*ctl
)
1505 put_page(ctl
->page
);
1510 static int fill_readdir_cache(struct inode
*dir
, struct dentry
*dn
,
1511 struct ceph_readdir_cache_control
*ctl
,
1512 struct ceph_mds_request
*req
)
1514 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1515 unsigned nsize
= PAGE_SIZE
/ sizeof(struct dentry
*);
1516 unsigned idx
= ctl
->index
% nsize
;
1517 pgoff_t pgoff
= ctl
->index
/ nsize
;
1519 if (!ctl
->page
|| pgoff
!= page_index(ctl
->page
)) {
1520 ceph_readdir_cache_release(ctl
);
1522 ctl
->page
= grab_cache_page(&dir
->i_data
, pgoff
);
1524 ctl
->page
= find_lock_page(&dir
->i_data
, pgoff
);
1527 return idx
== 0 ? -ENOMEM
: 0;
1529 /* reading/filling the cache are serialized by
1530 * i_mutex, no need to use page lock */
1531 unlock_page(ctl
->page
);
1532 ctl
->dentries
= kmap(ctl
->page
);
1534 memset(ctl
->dentries
, 0, PAGE_SIZE
);
1537 if (req
->r_dir_release_cnt
== atomic64_read(&ci
->i_release_count
) &&
1538 req
->r_dir_ordered_cnt
== atomic64_read(&ci
->i_ordered_count
)) {
1539 dout("readdir cache dn %p idx %d\n", dn
, ctl
->index
);
1540 ctl
->dentries
[idx
] = dn
;
1543 dout("disable readdir cache\n");
1549 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1550 struct ceph_mds_session
*session
)
1552 struct dentry
*parent
= req
->r_dentry
;
1553 struct ceph_inode_info
*ci
= ceph_inode(d_inode(parent
));
1554 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1558 int err
= 0, skipped
= 0, ret
, i
;
1559 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1560 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1563 struct ceph_readdir_cache_control cache_ctl
= {};
1565 if (test_bit(CEPH_MDS_R_ABORTED
, &req
->r_req_flags
))
1566 return readdir_prepopulate_inodes_only(req
, session
);
1568 if (rinfo
->hash_order
) {
1570 last_hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1572 strlen(req
->r_path2
));
1573 last_hash
= ceph_frag_value(last_hash
);
1574 } else if (rinfo
->offset_hash
) {
1575 /* mds understands offset_hash */
1576 WARN_ON_ONCE(req
->r_readdir_offset
!= 2);
1577 last_hash
= le32_to_cpu(rhead
->args
.readdir
.offset_hash
);
1581 if (rinfo
->dir_dir
&&
1582 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1583 dout("readdir_prepopulate got new frag %x -> %x\n",
1584 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1585 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1586 if (!rinfo
->hash_order
)
1587 req
->r_readdir_offset
= 2;
1590 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1591 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1592 rinfo
->dir_nr
, parent
);
1594 dout("readdir_prepopulate %d items under dn %p\n",
1595 rinfo
->dir_nr
, parent
);
1597 ceph_fill_dirfrag(d_inode(parent
), rinfo
->dir_dir
);
1599 if (ceph_frag_is_leftmost(frag
) &&
1600 req
->r_readdir_offset
== 2 &&
1601 !(rinfo
->hash_order
&& last_hash
)) {
1602 /* note dir version at start of readdir so we can
1603 * tell if any dentries get dropped */
1604 req
->r_dir_release_cnt
=
1605 atomic64_read(&ci
->i_release_count
);
1606 req
->r_dir_ordered_cnt
=
1607 atomic64_read(&ci
->i_ordered_count
);
1608 req
->r_readdir_cache_idx
= 0;
1612 cache_ctl
.index
= req
->r_readdir_cache_idx
;
1613 fpos_offset
= req
->r_readdir_offset
;
1615 /* FIXME: release caps/leases if error occurs */
1616 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1617 struct ceph_mds_reply_dir_entry
*rde
= rinfo
->dir_entries
+ i
;
1618 struct ceph_vino tvino
;
1620 dname
.name
= rde
->name
;
1621 dname
.len
= rde
->name_len
;
1622 dname
.hash
= full_name_hash(parent
, dname
.name
, dname
.len
);
1624 tvino
.ino
= le64_to_cpu(rde
->inode
.in
->ino
);
1625 tvino
.snap
= le64_to_cpu(rde
->inode
.in
->snapid
);
1627 if (rinfo
->hash_order
) {
1628 u32 hash
= ceph_str_hash(ci
->i_dir_layout
.dl_dir_hash
,
1629 rde
->name
, rde
->name_len
);
1630 hash
= ceph_frag_value(hash
);
1631 if (hash
!= last_hash
)
1634 rde
->offset
= ceph_make_fpos(hash
, fpos_offset
++, true);
1636 rde
->offset
= ceph_make_fpos(frag
, fpos_offset
++, false);
1640 dn
= d_lookup(parent
, &dname
);
1641 dout("d_lookup on parent=%p name=%.*s got %p\n",
1642 parent
, dname
.len
, dname
.name
, dn
);
1645 dn
= d_alloc(parent
, &dname
);
1646 dout("d_alloc %p '%.*s' = %p\n", parent
,
1647 dname
.len
, dname
.name
, dn
);
1649 dout("d_alloc badness\n");
1653 } else if (d_really_is_positive(dn
) &&
1654 (ceph_ino(d_inode(dn
)) != tvino
.ino
||
1655 ceph_snap(d_inode(dn
)) != tvino
.snap
)) {
1656 struct ceph_dentry_info
*di
= ceph_dentry(dn
);
1657 dout(" dn %p points to wrong inode %p\n",
1660 spin_lock(&dn
->d_lock
);
1661 if (di
->offset
> 0 &&
1662 di
->lease_shared_gen
==
1663 atomic_read(&ci
->i_shared_gen
)) {
1664 __ceph_dir_clear_ordered(ci
);
1667 spin_unlock(&dn
->d_lock
);
1675 if (d_really_is_positive(dn
)) {
1678 in
= ceph_get_inode(parent
->d_sb
, tvino
);
1680 dout("new_inode badness\n");
1688 ret
= fill_inode(in
, NULL
, &rde
->inode
, NULL
, session
,
1689 req
->r_request_started
, -1,
1690 &req
->r_caps_reservation
);
1692 pr_err("fill_inode badness on %p\n", in
);
1693 if (d_really_is_negative(dn
)) {
1694 /* avoid calling iput_final() in mds
1695 * dispatch threads */
1696 ceph_async_iput(in
);
1703 if (d_really_is_negative(dn
)) {
1704 if (ceph_security_xattr_deadlock(in
)) {
1705 dout(" skip splicing dn %p to inode %p"
1706 " (security xattr deadlock)\n", dn
, in
);
1707 ceph_async_iput(in
);
1712 err
= splice_dentry(&dn
, in
);
1717 ceph_dentry(dn
)->offset
= rde
->offset
;
1719 update_dentry_lease(d_inode(parent
), dn
,
1720 rde
->lease
, req
->r_session
,
1721 req
->r_request_started
);
1723 if (err
== 0 && skipped
== 0 && cache_ctl
.index
>= 0) {
1724 ret
= fill_readdir_cache(d_inode(parent
), dn
,
1733 if (err
== 0 && skipped
== 0) {
1734 set_bit(CEPH_MDS_R_DID_PREPOPULATE
, &req
->r_req_flags
);
1735 req
->r_readdir_cache_idx
= cache_ctl
.index
;
1737 ceph_readdir_cache_release(&cache_ctl
);
1738 dout("readdir_prepopulate done\n");
1742 bool ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1744 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1747 spin_lock(&ci
->i_ceph_lock
);
1748 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1749 i_size_write(inode
, size
);
1750 inode
->i_blocks
= calc_inode_blocks(size
);
1752 ret
= __ceph_should_report_size(ci
);
1754 spin_unlock(&ci
->i_ceph_lock
);
1759 * Put reference to inode, but avoid calling iput_final() in current thread.
1760 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1763 void ceph_async_iput(struct inode
*inode
)
1768 if (atomic_add_unless(&inode
->i_count
, -1, 1))
1770 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1771 &ceph_inode(inode
)->i_work
))
1773 /* queue work failed, i_count must be at least 2 */
1778 * Write back inode data in a worker thread. (This can't be done
1779 * in the message handler context.)
1781 void ceph_queue_writeback(struct inode
*inode
)
1783 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1784 set_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
);
1787 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1789 dout("ceph_queue_writeback %p\n", inode
);
1791 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1792 inode
, ci
->i_work_mask
);
1798 * queue an async invalidation
1800 void ceph_queue_invalidate(struct inode
*inode
)
1802 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1803 set_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
);
1806 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1807 &ceph_inode(inode
)->i_work
)) {
1808 dout("ceph_queue_invalidate %p\n", inode
);
1810 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1811 inode
, ci
->i_work_mask
);
1817 * Queue an async vmtruncate. If we fail to queue work, we will handle
1818 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1820 void ceph_queue_vmtruncate(struct inode
*inode
)
1822 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1823 set_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
);
1826 if (queue_work(ceph_inode_to_client(inode
)->inode_wq
,
1828 dout("ceph_queue_vmtruncate %p\n", inode
);
1830 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1831 inode
, ci
->i_work_mask
);
1836 static void ceph_do_invalidate_pages(struct inode
*inode
)
1838 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1839 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1843 mutex_lock(&ci
->i_truncate_mutex
);
1845 if (READ_ONCE(fsc
->mount_state
) == CEPH_MOUNT_SHUTDOWN
) {
1846 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1847 inode
, ceph_ino(inode
));
1848 mapping_set_error(inode
->i_mapping
, -EIO
);
1849 truncate_pagecache(inode
, 0);
1850 mutex_unlock(&ci
->i_truncate_mutex
);
1854 spin_lock(&ci
->i_ceph_lock
);
1855 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1856 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1857 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1858 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1860 spin_unlock(&ci
->i_ceph_lock
);
1861 mutex_unlock(&ci
->i_truncate_mutex
);
1864 orig_gen
= ci
->i_rdcache_gen
;
1865 spin_unlock(&ci
->i_ceph_lock
);
1867 if (invalidate_inode_pages2(inode
->i_mapping
) < 0) {
1868 pr_err("invalidate_pages %p fails\n", inode
);
1871 spin_lock(&ci
->i_ceph_lock
);
1872 if (orig_gen
== ci
->i_rdcache_gen
&&
1873 orig_gen
== ci
->i_rdcache_revoking
) {
1874 dout("invalidate_pages %p gen %d successful\n", inode
,
1876 ci
->i_rdcache_revoking
--;
1879 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1880 inode
, orig_gen
, ci
->i_rdcache_gen
,
1881 ci
->i_rdcache_revoking
);
1882 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1885 spin_unlock(&ci
->i_ceph_lock
);
1886 mutex_unlock(&ci
->i_truncate_mutex
);
1889 ceph_check_caps(ci
, 0, NULL
);
1893 * Make sure any pending truncation is applied before doing anything
1894 * that may depend on it.
1896 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1898 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1900 int wrbuffer_refs
, finish
= 0;
1902 mutex_lock(&ci
->i_truncate_mutex
);
1904 spin_lock(&ci
->i_ceph_lock
);
1905 if (ci
->i_truncate_pending
== 0) {
1906 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1907 spin_unlock(&ci
->i_ceph_lock
);
1908 mutex_unlock(&ci
->i_truncate_mutex
);
1913 * make sure any dirty snapped pages are flushed before we
1914 * possibly truncate them.. so write AND block!
1916 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1917 spin_unlock(&ci
->i_ceph_lock
);
1918 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1920 filemap_write_and_wait_range(&inode
->i_data
, 0,
1921 inode
->i_sb
->s_maxbytes
);
1925 /* there should be no reader or writer */
1926 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1928 to
= ci
->i_truncate_size
;
1929 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1930 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1931 ci
->i_truncate_pending
, to
);
1932 spin_unlock(&ci
->i_ceph_lock
);
1934 truncate_pagecache(inode
, to
);
1936 spin_lock(&ci
->i_ceph_lock
);
1937 if (to
== ci
->i_truncate_size
) {
1938 ci
->i_truncate_pending
= 0;
1941 spin_unlock(&ci
->i_ceph_lock
);
1945 mutex_unlock(&ci
->i_truncate_mutex
);
1947 if (wrbuffer_refs
== 0)
1948 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1950 wake_up_all(&ci
->i_cap_wq
);
1953 static void ceph_inode_work(struct work_struct
*work
)
1955 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1957 struct inode
*inode
= &ci
->vfs_inode
;
1959 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK
, &ci
->i_work_mask
)) {
1960 dout("writeback %p\n", inode
);
1961 filemap_fdatawrite(&inode
->i_data
);
1963 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES
, &ci
->i_work_mask
))
1964 ceph_do_invalidate_pages(inode
);
1966 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE
, &ci
->i_work_mask
))
1967 __ceph_do_pending_vmtruncate(inode
);
1975 static const struct inode_operations ceph_symlink_iops
= {
1976 .get_link
= simple_get_link
,
1977 .setattr
= ceph_setattr
,
1978 .getattr
= ceph_getattr
,
1979 .listxattr
= ceph_listxattr
,
1982 int __ceph_setattr(struct inode
*inode
, struct iattr
*attr
)
1984 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1985 const unsigned int ia_valid
= attr
->ia_valid
;
1986 struct ceph_mds_request
*req
;
1987 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1988 struct ceph_cap_flush
*prealloc_cf
;
1990 int release
= 0, dirtied
= 0;
1993 int inode_dirty_flags
= 0;
1994 bool lock_snap_rwsem
= false;
1996 prealloc_cf
= ceph_alloc_cap_flush();
2000 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
2003 ceph_free_cap_flush(prealloc_cf
);
2004 return PTR_ERR(req
);
2007 spin_lock(&ci
->i_ceph_lock
);
2008 issued
= __ceph_caps_issued(ci
, NULL
);
2010 if (!ci
->i_head_snapc
&&
2011 (issued
& (CEPH_CAP_ANY_EXCL
| CEPH_CAP_FILE_WR
))) {
2012 lock_snap_rwsem
= true;
2013 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
2014 spin_unlock(&ci
->i_ceph_lock
);
2015 down_read(&mdsc
->snap_rwsem
);
2016 spin_lock(&ci
->i_ceph_lock
);
2017 issued
= __ceph_caps_issued(ci
, NULL
);
2021 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
2023 if (ia_valid
& ATTR_UID
) {
2024 dout("setattr %p uid %d -> %d\n", inode
,
2025 from_kuid(&init_user_ns
, inode
->i_uid
),
2026 from_kuid(&init_user_ns
, attr
->ia_uid
));
2027 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2028 inode
->i_uid
= attr
->ia_uid
;
2029 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2030 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2031 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
2032 req
->r_args
.setattr
.uid
= cpu_to_le32(
2033 from_kuid(&init_user_ns
, attr
->ia_uid
));
2034 mask
|= CEPH_SETATTR_UID
;
2035 release
|= CEPH_CAP_AUTH_SHARED
;
2038 if (ia_valid
& ATTR_GID
) {
2039 dout("setattr %p gid %d -> %d\n", inode
,
2040 from_kgid(&init_user_ns
, inode
->i_gid
),
2041 from_kgid(&init_user_ns
, attr
->ia_gid
));
2042 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2043 inode
->i_gid
= attr
->ia_gid
;
2044 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2045 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2046 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
2047 req
->r_args
.setattr
.gid
= cpu_to_le32(
2048 from_kgid(&init_user_ns
, attr
->ia_gid
));
2049 mask
|= CEPH_SETATTR_GID
;
2050 release
|= CEPH_CAP_AUTH_SHARED
;
2053 if (ia_valid
& ATTR_MODE
) {
2054 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
2056 if (issued
& CEPH_CAP_AUTH_EXCL
) {
2057 inode
->i_mode
= attr
->ia_mode
;
2058 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2059 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
2060 attr
->ia_mode
!= inode
->i_mode
) {
2061 inode
->i_mode
= attr
->ia_mode
;
2062 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
2063 mask
|= CEPH_SETATTR_MODE
;
2064 release
|= CEPH_CAP_AUTH_SHARED
;
2068 if (ia_valid
& ATTR_ATIME
) {
2069 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode
,
2070 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
2071 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
2072 if (issued
& CEPH_CAP_FILE_EXCL
) {
2073 ci
->i_time_warp_seq
++;
2074 inode
->i_atime
= attr
->ia_atime
;
2075 dirtied
|= CEPH_CAP_FILE_EXCL
;
2076 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2077 timespec64_compare(&inode
->i_atime
,
2078 &attr
->ia_atime
) < 0) {
2079 inode
->i_atime
= attr
->ia_atime
;
2080 dirtied
|= CEPH_CAP_FILE_WR
;
2081 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2082 !timespec64_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
2083 ceph_encode_timespec64(&req
->r_args
.setattr
.atime
,
2085 mask
|= CEPH_SETATTR_ATIME
;
2086 release
|= CEPH_CAP_FILE_SHARED
|
2087 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2090 if (ia_valid
& ATTR_MTIME
) {
2091 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode
,
2092 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
2093 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
2094 if (issued
& CEPH_CAP_FILE_EXCL
) {
2095 ci
->i_time_warp_seq
++;
2096 inode
->i_mtime
= attr
->ia_mtime
;
2097 dirtied
|= CEPH_CAP_FILE_EXCL
;
2098 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
2099 timespec64_compare(&inode
->i_mtime
,
2100 &attr
->ia_mtime
) < 0) {
2101 inode
->i_mtime
= attr
->ia_mtime
;
2102 dirtied
|= CEPH_CAP_FILE_WR
;
2103 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2104 !timespec64_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
2105 ceph_encode_timespec64(&req
->r_args
.setattr
.mtime
,
2107 mask
|= CEPH_SETATTR_MTIME
;
2108 release
|= CEPH_CAP_FILE_SHARED
|
2109 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2112 if (ia_valid
& ATTR_SIZE
) {
2113 dout("setattr %p size %lld -> %lld\n", inode
,
2114 inode
->i_size
, attr
->ia_size
);
2115 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
2116 attr
->ia_size
> inode
->i_size
) {
2117 i_size_write(inode
, attr
->ia_size
);
2118 inode
->i_blocks
= calc_inode_blocks(attr
->ia_size
);
2119 ci
->i_reported_size
= attr
->ia_size
;
2120 dirtied
|= CEPH_CAP_FILE_EXCL
;
2121 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
2122 attr
->ia_size
!= inode
->i_size
) {
2123 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
2124 req
->r_args
.setattr
.old_size
=
2125 cpu_to_le64(inode
->i_size
);
2126 mask
|= CEPH_SETATTR_SIZE
;
2127 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_EXCL
|
2128 CEPH_CAP_FILE_RD
| CEPH_CAP_FILE_WR
;
2132 /* these do nothing */
2133 if (ia_valid
& ATTR_CTIME
) {
2134 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
2135 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
2136 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode
,
2137 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
2138 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
2139 only
? "ctime only" : "ignored");
2142 * if kernel wants to dirty ctime but nothing else,
2143 * we need to choose a cap to dirty under, or do
2144 * a almost-no-op setattr
2146 if (issued
& CEPH_CAP_AUTH_EXCL
)
2147 dirtied
|= CEPH_CAP_AUTH_EXCL
;
2148 else if (issued
& CEPH_CAP_FILE_EXCL
)
2149 dirtied
|= CEPH_CAP_FILE_EXCL
;
2150 else if (issued
& CEPH_CAP_XATTR_EXCL
)
2151 dirtied
|= CEPH_CAP_XATTR_EXCL
;
2153 mask
|= CEPH_SETATTR_CTIME
;
2156 if (ia_valid
& ATTR_FILE
)
2157 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
2160 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
,
2162 inode
->i_ctime
= attr
->ia_ctime
;
2166 spin_unlock(&ci
->i_ceph_lock
);
2167 if (lock_snap_rwsem
)
2168 up_read(&mdsc
->snap_rwsem
);
2170 if (inode_dirty_flags
)
2171 __mark_inode_dirty(inode
, inode_dirty_flags
);
2175 req
->r_inode
= inode
;
2177 req
->r_inode_drop
= release
;
2178 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
2179 req
->r_num_caps
= 1;
2180 req
->r_stamp
= attr
->ia_ctime
;
2181 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2183 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
2184 ceph_cap_string(dirtied
), mask
);
2186 ceph_mdsc_put_request(req
);
2187 ceph_free_cap_flush(prealloc_cf
);
2189 if (err
>= 0 && (mask
& CEPH_SETATTR_SIZE
))
2190 __ceph_do_pending_vmtruncate(inode
);
2198 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2200 struct inode
*inode
= d_inode(dentry
);
2201 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
2204 if (ceph_snap(inode
) != CEPH_NOSNAP
)
2207 err
= setattr_prepare(dentry
, attr
);
2211 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2212 attr
->ia_size
> max(inode
->i_size
, fsc
->max_file_size
))
2215 if ((attr
->ia_valid
& ATTR_SIZE
) &&
2216 ceph_quota_is_max_bytes_exceeded(inode
, attr
->ia_size
))
2219 err
= __ceph_setattr(inode
, attr
);
2221 if (err
>= 0 && (attr
->ia_valid
& ATTR_MODE
))
2222 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
2228 * Verify that we have a lease on the given mask. If not,
2229 * do a getattr against an mds.
2231 int __ceph_do_getattr(struct inode
*inode
, struct page
*locked_page
,
2232 int mask
, bool force
)
2234 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
2235 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
2236 struct ceph_mds_request
*req
;
2240 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
2241 dout("do_getattr inode %p SNAPDIR\n", inode
);
2245 dout("do_getattr inode %p mask %s mode 0%o\n",
2246 inode
, ceph_cap_string(mask
), inode
->i_mode
);
2247 if (!force
&& ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
2250 mode
= (mask
& CEPH_STAT_RSTAT
) ? USE_AUTH_MDS
: USE_ANY_MDS
;
2251 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, mode
);
2253 return PTR_ERR(req
);
2254 req
->r_inode
= inode
;
2256 req
->r_num_caps
= 1;
2257 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
2258 req
->r_locked_page
= locked_page
;
2259 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
2260 if (locked_page
&& err
== 0) {
2261 u64 inline_version
= req
->r_reply_info
.targeti
.inline_version
;
2262 if (inline_version
== 0) {
2263 /* the reply is supposed to contain inline data */
2265 } else if (inline_version
== CEPH_INLINE_NONE
) {
2268 err
= req
->r_reply_info
.targeti
.inline_len
;
2271 ceph_mdsc_put_request(req
);
2272 dout("do_getattr result=%d\n", err
);
2278 * Check inode permissions. We verify we have a valid value for
2279 * the AUTH cap, then call the generic handler.
2281 int ceph_permission(struct inode
*inode
, int mask
)
2285 if (mask
& MAY_NOT_BLOCK
)
2288 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
, false);
2291 err
= generic_permission(inode
, mask
);
2295 /* Craft a mask of needed caps given a set of requested statx attrs. */
2296 static int statx_to_caps(u32 want
)
2300 if (want
& (STATX_MODE
|STATX_UID
|STATX_GID
|STATX_CTIME
|STATX_BTIME
))
2301 mask
|= CEPH_CAP_AUTH_SHARED
;
2303 if (want
& (STATX_NLINK
|STATX_CTIME
))
2304 mask
|= CEPH_CAP_LINK_SHARED
;
2306 if (want
& (STATX_ATIME
|STATX_MTIME
|STATX_CTIME
|STATX_SIZE
|
2308 mask
|= CEPH_CAP_FILE_SHARED
;
2310 if (want
& (STATX_CTIME
))
2311 mask
|= CEPH_CAP_XATTR_SHARED
;
2317 * Get all the attributes. If we have sufficient caps for the requested attrs,
2318 * then we can avoid talking to the MDS at all.
2320 int ceph_getattr(const struct path
*path
, struct kstat
*stat
,
2321 u32 request_mask
, unsigned int flags
)
2323 struct inode
*inode
= d_inode(path
->dentry
);
2324 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2325 u32 valid_mask
= STATX_BASIC_STATS
;
2328 /* Skip the getattr altogether if we're asked not to sync */
2329 if (!(flags
& AT_STATX_DONT_SYNC
)) {
2330 err
= ceph_do_getattr(inode
, statx_to_caps(request_mask
),
2331 flags
& AT_STATX_FORCE_SYNC
);
2336 generic_fillattr(inode
, stat
);
2337 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
2340 * btime on newly-allocated inodes is 0, so if this is still set to
2341 * that, then assume that it's not valid.
2343 if (ci
->i_btime
.tv_sec
|| ci
->i_btime
.tv_nsec
) {
2344 stat
->btime
= ci
->i_btime
;
2345 valid_mask
|= STATX_BTIME
;
2348 if (ceph_snap(inode
) == CEPH_NOSNAP
)
2349 stat
->dev
= inode
->i_sb
->s_dev
;
2351 stat
->dev
= ci
->i_snapid_map
? ci
->i_snapid_map
->dev
: 0;
2353 if (S_ISDIR(inode
->i_mode
)) {
2354 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
2356 stat
->size
= ci
->i_rbytes
;
2358 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
2360 stat
->blksize
= 65536;
2362 * Some applications rely on the number of st_nlink
2363 * value on directories to be either 0 (if unlinked)
2364 * or 2 + number of subdirectories.
2366 if (stat
->nlink
== 1)
2367 /* '.' + '..' + subdirs */
2368 stat
->nlink
= 1 + 1 + ci
->i_subdirs
;
2371 stat
->result_mask
= request_mask
& valid_mask
;