1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/posix_acl.h>
15 #include "mds_client.h"
17 #include <linux/ceph/decode.h>
20 * Ceph inode operations
22 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23 * setattr, etc.), xattr helpers, and helpers for assimilating
24 * metadata returned by the MDS into our cache.
26 * Also define helpers for doing asynchronous writeback, invalidation,
27 * and truncation for the benefit of those who can't afford to block
28 * (typically because they are in the message handler path).
31 static const struct inode_operations ceph_symlink_iops
;
33 static void ceph_invalidate_work(struct work_struct
*work
);
34 static void ceph_writeback_work(struct work_struct
*work
);
35 static void ceph_vmtruncate_work(struct work_struct
*work
);
38 * find or create an inode, given the ceph ino number
40 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
42 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
43 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
47 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
50 ino_t t
= ceph_vino_to_ino(vino
);
52 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
54 return ERR_PTR(-ENOMEM
);
55 if (inode
->i_state
& I_NEW
) {
56 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
57 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
58 unlock_new_inode(inode
);
61 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
67 * get/constuct snapdir inode for a given directory
69 struct inode
*ceph_get_snapdir(struct inode
*parent
)
71 struct ceph_vino vino
= {
72 .ino
= ceph_ino(parent
),
75 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
76 struct ceph_inode_info
*ci
= ceph_inode(inode
);
78 BUG_ON(!S_ISDIR(parent
->i_mode
));
81 inode
->i_mode
= parent
->i_mode
;
82 inode
->i_uid
= parent
->i_uid
;
83 inode
->i_gid
= parent
->i_gid
;
84 inode
->i_op
= &ceph_dir_iops
;
85 inode
->i_fop
= &ceph_dir_fops
;
86 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
91 const struct inode_operations ceph_file_iops
= {
92 .permission
= ceph_permission
,
93 .setattr
= ceph_setattr
,
94 .getattr
= ceph_getattr
,
95 .setxattr
= ceph_setxattr
,
96 .getxattr
= ceph_getxattr
,
97 .listxattr
= ceph_listxattr
,
98 .removexattr
= ceph_removexattr
,
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 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
139 "frag %x\n", &ci
->vfs_inode
,
140 ceph_vinop(&ci
->vfs_inode
), f
);
141 return ERR_PTR(-ENOMEM
);
148 rb_link_node(&frag
->node
, parent
, p
);
149 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
151 dout("get_or_create_frag added %llx.%llx frag %x\n",
152 ceph_vinop(&ci
->vfs_inode
), f
);
157 * find a specific frag @f
159 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
161 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
164 struct ceph_inode_frag
*frag
=
165 rb_entry(n
, struct ceph_inode_frag
, node
);
166 int c
= ceph_frag_compare(f
, frag
->frag
);
178 * Choose frag containing the given value @v. If @pfrag is
179 * specified, copy the frag delegation info to the caller if
182 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
183 struct ceph_inode_frag
*pfrag
,
186 u32 t
= ceph_frag_make(0, 0);
187 struct ceph_inode_frag
*frag
;
194 mutex_lock(&ci
->i_fragtree_mutex
);
196 WARN_ON(!ceph_frag_contains_value(t
, v
));
197 frag
= __ceph_find_frag(ci
, t
);
199 break; /* t is a leaf */
200 if (frag
->split_by
== 0) {
202 memcpy(pfrag
, frag
, sizeof(*pfrag
));
209 nway
= 1 << frag
->split_by
;
210 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
211 frag
->split_by
, nway
);
212 for (i
= 0; i
< nway
; i
++) {
213 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
214 if (ceph_frag_contains_value(n
, v
)) {
221 dout("choose_frag(%x) = %x\n", v
, t
);
223 mutex_unlock(&ci
->i_fragtree_mutex
);
228 * Process dirfrag (delegation) info from the mds. Include leaf
229 * fragment in tree ONLY if ndist > 0. Otherwise, only
230 * branches/splits are included in i_fragtree)
232 static int ceph_fill_dirfrag(struct inode
*inode
,
233 struct ceph_mds_reply_dirfrag
*dirinfo
)
235 struct ceph_inode_info
*ci
= ceph_inode(inode
);
236 struct ceph_inode_frag
*frag
;
237 u32 id
= le32_to_cpu(dirinfo
->frag
);
238 int mds
= le32_to_cpu(dirinfo
->auth
);
239 int ndist
= le32_to_cpu(dirinfo
->ndist
);
243 mutex_lock(&ci
->i_fragtree_mutex
);
245 /* no delegation info needed. */
246 frag
= __ceph_find_frag(ci
, id
);
249 if (frag
->split_by
== 0) {
250 /* tree leaf, remove */
251 dout("fill_dirfrag removed %llx.%llx frag %x"
252 " (no ref)\n", ceph_vinop(inode
), id
);
253 rb_erase(&frag
->node
, &ci
->i_fragtree
);
256 /* tree branch, keep and clear */
257 dout("fill_dirfrag cleared %llx.%llx frag %x"
258 " referral\n", ceph_vinop(inode
), id
);
266 /* find/add this frag to store mds delegation info */
267 frag
= __get_or_create_frag(ci
, id
);
269 /* this is not the end of the world; we can continue
270 with bad/inaccurate delegation info */
271 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
272 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
278 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
279 for (i
= 0; i
< frag
->ndist
; i
++)
280 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
281 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
282 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
285 mutex_unlock(&ci
->i_fragtree_mutex
);
291 * initialize a newly allocated inode.
293 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
295 struct ceph_inode_info
*ci
;
298 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
302 dout("alloc_inode %p\n", &ci
->vfs_inode
);
304 spin_lock_init(&ci
->i_ceph_lock
);
307 ci
->i_time_warp_seq
= 0;
308 ci
->i_ceph_flags
= 0;
309 atomic_set(&ci
->i_release_count
, 1);
310 atomic_set(&ci
->i_complete_count
, 0);
311 ci
->i_symlink
= NULL
;
313 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
315 ci
->i_fragtree
= RB_ROOT
;
316 mutex_init(&ci
->i_fragtree_mutex
);
318 ci
->i_xattrs
.blob
= NULL
;
319 ci
->i_xattrs
.prealloc_blob
= NULL
;
320 ci
->i_xattrs
.dirty
= false;
321 ci
->i_xattrs
.index
= RB_ROOT
;
322 ci
->i_xattrs
.count
= 0;
323 ci
->i_xattrs
.names_size
= 0;
324 ci
->i_xattrs
.vals_size
= 0;
325 ci
->i_xattrs
.version
= 0;
326 ci
->i_xattrs
.index_version
= 0;
328 ci
->i_caps
= RB_ROOT
;
329 ci
->i_auth_cap
= NULL
;
330 ci
->i_dirty_caps
= 0;
331 ci
->i_flushing_caps
= 0;
332 INIT_LIST_HEAD(&ci
->i_dirty_item
);
333 INIT_LIST_HEAD(&ci
->i_flushing_item
);
334 ci
->i_cap_flush_seq
= 0;
335 ci
->i_cap_flush_last_tid
= 0;
336 memset(&ci
->i_cap_flush_tid
, 0, sizeof(ci
->i_cap_flush_tid
));
337 init_waitqueue_head(&ci
->i_cap_wq
);
338 ci
->i_hold_caps_min
= 0;
339 ci
->i_hold_caps_max
= 0;
340 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
341 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
342 ci
->i_head_snapc
= NULL
;
344 ci
->i_cap_exporting_issued
= 0;
346 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
347 ci
->i_nr_by_mode
[i
] = 0;
349 mutex_init(&ci
->i_truncate_mutex
);
350 ci
->i_truncate_seq
= 0;
351 ci
->i_truncate_size
= 0;
352 ci
->i_truncate_pending
= 0;
355 ci
->i_reported_size
= 0;
356 ci
->i_wanted_max_size
= 0;
357 ci
->i_requested_max_size
= 0;
361 ci
->i_rdcache_ref
= 0;
364 ci
->i_wrbuffer_ref
= 0;
365 ci
->i_wrbuffer_ref_head
= 0;
366 ci
->i_shared_gen
= 0;
367 ci
->i_rdcache_gen
= 0;
368 ci
->i_rdcache_revoking
= 0;
370 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
371 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
372 spin_lock_init(&ci
->i_unsafe_lock
);
374 ci
->i_snap_realm
= NULL
;
375 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
376 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
378 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
379 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
381 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
383 ceph_fscache_inode_init(ci
);
385 return &ci
->vfs_inode
;
388 static void ceph_i_callback(struct rcu_head
*head
)
390 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
391 struct ceph_inode_info
*ci
= ceph_inode(inode
);
393 kmem_cache_free(ceph_inode_cachep
, ci
);
396 void ceph_destroy_inode(struct inode
*inode
)
398 struct ceph_inode_info
*ci
= ceph_inode(inode
);
399 struct ceph_inode_frag
*frag
;
402 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
404 ceph_fscache_unregister_inode_cookie(ci
);
406 ceph_queue_caps_release(inode
);
409 * we may still have a snap_realm reference if there are stray
410 * caps in i_cap_exporting_issued or i_snap_caps.
412 if (ci
->i_snap_realm
) {
413 struct ceph_mds_client
*mdsc
=
414 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
415 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
417 dout(" dropping residual ref to snap realm %p\n", realm
);
418 spin_lock(&realm
->inodes_with_caps_lock
);
419 list_del_init(&ci
->i_snap_realm_item
);
420 spin_unlock(&realm
->inodes_with_caps_lock
);
421 ceph_put_snap_realm(mdsc
, realm
);
424 kfree(ci
->i_symlink
);
425 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
426 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
427 rb_erase(n
, &ci
->i_fragtree
);
431 __ceph_destroy_xattrs(ci
);
432 if (ci
->i_xattrs
.blob
)
433 ceph_buffer_put(ci
->i_xattrs
.blob
);
434 if (ci
->i_xattrs
.prealloc_blob
)
435 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
437 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
440 int ceph_drop_inode(struct inode
*inode
)
443 * Positve dentry and corresponding inode are always accompanied
444 * in MDS reply. So no need to keep inode in the cache after
445 * dropping all its aliases.
451 * Helpers to fill in size, ctime, mtime, and atime. We have to be
452 * careful because either the client or MDS may have more up to date
453 * info, depending on which capabilities are held, and whether
454 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
455 * and size are monotonically increasing, except when utimes() or
456 * truncate() increments the corresponding _seq values.)
458 int ceph_fill_file_size(struct inode
*inode
, int issued
,
459 u32 truncate_seq
, u64 truncate_size
, u64 size
)
461 struct ceph_inode_info
*ci
= ceph_inode(inode
);
464 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
465 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
466 dout("size %lld -> %llu\n", inode
->i_size
, size
);
467 inode
->i_size
= size
;
468 inode
->i_blocks
= (size
+ (1<<9) - 1) >> 9;
469 ci
->i_reported_size
= size
;
470 if (truncate_seq
!= ci
->i_truncate_seq
) {
471 dout("truncate_seq %u -> %u\n",
472 ci
->i_truncate_seq
, truncate_seq
);
473 ci
->i_truncate_seq
= truncate_seq
;
475 /* the MDS should have revoked these caps */
476 WARN_ON_ONCE(issued
& (CEPH_CAP_FILE_EXCL
|
479 CEPH_CAP_FILE_LAZYIO
));
481 * If we hold relevant caps, or in the case where we're
482 * not the only client referencing this file and we
483 * don't hold those caps, then we need to check whether
484 * the file is either opened or mmaped
486 if ((issued
& (CEPH_CAP_FILE_CACHE
|
487 CEPH_CAP_FILE_BUFFER
)) ||
488 mapping_mapped(inode
->i_mapping
) ||
489 __ceph_caps_file_wanted(ci
)) {
490 ci
->i_truncate_pending
++;
495 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
496 ci
->i_truncate_size
!= truncate_size
) {
497 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
499 ci
->i_truncate_size
= truncate_size
;
503 ceph_fscache_invalidate(inode
);
508 void ceph_fill_file_time(struct inode
*inode
, int issued
,
509 u64 time_warp_seq
, struct timespec
*ctime
,
510 struct timespec
*mtime
, struct timespec
*atime
)
512 struct ceph_inode_info
*ci
= ceph_inode(inode
);
515 if (issued
& (CEPH_CAP_FILE_EXCL
|
517 CEPH_CAP_FILE_BUFFER
|
519 CEPH_CAP_XATTR_EXCL
)) {
520 if (timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
521 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
522 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
523 ctime
->tv_sec
, ctime
->tv_nsec
);
524 inode
->i_ctime
= *ctime
;
526 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
527 /* the MDS did a utimes() */
528 dout("mtime %ld.%09ld -> %ld.%09ld "
530 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
531 mtime
->tv_sec
, mtime
->tv_nsec
,
532 ci
->i_time_warp_seq
, (int)time_warp_seq
);
534 inode
->i_mtime
= *mtime
;
535 inode
->i_atime
= *atime
;
536 ci
->i_time_warp_seq
= time_warp_seq
;
537 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
538 /* nobody did utimes(); take the max */
539 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
540 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
541 inode
->i_mtime
.tv_sec
,
542 inode
->i_mtime
.tv_nsec
,
543 mtime
->tv_sec
, mtime
->tv_nsec
);
544 inode
->i_mtime
= *mtime
;
546 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
547 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
548 inode
->i_atime
.tv_sec
,
549 inode
->i_atime
.tv_nsec
,
550 atime
->tv_sec
, atime
->tv_nsec
);
551 inode
->i_atime
= *atime
;
553 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
554 /* we did a utimes(); ignore mds values */
559 /* we have no write|excl caps; whatever the MDS says is true */
560 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
561 inode
->i_ctime
= *ctime
;
562 inode
->i_mtime
= *mtime
;
563 inode
->i_atime
= *atime
;
564 ci
->i_time_warp_seq
= time_warp_seq
;
569 if (warn
) /* time_warp_seq shouldn't go backwards */
570 dout("%p mds time_warp_seq %llu < %u\n",
571 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
575 * Populate an inode based on info from mds. May be called on new or
578 static int fill_inode(struct inode
*inode
,
579 struct ceph_mds_reply_info_in
*iinfo
,
580 struct ceph_mds_reply_dirfrag
*dirinfo
,
581 struct ceph_mds_session
*session
,
582 unsigned long ttl_from
, int cap_fmode
,
583 struct ceph_cap_reservation
*caps_reservation
)
585 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
586 struct ceph_inode_info
*ci
= ceph_inode(inode
);
588 int issued
= 0, implemented
;
589 struct timespec mtime
, atime
, ctime
;
591 struct ceph_inode_frag
*frag
;
592 struct rb_node
*rb_node
;
593 struct ceph_buffer
*xattr_blob
= NULL
;
597 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
598 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
602 * prealloc xattr data, if it looks like we'll need it. only
603 * if len > 4 (meaning there are actually xattrs; the first 4
604 * bytes are the xattr count).
606 if (iinfo
->xattr_len
> 4) {
607 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
609 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
613 spin_lock(&ci
->i_ceph_lock
);
616 * provided version will be odd if inode value is projected,
617 * even if stable. skip the update if we have newer stable
618 * info (ours>=theirs, e.g. due to racing mds replies), unless
619 * we are getting projected (unstable) info (in which case the
620 * version is odd, and we want ours>theirs).
626 if (le64_to_cpu(info
->version
) > 0 &&
627 (ci
->i_version
& ~1) >= le64_to_cpu(info
->version
))
630 issued
= __ceph_caps_issued(ci
, &implemented
);
631 issued
|= implemented
| __ceph_caps_dirty(ci
);
634 ci
->i_version
= le64_to_cpu(info
->version
);
636 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
638 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
639 inode
->i_mode
= le32_to_cpu(info
->mode
);
640 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
641 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
642 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
643 from_kuid(&init_user_ns
, inode
->i_uid
),
644 from_kgid(&init_user_ns
, inode
->i_gid
));
647 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
648 set_nlink(inode
, le32_to_cpu(info
->nlink
));
650 /* be careful with mtime, atime, size */
651 ceph_decode_timespec(&atime
, &info
->atime
);
652 ceph_decode_timespec(&mtime
, &info
->mtime
);
653 ceph_decode_timespec(&ctime
, &info
->ctime
);
654 queue_trunc
= ceph_fill_file_size(inode
, issued
,
655 le32_to_cpu(info
->truncate_seq
),
656 le64_to_cpu(info
->truncate_size
),
657 le64_to_cpu(info
->size
));
658 ceph_fill_file_time(inode
, issued
,
659 le32_to_cpu(info
->time_warp_seq
),
660 &ctime
, &mtime
, &atime
);
662 ci
->i_layout
= info
->layout
;
663 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
666 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
667 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 &&
668 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
669 if (ci
->i_xattrs
.blob
)
670 ceph_buffer_put(ci
->i_xattrs
.blob
);
671 ci
->i_xattrs
.blob
= xattr_blob
;
673 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
674 iinfo
->xattr_data
, iinfo
->xattr_len
);
675 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
676 ceph_forget_all_cached_acls(inode
);
680 inode
->i_mapping
->a_ops
= &ceph_aops
;
681 inode
->i_mapping
->backing_dev_info
=
682 &ceph_sb_to_client(inode
->i_sb
)->backing_dev_info
;
684 switch (inode
->i_mode
& S_IFMT
) {
689 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
690 inode
->i_op
= &ceph_file_iops
;
693 inode
->i_op
= &ceph_file_iops
;
694 inode
->i_fop
= &ceph_file_fops
;
697 inode
->i_op
= &ceph_symlink_iops
;
698 if (!ci
->i_symlink
) {
699 u32 symlen
= iinfo
->symlink_len
;
702 spin_unlock(&ci
->i_ceph_lock
);
705 if (WARN_ON(symlen
!= inode
->i_size
))
709 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
713 spin_lock(&ci
->i_ceph_lock
);
717 kfree(sym
); /* lost a race */
721 inode
->i_op
= &ceph_dir_iops
;
722 inode
->i_fop
= &ceph_dir_fops
;
724 ci
->i_dir_layout
= iinfo
->dir_layout
;
726 ci
->i_files
= le64_to_cpu(info
->files
);
727 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
728 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
729 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
730 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
731 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
734 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
735 ceph_vinop(inode
), inode
->i_mode
);
738 /* set dir completion flag? */
739 if (S_ISDIR(inode
->i_mode
) &&
740 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
741 ceph_snap(inode
) == CEPH_NOSNAP
&&
742 (le32_to_cpu(info
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) &&
743 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
744 !__ceph_dir_is_complete(ci
)) {
745 dout(" marking %p complete (empty)\n", inode
);
746 __ceph_dir_set_complete(ci
, atomic_read(&ci
->i_release_count
));
747 ci
->i_max_offset
= 2;
750 /* only update max_size on auth cap */
751 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
752 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
753 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
754 le64_to_cpu(info
->max_size
));
755 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
758 spin_unlock(&ci
->i_ceph_lock
);
760 /* queue truncate if we saw i_size decrease */
762 ceph_queue_vmtruncate(inode
);
764 /* populate frag tree */
765 /* FIXME: move me up, if/when version reflects fragtree changes */
766 nsplits
= le32_to_cpu(info
->fragtree
.nsplits
);
767 mutex_lock(&ci
->i_fragtree_mutex
);
768 rb_node
= rb_first(&ci
->i_fragtree
);
769 for (i
= 0; i
< nsplits
; i
++) {
770 u32 id
= le32_to_cpu(info
->fragtree
.splits
[i
].frag
);
773 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
774 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
775 if (frag
->frag
!= id
)
778 rb_node
= rb_next(rb_node
);
781 rb_node
= rb_next(rb_node
);
782 rb_erase(&frag
->node
, &ci
->i_fragtree
);
787 frag
= __get_or_create_frag(ci
, id
);
791 frag
->split_by
= le32_to_cpu(info
->fragtree
.splits
[i
].by
);
792 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
795 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
796 rb_node
= rb_next(rb_node
);
797 rb_erase(&frag
->node
, &ci
->i_fragtree
);
800 mutex_unlock(&ci
->i_fragtree_mutex
);
802 /* were we issued a capability? */
803 if (info
->cap
.caps
) {
804 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
805 ceph_add_cap(inode
, session
,
806 le64_to_cpu(info
->cap
.cap_id
),
808 le32_to_cpu(info
->cap
.caps
),
809 le32_to_cpu(info
->cap
.wanted
),
810 le32_to_cpu(info
->cap
.seq
),
811 le32_to_cpu(info
->cap
.mseq
),
812 le64_to_cpu(info
->cap
.realm
),
816 spin_lock(&ci
->i_ceph_lock
);
817 dout(" %p got snap_caps %s\n", inode
,
818 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
819 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
821 __ceph_get_fmode(ci
, cap_fmode
);
822 spin_unlock(&ci
->i_ceph_lock
);
824 } else if (cap_fmode
>= 0) {
825 pr_warning("mds issued no caps on %llx.%llx\n",
827 __ceph_get_fmode(ci
, cap_fmode
);
830 /* update delegation info? */
832 ceph_fill_dirfrag(inode
, dirinfo
);
838 ceph_buffer_put(xattr_blob
);
843 * caller should hold session s_mutex.
845 static void update_dentry_lease(struct dentry
*dentry
,
846 struct ceph_mds_reply_lease
*lease
,
847 struct ceph_mds_session
*session
,
848 unsigned long from_time
)
850 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
851 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
852 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
853 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
856 /* only track leases on regular dentries */
857 if (dentry
->d_op
!= &ceph_dentry_ops
)
860 spin_lock(&dentry
->d_lock
);
861 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
862 dentry
, duration
, ttl
);
864 /* make lease_rdcache_gen match directory */
865 dir
= dentry
->d_parent
->d_inode
;
866 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
871 if (di
->lease_gen
== session
->s_cap_gen
&&
872 time_before(ttl
, dentry
->d_time
))
873 goto out_unlock
; /* we already have a newer lease. */
875 if (di
->lease_session
&& di
->lease_session
!= session
)
878 ceph_dentry_lru_touch(dentry
);
880 if (!di
->lease_session
)
881 di
->lease_session
= ceph_get_mds_session(session
);
882 di
->lease_gen
= session
->s_cap_gen
;
883 di
->lease_seq
= le32_to_cpu(lease
->seq
);
884 di
->lease_renew_after
= half_ttl
;
885 di
->lease_renew_from
= 0;
886 dentry
->d_time
= ttl
;
888 spin_unlock(&dentry
->d_lock
);
893 * Set dentry's directory position based on the current dir's max, and
894 * order it in d_subdirs, so that dcache_readdir behaves.
896 * Always called under directory's i_mutex.
898 static void ceph_set_dentry_offset(struct dentry
*dn
)
900 struct dentry
*dir
= dn
->d_parent
;
901 struct inode
*inode
= dir
->d_inode
;
902 struct ceph_inode_info
*ci
;
903 struct ceph_dentry_info
*di
;
907 ci
= ceph_inode(inode
);
908 di
= ceph_dentry(dn
);
910 spin_lock(&ci
->i_ceph_lock
);
911 if (!__ceph_dir_is_complete(ci
)) {
912 spin_unlock(&ci
->i_ceph_lock
);
915 di
->offset
= ceph_inode(inode
)->i_max_offset
++;
916 spin_unlock(&ci
->i_ceph_lock
);
918 spin_lock(&dir
->d_lock
);
919 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
920 list_move(&dn
->d_u
.d_child
, &dir
->d_subdirs
);
921 dout("set_dentry_offset %p %lld (%p %p)\n", dn
, di
->offset
,
922 dn
->d_u
.d_child
.prev
, dn
->d_u
.d_child
.next
);
923 spin_unlock(&dn
->d_lock
);
924 spin_unlock(&dir
->d_lock
);
928 * splice a dentry to an inode.
929 * caller must hold directory i_mutex for this to be safe.
931 * we will only rehash the resulting dentry if @prehash is
932 * true; @prehash will be set to false (for the benefit of
933 * the caller) if we fail.
935 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
936 bool *prehash
, bool set_offset
)
938 struct dentry
*realdn
;
942 /* dn must be unhashed */
945 realdn
= d_materialise_unique(dn
, in
);
946 if (IS_ERR(realdn
)) {
947 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
948 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
950 *prehash
= false; /* don't rehash on error */
951 dn
= realdn
; /* note realdn contains the error */
954 dout("dn %p (%d) spliced with %p (%d) "
955 "inode %p ino %llx.%llx\n",
957 realdn
, d_count(realdn
),
958 realdn
->d_inode
, ceph_vinop(realdn
->d_inode
));
962 BUG_ON(!ceph_dentry(dn
));
963 dout("dn %p attached to %p ino %llx.%llx\n",
964 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
));
966 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
969 ceph_set_dentry_offset(dn
);
975 * Incorporate results into the local cache. This is either just
976 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
979 * A reply may contain
980 * a directory inode along with a dentry.
981 * and/or a target inode
983 * Called with snap_rwsem (read).
985 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
986 struct ceph_mds_session
*session
)
988 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
989 struct inode
*in
= NULL
;
990 struct ceph_mds_reply_inode
*ininfo
;
991 struct ceph_vino vino
;
992 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
995 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
996 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
1002 * If we resend completed ops to a recovering mds, we get no
1003 * trace. Since that is very rare, pretend this is the case
1004 * to ensure the 'no trace' handlers in the callers behave.
1006 * Fill in inodes unconditionally to avoid breaking cap
1009 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
1010 pr_info("fill_trace faking empty trace on %lld %s\n",
1011 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
1012 if (rinfo
->head
->is_dentry
) {
1013 rinfo
->head
->is_dentry
= 0;
1014 err
= fill_inode(req
->r_locked_dir
,
1015 &rinfo
->diri
, rinfo
->dirfrag
,
1016 session
, req
->r_request_started
, -1);
1018 if (rinfo
->head
->is_target
) {
1019 rinfo
->head
->is_target
= 0;
1020 ininfo
= rinfo
->targeti
.in
;
1021 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1022 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1023 in
= ceph_get_inode(sb
, vino
);
1024 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1025 session
, req
->r_request_started
,
1032 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
1033 dout("fill_trace reply is empty!\n");
1034 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
1035 ceph_invalidate_dir_request(req
);
1039 if (rinfo
->head
->is_dentry
) {
1040 struct inode
*dir
= req
->r_locked_dir
;
1043 err
= fill_inode(dir
, &rinfo
->diri
, rinfo
->dirfrag
,
1044 session
, req
->r_request_started
, -1,
1045 &req
->r_caps_reservation
);
1052 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1054 struct dentry
*dn
, *parent
;
1056 BUG_ON(!rinfo
->head
->is_target
);
1057 BUG_ON(req
->r_dentry
);
1059 parent
= d_find_any_alias(dir
);
1062 dname
.name
= rinfo
->dname
;
1063 dname
.len
= rinfo
->dname_len
;
1064 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1065 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1066 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1068 dn
= d_lookup(parent
, &dname
);
1069 dout("d_lookup on parent=%p name=%.*s got %p\n",
1070 parent
, dname
.len
, dname
.name
, dn
);
1073 dn
= d_alloc(parent
, &dname
);
1074 dout("d_alloc %p '%.*s' = %p\n", parent
,
1075 dname
.len
, dname
.name
, dn
);
1081 err
= ceph_init_dentry(dn
);
1087 } else if (dn
->d_inode
&&
1088 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1089 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1090 dout(" dn %p points to wrong inode %p\n",
1102 if (rinfo
->head
->is_target
) {
1103 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1104 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1106 in
= ceph_get_inode(sb
, vino
);
1111 req
->r_target_inode
= in
;
1113 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1114 session
, req
->r_request_started
,
1115 (!req
->r_aborted
&& rinfo
->head
->result
== 0) ?
1117 &req
->r_caps_reservation
);
1119 pr_err("fill_inode badness %p %llx.%llx\n",
1120 in
, ceph_vinop(in
));
1126 * ignore null lease/binding on snapdir ENOENT, or else we
1127 * will have trouble splicing in the virtual snapdir later
1129 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1130 req
->r_locked_dir
&&
1131 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1132 fsc
->mount_options
->snapdir_name
,
1133 req
->r_dentry
->d_name
.len
))) {
1135 * lookup link rename : null -> possibly existing inode
1136 * mknod symlink mkdir : null -> new inode
1137 * unlink : linked -> null
1139 struct inode
*dir
= req
->r_locked_dir
;
1140 struct dentry
*dn
= req
->r_dentry
;
1141 bool have_dir_cap
, have_lease
;
1145 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
1146 BUG_ON(ceph_ino(dir
) !=
1147 le64_to_cpu(rinfo
->diri
.in
->ino
));
1148 BUG_ON(ceph_snap(dir
) !=
1149 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1151 /* do we have a lease on the whole dir? */
1153 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1154 CEPH_CAP_FILE_SHARED
);
1156 /* do we have a dn lease? */
1157 have_lease
= have_dir_cap
||
1158 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1160 dout("fill_trace no dentry lease or dir cap\n");
1163 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1164 dout(" src %p '%.*s' dst %p '%.*s'\n",
1166 req
->r_old_dentry
->d_name
.len
,
1167 req
->r_old_dentry
->d_name
.name
,
1168 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1169 dout("fill_trace doing d_move %p -> %p\n",
1170 req
->r_old_dentry
, dn
);
1172 d_move(req
->r_old_dentry
, dn
);
1173 dout(" src %p '%.*s' dst %p '%.*s'\n",
1175 req
->r_old_dentry
->d_name
.len
,
1176 req
->r_old_dentry
->d_name
.name
,
1177 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1179 /* ensure target dentry is invalidated, despite
1180 rehashing bug in vfs_rename_dir */
1181 ceph_invalidate_dentry_lease(dn
);
1184 * d_move() puts the renamed dentry at the end of
1185 * d_subdirs. We need to assign it an appropriate
1186 * directory offset so we can behave when dir is
1189 ceph_set_dentry_offset(req
->r_old_dentry
);
1190 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1191 ceph_dentry(req
->r_old_dentry
)->offset
);
1193 dn
= req
->r_old_dentry
; /* use old_dentry */
1197 if (!rinfo
->head
->is_target
) {
1198 dout("fill_trace null dentry\n");
1200 dout("d_delete %p\n", dn
);
1203 dout("d_instantiate %p NULL\n", dn
);
1204 d_instantiate(dn
, NULL
);
1205 if (have_lease
&& d_unhashed(dn
))
1207 update_dentry_lease(dn
, rinfo
->dlease
,
1209 req
->r_request_started
);
1214 /* attach proper inode */
1217 dn
= splice_dentry(dn
, in
, &have_lease
, true);
1222 req
->r_dentry
= dn
; /* may have spliced */
1223 } else if (dn
->d_inode
&& dn
->d_inode
!= in
) {
1224 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1225 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
),
1231 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1232 req
->r_request_started
);
1233 dout(" final dn %p\n", dn
);
1234 } else if (!req
->r_aborted
&&
1235 (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1236 req
->r_op
== CEPH_MDS_OP_MKSNAP
)) {
1237 struct dentry
*dn
= req
->r_dentry
;
1239 /* fill out a snapdir LOOKUPSNAP dentry */
1241 BUG_ON(!req
->r_locked_dir
);
1242 BUG_ON(ceph_snap(req
->r_locked_dir
) != CEPH_SNAPDIR
);
1243 ininfo
= rinfo
->targeti
.in
;
1244 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1245 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1246 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1248 dn
= splice_dentry(dn
, in
, NULL
, true);
1253 req
->r_dentry
= dn
; /* may have spliced */
1256 dout("fill_trace done err=%d\n", err
);
1261 * Prepopulate our cache with readdir results, leases, etc.
1263 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1264 struct ceph_mds_session
*session
)
1266 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1269 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1270 struct ceph_vino vino
;
1274 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1275 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1277 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1280 dout("new_inode badness got %d\n", err
);
1283 rc
= fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1284 req
->r_request_started
, -1,
1285 &req
->r_caps_reservation
);
1287 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1296 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1297 struct ceph_mds_session
*session
)
1299 struct dentry
*parent
= req
->r_dentry
;
1300 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1304 int err
= 0, ret
, i
;
1305 struct inode
*snapdir
= NULL
;
1306 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1307 struct ceph_dentry_info
*di
;
1308 u64 r_readdir_offset
= req
->r_readdir_offset
;
1309 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1311 if (rinfo
->dir_dir
&&
1312 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1313 dout("readdir_prepopulate got new frag %x -> %x\n",
1314 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1315 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1316 if (ceph_frag_is_leftmost(frag
))
1317 r_readdir_offset
= 2;
1319 r_readdir_offset
= 0;
1323 return readdir_prepopulate_inodes_only(req
, session
);
1325 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1326 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1327 parent
= d_find_alias(snapdir
);
1328 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1329 rinfo
->dir_nr
, parent
);
1331 dout("readdir_prepopulate %d items under dn %p\n",
1332 rinfo
->dir_nr
, parent
);
1334 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1337 /* FIXME: release caps/leases if error occurs */
1338 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1339 struct ceph_vino vino
;
1341 dname
.name
= rinfo
->dir_dname
[i
];
1342 dname
.len
= rinfo
->dir_dname_len
[i
];
1343 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1345 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1346 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1349 dn
= d_lookup(parent
, &dname
);
1350 dout("d_lookup on parent=%p name=%.*s got %p\n",
1351 parent
, dname
.len
, dname
.name
, dn
);
1354 dn
= d_alloc(parent
, &dname
);
1355 dout("d_alloc %p '%.*s' = %p\n", parent
,
1356 dname
.len
, dname
.name
, dn
);
1358 dout("d_alloc badness\n");
1362 ret
= ceph_init_dentry(dn
);
1368 } else if (dn
->d_inode
&&
1369 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1370 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1371 dout(" dn %p points to wrong inode %p\n",
1377 /* reorder parent's d_subdirs */
1378 spin_lock(&parent
->d_lock
);
1379 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
1380 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1381 spin_unlock(&dn
->d_lock
);
1382 spin_unlock(&parent
->d_lock
);
1389 in
= ceph_get_inode(parent
->d_sb
, vino
);
1391 dout("new_inode badness\n");
1399 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1400 req
->r_request_started
, -1,
1401 &req
->r_caps_reservation
) < 0) {
1402 pr_err("fill_inode badness on %p\n", in
);
1410 dn
= splice_dentry(dn
, in
, NULL
, false);
1419 di
->offset
= ceph_make_fpos(frag
, i
+ r_readdir_offset
);
1421 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1423 req
->r_request_started
);
1429 req
->r_did_prepopulate
= true;
1436 dout("readdir_prepopulate done\n");
1440 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1442 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1445 spin_lock(&ci
->i_ceph_lock
);
1446 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1447 inode
->i_size
= size
;
1448 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1450 /* tell the MDS if we are approaching max_size */
1451 if ((size
<< 1) >= ci
->i_max_size
&&
1452 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1455 spin_unlock(&ci
->i_ceph_lock
);
1460 * Write back inode data in a worker thread. (This can't be done
1461 * in the message handler context.)
1463 void ceph_queue_writeback(struct inode
*inode
)
1466 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1467 &ceph_inode(inode
)->i_wb_work
)) {
1468 dout("ceph_queue_writeback %p\n", inode
);
1470 dout("ceph_queue_writeback %p failed\n", inode
);
1475 static void ceph_writeback_work(struct work_struct
*work
)
1477 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1479 struct inode
*inode
= &ci
->vfs_inode
;
1481 dout("writeback %p\n", inode
);
1482 filemap_fdatawrite(&inode
->i_data
);
1487 * queue an async invalidation
1489 void ceph_queue_invalidate(struct inode
*inode
)
1492 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1493 &ceph_inode(inode
)->i_pg_inv_work
)) {
1494 dout("ceph_queue_invalidate %p\n", inode
);
1496 dout("ceph_queue_invalidate %p failed\n", inode
);
1502 * Invalidate inode pages in a worker thread. (This can't be done
1503 * in the message handler context.)
1505 static void ceph_invalidate_work(struct work_struct
*work
)
1507 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1509 struct inode
*inode
= &ci
->vfs_inode
;
1513 mutex_lock(&ci
->i_truncate_mutex
);
1514 spin_lock(&ci
->i_ceph_lock
);
1515 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1516 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1517 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1518 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1520 spin_unlock(&ci
->i_ceph_lock
);
1521 mutex_unlock(&ci
->i_truncate_mutex
);
1524 orig_gen
= ci
->i_rdcache_gen
;
1525 spin_unlock(&ci
->i_ceph_lock
);
1527 truncate_inode_pages(inode
->i_mapping
, 0);
1529 spin_lock(&ci
->i_ceph_lock
);
1530 if (orig_gen
== ci
->i_rdcache_gen
&&
1531 orig_gen
== ci
->i_rdcache_revoking
) {
1532 dout("invalidate_pages %p gen %d successful\n", inode
,
1534 ci
->i_rdcache_revoking
--;
1537 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1538 inode
, orig_gen
, ci
->i_rdcache_gen
,
1539 ci
->i_rdcache_revoking
);
1540 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1543 spin_unlock(&ci
->i_ceph_lock
);
1544 mutex_unlock(&ci
->i_truncate_mutex
);
1547 ceph_check_caps(ci
, 0, NULL
);
1553 * called by trunc_wq;
1555 * We also truncate in a separate thread as well.
1557 static void ceph_vmtruncate_work(struct work_struct
*work
)
1559 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1561 struct inode
*inode
= &ci
->vfs_inode
;
1563 dout("vmtruncate_work %p\n", inode
);
1564 __ceph_do_pending_vmtruncate(inode
);
1569 * Queue an async vmtruncate. If we fail to queue work, we will handle
1570 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1572 void ceph_queue_vmtruncate(struct inode
*inode
)
1574 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1578 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1579 &ci
->i_vmtruncate_work
)) {
1580 dout("ceph_queue_vmtruncate %p\n", inode
);
1582 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1583 inode
, ci
->i_truncate_pending
);
1589 * Make sure any pending truncation is applied before doing anything
1590 * that may depend on it.
1592 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1594 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1596 int wrbuffer_refs
, finish
= 0;
1598 mutex_lock(&ci
->i_truncate_mutex
);
1600 spin_lock(&ci
->i_ceph_lock
);
1601 if (ci
->i_truncate_pending
== 0) {
1602 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1603 spin_unlock(&ci
->i_ceph_lock
);
1604 mutex_unlock(&ci
->i_truncate_mutex
);
1609 * make sure any dirty snapped pages are flushed before we
1610 * possibly truncate them.. so write AND block!
1612 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1613 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1615 spin_unlock(&ci
->i_ceph_lock
);
1616 filemap_write_and_wait_range(&inode
->i_data
, 0,
1617 inode
->i_sb
->s_maxbytes
);
1621 /* there should be no reader or writer */
1622 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1624 to
= ci
->i_truncate_size
;
1625 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1626 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1627 ci
->i_truncate_pending
, to
);
1628 spin_unlock(&ci
->i_ceph_lock
);
1630 truncate_inode_pages(inode
->i_mapping
, to
);
1632 spin_lock(&ci
->i_ceph_lock
);
1633 if (to
== ci
->i_truncate_size
) {
1634 ci
->i_truncate_pending
= 0;
1637 spin_unlock(&ci
->i_ceph_lock
);
1641 mutex_unlock(&ci
->i_truncate_mutex
);
1643 if (wrbuffer_refs
== 0)
1644 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1646 wake_up_all(&ci
->i_cap_wq
);
1652 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1654 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1655 nd_set_link(nd
, ci
->i_symlink
);
1659 static const struct inode_operations ceph_symlink_iops
= {
1660 .readlink
= generic_readlink
,
1661 .follow_link
= ceph_sym_follow_link
,
1662 .setattr
= ceph_setattr
,
1663 .getattr
= ceph_getattr
,
1664 .setxattr
= ceph_setxattr
,
1665 .getxattr
= ceph_getxattr
,
1666 .listxattr
= ceph_listxattr
,
1667 .removexattr
= ceph_removexattr
,
1673 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1675 struct inode
*inode
= dentry
->d_inode
;
1676 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1677 const unsigned int ia_valid
= attr
->ia_valid
;
1678 struct ceph_mds_request
*req
;
1679 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1681 int release
= 0, dirtied
= 0;
1684 int inode_dirty_flags
= 0;
1686 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1689 err
= inode_change_ok(inode
, attr
);
1693 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1696 return PTR_ERR(req
);
1698 spin_lock(&ci
->i_ceph_lock
);
1699 issued
= __ceph_caps_issued(ci
, NULL
);
1700 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1702 if (ia_valid
& ATTR_UID
) {
1703 dout("setattr %p uid %d -> %d\n", inode
,
1704 from_kuid(&init_user_ns
, inode
->i_uid
),
1705 from_kuid(&init_user_ns
, attr
->ia_uid
));
1706 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1707 inode
->i_uid
= attr
->ia_uid
;
1708 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1709 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1710 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1711 req
->r_args
.setattr
.uid
= cpu_to_le32(
1712 from_kuid(&init_user_ns
, attr
->ia_uid
));
1713 mask
|= CEPH_SETATTR_UID
;
1714 release
|= CEPH_CAP_AUTH_SHARED
;
1717 if (ia_valid
& ATTR_GID
) {
1718 dout("setattr %p gid %d -> %d\n", inode
,
1719 from_kgid(&init_user_ns
, inode
->i_gid
),
1720 from_kgid(&init_user_ns
, attr
->ia_gid
));
1721 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1722 inode
->i_gid
= attr
->ia_gid
;
1723 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1724 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1725 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1726 req
->r_args
.setattr
.gid
= cpu_to_le32(
1727 from_kgid(&init_user_ns
, attr
->ia_gid
));
1728 mask
|= CEPH_SETATTR_GID
;
1729 release
|= CEPH_CAP_AUTH_SHARED
;
1732 if (ia_valid
& ATTR_MODE
) {
1733 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1735 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1736 inode
->i_mode
= attr
->ia_mode
;
1737 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1738 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1739 attr
->ia_mode
!= inode
->i_mode
) {
1740 inode
->i_mode
= attr
->ia_mode
;
1741 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1742 mask
|= CEPH_SETATTR_MODE
;
1743 release
|= CEPH_CAP_AUTH_SHARED
;
1747 if (ia_valid
& ATTR_ATIME
) {
1748 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1749 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1750 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1751 if (issued
& CEPH_CAP_FILE_EXCL
) {
1752 ci
->i_time_warp_seq
++;
1753 inode
->i_atime
= attr
->ia_atime
;
1754 dirtied
|= CEPH_CAP_FILE_EXCL
;
1755 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1756 timespec_compare(&inode
->i_atime
,
1757 &attr
->ia_atime
) < 0) {
1758 inode
->i_atime
= attr
->ia_atime
;
1759 dirtied
|= CEPH_CAP_FILE_WR
;
1760 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1761 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1762 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1764 mask
|= CEPH_SETATTR_ATIME
;
1765 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1769 if (ia_valid
& ATTR_MTIME
) {
1770 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1771 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1772 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1773 if (issued
& CEPH_CAP_FILE_EXCL
) {
1774 ci
->i_time_warp_seq
++;
1775 inode
->i_mtime
= attr
->ia_mtime
;
1776 dirtied
|= CEPH_CAP_FILE_EXCL
;
1777 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1778 timespec_compare(&inode
->i_mtime
,
1779 &attr
->ia_mtime
) < 0) {
1780 inode
->i_mtime
= attr
->ia_mtime
;
1781 dirtied
|= CEPH_CAP_FILE_WR
;
1782 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1783 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1784 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1786 mask
|= CEPH_SETATTR_MTIME
;
1787 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1791 if (ia_valid
& ATTR_SIZE
) {
1792 dout("setattr %p size %lld -> %lld\n", inode
,
1793 inode
->i_size
, attr
->ia_size
);
1794 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1798 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1799 attr
->ia_size
> inode
->i_size
) {
1800 inode
->i_size
= attr
->ia_size
;
1802 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1803 inode
->i_ctime
= attr
->ia_ctime
;
1804 ci
->i_reported_size
= attr
->ia_size
;
1805 dirtied
|= CEPH_CAP_FILE_EXCL
;
1806 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1807 attr
->ia_size
!= inode
->i_size
) {
1808 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1809 req
->r_args
.setattr
.old_size
=
1810 cpu_to_le64(inode
->i_size
);
1811 mask
|= CEPH_SETATTR_SIZE
;
1812 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1817 /* these do nothing */
1818 if (ia_valid
& ATTR_CTIME
) {
1819 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1820 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1821 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1822 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1823 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1824 only
? "ctime only" : "ignored");
1825 inode
->i_ctime
= attr
->ia_ctime
;
1828 * if kernel wants to dirty ctime but nothing else,
1829 * we need to choose a cap to dirty under, or do
1830 * a almost-no-op setattr
1832 if (issued
& CEPH_CAP_AUTH_EXCL
)
1833 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1834 else if (issued
& CEPH_CAP_FILE_EXCL
)
1835 dirtied
|= CEPH_CAP_FILE_EXCL
;
1836 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1837 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1839 mask
|= CEPH_SETATTR_CTIME
;
1842 if (ia_valid
& ATTR_FILE
)
1843 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1846 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
);
1847 inode
->i_ctime
= CURRENT_TIME
;
1851 spin_unlock(&ci
->i_ceph_lock
);
1853 if (inode_dirty_flags
)
1854 __mark_inode_dirty(inode
, inode_dirty_flags
);
1856 if (ia_valid
& ATTR_MODE
) {
1857 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
1863 req
->r_inode
= inode
;
1865 req
->r_inode_drop
= release
;
1866 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1867 req
->r_num_caps
= 1;
1868 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1870 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1871 ceph_cap_string(dirtied
), mask
);
1873 ceph_mdsc_put_request(req
);
1874 if (mask
& CEPH_SETATTR_SIZE
)
1875 __ceph_do_pending_vmtruncate(inode
);
1878 spin_unlock(&ci
->i_ceph_lock
);
1880 ceph_mdsc_put_request(req
);
1885 * Verify that we have a lease on the given mask. If not,
1886 * do a getattr against an mds.
1888 int ceph_do_getattr(struct inode
*inode
, int mask
)
1890 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1891 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1892 struct ceph_mds_request
*req
;
1895 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1896 dout("do_getattr inode %p SNAPDIR\n", inode
);
1900 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1901 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1904 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1906 return PTR_ERR(req
);
1907 req
->r_inode
= inode
;
1909 req
->r_num_caps
= 1;
1910 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1911 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1912 ceph_mdsc_put_request(req
);
1913 dout("do_getattr result=%d\n", err
);
1919 * Check inode permissions. We verify we have a valid value for
1920 * the AUTH cap, then call the generic handler.
1922 int ceph_permission(struct inode
*inode
, int mask
)
1926 if (mask
& MAY_NOT_BLOCK
)
1929 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1932 err
= generic_permission(inode
, mask
);
1937 * Get all attributes. Hopefully somedata we'll have a statlite()
1938 * and can limit the fields we require to be accurate.
1940 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1943 struct inode
*inode
= dentry
->d_inode
;
1944 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1947 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1949 generic_fillattr(inode
, stat
);
1950 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
1951 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1952 stat
->dev
= ceph_snap(inode
);
1955 if (S_ISDIR(inode
->i_mode
)) {
1956 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
1958 stat
->size
= ci
->i_rbytes
;
1960 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
1962 stat
->blksize
= 65536;