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
));
749 /* only update max_size on auth cap */
750 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
751 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
752 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
753 le64_to_cpu(info
->max_size
));
754 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
757 spin_unlock(&ci
->i_ceph_lock
);
759 /* queue truncate if we saw i_size decrease */
761 ceph_queue_vmtruncate(inode
);
763 /* populate frag tree */
764 /* FIXME: move me up, if/when version reflects fragtree changes */
765 nsplits
= le32_to_cpu(info
->fragtree
.nsplits
);
766 mutex_lock(&ci
->i_fragtree_mutex
);
767 rb_node
= rb_first(&ci
->i_fragtree
);
768 for (i
= 0; i
< nsplits
; i
++) {
769 u32 id
= le32_to_cpu(info
->fragtree
.splits
[i
].frag
);
772 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
773 if (ceph_frag_compare(frag
->frag
, id
) >= 0) {
774 if (frag
->frag
!= id
)
777 rb_node
= rb_next(rb_node
);
780 rb_node
= rb_next(rb_node
);
781 rb_erase(&frag
->node
, &ci
->i_fragtree
);
786 frag
= __get_or_create_frag(ci
, id
);
790 frag
->split_by
= le32_to_cpu(info
->fragtree
.splits
[i
].by
);
791 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
794 frag
= rb_entry(rb_node
, struct ceph_inode_frag
, node
);
795 rb_node
= rb_next(rb_node
);
796 rb_erase(&frag
->node
, &ci
->i_fragtree
);
799 mutex_unlock(&ci
->i_fragtree_mutex
);
801 /* were we issued a capability? */
802 if (info
->cap
.caps
) {
803 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
804 ceph_add_cap(inode
, session
,
805 le64_to_cpu(info
->cap
.cap_id
),
807 le32_to_cpu(info
->cap
.caps
),
808 le32_to_cpu(info
->cap
.wanted
),
809 le32_to_cpu(info
->cap
.seq
),
810 le32_to_cpu(info
->cap
.mseq
),
811 le64_to_cpu(info
->cap
.realm
),
815 spin_lock(&ci
->i_ceph_lock
);
816 dout(" %p got snap_caps %s\n", inode
,
817 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
818 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
820 __ceph_get_fmode(ci
, cap_fmode
);
821 spin_unlock(&ci
->i_ceph_lock
);
823 } else if (cap_fmode
>= 0) {
824 pr_warn("mds issued no caps on %llx.%llx\n",
826 __ceph_get_fmode(ci
, cap_fmode
);
829 /* update delegation info? */
831 ceph_fill_dirfrag(inode
, dirinfo
);
837 ceph_buffer_put(xattr_blob
);
842 * caller should hold session s_mutex.
844 static void update_dentry_lease(struct dentry
*dentry
,
845 struct ceph_mds_reply_lease
*lease
,
846 struct ceph_mds_session
*session
,
847 unsigned long from_time
)
849 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
850 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
851 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
852 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
855 /* only track leases on regular dentries */
856 if (dentry
->d_op
!= &ceph_dentry_ops
)
859 spin_lock(&dentry
->d_lock
);
860 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
861 dentry
, duration
, ttl
);
863 /* make lease_rdcache_gen match directory */
864 dir
= dentry
->d_parent
->d_inode
;
865 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
870 if (di
->lease_gen
== session
->s_cap_gen
&&
871 time_before(ttl
, dentry
->d_time
))
872 goto out_unlock
; /* we already have a newer lease. */
874 if (di
->lease_session
&& di
->lease_session
!= session
)
877 ceph_dentry_lru_touch(dentry
);
879 if (!di
->lease_session
)
880 di
->lease_session
= ceph_get_mds_session(session
);
881 di
->lease_gen
= session
->s_cap_gen
;
882 di
->lease_seq
= le32_to_cpu(lease
->seq
);
883 di
->lease_renew_after
= half_ttl
;
884 di
->lease_renew_from
= 0;
885 dentry
->d_time
= ttl
;
887 spin_unlock(&dentry
->d_lock
);
892 * splice a dentry to an inode.
893 * caller must hold directory i_mutex for this to be safe.
895 * we will only rehash the resulting dentry if @prehash is
896 * true; @prehash will be set to false (for the benefit of
897 * the caller) if we fail.
899 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
902 struct dentry
*realdn
;
906 /* dn must be unhashed */
909 realdn
= d_materialise_unique(dn
, in
);
910 if (IS_ERR(realdn
)) {
911 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
912 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
914 *prehash
= false; /* don't rehash on error */
915 dn
= realdn
; /* note realdn contains the error */
918 dout("dn %p (%d) spliced with %p (%d) "
919 "inode %p ino %llx.%llx\n",
921 realdn
, d_count(realdn
),
922 realdn
->d_inode
, ceph_vinop(realdn
->d_inode
));
926 BUG_ON(!ceph_dentry(dn
));
927 dout("dn %p attached to %p ino %llx.%llx\n",
928 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
));
930 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
937 * Incorporate results into the local cache. This is either just
938 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
941 * A reply may contain
942 * a directory inode along with a dentry.
943 * and/or a target inode
945 * Called with snap_rwsem (read).
947 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
948 struct ceph_mds_session
*session
)
950 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
951 struct inode
*in
= NULL
;
952 struct ceph_vino vino
;
953 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
956 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
957 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
963 * If we resend completed ops to a recovering mds, we get no
964 * trace. Since that is very rare, pretend this is the case
965 * to ensure the 'no trace' handlers in the callers behave.
967 * Fill in inodes unconditionally to avoid breaking cap
970 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
971 pr_info("fill_trace faking empty trace on %lld %s\n",
972 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
973 if (rinfo
->head
->is_dentry
) {
974 rinfo
->head
->is_dentry
= 0;
975 err
= fill_inode(req
->r_locked_dir
,
976 &rinfo
->diri
, rinfo
->dirfrag
,
977 session
, req
->r_request_started
, -1);
979 if (rinfo
->head
->is_target
) {
980 rinfo
->head
->is_target
= 0;
981 ininfo
= rinfo
->targeti
.in
;
982 vino
.ino
= le64_to_cpu(ininfo
->ino
);
983 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
984 in
= ceph_get_inode(sb
, vino
);
985 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
986 session
, req
->r_request_started
,
993 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
994 dout("fill_trace reply is empty!\n");
995 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
996 ceph_invalidate_dir_request(req
);
1000 if (rinfo
->head
->is_dentry
) {
1001 struct inode
*dir
= req
->r_locked_dir
;
1004 err
= fill_inode(dir
, &rinfo
->diri
, rinfo
->dirfrag
,
1005 session
, req
->r_request_started
, -1,
1006 &req
->r_caps_reservation
);
1013 if (dir
&& req
->r_op
== CEPH_MDS_OP_LOOKUPNAME
) {
1015 struct dentry
*dn
, *parent
;
1017 BUG_ON(!rinfo
->head
->is_target
);
1018 BUG_ON(req
->r_dentry
);
1020 parent
= d_find_any_alias(dir
);
1023 dname
.name
= rinfo
->dname
;
1024 dname
.len
= rinfo
->dname_len
;
1025 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1026 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1027 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1029 dn
= d_lookup(parent
, &dname
);
1030 dout("d_lookup on parent=%p name=%.*s got %p\n",
1031 parent
, dname
.len
, dname
.name
, dn
);
1034 dn
= d_alloc(parent
, &dname
);
1035 dout("d_alloc %p '%.*s' = %p\n", parent
,
1036 dname
.len
, dname
.name
, dn
);
1042 err
= ceph_init_dentry(dn
);
1048 } else if (dn
->d_inode
&&
1049 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1050 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1051 dout(" dn %p points to wrong inode %p\n",
1063 if (rinfo
->head
->is_target
) {
1064 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1065 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1067 in
= ceph_get_inode(sb
, vino
);
1072 req
->r_target_inode
= in
;
1074 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1075 session
, req
->r_request_started
,
1076 (!req
->r_aborted
&& rinfo
->head
->result
== 0) ?
1078 &req
->r_caps_reservation
);
1080 pr_err("fill_inode badness %p %llx.%llx\n",
1081 in
, ceph_vinop(in
));
1087 * ignore null lease/binding on snapdir ENOENT, or else we
1088 * will have trouble splicing in the virtual snapdir later
1090 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1091 req
->r_locked_dir
&&
1092 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1093 fsc
->mount_options
->snapdir_name
,
1094 req
->r_dentry
->d_name
.len
))) {
1096 * lookup link rename : null -> possibly existing inode
1097 * mknod symlink mkdir : null -> new inode
1098 * unlink : linked -> null
1100 struct inode
*dir
= req
->r_locked_dir
;
1101 struct dentry
*dn
= req
->r_dentry
;
1102 bool have_dir_cap
, have_lease
;
1106 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
1107 BUG_ON(ceph_ino(dir
) !=
1108 le64_to_cpu(rinfo
->diri
.in
->ino
));
1109 BUG_ON(ceph_snap(dir
) !=
1110 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1112 /* do we have a lease on the whole dir? */
1114 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1115 CEPH_CAP_FILE_SHARED
);
1117 /* do we have a dn lease? */
1118 have_lease
= have_dir_cap
||
1119 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1121 dout("fill_trace no dentry lease or dir cap\n");
1124 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1125 struct inode
*olddir
= req
->r_old_dentry_dir
;
1128 dout(" src %p '%.*s' dst %p '%.*s'\n",
1130 req
->r_old_dentry
->d_name
.len
,
1131 req
->r_old_dentry
->d_name
.name
,
1132 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1133 dout("fill_trace doing d_move %p -> %p\n",
1134 req
->r_old_dentry
, dn
);
1136 d_move(req
->r_old_dentry
, dn
);
1137 dout(" src %p '%.*s' dst %p '%.*s'\n",
1139 req
->r_old_dentry
->d_name
.len
,
1140 req
->r_old_dentry
->d_name
.name
,
1141 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1143 /* ensure target dentry is invalidated, despite
1144 rehashing bug in vfs_rename_dir */
1145 ceph_invalidate_dentry_lease(dn
);
1147 /* d_move screws up sibling dentries' offsets */
1148 ceph_dir_clear_complete(dir
);
1149 ceph_dir_clear_complete(olddir
);
1151 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1152 ceph_dentry(req
->r_old_dentry
)->offset
);
1154 dn
= req
->r_old_dentry
; /* use old_dentry */
1158 if (!rinfo
->head
->is_target
) {
1159 dout("fill_trace null dentry\n");
1161 dout("d_delete %p\n", dn
);
1164 dout("d_instantiate %p NULL\n", dn
);
1165 d_instantiate(dn
, NULL
);
1166 if (have_lease
&& d_unhashed(dn
))
1168 update_dentry_lease(dn
, rinfo
->dlease
,
1170 req
->r_request_started
);
1175 /* attach proper inode */
1177 ceph_dir_clear_complete(dir
);
1179 dn
= splice_dentry(dn
, in
, &have_lease
);
1184 req
->r_dentry
= dn
; /* may have spliced */
1185 } else if (dn
->d_inode
&& dn
->d_inode
!= in
) {
1186 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1187 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
),
1193 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1194 req
->r_request_started
);
1195 dout(" final dn %p\n", dn
);
1196 } else if (!req
->r_aborted
&&
1197 (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1198 req
->r_op
== CEPH_MDS_OP_MKSNAP
)) {
1199 struct dentry
*dn
= req
->r_dentry
;
1200 struct inode
*dir
= req
->r_locked_dir
;
1202 /* fill out a snapdir LOOKUPSNAP dentry */
1205 BUG_ON(ceph_snap(dir
) != CEPH_SNAPDIR
);
1206 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1207 ceph_dir_clear_complete(dir
);
1209 dn
= splice_dentry(dn
, in
, NULL
);
1214 req
->r_dentry
= dn
; /* may have spliced */
1217 dout("fill_trace done err=%d\n", err
);
1222 * Prepopulate our cache with readdir results, leases, etc.
1224 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1225 struct ceph_mds_session
*session
)
1227 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1230 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1231 struct ceph_vino vino
;
1235 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1236 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1238 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1241 dout("new_inode badness got %d\n", err
);
1244 rc
= fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1245 req
->r_request_started
, -1,
1246 &req
->r_caps_reservation
);
1248 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1257 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1258 struct ceph_mds_session
*session
)
1260 struct dentry
*parent
= req
->r_dentry
;
1261 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1265 int err
= 0, ret
, i
;
1266 struct inode
*snapdir
= NULL
;
1267 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1268 struct ceph_dentry_info
*di
;
1269 u64 r_readdir_offset
= req
->r_readdir_offset
;
1270 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1272 if (rinfo
->dir_dir
&&
1273 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1274 dout("readdir_prepopulate got new frag %x -> %x\n",
1275 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1276 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1277 if (ceph_frag_is_leftmost(frag
))
1278 r_readdir_offset
= 2;
1280 r_readdir_offset
= 0;
1284 return readdir_prepopulate_inodes_only(req
, session
);
1286 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1287 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1288 parent
= d_find_alias(snapdir
);
1289 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1290 rinfo
->dir_nr
, parent
);
1292 dout("readdir_prepopulate %d items under dn %p\n",
1293 rinfo
->dir_nr
, parent
);
1295 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1298 /* FIXME: release caps/leases if error occurs */
1299 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1300 struct ceph_vino vino
;
1302 dname
.name
= rinfo
->dir_dname
[i
];
1303 dname
.len
= rinfo
->dir_dname_len
[i
];
1304 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1306 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1307 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1310 dn
= d_lookup(parent
, &dname
);
1311 dout("d_lookup on parent=%p name=%.*s got %p\n",
1312 parent
, dname
.len
, dname
.name
, dn
);
1315 dn
= d_alloc(parent
, &dname
);
1316 dout("d_alloc %p '%.*s' = %p\n", parent
,
1317 dname
.len
, dname
.name
, dn
);
1319 dout("d_alloc badness\n");
1323 ret
= ceph_init_dentry(dn
);
1329 } else if (dn
->d_inode
&&
1330 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1331 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1332 dout(" dn %p points to wrong inode %p\n",
1338 /* reorder parent's d_subdirs */
1339 spin_lock(&parent
->d_lock
);
1340 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
1341 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1342 spin_unlock(&dn
->d_lock
);
1343 spin_unlock(&parent
->d_lock
);
1350 in
= ceph_get_inode(parent
->d_sb
, vino
);
1352 dout("new_inode badness\n");
1360 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1361 req
->r_request_started
, -1,
1362 &req
->r_caps_reservation
) < 0) {
1363 pr_err("fill_inode badness on %p\n", in
);
1371 dn
= splice_dentry(dn
, in
, NULL
);
1380 di
->offset
= ceph_make_fpos(frag
, i
+ r_readdir_offset
);
1382 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1384 req
->r_request_started
);
1390 req
->r_did_prepopulate
= true;
1397 dout("readdir_prepopulate done\n");
1401 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1403 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1406 spin_lock(&ci
->i_ceph_lock
);
1407 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1408 inode
->i_size
= size
;
1409 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1411 /* tell the MDS if we are approaching max_size */
1412 if ((size
<< 1) >= ci
->i_max_size
&&
1413 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1416 spin_unlock(&ci
->i_ceph_lock
);
1421 * Write back inode data in a worker thread. (This can't be done
1422 * in the message handler context.)
1424 void ceph_queue_writeback(struct inode
*inode
)
1427 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1428 &ceph_inode(inode
)->i_wb_work
)) {
1429 dout("ceph_queue_writeback %p\n", inode
);
1431 dout("ceph_queue_writeback %p failed\n", inode
);
1436 static void ceph_writeback_work(struct work_struct
*work
)
1438 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1440 struct inode
*inode
= &ci
->vfs_inode
;
1442 dout("writeback %p\n", inode
);
1443 filemap_fdatawrite(&inode
->i_data
);
1448 * queue an async invalidation
1450 void ceph_queue_invalidate(struct inode
*inode
)
1453 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1454 &ceph_inode(inode
)->i_pg_inv_work
)) {
1455 dout("ceph_queue_invalidate %p\n", inode
);
1457 dout("ceph_queue_invalidate %p failed\n", inode
);
1463 * Invalidate inode pages in a worker thread. (This can't be done
1464 * in the message handler context.)
1466 static void ceph_invalidate_work(struct work_struct
*work
)
1468 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1470 struct inode
*inode
= &ci
->vfs_inode
;
1474 mutex_lock(&ci
->i_truncate_mutex
);
1475 spin_lock(&ci
->i_ceph_lock
);
1476 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1477 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1478 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1479 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1481 spin_unlock(&ci
->i_ceph_lock
);
1482 mutex_unlock(&ci
->i_truncate_mutex
);
1485 orig_gen
= ci
->i_rdcache_gen
;
1486 spin_unlock(&ci
->i_ceph_lock
);
1488 truncate_inode_pages(inode
->i_mapping
, 0);
1490 spin_lock(&ci
->i_ceph_lock
);
1491 if (orig_gen
== ci
->i_rdcache_gen
&&
1492 orig_gen
== ci
->i_rdcache_revoking
) {
1493 dout("invalidate_pages %p gen %d successful\n", inode
,
1495 ci
->i_rdcache_revoking
--;
1498 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1499 inode
, orig_gen
, ci
->i_rdcache_gen
,
1500 ci
->i_rdcache_revoking
);
1501 if (__ceph_caps_revoking_other(ci
, NULL
, CEPH_CAP_FILE_CACHE
))
1504 spin_unlock(&ci
->i_ceph_lock
);
1505 mutex_unlock(&ci
->i_truncate_mutex
);
1508 ceph_check_caps(ci
, 0, NULL
);
1514 * called by trunc_wq;
1516 * We also truncate in a separate thread as well.
1518 static void ceph_vmtruncate_work(struct work_struct
*work
)
1520 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1522 struct inode
*inode
= &ci
->vfs_inode
;
1524 dout("vmtruncate_work %p\n", inode
);
1525 __ceph_do_pending_vmtruncate(inode
);
1530 * Queue an async vmtruncate. If we fail to queue work, we will handle
1531 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1533 void ceph_queue_vmtruncate(struct inode
*inode
)
1535 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1539 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1540 &ci
->i_vmtruncate_work
)) {
1541 dout("ceph_queue_vmtruncate %p\n", inode
);
1543 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1544 inode
, ci
->i_truncate_pending
);
1550 * Make sure any pending truncation is applied before doing anything
1551 * that may depend on it.
1553 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1555 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1557 int wrbuffer_refs
, finish
= 0;
1559 mutex_lock(&ci
->i_truncate_mutex
);
1561 spin_lock(&ci
->i_ceph_lock
);
1562 if (ci
->i_truncate_pending
== 0) {
1563 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1564 spin_unlock(&ci
->i_ceph_lock
);
1565 mutex_unlock(&ci
->i_truncate_mutex
);
1570 * make sure any dirty snapped pages are flushed before we
1571 * possibly truncate them.. so write AND block!
1573 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1574 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1576 spin_unlock(&ci
->i_ceph_lock
);
1577 filemap_write_and_wait_range(&inode
->i_data
, 0,
1578 inode
->i_sb
->s_maxbytes
);
1582 /* there should be no reader or writer */
1583 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1585 to
= ci
->i_truncate_size
;
1586 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1587 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1588 ci
->i_truncate_pending
, to
);
1589 spin_unlock(&ci
->i_ceph_lock
);
1591 truncate_inode_pages(inode
->i_mapping
, to
);
1593 spin_lock(&ci
->i_ceph_lock
);
1594 if (to
== ci
->i_truncate_size
) {
1595 ci
->i_truncate_pending
= 0;
1598 spin_unlock(&ci
->i_ceph_lock
);
1602 mutex_unlock(&ci
->i_truncate_mutex
);
1604 if (wrbuffer_refs
== 0)
1605 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1607 wake_up_all(&ci
->i_cap_wq
);
1613 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1615 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1616 nd_set_link(nd
, ci
->i_symlink
);
1620 static const struct inode_operations ceph_symlink_iops
= {
1621 .readlink
= generic_readlink
,
1622 .follow_link
= ceph_sym_follow_link
,
1623 .setattr
= ceph_setattr
,
1624 .getattr
= ceph_getattr
,
1625 .setxattr
= ceph_setxattr
,
1626 .getxattr
= ceph_getxattr
,
1627 .listxattr
= ceph_listxattr
,
1628 .removexattr
= ceph_removexattr
,
1634 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1636 struct inode
*inode
= dentry
->d_inode
;
1637 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1638 const unsigned int ia_valid
= attr
->ia_valid
;
1639 struct ceph_mds_request
*req
;
1640 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1642 int release
= 0, dirtied
= 0;
1645 int inode_dirty_flags
= 0;
1647 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1650 err
= inode_change_ok(inode
, attr
);
1654 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1657 return PTR_ERR(req
);
1659 spin_lock(&ci
->i_ceph_lock
);
1660 issued
= __ceph_caps_issued(ci
, NULL
);
1661 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1663 if (ia_valid
& ATTR_UID
) {
1664 dout("setattr %p uid %d -> %d\n", inode
,
1665 from_kuid(&init_user_ns
, inode
->i_uid
),
1666 from_kuid(&init_user_ns
, attr
->ia_uid
));
1667 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1668 inode
->i_uid
= attr
->ia_uid
;
1669 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1670 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1671 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1672 req
->r_args
.setattr
.uid
= cpu_to_le32(
1673 from_kuid(&init_user_ns
, attr
->ia_uid
));
1674 mask
|= CEPH_SETATTR_UID
;
1675 release
|= CEPH_CAP_AUTH_SHARED
;
1678 if (ia_valid
& ATTR_GID
) {
1679 dout("setattr %p gid %d -> %d\n", inode
,
1680 from_kgid(&init_user_ns
, inode
->i_gid
),
1681 from_kgid(&init_user_ns
, attr
->ia_gid
));
1682 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1683 inode
->i_gid
= attr
->ia_gid
;
1684 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1685 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1686 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1687 req
->r_args
.setattr
.gid
= cpu_to_le32(
1688 from_kgid(&init_user_ns
, attr
->ia_gid
));
1689 mask
|= CEPH_SETATTR_GID
;
1690 release
|= CEPH_CAP_AUTH_SHARED
;
1693 if (ia_valid
& ATTR_MODE
) {
1694 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1696 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1697 inode
->i_mode
= attr
->ia_mode
;
1698 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1699 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1700 attr
->ia_mode
!= inode
->i_mode
) {
1701 inode
->i_mode
= attr
->ia_mode
;
1702 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1703 mask
|= CEPH_SETATTR_MODE
;
1704 release
|= CEPH_CAP_AUTH_SHARED
;
1708 if (ia_valid
& ATTR_ATIME
) {
1709 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1710 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1711 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1712 if (issued
& CEPH_CAP_FILE_EXCL
) {
1713 ci
->i_time_warp_seq
++;
1714 inode
->i_atime
= attr
->ia_atime
;
1715 dirtied
|= CEPH_CAP_FILE_EXCL
;
1716 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1717 timespec_compare(&inode
->i_atime
,
1718 &attr
->ia_atime
) < 0) {
1719 inode
->i_atime
= attr
->ia_atime
;
1720 dirtied
|= CEPH_CAP_FILE_WR
;
1721 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1722 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1723 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1725 mask
|= CEPH_SETATTR_ATIME
;
1726 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1730 if (ia_valid
& ATTR_MTIME
) {
1731 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1732 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1733 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1734 if (issued
& CEPH_CAP_FILE_EXCL
) {
1735 ci
->i_time_warp_seq
++;
1736 inode
->i_mtime
= attr
->ia_mtime
;
1737 dirtied
|= CEPH_CAP_FILE_EXCL
;
1738 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1739 timespec_compare(&inode
->i_mtime
,
1740 &attr
->ia_mtime
) < 0) {
1741 inode
->i_mtime
= attr
->ia_mtime
;
1742 dirtied
|= CEPH_CAP_FILE_WR
;
1743 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1744 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1745 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1747 mask
|= CEPH_SETATTR_MTIME
;
1748 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1752 if (ia_valid
& ATTR_SIZE
) {
1753 dout("setattr %p size %lld -> %lld\n", inode
,
1754 inode
->i_size
, attr
->ia_size
);
1755 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1759 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1760 attr
->ia_size
> inode
->i_size
) {
1761 inode
->i_size
= attr
->ia_size
;
1763 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1764 inode
->i_ctime
= attr
->ia_ctime
;
1765 ci
->i_reported_size
= attr
->ia_size
;
1766 dirtied
|= CEPH_CAP_FILE_EXCL
;
1767 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1768 attr
->ia_size
!= inode
->i_size
) {
1769 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1770 req
->r_args
.setattr
.old_size
=
1771 cpu_to_le64(inode
->i_size
);
1772 mask
|= CEPH_SETATTR_SIZE
;
1773 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1778 /* these do nothing */
1779 if (ia_valid
& ATTR_CTIME
) {
1780 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1781 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1782 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1783 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1784 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1785 only
? "ctime only" : "ignored");
1786 inode
->i_ctime
= attr
->ia_ctime
;
1789 * if kernel wants to dirty ctime but nothing else,
1790 * we need to choose a cap to dirty under, or do
1791 * a almost-no-op setattr
1793 if (issued
& CEPH_CAP_AUTH_EXCL
)
1794 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1795 else if (issued
& CEPH_CAP_FILE_EXCL
)
1796 dirtied
|= CEPH_CAP_FILE_EXCL
;
1797 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1798 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1800 mask
|= CEPH_SETATTR_CTIME
;
1803 if (ia_valid
& ATTR_FILE
)
1804 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1807 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
);
1808 inode
->i_ctime
= CURRENT_TIME
;
1812 spin_unlock(&ci
->i_ceph_lock
);
1814 if (inode_dirty_flags
)
1815 __mark_inode_dirty(inode
, inode_dirty_flags
);
1817 if (ia_valid
& ATTR_MODE
) {
1818 err
= posix_acl_chmod(inode
, attr
->ia_mode
);
1824 req
->r_inode
= inode
;
1826 req
->r_inode_drop
= release
;
1827 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1828 req
->r_num_caps
= 1;
1829 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1831 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1832 ceph_cap_string(dirtied
), mask
);
1834 ceph_mdsc_put_request(req
);
1835 if (mask
& CEPH_SETATTR_SIZE
)
1836 __ceph_do_pending_vmtruncate(inode
);
1839 spin_unlock(&ci
->i_ceph_lock
);
1841 ceph_mdsc_put_request(req
);
1846 * Verify that we have a lease on the given mask. If not,
1847 * do a getattr against an mds.
1849 int ceph_do_getattr(struct inode
*inode
, int mask
)
1851 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1852 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1853 struct ceph_mds_request
*req
;
1856 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1857 dout("do_getattr inode %p SNAPDIR\n", inode
);
1861 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1862 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1865 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1867 return PTR_ERR(req
);
1868 req
->r_inode
= inode
;
1870 req
->r_num_caps
= 1;
1871 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1872 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1873 ceph_mdsc_put_request(req
);
1874 dout("do_getattr result=%d\n", err
);
1880 * Check inode permissions. We verify we have a valid value for
1881 * the AUTH cap, then call the generic handler.
1883 int ceph_permission(struct inode
*inode
, int mask
)
1887 if (mask
& MAY_NOT_BLOCK
)
1890 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1893 err
= generic_permission(inode
, mask
);
1898 * Get all attributes. Hopefully somedata we'll have a statlite()
1899 * and can limit the fields we require to be accurate.
1901 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1904 struct inode
*inode
= dentry
->d_inode
;
1905 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1908 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1910 generic_fillattr(inode
, stat
);
1911 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
1912 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1913 stat
->dev
= ceph_snap(inode
);
1916 if (S_ISDIR(inode
->i_mode
)) {
1917 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
1919 stat
->size
= ci
->i_rbytes
;
1921 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
1923 stat
->blksize
= 65536;