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>
14 #include "mds_client.h"
16 #include <linux/ceph/decode.h>
19 * Ceph inode operations
21 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
22 * setattr, etc.), xattr helpers, and helpers for assimilating
23 * metadata returned by the MDS into our cache.
25 * Also define helpers for doing asynchronous writeback, invalidation,
26 * and truncation for the benefit of those who can't afford to block
27 * (typically because they are in the message handler path).
30 static const struct inode_operations ceph_symlink_iops
;
32 static void ceph_invalidate_work(struct work_struct
*work
);
33 static void ceph_writeback_work(struct work_struct
*work
);
34 static void ceph_vmtruncate_work(struct work_struct
*work
);
37 * find or create an inode, given the ceph ino number
39 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
41 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
42 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
46 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
49 ino_t t
= ceph_vino_to_ino(vino
);
51 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
53 return ERR_PTR(-ENOMEM
);
54 if (inode
->i_state
& I_NEW
) {
55 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
56 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
57 unlock_new_inode(inode
);
60 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
66 * get/constuct snapdir inode for a given directory
68 struct inode
*ceph_get_snapdir(struct inode
*parent
)
70 struct ceph_vino vino
= {
71 .ino
= ceph_ino(parent
),
74 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
75 struct ceph_inode_info
*ci
= ceph_inode(inode
);
77 BUG_ON(!S_ISDIR(parent
->i_mode
));
80 inode
->i_mode
= parent
->i_mode
;
81 inode
->i_uid
= parent
->i_uid
;
82 inode
->i_gid
= parent
->i_gid
;
83 inode
->i_op
= &ceph_dir_iops
;
84 inode
->i_fop
= &ceph_dir_fops
;
85 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
90 const struct inode_operations ceph_file_iops
= {
91 .permission
= ceph_permission
,
92 .setattr
= ceph_setattr
,
93 .getattr
= ceph_getattr
,
94 .setxattr
= ceph_setxattr
,
95 .getxattr
= ceph_getxattr
,
96 .listxattr
= ceph_listxattr
,
97 .removexattr
= ceph_removexattr
,
98 .get_acl
= ceph_get_acl
,
103 * We use a 'frag tree' to keep track of the MDS's directory fragments
104 * for a given inode (usually there is just a single fragment). We
105 * need to know when a child frag is delegated to a new MDS, or when
106 * it is flagged as replicated, so we can direct our requests
111 * find/create a frag in the tree
113 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
117 struct rb_node
*parent
= NULL
;
118 struct ceph_inode_frag
*frag
;
121 p
= &ci
->i_fragtree
.rb_node
;
124 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
125 c
= ceph_frag_compare(f
, frag
->frag
);
134 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
136 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
137 "frag %x\n", &ci
->vfs_inode
,
138 ceph_vinop(&ci
->vfs_inode
), f
);
139 return ERR_PTR(-ENOMEM
);
146 rb_link_node(&frag
->node
, parent
, p
);
147 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
149 dout("get_or_create_frag added %llx.%llx frag %x\n",
150 ceph_vinop(&ci
->vfs_inode
), f
);
155 * find a specific frag @f
157 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
159 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
162 struct ceph_inode_frag
*frag
=
163 rb_entry(n
, struct ceph_inode_frag
, node
);
164 int c
= ceph_frag_compare(f
, frag
->frag
);
176 * Choose frag containing the given value @v. If @pfrag is
177 * specified, copy the frag delegation info to the caller if
180 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
181 struct ceph_inode_frag
*pfrag
,
184 u32 t
= ceph_frag_make(0, 0);
185 struct ceph_inode_frag
*frag
;
192 mutex_lock(&ci
->i_fragtree_mutex
);
194 WARN_ON(!ceph_frag_contains_value(t
, v
));
195 frag
= __ceph_find_frag(ci
, t
);
197 break; /* t is a leaf */
198 if (frag
->split_by
== 0) {
200 memcpy(pfrag
, frag
, sizeof(*pfrag
));
207 nway
= 1 << frag
->split_by
;
208 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
209 frag
->split_by
, nway
);
210 for (i
= 0; i
< nway
; i
++) {
211 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
212 if (ceph_frag_contains_value(n
, v
)) {
219 dout("choose_frag(%x) = %x\n", v
, t
);
221 mutex_unlock(&ci
->i_fragtree_mutex
);
226 * Process dirfrag (delegation) info from the mds. Include leaf
227 * fragment in tree ONLY if ndist > 0. Otherwise, only
228 * branches/splits are included in i_fragtree)
230 static int ceph_fill_dirfrag(struct inode
*inode
,
231 struct ceph_mds_reply_dirfrag
*dirinfo
)
233 struct ceph_inode_info
*ci
= ceph_inode(inode
);
234 struct ceph_inode_frag
*frag
;
235 u32 id
= le32_to_cpu(dirinfo
->frag
);
236 int mds
= le32_to_cpu(dirinfo
->auth
);
237 int ndist
= le32_to_cpu(dirinfo
->ndist
);
241 mutex_lock(&ci
->i_fragtree_mutex
);
243 /* no delegation info needed. */
244 frag
= __ceph_find_frag(ci
, id
);
247 if (frag
->split_by
== 0) {
248 /* tree leaf, remove */
249 dout("fill_dirfrag removed %llx.%llx frag %x"
250 " (no ref)\n", ceph_vinop(inode
), id
);
251 rb_erase(&frag
->node
, &ci
->i_fragtree
);
254 /* tree branch, keep and clear */
255 dout("fill_dirfrag cleared %llx.%llx frag %x"
256 " referral\n", ceph_vinop(inode
), id
);
264 /* find/add this frag to store mds delegation info */
265 frag
= __get_or_create_frag(ci
, id
);
267 /* this is not the end of the world; we can continue
268 with bad/inaccurate delegation info */
269 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
270 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
276 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
277 for (i
= 0; i
< frag
->ndist
; i
++)
278 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
279 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
280 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
283 mutex_unlock(&ci
->i_fragtree_mutex
);
289 * initialize a newly allocated inode.
291 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
293 struct ceph_inode_info
*ci
;
296 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
300 dout("alloc_inode %p\n", &ci
->vfs_inode
);
302 spin_lock_init(&ci
->i_ceph_lock
);
305 ci
->i_time_warp_seq
= 0;
306 ci
->i_ceph_flags
= 0;
307 atomic_set(&ci
->i_release_count
, 1);
308 atomic_set(&ci
->i_complete_count
, 0);
309 ci
->i_symlink
= NULL
;
311 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
313 ci
->i_fragtree
= RB_ROOT
;
314 mutex_init(&ci
->i_fragtree_mutex
);
316 ci
->i_xattrs
.blob
= NULL
;
317 ci
->i_xattrs
.prealloc_blob
= NULL
;
318 ci
->i_xattrs
.dirty
= false;
319 ci
->i_xattrs
.index
= RB_ROOT
;
320 ci
->i_xattrs
.count
= 0;
321 ci
->i_xattrs
.names_size
= 0;
322 ci
->i_xattrs
.vals_size
= 0;
323 ci
->i_xattrs
.version
= 0;
324 ci
->i_xattrs
.index_version
= 0;
326 ci
->i_caps
= RB_ROOT
;
327 ci
->i_auth_cap
= NULL
;
328 ci
->i_dirty_caps
= 0;
329 ci
->i_flushing_caps
= 0;
330 INIT_LIST_HEAD(&ci
->i_dirty_item
);
331 INIT_LIST_HEAD(&ci
->i_flushing_item
);
332 ci
->i_cap_flush_seq
= 0;
333 ci
->i_cap_flush_last_tid
= 0;
334 memset(&ci
->i_cap_flush_tid
, 0, sizeof(ci
->i_cap_flush_tid
));
335 init_waitqueue_head(&ci
->i_cap_wq
);
336 ci
->i_hold_caps_min
= 0;
337 ci
->i_hold_caps_max
= 0;
338 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
339 ci
->i_cap_exporting_mds
= 0;
340 ci
->i_cap_exporting_mseq
= 0;
341 ci
->i_cap_exporting_issued
= 0;
342 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
343 ci
->i_head_snapc
= NULL
;
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 /* only update max_size on auth cap */
663 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
664 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
665 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
666 le64_to_cpu(info
->max_size
));
667 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
670 ci
->i_layout
= info
->layout
;
671 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
674 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
675 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 &&
676 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
677 if (ci
->i_xattrs
.blob
)
678 ceph_buffer_put(ci
->i_xattrs
.blob
);
679 ci
->i_xattrs
.blob
= xattr_blob
;
681 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
682 iinfo
->xattr_data
, iinfo
->xattr_len
);
683 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
684 ceph_forget_all_cached_acls(inode
);
688 inode
->i_mapping
->a_ops
= &ceph_aops
;
689 inode
->i_mapping
->backing_dev_info
=
690 &ceph_sb_to_client(inode
->i_sb
)->backing_dev_info
;
692 switch (inode
->i_mode
& S_IFMT
) {
697 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
698 inode
->i_op
= &ceph_file_iops
;
701 inode
->i_op
= &ceph_file_iops
;
702 inode
->i_fop
= &ceph_file_fops
;
705 inode
->i_op
= &ceph_symlink_iops
;
706 if (!ci
->i_symlink
) {
707 u32 symlen
= iinfo
->symlink_len
;
710 spin_unlock(&ci
->i_ceph_lock
);
713 if (WARN_ON(symlen
!= inode
->i_size
))
717 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
721 spin_lock(&ci
->i_ceph_lock
);
725 kfree(sym
); /* lost a race */
729 inode
->i_op
= &ceph_dir_iops
;
730 inode
->i_fop
= &ceph_dir_fops
;
732 ci
->i_dir_layout
= iinfo
->dir_layout
;
734 ci
->i_files
= le64_to_cpu(info
->files
);
735 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
736 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
737 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
738 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
739 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
742 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
743 ceph_vinop(inode
), inode
->i_mode
);
746 /* set dir completion flag? */
747 if (S_ISDIR(inode
->i_mode
) &&
748 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
749 ceph_snap(inode
) == CEPH_NOSNAP
&&
750 (le32_to_cpu(info
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) &&
751 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
752 !__ceph_dir_is_complete(ci
)) {
753 dout(" marking %p complete (empty)\n", inode
);
754 __ceph_dir_set_complete(ci
, atomic_read(&ci
->i_release_count
));
755 ci
->i_max_offset
= 2;
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
);
1053 if (rinfo
->head
->is_target
) {
1054 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1055 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1057 in
= ceph_get_inode(sb
, vino
);
1062 req
->r_target_inode
= in
;
1064 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
1065 session
, req
->r_request_started
,
1066 (le32_to_cpu(rinfo
->head
->result
) == 0) ?
1068 &req
->r_caps_reservation
);
1070 pr_err("fill_inode badness %p %llx.%llx\n",
1071 in
, ceph_vinop(in
));
1077 * ignore null lease/binding on snapdir ENOENT, or else we
1078 * will have trouble splicing in the virtual snapdir later
1080 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1081 req
->r_locked_dir
&&
1082 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1083 fsc
->mount_options
->snapdir_name
,
1084 req
->r_dentry
->d_name
.len
))) {
1086 * lookup link rename : null -> possibly existing inode
1087 * mknod symlink mkdir : null -> new inode
1088 * unlink : linked -> null
1090 struct inode
*dir
= req
->r_locked_dir
;
1091 struct dentry
*dn
= req
->r_dentry
;
1092 bool have_dir_cap
, have_lease
;
1096 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
1097 BUG_ON(ceph_ino(dir
) !=
1098 le64_to_cpu(rinfo
->diri
.in
->ino
));
1099 BUG_ON(ceph_snap(dir
) !=
1100 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1102 /* do we have a lease on the whole dir? */
1104 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1105 CEPH_CAP_FILE_SHARED
);
1107 /* do we have a dn lease? */
1108 have_lease
= have_dir_cap
||
1109 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1111 dout("fill_trace no dentry lease or dir cap\n");
1114 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1115 dout(" src %p '%.*s' dst %p '%.*s'\n",
1117 req
->r_old_dentry
->d_name
.len
,
1118 req
->r_old_dentry
->d_name
.name
,
1119 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1120 dout("fill_trace doing d_move %p -> %p\n",
1121 req
->r_old_dentry
, dn
);
1123 d_move(req
->r_old_dentry
, dn
);
1124 dout(" src %p '%.*s' dst %p '%.*s'\n",
1126 req
->r_old_dentry
->d_name
.len
,
1127 req
->r_old_dentry
->d_name
.name
,
1128 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1130 /* ensure target dentry is invalidated, despite
1131 rehashing bug in vfs_rename_dir */
1132 ceph_invalidate_dentry_lease(dn
);
1135 * d_move() puts the renamed dentry at the end of
1136 * d_subdirs. We need to assign it an appropriate
1137 * directory offset so we can behave when dir is
1140 ceph_set_dentry_offset(req
->r_old_dentry
);
1141 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1142 ceph_dentry(req
->r_old_dentry
)->offset
);
1144 dn
= req
->r_old_dentry
; /* use old_dentry */
1148 if (!rinfo
->head
->is_target
) {
1149 dout("fill_trace null dentry\n");
1151 dout("d_delete %p\n", dn
);
1154 dout("d_instantiate %p NULL\n", dn
);
1155 d_instantiate(dn
, NULL
);
1156 if (have_lease
&& d_unhashed(dn
))
1158 update_dentry_lease(dn
, rinfo
->dlease
,
1160 req
->r_request_started
);
1165 /* attach proper inode */
1168 dn
= splice_dentry(dn
, in
, &have_lease
, true);
1173 req
->r_dentry
= dn
; /* may have spliced */
1174 } else if (dn
->d_inode
&& dn
->d_inode
!= in
) {
1175 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1176 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
),
1182 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1183 req
->r_request_started
);
1184 dout(" final dn %p\n", dn
);
1185 } else if (!req
->r_aborted
&&
1186 (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1187 req
->r_op
== CEPH_MDS_OP_MKSNAP
)) {
1188 struct dentry
*dn
= req
->r_dentry
;
1190 /* fill out a snapdir LOOKUPSNAP dentry */
1192 BUG_ON(!req
->r_locked_dir
);
1193 BUG_ON(ceph_snap(req
->r_locked_dir
) != CEPH_SNAPDIR
);
1194 ininfo
= rinfo
->targeti
.in
;
1195 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1196 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1197 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1199 dn
= splice_dentry(dn
, in
, NULL
, true);
1204 req
->r_dentry
= dn
; /* may have spliced */
1207 dout("fill_trace done err=%d\n", err
);
1212 * Prepopulate our cache with readdir results, leases, etc.
1214 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1215 struct ceph_mds_session
*session
)
1217 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1220 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1221 struct ceph_vino vino
;
1225 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1226 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1228 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1231 dout("new_inode badness got %d\n", err
);
1234 rc
= fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1235 req
->r_request_started
, -1,
1236 &req
->r_caps_reservation
);
1238 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1247 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1248 struct ceph_mds_session
*session
)
1250 struct dentry
*parent
= req
->r_dentry
;
1251 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1255 int err
= 0, ret
, i
;
1256 struct inode
*snapdir
= NULL
;
1257 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1258 struct ceph_dentry_info
*di
;
1259 u64 r_readdir_offset
= req
->r_readdir_offset
;
1260 u32 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1262 if (rinfo
->dir_dir
&&
1263 le32_to_cpu(rinfo
->dir_dir
->frag
) != frag
) {
1264 dout("readdir_prepopulate got new frag %x -> %x\n",
1265 frag
, le32_to_cpu(rinfo
->dir_dir
->frag
));
1266 frag
= le32_to_cpu(rinfo
->dir_dir
->frag
);
1267 if (ceph_frag_is_leftmost(frag
))
1268 r_readdir_offset
= 2;
1270 r_readdir_offset
= 0;
1274 return readdir_prepopulate_inodes_only(req
, session
);
1276 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1277 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1278 parent
= d_find_alias(snapdir
);
1279 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1280 rinfo
->dir_nr
, parent
);
1282 dout("readdir_prepopulate %d items under dn %p\n",
1283 rinfo
->dir_nr
, parent
);
1285 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1288 /* FIXME: release caps/leases if error occurs */
1289 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1290 struct ceph_vino vino
;
1292 dname
.name
= rinfo
->dir_dname
[i
];
1293 dname
.len
= rinfo
->dir_dname_len
[i
];
1294 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1296 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1297 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1300 dn
= d_lookup(parent
, &dname
);
1301 dout("d_lookup on parent=%p name=%.*s got %p\n",
1302 parent
, dname
.len
, dname
.name
, dn
);
1305 dn
= d_alloc(parent
, &dname
);
1306 dout("d_alloc %p '%.*s' = %p\n", parent
,
1307 dname
.len
, dname
.name
, dn
);
1309 dout("d_alloc badness\n");
1313 ret
= ceph_init_dentry(dn
);
1319 } else if (dn
->d_inode
&&
1320 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1321 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1322 dout(" dn %p points to wrong inode %p\n",
1328 /* reorder parent's d_subdirs */
1329 spin_lock(&parent
->d_lock
);
1330 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
1331 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1332 spin_unlock(&dn
->d_lock
);
1333 spin_unlock(&parent
->d_lock
);
1340 in
= ceph_get_inode(parent
->d_sb
, vino
);
1342 dout("new_inode badness\n");
1350 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1351 req
->r_request_started
, -1,
1352 &req
->r_caps_reservation
) < 0) {
1353 pr_err("fill_inode badness on %p\n", in
);
1361 dn
= splice_dentry(dn
, in
, NULL
, false);
1370 di
->offset
= ceph_make_fpos(frag
, i
+ r_readdir_offset
);
1372 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1374 req
->r_request_started
);
1380 req
->r_did_prepopulate
= true;
1387 dout("readdir_prepopulate done\n");
1391 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1393 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1396 spin_lock(&ci
->i_ceph_lock
);
1397 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1398 inode
->i_size
= size
;
1399 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1401 /* tell the MDS if we are approaching max_size */
1402 if ((size
<< 1) >= ci
->i_max_size
&&
1403 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1406 spin_unlock(&ci
->i_ceph_lock
);
1411 * Write back inode data in a worker thread. (This can't be done
1412 * in the message handler context.)
1414 void ceph_queue_writeback(struct inode
*inode
)
1417 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1418 &ceph_inode(inode
)->i_wb_work
)) {
1419 dout("ceph_queue_writeback %p\n", inode
);
1421 dout("ceph_queue_writeback %p failed\n", inode
);
1426 static void ceph_writeback_work(struct work_struct
*work
)
1428 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1430 struct inode
*inode
= &ci
->vfs_inode
;
1432 dout("writeback %p\n", inode
);
1433 filemap_fdatawrite(&inode
->i_data
);
1438 * queue an async invalidation
1440 void ceph_queue_invalidate(struct inode
*inode
)
1443 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1444 &ceph_inode(inode
)->i_pg_inv_work
)) {
1445 dout("ceph_queue_invalidate %p\n", inode
);
1447 dout("ceph_queue_invalidate %p failed\n", inode
);
1453 * Invalidate inode pages in a worker thread. (This can't be done
1454 * in the message handler context.)
1456 static void ceph_invalidate_work(struct work_struct
*work
)
1458 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1460 struct inode
*inode
= &ci
->vfs_inode
;
1464 mutex_lock(&ci
->i_truncate_mutex
);
1465 spin_lock(&ci
->i_ceph_lock
);
1466 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1467 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1468 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1470 spin_unlock(&ci
->i_ceph_lock
);
1471 mutex_unlock(&ci
->i_truncate_mutex
);
1474 orig_gen
= ci
->i_rdcache_gen
;
1475 spin_unlock(&ci
->i_ceph_lock
);
1477 truncate_inode_pages(inode
->i_mapping
, 0);
1479 spin_lock(&ci
->i_ceph_lock
);
1480 if (orig_gen
== ci
->i_rdcache_gen
&&
1481 orig_gen
== ci
->i_rdcache_revoking
) {
1482 dout("invalidate_pages %p gen %d successful\n", inode
,
1484 ci
->i_rdcache_revoking
--;
1487 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1488 inode
, orig_gen
, ci
->i_rdcache_gen
,
1489 ci
->i_rdcache_revoking
);
1491 spin_unlock(&ci
->i_ceph_lock
);
1492 mutex_unlock(&ci
->i_truncate_mutex
);
1495 ceph_check_caps(ci
, 0, NULL
);
1502 * called by trunc_wq;
1504 * We also truncate in a separate thread as well.
1506 static void ceph_vmtruncate_work(struct work_struct
*work
)
1508 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1510 struct inode
*inode
= &ci
->vfs_inode
;
1512 dout("vmtruncate_work %p\n", inode
);
1513 __ceph_do_pending_vmtruncate(inode
);
1518 * Queue an async vmtruncate. If we fail to queue work, we will handle
1519 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1521 void ceph_queue_vmtruncate(struct inode
*inode
)
1523 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1527 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1528 &ci
->i_vmtruncate_work
)) {
1529 dout("ceph_queue_vmtruncate %p\n", inode
);
1531 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1532 inode
, ci
->i_truncate_pending
);
1538 * Make sure any pending truncation is applied before doing anything
1539 * that may depend on it.
1541 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1543 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1545 int wrbuffer_refs
, finish
= 0;
1547 mutex_lock(&ci
->i_truncate_mutex
);
1549 spin_lock(&ci
->i_ceph_lock
);
1550 if (ci
->i_truncate_pending
== 0) {
1551 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1552 spin_unlock(&ci
->i_ceph_lock
);
1553 mutex_unlock(&ci
->i_truncate_mutex
);
1558 * make sure any dirty snapped pages are flushed before we
1559 * possibly truncate them.. so write AND block!
1561 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1562 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1564 spin_unlock(&ci
->i_ceph_lock
);
1565 filemap_write_and_wait_range(&inode
->i_data
, 0,
1566 inode
->i_sb
->s_maxbytes
);
1570 /* there should be no reader or writer */
1571 WARN_ON_ONCE(ci
->i_rd_ref
|| ci
->i_wr_ref
);
1573 to
= ci
->i_truncate_size
;
1574 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1575 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1576 ci
->i_truncate_pending
, to
);
1577 spin_unlock(&ci
->i_ceph_lock
);
1579 truncate_inode_pages(inode
->i_mapping
, to
);
1581 spin_lock(&ci
->i_ceph_lock
);
1582 if (to
== ci
->i_truncate_size
) {
1583 ci
->i_truncate_pending
= 0;
1586 spin_unlock(&ci
->i_ceph_lock
);
1590 mutex_unlock(&ci
->i_truncate_mutex
);
1592 if (wrbuffer_refs
== 0)
1593 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1595 wake_up_all(&ci
->i_cap_wq
);
1601 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1603 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1604 nd_set_link(nd
, ci
->i_symlink
);
1608 static const struct inode_operations ceph_symlink_iops
= {
1609 .readlink
= generic_readlink
,
1610 .follow_link
= ceph_sym_follow_link
,
1611 .setattr
= ceph_setattr
,
1612 .getattr
= ceph_getattr
,
1613 .setxattr
= ceph_setxattr
,
1614 .getxattr
= ceph_getxattr
,
1615 .listxattr
= ceph_listxattr
,
1616 .removexattr
= ceph_removexattr
,
1617 .get_acl
= ceph_get_acl
,
1623 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1625 struct inode
*inode
= dentry
->d_inode
;
1626 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1627 struct inode
*parent_inode
;
1628 const unsigned int ia_valid
= attr
->ia_valid
;
1629 struct ceph_mds_request
*req
;
1630 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1632 int release
= 0, dirtied
= 0;
1635 int inode_dirty_flags
= 0;
1637 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1640 err
= inode_change_ok(inode
, attr
);
1644 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1647 return PTR_ERR(req
);
1649 spin_lock(&ci
->i_ceph_lock
);
1650 issued
= __ceph_caps_issued(ci
, NULL
);
1651 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1653 if (ia_valid
& ATTR_UID
) {
1654 dout("setattr %p uid %d -> %d\n", inode
,
1655 from_kuid(&init_user_ns
, inode
->i_uid
),
1656 from_kuid(&init_user_ns
, attr
->ia_uid
));
1657 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1658 inode
->i_uid
= attr
->ia_uid
;
1659 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1660 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1661 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1662 req
->r_args
.setattr
.uid
= cpu_to_le32(
1663 from_kuid(&init_user_ns
, attr
->ia_uid
));
1664 mask
|= CEPH_SETATTR_UID
;
1665 release
|= CEPH_CAP_AUTH_SHARED
;
1668 if (ia_valid
& ATTR_GID
) {
1669 dout("setattr %p gid %d -> %d\n", inode
,
1670 from_kgid(&init_user_ns
, inode
->i_gid
),
1671 from_kgid(&init_user_ns
, attr
->ia_gid
));
1672 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1673 inode
->i_gid
= attr
->ia_gid
;
1674 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1675 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1676 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1677 req
->r_args
.setattr
.gid
= cpu_to_le32(
1678 from_kgid(&init_user_ns
, attr
->ia_gid
));
1679 mask
|= CEPH_SETATTR_GID
;
1680 release
|= CEPH_CAP_AUTH_SHARED
;
1683 if (ia_valid
& ATTR_MODE
) {
1684 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1686 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1687 inode
->i_mode
= attr
->ia_mode
;
1688 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1689 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1690 attr
->ia_mode
!= inode
->i_mode
) {
1691 inode
->i_mode
= attr
->ia_mode
;
1692 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1693 mask
|= CEPH_SETATTR_MODE
;
1694 release
|= CEPH_CAP_AUTH_SHARED
;
1698 if (ia_valid
& ATTR_ATIME
) {
1699 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1700 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1701 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1702 if (issued
& CEPH_CAP_FILE_EXCL
) {
1703 ci
->i_time_warp_seq
++;
1704 inode
->i_atime
= attr
->ia_atime
;
1705 dirtied
|= CEPH_CAP_FILE_EXCL
;
1706 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1707 timespec_compare(&inode
->i_atime
,
1708 &attr
->ia_atime
) < 0) {
1709 inode
->i_atime
= attr
->ia_atime
;
1710 dirtied
|= CEPH_CAP_FILE_WR
;
1711 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1712 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1713 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1715 mask
|= CEPH_SETATTR_ATIME
;
1716 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1720 if (ia_valid
& ATTR_MTIME
) {
1721 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1722 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1723 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1724 if (issued
& CEPH_CAP_FILE_EXCL
) {
1725 ci
->i_time_warp_seq
++;
1726 inode
->i_mtime
= attr
->ia_mtime
;
1727 dirtied
|= CEPH_CAP_FILE_EXCL
;
1728 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1729 timespec_compare(&inode
->i_mtime
,
1730 &attr
->ia_mtime
) < 0) {
1731 inode
->i_mtime
= attr
->ia_mtime
;
1732 dirtied
|= CEPH_CAP_FILE_WR
;
1733 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1734 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1735 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1737 mask
|= CEPH_SETATTR_MTIME
;
1738 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1742 if (ia_valid
& ATTR_SIZE
) {
1743 dout("setattr %p size %lld -> %lld\n", inode
,
1744 inode
->i_size
, attr
->ia_size
);
1745 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1749 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1750 attr
->ia_size
> inode
->i_size
) {
1751 inode
->i_size
= attr
->ia_size
;
1753 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1754 inode
->i_ctime
= attr
->ia_ctime
;
1755 ci
->i_reported_size
= attr
->ia_size
;
1756 dirtied
|= CEPH_CAP_FILE_EXCL
;
1757 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1758 attr
->ia_size
!= inode
->i_size
) {
1759 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1760 req
->r_args
.setattr
.old_size
=
1761 cpu_to_le64(inode
->i_size
);
1762 mask
|= CEPH_SETATTR_SIZE
;
1763 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1768 /* these do nothing */
1769 if (ia_valid
& ATTR_CTIME
) {
1770 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1771 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1772 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1773 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1774 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1775 only
? "ctime only" : "ignored");
1776 inode
->i_ctime
= attr
->ia_ctime
;
1779 * if kernel wants to dirty ctime but nothing else,
1780 * we need to choose a cap to dirty under, or do
1781 * a almost-no-op setattr
1783 if (issued
& CEPH_CAP_AUTH_EXCL
)
1784 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1785 else if (issued
& CEPH_CAP_FILE_EXCL
)
1786 dirtied
|= CEPH_CAP_FILE_EXCL
;
1787 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1788 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1790 mask
|= CEPH_SETATTR_CTIME
;
1793 if (ia_valid
& ATTR_FILE
)
1794 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1797 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
);
1798 inode
->i_ctime
= CURRENT_TIME
;
1802 spin_unlock(&ci
->i_ceph_lock
);
1804 if (inode_dirty_flags
)
1805 __mark_inode_dirty(inode
, inode_dirty_flags
);
1807 if (ia_valid
& ATTR_MODE
) {
1808 err
= ceph_acl_chmod(dentry
, inode
);
1814 req
->r_inode
= inode
;
1816 req
->r_inode_drop
= release
;
1817 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1818 req
->r_num_caps
= 1;
1819 parent_inode
= ceph_get_dentry_parent_inode(dentry
);
1820 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
1823 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1824 ceph_cap_string(dirtied
), mask
);
1826 ceph_mdsc_put_request(req
);
1827 if (mask
& CEPH_SETATTR_SIZE
)
1828 __ceph_do_pending_vmtruncate(inode
);
1831 spin_unlock(&ci
->i_ceph_lock
);
1833 ceph_mdsc_put_request(req
);
1838 * Verify that we have a lease on the given mask. If not,
1839 * do a getattr against an mds.
1841 int ceph_do_getattr(struct inode
*inode
, int mask
)
1843 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1844 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1845 struct ceph_mds_request
*req
;
1848 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1849 dout("do_getattr inode %p SNAPDIR\n", inode
);
1853 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1854 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1857 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1859 return PTR_ERR(req
);
1860 req
->r_inode
= inode
;
1862 req
->r_num_caps
= 1;
1863 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1864 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1865 ceph_mdsc_put_request(req
);
1866 dout("do_getattr result=%d\n", err
);
1872 * Check inode permissions. We verify we have a valid value for
1873 * the AUTH cap, then call the generic handler.
1875 int ceph_permission(struct inode
*inode
, int mask
)
1879 if (mask
& MAY_NOT_BLOCK
)
1882 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1885 err
= generic_permission(inode
, mask
);
1890 * Get all attributes. Hopefully somedata we'll have a statlite()
1891 * and can limit the fields we require to be accurate.
1893 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1896 struct inode
*inode
= dentry
->d_inode
;
1897 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1900 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1902 generic_fillattr(inode
, stat
);
1903 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
1904 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1905 stat
->dev
= ceph_snap(inode
);
1908 if (S_ISDIR(inode
->i_mode
)) {
1909 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
1911 stat
->size
= ci
->i_rbytes
;
1913 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
1915 stat
->blksize
= 65536;