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/pagevec.h>
15 #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
,
102 * We use a 'frag tree' to keep track of the MDS's directory fragments
103 * for a given inode (usually there is just a single fragment). We
104 * need to know when a child frag is delegated to a new MDS, or when
105 * it is flagged as replicated, so we can direct our requests
110 * find/create a frag in the tree
112 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
116 struct rb_node
*parent
= NULL
;
117 struct ceph_inode_frag
*frag
;
120 p
= &ci
->i_fragtree
.rb_node
;
123 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
124 c
= ceph_frag_compare(f
, frag
->frag
);
133 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
135 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
136 "frag %x\n", &ci
->vfs_inode
,
137 ceph_vinop(&ci
->vfs_inode
), f
);
138 return ERR_PTR(-ENOMEM
);
145 rb_link_node(&frag
->node
, parent
, p
);
146 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
148 dout("get_or_create_frag added %llx.%llx frag %x\n",
149 ceph_vinop(&ci
->vfs_inode
), f
);
154 * find a specific frag @f
156 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
158 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
161 struct ceph_inode_frag
*frag
=
162 rb_entry(n
, struct ceph_inode_frag
, node
);
163 int c
= ceph_frag_compare(f
, frag
->frag
);
175 * Choose frag containing the given value @v. If @pfrag is
176 * specified, copy the frag delegation info to the caller if
179 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
180 struct ceph_inode_frag
*pfrag
,
183 u32 t
= ceph_frag_make(0, 0);
184 struct ceph_inode_frag
*frag
;
191 mutex_lock(&ci
->i_fragtree_mutex
);
193 WARN_ON(!ceph_frag_contains_value(t
, v
));
194 frag
= __ceph_find_frag(ci
, t
);
196 break; /* t is a leaf */
197 if (frag
->split_by
== 0) {
199 memcpy(pfrag
, frag
, sizeof(*pfrag
));
206 nway
= 1 << frag
->split_by
;
207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
208 frag
->split_by
, nway
);
209 for (i
= 0; i
< nway
; i
++) {
210 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
211 if (ceph_frag_contains_value(n
, v
)) {
218 dout("choose_frag(%x) = %x\n", v
, t
);
220 mutex_unlock(&ci
->i_fragtree_mutex
);
225 * Process dirfrag (delegation) info from the mds. Include leaf
226 * fragment in tree ONLY if ndist > 0. Otherwise, only
227 * branches/splits are included in i_fragtree)
229 static int ceph_fill_dirfrag(struct inode
*inode
,
230 struct ceph_mds_reply_dirfrag
*dirinfo
)
232 struct ceph_inode_info
*ci
= ceph_inode(inode
);
233 struct ceph_inode_frag
*frag
;
234 u32 id
= le32_to_cpu(dirinfo
->frag
);
235 int mds
= le32_to_cpu(dirinfo
->auth
);
236 int ndist
= le32_to_cpu(dirinfo
->ndist
);
240 mutex_lock(&ci
->i_fragtree_mutex
);
242 /* no delegation info needed. */
243 frag
= __ceph_find_frag(ci
, id
);
246 if (frag
->split_by
== 0) {
247 /* tree leaf, remove */
248 dout("fill_dirfrag removed %llx.%llx frag %x"
249 " (no ref)\n", ceph_vinop(inode
), id
);
250 rb_erase(&frag
->node
, &ci
->i_fragtree
);
253 /* tree branch, keep and clear */
254 dout("fill_dirfrag cleared %llx.%llx frag %x"
255 " referral\n", ceph_vinop(inode
), id
);
263 /* find/add this frag to store mds delegation info */
264 frag
= __get_or_create_frag(ci
, id
);
266 /* this is not the end of the world; we can continue
267 with bad/inaccurate delegation info */
268 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
269 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
275 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
276 for (i
= 0; i
< frag
->ndist
; i
++)
277 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
278 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
279 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
282 mutex_unlock(&ci
->i_fragtree_mutex
);
288 * initialize a newly allocated inode.
290 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
292 struct ceph_inode_info
*ci
;
295 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
299 dout("alloc_inode %p\n", &ci
->vfs_inode
);
302 ci
->i_time_warp_seq
= 0;
303 ci
->i_ceph_flags
= 0;
304 ci
->i_release_count
= 0;
305 ci
->i_symlink
= NULL
;
307 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
309 ci
->i_fragtree
= RB_ROOT
;
310 mutex_init(&ci
->i_fragtree_mutex
);
312 ci
->i_xattrs
.blob
= NULL
;
313 ci
->i_xattrs
.prealloc_blob
= NULL
;
314 ci
->i_xattrs
.dirty
= false;
315 ci
->i_xattrs
.index
= RB_ROOT
;
316 ci
->i_xattrs
.count
= 0;
317 ci
->i_xattrs
.names_size
= 0;
318 ci
->i_xattrs
.vals_size
= 0;
319 ci
->i_xattrs
.version
= 0;
320 ci
->i_xattrs
.index_version
= 0;
322 ci
->i_caps
= RB_ROOT
;
323 ci
->i_auth_cap
= NULL
;
324 ci
->i_dirty_caps
= 0;
325 ci
->i_flushing_caps
= 0;
326 INIT_LIST_HEAD(&ci
->i_dirty_item
);
327 INIT_LIST_HEAD(&ci
->i_flushing_item
);
328 ci
->i_cap_flush_seq
= 0;
329 ci
->i_cap_flush_last_tid
= 0;
330 memset(&ci
->i_cap_flush_tid
, 0, sizeof(ci
->i_cap_flush_tid
));
331 init_waitqueue_head(&ci
->i_cap_wq
);
332 ci
->i_hold_caps_min
= 0;
333 ci
->i_hold_caps_max
= 0;
334 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
335 ci
->i_cap_exporting_mds
= 0;
336 ci
->i_cap_exporting_mseq
= 0;
337 ci
->i_cap_exporting_issued
= 0;
338 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
339 ci
->i_head_snapc
= NULL
;
342 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
343 ci
->i_nr_by_mode
[i
] = 0;
345 ci
->i_truncate_seq
= 0;
346 ci
->i_truncate_size
= 0;
347 ci
->i_truncate_pending
= 0;
350 ci
->i_reported_size
= 0;
351 ci
->i_wanted_max_size
= 0;
352 ci
->i_requested_max_size
= 0;
356 ci
->i_rdcache_ref
= 0;
359 ci
->i_wrbuffer_ref
= 0;
360 ci
->i_wrbuffer_ref_head
= 0;
361 ci
->i_shared_gen
= 0;
362 ci
->i_rdcache_gen
= 0;
363 ci
->i_rdcache_revoking
= 0;
365 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
366 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
367 spin_lock_init(&ci
->i_unsafe_lock
);
369 ci
->i_snap_realm
= NULL
;
370 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
371 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
373 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
374 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
376 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
378 return &ci
->vfs_inode
;
381 static void ceph_i_callback(struct rcu_head
*head
)
383 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
384 struct ceph_inode_info
*ci
= ceph_inode(inode
);
386 INIT_LIST_HEAD(&inode
->i_dentry
);
387 kmem_cache_free(ceph_inode_cachep
, ci
);
390 void ceph_destroy_inode(struct inode
*inode
)
392 struct ceph_inode_info
*ci
= ceph_inode(inode
);
393 struct ceph_inode_frag
*frag
;
396 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
398 ceph_queue_caps_release(inode
);
401 * we may still have a snap_realm reference if there are stray
402 * caps in i_cap_exporting_issued or i_snap_caps.
404 if (ci
->i_snap_realm
) {
405 struct ceph_mds_client
*mdsc
=
406 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
407 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
409 dout(" dropping residual ref to snap realm %p\n", realm
);
410 spin_lock(&realm
->inodes_with_caps_lock
);
411 list_del_init(&ci
->i_snap_realm_item
);
412 spin_unlock(&realm
->inodes_with_caps_lock
);
413 ceph_put_snap_realm(mdsc
, realm
);
416 kfree(ci
->i_symlink
);
417 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
418 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
419 rb_erase(n
, &ci
->i_fragtree
);
423 __ceph_destroy_xattrs(ci
);
424 if (ci
->i_xattrs
.blob
)
425 ceph_buffer_put(ci
->i_xattrs
.blob
);
426 if (ci
->i_xattrs
.prealloc_blob
)
427 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
429 call_rcu(&inode
->i_rcu
, ceph_i_callback
);
434 * Helpers to fill in size, ctime, mtime, and atime. We have to be
435 * careful because either the client or MDS may have more up to date
436 * info, depending on which capabilities are held, and whether
437 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
438 * and size are monotonically increasing, except when utimes() or
439 * truncate() increments the corresponding _seq values.)
441 int ceph_fill_file_size(struct inode
*inode
, int issued
,
442 u32 truncate_seq
, u64 truncate_size
, u64 size
)
444 struct ceph_inode_info
*ci
= ceph_inode(inode
);
447 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
448 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
449 dout("size %lld -> %llu\n", inode
->i_size
, size
);
450 inode
->i_size
= size
;
451 inode
->i_blocks
= (size
+ (1<<9) - 1) >> 9;
452 ci
->i_reported_size
= size
;
453 if (truncate_seq
!= ci
->i_truncate_seq
) {
454 dout("truncate_seq %u -> %u\n",
455 ci
->i_truncate_seq
, truncate_seq
);
456 ci
->i_truncate_seq
= truncate_seq
;
458 * If we hold relevant caps, or in the case where we're
459 * not the only client referencing this file and we
460 * don't hold those caps, then we need to check whether
461 * the file is either opened or mmaped
463 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_RD
|
464 CEPH_CAP_FILE_WR
|CEPH_CAP_FILE_BUFFER
|
466 CEPH_CAP_FILE_LAZYIO
)) ||
467 mapping_mapped(inode
->i_mapping
) ||
468 __ceph_caps_file_wanted(ci
)) {
469 ci
->i_truncate_pending
++;
474 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
475 ci
->i_truncate_size
!= truncate_size
) {
476 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
478 ci
->i_truncate_size
= truncate_size
;
483 void ceph_fill_file_time(struct inode
*inode
, int issued
,
484 u64 time_warp_seq
, struct timespec
*ctime
,
485 struct timespec
*mtime
, struct timespec
*atime
)
487 struct ceph_inode_info
*ci
= ceph_inode(inode
);
490 if (issued
& (CEPH_CAP_FILE_EXCL
|
492 CEPH_CAP_FILE_BUFFER
|
494 CEPH_CAP_XATTR_EXCL
)) {
495 if (timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
496 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
497 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
498 ctime
->tv_sec
, ctime
->tv_nsec
);
499 inode
->i_ctime
= *ctime
;
501 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
502 /* the MDS did a utimes() */
503 dout("mtime %ld.%09ld -> %ld.%09ld "
505 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
506 mtime
->tv_sec
, mtime
->tv_nsec
,
507 ci
->i_time_warp_seq
, (int)time_warp_seq
);
509 inode
->i_mtime
= *mtime
;
510 inode
->i_atime
= *atime
;
511 ci
->i_time_warp_seq
= time_warp_seq
;
512 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
513 /* nobody did utimes(); take the max */
514 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
515 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
516 inode
->i_mtime
.tv_sec
,
517 inode
->i_mtime
.tv_nsec
,
518 mtime
->tv_sec
, mtime
->tv_nsec
);
519 inode
->i_mtime
= *mtime
;
521 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
522 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
523 inode
->i_atime
.tv_sec
,
524 inode
->i_atime
.tv_nsec
,
525 atime
->tv_sec
, atime
->tv_nsec
);
526 inode
->i_atime
= *atime
;
528 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
529 /* we did a utimes(); ignore mds values */
534 /* we have no write|excl caps; whatever the MDS says is true */
535 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
536 inode
->i_ctime
= *ctime
;
537 inode
->i_mtime
= *mtime
;
538 inode
->i_atime
= *atime
;
539 ci
->i_time_warp_seq
= time_warp_seq
;
544 if (warn
) /* time_warp_seq shouldn't go backwards */
545 dout("%p mds time_warp_seq %llu < %u\n",
546 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
550 * Populate an inode based on info from mds. May be called on new or
553 static int fill_inode(struct inode
*inode
,
554 struct ceph_mds_reply_info_in
*iinfo
,
555 struct ceph_mds_reply_dirfrag
*dirinfo
,
556 struct ceph_mds_session
*session
,
557 unsigned long ttl_from
, int cap_fmode
,
558 struct ceph_cap_reservation
*caps_reservation
)
560 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
561 struct ceph_inode_info
*ci
= ceph_inode(inode
);
563 int issued
= 0, implemented
;
564 int updating_inode
= 0;
565 struct timespec mtime
, atime
, ctime
;
567 struct ceph_buffer
*xattr_blob
= NULL
;
571 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
572 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
576 * prealloc xattr data, if it looks like we'll need it. only
577 * if len > 4 (meaning there are actually xattrs; the first 4
578 * bytes are the xattr count).
580 if (iinfo
->xattr_len
> 4) {
581 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
583 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
587 spin_lock(&inode
->i_lock
);
590 * provided version will be odd if inode value is projected,
591 * even if stable. skip the update if we have newer stable
592 * info (ours>=theirs, e.g. due to racing mds replies), unless
593 * we are getting projected (unstable) info (in which case the
594 * version is odd, and we want ours>theirs).
600 if (le64_to_cpu(info
->version
) > 0 &&
601 (ci
->i_version
& ~1) >= le64_to_cpu(info
->version
))
605 issued
= __ceph_caps_issued(ci
, &implemented
);
606 issued
|= implemented
| __ceph_caps_dirty(ci
);
609 ci
->i_version
= le64_to_cpu(info
->version
);
611 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
613 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
614 inode
->i_mode
= le32_to_cpu(info
->mode
);
615 inode
->i_uid
= le32_to_cpu(info
->uid
);
616 inode
->i_gid
= le32_to_cpu(info
->gid
);
617 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
618 inode
->i_uid
, inode
->i_gid
);
621 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
622 inode
->i_nlink
= le32_to_cpu(info
->nlink
);
624 /* be careful with mtime, atime, size */
625 ceph_decode_timespec(&atime
, &info
->atime
);
626 ceph_decode_timespec(&mtime
, &info
->mtime
);
627 ceph_decode_timespec(&ctime
, &info
->ctime
);
628 queue_trunc
= ceph_fill_file_size(inode
, issued
,
629 le32_to_cpu(info
->truncate_seq
),
630 le64_to_cpu(info
->truncate_size
),
631 le64_to_cpu(info
->size
));
632 ceph_fill_file_time(inode
, issued
,
633 le32_to_cpu(info
->time_warp_seq
),
634 &ctime
, &mtime
, &atime
);
636 /* only update max_size on auth cap */
637 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
638 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
639 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
640 le64_to_cpu(info
->max_size
));
641 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
644 ci
->i_layout
= info
->layout
;
645 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
648 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
649 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 &&
650 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
651 if (ci
->i_xattrs
.blob
)
652 ceph_buffer_put(ci
->i_xattrs
.blob
);
653 ci
->i_xattrs
.blob
= xattr_blob
;
655 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
656 iinfo
->xattr_data
, iinfo
->xattr_len
);
657 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
661 inode
->i_mapping
->a_ops
= &ceph_aops
;
662 inode
->i_mapping
->backing_dev_info
=
663 &ceph_sb_to_client(inode
->i_sb
)->backing_dev_info
;
665 switch (inode
->i_mode
& S_IFMT
) {
670 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
671 inode
->i_op
= &ceph_file_iops
;
674 inode
->i_op
= &ceph_file_iops
;
675 inode
->i_fop
= &ceph_file_fops
;
678 inode
->i_op
= &ceph_symlink_iops
;
679 if (!ci
->i_symlink
) {
680 int symlen
= iinfo
->symlink_len
;
683 BUG_ON(symlen
!= inode
->i_size
);
684 spin_unlock(&inode
->i_lock
);
687 sym
= kmalloc(symlen
+1, GFP_NOFS
);
690 memcpy(sym
, iinfo
->symlink
, symlen
);
693 spin_lock(&inode
->i_lock
);
697 kfree(sym
); /* lost a race */
701 inode
->i_op
= &ceph_dir_iops
;
702 inode
->i_fop
= &ceph_dir_fops
;
704 ci
->i_dir_layout
= iinfo
->dir_layout
;
706 ci
->i_files
= le64_to_cpu(info
->files
);
707 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
708 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
709 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
710 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
711 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
714 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
715 ceph_vinop(inode
), inode
->i_mode
);
719 spin_unlock(&inode
->i_lock
);
721 /* queue truncate if we saw i_size decrease */
723 ceph_queue_vmtruncate(inode
);
725 /* populate frag tree */
726 /* FIXME: move me up, if/when version reflects fragtree changes */
727 nsplits
= le32_to_cpu(info
->fragtree
.nsplits
);
728 mutex_lock(&ci
->i_fragtree_mutex
);
729 for (i
= 0; i
< nsplits
; i
++) {
730 u32 id
= le32_to_cpu(info
->fragtree
.splits
[i
].frag
);
731 struct ceph_inode_frag
*frag
= __get_or_create_frag(ci
, id
);
735 frag
->split_by
= le32_to_cpu(info
->fragtree
.splits
[i
].by
);
736 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
738 mutex_unlock(&ci
->i_fragtree_mutex
);
740 /* were we issued a capability? */
741 if (info
->cap
.caps
) {
742 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
743 ceph_add_cap(inode
, session
,
744 le64_to_cpu(info
->cap
.cap_id
),
746 le32_to_cpu(info
->cap
.caps
),
747 le32_to_cpu(info
->cap
.wanted
),
748 le32_to_cpu(info
->cap
.seq
),
749 le32_to_cpu(info
->cap
.mseq
),
750 le64_to_cpu(info
->cap
.realm
),
754 spin_lock(&inode
->i_lock
);
755 dout(" %p got snap_caps %s\n", inode
,
756 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
757 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
759 __ceph_get_fmode(ci
, cap_fmode
);
760 spin_unlock(&inode
->i_lock
);
762 } else if (cap_fmode
>= 0) {
763 pr_warning("mds issued no caps on %llx.%llx\n",
765 __ceph_get_fmode(ci
, cap_fmode
);
768 /* set dir completion flag? */
769 if (S_ISDIR(inode
->i_mode
) &&
770 updating_inode
&& /* didn't jump to no_change */
771 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
772 ceph_snap(inode
) == CEPH_NOSNAP
&&
773 (le32_to_cpu(info
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) &&
774 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
775 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
776 dout(" marking %p complete (empty)\n", inode
);
777 /* ci->i_ceph_flags |= CEPH_I_COMPLETE; */
778 ci
->i_max_offset
= 2;
781 /* update delegation info? */
783 ceph_fill_dirfrag(inode
, dirinfo
);
789 ceph_buffer_put(xattr_blob
);
794 * caller should hold session s_mutex.
796 static void update_dentry_lease(struct dentry
*dentry
,
797 struct ceph_mds_reply_lease
*lease
,
798 struct ceph_mds_session
*session
,
799 unsigned long from_time
)
801 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
802 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
803 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
804 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
807 /* only track leases on regular dentries */
808 if (dentry
->d_op
!= &ceph_dentry_ops
)
811 spin_lock(&dentry
->d_lock
);
812 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
813 dentry
, duration
, ttl
);
815 /* make lease_rdcache_gen match directory */
816 dir
= dentry
->d_parent
->d_inode
;
817 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
822 if (di
->lease_gen
== session
->s_cap_gen
&&
823 time_before(ttl
, dentry
->d_time
))
824 goto out_unlock
; /* we already have a newer lease. */
826 if (di
->lease_session
&& di
->lease_session
!= session
)
829 ceph_dentry_lru_touch(dentry
);
831 if (!di
->lease_session
)
832 di
->lease_session
= ceph_get_mds_session(session
);
833 di
->lease_gen
= session
->s_cap_gen
;
834 di
->lease_seq
= le32_to_cpu(lease
->seq
);
835 di
->lease_renew_after
= half_ttl
;
836 di
->lease_renew_from
= 0;
837 dentry
->d_time
= ttl
;
839 spin_unlock(&dentry
->d_lock
);
844 * Set dentry's directory position based on the current dir's max, and
845 * order it in d_subdirs, so that dcache_readdir behaves.
847 static void ceph_set_dentry_offset(struct dentry
*dn
)
849 struct dentry
*dir
= dn
->d_parent
;
850 struct inode
*inode
= dn
->d_parent
->d_inode
;
851 struct ceph_dentry_info
*di
;
855 di
= ceph_dentry(dn
);
857 spin_lock(&inode
->i_lock
);
858 if ((ceph_inode(inode
)->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
859 spin_unlock(&inode
->i_lock
);
862 di
->offset
= ceph_inode(inode
)->i_max_offset
++;
863 spin_unlock(&inode
->i_lock
);
865 spin_lock(&dir
->d_lock
);
866 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
867 list_move(&dn
->d_u
.d_child
, &dir
->d_subdirs
);
868 dout("set_dentry_offset %p %lld (%p %p)\n", dn
, di
->offset
,
869 dn
->d_u
.d_child
.prev
, dn
->d_u
.d_child
.next
);
870 spin_unlock(&dn
->d_lock
);
871 spin_unlock(&dir
->d_lock
);
875 * splice a dentry to an inode.
876 * caller must hold directory i_mutex for this to be safe.
878 * we will only rehash the resulting dentry if @prehash is
879 * true; @prehash will be set to false (for the benefit of
880 * the caller) if we fail.
882 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
883 bool *prehash
, bool set_offset
)
885 struct dentry
*realdn
;
889 /* dn must be unhashed */
892 realdn
= d_materialise_unique(dn
, in
);
893 if (IS_ERR(realdn
)) {
894 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
895 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
897 *prehash
= false; /* don't rehash on error */
898 dn
= realdn
; /* note realdn contains the error */
901 dout("dn %p (%d) spliced with %p (%d) "
902 "inode %p ino %llx.%llx\n",
904 realdn
, realdn
->d_count
,
905 realdn
->d_inode
, ceph_vinop(realdn
->d_inode
));
909 BUG_ON(!ceph_dentry(dn
));
910 dout("dn %p attached to %p ino %llx.%llx\n",
911 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
));
913 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
916 ceph_set_dentry_offset(dn
);
922 * Incorporate results into the local cache. This is either just
923 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
926 * A reply may contain
927 * a directory inode along with a dentry.
928 * and/or a target inode
930 * Called with snap_rwsem (read).
932 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
933 struct ceph_mds_session
*session
)
935 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
936 struct inode
*in
= NULL
;
937 struct ceph_mds_reply_inode
*ininfo
;
938 struct ceph_vino vino
;
939 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
943 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
944 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
950 * If we resend completed ops to a recovering mds, we get no
951 * trace. Since that is very rare, pretend this is the case
952 * to ensure the 'no trace' handlers in the callers behave.
954 * Fill in inodes unconditionally to avoid breaking cap
957 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
958 pr_info("fill_trace faking empty trace on %lld %s\n",
959 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
960 if (rinfo
->head
->is_dentry
) {
961 rinfo
->head
->is_dentry
= 0;
962 err
= fill_inode(req
->r_locked_dir
,
963 &rinfo
->diri
, rinfo
->dirfrag
,
964 session
, req
->r_request_started
, -1);
966 if (rinfo
->head
->is_target
) {
967 rinfo
->head
->is_target
= 0;
968 ininfo
= rinfo
->targeti
.in
;
969 vino
.ino
= le64_to_cpu(ininfo
->ino
);
970 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
971 in
= ceph_get_inode(sb
, vino
);
972 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
973 session
, req
->r_request_started
,
980 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
981 dout("fill_trace reply is empty!\n");
982 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
983 ceph_invalidate_dir_request(req
);
987 if (rinfo
->head
->is_dentry
) {
988 struct inode
*dir
= req
->r_locked_dir
;
990 err
= fill_inode(dir
, &rinfo
->diri
, rinfo
->dirfrag
,
991 session
, req
->r_request_started
, -1,
992 &req
->r_caps_reservation
);
998 * ignore null lease/binding on snapdir ENOENT, or else we
999 * will have trouble splicing in the virtual snapdir later
1001 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1002 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1003 fsc
->mount_options
->snapdir_name
,
1004 req
->r_dentry
->d_name
.len
))) {
1006 * lookup link rename : null -> possibly existing inode
1007 * mknod symlink mkdir : null -> new inode
1008 * unlink : linked -> null
1010 struct inode
*dir
= req
->r_locked_dir
;
1011 struct dentry
*dn
= req
->r_dentry
;
1012 bool have_dir_cap
, have_lease
;
1016 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
1017 BUG_ON(ceph_ino(dir
) !=
1018 le64_to_cpu(rinfo
->diri
.in
->ino
));
1019 BUG_ON(ceph_snap(dir
) !=
1020 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1022 /* do we have a lease on the whole dir? */
1024 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1025 CEPH_CAP_FILE_SHARED
);
1027 /* do we have a dn lease? */
1028 have_lease
= have_dir_cap
||
1029 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1031 dout("fill_trace no dentry lease or dir cap\n");
1034 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1035 dout(" src %p '%.*s' dst %p '%.*s'\n",
1037 req
->r_old_dentry
->d_name
.len
,
1038 req
->r_old_dentry
->d_name
.name
,
1039 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1040 dout("fill_trace doing d_move %p -> %p\n",
1041 req
->r_old_dentry
, dn
);
1043 d_move(req
->r_old_dentry
, dn
);
1044 dout(" src %p '%.*s' dst %p '%.*s'\n",
1046 req
->r_old_dentry
->d_name
.len
,
1047 req
->r_old_dentry
->d_name
.name
,
1048 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1050 /* ensure target dentry is invalidated, despite
1051 rehashing bug in vfs_rename_dir */
1052 ceph_invalidate_dentry_lease(dn
);
1055 * d_move() puts the renamed dentry at the end of
1056 * d_subdirs. We need to assign it an appropriate
1057 * directory offset so we can behave when holding
1060 ceph_set_dentry_offset(req
->r_old_dentry
);
1061 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1062 ceph_dentry(req
->r_old_dentry
)->offset
);
1064 dn
= req
->r_old_dentry
; /* use old_dentry */
1069 if (!rinfo
->head
->is_target
) {
1070 dout("fill_trace null dentry\n");
1072 dout("d_delete %p\n", dn
);
1075 dout("d_instantiate %p NULL\n", dn
);
1076 d_instantiate(dn
, NULL
);
1077 if (have_lease
&& d_unhashed(dn
))
1079 update_dentry_lease(dn
, rinfo
->dlease
,
1081 req
->r_request_started
);
1086 /* attach proper inode */
1087 ininfo
= rinfo
->targeti
.in
;
1088 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1089 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1092 in
= ceph_get_inode(sb
, vino
);
1094 pr_err("fill_trace bad get_inode "
1095 "%llx.%llx\n", vino
.ino
, vino
.snap
);
1100 dn
= splice_dentry(dn
, in
, &have_lease
, true);
1105 req
->r_dentry
= dn
; /* may have spliced */
1107 } else if (ceph_ino(in
) == vino
.ino
&&
1108 ceph_snap(in
) == vino
.snap
) {
1111 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1112 dn
, in
, ceph_ino(in
), ceph_snap(in
),
1113 vino
.ino
, vino
.snap
);
1119 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1120 req
->r_request_started
);
1121 dout(" final dn %p\n", dn
);
1123 } else if (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1124 req
->r_op
== CEPH_MDS_OP_MKSNAP
) {
1125 struct dentry
*dn
= req
->r_dentry
;
1127 /* fill out a snapdir LOOKUPSNAP dentry */
1129 BUG_ON(!req
->r_locked_dir
);
1130 BUG_ON(ceph_snap(req
->r_locked_dir
) != CEPH_SNAPDIR
);
1131 ininfo
= rinfo
->targeti
.in
;
1132 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1133 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1134 in
= ceph_get_inode(sb
, vino
);
1136 pr_err("fill_inode get_inode badness %llx.%llx\n",
1137 vino
.ino
, vino
.snap
);
1142 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1143 dn
= splice_dentry(dn
, in
, NULL
, true);
1148 req
->r_dentry
= dn
; /* may have spliced */
1150 rinfo
->head
->is_dentry
= 1; /* fool notrace handlers */
1153 if (rinfo
->head
->is_target
) {
1154 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1155 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1157 if (in
== NULL
|| ceph_ino(in
) != vino
.ino
||
1158 ceph_snap(in
) != vino
.snap
) {
1159 in
= ceph_get_inode(sb
, vino
);
1165 req
->r_target_inode
= in
;
1167 err
= fill_inode(in
,
1168 &rinfo
->targeti
, NULL
,
1169 session
, req
->r_request_started
,
1170 (le32_to_cpu(rinfo
->head
->result
) == 0) ?
1172 &req
->r_caps_reservation
);
1174 pr_err("fill_inode badness %p %llx.%llx\n",
1175 in
, ceph_vinop(in
));
1181 dout("fill_trace done err=%d\n", err
);
1186 * Prepopulate our cache with readdir results, leases, etc.
1188 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1189 struct ceph_mds_session
*session
)
1191 struct dentry
*parent
= req
->r_dentry
;
1192 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1197 struct inode
*snapdir
= NULL
;
1198 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1199 u64 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1200 struct ceph_dentry_info
*di
;
1202 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1203 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1204 parent
= d_find_alias(snapdir
);
1205 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1206 rinfo
->dir_nr
, parent
);
1208 dout("readdir_prepopulate %d items under dn %p\n",
1209 rinfo
->dir_nr
, parent
);
1211 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1214 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1215 struct ceph_vino vino
;
1217 dname
.name
= rinfo
->dir_dname
[i
];
1218 dname
.len
= rinfo
->dir_dname_len
[i
];
1219 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1221 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1222 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1225 dn
= d_lookup(parent
, &dname
);
1226 dout("d_lookup on parent=%p name=%.*s got %p\n",
1227 parent
, dname
.len
, dname
.name
, dn
);
1230 dn
= d_alloc(parent
, &dname
);
1231 dout("d_alloc %p '%.*s' = %p\n", parent
,
1232 dname
.len
, dname
.name
, dn
);
1234 dout("d_alloc badness\n");
1238 err
= ceph_init_dentry(dn
);
1243 } else if (dn
->d_inode
&&
1244 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1245 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1246 dout(" dn %p points to wrong inode %p\n",
1252 /* reorder parent's d_subdirs */
1253 spin_lock(&parent
->d_lock
);
1254 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
1255 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1256 spin_unlock(&dn
->d_lock
);
1257 spin_unlock(&parent
->d_lock
);
1261 di
->offset
= ceph_make_fpos(frag
, i
+ req
->r_readdir_offset
);
1267 in
= ceph_get_inode(parent
->d_sb
, vino
);
1269 dout("new_inode badness\n");
1275 dn
= splice_dentry(dn
, in
, NULL
, false);
1280 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1281 req
->r_request_started
, -1,
1282 &req
->r_caps_reservation
) < 0) {
1283 pr_err("fill_inode badness on %p\n", in
);
1287 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1289 req
->r_request_started
);
1294 req
->r_did_prepopulate
= true;
1301 dout("readdir_prepopulate done\n");
1305 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1307 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1310 spin_lock(&inode
->i_lock
);
1311 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1312 inode
->i_size
= size
;
1313 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1315 /* tell the MDS if we are approaching max_size */
1316 if ((size
<< 1) >= ci
->i_max_size
&&
1317 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1320 spin_unlock(&inode
->i_lock
);
1325 * Write back inode data in a worker thread. (This can't be done
1326 * in the message handler context.)
1328 void ceph_queue_writeback(struct inode
*inode
)
1330 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1331 &ceph_inode(inode
)->i_wb_work
)) {
1332 dout("ceph_queue_writeback %p\n", inode
);
1335 dout("ceph_queue_writeback %p failed\n", inode
);
1339 static void ceph_writeback_work(struct work_struct
*work
)
1341 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1343 struct inode
*inode
= &ci
->vfs_inode
;
1345 dout("writeback %p\n", inode
);
1346 filemap_fdatawrite(&inode
->i_data
);
1351 * queue an async invalidation
1353 void ceph_queue_invalidate(struct inode
*inode
)
1355 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1356 &ceph_inode(inode
)->i_pg_inv_work
)) {
1357 dout("ceph_queue_invalidate %p\n", inode
);
1360 dout("ceph_queue_invalidate %p failed\n", inode
);
1365 * invalidate any pages that are not dirty or under writeback. this
1366 * includes pages that are clean and mapped.
1368 static void ceph_invalidate_nondirty_pages(struct address_space
*mapping
)
1370 struct pagevec pvec
;
1374 pagevec_init(&pvec
, 0);
1375 while (pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
1376 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1377 struct page
*page
= pvec
.pages
[i
];
1380 (PageDirty(page
) || PageWriteback(page
));
1383 skip_page
= !trylock_page(page
);
1386 * We really shouldn't be looking at the ->index of an
1387 * unlocked page. But we're not allowed to lock these
1388 * pages. So we rely upon nobody altering the ->index
1389 * of this (pinned-by-us) page.
1391 index
= page
->index
;
1399 generic_error_remove_page(mapping
, page
);
1402 pagevec_release(&pvec
);
1408 * Invalidate inode pages in a worker thread. (This can't be done
1409 * in the message handler context.)
1411 static void ceph_invalidate_work(struct work_struct
*work
)
1413 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1415 struct inode
*inode
= &ci
->vfs_inode
;
1419 spin_lock(&inode
->i_lock
);
1420 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1421 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1422 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1424 spin_unlock(&inode
->i_lock
);
1427 orig_gen
= ci
->i_rdcache_gen
;
1428 spin_unlock(&inode
->i_lock
);
1430 ceph_invalidate_nondirty_pages(inode
->i_mapping
);
1432 spin_lock(&inode
->i_lock
);
1433 if (orig_gen
== ci
->i_rdcache_gen
&&
1434 orig_gen
== ci
->i_rdcache_revoking
) {
1435 dout("invalidate_pages %p gen %d successful\n", inode
,
1437 ci
->i_rdcache_revoking
--;
1440 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1441 inode
, orig_gen
, ci
->i_rdcache_gen
,
1442 ci
->i_rdcache_revoking
);
1444 spin_unlock(&inode
->i_lock
);
1447 ceph_check_caps(ci
, 0, NULL
);
1454 * called by trunc_wq; take i_mutex ourselves
1456 * We also truncate in a separate thread as well.
1458 static void ceph_vmtruncate_work(struct work_struct
*work
)
1460 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1462 struct inode
*inode
= &ci
->vfs_inode
;
1464 dout("vmtruncate_work %p\n", inode
);
1465 mutex_lock(&inode
->i_mutex
);
1466 __ceph_do_pending_vmtruncate(inode
);
1467 mutex_unlock(&inode
->i_mutex
);
1472 * Queue an async vmtruncate. If we fail to queue work, we will handle
1473 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1475 void ceph_queue_vmtruncate(struct inode
*inode
)
1477 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1479 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1480 &ci
->i_vmtruncate_work
)) {
1481 dout("ceph_queue_vmtruncate %p\n", inode
);
1484 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1485 inode
, ci
->i_truncate_pending
);
1490 * called with i_mutex held.
1492 * Make sure any pending truncation is applied before doing anything
1493 * that may depend on it.
1495 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1497 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1499 int wrbuffer_refs
, wake
= 0;
1502 spin_lock(&inode
->i_lock
);
1503 if (ci
->i_truncate_pending
== 0) {
1504 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1505 spin_unlock(&inode
->i_lock
);
1510 * make sure any dirty snapped pages are flushed before we
1511 * possibly truncate them.. so write AND block!
1513 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1514 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1516 spin_unlock(&inode
->i_lock
);
1517 filemap_write_and_wait_range(&inode
->i_data
, 0,
1518 inode
->i_sb
->s_maxbytes
);
1522 to
= ci
->i_truncate_size
;
1523 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1524 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1525 ci
->i_truncate_pending
, to
);
1526 spin_unlock(&inode
->i_lock
);
1528 truncate_inode_pages(inode
->i_mapping
, to
);
1530 spin_lock(&inode
->i_lock
);
1531 ci
->i_truncate_pending
--;
1532 if (ci
->i_truncate_pending
== 0)
1534 spin_unlock(&inode
->i_lock
);
1536 if (wrbuffer_refs
== 0)
1537 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1539 wake_up_all(&ci
->i_cap_wq
);
1546 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1548 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1549 nd_set_link(nd
, ci
->i_symlink
);
1553 static const struct inode_operations ceph_symlink_iops
= {
1554 .readlink
= generic_readlink
,
1555 .follow_link
= ceph_sym_follow_link
,
1561 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1563 struct inode
*inode
= dentry
->d_inode
;
1564 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1565 struct inode
*parent_inode
= dentry
->d_parent
->d_inode
;
1566 const unsigned int ia_valid
= attr
->ia_valid
;
1567 struct ceph_mds_request
*req
;
1568 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1570 int release
= 0, dirtied
= 0;
1573 int inode_dirty_flags
= 0;
1575 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1578 __ceph_do_pending_vmtruncate(inode
);
1580 err
= inode_change_ok(inode
, attr
);
1584 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1587 return PTR_ERR(req
);
1589 spin_lock(&inode
->i_lock
);
1590 issued
= __ceph_caps_issued(ci
, NULL
);
1591 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1593 if (ia_valid
& ATTR_UID
) {
1594 dout("setattr %p uid %d -> %d\n", inode
,
1595 inode
->i_uid
, attr
->ia_uid
);
1596 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1597 inode
->i_uid
= attr
->ia_uid
;
1598 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1599 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1600 attr
->ia_uid
!= inode
->i_uid
) {
1601 req
->r_args
.setattr
.uid
= cpu_to_le32(attr
->ia_uid
);
1602 mask
|= CEPH_SETATTR_UID
;
1603 release
|= CEPH_CAP_AUTH_SHARED
;
1606 if (ia_valid
& ATTR_GID
) {
1607 dout("setattr %p gid %d -> %d\n", inode
,
1608 inode
->i_gid
, attr
->ia_gid
);
1609 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1610 inode
->i_gid
= attr
->ia_gid
;
1611 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1612 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1613 attr
->ia_gid
!= inode
->i_gid
) {
1614 req
->r_args
.setattr
.gid
= cpu_to_le32(attr
->ia_gid
);
1615 mask
|= CEPH_SETATTR_GID
;
1616 release
|= CEPH_CAP_AUTH_SHARED
;
1619 if (ia_valid
& ATTR_MODE
) {
1620 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1622 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1623 inode
->i_mode
= attr
->ia_mode
;
1624 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1625 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1626 attr
->ia_mode
!= inode
->i_mode
) {
1627 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1628 mask
|= CEPH_SETATTR_MODE
;
1629 release
|= CEPH_CAP_AUTH_SHARED
;
1633 if (ia_valid
& ATTR_ATIME
) {
1634 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1635 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1636 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1637 if (issued
& CEPH_CAP_FILE_EXCL
) {
1638 ci
->i_time_warp_seq
++;
1639 inode
->i_atime
= attr
->ia_atime
;
1640 dirtied
|= CEPH_CAP_FILE_EXCL
;
1641 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1642 timespec_compare(&inode
->i_atime
,
1643 &attr
->ia_atime
) < 0) {
1644 inode
->i_atime
= attr
->ia_atime
;
1645 dirtied
|= CEPH_CAP_FILE_WR
;
1646 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1647 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1648 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1650 mask
|= CEPH_SETATTR_ATIME
;
1651 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1655 if (ia_valid
& ATTR_MTIME
) {
1656 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1657 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1658 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1659 if (issued
& CEPH_CAP_FILE_EXCL
) {
1660 ci
->i_time_warp_seq
++;
1661 inode
->i_mtime
= attr
->ia_mtime
;
1662 dirtied
|= CEPH_CAP_FILE_EXCL
;
1663 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1664 timespec_compare(&inode
->i_mtime
,
1665 &attr
->ia_mtime
) < 0) {
1666 inode
->i_mtime
= attr
->ia_mtime
;
1667 dirtied
|= CEPH_CAP_FILE_WR
;
1668 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1669 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1670 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1672 mask
|= CEPH_SETATTR_MTIME
;
1673 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1677 if (ia_valid
& ATTR_SIZE
) {
1678 dout("setattr %p size %lld -> %lld\n", inode
,
1679 inode
->i_size
, attr
->ia_size
);
1680 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1684 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1685 attr
->ia_size
> inode
->i_size
) {
1686 inode
->i_size
= attr
->ia_size
;
1688 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1689 inode
->i_ctime
= attr
->ia_ctime
;
1690 ci
->i_reported_size
= attr
->ia_size
;
1691 dirtied
|= CEPH_CAP_FILE_EXCL
;
1692 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1693 attr
->ia_size
!= inode
->i_size
) {
1694 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1695 req
->r_args
.setattr
.old_size
=
1696 cpu_to_le64(inode
->i_size
);
1697 mask
|= CEPH_SETATTR_SIZE
;
1698 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1703 /* these do nothing */
1704 if (ia_valid
& ATTR_CTIME
) {
1705 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1706 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1707 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1708 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1709 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1710 only
? "ctime only" : "ignored");
1711 inode
->i_ctime
= attr
->ia_ctime
;
1714 * if kernel wants to dirty ctime but nothing else,
1715 * we need to choose a cap to dirty under, or do
1716 * a almost-no-op setattr
1718 if (issued
& CEPH_CAP_AUTH_EXCL
)
1719 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1720 else if (issued
& CEPH_CAP_FILE_EXCL
)
1721 dirtied
|= CEPH_CAP_FILE_EXCL
;
1722 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1723 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1725 mask
|= CEPH_SETATTR_CTIME
;
1728 if (ia_valid
& ATTR_FILE
)
1729 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1732 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
);
1733 inode
->i_ctime
= CURRENT_TIME
;
1737 spin_unlock(&inode
->i_lock
);
1739 if (inode_dirty_flags
)
1740 __mark_inode_dirty(inode
, inode_dirty_flags
);
1743 req
->r_inode
= inode
;
1745 req
->r_inode_drop
= release
;
1746 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1747 req
->r_num_caps
= 1;
1748 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
1750 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1751 ceph_cap_string(dirtied
), mask
);
1753 ceph_mdsc_put_request(req
);
1754 __ceph_do_pending_vmtruncate(inode
);
1757 spin_unlock(&inode
->i_lock
);
1758 ceph_mdsc_put_request(req
);
1763 * Verify that we have a lease on the given mask. If not,
1764 * do a getattr against an mds.
1766 int ceph_do_getattr(struct inode
*inode
, int mask
)
1768 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1769 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1770 struct ceph_mds_request
*req
;
1773 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1774 dout("do_getattr inode %p SNAPDIR\n", inode
);
1778 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1779 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1782 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1784 return PTR_ERR(req
);
1785 req
->r_inode
= inode
;
1787 req
->r_num_caps
= 1;
1788 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1789 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1790 ceph_mdsc_put_request(req
);
1791 dout("do_getattr result=%d\n", err
);
1797 * Check inode permissions. We verify we have a valid value for
1798 * the AUTH cap, then call the generic handler.
1800 int ceph_permission(struct inode
*inode
, int mask
, unsigned int flags
)
1804 if (flags
& IPERM_FLAG_RCU
)
1807 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1810 err
= generic_permission(inode
, mask
, flags
, NULL
);
1815 * Get all attributes. Hopefully somedata we'll have a statlite()
1816 * and can limit the fields we require to be accurate.
1818 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1821 struct inode
*inode
= dentry
->d_inode
;
1822 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1825 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1827 generic_fillattr(inode
, stat
);
1828 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
1829 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1830 stat
->dev
= ceph_snap(inode
);
1833 if (S_ISDIR(inode
->i_mode
)) {
1834 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
1836 stat
->size
= ci
->i_rbytes
;
1838 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
1840 stat
->blksize
= 65536;