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"
15 #include <linux/ceph/decode.h>
18 * Ceph inode operations
20 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
21 * setattr, etc.), xattr helpers, and helpers for assimilating
22 * metadata returned by the MDS into our cache.
24 * Also define helpers for doing asynchronous writeback, invalidation,
25 * and truncation for the benefit of those who can't afford to block
26 * (typically because they are in the message handler path).
29 static const struct inode_operations ceph_symlink_iops
;
31 static void ceph_invalidate_work(struct work_struct
*work
);
32 static void ceph_writeback_work(struct work_struct
*work
);
33 static void ceph_vmtruncate_work(struct work_struct
*work
);
36 * find or create an inode, given the ceph ino number
38 static int ceph_set_ino_cb(struct inode
*inode
, void *data
)
40 ceph_inode(inode
)->i_vino
= *(struct ceph_vino
*)data
;
41 inode
->i_ino
= ceph_vino_to_ino(*(struct ceph_vino
*)data
);
45 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
48 ino_t t
= ceph_vino_to_ino(vino
);
50 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
52 return ERR_PTR(-ENOMEM
);
53 if (inode
->i_state
& I_NEW
) {
54 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
55 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
56 unlock_new_inode(inode
);
59 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
65 * get/constuct snapdir inode for a given directory
67 struct inode
*ceph_get_snapdir(struct inode
*parent
)
69 struct ceph_vino vino
= {
70 .ino
= ceph_ino(parent
),
73 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
74 struct ceph_inode_info
*ci
= ceph_inode(inode
);
76 BUG_ON(!S_ISDIR(parent
->i_mode
));
79 inode
->i_mode
= parent
->i_mode
;
80 inode
->i_uid
= parent
->i_uid
;
81 inode
->i_gid
= parent
->i_gid
;
82 inode
->i_op
= &ceph_dir_iops
;
83 inode
->i_fop
= &ceph_dir_fops
;
84 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
89 const struct inode_operations ceph_file_iops
= {
90 .permission
= ceph_permission
,
91 .setattr
= ceph_setattr
,
92 .getattr
= ceph_getattr
,
93 .setxattr
= ceph_setxattr
,
94 .getxattr
= ceph_getxattr
,
95 .listxattr
= ceph_listxattr
,
96 .removexattr
= ceph_removexattr
,
101 * We use a 'frag tree' to keep track of the MDS's directory fragments
102 * for a given inode (usually there is just a single fragment). We
103 * need to know when a child frag is delegated to a new MDS, or when
104 * it is flagged as replicated, so we can direct our requests
109 * find/create a frag in the tree
111 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
115 struct rb_node
*parent
= NULL
;
116 struct ceph_inode_frag
*frag
;
119 p
= &ci
->i_fragtree
.rb_node
;
122 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
123 c
= ceph_frag_compare(f
, frag
->frag
);
132 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
134 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
135 "frag %x\n", &ci
->vfs_inode
,
136 ceph_vinop(&ci
->vfs_inode
), f
);
137 return ERR_PTR(-ENOMEM
);
144 rb_link_node(&frag
->node
, parent
, p
);
145 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
147 dout("get_or_create_frag added %llx.%llx frag %x\n",
148 ceph_vinop(&ci
->vfs_inode
), f
);
153 * find a specific frag @f
155 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
157 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
160 struct ceph_inode_frag
*frag
=
161 rb_entry(n
, struct ceph_inode_frag
, node
);
162 int c
= ceph_frag_compare(f
, frag
->frag
);
174 * Choose frag containing the given value @v. If @pfrag is
175 * specified, copy the frag delegation info to the caller if
178 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
179 struct ceph_inode_frag
*pfrag
,
182 u32 t
= ceph_frag_make(0, 0);
183 struct ceph_inode_frag
*frag
;
190 mutex_lock(&ci
->i_fragtree_mutex
);
192 WARN_ON(!ceph_frag_contains_value(t
, v
));
193 frag
= __ceph_find_frag(ci
, t
);
195 break; /* t is a leaf */
196 if (frag
->split_by
== 0) {
198 memcpy(pfrag
, frag
, sizeof(*pfrag
));
205 nway
= 1 << frag
->split_by
;
206 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
207 frag
->split_by
, nway
);
208 for (i
= 0; i
< nway
; i
++) {
209 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
210 if (ceph_frag_contains_value(n
, v
)) {
217 dout("choose_frag(%x) = %x\n", v
, t
);
219 mutex_unlock(&ci
->i_fragtree_mutex
);
224 * Process dirfrag (delegation) info from the mds. Include leaf
225 * fragment in tree ONLY if ndist > 0. Otherwise, only
226 * branches/splits are included in i_fragtree)
228 static int ceph_fill_dirfrag(struct inode
*inode
,
229 struct ceph_mds_reply_dirfrag
*dirinfo
)
231 struct ceph_inode_info
*ci
= ceph_inode(inode
);
232 struct ceph_inode_frag
*frag
;
233 u32 id
= le32_to_cpu(dirinfo
->frag
);
234 int mds
= le32_to_cpu(dirinfo
->auth
);
235 int ndist
= le32_to_cpu(dirinfo
->ndist
);
239 mutex_lock(&ci
->i_fragtree_mutex
);
241 /* no delegation info needed. */
242 frag
= __ceph_find_frag(ci
, id
);
245 if (frag
->split_by
== 0) {
246 /* tree leaf, remove */
247 dout("fill_dirfrag removed %llx.%llx frag %x"
248 " (no ref)\n", ceph_vinop(inode
), id
);
249 rb_erase(&frag
->node
, &ci
->i_fragtree
);
252 /* tree branch, keep and clear */
253 dout("fill_dirfrag cleared %llx.%llx frag %x"
254 " referral\n", ceph_vinop(inode
), id
);
262 /* find/add this frag to store mds delegation info */
263 frag
= __get_or_create_frag(ci
, id
);
265 /* this is not the end of the world; we can continue
266 with bad/inaccurate delegation info */
267 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
268 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
274 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
275 for (i
= 0; i
< frag
->ndist
; i
++)
276 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
277 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
278 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
281 mutex_unlock(&ci
->i_fragtree_mutex
);
287 * initialize a newly allocated inode.
289 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
291 struct ceph_inode_info
*ci
;
294 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
298 dout("alloc_inode %p\n", &ci
->vfs_inode
);
300 spin_lock_init(&ci
->i_ceph_lock
);
303 ci
->i_time_warp_seq
= 0;
304 ci
->i_ceph_flags
= 0;
305 ci
->i_release_count
= 0;
306 ci
->i_symlink
= NULL
;
308 memset(&ci
->i_dir_layout
, 0, sizeof(ci
->i_dir_layout
));
310 ci
->i_fragtree
= RB_ROOT
;
311 mutex_init(&ci
->i_fragtree_mutex
);
313 ci
->i_xattrs
.blob
= NULL
;
314 ci
->i_xattrs
.prealloc_blob
= NULL
;
315 ci
->i_xattrs
.dirty
= false;
316 ci
->i_xattrs
.index
= RB_ROOT
;
317 ci
->i_xattrs
.count
= 0;
318 ci
->i_xattrs
.names_size
= 0;
319 ci
->i_xattrs
.vals_size
= 0;
320 ci
->i_xattrs
.version
= 0;
321 ci
->i_xattrs
.index_version
= 0;
323 ci
->i_caps
= RB_ROOT
;
324 ci
->i_auth_cap
= NULL
;
325 ci
->i_dirty_caps
= 0;
326 ci
->i_flushing_caps
= 0;
327 INIT_LIST_HEAD(&ci
->i_dirty_item
);
328 INIT_LIST_HEAD(&ci
->i_flushing_item
);
329 ci
->i_cap_flush_seq
= 0;
330 ci
->i_cap_flush_last_tid
= 0;
331 memset(&ci
->i_cap_flush_tid
, 0, sizeof(ci
->i_cap_flush_tid
));
332 init_waitqueue_head(&ci
->i_cap_wq
);
333 ci
->i_hold_caps_min
= 0;
334 ci
->i_hold_caps_max
= 0;
335 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
336 ci
->i_cap_exporting_mds
= 0;
337 ci
->i_cap_exporting_mseq
= 0;
338 ci
->i_cap_exporting_issued
= 0;
339 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
340 ci
->i_head_snapc
= NULL
;
343 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
344 ci
->i_nr_by_mode
[i
] = 0;
346 ci
->i_truncate_seq
= 0;
347 ci
->i_truncate_size
= 0;
348 ci
->i_truncate_pending
= 0;
351 ci
->i_reported_size
= 0;
352 ci
->i_wanted_max_size
= 0;
353 ci
->i_requested_max_size
= 0;
357 ci
->i_rdcache_ref
= 0;
360 ci
->i_wrbuffer_ref
= 0;
361 ci
->i_wrbuffer_ref_head
= 0;
362 ci
->i_shared_gen
= 0;
363 ci
->i_rdcache_gen
= 0;
364 ci
->i_rdcache_revoking
= 0;
366 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
367 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
368 spin_lock_init(&ci
->i_unsafe_lock
);
370 ci
->i_snap_realm
= NULL
;
371 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
372 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
374 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
375 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
377 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
379 return &ci
->vfs_inode
;
382 static void ceph_i_callback(struct rcu_head
*head
)
384 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
385 struct ceph_inode_info
*ci
= ceph_inode(inode
);
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 struct timespec mtime
, atime
, ctime
;
566 struct ceph_buffer
*xattr_blob
= NULL
;
570 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
571 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
575 * prealloc xattr data, if it looks like we'll need it. only
576 * if len > 4 (meaning there are actually xattrs; the first 4
577 * bytes are the xattr count).
579 if (iinfo
->xattr_len
> 4) {
580 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
582 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
586 spin_lock(&ci
->i_ceph_lock
);
589 * provided version will be odd if inode value is projected,
590 * even if stable. skip the update if we have newer stable
591 * info (ours>=theirs, e.g. due to racing mds replies), unless
592 * we are getting projected (unstable) info (in which case the
593 * version is odd, and we want ours>theirs).
599 if (le64_to_cpu(info
->version
) > 0 &&
600 (ci
->i_version
& ~1) >= le64_to_cpu(info
->version
))
603 issued
= __ceph_caps_issued(ci
, &implemented
);
604 issued
|= implemented
| __ceph_caps_dirty(ci
);
607 ci
->i_version
= le64_to_cpu(info
->version
);
609 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
611 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
612 inode
->i_mode
= le32_to_cpu(info
->mode
);
613 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(info
->uid
));
614 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(info
->gid
));
615 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
616 from_kuid(&init_user_ns
, inode
->i_uid
),
617 from_kgid(&init_user_ns
, inode
->i_gid
));
620 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
621 set_nlink(inode
, le32_to_cpu(info
->nlink
));
623 /* be careful with mtime, atime, size */
624 ceph_decode_timespec(&atime
, &info
->atime
);
625 ceph_decode_timespec(&mtime
, &info
->mtime
);
626 ceph_decode_timespec(&ctime
, &info
->ctime
);
627 queue_trunc
= ceph_fill_file_size(inode
, issued
,
628 le32_to_cpu(info
->truncate_seq
),
629 le64_to_cpu(info
->truncate_size
),
630 le64_to_cpu(info
->size
));
631 ceph_fill_file_time(inode
, issued
,
632 le32_to_cpu(info
->time_warp_seq
),
633 &ctime
, &mtime
, &atime
);
635 /* only update max_size on auth cap */
636 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
637 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
638 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
639 le64_to_cpu(info
->max_size
));
640 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
643 ci
->i_layout
= info
->layout
;
644 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
647 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
648 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 &&
649 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
650 if (ci
->i_xattrs
.blob
)
651 ceph_buffer_put(ci
->i_xattrs
.blob
);
652 ci
->i_xattrs
.blob
= xattr_blob
;
654 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
655 iinfo
->xattr_data
, iinfo
->xattr_len
);
656 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
660 inode
->i_mapping
->a_ops
= &ceph_aops
;
661 inode
->i_mapping
->backing_dev_info
=
662 &ceph_sb_to_client(inode
->i_sb
)->backing_dev_info
;
664 switch (inode
->i_mode
& S_IFMT
) {
669 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
670 inode
->i_op
= &ceph_file_iops
;
673 inode
->i_op
= &ceph_file_iops
;
674 inode
->i_fop
= &ceph_file_fops
;
677 inode
->i_op
= &ceph_symlink_iops
;
678 if (!ci
->i_symlink
) {
679 u32 symlen
= iinfo
->symlink_len
;
682 spin_unlock(&ci
->i_ceph_lock
);
685 if (WARN_ON(symlen
!= inode
->i_size
))
689 sym
= kstrndup(iinfo
->symlink
, symlen
, GFP_NOFS
);
693 spin_lock(&ci
->i_ceph_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
);
718 /* set dir completion flag? */
719 if (S_ISDIR(inode
->i_mode
) &&
720 ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
721 ceph_snap(inode
) == CEPH_NOSNAP
&&
722 (le32_to_cpu(info
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) &&
723 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
724 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
725 dout(" marking %p complete (empty)\n", inode
);
726 ci
->i_ceph_flags
|= CEPH_I_COMPLETE
;
727 ci
->i_max_offset
= 2;
730 spin_unlock(&ci
->i_ceph_lock
);
732 /* queue truncate if we saw i_size decrease */
734 ceph_queue_vmtruncate(inode
);
736 /* populate frag tree */
737 /* FIXME: move me up, if/when version reflects fragtree changes */
738 nsplits
= le32_to_cpu(info
->fragtree
.nsplits
);
739 mutex_lock(&ci
->i_fragtree_mutex
);
740 for (i
= 0; i
< nsplits
; i
++) {
741 u32 id
= le32_to_cpu(info
->fragtree
.splits
[i
].frag
);
742 struct ceph_inode_frag
*frag
= __get_or_create_frag(ci
, id
);
746 frag
->split_by
= le32_to_cpu(info
->fragtree
.splits
[i
].by
);
747 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
749 mutex_unlock(&ci
->i_fragtree_mutex
);
751 /* were we issued a capability? */
752 if (info
->cap
.caps
) {
753 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
754 ceph_add_cap(inode
, session
,
755 le64_to_cpu(info
->cap
.cap_id
),
757 le32_to_cpu(info
->cap
.caps
),
758 le32_to_cpu(info
->cap
.wanted
),
759 le32_to_cpu(info
->cap
.seq
),
760 le32_to_cpu(info
->cap
.mseq
),
761 le64_to_cpu(info
->cap
.realm
),
765 spin_lock(&ci
->i_ceph_lock
);
766 dout(" %p got snap_caps %s\n", inode
,
767 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
768 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
770 __ceph_get_fmode(ci
, cap_fmode
);
771 spin_unlock(&ci
->i_ceph_lock
);
773 } else if (cap_fmode
>= 0) {
774 pr_warning("mds issued no caps on %llx.%llx\n",
776 __ceph_get_fmode(ci
, cap_fmode
);
779 /* update delegation info? */
781 ceph_fill_dirfrag(inode
, dirinfo
);
787 ceph_buffer_put(xattr_blob
);
792 * caller should hold session s_mutex.
794 static void update_dentry_lease(struct dentry
*dentry
,
795 struct ceph_mds_reply_lease
*lease
,
796 struct ceph_mds_session
*session
,
797 unsigned long from_time
)
799 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
800 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
801 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
802 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
805 /* only track leases on regular dentries */
806 if (dentry
->d_op
!= &ceph_dentry_ops
)
809 spin_lock(&dentry
->d_lock
);
810 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
811 dentry
, duration
, ttl
);
813 /* make lease_rdcache_gen match directory */
814 dir
= dentry
->d_parent
->d_inode
;
815 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
820 if (di
->lease_gen
== session
->s_cap_gen
&&
821 time_before(ttl
, dentry
->d_time
))
822 goto out_unlock
; /* we already have a newer lease. */
824 if (di
->lease_session
&& di
->lease_session
!= session
)
827 ceph_dentry_lru_touch(dentry
);
829 if (!di
->lease_session
)
830 di
->lease_session
= ceph_get_mds_session(session
);
831 di
->lease_gen
= session
->s_cap_gen
;
832 di
->lease_seq
= le32_to_cpu(lease
->seq
);
833 di
->lease_renew_after
= half_ttl
;
834 di
->lease_renew_from
= 0;
835 dentry
->d_time
= ttl
;
837 spin_unlock(&dentry
->d_lock
);
842 * Set dentry's directory position based on the current dir's max, and
843 * order it in d_subdirs, so that dcache_readdir behaves.
845 * Always called under directory's i_mutex.
847 static void ceph_set_dentry_offset(struct dentry
*dn
)
849 struct dentry
*dir
= dn
->d_parent
;
850 struct inode
*inode
= dir
->d_inode
;
851 struct ceph_inode_info
*ci
;
852 struct ceph_dentry_info
*di
;
856 ci
= ceph_inode(inode
);
857 di
= ceph_dentry(dn
);
859 spin_lock(&ci
->i_ceph_lock
);
860 if ((ceph_inode(inode
)->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
861 spin_unlock(&ci
->i_ceph_lock
);
864 di
->offset
= ceph_inode(inode
)->i_max_offset
++;
865 spin_unlock(&ci
->i_ceph_lock
);
867 spin_lock(&dir
->d_lock
);
868 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
869 list_move(&dn
->d_u
.d_child
, &dir
->d_subdirs
);
870 dout("set_dentry_offset %p %lld (%p %p)\n", dn
, di
->offset
,
871 dn
->d_u
.d_child
.prev
, dn
->d_u
.d_child
.next
);
872 spin_unlock(&dn
->d_lock
);
873 spin_unlock(&dir
->d_lock
);
877 * splice a dentry to an inode.
878 * caller must hold directory i_mutex for this to be safe.
880 * we will only rehash the resulting dentry if @prehash is
881 * true; @prehash will be set to false (for the benefit of
882 * the caller) if we fail.
884 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
885 bool *prehash
, bool set_offset
)
887 struct dentry
*realdn
;
891 /* dn must be unhashed */
894 realdn
= d_materialise_unique(dn
, in
);
895 if (IS_ERR(realdn
)) {
896 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
897 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
899 *prehash
= false; /* don't rehash on error */
900 dn
= realdn
; /* note realdn contains the error */
903 dout("dn %p (%d) spliced with %p (%d) "
904 "inode %p ino %llx.%llx\n",
906 realdn
, realdn
->d_count
,
907 realdn
->d_inode
, ceph_vinop(realdn
->d_inode
));
911 BUG_ON(!ceph_dentry(dn
));
912 dout("dn %p attached to %p ino %llx.%llx\n",
913 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
));
915 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
918 ceph_set_dentry_offset(dn
);
924 * Incorporate results into the local cache. This is either just
925 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
928 * A reply may contain
929 * a directory inode along with a dentry.
930 * and/or a target inode
932 * Called with snap_rwsem (read).
934 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
935 struct ceph_mds_session
*session
)
937 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
938 struct inode
*in
= NULL
;
939 struct ceph_mds_reply_inode
*ininfo
;
940 struct ceph_vino vino
;
941 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
945 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
946 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
952 * If we resend completed ops to a recovering mds, we get no
953 * trace. Since that is very rare, pretend this is the case
954 * to ensure the 'no trace' handlers in the callers behave.
956 * Fill in inodes unconditionally to avoid breaking cap
959 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
960 pr_info("fill_trace faking empty trace on %lld %s\n",
961 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
962 if (rinfo
->head
->is_dentry
) {
963 rinfo
->head
->is_dentry
= 0;
964 err
= fill_inode(req
->r_locked_dir
,
965 &rinfo
->diri
, rinfo
->dirfrag
,
966 session
, req
->r_request_started
, -1);
968 if (rinfo
->head
->is_target
) {
969 rinfo
->head
->is_target
= 0;
970 ininfo
= rinfo
->targeti
.in
;
971 vino
.ino
= le64_to_cpu(ininfo
->ino
);
972 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
973 in
= ceph_get_inode(sb
, vino
);
974 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
975 session
, req
->r_request_started
,
982 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
983 dout("fill_trace reply is empty!\n");
984 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
985 ceph_invalidate_dir_request(req
);
989 if (rinfo
->head
->is_dentry
) {
990 struct inode
*dir
= req
->r_locked_dir
;
993 err
= fill_inode(dir
, &rinfo
->diri
, rinfo
->dirfrag
,
994 session
, req
->r_request_started
, -1,
995 &req
->r_caps_reservation
);
1004 * ignore null lease/binding on snapdir ENOENT, or else we
1005 * will have trouble splicing in the virtual snapdir later
1007 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
1008 req
->r_locked_dir
&&
1009 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
1010 fsc
->mount_options
->snapdir_name
,
1011 req
->r_dentry
->d_name
.len
))) {
1013 * lookup link rename : null -> possibly existing inode
1014 * mknod symlink mkdir : null -> new inode
1015 * unlink : linked -> null
1017 struct inode
*dir
= req
->r_locked_dir
;
1018 struct dentry
*dn
= req
->r_dentry
;
1019 bool have_dir_cap
, have_lease
;
1023 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
1024 BUG_ON(ceph_ino(dir
) !=
1025 le64_to_cpu(rinfo
->diri
.in
->ino
));
1026 BUG_ON(ceph_snap(dir
) !=
1027 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1029 /* do we have a lease on the whole dir? */
1031 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1032 CEPH_CAP_FILE_SHARED
);
1034 /* do we have a dn lease? */
1035 have_lease
= have_dir_cap
||
1036 le32_to_cpu(rinfo
->dlease
->duration_ms
);
1038 dout("fill_trace no dentry lease or dir cap\n");
1041 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1042 dout(" src %p '%.*s' dst %p '%.*s'\n",
1044 req
->r_old_dentry
->d_name
.len
,
1045 req
->r_old_dentry
->d_name
.name
,
1046 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1047 dout("fill_trace doing d_move %p -> %p\n",
1048 req
->r_old_dentry
, dn
);
1050 d_move(req
->r_old_dentry
, dn
);
1051 dout(" src %p '%.*s' dst %p '%.*s'\n",
1053 req
->r_old_dentry
->d_name
.len
,
1054 req
->r_old_dentry
->d_name
.name
,
1055 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1057 /* ensure target dentry is invalidated, despite
1058 rehashing bug in vfs_rename_dir */
1059 ceph_invalidate_dentry_lease(dn
);
1062 * d_move() puts the renamed dentry at the end of
1063 * d_subdirs. We need to assign it an appropriate
1064 * directory offset so we can behave when holding
1067 ceph_set_dentry_offset(req
->r_old_dentry
);
1068 dout("dn %p gets new offset %lld\n", req
->r_old_dentry
,
1069 ceph_dentry(req
->r_old_dentry
)->offset
);
1071 dn
= req
->r_old_dentry
; /* use old_dentry */
1076 if (!rinfo
->head
->is_target
) {
1077 dout("fill_trace null dentry\n");
1079 dout("d_delete %p\n", dn
);
1082 dout("d_instantiate %p NULL\n", dn
);
1083 d_instantiate(dn
, NULL
);
1084 if (have_lease
&& d_unhashed(dn
))
1086 update_dentry_lease(dn
, rinfo
->dlease
,
1088 req
->r_request_started
);
1093 /* attach proper inode */
1094 ininfo
= rinfo
->targeti
.in
;
1095 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1096 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1099 in
= ceph_get_inode(sb
, vino
);
1101 pr_err("fill_trace bad get_inode "
1102 "%llx.%llx\n", vino
.ino
, vino
.snap
);
1107 dn
= splice_dentry(dn
, in
, &have_lease
, true);
1112 req
->r_dentry
= dn
; /* may have spliced */
1114 } else if (ceph_ino(in
) == vino
.ino
&&
1115 ceph_snap(in
) == vino
.snap
) {
1118 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1119 dn
, in
, ceph_ino(in
), ceph_snap(in
),
1120 vino
.ino
, vino
.snap
);
1126 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1127 req
->r_request_started
);
1128 dout(" final dn %p\n", dn
);
1130 } else if ((req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1131 req
->r_op
== CEPH_MDS_OP_MKSNAP
) && !req
->r_aborted
) {
1132 struct dentry
*dn
= req
->r_dentry
;
1134 /* fill out a snapdir LOOKUPSNAP dentry */
1136 BUG_ON(!req
->r_locked_dir
);
1137 BUG_ON(ceph_snap(req
->r_locked_dir
) != CEPH_SNAPDIR
);
1138 ininfo
= rinfo
->targeti
.in
;
1139 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1140 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1141 in
= ceph_get_inode(sb
, vino
);
1143 pr_err("fill_inode get_inode badness %llx.%llx\n",
1144 vino
.ino
, vino
.snap
);
1149 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1150 dn
= splice_dentry(dn
, in
, NULL
, true);
1155 req
->r_dentry
= dn
; /* may have spliced */
1157 rinfo
->head
->is_dentry
= 1; /* fool notrace handlers */
1160 if (rinfo
->head
->is_target
) {
1161 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1162 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1164 if (in
== NULL
|| ceph_ino(in
) != vino
.ino
||
1165 ceph_snap(in
) != vino
.snap
) {
1166 in
= ceph_get_inode(sb
, vino
);
1172 req
->r_target_inode
= in
;
1174 err
= fill_inode(in
,
1175 &rinfo
->targeti
, NULL
,
1176 session
, req
->r_request_started
,
1177 (le32_to_cpu(rinfo
->head
->result
) == 0) ?
1179 &req
->r_caps_reservation
);
1181 pr_err("fill_inode badness %p %llx.%llx\n",
1182 in
, ceph_vinop(in
));
1188 dout("fill_trace done err=%d\n", err
);
1193 * Prepopulate our cache with readdir results, leases, etc.
1195 static int readdir_prepopulate_inodes_only(struct ceph_mds_request
*req
,
1196 struct ceph_mds_session
*session
)
1198 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1201 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1202 struct ceph_vino vino
;
1206 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1207 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1209 in
= ceph_get_inode(req
->r_dentry
->d_sb
, vino
);
1212 dout("new_inode badness got %d\n", err
);
1215 rc
= fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1216 req
->r_request_started
, -1,
1217 &req
->r_caps_reservation
);
1219 pr_err("fill_inode badness on %p got %d\n", in
, rc
);
1228 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1229 struct ceph_mds_session
*session
)
1231 struct dentry
*parent
= req
->r_dentry
;
1232 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1237 struct inode
*snapdir
= NULL
;
1238 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1239 u64 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1240 struct ceph_dentry_info
*di
;
1243 return readdir_prepopulate_inodes_only(req
, session
);
1245 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1246 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1247 parent
= d_find_alias(snapdir
);
1248 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1249 rinfo
->dir_nr
, parent
);
1251 dout("readdir_prepopulate %d items under dn %p\n",
1252 rinfo
->dir_nr
, parent
);
1254 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1257 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1258 struct ceph_vino vino
;
1260 dname
.name
= rinfo
->dir_dname
[i
];
1261 dname
.len
= rinfo
->dir_dname_len
[i
];
1262 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1264 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1265 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1268 dn
= d_lookup(parent
, &dname
);
1269 dout("d_lookup on parent=%p name=%.*s got %p\n",
1270 parent
, dname
.len
, dname
.name
, dn
);
1273 dn
= d_alloc(parent
, &dname
);
1274 dout("d_alloc %p '%.*s' = %p\n", parent
,
1275 dname
.len
, dname
.name
, dn
);
1277 dout("d_alloc badness\n");
1281 err
= ceph_init_dentry(dn
);
1286 } else if (dn
->d_inode
&&
1287 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1288 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1289 dout(" dn %p points to wrong inode %p\n",
1295 /* reorder parent's d_subdirs */
1296 spin_lock(&parent
->d_lock
);
1297 spin_lock_nested(&dn
->d_lock
, DENTRY_D_LOCK_NESTED
);
1298 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1299 spin_unlock(&dn
->d_lock
);
1300 spin_unlock(&parent
->d_lock
);
1304 di
->offset
= ceph_make_fpos(frag
, i
+ req
->r_readdir_offset
);
1310 in
= ceph_get_inode(parent
->d_sb
, vino
);
1312 dout("new_inode badness\n");
1318 dn
= splice_dentry(dn
, in
, NULL
, false);
1323 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1324 req
->r_request_started
, -1,
1325 &req
->r_caps_reservation
) < 0) {
1326 pr_err("fill_inode badness on %p\n", in
);
1330 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1332 req
->r_request_started
);
1337 req
->r_did_prepopulate
= true;
1344 dout("readdir_prepopulate done\n");
1348 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1350 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1353 spin_lock(&ci
->i_ceph_lock
);
1354 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1355 inode
->i_size
= size
;
1356 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1358 /* tell the MDS if we are approaching max_size */
1359 if ((size
<< 1) >= ci
->i_max_size
&&
1360 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1363 spin_unlock(&ci
->i_ceph_lock
);
1368 * Write back inode data in a worker thread. (This can't be done
1369 * in the message handler context.)
1371 void ceph_queue_writeback(struct inode
*inode
)
1374 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1375 &ceph_inode(inode
)->i_wb_work
)) {
1376 dout("ceph_queue_writeback %p\n", inode
);
1378 dout("ceph_queue_writeback %p failed\n", inode
);
1383 static void ceph_writeback_work(struct work_struct
*work
)
1385 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1387 struct inode
*inode
= &ci
->vfs_inode
;
1389 dout("writeback %p\n", inode
);
1390 filemap_fdatawrite(&inode
->i_data
);
1395 * queue an async invalidation
1397 void ceph_queue_invalidate(struct inode
*inode
)
1400 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1401 &ceph_inode(inode
)->i_pg_inv_work
)) {
1402 dout("ceph_queue_invalidate %p\n", inode
);
1404 dout("ceph_queue_invalidate %p failed\n", inode
);
1410 * Invalidate inode pages in a worker thread. (This can't be done
1411 * in the message handler context.)
1413 static void ceph_invalidate_work(struct work_struct
*work
)
1415 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1417 struct inode
*inode
= &ci
->vfs_inode
;
1421 spin_lock(&ci
->i_ceph_lock
);
1422 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1423 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1424 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1426 spin_unlock(&ci
->i_ceph_lock
);
1429 orig_gen
= ci
->i_rdcache_gen
;
1430 spin_unlock(&ci
->i_ceph_lock
);
1432 truncate_inode_pages(&inode
->i_data
, 0);
1434 spin_lock(&ci
->i_ceph_lock
);
1435 if (orig_gen
== ci
->i_rdcache_gen
&&
1436 orig_gen
== ci
->i_rdcache_revoking
) {
1437 dout("invalidate_pages %p gen %d successful\n", inode
,
1439 ci
->i_rdcache_revoking
--;
1442 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1443 inode
, orig_gen
, ci
->i_rdcache_gen
,
1444 ci
->i_rdcache_revoking
);
1446 spin_unlock(&ci
->i_ceph_lock
);
1449 ceph_check_caps(ci
, 0, NULL
);
1456 * called by trunc_wq; take i_mutex ourselves
1458 * We also truncate in a separate thread as well.
1460 static void ceph_vmtruncate_work(struct work_struct
*work
)
1462 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1464 struct inode
*inode
= &ci
->vfs_inode
;
1466 dout("vmtruncate_work %p\n", inode
);
1467 mutex_lock(&inode
->i_mutex
);
1468 __ceph_do_pending_vmtruncate(inode
);
1469 mutex_unlock(&inode
->i_mutex
);
1474 * Queue an async vmtruncate. If we fail to queue work, we will handle
1475 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1477 void ceph_queue_vmtruncate(struct inode
*inode
)
1479 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1482 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1483 &ci
->i_vmtruncate_work
)) {
1484 dout("ceph_queue_vmtruncate %p\n", inode
);
1486 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1487 inode
, ci
->i_truncate_pending
);
1493 * called with i_mutex held.
1495 * Make sure any pending truncation is applied before doing anything
1496 * that may depend on it.
1498 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1500 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1502 int wrbuffer_refs
, finish
= 0;
1505 spin_lock(&ci
->i_ceph_lock
);
1506 if (ci
->i_truncate_pending
== 0) {
1507 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1508 spin_unlock(&ci
->i_ceph_lock
);
1513 * make sure any dirty snapped pages are flushed before we
1514 * possibly truncate them.. so write AND block!
1516 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1517 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1519 spin_unlock(&ci
->i_ceph_lock
);
1520 filemap_write_and_wait_range(&inode
->i_data
, 0,
1521 inode
->i_sb
->s_maxbytes
);
1525 to
= ci
->i_truncate_size
;
1526 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1527 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1528 ci
->i_truncate_pending
, to
);
1529 spin_unlock(&ci
->i_ceph_lock
);
1531 truncate_inode_pages(inode
->i_mapping
, to
);
1533 spin_lock(&ci
->i_ceph_lock
);
1534 if (to
== ci
->i_truncate_size
) {
1535 ci
->i_truncate_pending
= 0;
1538 spin_unlock(&ci
->i_ceph_lock
);
1542 if (wrbuffer_refs
== 0)
1543 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1545 wake_up_all(&ci
->i_cap_wq
);
1552 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1554 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1555 nd_set_link(nd
, ci
->i_symlink
);
1559 static const struct inode_operations ceph_symlink_iops
= {
1560 .readlink
= generic_readlink
,
1561 .follow_link
= ceph_sym_follow_link
,
1567 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1569 struct inode
*inode
= dentry
->d_inode
;
1570 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1571 struct inode
*parent_inode
;
1572 const unsigned int ia_valid
= attr
->ia_valid
;
1573 struct ceph_mds_request
*req
;
1574 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1576 int release
= 0, dirtied
= 0;
1579 int inode_dirty_flags
= 0;
1581 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1584 __ceph_do_pending_vmtruncate(inode
);
1586 err
= inode_change_ok(inode
, attr
);
1590 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1593 return PTR_ERR(req
);
1595 spin_lock(&ci
->i_ceph_lock
);
1596 issued
= __ceph_caps_issued(ci
, NULL
);
1597 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1599 if (ia_valid
& ATTR_UID
) {
1600 dout("setattr %p uid %d -> %d\n", inode
,
1601 from_kuid(&init_user_ns
, inode
->i_uid
),
1602 from_kuid(&init_user_ns
, attr
->ia_uid
));
1603 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1604 inode
->i_uid
= attr
->ia_uid
;
1605 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1606 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1607 !uid_eq(attr
->ia_uid
, inode
->i_uid
)) {
1608 req
->r_args
.setattr
.uid
= cpu_to_le32(
1609 from_kuid(&init_user_ns
, attr
->ia_uid
));
1610 mask
|= CEPH_SETATTR_UID
;
1611 release
|= CEPH_CAP_AUTH_SHARED
;
1614 if (ia_valid
& ATTR_GID
) {
1615 dout("setattr %p gid %d -> %d\n", inode
,
1616 from_kgid(&init_user_ns
, inode
->i_gid
),
1617 from_kgid(&init_user_ns
, attr
->ia_gid
));
1618 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1619 inode
->i_gid
= attr
->ia_gid
;
1620 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1621 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1622 !gid_eq(attr
->ia_gid
, inode
->i_gid
)) {
1623 req
->r_args
.setattr
.gid
= cpu_to_le32(
1624 from_kgid(&init_user_ns
, attr
->ia_gid
));
1625 mask
|= CEPH_SETATTR_GID
;
1626 release
|= CEPH_CAP_AUTH_SHARED
;
1629 if (ia_valid
& ATTR_MODE
) {
1630 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1632 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1633 inode
->i_mode
= attr
->ia_mode
;
1634 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1635 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1636 attr
->ia_mode
!= inode
->i_mode
) {
1637 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1638 mask
|= CEPH_SETATTR_MODE
;
1639 release
|= CEPH_CAP_AUTH_SHARED
;
1643 if (ia_valid
& ATTR_ATIME
) {
1644 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1645 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1646 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1647 if (issued
& CEPH_CAP_FILE_EXCL
) {
1648 ci
->i_time_warp_seq
++;
1649 inode
->i_atime
= attr
->ia_atime
;
1650 dirtied
|= CEPH_CAP_FILE_EXCL
;
1651 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1652 timespec_compare(&inode
->i_atime
,
1653 &attr
->ia_atime
) < 0) {
1654 inode
->i_atime
= attr
->ia_atime
;
1655 dirtied
|= CEPH_CAP_FILE_WR
;
1656 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1657 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1658 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1660 mask
|= CEPH_SETATTR_ATIME
;
1661 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1665 if (ia_valid
& ATTR_MTIME
) {
1666 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1667 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1668 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1669 if (issued
& CEPH_CAP_FILE_EXCL
) {
1670 ci
->i_time_warp_seq
++;
1671 inode
->i_mtime
= attr
->ia_mtime
;
1672 dirtied
|= CEPH_CAP_FILE_EXCL
;
1673 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1674 timespec_compare(&inode
->i_mtime
,
1675 &attr
->ia_mtime
) < 0) {
1676 inode
->i_mtime
= attr
->ia_mtime
;
1677 dirtied
|= CEPH_CAP_FILE_WR
;
1678 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1679 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1680 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1682 mask
|= CEPH_SETATTR_MTIME
;
1683 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1687 if (ia_valid
& ATTR_SIZE
) {
1688 dout("setattr %p size %lld -> %lld\n", inode
,
1689 inode
->i_size
, attr
->ia_size
);
1690 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1694 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1695 attr
->ia_size
> inode
->i_size
) {
1696 inode
->i_size
= attr
->ia_size
;
1698 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1699 inode
->i_ctime
= attr
->ia_ctime
;
1700 ci
->i_reported_size
= attr
->ia_size
;
1701 dirtied
|= CEPH_CAP_FILE_EXCL
;
1702 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1703 attr
->ia_size
!= inode
->i_size
) {
1704 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1705 req
->r_args
.setattr
.old_size
=
1706 cpu_to_le64(inode
->i_size
);
1707 mask
|= CEPH_SETATTR_SIZE
;
1708 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1713 /* these do nothing */
1714 if (ia_valid
& ATTR_CTIME
) {
1715 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1716 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1717 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1718 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1719 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1720 only
? "ctime only" : "ignored");
1721 inode
->i_ctime
= attr
->ia_ctime
;
1724 * if kernel wants to dirty ctime but nothing else,
1725 * we need to choose a cap to dirty under, or do
1726 * a almost-no-op setattr
1728 if (issued
& CEPH_CAP_AUTH_EXCL
)
1729 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1730 else if (issued
& CEPH_CAP_FILE_EXCL
)
1731 dirtied
|= CEPH_CAP_FILE_EXCL
;
1732 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1733 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1735 mask
|= CEPH_SETATTR_CTIME
;
1738 if (ia_valid
& ATTR_FILE
)
1739 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1742 inode_dirty_flags
= __ceph_mark_dirty_caps(ci
, dirtied
);
1743 inode
->i_ctime
= CURRENT_TIME
;
1747 spin_unlock(&ci
->i_ceph_lock
);
1749 if (inode_dirty_flags
)
1750 __mark_inode_dirty(inode
, inode_dirty_flags
);
1753 req
->r_inode
= inode
;
1755 req
->r_inode_drop
= release
;
1756 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1757 req
->r_num_caps
= 1;
1758 parent_inode
= ceph_get_dentry_parent_inode(dentry
);
1759 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
1762 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1763 ceph_cap_string(dirtied
), mask
);
1765 ceph_mdsc_put_request(req
);
1766 __ceph_do_pending_vmtruncate(inode
);
1769 spin_unlock(&ci
->i_ceph_lock
);
1770 ceph_mdsc_put_request(req
);
1775 * Verify that we have a lease on the given mask. If not,
1776 * do a getattr against an mds.
1778 int ceph_do_getattr(struct inode
*inode
, int mask
)
1780 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1781 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1782 struct ceph_mds_request
*req
;
1785 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1786 dout("do_getattr inode %p SNAPDIR\n", inode
);
1790 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1791 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1794 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1796 return PTR_ERR(req
);
1797 req
->r_inode
= inode
;
1799 req
->r_num_caps
= 1;
1800 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1801 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1802 ceph_mdsc_put_request(req
);
1803 dout("do_getattr result=%d\n", err
);
1809 * Check inode permissions. We verify we have a valid value for
1810 * the AUTH cap, then call the generic handler.
1812 int ceph_permission(struct inode
*inode
, int mask
)
1816 if (mask
& MAY_NOT_BLOCK
)
1819 err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1822 err
= generic_permission(inode
, mask
);
1827 * Get all attributes. Hopefully somedata we'll have a statlite()
1828 * and can limit the fields we require to be accurate.
1830 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1833 struct inode
*inode
= dentry
->d_inode
;
1834 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1837 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1839 generic_fillattr(inode
, stat
);
1840 stat
->ino
= ceph_translate_ino(inode
->i_sb
, inode
->i_ino
);
1841 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1842 stat
->dev
= ceph_snap(inode
);
1845 if (S_ISDIR(inode
->i_mode
)) {
1846 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
),
1848 stat
->size
= ci
->i_rbytes
;
1850 stat
->size
= ci
->i_files
+ ci
->i_subdirs
;
1852 stat
->blksize
= 65536;