1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/smp_lock.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/vmalloc.h>
13 #include <linux/pagevec.h>
16 #include "mds_client.h"
17 #include <linux/ceph/decode.h>
20 * Ceph inode operations
22 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23 * setattr, etc.), xattr helpers, and helpers for assimilating
24 * metadata returned by the MDS into our cache.
26 * Also define helpers for doing asynchronous writeback, invalidation,
27 * and truncation for the benefit of those who can't afford to block
28 * (typically because they are in the message handler path).
31 static const struct inode_operations ceph_symlink_iops
;
33 static void ceph_invalidate_work(struct work_struct
*work
);
34 static void ceph_writeback_work(struct work_struct
*work
);
35 static void ceph_vmtruncate_work(struct work_struct
*work
);
38 * find or create an inode, given the ceph ino number
40 struct inode
*ceph_get_inode(struct super_block
*sb
, struct ceph_vino vino
)
43 ino_t t
= ceph_vino_to_ino(vino
);
45 inode
= iget5_locked(sb
, t
, ceph_ino_compare
, ceph_set_ino_cb
, &vino
);
47 return ERR_PTR(-ENOMEM
);
48 if (inode
->i_state
& I_NEW
) {
49 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
50 inode
, ceph_vinop(inode
), (u64
)inode
->i_ino
);
51 unlock_new_inode(inode
);
54 dout("get_inode on %lu=%llx.%llx got %p\n", inode
->i_ino
, vino
.ino
,
60 * get/constuct snapdir inode for a given directory
62 struct inode
*ceph_get_snapdir(struct inode
*parent
)
64 struct ceph_vino vino
= {
65 .ino
= ceph_ino(parent
),
68 struct inode
*inode
= ceph_get_inode(parent
->i_sb
, vino
);
69 struct ceph_inode_info
*ci
= ceph_inode(inode
);
71 BUG_ON(!S_ISDIR(parent
->i_mode
));
74 inode
->i_mode
= parent
->i_mode
;
75 inode
->i_uid
= parent
->i_uid
;
76 inode
->i_gid
= parent
->i_gid
;
77 inode
->i_op
= &ceph_dir_iops
;
78 inode
->i_fop
= &ceph_dir_fops
;
79 ci
->i_snap_caps
= CEPH_CAP_PIN
; /* so we can open */
84 const struct inode_operations ceph_file_iops
= {
85 .permission
= ceph_permission
,
86 .setattr
= ceph_setattr
,
87 .getattr
= ceph_getattr
,
88 .setxattr
= ceph_setxattr
,
89 .getxattr
= ceph_getxattr
,
90 .listxattr
= ceph_listxattr
,
91 .removexattr
= ceph_removexattr
,
96 * We use a 'frag tree' to keep track of the MDS's directory fragments
97 * for a given inode (usually there is just a single fragment). We
98 * need to know when a child frag is delegated to a new MDS, or when
99 * it is flagged as replicated, so we can direct our requests
104 * find/create a frag in the tree
106 static struct ceph_inode_frag
*__get_or_create_frag(struct ceph_inode_info
*ci
,
110 struct rb_node
*parent
= NULL
;
111 struct ceph_inode_frag
*frag
;
114 p
= &ci
->i_fragtree
.rb_node
;
117 frag
= rb_entry(parent
, struct ceph_inode_frag
, node
);
118 c
= ceph_frag_compare(f
, frag
->frag
);
127 frag
= kmalloc(sizeof(*frag
), GFP_NOFS
);
129 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
130 "frag %x\n", &ci
->vfs_inode
,
131 ceph_vinop(&ci
->vfs_inode
), f
);
132 return ERR_PTR(-ENOMEM
);
139 rb_link_node(&frag
->node
, parent
, p
);
140 rb_insert_color(&frag
->node
, &ci
->i_fragtree
);
142 dout("get_or_create_frag added %llx.%llx frag %x\n",
143 ceph_vinop(&ci
->vfs_inode
), f
);
148 * find a specific frag @f
150 struct ceph_inode_frag
*__ceph_find_frag(struct ceph_inode_info
*ci
, u32 f
)
152 struct rb_node
*n
= ci
->i_fragtree
.rb_node
;
155 struct ceph_inode_frag
*frag
=
156 rb_entry(n
, struct ceph_inode_frag
, node
);
157 int c
= ceph_frag_compare(f
, frag
->frag
);
169 * Choose frag containing the given value @v. If @pfrag is
170 * specified, copy the frag delegation info to the caller if
173 u32
ceph_choose_frag(struct ceph_inode_info
*ci
, u32 v
,
174 struct ceph_inode_frag
*pfrag
,
177 u32 t
= ceph_frag_make(0, 0);
178 struct ceph_inode_frag
*frag
;
185 mutex_lock(&ci
->i_fragtree_mutex
);
187 WARN_ON(!ceph_frag_contains_value(t
, v
));
188 frag
= __ceph_find_frag(ci
, t
);
190 break; /* t is a leaf */
191 if (frag
->split_by
== 0) {
193 memcpy(pfrag
, frag
, sizeof(*pfrag
));
200 nway
= 1 << frag
->split_by
;
201 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v
, t
,
202 frag
->split_by
, nway
);
203 for (i
= 0; i
< nway
; i
++) {
204 n
= ceph_frag_make_child(t
, frag
->split_by
, i
);
205 if (ceph_frag_contains_value(n
, v
)) {
212 dout("choose_frag(%x) = %x\n", v
, t
);
214 mutex_unlock(&ci
->i_fragtree_mutex
);
219 * Process dirfrag (delegation) info from the mds. Include leaf
220 * fragment in tree ONLY if ndist > 0. Otherwise, only
221 * branches/splits are included in i_fragtree)
223 static int ceph_fill_dirfrag(struct inode
*inode
,
224 struct ceph_mds_reply_dirfrag
*dirinfo
)
226 struct ceph_inode_info
*ci
= ceph_inode(inode
);
227 struct ceph_inode_frag
*frag
;
228 u32 id
= le32_to_cpu(dirinfo
->frag
);
229 int mds
= le32_to_cpu(dirinfo
->auth
);
230 int ndist
= le32_to_cpu(dirinfo
->ndist
);
234 mutex_lock(&ci
->i_fragtree_mutex
);
236 /* no delegation info needed. */
237 frag
= __ceph_find_frag(ci
, id
);
240 if (frag
->split_by
== 0) {
241 /* tree leaf, remove */
242 dout("fill_dirfrag removed %llx.%llx frag %x"
243 " (no ref)\n", ceph_vinop(inode
), id
);
244 rb_erase(&frag
->node
, &ci
->i_fragtree
);
247 /* tree branch, keep and clear */
248 dout("fill_dirfrag cleared %llx.%llx frag %x"
249 " referral\n", ceph_vinop(inode
), id
);
257 /* find/add this frag to store mds delegation info */
258 frag
= __get_or_create_frag(ci
, id
);
260 /* this is not the end of the world; we can continue
261 with bad/inaccurate delegation info */
262 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
263 ceph_vinop(inode
), le32_to_cpu(dirinfo
->frag
));
269 frag
->ndist
= min_t(u32
, ndist
, CEPH_MAX_DIRFRAG_REP
);
270 for (i
= 0; i
< frag
->ndist
; i
++)
271 frag
->dist
[i
] = le32_to_cpu(dirinfo
->dist
[i
]);
272 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
273 ceph_vinop(inode
), frag
->frag
, frag
->ndist
);
276 mutex_unlock(&ci
->i_fragtree_mutex
);
282 * initialize a newly allocated inode.
284 struct inode
*ceph_alloc_inode(struct super_block
*sb
)
286 struct ceph_inode_info
*ci
;
289 ci
= kmem_cache_alloc(ceph_inode_cachep
, GFP_NOFS
);
293 dout("alloc_inode %p\n", &ci
->vfs_inode
);
296 ci
->i_time_warp_seq
= 0;
297 ci
->i_ceph_flags
= 0;
298 ci
->i_release_count
= 0;
299 ci
->i_symlink
= NULL
;
301 ci
->i_fragtree
= RB_ROOT
;
302 mutex_init(&ci
->i_fragtree_mutex
);
304 ci
->i_xattrs
.blob
= NULL
;
305 ci
->i_xattrs
.prealloc_blob
= NULL
;
306 ci
->i_xattrs
.dirty
= false;
307 ci
->i_xattrs
.index
= RB_ROOT
;
308 ci
->i_xattrs
.count
= 0;
309 ci
->i_xattrs
.names_size
= 0;
310 ci
->i_xattrs
.vals_size
= 0;
311 ci
->i_xattrs
.version
= 0;
312 ci
->i_xattrs
.index_version
= 0;
314 ci
->i_caps
= RB_ROOT
;
315 ci
->i_auth_cap
= NULL
;
316 ci
->i_dirty_caps
= 0;
317 ci
->i_flushing_caps
= 0;
318 INIT_LIST_HEAD(&ci
->i_dirty_item
);
319 INIT_LIST_HEAD(&ci
->i_flushing_item
);
320 ci
->i_cap_flush_seq
= 0;
321 ci
->i_cap_flush_last_tid
= 0;
322 memset(&ci
->i_cap_flush_tid
, 0, sizeof(ci
->i_cap_flush_tid
));
323 init_waitqueue_head(&ci
->i_cap_wq
);
324 ci
->i_hold_caps_min
= 0;
325 ci
->i_hold_caps_max
= 0;
326 INIT_LIST_HEAD(&ci
->i_cap_delay_list
);
327 ci
->i_cap_exporting_mds
= 0;
328 ci
->i_cap_exporting_mseq
= 0;
329 ci
->i_cap_exporting_issued
= 0;
330 INIT_LIST_HEAD(&ci
->i_cap_snaps
);
331 ci
->i_head_snapc
= NULL
;
334 for (i
= 0; i
< CEPH_FILE_MODE_NUM
; i
++)
335 ci
->i_nr_by_mode
[i
] = 0;
337 ci
->i_truncate_seq
= 0;
338 ci
->i_truncate_size
= 0;
339 ci
->i_truncate_pending
= 0;
342 ci
->i_reported_size
= 0;
343 ci
->i_wanted_max_size
= 0;
344 ci
->i_requested_max_size
= 0;
348 ci
->i_rdcache_ref
= 0;
350 ci
->i_wrbuffer_ref
= 0;
351 ci
->i_wrbuffer_ref_head
= 0;
352 ci
->i_shared_gen
= 0;
353 ci
->i_rdcache_gen
= 0;
354 ci
->i_rdcache_revoking
= 0;
356 INIT_LIST_HEAD(&ci
->i_unsafe_writes
);
357 INIT_LIST_HEAD(&ci
->i_unsafe_dirops
);
358 spin_lock_init(&ci
->i_unsafe_lock
);
360 ci
->i_snap_realm
= NULL
;
361 INIT_LIST_HEAD(&ci
->i_snap_realm_item
);
362 INIT_LIST_HEAD(&ci
->i_snap_flush_item
);
364 INIT_WORK(&ci
->i_wb_work
, ceph_writeback_work
);
365 INIT_WORK(&ci
->i_pg_inv_work
, ceph_invalidate_work
);
367 INIT_WORK(&ci
->i_vmtruncate_work
, ceph_vmtruncate_work
);
369 return &ci
->vfs_inode
;
372 void ceph_destroy_inode(struct inode
*inode
)
374 struct ceph_inode_info
*ci
= ceph_inode(inode
);
375 struct ceph_inode_frag
*frag
;
378 dout("destroy_inode %p ino %llx.%llx\n", inode
, ceph_vinop(inode
));
380 ceph_queue_caps_release(inode
);
383 * we may still have a snap_realm reference if there are stray
384 * caps in i_cap_exporting_issued or i_snap_caps.
386 if (ci
->i_snap_realm
) {
387 struct ceph_mds_client
*mdsc
=
388 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
389 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
391 dout(" dropping residual ref to snap realm %p\n", realm
);
392 spin_lock(&realm
->inodes_with_caps_lock
);
393 list_del_init(&ci
->i_snap_realm_item
);
394 spin_unlock(&realm
->inodes_with_caps_lock
);
395 ceph_put_snap_realm(mdsc
, realm
);
398 kfree(ci
->i_symlink
);
399 while ((n
= rb_first(&ci
->i_fragtree
)) != NULL
) {
400 frag
= rb_entry(n
, struct ceph_inode_frag
, node
);
401 rb_erase(n
, &ci
->i_fragtree
);
405 __ceph_destroy_xattrs(ci
);
406 if (ci
->i_xattrs
.blob
)
407 ceph_buffer_put(ci
->i_xattrs
.blob
);
408 if (ci
->i_xattrs
.prealloc_blob
)
409 ceph_buffer_put(ci
->i_xattrs
.prealloc_blob
);
411 kmem_cache_free(ceph_inode_cachep
, ci
);
416 * Helpers to fill in size, ctime, mtime, and atime. We have to be
417 * careful because either the client or MDS may have more up to date
418 * info, depending on which capabilities are held, and whether
419 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
420 * and size are monotonically increasing, except when utimes() or
421 * truncate() increments the corresponding _seq values.)
423 int ceph_fill_file_size(struct inode
*inode
, int issued
,
424 u32 truncate_seq
, u64 truncate_size
, u64 size
)
426 struct ceph_inode_info
*ci
= ceph_inode(inode
);
429 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) > 0 ||
430 (truncate_seq
== ci
->i_truncate_seq
&& size
> inode
->i_size
)) {
431 dout("size %lld -> %llu\n", inode
->i_size
, size
);
432 inode
->i_size
= size
;
433 inode
->i_blocks
= (size
+ (1<<9) - 1) >> 9;
434 ci
->i_reported_size
= size
;
435 if (truncate_seq
!= ci
->i_truncate_seq
) {
436 dout("truncate_seq %u -> %u\n",
437 ci
->i_truncate_seq
, truncate_seq
);
438 ci
->i_truncate_seq
= truncate_seq
;
440 * If we hold relevant caps, or in the case where we're
441 * not the only client referencing this file and we
442 * don't hold those caps, then we need to check whether
443 * the file is either opened or mmaped
445 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_RD
|
446 CEPH_CAP_FILE_WR
|CEPH_CAP_FILE_BUFFER
|
448 CEPH_CAP_FILE_LAZYIO
)) ||
449 mapping_mapped(inode
->i_mapping
) ||
450 __ceph_caps_file_wanted(ci
)) {
451 ci
->i_truncate_pending
++;
456 if (ceph_seq_cmp(truncate_seq
, ci
->i_truncate_seq
) >= 0 &&
457 ci
->i_truncate_size
!= truncate_size
) {
458 dout("truncate_size %lld -> %llu\n", ci
->i_truncate_size
,
460 ci
->i_truncate_size
= truncate_size
;
465 void ceph_fill_file_time(struct inode
*inode
, int issued
,
466 u64 time_warp_seq
, struct timespec
*ctime
,
467 struct timespec
*mtime
, struct timespec
*atime
)
469 struct ceph_inode_info
*ci
= ceph_inode(inode
);
472 if (issued
& (CEPH_CAP_FILE_EXCL
|
474 CEPH_CAP_FILE_BUFFER
|
476 CEPH_CAP_XATTR_EXCL
)) {
477 if (timespec_compare(ctime
, &inode
->i_ctime
) > 0) {
478 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
479 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
480 ctime
->tv_sec
, ctime
->tv_nsec
);
481 inode
->i_ctime
= *ctime
;
483 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) > 0) {
484 /* the MDS did a utimes() */
485 dout("mtime %ld.%09ld -> %ld.%09ld "
487 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
488 mtime
->tv_sec
, mtime
->tv_nsec
,
489 ci
->i_time_warp_seq
, (int)time_warp_seq
);
491 inode
->i_mtime
= *mtime
;
492 inode
->i_atime
= *atime
;
493 ci
->i_time_warp_seq
= time_warp_seq
;
494 } else if (time_warp_seq
== ci
->i_time_warp_seq
) {
495 /* nobody did utimes(); take the max */
496 if (timespec_compare(mtime
, &inode
->i_mtime
) > 0) {
497 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
498 inode
->i_mtime
.tv_sec
,
499 inode
->i_mtime
.tv_nsec
,
500 mtime
->tv_sec
, mtime
->tv_nsec
);
501 inode
->i_mtime
= *mtime
;
503 if (timespec_compare(atime
, &inode
->i_atime
) > 0) {
504 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
505 inode
->i_atime
.tv_sec
,
506 inode
->i_atime
.tv_nsec
,
507 atime
->tv_sec
, atime
->tv_nsec
);
508 inode
->i_atime
= *atime
;
510 } else if (issued
& CEPH_CAP_FILE_EXCL
) {
511 /* we did a utimes(); ignore mds values */
516 /* we have no write|excl caps; whatever the MDS says is true */
517 if (ceph_seq_cmp(time_warp_seq
, ci
->i_time_warp_seq
) >= 0) {
518 inode
->i_ctime
= *ctime
;
519 inode
->i_mtime
= *mtime
;
520 inode
->i_atime
= *atime
;
521 ci
->i_time_warp_seq
= time_warp_seq
;
526 if (warn
) /* time_warp_seq shouldn't go backwards */
527 dout("%p mds time_warp_seq %llu < %u\n",
528 inode
, time_warp_seq
, ci
->i_time_warp_seq
);
532 * Populate an inode based on info from mds. May be called on new or
535 static int fill_inode(struct inode
*inode
,
536 struct ceph_mds_reply_info_in
*iinfo
,
537 struct ceph_mds_reply_dirfrag
*dirinfo
,
538 struct ceph_mds_session
*session
,
539 unsigned long ttl_from
, int cap_fmode
,
540 struct ceph_cap_reservation
*caps_reservation
)
542 struct ceph_mds_reply_inode
*info
= iinfo
->in
;
543 struct ceph_inode_info
*ci
= ceph_inode(inode
);
545 int issued
, implemented
;
546 struct timespec mtime
, atime
, ctime
;
548 struct ceph_buffer
*xattr_blob
= NULL
;
552 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
553 inode
, ceph_vinop(inode
), le64_to_cpu(info
->version
),
557 * prealloc xattr data, if it looks like we'll need it. only
558 * if len > 4 (meaning there are actually xattrs; the first 4
559 * bytes are the xattr count).
561 if (iinfo
->xattr_len
> 4) {
562 xattr_blob
= ceph_buffer_new(iinfo
->xattr_len
, GFP_NOFS
);
564 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
568 spin_lock(&inode
->i_lock
);
571 * provided version will be odd if inode value is projected,
572 * even if stable. skip the update if we have newer stable
573 * info (ours>=theirs, e.g. due to racing mds replies), unless
574 * we are getting projected (unstable) info (in which case the
575 * version is odd, and we want ours>theirs).
581 if (le64_to_cpu(info
->version
) > 0 &&
582 (ci
->i_version
& ~1) >= le64_to_cpu(info
->version
))
585 issued
= __ceph_caps_issued(ci
, &implemented
);
586 issued
|= implemented
| __ceph_caps_dirty(ci
);
589 ci
->i_version
= le64_to_cpu(info
->version
);
591 inode
->i_rdev
= le32_to_cpu(info
->rdev
);
593 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
594 inode
->i_mode
= le32_to_cpu(info
->mode
);
595 inode
->i_uid
= le32_to_cpu(info
->uid
);
596 inode
->i_gid
= le32_to_cpu(info
->gid
);
597 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
598 inode
->i_uid
, inode
->i_gid
);
601 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
602 inode
->i_nlink
= le32_to_cpu(info
->nlink
);
604 /* be careful with mtime, atime, size */
605 ceph_decode_timespec(&atime
, &info
->atime
);
606 ceph_decode_timespec(&mtime
, &info
->mtime
);
607 ceph_decode_timespec(&ctime
, &info
->ctime
);
608 queue_trunc
= ceph_fill_file_size(inode
, issued
,
609 le32_to_cpu(info
->truncate_seq
),
610 le64_to_cpu(info
->truncate_size
),
611 le64_to_cpu(info
->size
));
612 ceph_fill_file_time(inode
, issued
,
613 le32_to_cpu(info
->time_warp_seq
),
614 &ctime
, &mtime
, &atime
);
616 /* only update max_size on auth cap */
617 if ((info
->cap
.flags
& CEPH_CAP_FLAG_AUTH
) &&
618 ci
->i_max_size
!= le64_to_cpu(info
->max_size
)) {
619 dout("max_size %lld -> %llu\n", ci
->i_max_size
,
620 le64_to_cpu(info
->max_size
));
621 ci
->i_max_size
= le64_to_cpu(info
->max_size
);
624 ci
->i_layout
= info
->layout
;
625 inode
->i_blkbits
= fls(le32_to_cpu(info
->layout
.fl_stripe_unit
)) - 1;
628 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
629 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 &&
630 le64_to_cpu(info
->xattr_version
) > ci
->i_xattrs
.version
) {
631 if (ci
->i_xattrs
.blob
)
632 ceph_buffer_put(ci
->i_xattrs
.blob
);
633 ci
->i_xattrs
.blob
= xattr_blob
;
635 memcpy(ci
->i_xattrs
.blob
->vec
.iov_base
,
636 iinfo
->xattr_data
, iinfo
->xattr_len
);
637 ci
->i_xattrs
.version
= le64_to_cpu(info
->xattr_version
);
641 inode
->i_mapping
->a_ops
= &ceph_aops
;
642 inode
->i_mapping
->backing_dev_info
=
643 &ceph_sb_to_client(inode
->i_sb
)->backing_dev_info
;
645 switch (inode
->i_mode
& S_IFMT
) {
650 init_special_inode(inode
, inode
->i_mode
, inode
->i_rdev
);
651 inode
->i_op
= &ceph_file_iops
;
654 inode
->i_op
= &ceph_file_iops
;
655 inode
->i_fop
= &ceph_file_fops
;
658 inode
->i_op
= &ceph_symlink_iops
;
659 if (!ci
->i_symlink
) {
660 int symlen
= iinfo
->symlink_len
;
663 BUG_ON(symlen
!= inode
->i_size
);
664 spin_unlock(&inode
->i_lock
);
667 sym
= kmalloc(symlen
+1, GFP_NOFS
);
670 memcpy(sym
, iinfo
->symlink
, symlen
);
673 spin_lock(&inode
->i_lock
);
677 kfree(sym
); /* lost a race */
681 inode
->i_op
= &ceph_dir_iops
;
682 inode
->i_fop
= &ceph_dir_fops
;
684 ci
->i_files
= le64_to_cpu(info
->files
);
685 ci
->i_subdirs
= le64_to_cpu(info
->subdirs
);
686 ci
->i_rbytes
= le64_to_cpu(info
->rbytes
);
687 ci
->i_rfiles
= le64_to_cpu(info
->rfiles
);
688 ci
->i_rsubdirs
= le64_to_cpu(info
->rsubdirs
);
689 ceph_decode_timespec(&ci
->i_rctime
, &info
->rctime
);
691 /* set dir completion flag? */
692 if (ci
->i_files
== 0 && ci
->i_subdirs
== 0 &&
693 ceph_snap(inode
) == CEPH_NOSNAP
&&
694 (le32_to_cpu(info
->cap
.caps
) & CEPH_CAP_FILE_SHARED
) &&
695 (issued
& CEPH_CAP_FILE_EXCL
) == 0 &&
696 (ci
->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
697 dout(" marking %p complete (empty)\n", inode
);
698 ci
->i_ceph_flags
|= CEPH_I_COMPLETE
;
699 ci
->i_max_offset
= 2;
702 /* it may be better to set st_size in getattr instead? */
703 if (ceph_test_mount_opt(ceph_sb_to_client(inode
->i_sb
), RBYTES
))
704 inode
->i_size
= ci
->i_rbytes
;
707 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
708 ceph_vinop(inode
), inode
->i_mode
);
712 spin_unlock(&inode
->i_lock
);
714 /* queue truncate if we saw i_size decrease */
716 ceph_queue_vmtruncate(inode
);
718 /* populate frag tree */
719 /* FIXME: move me up, if/when version reflects fragtree changes */
720 nsplits
= le32_to_cpu(info
->fragtree
.nsplits
);
721 mutex_lock(&ci
->i_fragtree_mutex
);
722 for (i
= 0; i
< nsplits
; i
++) {
723 u32 id
= le32_to_cpu(info
->fragtree
.splits
[i
].frag
);
724 struct ceph_inode_frag
*frag
= __get_or_create_frag(ci
, id
);
728 frag
->split_by
= le32_to_cpu(info
->fragtree
.splits
[i
].by
);
729 dout(" frag %x split by %d\n", frag
->frag
, frag
->split_by
);
731 mutex_unlock(&ci
->i_fragtree_mutex
);
733 /* were we issued a capability? */
734 if (info
->cap
.caps
) {
735 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
736 ceph_add_cap(inode
, session
,
737 le64_to_cpu(info
->cap
.cap_id
),
739 le32_to_cpu(info
->cap
.caps
),
740 le32_to_cpu(info
->cap
.wanted
),
741 le32_to_cpu(info
->cap
.seq
),
742 le32_to_cpu(info
->cap
.mseq
),
743 le64_to_cpu(info
->cap
.realm
),
747 spin_lock(&inode
->i_lock
);
748 dout(" %p got snap_caps %s\n", inode
,
749 ceph_cap_string(le32_to_cpu(info
->cap
.caps
)));
750 ci
->i_snap_caps
|= le32_to_cpu(info
->cap
.caps
);
752 __ceph_get_fmode(ci
, cap_fmode
);
753 spin_unlock(&inode
->i_lock
);
755 } else if (cap_fmode
>= 0) {
756 pr_warning("mds issued no caps on %llx.%llx\n",
758 __ceph_get_fmode(ci
, cap_fmode
);
761 /* update delegation info? */
763 ceph_fill_dirfrag(inode
, dirinfo
);
769 ceph_buffer_put(xattr_blob
);
774 * caller should hold session s_mutex.
776 static void update_dentry_lease(struct dentry
*dentry
,
777 struct ceph_mds_reply_lease
*lease
,
778 struct ceph_mds_session
*session
,
779 unsigned long from_time
)
781 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
782 long unsigned duration
= le32_to_cpu(lease
->duration_ms
);
783 long unsigned ttl
= from_time
+ (duration
* HZ
) / 1000;
784 long unsigned half_ttl
= from_time
+ (duration
* HZ
/ 2) / 1000;
787 /* only track leases on regular dentries */
788 if (dentry
->d_op
!= &ceph_dentry_ops
)
791 spin_lock(&dentry
->d_lock
);
792 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
793 dentry
, le16_to_cpu(lease
->mask
), duration
, ttl
);
795 /* make lease_rdcache_gen match directory */
796 dir
= dentry
->d_parent
->d_inode
;
797 di
->lease_shared_gen
= ceph_inode(dir
)->i_shared_gen
;
799 if (lease
->mask
== 0)
802 if (di
->lease_gen
== session
->s_cap_gen
&&
803 time_before(ttl
, dentry
->d_time
))
804 goto out_unlock
; /* we already have a newer lease. */
806 if (di
->lease_session
&& di
->lease_session
!= session
)
809 ceph_dentry_lru_touch(dentry
);
811 if (!di
->lease_session
)
812 di
->lease_session
= ceph_get_mds_session(session
);
813 di
->lease_gen
= session
->s_cap_gen
;
814 di
->lease_seq
= le32_to_cpu(lease
->seq
);
815 di
->lease_renew_after
= half_ttl
;
816 di
->lease_renew_from
= 0;
817 dentry
->d_time
= ttl
;
819 spin_unlock(&dentry
->d_lock
);
824 * Set dentry's directory position based on the current dir's max, and
825 * order it in d_subdirs, so that dcache_readdir behaves.
827 static void ceph_set_dentry_offset(struct dentry
*dn
)
829 struct dentry
*dir
= dn
->d_parent
;
830 struct inode
*inode
= dn
->d_parent
->d_inode
;
831 struct ceph_dentry_info
*di
;
835 di
= ceph_dentry(dn
);
837 spin_lock(&inode
->i_lock
);
838 if ((ceph_inode(inode
)->i_ceph_flags
& CEPH_I_COMPLETE
) == 0) {
839 spin_unlock(&inode
->i_lock
);
842 di
->offset
= ceph_inode(inode
)->i_max_offset
++;
843 spin_unlock(&inode
->i_lock
);
845 spin_lock(&dcache_lock
);
846 spin_lock(&dn
->d_lock
);
847 list_move(&dn
->d_u
.d_child
, &dir
->d_subdirs
);
848 dout("set_dentry_offset %p %lld (%p %p)\n", dn
, di
->offset
,
849 dn
->d_u
.d_child
.prev
, dn
->d_u
.d_child
.next
);
850 spin_unlock(&dn
->d_lock
);
851 spin_unlock(&dcache_lock
);
855 * splice a dentry to an inode.
856 * caller must hold directory i_mutex for this to be safe.
858 * we will only rehash the resulting dentry if @prehash is
859 * true; @prehash will be set to false (for the benefit of
860 * the caller) if we fail.
862 static struct dentry
*splice_dentry(struct dentry
*dn
, struct inode
*in
,
863 bool *prehash
, bool set_offset
)
865 struct dentry
*realdn
;
869 /* dn must be unhashed */
872 realdn
= d_materialise_unique(dn
, in
);
873 if (IS_ERR(realdn
)) {
874 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
875 PTR_ERR(realdn
), dn
, in
, ceph_vinop(in
));
877 *prehash
= false; /* don't rehash on error */
878 dn
= realdn
; /* note realdn contains the error */
881 dout("dn %p (%d) spliced with %p (%d) "
882 "inode %p ino %llx.%llx\n",
883 dn
, atomic_read(&dn
->d_count
),
884 realdn
, atomic_read(&realdn
->d_count
),
885 realdn
->d_inode
, ceph_vinop(realdn
->d_inode
));
889 BUG_ON(!ceph_dentry(dn
));
890 dout("dn %p attached to %p ino %llx.%llx\n",
891 dn
, dn
->d_inode
, ceph_vinop(dn
->d_inode
));
893 if ((!prehash
|| *prehash
) && d_unhashed(dn
))
896 ceph_set_dentry_offset(dn
);
902 * Incorporate results into the local cache. This is either just
903 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
906 * A reply may contain
907 * a directory inode along with a dentry.
908 * and/or a target inode
910 * Called with snap_rwsem (read).
912 int ceph_fill_trace(struct super_block
*sb
, struct ceph_mds_request
*req
,
913 struct ceph_mds_session
*session
)
915 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
916 struct inode
*in
= NULL
;
917 struct ceph_mds_reply_inode
*ininfo
;
918 struct ceph_vino vino
;
919 struct ceph_fs_client
*fsc
= ceph_sb_to_client(sb
);
923 dout("fill_trace %p is_dentry %d is_target %d\n", req
,
924 rinfo
->head
->is_dentry
, rinfo
->head
->is_target
);
930 * If we resend completed ops to a recovering mds, we get no
931 * trace. Since that is very rare, pretend this is the case
932 * to ensure the 'no trace' handlers in the callers behave.
934 * Fill in inodes unconditionally to avoid breaking cap
937 if (rinfo
->head
->op
& CEPH_MDS_OP_WRITE
) {
938 pr_info("fill_trace faking empty trace on %lld %s\n",
939 req
->r_tid
, ceph_mds_op_name(rinfo
->head
->op
));
940 if (rinfo
->head
->is_dentry
) {
941 rinfo
->head
->is_dentry
= 0;
942 err
= fill_inode(req
->r_locked_dir
,
943 &rinfo
->diri
, rinfo
->dirfrag
,
944 session
, req
->r_request_started
, -1);
946 if (rinfo
->head
->is_target
) {
947 rinfo
->head
->is_target
= 0;
948 ininfo
= rinfo
->targeti
.in
;
949 vino
.ino
= le64_to_cpu(ininfo
->ino
);
950 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
951 in
= ceph_get_inode(sb
, vino
);
952 err
= fill_inode(in
, &rinfo
->targeti
, NULL
,
953 session
, req
->r_request_started
,
960 if (!rinfo
->head
->is_target
&& !rinfo
->head
->is_dentry
) {
961 dout("fill_trace reply is empty!\n");
962 if (rinfo
->head
->result
== 0 && req
->r_locked_dir
)
963 ceph_invalidate_dir_request(req
);
967 if (rinfo
->head
->is_dentry
) {
968 struct inode
*dir
= req
->r_locked_dir
;
970 err
= fill_inode(dir
, &rinfo
->diri
, rinfo
->dirfrag
,
971 session
, req
->r_request_started
, -1,
972 &req
->r_caps_reservation
);
978 * ignore null lease/binding on snapdir ENOENT, or else we
979 * will have trouble splicing in the virtual snapdir later
981 if (rinfo
->head
->is_dentry
&& !req
->r_aborted
&&
982 (rinfo
->head
->is_target
|| strncmp(req
->r_dentry
->d_name
.name
,
983 fsc
->mount_options
->snapdir_name
,
984 req
->r_dentry
->d_name
.len
))) {
986 * lookup link rename : null -> possibly existing inode
987 * mknod symlink mkdir : null -> new inode
988 * unlink : linked -> null
990 struct inode
*dir
= req
->r_locked_dir
;
991 struct dentry
*dn
= req
->r_dentry
;
992 bool have_dir_cap
, have_lease
;
996 BUG_ON(dn
->d_parent
->d_inode
!= dir
);
997 BUG_ON(ceph_ino(dir
) !=
998 le64_to_cpu(rinfo
->diri
.in
->ino
));
999 BUG_ON(ceph_snap(dir
) !=
1000 le64_to_cpu(rinfo
->diri
.in
->snapid
));
1002 /* do we have a lease on the whole dir? */
1004 (le32_to_cpu(rinfo
->diri
.in
->cap
.caps
) &
1005 CEPH_CAP_FILE_SHARED
);
1007 /* do we have a dn lease? */
1008 have_lease
= have_dir_cap
||
1009 (le16_to_cpu(rinfo
->dlease
->mask
) &
1013 dout("fill_trace no dentry lease or dir cap\n");
1016 if (req
->r_old_dentry
&& req
->r_op
== CEPH_MDS_OP_RENAME
) {
1017 dout(" src %p '%.*s' dst %p '%.*s'\n",
1019 req
->r_old_dentry
->d_name
.len
,
1020 req
->r_old_dentry
->d_name
.name
,
1021 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1022 dout("fill_trace doing d_move %p -> %p\n",
1023 req
->r_old_dentry
, dn
);
1025 /* d_move screws up d_subdirs order */
1026 ceph_i_clear(dir
, CEPH_I_COMPLETE
);
1028 d_move(req
->r_old_dentry
, dn
);
1029 dout(" src %p '%.*s' dst %p '%.*s'\n",
1031 req
->r_old_dentry
->d_name
.len
,
1032 req
->r_old_dentry
->d_name
.name
,
1033 dn
, dn
->d_name
.len
, dn
->d_name
.name
);
1035 /* ensure target dentry is invalidated, despite
1036 rehashing bug in vfs_rename_dir */
1037 ceph_invalidate_dentry_lease(dn
);
1039 /* take overwritten dentry's readdir offset */
1040 dout("dn %p gets %p offset %lld (old offset %lld)\n",
1041 req
->r_old_dentry
, dn
, ceph_dentry(dn
)->offset
,
1042 ceph_dentry(req
->r_old_dentry
)->offset
);
1043 ceph_dentry(req
->r_old_dentry
)->offset
=
1044 ceph_dentry(dn
)->offset
;
1046 dn
= req
->r_old_dentry
; /* use old_dentry */
1051 if (!rinfo
->head
->is_target
) {
1052 dout("fill_trace null dentry\n");
1054 dout("d_delete %p\n", dn
);
1057 dout("d_instantiate %p NULL\n", dn
);
1058 d_instantiate(dn
, NULL
);
1059 if (have_lease
&& d_unhashed(dn
))
1061 update_dentry_lease(dn
, rinfo
->dlease
,
1063 req
->r_request_started
);
1068 /* attach proper inode */
1069 ininfo
= rinfo
->targeti
.in
;
1070 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1071 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1074 in
= ceph_get_inode(sb
, vino
);
1076 pr_err("fill_trace bad get_inode "
1077 "%llx.%llx\n", vino
.ino
, vino
.snap
);
1082 dn
= splice_dentry(dn
, in
, &have_lease
, true);
1087 req
->r_dentry
= dn
; /* may have spliced */
1089 } else if (ceph_ino(in
) == vino
.ino
&&
1090 ceph_snap(in
) == vino
.snap
) {
1093 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1094 dn
, in
, ceph_ino(in
), ceph_snap(in
),
1095 vino
.ino
, vino
.snap
);
1101 update_dentry_lease(dn
, rinfo
->dlease
, session
,
1102 req
->r_request_started
);
1103 dout(" final dn %p\n", dn
);
1105 } else if (req
->r_op
== CEPH_MDS_OP_LOOKUPSNAP
||
1106 req
->r_op
== CEPH_MDS_OP_MKSNAP
) {
1107 struct dentry
*dn
= req
->r_dentry
;
1109 /* fill out a snapdir LOOKUPSNAP dentry */
1111 BUG_ON(!req
->r_locked_dir
);
1112 BUG_ON(ceph_snap(req
->r_locked_dir
) != CEPH_SNAPDIR
);
1113 ininfo
= rinfo
->targeti
.in
;
1114 vino
.ino
= le64_to_cpu(ininfo
->ino
);
1115 vino
.snap
= le64_to_cpu(ininfo
->snapid
);
1116 in
= ceph_get_inode(sb
, vino
);
1118 pr_err("fill_inode get_inode badness %llx.%llx\n",
1119 vino
.ino
, vino
.snap
);
1124 dout(" linking snapped dir %p to dn %p\n", in
, dn
);
1125 dn
= splice_dentry(dn
, in
, NULL
, true);
1130 req
->r_dentry
= dn
; /* may have spliced */
1132 rinfo
->head
->is_dentry
= 1; /* fool notrace handlers */
1135 if (rinfo
->head
->is_target
) {
1136 vino
.ino
= le64_to_cpu(rinfo
->targeti
.in
->ino
);
1137 vino
.snap
= le64_to_cpu(rinfo
->targeti
.in
->snapid
);
1139 if (in
== NULL
|| ceph_ino(in
) != vino
.ino
||
1140 ceph_snap(in
) != vino
.snap
) {
1141 in
= ceph_get_inode(sb
, vino
);
1147 req
->r_target_inode
= in
;
1149 err
= fill_inode(in
,
1150 &rinfo
->targeti
, NULL
,
1151 session
, req
->r_request_started
,
1152 (le32_to_cpu(rinfo
->head
->result
) == 0) ?
1154 &req
->r_caps_reservation
);
1156 pr_err("fill_inode badness %p %llx.%llx\n",
1157 in
, ceph_vinop(in
));
1163 dout("fill_trace done err=%d\n", err
);
1168 * Prepopulate our cache with readdir results, leases, etc.
1170 int ceph_readdir_prepopulate(struct ceph_mds_request
*req
,
1171 struct ceph_mds_session
*session
)
1173 struct dentry
*parent
= req
->r_dentry
;
1174 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1179 struct inode
*snapdir
= NULL
;
1180 struct ceph_mds_request_head
*rhead
= req
->r_request
->front
.iov_base
;
1181 u64 frag
= le32_to_cpu(rhead
->args
.readdir
.frag
);
1182 struct ceph_dentry_info
*di
;
1184 if (le32_to_cpu(rinfo
->head
->op
) == CEPH_MDS_OP_LSSNAP
) {
1185 snapdir
= ceph_get_snapdir(parent
->d_inode
);
1186 parent
= d_find_alias(snapdir
);
1187 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1188 rinfo
->dir_nr
, parent
);
1190 dout("readdir_prepopulate %d items under dn %p\n",
1191 rinfo
->dir_nr
, parent
);
1193 ceph_fill_dirfrag(parent
->d_inode
, rinfo
->dir_dir
);
1196 for (i
= 0; i
< rinfo
->dir_nr
; i
++) {
1197 struct ceph_vino vino
;
1199 dname
.name
= rinfo
->dir_dname
[i
];
1200 dname
.len
= rinfo
->dir_dname_len
[i
];
1201 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
1203 vino
.ino
= le64_to_cpu(rinfo
->dir_in
[i
].in
->ino
);
1204 vino
.snap
= le64_to_cpu(rinfo
->dir_in
[i
].in
->snapid
);
1207 dn
= d_lookup(parent
, &dname
);
1208 dout("d_lookup on parent=%p name=%.*s got %p\n",
1209 parent
, dname
.len
, dname
.name
, dn
);
1212 dn
= d_alloc(parent
, &dname
);
1213 dout("d_alloc %p '%.*s' = %p\n", parent
,
1214 dname
.len
, dname
.name
, dn
);
1216 dout("d_alloc badness\n");
1220 err
= ceph_init_dentry(dn
);
1225 } else if (dn
->d_inode
&&
1226 (ceph_ino(dn
->d_inode
) != vino
.ino
||
1227 ceph_snap(dn
->d_inode
) != vino
.snap
)) {
1228 dout(" dn %p points to wrong inode %p\n",
1234 /* reorder parent's d_subdirs */
1235 spin_lock(&dcache_lock
);
1236 spin_lock(&dn
->d_lock
);
1237 list_move(&dn
->d_u
.d_child
, &parent
->d_subdirs
);
1238 spin_unlock(&dn
->d_lock
);
1239 spin_unlock(&dcache_lock
);
1243 di
->offset
= ceph_make_fpos(frag
, i
+ req
->r_readdir_offset
);
1249 in
= ceph_get_inode(parent
->d_sb
, vino
);
1251 dout("new_inode badness\n");
1257 dn
= splice_dentry(dn
, in
, NULL
, false);
1262 if (fill_inode(in
, &rinfo
->dir_in
[i
], NULL
, session
,
1263 req
->r_request_started
, -1,
1264 &req
->r_caps_reservation
) < 0) {
1265 pr_err("fill_inode badness on %p\n", in
);
1269 update_dentry_lease(dn
, rinfo
->dir_dlease
[i
],
1271 req
->r_request_started
);
1276 req
->r_did_prepopulate
= true;
1283 dout("readdir_prepopulate done\n");
1287 int ceph_inode_set_size(struct inode
*inode
, loff_t size
)
1289 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1292 spin_lock(&inode
->i_lock
);
1293 dout("set_size %p %llu -> %llu\n", inode
, inode
->i_size
, size
);
1294 inode
->i_size
= size
;
1295 inode
->i_blocks
= (size
+ (1 << 9) - 1) >> 9;
1297 /* tell the MDS if we are approaching max_size */
1298 if ((size
<< 1) >= ci
->i_max_size
&&
1299 (ci
->i_reported_size
<< 1) < ci
->i_max_size
)
1302 spin_unlock(&inode
->i_lock
);
1307 * Write back inode data in a worker thread. (This can't be done
1308 * in the message handler context.)
1310 void ceph_queue_writeback(struct inode
*inode
)
1312 if (queue_work(ceph_inode_to_client(inode
)->wb_wq
,
1313 &ceph_inode(inode
)->i_wb_work
)) {
1314 dout("ceph_queue_writeback %p\n", inode
);
1317 dout("ceph_queue_writeback %p failed\n", inode
);
1321 static void ceph_writeback_work(struct work_struct
*work
)
1323 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1325 struct inode
*inode
= &ci
->vfs_inode
;
1327 dout("writeback %p\n", inode
);
1328 filemap_fdatawrite(&inode
->i_data
);
1333 * queue an async invalidation
1335 void ceph_queue_invalidate(struct inode
*inode
)
1337 if (queue_work(ceph_inode_to_client(inode
)->pg_inv_wq
,
1338 &ceph_inode(inode
)->i_pg_inv_work
)) {
1339 dout("ceph_queue_invalidate %p\n", inode
);
1342 dout("ceph_queue_invalidate %p failed\n", inode
);
1347 * invalidate any pages that are not dirty or under writeback. this
1348 * includes pages that are clean and mapped.
1350 static void ceph_invalidate_nondirty_pages(struct address_space
*mapping
)
1352 struct pagevec pvec
;
1356 pagevec_init(&pvec
, 0);
1357 while (pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
1358 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
1359 struct page
*page
= pvec
.pages
[i
];
1362 (PageDirty(page
) || PageWriteback(page
));
1365 skip_page
= !trylock_page(page
);
1368 * We really shouldn't be looking at the ->index of an
1369 * unlocked page. But we're not allowed to lock these
1370 * pages. So we rely upon nobody altering the ->index
1371 * of this (pinned-by-us) page.
1373 index
= page
->index
;
1381 generic_error_remove_page(mapping
, page
);
1384 pagevec_release(&pvec
);
1390 * Invalidate inode pages in a worker thread. (This can't be done
1391 * in the message handler context.)
1393 static void ceph_invalidate_work(struct work_struct
*work
)
1395 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1397 struct inode
*inode
= &ci
->vfs_inode
;
1401 spin_lock(&inode
->i_lock
);
1402 dout("invalidate_pages %p gen %d revoking %d\n", inode
,
1403 ci
->i_rdcache_gen
, ci
->i_rdcache_revoking
);
1404 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
1406 spin_unlock(&inode
->i_lock
);
1409 orig_gen
= ci
->i_rdcache_gen
;
1410 spin_unlock(&inode
->i_lock
);
1412 ceph_invalidate_nondirty_pages(inode
->i_mapping
);
1414 spin_lock(&inode
->i_lock
);
1415 if (orig_gen
== ci
->i_rdcache_gen
&&
1416 orig_gen
== ci
->i_rdcache_revoking
) {
1417 dout("invalidate_pages %p gen %d successful\n", inode
,
1419 ci
->i_rdcache_revoking
--;
1422 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1423 inode
, orig_gen
, ci
->i_rdcache_gen
,
1424 ci
->i_rdcache_revoking
);
1426 spin_unlock(&inode
->i_lock
);
1429 ceph_check_caps(ci
, 0, NULL
);
1436 * called by trunc_wq; take i_mutex ourselves
1438 * We also truncate in a separate thread as well.
1440 static void ceph_vmtruncate_work(struct work_struct
*work
)
1442 struct ceph_inode_info
*ci
= container_of(work
, struct ceph_inode_info
,
1444 struct inode
*inode
= &ci
->vfs_inode
;
1446 dout("vmtruncate_work %p\n", inode
);
1447 mutex_lock(&inode
->i_mutex
);
1448 __ceph_do_pending_vmtruncate(inode
);
1449 mutex_unlock(&inode
->i_mutex
);
1454 * Queue an async vmtruncate. If we fail to queue work, we will handle
1455 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1457 void ceph_queue_vmtruncate(struct inode
*inode
)
1459 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1461 if (queue_work(ceph_sb_to_client(inode
->i_sb
)->trunc_wq
,
1462 &ci
->i_vmtruncate_work
)) {
1463 dout("ceph_queue_vmtruncate %p\n", inode
);
1466 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1467 inode
, ci
->i_truncate_pending
);
1472 * called with i_mutex held.
1474 * Make sure any pending truncation is applied before doing anything
1475 * that may depend on it.
1477 void __ceph_do_pending_vmtruncate(struct inode
*inode
)
1479 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1481 int wrbuffer_refs
, wake
= 0;
1484 spin_lock(&inode
->i_lock
);
1485 if (ci
->i_truncate_pending
== 0) {
1486 dout("__do_pending_vmtruncate %p none pending\n", inode
);
1487 spin_unlock(&inode
->i_lock
);
1492 * make sure any dirty snapped pages are flushed before we
1493 * possibly truncate them.. so write AND block!
1495 if (ci
->i_wrbuffer_ref_head
< ci
->i_wrbuffer_ref
) {
1496 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1498 spin_unlock(&inode
->i_lock
);
1499 filemap_write_and_wait_range(&inode
->i_data
, 0,
1500 inode
->i_sb
->s_maxbytes
);
1504 to
= ci
->i_truncate_size
;
1505 wrbuffer_refs
= ci
->i_wrbuffer_ref
;
1506 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode
,
1507 ci
->i_truncate_pending
, to
);
1508 spin_unlock(&inode
->i_lock
);
1510 truncate_inode_pages(inode
->i_mapping
, to
);
1512 spin_lock(&inode
->i_lock
);
1513 ci
->i_truncate_pending
--;
1514 if (ci
->i_truncate_pending
== 0)
1516 spin_unlock(&inode
->i_lock
);
1518 if (wrbuffer_refs
== 0)
1519 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
1521 wake_up_all(&ci
->i_cap_wq
);
1528 static void *ceph_sym_follow_link(struct dentry
*dentry
, struct nameidata
*nd
)
1530 struct ceph_inode_info
*ci
= ceph_inode(dentry
->d_inode
);
1531 nd_set_link(nd
, ci
->i_symlink
);
1535 static const struct inode_operations ceph_symlink_iops
= {
1536 .readlink
= generic_readlink
,
1537 .follow_link
= ceph_sym_follow_link
,
1543 int ceph_setattr(struct dentry
*dentry
, struct iattr
*attr
)
1545 struct inode
*inode
= dentry
->d_inode
;
1546 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1547 struct inode
*parent_inode
= dentry
->d_parent
->d_inode
;
1548 const unsigned int ia_valid
= attr
->ia_valid
;
1549 struct ceph_mds_request
*req
;
1550 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(dentry
->d_sb
)->mdsc
;
1552 int release
= 0, dirtied
= 0;
1556 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1559 __ceph_do_pending_vmtruncate(inode
);
1561 err
= inode_change_ok(inode
, attr
);
1565 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_SETATTR
,
1568 return PTR_ERR(req
);
1570 spin_lock(&inode
->i_lock
);
1571 issued
= __ceph_caps_issued(ci
, NULL
);
1572 dout("setattr %p issued %s\n", inode
, ceph_cap_string(issued
));
1574 if (ia_valid
& ATTR_UID
) {
1575 dout("setattr %p uid %d -> %d\n", inode
,
1576 inode
->i_uid
, attr
->ia_uid
);
1577 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1578 inode
->i_uid
= attr
->ia_uid
;
1579 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1580 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1581 attr
->ia_uid
!= inode
->i_uid
) {
1582 req
->r_args
.setattr
.uid
= cpu_to_le32(attr
->ia_uid
);
1583 mask
|= CEPH_SETATTR_UID
;
1584 release
|= CEPH_CAP_AUTH_SHARED
;
1587 if (ia_valid
& ATTR_GID
) {
1588 dout("setattr %p gid %d -> %d\n", inode
,
1589 inode
->i_gid
, attr
->ia_gid
);
1590 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1591 inode
->i_gid
= attr
->ia_gid
;
1592 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1593 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1594 attr
->ia_gid
!= inode
->i_gid
) {
1595 req
->r_args
.setattr
.gid
= cpu_to_le32(attr
->ia_gid
);
1596 mask
|= CEPH_SETATTR_GID
;
1597 release
|= CEPH_CAP_AUTH_SHARED
;
1600 if (ia_valid
& ATTR_MODE
) {
1601 dout("setattr %p mode 0%o -> 0%o\n", inode
, inode
->i_mode
,
1603 if (issued
& CEPH_CAP_AUTH_EXCL
) {
1604 inode
->i_mode
= attr
->ia_mode
;
1605 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1606 } else if ((issued
& CEPH_CAP_AUTH_SHARED
) == 0 ||
1607 attr
->ia_mode
!= inode
->i_mode
) {
1608 req
->r_args
.setattr
.mode
= cpu_to_le32(attr
->ia_mode
);
1609 mask
|= CEPH_SETATTR_MODE
;
1610 release
|= CEPH_CAP_AUTH_SHARED
;
1614 if (ia_valid
& ATTR_ATIME
) {
1615 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode
,
1616 inode
->i_atime
.tv_sec
, inode
->i_atime
.tv_nsec
,
1617 attr
->ia_atime
.tv_sec
, attr
->ia_atime
.tv_nsec
);
1618 if (issued
& CEPH_CAP_FILE_EXCL
) {
1619 ci
->i_time_warp_seq
++;
1620 inode
->i_atime
= attr
->ia_atime
;
1621 dirtied
|= CEPH_CAP_FILE_EXCL
;
1622 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1623 timespec_compare(&inode
->i_atime
,
1624 &attr
->ia_atime
) < 0) {
1625 inode
->i_atime
= attr
->ia_atime
;
1626 dirtied
|= CEPH_CAP_FILE_WR
;
1627 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1628 !timespec_equal(&inode
->i_atime
, &attr
->ia_atime
)) {
1629 ceph_encode_timespec(&req
->r_args
.setattr
.atime
,
1631 mask
|= CEPH_SETATTR_ATIME
;
1632 release
|= CEPH_CAP_FILE_CACHE
| CEPH_CAP_FILE_RD
|
1636 if (ia_valid
& ATTR_MTIME
) {
1637 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode
,
1638 inode
->i_mtime
.tv_sec
, inode
->i_mtime
.tv_nsec
,
1639 attr
->ia_mtime
.tv_sec
, attr
->ia_mtime
.tv_nsec
);
1640 if (issued
& CEPH_CAP_FILE_EXCL
) {
1641 ci
->i_time_warp_seq
++;
1642 inode
->i_mtime
= attr
->ia_mtime
;
1643 dirtied
|= CEPH_CAP_FILE_EXCL
;
1644 } else if ((issued
& CEPH_CAP_FILE_WR
) &&
1645 timespec_compare(&inode
->i_mtime
,
1646 &attr
->ia_mtime
) < 0) {
1647 inode
->i_mtime
= attr
->ia_mtime
;
1648 dirtied
|= CEPH_CAP_FILE_WR
;
1649 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1650 !timespec_equal(&inode
->i_mtime
, &attr
->ia_mtime
)) {
1651 ceph_encode_timespec(&req
->r_args
.setattr
.mtime
,
1653 mask
|= CEPH_SETATTR_MTIME
;
1654 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1658 if (ia_valid
& ATTR_SIZE
) {
1659 dout("setattr %p size %lld -> %lld\n", inode
,
1660 inode
->i_size
, attr
->ia_size
);
1661 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
) {
1665 if ((issued
& CEPH_CAP_FILE_EXCL
) &&
1666 attr
->ia_size
> inode
->i_size
) {
1667 inode
->i_size
= attr
->ia_size
;
1669 (attr
->ia_size
+ (1 << 9) - 1) >> 9;
1670 inode
->i_ctime
= attr
->ia_ctime
;
1671 ci
->i_reported_size
= attr
->ia_size
;
1672 dirtied
|= CEPH_CAP_FILE_EXCL
;
1673 } else if ((issued
& CEPH_CAP_FILE_SHARED
) == 0 ||
1674 attr
->ia_size
!= inode
->i_size
) {
1675 req
->r_args
.setattr
.size
= cpu_to_le64(attr
->ia_size
);
1676 req
->r_args
.setattr
.old_size
=
1677 cpu_to_le64(inode
->i_size
);
1678 mask
|= CEPH_SETATTR_SIZE
;
1679 release
|= CEPH_CAP_FILE_SHARED
| CEPH_CAP_FILE_RD
|
1684 /* these do nothing */
1685 if (ia_valid
& ATTR_CTIME
) {
1686 bool only
= (ia_valid
& (ATTR_SIZE
|ATTR_MTIME
|ATTR_ATIME
|
1687 ATTR_MODE
|ATTR_UID
|ATTR_GID
)) == 0;
1688 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode
,
1689 inode
->i_ctime
.tv_sec
, inode
->i_ctime
.tv_nsec
,
1690 attr
->ia_ctime
.tv_sec
, attr
->ia_ctime
.tv_nsec
,
1691 only
? "ctime only" : "ignored");
1692 inode
->i_ctime
= attr
->ia_ctime
;
1695 * if kernel wants to dirty ctime but nothing else,
1696 * we need to choose a cap to dirty under, or do
1697 * a almost-no-op setattr
1699 if (issued
& CEPH_CAP_AUTH_EXCL
)
1700 dirtied
|= CEPH_CAP_AUTH_EXCL
;
1701 else if (issued
& CEPH_CAP_FILE_EXCL
)
1702 dirtied
|= CEPH_CAP_FILE_EXCL
;
1703 else if (issued
& CEPH_CAP_XATTR_EXCL
)
1704 dirtied
|= CEPH_CAP_XATTR_EXCL
;
1706 mask
|= CEPH_SETATTR_CTIME
;
1709 if (ia_valid
& ATTR_FILE
)
1710 dout("setattr %p ATTR_FILE ... hrm!\n", inode
);
1713 __ceph_mark_dirty_caps(ci
, dirtied
);
1714 inode
->i_ctime
= CURRENT_TIME
;
1718 spin_unlock(&inode
->i_lock
);
1721 req
->r_inode
= igrab(inode
);
1722 req
->r_inode_drop
= release
;
1723 req
->r_args
.setattr
.mask
= cpu_to_le32(mask
);
1724 req
->r_num_caps
= 1;
1725 err
= ceph_mdsc_do_request(mdsc
, parent_inode
, req
);
1727 dout("setattr %p result=%d (%s locally, %d remote)\n", inode
, err
,
1728 ceph_cap_string(dirtied
), mask
);
1730 ceph_mdsc_put_request(req
);
1731 __ceph_do_pending_vmtruncate(inode
);
1734 spin_unlock(&inode
->i_lock
);
1735 ceph_mdsc_put_request(req
);
1740 * Verify that we have a lease on the given mask. If not,
1741 * do a getattr against an mds.
1743 int ceph_do_getattr(struct inode
*inode
, int mask
)
1745 struct ceph_fs_client
*fsc
= ceph_sb_to_client(inode
->i_sb
);
1746 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1747 struct ceph_mds_request
*req
;
1750 if (ceph_snap(inode
) == CEPH_SNAPDIR
) {
1751 dout("do_getattr inode %p SNAPDIR\n", inode
);
1755 dout("do_getattr inode %p mask %s mode 0%o\n", inode
, ceph_cap_string(mask
), inode
->i_mode
);
1756 if (ceph_caps_issued_mask(ceph_inode(inode
), mask
, 1))
1759 req
= ceph_mdsc_create_request(mdsc
, CEPH_MDS_OP_GETATTR
, USE_ANY_MDS
);
1761 return PTR_ERR(req
);
1762 req
->r_inode
= igrab(inode
);
1763 req
->r_num_caps
= 1;
1764 req
->r_args
.getattr
.mask
= cpu_to_le32(mask
);
1765 err
= ceph_mdsc_do_request(mdsc
, NULL
, req
);
1766 ceph_mdsc_put_request(req
);
1767 dout("do_getattr result=%d\n", err
);
1773 * Check inode permissions. We verify we have a valid value for
1774 * the AUTH cap, then call the generic handler.
1776 int ceph_permission(struct inode
*inode
, int mask
)
1778 int err
= ceph_do_getattr(inode
, CEPH_CAP_AUTH_SHARED
);
1781 err
= generic_permission(inode
, mask
, NULL
);
1786 * Get all attributes. Hopefully somedata we'll have a statlite()
1787 * and can limit the fields we require to be accurate.
1789 int ceph_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
1792 struct inode
*inode
= dentry
->d_inode
;
1793 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1796 err
= ceph_do_getattr(inode
, CEPH_STAT_CAP_INODE_ALL
);
1798 generic_fillattr(inode
, stat
);
1799 stat
->ino
= inode
->i_ino
;
1800 if (ceph_snap(inode
) != CEPH_NOSNAP
)
1801 stat
->dev
= ceph_snap(inode
);
1804 if (S_ISDIR(inode
->i_mode
)) {
1805 stat
->size
= ci
->i_rbytes
;
1807 stat
->blksize
= 65536;