1 #include <linux/ceph/ceph_debug.h>
4 #include <linux/kernel.h>
5 #include <linux/sched.h>
6 #include <linux/slab.h>
7 #include <linux/vmalloc.h>
8 #include <linux/wait.h>
9 #include <linux/writeback.h>
12 #include "mds_client.h"
14 #include <linux/ceph/decode.h>
15 #include <linux/ceph/messenger.h>
18 * Capability management
20 * The Ceph metadata servers control client access to inode metadata
21 * and file data by issuing capabilities, granting clients permission
22 * to read and/or write both inode field and file data to OSDs
23 * (storage nodes). Each capability consists of a set of bits
24 * indicating which operations are allowed.
26 * If the client holds a *_SHARED cap, the client has a coherent value
27 * that can be safely read from the cached inode.
29 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30 * client is allowed to change inode attributes (e.g., file size,
31 * mtime), note its dirty state in the ceph_cap, and asynchronously
32 * flush that metadata change to the MDS.
34 * In the event of a conflicting operation (perhaps by another
35 * client), the MDS will revoke the conflicting client capabilities.
37 * In order for a client to cache an inode, it must hold a capability
38 * with at least one MDS server. When inodes are released, release
39 * notifications are batched and periodically sent en masse to the MDS
40 * cluster to release server state.
45 * Generate readable cap strings for debugging output.
47 #define MAX_CAP_STR 20
48 static char cap_str
[MAX_CAP_STR
][40];
49 static DEFINE_SPINLOCK(cap_str_lock
);
50 static int last_cap_str
;
52 static char *gcap_string(char *s
, int c
)
54 if (c
& CEPH_CAP_GSHARED
)
56 if (c
& CEPH_CAP_GEXCL
)
58 if (c
& CEPH_CAP_GCACHE
)
64 if (c
& CEPH_CAP_GBUFFER
)
66 if (c
& CEPH_CAP_GLAZYIO
)
71 const char *ceph_cap_string(int caps
)
77 spin_lock(&cap_str_lock
);
79 if (last_cap_str
== MAX_CAP_STR
)
81 spin_unlock(&cap_str_lock
);
85 if (caps
& CEPH_CAP_PIN
)
88 c
= (caps
>> CEPH_CAP_SAUTH
) & 3;
91 s
= gcap_string(s
, c
);
94 c
= (caps
>> CEPH_CAP_SLINK
) & 3;
97 s
= gcap_string(s
, c
);
100 c
= (caps
>> CEPH_CAP_SXATTR
) & 3;
103 s
= gcap_string(s
, c
);
106 c
= caps
>> CEPH_CAP_SFILE
;
109 s
= gcap_string(s
, c
);
118 void ceph_caps_init(struct ceph_mds_client
*mdsc
)
120 INIT_LIST_HEAD(&mdsc
->caps_list
);
121 spin_lock_init(&mdsc
->caps_list_lock
);
124 void ceph_caps_finalize(struct ceph_mds_client
*mdsc
)
126 struct ceph_cap
*cap
;
128 spin_lock(&mdsc
->caps_list_lock
);
129 while (!list_empty(&mdsc
->caps_list
)) {
130 cap
= list_first_entry(&mdsc
->caps_list
,
131 struct ceph_cap
, caps_item
);
132 list_del(&cap
->caps_item
);
133 kmem_cache_free(ceph_cap_cachep
, cap
);
135 mdsc
->caps_total_count
= 0;
136 mdsc
->caps_avail_count
= 0;
137 mdsc
->caps_use_count
= 0;
138 mdsc
->caps_reserve_count
= 0;
139 mdsc
->caps_min_count
= 0;
140 spin_unlock(&mdsc
->caps_list_lock
);
143 void ceph_adjust_min_caps(struct ceph_mds_client
*mdsc
, int delta
)
145 spin_lock(&mdsc
->caps_list_lock
);
146 mdsc
->caps_min_count
+= delta
;
147 BUG_ON(mdsc
->caps_min_count
< 0);
148 spin_unlock(&mdsc
->caps_list_lock
);
151 void ceph_reserve_caps(struct ceph_mds_client
*mdsc
,
152 struct ceph_cap_reservation
*ctx
, int need
)
155 struct ceph_cap
*cap
;
160 dout("reserve caps ctx=%p need=%d\n", ctx
, need
);
162 /* first reserve any caps that are already allocated */
163 spin_lock(&mdsc
->caps_list_lock
);
164 if (mdsc
->caps_avail_count
>= need
)
167 have
= mdsc
->caps_avail_count
;
168 mdsc
->caps_avail_count
-= have
;
169 mdsc
->caps_reserve_count
+= have
;
170 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
171 mdsc
->caps_reserve_count
+
172 mdsc
->caps_avail_count
);
173 spin_unlock(&mdsc
->caps_list_lock
);
175 for (i
= have
; i
< need
; i
++) {
176 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
179 list_add(&cap
->caps_item
, &newcaps
);
182 /* we didn't manage to reserve as much as we needed */
183 if (have
+ alloc
!= need
)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx
, need
, have
+ alloc
);
187 spin_lock(&mdsc
->caps_list_lock
);
188 mdsc
->caps_total_count
+= alloc
;
189 mdsc
->caps_reserve_count
+= alloc
;
190 list_splice(&newcaps
, &mdsc
->caps_list
);
192 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
193 mdsc
->caps_reserve_count
+
194 mdsc
->caps_avail_count
);
195 spin_unlock(&mdsc
->caps_list_lock
);
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
199 ctx
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
200 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
203 int ceph_unreserve_caps(struct ceph_mds_client
*mdsc
,
204 struct ceph_cap_reservation
*ctx
)
206 dout("unreserve caps ctx=%p count=%d\n", ctx
, ctx
->count
);
208 spin_lock(&mdsc
->caps_list_lock
);
209 BUG_ON(mdsc
->caps_reserve_count
< ctx
->count
);
210 mdsc
->caps_reserve_count
-= ctx
->count
;
211 mdsc
->caps_avail_count
+= ctx
->count
;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
214 mdsc
->caps_total_count
, mdsc
->caps_use_count
,
215 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
216 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
217 mdsc
->caps_reserve_count
+
218 mdsc
->caps_avail_count
);
219 spin_unlock(&mdsc
->caps_list_lock
);
224 struct ceph_cap
*ceph_get_cap(struct ceph_mds_client
*mdsc
,
225 struct ceph_cap_reservation
*ctx
)
227 struct ceph_cap
*cap
= NULL
;
229 /* temporary, until we do something about cap import/export */
231 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
233 spin_lock(&mdsc
->caps_list_lock
);
234 mdsc
->caps_use_count
++;
235 mdsc
->caps_total_count
++;
236 spin_unlock(&mdsc
->caps_list_lock
);
241 spin_lock(&mdsc
->caps_list_lock
);
242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
243 ctx
, ctx
->count
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
244 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
246 BUG_ON(ctx
->count
> mdsc
->caps_reserve_count
);
247 BUG_ON(list_empty(&mdsc
->caps_list
));
250 mdsc
->caps_reserve_count
--;
251 mdsc
->caps_use_count
++;
253 cap
= list_first_entry(&mdsc
->caps_list
, struct ceph_cap
, caps_item
);
254 list_del(&cap
->caps_item
);
256 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
257 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
258 spin_unlock(&mdsc
->caps_list_lock
);
262 void ceph_put_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
)
264 spin_lock(&mdsc
->caps_list_lock
);
265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
266 cap
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
267 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
268 mdsc
->caps_use_count
--;
270 * Keep some preallocated caps around (ceph_min_count), to
271 * avoid lots of free/alloc churn.
273 if (mdsc
->caps_avail_count
>= mdsc
->caps_reserve_count
+
274 mdsc
->caps_min_count
) {
275 mdsc
->caps_total_count
--;
276 kmem_cache_free(ceph_cap_cachep
, cap
);
278 mdsc
->caps_avail_count
++;
279 list_add(&cap
->caps_item
, &mdsc
->caps_list
);
282 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
283 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
284 spin_unlock(&mdsc
->caps_list_lock
);
287 void ceph_reservation_status(struct ceph_fs_client
*fsc
,
288 int *total
, int *avail
, int *used
, int *reserved
,
291 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
294 *total
= mdsc
->caps_total_count
;
296 *avail
= mdsc
->caps_avail_count
;
298 *used
= mdsc
->caps_use_count
;
300 *reserved
= mdsc
->caps_reserve_count
;
302 *min
= mdsc
->caps_min_count
;
306 * Find ceph_cap for given mds, if any.
308 * Called with i_ceph_lock held.
310 static struct ceph_cap
*__get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
312 struct ceph_cap
*cap
;
313 struct rb_node
*n
= ci
->i_caps
.rb_node
;
316 cap
= rb_entry(n
, struct ceph_cap
, ci_node
);
319 else if (mds
> cap
->mds
)
327 struct ceph_cap
*ceph_get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
329 struct ceph_cap
*cap
;
331 spin_lock(&ci
->i_ceph_lock
);
332 cap
= __get_cap_for_mds(ci
, mds
);
333 spin_unlock(&ci
->i_ceph_lock
);
338 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
340 static int __ceph_get_cap_mds(struct ceph_inode_info
*ci
)
342 struct ceph_cap
*cap
;
346 /* prefer mds with WR|BUFFER|EXCL caps */
347 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
348 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
350 if (cap
->issued
& (CEPH_CAP_FILE_WR
|
351 CEPH_CAP_FILE_BUFFER
|
358 int ceph_get_cap_mds(struct inode
*inode
)
360 struct ceph_inode_info
*ci
= ceph_inode(inode
);
362 spin_lock(&ci
->i_ceph_lock
);
363 mds
= __ceph_get_cap_mds(ceph_inode(inode
));
364 spin_unlock(&ci
->i_ceph_lock
);
369 * Called under i_ceph_lock.
371 static void __insert_cap_node(struct ceph_inode_info
*ci
,
372 struct ceph_cap
*new)
374 struct rb_node
**p
= &ci
->i_caps
.rb_node
;
375 struct rb_node
*parent
= NULL
;
376 struct ceph_cap
*cap
= NULL
;
380 cap
= rb_entry(parent
, struct ceph_cap
, ci_node
);
381 if (new->mds
< cap
->mds
)
383 else if (new->mds
> cap
->mds
)
389 rb_link_node(&new->ci_node
, parent
, p
);
390 rb_insert_color(&new->ci_node
, &ci
->i_caps
);
394 * (re)set cap hold timeouts, which control the delayed release
395 * of unused caps back to the MDS. Should be called on cap use.
397 static void __cap_set_timeouts(struct ceph_mds_client
*mdsc
,
398 struct ceph_inode_info
*ci
)
400 struct ceph_mount_options
*ma
= mdsc
->fsc
->mount_options
;
402 ci
->i_hold_caps_min
= round_jiffies(jiffies
+
403 ma
->caps_wanted_delay_min
* HZ
);
404 ci
->i_hold_caps_max
= round_jiffies(jiffies
+
405 ma
->caps_wanted_delay_max
* HZ
);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci
->vfs_inode
,
407 ci
->i_hold_caps_min
- jiffies
, ci
->i_hold_caps_max
- jiffies
);
411 * (Re)queue cap at the end of the delayed cap release list.
413 * If I_FLUSH is set, leave the inode at the front of the list.
415 * Caller holds i_ceph_lock
416 * -> we take mdsc->cap_delay_lock
418 static void __cap_delay_requeue(struct ceph_mds_client
*mdsc
,
419 struct ceph_inode_info
*ci
)
421 __cap_set_timeouts(mdsc
, ci
);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci
->vfs_inode
,
423 ci
->i_ceph_flags
, ci
->i_hold_caps_max
);
424 if (!mdsc
->stopping
) {
425 spin_lock(&mdsc
->cap_delay_lock
);
426 if (!list_empty(&ci
->i_cap_delay_list
)) {
427 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
429 list_del_init(&ci
->i_cap_delay_list
);
431 list_add_tail(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
433 spin_unlock(&mdsc
->cap_delay_lock
);
438 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
439 * indicating we should send a cap message to flush dirty metadata
440 * asap, and move to the front of the delayed cap list.
442 static void __cap_delay_requeue_front(struct ceph_mds_client
*mdsc
,
443 struct ceph_inode_info
*ci
)
445 dout("__cap_delay_requeue_front %p\n", &ci
->vfs_inode
);
446 spin_lock(&mdsc
->cap_delay_lock
);
447 ci
->i_ceph_flags
|= CEPH_I_FLUSH
;
448 if (!list_empty(&ci
->i_cap_delay_list
))
449 list_del_init(&ci
->i_cap_delay_list
);
450 list_add(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
451 spin_unlock(&mdsc
->cap_delay_lock
);
455 * Cancel delayed work on cap.
457 * Caller must hold i_ceph_lock.
459 static void __cap_delay_cancel(struct ceph_mds_client
*mdsc
,
460 struct ceph_inode_info
*ci
)
462 dout("__cap_delay_cancel %p\n", &ci
->vfs_inode
);
463 if (list_empty(&ci
->i_cap_delay_list
))
465 spin_lock(&mdsc
->cap_delay_lock
);
466 list_del_init(&ci
->i_cap_delay_list
);
467 spin_unlock(&mdsc
->cap_delay_lock
);
471 * Common issue checks for add_cap, handle_cap_grant.
473 static void __check_cap_issue(struct ceph_inode_info
*ci
, struct ceph_cap
*cap
,
476 unsigned had
= __ceph_caps_issued(ci
, NULL
);
479 * Each time we receive FILE_CACHE anew, we increment
482 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
483 (had
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0) {
488 * if we are newly issued FILE_SHARED, mark dir not complete; we
489 * don't know what happened to this directory while we didn't
492 if ((issued
& CEPH_CAP_FILE_SHARED
) &&
493 (had
& CEPH_CAP_FILE_SHARED
) == 0) {
495 if (S_ISDIR(ci
->vfs_inode
.i_mode
)) {
496 dout(" marking %p NOT complete\n", &ci
->vfs_inode
);
497 __ceph_dir_clear_complete(ci
);
503 * Add a capability under the given MDS session.
505 * Caller should hold session snap_rwsem (read) and s_mutex.
507 * @fmode is the open file mode, if we are opening a file, otherwise
508 * it is < 0. (This is so we can atomically add the cap and add an
509 * open file reference to it.)
511 void ceph_add_cap(struct inode
*inode
,
512 struct ceph_mds_session
*session
, u64 cap_id
,
513 int fmode
, unsigned issued
, unsigned wanted
,
514 unsigned seq
, unsigned mseq
, u64 realmino
, int flags
,
515 struct ceph_cap
**new_cap
)
517 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
518 struct ceph_inode_info
*ci
= ceph_inode(inode
);
519 struct ceph_cap
*cap
;
520 int mds
= session
->s_mds
;
523 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
524 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
527 * If we are opening the file, include file mode wanted bits
531 wanted
|= ceph_caps_for_mode(fmode
);
533 cap
= __get_cap_for_mds(ci
, mds
);
539 cap
->implemented
= 0;
545 __insert_cap_node(ci
, cap
);
547 /* add to session cap list */
548 cap
->session
= session
;
549 spin_lock(&session
->s_cap_lock
);
550 list_add_tail(&cap
->session_caps
, &session
->s_caps
);
551 session
->s_nr_caps
++;
552 spin_unlock(&session
->s_cap_lock
);
555 * auth mds of the inode changed. we received the cap export
556 * message, but still haven't received the cap import message.
557 * handle_cap_export() updated the new auth MDS' cap.
559 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
560 * a message that was send before the cap import message. So
563 if (ceph_seq_cmp(seq
, cap
->seq
) <= 0) {
564 WARN_ON(cap
!= ci
->i_auth_cap
);
565 WARN_ON(cap
->cap_id
!= cap_id
);
568 issued
|= cap
->issued
;
569 flags
|= CEPH_CAP_FLAG_AUTH
;
573 if (!ci
->i_snap_realm
) {
575 * add this inode to the appropriate snap realm
577 struct ceph_snap_realm
*realm
= ceph_lookup_snap_realm(mdsc
,
580 spin_lock(&realm
->inodes_with_caps_lock
);
581 ci
->i_snap_realm
= realm
;
582 list_add(&ci
->i_snap_realm_item
,
583 &realm
->inodes_with_caps
);
584 spin_unlock(&realm
->inodes_with_caps_lock
);
586 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
592 __check_cap_issue(ci
, cap
, issued
);
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
599 actual_wanted
= __ceph_caps_wanted(ci
);
600 if ((wanted
& ~actual_wanted
) ||
601 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued
), ceph_cap_string(wanted
),
604 ceph_cap_string(actual_wanted
));
605 __cap_delay_requeue(mdsc
, ci
);
608 if (flags
& CEPH_CAP_FLAG_AUTH
) {
609 if (ci
->i_auth_cap
== NULL
||
610 ceph_seq_cmp(ci
->i_auth_cap
->mseq
, mseq
) < 0) {
611 ci
->i_auth_cap
= cap
;
612 cap
->mds_wanted
= wanted
;
615 WARN_ON(ci
->i_auth_cap
== cap
);
618 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
619 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
620 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
621 cap
->cap_id
= cap_id
;
622 cap
->issued
= issued
;
623 cap
->implemented
|= issued
;
624 if (ceph_seq_cmp(mseq
, cap
->mseq
) > 0)
625 cap
->mds_wanted
= wanted
;
627 cap
->mds_wanted
|= wanted
;
629 cap
->issue_seq
= seq
;
631 cap
->cap_gen
= session
->s_cap_gen
;
634 __ceph_get_fmode(ci
, fmode
);
638 * Return true if cap has not timed out and belongs to the current
639 * generation of the MDS session (i.e. has not gone 'stale' due to
640 * us losing touch with the mds).
642 static int __cap_is_valid(struct ceph_cap
*cap
)
647 spin_lock(&cap
->session
->s_gen_ttl_lock
);
648 gen
= cap
->session
->s_cap_gen
;
649 ttl
= cap
->session
->s_cap_ttl
;
650 spin_unlock(&cap
->session
->s_gen_ttl_lock
);
652 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
653 dout("__cap_is_valid %p cap %p issued %s "
654 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
655 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
663 * Return set of valid cap bits issued to us. Note that caps time
664 * out, and may be invalidated in bulk if the client session times out
665 * and session->s_cap_gen is bumped.
667 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
669 int have
= ci
->i_snap_caps
;
670 struct ceph_cap
*cap
;
675 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
676 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
677 if (!__cap_is_valid(cap
))
679 dout("__ceph_caps_issued %p cap %p issued %s\n",
680 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
683 *implemented
|= cap
->implemented
;
686 * exclude caps issued by non-auth MDS, but are been revoking
687 * by the auth MDS. The non-auth MDS should be revoking/exporting
688 * these caps, but the message is delayed.
690 if (ci
->i_auth_cap
) {
691 cap
= ci
->i_auth_cap
;
692 have
&= ~cap
->implemented
| cap
->issued
;
698 * Get cap bits issued by caps other than @ocap
700 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
702 int have
= ci
->i_snap_caps
;
703 struct ceph_cap
*cap
;
706 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
707 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
710 if (!__cap_is_valid(cap
))
718 * Move a cap to the end of the LRU (oldest caps at list head, newest
721 static void __touch_cap(struct ceph_cap
*cap
)
723 struct ceph_mds_session
*s
= cap
->session
;
725 spin_lock(&s
->s_cap_lock
);
726 if (s
->s_cap_iterator
== NULL
) {
727 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
729 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
731 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
732 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
734 spin_unlock(&s
->s_cap_lock
);
738 * Check if we hold the given mask. If so, move the cap(s) to the
739 * front of their respective LRUs. (This is the preferred way for
740 * callers to check for caps they want.)
742 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
744 struct ceph_cap
*cap
;
746 int have
= ci
->i_snap_caps
;
748 if ((have
& mask
) == mask
) {
749 dout("__ceph_caps_issued_mask %p snap issued %s"
750 " (mask %s)\n", &ci
->vfs_inode
,
751 ceph_cap_string(have
),
752 ceph_cap_string(mask
));
756 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
757 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
758 if (!__cap_is_valid(cap
))
760 if ((cap
->issued
& mask
) == mask
) {
761 dout("__ceph_caps_issued_mask %p cap %p issued %s"
762 " (mask %s)\n", &ci
->vfs_inode
, cap
,
763 ceph_cap_string(cap
->issued
),
764 ceph_cap_string(mask
));
770 /* does a combination of caps satisfy mask? */
772 if ((have
& mask
) == mask
) {
773 dout("__ceph_caps_issued_mask %p combo issued %s"
774 " (mask %s)\n", &ci
->vfs_inode
,
775 ceph_cap_string(cap
->issued
),
776 ceph_cap_string(mask
));
780 /* touch this + preceding caps */
782 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
784 cap
= rb_entry(q
, struct ceph_cap
,
786 if (!__cap_is_valid(cap
))
799 * Return true if mask caps are currently being revoked by an MDS.
801 int __ceph_caps_revoking_other(struct ceph_inode_info
*ci
,
802 struct ceph_cap
*ocap
, int mask
)
804 struct ceph_cap
*cap
;
807 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
808 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
810 (cap
->implemented
& ~cap
->issued
& mask
))
816 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
818 struct inode
*inode
= &ci
->vfs_inode
;
821 spin_lock(&ci
->i_ceph_lock
);
822 ret
= __ceph_caps_revoking_other(ci
, NULL
, mask
);
823 spin_unlock(&ci
->i_ceph_lock
);
824 dout("ceph_caps_revoking %p %s = %d\n", inode
,
825 ceph_cap_string(mask
), ret
);
829 int __ceph_caps_used(struct ceph_inode_info
*ci
)
833 used
|= CEPH_CAP_PIN
;
835 used
|= CEPH_CAP_FILE_RD
;
836 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
837 used
|= CEPH_CAP_FILE_CACHE
;
839 used
|= CEPH_CAP_FILE_WR
;
840 if (ci
->i_wb_ref
|| ci
->i_wrbuffer_ref
)
841 used
|= CEPH_CAP_FILE_BUFFER
;
846 * wanted, by virtue of open file modes
848 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
852 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
853 if (ci
->i_nr_by_mode
[mode
])
854 want
|= ceph_caps_for_mode(mode
);
859 * Return caps we have registered with the MDS(s) as 'wanted'.
861 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
863 struct ceph_cap
*cap
;
867 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
868 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
869 if (!__cap_is_valid(cap
))
871 if (cap
== ci
->i_auth_cap
)
872 mds_wanted
|= cap
->mds_wanted
;
874 mds_wanted
|= (cap
->mds_wanted
& ~CEPH_CAP_ANY_FILE_WR
);
880 * called under i_ceph_lock
882 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
884 return !RB_EMPTY_ROOT(&ci
->i_caps
);
887 int ceph_is_any_caps(struct inode
*inode
)
889 struct ceph_inode_info
*ci
= ceph_inode(inode
);
892 spin_lock(&ci
->i_ceph_lock
);
893 ret
= __ceph_is_any_caps(ci
);
894 spin_unlock(&ci
->i_ceph_lock
);
899 static void drop_inode_snap_realm(struct ceph_inode_info
*ci
)
901 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
902 spin_lock(&realm
->inodes_with_caps_lock
);
903 list_del_init(&ci
->i_snap_realm_item
);
904 ci
->i_snap_realm_counter
++;
905 ci
->i_snap_realm
= NULL
;
906 spin_unlock(&realm
->inodes_with_caps_lock
);
907 ceph_put_snap_realm(ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
,
912 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
914 * caller should hold i_ceph_lock.
915 * caller will not hold session s_mutex if called from destroy_inode.
917 void __ceph_remove_cap(struct ceph_cap
*cap
, bool queue_release
)
919 struct ceph_mds_session
*session
= cap
->session
;
920 struct ceph_inode_info
*ci
= cap
->ci
;
921 struct ceph_mds_client
*mdsc
=
922 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
925 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
927 /* remove from session list */
928 spin_lock(&session
->s_cap_lock
);
929 if (session
->s_cap_iterator
== cap
) {
930 /* not yet, we are iterating over this very cap */
931 dout("__ceph_remove_cap delaying %p removal from session %p\n",
934 list_del_init(&cap
->session_caps
);
935 session
->s_nr_caps
--;
939 /* protect backpointer with s_cap_lock: see iterate_session_caps */
943 * s_cap_reconnect is protected by s_cap_lock. no one changes
944 * s_cap_gen while session is in the reconnect state.
947 (!session
->s_cap_reconnect
|| cap
->cap_gen
== session
->s_cap_gen
)) {
948 cap
->queue_release
= 1;
950 list_add_tail(&cap
->session_caps
,
951 &session
->s_cap_releases
);
952 session
->s_num_cap_releases
++;
956 cap
->queue_release
= 0;
958 cap
->cap_ino
= ci
->i_vino
.ino
;
960 spin_unlock(&session
->s_cap_lock
);
962 /* remove from inode list */
963 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
964 if (ci
->i_auth_cap
== cap
)
965 ci
->i_auth_cap
= NULL
;
968 ceph_put_cap(mdsc
, cap
);
970 /* when reconnect denied, we remove session caps forcibly,
971 * i_wr_ref can be non-zero. If there are ongoing write,
974 if (!__ceph_is_any_caps(ci
) && ci
->i_wr_ref
== 0 && ci
->i_snap_realm
)
975 drop_inode_snap_realm(ci
);
977 if (!__ceph_is_any_real_caps(ci
))
978 __cap_delay_cancel(mdsc
, ci
);
982 * Build and send a cap message to the given MDS.
984 * Caller should be holding s_mutex.
986 static int send_cap_msg(struct ceph_mds_session
*session
,
987 u64 ino
, u64 cid
, int op
,
988 int caps
, int wanted
, int dirty
,
989 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
990 u64 size
, u64 max_size
,
991 struct timespec
*mtime
, struct timespec
*atime
,
993 kuid_t uid
, kgid_t gid
, umode_t mode
,
995 struct ceph_buffer
*xattrs_buf
,
996 u64 follows
, bool inline_data
)
998 struct ceph_mds_caps
*fc
;
999 struct ceph_msg
*msg
;
1003 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
1004 " seq %u/%u mseq %u follows %lld size %llu/%llu"
1005 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
1006 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
1007 ceph_cap_string(dirty
),
1008 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
1009 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
1011 /* flock buffer size + inline version + inline data size */
1012 extra_len
= 4 + 8 + 4;
1013 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
) + extra_len
,
1018 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
1020 fc
= msg
->front
.iov_base
;
1021 memset(fc
, 0, sizeof(*fc
));
1023 fc
->cap_id
= cpu_to_le64(cid
);
1024 fc
->op
= cpu_to_le32(op
);
1025 fc
->seq
= cpu_to_le32(seq
);
1026 fc
->issue_seq
= cpu_to_le32(issue_seq
);
1027 fc
->migrate_seq
= cpu_to_le32(mseq
);
1028 fc
->caps
= cpu_to_le32(caps
);
1029 fc
->wanted
= cpu_to_le32(wanted
);
1030 fc
->dirty
= cpu_to_le32(dirty
);
1031 fc
->ino
= cpu_to_le64(ino
);
1032 fc
->snap_follows
= cpu_to_le64(follows
);
1034 fc
->size
= cpu_to_le64(size
);
1035 fc
->max_size
= cpu_to_le64(max_size
);
1037 ceph_encode_timespec(&fc
->mtime
, mtime
);
1039 ceph_encode_timespec(&fc
->atime
, atime
);
1040 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
1042 fc
->uid
= cpu_to_le32(from_kuid(&init_user_ns
, uid
));
1043 fc
->gid
= cpu_to_le32(from_kgid(&init_user_ns
, gid
));
1044 fc
->mode
= cpu_to_le32(mode
);
1047 /* flock buffer size */
1048 ceph_encode_32(&p
, 0);
1049 /* inline version */
1050 ceph_encode_64(&p
, inline_data
? 0 : CEPH_INLINE_NONE
);
1051 /* inline data size */
1052 ceph_encode_32(&p
, 0);
1054 fc
->xattr_version
= cpu_to_le64(xattr_version
);
1056 msg
->middle
= ceph_buffer_get(xattrs_buf
);
1057 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1058 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1061 ceph_con_send(&session
->s_con
, msg
);
1066 * Queue cap releases when an inode is dropped from our cache. Since
1067 * inode is about to be destroyed, there is no need for i_ceph_lock.
1069 void ceph_queue_caps_release(struct inode
*inode
)
1071 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1074 p
= rb_first(&ci
->i_caps
);
1076 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1078 __ceph_remove_cap(cap
, true);
1083 * Send a cap msg on the given inode. Update our caps state, then
1084 * drop i_ceph_lock and send the message.
1086 * Make note of max_size reported/requested from mds, revoked caps
1087 * that have now been implemented.
1089 * Make half-hearted attempt ot to invalidate page cache if we are
1090 * dropping RDCACHE. Note that this will leave behind locked pages
1091 * that we'll then need to deal with elsewhere.
1093 * Return non-zero if delayed release, or we experienced an error
1094 * such that the caller should requeue + retry later.
1096 * called with i_ceph_lock, then drops it.
1097 * caller should hold snap_rwsem (read), s_mutex.
1099 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1100 int op
, int used
, int want
, int retain
, int flushing
)
1101 __releases(cap
->ci
->i_ceph_lock
)
1103 struct ceph_inode_info
*ci
= cap
->ci
;
1104 struct inode
*inode
= &ci
->vfs_inode
;
1105 u64 cap_id
= cap
->cap_id
;
1106 int held
, revoking
, dropping
, keep
;
1107 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1109 struct timespec mtime
, atime
;
1114 struct ceph_mds_session
*session
;
1115 u64 xattr_version
= 0;
1116 struct ceph_buffer
*xattr_blob
= NULL
;
1123 held
= cap
->issued
| cap
->implemented
;
1124 revoking
= cap
->implemented
& ~cap
->issued
;
1125 retain
&= ~revoking
;
1126 dropping
= cap
->issued
& ~retain
;
1128 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1129 inode
, cap
, cap
->session
,
1130 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1131 ceph_cap_string(revoking
));
1132 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1134 session
= cap
->session
;
1136 /* don't release wanted unless we've waited a bit. */
1137 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1138 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1139 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1140 ceph_cap_string(cap
->issued
),
1141 ceph_cap_string(cap
->issued
& retain
),
1142 ceph_cap_string(cap
->mds_wanted
),
1143 ceph_cap_string(want
));
1144 want
|= cap
->mds_wanted
;
1145 retain
|= cap
->issued
;
1148 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1150 cap
->issued
&= retain
; /* drop bits we don't want */
1151 if (cap
->implemented
& ~cap
->issued
) {
1153 * Wake up any waiters on wanted -> needed transition.
1154 * This is due to the weird transition from buffered
1155 * to sync IO... we need to flush dirty pages _before_
1156 * allowing sync writes to avoid reordering.
1160 cap
->implemented
&= cap
->issued
| used
;
1161 cap
->mds_wanted
= want
;
1165 * assign a tid for flush operations so we can avoid
1166 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1167 * clean type races. track latest tid for every bit
1168 * so we can handle flush AxFw, flush Fw, and have the
1169 * first ack clean Ax.
1171 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1172 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1173 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1174 if (flushing
& (1 << i
))
1175 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1177 follows
= ci
->i_head_snapc
->seq
;
1182 keep
= cap
->implemented
;
1184 issue_seq
= cap
->issue_seq
;
1186 size
= inode
->i_size
;
1187 ci
->i_reported_size
= size
;
1188 max_size
= ci
->i_wanted_max_size
;
1189 ci
->i_requested_max_size
= max_size
;
1190 mtime
= inode
->i_mtime
;
1191 atime
= inode
->i_atime
;
1192 time_warp_seq
= ci
->i_time_warp_seq
;
1195 mode
= inode
->i_mode
;
1197 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1198 __ceph_build_xattrs_blob(ci
);
1199 xattr_blob
= ci
->i_xattrs
.blob
;
1200 xattr_version
= ci
->i_xattrs
.version
;
1203 inline_data
= ci
->i_inline_version
!= CEPH_INLINE_NONE
;
1205 spin_unlock(&ci
->i_ceph_lock
);
1207 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1208 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1209 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1210 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1211 follows
, inline_data
);
1213 dout("error sending cap msg, must requeue %p\n", inode
);
1218 wake_up_all(&ci
->i_cap_wq
);
1224 * When a snapshot is taken, clients accumulate dirty metadata on
1225 * inodes with capabilities in ceph_cap_snaps to describe the file
1226 * state at the time the snapshot was taken. This must be flushed
1227 * asynchronously back to the MDS once sync writes complete and dirty
1228 * data is written out.
1230 * Unless @kick is true, skip cap_snaps that were already sent to
1231 * the MDS (i.e., during this session).
1233 * Called under i_ceph_lock. Takes s_mutex as needed.
1235 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1236 struct ceph_mds_session
**psession
,
1238 __releases(ci
->i_ceph_lock
)
1239 __acquires(ci
->i_ceph_lock
)
1241 struct inode
*inode
= &ci
->vfs_inode
;
1243 struct ceph_cap_snap
*capsnap
;
1245 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1246 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1248 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1249 i_cap_snaps list, and skip these entries next time
1250 around to avoid an infinite loop */
1253 session
= *psession
;
1255 dout("__flush_snaps %p\n", inode
);
1257 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1258 /* avoid an infiniute loop after retry */
1259 if (capsnap
->follows
< next_follows
)
1262 * we need to wait for sync writes to complete and for dirty
1263 * pages to be written out.
1265 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1268 /* should be removed by ceph_try_drop_cap_snap() */
1269 BUG_ON(!capsnap
->need_flush
);
1271 /* pick mds, take s_mutex */
1272 if (ci
->i_auth_cap
== NULL
) {
1273 dout("no auth cap (migrating?), doing nothing\n");
1277 /* only flush each capsnap once */
1278 if (!kick
&& !list_empty(&capsnap
->flushing_item
)) {
1279 dout("already flushed %p, skipping\n", capsnap
);
1283 mds
= ci
->i_auth_cap
->session
->s_mds
;
1284 mseq
= ci
->i_auth_cap
->mseq
;
1286 if (session
&& session
->s_mds
!= mds
) {
1287 dout("oops, wrong session %p mutex\n", session
);
1291 mutex_unlock(&session
->s_mutex
);
1292 ceph_put_mds_session(session
);
1296 spin_unlock(&ci
->i_ceph_lock
);
1297 mutex_lock(&mdsc
->mutex
);
1298 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1299 mutex_unlock(&mdsc
->mutex
);
1301 dout("inverting session/ino locks on %p\n",
1303 mutex_lock(&session
->s_mutex
);
1306 * if session == NULL, we raced against a cap
1307 * deletion or migration. retry, and we'll
1308 * get a better @mds value next time.
1310 spin_lock(&ci
->i_ceph_lock
);
1314 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1315 atomic_inc(&capsnap
->nref
);
1316 if (list_empty(&capsnap
->flushing_item
))
1317 list_add_tail(&capsnap
->flushing_item
,
1318 &session
->s_cap_snaps_flushing
);
1319 spin_unlock(&ci
->i_ceph_lock
);
1321 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1322 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1323 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1324 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1325 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1327 &capsnap
->mtime
, &capsnap
->atime
,
1328 capsnap
->time_warp_seq
,
1329 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1330 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1331 capsnap
->follows
, capsnap
->inline_data
);
1333 next_follows
= capsnap
->follows
+ 1;
1334 ceph_put_cap_snap(capsnap
);
1336 spin_lock(&ci
->i_ceph_lock
);
1340 /* we flushed them all; remove this inode from the queue */
1341 spin_lock(&mdsc
->snap_flush_lock
);
1342 list_del_init(&ci
->i_snap_flush_item
);
1343 spin_unlock(&mdsc
->snap_flush_lock
);
1347 *psession
= session
;
1349 mutex_unlock(&session
->s_mutex
);
1350 ceph_put_mds_session(session
);
1354 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1356 spin_lock(&ci
->i_ceph_lock
);
1357 __ceph_flush_snaps(ci
, NULL
, 0);
1358 spin_unlock(&ci
->i_ceph_lock
);
1362 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1363 * Caller is then responsible for calling __mark_inode_dirty with the
1364 * returned flags value.
1366 int __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1368 struct ceph_mds_client
*mdsc
=
1369 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1370 struct inode
*inode
= &ci
->vfs_inode
;
1371 int was
= ci
->i_dirty_caps
;
1374 if (!ci
->i_auth_cap
) {
1375 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1376 "but no auth cap (session was closed?)\n",
1377 inode
, ceph_ino(inode
), ceph_cap_string(mask
));
1381 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1382 ceph_cap_string(mask
), ceph_cap_string(was
),
1383 ceph_cap_string(was
| mask
));
1384 ci
->i_dirty_caps
|= mask
;
1386 if (!ci
->i_head_snapc
) {
1387 WARN_ON_ONCE(!rwsem_is_locked(&mdsc
->snap_rwsem
));
1388 ci
->i_head_snapc
= ceph_get_snap_context(
1389 ci
->i_snap_realm
->cached_context
);
1391 dout(" inode %p now dirty snapc %p auth cap %p\n",
1392 &ci
->vfs_inode
, ci
->i_head_snapc
, ci
->i_auth_cap
);
1393 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1394 spin_lock(&mdsc
->cap_dirty_lock
);
1395 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1396 spin_unlock(&mdsc
->cap_dirty_lock
);
1397 if (ci
->i_flushing_caps
== 0) {
1399 dirty
|= I_DIRTY_SYNC
;
1402 BUG_ON(list_empty(&ci
->i_dirty_item
));
1403 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1404 (mask
& CEPH_CAP_FILE_BUFFER
))
1405 dirty
|= I_DIRTY_DATASYNC
;
1406 __cap_delay_requeue(mdsc
, ci
);
1411 * Add dirty inode to the flushing list. Assigned a seq number so we
1412 * can wait for caps to flush without starving.
1414 * Called under i_ceph_lock.
1416 static int __mark_caps_flushing(struct inode
*inode
,
1417 struct ceph_mds_session
*session
)
1419 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1420 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1423 BUG_ON(ci
->i_dirty_caps
== 0);
1424 BUG_ON(list_empty(&ci
->i_dirty_item
));
1426 flushing
= ci
->i_dirty_caps
;
1427 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1428 ceph_cap_string(flushing
),
1429 ceph_cap_string(ci
->i_flushing_caps
),
1430 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1431 ci
->i_flushing_caps
|= flushing
;
1432 ci
->i_dirty_caps
= 0;
1433 dout(" inode %p now !dirty\n", inode
);
1435 spin_lock(&mdsc
->cap_dirty_lock
);
1436 list_del_init(&ci
->i_dirty_item
);
1438 if (list_empty(&ci
->i_flushing_item
)) {
1439 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1440 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1441 mdsc
->num_cap_flushing
++;
1442 dout(" inode %p now flushing seq %lld\n", inode
,
1443 ci
->i_cap_flush_seq
);
1445 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1446 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1447 ci
->i_cap_flush_seq
);
1449 spin_unlock(&mdsc
->cap_dirty_lock
);
1455 * try to invalidate mapping pages without blocking.
1457 static int try_nonblocking_invalidate(struct inode
*inode
)
1459 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1460 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1462 spin_unlock(&ci
->i_ceph_lock
);
1463 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1464 spin_lock(&ci
->i_ceph_lock
);
1466 if (inode
->i_data
.nrpages
== 0 &&
1467 invalidating_gen
== ci
->i_rdcache_gen
) {
1469 dout("try_nonblocking_invalidate %p success\n", inode
);
1470 /* save any racing async invalidate some trouble */
1471 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
- 1;
1474 dout("try_nonblocking_invalidate %p failed\n", inode
);
1479 * Swiss army knife function to examine currently used and wanted
1480 * versus held caps. Release, flush, ack revoked caps to mds as
1483 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1484 * cap release further.
1485 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1486 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1489 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1490 struct ceph_mds_session
*session
)
1492 struct ceph_fs_client
*fsc
= ceph_inode_to_client(&ci
->vfs_inode
);
1493 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1494 struct inode
*inode
= &ci
->vfs_inode
;
1495 struct ceph_cap
*cap
;
1496 int file_wanted
, used
, cap_used
;
1497 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1498 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1499 int mds
= -1; /* keep track of how far we've gone through i_caps list
1500 to avoid an infinite loop on retry */
1502 int tried_invalidate
= 0;
1503 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1504 int queue_invalidate
= 0;
1505 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1507 /* if we are unmounting, flush any unused caps immediately. */
1511 spin_lock(&ci
->i_ceph_lock
);
1513 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1514 flags
|= CHECK_CAPS_FLUSH
;
1516 /* flush snaps first time around only */
1517 if (!list_empty(&ci
->i_cap_snaps
))
1518 __ceph_flush_snaps(ci
, &session
, 0);
1521 spin_lock(&ci
->i_ceph_lock
);
1523 file_wanted
= __ceph_caps_file_wanted(ci
);
1524 used
= __ceph_caps_used(ci
);
1525 issued
= __ceph_caps_issued(ci
, &implemented
);
1526 revoking
= implemented
& ~issued
;
1529 retain
= file_wanted
| used
| CEPH_CAP_PIN
;
1530 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1532 retain
|= CEPH_CAP_ANY
; /* be greedy */
1533 } else if (S_ISDIR(inode
->i_mode
) &&
1534 (issued
& CEPH_CAP_FILE_SHARED
) &&
1535 __ceph_dir_is_complete(ci
)) {
1537 * If a directory is complete, we want to keep
1538 * the exclusive cap. So that MDS does not end up
1539 * revoking the shared cap on every create/unlink
1542 want
= CEPH_CAP_ANY_SHARED
| CEPH_CAP_FILE_EXCL
;
1546 retain
|= CEPH_CAP_ANY_SHARED
;
1548 * keep RD only if we didn't have the file open RW,
1549 * because then the mds would revoke it anyway to
1550 * journal max_size=0.
1552 if (ci
->i_max_size
== 0)
1553 retain
|= CEPH_CAP_ANY_RD
;
1557 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1558 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1559 ceph_cap_string(file_wanted
),
1560 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1561 ceph_cap_string(ci
->i_flushing_caps
),
1562 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1563 ceph_cap_string(retain
),
1564 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1565 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1566 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1569 * If we no longer need to hold onto old our caps, and we may
1570 * have cached pages, but don't want them, then try to invalidate.
1571 * If we fail, it's because pages are locked.... try again later.
1573 if ((!is_delayed
|| mdsc
->stopping
) &&
1574 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1575 inode
->i_data
.nrpages
&& /* have cached pages */
1576 (file_wanted
== 0 || /* no open files */
1577 (revoking
& (CEPH_CAP_FILE_CACHE
|
1578 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1579 !tried_invalidate
) {
1580 dout("check_caps trying to invalidate on %p\n", inode
);
1581 if (try_nonblocking_invalidate(inode
) < 0) {
1582 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1583 CEPH_CAP_FILE_LAZYIO
)) {
1584 dout("check_caps queuing invalidate\n");
1585 queue_invalidate
= 1;
1586 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1588 dout("check_caps failed to invalidate pages\n");
1589 /* we failed to invalidate pages. check these
1590 caps again later. */
1592 __cap_set_timeouts(mdsc
, ci
);
1595 tried_invalidate
= 1;
1600 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1601 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1604 /* avoid looping forever */
1605 if (mds
>= cap
->mds
||
1606 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1609 /* NOTE: no side-effects allowed, until we take s_mutex */
1612 if (ci
->i_auth_cap
&& cap
!= ci
->i_auth_cap
)
1613 cap_used
&= ~ci
->i_auth_cap
->issued
;
1615 revoking
= cap
->implemented
& ~cap
->issued
;
1616 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1617 cap
->mds
, cap
, ceph_cap_string(cap
->issued
),
1618 ceph_cap_string(cap_used
),
1619 ceph_cap_string(cap
->implemented
),
1620 ceph_cap_string(revoking
));
1622 if (cap
== ci
->i_auth_cap
&&
1623 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1624 /* request larger max_size from MDS? */
1625 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1626 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1627 dout("requesting new max_size\n");
1631 /* approaching file_max? */
1632 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1633 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1634 dout("i_size approaching max_size\n");
1638 /* flush anything dirty? */
1639 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1641 dout("flushing dirty caps\n");
1645 /* completed revocation? going down and there are no caps? */
1646 if (revoking
&& (revoking
& cap_used
) == 0) {
1647 dout("completed revocation of %s\n",
1648 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1652 /* want more caps from mds? */
1653 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1656 /* things we might delay */
1657 if ((cap
->issued
& ~retain
) == 0 &&
1658 cap
->mds_wanted
== want
)
1659 continue; /* nope, all good */
1665 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1666 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1667 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1668 ceph_cap_string(cap
->issued
),
1669 ceph_cap_string(cap
->issued
& retain
),
1670 ceph_cap_string(cap
->mds_wanted
),
1671 ceph_cap_string(want
));
1677 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1678 dout(" skipping %p I_NOFLUSH set\n", inode
);
1682 if (session
&& session
!= cap
->session
) {
1683 dout("oops, wrong session %p mutex\n", session
);
1684 mutex_unlock(&session
->s_mutex
);
1688 session
= cap
->session
;
1689 if (mutex_trylock(&session
->s_mutex
) == 0) {
1690 dout("inverting session/ino locks on %p\n",
1692 spin_unlock(&ci
->i_ceph_lock
);
1693 if (took_snap_rwsem
) {
1694 up_read(&mdsc
->snap_rwsem
);
1695 took_snap_rwsem
= 0;
1697 mutex_lock(&session
->s_mutex
);
1701 /* take snap_rwsem after session mutex */
1702 if (!took_snap_rwsem
) {
1703 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1704 dout("inverting snap/in locks on %p\n",
1706 spin_unlock(&ci
->i_ceph_lock
);
1707 down_read(&mdsc
->snap_rwsem
);
1708 took_snap_rwsem
= 1;
1711 took_snap_rwsem
= 1;
1714 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1715 flushing
= __mark_caps_flushing(inode
, session
);
1719 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1722 /* __send_cap drops i_ceph_lock */
1723 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, cap_used
,
1724 want
, retain
, flushing
);
1725 goto retry
; /* retake i_ceph_lock and restart our cap scan. */
1729 * Reschedule delayed caps release if we delayed anything,
1732 if (delayed
&& is_delayed
)
1733 force_requeue
= 1; /* __send_cap delayed release; requeue */
1734 if (!delayed
&& !is_delayed
)
1735 __cap_delay_cancel(mdsc
, ci
);
1736 else if (!is_delayed
|| force_requeue
)
1737 __cap_delay_requeue(mdsc
, ci
);
1739 spin_unlock(&ci
->i_ceph_lock
);
1741 if (queue_invalidate
)
1742 ceph_queue_invalidate(inode
);
1745 mutex_unlock(&session
->s_mutex
);
1746 if (took_snap_rwsem
)
1747 up_read(&mdsc
->snap_rwsem
);
1751 * Try to flush dirty caps back to the auth mds.
1753 static int try_flush_caps(struct inode
*inode
, u16 flush_tid
[])
1755 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1756 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1757 struct ceph_mds_session
*session
= NULL
;
1761 spin_lock(&ci
->i_ceph_lock
);
1762 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1763 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1766 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1767 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1768 int used
= __ceph_caps_used(ci
);
1769 int want
= __ceph_caps_wanted(ci
);
1772 if (!session
|| session
!= cap
->session
) {
1773 spin_unlock(&ci
->i_ceph_lock
);
1775 mutex_unlock(&session
->s_mutex
);
1776 session
= cap
->session
;
1777 mutex_lock(&session
->s_mutex
);
1780 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1783 flushing
= __mark_caps_flushing(inode
, session
);
1785 /* __send_cap drops i_ceph_lock */
1786 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1787 cap
->issued
| cap
->implemented
, flushing
);
1789 spin_lock(&ci
->i_ceph_lock
);
1791 __cap_delay_requeue(mdsc
, ci
);
1794 flushing
= ci
->i_flushing_caps
;
1796 memcpy(flush_tid
, ci
->i_cap_flush_tid
,
1797 sizeof(ci
->i_cap_flush_tid
));
1799 spin_unlock(&ci
->i_ceph_lock
);
1801 mutex_unlock(&session
->s_mutex
);
1806 * Return true if we've flushed caps through the given flush_tid.
1808 static int caps_are_flushed(struct inode
*inode
, u16 flush_tid
[])
1810 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1813 spin_lock(&ci
->i_ceph_lock
);
1814 for (i
= 0; i
< CEPH_CAP_BITS
; i
++) {
1815 if (!(ci
->i_flushing_caps
& (1 << i
)))
1817 // tid only has 16 bits. we need to handle wrapping
1818 if ((s16
)(ci
->i_cap_flush_tid
[i
] - flush_tid
[i
]) <= 0) {
1819 /* still flushing this bit */
1824 spin_unlock(&ci
->i_ceph_lock
);
1829 * Wait on any unsafe replies for the given inode. First wait on the
1830 * newest request, and make that the upper bound. Then, if there are
1831 * more requests, keep waiting on the oldest as long as it is still older
1832 * than the original request.
1834 static void sync_write_wait(struct inode
*inode
)
1836 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1837 struct list_head
*head
= &ci
->i_unsafe_writes
;
1838 struct ceph_osd_request
*req
;
1841 if (!S_ISREG(inode
->i_mode
))
1844 spin_lock(&ci
->i_unsafe_lock
);
1845 if (list_empty(head
))
1848 /* set upper bound as _last_ entry in chain */
1849 req
= list_last_entry(head
, struct ceph_osd_request
,
1851 last_tid
= req
->r_tid
;
1854 ceph_osdc_get_request(req
);
1855 spin_unlock(&ci
->i_unsafe_lock
);
1856 dout("sync_write_wait on tid %llu (until %llu)\n",
1857 req
->r_tid
, last_tid
);
1858 wait_for_completion(&req
->r_safe_completion
);
1859 spin_lock(&ci
->i_unsafe_lock
);
1860 ceph_osdc_put_request(req
);
1863 * from here on look at first entry in chain, since we
1864 * only want to wait for anything older than last_tid
1866 if (list_empty(head
))
1868 req
= list_first_entry(head
, struct ceph_osd_request
,
1870 } while (req
->r_tid
< last_tid
);
1872 spin_unlock(&ci
->i_unsafe_lock
);
1876 * wait for any uncommitted directory operations to commit.
1878 static int unsafe_dirop_wait(struct inode
*inode
)
1880 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1881 struct list_head
*head
= &ci
->i_unsafe_dirops
;
1882 struct ceph_mds_request
*req
;
1886 if (!S_ISDIR(inode
->i_mode
))
1889 spin_lock(&ci
->i_unsafe_lock
);
1890 if (list_empty(head
))
1893 req
= list_last_entry(head
, struct ceph_mds_request
,
1895 last_tid
= req
->r_tid
;
1898 ceph_mdsc_get_request(req
);
1899 spin_unlock(&ci
->i_unsafe_lock
);
1901 dout("unsafe_dirop_wait %p wait on tid %llu (until %llu)\n",
1902 inode
, req
->r_tid
, last_tid
);
1903 ret
= !wait_for_completion_timeout(&req
->r_safe_completion
,
1904 ceph_timeout_jiffies(req
->r_timeout
));
1906 ret
= -EIO
; /* timed out */
1908 ceph_mdsc_put_request(req
);
1910 spin_lock(&ci
->i_unsafe_lock
);
1911 if (ret
|| list_empty(head
))
1913 req
= list_first_entry(head
, struct ceph_mds_request
,
1915 } while (req
->r_tid
< last_tid
);
1917 spin_unlock(&ci
->i_unsafe_lock
);
1921 int ceph_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1923 struct inode
*inode
= file
->f_mapping
->host
;
1924 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1925 u16 flush_tid
[CEPH_CAP_BITS
];
1929 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1930 sync_write_wait(inode
);
1932 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1939 mutex_lock(&inode
->i_mutex
);
1941 dirty
= try_flush_caps(inode
, flush_tid
);
1942 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1944 ret
= unsafe_dirop_wait(inode
);
1947 * only wait on non-file metadata writeback (the mds
1948 * can recover size and mtime, so we don't need to
1951 if (!ret
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1952 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1953 caps_are_flushed(inode
, flush_tid
));
1955 mutex_unlock(&inode
->i_mutex
);
1957 dout("fsync %p%s result=%d\n", inode
, datasync
? " datasync" : "", ret
);
1962 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1963 * queue inode for flush but don't do so immediately, because we can
1964 * get by with fewer MDS messages if we wait for data writeback to
1967 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1969 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1970 u16 flush_tid
[CEPH_CAP_BITS
];
1973 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1975 dout("write_inode %p wait=%d\n", inode
, wait
);
1977 dirty
= try_flush_caps(inode
, flush_tid
);
1979 err
= wait_event_interruptible(ci
->i_cap_wq
,
1980 caps_are_flushed(inode
, flush_tid
));
1982 struct ceph_mds_client
*mdsc
=
1983 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1985 spin_lock(&ci
->i_ceph_lock
);
1986 if (__ceph_caps_dirty(ci
))
1987 __cap_delay_requeue_front(mdsc
, ci
);
1988 spin_unlock(&ci
->i_ceph_lock
);
1994 * After a recovering MDS goes active, we need to resend any caps
1997 * Caller holds session->s_mutex.
1999 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
2000 struct ceph_mds_session
*session
)
2002 struct ceph_cap_snap
*capsnap
;
2004 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
2005 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
2007 struct ceph_inode_info
*ci
= capsnap
->ci
;
2008 struct inode
*inode
= &ci
->vfs_inode
;
2009 struct ceph_cap
*cap
;
2011 spin_lock(&ci
->i_ceph_lock
);
2012 cap
= ci
->i_auth_cap
;
2013 if (cap
&& cap
->session
== session
) {
2014 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
2016 __ceph_flush_snaps(ci
, &session
, 1);
2018 pr_err("%p auth cap %p not mds%d ???\n", inode
,
2019 cap
, session
->s_mds
);
2021 spin_unlock(&ci
->i_ceph_lock
);
2025 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
2026 struct ceph_mds_session
*session
)
2028 struct ceph_inode_info
*ci
;
2030 kick_flushing_capsnaps(mdsc
, session
);
2032 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
2033 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
2034 struct inode
*inode
= &ci
->vfs_inode
;
2035 struct ceph_cap
*cap
;
2038 spin_lock(&ci
->i_ceph_lock
);
2039 cap
= ci
->i_auth_cap
;
2040 if (cap
&& cap
->session
== session
) {
2041 dout("kick_flushing_caps %p cap %p %s\n", inode
,
2042 cap
, ceph_cap_string(ci
->i_flushing_caps
));
2043 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
2044 __ceph_caps_used(ci
),
2045 __ceph_caps_wanted(ci
),
2046 cap
->issued
| cap
->implemented
,
2047 ci
->i_flushing_caps
);
2049 spin_lock(&ci
->i_ceph_lock
);
2050 __cap_delay_requeue(mdsc
, ci
);
2051 spin_unlock(&ci
->i_ceph_lock
);
2054 pr_err("%p auth cap %p not mds%d ???\n", inode
,
2055 cap
, session
->s_mds
);
2056 spin_unlock(&ci
->i_ceph_lock
);
2061 static void kick_flushing_inode_caps(struct ceph_mds_client
*mdsc
,
2062 struct ceph_mds_session
*session
,
2063 struct inode
*inode
)
2065 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2066 struct ceph_cap
*cap
;
2069 spin_lock(&ci
->i_ceph_lock
);
2070 cap
= ci
->i_auth_cap
;
2071 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode
,
2072 ceph_cap_string(ci
->i_flushing_caps
), ci
->i_cap_flush_seq
);
2074 __ceph_flush_snaps(ci
, &session
, 1);
2076 if (ci
->i_flushing_caps
) {
2077 spin_lock(&mdsc
->cap_dirty_lock
);
2078 list_move_tail(&ci
->i_flushing_item
,
2079 &cap
->session
->s_cap_flushing
);
2080 spin_unlock(&mdsc
->cap_dirty_lock
);
2082 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
2083 __ceph_caps_used(ci
),
2084 __ceph_caps_wanted(ci
),
2085 cap
->issued
| cap
->implemented
,
2086 ci
->i_flushing_caps
);
2088 spin_lock(&ci
->i_ceph_lock
);
2089 __cap_delay_requeue(mdsc
, ci
);
2090 spin_unlock(&ci
->i_ceph_lock
);
2093 spin_unlock(&ci
->i_ceph_lock
);
2099 * Take references to capabilities we hold, so that we don't release
2100 * them to the MDS prematurely.
2102 * Protected by i_ceph_lock.
2104 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
,
2105 bool snap_rwsem_locked
)
2107 if (got
& CEPH_CAP_PIN
)
2109 if (got
& CEPH_CAP_FILE_RD
)
2111 if (got
& CEPH_CAP_FILE_CACHE
)
2112 ci
->i_rdcache_ref
++;
2113 if (got
& CEPH_CAP_FILE_WR
) {
2114 if (ci
->i_wr_ref
== 0 && !ci
->i_head_snapc
) {
2115 BUG_ON(!snap_rwsem_locked
);
2116 ci
->i_head_snapc
= ceph_get_snap_context(
2117 ci
->i_snap_realm
->cached_context
);
2121 if (got
& CEPH_CAP_FILE_BUFFER
) {
2122 if (ci
->i_wb_ref
== 0)
2123 ihold(&ci
->vfs_inode
);
2125 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2126 &ci
->vfs_inode
, ci
->i_wb_ref
-1, ci
->i_wb_ref
);
2131 * Try to grab cap references. Specify those refs we @want, and the
2132 * minimal set we @need. Also include the larger offset we are writing
2133 * to (when applicable), and check against max_size here as well.
2134 * Note that caller is responsible for ensuring max_size increases are
2135 * requested from the MDS.
2137 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
2138 loff_t endoff
, bool nonblock
, int *got
, int *err
)
2140 struct inode
*inode
= &ci
->vfs_inode
;
2141 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2143 int have
, implemented
;
2145 bool snap_rwsem_locked
= false;
2147 dout("get_cap_refs %p need %s want %s\n", inode
,
2148 ceph_cap_string(need
), ceph_cap_string(want
));
2151 spin_lock(&ci
->i_ceph_lock
);
2153 /* make sure file is actually open */
2154 file_wanted
= __ceph_caps_file_wanted(ci
);
2155 if ((file_wanted
& need
) == 0) {
2156 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2157 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2163 /* finish pending truncate */
2164 while (ci
->i_truncate_pending
) {
2165 spin_unlock(&ci
->i_ceph_lock
);
2166 if (snap_rwsem_locked
) {
2167 up_read(&mdsc
->snap_rwsem
);
2168 snap_rwsem_locked
= false;
2170 __ceph_do_pending_vmtruncate(inode
);
2171 spin_lock(&ci
->i_ceph_lock
);
2174 have
= __ceph_caps_issued(ci
, &implemented
);
2176 if (have
& need
& CEPH_CAP_FILE_WR
) {
2177 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2178 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2179 inode
, endoff
, ci
->i_max_size
);
2180 if (endoff
> ci
->i_requested_max_size
) {
2187 * If a sync write is in progress, we must wait, so that we
2188 * can get a final snapshot value for size+mtime.
2190 if (__ceph_have_pending_cap_snap(ci
)) {
2191 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2196 if ((have
& need
) == need
) {
2198 * Look at (implemented & ~have & not) so that we keep waiting
2199 * on transition from wanted -> needed caps. This is needed
2200 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2201 * going before a prior buffered writeback happens.
2203 int not = want
& ~(have
& need
);
2204 int revoking
= implemented
& ~have
;
2205 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2206 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2207 ceph_cap_string(revoking
));
2208 if ((revoking
& not) == 0) {
2209 if (!snap_rwsem_locked
&&
2210 !ci
->i_head_snapc
&&
2211 (need
& CEPH_CAP_FILE_WR
)) {
2212 if (!down_read_trylock(&mdsc
->snap_rwsem
)) {
2214 * we can not call down_read() when
2215 * task isn't in TASK_RUNNING state
2223 spin_unlock(&ci
->i_ceph_lock
);
2224 down_read(&mdsc
->snap_rwsem
);
2225 snap_rwsem_locked
= true;
2228 snap_rwsem_locked
= true;
2230 *got
= need
| (have
& want
);
2231 __take_cap_refs(ci
, *got
, true);
2235 int session_readonly
= false;
2236 if ((need
& CEPH_CAP_FILE_WR
) && ci
->i_auth_cap
) {
2237 struct ceph_mds_session
*s
= ci
->i_auth_cap
->session
;
2238 spin_lock(&s
->s_cap_lock
);
2239 session_readonly
= s
->s_readonly
;
2240 spin_unlock(&s
->s_cap_lock
);
2242 if (session_readonly
) {
2243 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2244 inode
, ceph_cap_string(need
), ci
->i_auth_cap
->mds
);
2250 dout("get_cap_refs %p have %s needed %s\n", inode
,
2251 ceph_cap_string(have
), ceph_cap_string(need
));
2254 spin_unlock(&ci
->i_ceph_lock
);
2255 if (snap_rwsem_locked
)
2256 up_read(&mdsc
->snap_rwsem
);
2258 dout("get_cap_refs %p ret %d got %s\n", inode
,
2259 ret
, ceph_cap_string(*got
));
2264 * Check the offset we are writing up to against our current
2265 * max_size. If necessary, tell the MDS we want to write to
2268 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2270 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2273 /* do we need to explicitly request a larger max_size? */
2274 spin_lock(&ci
->i_ceph_lock
);
2275 if (endoff
>= ci
->i_max_size
&& endoff
> ci
->i_wanted_max_size
) {
2276 dout("write %p at large endoff %llu, req max_size\n",
2278 ci
->i_wanted_max_size
= endoff
;
2280 /* duplicate ceph_check_caps()'s logic */
2281 if (ci
->i_auth_cap
&&
2282 (ci
->i_auth_cap
->issued
& CEPH_CAP_FILE_WR
) &&
2283 ci
->i_wanted_max_size
> ci
->i_max_size
&&
2284 ci
->i_wanted_max_size
> ci
->i_requested_max_size
)
2286 spin_unlock(&ci
->i_ceph_lock
);
2288 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2292 * Wait for caps, and take cap references. If we can't get a WR cap
2293 * due to a small max_size, make sure we check_max_size (and possibly
2294 * ask the mds) so we don't get hung up indefinitely.
2296 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
,
2297 loff_t endoff
, int *got
, struct page
**pinned_page
)
2299 int _got
, ret
, err
= 0;
2301 ret
= ceph_pool_perm_check(ci
, need
);
2307 check_max_size(&ci
->vfs_inode
, endoff
);
2311 ret
= try_get_cap_refs(ci
, need
, want
, endoff
,
2312 false, &_got
, &err
);
2319 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2320 try_get_cap_refs(ci
, need
, want
, endoff
,
2321 true, &_got
, &err
));
2330 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
2331 (_got
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
2332 i_size_read(&ci
->vfs_inode
) > 0) {
2334 find_get_page(ci
->vfs_inode
.i_mapping
, 0);
2336 if (PageUptodate(page
)) {
2337 *pinned_page
= page
;
2340 page_cache_release(page
);
2343 * drop cap refs first because getattr while
2344 * holding * caps refs can cause deadlock.
2346 ceph_put_cap_refs(ci
, _got
);
2350 * getattr request will bring inline data into
2353 ret
= __ceph_do_getattr(&ci
->vfs_inode
, NULL
,
2354 CEPH_STAT_CAP_INLINE_DATA
,
2368 * Take cap refs. Caller must already know we hold at least one ref
2369 * on the caps in question or we don't know this is safe.
2371 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2373 spin_lock(&ci
->i_ceph_lock
);
2374 __take_cap_refs(ci
, caps
, false);
2375 spin_unlock(&ci
->i_ceph_lock
);
2380 * drop cap_snap that is not associated with any snapshot.
2381 * we don't need to send FLUSHSNAP message for it.
2383 static int ceph_try_drop_cap_snap(struct ceph_cap_snap
*capsnap
)
2385 if (!capsnap
->need_flush
&&
2386 !capsnap
->writing
&& !capsnap
->dirty_pages
) {
2388 dout("dropping cap_snap %p follows %llu\n",
2389 capsnap
, capsnap
->follows
);
2390 ceph_put_snap_context(capsnap
->context
);
2391 list_del(&capsnap
->ci_item
);
2392 list_del(&capsnap
->flushing_item
);
2393 ceph_put_cap_snap(capsnap
);
2402 * If we released the last ref on any given cap, call ceph_check_caps
2403 * to release (or schedule a release).
2405 * If we are releasing a WR cap (from a sync write), finalize any affected
2406 * cap_snap, and wake up any waiters.
2408 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2410 struct inode
*inode
= &ci
->vfs_inode
;
2411 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2413 spin_lock(&ci
->i_ceph_lock
);
2414 if (had
& CEPH_CAP_PIN
)
2416 if (had
& CEPH_CAP_FILE_RD
)
2417 if (--ci
->i_rd_ref
== 0)
2419 if (had
& CEPH_CAP_FILE_CACHE
)
2420 if (--ci
->i_rdcache_ref
== 0)
2422 if (had
& CEPH_CAP_FILE_BUFFER
) {
2423 if (--ci
->i_wb_ref
== 0) {
2427 dout("put_cap_refs %p wb %d -> %d (?)\n",
2428 inode
, ci
->i_wb_ref
+1, ci
->i_wb_ref
);
2430 if (had
& CEPH_CAP_FILE_WR
)
2431 if (--ci
->i_wr_ref
== 0) {
2433 if (__ceph_have_pending_cap_snap(ci
)) {
2434 struct ceph_cap_snap
*capsnap
=
2435 list_last_entry(&ci
->i_cap_snaps
,
2436 struct ceph_cap_snap
,
2438 capsnap
->writing
= 0;
2439 if (ceph_try_drop_cap_snap(capsnap
))
2441 else if (__ceph_finish_cap_snap(ci
, capsnap
))
2445 if (ci
->i_wrbuffer_ref_head
== 0 &&
2446 ci
->i_dirty_caps
== 0 &&
2447 ci
->i_flushing_caps
== 0) {
2448 BUG_ON(!ci
->i_head_snapc
);
2449 ceph_put_snap_context(ci
->i_head_snapc
);
2450 ci
->i_head_snapc
= NULL
;
2452 /* see comment in __ceph_remove_cap() */
2453 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
)
2454 drop_inode_snap_realm(ci
);
2456 spin_unlock(&ci
->i_ceph_lock
);
2458 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2459 last
? " last" : "", put
? " put" : "");
2461 if (last
&& !flushsnaps
)
2462 ceph_check_caps(ci
, 0, NULL
);
2463 else if (flushsnaps
)
2464 ceph_flush_snaps(ci
);
2466 wake_up_all(&ci
->i_cap_wq
);
2472 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2473 * context. Adjust per-snap dirty page accounting as appropriate.
2474 * Once all dirty data for a cap_snap is flushed, flush snapped file
2475 * metadata back to the MDS. If we dropped the last ref, call
2478 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2479 struct ceph_snap_context
*snapc
)
2481 struct inode
*inode
= &ci
->vfs_inode
;
2483 int complete_capsnap
= 0;
2484 int drop_capsnap
= 0;
2486 struct ceph_cap_snap
*capsnap
= NULL
;
2488 spin_lock(&ci
->i_ceph_lock
);
2489 ci
->i_wrbuffer_ref
-= nr
;
2490 last
= !ci
->i_wrbuffer_ref
;
2492 if (ci
->i_head_snapc
== snapc
) {
2493 ci
->i_wrbuffer_ref_head
-= nr
;
2494 if (ci
->i_wrbuffer_ref_head
== 0 &&
2495 ci
->i_wr_ref
== 0 &&
2496 ci
->i_dirty_caps
== 0 &&
2497 ci
->i_flushing_caps
== 0) {
2498 BUG_ON(!ci
->i_head_snapc
);
2499 ceph_put_snap_context(ci
->i_head_snapc
);
2500 ci
->i_head_snapc
= NULL
;
2502 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2504 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2505 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2506 last
? " LAST" : "");
2508 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2509 if (capsnap
->context
== snapc
) {
2515 capsnap
->dirty_pages
-= nr
;
2516 if (capsnap
->dirty_pages
== 0) {
2517 complete_capsnap
= 1;
2518 drop_capsnap
= ceph_try_drop_cap_snap(capsnap
);
2520 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2521 " snap %lld %d/%d -> %d/%d %s%s\n",
2522 inode
, capsnap
, capsnap
->context
->seq
,
2523 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2524 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2525 last
? " (wrbuffer last)" : "",
2526 complete_capsnap
? " (complete capsnap)" : "");
2529 spin_unlock(&ci
->i_ceph_lock
);
2532 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2534 } else if (complete_capsnap
) {
2535 ceph_flush_snaps(ci
);
2536 wake_up_all(&ci
->i_cap_wq
);
2543 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2545 static void invalidate_aliases(struct inode
*inode
)
2547 struct dentry
*dn
, *prev
= NULL
;
2549 dout("invalidate_aliases inode %p\n", inode
);
2550 d_prune_aliases(inode
);
2552 * For non-directory inode, d_find_alias() only returns
2553 * hashed dentry. After calling d_invalidate(), the
2554 * dentry becomes unhashed.
2556 * For directory inode, d_find_alias() can return
2557 * unhashed dentry. But directory inode should have
2558 * one alias at most.
2560 while ((dn
= d_find_alias(inode
))) {
2575 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2576 * actually be a revocation if it specifies a smaller cap set.)
2578 * caller holds s_mutex and i_ceph_lock, we drop both.
2580 static void handle_cap_grant(struct ceph_mds_client
*mdsc
,
2581 struct inode
*inode
, struct ceph_mds_caps
*grant
,
2583 void *inline_data
, int inline_len
,
2584 struct ceph_buffer
*xattr_buf
,
2585 struct ceph_mds_session
*session
,
2586 struct ceph_cap
*cap
, int issued
)
2587 __releases(ci
->i_ceph_lock
)
2588 __releases(mdsc
->snap_rwsem
)
2590 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2591 int mds
= session
->s_mds
;
2592 int seq
= le32_to_cpu(grant
->seq
);
2593 int newcaps
= le32_to_cpu(grant
->caps
);
2594 int used
, wanted
, dirty
;
2595 u64 size
= le64_to_cpu(grant
->size
);
2596 u64 max_size
= le64_to_cpu(grant
->max_size
);
2597 struct timespec mtime
, atime
, ctime
;
2600 bool writeback
= false;
2601 bool queue_trunc
= false;
2602 bool queue_invalidate
= false;
2603 bool queue_revalidate
= false;
2604 bool deleted_inode
= false;
2605 bool fill_inline
= false;
2607 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2608 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2609 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2614 * auth mds of the inode changed. we received the cap export message,
2615 * but still haven't received the cap import message. handle_cap_export
2616 * updated the new auth MDS' cap.
2618 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2619 * that was sent before the cap import message. So don't remove caps.
2621 if (ceph_seq_cmp(seq
, cap
->seq
) <= 0) {
2622 WARN_ON(cap
!= ci
->i_auth_cap
);
2623 WARN_ON(cap
->cap_id
!= le64_to_cpu(grant
->cap_id
));
2625 newcaps
|= cap
->issued
;
2629 * If CACHE is being revoked, and we have no dirty buffers,
2630 * try to invalidate (once). (If there are dirty buffers, we
2631 * will invalidate _after_ writeback.)
2633 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2634 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2635 !ci
->i_wrbuffer_ref
) {
2636 if (try_nonblocking_invalidate(inode
)) {
2637 /* there were locked pages.. invalidate later
2638 in a separate thread. */
2639 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2640 queue_invalidate
= true;
2641 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2645 ceph_fscache_invalidate(inode
);
2648 /* side effects now are allowed */
2649 cap
->cap_gen
= session
->s_cap_gen
;
2652 __check_cap_issue(ci
, cap
, newcaps
);
2654 if ((newcaps
& CEPH_CAP_AUTH_SHARED
) &&
2655 (issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2656 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2657 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(grant
->uid
));
2658 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(grant
->gid
));
2659 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2660 from_kuid(&init_user_ns
, inode
->i_uid
),
2661 from_kgid(&init_user_ns
, inode
->i_gid
));
2664 if ((newcaps
& CEPH_CAP_AUTH_SHARED
) &&
2665 (issued
& CEPH_CAP_LINK_EXCL
) == 0) {
2666 set_nlink(inode
, le32_to_cpu(grant
->nlink
));
2667 if (inode
->i_nlink
== 0 &&
2668 (newcaps
& (CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
)))
2669 deleted_inode
= true;
2672 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2673 int len
= le32_to_cpu(grant
->xattr_len
);
2674 u64 version
= le64_to_cpu(grant
->xattr_version
);
2676 if (version
> ci
->i_xattrs
.version
) {
2677 dout(" got new xattrs v%llu on %p len %d\n",
2678 version
, inode
, len
);
2679 if (ci
->i_xattrs
.blob
)
2680 ceph_buffer_put(ci
->i_xattrs
.blob
);
2681 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2682 ci
->i_xattrs
.version
= version
;
2683 ceph_forget_all_cached_acls(inode
);
2687 /* Do we need to revalidate our fscache cookie. Don't bother on the
2688 * first cache cap as we already validate at cookie creation time. */
2689 if ((issued
& CEPH_CAP_FILE_CACHE
) && ci
->i_rdcache_gen
> 1)
2690 queue_revalidate
= true;
2692 if (newcaps
& CEPH_CAP_ANY_RD
) {
2693 /* ctime/mtime/atime? */
2694 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2695 ceph_decode_timespec(&atime
, &grant
->atime
);
2696 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2697 ceph_fill_file_time(inode
, issued
,
2698 le32_to_cpu(grant
->time_warp_seq
),
2699 &ctime
, &mtime
, &atime
);
2702 if (newcaps
& (CEPH_CAP_ANY_FILE_RD
| CEPH_CAP_ANY_FILE_WR
)) {
2703 /* file layout may have changed */
2704 ci
->i_layout
= grant
->layout
;
2705 /* size/truncate_seq? */
2706 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2707 le32_to_cpu(grant
->truncate_seq
),
2708 le64_to_cpu(grant
->truncate_size
),
2710 /* max size increase? */
2711 if (ci
->i_auth_cap
== cap
&& max_size
!= ci
->i_max_size
) {
2712 dout("max_size %lld -> %llu\n",
2713 ci
->i_max_size
, max_size
);
2714 ci
->i_max_size
= max_size
;
2715 if (max_size
>= ci
->i_wanted_max_size
) {
2716 ci
->i_wanted_max_size
= 0; /* reset */
2717 ci
->i_requested_max_size
= 0;
2723 /* check cap bits */
2724 wanted
= __ceph_caps_wanted(ci
);
2725 used
= __ceph_caps_used(ci
);
2726 dirty
= __ceph_caps_dirty(ci
);
2727 dout(" my wanted = %s, used = %s, dirty %s\n",
2728 ceph_cap_string(wanted
),
2729 ceph_cap_string(used
),
2730 ceph_cap_string(dirty
));
2731 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2732 dout("mds wanted %s -> %s\n",
2733 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2734 ceph_cap_string(wanted
));
2735 /* imported cap may not have correct mds_wanted */
2736 if (le32_to_cpu(grant
->op
) == CEPH_CAP_OP_IMPORT
)
2740 /* revocation, grant, or no-op? */
2741 if (cap
->issued
& ~newcaps
) {
2742 int revoking
= cap
->issued
& ~newcaps
;
2744 dout("revocation: %s -> %s (revoking %s)\n",
2745 ceph_cap_string(cap
->issued
),
2746 ceph_cap_string(newcaps
),
2747 ceph_cap_string(revoking
));
2748 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2749 writeback
= true; /* initiate writeback; will delay ack */
2750 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2751 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2753 ; /* do nothing yet, invalidation will be queued */
2754 else if (cap
== ci
->i_auth_cap
)
2755 check_caps
= 1; /* check auth cap only */
2757 check_caps
= 2; /* check all caps */
2758 cap
->issued
= newcaps
;
2759 cap
->implemented
|= newcaps
;
2760 } else if (cap
->issued
== newcaps
) {
2761 dout("caps unchanged: %s -> %s\n",
2762 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2764 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2765 ceph_cap_string(newcaps
));
2766 /* non-auth MDS is revoking the newly grant caps ? */
2767 if (cap
== ci
->i_auth_cap
&&
2768 __ceph_caps_revoking_other(ci
, cap
, newcaps
))
2771 cap
->issued
= newcaps
;
2772 cap
->implemented
|= newcaps
; /* add bits only, to
2773 * avoid stepping on a
2774 * pending revocation */
2777 BUG_ON(cap
->issued
& ~cap
->implemented
);
2779 if (inline_version
> 0 && inline_version
>= ci
->i_inline_version
) {
2780 ci
->i_inline_version
= inline_version
;
2781 if (ci
->i_inline_version
!= CEPH_INLINE_NONE
&&
2782 (newcaps
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)))
2786 spin_unlock(&ci
->i_ceph_lock
);
2788 if (le32_to_cpu(grant
->op
) == CEPH_CAP_OP_IMPORT
) {
2789 kick_flushing_inode_caps(mdsc
, session
, inode
);
2790 up_read(&mdsc
->snap_rwsem
);
2791 if (newcaps
& ~issued
)
2796 ceph_fill_inline_data(inode
, NULL
, inline_data
, inline_len
);
2799 ceph_queue_vmtruncate(inode
);
2800 ceph_queue_revalidate(inode
);
2801 } else if (queue_revalidate
)
2802 ceph_queue_revalidate(inode
);
2806 * queue inode for writeback: we can't actually call
2807 * filemap_write_and_wait, etc. from message handler
2810 ceph_queue_writeback(inode
);
2811 if (queue_invalidate
)
2812 ceph_queue_invalidate(inode
);
2814 invalidate_aliases(inode
);
2816 wake_up_all(&ci
->i_cap_wq
);
2818 if (check_caps
== 1)
2819 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2821 else if (check_caps
== 2)
2822 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2824 mutex_unlock(&session
->s_mutex
);
2828 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2829 * MDS has been safely committed.
2831 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2832 struct ceph_mds_caps
*m
,
2833 struct ceph_mds_session
*session
,
2834 struct ceph_cap
*cap
)
2835 __releases(ci
->i_ceph_lock
)
2837 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2838 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2839 unsigned seq
= le32_to_cpu(m
->seq
);
2840 int dirty
= le32_to_cpu(m
->dirty
);
2845 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2846 if ((dirty
& (1 << i
)) &&
2847 (u16
)flush_tid
== ci
->i_cap_flush_tid
[i
])
2850 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2851 " flushing %s -> %s\n",
2852 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2853 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2854 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2856 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2859 ci
->i_flushing_caps
&= ~cleaned
;
2861 spin_lock(&mdsc
->cap_dirty_lock
);
2862 if (ci
->i_flushing_caps
== 0) {
2863 list_del_init(&ci
->i_flushing_item
);
2864 if (!list_empty(&session
->s_cap_flushing
))
2865 dout(" mds%d still flushing cap on %p\n",
2867 &list_entry(session
->s_cap_flushing
.next
,
2868 struct ceph_inode_info
,
2869 i_flushing_item
)->vfs_inode
);
2870 mdsc
->num_cap_flushing
--;
2871 wake_up_all(&mdsc
->cap_flushing_wq
);
2872 dout(" inode %p now !flushing\n", inode
);
2874 if (ci
->i_dirty_caps
== 0) {
2875 dout(" inode %p now clean\n", inode
);
2876 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2878 if (ci
->i_wr_ref
== 0 &&
2879 ci
->i_wrbuffer_ref_head
== 0) {
2880 BUG_ON(!ci
->i_head_snapc
);
2881 ceph_put_snap_context(ci
->i_head_snapc
);
2882 ci
->i_head_snapc
= NULL
;
2885 BUG_ON(list_empty(&ci
->i_dirty_item
));
2888 spin_unlock(&mdsc
->cap_dirty_lock
);
2889 wake_up_all(&ci
->i_cap_wq
);
2892 spin_unlock(&ci
->i_ceph_lock
);
2898 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2899 * throw away our cap_snap.
2901 * Caller hold s_mutex.
2903 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2904 struct ceph_mds_caps
*m
,
2905 struct ceph_mds_session
*session
)
2907 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2908 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2909 u64 follows
= le64_to_cpu(m
->snap_follows
);
2910 struct ceph_cap_snap
*capsnap
;
2913 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2914 inode
, ci
, session
->s_mds
, follows
);
2916 spin_lock(&ci
->i_ceph_lock
);
2917 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2918 if (capsnap
->follows
== follows
) {
2919 if (capsnap
->flush_tid
!= flush_tid
) {
2920 dout(" cap_snap %p follows %lld tid %lld !="
2921 " %lld\n", capsnap
, follows
,
2922 flush_tid
, capsnap
->flush_tid
);
2925 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2926 dout(" removing %p cap_snap %p follows %lld\n",
2927 inode
, capsnap
, follows
);
2928 ceph_put_snap_context(capsnap
->context
);
2929 list_del(&capsnap
->ci_item
);
2930 list_del(&capsnap
->flushing_item
);
2931 ceph_put_cap_snap(capsnap
);
2932 wake_up_all(&mdsc
->cap_flushing_wq
);
2936 dout(" skipping cap_snap %p follows %lld\n",
2937 capsnap
, capsnap
->follows
);
2940 spin_unlock(&ci
->i_ceph_lock
);
2946 * Handle TRUNC from MDS, indicating file truncation.
2948 * caller hold s_mutex.
2950 static void handle_cap_trunc(struct inode
*inode
,
2951 struct ceph_mds_caps
*trunc
,
2952 struct ceph_mds_session
*session
)
2953 __releases(ci
->i_ceph_lock
)
2955 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2956 int mds
= session
->s_mds
;
2957 int seq
= le32_to_cpu(trunc
->seq
);
2958 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2959 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2960 u64 size
= le64_to_cpu(trunc
->size
);
2961 int implemented
= 0;
2962 int dirty
= __ceph_caps_dirty(ci
);
2963 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2964 int queue_trunc
= 0;
2966 issued
|= implemented
| dirty
;
2968 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2969 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2970 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2971 truncate_seq
, truncate_size
, size
);
2972 spin_unlock(&ci
->i_ceph_lock
);
2975 ceph_queue_vmtruncate(inode
);
2976 ceph_fscache_invalidate(inode
);
2981 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2982 * different one. If we are the most recent migration we've seen (as
2983 * indicated by mseq), make note of the migrating cap bits for the
2984 * duration (until we see the corresponding IMPORT).
2986 * caller holds s_mutex
2988 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2989 struct ceph_mds_cap_peer
*ph
,
2990 struct ceph_mds_session
*session
)
2992 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2993 struct ceph_mds_session
*tsession
= NULL
;
2994 struct ceph_cap
*cap
, *tcap
, *new_cap
= NULL
;
2995 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2997 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2998 unsigned t_seq
, t_mseq
;
3000 int mds
= session
->s_mds
;
3003 t_cap_id
= le64_to_cpu(ph
->cap_id
);
3004 t_seq
= le32_to_cpu(ph
->seq
);
3005 t_mseq
= le32_to_cpu(ph
->mseq
);
3006 target
= le32_to_cpu(ph
->mds
);
3008 t_cap_id
= t_seq
= t_mseq
= 0;
3012 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
3013 inode
, ci
, mds
, mseq
, target
);
3015 spin_lock(&ci
->i_ceph_lock
);
3016 cap
= __get_cap_for_mds(ci
, mds
);
3017 if (!cap
|| cap
->cap_id
!= le64_to_cpu(ex
->cap_id
))
3021 __ceph_remove_cap(cap
, false);
3026 * now we know we haven't received the cap import message yet
3027 * because the exported cap still exist.
3030 issued
= cap
->issued
;
3031 WARN_ON(issued
!= cap
->implemented
);
3033 tcap
= __get_cap_for_mds(ci
, target
);
3035 /* already have caps from the target */
3036 if (tcap
->cap_id
!= t_cap_id
||
3037 ceph_seq_cmp(tcap
->seq
, t_seq
) < 0) {
3038 dout(" updating import cap %p mds%d\n", tcap
, target
);
3039 tcap
->cap_id
= t_cap_id
;
3040 tcap
->seq
= t_seq
- 1;
3041 tcap
->issue_seq
= t_seq
- 1;
3042 tcap
->mseq
= t_mseq
;
3043 tcap
->issued
|= issued
;
3044 tcap
->implemented
|= issued
;
3045 if (cap
== ci
->i_auth_cap
)
3046 ci
->i_auth_cap
= tcap
;
3047 if (ci
->i_flushing_caps
&& ci
->i_auth_cap
== tcap
) {
3048 spin_lock(&mdsc
->cap_dirty_lock
);
3049 list_move_tail(&ci
->i_flushing_item
,
3050 &tcap
->session
->s_cap_flushing
);
3051 spin_unlock(&mdsc
->cap_dirty_lock
);
3054 __ceph_remove_cap(cap
, false);
3056 } else if (tsession
) {
3057 /* add placeholder for the export tagert */
3058 int flag
= (cap
== ci
->i_auth_cap
) ? CEPH_CAP_FLAG_AUTH
: 0;
3059 ceph_add_cap(inode
, tsession
, t_cap_id
, -1, issued
, 0,
3060 t_seq
- 1, t_mseq
, (u64
)-1, flag
, &new_cap
);
3062 __ceph_remove_cap(cap
, false);
3066 spin_unlock(&ci
->i_ceph_lock
);
3067 mutex_unlock(&session
->s_mutex
);
3069 /* open target session */
3070 tsession
= ceph_mdsc_open_export_target_session(mdsc
, target
);
3071 if (!IS_ERR(tsession
)) {
3073 mutex_lock(&session
->s_mutex
);
3074 mutex_lock_nested(&tsession
->s_mutex
,
3075 SINGLE_DEPTH_NESTING
);
3077 mutex_lock(&tsession
->s_mutex
);
3078 mutex_lock_nested(&session
->s_mutex
,
3079 SINGLE_DEPTH_NESTING
);
3081 new_cap
= ceph_get_cap(mdsc
, NULL
);
3090 spin_unlock(&ci
->i_ceph_lock
);
3091 mutex_unlock(&session
->s_mutex
);
3093 mutex_unlock(&tsession
->s_mutex
);
3094 ceph_put_mds_session(tsession
);
3097 ceph_put_cap(mdsc
, new_cap
);
3101 * Handle cap IMPORT.
3103 * caller holds s_mutex. acquires i_ceph_lock
3105 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
3106 struct inode
*inode
, struct ceph_mds_caps
*im
,
3107 struct ceph_mds_cap_peer
*ph
,
3108 struct ceph_mds_session
*session
,
3109 struct ceph_cap
**target_cap
, int *old_issued
)
3110 __acquires(ci
->i_ceph_lock
)
3112 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3113 struct ceph_cap
*cap
, *ocap
, *new_cap
= NULL
;
3114 int mds
= session
->s_mds
;
3116 unsigned caps
= le32_to_cpu(im
->caps
);
3117 unsigned wanted
= le32_to_cpu(im
->wanted
);
3118 unsigned seq
= le32_to_cpu(im
->seq
);
3119 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
3120 u64 realmino
= le64_to_cpu(im
->realm
);
3121 u64 cap_id
= le64_to_cpu(im
->cap_id
);
3126 p_cap_id
= le64_to_cpu(ph
->cap_id
);
3127 peer
= le32_to_cpu(ph
->mds
);
3133 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3134 inode
, ci
, mds
, mseq
, peer
);
3137 spin_lock(&ci
->i_ceph_lock
);
3138 cap
= __get_cap_for_mds(ci
, mds
);
3141 spin_unlock(&ci
->i_ceph_lock
);
3142 new_cap
= ceph_get_cap(mdsc
, NULL
);
3148 ceph_put_cap(mdsc
, new_cap
);
3153 __ceph_caps_issued(ci
, &issued
);
3154 issued
|= __ceph_caps_dirty(ci
);
3156 ceph_add_cap(inode
, session
, cap_id
, -1, caps
, wanted
, seq
, mseq
,
3157 realmino
, CEPH_CAP_FLAG_AUTH
, &new_cap
);
3159 ocap
= peer
>= 0 ? __get_cap_for_mds(ci
, peer
) : NULL
;
3160 if (ocap
&& ocap
->cap_id
== p_cap_id
) {
3161 dout(" remove export cap %p mds%d flags %d\n",
3162 ocap
, peer
, ph
->flags
);
3163 if ((ph
->flags
& CEPH_CAP_FLAG_AUTH
) &&
3164 (ocap
->seq
!= le32_to_cpu(ph
->seq
) ||
3165 ocap
->mseq
!= le32_to_cpu(ph
->mseq
))) {
3166 pr_err("handle_cap_import: mismatched seq/mseq: "
3167 "ino (%llx.%llx) mds%d seq %d mseq %d "
3168 "importer mds%d has peer seq %d mseq %d\n",
3169 ceph_vinop(inode
), peer
, ocap
->seq
,
3170 ocap
->mseq
, mds
, le32_to_cpu(ph
->seq
),
3171 le32_to_cpu(ph
->mseq
));
3173 __ceph_remove_cap(ocap
, (ph
->flags
& CEPH_CAP_FLAG_RELEASE
));
3176 /* make sure we re-request max_size, if necessary */
3177 ci
->i_wanted_max_size
= 0;
3178 ci
->i_requested_max_size
= 0;
3180 *old_issued
= issued
;
3185 * Handle a caps message from the MDS.
3187 * Identify the appropriate session, inode, and call the right handler
3188 * based on the cap op.
3190 void ceph_handle_caps(struct ceph_mds_session
*session
,
3191 struct ceph_msg
*msg
)
3193 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
3194 struct super_block
*sb
= mdsc
->fsc
->sb
;
3195 struct inode
*inode
;
3196 struct ceph_inode_info
*ci
;
3197 struct ceph_cap
*cap
;
3198 struct ceph_mds_caps
*h
;
3199 struct ceph_mds_cap_peer
*peer
= NULL
;
3200 struct ceph_snap_realm
*realm
;
3201 int mds
= session
->s_mds
;
3204 struct ceph_vino vino
;
3208 u64 inline_version
= 0;
3209 void *inline_data
= NULL
;
3212 size_t snaptrace_len
;
3215 dout("handle_caps from mds%d\n", mds
);
3218 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
3219 tid
= le64_to_cpu(msg
->hdr
.tid
);
3220 if (msg
->front
.iov_len
< sizeof(*h
))
3222 h
= msg
->front
.iov_base
;
3223 op
= le32_to_cpu(h
->op
);
3224 vino
.ino
= le64_to_cpu(h
->ino
);
3225 vino
.snap
= CEPH_NOSNAP
;
3226 cap_id
= le64_to_cpu(h
->cap_id
);
3227 seq
= le32_to_cpu(h
->seq
);
3228 mseq
= le32_to_cpu(h
->migrate_seq
);
3229 size
= le64_to_cpu(h
->size
);
3230 max_size
= le64_to_cpu(h
->max_size
);
3233 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
3234 p
= snaptrace
+ snaptrace_len
;
3236 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
3238 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
3239 if (p
+ flock_len
> end
)
3244 if (le16_to_cpu(msg
->hdr
.version
) >= 3) {
3245 if (op
== CEPH_CAP_OP_IMPORT
) {
3246 if (p
+ sizeof(*peer
) > end
)
3250 } else if (op
== CEPH_CAP_OP_EXPORT
) {
3251 /* recorded in unused fields */
3252 peer
= (void *)&h
->size
;
3256 if (le16_to_cpu(msg
->hdr
.version
) >= 4) {
3257 ceph_decode_64_safe(&p
, end
, inline_version
, bad
);
3258 ceph_decode_32_safe(&p
, end
, inline_len
, bad
);
3259 if (p
+ inline_len
> end
)
3266 inode
= ceph_find_inode(sb
, vino
);
3267 ci
= ceph_inode(inode
);
3268 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
3271 mutex_lock(&session
->s_mutex
);
3273 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
3277 dout(" i don't have ino %llx\n", vino
.ino
);
3279 if (op
== CEPH_CAP_OP_IMPORT
) {
3280 cap
= ceph_get_cap(mdsc
, NULL
);
3281 cap
->cap_ino
= vino
.ino
;
3282 cap
->queue_release
= 1;
3283 cap
->cap_id
= cap_id
;
3286 spin_lock(&session
->s_cap_lock
);
3287 list_add_tail(&cap
->session_caps
,
3288 &session
->s_cap_releases
);
3289 session
->s_num_cap_releases
++;
3290 spin_unlock(&session
->s_cap_lock
);
3292 goto flush_cap_releases
;
3295 /* these will work even if we don't have a cap yet */
3297 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
3298 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
3301 case CEPH_CAP_OP_EXPORT
:
3302 handle_cap_export(inode
, h
, peer
, session
);
3305 case CEPH_CAP_OP_IMPORT
:
3307 if (snaptrace_len
) {
3308 down_write(&mdsc
->snap_rwsem
);
3309 ceph_update_snap_trace(mdsc
, snaptrace
,
3310 snaptrace
+ snaptrace_len
,
3312 downgrade_write(&mdsc
->snap_rwsem
);
3314 down_read(&mdsc
->snap_rwsem
);
3316 handle_cap_import(mdsc
, inode
, h
, peer
, session
,
3318 handle_cap_grant(mdsc
, inode
, h
,
3319 inline_version
, inline_data
, inline_len
,
3320 msg
->middle
, session
, cap
, issued
);
3322 ceph_put_snap_realm(mdsc
, realm
);
3326 /* the rest require a cap */
3327 spin_lock(&ci
->i_ceph_lock
);
3328 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
3330 dout(" no cap on %p ino %llx.%llx from mds%d\n",
3331 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
3332 spin_unlock(&ci
->i_ceph_lock
);
3333 goto flush_cap_releases
;
3336 /* note that each of these drops i_ceph_lock for us */
3338 case CEPH_CAP_OP_REVOKE
:
3339 case CEPH_CAP_OP_GRANT
:
3340 __ceph_caps_issued(ci
, &issued
);
3341 issued
|= __ceph_caps_dirty(ci
);
3342 handle_cap_grant(mdsc
, inode
, h
,
3343 inline_version
, inline_data
, inline_len
,
3344 msg
->middle
, session
, cap
, issued
);
3347 case CEPH_CAP_OP_FLUSH_ACK
:
3348 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
3351 case CEPH_CAP_OP_TRUNC
:
3352 handle_cap_trunc(inode
, h
, session
);
3356 spin_unlock(&ci
->i_ceph_lock
);
3357 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
3358 ceph_cap_op_name(op
));
3365 * send any cap release message to try to move things
3366 * along for the mds (who clearly thinks we still have this
3369 ceph_send_cap_releases(mdsc
, session
);
3372 mutex_unlock(&session
->s_mutex
);
3378 pr_err("ceph_handle_caps: corrupt message\n");
3384 * Delayed work handler to process end of delayed cap release LRU list.
3386 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
3388 struct ceph_inode_info
*ci
;
3389 int flags
= CHECK_CAPS_NODELAY
;
3391 dout("check_delayed_caps\n");
3393 spin_lock(&mdsc
->cap_delay_lock
);
3394 if (list_empty(&mdsc
->cap_delay_list
))
3396 ci
= list_first_entry(&mdsc
->cap_delay_list
,
3397 struct ceph_inode_info
,
3399 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
3400 time_before(jiffies
, ci
->i_hold_caps_max
))
3402 list_del_init(&ci
->i_cap_delay_list
);
3403 spin_unlock(&mdsc
->cap_delay_lock
);
3404 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
3405 ceph_check_caps(ci
, flags
, NULL
);
3407 spin_unlock(&mdsc
->cap_delay_lock
);
3411 * Flush all dirty caps to the mds
3413 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
3415 struct ceph_inode_info
*ci
;
3416 struct inode
*inode
;
3418 dout("flush_dirty_caps\n");
3419 spin_lock(&mdsc
->cap_dirty_lock
);
3420 while (!list_empty(&mdsc
->cap_dirty
)) {
3421 ci
= list_first_entry(&mdsc
->cap_dirty
, struct ceph_inode_info
,
3423 inode
= &ci
->vfs_inode
;
3425 dout("flush_dirty_caps %p\n", inode
);
3426 spin_unlock(&mdsc
->cap_dirty_lock
);
3427 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
, NULL
);
3429 spin_lock(&mdsc
->cap_dirty_lock
);
3431 spin_unlock(&mdsc
->cap_dirty_lock
);
3432 dout("flush_dirty_caps done\n");
3436 * Drop open file reference. If we were the last open file,
3437 * we may need to release capabilities to the MDS (or schedule
3438 * their delayed release).
3440 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
3442 struct inode
*inode
= &ci
->vfs_inode
;
3445 spin_lock(&ci
->i_ceph_lock
);
3446 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
3447 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
3448 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
3449 if (--ci
->i_nr_by_mode
[fmode
] == 0)
3451 spin_unlock(&ci
->i_ceph_lock
);
3453 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
3454 ceph_check_caps(ci
, 0, NULL
);
3458 * Helpers for embedding cap and dentry lease releases into mds
3461 * @force is used by dentry_release (below) to force inclusion of a
3462 * record for the directory inode, even when there aren't any caps to
3465 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
3466 int mds
, int drop
, int unless
, int force
)
3468 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3469 struct ceph_cap
*cap
;
3470 struct ceph_mds_request_release
*rel
= *p
;
3474 spin_lock(&ci
->i_ceph_lock
);
3475 used
= __ceph_caps_used(ci
);
3476 dirty
= __ceph_caps_dirty(ci
);
3478 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3479 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
3480 ceph_cap_string(unless
));
3482 /* only drop unused, clean caps */
3483 drop
&= ~(used
| dirty
);
3485 cap
= __get_cap_for_mds(ci
, mds
);
3486 if (cap
&& __cap_is_valid(cap
)) {
3488 ((cap
->issued
& drop
) &&
3489 (cap
->issued
& unless
) == 0)) {
3490 if ((cap
->issued
& drop
) &&
3491 (cap
->issued
& unless
) == 0) {
3492 int wanted
= __ceph_caps_wanted(ci
);
3493 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0)
3494 wanted
|= cap
->mds_wanted
;
3495 dout("encode_inode_release %p cap %p "
3496 "%s -> %s, wanted %s -> %s\n", inode
, cap
,
3497 ceph_cap_string(cap
->issued
),
3498 ceph_cap_string(cap
->issued
& ~drop
),
3499 ceph_cap_string(cap
->mds_wanted
),
3500 ceph_cap_string(wanted
));
3502 cap
->issued
&= ~drop
;
3503 cap
->implemented
&= ~drop
;
3504 cap
->mds_wanted
= wanted
;
3506 dout("encode_inode_release %p cap %p %s"
3507 " (force)\n", inode
, cap
,
3508 ceph_cap_string(cap
->issued
));
3511 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3512 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3513 rel
->seq
= cpu_to_le32(cap
->seq
);
3514 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
);
3515 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3516 rel
->caps
= cpu_to_le32(cap
->implemented
);
3517 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3523 dout("encode_inode_release %p cap %p %s\n",
3524 inode
, cap
, ceph_cap_string(cap
->issued
));
3527 spin_unlock(&ci
->i_ceph_lock
);
3531 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3532 int mds
, int drop
, int unless
)
3534 struct inode
*dir
= d_inode(dentry
->d_parent
);
3535 struct ceph_mds_request_release
*rel
= *p
;
3536 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3541 * force an record for the directory caps if we have a dentry lease.
3542 * this is racy (can't take i_ceph_lock and d_lock together), but it
3543 * doesn't have to be perfect; the mds will revoke anything we don't
3546 spin_lock(&dentry
->d_lock
);
3547 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3549 spin_unlock(&dentry
->d_lock
);
3551 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3553 spin_lock(&dentry
->d_lock
);
3554 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3555 dout("encode_dentry_release %p mds%d seq %d\n",
3556 dentry
, mds
, (int)di
->lease_seq
);
3557 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3558 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3559 *p
+= dentry
->d_name
.len
;
3560 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3561 __ceph_mdsc_drop_dentry_lease(dentry
);
3563 spin_unlock(&dentry
->d_lock
);