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 static struct ceph_cap
*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 int 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_reservation
*caps_reservation
)
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
*new_cap
= NULL
;
520 struct ceph_cap
*cap
;
521 int mds
= session
->s_mds
;
524 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
525 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
528 * If we are opening the file, include file mode wanted bits
532 wanted
|= ceph_caps_for_mode(fmode
);
535 spin_lock(&ci
->i_ceph_lock
);
536 cap
= __get_cap_for_mds(ci
, mds
);
542 spin_unlock(&ci
->i_ceph_lock
);
543 new_cap
= get_cap(mdsc
, caps_reservation
);
550 cap
->implemented
= 0;
556 __insert_cap_node(ci
, cap
);
558 /* clear out old exporting info? (i.e. on cap import) */
559 if (ci
->i_cap_exporting_mds
== mds
) {
560 ci
->i_cap_exporting_issued
= 0;
561 ci
->i_cap_exporting_mseq
= 0;
562 ci
->i_cap_exporting_mds
= -1;
565 /* add to session cap list */
566 cap
->session
= session
;
567 spin_lock(&session
->s_cap_lock
);
568 list_add_tail(&cap
->session_caps
, &session
->s_caps
);
569 session
->s_nr_caps
++;
570 spin_unlock(&session
->s_cap_lock
);
572 ceph_put_cap(mdsc
, new_cap
);
574 if (!ci
->i_snap_realm
) {
576 * add this inode to the appropriate snap realm
578 struct ceph_snap_realm
*realm
= ceph_lookup_snap_realm(mdsc
,
581 ceph_get_snap_realm(mdsc
, realm
);
582 spin_lock(&realm
->inodes_with_caps_lock
);
583 ci
->i_snap_realm
= realm
;
584 list_add(&ci
->i_snap_realm_item
,
585 &realm
->inodes_with_caps
);
586 spin_unlock(&realm
->inodes_with_caps_lock
);
588 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
594 __check_cap_issue(ci
, cap
, issued
);
597 * If we are issued caps we don't want, or the mds' wanted
598 * value appears to be off, queue a check so we'll release
599 * later and/or update the mds wanted value.
601 actual_wanted
= __ceph_caps_wanted(ci
);
602 if ((wanted
& ~actual_wanted
) ||
603 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
604 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
605 ceph_cap_string(issued
), ceph_cap_string(wanted
),
606 ceph_cap_string(actual_wanted
));
607 __cap_delay_requeue(mdsc
, ci
);
610 if (flags
& CEPH_CAP_FLAG_AUTH
) {
611 if (ci
->i_auth_cap
== NULL
||
612 ceph_seq_cmp(ci
->i_auth_cap
->mseq
, mseq
) < 0)
613 ci
->i_auth_cap
= cap
;
614 ci
->i_cap_exporting_issued
= 0;
615 } else if (ci
->i_auth_cap
== cap
) {
616 ci
->i_auth_cap
= NULL
;
617 spin_lock(&mdsc
->cap_dirty_lock
);
618 if (!list_empty(&ci
->i_dirty_item
)) {
619 dout(" moving %p to cap_dirty_migrating\n", inode
);
620 list_move(&ci
->i_dirty_item
,
621 &mdsc
->cap_dirty_migrating
);
623 spin_unlock(&mdsc
->cap_dirty_lock
);
626 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
627 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
628 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
629 cap
->cap_id
= cap_id
;
630 cap
->issued
= issued
;
631 cap
->implemented
|= issued
;
632 if (ceph_seq_cmp(mseq
, cap
->mseq
) > 0)
633 cap
->mds_wanted
= wanted
;
635 cap
->mds_wanted
|= wanted
;
637 cap
->issue_seq
= seq
;
639 cap
->cap_gen
= session
->s_cap_gen
;
642 __ceph_get_fmode(ci
, fmode
);
643 spin_unlock(&ci
->i_ceph_lock
);
644 wake_up_all(&ci
->i_cap_wq
);
649 * Return true if cap has not timed out and belongs to the current
650 * generation of the MDS session (i.e. has not gone 'stale' due to
651 * us losing touch with the mds).
653 static int __cap_is_valid(struct ceph_cap
*cap
)
658 spin_lock(&cap
->session
->s_gen_ttl_lock
);
659 gen
= cap
->session
->s_cap_gen
;
660 ttl
= cap
->session
->s_cap_ttl
;
661 spin_unlock(&cap
->session
->s_gen_ttl_lock
);
663 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
664 dout("__cap_is_valid %p cap %p issued %s "
665 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
666 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
674 * Return set of valid cap bits issued to us. Note that caps time
675 * out, and may be invalidated in bulk if the client session times out
676 * and session->s_cap_gen is bumped.
678 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
680 int have
= ci
->i_snap_caps
| ci
->i_cap_exporting_issued
;
681 struct ceph_cap
*cap
;
686 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
687 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
688 if (!__cap_is_valid(cap
))
690 dout("__ceph_caps_issued %p cap %p issued %s\n",
691 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
694 *implemented
|= cap
->implemented
;
697 * exclude caps issued by non-auth MDS, but are been revoking
698 * by the auth MDS. The non-auth MDS should be revoking/exporting
699 * these caps, but the message is delayed.
701 if (ci
->i_auth_cap
) {
702 cap
= ci
->i_auth_cap
;
703 have
&= ~cap
->implemented
| cap
->issued
;
709 * Get cap bits issued by caps other than @ocap
711 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
713 int have
= ci
->i_snap_caps
;
714 struct ceph_cap
*cap
;
717 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
718 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
721 if (!__cap_is_valid(cap
))
729 * Move a cap to the end of the LRU (oldest caps at list head, newest
732 static void __touch_cap(struct ceph_cap
*cap
)
734 struct ceph_mds_session
*s
= cap
->session
;
736 spin_lock(&s
->s_cap_lock
);
737 if (s
->s_cap_iterator
== NULL
) {
738 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
740 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
742 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
743 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
745 spin_unlock(&s
->s_cap_lock
);
749 * Check if we hold the given mask. If so, move the cap(s) to the
750 * front of their respective LRUs. (This is the preferred way for
751 * callers to check for caps they want.)
753 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
755 struct ceph_cap
*cap
;
757 int have
= ci
->i_snap_caps
;
759 if ((have
& mask
) == mask
) {
760 dout("__ceph_caps_issued_mask %p snap issued %s"
761 " (mask %s)\n", &ci
->vfs_inode
,
762 ceph_cap_string(have
),
763 ceph_cap_string(mask
));
767 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
768 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
769 if (!__cap_is_valid(cap
))
771 if ((cap
->issued
& mask
) == mask
) {
772 dout("__ceph_caps_issued_mask %p cap %p issued %s"
773 " (mask %s)\n", &ci
->vfs_inode
, cap
,
774 ceph_cap_string(cap
->issued
),
775 ceph_cap_string(mask
));
781 /* does a combination of caps satisfy mask? */
783 if ((have
& mask
) == mask
) {
784 dout("__ceph_caps_issued_mask %p combo issued %s"
785 " (mask %s)\n", &ci
->vfs_inode
,
786 ceph_cap_string(cap
->issued
),
787 ceph_cap_string(mask
));
791 /* touch this + preceding caps */
793 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
795 cap
= rb_entry(q
, struct ceph_cap
,
797 if (!__cap_is_valid(cap
))
810 * Return true if mask caps are currently being revoked by an MDS.
812 int __ceph_caps_revoking_other(struct ceph_inode_info
*ci
,
813 struct ceph_cap
*ocap
, int mask
)
815 struct ceph_cap
*cap
;
818 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
819 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
821 (cap
->implemented
& ~cap
->issued
& mask
))
827 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
829 struct inode
*inode
= &ci
->vfs_inode
;
832 spin_lock(&ci
->i_ceph_lock
);
833 ret
= __ceph_caps_revoking_other(ci
, NULL
, mask
);
834 spin_unlock(&ci
->i_ceph_lock
);
835 dout("ceph_caps_revoking %p %s = %d\n", inode
,
836 ceph_cap_string(mask
), ret
);
840 int __ceph_caps_used(struct ceph_inode_info
*ci
)
844 used
|= CEPH_CAP_PIN
;
846 used
|= CEPH_CAP_FILE_RD
;
847 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
848 used
|= CEPH_CAP_FILE_CACHE
;
850 used
|= CEPH_CAP_FILE_WR
;
851 if (ci
->i_wb_ref
|| ci
->i_wrbuffer_ref
)
852 used
|= CEPH_CAP_FILE_BUFFER
;
857 * wanted, by virtue of open file modes
859 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
863 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
864 if (ci
->i_nr_by_mode
[mode
])
865 want
|= ceph_caps_for_mode(mode
);
870 * Return caps we have registered with the MDS(s) as 'wanted'.
872 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
874 struct ceph_cap
*cap
;
878 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
879 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
880 if (!__cap_is_valid(cap
))
882 mds_wanted
|= cap
->mds_wanted
;
888 * called under i_ceph_lock
890 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
892 return !RB_EMPTY_ROOT(&ci
->i_caps
) || ci
->i_cap_exporting_mds
>= 0;
895 int ceph_is_any_caps(struct inode
*inode
)
897 struct ceph_inode_info
*ci
= ceph_inode(inode
);
900 spin_lock(&ci
->i_ceph_lock
);
901 ret
= __ceph_is_any_caps(ci
);
902 spin_unlock(&ci
->i_ceph_lock
);
908 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
910 * caller should hold i_ceph_lock.
911 * caller will not hold session s_mutex if called from destroy_inode.
913 void __ceph_remove_cap(struct ceph_cap
*cap
, bool queue_release
)
915 struct ceph_mds_session
*session
= cap
->session
;
916 struct ceph_inode_info
*ci
= cap
->ci
;
917 struct ceph_mds_client
*mdsc
=
918 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
921 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
923 /* remove from session list */
924 spin_lock(&session
->s_cap_lock
);
926 * s_cap_reconnect is protected by s_cap_lock. no one changes
927 * s_cap_gen while session is in the reconnect state.
930 (!session
->s_cap_reconnect
||
931 cap
->cap_gen
== session
->s_cap_gen
))
932 __queue_cap_release(session
, ci
->i_vino
.ino
, cap
->cap_id
,
933 cap
->mseq
, cap
->issue_seq
);
935 if (session
->s_cap_iterator
== cap
) {
936 /* not yet, we are iterating over this very cap */
937 dout("__ceph_remove_cap delaying %p removal from session %p\n",
940 list_del_init(&cap
->session_caps
);
941 session
->s_nr_caps
--;
945 /* protect backpointer with s_cap_lock: see iterate_session_caps */
947 spin_unlock(&session
->s_cap_lock
);
949 /* remove from inode list */
950 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
951 if (ci
->i_auth_cap
== cap
)
952 ci
->i_auth_cap
= NULL
;
955 ceph_put_cap(mdsc
, cap
);
957 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
) {
958 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
959 spin_lock(&realm
->inodes_with_caps_lock
);
960 list_del_init(&ci
->i_snap_realm_item
);
961 ci
->i_snap_realm_counter
++;
962 ci
->i_snap_realm
= NULL
;
963 spin_unlock(&realm
->inodes_with_caps_lock
);
964 ceph_put_snap_realm(mdsc
, realm
);
966 if (!__ceph_is_any_real_caps(ci
))
967 __cap_delay_cancel(mdsc
, ci
);
971 * Build and send a cap message to the given MDS.
973 * Caller should be holding s_mutex.
975 static int send_cap_msg(struct ceph_mds_session
*session
,
976 u64 ino
, u64 cid
, int op
,
977 int caps
, int wanted
, int dirty
,
978 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
979 u64 size
, u64 max_size
,
980 struct timespec
*mtime
, struct timespec
*atime
,
982 kuid_t uid
, kgid_t gid
, umode_t mode
,
984 struct ceph_buffer
*xattrs_buf
,
987 struct ceph_mds_caps
*fc
;
988 struct ceph_msg
*msg
;
990 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
991 " seq %u/%u mseq %u follows %lld size %llu/%llu"
992 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
993 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
994 ceph_cap_string(dirty
),
995 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
996 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
998 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
), GFP_NOFS
, false);
1002 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
1004 fc
= msg
->front
.iov_base
;
1005 memset(fc
, 0, sizeof(*fc
));
1007 fc
->cap_id
= cpu_to_le64(cid
);
1008 fc
->op
= cpu_to_le32(op
);
1009 fc
->seq
= cpu_to_le32(seq
);
1010 fc
->issue_seq
= cpu_to_le32(issue_seq
);
1011 fc
->migrate_seq
= cpu_to_le32(mseq
);
1012 fc
->caps
= cpu_to_le32(caps
);
1013 fc
->wanted
= cpu_to_le32(wanted
);
1014 fc
->dirty
= cpu_to_le32(dirty
);
1015 fc
->ino
= cpu_to_le64(ino
);
1016 fc
->snap_follows
= cpu_to_le64(follows
);
1018 fc
->size
= cpu_to_le64(size
);
1019 fc
->max_size
= cpu_to_le64(max_size
);
1021 ceph_encode_timespec(&fc
->mtime
, mtime
);
1023 ceph_encode_timespec(&fc
->atime
, atime
);
1024 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
1026 fc
->uid
= cpu_to_le32(from_kuid(&init_user_ns
, uid
));
1027 fc
->gid
= cpu_to_le32(from_kgid(&init_user_ns
, gid
));
1028 fc
->mode
= cpu_to_le32(mode
);
1030 fc
->xattr_version
= cpu_to_le64(xattr_version
);
1032 msg
->middle
= ceph_buffer_get(xattrs_buf
);
1033 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1034 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1037 ceph_con_send(&session
->s_con
, msg
);
1041 void __queue_cap_release(struct ceph_mds_session
*session
,
1042 u64 ino
, u64 cap_id
, u32 migrate_seq
,
1045 struct ceph_msg
*msg
;
1046 struct ceph_mds_cap_release
*head
;
1047 struct ceph_mds_cap_item
*item
;
1049 BUG_ON(!session
->s_num_cap_releases
);
1050 msg
= list_first_entry(&session
->s_cap_releases
,
1051 struct ceph_msg
, list_head
);
1053 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1054 ino
, session
->s_mds
, msg
, session
->s_num_cap_releases
);
1056 BUG_ON(msg
->front
.iov_len
+ sizeof(*item
) > PAGE_CACHE_SIZE
);
1057 head
= msg
->front
.iov_base
;
1058 le32_add_cpu(&head
->num
, 1);
1059 item
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1060 item
->ino
= cpu_to_le64(ino
);
1061 item
->cap_id
= cpu_to_le64(cap_id
);
1062 item
->migrate_seq
= cpu_to_le32(migrate_seq
);
1063 item
->seq
= cpu_to_le32(issue_seq
);
1065 session
->s_num_cap_releases
--;
1067 msg
->front
.iov_len
+= sizeof(*item
);
1068 if (le32_to_cpu(head
->num
) == CEPH_CAPS_PER_RELEASE
) {
1069 dout(" release msg %p full\n", msg
);
1070 list_move_tail(&msg
->list_head
, &session
->s_cap_releases_done
);
1072 dout(" release msg %p at %d/%d (%d)\n", msg
,
1073 (int)le32_to_cpu(head
->num
),
1074 (int)CEPH_CAPS_PER_RELEASE
,
1075 (int)msg
->front
.iov_len
);
1080 * Queue cap releases when an inode is dropped from our cache. Since
1081 * inode is about to be destroyed, there is no need for i_ceph_lock.
1083 void ceph_queue_caps_release(struct inode
*inode
)
1085 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1088 p
= rb_first(&ci
->i_caps
);
1090 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1092 __ceph_remove_cap(cap
, true);
1097 * Send a cap msg on the given inode. Update our caps state, then
1098 * drop i_ceph_lock and send the message.
1100 * Make note of max_size reported/requested from mds, revoked caps
1101 * that have now been implemented.
1103 * Make half-hearted attempt ot to invalidate page cache if we are
1104 * dropping RDCACHE. Note that this will leave behind locked pages
1105 * that we'll then need to deal with elsewhere.
1107 * Return non-zero if delayed release, or we experienced an error
1108 * such that the caller should requeue + retry later.
1110 * called with i_ceph_lock, then drops it.
1111 * caller should hold snap_rwsem (read), s_mutex.
1113 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1114 int op
, int used
, int want
, int retain
, int flushing
,
1115 unsigned *pflush_tid
)
1116 __releases(cap
->ci
->i_ceph_lock
)
1118 struct ceph_inode_info
*ci
= cap
->ci
;
1119 struct inode
*inode
= &ci
->vfs_inode
;
1120 u64 cap_id
= cap
->cap_id
;
1121 int held
, revoking
, dropping
, keep
;
1122 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1124 struct timespec mtime
, atime
;
1129 struct ceph_mds_session
*session
;
1130 u64 xattr_version
= 0;
1131 struct ceph_buffer
*xattr_blob
= NULL
;
1137 held
= cap
->issued
| cap
->implemented
;
1138 revoking
= cap
->implemented
& ~cap
->issued
;
1139 retain
&= ~revoking
;
1140 dropping
= cap
->issued
& ~retain
;
1142 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1143 inode
, cap
, cap
->session
,
1144 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1145 ceph_cap_string(revoking
));
1146 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1148 session
= cap
->session
;
1150 /* don't release wanted unless we've waited a bit. */
1151 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1152 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1153 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1154 ceph_cap_string(cap
->issued
),
1155 ceph_cap_string(cap
->issued
& retain
),
1156 ceph_cap_string(cap
->mds_wanted
),
1157 ceph_cap_string(want
));
1158 want
|= cap
->mds_wanted
;
1159 retain
|= cap
->issued
;
1162 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1164 cap
->issued
&= retain
; /* drop bits we don't want */
1165 if (cap
->implemented
& ~cap
->issued
) {
1167 * Wake up any waiters on wanted -> needed transition.
1168 * This is due to the weird transition from buffered
1169 * to sync IO... we need to flush dirty pages _before_
1170 * allowing sync writes to avoid reordering.
1174 cap
->implemented
&= cap
->issued
| used
;
1175 cap
->mds_wanted
= want
;
1179 * assign a tid for flush operations so we can avoid
1180 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1181 * clean type races. track latest tid for every bit
1182 * so we can handle flush AxFw, flush Fw, and have the
1183 * first ack clean Ax.
1185 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1187 *pflush_tid
= flush_tid
;
1188 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1189 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1190 if (flushing
& (1 << i
))
1191 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1193 follows
= ci
->i_head_snapc
->seq
;
1198 keep
= cap
->implemented
;
1200 issue_seq
= cap
->issue_seq
;
1202 size
= inode
->i_size
;
1203 ci
->i_reported_size
= size
;
1204 max_size
= ci
->i_wanted_max_size
;
1205 ci
->i_requested_max_size
= max_size
;
1206 mtime
= inode
->i_mtime
;
1207 atime
= inode
->i_atime
;
1208 time_warp_seq
= ci
->i_time_warp_seq
;
1211 mode
= inode
->i_mode
;
1213 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1214 __ceph_build_xattrs_blob(ci
);
1215 xattr_blob
= ci
->i_xattrs
.blob
;
1216 xattr_version
= ci
->i_xattrs
.version
;
1219 spin_unlock(&ci
->i_ceph_lock
);
1221 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1222 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1223 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1224 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1227 dout("error sending cap msg, must requeue %p\n", inode
);
1232 wake_up_all(&ci
->i_cap_wq
);
1238 * When a snapshot is taken, clients accumulate dirty metadata on
1239 * inodes with capabilities in ceph_cap_snaps to describe the file
1240 * state at the time the snapshot was taken. This must be flushed
1241 * asynchronously back to the MDS once sync writes complete and dirty
1242 * data is written out.
1244 * Unless @again is true, skip cap_snaps that were already sent to
1245 * the MDS (i.e., during this session).
1247 * Called under i_ceph_lock. Takes s_mutex as needed.
1249 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1250 struct ceph_mds_session
**psession
,
1252 __releases(ci
->i_ceph_lock
)
1253 __acquires(ci
->i_ceph_lock
)
1255 struct inode
*inode
= &ci
->vfs_inode
;
1257 struct ceph_cap_snap
*capsnap
;
1259 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1260 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1262 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1263 i_cap_snaps list, and skip these entries next time
1264 around to avoid an infinite loop */
1267 session
= *psession
;
1269 dout("__flush_snaps %p\n", inode
);
1271 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1272 /* avoid an infiniute loop after retry */
1273 if (capsnap
->follows
< next_follows
)
1276 * we need to wait for sync writes to complete and for dirty
1277 * pages to be written out.
1279 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1283 * if cap writeback already occurred, we should have dropped
1284 * the capsnap in ceph_put_wrbuffer_cap_refs.
1286 BUG_ON(capsnap
->dirty
== 0);
1288 /* pick mds, take s_mutex */
1289 if (ci
->i_auth_cap
== NULL
) {
1290 dout("no auth cap (migrating?), doing nothing\n");
1294 /* only flush each capsnap once */
1295 if (!again
&& !list_empty(&capsnap
->flushing_item
)) {
1296 dout("already flushed %p, skipping\n", capsnap
);
1300 mds
= ci
->i_auth_cap
->session
->s_mds
;
1301 mseq
= ci
->i_auth_cap
->mseq
;
1303 if (session
&& session
->s_mds
!= mds
) {
1304 dout("oops, wrong session %p mutex\n", session
);
1305 mutex_unlock(&session
->s_mutex
);
1306 ceph_put_mds_session(session
);
1310 spin_unlock(&ci
->i_ceph_lock
);
1311 mutex_lock(&mdsc
->mutex
);
1312 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1313 mutex_unlock(&mdsc
->mutex
);
1315 dout("inverting session/ino locks on %p\n",
1317 mutex_lock(&session
->s_mutex
);
1320 * if session == NULL, we raced against a cap
1321 * deletion or migration. retry, and we'll
1322 * get a better @mds value next time.
1324 spin_lock(&ci
->i_ceph_lock
);
1328 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1329 atomic_inc(&capsnap
->nref
);
1330 if (!list_empty(&capsnap
->flushing_item
))
1331 list_del_init(&capsnap
->flushing_item
);
1332 list_add_tail(&capsnap
->flushing_item
,
1333 &session
->s_cap_snaps_flushing
);
1334 spin_unlock(&ci
->i_ceph_lock
);
1336 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1337 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1338 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1339 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1340 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1342 &capsnap
->mtime
, &capsnap
->atime
,
1343 capsnap
->time_warp_seq
,
1344 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1345 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1348 next_follows
= capsnap
->follows
+ 1;
1349 ceph_put_cap_snap(capsnap
);
1351 spin_lock(&ci
->i_ceph_lock
);
1355 /* we flushed them all; remove this inode from the queue */
1356 spin_lock(&mdsc
->snap_flush_lock
);
1357 list_del_init(&ci
->i_snap_flush_item
);
1358 spin_unlock(&mdsc
->snap_flush_lock
);
1362 *psession
= session
;
1364 mutex_unlock(&session
->s_mutex
);
1365 ceph_put_mds_session(session
);
1369 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1371 spin_lock(&ci
->i_ceph_lock
);
1372 __ceph_flush_snaps(ci
, NULL
, 0);
1373 spin_unlock(&ci
->i_ceph_lock
);
1377 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1378 * Caller is then responsible for calling __mark_inode_dirty with the
1379 * returned flags value.
1381 int __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1383 struct ceph_mds_client
*mdsc
=
1384 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1385 struct inode
*inode
= &ci
->vfs_inode
;
1386 int was
= ci
->i_dirty_caps
;
1389 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1390 ceph_cap_string(mask
), ceph_cap_string(was
),
1391 ceph_cap_string(was
| mask
));
1392 ci
->i_dirty_caps
|= mask
;
1394 if (!ci
->i_head_snapc
)
1395 ci
->i_head_snapc
= ceph_get_snap_context(
1396 ci
->i_snap_realm
->cached_context
);
1397 dout(" inode %p now dirty snapc %p auth cap %p\n",
1398 &ci
->vfs_inode
, ci
->i_head_snapc
, ci
->i_auth_cap
);
1399 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1400 spin_lock(&mdsc
->cap_dirty_lock
);
1402 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1404 list_add(&ci
->i_dirty_item
,
1405 &mdsc
->cap_dirty_migrating
);
1406 spin_unlock(&mdsc
->cap_dirty_lock
);
1407 if (ci
->i_flushing_caps
== 0) {
1409 dirty
|= I_DIRTY_SYNC
;
1412 BUG_ON(list_empty(&ci
->i_dirty_item
));
1413 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1414 (mask
& CEPH_CAP_FILE_BUFFER
))
1415 dirty
|= I_DIRTY_DATASYNC
;
1416 __cap_delay_requeue(mdsc
, ci
);
1421 * Add dirty inode to the flushing list. Assigned a seq number so we
1422 * can wait for caps to flush without starving.
1424 * Called under i_ceph_lock.
1426 static int __mark_caps_flushing(struct inode
*inode
,
1427 struct ceph_mds_session
*session
)
1429 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1430 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1433 BUG_ON(ci
->i_dirty_caps
== 0);
1434 BUG_ON(list_empty(&ci
->i_dirty_item
));
1436 flushing
= ci
->i_dirty_caps
;
1437 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1438 ceph_cap_string(flushing
),
1439 ceph_cap_string(ci
->i_flushing_caps
),
1440 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1441 ci
->i_flushing_caps
|= flushing
;
1442 ci
->i_dirty_caps
= 0;
1443 dout(" inode %p now !dirty\n", inode
);
1445 spin_lock(&mdsc
->cap_dirty_lock
);
1446 list_del_init(&ci
->i_dirty_item
);
1448 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1449 if (list_empty(&ci
->i_flushing_item
)) {
1450 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1451 mdsc
->num_cap_flushing
++;
1452 dout(" inode %p now flushing seq %lld\n", inode
,
1453 ci
->i_cap_flush_seq
);
1455 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1456 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1457 ci
->i_cap_flush_seq
);
1459 spin_unlock(&mdsc
->cap_dirty_lock
);
1465 * try to invalidate mapping pages without blocking.
1467 static int try_nonblocking_invalidate(struct inode
*inode
)
1469 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1470 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1472 spin_unlock(&ci
->i_ceph_lock
);
1473 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1474 spin_lock(&ci
->i_ceph_lock
);
1476 if (inode
->i_data
.nrpages
== 0 &&
1477 invalidating_gen
== ci
->i_rdcache_gen
) {
1479 dout("try_nonblocking_invalidate %p success\n", inode
);
1480 /* save any racing async invalidate some trouble */
1481 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
- 1;
1484 dout("try_nonblocking_invalidate %p failed\n", inode
);
1489 * Swiss army knife function to examine currently used and wanted
1490 * versus held caps. Release, flush, ack revoked caps to mds as
1493 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1494 * cap release further.
1495 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1496 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1499 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1500 struct ceph_mds_session
*session
)
1502 struct ceph_fs_client
*fsc
= ceph_inode_to_client(&ci
->vfs_inode
);
1503 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1504 struct inode
*inode
= &ci
->vfs_inode
;
1505 struct ceph_cap
*cap
;
1506 int file_wanted
, used
, cap_used
;
1507 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1508 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1509 int mds
= -1; /* keep track of how far we've gone through i_caps list
1510 to avoid an infinite loop on retry */
1512 int tried_invalidate
= 0;
1513 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1514 int queue_invalidate
= 0;
1515 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1517 /* if we are unmounting, flush any unused caps immediately. */
1521 spin_lock(&ci
->i_ceph_lock
);
1523 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1524 flags
|= CHECK_CAPS_FLUSH
;
1526 /* flush snaps first time around only */
1527 if (!list_empty(&ci
->i_cap_snaps
))
1528 __ceph_flush_snaps(ci
, &session
, 0);
1531 spin_lock(&ci
->i_ceph_lock
);
1533 file_wanted
= __ceph_caps_file_wanted(ci
);
1534 used
= __ceph_caps_used(ci
);
1535 want
= file_wanted
| used
;
1536 issued
= __ceph_caps_issued(ci
, &implemented
);
1537 revoking
= implemented
& ~issued
;
1539 retain
= want
| CEPH_CAP_PIN
;
1540 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1542 retain
|= CEPH_CAP_ANY
; /* be greedy */
1544 retain
|= CEPH_CAP_ANY_SHARED
;
1546 * keep RD only if we didn't have the file open RW,
1547 * because then the mds would revoke it anyway to
1548 * journal max_size=0.
1550 if (ci
->i_max_size
== 0)
1551 retain
|= CEPH_CAP_ANY_RD
;
1555 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1556 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1557 ceph_cap_string(file_wanted
),
1558 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1559 ceph_cap_string(ci
->i_flushing_caps
),
1560 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1561 ceph_cap_string(retain
),
1562 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1563 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1564 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1567 * If we no longer need to hold onto old our caps, and we may
1568 * have cached pages, but don't want them, then try to invalidate.
1569 * If we fail, it's because pages are locked.... try again later.
1571 if ((!is_delayed
|| mdsc
->stopping
) &&
1572 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1573 inode
->i_data
.nrpages
&& /* have cached pages */
1574 (file_wanted
== 0 || /* no open files */
1575 (revoking
& (CEPH_CAP_FILE_CACHE
|
1576 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1577 !tried_invalidate
) {
1578 dout("check_caps trying to invalidate on %p\n", inode
);
1579 if (try_nonblocking_invalidate(inode
) < 0) {
1580 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1581 CEPH_CAP_FILE_LAZYIO
)) {
1582 dout("check_caps queuing invalidate\n");
1583 queue_invalidate
= 1;
1584 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1586 dout("check_caps failed to invalidate pages\n");
1587 /* we failed to invalidate pages. check these
1588 caps again later. */
1590 __cap_set_timeouts(mdsc
, ci
);
1593 tried_invalidate
= 1;
1598 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1599 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1602 /* avoid looping forever */
1603 if (mds
>= cap
->mds
||
1604 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1607 /* NOTE: no side-effects allowed, until we take s_mutex */
1610 if (ci
->i_auth_cap
&& cap
!= ci
->i_auth_cap
)
1611 cap_used
&= ~ci
->i_auth_cap
->issued
;
1613 revoking
= cap
->implemented
& ~cap
->issued
;
1614 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1615 cap
->mds
, cap
, ceph_cap_string(cap
->issued
),
1616 ceph_cap_string(cap_used
),
1617 ceph_cap_string(cap
->implemented
),
1618 ceph_cap_string(revoking
));
1620 if (cap
== ci
->i_auth_cap
&&
1621 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1622 /* request larger max_size from MDS? */
1623 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1624 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1625 dout("requesting new max_size\n");
1629 /* approaching file_max? */
1630 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1631 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1632 dout("i_size approaching max_size\n");
1636 /* flush anything dirty? */
1637 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1639 dout("flushing dirty caps\n");
1643 /* completed revocation? going down and there are no caps? */
1644 if (revoking
&& (revoking
& cap_used
) == 0) {
1645 dout("completed revocation of %s\n",
1646 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1650 /* want more caps from mds? */
1651 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1654 /* things we might delay */
1655 if ((cap
->issued
& ~retain
) == 0 &&
1656 cap
->mds_wanted
== want
)
1657 continue; /* nope, all good */
1663 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1664 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1665 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1666 ceph_cap_string(cap
->issued
),
1667 ceph_cap_string(cap
->issued
& retain
),
1668 ceph_cap_string(cap
->mds_wanted
),
1669 ceph_cap_string(want
));
1675 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1676 dout(" skipping %p I_NOFLUSH set\n", inode
);
1680 if (session
&& session
!= cap
->session
) {
1681 dout("oops, wrong session %p mutex\n", session
);
1682 mutex_unlock(&session
->s_mutex
);
1686 session
= cap
->session
;
1687 if (mutex_trylock(&session
->s_mutex
) == 0) {
1688 dout("inverting session/ino locks on %p\n",
1690 spin_unlock(&ci
->i_ceph_lock
);
1691 if (took_snap_rwsem
) {
1692 up_read(&mdsc
->snap_rwsem
);
1693 took_snap_rwsem
= 0;
1695 mutex_lock(&session
->s_mutex
);
1699 /* take snap_rwsem after session mutex */
1700 if (!took_snap_rwsem
) {
1701 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1702 dout("inverting snap/in locks on %p\n",
1704 spin_unlock(&ci
->i_ceph_lock
);
1705 down_read(&mdsc
->snap_rwsem
);
1706 took_snap_rwsem
= 1;
1709 took_snap_rwsem
= 1;
1712 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1713 flushing
= __mark_caps_flushing(inode
, session
);
1717 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1720 /* __send_cap drops i_ceph_lock */
1721 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, cap_used
,
1722 want
, retain
, flushing
, NULL
);
1723 goto retry
; /* retake i_ceph_lock and restart our cap scan. */
1727 * Reschedule delayed caps release if we delayed anything,
1730 if (delayed
&& is_delayed
)
1731 force_requeue
= 1; /* __send_cap delayed release; requeue */
1732 if (!delayed
&& !is_delayed
)
1733 __cap_delay_cancel(mdsc
, ci
);
1734 else if (!is_delayed
|| force_requeue
)
1735 __cap_delay_requeue(mdsc
, ci
);
1737 spin_unlock(&ci
->i_ceph_lock
);
1739 if (queue_invalidate
)
1740 ceph_queue_invalidate(inode
);
1743 mutex_unlock(&session
->s_mutex
);
1744 if (took_snap_rwsem
)
1745 up_read(&mdsc
->snap_rwsem
);
1749 * Try to flush dirty caps back to the auth mds.
1751 static int try_flush_caps(struct inode
*inode
, unsigned *flush_tid
)
1753 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1754 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1756 struct ceph_mds_session
*session
= NULL
;
1759 spin_lock(&ci
->i_ceph_lock
);
1760 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1761 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1764 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1765 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1766 int used
= __ceph_caps_used(ci
);
1767 int want
= __ceph_caps_wanted(ci
);
1770 if (!session
|| session
!= cap
->session
) {
1771 spin_unlock(&ci
->i_ceph_lock
);
1773 mutex_unlock(&session
->s_mutex
);
1774 session
= cap
->session
;
1775 mutex_lock(&session
->s_mutex
);
1778 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1781 flushing
= __mark_caps_flushing(inode
, session
);
1783 /* __send_cap drops i_ceph_lock */
1784 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1785 cap
->issued
| cap
->implemented
, flushing
,
1790 spin_lock(&ci
->i_ceph_lock
);
1791 __cap_delay_requeue(mdsc
, ci
);
1794 spin_unlock(&ci
->i_ceph_lock
);
1797 mutex_unlock(&session
->s_mutex
);
1802 * Return true if we've flushed caps through the given flush_tid.
1804 static int caps_are_flushed(struct inode
*inode
, unsigned tid
)
1806 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1809 spin_lock(&ci
->i_ceph_lock
);
1810 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1811 if ((ci
->i_flushing_caps
& (1 << i
)) &&
1812 ci
->i_cap_flush_tid
[i
] <= tid
) {
1813 /* still flushing this bit */
1817 spin_unlock(&ci
->i_ceph_lock
);
1822 * Wait on any unsafe replies for the given inode. First wait on the
1823 * newest request, and make that the upper bound. Then, if there are
1824 * more requests, keep waiting on the oldest as long as it is still older
1825 * than the original request.
1827 static void sync_write_wait(struct inode
*inode
)
1829 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1830 struct list_head
*head
= &ci
->i_unsafe_writes
;
1831 struct ceph_osd_request
*req
;
1834 spin_lock(&ci
->i_unsafe_lock
);
1835 if (list_empty(head
))
1838 /* set upper bound as _last_ entry in chain */
1839 req
= list_entry(head
->prev
, struct ceph_osd_request
,
1841 last_tid
= req
->r_tid
;
1844 ceph_osdc_get_request(req
);
1845 spin_unlock(&ci
->i_unsafe_lock
);
1846 dout("sync_write_wait on tid %llu (until %llu)\n",
1847 req
->r_tid
, last_tid
);
1848 wait_for_completion(&req
->r_safe_completion
);
1849 spin_lock(&ci
->i_unsafe_lock
);
1850 ceph_osdc_put_request(req
);
1853 * from here on look at first entry in chain, since we
1854 * only want to wait for anything older than last_tid
1856 if (list_empty(head
))
1858 req
= list_entry(head
->next
, struct ceph_osd_request
,
1860 } while (req
->r_tid
< last_tid
);
1862 spin_unlock(&ci
->i_unsafe_lock
);
1865 int ceph_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1867 struct inode
*inode
= file
->f_mapping
->host
;
1868 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1873 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1874 sync_write_wait(inode
);
1876 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1879 mutex_lock(&inode
->i_mutex
);
1881 dirty
= try_flush_caps(inode
, &flush_tid
);
1882 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1885 * only wait on non-file metadata writeback (the mds
1886 * can recover size and mtime, so we don't need to
1889 if (!datasync
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1890 dout("fsync waiting for flush_tid %u\n", flush_tid
);
1891 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1892 caps_are_flushed(inode
, flush_tid
));
1895 dout("fsync %p%s done\n", inode
, datasync
? " datasync" : "");
1896 mutex_unlock(&inode
->i_mutex
);
1901 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1902 * queue inode for flush but don't do so immediately, because we can
1903 * get by with fewer MDS messages if we wait for data writeback to
1906 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1908 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1912 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1914 dout("write_inode %p wait=%d\n", inode
, wait
);
1916 dirty
= try_flush_caps(inode
, &flush_tid
);
1918 err
= wait_event_interruptible(ci
->i_cap_wq
,
1919 caps_are_flushed(inode
, flush_tid
));
1921 struct ceph_mds_client
*mdsc
=
1922 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1924 spin_lock(&ci
->i_ceph_lock
);
1925 if (__ceph_caps_dirty(ci
))
1926 __cap_delay_requeue_front(mdsc
, ci
);
1927 spin_unlock(&ci
->i_ceph_lock
);
1933 * After a recovering MDS goes active, we need to resend any caps
1936 * Caller holds session->s_mutex.
1938 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
1939 struct ceph_mds_session
*session
)
1941 struct ceph_cap_snap
*capsnap
;
1943 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
1944 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
1946 struct ceph_inode_info
*ci
= capsnap
->ci
;
1947 struct inode
*inode
= &ci
->vfs_inode
;
1948 struct ceph_cap
*cap
;
1950 spin_lock(&ci
->i_ceph_lock
);
1951 cap
= ci
->i_auth_cap
;
1952 if (cap
&& cap
->session
== session
) {
1953 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
1955 __ceph_flush_snaps(ci
, &session
, 1);
1957 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1958 cap
, session
->s_mds
);
1960 spin_unlock(&ci
->i_ceph_lock
);
1964 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
1965 struct ceph_mds_session
*session
)
1967 struct ceph_inode_info
*ci
;
1969 kick_flushing_capsnaps(mdsc
, session
);
1971 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
1972 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
1973 struct inode
*inode
= &ci
->vfs_inode
;
1974 struct ceph_cap
*cap
;
1977 spin_lock(&ci
->i_ceph_lock
);
1978 cap
= ci
->i_auth_cap
;
1979 if (cap
&& cap
->session
== session
) {
1980 dout("kick_flushing_caps %p cap %p %s\n", inode
,
1981 cap
, ceph_cap_string(ci
->i_flushing_caps
));
1982 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1983 __ceph_caps_used(ci
),
1984 __ceph_caps_wanted(ci
),
1985 cap
->issued
| cap
->implemented
,
1986 ci
->i_flushing_caps
, NULL
);
1988 spin_lock(&ci
->i_ceph_lock
);
1989 __cap_delay_requeue(mdsc
, ci
);
1990 spin_unlock(&ci
->i_ceph_lock
);
1993 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1994 cap
, session
->s_mds
);
1995 spin_unlock(&ci
->i_ceph_lock
);
2000 static void kick_flushing_inode_caps(struct ceph_mds_client
*mdsc
,
2001 struct ceph_mds_session
*session
,
2002 struct inode
*inode
)
2004 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2005 struct ceph_cap
*cap
;
2008 spin_lock(&ci
->i_ceph_lock
);
2009 cap
= ci
->i_auth_cap
;
2010 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode
,
2011 ceph_cap_string(ci
->i_flushing_caps
), ci
->i_cap_flush_seq
);
2013 __ceph_flush_snaps(ci
, &session
, 1);
2015 if (ci
->i_flushing_caps
) {
2016 spin_lock(&mdsc
->cap_dirty_lock
);
2017 list_move_tail(&ci
->i_flushing_item
,
2018 &cap
->session
->s_cap_flushing
);
2019 spin_unlock(&mdsc
->cap_dirty_lock
);
2021 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
2022 __ceph_caps_used(ci
),
2023 __ceph_caps_wanted(ci
),
2024 cap
->issued
| cap
->implemented
,
2025 ci
->i_flushing_caps
, NULL
);
2027 spin_lock(&ci
->i_ceph_lock
);
2028 __cap_delay_requeue(mdsc
, ci
);
2029 spin_unlock(&ci
->i_ceph_lock
);
2032 spin_unlock(&ci
->i_ceph_lock
);
2038 * Take references to capabilities we hold, so that we don't release
2039 * them to the MDS prematurely.
2041 * Protected by i_ceph_lock.
2043 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
)
2045 if (got
& CEPH_CAP_PIN
)
2047 if (got
& CEPH_CAP_FILE_RD
)
2049 if (got
& CEPH_CAP_FILE_CACHE
)
2050 ci
->i_rdcache_ref
++;
2051 if (got
& CEPH_CAP_FILE_WR
)
2053 if (got
& CEPH_CAP_FILE_BUFFER
) {
2054 if (ci
->i_wb_ref
== 0)
2055 ihold(&ci
->vfs_inode
);
2057 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2058 &ci
->vfs_inode
, ci
->i_wb_ref
-1, ci
->i_wb_ref
);
2063 * Try to grab cap references. Specify those refs we @want, and the
2064 * minimal set we @need. Also include the larger offset we are writing
2065 * to (when applicable), and check against max_size here as well.
2066 * Note that caller is responsible for ensuring max_size increases are
2067 * requested from the MDS.
2069 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
2070 int *got
, loff_t endoff
, int *check_max
, int *err
)
2072 struct inode
*inode
= &ci
->vfs_inode
;
2074 int have
, implemented
;
2077 dout("get_cap_refs %p need %s want %s\n", inode
,
2078 ceph_cap_string(need
), ceph_cap_string(want
));
2079 spin_lock(&ci
->i_ceph_lock
);
2081 /* make sure file is actually open */
2082 file_wanted
= __ceph_caps_file_wanted(ci
);
2083 if ((file_wanted
& need
) == 0) {
2084 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2085 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2091 /* finish pending truncate */
2092 while (ci
->i_truncate_pending
) {
2093 spin_unlock(&ci
->i_ceph_lock
);
2094 __ceph_do_pending_vmtruncate(inode
);
2095 spin_lock(&ci
->i_ceph_lock
);
2098 have
= __ceph_caps_issued(ci
, &implemented
);
2100 if (have
& need
& CEPH_CAP_FILE_WR
) {
2101 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2102 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2103 inode
, endoff
, ci
->i_max_size
);
2104 if (endoff
> ci
->i_requested_max_size
) {
2111 * If a sync write is in progress, we must wait, so that we
2112 * can get a final snapshot value for size+mtime.
2114 if (__ceph_have_pending_cap_snap(ci
)) {
2115 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2120 if ((have
& need
) == need
) {
2122 * Look at (implemented & ~have & not) so that we keep waiting
2123 * on transition from wanted -> needed caps. This is needed
2124 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2125 * going before a prior buffered writeback happens.
2127 int not = want
& ~(have
& need
);
2128 int revoking
= implemented
& ~have
;
2129 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2130 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2131 ceph_cap_string(revoking
));
2132 if ((revoking
& not) == 0) {
2133 *got
= need
| (have
& want
);
2134 __take_cap_refs(ci
, *got
);
2138 dout("get_cap_refs %p have %s needed %s\n", inode
,
2139 ceph_cap_string(have
), ceph_cap_string(need
));
2142 spin_unlock(&ci
->i_ceph_lock
);
2143 dout("get_cap_refs %p ret %d got %s\n", inode
,
2144 ret
, ceph_cap_string(*got
));
2149 * Check the offset we are writing up to against our current
2150 * max_size. If necessary, tell the MDS we want to write to
2153 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2155 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2158 /* do we need to explicitly request a larger max_size? */
2159 spin_lock(&ci
->i_ceph_lock
);
2160 if (endoff
>= ci
->i_max_size
&& endoff
> ci
->i_wanted_max_size
) {
2161 dout("write %p at large endoff %llu, req max_size\n",
2163 ci
->i_wanted_max_size
= endoff
;
2165 /* duplicate ceph_check_caps()'s logic */
2166 if (ci
->i_auth_cap
&&
2167 (ci
->i_auth_cap
->issued
& CEPH_CAP_FILE_WR
) &&
2168 ci
->i_wanted_max_size
> ci
->i_max_size
&&
2169 ci
->i_wanted_max_size
> ci
->i_requested_max_size
)
2171 spin_unlock(&ci
->i_ceph_lock
);
2173 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2177 * Wait for caps, and take cap references. If we can't get a WR cap
2178 * due to a small max_size, make sure we check_max_size (and possibly
2179 * ask the mds) so we don't get hung up indefinitely.
2181 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
, int *got
,
2184 int check_max
, ret
, err
;
2188 check_max_size(&ci
->vfs_inode
, endoff
);
2191 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2192 try_get_cap_refs(ci
, need
, want
,
2203 * Take cap refs. Caller must already know we hold at least one ref
2204 * on the caps in question or we don't know this is safe.
2206 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2208 spin_lock(&ci
->i_ceph_lock
);
2209 __take_cap_refs(ci
, caps
);
2210 spin_unlock(&ci
->i_ceph_lock
);
2216 * If we released the last ref on any given cap, call ceph_check_caps
2217 * to release (or schedule a release).
2219 * If we are releasing a WR cap (from a sync write), finalize any affected
2220 * cap_snap, and wake up any waiters.
2222 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2224 struct inode
*inode
= &ci
->vfs_inode
;
2225 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2226 struct ceph_cap_snap
*capsnap
;
2228 spin_lock(&ci
->i_ceph_lock
);
2229 if (had
& CEPH_CAP_PIN
)
2231 if (had
& CEPH_CAP_FILE_RD
)
2232 if (--ci
->i_rd_ref
== 0)
2234 if (had
& CEPH_CAP_FILE_CACHE
)
2235 if (--ci
->i_rdcache_ref
== 0)
2237 if (had
& CEPH_CAP_FILE_BUFFER
) {
2238 if (--ci
->i_wb_ref
== 0) {
2242 dout("put_cap_refs %p wb %d -> %d (?)\n",
2243 inode
, ci
->i_wb_ref
+1, ci
->i_wb_ref
);
2245 if (had
& CEPH_CAP_FILE_WR
)
2246 if (--ci
->i_wr_ref
== 0) {
2248 if (!list_empty(&ci
->i_cap_snaps
)) {
2249 capsnap
= list_first_entry(&ci
->i_cap_snaps
,
2250 struct ceph_cap_snap
,
2252 if (capsnap
->writing
) {
2253 capsnap
->writing
= 0;
2255 __ceph_finish_cap_snap(ci
,
2261 spin_unlock(&ci
->i_ceph_lock
);
2263 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2264 last
? " last" : "", put
? " put" : "");
2266 if (last
&& !flushsnaps
)
2267 ceph_check_caps(ci
, 0, NULL
);
2268 else if (flushsnaps
)
2269 ceph_flush_snaps(ci
);
2271 wake_up_all(&ci
->i_cap_wq
);
2277 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2278 * context. Adjust per-snap dirty page accounting as appropriate.
2279 * Once all dirty data for a cap_snap is flushed, flush snapped file
2280 * metadata back to the MDS. If we dropped the last ref, call
2283 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2284 struct ceph_snap_context
*snapc
)
2286 struct inode
*inode
= &ci
->vfs_inode
;
2288 int complete_capsnap
= 0;
2289 int drop_capsnap
= 0;
2291 struct ceph_cap_snap
*capsnap
= NULL
;
2293 spin_lock(&ci
->i_ceph_lock
);
2294 ci
->i_wrbuffer_ref
-= nr
;
2295 last
= !ci
->i_wrbuffer_ref
;
2297 if (ci
->i_head_snapc
== snapc
) {
2298 ci
->i_wrbuffer_ref_head
-= nr
;
2299 if (ci
->i_wrbuffer_ref_head
== 0 &&
2300 ci
->i_dirty_caps
== 0 && ci
->i_flushing_caps
== 0) {
2301 BUG_ON(!ci
->i_head_snapc
);
2302 ceph_put_snap_context(ci
->i_head_snapc
);
2303 ci
->i_head_snapc
= NULL
;
2305 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2307 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2308 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2309 last
? " LAST" : "");
2311 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2312 if (capsnap
->context
== snapc
) {
2318 capsnap
->dirty_pages
-= nr
;
2319 if (capsnap
->dirty_pages
== 0) {
2320 complete_capsnap
= 1;
2321 if (capsnap
->dirty
== 0)
2322 /* cap writeback completed before we created
2323 * the cap_snap; no FLUSHSNAP is needed */
2326 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2327 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2328 inode
, capsnap
, capsnap
->context
->seq
,
2329 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2330 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2331 last
? " (wrbuffer last)" : "",
2332 complete_capsnap
? " (complete capsnap)" : "",
2333 drop_capsnap
? " (drop capsnap)" : "");
2335 ceph_put_snap_context(capsnap
->context
);
2336 list_del(&capsnap
->ci_item
);
2337 list_del(&capsnap
->flushing_item
);
2338 ceph_put_cap_snap(capsnap
);
2342 spin_unlock(&ci
->i_ceph_lock
);
2345 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2347 } else if (complete_capsnap
) {
2348 ceph_flush_snaps(ci
);
2349 wake_up_all(&ci
->i_cap_wq
);
2356 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2358 static void invalidate_aliases(struct inode
*inode
)
2360 struct dentry
*dn
, *prev
= NULL
;
2362 dout("invalidate_aliases inode %p\n", inode
);
2363 d_prune_aliases(inode
);
2365 * For non-directory inode, d_find_alias() only returns
2366 * hashed dentry. After calling d_invalidate(), the
2367 * dentry becomes unhashed.
2369 * For directory inode, d_find_alias() can return
2370 * unhashed dentry. But directory inode should have
2371 * one alias at most.
2373 while ((dn
= d_find_alias(inode
))) {
2388 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2389 * actually be a revocation if it specifies a smaller cap set.)
2391 * caller holds s_mutex and i_ceph_lock, we drop both.
2395 * 1 - check_caps on auth cap only (writeback)
2396 * 2 - check_caps (ack revoke)
2398 static void handle_cap_grant(struct inode
*inode
, struct ceph_mds_caps
*grant
,
2399 struct ceph_mds_session
*session
,
2400 struct ceph_cap
*cap
,
2401 struct ceph_buffer
*xattr_buf
)
2402 __releases(ci
->i_ceph_lock
)
2404 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2405 int mds
= session
->s_mds
;
2406 int seq
= le32_to_cpu(grant
->seq
);
2407 int newcaps
= le32_to_cpu(grant
->caps
);
2408 int issued
, implemented
, used
, wanted
, dirty
;
2409 u64 size
= le64_to_cpu(grant
->size
);
2410 u64 max_size
= le64_to_cpu(grant
->max_size
);
2411 struct timespec mtime
, atime
, ctime
;
2415 int queue_invalidate
= 0;
2416 int deleted_inode
= 0;
2417 int queue_revalidate
= 0;
2419 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2420 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2421 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2425 * If CACHE is being revoked, and we have no dirty buffers,
2426 * try to invalidate (once). (If there are dirty buffers, we
2427 * will invalidate _after_ writeback.)
2429 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2430 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2431 !ci
->i_wrbuffer_ref
) {
2432 if (try_nonblocking_invalidate(inode
)) {
2433 /* there were locked pages.. invalidate later
2434 in a separate thread. */
2435 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2436 queue_invalidate
= 1;
2437 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2441 ceph_fscache_invalidate(inode
);
2444 /* side effects now are allowed */
2446 issued
= __ceph_caps_issued(ci
, &implemented
);
2447 issued
|= implemented
| __ceph_caps_dirty(ci
);
2449 cap
->cap_gen
= session
->s_cap_gen
;
2451 __check_cap_issue(ci
, cap
, newcaps
);
2453 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2454 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2455 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(grant
->uid
));
2456 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(grant
->gid
));
2457 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2458 from_kuid(&init_user_ns
, inode
->i_uid
),
2459 from_kgid(&init_user_ns
, inode
->i_gid
));
2462 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0) {
2463 set_nlink(inode
, le32_to_cpu(grant
->nlink
));
2464 if (inode
->i_nlink
== 0 &&
2465 (newcaps
& (CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
)))
2469 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2470 int len
= le32_to_cpu(grant
->xattr_len
);
2471 u64 version
= le64_to_cpu(grant
->xattr_version
);
2473 if (version
> ci
->i_xattrs
.version
) {
2474 dout(" got new xattrs v%llu on %p len %d\n",
2475 version
, inode
, len
);
2476 if (ci
->i_xattrs
.blob
)
2477 ceph_buffer_put(ci
->i_xattrs
.blob
);
2478 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2479 ci
->i_xattrs
.version
= version
;
2480 ceph_forget_all_cached_acls(inode
);
2484 /* Do we need to revalidate our fscache cookie. Don't bother on the
2485 * first cache cap as we already validate at cookie creation time. */
2486 if ((issued
& CEPH_CAP_FILE_CACHE
) && ci
->i_rdcache_gen
> 1)
2487 queue_revalidate
= 1;
2489 /* size/ctime/mtime/atime? */
2490 ceph_fill_file_size(inode
, issued
,
2491 le32_to_cpu(grant
->truncate_seq
),
2492 le64_to_cpu(grant
->truncate_size
), size
);
2493 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2494 ceph_decode_timespec(&atime
, &grant
->atime
);
2495 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2496 ceph_fill_file_time(inode
, issued
,
2497 le32_to_cpu(grant
->time_warp_seq
), &ctime
, &mtime
,
2500 /* max size increase? */
2501 if (ci
->i_auth_cap
== cap
&& max_size
!= ci
->i_max_size
) {
2502 dout("max_size %lld -> %llu\n", ci
->i_max_size
, max_size
);
2503 ci
->i_max_size
= max_size
;
2504 if (max_size
>= ci
->i_wanted_max_size
) {
2505 ci
->i_wanted_max_size
= 0; /* reset */
2506 ci
->i_requested_max_size
= 0;
2511 /* check cap bits */
2512 wanted
= __ceph_caps_wanted(ci
);
2513 used
= __ceph_caps_used(ci
);
2514 dirty
= __ceph_caps_dirty(ci
);
2515 dout(" my wanted = %s, used = %s, dirty %s\n",
2516 ceph_cap_string(wanted
),
2517 ceph_cap_string(used
),
2518 ceph_cap_string(dirty
));
2519 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2520 dout("mds wanted %s -> %s\n",
2521 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2522 ceph_cap_string(wanted
));
2523 /* imported cap may not have correct mds_wanted */
2524 if (le32_to_cpu(grant
->op
) == CEPH_CAP_OP_IMPORT
)
2530 /* file layout may have changed */
2531 ci
->i_layout
= grant
->layout
;
2533 /* revocation, grant, or no-op? */
2534 if (cap
->issued
& ~newcaps
) {
2535 int revoking
= cap
->issued
& ~newcaps
;
2537 dout("revocation: %s -> %s (revoking %s)\n",
2538 ceph_cap_string(cap
->issued
),
2539 ceph_cap_string(newcaps
),
2540 ceph_cap_string(revoking
));
2541 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2542 writeback
= 1; /* initiate writeback; will delay ack */
2543 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2544 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2546 ; /* do nothing yet, invalidation will be queued */
2547 else if (cap
== ci
->i_auth_cap
)
2548 check_caps
= 1; /* check auth cap only */
2550 check_caps
= 2; /* check all caps */
2551 cap
->issued
= newcaps
;
2552 cap
->implemented
|= newcaps
;
2553 } else if (cap
->issued
== newcaps
) {
2554 dout("caps unchanged: %s -> %s\n",
2555 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2557 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2558 ceph_cap_string(newcaps
));
2559 /* non-auth MDS is revoking the newly grant caps ? */
2560 if (cap
== ci
->i_auth_cap
&&
2561 __ceph_caps_revoking_other(ci
, cap
, newcaps
))
2564 cap
->issued
= newcaps
;
2565 cap
->implemented
|= newcaps
; /* add bits only, to
2566 * avoid stepping on a
2567 * pending revocation */
2570 BUG_ON(cap
->issued
& ~cap
->implemented
);
2572 spin_unlock(&ci
->i_ceph_lock
);
2576 * queue inode for writeback: we can't actually call
2577 * filemap_write_and_wait, etc. from message handler
2580 ceph_queue_writeback(inode
);
2581 if (queue_invalidate
)
2582 ceph_queue_invalidate(inode
);
2584 invalidate_aliases(inode
);
2585 if (queue_revalidate
)
2586 ceph_queue_revalidate(inode
);
2588 wake_up_all(&ci
->i_cap_wq
);
2590 if (check_caps
== 1)
2591 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2593 else if (check_caps
== 2)
2594 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2596 mutex_unlock(&session
->s_mutex
);
2600 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2601 * MDS has been safely committed.
2603 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2604 struct ceph_mds_caps
*m
,
2605 struct ceph_mds_session
*session
,
2606 struct ceph_cap
*cap
)
2607 __releases(ci
->i_ceph_lock
)
2609 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2610 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2611 unsigned seq
= le32_to_cpu(m
->seq
);
2612 int dirty
= le32_to_cpu(m
->dirty
);
2617 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2618 if ((dirty
& (1 << i
)) &&
2619 flush_tid
== ci
->i_cap_flush_tid
[i
])
2622 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2623 " flushing %s -> %s\n",
2624 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2625 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2626 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2628 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2631 ci
->i_flushing_caps
&= ~cleaned
;
2633 spin_lock(&mdsc
->cap_dirty_lock
);
2634 if (ci
->i_flushing_caps
== 0) {
2635 list_del_init(&ci
->i_flushing_item
);
2636 if (!list_empty(&session
->s_cap_flushing
))
2637 dout(" mds%d still flushing cap on %p\n",
2639 &list_entry(session
->s_cap_flushing
.next
,
2640 struct ceph_inode_info
,
2641 i_flushing_item
)->vfs_inode
);
2642 mdsc
->num_cap_flushing
--;
2643 wake_up_all(&mdsc
->cap_flushing_wq
);
2644 dout(" inode %p now !flushing\n", inode
);
2646 if (ci
->i_dirty_caps
== 0) {
2647 dout(" inode %p now clean\n", inode
);
2648 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2650 if (ci
->i_wrbuffer_ref_head
== 0) {
2651 BUG_ON(!ci
->i_head_snapc
);
2652 ceph_put_snap_context(ci
->i_head_snapc
);
2653 ci
->i_head_snapc
= NULL
;
2656 BUG_ON(list_empty(&ci
->i_dirty_item
));
2659 spin_unlock(&mdsc
->cap_dirty_lock
);
2660 wake_up_all(&ci
->i_cap_wq
);
2663 spin_unlock(&ci
->i_ceph_lock
);
2669 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2670 * throw away our cap_snap.
2672 * Caller hold s_mutex.
2674 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2675 struct ceph_mds_caps
*m
,
2676 struct ceph_mds_session
*session
)
2678 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2679 u64 follows
= le64_to_cpu(m
->snap_follows
);
2680 struct ceph_cap_snap
*capsnap
;
2683 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2684 inode
, ci
, session
->s_mds
, follows
);
2686 spin_lock(&ci
->i_ceph_lock
);
2687 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2688 if (capsnap
->follows
== follows
) {
2689 if (capsnap
->flush_tid
!= flush_tid
) {
2690 dout(" cap_snap %p follows %lld tid %lld !="
2691 " %lld\n", capsnap
, follows
,
2692 flush_tid
, capsnap
->flush_tid
);
2695 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2696 dout(" removing %p cap_snap %p follows %lld\n",
2697 inode
, capsnap
, follows
);
2698 ceph_put_snap_context(capsnap
->context
);
2699 list_del(&capsnap
->ci_item
);
2700 list_del(&capsnap
->flushing_item
);
2701 ceph_put_cap_snap(capsnap
);
2705 dout(" skipping cap_snap %p follows %lld\n",
2706 capsnap
, capsnap
->follows
);
2709 spin_unlock(&ci
->i_ceph_lock
);
2715 * Handle TRUNC from MDS, indicating file truncation.
2717 * caller hold s_mutex.
2719 static void handle_cap_trunc(struct inode
*inode
,
2720 struct ceph_mds_caps
*trunc
,
2721 struct ceph_mds_session
*session
)
2722 __releases(ci
->i_ceph_lock
)
2724 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2725 int mds
= session
->s_mds
;
2726 int seq
= le32_to_cpu(trunc
->seq
);
2727 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2728 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2729 u64 size
= le64_to_cpu(trunc
->size
);
2730 int implemented
= 0;
2731 int dirty
= __ceph_caps_dirty(ci
);
2732 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2733 int queue_trunc
= 0;
2735 issued
|= implemented
| dirty
;
2737 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2738 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2739 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2740 truncate_seq
, truncate_size
, size
);
2741 spin_unlock(&ci
->i_ceph_lock
);
2744 ceph_queue_vmtruncate(inode
);
2745 ceph_fscache_invalidate(inode
);
2750 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2751 * different one. If we are the most recent migration we've seen (as
2752 * indicated by mseq), make note of the migrating cap bits for the
2753 * duration (until we see the corresponding IMPORT).
2755 * caller holds s_mutex
2757 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2758 struct ceph_mds_session
*session
,
2759 int *open_target_sessions
)
2761 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2762 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2763 int mds
= session
->s_mds
;
2764 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2765 struct ceph_cap
*cap
= NULL
, *t
;
2769 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2770 inode
, ci
, mds
, mseq
);
2772 spin_lock(&ci
->i_ceph_lock
);
2774 /* make sure we haven't seen a higher mseq */
2775 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
2776 t
= rb_entry(p
, struct ceph_cap
, ci_node
);
2777 if (ceph_seq_cmp(t
->mseq
, mseq
) > 0) {
2778 dout(" higher mseq on cap from mds%d\n",
2782 if (t
->session
->s_mds
== mds
)
2789 ci
->i_cap_exporting_mds
= mds
;
2790 ci
->i_cap_exporting_mseq
= mseq
;
2791 ci
->i_cap_exporting_issued
= cap
->issued
;
2794 * make sure we have open sessions with all possible
2795 * export targets, so that we get the matching IMPORT
2797 *open_target_sessions
= 1;
2800 * we can't flush dirty caps that we've seen the
2801 * EXPORT but no IMPORT for
2803 spin_lock(&mdsc
->cap_dirty_lock
);
2804 if (!list_empty(&ci
->i_dirty_item
)) {
2805 dout(" moving %p to cap_dirty_migrating\n",
2807 list_move(&ci
->i_dirty_item
,
2808 &mdsc
->cap_dirty_migrating
);
2810 spin_unlock(&mdsc
->cap_dirty_lock
);
2812 __ceph_remove_cap(cap
, false);
2814 /* else, we already released it */
2816 spin_unlock(&ci
->i_ceph_lock
);
2820 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2823 * caller holds s_mutex.
2825 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
2826 struct inode
*inode
, struct ceph_mds_caps
*im
,
2827 struct ceph_mds_cap_peer
*ph
,
2828 struct ceph_mds_session
*session
,
2829 void *snaptrace
, int snaptrace_len
)
2831 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2832 struct ceph_cap
*cap
;
2833 int mds
= session
->s_mds
;
2834 unsigned issued
= le32_to_cpu(im
->caps
);
2835 unsigned wanted
= le32_to_cpu(im
->wanted
);
2836 unsigned seq
= le32_to_cpu(im
->seq
);
2837 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
2838 u64 realmino
= le64_to_cpu(im
->realm
);
2839 u64 cap_id
= le64_to_cpu(im
->cap_id
);
2844 p_cap_id
= le64_to_cpu(ph
->cap_id
);
2845 peer
= le32_to_cpu(ph
->mds
);
2851 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
2852 inode
, ci
, mds
, mseq
, peer
);
2854 spin_lock(&ci
->i_ceph_lock
);
2855 cap
= peer
>= 0 ? __get_cap_for_mds(ci
, peer
) : NULL
;
2856 if (cap
&& cap
->cap_id
== p_cap_id
) {
2857 dout(" remove export cap %p mds%d flags %d\n",
2858 cap
, peer
, ph
->flags
);
2859 if ((ph
->flags
& CEPH_CAP_FLAG_AUTH
) &&
2860 (cap
->seq
!= le32_to_cpu(ph
->seq
) ||
2861 cap
->mseq
!= le32_to_cpu(ph
->mseq
))) {
2862 pr_err("handle_cap_import: mismatched seq/mseq: "
2863 "ino (%llx.%llx) mds%d seq %d mseq %d "
2864 "importer mds%d has peer seq %d mseq %d\n",
2865 ceph_vinop(inode
), peer
, cap
->seq
,
2866 cap
->mseq
, mds
, le32_to_cpu(ph
->seq
),
2867 le32_to_cpu(ph
->mseq
));
2869 ci
->i_cap_exporting_issued
= cap
->issued
;
2870 __ceph_remove_cap(cap
, (ph
->flags
& CEPH_CAP_FLAG_RELEASE
));
2873 /* make sure we re-request max_size, if necessary */
2874 ci
->i_wanted_max_size
= 0;
2875 ci
->i_requested_max_size
= 0;
2876 spin_unlock(&ci
->i_ceph_lock
);
2878 down_write(&mdsc
->snap_rwsem
);
2879 ceph_update_snap_trace(mdsc
, snaptrace
, snaptrace
+snaptrace_len
,
2881 downgrade_write(&mdsc
->snap_rwsem
);
2882 ceph_add_cap(inode
, session
, cap_id
, -1,
2883 issued
, wanted
, seq
, mseq
, realmino
, CEPH_CAP_FLAG_AUTH
,
2884 NULL
/* no caps context */);
2885 kick_flushing_inode_caps(mdsc
, session
, inode
);
2886 up_read(&mdsc
->snap_rwsem
);
2891 * Handle a caps message from the MDS.
2893 * Identify the appropriate session, inode, and call the right handler
2894 * based on the cap op.
2896 void ceph_handle_caps(struct ceph_mds_session
*session
,
2897 struct ceph_msg
*msg
)
2899 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2900 struct super_block
*sb
= mdsc
->fsc
->sb
;
2901 struct inode
*inode
;
2902 struct ceph_inode_info
*ci
;
2903 struct ceph_cap
*cap
;
2904 struct ceph_mds_caps
*h
;
2905 struct ceph_mds_cap_peer
*peer
= NULL
;
2906 int mds
= session
->s_mds
;
2909 struct ceph_vino vino
;
2914 size_t snaptrace_len
;
2918 int open_target_sessions
= 0;
2920 dout("handle_caps from mds%d\n", mds
);
2923 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
2924 tid
= le64_to_cpu(msg
->hdr
.tid
);
2925 if (msg
->front
.iov_len
< sizeof(*h
))
2927 h
= msg
->front
.iov_base
;
2928 op
= le32_to_cpu(h
->op
);
2929 vino
.ino
= le64_to_cpu(h
->ino
);
2930 vino
.snap
= CEPH_NOSNAP
;
2931 cap_id
= le64_to_cpu(h
->cap_id
);
2932 seq
= le32_to_cpu(h
->seq
);
2933 mseq
= le32_to_cpu(h
->migrate_seq
);
2934 size
= le64_to_cpu(h
->size
);
2935 max_size
= le64_to_cpu(h
->max_size
);
2938 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
2940 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
2941 void *p
= snaptrace
+ snaptrace_len
;
2942 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
2943 if (p
+ flock_len
> end
)
2951 if (le16_to_cpu(msg
->hdr
.version
) >= 3) {
2952 if (op
== CEPH_CAP_OP_IMPORT
) {
2953 void *p
= flock
+ flock_len
;
2954 if (p
+ sizeof(*peer
) > end
)
2960 mutex_lock(&session
->s_mutex
);
2962 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
2965 if (op
== CEPH_CAP_OP_IMPORT
)
2966 ceph_add_cap_releases(mdsc
, session
);
2969 inode
= ceph_find_inode(sb
, vino
);
2970 ci
= ceph_inode(inode
);
2971 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
2974 dout(" i don't have ino %llx\n", vino
.ino
);
2976 if (op
== CEPH_CAP_OP_IMPORT
) {
2977 spin_lock(&session
->s_cap_lock
);
2978 __queue_cap_release(session
, vino
.ino
, cap_id
,
2980 spin_unlock(&session
->s_cap_lock
);
2982 goto flush_cap_releases
;
2985 /* these will work even if we don't have a cap yet */
2987 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
2988 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
2991 case CEPH_CAP_OP_EXPORT
:
2992 handle_cap_export(inode
, h
, session
, &open_target_sessions
);
2995 case CEPH_CAP_OP_IMPORT
:
2996 handle_cap_import(mdsc
, inode
, h
, peer
, session
,
2997 snaptrace
, snaptrace_len
);
3000 /* the rest require a cap */
3001 spin_lock(&ci
->i_ceph_lock
);
3002 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
3004 dout(" no cap on %p ino %llx.%llx from mds%d\n",
3005 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
3006 spin_unlock(&ci
->i_ceph_lock
);
3007 goto flush_cap_releases
;
3010 /* note that each of these drops i_ceph_lock for us */
3012 case CEPH_CAP_OP_REVOKE
:
3013 case CEPH_CAP_OP_GRANT
:
3014 case CEPH_CAP_OP_IMPORT
:
3015 handle_cap_grant(inode
, h
, session
, cap
, msg
->middle
);
3018 case CEPH_CAP_OP_FLUSH_ACK
:
3019 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
3022 case CEPH_CAP_OP_TRUNC
:
3023 handle_cap_trunc(inode
, h
, session
);
3027 spin_unlock(&ci
->i_ceph_lock
);
3028 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
3029 ceph_cap_op_name(op
));
3036 * send any full release message to try to move things
3037 * along for the mds (who clearly thinks we still have this
3040 ceph_add_cap_releases(mdsc
, session
);
3041 ceph_send_cap_releases(mdsc
, session
);
3044 mutex_unlock(&session
->s_mutex
);
3048 if (open_target_sessions
)
3049 ceph_mdsc_open_export_target_sessions(mdsc
, session
);
3053 pr_err("ceph_handle_caps: corrupt message\n");
3059 * Delayed work handler to process end of delayed cap release LRU list.
3061 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
3063 struct ceph_inode_info
*ci
;
3064 int flags
= CHECK_CAPS_NODELAY
;
3066 dout("check_delayed_caps\n");
3068 spin_lock(&mdsc
->cap_delay_lock
);
3069 if (list_empty(&mdsc
->cap_delay_list
))
3071 ci
= list_first_entry(&mdsc
->cap_delay_list
,
3072 struct ceph_inode_info
,
3074 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
3075 time_before(jiffies
, ci
->i_hold_caps_max
))
3077 list_del_init(&ci
->i_cap_delay_list
);
3078 spin_unlock(&mdsc
->cap_delay_lock
);
3079 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
3080 ceph_check_caps(ci
, flags
, NULL
);
3082 spin_unlock(&mdsc
->cap_delay_lock
);
3086 * Flush all dirty caps to the mds
3088 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
3090 struct ceph_inode_info
*ci
;
3091 struct inode
*inode
;
3093 dout("flush_dirty_caps\n");
3094 spin_lock(&mdsc
->cap_dirty_lock
);
3095 while (!list_empty(&mdsc
->cap_dirty
)) {
3096 ci
= list_first_entry(&mdsc
->cap_dirty
, struct ceph_inode_info
,
3098 inode
= &ci
->vfs_inode
;
3100 dout("flush_dirty_caps %p\n", inode
);
3101 spin_unlock(&mdsc
->cap_dirty_lock
);
3102 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
, NULL
);
3104 spin_lock(&mdsc
->cap_dirty_lock
);
3106 spin_unlock(&mdsc
->cap_dirty_lock
);
3107 dout("flush_dirty_caps done\n");
3111 * Drop open file reference. If we were the last open file,
3112 * we may need to release capabilities to the MDS (or schedule
3113 * their delayed release).
3115 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
3117 struct inode
*inode
= &ci
->vfs_inode
;
3120 spin_lock(&ci
->i_ceph_lock
);
3121 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
3122 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
3123 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
3124 if (--ci
->i_nr_by_mode
[fmode
] == 0)
3126 spin_unlock(&ci
->i_ceph_lock
);
3128 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
3129 ceph_check_caps(ci
, 0, NULL
);
3133 * Helpers for embedding cap and dentry lease releases into mds
3136 * @force is used by dentry_release (below) to force inclusion of a
3137 * record for the directory inode, even when there aren't any caps to
3140 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
3141 int mds
, int drop
, int unless
, int force
)
3143 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3144 struct ceph_cap
*cap
;
3145 struct ceph_mds_request_release
*rel
= *p
;
3149 spin_lock(&ci
->i_ceph_lock
);
3150 used
= __ceph_caps_used(ci
);
3151 dirty
= __ceph_caps_dirty(ci
);
3153 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3154 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
3155 ceph_cap_string(unless
));
3157 /* only drop unused, clean caps */
3158 drop
&= ~(used
| dirty
);
3160 cap
= __get_cap_for_mds(ci
, mds
);
3161 if (cap
&& __cap_is_valid(cap
)) {
3163 ((cap
->issued
& drop
) &&
3164 (cap
->issued
& unless
) == 0)) {
3165 if ((cap
->issued
& drop
) &&
3166 (cap
->issued
& unless
) == 0) {
3167 int wanted
= __ceph_caps_wanted(ci
);
3168 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0)
3169 wanted
|= cap
->mds_wanted
;
3170 dout("encode_inode_release %p cap %p "
3171 "%s -> %s, wanted %s -> %s\n", inode
, cap
,
3172 ceph_cap_string(cap
->issued
),
3173 ceph_cap_string(cap
->issued
& ~drop
),
3174 ceph_cap_string(cap
->mds_wanted
),
3175 ceph_cap_string(wanted
));
3177 cap
->issued
&= ~drop
;
3178 cap
->implemented
&= ~drop
;
3179 cap
->mds_wanted
= wanted
;
3181 dout("encode_inode_release %p cap %p %s"
3182 " (force)\n", inode
, cap
,
3183 ceph_cap_string(cap
->issued
));
3186 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3187 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3188 rel
->seq
= cpu_to_le32(cap
->seq
);
3189 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
),
3190 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3191 rel
->caps
= cpu_to_le32(cap
->issued
);
3192 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3198 dout("encode_inode_release %p cap %p %s\n",
3199 inode
, cap
, ceph_cap_string(cap
->issued
));
3202 spin_unlock(&ci
->i_ceph_lock
);
3206 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3207 int mds
, int drop
, int unless
)
3209 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3210 struct ceph_mds_request_release
*rel
= *p
;
3211 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3216 * force an record for the directory caps if we have a dentry lease.
3217 * this is racy (can't take i_ceph_lock and d_lock together), but it
3218 * doesn't have to be perfect; the mds will revoke anything we don't
3221 spin_lock(&dentry
->d_lock
);
3222 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3224 spin_unlock(&dentry
->d_lock
);
3226 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3228 spin_lock(&dentry
->d_lock
);
3229 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3230 dout("encode_dentry_release %p mds%d seq %d\n",
3231 dentry
, mds
, (int)di
->lease_seq
);
3232 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3233 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3234 *p
+= dentry
->d_name
.len
;
3235 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3236 __ceph_mdsc_drop_dentry_lease(dentry
);
3238 spin_unlock(&dentry
->d_lock
);