1 #include "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>
13 #include "messenger.h"
16 * Capability management
18 * The Ceph metadata servers control client access to inode metadata
19 * and file data by issuing capabilities, granting clients permission
20 * to read and/or write both inode field and file data to OSDs
21 * (storage nodes). Each capability consists of a set of bits
22 * indicating which operations are allowed.
24 * If the client holds a *_SHARED cap, the client has a coherent value
25 * that can be safely read from the cached inode.
27 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
28 * client is allowed to change inode attributes (e.g., file size,
29 * mtime), note its dirty state in the ceph_cap, and asynchronously
30 * flush that metadata change to the MDS.
32 * In the event of a conflicting operation (perhaps by another
33 * client), the MDS will revoke the conflicting client capabilities.
35 * In order for a client to cache an inode, it must hold a capability
36 * with at least one MDS server. When inodes are released, release
37 * notifications are batched and periodically sent en masse to the MDS
38 * cluster to release server state.
43 * Generate readable cap strings for debugging output.
45 #define MAX_CAP_STR 20
46 static char cap_str
[MAX_CAP_STR
][40];
47 static DEFINE_SPINLOCK(cap_str_lock
);
48 static int last_cap_str
;
50 static char *gcap_string(char *s
, int c
)
52 if (c
& CEPH_CAP_GSHARED
)
54 if (c
& CEPH_CAP_GEXCL
)
56 if (c
& CEPH_CAP_GCACHE
)
62 if (c
& CEPH_CAP_GBUFFER
)
64 if (c
& CEPH_CAP_GLAZYIO
)
69 const char *ceph_cap_string(int caps
)
75 spin_lock(&cap_str_lock
);
77 if (last_cap_str
== MAX_CAP_STR
)
79 spin_unlock(&cap_str_lock
);
83 if (caps
& CEPH_CAP_PIN
)
86 c
= (caps
>> CEPH_CAP_SAUTH
) & 3;
89 s
= gcap_string(s
, c
);
92 c
= (caps
>> CEPH_CAP_SLINK
) & 3;
95 s
= gcap_string(s
, c
);
98 c
= (caps
>> CEPH_CAP_SXATTR
) & 3;
101 s
= gcap_string(s
, c
);
104 c
= caps
>> CEPH_CAP_SFILE
;
107 s
= gcap_string(s
, c
);
116 void ceph_caps_init(struct ceph_mds_client
*mdsc
)
118 INIT_LIST_HEAD(&mdsc
->caps_list
);
119 spin_lock_init(&mdsc
->caps_list_lock
);
122 void ceph_caps_finalize(struct ceph_mds_client
*mdsc
)
124 struct ceph_cap
*cap
;
126 spin_lock(&mdsc
->caps_list_lock
);
127 while (!list_empty(&mdsc
->caps_list
)) {
128 cap
= list_first_entry(&mdsc
->caps_list
,
129 struct ceph_cap
, caps_item
);
130 list_del(&cap
->caps_item
);
131 kmem_cache_free(ceph_cap_cachep
, cap
);
133 mdsc
->caps_total_count
= 0;
134 mdsc
->caps_avail_count
= 0;
135 mdsc
->caps_use_count
= 0;
136 mdsc
->caps_reserve_count
= 0;
137 mdsc
->caps_min_count
= 0;
138 spin_unlock(&mdsc
->caps_list_lock
);
141 void ceph_adjust_min_caps(struct ceph_mds_client
*mdsc
, int delta
)
143 spin_lock(&mdsc
->caps_list_lock
);
144 mdsc
->caps_min_count
+= delta
;
145 BUG_ON(mdsc
->caps_min_count
< 0);
146 spin_unlock(&mdsc
->caps_list_lock
);
149 int ceph_reserve_caps(struct ceph_mds_client
*mdsc
,
150 struct ceph_cap_reservation
*ctx
, int need
)
153 struct ceph_cap
*cap
;
159 dout("reserve caps ctx=%p need=%d\n", ctx
, need
);
161 /* first reserve any caps that are already allocated */
162 spin_lock(&mdsc
->caps_list_lock
);
163 if (mdsc
->caps_avail_count
>= need
)
166 have
= mdsc
->caps_avail_count
;
167 mdsc
->caps_avail_count
-= have
;
168 mdsc
->caps_reserve_count
+= have
;
169 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
170 mdsc
->caps_reserve_count
+
171 mdsc
->caps_avail_count
);
172 spin_unlock(&mdsc
->caps_list_lock
);
174 for (i
= have
; i
< need
; i
++) {
175 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
178 goto out_alloc_count
;
180 list_add(&cap
->caps_item
, &newcaps
);
183 BUG_ON(have
+ alloc
!= need
);
185 spin_lock(&mdsc
->caps_list_lock
);
186 mdsc
->caps_total_count
+= alloc
;
187 mdsc
->caps_reserve_count
+= alloc
;
188 list_splice(&newcaps
, &mdsc
->caps_list
);
190 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
191 mdsc
->caps_reserve_count
+
192 mdsc
->caps_avail_count
);
193 spin_unlock(&mdsc
->caps_list_lock
);
196 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
197 ctx
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
198 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
202 /* we didn't manage to reserve as much as we needed */
203 pr_warning("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
208 int ceph_unreserve_caps(struct ceph_mds_client
*mdsc
,
209 struct ceph_cap_reservation
*ctx
)
211 dout("unreserve caps ctx=%p count=%d\n", ctx
, ctx
->count
);
213 spin_lock(&mdsc
->caps_list_lock
);
214 BUG_ON(mdsc
->caps_reserve_count
< ctx
->count
);
215 mdsc
->caps_reserve_count
-= ctx
->count
;
216 mdsc
->caps_avail_count
+= ctx
->count
;
218 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
219 mdsc
->caps_total_count
, mdsc
->caps_use_count
,
220 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
221 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
222 mdsc
->caps_reserve_count
+
223 mdsc
->caps_avail_count
);
224 spin_unlock(&mdsc
->caps_list_lock
);
229 static struct ceph_cap
*get_cap(struct ceph_mds_client
*mdsc
,
230 struct ceph_cap_reservation
*ctx
)
232 struct ceph_cap
*cap
= NULL
;
234 /* temporary, until we do something about cap import/export */
236 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
238 mdsc
->caps_use_count
++;
239 mdsc
->caps_total_count
++;
244 spin_lock(&mdsc
->caps_list_lock
);
245 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
246 ctx
, ctx
->count
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
247 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
249 BUG_ON(ctx
->count
> mdsc
->caps_reserve_count
);
250 BUG_ON(list_empty(&mdsc
->caps_list
));
253 mdsc
->caps_reserve_count
--;
254 mdsc
->caps_use_count
++;
256 cap
= list_first_entry(&mdsc
->caps_list
, struct ceph_cap
, caps_item
);
257 list_del(&cap
->caps_item
);
259 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
260 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
261 spin_unlock(&mdsc
->caps_list_lock
);
265 void ceph_put_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
)
267 spin_lock(&mdsc
->caps_list_lock
);
268 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
269 cap
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
270 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
271 mdsc
->caps_use_count
--;
273 * Keep some preallocated caps around (ceph_min_count), to
274 * avoid lots of free/alloc churn.
276 if (mdsc
->caps_avail_count
>= mdsc
->caps_reserve_count
+
277 mdsc
->caps_min_count
) {
278 mdsc
->caps_total_count
--;
279 kmem_cache_free(ceph_cap_cachep
, cap
);
281 mdsc
->caps_avail_count
++;
282 list_add(&cap
->caps_item
, &mdsc
->caps_list
);
285 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
286 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
287 spin_unlock(&mdsc
->caps_list_lock
);
290 void ceph_reservation_status(struct ceph_client
*client
,
291 int *total
, int *avail
, int *used
, int *reserved
,
294 struct ceph_mds_client
*mdsc
= &client
->mdsc
;
297 *total
= mdsc
->caps_total_count
;
299 *avail
= mdsc
->caps_avail_count
;
301 *used
= mdsc
->caps_use_count
;
303 *reserved
= mdsc
->caps_reserve_count
;
305 *min
= mdsc
->caps_min_count
;
309 * Find ceph_cap for given mds, if any.
311 * Called with i_lock held.
313 static struct ceph_cap
*__get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
315 struct ceph_cap
*cap
;
316 struct rb_node
*n
= ci
->i_caps
.rb_node
;
319 cap
= rb_entry(n
, struct ceph_cap
, ci_node
);
322 else if (mds
> cap
->mds
)
330 struct ceph_cap
*ceph_get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
332 struct ceph_cap
*cap
;
334 spin_lock(&ci
->vfs_inode
.i_lock
);
335 cap
= __get_cap_for_mds(ci
, mds
);
336 spin_unlock(&ci
->vfs_inode
.i_lock
);
341 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
343 static int __ceph_get_cap_mds(struct ceph_inode_info
*ci
)
345 struct ceph_cap
*cap
;
349 /* prefer mds with WR|BUFFER|EXCL caps */
350 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
351 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
353 if (cap
->issued
& (CEPH_CAP_FILE_WR
|
354 CEPH_CAP_FILE_BUFFER
|
361 int ceph_get_cap_mds(struct inode
*inode
)
364 spin_lock(&inode
->i_lock
);
365 mds
= __ceph_get_cap_mds(ceph_inode(inode
));
366 spin_unlock(&inode
->i_lock
);
371 * Called under i_lock.
373 static void __insert_cap_node(struct ceph_inode_info
*ci
,
374 struct ceph_cap
*new)
376 struct rb_node
**p
= &ci
->i_caps
.rb_node
;
377 struct rb_node
*parent
= NULL
;
378 struct ceph_cap
*cap
= NULL
;
382 cap
= rb_entry(parent
, struct ceph_cap
, ci_node
);
383 if (new->mds
< cap
->mds
)
385 else if (new->mds
> cap
->mds
)
391 rb_link_node(&new->ci_node
, parent
, p
);
392 rb_insert_color(&new->ci_node
, &ci
->i_caps
);
396 * (re)set cap hold timeouts, which control the delayed release
397 * of unused caps back to the MDS. Should be called on cap use.
399 static void __cap_set_timeouts(struct ceph_mds_client
*mdsc
,
400 struct ceph_inode_info
*ci
)
402 struct ceph_mount_args
*ma
= mdsc
->client
->mount_args
;
404 ci
->i_hold_caps_min
= round_jiffies(jiffies
+
405 ma
->caps_wanted_delay_min
* HZ
);
406 ci
->i_hold_caps_max
= round_jiffies(jiffies
+
407 ma
->caps_wanted_delay_max
* HZ
);
408 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci
->vfs_inode
,
409 ci
->i_hold_caps_min
- jiffies
, ci
->i_hold_caps_max
- jiffies
);
413 * (Re)queue cap at the end of the delayed cap release list.
415 * If I_FLUSH is set, leave the inode at the front of the list.
417 * Caller holds i_lock
418 * -> we take mdsc->cap_delay_lock
420 static void __cap_delay_requeue(struct ceph_mds_client
*mdsc
,
421 struct ceph_inode_info
*ci
)
423 __cap_set_timeouts(mdsc
, ci
);
424 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci
->vfs_inode
,
425 ci
->i_ceph_flags
, ci
->i_hold_caps_max
);
426 if (!mdsc
->stopping
) {
427 spin_lock(&mdsc
->cap_delay_lock
);
428 if (!list_empty(&ci
->i_cap_delay_list
)) {
429 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
431 list_del_init(&ci
->i_cap_delay_list
);
433 list_add_tail(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
435 spin_unlock(&mdsc
->cap_delay_lock
);
440 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
441 * indicating we should send a cap message to flush dirty metadata
442 * asap, and move to the front of the delayed cap list.
444 static void __cap_delay_requeue_front(struct ceph_mds_client
*mdsc
,
445 struct ceph_inode_info
*ci
)
447 dout("__cap_delay_requeue_front %p\n", &ci
->vfs_inode
);
448 spin_lock(&mdsc
->cap_delay_lock
);
449 ci
->i_ceph_flags
|= CEPH_I_FLUSH
;
450 if (!list_empty(&ci
->i_cap_delay_list
))
451 list_del_init(&ci
->i_cap_delay_list
);
452 list_add(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
453 spin_unlock(&mdsc
->cap_delay_lock
);
457 * Cancel delayed work on cap.
459 * Caller must hold i_lock.
461 static void __cap_delay_cancel(struct ceph_mds_client
*mdsc
,
462 struct ceph_inode_info
*ci
)
464 dout("__cap_delay_cancel %p\n", &ci
->vfs_inode
);
465 if (list_empty(&ci
->i_cap_delay_list
))
467 spin_lock(&mdsc
->cap_delay_lock
);
468 list_del_init(&ci
->i_cap_delay_list
);
469 spin_unlock(&mdsc
->cap_delay_lock
);
473 * Common issue checks for add_cap, handle_cap_grant.
475 static void __check_cap_issue(struct ceph_inode_info
*ci
, struct ceph_cap
*cap
,
478 unsigned had
= __ceph_caps_issued(ci
, NULL
);
481 * Each time we receive FILE_CACHE anew, we increment
484 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
485 (had
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
489 * if we are newly issued FILE_SHARED, clear I_COMPLETE; we
490 * don't know what happened to this directory while we didn't
493 if ((issued
& CEPH_CAP_FILE_SHARED
) &&
494 (had
& CEPH_CAP_FILE_SHARED
) == 0) {
496 if (S_ISDIR(ci
->vfs_inode
.i_mode
)) {
497 dout(" marking %p NOT complete\n", &ci
->vfs_inode
);
498 ci
->i_ceph_flags
&= ~CEPH_I_COMPLETE
;
504 * Add a capability under the given MDS session.
506 * Caller should hold session snap_rwsem (read) and s_mutex.
508 * @fmode is the open file mode, if we are opening a file, otherwise
509 * it is < 0. (This is so we can atomically add the cap and add an
510 * open file reference to it.)
512 int ceph_add_cap(struct inode
*inode
,
513 struct ceph_mds_session
*session
, u64 cap_id
,
514 int fmode
, unsigned issued
, unsigned wanted
,
515 unsigned seq
, unsigned mseq
, u64 realmino
, int flags
,
516 struct ceph_cap_reservation
*caps_reservation
)
518 struct ceph_mds_client
*mdsc
= &ceph_inode_to_client(inode
)->mdsc
;
519 struct ceph_inode_info
*ci
= ceph_inode(inode
);
520 struct ceph_cap
*new_cap
= NULL
;
521 struct ceph_cap
*cap
;
522 int mds
= session
->s_mds
;
525 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
526 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
529 * If we are opening the file, include file mode wanted bits
533 wanted
|= ceph_caps_for_mode(fmode
);
536 spin_lock(&inode
->i_lock
);
537 cap
= __get_cap_for_mds(ci
, mds
);
543 spin_unlock(&inode
->i_lock
);
544 new_cap
= get_cap(mdsc
, caps_reservation
);
551 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
);
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 ceph_get_snap_realm(mdsc
, realm
);
581 spin_lock(&realm
->inodes_with_caps_lock
);
582 ci
->i_snap_realm
= realm
;
583 list_add(&ci
->i_snap_realm_item
,
584 &realm
->inodes_with_caps
);
585 spin_unlock(&realm
->inodes_with_caps_lock
);
587 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
593 __check_cap_issue(ci
, cap
, issued
);
596 * If we are issued caps we don't want, or the mds' wanted
597 * value appears to be off, queue a check so we'll release
598 * later and/or update the mds wanted value.
600 actual_wanted
= __ceph_caps_wanted(ci
);
601 if ((wanted
& ~actual_wanted
) ||
602 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
603 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
604 ceph_cap_string(issued
), ceph_cap_string(wanted
),
605 ceph_cap_string(actual_wanted
));
606 __cap_delay_requeue(mdsc
, ci
);
609 if (flags
& CEPH_CAP_FLAG_AUTH
)
610 ci
->i_auth_cap
= cap
;
611 else if (ci
->i_auth_cap
== cap
)
612 ci
->i_auth_cap
= NULL
;
614 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
615 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
616 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
617 cap
->cap_id
= cap_id
;
618 cap
->issued
= issued
;
619 cap
->implemented
|= issued
;
620 cap
->mds_wanted
|= wanted
;
622 cap
->issue_seq
= seq
;
624 cap
->cap_gen
= session
->s_cap_gen
;
627 __ceph_get_fmode(ci
, fmode
);
628 spin_unlock(&inode
->i_lock
);
629 wake_up_all(&ci
->i_cap_wq
);
634 * Return true if cap has not timed out and belongs to the current
635 * generation of the MDS session (i.e. has not gone 'stale' due to
636 * us losing touch with the mds).
638 static int __cap_is_valid(struct ceph_cap
*cap
)
643 spin_lock(&cap
->session
->s_cap_lock
);
644 gen
= cap
->session
->s_cap_gen
;
645 ttl
= cap
->session
->s_cap_ttl
;
646 spin_unlock(&cap
->session
->s_cap_lock
);
648 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
649 dout("__cap_is_valid %p cap %p issued %s "
650 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
651 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
659 * Return set of valid cap bits issued to us. Note that caps time
660 * out, and may be invalidated in bulk if the client session times out
661 * and session->s_cap_gen is bumped.
663 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
665 int have
= ci
->i_snap_caps
| ci
->i_cap_exporting_issued
;
666 struct ceph_cap
*cap
;
671 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
672 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
673 if (!__cap_is_valid(cap
))
675 dout("__ceph_caps_issued %p cap %p issued %s\n",
676 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
679 *implemented
|= cap
->implemented
;
685 * Get cap bits issued by caps other than @ocap
687 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
689 int have
= ci
->i_snap_caps
;
690 struct ceph_cap
*cap
;
693 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
694 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
697 if (!__cap_is_valid(cap
))
705 * Move a cap to the end of the LRU (oldest caps at list head, newest
708 static void __touch_cap(struct ceph_cap
*cap
)
710 struct ceph_mds_session
*s
= cap
->session
;
712 spin_lock(&s
->s_cap_lock
);
713 if (s
->s_cap_iterator
== NULL
) {
714 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
716 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
718 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
719 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
721 spin_unlock(&s
->s_cap_lock
);
725 * Check if we hold the given mask. If so, move the cap(s) to the
726 * front of their respective LRUs. (This is the preferred way for
727 * callers to check for caps they want.)
729 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
731 struct ceph_cap
*cap
;
733 int have
= ci
->i_snap_caps
;
735 if ((have
& mask
) == mask
) {
736 dout("__ceph_caps_issued_mask %p snap issued %s"
737 " (mask %s)\n", &ci
->vfs_inode
,
738 ceph_cap_string(have
),
739 ceph_cap_string(mask
));
743 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
744 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
745 if (!__cap_is_valid(cap
))
747 if ((cap
->issued
& mask
) == mask
) {
748 dout("__ceph_caps_issued_mask %p cap %p issued %s"
749 " (mask %s)\n", &ci
->vfs_inode
, cap
,
750 ceph_cap_string(cap
->issued
),
751 ceph_cap_string(mask
));
757 /* does a combination of caps satisfy mask? */
759 if ((have
& mask
) == mask
) {
760 dout("__ceph_caps_issued_mask %p combo issued %s"
761 " (mask %s)\n", &ci
->vfs_inode
,
762 ceph_cap_string(cap
->issued
),
763 ceph_cap_string(mask
));
767 /* touch this + preceeding caps */
769 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
771 cap
= rb_entry(q
, struct ceph_cap
,
773 if (!__cap_is_valid(cap
))
786 * Return true if mask caps are currently being revoked by an MDS.
788 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
790 struct inode
*inode
= &ci
->vfs_inode
;
791 struct ceph_cap
*cap
;
795 spin_lock(&inode
->i_lock
);
796 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
797 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
798 if (__cap_is_valid(cap
) &&
799 (cap
->implemented
& ~cap
->issued
& mask
)) {
804 spin_unlock(&inode
->i_lock
);
805 dout("ceph_caps_revoking %p %s = %d\n", inode
,
806 ceph_cap_string(mask
), ret
);
810 int __ceph_caps_used(struct ceph_inode_info
*ci
)
814 used
|= CEPH_CAP_PIN
;
816 used
|= CEPH_CAP_FILE_RD
;
817 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
818 used
|= CEPH_CAP_FILE_CACHE
;
820 used
|= CEPH_CAP_FILE_WR
;
821 if (ci
->i_wrbuffer_ref
)
822 used
|= CEPH_CAP_FILE_BUFFER
;
827 * wanted, by virtue of open file modes
829 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
833 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
834 if (ci
->i_nr_by_mode
[mode
])
835 want
|= ceph_caps_for_mode(mode
);
840 * Return caps we have registered with the MDS(s) as 'wanted'.
842 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
844 struct ceph_cap
*cap
;
848 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
849 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
850 if (!__cap_is_valid(cap
))
852 mds_wanted
|= cap
->mds_wanted
;
858 * called under i_lock
860 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
862 return !RB_EMPTY_ROOT(&ci
->i_caps
) || ci
->i_cap_exporting_mds
>= 0;
866 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
868 * caller should hold i_lock.
869 * caller will not hold session s_mutex if called from destroy_inode.
871 void __ceph_remove_cap(struct ceph_cap
*cap
)
873 struct ceph_mds_session
*session
= cap
->session
;
874 struct ceph_inode_info
*ci
= cap
->ci
;
875 struct ceph_mds_client
*mdsc
=
876 &ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
879 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
881 /* remove from session list */
882 spin_lock(&session
->s_cap_lock
);
883 if (session
->s_cap_iterator
== cap
) {
884 /* not yet, we are iterating over this very cap */
885 dout("__ceph_remove_cap delaying %p removal from session %p\n",
888 list_del_init(&cap
->session_caps
);
889 session
->s_nr_caps
--;
893 /* protect backpointer with s_cap_lock: see iterate_session_caps */
895 spin_unlock(&session
->s_cap_lock
);
897 /* remove from inode list */
898 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
899 if (ci
->i_auth_cap
== cap
)
900 ci
->i_auth_cap
= NULL
;
903 ceph_put_cap(mdsc
, cap
);
905 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
) {
906 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
907 spin_lock(&realm
->inodes_with_caps_lock
);
908 list_del_init(&ci
->i_snap_realm_item
);
909 ci
->i_snap_realm_counter
++;
910 ci
->i_snap_realm
= NULL
;
911 spin_unlock(&realm
->inodes_with_caps_lock
);
912 ceph_put_snap_realm(mdsc
, realm
);
914 if (!__ceph_is_any_real_caps(ci
))
915 __cap_delay_cancel(mdsc
, ci
);
919 * Build and send a cap message to the given MDS.
921 * Caller should be holding s_mutex.
923 static int send_cap_msg(struct ceph_mds_session
*session
,
924 u64 ino
, u64 cid
, int op
,
925 int caps
, int wanted
, int dirty
,
926 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
927 u64 size
, u64 max_size
,
928 struct timespec
*mtime
, struct timespec
*atime
,
930 uid_t uid
, gid_t gid
, mode_t mode
,
932 struct ceph_buffer
*xattrs_buf
,
935 struct ceph_mds_caps
*fc
;
936 struct ceph_msg
*msg
;
938 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
939 " seq %u/%u mseq %u follows %lld size %llu/%llu"
940 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
941 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
942 ceph_cap_string(dirty
),
943 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
944 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
946 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
), GFP_NOFS
);
950 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
952 fc
= msg
->front
.iov_base
;
953 memset(fc
, 0, sizeof(*fc
));
955 fc
->cap_id
= cpu_to_le64(cid
);
956 fc
->op
= cpu_to_le32(op
);
957 fc
->seq
= cpu_to_le32(seq
);
958 fc
->issue_seq
= cpu_to_le32(issue_seq
);
959 fc
->migrate_seq
= cpu_to_le32(mseq
);
960 fc
->caps
= cpu_to_le32(caps
);
961 fc
->wanted
= cpu_to_le32(wanted
);
962 fc
->dirty
= cpu_to_le32(dirty
);
963 fc
->ino
= cpu_to_le64(ino
);
964 fc
->snap_follows
= cpu_to_le64(follows
);
966 fc
->size
= cpu_to_le64(size
);
967 fc
->max_size
= cpu_to_le64(max_size
);
969 ceph_encode_timespec(&fc
->mtime
, mtime
);
971 ceph_encode_timespec(&fc
->atime
, atime
);
972 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
974 fc
->uid
= cpu_to_le32(uid
);
975 fc
->gid
= cpu_to_le32(gid
);
976 fc
->mode
= cpu_to_le32(mode
);
978 fc
->xattr_version
= cpu_to_le64(xattr_version
);
980 msg
->middle
= ceph_buffer_get(xattrs_buf
);
981 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
982 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
985 ceph_con_send(&session
->s_con
, msg
);
989 static void __queue_cap_release(struct ceph_mds_session
*session
,
990 u64 ino
, u64 cap_id
, u32 migrate_seq
,
993 struct ceph_msg
*msg
;
994 struct ceph_mds_cap_release
*head
;
995 struct ceph_mds_cap_item
*item
;
997 spin_lock(&session
->s_cap_lock
);
998 BUG_ON(!session
->s_num_cap_releases
);
999 msg
= list_first_entry(&session
->s_cap_releases
,
1000 struct ceph_msg
, list_head
);
1002 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1003 ino
, session
->s_mds
, msg
, session
->s_num_cap_releases
);
1005 BUG_ON(msg
->front
.iov_len
+ sizeof(*item
) > PAGE_CACHE_SIZE
);
1006 head
= msg
->front
.iov_base
;
1007 head
->num
= cpu_to_le32(le32_to_cpu(head
->num
) + 1);
1008 item
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1009 item
->ino
= cpu_to_le64(ino
);
1010 item
->cap_id
= cpu_to_le64(cap_id
);
1011 item
->migrate_seq
= cpu_to_le32(migrate_seq
);
1012 item
->seq
= cpu_to_le32(issue_seq
);
1014 session
->s_num_cap_releases
--;
1016 msg
->front
.iov_len
+= sizeof(*item
);
1017 if (le32_to_cpu(head
->num
) == CEPH_CAPS_PER_RELEASE
) {
1018 dout(" release msg %p full\n", msg
);
1019 list_move_tail(&msg
->list_head
, &session
->s_cap_releases_done
);
1021 dout(" release msg %p at %d/%d (%d)\n", msg
,
1022 (int)le32_to_cpu(head
->num
),
1023 (int)CEPH_CAPS_PER_RELEASE
,
1024 (int)msg
->front
.iov_len
);
1026 spin_unlock(&session
->s_cap_lock
);
1030 * Queue cap releases when an inode is dropped from our cache. Since
1031 * inode is about to be destroyed, there is no need for i_lock.
1033 void ceph_queue_caps_release(struct inode
*inode
)
1035 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1038 p
= rb_first(&ci
->i_caps
);
1040 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1041 struct ceph_mds_session
*session
= cap
->session
;
1043 __queue_cap_release(session
, ceph_ino(inode
), cap
->cap_id
,
1044 cap
->mseq
, cap
->issue_seq
);
1046 __ceph_remove_cap(cap
);
1051 * Send a cap msg on the given inode. Update our caps state, then
1052 * drop i_lock and send the message.
1054 * Make note of max_size reported/requested from mds, revoked caps
1055 * that have now been implemented.
1057 * Make half-hearted attempt ot to invalidate page cache if we are
1058 * dropping RDCACHE. Note that this will leave behind locked pages
1059 * that we'll then need to deal with elsewhere.
1061 * Return non-zero if delayed release, or we experienced an error
1062 * such that the caller should requeue + retry later.
1064 * called with i_lock, then drops it.
1065 * caller should hold snap_rwsem (read), s_mutex.
1067 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1068 int op
, int used
, int want
, int retain
, int flushing
,
1069 unsigned *pflush_tid
)
1070 __releases(cap
->ci
->vfs_inode
->i_lock
)
1072 struct ceph_inode_info
*ci
= cap
->ci
;
1073 struct inode
*inode
= &ci
->vfs_inode
;
1074 u64 cap_id
= cap
->cap_id
;
1075 int held
, revoking
, dropping
, keep
;
1076 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1078 struct timespec mtime
, atime
;
1083 struct ceph_mds_session
*session
;
1084 u64 xattr_version
= 0;
1085 struct ceph_buffer
*xattr_blob
= NULL
;
1091 held
= cap
->issued
| cap
->implemented
;
1092 revoking
= cap
->implemented
& ~cap
->issued
;
1093 retain
&= ~revoking
;
1094 dropping
= cap
->issued
& ~retain
;
1096 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1097 inode
, cap
, cap
->session
,
1098 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1099 ceph_cap_string(revoking
));
1100 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1102 session
= cap
->session
;
1104 /* don't release wanted unless we've waited a bit. */
1105 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1106 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1107 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1108 ceph_cap_string(cap
->issued
),
1109 ceph_cap_string(cap
->issued
& retain
),
1110 ceph_cap_string(cap
->mds_wanted
),
1111 ceph_cap_string(want
));
1112 want
|= cap
->mds_wanted
;
1113 retain
|= cap
->issued
;
1116 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1118 cap
->issued
&= retain
; /* drop bits we don't want */
1119 if (cap
->implemented
& ~cap
->issued
) {
1121 * Wake up any waiters on wanted -> needed transition.
1122 * This is due to the weird transition from buffered
1123 * to sync IO... we need to flush dirty pages _before_
1124 * allowing sync writes to avoid reordering.
1128 cap
->implemented
&= cap
->issued
| used
;
1129 cap
->mds_wanted
= want
;
1133 * assign a tid for flush operations so we can avoid
1134 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1135 * clean type races. track latest tid for every bit
1136 * so we can handle flush AxFw, flush Fw, and have the
1137 * first ack clean Ax.
1139 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1141 *pflush_tid
= flush_tid
;
1142 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1143 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1144 if (flushing
& (1 << i
))
1145 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1147 follows
= ci
->i_head_snapc
->seq
;
1152 keep
= cap
->implemented
;
1154 issue_seq
= cap
->issue_seq
;
1156 size
= inode
->i_size
;
1157 ci
->i_reported_size
= size
;
1158 max_size
= ci
->i_wanted_max_size
;
1159 ci
->i_requested_max_size
= max_size
;
1160 mtime
= inode
->i_mtime
;
1161 atime
= inode
->i_atime
;
1162 time_warp_seq
= ci
->i_time_warp_seq
;
1165 mode
= inode
->i_mode
;
1167 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1168 __ceph_build_xattrs_blob(ci
);
1169 xattr_blob
= ci
->i_xattrs
.blob
;
1170 xattr_version
= ci
->i_xattrs
.version
;
1173 spin_unlock(&inode
->i_lock
);
1175 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1176 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1177 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1178 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1181 dout("error sending cap msg, must requeue %p\n", inode
);
1186 wake_up_all(&ci
->i_cap_wq
);
1192 * When a snapshot is taken, clients accumulate dirty metadata on
1193 * inodes with capabilities in ceph_cap_snaps to describe the file
1194 * state at the time the snapshot was taken. This must be flushed
1195 * asynchronously back to the MDS once sync writes complete and dirty
1196 * data is written out.
1198 * Unless @again is true, skip cap_snaps that were already sent to
1199 * the MDS (i.e., during this session).
1201 * Called under i_lock. Takes s_mutex as needed.
1203 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1204 struct ceph_mds_session
**psession
,
1206 __releases(ci
->vfs_inode
->i_lock
)
1207 __acquires(ci
->vfs_inode
->i_lock
)
1209 struct inode
*inode
= &ci
->vfs_inode
;
1211 struct ceph_cap_snap
*capsnap
;
1213 struct ceph_mds_client
*mdsc
= &ceph_inode_to_client(inode
)->mdsc
;
1214 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1216 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1217 i_cap_snaps list, and skip these entries next time
1218 around to avoid an infinite loop */
1221 session
= *psession
;
1223 dout("__flush_snaps %p\n", inode
);
1225 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1226 /* avoid an infiniute loop after retry */
1227 if (capsnap
->follows
< next_follows
)
1230 * we need to wait for sync writes to complete and for dirty
1231 * pages to be written out.
1233 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1237 * if cap writeback already occurred, we should have dropped
1238 * the capsnap in ceph_put_wrbuffer_cap_refs.
1240 BUG_ON(capsnap
->dirty
== 0);
1242 /* pick mds, take s_mutex */
1243 if (ci
->i_auth_cap
== NULL
) {
1244 dout("no auth cap (migrating?), doing nothing\n");
1248 /* only flush each capsnap once */
1249 if (!again
&& !list_empty(&capsnap
->flushing_item
)) {
1250 dout("already flushed %p, skipping\n", capsnap
);
1254 mds
= ci
->i_auth_cap
->session
->s_mds
;
1255 mseq
= ci
->i_auth_cap
->mseq
;
1257 if (session
&& session
->s_mds
!= mds
) {
1258 dout("oops, wrong session %p mutex\n", session
);
1259 mutex_unlock(&session
->s_mutex
);
1260 ceph_put_mds_session(session
);
1264 spin_unlock(&inode
->i_lock
);
1265 mutex_lock(&mdsc
->mutex
);
1266 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1267 mutex_unlock(&mdsc
->mutex
);
1269 dout("inverting session/ino locks on %p\n",
1271 mutex_lock(&session
->s_mutex
);
1274 * if session == NULL, we raced against a cap
1275 * deletion or migration. retry, and we'll
1276 * get a better @mds value next time.
1278 spin_lock(&inode
->i_lock
);
1282 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1283 atomic_inc(&capsnap
->nref
);
1284 if (!list_empty(&capsnap
->flushing_item
))
1285 list_del_init(&capsnap
->flushing_item
);
1286 list_add_tail(&capsnap
->flushing_item
,
1287 &session
->s_cap_snaps_flushing
);
1288 spin_unlock(&inode
->i_lock
);
1290 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1291 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1292 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1293 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1294 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1296 &capsnap
->mtime
, &capsnap
->atime
,
1297 capsnap
->time_warp_seq
,
1298 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1299 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1302 next_follows
= capsnap
->follows
+ 1;
1303 ceph_put_cap_snap(capsnap
);
1305 spin_lock(&inode
->i_lock
);
1309 /* we flushed them all; remove this inode from the queue */
1310 spin_lock(&mdsc
->snap_flush_lock
);
1311 list_del_init(&ci
->i_snap_flush_item
);
1312 spin_unlock(&mdsc
->snap_flush_lock
);
1316 *psession
= session
;
1318 mutex_unlock(&session
->s_mutex
);
1319 ceph_put_mds_session(session
);
1323 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1325 struct inode
*inode
= &ci
->vfs_inode
;
1327 spin_lock(&inode
->i_lock
);
1328 __ceph_flush_snaps(ci
, NULL
, 0);
1329 spin_unlock(&inode
->i_lock
);
1333 * Mark caps dirty. If inode is newly dirty, add to the global dirty
1336 void __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1338 struct ceph_mds_client
*mdsc
=
1339 &ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1340 struct inode
*inode
= &ci
->vfs_inode
;
1341 int was
= ci
->i_dirty_caps
;
1344 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1345 ceph_cap_string(mask
), ceph_cap_string(was
),
1346 ceph_cap_string(was
| mask
));
1347 ci
->i_dirty_caps
|= mask
;
1349 if (!ci
->i_head_snapc
)
1350 ci
->i_head_snapc
= ceph_get_snap_context(
1351 ci
->i_snap_realm
->cached_context
);
1352 dout(" inode %p now dirty snapc %p\n", &ci
->vfs_inode
,
1354 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1355 spin_lock(&mdsc
->cap_dirty_lock
);
1356 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1357 spin_unlock(&mdsc
->cap_dirty_lock
);
1358 if (ci
->i_flushing_caps
== 0) {
1360 dirty
|= I_DIRTY_SYNC
;
1363 BUG_ON(list_empty(&ci
->i_dirty_item
));
1364 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1365 (mask
& CEPH_CAP_FILE_BUFFER
))
1366 dirty
|= I_DIRTY_DATASYNC
;
1368 __mark_inode_dirty(inode
, dirty
);
1369 __cap_delay_requeue(mdsc
, ci
);
1373 * Add dirty inode to the flushing list. Assigned a seq number so we
1374 * can wait for caps to flush without starving.
1376 * Called under i_lock.
1378 static int __mark_caps_flushing(struct inode
*inode
,
1379 struct ceph_mds_session
*session
)
1381 struct ceph_mds_client
*mdsc
= &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1382 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1385 BUG_ON(ci
->i_dirty_caps
== 0);
1386 BUG_ON(list_empty(&ci
->i_dirty_item
));
1388 flushing
= ci
->i_dirty_caps
;
1389 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1390 ceph_cap_string(flushing
),
1391 ceph_cap_string(ci
->i_flushing_caps
),
1392 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1393 ci
->i_flushing_caps
|= flushing
;
1394 ci
->i_dirty_caps
= 0;
1395 dout(" inode %p now !dirty\n", inode
);
1397 spin_lock(&mdsc
->cap_dirty_lock
);
1398 list_del_init(&ci
->i_dirty_item
);
1400 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1401 if (list_empty(&ci
->i_flushing_item
)) {
1402 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1403 mdsc
->num_cap_flushing
++;
1404 dout(" inode %p now flushing seq %lld\n", inode
,
1405 ci
->i_cap_flush_seq
);
1407 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1408 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1409 ci
->i_cap_flush_seq
);
1411 spin_unlock(&mdsc
->cap_dirty_lock
);
1417 * try to invalidate mapping pages without blocking.
1419 static int mapping_is_empty(struct address_space
*mapping
)
1421 struct page
*page
= find_get_page(mapping
, 0);
1430 static int try_nonblocking_invalidate(struct inode
*inode
)
1432 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1433 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1435 spin_unlock(&inode
->i_lock
);
1436 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1437 spin_lock(&inode
->i_lock
);
1439 if (mapping_is_empty(&inode
->i_data
) &&
1440 invalidating_gen
== ci
->i_rdcache_gen
) {
1442 dout("try_nonblocking_invalidate %p success\n", inode
);
1443 ci
->i_rdcache_gen
= 0;
1444 ci
->i_rdcache_revoking
= 0;
1447 dout("try_nonblocking_invalidate %p failed\n", inode
);
1452 * Swiss army knife function to examine currently used and wanted
1453 * versus held caps. Release, flush, ack revoked caps to mds as
1456 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1457 * cap release further.
1458 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1459 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1462 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1463 struct ceph_mds_session
*session
)
1465 struct ceph_client
*client
= ceph_inode_to_client(&ci
->vfs_inode
);
1466 struct ceph_mds_client
*mdsc
= &client
->mdsc
;
1467 struct inode
*inode
= &ci
->vfs_inode
;
1468 struct ceph_cap
*cap
;
1469 int file_wanted
, used
;
1470 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1471 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1472 int mds
= -1; /* keep track of how far we've gone through i_caps list
1473 to avoid an infinite loop on retry */
1475 int tried_invalidate
= 0;
1476 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1477 int queue_invalidate
= 0;
1478 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1480 /* if we are unmounting, flush any unused caps immediately. */
1484 spin_lock(&inode
->i_lock
);
1486 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1487 flags
|= CHECK_CAPS_FLUSH
;
1489 /* flush snaps first time around only */
1490 if (!list_empty(&ci
->i_cap_snaps
))
1491 __ceph_flush_snaps(ci
, &session
, 0);
1494 spin_lock(&inode
->i_lock
);
1496 file_wanted
= __ceph_caps_file_wanted(ci
);
1497 used
= __ceph_caps_used(ci
);
1498 want
= file_wanted
| used
;
1499 issued
= __ceph_caps_issued(ci
, &implemented
);
1500 revoking
= implemented
& ~issued
;
1502 retain
= want
| CEPH_CAP_PIN
;
1503 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1505 retain
|= CEPH_CAP_ANY
; /* be greedy */
1507 retain
|= CEPH_CAP_ANY_SHARED
;
1509 * keep RD only if we didn't have the file open RW,
1510 * because then the mds would revoke it anyway to
1511 * journal max_size=0.
1513 if (ci
->i_max_size
== 0)
1514 retain
|= CEPH_CAP_ANY_RD
;
1518 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1519 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1520 ceph_cap_string(file_wanted
),
1521 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1522 ceph_cap_string(ci
->i_flushing_caps
),
1523 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1524 ceph_cap_string(retain
),
1525 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1526 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1527 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1530 * If we no longer need to hold onto old our caps, and we may
1531 * have cached pages, but don't want them, then try to invalidate.
1532 * If we fail, it's because pages are locked.... try again later.
1534 if ((!is_delayed
|| mdsc
->stopping
) &&
1535 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1536 ci
->i_rdcache_gen
&& /* may have cached pages */
1537 (file_wanted
== 0 || /* no open files */
1538 (revoking
& (CEPH_CAP_FILE_CACHE
|
1539 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1540 !tried_invalidate
) {
1541 dout("check_caps trying to invalidate on %p\n", inode
);
1542 if (try_nonblocking_invalidate(inode
) < 0) {
1543 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1544 CEPH_CAP_FILE_LAZYIO
)) {
1545 dout("check_caps queuing invalidate\n");
1546 queue_invalidate
= 1;
1547 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1549 dout("check_caps failed to invalidate pages\n");
1550 /* we failed to invalidate pages. check these
1551 caps again later. */
1553 __cap_set_timeouts(mdsc
, ci
);
1556 tried_invalidate
= 1;
1561 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1562 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1565 /* avoid looping forever */
1566 if (mds
>= cap
->mds
||
1567 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1570 /* NOTE: no side-effects allowed, until we take s_mutex */
1572 revoking
= cap
->implemented
& ~cap
->issued
;
1574 dout(" mds%d revoking %s\n", cap
->mds
,
1575 ceph_cap_string(revoking
));
1577 if (cap
== ci
->i_auth_cap
&&
1578 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1579 /* request larger max_size from MDS? */
1580 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1581 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1582 dout("requesting new max_size\n");
1586 /* approaching file_max? */
1587 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1588 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1589 dout("i_size approaching max_size\n");
1593 /* flush anything dirty? */
1594 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1596 dout("flushing dirty caps\n");
1600 /* completed revocation? going down and there are no caps? */
1601 if (revoking
&& (revoking
& used
) == 0) {
1602 dout("completed revocation of %s\n",
1603 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1607 /* want more caps from mds? */
1608 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1611 /* things we might delay */
1612 if ((cap
->issued
& ~retain
) == 0 &&
1613 cap
->mds_wanted
== want
)
1614 continue; /* nope, all good */
1620 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1621 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1622 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1623 ceph_cap_string(cap
->issued
),
1624 ceph_cap_string(cap
->issued
& retain
),
1625 ceph_cap_string(cap
->mds_wanted
),
1626 ceph_cap_string(want
));
1632 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1633 dout(" skipping %p I_NOFLUSH set\n", inode
);
1637 if (session
&& session
!= cap
->session
) {
1638 dout("oops, wrong session %p mutex\n", session
);
1639 mutex_unlock(&session
->s_mutex
);
1643 session
= cap
->session
;
1644 if (mutex_trylock(&session
->s_mutex
) == 0) {
1645 dout("inverting session/ino locks on %p\n",
1647 spin_unlock(&inode
->i_lock
);
1648 if (took_snap_rwsem
) {
1649 up_read(&mdsc
->snap_rwsem
);
1650 took_snap_rwsem
= 0;
1652 mutex_lock(&session
->s_mutex
);
1656 /* take snap_rwsem after session mutex */
1657 if (!took_snap_rwsem
) {
1658 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1659 dout("inverting snap/in locks on %p\n",
1661 spin_unlock(&inode
->i_lock
);
1662 down_read(&mdsc
->snap_rwsem
);
1663 took_snap_rwsem
= 1;
1666 took_snap_rwsem
= 1;
1669 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1670 flushing
= __mark_caps_flushing(inode
, session
);
1672 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1675 /* __send_cap drops i_lock */
1676 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, used
, want
,
1677 retain
, flushing
, NULL
);
1678 goto retry
; /* retake i_lock and restart our cap scan. */
1682 * Reschedule delayed caps release if we delayed anything,
1685 if (delayed
&& is_delayed
)
1686 force_requeue
= 1; /* __send_cap delayed release; requeue */
1687 if (!delayed
&& !is_delayed
)
1688 __cap_delay_cancel(mdsc
, ci
);
1689 else if (!is_delayed
|| force_requeue
)
1690 __cap_delay_requeue(mdsc
, ci
);
1692 spin_unlock(&inode
->i_lock
);
1694 if (queue_invalidate
)
1695 ceph_queue_invalidate(inode
);
1698 mutex_unlock(&session
->s_mutex
);
1699 if (took_snap_rwsem
)
1700 up_read(&mdsc
->snap_rwsem
);
1704 * Try to flush dirty caps back to the auth mds.
1706 static int try_flush_caps(struct inode
*inode
, struct ceph_mds_session
*session
,
1707 unsigned *flush_tid
)
1709 struct ceph_mds_client
*mdsc
= &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1710 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1711 int unlock_session
= session
? 0 : 1;
1715 spin_lock(&inode
->i_lock
);
1716 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1717 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1720 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1721 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1722 int used
= __ceph_caps_used(ci
);
1723 int want
= __ceph_caps_wanted(ci
);
1727 spin_unlock(&inode
->i_lock
);
1728 session
= cap
->session
;
1729 mutex_lock(&session
->s_mutex
);
1732 BUG_ON(session
!= cap
->session
);
1733 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1736 flushing
= __mark_caps_flushing(inode
, session
);
1738 /* __send_cap drops i_lock */
1739 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1740 cap
->issued
| cap
->implemented
, flushing
,
1745 spin_lock(&inode
->i_lock
);
1746 __cap_delay_requeue(mdsc
, ci
);
1749 spin_unlock(&inode
->i_lock
);
1751 if (session
&& unlock_session
)
1752 mutex_unlock(&session
->s_mutex
);
1757 * Return true if we've flushed caps through the given flush_tid.
1759 static int caps_are_flushed(struct inode
*inode
, unsigned tid
)
1761 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1764 spin_lock(&inode
->i_lock
);
1765 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1766 if ((ci
->i_flushing_caps
& (1 << i
)) &&
1767 ci
->i_cap_flush_tid
[i
] <= tid
) {
1768 /* still flushing this bit */
1772 spin_unlock(&inode
->i_lock
);
1777 * Wait on any unsafe replies for the given inode. First wait on the
1778 * newest request, and make that the upper bound. Then, if there are
1779 * more requests, keep waiting on the oldest as long as it is still older
1780 * than the original request.
1782 static void sync_write_wait(struct inode
*inode
)
1784 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1785 struct list_head
*head
= &ci
->i_unsafe_writes
;
1786 struct ceph_osd_request
*req
;
1789 spin_lock(&ci
->i_unsafe_lock
);
1790 if (list_empty(head
))
1793 /* set upper bound as _last_ entry in chain */
1794 req
= list_entry(head
->prev
, struct ceph_osd_request
,
1796 last_tid
= req
->r_tid
;
1799 ceph_osdc_get_request(req
);
1800 spin_unlock(&ci
->i_unsafe_lock
);
1801 dout("sync_write_wait on tid %llu (until %llu)\n",
1802 req
->r_tid
, last_tid
);
1803 wait_for_completion(&req
->r_safe_completion
);
1804 spin_lock(&ci
->i_unsafe_lock
);
1805 ceph_osdc_put_request(req
);
1808 * from here on look at first entry in chain, since we
1809 * only want to wait for anything older than last_tid
1811 if (list_empty(head
))
1813 req
= list_entry(head
->next
, struct ceph_osd_request
,
1815 } while (req
->r_tid
< last_tid
);
1817 spin_unlock(&ci
->i_unsafe_lock
);
1820 int ceph_fsync(struct file
*file
, int datasync
)
1822 struct inode
*inode
= file
->f_mapping
->host
;
1823 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1828 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1829 sync_write_wait(inode
);
1831 ret
= filemap_write_and_wait(inode
->i_mapping
);
1835 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1836 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1839 * only wait on non-file metadata writeback (the mds
1840 * can recover size and mtime, so we don't need to
1843 if (!datasync
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1844 dout("fsync waiting for flush_tid %u\n", flush_tid
);
1845 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1846 caps_are_flushed(inode
, flush_tid
));
1849 dout("fsync %p%s done\n", inode
, datasync
? " datasync" : "");
1854 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1855 * queue inode for flush but don't do so immediately, because we can
1856 * get by with fewer MDS messages if we wait for data writeback to
1859 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1861 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1865 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1867 dout("write_inode %p wait=%d\n", inode
, wait
);
1869 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1871 err
= wait_event_interruptible(ci
->i_cap_wq
,
1872 caps_are_flushed(inode
, flush_tid
));
1874 struct ceph_mds_client
*mdsc
=
1875 &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1877 spin_lock(&inode
->i_lock
);
1878 if (__ceph_caps_dirty(ci
))
1879 __cap_delay_requeue_front(mdsc
, ci
);
1880 spin_unlock(&inode
->i_lock
);
1886 * After a recovering MDS goes active, we need to resend any caps
1889 * Caller holds session->s_mutex.
1891 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
1892 struct ceph_mds_session
*session
)
1894 struct ceph_cap_snap
*capsnap
;
1896 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
1897 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
1899 struct ceph_inode_info
*ci
= capsnap
->ci
;
1900 struct inode
*inode
= &ci
->vfs_inode
;
1901 struct ceph_cap
*cap
;
1903 spin_lock(&inode
->i_lock
);
1904 cap
= ci
->i_auth_cap
;
1905 if (cap
&& cap
->session
== session
) {
1906 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
1908 __ceph_flush_snaps(ci
, &session
, 1);
1910 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1911 cap
, session
->s_mds
);
1913 spin_unlock(&inode
->i_lock
);
1917 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
1918 struct ceph_mds_session
*session
)
1920 struct ceph_inode_info
*ci
;
1922 kick_flushing_capsnaps(mdsc
, session
);
1924 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
1925 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
1926 struct inode
*inode
= &ci
->vfs_inode
;
1927 struct ceph_cap
*cap
;
1930 spin_lock(&inode
->i_lock
);
1931 cap
= ci
->i_auth_cap
;
1932 if (cap
&& cap
->session
== session
) {
1933 dout("kick_flushing_caps %p cap %p %s\n", inode
,
1934 cap
, ceph_cap_string(ci
->i_flushing_caps
));
1935 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1936 __ceph_caps_used(ci
),
1937 __ceph_caps_wanted(ci
),
1938 cap
->issued
| cap
->implemented
,
1939 ci
->i_flushing_caps
, NULL
);
1941 spin_lock(&inode
->i_lock
);
1942 __cap_delay_requeue(mdsc
, ci
);
1943 spin_unlock(&inode
->i_lock
);
1946 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1947 cap
, session
->s_mds
);
1948 spin_unlock(&inode
->i_lock
);
1955 * Take references to capabilities we hold, so that we don't release
1956 * them to the MDS prematurely.
1958 * Protected by i_lock.
1960 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
)
1962 if (got
& CEPH_CAP_PIN
)
1964 if (got
& CEPH_CAP_FILE_RD
)
1966 if (got
& CEPH_CAP_FILE_CACHE
)
1967 ci
->i_rdcache_ref
++;
1968 if (got
& CEPH_CAP_FILE_WR
)
1970 if (got
& CEPH_CAP_FILE_BUFFER
) {
1971 if (ci
->i_wrbuffer_ref
== 0)
1972 igrab(&ci
->vfs_inode
);
1973 ci
->i_wrbuffer_ref
++;
1974 dout("__take_cap_refs %p wrbuffer %d -> %d (?)\n",
1975 &ci
->vfs_inode
, ci
->i_wrbuffer_ref
-1, ci
->i_wrbuffer_ref
);
1980 * Try to grab cap references. Specify those refs we @want, and the
1981 * minimal set we @need. Also include the larger offset we are writing
1982 * to (when applicable), and check against max_size here as well.
1983 * Note that caller is responsible for ensuring max_size increases are
1984 * requested from the MDS.
1986 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
1987 int *got
, loff_t endoff
, int *check_max
, int *err
)
1989 struct inode
*inode
= &ci
->vfs_inode
;
1991 int have
, implemented
;
1994 dout("get_cap_refs %p need %s want %s\n", inode
,
1995 ceph_cap_string(need
), ceph_cap_string(want
));
1996 spin_lock(&inode
->i_lock
);
1998 /* make sure file is actually open */
1999 file_wanted
= __ceph_caps_file_wanted(ci
);
2000 if ((file_wanted
& need
) == 0) {
2001 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2002 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2008 if (need
& CEPH_CAP_FILE_WR
) {
2009 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2010 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2011 inode
, endoff
, ci
->i_max_size
);
2012 if (endoff
> ci
->i_wanted_max_size
) {
2019 * If a sync write is in progress, we must wait, so that we
2020 * can get a final snapshot value for size+mtime.
2022 if (__ceph_have_pending_cap_snap(ci
)) {
2023 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2027 have
= __ceph_caps_issued(ci
, &implemented
);
2030 * disallow writes while a truncate is pending
2032 if (ci
->i_truncate_pending
)
2033 have
&= ~CEPH_CAP_FILE_WR
;
2035 if ((have
& need
) == need
) {
2037 * Look at (implemented & ~have & not) so that we keep waiting
2038 * on transition from wanted -> needed caps. This is needed
2039 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2040 * going before a prior buffered writeback happens.
2042 int not = want
& ~(have
& need
);
2043 int revoking
= implemented
& ~have
;
2044 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2045 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2046 ceph_cap_string(revoking
));
2047 if ((revoking
& not) == 0) {
2048 *got
= need
| (have
& want
);
2049 __take_cap_refs(ci
, *got
);
2053 dout("get_cap_refs %p have %s needed %s\n", inode
,
2054 ceph_cap_string(have
), ceph_cap_string(need
));
2057 spin_unlock(&inode
->i_lock
);
2058 dout("get_cap_refs %p ret %d got %s\n", inode
,
2059 ret
, ceph_cap_string(*got
));
2064 * Check the offset we are writing up to against our current
2065 * max_size. If necessary, tell the MDS we want to write to
2068 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2070 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2073 /* do we need to explicitly request a larger max_size? */
2074 spin_lock(&inode
->i_lock
);
2075 if ((endoff
>= ci
->i_max_size
||
2076 endoff
> (inode
->i_size
<< 1)) &&
2077 endoff
> ci
->i_wanted_max_size
) {
2078 dout("write %p at large endoff %llu, req max_size\n",
2080 ci
->i_wanted_max_size
= endoff
;
2083 spin_unlock(&inode
->i_lock
);
2085 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2089 * Wait for caps, and take cap references. If we can't get a WR cap
2090 * due to a small max_size, make sure we check_max_size (and possibly
2091 * ask the mds) so we don't get hung up indefinitely.
2093 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
, int *got
,
2096 int check_max
, ret
, err
;
2100 check_max_size(&ci
->vfs_inode
, endoff
);
2103 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2104 try_get_cap_refs(ci
, need
, want
,
2115 * Take cap refs. Caller must already know we hold at least one ref
2116 * on the caps in question or we don't know this is safe.
2118 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2120 spin_lock(&ci
->vfs_inode
.i_lock
);
2121 __take_cap_refs(ci
, caps
);
2122 spin_unlock(&ci
->vfs_inode
.i_lock
);
2128 * If we released the last ref on any given cap, call ceph_check_caps
2129 * to release (or schedule a release).
2131 * If we are releasing a WR cap (from a sync write), finalize any affected
2132 * cap_snap, and wake up any waiters.
2134 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2136 struct inode
*inode
= &ci
->vfs_inode
;
2137 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2138 struct ceph_cap_snap
*capsnap
;
2140 spin_lock(&inode
->i_lock
);
2141 if (had
& CEPH_CAP_PIN
)
2143 if (had
& CEPH_CAP_FILE_RD
)
2144 if (--ci
->i_rd_ref
== 0)
2146 if (had
& CEPH_CAP_FILE_CACHE
)
2147 if (--ci
->i_rdcache_ref
== 0)
2149 if (had
& CEPH_CAP_FILE_BUFFER
) {
2150 if (--ci
->i_wrbuffer_ref
== 0) {
2154 dout("put_cap_refs %p wrbuffer %d -> %d (?)\n",
2155 inode
, ci
->i_wrbuffer_ref
+1, ci
->i_wrbuffer_ref
);
2157 if (had
& CEPH_CAP_FILE_WR
)
2158 if (--ci
->i_wr_ref
== 0) {
2160 if (!list_empty(&ci
->i_cap_snaps
)) {
2161 capsnap
= list_first_entry(&ci
->i_cap_snaps
,
2162 struct ceph_cap_snap
,
2164 if (capsnap
->writing
) {
2165 capsnap
->writing
= 0;
2167 __ceph_finish_cap_snap(ci
,
2173 spin_unlock(&inode
->i_lock
);
2175 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2176 last
? " last" : "", put
? " put" : "");
2178 if (last
&& !flushsnaps
)
2179 ceph_check_caps(ci
, 0, NULL
);
2180 else if (flushsnaps
)
2181 ceph_flush_snaps(ci
);
2183 wake_up_all(&ci
->i_cap_wq
);
2189 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2190 * context. Adjust per-snap dirty page accounting as appropriate.
2191 * Once all dirty data for a cap_snap is flushed, flush snapped file
2192 * metadata back to the MDS. If we dropped the last ref, call
2195 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2196 struct ceph_snap_context
*snapc
)
2198 struct inode
*inode
= &ci
->vfs_inode
;
2200 int complete_capsnap
= 0;
2201 int drop_capsnap
= 0;
2203 struct ceph_cap_snap
*capsnap
= NULL
;
2205 spin_lock(&inode
->i_lock
);
2206 ci
->i_wrbuffer_ref
-= nr
;
2207 last
= !ci
->i_wrbuffer_ref
;
2209 if (ci
->i_head_snapc
== snapc
) {
2210 ci
->i_wrbuffer_ref_head
-= nr
;
2211 if (ci
->i_wrbuffer_ref_head
== 0 &&
2212 ci
->i_dirty_caps
== 0 && ci
->i_flushing_caps
== 0) {
2213 BUG_ON(!ci
->i_head_snapc
);
2214 ceph_put_snap_context(ci
->i_head_snapc
);
2215 ci
->i_head_snapc
= NULL
;
2217 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2219 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2220 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2221 last
? " LAST" : "");
2223 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2224 if (capsnap
->context
== snapc
) {
2230 capsnap
->dirty_pages
-= nr
;
2231 if (capsnap
->dirty_pages
== 0) {
2232 complete_capsnap
= 1;
2233 if (capsnap
->dirty
== 0)
2234 /* cap writeback completed before we created
2235 * the cap_snap; no FLUSHSNAP is needed */
2238 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2239 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2240 inode
, capsnap
, capsnap
->context
->seq
,
2241 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2242 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2243 last
? " (wrbuffer last)" : "",
2244 complete_capsnap
? " (complete capsnap)" : "",
2245 drop_capsnap
? " (drop capsnap)" : "");
2247 ceph_put_snap_context(capsnap
->context
);
2248 list_del(&capsnap
->ci_item
);
2249 list_del(&capsnap
->flushing_item
);
2250 ceph_put_cap_snap(capsnap
);
2254 spin_unlock(&inode
->i_lock
);
2257 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2259 } else if (complete_capsnap
) {
2260 ceph_flush_snaps(ci
);
2261 wake_up_all(&ci
->i_cap_wq
);
2268 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2269 * actually be a revocation if it specifies a smaller cap set.)
2271 * caller holds s_mutex and i_lock, we drop both.
2275 * 1 - check_caps on auth cap only (writeback)
2276 * 2 - check_caps (ack revoke)
2278 static void handle_cap_grant(struct inode
*inode
, struct ceph_mds_caps
*grant
,
2279 struct ceph_mds_session
*session
,
2280 struct ceph_cap
*cap
,
2281 struct ceph_buffer
*xattr_buf
)
2282 __releases(inode
->i_lock
)
2284 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2285 int mds
= session
->s_mds
;
2286 int seq
= le32_to_cpu(grant
->seq
);
2287 int newcaps
= le32_to_cpu(grant
->caps
);
2288 int issued
, implemented
, used
, wanted
, dirty
;
2289 u64 size
= le64_to_cpu(grant
->size
);
2290 u64 max_size
= le64_to_cpu(grant
->max_size
);
2291 struct timespec mtime
, atime
, ctime
;
2295 int revoked_rdcache
= 0;
2296 int queue_invalidate
= 0;
2298 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2299 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2300 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2304 * If CACHE is being revoked, and we have no dirty buffers,
2305 * try to invalidate (once). (If there are dirty buffers, we
2306 * will invalidate _after_ writeback.)
2308 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2309 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2310 !ci
->i_wrbuffer_ref
) {
2311 if (try_nonblocking_invalidate(inode
) == 0) {
2312 revoked_rdcache
= 1;
2314 /* there were locked pages.. invalidate later
2315 in a separate thread. */
2316 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2317 queue_invalidate
= 1;
2318 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2323 /* side effects now are allowed */
2325 issued
= __ceph_caps_issued(ci
, &implemented
);
2326 issued
|= implemented
| __ceph_caps_dirty(ci
);
2328 cap
->cap_gen
= session
->s_cap_gen
;
2330 __check_cap_issue(ci
, cap
, newcaps
);
2332 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2333 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2334 inode
->i_uid
= le32_to_cpu(grant
->uid
);
2335 inode
->i_gid
= le32_to_cpu(grant
->gid
);
2336 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2337 inode
->i_uid
, inode
->i_gid
);
2340 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
2341 inode
->i_nlink
= le32_to_cpu(grant
->nlink
);
2343 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2344 int len
= le32_to_cpu(grant
->xattr_len
);
2345 u64 version
= le64_to_cpu(grant
->xattr_version
);
2347 if (version
> ci
->i_xattrs
.version
) {
2348 dout(" got new xattrs v%llu on %p len %d\n",
2349 version
, inode
, len
);
2350 if (ci
->i_xattrs
.blob
)
2351 ceph_buffer_put(ci
->i_xattrs
.blob
);
2352 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2353 ci
->i_xattrs
.version
= version
;
2357 /* size/ctime/mtime/atime? */
2358 ceph_fill_file_size(inode
, issued
,
2359 le32_to_cpu(grant
->truncate_seq
),
2360 le64_to_cpu(grant
->truncate_size
), size
);
2361 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2362 ceph_decode_timespec(&atime
, &grant
->atime
);
2363 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2364 ceph_fill_file_time(inode
, issued
,
2365 le32_to_cpu(grant
->time_warp_seq
), &ctime
, &mtime
,
2368 /* max size increase? */
2369 if (max_size
!= ci
->i_max_size
) {
2370 dout("max_size %lld -> %llu\n", ci
->i_max_size
, max_size
);
2371 ci
->i_max_size
= max_size
;
2372 if (max_size
>= ci
->i_wanted_max_size
) {
2373 ci
->i_wanted_max_size
= 0; /* reset */
2374 ci
->i_requested_max_size
= 0;
2379 /* check cap bits */
2380 wanted
= __ceph_caps_wanted(ci
);
2381 used
= __ceph_caps_used(ci
);
2382 dirty
= __ceph_caps_dirty(ci
);
2383 dout(" my wanted = %s, used = %s, dirty %s\n",
2384 ceph_cap_string(wanted
),
2385 ceph_cap_string(used
),
2386 ceph_cap_string(dirty
));
2387 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2388 dout("mds wanted %s -> %s\n",
2389 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2390 ceph_cap_string(wanted
));
2391 grant
->wanted
= cpu_to_le32(wanted
);
2396 /* file layout may have changed */
2397 ci
->i_layout
= grant
->layout
;
2399 /* revocation, grant, or no-op? */
2400 if (cap
->issued
& ~newcaps
) {
2401 int revoking
= cap
->issued
& ~newcaps
;
2403 dout("revocation: %s -> %s (revoking %s)\n",
2404 ceph_cap_string(cap
->issued
),
2405 ceph_cap_string(newcaps
),
2406 ceph_cap_string(revoking
));
2407 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2408 writeback
= 1; /* initiate writeback; will delay ack */
2409 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2410 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2412 ; /* do nothing yet, invalidation will be queued */
2413 else if (cap
== ci
->i_auth_cap
)
2414 check_caps
= 1; /* check auth cap only */
2416 check_caps
= 2; /* check all caps */
2417 cap
->issued
= newcaps
;
2418 cap
->implemented
|= newcaps
;
2419 } else if (cap
->issued
== newcaps
) {
2420 dout("caps unchanged: %s -> %s\n",
2421 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2423 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2424 ceph_cap_string(newcaps
));
2425 cap
->issued
= newcaps
;
2426 cap
->implemented
|= newcaps
; /* add bits only, to
2427 * avoid stepping on a
2428 * pending revocation */
2431 BUG_ON(cap
->issued
& ~cap
->implemented
);
2433 spin_unlock(&inode
->i_lock
);
2436 * queue inode for writeback: we can't actually call
2437 * filemap_write_and_wait, etc. from message handler
2440 ceph_queue_writeback(inode
);
2441 if (queue_invalidate
)
2442 ceph_queue_invalidate(inode
);
2444 wake_up_all(&ci
->i_cap_wq
);
2446 if (check_caps
== 1)
2447 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2449 else if (check_caps
== 2)
2450 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2452 mutex_unlock(&session
->s_mutex
);
2456 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2457 * MDS has been safely committed.
2459 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2460 struct ceph_mds_caps
*m
,
2461 struct ceph_mds_session
*session
,
2462 struct ceph_cap
*cap
)
2463 __releases(inode
->i_lock
)
2465 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2466 struct ceph_mds_client
*mdsc
= &ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2467 unsigned seq
= le32_to_cpu(m
->seq
);
2468 int dirty
= le32_to_cpu(m
->dirty
);
2473 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2474 if ((dirty
& (1 << i
)) &&
2475 flush_tid
== ci
->i_cap_flush_tid
[i
])
2478 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2479 " flushing %s -> %s\n",
2480 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2481 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2482 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2484 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2487 ci
->i_flushing_caps
&= ~cleaned
;
2489 spin_lock(&mdsc
->cap_dirty_lock
);
2490 if (ci
->i_flushing_caps
== 0) {
2491 list_del_init(&ci
->i_flushing_item
);
2492 if (!list_empty(&session
->s_cap_flushing
))
2493 dout(" mds%d still flushing cap on %p\n",
2495 &list_entry(session
->s_cap_flushing
.next
,
2496 struct ceph_inode_info
,
2497 i_flushing_item
)->vfs_inode
);
2498 mdsc
->num_cap_flushing
--;
2499 wake_up_all(&mdsc
->cap_flushing_wq
);
2500 dout(" inode %p now !flushing\n", inode
);
2502 if (ci
->i_dirty_caps
== 0) {
2503 dout(" inode %p now clean\n", inode
);
2504 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2506 if (ci
->i_wrbuffer_ref_head
== 0) {
2507 BUG_ON(!ci
->i_head_snapc
);
2508 ceph_put_snap_context(ci
->i_head_snapc
);
2509 ci
->i_head_snapc
= NULL
;
2512 BUG_ON(list_empty(&ci
->i_dirty_item
));
2515 spin_unlock(&mdsc
->cap_dirty_lock
);
2516 wake_up_all(&ci
->i_cap_wq
);
2519 spin_unlock(&inode
->i_lock
);
2525 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2526 * throw away our cap_snap.
2528 * Caller hold s_mutex.
2530 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2531 struct ceph_mds_caps
*m
,
2532 struct ceph_mds_session
*session
)
2534 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2535 u64 follows
= le64_to_cpu(m
->snap_follows
);
2536 struct ceph_cap_snap
*capsnap
;
2539 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2540 inode
, ci
, session
->s_mds
, follows
);
2542 spin_lock(&inode
->i_lock
);
2543 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2544 if (capsnap
->follows
== follows
) {
2545 if (capsnap
->flush_tid
!= flush_tid
) {
2546 dout(" cap_snap %p follows %lld tid %lld !="
2547 " %lld\n", capsnap
, follows
,
2548 flush_tid
, capsnap
->flush_tid
);
2551 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2552 dout(" removing %p cap_snap %p follows %lld\n",
2553 inode
, capsnap
, follows
);
2554 ceph_put_snap_context(capsnap
->context
);
2555 list_del(&capsnap
->ci_item
);
2556 list_del(&capsnap
->flushing_item
);
2557 ceph_put_cap_snap(capsnap
);
2561 dout(" skipping cap_snap %p follows %lld\n",
2562 capsnap
, capsnap
->follows
);
2565 spin_unlock(&inode
->i_lock
);
2571 * Handle TRUNC from MDS, indicating file truncation.
2573 * caller hold s_mutex.
2575 static void handle_cap_trunc(struct inode
*inode
,
2576 struct ceph_mds_caps
*trunc
,
2577 struct ceph_mds_session
*session
)
2578 __releases(inode
->i_lock
)
2580 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2581 int mds
= session
->s_mds
;
2582 int seq
= le32_to_cpu(trunc
->seq
);
2583 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2584 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2585 u64 size
= le64_to_cpu(trunc
->size
);
2586 int implemented
= 0;
2587 int dirty
= __ceph_caps_dirty(ci
);
2588 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2589 int queue_trunc
= 0;
2591 issued
|= implemented
| dirty
;
2593 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2594 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2595 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2596 truncate_seq
, truncate_size
, size
);
2597 spin_unlock(&inode
->i_lock
);
2600 ceph_queue_vmtruncate(inode
);
2604 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2605 * different one. If we are the most recent migration we've seen (as
2606 * indicated by mseq), make note of the migrating cap bits for the
2607 * duration (until we see the corresponding IMPORT).
2609 * caller holds s_mutex
2611 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2612 struct ceph_mds_session
*session
,
2613 int *open_target_sessions
)
2615 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2616 int mds
= session
->s_mds
;
2617 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2618 struct ceph_cap
*cap
= NULL
, *t
;
2622 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2623 inode
, ci
, mds
, mseq
);
2625 spin_lock(&inode
->i_lock
);
2627 /* make sure we haven't seen a higher mseq */
2628 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
2629 t
= rb_entry(p
, struct ceph_cap
, ci_node
);
2630 if (ceph_seq_cmp(t
->mseq
, mseq
) > 0) {
2631 dout(" higher mseq on cap from mds%d\n",
2635 if (t
->session
->s_mds
== mds
)
2642 ci
->i_cap_exporting_mds
= mds
;
2643 ci
->i_cap_exporting_mseq
= mseq
;
2644 ci
->i_cap_exporting_issued
= cap
->issued
;
2647 * make sure we have open sessions with all possible
2648 * export targets, so that we get the matching IMPORT
2650 *open_target_sessions
= 1;
2652 __ceph_remove_cap(cap
);
2654 /* else, we already released it */
2656 spin_unlock(&inode
->i_lock
);
2660 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2663 * caller holds s_mutex.
2665 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
2666 struct inode
*inode
, struct ceph_mds_caps
*im
,
2667 struct ceph_mds_session
*session
,
2668 void *snaptrace
, int snaptrace_len
)
2670 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2671 int mds
= session
->s_mds
;
2672 unsigned issued
= le32_to_cpu(im
->caps
);
2673 unsigned wanted
= le32_to_cpu(im
->wanted
);
2674 unsigned seq
= le32_to_cpu(im
->seq
);
2675 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
2676 u64 realmino
= le64_to_cpu(im
->realm
);
2677 u64 cap_id
= le64_to_cpu(im
->cap_id
);
2679 if (ci
->i_cap_exporting_mds
>= 0 &&
2680 ceph_seq_cmp(ci
->i_cap_exporting_mseq
, mseq
) < 0) {
2681 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2682 " - cleared exporting from mds%d\n",
2683 inode
, ci
, mds
, mseq
,
2684 ci
->i_cap_exporting_mds
);
2685 ci
->i_cap_exporting_issued
= 0;
2686 ci
->i_cap_exporting_mseq
= 0;
2687 ci
->i_cap_exporting_mds
= -1;
2689 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2690 inode
, ci
, mds
, mseq
);
2693 down_write(&mdsc
->snap_rwsem
);
2694 ceph_update_snap_trace(mdsc
, snaptrace
, snaptrace
+snaptrace_len
,
2696 downgrade_write(&mdsc
->snap_rwsem
);
2697 ceph_add_cap(inode
, session
, cap_id
, -1,
2698 issued
, wanted
, seq
, mseq
, realmino
, CEPH_CAP_FLAG_AUTH
,
2699 NULL
/* no caps context */);
2700 try_flush_caps(inode
, session
, NULL
);
2701 up_read(&mdsc
->snap_rwsem
);
2705 * Handle a caps message from the MDS.
2707 * Identify the appropriate session, inode, and call the right handler
2708 * based on the cap op.
2710 void ceph_handle_caps(struct ceph_mds_session
*session
,
2711 struct ceph_msg
*msg
)
2713 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2714 struct super_block
*sb
= mdsc
->client
->sb
;
2715 struct inode
*inode
;
2716 struct ceph_cap
*cap
;
2717 struct ceph_mds_caps
*h
;
2718 int mds
= session
->s_mds
;
2721 struct ceph_vino vino
;
2726 size_t snaptrace_len
;
2729 int open_target_sessions
= 0;
2731 dout("handle_caps from mds%d\n", mds
);
2734 tid
= le64_to_cpu(msg
->hdr
.tid
);
2735 if (msg
->front
.iov_len
< sizeof(*h
))
2737 h
= msg
->front
.iov_base
;
2738 op
= le32_to_cpu(h
->op
);
2739 vino
.ino
= le64_to_cpu(h
->ino
);
2740 vino
.snap
= CEPH_NOSNAP
;
2741 cap_id
= le64_to_cpu(h
->cap_id
);
2742 seq
= le32_to_cpu(h
->seq
);
2743 mseq
= le32_to_cpu(h
->migrate_seq
);
2744 size
= le64_to_cpu(h
->size
);
2745 max_size
= le64_to_cpu(h
->max_size
);
2748 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
2750 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
2753 p
= snaptrace
+ snaptrace_len
;
2754 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
2755 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
2762 mutex_lock(&session
->s_mutex
);
2764 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
2768 inode
= ceph_find_inode(sb
, vino
);
2769 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
2772 dout(" i don't have ino %llx\n", vino
.ino
);
2774 if (op
== CEPH_CAP_OP_IMPORT
)
2775 __queue_cap_release(session
, vino
.ino
, cap_id
,
2779 * send any full release message to try to move things
2780 * along for the mds (who clearly thinks we still have this
2783 ceph_add_cap_releases(mdsc
, session
);
2784 ceph_send_cap_releases(mdsc
, session
);
2788 /* these will work even if we don't have a cap yet */
2790 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
2791 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
2794 case CEPH_CAP_OP_EXPORT
:
2795 handle_cap_export(inode
, h
, session
, &open_target_sessions
);
2798 case CEPH_CAP_OP_IMPORT
:
2799 handle_cap_import(mdsc
, inode
, h
, session
,
2800 snaptrace
, snaptrace_len
);
2801 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_NODELAY
,
2806 /* the rest require a cap */
2807 spin_lock(&inode
->i_lock
);
2808 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
2810 dout(" no cap on %p ino %llx.%llx from mds%d\n",
2811 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
2812 spin_unlock(&inode
->i_lock
);
2816 /* note that each of these drops i_lock for us */
2818 case CEPH_CAP_OP_REVOKE
:
2819 case CEPH_CAP_OP_GRANT
:
2820 handle_cap_grant(inode
, h
, session
, cap
, msg
->middle
);
2823 case CEPH_CAP_OP_FLUSH_ACK
:
2824 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
2827 case CEPH_CAP_OP_TRUNC
:
2828 handle_cap_trunc(inode
, h
, session
);
2832 spin_unlock(&inode
->i_lock
);
2833 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
2834 ceph_cap_op_name(op
));
2838 mutex_unlock(&session
->s_mutex
);
2842 if (open_target_sessions
)
2843 ceph_mdsc_open_export_target_sessions(mdsc
, session
);
2847 pr_err("ceph_handle_caps: corrupt message\n");
2853 * Delayed work handler to process end of delayed cap release LRU list.
2855 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
2857 struct ceph_inode_info
*ci
;
2858 int flags
= CHECK_CAPS_NODELAY
;
2860 dout("check_delayed_caps\n");
2862 spin_lock(&mdsc
->cap_delay_lock
);
2863 if (list_empty(&mdsc
->cap_delay_list
))
2865 ci
= list_first_entry(&mdsc
->cap_delay_list
,
2866 struct ceph_inode_info
,
2868 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
2869 time_before(jiffies
, ci
->i_hold_caps_max
))
2871 list_del_init(&ci
->i_cap_delay_list
);
2872 spin_unlock(&mdsc
->cap_delay_lock
);
2873 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
2874 ceph_check_caps(ci
, flags
, NULL
);
2876 spin_unlock(&mdsc
->cap_delay_lock
);
2880 * Flush all dirty caps to the mds
2882 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
2884 struct ceph_inode_info
*ci
, *nci
= NULL
;
2885 struct inode
*inode
, *ninode
= NULL
;
2886 struct list_head
*p
, *n
;
2888 dout("flush_dirty_caps\n");
2889 spin_lock(&mdsc
->cap_dirty_lock
);
2890 list_for_each_safe(p
, n
, &mdsc
->cap_dirty
) {
2894 ci
->i_ceph_flags
&= ~CEPH_I_NOFLUSH
;
2895 dout("flush_dirty_caps inode %p (was next inode)\n",
2898 ci
= list_entry(p
, struct ceph_inode_info
,
2900 inode
= igrab(&ci
->vfs_inode
);
2902 dout("flush_dirty_caps inode %p\n", inode
);
2904 if (n
!= &mdsc
->cap_dirty
) {
2905 nci
= list_entry(n
, struct ceph_inode_info
,
2907 ninode
= igrab(&nci
->vfs_inode
);
2909 nci
->i_ceph_flags
|= CEPH_I_NOFLUSH
;
2910 dout("flush_dirty_caps next inode %p, noflush\n",
2916 spin_unlock(&mdsc
->cap_dirty_lock
);
2918 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
,
2922 spin_lock(&mdsc
->cap_dirty_lock
);
2924 spin_unlock(&mdsc
->cap_dirty_lock
);
2928 * Drop open file reference. If we were the last open file,
2929 * we may need to release capabilities to the MDS (or schedule
2930 * their delayed release).
2932 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
2934 struct inode
*inode
= &ci
->vfs_inode
;
2937 spin_lock(&inode
->i_lock
);
2938 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
2939 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
2940 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
2941 if (--ci
->i_nr_by_mode
[fmode
] == 0)
2943 spin_unlock(&inode
->i_lock
);
2945 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
2946 ceph_check_caps(ci
, 0, NULL
);
2950 * Helpers for embedding cap and dentry lease releases into mds
2953 * @force is used by dentry_release (below) to force inclusion of a
2954 * record for the directory inode, even when there aren't any caps to
2957 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
2958 int mds
, int drop
, int unless
, int force
)
2960 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2961 struct ceph_cap
*cap
;
2962 struct ceph_mds_request_release
*rel
= *p
;
2966 spin_lock(&inode
->i_lock
);
2967 used
= __ceph_caps_used(ci
);
2968 dirty
= __ceph_caps_dirty(ci
);
2970 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
2971 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
2972 ceph_cap_string(unless
));
2974 /* only drop unused, clean caps */
2975 drop
&= ~(used
| dirty
);
2977 cap
= __get_cap_for_mds(ci
, mds
);
2978 if (cap
&& __cap_is_valid(cap
)) {
2980 ((cap
->issued
& drop
) &&
2981 (cap
->issued
& unless
) == 0)) {
2982 if ((cap
->issued
& drop
) &&
2983 (cap
->issued
& unless
) == 0) {
2984 dout("encode_inode_release %p cap %p %s -> "
2986 ceph_cap_string(cap
->issued
),
2987 ceph_cap_string(cap
->issued
& ~drop
));
2988 cap
->issued
&= ~drop
;
2989 cap
->implemented
&= ~drop
;
2990 if (ci
->i_ceph_flags
& CEPH_I_NODELAY
) {
2991 int wanted
= __ceph_caps_wanted(ci
);
2992 dout(" wanted %s -> %s (act %s)\n",
2993 ceph_cap_string(cap
->mds_wanted
),
2994 ceph_cap_string(cap
->mds_wanted
&
2996 ceph_cap_string(wanted
));
2997 cap
->mds_wanted
&= wanted
;
3000 dout("encode_inode_release %p cap %p %s"
3001 " (force)\n", inode
, cap
,
3002 ceph_cap_string(cap
->issued
));
3005 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3006 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3007 rel
->seq
= cpu_to_le32(cap
->seq
);
3008 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
),
3009 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3010 rel
->caps
= cpu_to_le32(cap
->issued
);
3011 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3017 dout("encode_inode_release %p cap %p %s\n",
3018 inode
, cap
, ceph_cap_string(cap
->issued
));
3021 spin_unlock(&inode
->i_lock
);
3025 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3026 int mds
, int drop
, int unless
)
3028 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3029 struct ceph_mds_request_release
*rel
= *p
;
3030 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3035 * force an record for the directory caps if we have a dentry lease.
3036 * this is racy (can't take i_lock and d_lock together), but it
3037 * doesn't have to be perfect; the mds will revoke anything we don't
3040 spin_lock(&dentry
->d_lock
);
3041 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3043 spin_unlock(&dentry
->d_lock
);
3045 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3047 spin_lock(&dentry
->d_lock
);
3048 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3049 dout("encode_dentry_release %p mds%d seq %d\n",
3050 dentry
, mds
, (int)di
->lease_seq
);
3051 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3052 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3053 *p
+= dentry
->d_name
.len
;
3054 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3055 __ceph_mdsc_drop_dentry_lease(dentry
);
3057 spin_unlock(&dentry
->d_lock
);