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"
13 #include <linux/ceph/decode.h>
14 #include <linux/ceph/messenger.h>
17 * Capability management
19 * The Ceph metadata servers control client access to inode metadata
20 * and file data by issuing capabilities, granting clients permission
21 * to read and/or write both inode field and file data to OSDs
22 * (storage nodes). Each capability consists of a set of bits
23 * indicating which operations are allowed.
25 * If the client holds a *_SHARED cap, the client has a coherent value
26 * that can be safely read from the cached inode.
28 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
29 * client is allowed to change inode attributes (e.g., file size,
30 * mtime), note its dirty state in the ceph_cap, and asynchronously
31 * flush that metadata change to the MDS.
33 * In the event of a conflicting operation (perhaps by another
34 * client), the MDS will revoke the conflicting client capabilities.
36 * In order for a client to cache an inode, it must hold a capability
37 * with at least one MDS server. When inodes are released, release
38 * notifications are batched and periodically sent en masse to the MDS
39 * cluster to release server state.
44 * Generate readable cap strings for debugging output.
46 #define MAX_CAP_STR 20
47 static char cap_str
[MAX_CAP_STR
][40];
48 static DEFINE_SPINLOCK(cap_str_lock
);
49 static int last_cap_str
;
51 static char *gcap_string(char *s
, int c
)
53 if (c
& CEPH_CAP_GSHARED
)
55 if (c
& CEPH_CAP_GEXCL
)
57 if (c
& CEPH_CAP_GCACHE
)
63 if (c
& CEPH_CAP_GBUFFER
)
65 if (c
& CEPH_CAP_GLAZYIO
)
70 const char *ceph_cap_string(int caps
)
76 spin_lock(&cap_str_lock
);
78 if (last_cap_str
== MAX_CAP_STR
)
80 spin_unlock(&cap_str_lock
);
84 if (caps
& CEPH_CAP_PIN
)
87 c
= (caps
>> CEPH_CAP_SAUTH
) & 3;
90 s
= gcap_string(s
, c
);
93 c
= (caps
>> CEPH_CAP_SLINK
) & 3;
96 s
= gcap_string(s
, c
);
99 c
= (caps
>> CEPH_CAP_SXATTR
) & 3;
102 s
= gcap_string(s
, c
);
105 c
= caps
>> CEPH_CAP_SFILE
;
108 s
= gcap_string(s
, c
);
117 void ceph_caps_init(struct ceph_mds_client
*mdsc
)
119 INIT_LIST_HEAD(&mdsc
->caps_list
);
120 spin_lock_init(&mdsc
->caps_list_lock
);
123 void ceph_caps_finalize(struct ceph_mds_client
*mdsc
)
125 struct ceph_cap
*cap
;
127 spin_lock(&mdsc
->caps_list_lock
);
128 while (!list_empty(&mdsc
->caps_list
)) {
129 cap
= list_first_entry(&mdsc
->caps_list
,
130 struct ceph_cap
, caps_item
);
131 list_del(&cap
->caps_item
);
132 kmem_cache_free(ceph_cap_cachep
, cap
);
134 mdsc
->caps_total_count
= 0;
135 mdsc
->caps_avail_count
= 0;
136 mdsc
->caps_use_count
= 0;
137 mdsc
->caps_reserve_count
= 0;
138 mdsc
->caps_min_count
= 0;
139 spin_unlock(&mdsc
->caps_list_lock
);
142 void ceph_adjust_min_caps(struct ceph_mds_client
*mdsc
, int delta
)
144 spin_lock(&mdsc
->caps_list_lock
);
145 mdsc
->caps_min_count
+= delta
;
146 BUG_ON(mdsc
->caps_min_count
< 0);
147 spin_unlock(&mdsc
->caps_list_lock
);
150 void ceph_reserve_caps(struct ceph_mds_client
*mdsc
,
151 struct ceph_cap_reservation
*ctx
, int need
)
154 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 list_add(&cap
->caps_item
, &newcaps
);
181 /* we didn't manage to reserve as much as we needed */
182 if (have
+ alloc
!= need
)
183 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
184 ctx
, need
, have
+ alloc
);
186 spin_lock(&mdsc
->caps_list_lock
);
187 mdsc
->caps_total_count
+= alloc
;
188 mdsc
->caps_reserve_count
+= alloc
;
189 list_splice(&newcaps
, &mdsc
->caps_list
);
191 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
192 mdsc
->caps_reserve_count
+
193 mdsc
->caps_avail_count
);
194 spin_unlock(&mdsc
->caps_list_lock
);
197 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
198 ctx
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
199 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
202 int ceph_unreserve_caps(struct ceph_mds_client
*mdsc
,
203 struct ceph_cap_reservation
*ctx
)
205 dout("unreserve caps ctx=%p count=%d\n", ctx
, ctx
->count
);
207 spin_lock(&mdsc
->caps_list_lock
);
208 BUG_ON(mdsc
->caps_reserve_count
< ctx
->count
);
209 mdsc
->caps_reserve_count
-= ctx
->count
;
210 mdsc
->caps_avail_count
+= ctx
->count
;
212 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
213 mdsc
->caps_total_count
, mdsc
->caps_use_count
,
214 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
215 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
216 mdsc
->caps_reserve_count
+
217 mdsc
->caps_avail_count
);
218 spin_unlock(&mdsc
->caps_list_lock
);
223 static struct ceph_cap
*get_cap(struct ceph_mds_client
*mdsc
,
224 struct ceph_cap_reservation
*ctx
)
226 struct ceph_cap
*cap
= NULL
;
228 /* temporary, until we do something about cap import/export */
230 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
232 spin_lock(&mdsc
->caps_list_lock
);
233 mdsc
->caps_use_count
++;
234 mdsc
->caps_total_count
++;
235 spin_unlock(&mdsc
->caps_list_lock
);
240 spin_lock(&mdsc
->caps_list_lock
);
241 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
242 ctx
, ctx
->count
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
243 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
245 BUG_ON(ctx
->count
> mdsc
->caps_reserve_count
);
246 BUG_ON(list_empty(&mdsc
->caps_list
));
249 mdsc
->caps_reserve_count
--;
250 mdsc
->caps_use_count
++;
252 cap
= list_first_entry(&mdsc
->caps_list
, struct ceph_cap
, caps_item
);
253 list_del(&cap
->caps_item
);
255 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
256 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
257 spin_unlock(&mdsc
->caps_list_lock
);
261 void ceph_put_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
)
263 spin_lock(&mdsc
->caps_list_lock
);
264 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
265 cap
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
266 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
267 mdsc
->caps_use_count
--;
269 * Keep some preallocated caps around (ceph_min_count), to
270 * avoid lots of free/alloc churn.
272 if (mdsc
->caps_avail_count
>= mdsc
->caps_reserve_count
+
273 mdsc
->caps_min_count
) {
274 mdsc
->caps_total_count
--;
275 kmem_cache_free(ceph_cap_cachep
, cap
);
277 mdsc
->caps_avail_count
++;
278 list_add(&cap
->caps_item
, &mdsc
->caps_list
);
281 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
282 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
283 spin_unlock(&mdsc
->caps_list_lock
);
286 void ceph_reservation_status(struct ceph_fs_client
*fsc
,
287 int *total
, int *avail
, int *used
, int *reserved
,
290 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
293 *total
= mdsc
->caps_total_count
;
295 *avail
= mdsc
->caps_avail_count
;
297 *used
= mdsc
->caps_use_count
;
299 *reserved
= mdsc
->caps_reserve_count
;
301 *min
= mdsc
->caps_min_count
;
305 * Find ceph_cap for given mds, if any.
307 * Called with i_ceph_lock held.
309 static struct ceph_cap
*__get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
311 struct ceph_cap
*cap
;
312 struct rb_node
*n
= ci
->i_caps
.rb_node
;
315 cap
= rb_entry(n
, struct ceph_cap
, ci_node
);
318 else if (mds
> cap
->mds
)
326 struct ceph_cap
*ceph_get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
328 struct ceph_cap
*cap
;
330 spin_lock(&ci
->i_ceph_lock
);
331 cap
= __get_cap_for_mds(ci
, mds
);
332 spin_unlock(&ci
->i_ceph_lock
);
337 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
339 static int __ceph_get_cap_mds(struct ceph_inode_info
*ci
)
341 struct ceph_cap
*cap
;
345 /* prefer mds with WR|BUFFER|EXCL caps */
346 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
347 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
349 if (cap
->issued
& (CEPH_CAP_FILE_WR
|
350 CEPH_CAP_FILE_BUFFER
|
357 int ceph_get_cap_mds(struct inode
*inode
)
359 struct ceph_inode_info
*ci
= ceph_inode(inode
);
361 spin_lock(&ci
->i_ceph_lock
);
362 mds
= __ceph_get_cap_mds(ceph_inode(inode
));
363 spin_unlock(&ci
->i_ceph_lock
);
368 * Called under i_ceph_lock.
370 static void __insert_cap_node(struct ceph_inode_info
*ci
,
371 struct ceph_cap
*new)
373 struct rb_node
**p
= &ci
->i_caps
.rb_node
;
374 struct rb_node
*parent
= NULL
;
375 struct ceph_cap
*cap
= NULL
;
379 cap
= rb_entry(parent
, struct ceph_cap
, ci_node
);
380 if (new->mds
< cap
->mds
)
382 else if (new->mds
> cap
->mds
)
388 rb_link_node(&new->ci_node
, parent
, p
);
389 rb_insert_color(&new->ci_node
, &ci
->i_caps
);
393 * (re)set cap hold timeouts, which control the delayed release
394 * of unused caps back to the MDS. Should be called on cap use.
396 static void __cap_set_timeouts(struct ceph_mds_client
*mdsc
,
397 struct ceph_inode_info
*ci
)
399 struct ceph_mount_options
*ma
= mdsc
->fsc
->mount_options
;
401 ci
->i_hold_caps_min
= round_jiffies(jiffies
+
402 ma
->caps_wanted_delay_min
* HZ
);
403 ci
->i_hold_caps_max
= round_jiffies(jiffies
+
404 ma
->caps_wanted_delay_max
* HZ
);
405 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci
->vfs_inode
,
406 ci
->i_hold_caps_min
- jiffies
, ci
->i_hold_caps_max
- jiffies
);
410 * (Re)queue cap at the end of the delayed cap release list.
412 * If I_FLUSH is set, leave the inode at the front of the list.
414 * Caller holds i_ceph_lock
415 * -> we take mdsc->cap_delay_lock
417 static void __cap_delay_requeue(struct ceph_mds_client
*mdsc
,
418 struct ceph_inode_info
*ci
)
420 __cap_set_timeouts(mdsc
, ci
);
421 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci
->vfs_inode
,
422 ci
->i_ceph_flags
, ci
->i_hold_caps_max
);
423 if (!mdsc
->stopping
) {
424 spin_lock(&mdsc
->cap_delay_lock
);
425 if (!list_empty(&ci
->i_cap_delay_list
)) {
426 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
428 list_del_init(&ci
->i_cap_delay_list
);
430 list_add_tail(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
432 spin_unlock(&mdsc
->cap_delay_lock
);
437 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
438 * indicating we should send a cap message to flush dirty metadata
439 * asap, and move to the front of the delayed cap list.
441 static void __cap_delay_requeue_front(struct ceph_mds_client
*mdsc
,
442 struct ceph_inode_info
*ci
)
444 dout("__cap_delay_requeue_front %p\n", &ci
->vfs_inode
);
445 spin_lock(&mdsc
->cap_delay_lock
);
446 ci
->i_ceph_flags
|= CEPH_I_FLUSH
;
447 if (!list_empty(&ci
->i_cap_delay_list
))
448 list_del_init(&ci
->i_cap_delay_list
);
449 list_add(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
450 spin_unlock(&mdsc
->cap_delay_lock
);
454 * Cancel delayed work on cap.
456 * Caller must hold i_ceph_lock.
458 static void __cap_delay_cancel(struct ceph_mds_client
*mdsc
,
459 struct ceph_inode_info
*ci
)
461 dout("__cap_delay_cancel %p\n", &ci
->vfs_inode
);
462 if (list_empty(&ci
->i_cap_delay_list
))
464 spin_lock(&mdsc
->cap_delay_lock
);
465 list_del_init(&ci
->i_cap_delay_list
);
466 spin_unlock(&mdsc
->cap_delay_lock
);
470 * Common issue checks for add_cap, handle_cap_grant.
472 static void __check_cap_issue(struct ceph_inode_info
*ci
, struct ceph_cap
*cap
,
475 unsigned had
= __ceph_caps_issued(ci
, NULL
);
478 * Each time we receive FILE_CACHE anew, we increment
481 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
482 (had
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
486 * if we are newly issued FILE_SHARED, mark dir not complete; we
487 * don't know what happened to this directory while we didn't
490 if ((issued
& CEPH_CAP_FILE_SHARED
) &&
491 (had
& CEPH_CAP_FILE_SHARED
) == 0) {
493 if (S_ISDIR(ci
->vfs_inode
.i_mode
)) {
494 dout(" marking %p NOT complete\n", &ci
->vfs_inode
);
495 __ceph_dir_clear_complete(ci
);
501 * Add a capability under the given MDS session.
503 * Caller should hold session snap_rwsem (read) and s_mutex.
505 * @fmode is the open file mode, if we are opening a file, otherwise
506 * it is < 0. (This is so we can atomically add the cap and add an
507 * open file reference to it.)
509 int ceph_add_cap(struct inode
*inode
,
510 struct ceph_mds_session
*session
, u64 cap_id
,
511 int fmode
, unsigned issued
, unsigned wanted
,
512 unsigned seq
, unsigned mseq
, u64 realmino
, int flags
,
513 struct ceph_cap_reservation
*caps_reservation
)
515 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
516 struct ceph_inode_info
*ci
= ceph_inode(inode
);
517 struct ceph_cap
*new_cap
= NULL
;
518 struct ceph_cap
*cap
;
519 int mds
= session
->s_mds
;
522 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
523 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
526 * If we are opening the file, include file mode wanted bits
530 wanted
|= ceph_caps_for_mode(fmode
);
533 spin_lock(&ci
->i_ceph_lock
);
534 cap
= __get_cap_for_mds(ci
, mds
);
540 spin_unlock(&ci
->i_ceph_lock
);
541 new_cap
= get_cap(mdsc
, caps_reservation
);
548 cap
->implemented
= 0;
554 __insert_cap_node(ci
, cap
);
556 /* clear out old exporting info? (i.e. on cap import) */
557 if (ci
->i_cap_exporting_mds
== mds
) {
558 ci
->i_cap_exporting_issued
= 0;
559 ci
->i_cap_exporting_mseq
= 0;
560 ci
->i_cap_exporting_mds
= -1;
563 /* add to session cap list */
564 cap
->session
= session
;
565 spin_lock(&session
->s_cap_lock
);
566 list_add_tail(&cap
->session_caps
, &session
->s_caps
);
567 session
->s_nr_caps
++;
568 spin_unlock(&session
->s_cap_lock
);
570 ceph_put_cap(mdsc
, new_cap
);
572 if (!ci
->i_snap_realm
) {
574 * add this inode to the appropriate snap realm
576 struct ceph_snap_realm
*realm
= ceph_lookup_snap_realm(mdsc
,
579 ceph_get_snap_realm(mdsc
, realm
);
580 spin_lock(&realm
->inodes_with_caps_lock
);
581 ci
->i_snap_realm
= realm
;
582 list_add(&ci
->i_snap_realm_item
,
583 &realm
->inodes_with_caps
);
584 spin_unlock(&realm
->inodes_with_caps_lock
);
586 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
592 __check_cap_issue(ci
, cap
, issued
);
595 * If we are issued caps we don't want, or the mds' wanted
596 * value appears to be off, queue a check so we'll release
597 * later and/or update the mds wanted value.
599 actual_wanted
= __ceph_caps_wanted(ci
);
600 if ((wanted
& ~actual_wanted
) ||
601 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
602 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
603 ceph_cap_string(issued
), ceph_cap_string(wanted
),
604 ceph_cap_string(actual_wanted
));
605 __cap_delay_requeue(mdsc
, ci
);
608 if (flags
& CEPH_CAP_FLAG_AUTH
) {
609 if (ci
->i_auth_cap
== NULL
||
610 ceph_seq_cmp(ci
->i_auth_cap
->mseq
, mseq
) < 0)
611 ci
->i_auth_cap
= cap
;
612 } else if (ci
->i_auth_cap
== cap
) {
613 ci
->i_auth_cap
= NULL
;
614 spin_lock(&mdsc
->cap_dirty_lock
);
615 if (!list_empty(&ci
->i_dirty_item
)) {
616 dout(" moving %p to cap_dirty_migrating\n", inode
);
617 list_move(&ci
->i_dirty_item
,
618 &mdsc
->cap_dirty_migrating
);
620 spin_unlock(&mdsc
->cap_dirty_lock
);
623 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
624 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
625 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
626 cap
->cap_id
= cap_id
;
627 cap
->issued
= issued
;
628 cap
->implemented
|= issued
;
629 if (mseq
> cap
->mseq
)
630 cap
->mds_wanted
= wanted
;
632 cap
->mds_wanted
|= wanted
;
634 cap
->issue_seq
= seq
;
636 cap
->cap_gen
= session
->s_cap_gen
;
639 __ceph_get_fmode(ci
, fmode
);
640 spin_unlock(&ci
->i_ceph_lock
);
641 wake_up_all(&ci
->i_cap_wq
);
646 * Return true if cap has not timed out and belongs to the current
647 * generation of the MDS session (i.e. has not gone 'stale' due to
648 * us losing touch with the mds).
650 static int __cap_is_valid(struct ceph_cap
*cap
)
655 spin_lock(&cap
->session
->s_gen_ttl_lock
);
656 gen
= cap
->session
->s_cap_gen
;
657 ttl
= cap
->session
->s_cap_ttl
;
658 spin_unlock(&cap
->session
->s_gen_ttl_lock
);
660 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
661 dout("__cap_is_valid %p cap %p issued %s "
662 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
663 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
671 * Return set of valid cap bits issued to us. Note that caps time
672 * out, and may be invalidated in bulk if the client session times out
673 * and session->s_cap_gen is bumped.
675 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
677 int have
= ci
->i_snap_caps
| ci
->i_cap_exporting_issued
;
678 struct ceph_cap
*cap
;
683 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
684 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
685 if (!__cap_is_valid(cap
))
687 dout("__ceph_caps_issued %p cap %p issued %s\n",
688 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
691 *implemented
|= cap
->implemented
;
697 * Get cap bits issued by caps other than @ocap
699 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
701 int have
= ci
->i_snap_caps
;
702 struct ceph_cap
*cap
;
705 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
706 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
709 if (!__cap_is_valid(cap
))
717 * Move a cap to the end of the LRU (oldest caps at list head, newest
720 static void __touch_cap(struct ceph_cap
*cap
)
722 struct ceph_mds_session
*s
= cap
->session
;
724 spin_lock(&s
->s_cap_lock
);
725 if (s
->s_cap_iterator
== NULL
) {
726 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
728 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
730 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
731 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
733 spin_unlock(&s
->s_cap_lock
);
737 * Check if we hold the given mask. If so, move the cap(s) to the
738 * front of their respective LRUs. (This is the preferred way for
739 * callers to check for caps they want.)
741 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
743 struct ceph_cap
*cap
;
745 int have
= ci
->i_snap_caps
;
747 if ((have
& mask
) == mask
) {
748 dout("__ceph_caps_issued_mask %p snap issued %s"
749 " (mask %s)\n", &ci
->vfs_inode
,
750 ceph_cap_string(have
),
751 ceph_cap_string(mask
));
755 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
756 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
757 if (!__cap_is_valid(cap
))
759 if ((cap
->issued
& mask
) == mask
) {
760 dout("__ceph_caps_issued_mask %p cap %p issued %s"
761 " (mask %s)\n", &ci
->vfs_inode
, cap
,
762 ceph_cap_string(cap
->issued
),
763 ceph_cap_string(mask
));
769 /* does a combination of caps satisfy mask? */
771 if ((have
& mask
) == mask
) {
772 dout("__ceph_caps_issued_mask %p combo issued %s"
773 " (mask %s)\n", &ci
->vfs_inode
,
774 ceph_cap_string(cap
->issued
),
775 ceph_cap_string(mask
));
779 /* touch this + preceding caps */
781 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
783 cap
= rb_entry(q
, struct ceph_cap
,
785 if (!__cap_is_valid(cap
))
798 * Return true if mask caps are currently being revoked by an MDS.
800 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
802 struct inode
*inode
= &ci
->vfs_inode
;
803 struct ceph_cap
*cap
;
807 spin_lock(&ci
->i_ceph_lock
);
808 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
809 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
810 if (__cap_is_valid(cap
) &&
811 (cap
->implemented
& ~cap
->issued
& mask
)) {
816 spin_unlock(&ci
->i_ceph_lock
);
817 dout("ceph_caps_revoking %p %s = %d\n", inode
,
818 ceph_cap_string(mask
), ret
);
822 int __ceph_caps_used(struct ceph_inode_info
*ci
)
826 used
|= CEPH_CAP_PIN
;
828 used
|= CEPH_CAP_FILE_RD
;
829 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
830 used
|= CEPH_CAP_FILE_CACHE
;
832 used
|= CEPH_CAP_FILE_WR
;
833 if (ci
->i_wb_ref
|| ci
->i_wrbuffer_ref
)
834 used
|= CEPH_CAP_FILE_BUFFER
;
839 * wanted, by virtue of open file modes
841 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
845 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
846 if (ci
->i_nr_by_mode
[mode
])
847 want
|= ceph_caps_for_mode(mode
);
852 * Return caps we have registered with the MDS(s) as 'wanted'.
854 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
856 struct ceph_cap
*cap
;
860 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
861 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
862 if (!__cap_is_valid(cap
))
864 mds_wanted
|= cap
->mds_wanted
;
870 * called under i_ceph_lock
872 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
874 return !RB_EMPTY_ROOT(&ci
->i_caps
) || ci
->i_cap_exporting_mds
>= 0;
878 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
880 * caller should hold i_ceph_lock.
881 * caller will not hold session s_mutex if called from destroy_inode.
883 void __ceph_remove_cap(struct ceph_cap
*cap
)
885 struct ceph_mds_session
*session
= cap
->session
;
886 struct ceph_inode_info
*ci
= cap
->ci
;
887 struct ceph_mds_client
*mdsc
=
888 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
891 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
893 /* remove from session list */
894 spin_lock(&session
->s_cap_lock
);
895 if (session
->s_cap_iterator
== cap
) {
896 /* not yet, we are iterating over this very cap */
897 dout("__ceph_remove_cap delaying %p removal from session %p\n",
900 list_del_init(&cap
->session_caps
);
901 session
->s_nr_caps
--;
905 /* protect backpointer with s_cap_lock: see iterate_session_caps */
907 spin_unlock(&session
->s_cap_lock
);
909 /* remove from inode list */
910 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
911 if (ci
->i_auth_cap
== cap
)
912 ci
->i_auth_cap
= NULL
;
915 ceph_put_cap(mdsc
, cap
);
917 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
) {
918 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
919 spin_lock(&realm
->inodes_with_caps_lock
);
920 list_del_init(&ci
->i_snap_realm_item
);
921 ci
->i_snap_realm_counter
++;
922 ci
->i_snap_realm
= NULL
;
923 spin_unlock(&realm
->inodes_with_caps_lock
);
924 ceph_put_snap_realm(mdsc
, realm
);
926 if (!__ceph_is_any_real_caps(ci
))
927 __cap_delay_cancel(mdsc
, ci
);
931 * Build and send a cap message to the given MDS.
933 * Caller should be holding s_mutex.
935 static int send_cap_msg(struct ceph_mds_session
*session
,
936 u64 ino
, u64 cid
, int op
,
937 int caps
, int wanted
, int dirty
,
938 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
939 u64 size
, u64 max_size
,
940 struct timespec
*mtime
, struct timespec
*atime
,
942 kuid_t uid
, kgid_t gid
, umode_t mode
,
944 struct ceph_buffer
*xattrs_buf
,
947 struct ceph_mds_caps
*fc
;
948 struct ceph_msg
*msg
;
950 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
951 " seq %u/%u mseq %u follows %lld size %llu/%llu"
952 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
953 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
954 ceph_cap_string(dirty
),
955 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
956 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
958 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
), GFP_NOFS
, false);
962 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
964 fc
= msg
->front
.iov_base
;
965 memset(fc
, 0, sizeof(*fc
));
967 fc
->cap_id
= cpu_to_le64(cid
);
968 fc
->op
= cpu_to_le32(op
);
969 fc
->seq
= cpu_to_le32(seq
);
970 fc
->issue_seq
= cpu_to_le32(issue_seq
);
971 fc
->migrate_seq
= cpu_to_le32(mseq
);
972 fc
->caps
= cpu_to_le32(caps
);
973 fc
->wanted
= cpu_to_le32(wanted
);
974 fc
->dirty
= cpu_to_le32(dirty
);
975 fc
->ino
= cpu_to_le64(ino
);
976 fc
->snap_follows
= cpu_to_le64(follows
);
978 fc
->size
= cpu_to_le64(size
);
979 fc
->max_size
= cpu_to_le64(max_size
);
981 ceph_encode_timespec(&fc
->mtime
, mtime
);
983 ceph_encode_timespec(&fc
->atime
, atime
);
984 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
986 fc
->uid
= cpu_to_le32(from_kuid(&init_user_ns
, uid
));
987 fc
->gid
= cpu_to_le32(from_kgid(&init_user_ns
, gid
));
988 fc
->mode
= cpu_to_le32(mode
);
990 fc
->xattr_version
= cpu_to_le64(xattr_version
);
992 msg
->middle
= ceph_buffer_get(xattrs_buf
);
993 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
994 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
997 ceph_con_send(&session
->s_con
, msg
);
1001 void __queue_cap_release(struct ceph_mds_session
*session
,
1002 u64 ino
, u64 cap_id
, u32 migrate_seq
,
1005 struct ceph_msg
*msg
;
1006 struct ceph_mds_cap_release
*head
;
1007 struct ceph_mds_cap_item
*item
;
1009 spin_lock(&session
->s_cap_lock
);
1010 BUG_ON(!session
->s_num_cap_releases
);
1011 msg
= list_first_entry(&session
->s_cap_releases
,
1012 struct ceph_msg
, list_head
);
1014 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1015 ino
, session
->s_mds
, msg
, session
->s_num_cap_releases
);
1017 BUG_ON(msg
->front
.iov_len
+ sizeof(*item
) > PAGE_CACHE_SIZE
);
1018 head
= msg
->front
.iov_base
;
1019 le32_add_cpu(&head
->num
, 1);
1020 item
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1021 item
->ino
= cpu_to_le64(ino
);
1022 item
->cap_id
= cpu_to_le64(cap_id
);
1023 item
->migrate_seq
= cpu_to_le32(migrate_seq
);
1024 item
->seq
= cpu_to_le32(issue_seq
);
1026 session
->s_num_cap_releases
--;
1028 msg
->front
.iov_len
+= sizeof(*item
);
1029 if (le32_to_cpu(head
->num
) == CEPH_CAPS_PER_RELEASE
) {
1030 dout(" release msg %p full\n", msg
);
1031 list_move_tail(&msg
->list_head
, &session
->s_cap_releases_done
);
1033 dout(" release msg %p at %d/%d (%d)\n", msg
,
1034 (int)le32_to_cpu(head
->num
),
1035 (int)CEPH_CAPS_PER_RELEASE
,
1036 (int)msg
->front
.iov_len
);
1038 spin_unlock(&session
->s_cap_lock
);
1042 * Queue cap releases when an inode is dropped from our cache. Since
1043 * inode is about to be destroyed, there is no need for i_ceph_lock.
1045 void ceph_queue_caps_release(struct inode
*inode
)
1047 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1050 p
= rb_first(&ci
->i_caps
);
1052 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1053 struct ceph_mds_session
*session
= cap
->session
;
1055 __queue_cap_release(session
, ceph_ino(inode
), cap
->cap_id
,
1056 cap
->mseq
, cap
->issue_seq
);
1058 __ceph_remove_cap(cap
);
1063 * Send a cap msg on the given inode. Update our caps state, then
1064 * drop i_ceph_lock and send the message.
1066 * Make note of max_size reported/requested from mds, revoked caps
1067 * that have now been implemented.
1069 * Make half-hearted attempt ot to invalidate page cache if we are
1070 * dropping RDCACHE. Note that this will leave behind locked pages
1071 * that we'll then need to deal with elsewhere.
1073 * Return non-zero if delayed release, or we experienced an error
1074 * such that the caller should requeue + retry later.
1076 * called with i_ceph_lock, then drops it.
1077 * caller should hold snap_rwsem (read), s_mutex.
1079 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1080 int op
, int used
, int want
, int retain
, int flushing
,
1081 unsigned *pflush_tid
)
1082 __releases(cap
->ci
->i_ceph_lock
)
1084 struct ceph_inode_info
*ci
= cap
->ci
;
1085 struct inode
*inode
= &ci
->vfs_inode
;
1086 u64 cap_id
= cap
->cap_id
;
1087 int held
, revoking
, dropping
, keep
;
1088 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1090 struct timespec mtime
, atime
;
1095 struct ceph_mds_session
*session
;
1096 u64 xattr_version
= 0;
1097 struct ceph_buffer
*xattr_blob
= NULL
;
1103 held
= cap
->issued
| cap
->implemented
;
1104 revoking
= cap
->implemented
& ~cap
->issued
;
1105 retain
&= ~revoking
;
1106 dropping
= cap
->issued
& ~retain
;
1108 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1109 inode
, cap
, cap
->session
,
1110 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1111 ceph_cap_string(revoking
));
1112 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1114 session
= cap
->session
;
1116 /* don't release wanted unless we've waited a bit. */
1117 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1118 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1119 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1120 ceph_cap_string(cap
->issued
),
1121 ceph_cap_string(cap
->issued
& retain
),
1122 ceph_cap_string(cap
->mds_wanted
),
1123 ceph_cap_string(want
));
1124 want
|= cap
->mds_wanted
;
1125 retain
|= cap
->issued
;
1128 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1130 cap
->issued
&= retain
; /* drop bits we don't want */
1131 if (cap
->implemented
& ~cap
->issued
) {
1133 * Wake up any waiters on wanted -> needed transition.
1134 * This is due to the weird transition from buffered
1135 * to sync IO... we need to flush dirty pages _before_
1136 * allowing sync writes to avoid reordering.
1140 cap
->implemented
&= cap
->issued
| used
;
1141 cap
->mds_wanted
= want
;
1145 * assign a tid for flush operations so we can avoid
1146 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1147 * clean type races. track latest tid for every bit
1148 * so we can handle flush AxFw, flush Fw, and have the
1149 * first ack clean Ax.
1151 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1153 *pflush_tid
= flush_tid
;
1154 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1155 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1156 if (flushing
& (1 << i
))
1157 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1159 follows
= ci
->i_head_snapc
->seq
;
1164 keep
= cap
->implemented
;
1166 issue_seq
= cap
->issue_seq
;
1168 size
= inode
->i_size
;
1169 ci
->i_reported_size
= size
;
1170 max_size
= ci
->i_wanted_max_size
;
1171 ci
->i_requested_max_size
= max_size
;
1172 mtime
= inode
->i_mtime
;
1173 atime
= inode
->i_atime
;
1174 time_warp_seq
= ci
->i_time_warp_seq
;
1177 mode
= inode
->i_mode
;
1179 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1180 __ceph_build_xattrs_blob(ci
);
1181 xattr_blob
= ci
->i_xattrs
.blob
;
1182 xattr_version
= ci
->i_xattrs
.version
;
1185 spin_unlock(&ci
->i_ceph_lock
);
1187 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1188 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1189 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1190 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1193 dout("error sending cap msg, must requeue %p\n", inode
);
1198 wake_up_all(&ci
->i_cap_wq
);
1204 * When a snapshot is taken, clients accumulate dirty metadata on
1205 * inodes with capabilities in ceph_cap_snaps to describe the file
1206 * state at the time the snapshot was taken. This must be flushed
1207 * asynchronously back to the MDS once sync writes complete and dirty
1208 * data is written out.
1210 * Unless @again is true, skip cap_snaps that were already sent to
1211 * the MDS (i.e., during this session).
1213 * Called under i_ceph_lock. Takes s_mutex as needed.
1215 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1216 struct ceph_mds_session
**psession
,
1218 __releases(ci
->i_ceph_lock
)
1219 __acquires(ci
->i_ceph_lock
)
1221 struct inode
*inode
= &ci
->vfs_inode
;
1223 struct ceph_cap_snap
*capsnap
;
1225 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1226 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1228 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1229 i_cap_snaps list, and skip these entries next time
1230 around to avoid an infinite loop */
1233 session
= *psession
;
1235 dout("__flush_snaps %p\n", inode
);
1237 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1238 /* avoid an infiniute loop after retry */
1239 if (capsnap
->follows
< next_follows
)
1242 * we need to wait for sync writes to complete and for dirty
1243 * pages to be written out.
1245 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1249 * if cap writeback already occurred, we should have dropped
1250 * the capsnap in ceph_put_wrbuffer_cap_refs.
1252 BUG_ON(capsnap
->dirty
== 0);
1254 /* pick mds, take s_mutex */
1255 if (ci
->i_auth_cap
== NULL
) {
1256 dout("no auth cap (migrating?), doing nothing\n");
1260 /* only flush each capsnap once */
1261 if (!again
&& !list_empty(&capsnap
->flushing_item
)) {
1262 dout("already flushed %p, skipping\n", capsnap
);
1266 mds
= ci
->i_auth_cap
->session
->s_mds
;
1267 mseq
= ci
->i_auth_cap
->mseq
;
1269 if (session
&& session
->s_mds
!= mds
) {
1270 dout("oops, wrong session %p mutex\n", session
);
1271 mutex_unlock(&session
->s_mutex
);
1272 ceph_put_mds_session(session
);
1276 spin_unlock(&ci
->i_ceph_lock
);
1277 mutex_lock(&mdsc
->mutex
);
1278 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1279 mutex_unlock(&mdsc
->mutex
);
1281 dout("inverting session/ino locks on %p\n",
1283 mutex_lock(&session
->s_mutex
);
1286 * if session == NULL, we raced against a cap
1287 * deletion or migration. retry, and we'll
1288 * get a better @mds value next time.
1290 spin_lock(&ci
->i_ceph_lock
);
1294 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1295 atomic_inc(&capsnap
->nref
);
1296 if (!list_empty(&capsnap
->flushing_item
))
1297 list_del_init(&capsnap
->flushing_item
);
1298 list_add_tail(&capsnap
->flushing_item
,
1299 &session
->s_cap_snaps_flushing
);
1300 spin_unlock(&ci
->i_ceph_lock
);
1302 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1303 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1304 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1305 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1306 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1308 &capsnap
->mtime
, &capsnap
->atime
,
1309 capsnap
->time_warp_seq
,
1310 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1311 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1314 next_follows
= capsnap
->follows
+ 1;
1315 ceph_put_cap_snap(capsnap
);
1317 spin_lock(&ci
->i_ceph_lock
);
1321 /* we flushed them all; remove this inode from the queue */
1322 spin_lock(&mdsc
->snap_flush_lock
);
1323 list_del_init(&ci
->i_snap_flush_item
);
1324 spin_unlock(&mdsc
->snap_flush_lock
);
1328 *psession
= session
;
1330 mutex_unlock(&session
->s_mutex
);
1331 ceph_put_mds_session(session
);
1335 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1337 spin_lock(&ci
->i_ceph_lock
);
1338 __ceph_flush_snaps(ci
, NULL
, 0);
1339 spin_unlock(&ci
->i_ceph_lock
);
1343 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1344 * Caller is then responsible for calling __mark_inode_dirty with the
1345 * returned flags value.
1347 int __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1349 struct ceph_mds_client
*mdsc
=
1350 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1351 struct inode
*inode
= &ci
->vfs_inode
;
1352 int was
= ci
->i_dirty_caps
;
1355 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1356 ceph_cap_string(mask
), ceph_cap_string(was
),
1357 ceph_cap_string(was
| mask
));
1358 ci
->i_dirty_caps
|= mask
;
1360 if (!ci
->i_head_snapc
)
1361 ci
->i_head_snapc
= ceph_get_snap_context(
1362 ci
->i_snap_realm
->cached_context
);
1363 dout(" inode %p now dirty snapc %p auth cap %p\n",
1364 &ci
->vfs_inode
, ci
->i_head_snapc
, ci
->i_auth_cap
);
1365 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1366 spin_lock(&mdsc
->cap_dirty_lock
);
1368 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1370 list_add(&ci
->i_dirty_item
,
1371 &mdsc
->cap_dirty_migrating
);
1372 spin_unlock(&mdsc
->cap_dirty_lock
);
1373 if (ci
->i_flushing_caps
== 0) {
1375 dirty
|= I_DIRTY_SYNC
;
1378 BUG_ON(list_empty(&ci
->i_dirty_item
));
1379 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1380 (mask
& CEPH_CAP_FILE_BUFFER
))
1381 dirty
|= I_DIRTY_DATASYNC
;
1382 __cap_delay_requeue(mdsc
, ci
);
1387 * Add dirty inode to the flushing list. Assigned a seq number so we
1388 * can wait for caps to flush without starving.
1390 * Called under i_ceph_lock.
1392 static int __mark_caps_flushing(struct inode
*inode
,
1393 struct ceph_mds_session
*session
)
1395 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1396 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1399 BUG_ON(ci
->i_dirty_caps
== 0);
1400 BUG_ON(list_empty(&ci
->i_dirty_item
));
1402 flushing
= ci
->i_dirty_caps
;
1403 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1404 ceph_cap_string(flushing
),
1405 ceph_cap_string(ci
->i_flushing_caps
),
1406 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1407 ci
->i_flushing_caps
|= flushing
;
1408 ci
->i_dirty_caps
= 0;
1409 dout(" inode %p now !dirty\n", inode
);
1411 spin_lock(&mdsc
->cap_dirty_lock
);
1412 list_del_init(&ci
->i_dirty_item
);
1414 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1415 if (list_empty(&ci
->i_flushing_item
)) {
1416 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1417 mdsc
->num_cap_flushing
++;
1418 dout(" inode %p now flushing seq %lld\n", inode
,
1419 ci
->i_cap_flush_seq
);
1421 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1422 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1423 ci
->i_cap_flush_seq
);
1425 spin_unlock(&mdsc
->cap_dirty_lock
);
1431 * try to invalidate mapping pages without blocking.
1433 static int try_nonblocking_invalidate(struct inode
*inode
)
1435 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1436 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1438 spin_unlock(&ci
->i_ceph_lock
);
1439 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1440 spin_lock(&ci
->i_ceph_lock
);
1442 if (inode
->i_data
.nrpages
== 0 &&
1443 invalidating_gen
== ci
->i_rdcache_gen
) {
1445 dout("try_nonblocking_invalidate %p success\n", inode
);
1446 /* save any racing async invalidate some trouble */
1447 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
- 1;
1450 dout("try_nonblocking_invalidate %p failed\n", inode
);
1455 * Swiss army knife function to examine currently used and wanted
1456 * versus held caps. Release, flush, ack revoked caps to mds as
1459 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1460 * cap release further.
1461 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1462 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1465 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1466 struct ceph_mds_session
*session
)
1468 struct ceph_fs_client
*fsc
= ceph_inode_to_client(&ci
->vfs_inode
);
1469 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1470 struct inode
*inode
= &ci
->vfs_inode
;
1471 struct ceph_cap
*cap
;
1472 int file_wanted
, used
, cap_used
;
1473 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1474 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1475 int mds
= -1; /* keep track of how far we've gone through i_caps list
1476 to avoid an infinite loop on retry */
1478 int tried_invalidate
= 0;
1479 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1480 int queue_invalidate
= 0;
1481 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1483 /* if we are unmounting, flush any unused caps immediately. */
1487 spin_lock(&ci
->i_ceph_lock
);
1489 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1490 flags
|= CHECK_CAPS_FLUSH
;
1492 /* flush snaps first time around only */
1493 if (!list_empty(&ci
->i_cap_snaps
))
1494 __ceph_flush_snaps(ci
, &session
, 0);
1497 spin_lock(&ci
->i_ceph_lock
);
1499 file_wanted
= __ceph_caps_file_wanted(ci
);
1500 used
= __ceph_caps_used(ci
);
1501 want
= file_wanted
| used
;
1502 issued
= __ceph_caps_issued(ci
, &implemented
);
1503 revoking
= implemented
& ~issued
;
1505 retain
= want
| CEPH_CAP_PIN
;
1506 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1508 retain
|= CEPH_CAP_ANY
; /* be greedy */
1510 retain
|= CEPH_CAP_ANY_SHARED
;
1512 * keep RD only if we didn't have the file open RW,
1513 * because then the mds would revoke it anyway to
1514 * journal max_size=0.
1516 if (ci
->i_max_size
== 0)
1517 retain
|= CEPH_CAP_ANY_RD
;
1521 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1522 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1523 ceph_cap_string(file_wanted
),
1524 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1525 ceph_cap_string(ci
->i_flushing_caps
),
1526 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1527 ceph_cap_string(retain
),
1528 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1529 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1530 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1533 * If we no longer need to hold onto old our caps, and we may
1534 * have cached pages, but don't want them, then try to invalidate.
1535 * If we fail, it's because pages are locked.... try again later.
1537 if ((!is_delayed
|| mdsc
->stopping
) &&
1538 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1539 inode
->i_data
.nrpages
&& /* have cached pages */
1540 (file_wanted
== 0 || /* no open files */
1541 (revoking
& (CEPH_CAP_FILE_CACHE
|
1542 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1543 !tried_invalidate
) {
1544 dout("check_caps trying to invalidate on %p\n", inode
);
1545 if (try_nonblocking_invalidate(inode
) < 0) {
1546 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1547 CEPH_CAP_FILE_LAZYIO
)) {
1548 dout("check_caps queuing invalidate\n");
1549 queue_invalidate
= 1;
1550 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1552 dout("check_caps failed to invalidate pages\n");
1553 /* we failed to invalidate pages. check these
1554 caps again later. */
1556 __cap_set_timeouts(mdsc
, ci
);
1559 tried_invalidate
= 1;
1564 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1565 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1568 /* avoid looping forever */
1569 if (mds
>= cap
->mds
||
1570 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1573 /* NOTE: no side-effects allowed, until we take s_mutex */
1576 if (ci
->i_auth_cap
&& cap
!= ci
->i_auth_cap
)
1577 cap_used
&= ~ci
->i_auth_cap
->issued
;
1579 revoking
= cap
->implemented
& ~cap
->issued
;
1580 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1581 cap
->mds
, cap
, ceph_cap_string(cap
->issued
),
1582 ceph_cap_string(cap_used
),
1583 ceph_cap_string(cap
->implemented
),
1584 ceph_cap_string(revoking
));
1586 if (cap
== ci
->i_auth_cap
&&
1587 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1588 /* request larger max_size from MDS? */
1589 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1590 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1591 dout("requesting new max_size\n");
1595 /* approaching file_max? */
1596 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1597 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1598 dout("i_size approaching max_size\n");
1602 /* flush anything dirty? */
1603 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1605 dout("flushing dirty caps\n");
1609 /* completed revocation? going down and there are no caps? */
1610 if (revoking
&& (revoking
& cap_used
) == 0) {
1611 dout("completed revocation of %s\n",
1612 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1616 /* want more caps from mds? */
1617 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1620 /* things we might delay */
1621 if ((cap
->issued
& ~retain
) == 0 &&
1622 cap
->mds_wanted
== want
)
1623 continue; /* nope, all good */
1629 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1630 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1631 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1632 ceph_cap_string(cap
->issued
),
1633 ceph_cap_string(cap
->issued
& retain
),
1634 ceph_cap_string(cap
->mds_wanted
),
1635 ceph_cap_string(want
));
1641 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1642 dout(" skipping %p I_NOFLUSH set\n", inode
);
1646 if (session
&& session
!= cap
->session
) {
1647 dout("oops, wrong session %p mutex\n", session
);
1648 mutex_unlock(&session
->s_mutex
);
1652 session
= cap
->session
;
1653 if (mutex_trylock(&session
->s_mutex
) == 0) {
1654 dout("inverting session/ino locks on %p\n",
1656 spin_unlock(&ci
->i_ceph_lock
);
1657 if (took_snap_rwsem
) {
1658 up_read(&mdsc
->snap_rwsem
);
1659 took_snap_rwsem
= 0;
1661 mutex_lock(&session
->s_mutex
);
1665 /* take snap_rwsem after session mutex */
1666 if (!took_snap_rwsem
) {
1667 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1668 dout("inverting snap/in locks on %p\n",
1670 spin_unlock(&ci
->i_ceph_lock
);
1671 down_read(&mdsc
->snap_rwsem
);
1672 took_snap_rwsem
= 1;
1675 took_snap_rwsem
= 1;
1678 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1679 flushing
= __mark_caps_flushing(inode
, session
);
1683 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1686 /* __send_cap drops i_ceph_lock */
1687 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, cap_used
,
1688 want
, retain
, flushing
, NULL
);
1689 goto retry
; /* retake i_ceph_lock and restart our cap scan. */
1693 * Reschedule delayed caps release if we delayed anything,
1696 if (delayed
&& is_delayed
)
1697 force_requeue
= 1; /* __send_cap delayed release; requeue */
1698 if (!delayed
&& !is_delayed
)
1699 __cap_delay_cancel(mdsc
, ci
);
1700 else if (!is_delayed
|| force_requeue
)
1701 __cap_delay_requeue(mdsc
, ci
);
1703 spin_unlock(&ci
->i_ceph_lock
);
1705 if (queue_invalidate
)
1706 ceph_queue_invalidate(inode
);
1709 mutex_unlock(&session
->s_mutex
);
1710 if (took_snap_rwsem
)
1711 up_read(&mdsc
->snap_rwsem
);
1715 * Try to flush dirty caps back to the auth mds.
1717 static int try_flush_caps(struct inode
*inode
, struct ceph_mds_session
*session
,
1718 unsigned *flush_tid
)
1720 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1721 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1722 int unlock_session
= session
? 0 : 1;
1726 spin_lock(&ci
->i_ceph_lock
);
1727 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1728 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1731 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1732 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1733 int used
= __ceph_caps_used(ci
);
1734 int want
= __ceph_caps_wanted(ci
);
1738 spin_unlock(&ci
->i_ceph_lock
);
1739 session
= cap
->session
;
1740 mutex_lock(&session
->s_mutex
);
1743 BUG_ON(session
!= cap
->session
);
1744 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1747 flushing
= __mark_caps_flushing(inode
, session
);
1749 /* __send_cap drops i_ceph_lock */
1750 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1751 cap
->issued
| cap
->implemented
, flushing
,
1756 spin_lock(&ci
->i_ceph_lock
);
1757 __cap_delay_requeue(mdsc
, ci
);
1760 spin_unlock(&ci
->i_ceph_lock
);
1762 if (session
&& unlock_session
)
1763 mutex_unlock(&session
->s_mutex
);
1768 * Return true if we've flushed caps through the given flush_tid.
1770 static int caps_are_flushed(struct inode
*inode
, unsigned tid
)
1772 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1775 spin_lock(&ci
->i_ceph_lock
);
1776 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1777 if ((ci
->i_flushing_caps
& (1 << i
)) &&
1778 ci
->i_cap_flush_tid
[i
] <= tid
) {
1779 /* still flushing this bit */
1783 spin_unlock(&ci
->i_ceph_lock
);
1788 * Wait on any unsafe replies for the given inode. First wait on the
1789 * newest request, and make that the upper bound. Then, if there are
1790 * more requests, keep waiting on the oldest as long as it is still older
1791 * than the original request.
1793 static void sync_write_wait(struct inode
*inode
)
1795 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1796 struct list_head
*head
= &ci
->i_unsafe_writes
;
1797 struct ceph_osd_request
*req
;
1800 spin_lock(&ci
->i_unsafe_lock
);
1801 if (list_empty(head
))
1804 /* set upper bound as _last_ entry in chain */
1805 req
= list_entry(head
->prev
, struct ceph_osd_request
,
1807 last_tid
= req
->r_tid
;
1810 ceph_osdc_get_request(req
);
1811 spin_unlock(&ci
->i_unsafe_lock
);
1812 dout("sync_write_wait on tid %llu (until %llu)\n",
1813 req
->r_tid
, last_tid
);
1814 wait_for_completion(&req
->r_safe_completion
);
1815 spin_lock(&ci
->i_unsafe_lock
);
1816 ceph_osdc_put_request(req
);
1819 * from here on look at first entry in chain, since we
1820 * only want to wait for anything older than last_tid
1822 if (list_empty(head
))
1824 req
= list_entry(head
->next
, struct ceph_osd_request
,
1826 } while (req
->r_tid
< last_tid
);
1828 spin_unlock(&ci
->i_unsafe_lock
);
1831 int ceph_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1833 struct inode
*inode
= file
->f_mapping
->host
;
1834 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1839 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1840 sync_write_wait(inode
);
1842 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1845 mutex_lock(&inode
->i_mutex
);
1847 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1848 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1851 * only wait on non-file metadata writeback (the mds
1852 * can recover size and mtime, so we don't need to
1855 if (!datasync
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1856 dout("fsync waiting for flush_tid %u\n", flush_tid
);
1857 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1858 caps_are_flushed(inode
, flush_tid
));
1861 dout("fsync %p%s done\n", inode
, datasync
? " datasync" : "");
1862 mutex_unlock(&inode
->i_mutex
);
1867 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1868 * queue inode for flush but don't do so immediately, because we can
1869 * get by with fewer MDS messages if we wait for data writeback to
1872 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1874 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1878 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1880 dout("write_inode %p wait=%d\n", inode
, wait
);
1882 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1884 err
= wait_event_interruptible(ci
->i_cap_wq
,
1885 caps_are_flushed(inode
, flush_tid
));
1887 struct ceph_mds_client
*mdsc
=
1888 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1890 spin_lock(&ci
->i_ceph_lock
);
1891 if (__ceph_caps_dirty(ci
))
1892 __cap_delay_requeue_front(mdsc
, ci
);
1893 spin_unlock(&ci
->i_ceph_lock
);
1899 * After a recovering MDS goes active, we need to resend any caps
1902 * Caller holds session->s_mutex.
1904 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
1905 struct ceph_mds_session
*session
)
1907 struct ceph_cap_snap
*capsnap
;
1909 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
1910 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
1912 struct ceph_inode_info
*ci
= capsnap
->ci
;
1913 struct inode
*inode
= &ci
->vfs_inode
;
1914 struct ceph_cap
*cap
;
1916 spin_lock(&ci
->i_ceph_lock
);
1917 cap
= ci
->i_auth_cap
;
1918 if (cap
&& cap
->session
== session
) {
1919 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
1921 __ceph_flush_snaps(ci
, &session
, 1);
1923 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1924 cap
, session
->s_mds
);
1926 spin_unlock(&ci
->i_ceph_lock
);
1930 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
1931 struct ceph_mds_session
*session
)
1933 struct ceph_inode_info
*ci
;
1935 kick_flushing_capsnaps(mdsc
, session
);
1937 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
1938 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
1939 struct inode
*inode
= &ci
->vfs_inode
;
1940 struct ceph_cap
*cap
;
1943 spin_lock(&ci
->i_ceph_lock
);
1944 cap
= ci
->i_auth_cap
;
1945 if (cap
&& cap
->session
== session
) {
1946 dout("kick_flushing_caps %p cap %p %s\n", inode
,
1947 cap
, ceph_cap_string(ci
->i_flushing_caps
));
1948 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1949 __ceph_caps_used(ci
),
1950 __ceph_caps_wanted(ci
),
1951 cap
->issued
| cap
->implemented
,
1952 ci
->i_flushing_caps
, NULL
);
1954 spin_lock(&ci
->i_ceph_lock
);
1955 __cap_delay_requeue(mdsc
, ci
);
1956 spin_unlock(&ci
->i_ceph_lock
);
1959 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1960 cap
, session
->s_mds
);
1961 spin_unlock(&ci
->i_ceph_lock
);
1966 static void kick_flushing_inode_caps(struct ceph_mds_client
*mdsc
,
1967 struct ceph_mds_session
*session
,
1968 struct inode
*inode
)
1970 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1971 struct ceph_cap
*cap
;
1974 spin_lock(&ci
->i_ceph_lock
);
1975 cap
= ci
->i_auth_cap
;
1976 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode
,
1977 ceph_cap_string(ci
->i_flushing_caps
), ci
->i_cap_flush_seq
);
1979 __ceph_flush_snaps(ci
, &session
, 1);
1981 if (ci
->i_flushing_caps
) {
1982 spin_lock(&mdsc
->cap_dirty_lock
);
1983 list_move_tail(&ci
->i_flushing_item
,
1984 &cap
->session
->s_cap_flushing
);
1985 spin_unlock(&mdsc
->cap_dirty_lock
);
1987 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1988 __ceph_caps_used(ci
),
1989 __ceph_caps_wanted(ci
),
1990 cap
->issued
| cap
->implemented
,
1991 ci
->i_flushing_caps
, NULL
);
1993 spin_lock(&ci
->i_ceph_lock
);
1994 __cap_delay_requeue(mdsc
, ci
);
1995 spin_unlock(&ci
->i_ceph_lock
);
1998 spin_unlock(&ci
->i_ceph_lock
);
2004 * Take references to capabilities we hold, so that we don't release
2005 * them to the MDS prematurely.
2007 * Protected by i_ceph_lock.
2009 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
)
2011 if (got
& CEPH_CAP_PIN
)
2013 if (got
& CEPH_CAP_FILE_RD
)
2015 if (got
& CEPH_CAP_FILE_CACHE
)
2016 ci
->i_rdcache_ref
++;
2017 if (got
& CEPH_CAP_FILE_WR
)
2019 if (got
& CEPH_CAP_FILE_BUFFER
) {
2020 if (ci
->i_wb_ref
== 0)
2021 ihold(&ci
->vfs_inode
);
2023 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2024 &ci
->vfs_inode
, ci
->i_wb_ref
-1, ci
->i_wb_ref
);
2029 * Try to grab cap references. Specify those refs we @want, and the
2030 * minimal set we @need. Also include the larger offset we are writing
2031 * to (when applicable), and check against max_size here as well.
2032 * Note that caller is responsible for ensuring max_size increases are
2033 * requested from the MDS.
2035 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
2036 int *got
, loff_t endoff
, int *check_max
, int *err
)
2038 struct inode
*inode
= &ci
->vfs_inode
;
2040 int have
, implemented
;
2043 dout("get_cap_refs %p need %s want %s\n", inode
,
2044 ceph_cap_string(need
), ceph_cap_string(want
));
2045 spin_lock(&ci
->i_ceph_lock
);
2047 /* make sure file is actually open */
2048 file_wanted
= __ceph_caps_file_wanted(ci
);
2049 if ((file_wanted
& need
) == 0) {
2050 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2051 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2057 /* finish pending truncate */
2058 while (ci
->i_truncate_pending
) {
2059 spin_unlock(&ci
->i_ceph_lock
);
2060 __ceph_do_pending_vmtruncate(inode
, !(need
& CEPH_CAP_FILE_WR
));
2061 spin_lock(&ci
->i_ceph_lock
);
2064 if (need
& CEPH_CAP_FILE_WR
) {
2065 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2066 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2067 inode
, endoff
, ci
->i_max_size
);
2068 if (endoff
> ci
->i_wanted_max_size
) {
2075 * If a sync write is in progress, we must wait, so that we
2076 * can get a final snapshot value for size+mtime.
2078 if (__ceph_have_pending_cap_snap(ci
)) {
2079 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2083 have
= __ceph_caps_issued(ci
, &implemented
);
2085 if ((have
& need
) == need
) {
2087 * Look at (implemented & ~have & not) so that we keep waiting
2088 * on transition from wanted -> needed caps. This is needed
2089 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2090 * going before a prior buffered writeback happens.
2092 int not = want
& ~(have
& need
);
2093 int revoking
= implemented
& ~have
;
2094 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2095 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2096 ceph_cap_string(revoking
));
2097 if ((revoking
& not) == 0) {
2098 *got
= need
| (have
& want
);
2099 __take_cap_refs(ci
, *got
);
2103 dout("get_cap_refs %p have %s needed %s\n", inode
,
2104 ceph_cap_string(have
), ceph_cap_string(need
));
2107 spin_unlock(&ci
->i_ceph_lock
);
2108 dout("get_cap_refs %p ret %d got %s\n", inode
,
2109 ret
, ceph_cap_string(*got
));
2114 * Check the offset we are writing up to against our current
2115 * max_size. If necessary, tell the MDS we want to write to
2118 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2120 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2123 /* do we need to explicitly request a larger max_size? */
2124 spin_lock(&ci
->i_ceph_lock
);
2125 if ((endoff
>= ci
->i_max_size
||
2126 endoff
> (inode
->i_size
<< 1)) &&
2127 endoff
> ci
->i_wanted_max_size
) {
2128 dout("write %p at large endoff %llu, req max_size\n",
2130 ci
->i_wanted_max_size
= endoff
;
2133 spin_unlock(&ci
->i_ceph_lock
);
2135 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2139 * Wait for caps, and take cap references. If we can't get a WR cap
2140 * due to a small max_size, make sure we check_max_size (and possibly
2141 * ask the mds) so we don't get hung up indefinitely.
2143 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
, int *got
,
2146 int check_max
, ret
, err
;
2150 check_max_size(&ci
->vfs_inode
, endoff
);
2153 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2154 try_get_cap_refs(ci
, need
, want
,
2165 * Take cap refs. Caller must already know we hold at least one ref
2166 * on the caps in question or we don't know this is safe.
2168 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2170 spin_lock(&ci
->i_ceph_lock
);
2171 __take_cap_refs(ci
, caps
);
2172 spin_unlock(&ci
->i_ceph_lock
);
2178 * If we released the last ref on any given cap, call ceph_check_caps
2179 * to release (or schedule a release).
2181 * If we are releasing a WR cap (from a sync write), finalize any affected
2182 * cap_snap, and wake up any waiters.
2184 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2186 struct inode
*inode
= &ci
->vfs_inode
;
2187 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2188 struct ceph_cap_snap
*capsnap
;
2190 spin_lock(&ci
->i_ceph_lock
);
2191 if (had
& CEPH_CAP_PIN
)
2193 if (had
& CEPH_CAP_FILE_RD
)
2194 if (--ci
->i_rd_ref
== 0)
2196 if (had
& CEPH_CAP_FILE_CACHE
)
2197 if (--ci
->i_rdcache_ref
== 0)
2199 if (had
& CEPH_CAP_FILE_BUFFER
) {
2200 if (--ci
->i_wb_ref
== 0) {
2204 dout("put_cap_refs %p wb %d -> %d (?)\n",
2205 inode
, ci
->i_wb_ref
+1, ci
->i_wb_ref
);
2207 if (had
& CEPH_CAP_FILE_WR
)
2208 if (--ci
->i_wr_ref
== 0) {
2210 if (!list_empty(&ci
->i_cap_snaps
)) {
2211 capsnap
= list_first_entry(&ci
->i_cap_snaps
,
2212 struct ceph_cap_snap
,
2214 if (capsnap
->writing
) {
2215 capsnap
->writing
= 0;
2217 __ceph_finish_cap_snap(ci
,
2223 spin_unlock(&ci
->i_ceph_lock
);
2225 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2226 last
? " last" : "", put
? " put" : "");
2228 if (last
&& !flushsnaps
)
2229 ceph_check_caps(ci
, 0, NULL
);
2230 else if (flushsnaps
)
2231 ceph_flush_snaps(ci
);
2233 wake_up_all(&ci
->i_cap_wq
);
2239 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2240 * context. Adjust per-snap dirty page accounting as appropriate.
2241 * Once all dirty data for a cap_snap is flushed, flush snapped file
2242 * metadata back to the MDS. If we dropped the last ref, call
2245 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2246 struct ceph_snap_context
*snapc
)
2248 struct inode
*inode
= &ci
->vfs_inode
;
2250 int complete_capsnap
= 0;
2251 int drop_capsnap
= 0;
2253 struct ceph_cap_snap
*capsnap
= NULL
;
2255 spin_lock(&ci
->i_ceph_lock
);
2256 ci
->i_wrbuffer_ref
-= nr
;
2257 last
= !ci
->i_wrbuffer_ref
;
2259 if (ci
->i_head_snapc
== snapc
) {
2260 ci
->i_wrbuffer_ref_head
-= nr
;
2261 if (ci
->i_wrbuffer_ref_head
== 0 &&
2262 ci
->i_dirty_caps
== 0 && ci
->i_flushing_caps
== 0) {
2263 BUG_ON(!ci
->i_head_snapc
);
2264 ceph_put_snap_context(ci
->i_head_snapc
);
2265 ci
->i_head_snapc
= NULL
;
2267 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2269 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2270 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2271 last
? " LAST" : "");
2273 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2274 if (capsnap
->context
== snapc
) {
2280 capsnap
->dirty_pages
-= nr
;
2281 if (capsnap
->dirty_pages
== 0) {
2282 complete_capsnap
= 1;
2283 if (capsnap
->dirty
== 0)
2284 /* cap writeback completed before we created
2285 * the cap_snap; no FLUSHSNAP is needed */
2288 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2289 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2290 inode
, capsnap
, capsnap
->context
->seq
,
2291 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2292 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2293 last
? " (wrbuffer last)" : "",
2294 complete_capsnap
? " (complete capsnap)" : "",
2295 drop_capsnap
? " (drop capsnap)" : "");
2297 ceph_put_snap_context(capsnap
->context
);
2298 list_del(&capsnap
->ci_item
);
2299 list_del(&capsnap
->flushing_item
);
2300 ceph_put_cap_snap(capsnap
);
2304 spin_unlock(&ci
->i_ceph_lock
);
2307 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2309 } else if (complete_capsnap
) {
2310 ceph_flush_snaps(ci
);
2311 wake_up_all(&ci
->i_cap_wq
);
2318 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2319 * actually be a revocation if it specifies a smaller cap set.)
2321 * caller holds s_mutex and i_ceph_lock, we drop both.
2325 * 1 - check_caps on auth cap only (writeback)
2326 * 2 - check_caps (ack revoke)
2328 static void handle_cap_grant(struct inode
*inode
, struct ceph_mds_caps
*grant
,
2329 struct ceph_mds_session
*session
,
2330 struct ceph_cap
*cap
,
2331 struct ceph_buffer
*xattr_buf
)
2332 __releases(ci
->i_ceph_lock
)
2334 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2335 int mds
= session
->s_mds
;
2336 int seq
= le32_to_cpu(grant
->seq
);
2337 int newcaps
= le32_to_cpu(grant
->caps
);
2338 int issued
, implemented
, used
, wanted
, dirty
;
2339 u64 size
= le64_to_cpu(grant
->size
);
2340 u64 max_size
= le64_to_cpu(grant
->max_size
);
2341 struct timespec mtime
, atime
, ctime
;
2345 int revoked_rdcache
= 0;
2346 int queue_invalidate
= 0;
2348 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2349 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2350 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2354 * If CACHE is being revoked, and we have no dirty buffers,
2355 * try to invalidate (once). (If there are dirty buffers, we
2356 * will invalidate _after_ writeback.)
2358 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2359 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2360 !ci
->i_wrbuffer_ref
) {
2361 if (try_nonblocking_invalidate(inode
) == 0) {
2362 revoked_rdcache
= 1;
2364 /* there were locked pages.. invalidate later
2365 in a separate thread. */
2366 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2367 queue_invalidate
= 1;
2368 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2373 /* side effects now are allowed */
2375 issued
= __ceph_caps_issued(ci
, &implemented
);
2376 issued
|= implemented
| __ceph_caps_dirty(ci
);
2378 cap
->cap_gen
= session
->s_cap_gen
;
2380 __check_cap_issue(ci
, cap
, newcaps
);
2382 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2383 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2384 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(grant
->uid
));
2385 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(grant
->gid
));
2386 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2387 from_kuid(&init_user_ns
, inode
->i_uid
),
2388 from_kgid(&init_user_ns
, inode
->i_gid
));
2391 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
2392 set_nlink(inode
, le32_to_cpu(grant
->nlink
));
2394 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2395 int len
= le32_to_cpu(grant
->xattr_len
);
2396 u64 version
= le64_to_cpu(grant
->xattr_version
);
2398 if (version
> ci
->i_xattrs
.version
) {
2399 dout(" got new xattrs v%llu on %p len %d\n",
2400 version
, inode
, len
);
2401 if (ci
->i_xattrs
.blob
)
2402 ceph_buffer_put(ci
->i_xattrs
.blob
);
2403 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2404 ci
->i_xattrs
.version
= version
;
2408 /* size/ctime/mtime/atime? */
2409 ceph_fill_file_size(inode
, issued
,
2410 le32_to_cpu(grant
->truncate_seq
),
2411 le64_to_cpu(grant
->truncate_size
), size
);
2412 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2413 ceph_decode_timespec(&atime
, &grant
->atime
);
2414 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2415 ceph_fill_file_time(inode
, issued
,
2416 le32_to_cpu(grant
->time_warp_seq
), &ctime
, &mtime
,
2419 /* max size increase? */
2420 if (ci
->i_auth_cap
== cap
&& max_size
!= ci
->i_max_size
) {
2421 dout("max_size %lld -> %llu\n", ci
->i_max_size
, max_size
);
2422 ci
->i_max_size
= max_size
;
2423 if (max_size
>= ci
->i_wanted_max_size
) {
2424 ci
->i_wanted_max_size
= 0; /* reset */
2425 ci
->i_requested_max_size
= 0;
2430 /* check cap bits */
2431 wanted
= __ceph_caps_wanted(ci
);
2432 used
= __ceph_caps_used(ci
);
2433 dirty
= __ceph_caps_dirty(ci
);
2434 dout(" my wanted = %s, used = %s, dirty %s\n",
2435 ceph_cap_string(wanted
),
2436 ceph_cap_string(used
),
2437 ceph_cap_string(dirty
));
2438 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2439 dout("mds wanted %s -> %s\n",
2440 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2441 ceph_cap_string(wanted
));
2442 /* imported cap may not have correct mds_wanted */
2443 if (le32_to_cpu(grant
->op
) == CEPH_CAP_OP_IMPORT
)
2449 /* file layout may have changed */
2450 ci
->i_layout
= grant
->layout
;
2452 /* revocation, grant, or no-op? */
2453 if (cap
->issued
& ~newcaps
) {
2454 int revoking
= cap
->issued
& ~newcaps
;
2456 dout("revocation: %s -> %s (revoking %s)\n",
2457 ceph_cap_string(cap
->issued
),
2458 ceph_cap_string(newcaps
),
2459 ceph_cap_string(revoking
));
2460 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2461 writeback
= 1; /* initiate writeback; will delay ack */
2462 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2463 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2465 ; /* do nothing yet, invalidation will be queued */
2466 else if (cap
== ci
->i_auth_cap
)
2467 check_caps
= 1; /* check auth cap only */
2469 check_caps
= 2; /* check all caps */
2470 cap
->issued
= newcaps
;
2471 cap
->implemented
|= newcaps
;
2472 } else if (cap
->issued
== newcaps
) {
2473 dout("caps unchanged: %s -> %s\n",
2474 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2476 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2477 ceph_cap_string(newcaps
));
2478 cap
->issued
= newcaps
;
2479 cap
->implemented
|= newcaps
; /* add bits only, to
2480 * avoid stepping on a
2481 * pending revocation */
2484 BUG_ON(cap
->issued
& ~cap
->implemented
);
2486 spin_unlock(&ci
->i_ceph_lock
);
2489 * queue inode for writeback: we can't actually call
2490 * filemap_write_and_wait, etc. from message handler
2493 ceph_queue_writeback(inode
);
2494 if (queue_invalidate
)
2495 ceph_queue_invalidate(inode
);
2497 wake_up_all(&ci
->i_cap_wq
);
2499 if (check_caps
== 1)
2500 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2502 else if (check_caps
== 2)
2503 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2505 mutex_unlock(&session
->s_mutex
);
2509 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2510 * MDS has been safely committed.
2512 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2513 struct ceph_mds_caps
*m
,
2514 struct ceph_mds_session
*session
,
2515 struct ceph_cap
*cap
)
2516 __releases(ci
->i_ceph_lock
)
2518 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2519 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2520 unsigned seq
= le32_to_cpu(m
->seq
);
2521 int dirty
= le32_to_cpu(m
->dirty
);
2526 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2527 if ((dirty
& (1 << i
)) &&
2528 flush_tid
== ci
->i_cap_flush_tid
[i
])
2531 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2532 " flushing %s -> %s\n",
2533 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2534 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2535 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2537 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2540 ci
->i_flushing_caps
&= ~cleaned
;
2542 spin_lock(&mdsc
->cap_dirty_lock
);
2543 if (ci
->i_flushing_caps
== 0) {
2544 list_del_init(&ci
->i_flushing_item
);
2545 if (!list_empty(&session
->s_cap_flushing
))
2546 dout(" mds%d still flushing cap on %p\n",
2548 &list_entry(session
->s_cap_flushing
.next
,
2549 struct ceph_inode_info
,
2550 i_flushing_item
)->vfs_inode
);
2551 mdsc
->num_cap_flushing
--;
2552 wake_up_all(&mdsc
->cap_flushing_wq
);
2553 dout(" inode %p now !flushing\n", inode
);
2555 if (ci
->i_dirty_caps
== 0) {
2556 dout(" inode %p now clean\n", inode
);
2557 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2559 if (ci
->i_wrbuffer_ref_head
== 0) {
2560 BUG_ON(!ci
->i_head_snapc
);
2561 ceph_put_snap_context(ci
->i_head_snapc
);
2562 ci
->i_head_snapc
= NULL
;
2565 BUG_ON(list_empty(&ci
->i_dirty_item
));
2568 spin_unlock(&mdsc
->cap_dirty_lock
);
2569 wake_up_all(&ci
->i_cap_wq
);
2572 spin_unlock(&ci
->i_ceph_lock
);
2578 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2579 * throw away our cap_snap.
2581 * Caller hold s_mutex.
2583 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2584 struct ceph_mds_caps
*m
,
2585 struct ceph_mds_session
*session
)
2587 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2588 u64 follows
= le64_to_cpu(m
->snap_follows
);
2589 struct ceph_cap_snap
*capsnap
;
2592 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2593 inode
, ci
, session
->s_mds
, follows
);
2595 spin_lock(&ci
->i_ceph_lock
);
2596 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2597 if (capsnap
->follows
== follows
) {
2598 if (capsnap
->flush_tid
!= flush_tid
) {
2599 dout(" cap_snap %p follows %lld tid %lld !="
2600 " %lld\n", capsnap
, follows
,
2601 flush_tid
, capsnap
->flush_tid
);
2604 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2605 dout(" removing %p cap_snap %p follows %lld\n",
2606 inode
, capsnap
, follows
);
2607 ceph_put_snap_context(capsnap
->context
);
2608 list_del(&capsnap
->ci_item
);
2609 list_del(&capsnap
->flushing_item
);
2610 ceph_put_cap_snap(capsnap
);
2614 dout(" skipping cap_snap %p follows %lld\n",
2615 capsnap
, capsnap
->follows
);
2618 spin_unlock(&ci
->i_ceph_lock
);
2624 * Handle TRUNC from MDS, indicating file truncation.
2626 * caller hold s_mutex.
2628 static void handle_cap_trunc(struct inode
*inode
,
2629 struct ceph_mds_caps
*trunc
,
2630 struct ceph_mds_session
*session
)
2631 __releases(ci
->i_ceph_lock
)
2633 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2634 int mds
= session
->s_mds
;
2635 int seq
= le32_to_cpu(trunc
->seq
);
2636 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2637 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2638 u64 size
= le64_to_cpu(trunc
->size
);
2639 int implemented
= 0;
2640 int dirty
= __ceph_caps_dirty(ci
);
2641 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2642 int queue_trunc
= 0;
2644 issued
|= implemented
| dirty
;
2646 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2647 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2648 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2649 truncate_seq
, truncate_size
, size
);
2650 spin_unlock(&ci
->i_ceph_lock
);
2653 ceph_queue_vmtruncate(inode
);
2657 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2658 * different one. If we are the most recent migration we've seen (as
2659 * indicated by mseq), make note of the migrating cap bits for the
2660 * duration (until we see the corresponding IMPORT).
2662 * caller holds s_mutex
2664 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2665 struct ceph_mds_session
*session
,
2666 int *open_target_sessions
)
2668 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2669 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2670 int mds
= session
->s_mds
;
2671 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2672 struct ceph_cap
*cap
= NULL
, *t
;
2676 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2677 inode
, ci
, mds
, mseq
);
2679 spin_lock(&ci
->i_ceph_lock
);
2681 /* make sure we haven't seen a higher mseq */
2682 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
2683 t
= rb_entry(p
, struct ceph_cap
, ci_node
);
2684 if (ceph_seq_cmp(t
->mseq
, mseq
) > 0) {
2685 dout(" higher mseq on cap from mds%d\n",
2689 if (t
->session
->s_mds
== mds
)
2696 ci
->i_cap_exporting_mds
= mds
;
2697 ci
->i_cap_exporting_mseq
= mseq
;
2698 ci
->i_cap_exporting_issued
= cap
->issued
;
2701 * make sure we have open sessions with all possible
2702 * export targets, so that we get the matching IMPORT
2704 *open_target_sessions
= 1;
2707 * we can't flush dirty caps that we've seen the
2708 * EXPORT but no IMPORT for
2710 spin_lock(&mdsc
->cap_dirty_lock
);
2711 if (!list_empty(&ci
->i_dirty_item
)) {
2712 dout(" moving %p to cap_dirty_migrating\n",
2714 list_move(&ci
->i_dirty_item
,
2715 &mdsc
->cap_dirty_migrating
);
2717 spin_unlock(&mdsc
->cap_dirty_lock
);
2719 __ceph_remove_cap(cap
);
2721 /* else, we already released it */
2723 spin_unlock(&ci
->i_ceph_lock
);
2727 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2730 * caller holds s_mutex.
2732 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
2733 struct inode
*inode
, struct ceph_mds_caps
*im
,
2734 struct ceph_mds_session
*session
,
2735 void *snaptrace
, int snaptrace_len
)
2737 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2738 int mds
= session
->s_mds
;
2739 unsigned issued
= le32_to_cpu(im
->caps
);
2740 unsigned wanted
= le32_to_cpu(im
->wanted
);
2741 unsigned seq
= le32_to_cpu(im
->seq
);
2742 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
2743 u64 realmino
= le64_to_cpu(im
->realm
);
2744 u64 cap_id
= le64_to_cpu(im
->cap_id
);
2746 if (ci
->i_cap_exporting_mds
>= 0 &&
2747 ceph_seq_cmp(ci
->i_cap_exporting_mseq
, mseq
) < 0) {
2748 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2749 " - cleared exporting from mds%d\n",
2750 inode
, ci
, mds
, mseq
,
2751 ci
->i_cap_exporting_mds
);
2752 ci
->i_cap_exporting_issued
= 0;
2753 ci
->i_cap_exporting_mseq
= 0;
2754 ci
->i_cap_exporting_mds
= -1;
2756 spin_lock(&mdsc
->cap_dirty_lock
);
2757 if (!list_empty(&ci
->i_dirty_item
)) {
2758 dout(" moving %p back to cap_dirty\n", inode
);
2759 list_move(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
2761 spin_unlock(&mdsc
->cap_dirty_lock
);
2763 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2764 inode
, ci
, mds
, mseq
);
2767 down_write(&mdsc
->snap_rwsem
);
2768 ceph_update_snap_trace(mdsc
, snaptrace
, snaptrace
+snaptrace_len
,
2770 downgrade_write(&mdsc
->snap_rwsem
);
2771 ceph_add_cap(inode
, session
, cap_id
, -1,
2772 issued
, wanted
, seq
, mseq
, realmino
, CEPH_CAP_FLAG_AUTH
,
2773 NULL
/* no caps context */);
2774 kick_flushing_inode_caps(mdsc
, session
, inode
);
2775 up_read(&mdsc
->snap_rwsem
);
2777 /* make sure we re-request max_size, if necessary */
2778 spin_lock(&ci
->i_ceph_lock
);
2779 ci
->i_wanted_max_size
= 0; /* reset */
2780 ci
->i_requested_max_size
= 0;
2781 spin_unlock(&ci
->i_ceph_lock
);
2785 * Handle a caps message from the MDS.
2787 * Identify the appropriate session, inode, and call the right handler
2788 * based on the cap op.
2790 void ceph_handle_caps(struct ceph_mds_session
*session
,
2791 struct ceph_msg
*msg
)
2793 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2794 struct super_block
*sb
= mdsc
->fsc
->sb
;
2795 struct inode
*inode
;
2796 struct ceph_inode_info
*ci
;
2797 struct ceph_cap
*cap
;
2798 struct ceph_mds_caps
*h
;
2799 int mds
= session
->s_mds
;
2802 struct ceph_vino vino
;
2807 size_t snaptrace_len
;
2810 int open_target_sessions
= 0;
2812 dout("handle_caps from mds%d\n", mds
);
2815 tid
= le64_to_cpu(msg
->hdr
.tid
);
2816 if (msg
->front
.iov_len
< sizeof(*h
))
2818 h
= msg
->front
.iov_base
;
2819 op
= le32_to_cpu(h
->op
);
2820 vino
.ino
= le64_to_cpu(h
->ino
);
2821 vino
.snap
= CEPH_NOSNAP
;
2822 cap_id
= le64_to_cpu(h
->cap_id
);
2823 seq
= le32_to_cpu(h
->seq
);
2824 mseq
= le32_to_cpu(h
->migrate_seq
);
2825 size
= le64_to_cpu(h
->size
);
2826 max_size
= le64_to_cpu(h
->max_size
);
2829 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
2831 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
2834 p
= snaptrace
+ snaptrace_len
;
2835 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
2836 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
2843 mutex_lock(&session
->s_mutex
);
2845 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
2848 if (op
== CEPH_CAP_OP_IMPORT
)
2849 ceph_add_cap_releases(mdsc
, session
);
2852 inode
= ceph_find_inode(sb
, vino
);
2853 ci
= ceph_inode(inode
);
2854 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
2857 dout(" i don't have ino %llx\n", vino
.ino
);
2859 if (op
== CEPH_CAP_OP_IMPORT
)
2860 __queue_cap_release(session
, vino
.ino
, cap_id
,
2862 goto flush_cap_releases
;
2865 /* these will work even if we don't have a cap yet */
2867 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
2868 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
2871 case CEPH_CAP_OP_EXPORT
:
2872 handle_cap_export(inode
, h
, session
, &open_target_sessions
);
2875 case CEPH_CAP_OP_IMPORT
:
2876 handle_cap_import(mdsc
, inode
, h
, session
,
2877 snaptrace
, snaptrace_len
);
2880 /* the rest require a cap */
2881 spin_lock(&ci
->i_ceph_lock
);
2882 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
2884 dout(" no cap on %p ino %llx.%llx from mds%d\n",
2885 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
2886 spin_unlock(&ci
->i_ceph_lock
);
2887 goto flush_cap_releases
;
2890 /* note that each of these drops i_ceph_lock for us */
2892 case CEPH_CAP_OP_REVOKE
:
2893 case CEPH_CAP_OP_GRANT
:
2894 case CEPH_CAP_OP_IMPORT
:
2895 handle_cap_grant(inode
, h
, session
, cap
, msg
->middle
);
2898 case CEPH_CAP_OP_FLUSH_ACK
:
2899 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
2902 case CEPH_CAP_OP_TRUNC
:
2903 handle_cap_trunc(inode
, h
, session
);
2907 spin_unlock(&ci
->i_ceph_lock
);
2908 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
2909 ceph_cap_op_name(op
));
2916 * send any full release message to try to move things
2917 * along for the mds (who clearly thinks we still have this
2920 ceph_add_cap_releases(mdsc
, session
);
2921 ceph_send_cap_releases(mdsc
, session
);
2924 mutex_unlock(&session
->s_mutex
);
2928 if (open_target_sessions
)
2929 ceph_mdsc_open_export_target_sessions(mdsc
, session
);
2933 pr_err("ceph_handle_caps: corrupt message\n");
2939 * Delayed work handler to process end of delayed cap release LRU list.
2941 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
2943 struct ceph_inode_info
*ci
;
2944 int flags
= CHECK_CAPS_NODELAY
;
2946 dout("check_delayed_caps\n");
2948 spin_lock(&mdsc
->cap_delay_lock
);
2949 if (list_empty(&mdsc
->cap_delay_list
))
2951 ci
= list_first_entry(&mdsc
->cap_delay_list
,
2952 struct ceph_inode_info
,
2954 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
2955 time_before(jiffies
, ci
->i_hold_caps_max
))
2957 list_del_init(&ci
->i_cap_delay_list
);
2958 spin_unlock(&mdsc
->cap_delay_lock
);
2959 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
2960 ceph_check_caps(ci
, flags
, NULL
);
2962 spin_unlock(&mdsc
->cap_delay_lock
);
2966 * Flush all dirty caps to the mds
2968 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
2970 struct ceph_inode_info
*ci
;
2971 struct inode
*inode
;
2973 dout("flush_dirty_caps\n");
2974 spin_lock(&mdsc
->cap_dirty_lock
);
2975 while (!list_empty(&mdsc
->cap_dirty
)) {
2976 ci
= list_first_entry(&mdsc
->cap_dirty
, struct ceph_inode_info
,
2978 inode
= &ci
->vfs_inode
;
2980 dout("flush_dirty_caps %p\n", inode
);
2981 spin_unlock(&mdsc
->cap_dirty_lock
);
2982 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
, NULL
);
2984 spin_lock(&mdsc
->cap_dirty_lock
);
2986 spin_unlock(&mdsc
->cap_dirty_lock
);
2987 dout("flush_dirty_caps done\n");
2991 * Drop open file reference. If we were the last open file,
2992 * we may need to release capabilities to the MDS (or schedule
2993 * their delayed release).
2995 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
2997 struct inode
*inode
= &ci
->vfs_inode
;
3000 spin_lock(&ci
->i_ceph_lock
);
3001 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
3002 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
3003 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
3004 if (--ci
->i_nr_by_mode
[fmode
] == 0)
3006 spin_unlock(&ci
->i_ceph_lock
);
3008 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
3009 ceph_check_caps(ci
, 0, NULL
);
3013 * Helpers for embedding cap and dentry lease releases into mds
3016 * @force is used by dentry_release (below) to force inclusion of a
3017 * record for the directory inode, even when there aren't any caps to
3020 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
3021 int mds
, int drop
, int unless
, int force
)
3023 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3024 struct ceph_cap
*cap
;
3025 struct ceph_mds_request_release
*rel
= *p
;
3029 spin_lock(&ci
->i_ceph_lock
);
3030 used
= __ceph_caps_used(ci
);
3031 dirty
= __ceph_caps_dirty(ci
);
3033 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3034 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
3035 ceph_cap_string(unless
));
3037 /* only drop unused, clean caps */
3038 drop
&= ~(used
| dirty
);
3040 cap
= __get_cap_for_mds(ci
, mds
);
3041 if (cap
&& __cap_is_valid(cap
)) {
3043 ((cap
->issued
& drop
) &&
3044 (cap
->issued
& unless
) == 0)) {
3045 if ((cap
->issued
& drop
) &&
3046 (cap
->issued
& unless
) == 0) {
3047 int wanted
= __ceph_caps_wanted(ci
);
3048 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0)
3049 wanted
|= cap
->mds_wanted
;
3050 dout("encode_inode_release %p cap %p "
3051 "%s -> %s, wanted %s -> %s\n", inode
, cap
,
3052 ceph_cap_string(cap
->issued
),
3053 ceph_cap_string(cap
->issued
& ~drop
),
3054 ceph_cap_string(cap
->mds_wanted
),
3055 ceph_cap_string(wanted
));
3057 cap
->issued
&= ~drop
;
3058 cap
->implemented
&= ~drop
;
3059 cap
->mds_wanted
= wanted
;
3061 dout("encode_inode_release %p cap %p %s"
3062 " (force)\n", inode
, cap
,
3063 ceph_cap_string(cap
->issued
));
3066 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3067 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3068 rel
->seq
= cpu_to_le32(cap
->seq
);
3069 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
),
3070 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3071 rel
->caps
= cpu_to_le32(cap
->issued
);
3072 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3078 dout("encode_inode_release %p cap %p %s\n",
3079 inode
, cap
, ceph_cap_string(cap
->issued
));
3082 spin_unlock(&ci
->i_ceph_lock
);
3086 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3087 int mds
, int drop
, int unless
)
3089 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3090 struct ceph_mds_request_release
*rel
= *p
;
3091 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3096 * force an record for the directory caps if we have a dentry lease.
3097 * this is racy (can't take i_ceph_lock and d_lock together), but it
3098 * doesn't have to be perfect; the mds will revoke anything we don't
3101 spin_lock(&dentry
->d_lock
);
3102 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3104 spin_unlock(&dentry
->d_lock
);
3106 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3108 spin_lock(&dentry
->d_lock
);
3109 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3110 dout("encode_dentry_release %p mds%d seq %d\n",
3111 dentry
, mds
, (int)di
->lease_seq
);
3112 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3113 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3114 *p
+= dentry
->d_name
.len
;
3115 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3116 __ceph_mdsc_drop_dentry_lease(dentry
);
3118 spin_unlock(&dentry
->d_lock
);