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 int ceph_reserve_caps(struct ceph_mds_client
*mdsc
,
151 struct ceph_cap_reservation
*ctx
, int need
)
154 struct ceph_cap
*cap
;
160 dout("reserve caps ctx=%p need=%d\n", ctx
, need
);
162 /* first reserve any caps that are already allocated */
163 spin_lock(&mdsc
->caps_list_lock
);
164 if (mdsc
->caps_avail_count
>= need
)
167 have
= mdsc
->caps_avail_count
;
168 mdsc
->caps_avail_count
-= have
;
169 mdsc
->caps_reserve_count
+= have
;
170 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
171 mdsc
->caps_reserve_count
+
172 mdsc
->caps_avail_count
);
173 spin_unlock(&mdsc
->caps_list_lock
);
175 for (i
= have
; i
< need
; i
++) {
176 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
179 goto out_alloc_count
;
181 list_add(&cap
->caps_item
, &newcaps
);
184 BUG_ON(have
+ alloc
!= need
);
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
);
203 /* we didn't manage to reserve as much as we needed */
204 pr_warning("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
209 int ceph_unreserve_caps(struct ceph_mds_client
*mdsc
,
210 struct ceph_cap_reservation
*ctx
)
212 dout("unreserve caps ctx=%p count=%d\n", ctx
, ctx
->count
);
214 spin_lock(&mdsc
->caps_list_lock
);
215 BUG_ON(mdsc
->caps_reserve_count
< ctx
->count
);
216 mdsc
->caps_reserve_count
-= ctx
->count
;
217 mdsc
->caps_avail_count
+= ctx
->count
;
219 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
220 mdsc
->caps_total_count
, mdsc
->caps_use_count
,
221 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
222 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
223 mdsc
->caps_reserve_count
+
224 mdsc
->caps_avail_count
);
225 spin_unlock(&mdsc
->caps_list_lock
);
230 static struct ceph_cap
*get_cap(struct ceph_mds_client
*mdsc
,
231 struct ceph_cap_reservation
*ctx
)
233 struct ceph_cap
*cap
= NULL
;
235 /* temporary, until we do something about cap import/export */
237 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
239 spin_lock(&mdsc
->caps_list_lock
);
240 mdsc
->caps_use_count
++;
241 mdsc
->caps_total_count
++;
242 spin_unlock(&mdsc
->caps_list_lock
);
247 spin_lock(&mdsc
->caps_list_lock
);
248 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
249 ctx
, ctx
->count
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
250 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
252 BUG_ON(ctx
->count
> mdsc
->caps_reserve_count
);
253 BUG_ON(list_empty(&mdsc
->caps_list
));
256 mdsc
->caps_reserve_count
--;
257 mdsc
->caps_use_count
++;
259 cap
= list_first_entry(&mdsc
->caps_list
, struct ceph_cap
, caps_item
);
260 list_del(&cap
->caps_item
);
262 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
263 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
264 spin_unlock(&mdsc
->caps_list_lock
);
268 void ceph_put_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
)
270 spin_lock(&mdsc
->caps_list_lock
);
271 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
272 cap
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
273 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
274 mdsc
->caps_use_count
--;
276 * Keep some preallocated caps around (ceph_min_count), to
277 * avoid lots of free/alloc churn.
279 if (mdsc
->caps_avail_count
>= mdsc
->caps_reserve_count
+
280 mdsc
->caps_min_count
) {
281 mdsc
->caps_total_count
--;
282 kmem_cache_free(ceph_cap_cachep
, cap
);
284 mdsc
->caps_avail_count
++;
285 list_add(&cap
->caps_item
, &mdsc
->caps_list
);
288 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
289 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
290 spin_unlock(&mdsc
->caps_list_lock
);
293 void ceph_reservation_status(struct ceph_fs_client
*fsc
,
294 int *total
, int *avail
, int *used
, int *reserved
,
297 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
300 *total
= mdsc
->caps_total_count
;
302 *avail
= mdsc
->caps_avail_count
;
304 *used
= mdsc
->caps_use_count
;
306 *reserved
= mdsc
->caps_reserve_count
;
308 *min
= mdsc
->caps_min_count
;
312 * Find ceph_cap for given mds, if any.
314 * Called with i_ceph_lock held.
316 static struct ceph_cap
*__get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
318 struct ceph_cap
*cap
;
319 struct rb_node
*n
= ci
->i_caps
.rb_node
;
322 cap
= rb_entry(n
, struct ceph_cap
, ci_node
);
325 else if (mds
> cap
->mds
)
333 struct ceph_cap
*ceph_get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
335 struct ceph_cap
*cap
;
337 spin_lock(&ci
->i_ceph_lock
);
338 cap
= __get_cap_for_mds(ci
, mds
);
339 spin_unlock(&ci
->i_ceph_lock
);
344 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
346 static int __ceph_get_cap_mds(struct ceph_inode_info
*ci
)
348 struct ceph_cap
*cap
;
352 /* prefer mds with WR|BUFFER|EXCL caps */
353 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
354 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
356 if (cap
->issued
& (CEPH_CAP_FILE_WR
|
357 CEPH_CAP_FILE_BUFFER
|
364 int ceph_get_cap_mds(struct inode
*inode
)
366 struct ceph_inode_info
*ci
= ceph_inode(inode
);
368 spin_lock(&ci
->i_ceph_lock
);
369 mds
= __ceph_get_cap_mds(ceph_inode(inode
));
370 spin_unlock(&ci
->i_ceph_lock
);
375 * Called under i_ceph_lock.
377 static void __insert_cap_node(struct ceph_inode_info
*ci
,
378 struct ceph_cap
*new)
380 struct rb_node
**p
= &ci
->i_caps
.rb_node
;
381 struct rb_node
*parent
= NULL
;
382 struct ceph_cap
*cap
= NULL
;
386 cap
= rb_entry(parent
, struct ceph_cap
, ci_node
);
387 if (new->mds
< cap
->mds
)
389 else if (new->mds
> cap
->mds
)
395 rb_link_node(&new->ci_node
, parent
, p
);
396 rb_insert_color(&new->ci_node
, &ci
->i_caps
);
400 * (re)set cap hold timeouts, which control the delayed release
401 * of unused caps back to the MDS. Should be called on cap use.
403 static void __cap_set_timeouts(struct ceph_mds_client
*mdsc
,
404 struct ceph_inode_info
*ci
)
406 struct ceph_mount_options
*ma
= mdsc
->fsc
->mount_options
;
408 ci
->i_hold_caps_min
= round_jiffies(jiffies
+
409 ma
->caps_wanted_delay_min
* HZ
);
410 ci
->i_hold_caps_max
= round_jiffies(jiffies
+
411 ma
->caps_wanted_delay_max
* HZ
);
412 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci
->vfs_inode
,
413 ci
->i_hold_caps_min
- jiffies
, ci
->i_hold_caps_max
- jiffies
);
417 * (Re)queue cap at the end of the delayed cap release list.
419 * If I_FLUSH is set, leave the inode at the front of the list.
421 * Caller holds i_ceph_lock
422 * -> we take mdsc->cap_delay_lock
424 static void __cap_delay_requeue(struct ceph_mds_client
*mdsc
,
425 struct ceph_inode_info
*ci
)
427 __cap_set_timeouts(mdsc
, ci
);
428 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci
->vfs_inode
,
429 ci
->i_ceph_flags
, ci
->i_hold_caps_max
);
430 if (!mdsc
->stopping
) {
431 spin_lock(&mdsc
->cap_delay_lock
);
432 if (!list_empty(&ci
->i_cap_delay_list
)) {
433 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
435 list_del_init(&ci
->i_cap_delay_list
);
437 list_add_tail(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
439 spin_unlock(&mdsc
->cap_delay_lock
);
444 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
445 * indicating we should send a cap message to flush dirty metadata
446 * asap, and move to the front of the delayed cap list.
448 static void __cap_delay_requeue_front(struct ceph_mds_client
*mdsc
,
449 struct ceph_inode_info
*ci
)
451 dout("__cap_delay_requeue_front %p\n", &ci
->vfs_inode
);
452 spin_lock(&mdsc
->cap_delay_lock
);
453 ci
->i_ceph_flags
|= CEPH_I_FLUSH
;
454 if (!list_empty(&ci
->i_cap_delay_list
))
455 list_del_init(&ci
->i_cap_delay_list
);
456 list_add(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
457 spin_unlock(&mdsc
->cap_delay_lock
);
461 * Cancel delayed work on cap.
463 * Caller must hold i_ceph_lock.
465 static void __cap_delay_cancel(struct ceph_mds_client
*mdsc
,
466 struct ceph_inode_info
*ci
)
468 dout("__cap_delay_cancel %p\n", &ci
->vfs_inode
);
469 if (list_empty(&ci
->i_cap_delay_list
))
471 spin_lock(&mdsc
->cap_delay_lock
);
472 list_del_init(&ci
->i_cap_delay_list
);
473 spin_unlock(&mdsc
->cap_delay_lock
);
477 * Common issue checks for add_cap, handle_cap_grant.
479 static void __check_cap_issue(struct ceph_inode_info
*ci
, struct ceph_cap
*cap
,
482 unsigned had
= __ceph_caps_issued(ci
, NULL
);
485 * Each time we receive FILE_CACHE anew, we increment
488 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
489 (had
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
493 * if we are newly issued FILE_SHARED, clear D_COMPLETE; we
494 * don't know what happened to this directory while we didn't
497 if ((issued
& CEPH_CAP_FILE_SHARED
) &&
498 (had
& CEPH_CAP_FILE_SHARED
) == 0) {
500 if (S_ISDIR(ci
->vfs_inode
.i_mode
))
501 ceph_dir_clear_complete(&ci
->vfs_inode
);
506 * Add a capability under the given MDS session.
508 * Caller should hold session snap_rwsem (read) and s_mutex.
510 * @fmode is the open file mode, if we are opening a file, otherwise
511 * it is < 0. (This is so we can atomically add the cap and add an
512 * open file reference to it.)
514 int ceph_add_cap(struct inode
*inode
,
515 struct ceph_mds_session
*session
, u64 cap_id
,
516 int fmode
, unsigned issued
, unsigned wanted
,
517 unsigned seq
, unsigned mseq
, u64 realmino
, int flags
,
518 struct ceph_cap_reservation
*caps_reservation
)
520 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
521 struct ceph_inode_info
*ci
= ceph_inode(inode
);
522 struct ceph_cap
*new_cap
= NULL
;
523 struct ceph_cap
*cap
;
524 int mds
= session
->s_mds
;
527 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
528 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
531 * If we are opening the file, include file mode wanted bits
535 wanted
|= ceph_caps_for_mode(fmode
);
538 spin_lock(&ci
->i_ceph_lock
);
539 cap
= __get_cap_for_mds(ci
, mds
);
545 spin_unlock(&ci
->i_ceph_lock
);
546 new_cap
= get_cap(mdsc
, caps_reservation
);
553 cap
->implemented
= 0;
558 __insert_cap_node(ci
, cap
);
560 /* clear out old exporting info? (i.e. on cap import) */
561 if (ci
->i_cap_exporting_mds
== mds
) {
562 ci
->i_cap_exporting_issued
= 0;
563 ci
->i_cap_exporting_mseq
= 0;
564 ci
->i_cap_exporting_mds
= -1;
567 /* add to session cap list */
568 cap
->session
= session
;
569 spin_lock(&session
->s_cap_lock
);
570 list_add_tail(&cap
->session_caps
, &session
->s_caps
);
571 session
->s_nr_caps
++;
572 spin_unlock(&session
->s_cap_lock
);
574 ceph_put_cap(mdsc
, new_cap
);
576 if (!ci
->i_snap_realm
) {
578 * add this inode to the appropriate snap realm
580 struct ceph_snap_realm
*realm
= ceph_lookup_snap_realm(mdsc
,
583 ceph_get_snap_realm(mdsc
, realm
);
584 spin_lock(&realm
->inodes_with_caps_lock
);
585 ci
->i_snap_realm
= realm
;
586 list_add(&ci
->i_snap_realm_item
,
587 &realm
->inodes_with_caps
);
588 spin_unlock(&realm
->inodes_with_caps_lock
);
590 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
596 __check_cap_issue(ci
, cap
, issued
);
599 * If we are issued caps we don't want, or the mds' wanted
600 * value appears to be off, queue a check so we'll release
601 * later and/or update the mds wanted value.
603 actual_wanted
= __ceph_caps_wanted(ci
);
604 if ((wanted
& ~actual_wanted
) ||
605 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
606 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
607 ceph_cap_string(issued
), ceph_cap_string(wanted
),
608 ceph_cap_string(actual_wanted
));
609 __cap_delay_requeue(mdsc
, ci
);
612 if (flags
& CEPH_CAP_FLAG_AUTH
)
613 ci
->i_auth_cap
= cap
;
614 else if (ci
->i_auth_cap
== cap
) {
615 ci
->i_auth_cap
= NULL
;
616 spin_lock(&mdsc
->cap_dirty_lock
);
617 if (!list_empty(&ci
->i_dirty_item
)) {
618 dout(" moving %p to cap_dirty_migrating\n", inode
);
619 list_move(&ci
->i_dirty_item
,
620 &mdsc
->cap_dirty_migrating
);
622 spin_unlock(&mdsc
->cap_dirty_lock
);
625 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
626 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
627 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
628 cap
->cap_id
= cap_id
;
629 cap
->issued
= issued
;
630 cap
->implemented
|= issued
;
631 cap
->mds_wanted
|= wanted
;
633 cap
->issue_seq
= seq
;
635 cap
->cap_gen
= session
->s_cap_gen
;
638 __ceph_get_fmode(ci
, fmode
);
639 spin_unlock(&ci
->i_ceph_lock
);
640 wake_up_all(&ci
->i_cap_wq
);
645 * Return true if cap has not timed out and belongs to the current
646 * generation of the MDS session (i.e. has not gone 'stale' due to
647 * us losing touch with the mds).
649 static int __cap_is_valid(struct ceph_cap
*cap
)
654 spin_lock(&cap
->session
->s_gen_ttl_lock
);
655 gen
= cap
->session
->s_cap_gen
;
656 ttl
= cap
->session
->s_cap_ttl
;
657 spin_unlock(&cap
->session
->s_gen_ttl_lock
);
659 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
660 dout("__cap_is_valid %p cap %p issued %s "
661 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
662 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
670 * Return set of valid cap bits issued to us. Note that caps time
671 * out, and may be invalidated in bulk if the client session times out
672 * and session->s_cap_gen is bumped.
674 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
676 int have
= ci
->i_snap_caps
| ci
->i_cap_exporting_issued
;
677 struct ceph_cap
*cap
;
682 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
683 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
684 if (!__cap_is_valid(cap
))
686 dout("__ceph_caps_issued %p cap %p issued %s\n",
687 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
690 *implemented
|= cap
->implemented
;
696 * Get cap bits issued by caps other than @ocap
698 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
700 int have
= ci
->i_snap_caps
;
701 struct ceph_cap
*cap
;
704 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
705 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
708 if (!__cap_is_valid(cap
))
716 * Move a cap to the end of the LRU (oldest caps at list head, newest
719 static void __touch_cap(struct ceph_cap
*cap
)
721 struct ceph_mds_session
*s
= cap
->session
;
723 spin_lock(&s
->s_cap_lock
);
724 if (s
->s_cap_iterator
== NULL
) {
725 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
727 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
729 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
730 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
732 spin_unlock(&s
->s_cap_lock
);
736 * Check if we hold the given mask. If so, move the cap(s) to the
737 * front of their respective LRUs. (This is the preferred way for
738 * callers to check for caps they want.)
740 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
742 struct ceph_cap
*cap
;
744 int have
= ci
->i_snap_caps
;
746 if ((have
& mask
) == mask
) {
747 dout("__ceph_caps_issued_mask %p snap issued %s"
748 " (mask %s)\n", &ci
->vfs_inode
,
749 ceph_cap_string(have
),
750 ceph_cap_string(mask
));
754 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
755 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
756 if (!__cap_is_valid(cap
))
758 if ((cap
->issued
& mask
) == mask
) {
759 dout("__ceph_caps_issued_mask %p cap %p issued %s"
760 " (mask %s)\n", &ci
->vfs_inode
, cap
,
761 ceph_cap_string(cap
->issued
),
762 ceph_cap_string(mask
));
768 /* does a combination of caps satisfy mask? */
770 if ((have
& mask
) == mask
) {
771 dout("__ceph_caps_issued_mask %p combo issued %s"
772 " (mask %s)\n", &ci
->vfs_inode
,
773 ceph_cap_string(cap
->issued
),
774 ceph_cap_string(mask
));
778 /* touch this + preceding caps */
780 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
782 cap
= rb_entry(q
, struct ceph_cap
,
784 if (!__cap_is_valid(cap
))
797 * Return true if mask caps are currently being revoked by an MDS.
799 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
801 struct inode
*inode
= &ci
->vfs_inode
;
802 struct ceph_cap
*cap
;
806 spin_lock(&ci
->i_ceph_lock
);
807 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
808 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
809 if (__cap_is_valid(cap
) &&
810 (cap
->implemented
& ~cap
->issued
& mask
)) {
815 spin_unlock(&ci
->i_ceph_lock
);
816 dout("ceph_caps_revoking %p %s = %d\n", inode
,
817 ceph_cap_string(mask
), ret
);
821 int __ceph_caps_used(struct ceph_inode_info
*ci
)
825 used
|= CEPH_CAP_PIN
;
827 used
|= CEPH_CAP_FILE_RD
;
828 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
829 used
|= CEPH_CAP_FILE_CACHE
;
831 used
|= CEPH_CAP_FILE_WR
;
832 if (ci
->i_wb_ref
|| ci
->i_wrbuffer_ref
)
833 used
|= CEPH_CAP_FILE_BUFFER
;
838 * wanted, by virtue of open file modes
840 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
844 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
845 if (ci
->i_nr_by_mode
[mode
])
846 want
|= ceph_caps_for_mode(mode
);
851 * Return caps we have registered with the MDS(s) as 'wanted'.
853 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
855 struct ceph_cap
*cap
;
859 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
860 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
861 if (!__cap_is_valid(cap
))
863 mds_wanted
|= cap
->mds_wanted
;
869 * called under i_ceph_lock
871 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
873 return !RB_EMPTY_ROOT(&ci
->i_caps
) || ci
->i_cap_exporting_mds
>= 0;
877 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
879 * caller should hold i_ceph_lock.
880 * caller will not hold session s_mutex if called from destroy_inode.
882 void __ceph_remove_cap(struct ceph_cap
*cap
)
884 struct ceph_mds_session
*session
= cap
->session
;
885 struct ceph_inode_info
*ci
= cap
->ci
;
886 struct ceph_mds_client
*mdsc
=
887 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
890 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
892 /* remove from session list */
893 spin_lock(&session
->s_cap_lock
);
894 if (session
->s_cap_iterator
== cap
) {
895 /* not yet, we are iterating over this very cap */
896 dout("__ceph_remove_cap delaying %p removal from session %p\n",
899 list_del_init(&cap
->session_caps
);
900 session
->s_nr_caps
--;
904 /* protect backpointer with s_cap_lock: see iterate_session_caps */
906 spin_unlock(&session
->s_cap_lock
);
908 /* remove from inode list */
909 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
910 if (ci
->i_auth_cap
== cap
)
911 ci
->i_auth_cap
= NULL
;
914 ceph_put_cap(mdsc
, cap
);
916 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
) {
917 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
918 spin_lock(&realm
->inodes_with_caps_lock
);
919 list_del_init(&ci
->i_snap_realm_item
);
920 ci
->i_snap_realm_counter
++;
921 ci
->i_snap_realm
= NULL
;
922 spin_unlock(&realm
->inodes_with_caps_lock
);
923 ceph_put_snap_realm(mdsc
, realm
);
925 if (!__ceph_is_any_real_caps(ci
))
926 __cap_delay_cancel(mdsc
, ci
);
930 * Build and send a cap message to the given MDS.
932 * Caller should be holding s_mutex.
934 static int send_cap_msg(struct ceph_mds_session
*session
,
935 u64 ino
, u64 cid
, int op
,
936 int caps
, int wanted
, int dirty
,
937 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
938 u64 size
, u64 max_size
,
939 struct timespec
*mtime
, struct timespec
*atime
,
941 uid_t uid
, gid_t gid
, umode_t mode
,
943 struct ceph_buffer
*xattrs_buf
,
946 struct ceph_mds_caps
*fc
;
947 struct ceph_msg
*msg
;
949 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
950 " seq %u/%u mseq %u follows %lld size %llu/%llu"
951 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
952 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
953 ceph_cap_string(dirty
),
954 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
955 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
957 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
), GFP_NOFS
, false);
961 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
963 fc
= msg
->front
.iov_base
;
964 memset(fc
, 0, sizeof(*fc
));
966 fc
->cap_id
= cpu_to_le64(cid
);
967 fc
->op
= cpu_to_le32(op
);
968 fc
->seq
= cpu_to_le32(seq
);
969 fc
->issue_seq
= cpu_to_le32(issue_seq
);
970 fc
->migrate_seq
= cpu_to_le32(mseq
);
971 fc
->caps
= cpu_to_le32(caps
);
972 fc
->wanted
= cpu_to_le32(wanted
);
973 fc
->dirty
= cpu_to_le32(dirty
);
974 fc
->ino
= cpu_to_le64(ino
);
975 fc
->snap_follows
= cpu_to_le64(follows
);
977 fc
->size
= cpu_to_le64(size
);
978 fc
->max_size
= cpu_to_le64(max_size
);
980 ceph_encode_timespec(&fc
->mtime
, mtime
);
982 ceph_encode_timespec(&fc
->atime
, atime
);
983 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
985 fc
->uid
= cpu_to_le32(uid
);
986 fc
->gid
= cpu_to_le32(gid
);
987 fc
->mode
= cpu_to_le32(mode
);
989 fc
->xattr_version
= cpu_to_le64(xattr_version
);
991 msg
->middle
= ceph_buffer_get(xattrs_buf
);
992 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
993 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
996 ceph_con_send(&session
->s_con
, msg
);
1000 static void __queue_cap_release(struct ceph_mds_session
*session
,
1001 u64 ino
, u64 cap_id
, u32 migrate_seq
,
1004 struct ceph_msg
*msg
;
1005 struct ceph_mds_cap_release
*head
;
1006 struct ceph_mds_cap_item
*item
;
1008 spin_lock(&session
->s_cap_lock
);
1009 BUG_ON(!session
->s_num_cap_releases
);
1010 msg
= list_first_entry(&session
->s_cap_releases
,
1011 struct ceph_msg
, list_head
);
1013 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1014 ino
, session
->s_mds
, msg
, session
->s_num_cap_releases
);
1016 BUG_ON(msg
->front
.iov_len
+ sizeof(*item
) > PAGE_CACHE_SIZE
);
1017 head
= msg
->front
.iov_base
;
1018 le32_add_cpu(&head
->num
, 1);
1019 item
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1020 item
->ino
= cpu_to_le64(ino
);
1021 item
->cap_id
= cpu_to_le64(cap_id
);
1022 item
->migrate_seq
= cpu_to_le32(migrate_seq
);
1023 item
->seq
= cpu_to_le32(issue_seq
);
1025 session
->s_num_cap_releases
--;
1027 msg
->front
.iov_len
+= sizeof(*item
);
1028 if (le32_to_cpu(head
->num
) == CEPH_CAPS_PER_RELEASE
) {
1029 dout(" release msg %p full\n", msg
);
1030 list_move_tail(&msg
->list_head
, &session
->s_cap_releases_done
);
1032 dout(" release msg %p at %d/%d (%d)\n", msg
,
1033 (int)le32_to_cpu(head
->num
),
1034 (int)CEPH_CAPS_PER_RELEASE
,
1035 (int)msg
->front
.iov_len
);
1037 spin_unlock(&session
->s_cap_lock
);
1041 * Queue cap releases when an inode is dropped from our cache. Since
1042 * inode is about to be destroyed, there is no need for i_ceph_lock.
1044 void ceph_queue_caps_release(struct inode
*inode
)
1046 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1049 p
= rb_first(&ci
->i_caps
);
1051 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1052 struct ceph_mds_session
*session
= cap
->session
;
1054 __queue_cap_release(session
, ceph_ino(inode
), cap
->cap_id
,
1055 cap
->mseq
, cap
->issue_seq
);
1057 __ceph_remove_cap(cap
);
1062 * Send a cap msg on the given inode. Update our caps state, then
1063 * drop i_ceph_lock and send the message.
1065 * Make note of max_size reported/requested from mds, revoked caps
1066 * that have now been implemented.
1068 * Make half-hearted attempt ot to invalidate page cache if we are
1069 * dropping RDCACHE. Note that this will leave behind locked pages
1070 * that we'll then need to deal with elsewhere.
1072 * Return non-zero if delayed release, or we experienced an error
1073 * such that the caller should requeue + retry later.
1075 * called with i_ceph_lock, then drops it.
1076 * caller should hold snap_rwsem (read), s_mutex.
1078 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1079 int op
, int used
, int want
, int retain
, int flushing
,
1080 unsigned *pflush_tid
)
1081 __releases(cap
->ci
->i_ceph_lock
)
1083 struct ceph_inode_info
*ci
= cap
->ci
;
1084 struct inode
*inode
= &ci
->vfs_inode
;
1085 u64 cap_id
= cap
->cap_id
;
1086 int held
, revoking
, dropping
, keep
;
1087 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1089 struct timespec mtime
, atime
;
1094 struct ceph_mds_session
*session
;
1095 u64 xattr_version
= 0;
1096 struct ceph_buffer
*xattr_blob
= NULL
;
1102 held
= cap
->issued
| cap
->implemented
;
1103 revoking
= cap
->implemented
& ~cap
->issued
;
1104 retain
&= ~revoking
;
1105 dropping
= cap
->issued
& ~retain
;
1107 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1108 inode
, cap
, cap
->session
,
1109 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1110 ceph_cap_string(revoking
));
1111 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1113 session
= cap
->session
;
1115 /* don't release wanted unless we've waited a bit. */
1116 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1117 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1118 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1119 ceph_cap_string(cap
->issued
),
1120 ceph_cap_string(cap
->issued
& retain
),
1121 ceph_cap_string(cap
->mds_wanted
),
1122 ceph_cap_string(want
));
1123 want
|= cap
->mds_wanted
;
1124 retain
|= cap
->issued
;
1127 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1129 cap
->issued
&= retain
; /* drop bits we don't want */
1130 if (cap
->implemented
& ~cap
->issued
) {
1132 * Wake up any waiters on wanted -> needed transition.
1133 * This is due to the weird transition from buffered
1134 * to sync IO... we need to flush dirty pages _before_
1135 * allowing sync writes to avoid reordering.
1139 cap
->implemented
&= cap
->issued
| used
;
1140 cap
->mds_wanted
= want
;
1144 * assign a tid for flush operations so we can avoid
1145 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1146 * clean type races. track latest tid for every bit
1147 * so we can handle flush AxFw, flush Fw, and have the
1148 * first ack clean Ax.
1150 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1152 *pflush_tid
= flush_tid
;
1153 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1154 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1155 if (flushing
& (1 << i
))
1156 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1158 follows
= ci
->i_head_snapc
->seq
;
1163 keep
= cap
->implemented
;
1165 issue_seq
= cap
->issue_seq
;
1167 size
= inode
->i_size
;
1168 ci
->i_reported_size
= size
;
1169 max_size
= ci
->i_wanted_max_size
;
1170 ci
->i_requested_max_size
= max_size
;
1171 mtime
= inode
->i_mtime
;
1172 atime
= inode
->i_atime
;
1173 time_warp_seq
= ci
->i_time_warp_seq
;
1176 mode
= inode
->i_mode
;
1178 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1179 __ceph_build_xattrs_blob(ci
);
1180 xattr_blob
= ci
->i_xattrs
.blob
;
1181 xattr_version
= ci
->i_xattrs
.version
;
1184 spin_unlock(&ci
->i_ceph_lock
);
1186 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1187 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1188 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1189 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1192 dout("error sending cap msg, must requeue %p\n", inode
);
1197 wake_up_all(&ci
->i_cap_wq
);
1203 * When a snapshot is taken, clients accumulate dirty metadata on
1204 * inodes with capabilities in ceph_cap_snaps to describe the file
1205 * state at the time the snapshot was taken. This must be flushed
1206 * asynchronously back to the MDS once sync writes complete and dirty
1207 * data is written out.
1209 * Unless @again is true, skip cap_snaps that were already sent to
1210 * the MDS (i.e., during this session).
1212 * Called under i_ceph_lock. Takes s_mutex as needed.
1214 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1215 struct ceph_mds_session
**psession
,
1217 __releases(ci
->i_ceph_lock
)
1218 __acquires(ci
->i_ceph_lock
)
1220 struct inode
*inode
= &ci
->vfs_inode
;
1222 struct ceph_cap_snap
*capsnap
;
1224 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1225 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1227 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1228 i_cap_snaps list, and skip these entries next time
1229 around to avoid an infinite loop */
1232 session
= *psession
;
1234 dout("__flush_snaps %p\n", inode
);
1236 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1237 /* avoid an infiniute loop after retry */
1238 if (capsnap
->follows
< next_follows
)
1241 * we need to wait for sync writes to complete and for dirty
1242 * pages to be written out.
1244 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1248 * if cap writeback already occurred, we should have dropped
1249 * the capsnap in ceph_put_wrbuffer_cap_refs.
1251 BUG_ON(capsnap
->dirty
== 0);
1253 /* pick mds, take s_mutex */
1254 if (ci
->i_auth_cap
== NULL
) {
1255 dout("no auth cap (migrating?), doing nothing\n");
1259 /* only flush each capsnap once */
1260 if (!again
&& !list_empty(&capsnap
->flushing_item
)) {
1261 dout("already flushed %p, skipping\n", capsnap
);
1265 mds
= ci
->i_auth_cap
->session
->s_mds
;
1266 mseq
= ci
->i_auth_cap
->mseq
;
1268 if (session
&& session
->s_mds
!= mds
) {
1269 dout("oops, wrong session %p mutex\n", session
);
1270 mutex_unlock(&session
->s_mutex
);
1271 ceph_put_mds_session(session
);
1275 spin_unlock(&ci
->i_ceph_lock
);
1276 mutex_lock(&mdsc
->mutex
);
1277 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1278 mutex_unlock(&mdsc
->mutex
);
1280 dout("inverting session/ino locks on %p\n",
1282 mutex_lock(&session
->s_mutex
);
1285 * if session == NULL, we raced against a cap
1286 * deletion or migration. retry, and we'll
1287 * get a better @mds value next time.
1289 spin_lock(&ci
->i_ceph_lock
);
1293 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1294 atomic_inc(&capsnap
->nref
);
1295 if (!list_empty(&capsnap
->flushing_item
))
1296 list_del_init(&capsnap
->flushing_item
);
1297 list_add_tail(&capsnap
->flushing_item
,
1298 &session
->s_cap_snaps_flushing
);
1299 spin_unlock(&ci
->i_ceph_lock
);
1301 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1302 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1303 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1304 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1305 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1307 &capsnap
->mtime
, &capsnap
->atime
,
1308 capsnap
->time_warp_seq
,
1309 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1310 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1313 next_follows
= capsnap
->follows
+ 1;
1314 ceph_put_cap_snap(capsnap
);
1316 spin_lock(&ci
->i_ceph_lock
);
1320 /* we flushed them all; remove this inode from the queue */
1321 spin_lock(&mdsc
->snap_flush_lock
);
1322 list_del_init(&ci
->i_snap_flush_item
);
1323 spin_unlock(&mdsc
->snap_flush_lock
);
1327 *psession
= session
;
1329 mutex_unlock(&session
->s_mutex
);
1330 ceph_put_mds_session(session
);
1334 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1336 spin_lock(&ci
->i_ceph_lock
);
1337 __ceph_flush_snaps(ci
, NULL
, 0);
1338 spin_unlock(&ci
->i_ceph_lock
);
1342 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1343 * Caller is then responsible for calling __mark_inode_dirty with the
1344 * returned flags value.
1346 int __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1348 struct ceph_mds_client
*mdsc
=
1349 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1350 struct inode
*inode
= &ci
->vfs_inode
;
1351 int was
= ci
->i_dirty_caps
;
1354 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1355 ceph_cap_string(mask
), ceph_cap_string(was
),
1356 ceph_cap_string(was
| mask
));
1357 ci
->i_dirty_caps
|= mask
;
1359 if (!ci
->i_head_snapc
)
1360 ci
->i_head_snapc
= ceph_get_snap_context(
1361 ci
->i_snap_realm
->cached_context
);
1362 dout(" inode %p now dirty snapc %p auth cap %p\n",
1363 &ci
->vfs_inode
, ci
->i_head_snapc
, ci
->i_auth_cap
);
1364 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1365 spin_lock(&mdsc
->cap_dirty_lock
);
1367 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1369 list_add(&ci
->i_dirty_item
,
1370 &mdsc
->cap_dirty_migrating
);
1371 spin_unlock(&mdsc
->cap_dirty_lock
);
1372 if (ci
->i_flushing_caps
== 0) {
1374 dirty
|= I_DIRTY_SYNC
;
1377 BUG_ON(list_empty(&ci
->i_dirty_item
));
1378 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1379 (mask
& CEPH_CAP_FILE_BUFFER
))
1380 dirty
|= I_DIRTY_DATASYNC
;
1381 __cap_delay_requeue(mdsc
, ci
);
1386 * Add dirty inode to the flushing list. Assigned a seq number so we
1387 * can wait for caps to flush without starving.
1389 * Called under i_ceph_lock.
1391 static int __mark_caps_flushing(struct inode
*inode
,
1392 struct ceph_mds_session
*session
)
1394 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1395 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1398 BUG_ON(ci
->i_dirty_caps
== 0);
1399 BUG_ON(list_empty(&ci
->i_dirty_item
));
1401 flushing
= ci
->i_dirty_caps
;
1402 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1403 ceph_cap_string(flushing
),
1404 ceph_cap_string(ci
->i_flushing_caps
),
1405 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1406 ci
->i_flushing_caps
|= flushing
;
1407 ci
->i_dirty_caps
= 0;
1408 dout(" inode %p now !dirty\n", inode
);
1410 spin_lock(&mdsc
->cap_dirty_lock
);
1411 list_del_init(&ci
->i_dirty_item
);
1413 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1414 if (list_empty(&ci
->i_flushing_item
)) {
1415 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1416 mdsc
->num_cap_flushing
++;
1417 dout(" inode %p now flushing seq %lld\n", inode
,
1418 ci
->i_cap_flush_seq
);
1420 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1421 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1422 ci
->i_cap_flush_seq
);
1424 spin_unlock(&mdsc
->cap_dirty_lock
);
1430 * try to invalidate mapping pages without blocking.
1432 static int try_nonblocking_invalidate(struct inode
*inode
)
1434 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1435 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1437 spin_unlock(&ci
->i_ceph_lock
);
1438 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1439 spin_lock(&ci
->i_ceph_lock
);
1441 if (inode
->i_data
.nrpages
== 0 &&
1442 invalidating_gen
== ci
->i_rdcache_gen
) {
1444 dout("try_nonblocking_invalidate %p success\n", inode
);
1445 /* save any racing async invalidate some trouble */
1446 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
- 1;
1449 dout("try_nonblocking_invalidate %p failed\n", inode
);
1454 * Swiss army knife function to examine currently used and wanted
1455 * versus held caps. Release, flush, ack revoked caps to mds as
1458 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1459 * cap release further.
1460 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1461 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1464 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1465 struct ceph_mds_session
*session
)
1467 struct ceph_fs_client
*fsc
= ceph_inode_to_client(&ci
->vfs_inode
);
1468 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1469 struct inode
*inode
= &ci
->vfs_inode
;
1470 struct ceph_cap
*cap
;
1471 int file_wanted
, used
, cap_used
;
1472 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1473 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1474 int mds
= -1; /* keep track of how far we've gone through i_caps list
1475 to avoid an infinite loop on retry */
1477 int tried_invalidate
= 0;
1478 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1479 int queue_invalidate
= 0;
1480 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1482 /* if we are unmounting, flush any unused caps immediately. */
1486 spin_lock(&ci
->i_ceph_lock
);
1488 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1489 flags
|= CHECK_CAPS_FLUSH
;
1491 /* flush snaps first time around only */
1492 if (!list_empty(&ci
->i_cap_snaps
))
1493 __ceph_flush_snaps(ci
, &session
, 0);
1496 spin_lock(&ci
->i_ceph_lock
);
1498 file_wanted
= __ceph_caps_file_wanted(ci
);
1499 used
= __ceph_caps_used(ci
);
1500 want
= file_wanted
| used
;
1501 issued
= __ceph_caps_issued(ci
, &implemented
);
1502 revoking
= implemented
& ~issued
;
1504 retain
= want
| CEPH_CAP_PIN
;
1505 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1507 retain
|= CEPH_CAP_ANY
; /* be greedy */
1509 retain
|= CEPH_CAP_ANY_SHARED
;
1511 * keep RD only if we didn't have the file open RW,
1512 * because then the mds would revoke it anyway to
1513 * journal max_size=0.
1515 if (ci
->i_max_size
== 0)
1516 retain
|= CEPH_CAP_ANY_RD
;
1520 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1521 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1522 ceph_cap_string(file_wanted
),
1523 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1524 ceph_cap_string(ci
->i_flushing_caps
),
1525 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1526 ceph_cap_string(retain
),
1527 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1528 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1529 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1532 * If we no longer need to hold onto old our caps, and we may
1533 * have cached pages, but don't want them, then try to invalidate.
1534 * If we fail, it's because pages are locked.... try again later.
1536 if ((!is_delayed
|| mdsc
->stopping
) &&
1537 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1538 inode
->i_data
.nrpages
&& /* have cached pages */
1539 (file_wanted
== 0 || /* no open files */
1540 (revoking
& (CEPH_CAP_FILE_CACHE
|
1541 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1542 !tried_invalidate
) {
1543 dout("check_caps trying to invalidate on %p\n", inode
);
1544 if (try_nonblocking_invalidate(inode
) < 0) {
1545 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1546 CEPH_CAP_FILE_LAZYIO
)) {
1547 dout("check_caps queuing invalidate\n");
1548 queue_invalidate
= 1;
1549 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1551 dout("check_caps failed to invalidate pages\n");
1552 /* we failed to invalidate pages. check these
1553 caps again later. */
1555 __cap_set_timeouts(mdsc
, ci
);
1558 tried_invalidate
= 1;
1563 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1564 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1567 /* avoid looping forever */
1568 if (mds
>= cap
->mds
||
1569 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1572 /* NOTE: no side-effects allowed, until we take s_mutex */
1575 if (ci
->i_auth_cap
&& cap
!= ci
->i_auth_cap
)
1576 cap_used
&= ~ci
->i_auth_cap
->issued
;
1578 revoking
= cap
->implemented
& ~cap
->issued
;
1579 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1580 cap
->mds
, cap
, ceph_cap_string(cap
->issued
),
1581 ceph_cap_string(cap_used
),
1582 ceph_cap_string(cap
->implemented
),
1583 ceph_cap_string(revoking
));
1585 if (cap
== ci
->i_auth_cap
&&
1586 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1587 /* request larger max_size from MDS? */
1588 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1589 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1590 dout("requesting new max_size\n");
1594 /* approaching file_max? */
1595 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1596 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1597 dout("i_size approaching max_size\n");
1601 /* flush anything dirty? */
1602 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1604 dout("flushing dirty caps\n");
1608 /* completed revocation? going down and there are no caps? */
1609 if (revoking
&& (revoking
& cap_used
) == 0) {
1610 dout("completed revocation of %s\n",
1611 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1615 /* want more caps from mds? */
1616 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1619 /* things we might delay */
1620 if ((cap
->issued
& ~retain
) == 0 &&
1621 cap
->mds_wanted
== want
)
1622 continue; /* nope, all good */
1628 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1629 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1630 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1631 ceph_cap_string(cap
->issued
),
1632 ceph_cap_string(cap
->issued
& retain
),
1633 ceph_cap_string(cap
->mds_wanted
),
1634 ceph_cap_string(want
));
1640 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1641 dout(" skipping %p I_NOFLUSH set\n", inode
);
1645 if (session
&& session
!= cap
->session
) {
1646 dout("oops, wrong session %p mutex\n", session
);
1647 mutex_unlock(&session
->s_mutex
);
1651 session
= cap
->session
;
1652 if (mutex_trylock(&session
->s_mutex
) == 0) {
1653 dout("inverting session/ino locks on %p\n",
1655 spin_unlock(&ci
->i_ceph_lock
);
1656 if (took_snap_rwsem
) {
1657 up_read(&mdsc
->snap_rwsem
);
1658 took_snap_rwsem
= 0;
1660 mutex_lock(&session
->s_mutex
);
1664 /* take snap_rwsem after session mutex */
1665 if (!took_snap_rwsem
) {
1666 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1667 dout("inverting snap/in locks on %p\n",
1669 spin_unlock(&ci
->i_ceph_lock
);
1670 down_read(&mdsc
->snap_rwsem
);
1671 took_snap_rwsem
= 1;
1674 took_snap_rwsem
= 1;
1677 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1678 flushing
= __mark_caps_flushing(inode
, session
);
1682 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1685 /* __send_cap drops i_ceph_lock */
1686 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, cap_used
,
1687 want
, retain
, flushing
, NULL
);
1688 goto retry
; /* retake i_ceph_lock and restart our cap scan. */
1692 * Reschedule delayed caps release if we delayed anything,
1695 if (delayed
&& is_delayed
)
1696 force_requeue
= 1; /* __send_cap delayed release; requeue */
1697 if (!delayed
&& !is_delayed
)
1698 __cap_delay_cancel(mdsc
, ci
);
1699 else if (!is_delayed
|| force_requeue
)
1700 __cap_delay_requeue(mdsc
, ci
);
1702 spin_unlock(&ci
->i_ceph_lock
);
1704 if (queue_invalidate
)
1705 ceph_queue_invalidate(inode
);
1708 mutex_unlock(&session
->s_mutex
);
1709 if (took_snap_rwsem
)
1710 up_read(&mdsc
->snap_rwsem
);
1714 * Try to flush dirty caps back to the auth mds.
1716 static int try_flush_caps(struct inode
*inode
, struct ceph_mds_session
*session
,
1717 unsigned *flush_tid
)
1719 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1720 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1721 int unlock_session
= session
? 0 : 1;
1725 spin_lock(&ci
->i_ceph_lock
);
1726 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1727 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1730 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1731 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1732 int used
= __ceph_caps_used(ci
);
1733 int want
= __ceph_caps_wanted(ci
);
1737 spin_unlock(&ci
->i_ceph_lock
);
1738 session
= cap
->session
;
1739 mutex_lock(&session
->s_mutex
);
1742 BUG_ON(session
!= cap
->session
);
1743 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1746 flushing
= __mark_caps_flushing(inode
, session
);
1748 /* __send_cap drops i_ceph_lock */
1749 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1750 cap
->issued
| cap
->implemented
, flushing
,
1755 spin_lock(&ci
->i_ceph_lock
);
1756 __cap_delay_requeue(mdsc
, ci
);
1759 spin_unlock(&ci
->i_ceph_lock
);
1761 if (session
&& unlock_session
)
1762 mutex_unlock(&session
->s_mutex
);
1767 * Return true if we've flushed caps through the given flush_tid.
1769 static int caps_are_flushed(struct inode
*inode
, unsigned tid
)
1771 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1774 spin_lock(&ci
->i_ceph_lock
);
1775 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1776 if ((ci
->i_flushing_caps
& (1 << i
)) &&
1777 ci
->i_cap_flush_tid
[i
] <= tid
) {
1778 /* still flushing this bit */
1782 spin_unlock(&ci
->i_ceph_lock
);
1787 * Wait on any unsafe replies for the given inode. First wait on the
1788 * newest request, and make that the upper bound. Then, if there are
1789 * more requests, keep waiting on the oldest as long as it is still older
1790 * than the original request.
1792 static void sync_write_wait(struct inode
*inode
)
1794 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1795 struct list_head
*head
= &ci
->i_unsafe_writes
;
1796 struct ceph_osd_request
*req
;
1799 spin_lock(&ci
->i_unsafe_lock
);
1800 if (list_empty(head
))
1803 /* set upper bound as _last_ entry in chain */
1804 req
= list_entry(head
->prev
, struct ceph_osd_request
,
1806 last_tid
= req
->r_tid
;
1809 ceph_osdc_get_request(req
);
1810 spin_unlock(&ci
->i_unsafe_lock
);
1811 dout("sync_write_wait on tid %llu (until %llu)\n",
1812 req
->r_tid
, last_tid
);
1813 wait_for_completion(&req
->r_safe_completion
);
1814 spin_lock(&ci
->i_unsafe_lock
);
1815 ceph_osdc_put_request(req
);
1818 * from here on look at first entry in chain, since we
1819 * only want to wait for anything older than last_tid
1821 if (list_empty(head
))
1823 req
= list_entry(head
->next
, struct ceph_osd_request
,
1825 } while (req
->r_tid
< last_tid
);
1827 spin_unlock(&ci
->i_unsafe_lock
);
1830 int ceph_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1832 struct inode
*inode
= file
->f_mapping
->host
;
1833 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1838 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1839 sync_write_wait(inode
);
1841 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1844 mutex_lock(&inode
->i_mutex
);
1846 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1847 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1850 * only wait on non-file metadata writeback (the mds
1851 * can recover size and mtime, so we don't need to
1854 if (!datasync
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1855 dout("fsync waiting for flush_tid %u\n", flush_tid
);
1856 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1857 caps_are_flushed(inode
, flush_tid
));
1860 dout("fsync %p%s done\n", inode
, datasync
? " datasync" : "");
1861 mutex_unlock(&inode
->i_mutex
);
1866 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1867 * queue inode for flush but don't do so immediately, because we can
1868 * get by with fewer MDS messages if we wait for data writeback to
1871 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1873 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1877 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1879 dout("write_inode %p wait=%d\n", inode
, wait
);
1881 dirty
= try_flush_caps(inode
, NULL
, &flush_tid
);
1883 err
= wait_event_interruptible(ci
->i_cap_wq
,
1884 caps_are_flushed(inode
, flush_tid
));
1886 struct ceph_mds_client
*mdsc
=
1887 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1889 spin_lock(&ci
->i_ceph_lock
);
1890 if (__ceph_caps_dirty(ci
))
1891 __cap_delay_requeue_front(mdsc
, ci
);
1892 spin_unlock(&ci
->i_ceph_lock
);
1898 * After a recovering MDS goes active, we need to resend any caps
1901 * Caller holds session->s_mutex.
1903 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
1904 struct ceph_mds_session
*session
)
1906 struct ceph_cap_snap
*capsnap
;
1908 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
1909 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
1911 struct ceph_inode_info
*ci
= capsnap
->ci
;
1912 struct inode
*inode
= &ci
->vfs_inode
;
1913 struct ceph_cap
*cap
;
1915 spin_lock(&ci
->i_ceph_lock
);
1916 cap
= ci
->i_auth_cap
;
1917 if (cap
&& cap
->session
== session
) {
1918 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
1920 __ceph_flush_snaps(ci
, &session
, 1);
1922 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1923 cap
, session
->s_mds
);
1925 spin_unlock(&ci
->i_ceph_lock
);
1929 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
1930 struct ceph_mds_session
*session
)
1932 struct ceph_inode_info
*ci
;
1934 kick_flushing_capsnaps(mdsc
, session
);
1936 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
1937 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
1938 struct inode
*inode
= &ci
->vfs_inode
;
1939 struct ceph_cap
*cap
;
1942 spin_lock(&ci
->i_ceph_lock
);
1943 cap
= ci
->i_auth_cap
;
1944 if (cap
&& cap
->session
== session
) {
1945 dout("kick_flushing_caps %p cap %p %s\n", inode
,
1946 cap
, ceph_cap_string(ci
->i_flushing_caps
));
1947 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1948 __ceph_caps_used(ci
),
1949 __ceph_caps_wanted(ci
),
1950 cap
->issued
| cap
->implemented
,
1951 ci
->i_flushing_caps
, NULL
);
1953 spin_lock(&ci
->i_ceph_lock
);
1954 __cap_delay_requeue(mdsc
, ci
);
1955 spin_unlock(&ci
->i_ceph_lock
);
1958 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1959 cap
, session
->s_mds
);
1960 spin_unlock(&ci
->i_ceph_lock
);
1965 static void kick_flushing_inode_caps(struct ceph_mds_client
*mdsc
,
1966 struct ceph_mds_session
*session
,
1967 struct inode
*inode
)
1969 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1970 struct ceph_cap
*cap
;
1973 spin_lock(&ci
->i_ceph_lock
);
1974 cap
= ci
->i_auth_cap
;
1975 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode
,
1976 ceph_cap_string(ci
->i_flushing_caps
), ci
->i_cap_flush_seq
);
1977 __ceph_flush_snaps(ci
, &session
, 1);
1978 if (ci
->i_flushing_caps
) {
1979 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1980 __ceph_caps_used(ci
),
1981 __ceph_caps_wanted(ci
),
1982 cap
->issued
| cap
->implemented
,
1983 ci
->i_flushing_caps
, NULL
);
1985 spin_lock(&ci
->i_ceph_lock
);
1986 __cap_delay_requeue(mdsc
, ci
);
1987 spin_unlock(&ci
->i_ceph_lock
);
1990 spin_unlock(&ci
->i_ceph_lock
);
1996 * Take references to capabilities we hold, so that we don't release
1997 * them to the MDS prematurely.
1999 * Protected by i_ceph_lock.
2001 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
)
2003 if (got
& CEPH_CAP_PIN
)
2005 if (got
& CEPH_CAP_FILE_RD
)
2007 if (got
& CEPH_CAP_FILE_CACHE
)
2008 ci
->i_rdcache_ref
++;
2009 if (got
& CEPH_CAP_FILE_WR
)
2011 if (got
& CEPH_CAP_FILE_BUFFER
) {
2012 if (ci
->i_wb_ref
== 0)
2013 ihold(&ci
->vfs_inode
);
2015 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2016 &ci
->vfs_inode
, ci
->i_wb_ref
-1, ci
->i_wb_ref
);
2021 * Try to grab cap references. Specify those refs we @want, and the
2022 * minimal set we @need. Also include the larger offset we are writing
2023 * to (when applicable), and check against max_size here as well.
2024 * Note that caller is responsible for ensuring max_size increases are
2025 * requested from the MDS.
2027 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
2028 int *got
, loff_t endoff
, int *check_max
, int *err
)
2030 struct inode
*inode
= &ci
->vfs_inode
;
2032 int have
, implemented
;
2035 dout("get_cap_refs %p need %s want %s\n", inode
,
2036 ceph_cap_string(need
), ceph_cap_string(want
));
2037 spin_lock(&ci
->i_ceph_lock
);
2039 /* make sure file is actually open */
2040 file_wanted
= __ceph_caps_file_wanted(ci
);
2041 if ((file_wanted
& need
) == 0) {
2042 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2043 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2049 if (need
& CEPH_CAP_FILE_WR
) {
2050 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2051 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2052 inode
, endoff
, ci
->i_max_size
);
2053 if (endoff
> ci
->i_wanted_max_size
) {
2060 * If a sync write is in progress, we must wait, so that we
2061 * can get a final snapshot value for size+mtime.
2063 if (__ceph_have_pending_cap_snap(ci
)) {
2064 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2068 have
= __ceph_caps_issued(ci
, &implemented
);
2071 * disallow writes while a truncate is pending
2073 if (ci
->i_truncate_pending
)
2074 have
&= ~CEPH_CAP_FILE_WR
;
2076 if ((have
& need
) == need
) {
2078 * Look at (implemented & ~have & not) so that we keep waiting
2079 * on transition from wanted -> needed caps. This is needed
2080 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2081 * going before a prior buffered writeback happens.
2083 int not = want
& ~(have
& need
);
2084 int revoking
= implemented
& ~have
;
2085 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2086 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2087 ceph_cap_string(revoking
));
2088 if ((revoking
& not) == 0) {
2089 *got
= need
| (have
& want
);
2090 __take_cap_refs(ci
, *got
);
2094 dout("get_cap_refs %p have %s needed %s\n", inode
,
2095 ceph_cap_string(have
), ceph_cap_string(need
));
2098 spin_unlock(&ci
->i_ceph_lock
);
2099 dout("get_cap_refs %p ret %d got %s\n", inode
,
2100 ret
, ceph_cap_string(*got
));
2105 * Check the offset we are writing up to against our current
2106 * max_size. If necessary, tell the MDS we want to write to
2109 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2111 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2114 /* do we need to explicitly request a larger max_size? */
2115 spin_lock(&ci
->i_ceph_lock
);
2116 if ((endoff
>= ci
->i_max_size
||
2117 endoff
> (inode
->i_size
<< 1)) &&
2118 endoff
> ci
->i_wanted_max_size
) {
2119 dout("write %p at large endoff %llu, req max_size\n",
2121 ci
->i_wanted_max_size
= endoff
;
2124 spin_unlock(&ci
->i_ceph_lock
);
2126 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2130 * Wait for caps, and take cap references. If we can't get a WR cap
2131 * due to a small max_size, make sure we check_max_size (and possibly
2132 * ask the mds) so we don't get hung up indefinitely.
2134 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
, int *got
,
2137 int check_max
, ret
, err
;
2141 check_max_size(&ci
->vfs_inode
, endoff
);
2144 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2145 try_get_cap_refs(ci
, need
, want
,
2156 * Take cap refs. Caller must already know we hold at least one ref
2157 * on the caps in question or we don't know this is safe.
2159 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2161 spin_lock(&ci
->i_ceph_lock
);
2162 __take_cap_refs(ci
, caps
);
2163 spin_unlock(&ci
->i_ceph_lock
);
2169 * If we released the last ref on any given cap, call ceph_check_caps
2170 * to release (or schedule a release).
2172 * If we are releasing a WR cap (from a sync write), finalize any affected
2173 * cap_snap, and wake up any waiters.
2175 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2177 struct inode
*inode
= &ci
->vfs_inode
;
2178 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2179 struct ceph_cap_snap
*capsnap
;
2181 spin_lock(&ci
->i_ceph_lock
);
2182 if (had
& CEPH_CAP_PIN
)
2184 if (had
& CEPH_CAP_FILE_RD
)
2185 if (--ci
->i_rd_ref
== 0)
2187 if (had
& CEPH_CAP_FILE_CACHE
)
2188 if (--ci
->i_rdcache_ref
== 0)
2190 if (had
& CEPH_CAP_FILE_BUFFER
) {
2191 if (--ci
->i_wb_ref
== 0) {
2195 dout("put_cap_refs %p wb %d -> %d (?)\n",
2196 inode
, ci
->i_wb_ref
+1, ci
->i_wb_ref
);
2198 if (had
& CEPH_CAP_FILE_WR
)
2199 if (--ci
->i_wr_ref
== 0) {
2201 if (!list_empty(&ci
->i_cap_snaps
)) {
2202 capsnap
= list_first_entry(&ci
->i_cap_snaps
,
2203 struct ceph_cap_snap
,
2205 if (capsnap
->writing
) {
2206 capsnap
->writing
= 0;
2208 __ceph_finish_cap_snap(ci
,
2214 spin_unlock(&ci
->i_ceph_lock
);
2216 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2217 last
? " last" : "", put
? " put" : "");
2219 if (last
&& !flushsnaps
)
2220 ceph_check_caps(ci
, 0, NULL
);
2221 else if (flushsnaps
)
2222 ceph_flush_snaps(ci
);
2224 wake_up_all(&ci
->i_cap_wq
);
2230 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2231 * context. Adjust per-snap dirty page accounting as appropriate.
2232 * Once all dirty data for a cap_snap is flushed, flush snapped file
2233 * metadata back to the MDS. If we dropped the last ref, call
2236 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2237 struct ceph_snap_context
*snapc
)
2239 struct inode
*inode
= &ci
->vfs_inode
;
2241 int complete_capsnap
= 0;
2242 int drop_capsnap
= 0;
2244 struct ceph_cap_snap
*capsnap
= NULL
;
2246 spin_lock(&ci
->i_ceph_lock
);
2247 ci
->i_wrbuffer_ref
-= nr
;
2248 last
= !ci
->i_wrbuffer_ref
;
2250 if (ci
->i_head_snapc
== snapc
) {
2251 ci
->i_wrbuffer_ref_head
-= nr
;
2252 if (ci
->i_wrbuffer_ref_head
== 0 &&
2253 ci
->i_dirty_caps
== 0 && ci
->i_flushing_caps
== 0) {
2254 BUG_ON(!ci
->i_head_snapc
);
2255 ceph_put_snap_context(ci
->i_head_snapc
);
2256 ci
->i_head_snapc
= NULL
;
2258 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2260 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2261 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2262 last
? " LAST" : "");
2264 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2265 if (capsnap
->context
== snapc
) {
2271 capsnap
->dirty_pages
-= nr
;
2272 if (capsnap
->dirty_pages
== 0) {
2273 complete_capsnap
= 1;
2274 if (capsnap
->dirty
== 0)
2275 /* cap writeback completed before we created
2276 * the cap_snap; no FLUSHSNAP is needed */
2279 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2280 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2281 inode
, capsnap
, capsnap
->context
->seq
,
2282 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2283 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2284 last
? " (wrbuffer last)" : "",
2285 complete_capsnap
? " (complete capsnap)" : "",
2286 drop_capsnap
? " (drop capsnap)" : "");
2288 ceph_put_snap_context(capsnap
->context
);
2289 list_del(&capsnap
->ci_item
);
2290 list_del(&capsnap
->flushing_item
);
2291 ceph_put_cap_snap(capsnap
);
2295 spin_unlock(&ci
->i_ceph_lock
);
2298 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2300 } else if (complete_capsnap
) {
2301 ceph_flush_snaps(ci
);
2302 wake_up_all(&ci
->i_cap_wq
);
2309 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2310 * actually be a revocation if it specifies a smaller cap set.)
2312 * caller holds s_mutex and i_ceph_lock, we drop both.
2316 * 1 - check_caps on auth cap only (writeback)
2317 * 2 - check_caps (ack revoke)
2319 static void handle_cap_grant(struct inode
*inode
, struct ceph_mds_caps
*grant
,
2320 struct ceph_mds_session
*session
,
2321 struct ceph_cap
*cap
,
2322 struct ceph_buffer
*xattr_buf
)
2323 __releases(ci
->i_ceph_lock
)
2325 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2326 int mds
= session
->s_mds
;
2327 int seq
= le32_to_cpu(grant
->seq
);
2328 int newcaps
= le32_to_cpu(grant
->caps
);
2329 int issued
, implemented
, used
, wanted
, dirty
;
2330 u64 size
= le64_to_cpu(grant
->size
);
2331 u64 max_size
= le64_to_cpu(grant
->max_size
);
2332 struct timespec mtime
, atime
, ctime
;
2336 int revoked_rdcache
= 0;
2337 int queue_invalidate
= 0;
2339 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2340 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2341 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2345 * If CACHE is being revoked, and we have no dirty buffers,
2346 * try to invalidate (once). (If there are dirty buffers, we
2347 * will invalidate _after_ writeback.)
2349 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2350 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2351 !ci
->i_wrbuffer_ref
) {
2352 if (try_nonblocking_invalidate(inode
) == 0) {
2353 revoked_rdcache
= 1;
2355 /* there were locked pages.. invalidate later
2356 in a separate thread. */
2357 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2358 queue_invalidate
= 1;
2359 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2364 /* side effects now are allowed */
2366 issued
= __ceph_caps_issued(ci
, &implemented
);
2367 issued
|= implemented
| __ceph_caps_dirty(ci
);
2369 cap
->cap_gen
= session
->s_cap_gen
;
2371 __check_cap_issue(ci
, cap
, newcaps
);
2373 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2374 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2375 inode
->i_uid
= le32_to_cpu(grant
->uid
);
2376 inode
->i_gid
= le32_to_cpu(grant
->gid
);
2377 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2378 inode
->i_uid
, inode
->i_gid
);
2381 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0)
2382 set_nlink(inode
, le32_to_cpu(grant
->nlink
));
2384 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2385 int len
= le32_to_cpu(grant
->xattr_len
);
2386 u64 version
= le64_to_cpu(grant
->xattr_version
);
2388 if (version
> ci
->i_xattrs
.version
) {
2389 dout(" got new xattrs v%llu on %p len %d\n",
2390 version
, inode
, len
);
2391 if (ci
->i_xattrs
.blob
)
2392 ceph_buffer_put(ci
->i_xattrs
.blob
);
2393 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2394 ci
->i_xattrs
.version
= version
;
2398 /* size/ctime/mtime/atime? */
2399 ceph_fill_file_size(inode
, issued
,
2400 le32_to_cpu(grant
->truncate_seq
),
2401 le64_to_cpu(grant
->truncate_size
), size
);
2402 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2403 ceph_decode_timespec(&atime
, &grant
->atime
);
2404 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2405 ceph_fill_file_time(inode
, issued
,
2406 le32_to_cpu(grant
->time_warp_seq
), &ctime
, &mtime
,
2409 /* max size increase? */
2410 if (ci
->i_auth_cap
== cap
&& max_size
!= ci
->i_max_size
) {
2411 dout("max_size %lld -> %llu\n", ci
->i_max_size
, max_size
);
2412 ci
->i_max_size
= max_size
;
2413 if (max_size
>= ci
->i_wanted_max_size
) {
2414 ci
->i_wanted_max_size
= 0; /* reset */
2415 ci
->i_requested_max_size
= 0;
2420 /* check cap bits */
2421 wanted
= __ceph_caps_wanted(ci
);
2422 used
= __ceph_caps_used(ci
);
2423 dirty
= __ceph_caps_dirty(ci
);
2424 dout(" my wanted = %s, used = %s, dirty %s\n",
2425 ceph_cap_string(wanted
),
2426 ceph_cap_string(used
),
2427 ceph_cap_string(dirty
));
2428 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2429 dout("mds wanted %s -> %s\n",
2430 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2431 ceph_cap_string(wanted
));
2432 grant
->wanted
= cpu_to_le32(wanted
);
2437 /* file layout may have changed */
2438 ci
->i_layout
= grant
->layout
;
2440 /* revocation, grant, or no-op? */
2441 if (cap
->issued
& ~newcaps
) {
2442 int revoking
= cap
->issued
& ~newcaps
;
2444 dout("revocation: %s -> %s (revoking %s)\n",
2445 ceph_cap_string(cap
->issued
),
2446 ceph_cap_string(newcaps
),
2447 ceph_cap_string(revoking
));
2448 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2449 writeback
= 1; /* initiate writeback; will delay ack */
2450 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2451 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2453 ; /* do nothing yet, invalidation will be queued */
2454 else if (cap
== ci
->i_auth_cap
)
2455 check_caps
= 1; /* check auth cap only */
2457 check_caps
= 2; /* check all caps */
2458 cap
->issued
= newcaps
;
2459 cap
->implemented
|= newcaps
;
2460 } else if (cap
->issued
== newcaps
) {
2461 dout("caps unchanged: %s -> %s\n",
2462 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2464 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2465 ceph_cap_string(newcaps
));
2466 cap
->issued
= newcaps
;
2467 cap
->implemented
|= newcaps
; /* add bits only, to
2468 * avoid stepping on a
2469 * pending revocation */
2472 BUG_ON(cap
->issued
& ~cap
->implemented
);
2474 spin_unlock(&ci
->i_ceph_lock
);
2477 * queue inode for writeback: we can't actually call
2478 * filemap_write_and_wait, etc. from message handler
2481 ceph_queue_writeback(inode
);
2482 if (queue_invalidate
)
2483 ceph_queue_invalidate(inode
);
2485 wake_up_all(&ci
->i_cap_wq
);
2487 if (check_caps
== 1)
2488 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2490 else if (check_caps
== 2)
2491 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2493 mutex_unlock(&session
->s_mutex
);
2497 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2498 * MDS has been safely committed.
2500 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2501 struct ceph_mds_caps
*m
,
2502 struct ceph_mds_session
*session
,
2503 struct ceph_cap
*cap
)
2504 __releases(ci
->i_ceph_lock
)
2506 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2507 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2508 unsigned seq
= le32_to_cpu(m
->seq
);
2509 int dirty
= le32_to_cpu(m
->dirty
);
2514 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2515 if ((dirty
& (1 << i
)) &&
2516 flush_tid
== ci
->i_cap_flush_tid
[i
])
2519 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2520 " flushing %s -> %s\n",
2521 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2522 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2523 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2525 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2528 ci
->i_flushing_caps
&= ~cleaned
;
2530 spin_lock(&mdsc
->cap_dirty_lock
);
2531 if (ci
->i_flushing_caps
== 0) {
2532 list_del_init(&ci
->i_flushing_item
);
2533 if (!list_empty(&session
->s_cap_flushing
))
2534 dout(" mds%d still flushing cap on %p\n",
2536 &list_entry(session
->s_cap_flushing
.next
,
2537 struct ceph_inode_info
,
2538 i_flushing_item
)->vfs_inode
);
2539 mdsc
->num_cap_flushing
--;
2540 wake_up_all(&mdsc
->cap_flushing_wq
);
2541 dout(" inode %p now !flushing\n", inode
);
2543 if (ci
->i_dirty_caps
== 0) {
2544 dout(" inode %p now clean\n", inode
);
2545 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2547 if (ci
->i_wrbuffer_ref_head
== 0) {
2548 BUG_ON(!ci
->i_head_snapc
);
2549 ceph_put_snap_context(ci
->i_head_snapc
);
2550 ci
->i_head_snapc
= NULL
;
2553 BUG_ON(list_empty(&ci
->i_dirty_item
));
2556 spin_unlock(&mdsc
->cap_dirty_lock
);
2557 wake_up_all(&ci
->i_cap_wq
);
2560 spin_unlock(&ci
->i_ceph_lock
);
2566 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2567 * throw away our cap_snap.
2569 * Caller hold s_mutex.
2571 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2572 struct ceph_mds_caps
*m
,
2573 struct ceph_mds_session
*session
)
2575 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2576 u64 follows
= le64_to_cpu(m
->snap_follows
);
2577 struct ceph_cap_snap
*capsnap
;
2580 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2581 inode
, ci
, session
->s_mds
, follows
);
2583 spin_lock(&ci
->i_ceph_lock
);
2584 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2585 if (capsnap
->follows
== follows
) {
2586 if (capsnap
->flush_tid
!= flush_tid
) {
2587 dout(" cap_snap %p follows %lld tid %lld !="
2588 " %lld\n", capsnap
, follows
,
2589 flush_tid
, capsnap
->flush_tid
);
2592 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2593 dout(" removing %p cap_snap %p follows %lld\n",
2594 inode
, capsnap
, follows
);
2595 ceph_put_snap_context(capsnap
->context
);
2596 list_del(&capsnap
->ci_item
);
2597 list_del(&capsnap
->flushing_item
);
2598 ceph_put_cap_snap(capsnap
);
2602 dout(" skipping cap_snap %p follows %lld\n",
2603 capsnap
, capsnap
->follows
);
2606 spin_unlock(&ci
->i_ceph_lock
);
2612 * Handle TRUNC from MDS, indicating file truncation.
2614 * caller hold s_mutex.
2616 static void handle_cap_trunc(struct inode
*inode
,
2617 struct ceph_mds_caps
*trunc
,
2618 struct ceph_mds_session
*session
)
2619 __releases(ci
->i_ceph_lock
)
2621 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2622 int mds
= session
->s_mds
;
2623 int seq
= le32_to_cpu(trunc
->seq
);
2624 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2625 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2626 u64 size
= le64_to_cpu(trunc
->size
);
2627 int implemented
= 0;
2628 int dirty
= __ceph_caps_dirty(ci
);
2629 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2630 int queue_trunc
= 0;
2632 issued
|= implemented
| dirty
;
2634 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2635 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2636 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2637 truncate_seq
, truncate_size
, size
);
2638 spin_unlock(&ci
->i_ceph_lock
);
2641 ceph_queue_vmtruncate(inode
);
2645 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2646 * different one. If we are the most recent migration we've seen (as
2647 * indicated by mseq), make note of the migrating cap bits for the
2648 * duration (until we see the corresponding IMPORT).
2650 * caller holds s_mutex
2652 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2653 struct ceph_mds_session
*session
,
2654 int *open_target_sessions
)
2656 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2657 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2658 int mds
= session
->s_mds
;
2659 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2660 struct ceph_cap
*cap
= NULL
, *t
;
2664 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2665 inode
, ci
, mds
, mseq
);
2667 spin_lock(&ci
->i_ceph_lock
);
2669 /* make sure we haven't seen a higher mseq */
2670 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
2671 t
= rb_entry(p
, struct ceph_cap
, ci_node
);
2672 if (ceph_seq_cmp(t
->mseq
, mseq
) > 0) {
2673 dout(" higher mseq on cap from mds%d\n",
2677 if (t
->session
->s_mds
== mds
)
2684 ci
->i_cap_exporting_mds
= mds
;
2685 ci
->i_cap_exporting_mseq
= mseq
;
2686 ci
->i_cap_exporting_issued
= cap
->issued
;
2689 * make sure we have open sessions with all possible
2690 * export targets, so that we get the matching IMPORT
2692 *open_target_sessions
= 1;
2695 * we can't flush dirty caps that we've seen the
2696 * EXPORT but no IMPORT for
2698 spin_lock(&mdsc
->cap_dirty_lock
);
2699 if (!list_empty(&ci
->i_dirty_item
)) {
2700 dout(" moving %p to cap_dirty_migrating\n",
2702 list_move(&ci
->i_dirty_item
,
2703 &mdsc
->cap_dirty_migrating
);
2705 spin_unlock(&mdsc
->cap_dirty_lock
);
2707 __ceph_remove_cap(cap
);
2709 /* else, we already released it */
2711 spin_unlock(&ci
->i_ceph_lock
);
2715 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2718 * caller holds s_mutex.
2720 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
2721 struct inode
*inode
, struct ceph_mds_caps
*im
,
2722 struct ceph_mds_session
*session
,
2723 void *snaptrace
, int snaptrace_len
)
2725 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2726 int mds
= session
->s_mds
;
2727 unsigned issued
= le32_to_cpu(im
->caps
);
2728 unsigned wanted
= le32_to_cpu(im
->wanted
);
2729 unsigned seq
= le32_to_cpu(im
->seq
);
2730 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
2731 u64 realmino
= le64_to_cpu(im
->realm
);
2732 u64 cap_id
= le64_to_cpu(im
->cap_id
);
2734 if (ci
->i_cap_exporting_mds
>= 0 &&
2735 ceph_seq_cmp(ci
->i_cap_exporting_mseq
, mseq
) < 0) {
2736 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2737 " - cleared exporting from mds%d\n",
2738 inode
, ci
, mds
, mseq
,
2739 ci
->i_cap_exporting_mds
);
2740 ci
->i_cap_exporting_issued
= 0;
2741 ci
->i_cap_exporting_mseq
= 0;
2742 ci
->i_cap_exporting_mds
= -1;
2744 spin_lock(&mdsc
->cap_dirty_lock
);
2745 if (!list_empty(&ci
->i_dirty_item
)) {
2746 dout(" moving %p back to cap_dirty\n", inode
);
2747 list_move(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
2749 spin_unlock(&mdsc
->cap_dirty_lock
);
2751 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2752 inode
, ci
, mds
, mseq
);
2755 down_write(&mdsc
->snap_rwsem
);
2756 ceph_update_snap_trace(mdsc
, snaptrace
, snaptrace
+snaptrace_len
,
2758 downgrade_write(&mdsc
->snap_rwsem
);
2759 ceph_add_cap(inode
, session
, cap_id
, -1,
2760 issued
, wanted
, seq
, mseq
, realmino
, CEPH_CAP_FLAG_AUTH
,
2761 NULL
/* no caps context */);
2762 kick_flushing_inode_caps(mdsc
, session
, inode
);
2763 up_read(&mdsc
->snap_rwsem
);
2765 /* make sure we re-request max_size, if necessary */
2766 spin_lock(&ci
->i_ceph_lock
);
2767 ci
->i_wanted_max_size
= 0; /* reset */
2768 ci
->i_requested_max_size
= 0;
2769 spin_unlock(&ci
->i_ceph_lock
);
2773 * Handle a caps message from the MDS.
2775 * Identify the appropriate session, inode, and call the right handler
2776 * based on the cap op.
2778 void ceph_handle_caps(struct ceph_mds_session
*session
,
2779 struct ceph_msg
*msg
)
2781 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2782 struct super_block
*sb
= mdsc
->fsc
->sb
;
2783 struct inode
*inode
;
2784 struct ceph_inode_info
*ci
;
2785 struct ceph_cap
*cap
;
2786 struct ceph_mds_caps
*h
;
2787 int mds
= session
->s_mds
;
2790 struct ceph_vino vino
;
2795 size_t snaptrace_len
;
2798 int open_target_sessions
= 0;
2800 dout("handle_caps from mds%d\n", mds
);
2803 tid
= le64_to_cpu(msg
->hdr
.tid
);
2804 if (msg
->front
.iov_len
< sizeof(*h
))
2806 h
= msg
->front
.iov_base
;
2807 op
= le32_to_cpu(h
->op
);
2808 vino
.ino
= le64_to_cpu(h
->ino
);
2809 vino
.snap
= CEPH_NOSNAP
;
2810 cap_id
= le64_to_cpu(h
->cap_id
);
2811 seq
= le32_to_cpu(h
->seq
);
2812 mseq
= le32_to_cpu(h
->migrate_seq
);
2813 size
= le64_to_cpu(h
->size
);
2814 max_size
= le64_to_cpu(h
->max_size
);
2817 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
2819 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
2822 p
= snaptrace
+ snaptrace_len
;
2823 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
2824 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
2831 mutex_lock(&session
->s_mutex
);
2833 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
2837 inode
= ceph_find_inode(sb
, vino
);
2838 ci
= ceph_inode(inode
);
2839 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
2842 dout(" i don't have ino %llx\n", vino
.ino
);
2844 if (op
== CEPH_CAP_OP_IMPORT
)
2845 __queue_cap_release(session
, vino
.ino
, cap_id
,
2847 goto flush_cap_releases
;
2850 /* these will work even if we don't have a cap yet */
2852 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
2853 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
2856 case CEPH_CAP_OP_EXPORT
:
2857 handle_cap_export(inode
, h
, session
, &open_target_sessions
);
2860 case CEPH_CAP_OP_IMPORT
:
2861 handle_cap_import(mdsc
, inode
, h
, session
,
2862 snaptrace
, snaptrace_len
);
2865 /* the rest require a cap */
2866 spin_lock(&ci
->i_ceph_lock
);
2867 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
2869 dout(" no cap on %p ino %llx.%llx from mds%d\n",
2870 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
2871 spin_unlock(&ci
->i_ceph_lock
);
2872 goto flush_cap_releases
;
2875 /* note that each of these drops i_ceph_lock for us */
2877 case CEPH_CAP_OP_REVOKE
:
2878 case CEPH_CAP_OP_GRANT
:
2879 case CEPH_CAP_OP_IMPORT
:
2880 handle_cap_grant(inode
, h
, session
, cap
, msg
->middle
);
2883 case CEPH_CAP_OP_FLUSH_ACK
:
2884 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
2887 case CEPH_CAP_OP_TRUNC
:
2888 handle_cap_trunc(inode
, h
, session
);
2892 spin_unlock(&ci
->i_ceph_lock
);
2893 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
2894 ceph_cap_op_name(op
));
2901 * send any full release message to try to move things
2902 * along for the mds (who clearly thinks we still have this
2905 ceph_add_cap_releases(mdsc
, session
);
2906 ceph_send_cap_releases(mdsc
, session
);
2909 mutex_unlock(&session
->s_mutex
);
2913 if (open_target_sessions
)
2914 ceph_mdsc_open_export_target_sessions(mdsc
, session
);
2918 pr_err("ceph_handle_caps: corrupt message\n");
2924 * Delayed work handler to process end of delayed cap release LRU list.
2926 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
2928 struct ceph_inode_info
*ci
;
2929 int flags
= CHECK_CAPS_NODELAY
;
2931 dout("check_delayed_caps\n");
2933 spin_lock(&mdsc
->cap_delay_lock
);
2934 if (list_empty(&mdsc
->cap_delay_list
))
2936 ci
= list_first_entry(&mdsc
->cap_delay_list
,
2937 struct ceph_inode_info
,
2939 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
2940 time_before(jiffies
, ci
->i_hold_caps_max
))
2942 list_del_init(&ci
->i_cap_delay_list
);
2943 spin_unlock(&mdsc
->cap_delay_lock
);
2944 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
2945 ceph_check_caps(ci
, flags
, NULL
);
2947 spin_unlock(&mdsc
->cap_delay_lock
);
2951 * Flush all dirty caps to the mds
2953 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
2955 struct ceph_inode_info
*ci
;
2956 struct inode
*inode
;
2958 dout("flush_dirty_caps\n");
2959 spin_lock(&mdsc
->cap_dirty_lock
);
2960 while (!list_empty(&mdsc
->cap_dirty
)) {
2961 ci
= list_first_entry(&mdsc
->cap_dirty
, struct ceph_inode_info
,
2963 inode
= &ci
->vfs_inode
;
2965 dout("flush_dirty_caps %p\n", inode
);
2966 spin_unlock(&mdsc
->cap_dirty_lock
);
2967 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
, NULL
);
2969 spin_lock(&mdsc
->cap_dirty_lock
);
2971 spin_unlock(&mdsc
->cap_dirty_lock
);
2972 dout("flush_dirty_caps done\n");
2976 * Drop open file reference. If we were the last open file,
2977 * we may need to release capabilities to the MDS (or schedule
2978 * their delayed release).
2980 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
2982 struct inode
*inode
= &ci
->vfs_inode
;
2985 spin_lock(&ci
->i_ceph_lock
);
2986 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
2987 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
2988 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
2989 if (--ci
->i_nr_by_mode
[fmode
] == 0)
2991 spin_unlock(&ci
->i_ceph_lock
);
2993 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
2994 ceph_check_caps(ci
, 0, NULL
);
2998 * Helpers for embedding cap and dentry lease releases into mds
3001 * @force is used by dentry_release (below) to force inclusion of a
3002 * record for the directory inode, even when there aren't any caps to
3005 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
3006 int mds
, int drop
, int unless
, int force
)
3008 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3009 struct ceph_cap
*cap
;
3010 struct ceph_mds_request_release
*rel
= *p
;
3014 spin_lock(&ci
->i_ceph_lock
);
3015 used
= __ceph_caps_used(ci
);
3016 dirty
= __ceph_caps_dirty(ci
);
3018 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3019 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
3020 ceph_cap_string(unless
));
3022 /* only drop unused, clean caps */
3023 drop
&= ~(used
| dirty
);
3025 cap
= __get_cap_for_mds(ci
, mds
);
3026 if (cap
&& __cap_is_valid(cap
)) {
3028 ((cap
->issued
& drop
) &&
3029 (cap
->issued
& unless
) == 0)) {
3030 if ((cap
->issued
& drop
) &&
3031 (cap
->issued
& unless
) == 0) {
3032 dout("encode_inode_release %p cap %p %s -> "
3034 ceph_cap_string(cap
->issued
),
3035 ceph_cap_string(cap
->issued
& ~drop
));
3036 cap
->issued
&= ~drop
;
3037 cap
->implemented
&= ~drop
;
3038 if (ci
->i_ceph_flags
& CEPH_I_NODELAY
) {
3039 int wanted
= __ceph_caps_wanted(ci
);
3040 dout(" wanted %s -> %s (act %s)\n",
3041 ceph_cap_string(cap
->mds_wanted
),
3042 ceph_cap_string(cap
->mds_wanted
&
3044 ceph_cap_string(wanted
));
3045 cap
->mds_wanted
&= wanted
;
3048 dout("encode_inode_release %p cap %p %s"
3049 " (force)\n", inode
, cap
,
3050 ceph_cap_string(cap
->issued
));
3053 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3054 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3055 rel
->seq
= cpu_to_le32(cap
->seq
);
3056 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
),
3057 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3058 rel
->caps
= cpu_to_le32(cap
->issued
);
3059 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3065 dout("encode_inode_release %p cap %p %s\n",
3066 inode
, cap
, ceph_cap_string(cap
->issued
));
3069 spin_unlock(&ci
->i_ceph_lock
);
3073 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3074 int mds
, int drop
, int unless
)
3076 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3077 struct ceph_mds_request_release
*rel
= *p
;
3078 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3083 * force an record for the directory caps if we have a dentry lease.
3084 * this is racy (can't take i_ceph_lock and d_lock together), but it
3085 * doesn't have to be perfect; the mds will revoke anything we don't
3088 spin_lock(&dentry
->d_lock
);
3089 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3091 spin_unlock(&dentry
->d_lock
);
3093 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3095 spin_lock(&dentry
->d_lock
);
3096 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3097 dout("encode_dentry_release %p mds%d seq %d\n",
3098 dentry
, mds
, (int)di
->lease_seq
);
3099 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3100 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3101 *p
+= dentry
->d_name
.len
;
3102 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3103 __ceph_mdsc_drop_dentry_lease(dentry
);
3105 spin_unlock(&dentry
->d_lock
);