1 #include <linux/ceph/ceph_debug.h>
4 #include <linux/kernel.h>
5 #include <linux/sched.h>
6 #include <linux/slab.h>
7 #include <linux/vmalloc.h>
8 #include <linux/wait.h>
9 #include <linux/writeback.h>
12 #include "mds_client.h"
14 #include <linux/ceph/decode.h>
15 #include <linux/ceph/messenger.h>
18 * Capability management
20 * The Ceph metadata servers control client access to inode metadata
21 * and file data by issuing capabilities, granting clients permission
22 * to read and/or write both inode field and file data to OSDs
23 * (storage nodes). Each capability consists of a set of bits
24 * indicating which operations are allowed.
26 * If the client holds a *_SHARED cap, the client has a coherent value
27 * that can be safely read from the cached inode.
29 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
30 * client is allowed to change inode attributes (e.g., file size,
31 * mtime), note its dirty state in the ceph_cap, and asynchronously
32 * flush that metadata change to the MDS.
34 * In the event of a conflicting operation (perhaps by another
35 * client), the MDS will revoke the conflicting client capabilities.
37 * In order for a client to cache an inode, it must hold a capability
38 * with at least one MDS server. When inodes are released, release
39 * notifications are batched and periodically sent en masse to the MDS
40 * cluster to release server state.
45 * Generate readable cap strings for debugging output.
47 #define MAX_CAP_STR 20
48 static char cap_str
[MAX_CAP_STR
][40];
49 static DEFINE_SPINLOCK(cap_str_lock
);
50 static int last_cap_str
;
52 static char *gcap_string(char *s
, int c
)
54 if (c
& CEPH_CAP_GSHARED
)
56 if (c
& CEPH_CAP_GEXCL
)
58 if (c
& CEPH_CAP_GCACHE
)
64 if (c
& CEPH_CAP_GBUFFER
)
66 if (c
& CEPH_CAP_GLAZYIO
)
71 const char *ceph_cap_string(int caps
)
77 spin_lock(&cap_str_lock
);
79 if (last_cap_str
== MAX_CAP_STR
)
81 spin_unlock(&cap_str_lock
);
85 if (caps
& CEPH_CAP_PIN
)
88 c
= (caps
>> CEPH_CAP_SAUTH
) & 3;
91 s
= gcap_string(s
, c
);
94 c
= (caps
>> CEPH_CAP_SLINK
) & 3;
97 s
= gcap_string(s
, c
);
100 c
= (caps
>> CEPH_CAP_SXATTR
) & 3;
103 s
= gcap_string(s
, c
);
106 c
= caps
>> CEPH_CAP_SFILE
;
109 s
= gcap_string(s
, c
);
118 void ceph_caps_init(struct ceph_mds_client
*mdsc
)
120 INIT_LIST_HEAD(&mdsc
->caps_list
);
121 spin_lock_init(&mdsc
->caps_list_lock
);
124 void ceph_caps_finalize(struct ceph_mds_client
*mdsc
)
126 struct ceph_cap
*cap
;
128 spin_lock(&mdsc
->caps_list_lock
);
129 while (!list_empty(&mdsc
->caps_list
)) {
130 cap
= list_first_entry(&mdsc
->caps_list
,
131 struct ceph_cap
, caps_item
);
132 list_del(&cap
->caps_item
);
133 kmem_cache_free(ceph_cap_cachep
, cap
);
135 mdsc
->caps_total_count
= 0;
136 mdsc
->caps_avail_count
= 0;
137 mdsc
->caps_use_count
= 0;
138 mdsc
->caps_reserve_count
= 0;
139 mdsc
->caps_min_count
= 0;
140 spin_unlock(&mdsc
->caps_list_lock
);
143 void ceph_adjust_min_caps(struct ceph_mds_client
*mdsc
, int delta
)
145 spin_lock(&mdsc
->caps_list_lock
);
146 mdsc
->caps_min_count
+= delta
;
147 BUG_ON(mdsc
->caps_min_count
< 0);
148 spin_unlock(&mdsc
->caps_list_lock
);
151 void ceph_reserve_caps(struct ceph_mds_client
*mdsc
,
152 struct ceph_cap_reservation
*ctx
, int need
)
155 struct ceph_cap
*cap
;
160 dout("reserve caps ctx=%p need=%d\n", ctx
, need
);
162 /* first reserve any caps that are already allocated */
163 spin_lock(&mdsc
->caps_list_lock
);
164 if (mdsc
->caps_avail_count
>= need
)
167 have
= mdsc
->caps_avail_count
;
168 mdsc
->caps_avail_count
-= have
;
169 mdsc
->caps_reserve_count
+= have
;
170 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
171 mdsc
->caps_reserve_count
+
172 mdsc
->caps_avail_count
);
173 spin_unlock(&mdsc
->caps_list_lock
);
175 for (i
= have
; i
< need
; i
++) {
176 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
179 list_add(&cap
->caps_item
, &newcaps
);
182 /* we didn't manage to reserve as much as we needed */
183 if (have
+ alloc
!= need
)
184 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
185 ctx
, need
, have
+ alloc
);
187 spin_lock(&mdsc
->caps_list_lock
);
188 mdsc
->caps_total_count
+= alloc
;
189 mdsc
->caps_reserve_count
+= alloc
;
190 list_splice(&newcaps
, &mdsc
->caps_list
);
192 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
193 mdsc
->caps_reserve_count
+
194 mdsc
->caps_avail_count
);
195 spin_unlock(&mdsc
->caps_list_lock
);
198 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
199 ctx
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
200 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
203 int ceph_unreserve_caps(struct ceph_mds_client
*mdsc
,
204 struct ceph_cap_reservation
*ctx
)
206 dout("unreserve caps ctx=%p count=%d\n", ctx
, ctx
->count
);
208 spin_lock(&mdsc
->caps_list_lock
);
209 BUG_ON(mdsc
->caps_reserve_count
< ctx
->count
);
210 mdsc
->caps_reserve_count
-= ctx
->count
;
211 mdsc
->caps_avail_count
+= ctx
->count
;
213 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
214 mdsc
->caps_total_count
, mdsc
->caps_use_count
,
215 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
216 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
217 mdsc
->caps_reserve_count
+
218 mdsc
->caps_avail_count
);
219 spin_unlock(&mdsc
->caps_list_lock
);
224 static struct ceph_cap
*get_cap(struct ceph_mds_client
*mdsc
,
225 struct ceph_cap_reservation
*ctx
)
227 struct ceph_cap
*cap
= NULL
;
229 /* temporary, until we do something about cap import/export */
231 cap
= kmem_cache_alloc(ceph_cap_cachep
, GFP_NOFS
);
233 spin_lock(&mdsc
->caps_list_lock
);
234 mdsc
->caps_use_count
++;
235 mdsc
->caps_total_count
++;
236 spin_unlock(&mdsc
->caps_list_lock
);
241 spin_lock(&mdsc
->caps_list_lock
);
242 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
243 ctx
, ctx
->count
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
244 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
246 BUG_ON(ctx
->count
> mdsc
->caps_reserve_count
);
247 BUG_ON(list_empty(&mdsc
->caps_list
));
250 mdsc
->caps_reserve_count
--;
251 mdsc
->caps_use_count
++;
253 cap
= list_first_entry(&mdsc
->caps_list
, struct ceph_cap
, caps_item
);
254 list_del(&cap
->caps_item
);
256 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
257 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
258 spin_unlock(&mdsc
->caps_list_lock
);
262 void ceph_put_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
)
264 spin_lock(&mdsc
->caps_list_lock
);
265 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
266 cap
, mdsc
->caps_total_count
, mdsc
->caps_use_count
,
267 mdsc
->caps_reserve_count
, mdsc
->caps_avail_count
);
268 mdsc
->caps_use_count
--;
270 * Keep some preallocated caps around (ceph_min_count), to
271 * avoid lots of free/alloc churn.
273 if (mdsc
->caps_avail_count
>= mdsc
->caps_reserve_count
+
274 mdsc
->caps_min_count
) {
275 mdsc
->caps_total_count
--;
276 kmem_cache_free(ceph_cap_cachep
, cap
);
278 mdsc
->caps_avail_count
++;
279 list_add(&cap
->caps_item
, &mdsc
->caps_list
);
282 BUG_ON(mdsc
->caps_total_count
!= mdsc
->caps_use_count
+
283 mdsc
->caps_reserve_count
+ mdsc
->caps_avail_count
);
284 spin_unlock(&mdsc
->caps_list_lock
);
287 void ceph_reservation_status(struct ceph_fs_client
*fsc
,
288 int *total
, int *avail
, int *used
, int *reserved
,
291 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
294 *total
= mdsc
->caps_total_count
;
296 *avail
= mdsc
->caps_avail_count
;
298 *used
= mdsc
->caps_use_count
;
300 *reserved
= mdsc
->caps_reserve_count
;
302 *min
= mdsc
->caps_min_count
;
306 * Find ceph_cap for given mds, if any.
308 * Called with i_ceph_lock held.
310 static struct ceph_cap
*__get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
312 struct ceph_cap
*cap
;
313 struct rb_node
*n
= ci
->i_caps
.rb_node
;
316 cap
= rb_entry(n
, struct ceph_cap
, ci_node
);
319 else if (mds
> cap
->mds
)
327 struct ceph_cap
*ceph_get_cap_for_mds(struct ceph_inode_info
*ci
, int mds
)
329 struct ceph_cap
*cap
;
331 spin_lock(&ci
->i_ceph_lock
);
332 cap
= __get_cap_for_mds(ci
, mds
);
333 spin_unlock(&ci
->i_ceph_lock
);
338 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
340 static int __ceph_get_cap_mds(struct ceph_inode_info
*ci
)
342 struct ceph_cap
*cap
;
346 /* prefer mds with WR|BUFFER|EXCL caps */
347 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
348 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
350 if (cap
->issued
& (CEPH_CAP_FILE_WR
|
351 CEPH_CAP_FILE_BUFFER
|
358 int ceph_get_cap_mds(struct inode
*inode
)
360 struct ceph_inode_info
*ci
= ceph_inode(inode
);
362 spin_lock(&ci
->i_ceph_lock
);
363 mds
= __ceph_get_cap_mds(ceph_inode(inode
));
364 spin_unlock(&ci
->i_ceph_lock
);
369 * Called under i_ceph_lock.
371 static void __insert_cap_node(struct ceph_inode_info
*ci
,
372 struct ceph_cap
*new)
374 struct rb_node
**p
= &ci
->i_caps
.rb_node
;
375 struct rb_node
*parent
= NULL
;
376 struct ceph_cap
*cap
= NULL
;
380 cap
= rb_entry(parent
, struct ceph_cap
, ci_node
);
381 if (new->mds
< cap
->mds
)
383 else if (new->mds
> cap
->mds
)
389 rb_link_node(&new->ci_node
, parent
, p
);
390 rb_insert_color(&new->ci_node
, &ci
->i_caps
);
394 * (re)set cap hold timeouts, which control the delayed release
395 * of unused caps back to the MDS. Should be called on cap use.
397 static void __cap_set_timeouts(struct ceph_mds_client
*mdsc
,
398 struct ceph_inode_info
*ci
)
400 struct ceph_mount_options
*ma
= mdsc
->fsc
->mount_options
;
402 ci
->i_hold_caps_min
= round_jiffies(jiffies
+
403 ma
->caps_wanted_delay_min
* HZ
);
404 ci
->i_hold_caps_max
= round_jiffies(jiffies
+
405 ma
->caps_wanted_delay_max
* HZ
);
406 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci
->vfs_inode
,
407 ci
->i_hold_caps_min
- jiffies
, ci
->i_hold_caps_max
- jiffies
);
411 * (Re)queue cap at the end of the delayed cap release list.
413 * If I_FLUSH is set, leave the inode at the front of the list.
415 * Caller holds i_ceph_lock
416 * -> we take mdsc->cap_delay_lock
418 static void __cap_delay_requeue(struct ceph_mds_client
*mdsc
,
419 struct ceph_inode_info
*ci
)
421 __cap_set_timeouts(mdsc
, ci
);
422 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci
->vfs_inode
,
423 ci
->i_ceph_flags
, ci
->i_hold_caps_max
);
424 if (!mdsc
->stopping
) {
425 spin_lock(&mdsc
->cap_delay_lock
);
426 if (!list_empty(&ci
->i_cap_delay_list
)) {
427 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
429 list_del_init(&ci
->i_cap_delay_list
);
431 list_add_tail(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
433 spin_unlock(&mdsc
->cap_delay_lock
);
438 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
439 * indicating we should send a cap message to flush dirty metadata
440 * asap, and move to the front of the delayed cap list.
442 static void __cap_delay_requeue_front(struct ceph_mds_client
*mdsc
,
443 struct ceph_inode_info
*ci
)
445 dout("__cap_delay_requeue_front %p\n", &ci
->vfs_inode
);
446 spin_lock(&mdsc
->cap_delay_lock
);
447 ci
->i_ceph_flags
|= CEPH_I_FLUSH
;
448 if (!list_empty(&ci
->i_cap_delay_list
))
449 list_del_init(&ci
->i_cap_delay_list
);
450 list_add(&ci
->i_cap_delay_list
, &mdsc
->cap_delay_list
);
451 spin_unlock(&mdsc
->cap_delay_lock
);
455 * Cancel delayed work on cap.
457 * Caller must hold i_ceph_lock.
459 static void __cap_delay_cancel(struct ceph_mds_client
*mdsc
,
460 struct ceph_inode_info
*ci
)
462 dout("__cap_delay_cancel %p\n", &ci
->vfs_inode
);
463 if (list_empty(&ci
->i_cap_delay_list
))
465 spin_lock(&mdsc
->cap_delay_lock
);
466 list_del_init(&ci
->i_cap_delay_list
);
467 spin_unlock(&mdsc
->cap_delay_lock
);
471 * Common issue checks for add_cap, handle_cap_grant.
473 static void __check_cap_issue(struct ceph_inode_info
*ci
, struct ceph_cap
*cap
,
476 unsigned had
= __ceph_caps_issued(ci
, NULL
);
479 * Each time we receive FILE_CACHE anew, we increment
482 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) &&
483 (had
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0) {
488 * if we are newly issued FILE_SHARED, mark dir not complete; we
489 * don't know what happened to this directory while we didn't
492 if ((issued
& CEPH_CAP_FILE_SHARED
) &&
493 (had
& CEPH_CAP_FILE_SHARED
) == 0) {
495 if (S_ISDIR(ci
->vfs_inode
.i_mode
)) {
496 dout(" marking %p NOT complete\n", &ci
->vfs_inode
);
497 __ceph_dir_clear_complete(ci
);
503 * Add a capability under the given MDS session.
505 * Caller should hold session snap_rwsem (read) and s_mutex.
507 * @fmode is the open file mode, if we are opening a file, otherwise
508 * it is < 0. (This is so we can atomically add the cap and add an
509 * open file reference to it.)
511 int ceph_add_cap(struct inode
*inode
,
512 struct ceph_mds_session
*session
, u64 cap_id
,
513 int fmode
, unsigned issued
, unsigned wanted
,
514 unsigned seq
, unsigned mseq
, u64 realmino
, int flags
,
515 struct ceph_cap_reservation
*caps_reservation
)
517 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
518 struct ceph_inode_info
*ci
= ceph_inode(inode
);
519 struct ceph_cap
*new_cap
= NULL
;
520 struct ceph_cap
*cap
;
521 int mds
= session
->s_mds
;
524 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode
,
525 session
->s_mds
, cap_id
, ceph_cap_string(issued
), seq
);
528 * If we are opening the file, include file mode wanted bits
532 wanted
|= ceph_caps_for_mode(fmode
);
535 spin_lock(&ci
->i_ceph_lock
);
536 cap
= __get_cap_for_mds(ci
, mds
);
542 spin_unlock(&ci
->i_ceph_lock
);
543 new_cap
= get_cap(mdsc
, caps_reservation
);
550 cap
->implemented
= 0;
556 __insert_cap_node(ci
, cap
);
558 /* add to session cap list */
559 cap
->session
= session
;
560 spin_lock(&session
->s_cap_lock
);
561 list_add_tail(&cap
->session_caps
, &session
->s_caps
);
562 session
->s_nr_caps
++;
563 spin_unlock(&session
->s_cap_lock
);
566 ceph_put_cap(mdsc
, new_cap
);
569 * auth mds of the inode changed. we received the cap export
570 * message, but still haven't received the cap import message.
571 * handle_cap_export() updated the new auth MDS' cap.
573 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
574 * a message that was send before the cap import message. So
577 if (ceph_seq_cmp(seq
, cap
->seq
) <= 0) {
578 WARN_ON(cap
!= ci
->i_auth_cap
);
579 WARN_ON(cap
->cap_id
!= cap_id
);
582 issued
|= cap
->issued
;
583 flags
|= CEPH_CAP_FLAG_AUTH
;
587 if (!ci
->i_snap_realm
) {
589 * add this inode to the appropriate snap realm
591 struct ceph_snap_realm
*realm
= ceph_lookup_snap_realm(mdsc
,
594 ceph_get_snap_realm(mdsc
, realm
);
595 spin_lock(&realm
->inodes_with_caps_lock
);
596 ci
->i_snap_realm
= realm
;
597 list_add(&ci
->i_snap_realm_item
,
598 &realm
->inodes_with_caps
);
599 spin_unlock(&realm
->inodes_with_caps_lock
);
601 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
607 __check_cap_issue(ci
, cap
, issued
);
610 * If we are issued caps we don't want, or the mds' wanted
611 * value appears to be off, queue a check so we'll release
612 * later and/or update the mds wanted value.
614 actual_wanted
= __ceph_caps_wanted(ci
);
615 if ((wanted
& ~actual_wanted
) ||
616 (issued
& ~actual_wanted
& CEPH_CAP_ANY_WR
)) {
617 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
618 ceph_cap_string(issued
), ceph_cap_string(wanted
),
619 ceph_cap_string(actual_wanted
));
620 __cap_delay_requeue(mdsc
, ci
);
623 if (flags
& CEPH_CAP_FLAG_AUTH
) {
624 if (ci
->i_auth_cap
== NULL
||
625 ceph_seq_cmp(ci
->i_auth_cap
->mseq
, mseq
) < 0)
626 ci
->i_auth_cap
= cap
;
627 ci
->i_cap_exporting_issued
= 0;
629 WARN_ON(ci
->i_auth_cap
== cap
);
632 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
633 inode
, ceph_vinop(inode
), cap
, ceph_cap_string(issued
),
634 ceph_cap_string(issued
|cap
->issued
), seq
, mds
);
635 cap
->cap_id
= cap_id
;
636 cap
->issued
= issued
;
637 cap
->implemented
|= issued
;
638 if (ceph_seq_cmp(mseq
, cap
->mseq
) > 0)
639 cap
->mds_wanted
= wanted
;
641 cap
->mds_wanted
|= wanted
;
643 cap
->issue_seq
= seq
;
645 cap
->cap_gen
= session
->s_cap_gen
;
648 __ceph_get_fmode(ci
, fmode
);
649 spin_unlock(&ci
->i_ceph_lock
);
650 wake_up_all(&ci
->i_cap_wq
);
655 * Return true if cap has not timed out and belongs to the current
656 * generation of the MDS session (i.e. has not gone 'stale' due to
657 * us losing touch with the mds).
659 static int __cap_is_valid(struct ceph_cap
*cap
)
664 spin_lock(&cap
->session
->s_gen_ttl_lock
);
665 gen
= cap
->session
->s_cap_gen
;
666 ttl
= cap
->session
->s_cap_ttl
;
667 spin_unlock(&cap
->session
->s_gen_ttl_lock
);
669 if (cap
->cap_gen
< gen
|| time_after_eq(jiffies
, ttl
)) {
670 dout("__cap_is_valid %p cap %p issued %s "
671 "but STALE (gen %u vs %u)\n", &cap
->ci
->vfs_inode
,
672 cap
, ceph_cap_string(cap
->issued
), cap
->cap_gen
, gen
);
680 * Return set of valid cap bits issued to us. Note that caps time
681 * out, and may be invalidated in bulk if the client session times out
682 * and session->s_cap_gen is bumped.
684 int __ceph_caps_issued(struct ceph_inode_info
*ci
, int *implemented
)
686 int have
= ci
->i_snap_caps
| ci
->i_cap_exporting_issued
;
687 struct ceph_cap
*cap
;
692 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
693 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
694 if (!__cap_is_valid(cap
))
696 dout("__ceph_caps_issued %p cap %p issued %s\n",
697 &ci
->vfs_inode
, cap
, ceph_cap_string(cap
->issued
));
700 *implemented
|= cap
->implemented
;
703 * exclude caps issued by non-auth MDS, but are been revoking
704 * by the auth MDS. The non-auth MDS should be revoking/exporting
705 * these caps, but the message is delayed.
707 if (ci
->i_auth_cap
) {
708 cap
= ci
->i_auth_cap
;
709 have
&= ~cap
->implemented
| cap
->issued
;
715 * Get cap bits issued by caps other than @ocap
717 int __ceph_caps_issued_other(struct ceph_inode_info
*ci
, struct ceph_cap
*ocap
)
719 int have
= ci
->i_snap_caps
;
720 struct ceph_cap
*cap
;
723 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
724 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
727 if (!__cap_is_valid(cap
))
735 * Move a cap to the end of the LRU (oldest caps at list head, newest
738 static void __touch_cap(struct ceph_cap
*cap
)
740 struct ceph_mds_session
*s
= cap
->session
;
742 spin_lock(&s
->s_cap_lock
);
743 if (s
->s_cap_iterator
== NULL
) {
744 dout("__touch_cap %p cap %p mds%d\n", &cap
->ci
->vfs_inode
, cap
,
746 list_move_tail(&cap
->session_caps
, &s
->s_caps
);
748 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
749 &cap
->ci
->vfs_inode
, cap
, s
->s_mds
);
751 spin_unlock(&s
->s_cap_lock
);
755 * Check if we hold the given mask. If so, move the cap(s) to the
756 * front of their respective LRUs. (This is the preferred way for
757 * callers to check for caps they want.)
759 int __ceph_caps_issued_mask(struct ceph_inode_info
*ci
, int mask
, int touch
)
761 struct ceph_cap
*cap
;
763 int have
= ci
->i_snap_caps
;
765 if ((have
& mask
) == mask
) {
766 dout("__ceph_caps_issued_mask %p snap issued %s"
767 " (mask %s)\n", &ci
->vfs_inode
,
768 ceph_cap_string(have
),
769 ceph_cap_string(mask
));
773 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
774 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
775 if (!__cap_is_valid(cap
))
777 if ((cap
->issued
& mask
) == mask
) {
778 dout("__ceph_caps_issued_mask %p cap %p issued %s"
779 " (mask %s)\n", &ci
->vfs_inode
, cap
,
780 ceph_cap_string(cap
->issued
),
781 ceph_cap_string(mask
));
787 /* does a combination of caps satisfy mask? */
789 if ((have
& mask
) == mask
) {
790 dout("__ceph_caps_issued_mask %p combo issued %s"
791 " (mask %s)\n", &ci
->vfs_inode
,
792 ceph_cap_string(cap
->issued
),
793 ceph_cap_string(mask
));
797 /* touch this + preceding caps */
799 for (q
= rb_first(&ci
->i_caps
); q
!= p
;
801 cap
= rb_entry(q
, struct ceph_cap
,
803 if (!__cap_is_valid(cap
))
816 * Return true if mask caps are currently being revoked by an MDS.
818 int __ceph_caps_revoking_other(struct ceph_inode_info
*ci
,
819 struct ceph_cap
*ocap
, int mask
)
821 struct ceph_cap
*cap
;
824 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
825 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
827 (cap
->implemented
& ~cap
->issued
& mask
))
833 int ceph_caps_revoking(struct ceph_inode_info
*ci
, int mask
)
835 struct inode
*inode
= &ci
->vfs_inode
;
838 spin_lock(&ci
->i_ceph_lock
);
839 ret
= __ceph_caps_revoking_other(ci
, NULL
, mask
);
840 spin_unlock(&ci
->i_ceph_lock
);
841 dout("ceph_caps_revoking %p %s = %d\n", inode
,
842 ceph_cap_string(mask
), ret
);
846 int __ceph_caps_used(struct ceph_inode_info
*ci
)
850 used
|= CEPH_CAP_PIN
;
852 used
|= CEPH_CAP_FILE_RD
;
853 if (ci
->i_rdcache_ref
|| ci
->vfs_inode
.i_data
.nrpages
)
854 used
|= CEPH_CAP_FILE_CACHE
;
856 used
|= CEPH_CAP_FILE_WR
;
857 if (ci
->i_wb_ref
|| ci
->i_wrbuffer_ref
)
858 used
|= CEPH_CAP_FILE_BUFFER
;
863 * wanted, by virtue of open file modes
865 int __ceph_caps_file_wanted(struct ceph_inode_info
*ci
)
869 for (mode
= 0; mode
< CEPH_FILE_MODE_NUM
; mode
++)
870 if (ci
->i_nr_by_mode
[mode
])
871 want
|= ceph_caps_for_mode(mode
);
876 * Return caps we have registered with the MDS(s) as 'wanted'.
878 int __ceph_caps_mds_wanted(struct ceph_inode_info
*ci
)
880 struct ceph_cap
*cap
;
884 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
885 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
886 if (!__cap_is_valid(cap
))
888 mds_wanted
|= cap
->mds_wanted
;
894 * called under i_ceph_lock
896 static int __ceph_is_any_caps(struct ceph_inode_info
*ci
)
898 return !RB_EMPTY_ROOT(&ci
->i_caps
) || ci
->i_cap_exporting_issued
;
901 int ceph_is_any_caps(struct inode
*inode
)
903 struct ceph_inode_info
*ci
= ceph_inode(inode
);
906 spin_lock(&ci
->i_ceph_lock
);
907 ret
= __ceph_is_any_caps(ci
);
908 spin_unlock(&ci
->i_ceph_lock
);
914 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
916 * caller should hold i_ceph_lock.
917 * caller will not hold session s_mutex if called from destroy_inode.
919 void __ceph_remove_cap(struct ceph_cap
*cap
, bool queue_release
)
921 struct ceph_mds_session
*session
= cap
->session
;
922 struct ceph_inode_info
*ci
= cap
->ci
;
923 struct ceph_mds_client
*mdsc
=
924 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
927 dout("__ceph_remove_cap %p from %p\n", cap
, &ci
->vfs_inode
);
929 /* remove from session list */
930 spin_lock(&session
->s_cap_lock
);
932 * s_cap_reconnect is protected by s_cap_lock. no one changes
933 * s_cap_gen while session is in the reconnect state.
936 (!session
->s_cap_reconnect
||
937 cap
->cap_gen
== session
->s_cap_gen
))
938 __queue_cap_release(session
, ci
->i_vino
.ino
, cap
->cap_id
,
939 cap
->mseq
, cap
->issue_seq
);
941 if (session
->s_cap_iterator
== cap
) {
942 /* not yet, we are iterating over this very cap */
943 dout("__ceph_remove_cap delaying %p removal from session %p\n",
946 list_del_init(&cap
->session_caps
);
947 session
->s_nr_caps
--;
951 /* protect backpointer with s_cap_lock: see iterate_session_caps */
953 spin_unlock(&session
->s_cap_lock
);
955 /* remove from inode list */
956 rb_erase(&cap
->ci_node
, &ci
->i_caps
);
957 if (ci
->i_auth_cap
== cap
)
958 ci
->i_auth_cap
= NULL
;
961 ceph_put_cap(mdsc
, cap
);
963 if (!__ceph_is_any_caps(ci
) && ci
->i_snap_realm
) {
964 struct ceph_snap_realm
*realm
= ci
->i_snap_realm
;
965 spin_lock(&realm
->inodes_with_caps_lock
);
966 list_del_init(&ci
->i_snap_realm_item
);
967 ci
->i_snap_realm_counter
++;
968 ci
->i_snap_realm
= NULL
;
969 spin_unlock(&realm
->inodes_with_caps_lock
);
970 ceph_put_snap_realm(mdsc
, realm
);
972 if (!__ceph_is_any_real_caps(ci
))
973 __cap_delay_cancel(mdsc
, ci
);
977 * Build and send a cap message to the given MDS.
979 * Caller should be holding s_mutex.
981 static int send_cap_msg(struct ceph_mds_session
*session
,
982 u64 ino
, u64 cid
, int op
,
983 int caps
, int wanted
, int dirty
,
984 u32 seq
, u64 flush_tid
, u32 issue_seq
, u32 mseq
,
985 u64 size
, u64 max_size
,
986 struct timespec
*mtime
, struct timespec
*atime
,
988 kuid_t uid
, kgid_t gid
, umode_t mode
,
990 struct ceph_buffer
*xattrs_buf
,
993 struct ceph_mds_caps
*fc
;
994 struct ceph_msg
*msg
;
996 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
997 " seq %u/%u mseq %u follows %lld size %llu/%llu"
998 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op
),
999 cid
, ino
, ceph_cap_string(caps
), ceph_cap_string(wanted
),
1000 ceph_cap_string(dirty
),
1001 seq
, issue_seq
, mseq
, follows
, size
, max_size
,
1002 xattr_version
, xattrs_buf
? (int)xattrs_buf
->vec
.iov_len
: 0);
1004 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPS
, sizeof(*fc
), GFP_NOFS
, false);
1008 msg
->hdr
.tid
= cpu_to_le64(flush_tid
);
1010 fc
= msg
->front
.iov_base
;
1011 memset(fc
, 0, sizeof(*fc
));
1013 fc
->cap_id
= cpu_to_le64(cid
);
1014 fc
->op
= cpu_to_le32(op
);
1015 fc
->seq
= cpu_to_le32(seq
);
1016 fc
->issue_seq
= cpu_to_le32(issue_seq
);
1017 fc
->migrate_seq
= cpu_to_le32(mseq
);
1018 fc
->caps
= cpu_to_le32(caps
);
1019 fc
->wanted
= cpu_to_le32(wanted
);
1020 fc
->dirty
= cpu_to_le32(dirty
);
1021 fc
->ino
= cpu_to_le64(ino
);
1022 fc
->snap_follows
= cpu_to_le64(follows
);
1024 fc
->size
= cpu_to_le64(size
);
1025 fc
->max_size
= cpu_to_le64(max_size
);
1027 ceph_encode_timespec(&fc
->mtime
, mtime
);
1029 ceph_encode_timespec(&fc
->atime
, atime
);
1030 fc
->time_warp_seq
= cpu_to_le32(time_warp_seq
);
1032 fc
->uid
= cpu_to_le32(from_kuid(&init_user_ns
, uid
));
1033 fc
->gid
= cpu_to_le32(from_kgid(&init_user_ns
, gid
));
1034 fc
->mode
= cpu_to_le32(mode
);
1036 fc
->xattr_version
= cpu_to_le64(xattr_version
);
1038 msg
->middle
= ceph_buffer_get(xattrs_buf
);
1039 fc
->xattr_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1040 msg
->hdr
.middle_len
= cpu_to_le32(xattrs_buf
->vec
.iov_len
);
1043 ceph_con_send(&session
->s_con
, msg
);
1047 void __queue_cap_release(struct ceph_mds_session
*session
,
1048 u64 ino
, u64 cap_id
, u32 migrate_seq
,
1051 struct ceph_msg
*msg
;
1052 struct ceph_mds_cap_release
*head
;
1053 struct ceph_mds_cap_item
*item
;
1055 BUG_ON(!session
->s_num_cap_releases
);
1056 msg
= list_first_entry(&session
->s_cap_releases
,
1057 struct ceph_msg
, list_head
);
1059 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1060 ino
, session
->s_mds
, msg
, session
->s_num_cap_releases
);
1062 BUG_ON(msg
->front
.iov_len
+ sizeof(*item
) > PAGE_CACHE_SIZE
);
1063 head
= msg
->front
.iov_base
;
1064 le32_add_cpu(&head
->num
, 1);
1065 item
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1066 item
->ino
= cpu_to_le64(ino
);
1067 item
->cap_id
= cpu_to_le64(cap_id
);
1068 item
->migrate_seq
= cpu_to_le32(migrate_seq
);
1069 item
->seq
= cpu_to_le32(issue_seq
);
1071 session
->s_num_cap_releases
--;
1073 msg
->front
.iov_len
+= sizeof(*item
);
1074 if (le32_to_cpu(head
->num
) == CEPH_CAPS_PER_RELEASE
) {
1075 dout(" release msg %p full\n", msg
);
1076 list_move_tail(&msg
->list_head
, &session
->s_cap_releases_done
);
1078 dout(" release msg %p at %d/%d (%d)\n", msg
,
1079 (int)le32_to_cpu(head
->num
),
1080 (int)CEPH_CAPS_PER_RELEASE
,
1081 (int)msg
->front
.iov_len
);
1086 * Queue cap releases when an inode is dropped from our cache. Since
1087 * inode is about to be destroyed, there is no need for i_ceph_lock.
1089 void ceph_queue_caps_release(struct inode
*inode
)
1091 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1094 p
= rb_first(&ci
->i_caps
);
1096 struct ceph_cap
*cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1098 __ceph_remove_cap(cap
, true);
1103 * Send a cap msg on the given inode. Update our caps state, then
1104 * drop i_ceph_lock and send the message.
1106 * Make note of max_size reported/requested from mds, revoked caps
1107 * that have now been implemented.
1109 * Make half-hearted attempt ot to invalidate page cache if we are
1110 * dropping RDCACHE. Note that this will leave behind locked pages
1111 * that we'll then need to deal with elsewhere.
1113 * Return non-zero if delayed release, or we experienced an error
1114 * such that the caller should requeue + retry later.
1116 * called with i_ceph_lock, then drops it.
1117 * caller should hold snap_rwsem (read), s_mutex.
1119 static int __send_cap(struct ceph_mds_client
*mdsc
, struct ceph_cap
*cap
,
1120 int op
, int used
, int want
, int retain
, int flushing
,
1121 unsigned *pflush_tid
)
1122 __releases(cap
->ci
->i_ceph_lock
)
1124 struct ceph_inode_info
*ci
= cap
->ci
;
1125 struct inode
*inode
= &ci
->vfs_inode
;
1126 u64 cap_id
= cap
->cap_id
;
1127 int held
, revoking
, dropping
, keep
;
1128 u64 seq
, issue_seq
, mseq
, time_warp_seq
, follows
;
1130 struct timespec mtime
, atime
;
1135 struct ceph_mds_session
*session
;
1136 u64 xattr_version
= 0;
1137 struct ceph_buffer
*xattr_blob
= NULL
;
1143 held
= cap
->issued
| cap
->implemented
;
1144 revoking
= cap
->implemented
& ~cap
->issued
;
1145 retain
&= ~revoking
;
1146 dropping
= cap
->issued
& ~retain
;
1148 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1149 inode
, cap
, cap
->session
,
1150 ceph_cap_string(held
), ceph_cap_string(held
& retain
),
1151 ceph_cap_string(revoking
));
1152 BUG_ON((retain
& CEPH_CAP_PIN
) == 0);
1154 session
= cap
->session
;
1156 /* don't release wanted unless we've waited a bit. */
1157 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1158 time_before(jiffies
, ci
->i_hold_caps_min
)) {
1159 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1160 ceph_cap_string(cap
->issued
),
1161 ceph_cap_string(cap
->issued
& retain
),
1162 ceph_cap_string(cap
->mds_wanted
),
1163 ceph_cap_string(want
));
1164 want
|= cap
->mds_wanted
;
1165 retain
|= cap
->issued
;
1168 ci
->i_ceph_flags
&= ~(CEPH_I_NODELAY
| CEPH_I_FLUSH
);
1170 cap
->issued
&= retain
; /* drop bits we don't want */
1171 if (cap
->implemented
& ~cap
->issued
) {
1173 * Wake up any waiters on wanted -> needed transition.
1174 * This is due to the weird transition from buffered
1175 * to sync IO... we need to flush dirty pages _before_
1176 * allowing sync writes to avoid reordering.
1180 cap
->implemented
&= cap
->issued
| used
;
1181 cap
->mds_wanted
= want
;
1185 * assign a tid for flush operations so we can avoid
1186 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1187 * clean type races. track latest tid for every bit
1188 * so we can handle flush AxFw, flush Fw, and have the
1189 * first ack clean Ax.
1191 flush_tid
= ++ci
->i_cap_flush_last_tid
;
1193 *pflush_tid
= flush_tid
;
1194 dout(" cap_flush_tid %d\n", (int)flush_tid
);
1195 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1196 if (flushing
& (1 << i
))
1197 ci
->i_cap_flush_tid
[i
] = flush_tid
;
1199 follows
= ci
->i_head_snapc
->seq
;
1204 keep
= cap
->implemented
;
1206 issue_seq
= cap
->issue_seq
;
1208 size
= inode
->i_size
;
1209 ci
->i_reported_size
= size
;
1210 max_size
= ci
->i_wanted_max_size
;
1211 ci
->i_requested_max_size
= max_size
;
1212 mtime
= inode
->i_mtime
;
1213 atime
= inode
->i_atime
;
1214 time_warp_seq
= ci
->i_time_warp_seq
;
1217 mode
= inode
->i_mode
;
1219 if (flushing
& CEPH_CAP_XATTR_EXCL
) {
1220 __ceph_build_xattrs_blob(ci
);
1221 xattr_blob
= ci
->i_xattrs
.blob
;
1222 xattr_version
= ci
->i_xattrs
.version
;
1225 spin_unlock(&ci
->i_ceph_lock
);
1227 ret
= send_cap_msg(session
, ceph_vino(inode
).ino
, cap_id
,
1228 op
, keep
, want
, flushing
, seq
, flush_tid
, issue_seq
, mseq
,
1229 size
, max_size
, &mtime
, &atime
, time_warp_seq
,
1230 uid
, gid
, mode
, xattr_version
, xattr_blob
,
1233 dout("error sending cap msg, must requeue %p\n", inode
);
1238 wake_up_all(&ci
->i_cap_wq
);
1244 * When a snapshot is taken, clients accumulate dirty metadata on
1245 * inodes with capabilities in ceph_cap_snaps to describe the file
1246 * state at the time the snapshot was taken. This must be flushed
1247 * asynchronously back to the MDS once sync writes complete and dirty
1248 * data is written out.
1250 * Unless @again is true, skip cap_snaps that were already sent to
1251 * the MDS (i.e., during this session).
1253 * Called under i_ceph_lock. Takes s_mutex as needed.
1255 void __ceph_flush_snaps(struct ceph_inode_info
*ci
,
1256 struct ceph_mds_session
**psession
,
1258 __releases(ci
->i_ceph_lock
)
1259 __acquires(ci
->i_ceph_lock
)
1261 struct inode
*inode
= &ci
->vfs_inode
;
1263 struct ceph_cap_snap
*capsnap
;
1265 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1266 struct ceph_mds_session
*session
= NULL
; /* if session != NULL, we hold
1268 u64 next_follows
= 0; /* keep track of how far we've gotten through the
1269 i_cap_snaps list, and skip these entries next time
1270 around to avoid an infinite loop */
1273 session
= *psession
;
1275 dout("__flush_snaps %p\n", inode
);
1277 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
1278 /* avoid an infiniute loop after retry */
1279 if (capsnap
->follows
< next_follows
)
1282 * we need to wait for sync writes to complete and for dirty
1283 * pages to be written out.
1285 if (capsnap
->dirty_pages
|| capsnap
->writing
)
1289 * if cap writeback already occurred, we should have dropped
1290 * the capsnap in ceph_put_wrbuffer_cap_refs.
1292 BUG_ON(capsnap
->dirty
== 0);
1294 /* pick mds, take s_mutex */
1295 if (ci
->i_auth_cap
== NULL
) {
1296 dout("no auth cap (migrating?), doing nothing\n");
1300 /* only flush each capsnap once */
1301 if (!again
&& !list_empty(&capsnap
->flushing_item
)) {
1302 dout("already flushed %p, skipping\n", capsnap
);
1306 mds
= ci
->i_auth_cap
->session
->s_mds
;
1307 mseq
= ci
->i_auth_cap
->mseq
;
1309 if (session
&& session
->s_mds
!= mds
) {
1310 dout("oops, wrong session %p mutex\n", session
);
1311 mutex_unlock(&session
->s_mutex
);
1312 ceph_put_mds_session(session
);
1316 spin_unlock(&ci
->i_ceph_lock
);
1317 mutex_lock(&mdsc
->mutex
);
1318 session
= __ceph_lookup_mds_session(mdsc
, mds
);
1319 mutex_unlock(&mdsc
->mutex
);
1321 dout("inverting session/ino locks on %p\n",
1323 mutex_lock(&session
->s_mutex
);
1326 * if session == NULL, we raced against a cap
1327 * deletion or migration. retry, and we'll
1328 * get a better @mds value next time.
1330 spin_lock(&ci
->i_ceph_lock
);
1334 capsnap
->flush_tid
= ++ci
->i_cap_flush_last_tid
;
1335 atomic_inc(&capsnap
->nref
);
1336 if (!list_empty(&capsnap
->flushing_item
))
1337 list_del_init(&capsnap
->flushing_item
);
1338 list_add_tail(&capsnap
->flushing_item
,
1339 &session
->s_cap_snaps_flushing
);
1340 spin_unlock(&ci
->i_ceph_lock
);
1342 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1343 inode
, capsnap
, capsnap
->follows
, capsnap
->flush_tid
);
1344 send_cap_msg(session
, ceph_vino(inode
).ino
, 0,
1345 CEPH_CAP_OP_FLUSHSNAP
, capsnap
->issued
, 0,
1346 capsnap
->dirty
, 0, capsnap
->flush_tid
, 0, mseq
,
1348 &capsnap
->mtime
, &capsnap
->atime
,
1349 capsnap
->time_warp_seq
,
1350 capsnap
->uid
, capsnap
->gid
, capsnap
->mode
,
1351 capsnap
->xattr_version
, capsnap
->xattr_blob
,
1354 next_follows
= capsnap
->follows
+ 1;
1355 ceph_put_cap_snap(capsnap
);
1357 spin_lock(&ci
->i_ceph_lock
);
1361 /* we flushed them all; remove this inode from the queue */
1362 spin_lock(&mdsc
->snap_flush_lock
);
1363 list_del_init(&ci
->i_snap_flush_item
);
1364 spin_unlock(&mdsc
->snap_flush_lock
);
1368 *psession
= session
;
1370 mutex_unlock(&session
->s_mutex
);
1371 ceph_put_mds_session(session
);
1375 static void ceph_flush_snaps(struct ceph_inode_info
*ci
)
1377 spin_lock(&ci
->i_ceph_lock
);
1378 __ceph_flush_snaps(ci
, NULL
, 0);
1379 spin_unlock(&ci
->i_ceph_lock
);
1383 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1384 * Caller is then responsible for calling __mark_inode_dirty with the
1385 * returned flags value.
1387 int __ceph_mark_dirty_caps(struct ceph_inode_info
*ci
, int mask
)
1389 struct ceph_mds_client
*mdsc
=
1390 ceph_sb_to_client(ci
->vfs_inode
.i_sb
)->mdsc
;
1391 struct inode
*inode
= &ci
->vfs_inode
;
1392 int was
= ci
->i_dirty_caps
;
1395 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci
->vfs_inode
,
1396 ceph_cap_string(mask
), ceph_cap_string(was
),
1397 ceph_cap_string(was
| mask
));
1398 ci
->i_dirty_caps
|= mask
;
1400 if (!ci
->i_head_snapc
)
1401 ci
->i_head_snapc
= ceph_get_snap_context(
1402 ci
->i_snap_realm
->cached_context
);
1403 dout(" inode %p now dirty snapc %p auth cap %p\n",
1404 &ci
->vfs_inode
, ci
->i_head_snapc
, ci
->i_auth_cap
);
1405 WARN_ON(!ci
->i_auth_cap
);
1406 BUG_ON(!list_empty(&ci
->i_dirty_item
));
1407 spin_lock(&mdsc
->cap_dirty_lock
);
1408 list_add(&ci
->i_dirty_item
, &mdsc
->cap_dirty
);
1409 spin_unlock(&mdsc
->cap_dirty_lock
);
1410 if (ci
->i_flushing_caps
== 0) {
1412 dirty
|= I_DIRTY_SYNC
;
1415 BUG_ON(list_empty(&ci
->i_dirty_item
));
1416 if (((was
| ci
->i_flushing_caps
) & CEPH_CAP_FILE_BUFFER
) &&
1417 (mask
& CEPH_CAP_FILE_BUFFER
))
1418 dirty
|= I_DIRTY_DATASYNC
;
1419 __cap_delay_requeue(mdsc
, ci
);
1424 * Add dirty inode to the flushing list. Assigned a seq number so we
1425 * can wait for caps to flush without starving.
1427 * Called under i_ceph_lock.
1429 static int __mark_caps_flushing(struct inode
*inode
,
1430 struct ceph_mds_session
*session
)
1432 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1433 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1436 BUG_ON(ci
->i_dirty_caps
== 0);
1437 BUG_ON(list_empty(&ci
->i_dirty_item
));
1439 flushing
= ci
->i_dirty_caps
;
1440 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1441 ceph_cap_string(flushing
),
1442 ceph_cap_string(ci
->i_flushing_caps
),
1443 ceph_cap_string(ci
->i_flushing_caps
| flushing
));
1444 ci
->i_flushing_caps
|= flushing
;
1445 ci
->i_dirty_caps
= 0;
1446 dout(" inode %p now !dirty\n", inode
);
1448 spin_lock(&mdsc
->cap_dirty_lock
);
1449 list_del_init(&ci
->i_dirty_item
);
1451 ci
->i_cap_flush_seq
= ++mdsc
->cap_flush_seq
;
1452 if (list_empty(&ci
->i_flushing_item
)) {
1453 list_add_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1454 mdsc
->num_cap_flushing
++;
1455 dout(" inode %p now flushing seq %lld\n", inode
,
1456 ci
->i_cap_flush_seq
);
1458 list_move_tail(&ci
->i_flushing_item
, &session
->s_cap_flushing
);
1459 dout(" inode %p now flushing (more) seq %lld\n", inode
,
1460 ci
->i_cap_flush_seq
);
1462 spin_unlock(&mdsc
->cap_dirty_lock
);
1468 * try to invalidate mapping pages without blocking.
1470 static int try_nonblocking_invalidate(struct inode
*inode
)
1472 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1473 u32 invalidating_gen
= ci
->i_rdcache_gen
;
1475 spin_unlock(&ci
->i_ceph_lock
);
1476 invalidate_mapping_pages(&inode
->i_data
, 0, -1);
1477 spin_lock(&ci
->i_ceph_lock
);
1479 if (inode
->i_data
.nrpages
== 0 &&
1480 invalidating_gen
== ci
->i_rdcache_gen
) {
1482 dout("try_nonblocking_invalidate %p success\n", inode
);
1483 /* save any racing async invalidate some trouble */
1484 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
- 1;
1487 dout("try_nonblocking_invalidate %p failed\n", inode
);
1492 * Swiss army knife function to examine currently used and wanted
1493 * versus held caps. Release, flush, ack revoked caps to mds as
1496 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1497 * cap release further.
1498 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1499 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1502 void ceph_check_caps(struct ceph_inode_info
*ci
, int flags
,
1503 struct ceph_mds_session
*session
)
1505 struct ceph_fs_client
*fsc
= ceph_inode_to_client(&ci
->vfs_inode
);
1506 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1507 struct inode
*inode
= &ci
->vfs_inode
;
1508 struct ceph_cap
*cap
;
1509 int file_wanted
, used
, cap_used
;
1510 int took_snap_rwsem
= 0; /* true if mdsc->snap_rwsem held */
1511 int issued
, implemented
, want
, retain
, revoking
, flushing
= 0;
1512 int mds
= -1; /* keep track of how far we've gone through i_caps list
1513 to avoid an infinite loop on retry */
1515 int tried_invalidate
= 0;
1516 int delayed
= 0, sent
= 0, force_requeue
= 0, num
;
1517 int queue_invalidate
= 0;
1518 int is_delayed
= flags
& CHECK_CAPS_NODELAY
;
1520 /* if we are unmounting, flush any unused caps immediately. */
1524 spin_lock(&ci
->i_ceph_lock
);
1526 if (ci
->i_ceph_flags
& CEPH_I_FLUSH
)
1527 flags
|= CHECK_CAPS_FLUSH
;
1529 /* flush snaps first time around only */
1530 if (!list_empty(&ci
->i_cap_snaps
))
1531 __ceph_flush_snaps(ci
, &session
, 0);
1534 spin_lock(&ci
->i_ceph_lock
);
1536 file_wanted
= __ceph_caps_file_wanted(ci
);
1537 used
= __ceph_caps_used(ci
);
1538 want
= file_wanted
| used
;
1539 issued
= __ceph_caps_issued(ci
, &implemented
);
1540 revoking
= implemented
& ~issued
;
1542 retain
= want
| CEPH_CAP_PIN
;
1543 if (!mdsc
->stopping
&& inode
->i_nlink
> 0) {
1545 retain
|= CEPH_CAP_ANY
; /* be greedy */
1547 retain
|= CEPH_CAP_ANY_SHARED
;
1549 * keep RD only if we didn't have the file open RW,
1550 * because then the mds would revoke it anyway to
1551 * journal max_size=0.
1553 if (ci
->i_max_size
== 0)
1554 retain
|= CEPH_CAP_ANY_RD
;
1558 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1559 " issued %s revoking %s retain %s %s%s%s\n", inode
,
1560 ceph_cap_string(file_wanted
),
1561 ceph_cap_string(used
), ceph_cap_string(ci
->i_dirty_caps
),
1562 ceph_cap_string(ci
->i_flushing_caps
),
1563 ceph_cap_string(issued
), ceph_cap_string(revoking
),
1564 ceph_cap_string(retain
),
1565 (flags
& CHECK_CAPS_AUTHONLY
) ? " AUTHONLY" : "",
1566 (flags
& CHECK_CAPS_NODELAY
) ? " NODELAY" : "",
1567 (flags
& CHECK_CAPS_FLUSH
) ? " FLUSH" : "");
1570 * If we no longer need to hold onto old our caps, and we may
1571 * have cached pages, but don't want them, then try to invalidate.
1572 * If we fail, it's because pages are locked.... try again later.
1574 if ((!is_delayed
|| mdsc
->stopping
) &&
1575 ci
->i_wrbuffer_ref
== 0 && /* no dirty pages... */
1576 inode
->i_data
.nrpages
&& /* have cached pages */
1577 (file_wanted
== 0 || /* no open files */
1578 (revoking
& (CEPH_CAP_FILE_CACHE
|
1579 CEPH_CAP_FILE_LAZYIO
))) && /* or revoking cache */
1580 !tried_invalidate
) {
1581 dout("check_caps trying to invalidate on %p\n", inode
);
1582 if (try_nonblocking_invalidate(inode
) < 0) {
1583 if (revoking
& (CEPH_CAP_FILE_CACHE
|
1584 CEPH_CAP_FILE_LAZYIO
)) {
1585 dout("check_caps queuing invalidate\n");
1586 queue_invalidate
= 1;
1587 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
1589 dout("check_caps failed to invalidate pages\n");
1590 /* we failed to invalidate pages. check these
1591 caps again later. */
1593 __cap_set_timeouts(mdsc
, ci
);
1596 tried_invalidate
= 1;
1601 for (p
= rb_first(&ci
->i_caps
); p
; p
= rb_next(p
)) {
1602 cap
= rb_entry(p
, struct ceph_cap
, ci_node
);
1605 /* avoid looping forever */
1606 if (mds
>= cap
->mds
||
1607 ((flags
& CHECK_CAPS_AUTHONLY
) && cap
!= ci
->i_auth_cap
))
1610 /* NOTE: no side-effects allowed, until we take s_mutex */
1613 if (ci
->i_auth_cap
&& cap
!= ci
->i_auth_cap
)
1614 cap_used
&= ~ci
->i_auth_cap
->issued
;
1616 revoking
= cap
->implemented
& ~cap
->issued
;
1617 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1618 cap
->mds
, cap
, ceph_cap_string(cap
->issued
),
1619 ceph_cap_string(cap_used
),
1620 ceph_cap_string(cap
->implemented
),
1621 ceph_cap_string(revoking
));
1623 if (cap
== ci
->i_auth_cap
&&
1624 (cap
->issued
& CEPH_CAP_FILE_WR
)) {
1625 /* request larger max_size from MDS? */
1626 if (ci
->i_wanted_max_size
> ci
->i_max_size
&&
1627 ci
->i_wanted_max_size
> ci
->i_requested_max_size
) {
1628 dout("requesting new max_size\n");
1632 /* approaching file_max? */
1633 if ((inode
->i_size
<< 1) >= ci
->i_max_size
&&
1634 (ci
->i_reported_size
<< 1) < ci
->i_max_size
) {
1635 dout("i_size approaching max_size\n");
1639 /* flush anything dirty? */
1640 if (cap
== ci
->i_auth_cap
&& (flags
& CHECK_CAPS_FLUSH
) &&
1642 dout("flushing dirty caps\n");
1646 /* completed revocation? going down and there are no caps? */
1647 if (revoking
&& (revoking
& cap_used
) == 0) {
1648 dout("completed revocation of %s\n",
1649 ceph_cap_string(cap
->implemented
& ~cap
->issued
));
1653 /* want more caps from mds? */
1654 if (want
& ~(cap
->mds_wanted
| cap
->issued
))
1657 /* things we might delay */
1658 if ((cap
->issued
& ~retain
) == 0 &&
1659 cap
->mds_wanted
== want
)
1660 continue; /* nope, all good */
1666 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0 &&
1667 time_before(jiffies
, ci
->i_hold_caps_max
)) {
1668 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1669 ceph_cap_string(cap
->issued
),
1670 ceph_cap_string(cap
->issued
& retain
),
1671 ceph_cap_string(cap
->mds_wanted
),
1672 ceph_cap_string(want
));
1678 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1679 dout(" skipping %p I_NOFLUSH set\n", inode
);
1683 if (session
&& session
!= cap
->session
) {
1684 dout("oops, wrong session %p mutex\n", session
);
1685 mutex_unlock(&session
->s_mutex
);
1689 session
= cap
->session
;
1690 if (mutex_trylock(&session
->s_mutex
) == 0) {
1691 dout("inverting session/ino locks on %p\n",
1693 spin_unlock(&ci
->i_ceph_lock
);
1694 if (took_snap_rwsem
) {
1695 up_read(&mdsc
->snap_rwsem
);
1696 took_snap_rwsem
= 0;
1698 mutex_lock(&session
->s_mutex
);
1702 /* take snap_rwsem after session mutex */
1703 if (!took_snap_rwsem
) {
1704 if (down_read_trylock(&mdsc
->snap_rwsem
) == 0) {
1705 dout("inverting snap/in locks on %p\n",
1707 spin_unlock(&ci
->i_ceph_lock
);
1708 down_read(&mdsc
->snap_rwsem
);
1709 took_snap_rwsem
= 1;
1712 took_snap_rwsem
= 1;
1715 if (cap
== ci
->i_auth_cap
&& ci
->i_dirty_caps
)
1716 flushing
= __mark_caps_flushing(inode
, session
);
1720 mds
= cap
->mds
; /* remember mds, so we don't repeat */
1723 /* __send_cap drops i_ceph_lock */
1724 delayed
+= __send_cap(mdsc
, cap
, CEPH_CAP_OP_UPDATE
, cap_used
,
1725 want
, retain
, flushing
, NULL
);
1726 goto retry
; /* retake i_ceph_lock and restart our cap scan. */
1730 * Reschedule delayed caps release if we delayed anything,
1733 if (delayed
&& is_delayed
)
1734 force_requeue
= 1; /* __send_cap delayed release; requeue */
1735 if (!delayed
&& !is_delayed
)
1736 __cap_delay_cancel(mdsc
, ci
);
1737 else if (!is_delayed
|| force_requeue
)
1738 __cap_delay_requeue(mdsc
, ci
);
1740 spin_unlock(&ci
->i_ceph_lock
);
1742 if (queue_invalidate
)
1743 ceph_queue_invalidate(inode
);
1746 mutex_unlock(&session
->s_mutex
);
1747 if (took_snap_rwsem
)
1748 up_read(&mdsc
->snap_rwsem
);
1752 * Try to flush dirty caps back to the auth mds.
1754 static int try_flush_caps(struct inode
*inode
, unsigned *flush_tid
)
1756 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1757 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1759 struct ceph_mds_session
*session
= NULL
;
1762 spin_lock(&ci
->i_ceph_lock
);
1763 if (ci
->i_ceph_flags
& CEPH_I_NOFLUSH
) {
1764 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode
);
1767 if (ci
->i_dirty_caps
&& ci
->i_auth_cap
) {
1768 struct ceph_cap
*cap
= ci
->i_auth_cap
;
1769 int used
= __ceph_caps_used(ci
);
1770 int want
= __ceph_caps_wanted(ci
);
1773 if (!session
|| session
!= cap
->session
) {
1774 spin_unlock(&ci
->i_ceph_lock
);
1776 mutex_unlock(&session
->s_mutex
);
1777 session
= cap
->session
;
1778 mutex_lock(&session
->s_mutex
);
1781 if (cap
->session
->s_state
< CEPH_MDS_SESSION_OPEN
)
1784 flushing
= __mark_caps_flushing(inode
, session
);
1786 /* __send_cap drops i_ceph_lock */
1787 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
, used
, want
,
1788 cap
->issued
| cap
->implemented
, flushing
,
1793 spin_lock(&ci
->i_ceph_lock
);
1794 __cap_delay_requeue(mdsc
, ci
);
1797 spin_unlock(&ci
->i_ceph_lock
);
1800 mutex_unlock(&session
->s_mutex
);
1805 * Return true if we've flushed caps through the given flush_tid.
1807 static int caps_are_flushed(struct inode
*inode
, unsigned tid
)
1809 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1812 spin_lock(&ci
->i_ceph_lock
);
1813 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
1814 if ((ci
->i_flushing_caps
& (1 << i
)) &&
1815 ci
->i_cap_flush_tid
[i
] <= tid
) {
1816 /* still flushing this bit */
1820 spin_unlock(&ci
->i_ceph_lock
);
1825 * Wait on any unsafe replies for the given inode. First wait on the
1826 * newest request, and make that the upper bound. Then, if there are
1827 * more requests, keep waiting on the oldest as long as it is still older
1828 * than the original request.
1830 static void sync_write_wait(struct inode
*inode
)
1832 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1833 struct list_head
*head
= &ci
->i_unsafe_writes
;
1834 struct ceph_osd_request
*req
;
1837 spin_lock(&ci
->i_unsafe_lock
);
1838 if (list_empty(head
))
1841 /* set upper bound as _last_ entry in chain */
1842 req
= list_entry(head
->prev
, struct ceph_osd_request
,
1844 last_tid
= req
->r_tid
;
1847 ceph_osdc_get_request(req
);
1848 spin_unlock(&ci
->i_unsafe_lock
);
1849 dout("sync_write_wait on tid %llu (until %llu)\n",
1850 req
->r_tid
, last_tid
);
1851 wait_for_completion(&req
->r_safe_completion
);
1852 spin_lock(&ci
->i_unsafe_lock
);
1853 ceph_osdc_put_request(req
);
1856 * from here on look at first entry in chain, since we
1857 * only want to wait for anything older than last_tid
1859 if (list_empty(head
))
1861 req
= list_entry(head
->next
, struct ceph_osd_request
,
1863 } while (req
->r_tid
< last_tid
);
1865 spin_unlock(&ci
->i_unsafe_lock
);
1868 int ceph_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
1870 struct inode
*inode
= file
->f_mapping
->host
;
1871 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1876 dout("fsync %p%s\n", inode
, datasync
? " datasync" : "");
1877 sync_write_wait(inode
);
1879 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
1882 mutex_lock(&inode
->i_mutex
);
1884 dirty
= try_flush_caps(inode
, &flush_tid
);
1885 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty
));
1888 * only wait on non-file metadata writeback (the mds
1889 * can recover size and mtime, so we don't need to
1892 if (!datasync
&& (dirty
& ~CEPH_CAP_ANY_FILE_WR
)) {
1893 dout("fsync waiting for flush_tid %u\n", flush_tid
);
1894 ret
= wait_event_interruptible(ci
->i_cap_wq
,
1895 caps_are_flushed(inode
, flush_tid
));
1898 dout("fsync %p%s done\n", inode
, datasync
? " datasync" : "");
1899 mutex_unlock(&inode
->i_mutex
);
1904 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1905 * queue inode for flush but don't do so immediately, because we can
1906 * get by with fewer MDS messages if we wait for data writeback to
1909 int ceph_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1911 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1915 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
1917 dout("write_inode %p wait=%d\n", inode
, wait
);
1919 dirty
= try_flush_caps(inode
, &flush_tid
);
1921 err
= wait_event_interruptible(ci
->i_cap_wq
,
1922 caps_are_flushed(inode
, flush_tid
));
1924 struct ceph_mds_client
*mdsc
=
1925 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1927 spin_lock(&ci
->i_ceph_lock
);
1928 if (__ceph_caps_dirty(ci
))
1929 __cap_delay_requeue_front(mdsc
, ci
);
1930 spin_unlock(&ci
->i_ceph_lock
);
1936 * After a recovering MDS goes active, we need to resend any caps
1939 * Caller holds session->s_mutex.
1941 static void kick_flushing_capsnaps(struct ceph_mds_client
*mdsc
,
1942 struct ceph_mds_session
*session
)
1944 struct ceph_cap_snap
*capsnap
;
1946 dout("kick_flushing_capsnaps mds%d\n", session
->s_mds
);
1947 list_for_each_entry(capsnap
, &session
->s_cap_snaps_flushing
,
1949 struct ceph_inode_info
*ci
= capsnap
->ci
;
1950 struct inode
*inode
= &ci
->vfs_inode
;
1951 struct ceph_cap
*cap
;
1953 spin_lock(&ci
->i_ceph_lock
);
1954 cap
= ci
->i_auth_cap
;
1955 if (cap
&& cap
->session
== session
) {
1956 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode
,
1958 __ceph_flush_snaps(ci
, &session
, 1);
1960 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1961 cap
, session
->s_mds
);
1963 spin_unlock(&ci
->i_ceph_lock
);
1967 void ceph_kick_flushing_caps(struct ceph_mds_client
*mdsc
,
1968 struct ceph_mds_session
*session
)
1970 struct ceph_inode_info
*ci
;
1972 kick_flushing_capsnaps(mdsc
, session
);
1974 dout("kick_flushing_caps mds%d\n", session
->s_mds
);
1975 list_for_each_entry(ci
, &session
->s_cap_flushing
, i_flushing_item
) {
1976 struct inode
*inode
= &ci
->vfs_inode
;
1977 struct ceph_cap
*cap
;
1980 spin_lock(&ci
->i_ceph_lock
);
1981 cap
= ci
->i_auth_cap
;
1982 if (cap
&& cap
->session
== session
) {
1983 dout("kick_flushing_caps %p cap %p %s\n", inode
,
1984 cap
, ceph_cap_string(ci
->i_flushing_caps
));
1985 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
1986 __ceph_caps_used(ci
),
1987 __ceph_caps_wanted(ci
),
1988 cap
->issued
| cap
->implemented
,
1989 ci
->i_flushing_caps
, NULL
);
1991 spin_lock(&ci
->i_ceph_lock
);
1992 __cap_delay_requeue(mdsc
, ci
);
1993 spin_unlock(&ci
->i_ceph_lock
);
1996 pr_err("%p auth cap %p not mds%d ???\n", inode
,
1997 cap
, session
->s_mds
);
1998 spin_unlock(&ci
->i_ceph_lock
);
2003 static void kick_flushing_inode_caps(struct ceph_mds_client
*mdsc
,
2004 struct ceph_mds_session
*session
,
2005 struct inode
*inode
)
2007 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2008 struct ceph_cap
*cap
;
2011 spin_lock(&ci
->i_ceph_lock
);
2012 cap
= ci
->i_auth_cap
;
2013 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode
,
2014 ceph_cap_string(ci
->i_flushing_caps
), ci
->i_cap_flush_seq
);
2016 __ceph_flush_snaps(ci
, &session
, 1);
2018 if (ci
->i_flushing_caps
) {
2019 spin_lock(&mdsc
->cap_dirty_lock
);
2020 list_move_tail(&ci
->i_flushing_item
,
2021 &cap
->session
->s_cap_flushing
);
2022 spin_unlock(&mdsc
->cap_dirty_lock
);
2024 delayed
= __send_cap(mdsc
, cap
, CEPH_CAP_OP_FLUSH
,
2025 __ceph_caps_used(ci
),
2026 __ceph_caps_wanted(ci
),
2027 cap
->issued
| cap
->implemented
,
2028 ci
->i_flushing_caps
, NULL
);
2030 spin_lock(&ci
->i_ceph_lock
);
2031 __cap_delay_requeue(mdsc
, ci
);
2032 spin_unlock(&ci
->i_ceph_lock
);
2035 spin_unlock(&ci
->i_ceph_lock
);
2041 * Take references to capabilities we hold, so that we don't release
2042 * them to the MDS prematurely.
2044 * Protected by i_ceph_lock.
2046 static void __take_cap_refs(struct ceph_inode_info
*ci
, int got
)
2048 if (got
& CEPH_CAP_PIN
)
2050 if (got
& CEPH_CAP_FILE_RD
)
2052 if (got
& CEPH_CAP_FILE_CACHE
)
2053 ci
->i_rdcache_ref
++;
2054 if (got
& CEPH_CAP_FILE_WR
)
2056 if (got
& CEPH_CAP_FILE_BUFFER
) {
2057 if (ci
->i_wb_ref
== 0)
2058 ihold(&ci
->vfs_inode
);
2060 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2061 &ci
->vfs_inode
, ci
->i_wb_ref
-1, ci
->i_wb_ref
);
2066 * Try to grab cap references. Specify those refs we @want, and the
2067 * minimal set we @need. Also include the larger offset we are writing
2068 * to (when applicable), and check against max_size here as well.
2069 * Note that caller is responsible for ensuring max_size increases are
2070 * requested from the MDS.
2072 static int try_get_cap_refs(struct ceph_inode_info
*ci
, int need
, int want
,
2073 int *got
, loff_t endoff
, int *check_max
, int *err
)
2075 struct inode
*inode
= &ci
->vfs_inode
;
2077 int have
, implemented
;
2080 dout("get_cap_refs %p need %s want %s\n", inode
,
2081 ceph_cap_string(need
), ceph_cap_string(want
));
2082 spin_lock(&ci
->i_ceph_lock
);
2084 /* make sure file is actually open */
2085 file_wanted
= __ceph_caps_file_wanted(ci
);
2086 if ((file_wanted
& need
) == 0) {
2087 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2088 ceph_cap_string(need
), ceph_cap_string(file_wanted
));
2094 /* finish pending truncate */
2095 while (ci
->i_truncate_pending
) {
2096 spin_unlock(&ci
->i_ceph_lock
);
2097 __ceph_do_pending_vmtruncate(inode
);
2098 spin_lock(&ci
->i_ceph_lock
);
2101 have
= __ceph_caps_issued(ci
, &implemented
);
2103 if (have
& need
& CEPH_CAP_FILE_WR
) {
2104 if (endoff
>= 0 && endoff
> (loff_t
)ci
->i_max_size
) {
2105 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2106 inode
, endoff
, ci
->i_max_size
);
2107 if (endoff
> ci
->i_requested_max_size
) {
2114 * If a sync write is in progress, we must wait, so that we
2115 * can get a final snapshot value for size+mtime.
2117 if (__ceph_have_pending_cap_snap(ci
)) {
2118 dout("get_cap_refs %p cap_snap_pending\n", inode
);
2123 if ((have
& need
) == need
) {
2125 * Look at (implemented & ~have & not) so that we keep waiting
2126 * on transition from wanted -> needed caps. This is needed
2127 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2128 * going before a prior buffered writeback happens.
2130 int not = want
& ~(have
& need
);
2131 int revoking
= implemented
& ~have
;
2132 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2133 inode
, ceph_cap_string(have
), ceph_cap_string(not),
2134 ceph_cap_string(revoking
));
2135 if ((revoking
& not) == 0) {
2136 *got
= need
| (have
& want
);
2137 __take_cap_refs(ci
, *got
);
2141 dout("get_cap_refs %p have %s needed %s\n", inode
,
2142 ceph_cap_string(have
), ceph_cap_string(need
));
2145 spin_unlock(&ci
->i_ceph_lock
);
2146 dout("get_cap_refs %p ret %d got %s\n", inode
,
2147 ret
, ceph_cap_string(*got
));
2152 * Check the offset we are writing up to against our current
2153 * max_size. If necessary, tell the MDS we want to write to
2156 static void check_max_size(struct inode
*inode
, loff_t endoff
)
2158 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2161 /* do we need to explicitly request a larger max_size? */
2162 spin_lock(&ci
->i_ceph_lock
);
2163 if (endoff
>= ci
->i_max_size
&& endoff
> ci
->i_wanted_max_size
) {
2164 dout("write %p at large endoff %llu, req max_size\n",
2166 ci
->i_wanted_max_size
= endoff
;
2168 /* duplicate ceph_check_caps()'s logic */
2169 if (ci
->i_auth_cap
&&
2170 (ci
->i_auth_cap
->issued
& CEPH_CAP_FILE_WR
) &&
2171 ci
->i_wanted_max_size
> ci
->i_max_size
&&
2172 ci
->i_wanted_max_size
> ci
->i_requested_max_size
)
2174 spin_unlock(&ci
->i_ceph_lock
);
2176 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2180 * Wait for caps, and take cap references. If we can't get a WR cap
2181 * due to a small max_size, make sure we check_max_size (and possibly
2182 * ask the mds) so we don't get hung up indefinitely.
2184 int ceph_get_caps(struct ceph_inode_info
*ci
, int need
, int want
, int *got
,
2187 int check_max
, ret
, err
;
2191 check_max_size(&ci
->vfs_inode
, endoff
);
2194 ret
= wait_event_interruptible(ci
->i_cap_wq
,
2195 try_get_cap_refs(ci
, need
, want
,
2206 * Take cap refs. Caller must already know we hold at least one ref
2207 * on the caps in question or we don't know this is safe.
2209 void ceph_get_cap_refs(struct ceph_inode_info
*ci
, int caps
)
2211 spin_lock(&ci
->i_ceph_lock
);
2212 __take_cap_refs(ci
, caps
);
2213 spin_unlock(&ci
->i_ceph_lock
);
2219 * If we released the last ref on any given cap, call ceph_check_caps
2220 * to release (or schedule a release).
2222 * If we are releasing a WR cap (from a sync write), finalize any affected
2223 * cap_snap, and wake up any waiters.
2225 void ceph_put_cap_refs(struct ceph_inode_info
*ci
, int had
)
2227 struct inode
*inode
= &ci
->vfs_inode
;
2228 int last
= 0, put
= 0, flushsnaps
= 0, wake
= 0;
2229 struct ceph_cap_snap
*capsnap
;
2231 spin_lock(&ci
->i_ceph_lock
);
2232 if (had
& CEPH_CAP_PIN
)
2234 if (had
& CEPH_CAP_FILE_RD
)
2235 if (--ci
->i_rd_ref
== 0)
2237 if (had
& CEPH_CAP_FILE_CACHE
)
2238 if (--ci
->i_rdcache_ref
== 0)
2240 if (had
& CEPH_CAP_FILE_BUFFER
) {
2241 if (--ci
->i_wb_ref
== 0) {
2245 dout("put_cap_refs %p wb %d -> %d (?)\n",
2246 inode
, ci
->i_wb_ref
+1, ci
->i_wb_ref
);
2248 if (had
& CEPH_CAP_FILE_WR
)
2249 if (--ci
->i_wr_ref
== 0) {
2251 if (!list_empty(&ci
->i_cap_snaps
)) {
2252 capsnap
= list_first_entry(&ci
->i_cap_snaps
,
2253 struct ceph_cap_snap
,
2255 if (capsnap
->writing
) {
2256 capsnap
->writing
= 0;
2258 __ceph_finish_cap_snap(ci
,
2264 spin_unlock(&ci
->i_ceph_lock
);
2266 dout("put_cap_refs %p had %s%s%s\n", inode
, ceph_cap_string(had
),
2267 last
? " last" : "", put
? " put" : "");
2269 if (last
&& !flushsnaps
)
2270 ceph_check_caps(ci
, 0, NULL
);
2271 else if (flushsnaps
)
2272 ceph_flush_snaps(ci
);
2274 wake_up_all(&ci
->i_cap_wq
);
2280 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2281 * context. Adjust per-snap dirty page accounting as appropriate.
2282 * Once all dirty data for a cap_snap is flushed, flush snapped file
2283 * metadata back to the MDS. If we dropped the last ref, call
2286 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info
*ci
, int nr
,
2287 struct ceph_snap_context
*snapc
)
2289 struct inode
*inode
= &ci
->vfs_inode
;
2291 int complete_capsnap
= 0;
2292 int drop_capsnap
= 0;
2294 struct ceph_cap_snap
*capsnap
= NULL
;
2296 spin_lock(&ci
->i_ceph_lock
);
2297 ci
->i_wrbuffer_ref
-= nr
;
2298 last
= !ci
->i_wrbuffer_ref
;
2300 if (ci
->i_head_snapc
== snapc
) {
2301 ci
->i_wrbuffer_ref_head
-= nr
;
2302 if (ci
->i_wrbuffer_ref_head
== 0 &&
2303 ci
->i_dirty_caps
== 0 && ci
->i_flushing_caps
== 0) {
2304 BUG_ON(!ci
->i_head_snapc
);
2305 ceph_put_snap_context(ci
->i_head_snapc
);
2306 ci
->i_head_snapc
= NULL
;
2308 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2310 ci
->i_wrbuffer_ref
+nr
, ci
->i_wrbuffer_ref_head
+nr
,
2311 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
2312 last
? " LAST" : "");
2314 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2315 if (capsnap
->context
== snapc
) {
2321 capsnap
->dirty_pages
-= nr
;
2322 if (capsnap
->dirty_pages
== 0) {
2323 complete_capsnap
= 1;
2324 if (capsnap
->dirty
== 0)
2325 /* cap writeback completed before we created
2326 * the cap_snap; no FLUSHSNAP is needed */
2329 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2330 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2331 inode
, capsnap
, capsnap
->context
->seq
,
2332 ci
->i_wrbuffer_ref
+nr
, capsnap
->dirty_pages
+ nr
,
2333 ci
->i_wrbuffer_ref
, capsnap
->dirty_pages
,
2334 last
? " (wrbuffer last)" : "",
2335 complete_capsnap
? " (complete capsnap)" : "",
2336 drop_capsnap
? " (drop capsnap)" : "");
2338 ceph_put_snap_context(capsnap
->context
);
2339 list_del(&capsnap
->ci_item
);
2340 list_del(&capsnap
->flushing_item
);
2341 ceph_put_cap_snap(capsnap
);
2345 spin_unlock(&ci
->i_ceph_lock
);
2348 ceph_check_caps(ci
, CHECK_CAPS_AUTHONLY
, NULL
);
2350 } else if (complete_capsnap
) {
2351 ceph_flush_snaps(ci
);
2352 wake_up_all(&ci
->i_cap_wq
);
2359 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2361 static void invalidate_aliases(struct inode
*inode
)
2363 struct dentry
*dn
, *prev
= NULL
;
2365 dout("invalidate_aliases inode %p\n", inode
);
2366 d_prune_aliases(inode
);
2368 * For non-directory inode, d_find_alias() only returns
2369 * hashed dentry. After calling d_invalidate(), the
2370 * dentry becomes unhashed.
2372 * For directory inode, d_find_alias() can return
2373 * unhashed dentry. But directory inode should have
2374 * one alias at most.
2376 while ((dn
= d_find_alias(inode
))) {
2391 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2392 * actually be a revocation if it specifies a smaller cap set.)
2394 * caller holds s_mutex and i_ceph_lock, we drop both.
2398 * 1 - check_caps on auth cap only (writeback)
2399 * 2 - check_caps (ack revoke)
2401 static void handle_cap_grant(struct inode
*inode
, struct ceph_mds_caps
*grant
,
2402 struct ceph_mds_session
*session
,
2403 struct ceph_cap
*cap
,
2404 struct ceph_buffer
*xattr_buf
)
2405 __releases(ci
->i_ceph_lock
)
2407 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2408 int mds
= session
->s_mds
;
2409 int seq
= le32_to_cpu(grant
->seq
);
2410 int newcaps
= le32_to_cpu(grant
->caps
);
2411 int issued
, implemented
, used
, wanted
, dirty
;
2412 u64 size
= le64_to_cpu(grant
->size
);
2413 u64 max_size
= le64_to_cpu(grant
->max_size
);
2414 struct timespec mtime
, atime
, ctime
;
2418 int queue_invalidate
= 0;
2419 int deleted_inode
= 0;
2420 int queue_revalidate
= 0;
2422 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2423 inode
, cap
, mds
, seq
, ceph_cap_string(newcaps
));
2424 dout(" size %llu max_size %llu, i_size %llu\n", size
, max_size
,
2429 * auth mds of the inode changed. we received the cap export message,
2430 * but still haven't received the cap import message. handle_cap_export
2431 * updated the new auth MDS' cap.
2433 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
2434 * that was sent before the cap import message. So don't remove caps.
2436 if (ceph_seq_cmp(seq
, cap
->seq
) <= 0) {
2437 WARN_ON(cap
!= ci
->i_auth_cap
);
2438 WARN_ON(cap
->cap_id
!= le64_to_cpu(grant
->cap_id
));
2440 newcaps
|= cap
->issued
;
2444 * If CACHE is being revoked, and we have no dirty buffers,
2445 * try to invalidate (once). (If there are dirty buffers, we
2446 * will invalidate _after_ writeback.)
2448 if (((cap
->issued
& ~newcaps
) & CEPH_CAP_FILE_CACHE
) &&
2449 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2450 !ci
->i_wrbuffer_ref
) {
2451 if (try_nonblocking_invalidate(inode
)) {
2452 /* there were locked pages.. invalidate later
2453 in a separate thread. */
2454 if (ci
->i_rdcache_revoking
!= ci
->i_rdcache_gen
) {
2455 queue_invalidate
= 1;
2456 ci
->i_rdcache_revoking
= ci
->i_rdcache_gen
;
2460 ceph_fscache_invalidate(inode
);
2463 /* side effects now are allowed */
2465 issued
= __ceph_caps_issued(ci
, &implemented
);
2466 issued
|= implemented
| __ceph_caps_dirty(ci
);
2468 cap
->cap_gen
= session
->s_cap_gen
;
2471 __check_cap_issue(ci
, cap
, newcaps
);
2473 if ((issued
& CEPH_CAP_AUTH_EXCL
) == 0) {
2474 inode
->i_mode
= le32_to_cpu(grant
->mode
);
2475 inode
->i_uid
= make_kuid(&init_user_ns
, le32_to_cpu(grant
->uid
));
2476 inode
->i_gid
= make_kgid(&init_user_ns
, le32_to_cpu(grant
->gid
));
2477 dout("%p mode 0%o uid.gid %d.%d\n", inode
, inode
->i_mode
,
2478 from_kuid(&init_user_ns
, inode
->i_uid
),
2479 from_kgid(&init_user_ns
, inode
->i_gid
));
2482 if ((issued
& CEPH_CAP_LINK_EXCL
) == 0) {
2483 set_nlink(inode
, le32_to_cpu(grant
->nlink
));
2484 if (inode
->i_nlink
== 0 &&
2485 (newcaps
& (CEPH_CAP_LINK_SHARED
| CEPH_CAP_LINK_EXCL
)))
2489 if ((issued
& CEPH_CAP_XATTR_EXCL
) == 0 && grant
->xattr_len
) {
2490 int len
= le32_to_cpu(grant
->xattr_len
);
2491 u64 version
= le64_to_cpu(grant
->xattr_version
);
2493 if (version
> ci
->i_xattrs
.version
) {
2494 dout(" got new xattrs v%llu on %p len %d\n",
2495 version
, inode
, len
);
2496 if (ci
->i_xattrs
.blob
)
2497 ceph_buffer_put(ci
->i_xattrs
.blob
);
2498 ci
->i_xattrs
.blob
= ceph_buffer_get(xattr_buf
);
2499 ci
->i_xattrs
.version
= version
;
2500 ceph_forget_all_cached_acls(inode
);
2504 /* Do we need to revalidate our fscache cookie. Don't bother on the
2505 * first cache cap as we already validate at cookie creation time. */
2506 if ((issued
& CEPH_CAP_FILE_CACHE
) && ci
->i_rdcache_gen
> 1)
2507 queue_revalidate
= 1;
2509 /* size/ctime/mtime/atime? */
2510 ceph_fill_file_size(inode
, issued
,
2511 le32_to_cpu(grant
->truncate_seq
),
2512 le64_to_cpu(grant
->truncate_size
), size
);
2513 ceph_decode_timespec(&mtime
, &grant
->mtime
);
2514 ceph_decode_timespec(&atime
, &grant
->atime
);
2515 ceph_decode_timespec(&ctime
, &grant
->ctime
);
2516 ceph_fill_file_time(inode
, issued
,
2517 le32_to_cpu(grant
->time_warp_seq
), &ctime
, &mtime
,
2521 /* file layout may have changed */
2522 ci
->i_layout
= grant
->layout
;
2524 /* max size increase? */
2525 if (ci
->i_auth_cap
== cap
&& max_size
!= ci
->i_max_size
) {
2526 dout("max_size %lld -> %llu\n", ci
->i_max_size
, max_size
);
2527 ci
->i_max_size
= max_size
;
2528 if (max_size
>= ci
->i_wanted_max_size
) {
2529 ci
->i_wanted_max_size
= 0; /* reset */
2530 ci
->i_requested_max_size
= 0;
2535 /* check cap bits */
2536 wanted
= __ceph_caps_wanted(ci
);
2537 used
= __ceph_caps_used(ci
);
2538 dirty
= __ceph_caps_dirty(ci
);
2539 dout(" my wanted = %s, used = %s, dirty %s\n",
2540 ceph_cap_string(wanted
),
2541 ceph_cap_string(used
),
2542 ceph_cap_string(dirty
));
2543 if (wanted
!= le32_to_cpu(grant
->wanted
)) {
2544 dout("mds wanted %s -> %s\n",
2545 ceph_cap_string(le32_to_cpu(grant
->wanted
)),
2546 ceph_cap_string(wanted
));
2547 /* imported cap may not have correct mds_wanted */
2548 if (le32_to_cpu(grant
->op
) == CEPH_CAP_OP_IMPORT
)
2552 /* revocation, grant, or no-op? */
2553 if (cap
->issued
& ~newcaps
) {
2554 int revoking
= cap
->issued
& ~newcaps
;
2556 dout("revocation: %s -> %s (revoking %s)\n",
2557 ceph_cap_string(cap
->issued
),
2558 ceph_cap_string(newcaps
),
2559 ceph_cap_string(revoking
));
2560 if (revoking
& used
& CEPH_CAP_FILE_BUFFER
)
2561 writeback
= 1; /* initiate writeback; will delay ack */
2562 else if (revoking
== CEPH_CAP_FILE_CACHE
&&
2563 (newcaps
& CEPH_CAP_FILE_LAZYIO
) == 0 &&
2565 ; /* do nothing yet, invalidation will be queued */
2566 else if (cap
== ci
->i_auth_cap
)
2567 check_caps
= 1; /* check auth cap only */
2569 check_caps
= 2; /* check all caps */
2570 cap
->issued
= newcaps
;
2571 cap
->implemented
|= newcaps
;
2572 } else if (cap
->issued
== newcaps
) {
2573 dout("caps unchanged: %s -> %s\n",
2574 ceph_cap_string(cap
->issued
), ceph_cap_string(newcaps
));
2576 dout("grant: %s -> %s\n", ceph_cap_string(cap
->issued
),
2577 ceph_cap_string(newcaps
));
2578 /* non-auth MDS is revoking the newly grant caps ? */
2579 if (cap
== ci
->i_auth_cap
&&
2580 __ceph_caps_revoking_other(ci
, cap
, newcaps
))
2583 cap
->issued
= newcaps
;
2584 cap
->implemented
|= newcaps
; /* add bits only, to
2585 * avoid stepping on a
2586 * pending revocation */
2589 BUG_ON(cap
->issued
& ~cap
->implemented
);
2591 spin_unlock(&ci
->i_ceph_lock
);
2595 * queue inode for writeback: we can't actually call
2596 * filemap_write_and_wait, etc. from message handler
2599 ceph_queue_writeback(inode
);
2600 if (queue_invalidate
)
2601 ceph_queue_invalidate(inode
);
2603 invalidate_aliases(inode
);
2604 if (queue_revalidate
)
2605 ceph_queue_revalidate(inode
);
2607 wake_up_all(&ci
->i_cap_wq
);
2609 if (check_caps
== 1)
2610 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_AUTHONLY
,
2612 else if (check_caps
== 2)
2613 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
, session
);
2615 mutex_unlock(&session
->s_mutex
);
2619 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2620 * MDS has been safely committed.
2622 static void handle_cap_flush_ack(struct inode
*inode
, u64 flush_tid
,
2623 struct ceph_mds_caps
*m
,
2624 struct ceph_mds_session
*session
,
2625 struct ceph_cap
*cap
)
2626 __releases(ci
->i_ceph_lock
)
2628 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2629 struct ceph_mds_client
*mdsc
= ceph_sb_to_client(inode
->i_sb
)->mdsc
;
2630 unsigned seq
= le32_to_cpu(m
->seq
);
2631 int dirty
= le32_to_cpu(m
->dirty
);
2636 for (i
= 0; i
< CEPH_CAP_BITS
; i
++)
2637 if ((dirty
& (1 << i
)) &&
2638 flush_tid
== ci
->i_cap_flush_tid
[i
])
2641 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2642 " flushing %s -> %s\n",
2643 inode
, session
->s_mds
, seq
, ceph_cap_string(dirty
),
2644 ceph_cap_string(cleaned
), ceph_cap_string(ci
->i_flushing_caps
),
2645 ceph_cap_string(ci
->i_flushing_caps
& ~cleaned
));
2647 if (ci
->i_flushing_caps
== (ci
->i_flushing_caps
& ~cleaned
))
2650 ci
->i_flushing_caps
&= ~cleaned
;
2652 spin_lock(&mdsc
->cap_dirty_lock
);
2653 if (ci
->i_flushing_caps
== 0) {
2654 list_del_init(&ci
->i_flushing_item
);
2655 if (!list_empty(&session
->s_cap_flushing
))
2656 dout(" mds%d still flushing cap on %p\n",
2658 &list_entry(session
->s_cap_flushing
.next
,
2659 struct ceph_inode_info
,
2660 i_flushing_item
)->vfs_inode
);
2661 mdsc
->num_cap_flushing
--;
2662 wake_up_all(&mdsc
->cap_flushing_wq
);
2663 dout(" inode %p now !flushing\n", inode
);
2665 if (ci
->i_dirty_caps
== 0) {
2666 dout(" inode %p now clean\n", inode
);
2667 BUG_ON(!list_empty(&ci
->i_dirty_item
));
2669 if (ci
->i_wrbuffer_ref_head
== 0) {
2670 BUG_ON(!ci
->i_head_snapc
);
2671 ceph_put_snap_context(ci
->i_head_snapc
);
2672 ci
->i_head_snapc
= NULL
;
2675 BUG_ON(list_empty(&ci
->i_dirty_item
));
2678 spin_unlock(&mdsc
->cap_dirty_lock
);
2679 wake_up_all(&ci
->i_cap_wq
);
2682 spin_unlock(&ci
->i_ceph_lock
);
2688 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2689 * throw away our cap_snap.
2691 * Caller hold s_mutex.
2693 static void handle_cap_flushsnap_ack(struct inode
*inode
, u64 flush_tid
,
2694 struct ceph_mds_caps
*m
,
2695 struct ceph_mds_session
*session
)
2697 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2698 u64 follows
= le64_to_cpu(m
->snap_follows
);
2699 struct ceph_cap_snap
*capsnap
;
2702 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2703 inode
, ci
, session
->s_mds
, follows
);
2705 spin_lock(&ci
->i_ceph_lock
);
2706 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
2707 if (capsnap
->follows
== follows
) {
2708 if (capsnap
->flush_tid
!= flush_tid
) {
2709 dout(" cap_snap %p follows %lld tid %lld !="
2710 " %lld\n", capsnap
, follows
,
2711 flush_tid
, capsnap
->flush_tid
);
2714 WARN_ON(capsnap
->dirty_pages
|| capsnap
->writing
);
2715 dout(" removing %p cap_snap %p follows %lld\n",
2716 inode
, capsnap
, follows
);
2717 ceph_put_snap_context(capsnap
->context
);
2718 list_del(&capsnap
->ci_item
);
2719 list_del(&capsnap
->flushing_item
);
2720 ceph_put_cap_snap(capsnap
);
2724 dout(" skipping cap_snap %p follows %lld\n",
2725 capsnap
, capsnap
->follows
);
2728 spin_unlock(&ci
->i_ceph_lock
);
2734 * Handle TRUNC from MDS, indicating file truncation.
2736 * caller hold s_mutex.
2738 static void handle_cap_trunc(struct inode
*inode
,
2739 struct ceph_mds_caps
*trunc
,
2740 struct ceph_mds_session
*session
)
2741 __releases(ci
->i_ceph_lock
)
2743 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2744 int mds
= session
->s_mds
;
2745 int seq
= le32_to_cpu(trunc
->seq
);
2746 u32 truncate_seq
= le32_to_cpu(trunc
->truncate_seq
);
2747 u64 truncate_size
= le64_to_cpu(trunc
->truncate_size
);
2748 u64 size
= le64_to_cpu(trunc
->size
);
2749 int implemented
= 0;
2750 int dirty
= __ceph_caps_dirty(ci
);
2751 int issued
= __ceph_caps_issued(ceph_inode(inode
), &implemented
);
2752 int queue_trunc
= 0;
2754 issued
|= implemented
| dirty
;
2756 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2757 inode
, mds
, seq
, truncate_size
, truncate_seq
);
2758 queue_trunc
= ceph_fill_file_size(inode
, issued
,
2759 truncate_seq
, truncate_size
, size
);
2760 spin_unlock(&ci
->i_ceph_lock
);
2763 ceph_queue_vmtruncate(inode
);
2764 ceph_fscache_invalidate(inode
);
2769 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2770 * different one. If we are the most recent migration we've seen (as
2771 * indicated by mseq), make note of the migrating cap bits for the
2772 * duration (until we see the corresponding IMPORT).
2774 * caller holds s_mutex
2776 static void handle_cap_export(struct inode
*inode
, struct ceph_mds_caps
*ex
,
2777 struct ceph_mds_cap_peer
*ph
,
2778 struct ceph_mds_session
*session
)
2780 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
2781 struct ceph_mds_session
*tsession
= NULL
;
2782 struct ceph_cap
*cap
, *tcap
;
2783 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2785 unsigned mseq
= le32_to_cpu(ex
->migrate_seq
);
2786 unsigned t_seq
, t_mseq
;
2788 int mds
= session
->s_mds
;
2791 t_cap_id
= le64_to_cpu(ph
->cap_id
);
2792 t_seq
= le32_to_cpu(ph
->seq
);
2793 t_mseq
= le32_to_cpu(ph
->mseq
);
2794 target
= le32_to_cpu(ph
->mds
);
2796 t_cap_id
= t_seq
= t_mseq
= 0;
2800 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
2801 inode
, ci
, mds
, mseq
, target
);
2803 spin_lock(&ci
->i_ceph_lock
);
2804 cap
= __get_cap_for_mds(ci
, mds
);
2809 __ceph_remove_cap(cap
, false);
2814 * now we know we haven't received the cap import message yet
2815 * because the exported cap still exist.
2818 issued
= cap
->issued
;
2819 WARN_ON(issued
!= cap
->implemented
);
2821 tcap
= __get_cap_for_mds(ci
, target
);
2823 /* already have caps from the target */
2824 if (tcap
->cap_id
!= t_cap_id
||
2825 ceph_seq_cmp(tcap
->seq
, t_seq
) < 0) {
2826 dout(" updating import cap %p mds%d\n", tcap
, target
);
2827 tcap
->cap_id
= t_cap_id
;
2828 tcap
->seq
= t_seq
- 1;
2829 tcap
->issue_seq
= t_seq
- 1;
2830 tcap
->mseq
= t_mseq
;
2831 tcap
->issued
|= issued
;
2832 tcap
->implemented
|= issued
;
2833 if (cap
== ci
->i_auth_cap
)
2834 ci
->i_auth_cap
= tcap
;
2835 if (ci
->i_flushing_caps
&& ci
->i_auth_cap
== tcap
) {
2836 spin_lock(&mdsc
->cap_dirty_lock
);
2837 list_move_tail(&ci
->i_flushing_item
,
2838 &tcap
->session
->s_cap_flushing
);
2839 spin_unlock(&mdsc
->cap_dirty_lock
);
2842 __ceph_remove_cap(cap
, false);
2847 int flag
= (cap
== ci
->i_auth_cap
) ? CEPH_CAP_FLAG_AUTH
: 0;
2848 spin_unlock(&ci
->i_ceph_lock
);
2849 /* add placeholder for the export tagert */
2850 ceph_add_cap(inode
, tsession
, t_cap_id
, -1, issued
, 0,
2851 t_seq
- 1, t_mseq
, (u64
)-1, flag
, NULL
);
2855 spin_unlock(&ci
->i_ceph_lock
);
2856 mutex_unlock(&session
->s_mutex
);
2858 /* open target session */
2859 tsession
= ceph_mdsc_open_export_target_session(mdsc
, target
);
2860 if (!IS_ERR(tsession
)) {
2862 mutex_lock(&session
->s_mutex
);
2863 mutex_lock_nested(&tsession
->s_mutex
,
2864 SINGLE_DEPTH_NESTING
);
2866 mutex_lock(&tsession
->s_mutex
);
2867 mutex_lock_nested(&session
->s_mutex
,
2868 SINGLE_DEPTH_NESTING
);
2870 ceph_add_cap_releases(mdsc
, tsession
);
2879 spin_unlock(&ci
->i_ceph_lock
);
2880 mutex_unlock(&session
->s_mutex
);
2882 mutex_unlock(&tsession
->s_mutex
);
2883 ceph_put_mds_session(tsession
);
2888 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2891 * caller holds s_mutex.
2893 static void handle_cap_import(struct ceph_mds_client
*mdsc
,
2894 struct inode
*inode
, struct ceph_mds_caps
*im
,
2895 struct ceph_mds_cap_peer
*ph
,
2896 struct ceph_mds_session
*session
,
2897 void *snaptrace
, int snaptrace_len
)
2899 struct ceph_inode_info
*ci
= ceph_inode(inode
);
2900 struct ceph_cap
*cap
;
2901 int mds
= session
->s_mds
;
2902 unsigned issued
= le32_to_cpu(im
->caps
);
2903 unsigned wanted
= le32_to_cpu(im
->wanted
);
2904 unsigned seq
= le32_to_cpu(im
->seq
);
2905 unsigned mseq
= le32_to_cpu(im
->migrate_seq
);
2906 u64 realmino
= le64_to_cpu(im
->realm
);
2907 u64 cap_id
= le64_to_cpu(im
->cap_id
);
2912 p_cap_id
= le64_to_cpu(ph
->cap_id
);
2913 peer
= le32_to_cpu(ph
->mds
);
2919 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
2920 inode
, ci
, mds
, mseq
, peer
);
2922 spin_lock(&ci
->i_ceph_lock
);
2923 cap
= peer
>= 0 ? __get_cap_for_mds(ci
, peer
) : NULL
;
2924 if (cap
&& cap
->cap_id
== p_cap_id
) {
2925 dout(" remove export cap %p mds%d flags %d\n",
2926 cap
, peer
, ph
->flags
);
2927 if ((ph
->flags
& CEPH_CAP_FLAG_AUTH
) &&
2928 (cap
->seq
!= le32_to_cpu(ph
->seq
) ||
2929 cap
->mseq
!= le32_to_cpu(ph
->mseq
))) {
2930 pr_err("handle_cap_import: mismatched seq/mseq: "
2931 "ino (%llx.%llx) mds%d seq %d mseq %d "
2932 "importer mds%d has peer seq %d mseq %d\n",
2933 ceph_vinop(inode
), peer
, cap
->seq
,
2934 cap
->mseq
, mds
, le32_to_cpu(ph
->seq
),
2935 le32_to_cpu(ph
->mseq
));
2937 ci
->i_cap_exporting_issued
= cap
->issued
;
2938 __ceph_remove_cap(cap
, (ph
->flags
& CEPH_CAP_FLAG_RELEASE
));
2941 /* make sure we re-request max_size, if necessary */
2942 ci
->i_wanted_max_size
= 0;
2943 ci
->i_requested_max_size
= 0;
2944 spin_unlock(&ci
->i_ceph_lock
);
2946 down_write(&mdsc
->snap_rwsem
);
2947 ceph_update_snap_trace(mdsc
, snaptrace
, snaptrace
+snaptrace_len
,
2949 downgrade_write(&mdsc
->snap_rwsem
);
2950 ceph_add_cap(inode
, session
, cap_id
, -1,
2951 issued
, wanted
, seq
, mseq
, realmino
, CEPH_CAP_FLAG_AUTH
,
2952 NULL
/* no caps context */);
2953 kick_flushing_inode_caps(mdsc
, session
, inode
);
2954 up_read(&mdsc
->snap_rwsem
);
2959 * Handle a caps message from the MDS.
2961 * Identify the appropriate session, inode, and call the right handler
2962 * based on the cap op.
2964 void ceph_handle_caps(struct ceph_mds_session
*session
,
2965 struct ceph_msg
*msg
)
2967 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2968 struct super_block
*sb
= mdsc
->fsc
->sb
;
2969 struct inode
*inode
;
2970 struct ceph_inode_info
*ci
;
2971 struct ceph_cap
*cap
;
2972 struct ceph_mds_caps
*h
;
2973 struct ceph_mds_cap_peer
*peer
= NULL
;
2974 int mds
= session
->s_mds
;
2977 struct ceph_vino vino
;
2982 size_t snaptrace_len
;
2987 dout("handle_caps from mds%d\n", mds
);
2990 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
2991 tid
= le64_to_cpu(msg
->hdr
.tid
);
2992 if (msg
->front
.iov_len
< sizeof(*h
))
2994 h
= msg
->front
.iov_base
;
2995 op
= le32_to_cpu(h
->op
);
2996 vino
.ino
= le64_to_cpu(h
->ino
);
2997 vino
.snap
= CEPH_NOSNAP
;
2998 cap_id
= le64_to_cpu(h
->cap_id
);
2999 seq
= le32_to_cpu(h
->seq
);
3000 mseq
= le32_to_cpu(h
->migrate_seq
);
3001 size
= le64_to_cpu(h
->size
);
3002 max_size
= le64_to_cpu(h
->max_size
);
3005 snaptrace_len
= le32_to_cpu(h
->snap_trace_len
);
3007 if (le16_to_cpu(msg
->hdr
.version
) >= 2) {
3008 void *p
= snaptrace
+ snaptrace_len
;
3009 ceph_decode_32_safe(&p
, end
, flock_len
, bad
);
3010 if (p
+ flock_len
> end
)
3018 if (le16_to_cpu(msg
->hdr
.version
) >= 3) {
3019 if (op
== CEPH_CAP_OP_IMPORT
) {
3020 void *p
= flock
+ flock_len
;
3021 if (p
+ sizeof(*peer
) > end
)
3024 } else if (op
== CEPH_CAP_OP_EXPORT
) {
3025 /* recorded in unused fields */
3026 peer
= (void *)&h
->size
;
3030 mutex_lock(&session
->s_mutex
);
3032 dout(" mds%d seq %lld cap seq %u\n", session
->s_mds
, session
->s_seq
,
3035 if (op
== CEPH_CAP_OP_IMPORT
)
3036 ceph_add_cap_releases(mdsc
, session
);
3039 inode
= ceph_find_inode(sb
, vino
);
3040 ci
= ceph_inode(inode
);
3041 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op
), vino
.ino
,
3044 dout(" i don't have ino %llx\n", vino
.ino
);
3046 if (op
== CEPH_CAP_OP_IMPORT
) {
3047 spin_lock(&session
->s_cap_lock
);
3048 __queue_cap_release(session
, vino
.ino
, cap_id
,
3050 spin_unlock(&session
->s_cap_lock
);
3052 goto flush_cap_releases
;
3055 /* these will work even if we don't have a cap yet */
3057 case CEPH_CAP_OP_FLUSHSNAP_ACK
:
3058 handle_cap_flushsnap_ack(inode
, tid
, h
, session
);
3061 case CEPH_CAP_OP_EXPORT
:
3062 handle_cap_export(inode
, h
, peer
, session
);
3065 case CEPH_CAP_OP_IMPORT
:
3066 handle_cap_import(mdsc
, inode
, h
, peer
, session
,
3067 snaptrace
, snaptrace_len
);
3070 /* the rest require a cap */
3071 spin_lock(&ci
->i_ceph_lock
);
3072 cap
= __get_cap_for_mds(ceph_inode(inode
), mds
);
3074 dout(" no cap on %p ino %llx.%llx from mds%d\n",
3075 inode
, ceph_ino(inode
), ceph_snap(inode
), mds
);
3076 spin_unlock(&ci
->i_ceph_lock
);
3077 goto flush_cap_releases
;
3080 /* note that each of these drops i_ceph_lock for us */
3082 case CEPH_CAP_OP_REVOKE
:
3083 case CEPH_CAP_OP_GRANT
:
3084 case CEPH_CAP_OP_IMPORT
:
3085 handle_cap_grant(inode
, h
, session
, cap
, msg
->middle
);
3088 case CEPH_CAP_OP_FLUSH_ACK
:
3089 handle_cap_flush_ack(inode
, tid
, h
, session
, cap
);
3092 case CEPH_CAP_OP_TRUNC
:
3093 handle_cap_trunc(inode
, h
, session
);
3097 spin_unlock(&ci
->i_ceph_lock
);
3098 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op
,
3099 ceph_cap_op_name(op
));
3106 * send any full release message to try to move things
3107 * along for the mds (who clearly thinks we still have this
3110 ceph_add_cap_releases(mdsc
, session
);
3111 ceph_send_cap_releases(mdsc
, session
);
3114 mutex_unlock(&session
->s_mutex
);
3121 pr_err("ceph_handle_caps: corrupt message\n");
3127 * Delayed work handler to process end of delayed cap release LRU list.
3129 void ceph_check_delayed_caps(struct ceph_mds_client
*mdsc
)
3131 struct ceph_inode_info
*ci
;
3132 int flags
= CHECK_CAPS_NODELAY
;
3134 dout("check_delayed_caps\n");
3136 spin_lock(&mdsc
->cap_delay_lock
);
3137 if (list_empty(&mdsc
->cap_delay_list
))
3139 ci
= list_first_entry(&mdsc
->cap_delay_list
,
3140 struct ceph_inode_info
,
3142 if ((ci
->i_ceph_flags
& CEPH_I_FLUSH
) == 0 &&
3143 time_before(jiffies
, ci
->i_hold_caps_max
))
3145 list_del_init(&ci
->i_cap_delay_list
);
3146 spin_unlock(&mdsc
->cap_delay_lock
);
3147 dout("check_delayed_caps on %p\n", &ci
->vfs_inode
);
3148 ceph_check_caps(ci
, flags
, NULL
);
3150 spin_unlock(&mdsc
->cap_delay_lock
);
3154 * Flush all dirty caps to the mds
3156 void ceph_flush_dirty_caps(struct ceph_mds_client
*mdsc
)
3158 struct ceph_inode_info
*ci
;
3159 struct inode
*inode
;
3161 dout("flush_dirty_caps\n");
3162 spin_lock(&mdsc
->cap_dirty_lock
);
3163 while (!list_empty(&mdsc
->cap_dirty
)) {
3164 ci
= list_first_entry(&mdsc
->cap_dirty
, struct ceph_inode_info
,
3166 inode
= &ci
->vfs_inode
;
3168 dout("flush_dirty_caps %p\n", inode
);
3169 spin_unlock(&mdsc
->cap_dirty_lock
);
3170 ceph_check_caps(ci
, CHECK_CAPS_NODELAY
|CHECK_CAPS_FLUSH
, NULL
);
3172 spin_lock(&mdsc
->cap_dirty_lock
);
3174 spin_unlock(&mdsc
->cap_dirty_lock
);
3175 dout("flush_dirty_caps done\n");
3179 * Drop open file reference. If we were the last open file,
3180 * we may need to release capabilities to the MDS (or schedule
3181 * their delayed release).
3183 void ceph_put_fmode(struct ceph_inode_info
*ci
, int fmode
)
3185 struct inode
*inode
= &ci
->vfs_inode
;
3188 spin_lock(&ci
->i_ceph_lock
);
3189 dout("put_fmode %p fmode %d %d -> %d\n", inode
, fmode
,
3190 ci
->i_nr_by_mode
[fmode
], ci
->i_nr_by_mode
[fmode
]-1);
3191 BUG_ON(ci
->i_nr_by_mode
[fmode
] == 0);
3192 if (--ci
->i_nr_by_mode
[fmode
] == 0)
3194 spin_unlock(&ci
->i_ceph_lock
);
3196 if (last
&& ci
->i_vino
.snap
== CEPH_NOSNAP
)
3197 ceph_check_caps(ci
, 0, NULL
);
3201 * Helpers for embedding cap and dentry lease releases into mds
3204 * @force is used by dentry_release (below) to force inclusion of a
3205 * record for the directory inode, even when there aren't any caps to
3208 int ceph_encode_inode_release(void **p
, struct inode
*inode
,
3209 int mds
, int drop
, int unless
, int force
)
3211 struct ceph_inode_info
*ci
= ceph_inode(inode
);
3212 struct ceph_cap
*cap
;
3213 struct ceph_mds_request_release
*rel
= *p
;
3217 spin_lock(&ci
->i_ceph_lock
);
3218 used
= __ceph_caps_used(ci
);
3219 dirty
= __ceph_caps_dirty(ci
);
3221 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3222 inode
, mds
, ceph_cap_string(used
|dirty
), ceph_cap_string(drop
),
3223 ceph_cap_string(unless
));
3225 /* only drop unused, clean caps */
3226 drop
&= ~(used
| dirty
);
3228 cap
= __get_cap_for_mds(ci
, mds
);
3229 if (cap
&& __cap_is_valid(cap
)) {
3231 ((cap
->issued
& drop
) &&
3232 (cap
->issued
& unless
) == 0)) {
3233 if ((cap
->issued
& drop
) &&
3234 (cap
->issued
& unless
) == 0) {
3235 int wanted
= __ceph_caps_wanted(ci
);
3236 if ((ci
->i_ceph_flags
& CEPH_I_NODELAY
) == 0)
3237 wanted
|= cap
->mds_wanted
;
3238 dout("encode_inode_release %p cap %p "
3239 "%s -> %s, wanted %s -> %s\n", inode
, cap
,
3240 ceph_cap_string(cap
->issued
),
3241 ceph_cap_string(cap
->issued
& ~drop
),
3242 ceph_cap_string(cap
->mds_wanted
),
3243 ceph_cap_string(wanted
));
3245 cap
->issued
&= ~drop
;
3246 cap
->implemented
&= ~drop
;
3247 cap
->mds_wanted
= wanted
;
3249 dout("encode_inode_release %p cap %p %s"
3250 " (force)\n", inode
, cap
,
3251 ceph_cap_string(cap
->issued
));
3254 rel
->ino
= cpu_to_le64(ceph_ino(inode
));
3255 rel
->cap_id
= cpu_to_le64(cap
->cap_id
);
3256 rel
->seq
= cpu_to_le32(cap
->seq
);
3257 rel
->issue_seq
= cpu_to_le32(cap
->issue_seq
),
3258 rel
->mseq
= cpu_to_le32(cap
->mseq
);
3259 rel
->caps
= cpu_to_le32(cap
->issued
);
3260 rel
->wanted
= cpu_to_le32(cap
->mds_wanted
);
3266 dout("encode_inode_release %p cap %p %s\n",
3267 inode
, cap
, ceph_cap_string(cap
->issued
));
3270 spin_unlock(&ci
->i_ceph_lock
);
3274 int ceph_encode_dentry_release(void **p
, struct dentry
*dentry
,
3275 int mds
, int drop
, int unless
)
3277 struct inode
*dir
= dentry
->d_parent
->d_inode
;
3278 struct ceph_mds_request_release
*rel
= *p
;
3279 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
3284 * force an record for the directory caps if we have a dentry lease.
3285 * this is racy (can't take i_ceph_lock and d_lock together), but it
3286 * doesn't have to be perfect; the mds will revoke anything we don't
3289 spin_lock(&dentry
->d_lock
);
3290 if (di
->lease_session
&& di
->lease_session
->s_mds
== mds
)
3292 spin_unlock(&dentry
->d_lock
);
3294 ret
= ceph_encode_inode_release(p
, dir
, mds
, drop
, unless
, force
);
3296 spin_lock(&dentry
->d_lock
);
3297 if (ret
&& di
->lease_session
&& di
->lease_session
->s_mds
== mds
) {
3298 dout("encode_dentry_release %p mds%d seq %d\n",
3299 dentry
, mds
, (int)di
->lease_seq
);
3300 rel
->dname_len
= cpu_to_le32(dentry
->d_name
.len
);
3301 memcpy(*p
, dentry
->d_name
.name
, dentry
->d_name
.len
);
3302 *p
+= dentry
->d_name
.len
;
3303 rel
->dname_seq
= cpu_to_le32(di
->lease_seq
);
3304 __ceph_mdsc_drop_dentry_lease(dentry
);
3306 spin_unlock(&dentry
->d_lock
);