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, mark dir not 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 dout(" marking %p NOT complete\n", &ci->vfs_inode);
502 __ceph_dir_clear_complete(ci);
508 * Add a capability under the given MDS session.
510 * Caller should hold session snap_rwsem (read) and s_mutex.
512 * @fmode is the open file mode, if we are opening a file, otherwise
513 * it is < 0. (This is so we can atomically add the cap and add an
514 * open file reference to it.)
516 int ceph_add_cap(struct inode *inode,
517 struct ceph_mds_session *session, u64 cap_id,
518 int fmode, unsigned issued, unsigned wanted,
519 unsigned seq, unsigned mseq, u64 realmino, int flags,
520 struct ceph_cap_reservation *caps_reservation)
522 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
523 struct ceph_inode_info *ci = ceph_inode(inode);
524 struct ceph_cap *new_cap = NULL;
525 struct ceph_cap *cap;
526 int mds = session->s_mds;
529 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
530 session->s_mds, cap_id, ceph_cap_string(issued), seq);
533 * If we are opening the file, include file mode wanted bits
537 wanted |= ceph_caps_for_mode(fmode);
540 spin_lock(&ci->i_ceph_lock);
541 cap = __get_cap_for_mds(ci, mds);
547 spin_unlock(&ci->i_ceph_lock);
548 new_cap = get_cap(mdsc, caps_reservation);
555 cap->implemented = 0;
561 __insert_cap_node(ci, cap);
563 /* clear out old exporting info? (i.e. on cap import) */
564 if (ci->i_cap_exporting_mds == mds) {
565 ci->i_cap_exporting_issued = 0;
566 ci->i_cap_exporting_mseq = 0;
567 ci->i_cap_exporting_mds = -1;
570 /* add to session cap list */
571 cap->session = session;
572 spin_lock(&session->s_cap_lock);
573 list_add_tail(&cap->session_caps, &session->s_caps);
574 session->s_nr_caps++;
575 spin_unlock(&session->s_cap_lock);
577 ceph_put_cap(mdsc, new_cap);
579 if (!ci->i_snap_realm) {
581 * add this inode to the appropriate snap realm
583 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
586 ceph_get_snap_realm(mdsc, realm);
587 spin_lock(&realm->inodes_with_caps_lock);
588 ci->i_snap_realm = realm;
589 list_add(&ci->i_snap_realm_item,
590 &realm->inodes_with_caps);
591 spin_unlock(&realm->inodes_with_caps_lock);
593 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
599 __check_cap_issue(ci, cap, issued);
602 * If we are issued caps we don't want, or the mds' wanted
603 * value appears to be off, queue a check so we'll release
604 * later and/or update the mds wanted value.
606 actual_wanted = __ceph_caps_wanted(ci);
607 if ((wanted & ~actual_wanted) ||
608 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
609 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
610 ceph_cap_string(issued), ceph_cap_string(wanted),
611 ceph_cap_string(actual_wanted));
612 __cap_delay_requeue(mdsc, ci);
615 if (flags & CEPH_CAP_FLAG_AUTH) {
616 if (ci->i_auth_cap == NULL ||
617 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0)
618 ci->i_auth_cap = cap;
619 } else if (ci->i_auth_cap == cap) {
620 ci->i_auth_cap = NULL;
621 spin_lock(&mdsc->cap_dirty_lock);
622 if (!list_empty(&ci->i_dirty_item)) {
623 dout(" moving %p to cap_dirty_migrating\n", inode);
624 list_move(&ci->i_dirty_item,
625 &mdsc->cap_dirty_migrating);
627 spin_unlock(&mdsc->cap_dirty_lock);
630 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
631 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
632 ceph_cap_string(issued|cap->issued), seq, mds);
633 cap->cap_id = cap_id;
634 cap->issued = issued;
635 cap->implemented |= issued;
636 if (mseq > cap->mseq)
637 cap->mds_wanted = wanted;
639 cap->mds_wanted |= wanted;
641 cap->issue_seq = seq;
643 cap->cap_gen = session->s_cap_gen;
646 __ceph_get_fmode(ci, fmode);
647 spin_unlock(&ci->i_ceph_lock);
648 wake_up_all(&ci->i_cap_wq);
653 * Return true if cap has not timed out and belongs to the current
654 * generation of the MDS session (i.e. has not gone 'stale' due to
655 * us losing touch with the mds).
657 static int __cap_is_valid(struct ceph_cap *cap)
662 spin_lock(&cap->session->s_gen_ttl_lock);
663 gen = cap->session->s_cap_gen;
664 ttl = cap->session->s_cap_ttl;
665 spin_unlock(&cap->session->s_gen_ttl_lock);
667 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
668 dout("__cap_is_valid %p cap %p issued %s "
669 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
670 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
678 * Return set of valid cap bits issued to us. Note that caps time
679 * out, and may be invalidated in bulk if the client session times out
680 * and session->s_cap_gen is bumped.
682 int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
684 int have = ci->i_snap_caps | ci->i_cap_exporting_issued;
685 struct ceph_cap *cap;
690 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
691 cap = rb_entry(p, struct ceph_cap, ci_node);
692 if (!__cap_is_valid(cap))
694 dout("__ceph_caps_issued %p cap %p issued %s\n",
695 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
698 *implemented |= cap->implemented;
704 * Get cap bits issued by caps other than @ocap
706 int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
708 int have = ci->i_snap_caps;
709 struct ceph_cap *cap;
712 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
713 cap = rb_entry(p, struct ceph_cap, ci_node);
716 if (!__cap_is_valid(cap))
724 * Move a cap to the end of the LRU (oldest caps at list head, newest
727 static void __touch_cap(struct ceph_cap *cap)
729 struct ceph_mds_session *s = cap->session;
731 spin_lock(&s->s_cap_lock);
732 if (s->s_cap_iterator == NULL) {
733 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
735 list_move_tail(&cap->session_caps, &s->s_caps);
737 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
738 &cap->ci->vfs_inode, cap, s->s_mds);
740 spin_unlock(&s->s_cap_lock);
744 * Check if we hold the given mask. If so, move the cap(s) to the
745 * front of their respective LRUs. (This is the preferred way for
746 * callers to check for caps they want.)
748 int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
750 struct ceph_cap *cap;
752 int have = ci->i_snap_caps;
754 if ((have & mask) == mask) {
755 dout("__ceph_caps_issued_mask %p snap issued %s"
756 " (mask %s)\n", &ci->vfs_inode,
757 ceph_cap_string(have),
758 ceph_cap_string(mask));
762 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
763 cap = rb_entry(p, struct ceph_cap, ci_node);
764 if (!__cap_is_valid(cap))
766 if ((cap->issued & mask) == mask) {
767 dout("__ceph_caps_issued_mask %p cap %p issued %s"
768 " (mask %s)\n", &ci->vfs_inode, cap,
769 ceph_cap_string(cap->issued),
770 ceph_cap_string(mask));
776 /* does a combination of caps satisfy mask? */
778 if ((have & mask) == mask) {
779 dout("__ceph_caps_issued_mask %p combo issued %s"
780 " (mask %s)\n", &ci->vfs_inode,
781 ceph_cap_string(cap->issued),
782 ceph_cap_string(mask));
786 /* touch this + preceding caps */
788 for (q = rb_first(&ci->i_caps); q != p;
790 cap = rb_entry(q, struct ceph_cap,
792 if (!__cap_is_valid(cap))
805 * Return true if mask caps are currently being revoked by an MDS.
807 int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
809 struct inode *inode = &ci->vfs_inode;
810 struct ceph_cap *cap;
814 spin_lock(&ci->i_ceph_lock);
815 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
816 cap = rb_entry(p, struct ceph_cap, ci_node);
817 if (__cap_is_valid(cap) &&
818 (cap->implemented & ~cap->issued & mask)) {
823 spin_unlock(&ci->i_ceph_lock);
824 dout("ceph_caps_revoking %p %s = %d\n", inode,
825 ceph_cap_string(mask), ret);
829 int __ceph_caps_used(struct ceph_inode_info *ci)
833 used |= CEPH_CAP_PIN;
835 used |= CEPH_CAP_FILE_RD;
836 if (ci->i_rdcache_ref || ci->vfs_inode.i_data.nrpages)
837 used |= CEPH_CAP_FILE_CACHE;
839 used |= CEPH_CAP_FILE_WR;
840 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
841 used |= CEPH_CAP_FILE_BUFFER;
846 * wanted, by virtue of open file modes
848 int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
852 for (mode = 0; mode < CEPH_FILE_MODE_NUM; mode++)
853 if (ci->i_nr_by_mode[mode])
854 want |= ceph_caps_for_mode(mode);
859 * Return caps we have registered with the MDS(s) as 'wanted'.
861 int __ceph_caps_mds_wanted(struct ceph_inode_info *ci)
863 struct ceph_cap *cap;
867 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
868 cap = rb_entry(p, struct ceph_cap, ci_node);
869 if (!__cap_is_valid(cap))
871 mds_wanted |= cap->mds_wanted;
877 * called under i_ceph_lock
879 static int __ceph_is_any_caps(struct ceph_inode_info *ci)
881 return !RB_EMPTY_ROOT(&ci->i_caps) || ci->i_cap_exporting_mds >= 0;
885 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
887 * caller should hold i_ceph_lock.
888 * caller will not hold session s_mutex if called from destroy_inode.
890 void __ceph_remove_cap(struct ceph_cap *cap)
892 struct ceph_mds_session *session = cap->session;
893 struct ceph_inode_info *ci = cap->ci;
894 struct ceph_mds_client *mdsc =
895 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
898 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
900 /* remove from session list */
901 spin_lock(&session->s_cap_lock);
902 if (session->s_cap_iterator == cap) {
903 /* not yet, we are iterating over this very cap */
904 dout("__ceph_remove_cap delaying %p removal from session %p\n",
907 list_del_init(&cap->session_caps);
908 session->s_nr_caps--;
912 /* protect backpointer with s_cap_lock: see iterate_session_caps */
914 spin_unlock(&session->s_cap_lock);
916 /* remove from inode list */
917 rb_erase(&cap->ci_node, &ci->i_caps);
918 if (ci->i_auth_cap == cap)
919 ci->i_auth_cap = NULL;
922 ceph_put_cap(mdsc, cap);
924 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm) {
925 struct ceph_snap_realm *realm = ci->i_snap_realm;
926 spin_lock(&realm->inodes_with_caps_lock);
927 list_del_init(&ci->i_snap_realm_item);
928 ci->i_snap_realm_counter++;
929 ci->i_snap_realm = NULL;
930 spin_unlock(&realm->inodes_with_caps_lock);
931 ceph_put_snap_realm(mdsc, realm);
933 if (!__ceph_is_any_real_caps(ci))
934 __cap_delay_cancel(mdsc, ci);
938 * Build and send a cap message to the given MDS.
940 * Caller should be holding s_mutex.
942 static int send_cap_msg(struct ceph_mds_session *session,
943 u64 ino, u64 cid, int op,
944 int caps, int wanted, int dirty,
945 u32 seq, u64 flush_tid, u32 issue_seq, u32 mseq,
946 u64 size, u64 max_size,
947 struct timespec *mtime, struct timespec *atime,
949 kuid_t uid, kgid_t gid, umode_t mode,
951 struct ceph_buffer *xattrs_buf,
954 struct ceph_mds_caps *fc;
955 struct ceph_msg *msg;
957 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
958 " seq %u/%u mseq %u follows %lld size %llu/%llu"
959 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(op),
960 cid, ino, ceph_cap_string(caps), ceph_cap_string(wanted),
961 ceph_cap_string(dirty),
962 seq, issue_seq, mseq, follows, size, max_size,
963 xattr_version, xattrs_buf ? (int)xattrs_buf->vec.iov_len : 0);
965 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc), GFP_NOFS, false);
969 msg->hdr.tid = cpu_to_le64(flush_tid);
971 fc = msg->front.iov_base;
972 memset(fc, 0, sizeof(*fc));
974 fc->cap_id = cpu_to_le64(cid);
975 fc->op = cpu_to_le32(op);
976 fc->seq = cpu_to_le32(seq);
977 fc->issue_seq = cpu_to_le32(issue_seq);
978 fc->migrate_seq = cpu_to_le32(mseq);
979 fc->caps = cpu_to_le32(caps);
980 fc->wanted = cpu_to_le32(wanted);
981 fc->dirty = cpu_to_le32(dirty);
982 fc->ino = cpu_to_le64(ino);
983 fc->snap_follows = cpu_to_le64(follows);
985 fc->size = cpu_to_le64(size);
986 fc->max_size = cpu_to_le64(max_size);
988 ceph_encode_timespec(&fc->mtime, mtime);
990 ceph_encode_timespec(&fc->atime, atime);
991 fc->time_warp_seq = cpu_to_le32(time_warp_seq);
993 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, uid));
994 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, gid));
995 fc->mode = cpu_to_le32(mode);
997 fc->xattr_version = cpu_to_le64(xattr_version);
999 msg->middle = ceph_buffer_get(xattrs_buf);
1000 fc->xattr_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1001 msg->hdr.middle_len = cpu_to_le32(xattrs_buf->vec.iov_len);
1004 ceph_con_send(&session->s_con, msg);
1008 void __queue_cap_release(struct ceph_mds_session *session,
1009 u64 ino, u64 cap_id, u32 migrate_seq,
1012 struct ceph_msg *msg;
1013 struct ceph_mds_cap_release *head;
1014 struct ceph_mds_cap_item *item;
1016 spin_lock(&session->s_cap_lock);
1017 BUG_ON(!session->s_num_cap_releases);
1018 msg = list_first_entry(&session->s_cap_releases,
1019 struct ceph_msg, list_head);
1021 dout(" adding %llx release to mds%d msg %p (%d left)\n",
1022 ino, session->s_mds, msg, session->s_num_cap_releases);
1024 BUG_ON(msg->front.iov_len + sizeof(*item) > PAGE_CACHE_SIZE);
1025 head = msg->front.iov_base;
1026 le32_add_cpu(&head->num, 1);
1027 item = msg->front.iov_base + msg->front.iov_len;
1028 item->ino = cpu_to_le64(ino);
1029 item->cap_id = cpu_to_le64(cap_id);
1030 item->migrate_seq = cpu_to_le32(migrate_seq);
1031 item->seq = cpu_to_le32(issue_seq);
1033 session->s_num_cap_releases--;
1035 msg->front.iov_len += sizeof(*item);
1036 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1037 dout(" release msg %p full\n", msg);
1038 list_move_tail(&msg->list_head, &session->s_cap_releases_done);
1040 dout(" release msg %p at %d/%d (%d)\n", msg,
1041 (int)le32_to_cpu(head->num),
1042 (int)CEPH_CAPS_PER_RELEASE,
1043 (int)msg->front.iov_len);
1045 spin_unlock(&session->s_cap_lock);
1049 * Queue cap releases when an inode is dropped from our cache. Since
1050 * inode is about to be destroyed, there is no need for i_ceph_lock.
1052 void ceph_queue_caps_release(struct inode *inode)
1054 struct ceph_inode_info *ci = ceph_inode(inode);
1057 p = rb_first(&ci->i_caps);
1059 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1060 struct ceph_mds_session *session = cap->session;
1062 __queue_cap_release(session, ceph_ino(inode), cap->cap_id,
1063 cap->mseq, cap->issue_seq);
1065 __ceph_remove_cap(cap);
1070 * Send a cap msg on the given inode. Update our caps state, then
1071 * drop i_ceph_lock and send the message.
1073 * Make note of max_size reported/requested from mds, revoked caps
1074 * that have now been implemented.
1076 * Make half-hearted attempt ot to invalidate page cache if we are
1077 * dropping RDCACHE. Note that this will leave behind locked pages
1078 * that we'll then need to deal with elsewhere.
1080 * Return non-zero if delayed release, or we experienced an error
1081 * such that the caller should requeue + retry later.
1083 * called with i_ceph_lock, then drops it.
1084 * caller should hold snap_rwsem (read), s_mutex.
1086 static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1087 int op, int used, int want, int retain, int flushing,
1088 unsigned *pflush_tid)
1089 __releases(cap->ci->i_ceph_lock)
1091 struct ceph_inode_info *ci = cap->ci;
1092 struct inode *inode = &ci->vfs_inode;
1093 u64 cap_id = cap->cap_id;
1094 int held, revoking, dropping, keep;
1095 u64 seq, issue_seq, mseq, time_warp_seq, follows;
1097 struct timespec mtime, atime;
1102 struct ceph_mds_session *session;
1103 u64 xattr_version = 0;
1104 struct ceph_buffer *xattr_blob = NULL;
1110 held = cap->issued | cap->implemented;
1111 revoking = cap->implemented & ~cap->issued;
1112 retain &= ~revoking;
1113 dropping = cap->issued & ~retain;
1115 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1116 inode, cap, cap->session,
1117 ceph_cap_string(held), ceph_cap_string(held & retain),
1118 ceph_cap_string(revoking));
1119 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1121 session = cap->session;
1123 /* don't release wanted unless we've waited a bit. */
1124 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1125 time_before(jiffies, ci->i_hold_caps_min)) {
1126 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1127 ceph_cap_string(cap->issued),
1128 ceph_cap_string(cap->issued & retain),
1129 ceph_cap_string(cap->mds_wanted),
1130 ceph_cap_string(want));
1131 want |= cap->mds_wanted;
1132 retain |= cap->issued;
1135 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1137 cap->issued &= retain; /* drop bits we don't want */
1138 if (cap->implemented & ~cap->issued) {
1140 * Wake up any waiters on wanted -> needed transition.
1141 * This is due to the weird transition from buffered
1142 * to sync IO... we need to flush dirty pages _before_
1143 * allowing sync writes to avoid reordering.
1147 cap->implemented &= cap->issued | used;
1148 cap->mds_wanted = want;
1152 * assign a tid for flush operations so we can avoid
1153 * flush1 -> dirty1 -> flush2 -> flushack1 -> mark
1154 * clean type races. track latest tid for every bit
1155 * so we can handle flush AxFw, flush Fw, and have the
1156 * first ack clean Ax.
1158 flush_tid = ++ci->i_cap_flush_last_tid;
1160 *pflush_tid = flush_tid;
1161 dout(" cap_flush_tid %d\n", (int)flush_tid);
1162 for (i = 0; i < CEPH_CAP_BITS; i++)
1163 if (flushing & (1 << i))
1164 ci->i_cap_flush_tid[i] = flush_tid;
1166 follows = ci->i_head_snapc->seq;
1171 keep = cap->implemented;
1173 issue_seq = cap->issue_seq;
1175 size = inode->i_size;
1176 ci->i_reported_size = size;
1177 max_size = ci->i_wanted_max_size;
1178 ci->i_requested_max_size = max_size;
1179 mtime = inode->i_mtime;
1180 atime = inode->i_atime;
1181 time_warp_seq = ci->i_time_warp_seq;
1184 mode = inode->i_mode;
1186 if (flushing & CEPH_CAP_XATTR_EXCL) {
1187 __ceph_build_xattrs_blob(ci);
1188 xattr_blob = ci->i_xattrs.blob;
1189 xattr_version = ci->i_xattrs.version;
1192 spin_unlock(&ci->i_ceph_lock);
1194 ret = send_cap_msg(session, ceph_vino(inode).ino, cap_id,
1195 op, keep, want, flushing, seq, flush_tid, issue_seq, mseq,
1196 size, max_size, &mtime, &atime, time_warp_seq,
1197 uid, gid, mode, xattr_version, xattr_blob,
1200 dout("error sending cap msg, must requeue %p\n", inode);
1205 wake_up_all(&ci->i_cap_wq);
1211 * When a snapshot is taken, clients accumulate dirty metadata on
1212 * inodes with capabilities in ceph_cap_snaps to describe the file
1213 * state at the time the snapshot was taken. This must be flushed
1214 * asynchronously back to the MDS once sync writes complete and dirty
1215 * data is written out.
1217 * Unless @again is true, skip cap_snaps that were already sent to
1218 * the MDS (i.e., during this session).
1220 * Called under i_ceph_lock. Takes s_mutex as needed.
1222 void __ceph_flush_snaps(struct ceph_inode_info *ci,
1223 struct ceph_mds_session **psession,
1225 __releases(ci->i_ceph_lock)
1226 __acquires(ci->i_ceph_lock)
1228 struct inode *inode = &ci->vfs_inode;
1230 struct ceph_cap_snap *capsnap;
1232 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1233 struct ceph_mds_session *session = NULL; /* if session != NULL, we hold
1235 u64 next_follows = 0; /* keep track of how far we've gotten through the
1236 i_cap_snaps list, and skip these entries next time
1237 around to avoid an infinite loop */
1240 session = *psession;
1242 dout("__flush_snaps %p\n", inode);
1244 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1245 /* avoid an infiniute loop after retry */
1246 if (capsnap->follows < next_follows)
1249 * we need to wait for sync writes to complete and for dirty
1250 * pages to be written out.
1252 if (capsnap->dirty_pages || capsnap->writing)
1256 * if cap writeback already occurred, we should have dropped
1257 * the capsnap in ceph_put_wrbuffer_cap_refs.
1259 BUG_ON(capsnap->dirty == 0);
1261 /* pick mds, take s_mutex */
1262 if (ci->i_auth_cap == NULL) {
1263 dout("no auth cap (migrating?), doing nothing\n");
1267 /* only flush each capsnap once */
1268 if (!again && !list_empty(&capsnap->flushing_item)) {
1269 dout("already flushed %p, skipping\n", capsnap);
1273 mds = ci->i_auth_cap->session->s_mds;
1274 mseq = ci->i_auth_cap->mseq;
1276 if (session && session->s_mds != mds) {
1277 dout("oops, wrong session %p mutex\n", session);
1278 mutex_unlock(&session->s_mutex);
1279 ceph_put_mds_session(session);
1283 spin_unlock(&ci->i_ceph_lock);
1284 mutex_lock(&mdsc->mutex);
1285 session = __ceph_lookup_mds_session(mdsc, mds);
1286 mutex_unlock(&mdsc->mutex);
1288 dout("inverting session/ino locks on %p\n",
1290 mutex_lock(&session->s_mutex);
1293 * if session == NULL, we raced against a cap
1294 * deletion or migration. retry, and we'll
1295 * get a better @mds value next time.
1297 spin_lock(&ci->i_ceph_lock);
1301 capsnap->flush_tid = ++ci->i_cap_flush_last_tid;
1302 atomic_inc(&capsnap->nref);
1303 if (!list_empty(&capsnap->flushing_item))
1304 list_del_init(&capsnap->flushing_item);
1305 list_add_tail(&capsnap->flushing_item,
1306 &session->s_cap_snaps_flushing);
1307 spin_unlock(&ci->i_ceph_lock);
1309 dout("flush_snaps %p cap_snap %p follows %lld tid %llu\n",
1310 inode, capsnap, capsnap->follows, capsnap->flush_tid);
1311 send_cap_msg(session, ceph_vino(inode).ino, 0,
1312 CEPH_CAP_OP_FLUSHSNAP, capsnap->issued, 0,
1313 capsnap->dirty, 0, capsnap->flush_tid, 0, mseq,
1315 &capsnap->mtime, &capsnap->atime,
1316 capsnap->time_warp_seq,
1317 capsnap->uid, capsnap->gid, capsnap->mode,
1318 capsnap->xattr_version, capsnap->xattr_blob,
1321 next_follows = capsnap->follows + 1;
1322 ceph_put_cap_snap(capsnap);
1324 spin_lock(&ci->i_ceph_lock);
1328 /* we flushed them all; remove this inode from the queue */
1329 spin_lock(&mdsc->snap_flush_lock);
1330 list_del_init(&ci->i_snap_flush_item);
1331 spin_unlock(&mdsc->snap_flush_lock);
1335 *psession = session;
1337 mutex_unlock(&session->s_mutex);
1338 ceph_put_mds_session(session);
1342 static void ceph_flush_snaps(struct ceph_inode_info *ci)
1344 spin_lock(&ci->i_ceph_lock);
1345 __ceph_flush_snaps(ci, NULL, 0);
1346 spin_unlock(&ci->i_ceph_lock);
1350 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1351 * Caller is then responsible for calling __mark_inode_dirty with the
1352 * returned flags value.
1354 int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask)
1356 struct ceph_mds_client *mdsc =
1357 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
1358 struct inode *inode = &ci->vfs_inode;
1359 int was = ci->i_dirty_caps;
1362 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1363 ceph_cap_string(mask), ceph_cap_string(was),
1364 ceph_cap_string(was | mask));
1365 ci->i_dirty_caps |= mask;
1367 if (!ci->i_head_snapc)
1368 ci->i_head_snapc = ceph_get_snap_context(
1369 ci->i_snap_realm->cached_context);
1370 dout(" inode %p now dirty snapc %p auth cap %p\n",
1371 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
1372 BUG_ON(!list_empty(&ci->i_dirty_item));
1373 spin_lock(&mdsc->cap_dirty_lock);
1375 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1377 list_add(&ci->i_dirty_item,
1378 &mdsc->cap_dirty_migrating);
1379 spin_unlock(&mdsc->cap_dirty_lock);
1380 if (ci->i_flushing_caps == 0) {
1382 dirty |= I_DIRTY_SYNC;
1385 BUG_ON(list_empty(&ci->i_dirty_item));
1386 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1387 (mask & CEPH_CAP_FILE_BUFFER))
1388 dirty |= I_DIRTY_DATASYNC;
1389 __cap_delay_requeue(mdsc, ci);
1394 * Add dirty inode to the flushing list. Assigned a seq number so we
1395 * can wait for caps to flush without starving.
1397 * Called under i_ceph_lock.
1399 static int __mark_caps_flushing(struct inode *inode,
1400 struct ceph_mds_session *session)
1402 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1403 struct ceph_inode_info *ci = ceph_inode(inode);
1406 BUG_ON(ci->i_dirty_caps == 0);
1407 BUG_ON(list_empty(&ci->i_dirty_item));
1409 flushing = ci->i_dirty_caps;
1410 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1411 ceph_cap_string(flushing),
1412 ceph_cap_string(ci->i_flushing_caps),
1413 ceph_cap_string(ci->i_flushing_caps | flushing));
1414 ci->i_flushing_caps |= flushing;
1415 ci->i_dirty_caps = 0;
1416 dout(" inode %p now !dirty\n", inode);
1418 spin_lock(&mdsc->cap_dirty_lock);
1419 list_del_init(&ci->i_dirty_item);
1421 ci->i_cap_flush_seq = ++mdsc->cap_flush_seq;
1422 if (list_empty(&ci->i_flushing_item)) {
1423 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1424 mdsc->num_cap_flushing++;
1425 dout(" inode %p now flushing seq %lld\n", inode,
1426 ci->i_cap_flush_seq);
1428 list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1429 dout(" inode %p now flushing (more) seq %lld\n", inode,
1430 ci->i_cap_flush_seq);
1432 spin_unlock(&mdsc->cap_dirty_lock);
1438 * try to invalidate mapping pages without blocking.
1440 static int try_nonblocking_invalidate(struct inode *inode)
1442 struct ceph_inode_info *ci = ceph_inode(inode);
1443 u32 invalidating_gen = ci->i_rdcache_gen;
1445 spin_unlock(&ci->i_ceph_lock);
1446 invalidate_mapping_pages(&inode->i_data, 0, -1);
1447 spin_lock(&ci->i_ceph_lock);
1449 if (inode->i_data.nrpages == 0 &&
1450 invalidating_gen == ci->i_rdcache_gen) {
1452 dout("try_nonblocking_invalidate %p success\n", inode);
1453 /* save any racing async invalidate some trouble */
1454 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
1457 dout("try_nonblocking_invalidate %p failed\n", inode);
1462 * Swiss army knife function to examine currently used and wanted
1463 * versus held caps. Release, flush, ack revoked caps to mds as
1466 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1467 * cap release further.
1468 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1469 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1472 void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1473 struct ceph_mds_session *session)
1475 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1476 struct ceph_mds_client *mdsc = fsc->mdsc;
1477 struct inode *inode = &ci->vfs_inode;
1478 struct ceph_cap *cap;
1479 int file_wanted, used, cap_used;
1480 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
1481 int issued, implemented, want, retain, revoking, flushing = 0;
1482 int mds = -1; /* keep track of how far we've gone through i_caps list
1483 to avoid an infinite loop on retry */
1485 int tried_invalidate = 0;
1486 int delayed = 0, sent = 0, force_requeue = 0, num;
1487 int queue_invalidate = 0;
1488 int is_delayed = flags & CHECK_CAPS_NODELAY;
1490 /* if we are unmounting, flush any unused caps immediately. */
1494 spin_lock(&ci->i_ceph_lock);
1496 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1497 flags |= CHECK_CAPS_FLUSH;
1499 /* flush snaps first time around only */
1500 if (!list_empty(&ci->i_cap_snaps))
1501 __ceph_flush_snaps(ci, &session, 0);
1504 spin_lock(&ci->i_ceph_lock);
1506 file_wanted = __ceph_caps_file_wanted(ci);
1507 used = __ceph_caps_used(ci);
1508 want = file_wanted | used;
1509 issued = __ceph_caps_issued(ci, &implemented);
1510 revoking = implemented & ~issued;
1512 retain = want | CEPH_CAP_PIN;
1513 if (!mdsc->stopping && inode->i_nlink > 0) {
1515 retain |= CEPH_CAP_ANY; /* be greedy */
1517 retain |= CEPH_CAP_ANY_SHARED;
1519 * keep RD only if we didn't have the file open RW,
1520 * because then the mds would revoke it anyway to
1521 * journal max_size=0.
1523 if (ci->i_max_size == 0)
1524 retain |= CEPH_CAP_ANY_RD;
1528 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1529 " issued %s revoking %s retain %s %s%s%s\n", inode,
1530 ceph_cap_string(file_wanted),
1531 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1532 ceph_cap_string(ci->i_flushing_caps),
1533 ceph_cap_string(issued), ceph_cap_string(revoking),
1534 ceph_cap_string(retain),
1535 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1536 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1537 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1540 * If we no longer need to hold onto old our caps, and we may
1541 * have cached pages, but don't want them, then try to invalidate.
1542 * If we fail, it's because pages are locked.... try again later.
1544 if ((!is_delayed || mdsc->stopping) &&
1545 ci->i_wrbuffer_ref == 0 && /* no dirty pages... */
1546 inode->i_data.nrpages && /* have cached pages */
1547 (file_wanted == 0 || /* no open files */
1548 (revoking & (CEPH_CAP_FILE_CACHE|
1549 CEPH_CAP_FILE_LAZYIO))) && /* or revoking cache */
1550 !tried_invalidate) {
1551 dout("check_caps trying to invalidate on %p\n", inode);
1552 if (try_nonblocking_invalidate(inode) < 0) {
1553 if (revoking & (CEPH_CAP_FILE_CACHE|
1554 CEPH_CAP_FILE_LAZYIO)) {
1555 dout("check_caps queuing invalidate\n");
1556 queue_invalidate = 1;
1557 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1559 dout("check_caps failed to invalidate pages\n");
1560 /* we failed to invalidate pages. check these
1561 caps again later. */
1563 __cap_set_timeouts(mdsc, ci);
1566 tried_invalidate = 1;
1571 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1572 cap = rb_entry(p, struct ceph_cap, ci_node);
1575 /* avoid looping forever */
1576 if (mds >= cap->mds ||
1577 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1580 /* NOTE: no side-effects allowed, until we take s_mutex */
1583 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1584 cap_used &= ~ci->i_auth_cap->issued;
1586 revoking = cap->implemented & ~cap->issued;
1587 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1588 cap->mds, cap, ceph_cap_string(cap->issued),
1589 ceph_cap_string(cap_used),
1590 ceph_cap_string(cap->implemented),
1591 ceph_cap_string(revoking));
1593 if (cap == ci->i_auth_cap &&
1594 (cap->issued & CEPH_CAP_FILE_WR)) {
1595 /* request larger max_size from MDS? */
1596 if (ci->i_wanted_max_size > ci->i_max_size &&
1597 ci->i_wanted_max_size > ci->i_requested_max_size) {
1598 dout("requesting new max_size\n");
1602 /* approaching file_max? */
1603 if ((inode->i_size << 1) >= ci->i_max_size &&
1604 (ci->i_reported_size << 1) < ci->i_max_size) {
1605 dout("i_size approaching max_size\n");
1609 /* flush anything dirty? */
1610 if (cap == ci->i_auth_cap && (flags & CHECK_CAPS_FLUSH) &&
1612 dout("flushing dirty caps\n");
1616 /* completed revocation? going down and there are no caps? */
1617 if (revoking && (revoking & cap_used) == 0) {
1618 dout("completed revocation of %s\n",
1619 ceph_cap_string(cap->implemented & ~cap->issued));
1623 /* want more caps from mds? */
1624 if (want & ~(cap->mds_wanted | cap->issued))
1627 /* things we might delay */
1628 if ((cap->issued & ~retain) == 0 &&
1629 cap->mds_wanted == want)
1630 continue; /* nope, all good */
1636 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1637 time_before(jiffies, ci->i_hold_caps_max)) {
1638 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1639 ceph_cap_string(cap->issued),
1640 ceph_cap_string(cap->issued & retain),
1641 ceph_cap_string(cap->mds_wanted),
1642 ceph_cap_string(want));
1648 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1649 dout(" skipping %p I_NOFLUSH set\n", inode);
1653 if (session && session != cap->session) {
1654 dout("oops, wrong session %p mutex\n", session);
1655 mutex_unlock(&session->s_mutex);
1659 session = cap->session;
1660 if (mutex_trylock(&session->s_mutex) == 0) {
1661 dout("inverting session/ino locks on %p\n",
1663 spin_unlock(&ci->i_ceph_lock);
1664 if (took_snap_rwsem) {
1665 up_read(&mdsc->snap_rwsem);
1666 took_snap_rwsem = 0;
1668 mutex_lock(&session->s_mutex);
1672 /* take snap_rwsem after session mutex */
1673 if (!took_snap_rwsem) {
1674 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
1675 dout("inverting snap/in locks on %p\n",
1677 spin_unlock(&ci->i_ceph_lock);
1678 down_read(&mdsc->snap_rwsem);
1679 took_snap_rwsem = 1;
1682 took_snap_rwsem = 1;
1685 if (cap == ci->i_auth_cap && ci->i_dirty_caps)
1686 flushing = __mark_caps_flushing(inode, session);
1690 mds = cap->mds; /* remember mds, so we don't repeat */
1693 /* __send_cap drops i_ceph_lock */
1694 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, cap_used,
1695 want, retain, flushing, NULL);
1696 goto retry; /* retake i_ceph_lock and restart our cap scan. */
1700 * Reschedule delayed caps release if we delayed anything,
1703 if (delayed && is_delayed)
1704 force_requeue = 1; /* __send_cap delayed release; requeue */
1705 if (!delayed && !is_delayed)
1706 __cap_delay_cancel(mdsc, ci);
1707 else if (!is_delayed || force_requeue)
1708 __cap_delay_requeue(mdsc, ci);
1710 spin_unlock(&ci->i_ceph_lock);
1712 if (queue_invalidate)
1713 ceph_queue_invalidate(inode);
1716 mutex_unlock(&session->s_mutex);
1717 if (took_snap_rwsem)
1718 up_read(&mdsc->snap_rwsem);
1722 * Try to flush dirty caps back to the auth mds.
1724 static int try_flush_caps(struct inode *inode, struct ceph_mds_session *session,
1725 unsigned *flush_tid)
1727 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1728 struct ceph_inode_info *ci = ceph_inode(inode);
1729 int unlock_session = session ? 0 : 1;
1733 spin_lock(&ci->i_ceph_lock);
1734 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
1735 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
1738 if (ci->i_dirty_caps && ci->i_auth_cap) {
1739 struct ceph_cap *cap = ci->i_auth_cap;
1740 int used = __ceph_caps_used(ci);
1741 int want = __ceph_caps_wanted(ci);
1745 spin_unlock(&ci->i_ceph_lock);
1746 session = cap->session;
1747 mutex_lock(&session->s_mutex);
1750 BUG_ON(session != cap->session);
1751 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN)
1754 flushing = __mark_caps_flushing(inode, session);
1756 /* __send_cap drops i_ceph_lock */
1757 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, used, want,
1758 cap->issued | cap->implemented, flushing,
1763 spin_lock(&ci->i_ceph_lock);
1764 __cap_delay_requeue(mdsc, ci);
1767 spin_unlock(&ci->i_ceph_lock);
1769 if (session && unlock_session)
1770 mutex_unlock(&session->s_mutex);
1775 * Return true if we've flushed caps through the given flush_tid.
1777 static int caps_are_flushed(struct inode *inode, unsigned tid)
1779 struct ceph_inode_info *ci = ceph_inode(inode);
1782 spin_lock(&ci->i_ceph_lock);
1783 for (i = 0; i < CEPH_CAP_BITS; i++)
1784 if ((ci->i_flushing_caps & (1 << i)) &&
1785 ci->i_cap_flush_tid[i] <= tid) {
1786 /* still flushing this bit */
1790 spin_unlock(&ci->i_ceph_lock);
1795 * Wait on any unsafe replies for the given inode. First wait on the
1796 * newest request, and make that the upper bound. Then, if there are
1797 * more requests, keep waiting on the oldest as long as it is still older
1798 * than the original request.
1800 static void sync_write_wait(struct inode *inode)
1802 struct ceph_inode_info *ci = ceph_inode(inode);
1803 struct list_head *head = &ci->i_unsafe_writes;
1804 struct ceph_osd_request *req;
1807 spin_lock(&ci->i_unsafe_lock);
1808 if (list_empty(head))
1811 /* set upper bound as _last_ entry in chain */
1812 req = list_entry(head->prev, struct ceph_osd_request,
1814 last_tid = req->r_tid;
1817 ceph_osdc_get_request(req);
1818 spin_unlock(&ci->i_unsafe_lock);
1819 dout("sync_write_wait on tid %llu (until %llu)\n",
1820 req->r_tid, last_tid);
1821 wait_for_completion(&req->r_safe_completion);
1822 spin_lock(&ci->i_unsafe_lock);
1823 ceph_osdc_put_request(req);
1826 * from here on look at first entry in chain, since we
1827 * only want to wait for anything older than last_tid
1829 if (list_empty(head))
1831 req = list_entry(head->next, struct ceph_osd_request,
1833 } while (req->r_tid < last_tid);
1835 spin_unlock(&ci->i_unsafe_lock);
1838 int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1840 struct inode *inode = file->f_mapping->host;
1841 struct ceph_inode_info *ci = ceph_inode(inode);
1846 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
1847 sync_write_wait(inode);
1849 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
1852 mutex_lock(&inode->i_mutex);
1854 dirty = try_flush_caps(inode, NULL, &flush_tid);
1855 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
1858 * only wait on non-file metadata writeback (the mds
1859 * can recover size and mtime, so we don't need to
1862 if (!datasync && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
1863 dout("fsync waiting for flush_tid %u\n", flush_tid);
1864 ret = wait_event_interruptible(ci->i_cap_wq,
1865 caps_are_flushed(inode, flush_tid));
1868 dout("fsync %p%s done\n", inode, datasync ? " datasync" : "");
1869 mutex_unlock(&inode->i_mutex);
1874 * Flush any dirty caps back to the mds. If we aren't asked to wait,
1875 * queue inode for flush but don't do so immediately, because we can
1876 * get by with fewer MDS messages if we wait for data writeback to
1879 int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
1881 struct ceph_inode_info *ci = ceph_inode(inode);
1885 int wait = wbc->sync_mode == WB_SYNC_ALL;
1887 dout("write_inode %p wait=%d\n", inode, wait);
1889 dirty = try_flush_caps(inode, NULL, &flush_tid);
1891 err = wait_event_interruptible(ci->i_cap_wq,
1892 caps_are_flushed(inode, flush_tid));
1894 struct ceph_mds_client *mdsc =
1895 ceph_sb_to_client(inode->i_sb)->mdsc;
1897 spin_lock(&ci->i_ceph_lock);
1898 if (__ceph_caps_dirty(ci))
1899 __cap_delay_requeue_front(mdsc, ci);
1900 spin_unlock(&ci->i_ceph_lock);
1906 * After a recovering MDS goes active, we need to resend any caps
1909 * Caller holds session->s_mutex.
1911 static void kick_flushing_capsnaps(struct ceph_mds_client *mdsc,
1912 struct ceph_mds_session *session)
1914 struct ceph_cap_snap *capsnap;
1916 dout("kick_flushing_capsnaps mds%d\n", session->s_mds);
1917 list_for_each_entry(capsnap, &session->s_cap_snaps_flushing,
1919 struct ceph_inode_info *ci = capsnap->ci;
1920 struct inode *inode = &ci->vfs_inode;
1921 struct ceph_cap *cap;
1923 spin_lock(&ci->i_ceph_lock);
1924 cap = ci->i_auth_cap;
1925 if (cap && cap->session == session) {
1926 dout("kick_flushing_caps %p cap %p capsnap %p\n", inode,
1928 __ceph_flush_snaps(ci, &session, 1);
1930 pr_err("%p auth cap %p not mds%d ???\n", inode,
1931 cap, session->s_mds);
1933 spin_unlock(&ci->i_ceph_lock);
1937 void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
1938 struct ceph_mds_session *session)
1940 struct ceph_inode_info *ci;
1942 kick_flushing_capsnaps(mdsc, session);
1944 dout("kick_flushing_caps mds%d\n", session->s_mds);
1945 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
1946 struct inode *inode = &ci->vfs_inode;
1947 struct ceph_cap *cap;
1950 spin_lock(&ci->i_ceph_lock);
1951 cap = ci->i_auth_cap;
1952 if (cap && cap->session == session) {
1953 dout("kick_flushing_caps %p cap %p %s\n", inode,
1954 cap, ceph_cap_string(ci->i_flushing_caps));
1955 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1956 __ceph_caps_used(ci),
1957 __ceph_caps_wanted(ci),
1958 cap->issued | cap->implemented,
1959 ci->i_flushing_caps, NULL);
1961 spin_lock(&ci->i_ceph_lock);
1962 __cap_delay_requeue(mdsc, ci);
1963 spin_unlock(&ci->i_ceph_lock);
1966 pr_err("%p auth cap %p not mds%d ???\n", inode,
1967 cap, session->s_mds);
1968 spin_unlock(&ci->i_ceph_lock);
1973 static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
1974 struct ceph_mds_session *session,
1975 struct inode *inode)
1977 struct ceph_inode_info *ci = ceph_inode(inode);
1978 struct ceph_cap *cap;
1981 spin_lock(&ci->i_ceph_lock);
1982 cap = ci->i_auth_cap;
1983 dout("kick_flushing_inode_caps %p flushing %s flush_seq %lld\n", inode,
1984 ceph_cap_string(ci->i_flushing_caps), ci->i_cap_flush_seq);
1986 __ceph_flush_snaps(ci, &session, 1);
1988 if (ci->i_flushing_caps) {
1989 spin_lock(&mdsc->cap_dirty_lock);
1990 list_move_tail(&ci->i_flushing_item,
1991 &cap->session->s_cap_flushing);
1992 spin_unlock(&mdsc->cap_dirty_lock);
1994 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
1995 __ceph_caps_used(ci),
1996 __ceph_caps_wanted(ci),
1997 cap->issued | cap->implemented,
1998 ci->i_flushing_caps, NULL);
2000 spin_lock(&ci->i_ceph_lock);
2001 __cap_delay_requeue(mdsc, ci);
2002 spin_unlock(&ci->i_ceph_lock);
2005 spin_unlock(&ci->i_ceph_lock);
2011 * Take references to capabilities we hold, so that we don't release
2012 * them to the MDS prematurely.
2014 * Protected by i_ceph_lock.
2016 static void __take_cap_refs(struct ceph_inode_info *ci, int got)
2018 if (got & CEPH_CAP_PIN)
2020 if (got & CEPH_CAP_FILE_RD)
2022 if (got & CEPH_CAP_FILE_CACHE)
2023 ci->i_rdcache_ref++;
2024 if (got & CEPH_CAP_FILE_WR)
2026 if (got & CEPH_CAP_FILE_BUFFER) {
2027 if (ci->i_wb_ref == 0)
2028 ihold(&ci->vfs_inode);
2030 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2031 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
2036 * Try to grab cap references. Specify those refs we @want, and the
2037 * minimal set we @need. Also include the larger offset we are writing
2038 * to (when applicable), and check against max_size here as well.
2039 * Note that caller is responsible for ensuring max_size increases are
2040 * requested from the MDS.
2042 static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2043 int *got, loff_t endoff, int *check_max, int *err)
2045 struct inode *inode = &ci->vfs_inode;
2047 int have, implemented;
2050 dout("get_cap_refs %p need %s want %s\n", inode,
2051 ceph_cap_string(need), ceph_cap_string(want));
2052 spin_lock(&ci->i_ceph_lock);
2054 /* make sure file is actually open */
2055 file_wanted = __ceph_caps_file_wanted(ci);
2056 if ((file_wanted & need) == 0) {
2057 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2058 ceph_cap_string(need), ceph_cap_string(file_wanted));
2064 /* finish pending truncate */
2065 while (ci->i_truncate_pending) {
2066 spin_unlock(&ci->i_ceph_lock);
2067 __ceph_do_pending_vmtruncate(inode, !(need & CEPH_CAP_FILE_WR));
2068 spin_lock(&ci->i_ceph_lock);
2071 if (need & CEPH_CAP_FILE_WR) {
2072 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2073 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2074 inode, endoff, ci->i_max_size);
2075 if (endoff > ci->i_wanted_max_size) {
2082 * If a sync write is in progress, we must wait, so that we
2083 * can get a final snapshot value for size+mtime.
2085 if (__ceph_have_pending_cap_snap(ci)) {
2086 dout("get_cap_refs %p cap_snap_pending\n", inode);
2090 have = __ceph_caps_issued(ci, &implemented);
2092 if ((have & need) == need) {
2094 * Look at (implemented & ~have & not) so that we keep waiting
2095 * on transition from wanted -> needed caps. This is needed
2096 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2097 * going before a prior buffered writeback happens.
2099 int not = want & ~(have & need);
2100 int revoking = implemented & ~have;
2101 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2102 inode, ceph_cap_string(have), ceph_cap_string(not),
2103 ceph_cap_string(revoking));
2104 if ((revoking & not) == 0) {
2105 *got = need | (have & want);
2106 __take_cap_refs(ci, *got);
2110 dout("get_cap_refs %p have %s needed %s\n", inode,
2111 ceph_cap_string(have), ceph_cap_string(need));
2114 spin_unlock(&ci->i_ceph_lock);
2115 dout("get_cap_refs %p ret %d got %s\n", inode,
2116 ret, ceph_cap_string(*got));
2121 * Check the offset we are writing up to against our current
2122 * max_size. If necessary, tell the MDS we want to write to
2125 static void check_max_size(struct inode *inode, loff_t endoff)
2127 struct ceph_inode_info *ci = ceph_inode(inode);
2130 /* do we need to explicitly request a larger max_size? */
2131 spin_lock(&ci->i_ceph_lock);
2132 if ((endoff >= ci->i_max_size ||
2133 endoff > (inode->i_size << 1)) &&
2134 endoff > ci->i_wanted_max_size) {
2135 dout("write %p at large endoff %llu, req max_size\n",
2137 ci->i_wanted_max_size = endoff;
2140 spin_unlock(&ci->i_ceph_lock);
2142 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2146 * Wait for caps, and take cap references. If we can't get a WR cap
2147 * due to a small max_size, make sure we check_max_size (and possibly
2148 * ask the mds) so we don't get hung up indefinitely.
2150 int ceph_get_caps(struct ceph_inode_info *ci, int need, int want, int *got,
2153 int check_max, ret, err;
2157 check_max_size(&ci->vfs_inode, endoff);
2160 ret = wait_event_interruptible(ci->i_cap_wq,
2161 try_get_cap_refs(ci, need, want,
2172 * Take cap refs. Caller must already know we hold at least one ref
2173 * on the caps in question or we don't know this is safe.
2175 void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2177 spin_lock(&ci->i_ceph_lock);
2178 __take_cap_refs(ci, caps);
2179 spin_unlock(&ci->i_ceph_lock);
2185 * If we released the last ref on any given cap, call ceph_check_caps
2186 * to release (or schedule a release).
2188 * If we are releasing a WR cap (from a sync write), finalize any affected
2189 * cap_snap, and wake up any waiters.
2191 void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2193 struct inode *inode = &ci->vfs_inode;
2194 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2195 struct ceph_cap_snap *capsnap;
2197 spin_lock(&ci->i_ceph_lock);
2198 if (had & CEPH_CAP_PIN)
2200 if (had & CEPH_CAP_FILE_RD)
2201 if (--ci->i_rd_ref == 0)
2203 if (had & CEPH_CAP_FILE_CACHE)
2204 if (--ci->i_rdcache_ref == 0)
2206 if (had & CEPH_CAP_FILE_BUFFER) {
2207 if (--ci->i_wb_ref == 0) {
2211 dout("put_cap_refs %p wb %d -> %d (?)\n",
2212 inode, ci->i_wb_ref+1, ci->i_wb_ref);
2214 if (had & CEPH_CAP_FILE_WR)
2215 if (--ci->i_wr_ref == 0) {
2217 if (!list_empty(&ci->i_cap_snaps)) {
2218 capsnap = list_first_entry(&ci->i_cap_snaps,
2219 struct ceph_cap_snap,
2221 if (capsnap->writing) {
2222 capsnap->writing = 0;
2224 __ceph_finish_cap_snap(ci,
2230 spin_unlock(&ci->i_ceph_lock);
2232 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2233 last ? " last" : "", put ? " put" : "");
2235 if (last && !flushsnaps)
2236 ceph_check_caps(ci, 0, NULL);
2237 else if (flushsnaps)
2238 ceph_flush_snaps(ci);
2240 wake_up_all(&ci->i_cap_wq);
2246 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2247 * context. Adjust per-snap dirty page accounting as appropriate.
2248 * Once all dirty data for a cap_snap is flushed, flush snapped file
2249 * metadata back to the MDS. If we dropped the last ref, call
2252 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2253 struct ceph_snap_context *snapc)
2255 struct inode *inode = &ci->vfs_inode;
2257 int complete_capsnap = 0;
2258 int drop_capsnap = 0;
2260 struct ceph_cap_snap *capsnap = NULL;
2262 spin_lock(&ci->i_ceph_lock);
2263 ci->i_wrbuffer_ref -= nr;
2264 last = !ci->i_wrbuffer_ref;
2266 if (ci->i_head_snapc == snapc) {
2267 ci->i_wrbuffer_ref_head -= nr;
2268 if (ci->i_wrbuffer_ref_head == 0 &&
2269 ci->i_dirty_caps == 0 && ci->i_flushing_caps == 0) {
2270 BUG_ON(!ci->i_head_snapc);
2271 ceph_put_snap_context(ci->i_head_snapc);
2272 ci->i_head_snapc = NULL;
2274 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2276 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2277 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2278 last ? " LAST" : "");
2280 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2281 if (capsnap->context == snapc) {
2287 capsnap->dirty_pages -= nr;
2288 if (capsnap->dirty_pages == 0) {
2289 complete_capsnap = 1;
2290 if (capsnap->dirty == 0)
2291 /* cap writeback completed before we created
2292 * the cap_snap; no FLUSHSNAP is needed */
2295 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2296 " snap %lld %d/%d -> %d/%d %s%s%s\n",
2297 inode, capsnap, capsnap->context->seq,
2298 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2299 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2300 last ? " (wrbuffer last)" : "",
2301 complete_capsnap ? " (complete capsnap)" : "",
2302 drop_capsnap ? " (drop capsnap)" : "");
2304 ceph_put_snap_context(capsnap->context);
2305 list_del(&capsnap->ci_item);
2306 list_del(&capsnap->flushing_item);
2307 ceph_put_cap_snap(capsnap);
2311 spin_unlock(&ci->i_ceph_lock);
2314 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2316 } else if (complete_capsnap) {
2317 ceph_flush_snaps(ci);
2318 wake_up_all(&ci->i_cap_wq);
2325 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
2326 * actually be a revocation if it specifies a smaller cap set.)
2328 * caller holds s_mutex and i_ceph_lock, we drop both.
2332 * 1 - check_caps on auth cap only (writeback)
2333 * 2 - check_caps (ack revoke)
2335 static void handle_cap_grant(struct inode *inode, struct ceph_mds_caps *grant,
2336 struct ceph_mds_session *session,
2337 struct ceph_cap *cap,
2338 struct ceph_buffer *xattr_buf)
2339 __releases(ci->i_ceph_lock)
2341 struct ceph_inode_info *ci = ceph_inode(inode);
2342 int mds = session->s_mds;
2343 int seq = le32_to_cpu(grant->seq);
2344 int newcaps = le32_to_cpu(grant->caps);
2345 int issued, implemented, used, wanted, dirty;
2346 u64 size = le64_to_cpu(grant->size);
2347 u64 max_size = le64_to_cpu(grant->max_size);
2348 struct timespec mtime, atime, ctime;
2352 int revoked_rdcache = 0;
2353 int queue_invalidate = 0;
2355 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
2356 inode, cap, mds, seq, ceph_cap_string(newcaps));
2357 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
2361 * If CACHE is being revoked, and we have no dirty buffers,
2362 * try to invalidate (once). (If there are dirty buffers, we
2363 * will invalidate _after_ writeback.)
2365 if (((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
2366 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2367 !ci->i_wrbuffer_ref) {
2368 if (try_nonblocking_invalidate(inode) == 0) {
2369 revoked_rdcache = 1;
2371 /* there were locked pages.. invalidate later
2372 in a separate thread. */
2373 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2374 queue_invalidate = 1;
2375 ci->i_rdcache_revoking = ci->i_rdcache_gen;
2380 /* side effects now are allowed */
2382 issued = __ceph_caps_issued(ci, &implemented);
2383 issued |= implemented | __ceph_caps_dirty(ci);
2385 cap->cap_gen = session->s_cap_gen;
2387 __check_cap_issue(ci, cap, newcaps);
2389 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
2390 inode->i_mode = le32_to_cpu(grant->mode);
2391 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
2392 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
2393 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
2394 from_kuid(&init_user_ns, inode->i_uid),
2395 from_kgid(&init_user_ns, inode->i_gid));
2398 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
2399 set_nlink(inode, le32_to_cpu(grant->nlink));
2401 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
2402 int len = le32_to_cpu(grant->xattr_len);
2403 u64 version = le64_to_cpu(grant->xattr_version);
2405 if (version > ci->i_xattrs.version) {
2406 dout(" got new xattrs v%llu on %p len %d\n",
2407 version, inode, len);
2408 if (ci->i_xattrs.blob)
2409 ceph_buffer_put(ci->i_xattrs.blob);
2410 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
2411 ci->i_xattrs.version = version;
2415 /* size/ctime/mtime/atime? */
2416 ceph_fill_file_size(inode, issued,
2417 le32_to_cpu(grant->truncate_seq),
2418 le64_to_cpu(grant->truncate_size), size);
2419 ceph_decode_timespec(&mtime, &grant->mtime);
2420 ceph_decode_timespec(&atime, &grant->atime);
2421 ceph_decode_timespec(&ctime, &grant->ctime);
2422 ceph_fill_file_time(inode, issued,
2423 le32_to_cpu(grant->time_warp_seq), &ctime, &mtime,
2426 /* max size increase? */
2427 if (ci->i_auth_cap == cap && max_size != ci->i_max_size) {
2428 dout("max_size %lld -> %llu\n", ci->i_max_size, max_size);
2429 ci->i_max_size = max_size;
2430 if (max_size >= ci->i_wanted_max_size) {
2431 ci->i_wanted_max_size = 0; /* reset */
2432 ci->i_requested_max_size = 0;
2437 /* check cap bits */
2438 wanted = __ceph_caps_wanted(ci);
2439 used = __ceph_caps_used(ci);
2440 dirty = __ceph_caps_dirty(ci);
2441 dout(" my wanted = %s, used = %s, dirty %s\n",
2442 ceph_cap_string(wanted),
2443 ceph_cap_string(used),
2444 ceph_cap_string(dirty));
2445 if (wanted != le32_to_cpu(grant->wanted)) {
2446 dout("mds wanted %s -> %s\n",
2447 ceph_cap_string(le32_to_cpu(grant->wanted)),
2448 ceph_cap_string(wanted));
2449 /* imported cap may not have correct mds_wanted */
2450 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
2456 /* file layout may have changed */
2457 ci->i_layout = grant->layout;
2459 /* revocation, grant, or no-op? */
2460 if (cap->issued & ~newcaps) {
2461 int revoking = cap->issued & ~newcaps;
2463 dout("revocation: %s -> %s (revoking %s)\n",
2464 ceph_cap_string(cap->issued),
2465 ceph_cap_string(newcaps),
2466 ceph_cap_string(revoking));
2467 if (revoking & used & CEPH_CAP_FILE_BUFFER)
2468 writeback = 1; /* initiate writeback; will delay ack */
2469 else if (revoking == CEPH_CAP_FILE_CACHE &&
2470 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
2472 ; /* do nothing yet, invalidation will be queued */
2473 else if (cap == ci->i_auth_cap)
2474 check_caps = 1; /* check auth cap only */
2476 check_caps = 2; /* check all caps */
2477 cap->issued = newcaps;
2478 cap->implemented |= newcaps;
2479 } else if (cap->issued == newcaps) {
2480 dout("caps unchanged: %s -> %s\n",
2481 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
2483 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
2484 ceph_cap_string(newcaps));
2485 cap->issued = newcaps;
2486 cap->implemented |= newcaps; /* add bits only, to
2487 * avoid stepping on a
2488 * pending revocation */
2491 BUG_ON(cap->issued & ~cap->implemented);
2493 spin_unlock(&ci->i_ceph_lock);
2496 * queue inode for writeback: we can't actually call
2497 * filemap_write_and_wait, etc. from message handler
2500 ceph_queue_writeback(inode);
2501 if (queue_invalidate)
2502 ceph_queue_invalidate(inode);
2504 wake_up_all(&ci->i_cap_wq);
2506 if (check_caps == 1)
2507 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
2509 else if (check_caps == 2)
2510 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
2512 mutex_unlock(&session->s_mutex);
2516 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
2517 * MDS has been safely committed.
2519 static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
2520 struct ceph_mds_caps *m,
2521 struct ceph_mds_session *session,
2522 struct ceph_cap *cap)
2523 __releases(ci->i_ceph_lock)
2525 struct ceph_inode_info *ci = ceph_inode(inode);
2526 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2527 unsigned seq = le32_to_cpu(m->seq);
2528 int dirty = le32_to_cpu(m->dirty);
2533 for (i = 0; i < CEPH_CAP_BITS; i++)
2534 if ((dirty & (1 << i)) &&
2535 flush_tid == ci->i_cap_flush_tid[i])
2538 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
2539 " flushing %s -> %s\n",
2540 inode, session->s_mds, seq, ceph_cap_string(dirty),
2541 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
2542 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
2544 if (ci->i_flushing_caps == (ci->i_flushing_caps & ~cleaned))
2547 ci->i_flushing_caps &= ~cleaned;
2549 spin_lock(&mdsc->cap_dirty_lock);
2550 if (ci->i_flushing_caps == 0) {
2551 list_del_init(&ci->i_flushing_item);
2552 if (!list_empty(&session->s_cap_flushing))
2553 dout(" mds%d still flushing cap on %p\n",
2555 &list_entry(session->s_cap_flushing.next,
2556 struct ceph_inode_info,
2557 i_flushing_item)->vfs_inode);
2558 mdsc->num_cap_flushing--;
2559 wake_up_all(&mdsc->cap_flushing_wq);
2560 dout(" inode %p now !flushing\n", inode);
2562 if (ci->i_dirty_caps == 0) {
2563 dout(" inode %p now clean\n", inode);
2564 BUG_ON(!list_empty(&ci->i_dirty_item));
2566 if (ci->i_wrbuffer_ref_head == 0) {
2567 BUG_ON(!ci->i_head_snapc);
2568 ceph_put_snap_context(ci->i_head_snapc);
2569 ci->i_head_snapc = NULL;
2572 BUG_ON(list_empty(&ci->i_dirty_item));
2575 spin_unlock(&mdsc->cap_dirty_lock);
2576 wake_up_all(&ci->i_cap_wq);
2579 spin_unlock(&ci->i_ceph_lock);
2585 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
2586 * throw away our cap_snap.
2588 * Caller hold s_mutex.
2590 static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
2591 struct ceph_mds_caps *m,
2592 struct ceph_mds_session *session)
2594 struct ceph_inode_info *ci = ceph_inode(inode);
2595 u64 follows = le64_to_cpu(m->snap_follows);
2596 struct ceph_cap_snap *capsnap;
2599 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
2600 inode, ci, session->s_mds, follows);
2602 spin_lock(&ci->i_ceph_lock);
2603 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2604 if (capsnap->follows == follows) {
2605 if (capsnap->flush_tid != flush_tid) {
2606 dout(" cap_snap %p follows %lld tid %lld !="
2607 " %lld\n", capsnap, follows,
2608 flush_tid, capsnap->flush_tid);
2611 WARN_ON(capsnap->dirty_pages || capsnap->writing);
2612 dout(" removing %p cap_snap %p follows %lld\n",
2613 inode, capsnap, follows);
2614 ceph_put_snap_context(capsnap->context);
2615 list_del(&capsnap->ci_item);
2616 list_del(&capsnap->flushing_item);
2617 ceph_put_cap_snap(capsnap);
2621 dout(" skipping cap_snap %p follows %lld\n",
2622 capsnap, capsnap->follows);
2625 spin_unlock(&ci->i_ceph_lock);
2631 * Handle TRUNC from MDS, indicating file truncation.
2633 * caller hold s_mutex.
2635 static void handle_cap_trunc(struct inode *inode,
2636 struct ceph_mds_caps *trunc,
2637 struct ceph_mds_session *session)
2638 __releases(ci->i_ceph_lock)
2640 struct ceph_inode_info *ci = ceph_inode(inode);
2641 int mds = session->s_mds;
2642 int seq = le32_to_cpu(trunc->seq);
2643 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
2644 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
2645 u64 size = le64_to_cpu(trunc->size);
2646 int implemented = 0;
2647 int dirty = __ceph_caps_dirty(ci);
2648 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
2649 int queue_trunc = 0;
2651 issued |= implemented | dirty;
2653 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
2654 inode, mds, seq, truncate_size, truncate_seq);
2655 queue_trunc = ceph_fill_file_size(inode, issued,
2656 truncate_seq, truncate_size, size);
2657 spin_unlock(&ci->i_ceph_lock);
2660 ceph_queue_vmtruncate(inode);
2664 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
2665 * different one. If we are the most recent migration we've seen (as
2666 * indicated by mseq), make note of the migrating cap bits for the
2667 * duration (until we see the corresponding IMPORT).
2669 * caller holds s_mutex
2671 static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
2672 struct ceph_mds_session *session,
2673 int *open_target_sessions)
2675 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
2676 struct ceph_inode_info *ci = ceph_inode(inode);
2677 int mds = session->s_mds;
2678 unsigned mseq = le32_to_cpu(ex->migrate_seq);
2679 struct ceph_cap *cap = NULL, *t;
2683 dout("handle_cap_export inode %p ci %p mds%d mseq %d\n",
2684 inode, ci, mds, mseq);
2686 spin_lock(&ci->i_ceph_lock);
2688 /* make sure we haven't seen a higher mseq */
2689 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
2690 t = rb_entry(p, struct ceph_cap, ci_node);
2691 if (ceph_seq_cmp(t->mseq, mseq) > 0) {
2692 dout(" higher mseq on cap from mds%d\n",
2696 if (t->session->s_mds == mds)
2703 ci->i_cap_exporting_mds = mds;
2704 ci->i_cap_exporting_mseq = mseq;
2705 ci->i_cap_exporting_issued = cap->issued;
2708 * make sure we have open sessions with all possible
2709 * export targets, so that we get the matching IMPORT
2711 *open_target_sessions = 1;
2714 * we can't flush dirty caps that we've seen the
2715 * EXPORT but no IMPORT for
2717 spin_lock(&mdsc->cap_dirty_lock);
2718 if (!list_empty(&ci->i_dirty_item)) {
2719 dout(" moving %p to cap_dirty_migrating\n",
2721 list_move(&ci->i_dirty_item,
2722 &mdsc->cap_dirty_migrating);
2724 spin_unlock(&mdsc->cap_dirty_lock);
2726 __ceph_remove_cap(cap);
2728 /* else, we already released it */
2730 spin_unlock(&ci->i_ceph_lock);
2734 * Handle cap IMPORT. If there are temp bits from an older EXPORT,
2737 * caller holds s_mutex.
2739 static void handle_cap_import(struct ceph_mds_client *mdsc,
2740 struct inode *inode, struct ceph_mds_caps *im,
2741 struct ceph_mds_session *session,
2742 void *snaptrace, int snaptrace_len)
2744 struct ceph_inode_info *ci = ceph_inode(inode);
2745 int mds = session->s_mds;
2746 unsigned issued = le32_to_cpu(im->caps);
2747 unsigned wanted = le32_to_cpu(im->wanted);
2748 unsigned seq = le32_to_cpu(im->seq);
2749 unsigned mseq = le32_to_cpu(im->migrate_seq);
2750 u64 realmino = le64_to_cpu(im->realm);
2751 u64 cap_id = le64_to_cpu(im->cap_id);
2753 if (ci->i_cap_exporting_mds >= 0 &&
2754 ceph_seq_cmp(ci->i_cap_exporting_mseq, mseq) < 0) {
2755 dout("handle_cap_import inode %p ci %p mds%d mseq %d"
2756 " - cleared exporting from mds%d\n",
2757 inode, ci, mds, mseq,
2758 ci->i_cap_exporting_mds);
2759 ci->i_cap_exporting_issued = 0;
2760 ci->i_cap_exporting_mseq = 0;
2761 ci->i_cap_exporting_mds = -1;
2763 spin_lock(&mdsc->cap_dirty_lock);
2764 if (!list_empty(&ci->i_dirty_item)) {
2765 dout(" moving %p back to cap_dirty\n", inode);
2766 list_move(&ci->i_dirty_item, &mdsc->cap_dirty);
2768 spin_unlock(&mdsc->cap_dirty_lock);
2770 dout("handle_cap_import inode %p ci %p mds%d mseq %d\n",
2771 inode, ci, mds, mseq);
2774 down_write(&mdsc->snap_rwsem);
2775 ceph_update_snap_trace(mdsc, snaptrace, snaptrace+snaptrace_len,
2777 downgrade_write(&mdsc->snap_rwsem);
2778 ceph_add_cap(inode, session, cap_id, -1,
2779 issued, wanted, seq, mseq, realmino, CEPH_CAP_FLAG_AUTH,
2780 NULL /* no caps context */);
2781 kick_flushing_inode_caps(mdsc, session, inode);
2782 up_read(&mdsc->snap_rwsem);
2784 /* make sure we re-request max_size, if necessary */
2785 spin_lock(&ci->i_ceph_lock);
2786 ci->i_wanted_max_size = 0; /* reset */
2787 ci->i_requested_max_size = 0;
2788 spin_unlock(&ci->i_ceph_lock);
2792 * Handle a caps message from the MDS.
2794 * Identify the appropriate session, inode, and call the right handler
2795 * based on the cap op.
2797 void ceph_handle_caps(struct ceph_mds_session *session,
2798 struct ceph_msg *msg)
2800 struct ceph_mds_client *mdsc = session->s_mdsc;
2801 struct super_block *sb = mdsc->fsc->sb;
2802 struct inode *inode;
2803 struct ceph_inode_info *ci;
2804 struct ceph_cap *cap;
2805 struct ceph_mds_caps *h;
2806 int mds = session->s_mds;
2809 struct ceph_vino vino;
2814 size_t snaptrace_len;
2817 int open_target_sessions = 0;
2819 dout("handle_caps from mds%d\n", mds);
2822 tid = le64_to_cpu(msg->hdr.tid);
2823 if (msg->front.iov_len < sizeof(*h))
2825 h = msg->front.iov_base;
2826 op = le32_to_cpu(h->op);
2827 vino.ino = le64_to_cpu(h->ino);
2828 vino.snap = CEPH_NOSNAP;
2829 cap_id = le64_to_cpu(h->cap_id);
2830 seq = le32_to_cpu(h->seq);
2831 mseq = le32_to_cpu(h->migrate_seq);
2832 size = le64_to_cpu(h->size);
2833 max_size = le64_to_cpu(h->max_size);
2836 snaptrace_len = le32_to_cpu(h->snap_trace_len);
2838 if (le16_to_cpu(msg->hdr.version) >= 2) {
2841 p = snaptrace + snaptrace_len;
2842 end = msg->front.iov_base + msg->front.iov_len;
2843 ceph_decode_32_safe(&p, end, flock_len, bad);
2850 mutex_lock(&session->s_mutex);
2852 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
2855 if (op == CEPH_CAP_OP_IMPORT)
2856 ceph_add_cap_releases(mdsc, session);
2859 inode = ceph_find_inode(sb, vino);
2860 ci = ceph_inode(inode);
2861 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
2864 dout(" i don't have ino %llx\n", vino.ino);
2866 if (op == CEPH_CAP_OP_IMPORT)
2867 __queue_cap_release(session, vino.ino, cap_id,
2869 goto flush_cap_releases;
2872 /* these will work even if we don't have a cap yet */
2874 case CEPH_CAP_OP_FLUSHSNAP_ACK:
2875 handle_cap_flushsnap_ack(inode, tid, h, session);
2878 case CEPH_CAP_OP_EXPORT:
2879 handle_cap_export(inode, h, session, &open_target_sessions);
2882 case CEPH_CAP_OP_IMPORT:
2883 handle_cap_import(mdsc, inode, h, session,
2884 snaptrace, snaptrace_len);
2887 /* the rest require a cap */
2888 spin_lock(&ci->i_ceph_lock);
2889 cap = __get_cap_for_mds(ceph_inode(inode), mds);
2891 dout(" no cap on %p ino %llx.%llx from mds%d\n",
2892 inode, ceph_ino(inode), ceph_snap(inode), mds);
2893 spin_unlock(&ci->i_ceph_lock);
2894 goto flush_cap_releases;
2897 /* note that each of these drops i_ceph_lock for us */
2899 case CEPH_CAP_OP_REVOKE:
2900 case CEPH_CAP_OP_GRANT:
2901 case CEPH_CAP_OP_IMPORT:
2902 handle_cap_grant(inode, h, session, cap, msg->middle);
2905 case CEPH_CAP_OP_FLUSH_ACK:
2906 handle_cap_flush_ack(inode, tid, h, session, cap);
2909 case CEPH_CAP_OP_TRUNC:
2910 handle_cap_trunc(inode, h, session);
2914 spin_unlock(&ci->i_ceph_lock);
2915 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
2916 ceph_cap_op_name(op));
2923 * send any full release message to try to move things
2924 * along for the mds (who clearly thinks we still have this
2927 ceph_add_cap_releases(mdsc, session);
2928 ceph_send_cap_releases(mdsc, session);
2931 mutex_unlock(&session->s_mutex);
2935 if (open_target_sessions)
2936 ceph_mdsc_open_export_target_sessions(mdsc, session);
2940 pr_err("ceph_handle_caps: corrupt message\n");
2946 * Delayed work handler to process end of delayed cap release LRU list.
2948 void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
2950 struct ceph_inode_info *ci;
2951 int flags = CHECK_CAPS_NODELAY;
2953 dout("check_delayed_caps\n");
2955 spin_lock(&mdsc->cap_delay_lock);
2956 if (list_empty(&mdsc->cap_delay_list))
2958 ci = list_first_entry(&mdsc->cap_delay_list,
2959 struct ceph_inode_info,
2961 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
2962 time_before(jiffies, ci->i_hold_caps_max))
2964 list_del_init(&ci->i_cap_delay_list);
2965 spin_unlock(&mdsc->cap_delay_lock);
2966 dout("check_delayed_caps on %p\n", &ci->vfs_inode);
2967 ceph_check_caps(ci, flags, NULL);
2969 spin_unlock(&mdsc->cap_delay_lock);
2973 * Flush all dirty caps to the mds
2975 void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
2977 struct ceph_inode_info *ci;
2978 struct inode *inode;
2980 dout("flush_dirty_caps\n");
2981 spin_lock(&mdsc->cap_dirty_lock);
2982 while (!list_empty(&mdsc->cap_dirty)) {
2983 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
2985 inode = &ci->vfs_inode;
2987 dout("flush_dirty_caps %p\n", inode);
2988 spin_unlock(&mdsc->cap_dirty_lock);
2989 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
2991 spin_lock(&mdsc->cap_dirty_lock);
2993 spin_unlock(&mdsc->cap_dirty_lock);
2994 dout("flush_dirty_caps done\n");
2998 * Drop open file reference. If we were the last open file,
2999 * we may need to release capabilities to the MDS (or schedule
3000 * their delayed release).
3002 void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
3004 struct inode *inode = &ci->vfs_inode;
3007 spin_lock(&ci->i_ceph_lock);
3008 dout("put_fmode %p fmode %d %d -> %d\n", inode, fmode,
3009 ci->i_nr_by_mode[fmode], ci->i_nr_by_mode[fmode]-1);
3010 BUG_ON(ci->i_nr_by_mode[fmode] == 0);
3011 if (--ci->i_nr_by_mode[fmode] == 0)
3013 spin_unlock(&ci->i_ceph_lock);
3015 if (last && ci->i_vino.snap == CEPH_NOSNAP)
3016 ceph_check_caps(ci, 0, NULL);
3020 * Helpers for embedding cap and dentry lease releases into mds
3023 * @force is used by dentry_release (below) to force inclusion of a
3024 * record for the directory inode, even when there aren't any caps to
3027 int ceph_encode_inode_release(void **p, struct inode *inode,
3028 int mds, int drop, int unless, int force)
3030 struct ceph_inode_info *ci = ceph_inode(inode);
3031 struct ceph_cap *cap;
3032 struct ceph_mds_request_release *rel = *p;
3036 spin_lock(&ci->i_ceph_lock);
3037 used = __ceph_caps_used(ci);
3038 dirty = __ceph_caps_dirty(ci);
3040 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
3041 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
3042 ceph_cap_string(unless));
3044 /* only drop unused, clean caps */
3045 drop &= ~(used | dirty);
3047 cap = __get_cap_for_mds(ci, mds);
3048 if (cap && __cap_is_valid(cap)) {
3050 ((cap->issued & drop) &&
3051 (cap->issued & unless) == 0)) {
3052 if ((cap->issued & drop) &&
3053 (cap->issued & unless) == 0) {
3054 int wanted = __ceph_caps_wanted(ci);
3055 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
3056 wanted |= cap->mds_wanted;
3057 dout("encode_inode_release %p cap %p "
3058 "%s -> %s, wanted %s -> %s\n", inode, cap,
3059 ceph_cap_string(cap->issued),
3060 ceph_cap_string(cap->issued & ~drop),
3061 ceph_cap_string(cap->mds_wanted),
3062 ceph_cap_string(wanted));
3064 cap->issued &= ~drop;
3065 cap->implemented &= ~drop;
3066 cap->mds_wanted = wanted;
3068 dout("encode_inode_release %p cap %p %s"
3069 " (force)\n", inode, cap,
3070 ceph_cap_string(cap->issued));
3073 rel->ino = cpu_to_le64(ceph_ino(inode));
3074 rel->cap_id = cpu_to_le64(cap->cap_id);
3075 rel->seq = cpu_to_le32(cap->seq);
3076 rel->issue_seq = cpu_to_le32(cap->issue_seq),
3077 rel->mseq = cpu_to_le32(cap->mseq);
3078 rel->caps = cpu_to_le32(cap->issued);
3079 rel->wanted = cpu_to_le32(cap->mds_wanted);
3085 dout("encode_inode_release %p cap %p %s\n",
3086 inode, cap, ceph_cap_string(cap->issued));
3089 spin_unlock(&ci->i_ceph_lock);
3093 int ceph_encode_dentry_release(void **p, struct dentry *dentry,
3094 int mds, int drop, int unless)
3096 struct inode *dir = dentry->d_parent->d_inode;
3097 struct ceph_mds_request_release *rel = *p;
3098 struct ceph_dentry_info *di = ceph_dentry(dentry);
3103 * force an record for the directory caps if we have a dentry lease.
3104 * this is racy (can't take i_ceph_lock and d_lock together), but it
3105 * doesn't have to be perfect; the mds will revoke anything we don't
3108 spin_lock(&dentry->d_lock);
3109 if (di->lease_session && di->lease_session->s_mds == mds)
3111 spin_unlock(&dentry->d_lock);
3113 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
3115 spin_lock(&dentry->d_lock);
3116 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
3117 dout("encode_dentry_release %p mds%d seq %d\n",
3118 dentry, mds, (int)di->lease_seq);
3119 rel->dname_len = cpu_to_le32(dentry->d_name.len);
3120 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
3121 *p += dentry->d_name.len;
3122 rel->dname_seq = cpu_to_le32(di->lease_seq);
3123 __ceph_mdsc_drop_dentry_lease(dentry);
3125 spin_unlock(&dentry->d_lock);