1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
5 #include <linux/kernel.h>
6 #include <linux/sched/signal.h>
7 #include <linux/slab.h>
8 #include <linux/vmalloc.h>
9 #include <linux/wait.h>
10 #include <linux/writeback.h>
13 #include "mds_client.h"
15 #include <linux/ceph/decode.h>
16 #include <linux/ceph/messenger.h>
19 * Capability management
21 * The Ceph metadata servers control client access to inode metadata
22 * and file data by issuing capabilities, granting clients permission
23 * to read and/or write both inode field and file data to OSDs
24 * (storage nodes). Each capability consists of a set of bits
25 * indicating which operations are allowed.
27 * If the client holds a *_SHARED cap, the client has a coherent value
28 * that can be safely read from the cached inode.
30 * In the case of a *_EXCL (exclusive) or FILE_WR capabilities, the
31 * client is allowed to change inode attributes (e.g., file size,
32 * mtime), note its dirty state in the ceph_cap, and asynchronously
33 * flush that metadata change to the MDS.
35 * In the event of a conflicting operation (perhaps by another
36 * client), the MDS will revoke the conflicting client capabilities.
38 * In order for a client to cache an inode, it must hold a capability
39 * with at least one MDS server. When inodes are released, release
40 * notifications are batched and periodically sent en masse to the MDS
41 * cluster to release server state.
44 static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc);
45 static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
46 struct ceph_mds_session *session,
47 struct ceph_inode_info *ci,
48 u64 oldest_flush_tid);
51 * Generate readable cap strings for debugging output.
53 #define MAX_CAP_STR 20
54 static char cap_str[MAX_CAP_STR][40];
55 static DEFINE_SPINLOCK(cap_str_lock);
56 static int last_cap_str;
58 static char *gcap_string(char *s, int c)
60 if (c & CEPH_CAP_GSHARED)
62 if (c & CEPH_CAP_GEXCL)
64 if (c & CEPH_CAP_GCACHE)
70 if (c & CEPH_CAP_GBUFFER)
72 if (c & CEPH_CAP_GLAZYIO)
77 const char *ceph_cap_string(int caps)
83 spin_lock(&cap_str_lock);
85 if (last_cap_str == MAX_CAP_STR)
87 spin_unlock(&cap_str_lock);
91 if (caps & CEPH_CAP_PIN)
94 c = (caps >> CEPH_CAP_SAUTH) & 3;
97 s = gcap_string(s, c);
100 c = (caps >> CEPH_CAP_SLINK) & 3;
103 s = gcap_string(s, c);
106 c = (caps >> CEPH_CAP_SXATTR) & 3;
109 s = gcap_string(s, c);
112 c = caps >> CEPH_CAP_SFILE;
115 s = gcap_string(s, c);
124 void ceph_caps_init(struct ceph_mds_client *mdsc)
126 INIT_LIST_HEAD(&mdsc->caps_list);
127 spin_lock_init(&mdsc->caps_list_lock);
130 void ceph_caps_finalize(struct ceph_mds_client *mdsc)
132 struct ceph_cap *cap;
134 spin_lock(&mdsc->caps_list_lock);
135 while (!list_empty(&mdsc->caps_list)) {
136 cap = list_first_entry(&mdsc->caps_list,
137 struct ceph_cap, caps_item);
138 list_del(&cap->caps_item);
139 kmem_cache_free(ceph_cap_cachep, cap);
141 mdsc->caps_total_count = 0;
142 mdsc->caps_avail_count = 0;
143 mdsc->caps_use_count = 0;
144 mdsc->caps_reserve_count = 0;
145 mdsc->caps_min_count = 0;
146 spin_unlock(&mdsc->caps_list_lock);
149 void ceph_adjust_min_caps(struct ceph_mds_client *mdsc, int delta)
151 spin_lock(&mdsc->caps_list_lock);
152 mdsc->caps_min_count += delta;
153 BUG_ON(mdsc->caps_min_count < 0);
154 spin_unlock(&mdsc->caps_list_lock);
158 * Called under mdsc->mutex.
160 int ceph_reserve_caps(struct ceph_mds_client *mdsc,
161 struct ceph_cap_reservation *ctx, int need)
164 struct ceph_cap *cap;
168 bool trimmed = false;
169 struct ceph_mds_session *s;
172 dout("reserve caps ctx=%p need=%d\n", ctx, need);
174 /* first reserve any caps that are already allocated */
175 spin_lock(&mdsc->caps_list_lock);
176 if (mdsc->caps_avail_count >= need)
179 have = mdsc->caps_avail_count;
180 mdsc->caps_avail_count -= have;
181 mdsc->caps_reserve_count += have;
182 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
183 mdsc->caps_reserve_count +
184 mdsc->caps_avail_count);
185 spin_unlock(&mdsc->caps_list_lock);
187 for (i = have; i < need; ) {
188 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
190 list_add(&cap->caps_item, &newcaps);
197 for (j = 0; j < mdsc->max_sessions; j++) {
198 s = __ceph_lookup_mds_session(mdsc, j);
201 mutex_unlock(&mdsc->mutex);
203 mutex_lock(&s->s_mutex);
204 max_caps = s->s_nr_caps - (need - i);
205 ceph_trim_caps(mdsc, s, max_caps);
206 mutex_unlock(&s->s_mutex);
208 ceph_put_mds_session(s);
209 mutex_lock(&mdsc->mutex);
213 spin_lock(&mdsc->caps_list_lock);
214 if (mdsc->caps_avail_count) {
216 if (mdsc->caps_avail_count >= need - i)
217 more_have = need - i;
219 more_have = mdsc->caps_avail_count;
223 mdsc->caps_avail_count -= more_have;
224 mdsc->caps_reserve_count += more_have;
227 spin_unlock(&mdsc->caps_list_lock);
232 pr_warn("reserve caps ctx=%p ENOMEM need=%d got=%d\n",
233 ctx, need, have + alloc);
236 BUG_ON(have + alloc != need);
238 spin_lock(&mdsc->caps_list_lock);
239 mdsc->caps_total_count += alloc;
240 mdsc->caps_reserve_count += alloc;
241 list_splice(&newcaps, &mdsc->caps_list);
243 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
244 mdsc->caps_reserve_count +
245 mdsc->caps_avail_count);
246 spin_unlock(&mdsc->caps_list_lock);
249 dout("reserve caps ctx=%p %d = %d used + %d resv + %d avail\n",
250 ctx, mdsc->caps_total_count, mdsc->caps_use_count,
251 mdsc->caps_reserve_count, mdsc->caps_avail_count);
256 spin_lock(&mdsc->caps_list_lock);
257 mdsc->caps_avail_count += have;
258 mdsc->caps_reserve_count -= have;
260 while (!list_empty(&newcaps)) {
261 cap = list_first_entry(&newcaps,
262 struct ceph_cap, caps_item);
263 list_del(&cap->caps_item);
265 /* Keep some preallocated caps around (ceph_min_count), to
266 * avoid lots of free/alloc churn. */
267 if (mdsc->caps_avail_count >=
268 mdsc->caps_reserve_count + mdsc->caps_min_count) {
269 kmem_cache_free(ceph_cap_cachep, cap);
271 mdsc->caps_avail_count++;
272 mdsc->caps_total_count++;
273 list_add(&cap->caps_item, &mdsc->caps_list);
277 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
278 mdsc->caps_reserve_count +
279 mdsc->caps_avail_count);
280 spin_unlock(&mdsc->caps_list_lock);
284 int ceph_unreserve_caps(struct ceph_mds_client *mdsc,
285 struct ceph_cap_reservation *ctx)
288 struct ceph_cap *cap;
290 dout("unreserve caps ctx=%p count=%d\n", ctx, ctx->count);
292 spin_lock(&mdsc->caps_list_lock);
293 BUG_ON(mdsc->caps_reserve_count < ctx->count);
294 mdsc->caps_reserve_count -= ctx->count;
295 if (mdsc->caps_avail_count >=
296 mdsc->caps_reserve_count + mdsc->caps_min_count) {
297 mdsc->caps_total_count -= ctx->count;
298 for (i = 0; i < ctx->count; i++) {
299 cap = list_first_entry(&mdsc->caps_list,
300 struct ceph_cap, caps_item);
301 list_del(&cap->caps_item);
302 kmem_cache_free(ceph_cap_cachep, cap);
305 mdsc->caps_avail_count += ctx->count;
308 dout("unreserve caps %d = %d used + %d resv + %d avail\n",
309 mdsc->caps_total_count, mdsc->caps_use_count,
310 mdsc->caps_reserve_count, mdsc->caps_avail_count);
311 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
312 mdsc->caps_reserve_count +
313 mdsc->caps_avail_count);
314 spin_unlock(&mdsc->caps_list_lock);
319 struct ceph_cap *ceph_get_cap(struct ceph_mds_client *mdsc,
320 struct ceph_cap_reservation *ctx)
322 struct ceph_cap *cap = NULL;
324 /* temporary, until we do something about cap import/export */
326 cap = kmem_cache_alloc(ceph_cap_cachep, GFP_NOFS);
328 spin_lock(&mdsc->caps_list_lock);
329 mdsc->caps_use_count++;
330 mdsc->caps_total_count++;
331 spin_unlock(&mdsc->caps_list_lock);
333 spin_lock(&mdsc->caps_list_lock);
334 if (mdsc->caps_avail_count) {
335 BUG_ON(list_empty(&mdsc->caps_list));
337 mdsc->caps_avail_count--;
338 mdsc->caps_use_count++;
339 cap = list_first_entry(&mdsc->caps_list,
340 struct ceph_cap, caps_item);
341 list_del(&cap->caps_item);
343 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
344 mdsc->caps_reserve_count + mdsc->caps_avail_count);
346 spin_unlock(&mdsc->caps_list_lock);
352 spin_lock(&mdsc->caps_list_lock);
353 dout("get_cap ctx=%p (%d) %d = %d used + %d resv + %d avail\n",
354 ctx, ctx->count, mdsc->caps_total_count, mdsc->caps_use_count,
355 mdsc->caps_reserve_count, mdsc->caps_avail_count);
357 BUG_ON(ctx->count > mdsc->caps_reserve_count);
358 BUG_ON(list_empty(&mdsc->caps_list));
361 mdsc->caps_reserve_count--;
362 mdsc->caps_use_count++;
364 cap = list_first_entry(&mdsc->caps_list, struct ceph_cap, caps_item);
365 list_del(&cap->caps_item);
367 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
368 mdsc->caps_reserve_count + mdsc->caps_avail_count);
369 spin_unlock(&mdsc->caps_list_lock);
373 void ceph_put_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap)
375 spin_lock(&mdsc->caps_list_lock);
376 dout("put_cap %p %d = %d used + %d resv + %d avail\n",
377 cap, mdsc->caps_total_count, mdsc->caps_use_count,
378 mdsc->caps_reserve_count, mdsc->caps_avail_count);
379 mdsc->caps_use_count--;
381 * Keep some preallocated caps around (ceph_min_count), to
382 * avoid lots of free/alloc churn.
384 if (mdsc->caps_avail_count >= mdsc->caps_reserve_count +
385 mdsc->caps_min_count) {
386 mdsc->caps_total_count--;
387 kmem_cache_free(ceph_cap_cachep, cap);
389 mdsc->caps_avail_count++;
390 list_add(&cap->caps_item, &mdsc->caps_list);
393 BUG_ON(mdsc->caps_total_count != mdsc->caps_use_count +
394 mdsc->caps_reserve_count + mdsc->caps_avail_count);
395 spin_unlock(&mdsc->caps_list_lock);
398 void ceph_reservation_status(struct ceph_fs_client *fsc,
399 int *total, int *avail, int *used, int *reserved,
402 struct ceph_mds_client *mdsc = fsc->mdsc;
404 spin_lock(&mdsc->caps_list_lock);
407 *total = mdsc->caps_total_count;
409 *avail = mdsc->caps_avail_count;
411 *used = mdsc->caps_use_count;
413 *reserved = mdsc->caps_reserve_count;
415 *min = mdsc->caps_min_count;
417 spin_unlock(&mdsc->caps_list_lock);
421 * Find ceph_cap for given mds, if any.
423 * Called with i_ceph_lock held.
425 static struct ceph_cap *__get_cap_for_mds(struct ceph_inode_info *ci, int mds)
427 struct ceph_cap *cap;
428 struct rb_node *n = ci->i_caps.rb_node;
431 cap = rb_entry(n, struct ceph_cap, ci_node);
434 else if (mds > cap->mds)
442 struct ceph_cap *ceph_get_cap_for_mds(struct ceph_inode_info *ci, int mds)
444 struct ceph_cap *cap;
446 spin_lock(&ci->i_ceph_lock);
447 cap = __get_cap_for_mds(ci, mds);
448 spin_unlock(&ci->i_ceph_lock);
453 * Return id of any MDS with a cap, preferably FILE_WR|BUFFER|EXCL, else -1.
455 static int __ceph_get_cap_mds(struct ceph_inode_info *ci)
457 struct ceph_cap *cap;
461 /* prefer mds with WR|BUFFER|EXCL caps */
462 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
463 cap = rb_entry(p, struct ceph_cap, ci_node);
465 if (cap->issued & (CEPH_CAP_FILE_WR |
466 CEPH_CAP_FILE_BUFFER |
473 int ceph_get_cap_mds(struct inode *inode)
475 struct ceph_inode_info *ci = ceph_inode(inode);
477 spin_lock(&ci->i_ceph_lock);
478 mds = __ceph_get_cap_mds(ceph_inode(inode));
479 spin_unlock(&ci->i_ceph_lock);
484 * Called under i_ceph_lock.
486 static void __insert_cap_node(struct ceph_inode_info *ci,
487 struct ceph_cap *new)
489 struct rb_node **p = &ci->i_caps.rb_node;
490 struct rb_node *parent = NULL;
491 struct ceph_cap *cap = NULL;
495 cap = rb_entry(parent, struct ceph_cap, ci_node);
496 if (new->mds < cap->mds)
498 else if (new->mds > cap->mds)
504 rb_link_node(&new->ci_node, parent, p);
505 rb_insert_color(&new->ci_node, &ci->i_caps);
509 * (re)set cap hold timeouts, which control the delayed release
510 * of unused caps back to the MDS. Should be called on cap use.
512 static void __cap_set_timeouts(struct ceph_mds_client *mdsc,
513 struct ceph_inode_info *ci)
515 struct ceph_mount_options *ma = mdsc->fsc->mount_options;
517 ci->i_hold_caps_min = round_jiffies(jiffies +
518 ma->caps_wanted_delay_min * HZ);
519 ci->i_hold_caps_max = round_jiffies(jiffies +
520 ma->caps_wanted_delay_max * HZ);
521 dout("__cap_set_timeouts %p min %lu max %lu\n", &ci->vfs_inode,
522 ci->i_hold_caps_min - jiffies, ci->i_hold_caps_max - jiffies);
526 * (Re)queue cap at the end of the delayed cap release list.
528 * If I_FLUSH is set, leave the inode at the front of the list.
530 * Caller holds i_ceph_lock
531 * -> we take mdsc->cap_delay_lock
533 static void __cap_delay_requeue(struct ceph_mds_client *mdsc,
534 struct ceph_inode_info *ci)
536 __cap_set_timeouts(mdsc, ci);
537 dout("__cap_delay_requeue %p flags %d at %lu\n", &ci->vfs_inode,
538 ci->i_ceph_flags, ci->i_hold_caps_max);
539 if (!mdsc->stopping) {
540 spin_lock(&mdsc->cap_delay_lock);
541 if (!list_empty(&ci->i_cap_delay_list)) {
542 if (ci->i_ceph_flags & CEPH_I_FLUSH)
544 list_del_init(&ci->i_cap_delay_list);
546 list_add_tail(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
548 spin_unlock(&mdsc->cap_delay_lock);
553 * Queue an inode for immediate writeback. Mark inode with I_FLUSH,
554 * indicating we should send a cap message to flush dirty metadata
555 * asap, and move to the front of the delayed cap list.
557 static void __cap_delay_requeue_front(struct ceph_mds_client *mdsc,
558 struct ceph_inode_info *ci)
560 dout("__cap_delay_requeue_front %p\n", &ci->vfs_inode);
561 spin_lock(&mdsc->cap_delay_lock);
562 ci->i_ceph_flags |= CEPH_I_FLUSH;
563 if (!list_empty(&ci->i_cap_delay_list))
564 list_del_init(&ci->i_cap_delay_list);
565 list_add(&ci->i_cap_delay_list, &mdsc->cap_delay_list);
566 spin_unlock(&mdsc->cap_delay_lock);
570 * Cancel delayed work on cap.
572 * Caller must hold i_ceph_lock.
574 static void __cap_delay_cancel(struct ceph_mds_client *mdsc,
575 struct ceph_inode_info *ci)
577 dout("__cap_delay_cancel %p\n", &ci->vfs_inode);
578 if (list_empty(&ci->i_cap_delay_list))
580 spin_lock(&mdsc->cap_delay_lock);
581 list_del_init(&ci->i_cap_delay_list);
582 spin_unlock(&mdsc->cap_delay_lock);
586 * Common issue checks for add_cap, handle_cap_grant.
588 static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
591 unsigned had = __ceph_caps_issued(ci, NULL);
594 * Each time we receive FILE_CACHE anew, we increment
597 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
598 (had & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0) {
603 * If FILE_SHARED is newly issued, mark dir not complete. We don't
604 * know what happened to this directory while we didn't have the cap.
605 * If FILE_SHARED is being revoked, also mark dir not complete. It
606 * stops on-going cached readdir.
608 if ((issued & CEPH_CAP_FILE_SHARED) != (had & CEPH_CAP_FILE_SHARED)) {
609 if (issued & CEPH_CAP_FILE_SHARED)
610 atomic_inc(&ci->i_shared_gen);
611 if (S_ISDIR(ci->vfs_inode.i_mode)) {
612 dout(" marking %p NOT complete\n", &ci->vfs_inode);
613 __ceph_dir_clear_complete(ci);
619 * Add a capability under the given MDS session.
621 * Caller should hold session snap_rwsem (read) and s_mutex.
623 * @fmode is the open file mode, if we are opening a file, otherwise
624 * it is < 0. (This is so we can atomically add the cap and add an
625 * open file reference to it.)
627 void ceph_add_cap(struct inode *inode,
628 struct ceph_mds_session *session, u64 cap_id,
629 int fmode, unsigned issued, unsigned wanted,
630 unsigned seq, unsigned mseq, u64 realmino, int flags,
631 struct ceph_cap **new_cap)
633 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
634 struct ceph_inode_info *ci = ceph_inode(inode);
635 struct ceph_cap *cap;
636 int mds = session->s_mds;
639 dout("add_cap %p mds%d cap %llx %s seq %d\n", inode,
640 session->s_mds, cap_id, ceph_cap_string(issued), seq);
643 * If we are opening the file, include file mode wanted bits
647 wanted |= ceph_caps_for_mode(fmode);
649 cap = __get_cap_for_mds(ci, mds);
655 cap->implemented = 0;
661 __insert_cap_node(ci, cap);
663 /* add to session cap list */
664 cap->session = session;
665 spin_lock(&session->s_cap_lock);
666 list_add_tail(&cap->session_caps, &session->s_caps);
667 session->s_nr_caps++;
668 spin_unlock(&session->s_cap_lock);
671 * auth mds of the inode changed. we received the cap export
672 * message, but still haven't received the cap import message.
673 * handle_cap_export() updated the new auth MDS' cap.
675 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing
676 * a message that was send before the cap import message. So
679 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
680 WARN_ON(cap != ci->i_auth_cap);
681 WARN_ON(cap->cap_id != cap_id);
684 issued |= cap->issued;
685 flags |= CEPH_CAP_FLAG_AUTH;
689 if (!ci->i_snap_realm ||
690 ((flags & CEPH_CAP_FLAG_AUTH) &&
691 realmino != (u64)-1 && ci->i_snap_realm->ino != realmino)) {
693 * add this inode to the appropriate snap realm
695 struct ceph_snap_realm *realm = ceph_lookup_snap_realm(mdsc,
698 struct ceph_snap_realm *oldrealm = ci->i_snap_realm;
700 spin_lock(&oldrealm->inodes_with_caps_lock);
701 list_del_init(&ci->i_snap_realm_item);
702 spin_unlock(&oldrealm->inodes_with_caps_lock);
705 spin_lock(&realm->inodes_with_caps_lock);
706 list_add(&ci->i_snap_realm_item,
707 &realm->inodes_with_caps);
708 ci->i_snap_realm = realm;
709 if (realm->ino == ci->i_vino.ino)
710 realm->inode = inode;
711 spin_unlock(&realm->inodes_with_caps_lock);
714 ceph_put_snap_realm(mdsc, oldrealm);
716 pr_err("ceph_add_cap: couldn't find snap realm %llx\n",
722 __check_cap_issue(ci, cap, issued);
725 * If we are issued caps we don't want, or the mds' wanted
726 * value appears to be off, queue a check so we'll release
727 * later and/or update the mds wanted value.
729 actual_wanted = __ceph_caps_wanted(ci);
730 if ((wanted & ~actual_wanted) ||
731 (issued & ~actual_wanted & CEPH_CAP_ANY_WR)) {
732 dout(" issued %s, mds wanted %s, actual %s, queueing\n",
733 ceph_cap_string(issued), ceph_cap_string(wanted),
734 ceph_cap_string(actual_wanted));
735 __cap_delay_requeue(mdsc, ci);
738 if (flags & CEPH_CAP_FLAG_AUTH) {
739 if (!ci->i_auth_cap ||
740 ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
741 ci->i_auth_cap = cap;
742 cap->mds_wanted = wanted;
745 WARN_ON(ci->i_auth_cap == cap);
748 dout("add_cap inode %p (%llx.%llx) cap %p %s now %s seq %d mds%d\n",
749 inode, ceph_vinop(inode), cap, ceph_cap_string(issued),
750 ceph_cap_string(issued|cap->issued), seq, mds);
751 cap->cap_id = cap_id;
752 cap->issued = issued;
753 cap->implemented |= issued;
754 if (ceph_seq_cmp(mseq, cap->mseq) > 0)
755 cap->mds_wanted = wanted;
757 cap->mds_wanted |= wanted;
759 cap->issue_seq = seq;
761 cap->cap_gen = session->s_cap_gen;
764 __ceph_get_fmode(ci, fmode);
768 * Return true if cap has not timed out and belongs to the current
769 * generation of the MDS session (i.e. has not gone 'stale' due to
770 * us losing touch with the mds).
772 static int __cap_is_valid(struct ceph_cap *cap)
777 spin_lock(&cap->session->s_gen_ttl_lock);
778 gen = cap->session->s_cap_gen;
779 ttl = cap->session->s_cap_ttl;
780 spin_unlock(&cap->session->s_gen_ttl_lock);
782 if (cap->cap_gen < gen || time_after_eq(jiffies, ttl)) {
783 dout("__cap_is_valid %p cap %p issued %s "
784 "but STALE (gen %u vs %u)\n", &cap->ci->vfs_inode,
785 cap, ceph_cap_string(cap->issued), cap->cap_gen, gen);
793 * Return set of valid cap bits issued to us. Note that caps time
794 * out, and may be invalidated in bulk if the client session times out
795 * and session->s_cap_gen is bumped.
797 int __ceph_caps_issued(struct ceph_inode_info *ci, int *implemented)
799 int have = ci->i_snap_caps;
800 struct ceph_cap *cap;
805 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
806 cap = rb_entry(p, struct ceph_cap, ci_node);
807 if (!__cap_is_valid(cap))
809 dout("__ceph_caps_issued %p cap %p issued %s\n",
810 &ci->vfs_inode, cap, ceph_cap_string(cap->issued));
813 *implemented |= cap->implemented;
816 * exclude caps issued by non-auth MDS, but are been revoking
817 * by the auth MDS. The non-auth MDS should be revoking/exporting
818 * these caps, but the message is delayed.
820 if (ci->i_auth_cap) {
821 cap = ci->i_auth_cap;
822 have &= ~cap->implemented | cap->issued;
828 * Get cap bits issued by caps other than @ocap
830 int __ceph_caps_issued_other(struct ceph_inode_info *ci, struct ceph_cap *ocap)
832 int have = ci->i_snap_caps;
833 struct ceph_cap *cap;
836 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
837 cap = rb_entry(p, struct ceph_cap, ci_node);
840 if (!__cap_is_valid(cap))
848 * Move a cap to the end of the LRU (oldest caps at list head, newest
851 static void __touch_cap(struct ceph_cap *cap)
853 struct ceph_mds_session *s = cap->session;
855 spin_lock(&s->s_cap_lock);
856 if (!s->s_cap_iterator) {
857 dout("__touch_cap %p cap %p mds%d\n", &cap->ci->vfs_inode, cap,
859 list_move_tail(&cap->session_caps, &s->s_caps);
861 dout("__touch_cap %p cap %p mds%d NOP, iterating over caps\n",
862 &cap->ci->vfs_inode, cap, s->s_mds);
864 spin_unlock(&s->s_cap_lock);
868 * Check if we hold the given mask. If so, move the cap(s) to the
869 * front of their respective LRUs. (This is the preferred way for
870 * callers to check for caps they want.)
872 int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
874 struct ceph_cap *cap;
876 int have = ci->i_snap_caps;
878 if ((have & mask) == mask) {
879 dout("__ceph_caps_issued_mask %p snap issued %s"
880 " (mask %s)\n", &ci->vfs_inode,
881 ceph_cap_string(have),
882 ceph_cap_string(mask));
886 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
887 cap = rb_entry(p, struct ceph_cap, ci_node);
888 if (!__cap_is_valid(cap))
890 if ((cap->issued & mask) == mask) {
891 dout("__ceph_caps_issued_mask %p cap %p issued %s"
892 " (mask %s)\n", &ci->vfs_inode, cap,
893 ceph_cap_string(cap->issued),
894 ceph_cap_string(mask));
900 /* does a combination of caps satisfy mask? */
902 if ((have & mask) == mask) {
903 dout("__ceph_caps_issued_mask %p combo issued %s"
904 " (mask %s)\n", &ci->vfs_inode,
905 ceph_cap_string(cap->issued),
906 ceph_cap_string(mask));
910 /* touch this + preceding caps */
912 for (q = rb_first(&ci->i_caps); q != p;
914 cap = rb_entry(q, struct ceph_cap,
916 if (!__cap_is_valid(cap))
929 * Return true if mask caps are currently being revoked by an MDS.
931 int __ceph_caps_revoking_other(struct ceph_inode_info *ci,
932 struct ceph_cap *ocap, int mask)
934 struct ceph_cap *cap;
937 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
938 cap = rb_entry(p, struct ceph_cap, ci_node);
940 (cap->implemented & ~cap->issued & mask))
946 int ceph_caps_revoking(struct ceph_inode_info *ci, int mask)
948 struct inode *inode = &ci->vfs_inode;
951 spin_lock(&ci->i_ceph_lock);
952 ret = __ceph_caps_revoking_other(ci, NULL, mask);
953 spin_unlock(&ci->i_ceph_lock);
954 dout("ceph_caps_revoking %p %s = %d\n", inode,
955 ceph_cap_string(mask), ret);
959 int __ceph_caps_used(struct ceph_inode_info *ci)
963 used |= CEPH_CAP_PIN;
965 used |= CEPH_CAP_FILE_RD;
966 if (ci->i_rdcache_ref ||
967 (!S_ISDIR(ci->vfs_inode.i_mode) && /* ignore readdir cache */
968 ci->vfs_inode.i_data.nrpages))
969 used |= CEPH_CAP_FILE_CACHE;
971 used |= CEPH_CAP_FILE_WR;
972 if (ci->i_wb_ref || ci->i_wrbuffer_ref)
973 used |= CEPH_CAP_FILE_BUFFER;
978 * wanted, by virtue of open file modes
980 int __ceph_caps_file_wanted(struct ceph_inode_info *ci)
983 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
984 if (ci->i_nr_by_mode[i])
989 return ceph_caps_for_mode(bits >> 1);
993 * Return caps we have registered with the MDS(s) as 'wanted'.
995 int __ceph_caps_mds_wanted(struct ceph_inode_info *ci, bool check)
997 struct ceph_cap *cap;
1001 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1002 cap = rb_entry(p, struct ceph_cap, ci_node);
1003 if (check && !__cap_is_valid(cap))
1005 if (cap == ci->i_auth_cap)
1006 mds_wanted |= cap->mds_wanted;
1008 mds_wanted |= (cap->mds_wanted & ~CEPH_CAP_ANY_FILE_WR);
1014 * called under i_ceph_lock
1016 static int __ceph_is_single_caps(struct ceph_inode_info *ci)
1018 return rb_first(&ci->i_caps) == rb_last(&ci->i_caps);
1021 static int __ceph_is_any_caps(struct ceph_inode_info *ci)
1023 return !RB_EMPTY_ROOT(&ci->i_caps);
1026 int ceph_is_any_caps(struct inode *inode)
1028 struct ceph_inode_info *ci = ceph_inode(inode);
1031 spin_lock(&ci->i_ceph_lock);
1032 ret = __ceph_is_any_caps(ci);
1033 spin_unlock(&ci->i_ceph_lock);
1038 static void drop_inode_snap_realm(struct ceph_inode_info *ci)
1040 struct ceph_snap_realm *realm = ci->i_snap_realm;
1041 spin_lock(&realm->inodes_with_caps_lock);
1042 list_del_init(&ci->i_snap_realm_item);
1043 ci->i_snap_realm_counter++;
1044 ci->i_snap_realm = NULL;
1045 spin_unlock(&realm->inodes_with_caps_lock);
1046 ceph_put_snap_realm(ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc,
1051 * Remove a cap. Take steps to deal with a racing iterate_session_caps.
1053 * caller should hold i_ceph_lock.
1054 * caller will not hold session s_mutex if called from destroy_inode.
1056 void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
1058 struct ceph_mds_session *session = cap->session;
1059 struct ceph_inode_info *ci = cap->ci;
1060 struct ceph_mds_client *mdsc =
1061 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
1064 dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode);
1066 /* remove from session list */
1067 spin_lock(&session->s_cap_lock);
1068 if (session->s_cap_iterator == cap) {
1069 /* not yet, we are iterating over this very cap */
1070 dout("__ceph_remove_cap delaying %p removal from session %p\n",
1073 list_del_init(&cap->session_caps);
1074 session->s_nr_caps--;
1075 cap->session = NULL;
1078 /* protect backpointer with s_cap_lock: see iterate_session_caps */
1082 * s_cap_reconnect is protected by s_cap_lock. no one changes
1083 * s_cap_gen while session is in the reconnect state.
1085 if (queue_release &&
1086 (!session->s_cap_reconnect || cap->cap_gen == session->s_cap_gen)) {
1087 cap->queue_release = 1;
1089 list_add_tail(&cap->session_caps,
1090 &session->s_cap_releases);
1091 session->s_num_cap_releases++;
1095 cap->queue_release = 0;
1097 cap->cap_ino = ci->i_vino.ino;
1099 spin_unlock(&session->s_cap_lock);
1101 /* remove from inode list */
1102 rb_erase(&cap->ci_node, &ci->i_caps);
1103 if (ci->i_auth_cap == cap)
1104 ci->i_auth_cap = NULL;
1107 ceph_put_cap(mdsc, cap);
1109 /* when reconnect denied, we remove session caps forcibly,
1110 * i_wr_ref can be non-zero. If there are ongoing write,
1111 * keep i_snap_realm.
1113 if (!__ceph_is_any_caps(ci) && ci->i_wr_ref == 0 && ci->i_snap_realm)
1114 drop_inode_snap_realm(ci);
1116 if (!__ceph_is_any_real_caps(ci))
1117 __cap_delay_cancel(mdsc, ci);
1120 struct cap_msg_args {
1121 struct ceph_mds_session *session;
1122 u64 ino, cid, follows;
1123 u64 flush_tid, oldest_flush_tid, size, max_size;
1125 struct ceph_buffer *xattr_buf;
1126 struct timespec atime, mtime, ctime;
1127 int op, caps, wanted, dirty;
1128 u32 seq, issue_seq, mseq, time_warp_seq;
1137 * Build and send a cap message to the given MDS.
1139 * Caller should be holding s_mutex.
1141 static int send_cap_msg(struct cap_msg_args *arg)
1143 struct ceph_mds_caps *fc;
1144 struct ceph_msg *msg;
1147 struct timespec zerotime = {0};
1148 struct ceph_osd_client *osdc = &arg->session->s_mdsc->fsc->client->osdc;
1150 dout("send_cap_msg %s %llx %llx caps %s wanted %s dirty %s"
1151 " seq %u/%u tid %llu/%llu mseq %u follows %lld size %llu/%llu"
1152 " xattr_ver %llu xattr_len %d\n", ceph_cap_op_name(arg->op),
1153 arg->cid, arg->ino, ceph_cap_string(arg->caps),
1154 ceph_cap_string(arg->wanted), ceph_cap_string(arg->dirty),
1155 arg->seq, arg->issue_seq, arg->flush_tid, arg->oldest_flush_tid,
1156 arg->mseq, arg->follows, arg->size, arg->max_size,
1158 arg->xattr_buf ? (int)arg->xattr_buf->vec.iov_len : 0);
1160 /* flock buffer size + inline version + inline data size +
1161 * osd_epoch_barrier + oldest_flush_tid */
1162 extra_len = 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4;
1163 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, sizeof(*fc) + extra_len,
1168 msg->hdr.version = cpu_to_le16(10);
1169 msg->hdr.tid = cpu_to_le64(arg->flush_tid);
1171 fc = msg->front.iov_base;
1172 memset(fc, 0, sizeof(*fc));
1174 fc->cap_id = cpu_to_le64(arg->cid);
1175 fc->op = cpu_to_le32(arg->op);
1176 fc->seq = cpu_to_le32(arg->seq);
1177 fc->issue_seq = cpu_to_le32(arg->issue_seq);
1178 fc->migrate_seq = cpu_to_le32(arg->mseq);
1179 fc->caps = cpu_to_le32(arg->caps);
1180 fc->wanted = cpu_to_le32(arg->wanted);
1181 fc->dirty = cpu_to_le32(arg->dirty);
1182 fc->ino = cpu_to_le64(arg->ino);
1183 fc->snap_follows = cpu_to_le64(arg->follows);
1185 fc->size = cpu_to_le64(arg->size);
1186 fc->max_size = cpu_to_le64(arg->max_size);
1187 ceph_encode_timespec(&fc->mtime, &arg->mtime);
1188 ceph_encode_timespec(&fc->atime, &arg->atime);
1189 ceph_encode_timespec(&fc->ctime, &arg->ctime);
1190 fc->time_warp_seq = cpu_to_le32(arg->time_warp_seq);
1192 fc->uid = cpu_to_le32(from_kuid(&init_user_ns, arg->uid));
1193 fc->gid = cpu_to_le32(from_kgid(&init_user_ns, arg->gid));
1194 fc->mode = cpu_to_le32(arg->mode);
1196 fc->xattr_version = cpu_to_le64(arg->xattr_version);
1197 if (arg->xattr_buf) {
1198 msg->middle = ceph_buffer_get(arg->xattr_buf);
1199 fc->xattr_len = cpu_to_le32(arg->xattr_buf->vec.iov_len);
1200 msg->hdr.middle_len = cpu_to_le32(arg->xattr_buf->vec.iov_len);
1204 /* flock buffer size (version 2) */
1205 ceph_encode_32(&p, 0);
1206 /* inline version (version 4) */
1207 ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE);
1208 /* inline data size */
1209 ceph_encode_32(&p, 0);
1211 * osd_epoch_barrier (version 5)
1212 * The epoch_barrier is protected osdc->lock, so READ_ONCE here in
1213 * case it was recently changed
1215 ceph_encode_32(&p, READ_ONCE(osdc->epoch_barrier));
1216 /* oldest_flush_tid (version 6) */
1217 ceph_encode_64(&p, arg->oldest_flush_tid);
1220 * caller_uid/caller_gid (version 7)
1222 * Currently, we don't properly track which caller dirtied the caps
1223 * last, and force a flush of them when there is a conflict. For now,
1224 * just set this to 0:0, to emulate how the MDS has worked up to now.
1226 ceph_encode_32(&p, 0);
1227 ceph_encode_32(&p, 0);
1229 /* pool namespace (version 8) (mds always ignores this) */
1230 ceph_encode_32(&p, 0);
1233 * btime and change_attr (version 9)
1235 * We just zero these out for now, as the MDS ignores them unless
1236 * the requisite feature flags are set (which we don't do yet).
1238 ceph_encode_timespec(p, &zerotime);
1239 p += sizeof(struct ceph_timespec);
1240 ceph_encode_64(&p, 0);
1242 /* Advisory flags (version 10) */
1243 ceph_encode_32(&p, arg->flags);
1245 ceph_con_send(&arg->session->s_con, msg);
1250 * Queue cap releases when an inode is dropped from our cache. Since
1251 * inode is about to be destroyed, there is no need for i_ceph_lock.
1253 void ceph_queue_caps_release(struct inode *inode)
1255 struct ceph_inode_info *ci = ceph_inode(inode);
1258 p = rb_first(&ci->i_caps);
1260 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
1262 __ceph_remove_cap(cap, true);
1267 * Send a cap msg on the given inode. Update our caps state, then
1268 * drop i_ceph_lock and send the message.
1270 * Make note of max_size reported/requested from mds, revoked caps
1271 * that have now been implemented.
1273 * Make half-hearted attempt ot to invalidate page cache if we are
1274 * dropping RDCACHE. Note that this will leave behind locked pages
1275 * that we'll then need to deal with elsewhere.
1277 * Return non-zero if delayed release, or we experienced an error
1278 * such that the caller should requeue + retry later.
1280 * called with i_ceph_lock, then drops it.
1281 * caller should hold snap_rwsem (read), s_mutex.
1283 static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
1284 int op, bool sync, int used, int want, int retain,
1285 int flushing, u64 flush_tid, u64 oldest_flush_tid)
1286 __releases(cap->ci->i_ceph_lock)
1288 struct ceph_inode_info *ci = cap->ci;
1289 struct inode *inode = &ci->vfs_inode;
1290 struct cap_msg_args arg;
1296 held = cap->issued | cap->implemented;
1297 revoking = cap->implemented & ~cap->issued;
1298 retain &= ~revoking;
1300 dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
1301 inode, cap, cap->session,
1302 ceph_cap_string(held), ceph_cap_string(held & retain),
1303 ceph_cap_string(revoking));
1304 BUG_ON((retain & CEPH_CAP_PIN) == 0);
1306 arg.session = cap->session;
1308 /* don't release wanted unless we've waited a bit. */
1309 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1310 time_before(jiffies, ci->i_hold_caps_min)) {
1311 dout(" delaying issued %s -> %s, wanted %s -> %s on send\n",
1312 ceph_cap_string(cap->issued),
1313 ceph_cap_string(cap->issued & retain),
1314 ceph_cap_string(cap->mds_wanted),
1315 ceph_cap_string(want));
1316 want |= cap->mds_wanted;
1317 retain |= cap->issued;
1320 ci->i_ceph_flags &= ~(CEPH_I_NODELAY | CEPH_I_FLUSH);
1321 if (want & ~cap->mds_wanted) {
1322 /* user space may open/close single file frequently.
1323 * This avoids droping mds_wanted immediately after
1324 * requesting new mds_wanted.
1326 __cap_set_timeouts(mdsc, ci);
1329 cap->issued &= retain; /* drop bits we don't want */
1330 if (cap->implemented & ~cap->issued) {
1332 * Wake up any waiters on wanted -> needed transition.
1333 * This is due to the weird transition from buffered
1334 * to sync IO... we need to flush dirty pages _before_
1335 * allowing sync writes to avoid reordering.
1339 cap->implemented &= cap->issued | used;
1340 cap->mds_wanted = want;
1342 arg.ino = ceph_vino(inode).ino;
1343 arg.cid = cap->cap_id;
1344 arg.follows = flushing ? ci->i_head_snapc->seq : 0;
1345 arg.flush_tid = flush_tid;
1346 arg.oldest_flush_tid = oldest_flush_tid;
1348 arg.size = inode->i_size;
1349 ci->i_reported_size = arg.size;
1350 arg.max_size = ci->i_wanted_max_size;
1351 ci->i_requested_max_size = arg.max_size;
1353 if (flushing & CEPH_CAP_XATTR_EXCL) {
1354 __ceph_build_xattrs_blob(ci);
1355 arg.xattr_version = ci->i_xattrs.version;
1356 arg.xattr_buf = ci->i_xattrs.blob;
1358 arg.xattr_buf = NULL;
1361 arg.mtime = inode->i_mtime;
1362 arg.atime = inode->i_atime;
1363 arg.ctime = inode->i_ctime;
1366 arg.caps = cap->implemented;
1368 arg.dirty = flushing;
1371 arg.issue_seq = cap->issue_seq;
1372 arg.mseq = cap->mseq;
1373 arg.time_warp_seq = ci->i_time_warp_seq;
1375 arg.uid = inode->i_uid;
1376 arg.gid = inode->i_gid;
1377 arg.mode = inode->i_mode;
1379 arg.inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
1380 if (list_empty(&ci->i_cap_snaps))
1381 arg.flags = CEPH_CLIENT_CAPS_NO_CAPSNAP;
1383 arg.flags = CEPH_CLIENT_CAPS_PENDING_CAPSNAP;
1385 arg.flags |= CEPH_CLIENT_CAPS_SYNC;
1387 spin_unlock(&ci->i_ceph_lock);
1389 ret = send_cap_msg(&arg);
1391 dout("error sending cap msg, must requeue %p\n", inode);
1396 wake_up_all(&ci->i_cap_wq);
1401 static inline int __send_flush_snap(struct inode *inode,
1402 struct ceph_mds_session *session,
1403 struct ceph_cap_snap *capsnap,
1404 u32 mseq, u64 oldest_flush_tid)
1406 struct cap_msg_args arg;
1408 arg.session = session;
1409 arg.ino = ceph_vino(inode).ino;
1411 arg.follows = capsnap->follows;
1412 arg.flush_tid = capsnap->cap_flush.tid;
1413 arg.oldest_flush_tid = oldest_flush_tid;
1415 arg.size = capsnap->size;
1417 arg.xattr_version = capsnap->xattr_version;
1418 arg.xattr_buf = capsnap->xattr_blob;
1420 arg.atime = capsnap->atime;
1421 arg.mtime = capsnap->mtime;
1422 arg.ctime = capsnap->ctime;
1424 arg.op = CEPH_CAP_OP_FLUSHSNAP;
1425 arg.caps = capsnap->issued;
1427 arg.dirty = capsnap->dirty;
1432 arg.time_warp_seq = capsnap->time_warp_seq;
1434 arg.uid = capsnap->uid;
1435 arg.gid = capsnap->gid;
1436 arg.mode = capsnap->mode;
1438 arg.inline_data = capsnap->inline_data;
1441 return send_cap_msg(&arg);
1445 * When a snapshot is taken, clients accumulate dirty metadata on
1446 * inodes with capabilities in ceph_cap_snaps to describe the file
1447 * state at the time the snapshot was taken. This must be flushed
1448 * asynchronously back to the MDS once sync writes complete and dirty
1449 * data is written out.
1451 * Called under i_ceph_lock. Takes s_mutex as needed.
1453 static void __ceph_flush_snaps(struct ceph_inode_info *ci,
1454 struct ceph_mds_session *session)
1455 __releases(ci->i_ceph_lock)
1456 __acquires(ci->i_ceph_lock)
1458 struct inode *inode = &ci->vfs_inode;
1459 struct ceph_mds_client *mdsc = session->s_mdsc;
1460 struct ceph_cap_snap *capsnap;
1461 u64 oldest_flush_tid = 0;
1462 u64 first_tid = 1, last_tid = 0;
1464 dout("__flush_snaps %p session %p\n", inode, session);
1466 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
1468 * we need to wait for sync writes to complete and for dirty
1469 * pages to be written out.
1471 if (capsnap->dirty_pages || capsnap->writing)
1474 /* should be removed by ceph_try_drop_cap_snap() */
1475 BUG_ON(!capsnap->need_flush);
1477 /* only flush each capsnap once */
1478 if (capsnap->cap_flush.tid > 0) {
1479 dout(" already flushed %p, skipping\n", capsnap);
1483 spin_lock(&mdsc->cap_dirty_lock);
1484 capsnap->cap_flush.tid = ++mdsc->last_cap_flush_tid;
1485 list_add_tail(&capsnap->cap_flush.g_list,
1486 &mdsc->cap_flush_list);
1487 if (oldest_flush_tid == 0)
1488 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1489 if (list_empty(&ci->i_flushing_item)) {
1490 list_add_tail(&ci->i_flushing_item,
1491 &session->s_cap_flushing);
1493 spin_unlock(&mdsc->cap_dirty_lock);
1495 list_add_tail(&capsnap->cap_flush.i_list,
1496 &ci->i_cap_flush_list);
1499 first_tid = capsnap->cap_flush.tid;
1500 last_tid = capsnap->cap_flush.tid;
1503 ci->i_ceph_flags &= ~CEPH_I_FLUSH_SNAPS;
1505 while (first_tid <= last_tid) {
1506 struct ceph_cap *cap = ci->i_auth_cap;
1507 struct ceph_cap_flush *cf;
1510 if (!(cap && cap->session == session)) {
1511 dout("__flush_snaps %p auth cap %p not mds%d, "
1512 "stop\n", inode, cap, session->s_mds);
1517 list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
1518 if (cf->tid >= first_tid) {
1526 first_tid = cf->tid + 1;
1528 capsnap = container_of(cf, struct ceph_cap_snap, cap_flush);
1529 refcount_inc(&capsnap->nref);
1530 spin_unlock(&ci->i_ceph_lock);
1532 dout("__flush_snaps %p capsnap %p tid %llu %s\n",
1533 inode, capsnap, cf->tid, ceph_cap_string(capsnap->dirty));
1535 ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
1538 pr_err("__flush_snaps: error sending cap flushsnap, "
1539 "ino (%llx.%llx) tid %llu follows %llu\n",
1540 ceph_vinop(inode), cf->tid, capsnap->follows);
1543 ceph_put_cap_snap(capsnap);
1544 spin_lock(&ci->i_ceph_lock);
1548 void ceph_flush_snaps(struct ceph_inode_info *ci,
1549 struct ceph_mds_session **psession)
1551 struct inode *inode = &ci->vfs_inode;
1552 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1553 struct ceph_mds_session *session = NULL;
1556 dout("ceph_flush_snaps %p\n", inode);
1558 session = *psession;
1560 spin_lock(&ci->i_ceph_lock);
1561 if (!(ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)) {
1562 dout(" no capsnap needs flush, doing nothing\n");
1565 if (!ci->i_auth_cap) {
1566 dout(" no auth cap (migrating?), doing nothing\n");
1570 mds = ci->i_auth_cap->session->s_mds;
1571 if (session && session->s_mds != mds) {
1572 dout(" oops, wrong session %p mutex\n", session);
1573 mutex_unlock(&session->s_mutex);
1574 ceph_put_mds_session(session);
1578 spin_unlock(&ci->i_ceph_lock);
1579 mutex_lock(&mdsc->mutex);
1580 session = __ceph_lookup_mds_session(mdsc, mds);
1581 mutex_unlock(&mdsc->mutex);
1583 dout(" inverting session/ino locks on %p\n", session);
1584 mutex_lock(&session->s_mutex);
1589 // make sure flushsnap messages are sent in proper order.
1590 if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
1591 __kick_flushing_caps(mdsc, session, ci, 0);
1592 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
1595 __ceph_flush_snaps(ci, session);
1597 spin_unlock(&ci->i_ceph_lock);
1600 *psession = session;
1601 } else if (session) {
1602 mutex_unlock(&session->s_mutex);
1603 ceph_put_mds_session(session);
1605 /* we flushed them all; remove this inode from the queue */
1606 spin_lock(&mdsc->snap_flush_lock);
1607 list_del_init(&ci->i_snap_flush_item);
1608 spin_unlock(&mdsc->snap_flush_lock);
1612 * Mark caps dirty. If inode is newly dirty, return the dirty flags.
1613 * Caller is then responsible for calling __mark_inode_dirty with the
1614 * returned flags value.
1616 int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
1617 struct ceph_cap_flush **pcf)
1619 struct ceph_mds_client *mdsc =
1620 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
1621 struct inode *inode = &ci->vfs_inode;
1622 int was = ci->i_dirty_caps;
1625 if (!ci->i_auth_cap) {
1626 pr_warn("__mark_dirty_caps %p %llx mask %s, "
1627 "but no auth cap (session was closed?)\n",
1628 inode, ceph_ino(inode), ceph_cap_string(mask));
1632 dout("__mark_dirty_caps %p %s dirty %s -> %s\n", &ci->vfs_inode,
1633 ceph_cap_string(mask), ceph_cap_string(was),
1634 ceph_cap_string(was | mask));
1635 ci->i_dirty_caps |= mask;
1637 WARN_ON_ONCE(ci->i_prealloc_cap_flush);
1638 swap(ci->i_prealloc_cap_flush, *pcf);
1640 if (!ci->i_head_snapc) {
1641 WARN_ON_ONCE(!rwsem_is_locked(&mdsc->snap_rwsem));
1642 ci->i_head_snapc = ceph_get_snap_context(
1643 ci->i_snap_realm->cached_context);
1645 dout(" inode %p now dirty snapc %p auth cap %p\n",
1646 &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
1647 BUG_ON(!list_empty(&ci->i_dirty_item));
1648 spin_lock(&mdsc->cap_dirty_lock);
1649 list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
1650 spin_unlock(&mdsc->cap_dirty_lock);
1651 if (ci->i_flushing_caps == 0) {
1653 dirty |= I_DIRTY_SYNC;
1656 WARN_ON_ONCE(!ci->i_prealloc_cap_flush);
1658 BUG_ON(list_empty(&ci->i_dirty_item));
1659 if (((was | ci->i_flushing_caps) & CEPH_CAP_FILE_BUFFER) &&
1660 (mask & CEPH_CAP_FILE_BUFFER))
1661 dirty |= I_DIRTY_DATASYNC;
1662 __cap_delay_requeue(mdsc, ci);
1666 struct ceph_cap_flush *ceph_alloc_cap_flush(void)
1668 return kmem_cache_alloc(ceph_cap_flush_cachep, GFP_KERNEL);
1671 void ceph_free_cap_flush(struct ceph_cap_flush *cf)
1674 kmem_cache_free(ceph_cap_flush_cachep, cf);
1677 static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc)
1679 if (!list_empty(&mdsc->cap_flush_list)) {
1680 struct ceph_cap_flush *cf =
1681 list_first_entry(&mdsc->cap_flush_list,
1682 struct ceph_cap_flush, g_list);
1689 * Remove cap_flush from the mdsc's or inode's flushing cap list.
1690 * Return true if caller needs to wake up flush waiters.
1692 static bool __finish_cap_flush(struct ceph_mds_client *mdsc,
1693 struct ceph_inode_info *ci,
1694 struct ceph_cap_flush *cf)
1696 struct ceph_cap_flush *prev;
1697 bool wake = cf->wake;
1699 /* are there older pending cap flushes? */
1700 if (wake && cf->g_list.prev != &mdsc->cap_flush_list) {
1701 prev = list_prev_entry(cf, g_list);
1705 list_del(&cf->g_list);
1707 if (wake && cf->i_list.prev != &ci->i_cap_flush_list) {
1708 prev = list_prev_entry(cf, i_list);
1712 list_del(&cf->i_list);
1720 * Add dirty inode to the flushing list. Assigned a seq number so we
1721 * can wait for caps to flush without starving.
1723 * Called under i_ceph_lock.
1725 static int __mark_caps_flushing(struct inode *inode,
1726 struct ceph_mds_session *session, bool wake,
1727 u64 *flush_tid, u64 *oldest_flush_tid)
1729 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1730 struct ceph_inode_info *ci = ceph_inode(inode);
1731 struct ceph_cap_flush *cf = NULL;
1734 BUG_ON(ci->i_dirty_caps == 0);
1735 BUG_ON(list_empty(&ci->i_dirty_item));
1736 BUG_ON(!ci->i_prealloc_cap_flush);
1738 flushing = ci->i_dirty_caps;
1739 dout("__mark_caps_flushing flushing %s, flushing_caps %s -> %s\n",
1740 ceph_cap_string(flushing),
1741 ceph_cap_string(ci->i_flushing_caps),
1742 ceph_cap_string(ci->i_flushing_caps | flushing));
1743 ci->i_flushing_caps |= flushing;
1744 ci->i_dirty_caps = 0;
1745 dout(" inode %p now !dirty\n", inode);
1747 swap(cf, ci->i_prealloc_cap_flush);
1748 cf->caps = flushing;
1751 spin_lock(&mdsc->cap_dirty_lock);
1752 list_del_init(&ci->i_dirty_item);
1754 cf->tid = ++mdsc->last_cap_flush_tid;
1755 list_add_tail(&cf->g_list, &mdsc->cap_flush_list);
1756 *oldest_flush_tid = __get_oldest_flush_tid(mdsc);
1758 if (list_empty(&ci->i_flushing_item)) {
1759 list_add_tail(&ci->i_flushing_item, &session->s_cap_flushing);
1760 mdsc->num_cap_flushing++;
1762 spin_unlock(&mdsc->cap_dirty_lock);
1764 list_add_tail(&cf->i_list, &ci->i_cap_flush_list);
1766 *flush_tid = cf->tid;
1771 * try to invalidate mapping pages without blocking.
1773 static int try_nonblocking_invalidate(struct inode *inode)
1775 struct ceph_inode_info *ci = ceph_inode(inode);
1776 u32 invalidating_gen = ci->i_rdcache_gen;
1778 spin_unlock(&ci->i_ceph_lock);
1779 invalidate_mapping_pages(&inode->i_data, 0, -1);
1780 spin_lock(&ci->i_ceph_lock);
1782 if (inode->i_data.nrpages == 0 &&
1783 invalidating_gen == ci->i_rdcache_gen) {
1785 dout("try_nonblocking_invalidate %p success\n", inode);
1786 /* save any racing async invalidate some trouble */
1787 ci->i_rdcache_revoking = ci->i_rdcache_gen - 1;
1790 dout("try_nonblocking_invalidate %p failed\n", inode);
1794 bool __ceph_should_report_size(struct ceph_inode_info *ci)
1796 loff_t size = ci->vfs_inode.i_size;
1797 /* mds will adjust max size according to the reported size */
1798 if (ci->i_flushing_caps & CEPH_CAP_FILE_WR)
1800 if (size >= ci->i_max_size)
1802 /* half of previous max_size increment has been used */
1803 if (ci->i_max_size > ci->i_reported_size &&
1804 (size << 1) >= ci->i_max_size + ci->i_reported_size)
1810 * Swiss army knife function to examine currently used and wanted
1811 * versus held caps. Release, flush, ack revoked caps to mds as
1814 * CHECK_CAPS_NODELAY - caller is delayed work and we should not delay
1815 * cap release further.
1816 * CHECK_CAPS_AUTHONLY - we should only check the auth cap
1817 * CHECK_CAPS_FLUSH - we should flush any dirty caps immediately, without
1820 void ceph_check_caps(struct ceph_inode_info *ci, int flags,
1821 struct ceph_mds_session *session)
1823 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1824 struct ceph_mds_client *mdsc = fsc->mdsc;
1825 struct inode *inode = &ci->vfs_inode;
1826 struct ceph_cap *cap;
1827 u64 flush_tid, oldest_flush_tid;
1828 int file_wanted, used, cap_used;
1829 int took_snap_rwsem = 0; /* true if mdsc->snap_rwsem held */
1830 int issued, implemented, want, retain, revoking, flushing = 0;
1831 int mds = -1; /* keep track of how far we've gone through i_caps list
1832 to avoid an infinite loop on retry */
1834 int delayed = 0, sent = 0;
1835 bool no_delay = flags & CHECK_CAPS_NODELAY;
1836 bool queue_invalidate = false;
1837 bool tried_invalidate = false;
1839 /* if we are unmounting, flush any unused caps immediately. */
1843 spin_lock(&ci->i_ceph_lock);
1845 if (ci->i_ceph_flags & CEPH_I_FLUSH)
1846 flags |= CHECK_CAPS_FLUSH;
1848 if (!(flags & CHECK_CAPS_AUTHONLY) ||
1849 (ci->i_auth_cap && __ceph_is_single_caps(ci)))
1850 __cap_delay_cancel(mdsc, ci);
1854 spin_lock(&ci->i_ceph_lock);
1856 file_wanted = __ceph_caps_file_wanted(ci);
1857 used = __ceph_caps_used(ci);
1858 issued = __ceph_caps_issued(ci, &implemented);
1859 revoking = implemented & ~issued;
1862 retain = file_wanted | used | CEPH_CAP_PIN;
1863 if (!mdsc->stopping && inode->i_nlink > 0) {
1865 retain |= CEPH_CAP_ANY; /* be greedy */
1866 } else if (S_ISDIR(inode->i_mode) &&
1867 (issued & CEPH_CAP_FILE_SHARED) &&
1868 __ceph_dir_is_complete(ci)) {
1870 * If a directory is complete, we want to keep
1871 * the exclusive cap. So that MDS does not end up
1872 * revoking the shared cap on every create/unlink
1875 want = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL;
1879 retain |= CEPH_CAP_ANY_SHARED;
1881 * keep RD only if we didn't have the file open RW,
1882 * because then the mds would revoke it anyway to
1883 * journal max_size=0.
1885 if (ci->i_max_size == 0)
1886 retain |= CEPH_CAP_ANY_RD;
1890 dout("check_caps %p file_want %s used %s dirty %s flushing %s"
1891 " issued %s revoking %s retain %s %s%s%s\n", inode,
1892 ceph_cap_string(file_wanted),
1893 ceph_cap_string(used), ceph_cap_string(ci->i_dirty_caps),
1894 ceph_cap_string(ci->i_flushing_caps),
1895 ceph_cap_string(issued), ceph_cap_string(revoking),
1896 ceph_cap_string(retain),
1897 (flags & CHECK_CAPS_AUTHONLY) ? " AUTHONLY" : "",
1898 (flags & CHECK_CAPS_NODELAY) ? " NODELAY" : "",
1899 (flags & CHECK_CAPS_FLUSH) ? " FLUSH" : "");
1902 * If we no longer need to hold onto old our caps, and we may
1903 * have cached pages, but don't want them, then try to invalidate.
1904 * If we fail, it's because pages are locked.... try again later.
1906 if ((!no_delay || mdsc->stopping) &&
1907 !S_ISDIR(inode->i_mode) && /* ignore readdir cache */
1908 !(ci->i_wb_ref || ci->i_wrbuffer_ref) && /* no dirty pages... */
1909 inode->i_data.nrpages && /* have cached pages */
1910 (revoking & (CEPH_CAP_FILE_CACHE|
1911 CEPH_CAP_FILE_LAZYIO)) && /* or revoking cache */
1912 !tried_invalidate) {
1913 dout("check_caps trying to invalidate on %p\n", inode);
1914 if (try_nonblocking_invalidate(inode) < 0) {
1915 dout("check_caps queuing invalidate\n");
1916 queue_invalidate = true;
1917 ci->i_rdcache_revoking = ci->i_rdcache_gen;
1919 tried_invalidate = true;
1923 for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
1924 cap = rb_entry(p, struct ceph_cap, ci_node);
1926 /* avoid looping forever */
1927 if (mds >= cap->mds ||
1928 ((flags & CHECK_CAPS_AUTHONLY) && cap != ci->i_auth_cap))
1931 /* NOTE: no side-effects allowed, until we take s_mutex */
1934 if (ci->i_auth_cap && cap != ci->i_auth_cap)
1935 cap_used &= ~ci->i_auth_cap->issued;
1937 revoking = cap->implemented & ~cap->issued;
1938 dout(" mds%d cap %p used %s issued %s implemented %s revoking %s\n",
1939 cap->mds, cap, ceph_cap_string(cap_used),
1940 ceph_cap_string(cap->issued),
1941 ceph_cap_string(cap->implemented),
1942 ceph_cap_string(revoking));
1944 if (cap == ci->i_auth_cap &&
1945 (cap->issued & CEPH_CAP_FILE_WR)) {
1946 /* request larger max_size from MDS? */
1947 if (ci->i_wanted_max_size > ci->i_max_size &&
1948 ci->i_wanted_max_size > ci->i_requested_max_size) {
1949 dout("requesting new max_size\n");
1953 /* approaching file_max? */
1954 if (__ceph_should_report_size(ci)) {
1955 dout("i_size approaching max_size\n");
1959 /* flush anything dirty? */
1960 if (cap == ci->i_auth_cap) {
1961 if ((flags & CHECK_CAPS_FLUSH) && ci->i_dirty_caps) {
1962 dout("flushing dirty caps\n");
1965 if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS) {
1966 dout("flushing snap caps\n");
1971 /* completed revocation? going down and there are no caps? */
1972 if (revoking && (revoking & cap_used) == 0) {
1973 dout("completed revocation of %s\n",
1974 ceph_cap_string(cap->implemented & ~cap->issued));
1978 /* want more caps from mds? */
1979 if (want & ~(cap->mds_wanted | cap->issued))
1982 /* things we might delay */
1983 if ((cap->issued & ~retain) == 0 &&
1984 cap->mds_wanted == want)
1985 continue; /* nope, all good */
1991 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0 &&
1992 time_before(jiffies, ci->i_hold_caps_max)) {
1993 dout(" delaying issued %s -> %s, wanted %s -> %s\n",
1994 ceph_cap_string(cap->issued),
1995 ceph_cap_string(cap->issued & retain),
1996 ceph_cap_string(cap->mds_wanted),
1997 ceph_cap_string(want));
2003 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
2004 dout(" skipping %p I_NOFLUSH set\n", inode);
2008 if (session && session != cap->session) {
2009 dout("oops, wrong session %p mutex\n", session);
2010 mutex_unlock(&session->s_mutex);
2014 session = cap->session;
2015 if (mutex_trylock(&session->s_mutex) == 0) {
2016 dout("inverting session/ino locks on %p\n",
2018 spin_unlock(&ci->i_ceph_lock);
2019 if (took_snap_rwsem) {
2020 up_read(&mdsc->snap_rwsem);
2021 took_snap_rwsem = 0;
2023 mutex_lock(&session->s_mutex);
2028 /* kick flushing and flush snaps before sending normal
2030 if (cap == ci->i_auth_cap &&
2032 (CEPH_I_KICK_FLUSH | CEPH_I_FLUSH_SNAPS))) {
2033 if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
2034 __kick_flushing_caps(mdsc, session, ci, 0);
2035 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2037 if (ci->i_ceph_flags & CEPH_I_FLUSH_SNAPS)
2038 __ceph_flush_snaps(ci, session);
2043 /* take snap_rwsem after session mutex */
2044 if (!took_snap_rwsem) {
2045 if (down_read_trylock(&mdsc->snap_rwsem) == 0) {
2046 dout("inverting snap/in locks on %p\n",
2048 spin_unlock(&ci->i_ceph_lock);
2049 down_read(&mdsc->snap_rwsem);
2050 took_snap_rwsem = 1;
2053 took_snap_rwsem = 1;
2056 if (cap == ci->i_auth_cap && ci->i_dirty_caps) {
2057 flushing = __mark_caps_flushing(inode, session, false,
2063 spin_lock(&mdsc->cap_dirty_lock);
2064 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2065 spin_unlock(&mdsc->cap_dirty_lock);
2068 mds = cap->mds; /* remember mds, so we don't repeat */
2071 /* __send_cap drops i_ceph_lock */
2072 delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, false,
2073 cap_used, want, retain, flushing,
2074 flush_tid, oldest_flush_tid);
2075 goto retry; /* retake i_ceph_lock and restart our cap scan. */
2078 /* Reschedule delayed caps release if we delayed anything */
2080 __cap_delay_requeue(mdsc, ci);
2082 spin_unlock(&ci->i_ceph_lock);
2084 if (queue_invalidate)
2085 ceph_queue_invalidate(inode);
2088 mutex_unlock(&session->s_mutex);
2089 if (took_snap_rwsem)
2090 up_read(&mdsc->snap_rwsem);
2094 * Try to flush dirty caps back to the auth mds.
2096 static int try_flush_caps(struct inode *inode, u64 *ptid)
2098 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2099 struct ceph_inode_info *ci = ceph_inode(inode);
2100 struct ceph_mds_session *session = NULL;
2102 u64 flush_tid = 0, oldest_flush_tid = 0;
2105 spin_lock(&ci->i_ceph_lock);
2106 if (ci->i_ceph_flags & CEPH_I_NOFLUSH) {
2107 spin_unlock(&ci->i_ceph_lock);
2108 dout("try_flush_caps skipping %p I_NOFLUSH set\n", inode);
2111 if (ci->i_dirty_caps && ci->i_auth_cap) {
2112 struct ceph_cap *cap = ci->i_auth_cap;
2113 int used = __ceph_caps_used(ci);
2114 int want = __ceph_caps_wanted(ci);
2117 if (!session || session != cap->session) {
2118 spin_unlock(&ci->i_ceph_lock);
2120 mutex_unlock(&session->s_mutex);
2121 session = cap->session;
2122 mutex_lock(&session->s_mutex);
2125 if (cap->session->s_state < CEPH_MDS_SESSION_OPEN) {
2126 spin_unlock(&ci->i_ceph_lock);
2130 flushing = __mark_caps_flushing(inode, session, true,
2131 &flush_tid, &oldest_flush_tid);
2133 /* __send_cap drops i_ceph_lock */
2134 delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, true,
2135 used, want, (cap->issued | cap->implemented),
2136 flushing, flush_tid, oldest_flush_tid);
2139 spin_lock(&ci->i_ceph_lock);
2140 __cap_delay_requeue(mdsc, ci);
2141 spin_unlock(&ci->i_ceph_lock);
2144 if (!list_empty(&ci->i_cap_flush_list)) {
2145 struct ceph_cap_flush *cf =
2146 list_last_entry(&ci->i_cap_flush_list,
2147 struct ceph_cap_flush, i_list);
2149 flush_tid = cf->tid;
2151 flushing = ci->i_flushing_caps;
2152 spin_unlock(&ci->i_ceph_lock);
2156 mutex_unlock(&session->s_mutex);
2163 * Return true if we've flushed caps through the given flush_tid.
2165 static int caps_are_flushed(struct inode *inode, u64 flush_tid)
2167 struct ceph_inode_info *ci = ceph_inode(inode);
2170 spin_lock(&ci->i_ceph_lock);
2171 if (!list_empty(&ci->i_cap_flush_list)) {
2172 struct ceph_cap_flush * cf =
2173 list_first_entry(&ci->i_cap_flush_list,
2174 struct ceph_cap_flush, i_list);
2175 if (cf->tid <= flush_tid)
2178 spin_unlock(&ci->i_ceph_lock);
2183 * wait for any unsafe requests to complete.
2185 static int unsafe_request_wait(struct inode *inode)
2187 struct ceph_inode_info *ci = ceph_inode(inode);
2188 struct ceph_mds_request *req1 = NULL, *req2 = NULL;
2191 spin_lock(&ci->i_unsafe_lock);
2192 if (S_ISDIR(inode->i_mode) && !list_empty(&ci->i_unsafe_dirops)) {
2193 req1 = list_last_entry(&ci->i_unsafe_dirops,
2194 struct ceph_mds_request,
2196 ceph_mdsc_get_request(req1);
2198 if (!list_empty(&ci->i_unsafe_iops)) {
2199 req2 = list_last_entry(&ci->i_unsafe_iops,
2200 struct ceph_mds_request,
2201 r_unsafe_target_item);
2202 ceph_mdsc_get_request(req2);
2204 spin_unlock(&ci->i_unsafe_lock);
2206 dout("unsafe_request_wait %p wait on tid %llu %llu\n",
2207 inode, req1 ? req1->r_tid : 0ULL, req2 ? req2->r_tid : 0ULL);
2209 ret = !wait_for_completion_timeout(&req1->r_safe_completion,
2210 ceph_timeout_jiffies(req1->r_timeout));
2213 ceph_mdsc_put_request(req1);
2216 ret = !wait_for_completion_timeout(&req2->r_safe_completion,
2217 ceph_timeout_jiffies(req2->r_timeout));
2220 ceph_mdsc_put_request(req2);
2225 int ceph_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2227 struct inode *inode = file->f_mapping->host;
2228 struct ceph_inode_info *ci = ceph_inode(inode);
2233 dout("fsync %p%s\n", inode, datasync ? " datasync" : "");
2235 ret = file_write_and_wait_range(file, start, end);
2244 dirty = try_flush_caps(inode, &flush_tid);
2245 dout("fsync dirty caps are %s\n", ceph_cap_string(dirty));
2247 ret = unsafe_request_wait(inode);
2250 * only wait on non-file metadata writeback (the mds
2251 * can recover size and mtime, so we don't need to
2254 if (!ret && (dirty & ~CEPH_CAP_ANY_FILE_WR)) {
2255 ret = wait_event_interruptible(ci->i_cap_wq,
2256 caps_are_flushed(inode, flush_tid));
2258 inode_unlock(inode);
2260 dout("fsync %p%s result=%d\n", inode, datasync ? " datasync" : "", ret);
2265 * Flush any dirty caps back to the mds. If we aren't asked to wait,
2266 * queue inode for flush but don't do so immediately, because we can
2267 * get by with fewer MDS messages if we wait for data writeback to
2270 int ceph_write_inode(struct inode *inode, struct writeback_control *wbc)
2272 struct ceph_inode_info *ci = ceph_inode(inode);
2276 int wait = (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync);
2278 dout("write_inode %p wait=%d\n", inode, wait);
2280 dirty = try_flush_caps(inode, &flush_tid);
2282 err = wait_event_interruptible(ci->i_cap_wq,
2283 caps_are_flushed(inode, flush_tid));
2285 struct ceph_mds_client *mdsc =
2286 ceph_sb_to_client(inode->i_sb)->mdsc;
2288 spin_lock(&ci->i_ceph_lock);
2289 if (__ceph_caps_dirty(ci))
2290 __cap_delay_requeue_front(mdsc, ci);
2291 spin_unlock(&ci->i_ceph_lock);
2296 static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
2297 struct ceph_mds_session *session,
2298 struct ceph_inode_info *ci,
2299 u64 oldest_flush_tid)
2300 __releases(ci->i_ceph_lock)
2301 __acquires(ci->i_ceph_lock)
2303 struct inode *inode = &ci->vfs_inode;
2304 struct ceph_cap *cap;
2305 struct ceph_cap_flush *cf;
2309 list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
2310 if (cf->tid < first_tid)
2313 cap = ci->i_auth_cap;
2314 if (!(cap && cap->session == session)) {
2315 pr_err("%p auth cap %p not mds%d ???\n",
2316 inode, cap, session->s_mds);
2320 first_tid = cf->tid + 1;
2323 dout("kick_flushing_caps %p cap %p tid %llu %s\n",
2324 inode, cap, cf->tid, ceph_cap_string(cf->caps));
2325 ci->i_ceph_flags |= CEPH_I_NODELAY;
2326 ret = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
2327 false, __ceph_caps_used(ci),
2328 __ceph_caps_wanted(ci),
2329 cap->issued | cap->implemented,
2330 cf->caps, cf->tid, oldest_flush_tid);
2332 pr_err("kick_flushing_caps: error sending "
2333 "cap flush, ino (%llx.%llx) "
2334 "tid %llu flushing %s\n",
2335 ceph_vinop(inode), cf->tid,
2336 ceph_cap_string(cf->caps));
2339 struct ceph_cap_snap *capsnap =
2340 container_of(cf, struct ceph_cap_snap,
2342 dout("kick_flushing_caps %p capsnap %p tid %llu %s\n",
2343 inode, capsnap, cf->tid,
2344 ceph_cap_string(capsnap->dirty));
2346 refcount_inc(&capsnap->nref);
2347 spin_unlock(&ci->i_ceph_lock);
2349 ret = __send_flush_snap(inode, session, capsnap, cap->mseq,
2352 pr_err("kick_flushing_caps: error sending "
2353 "cap flushsnap, ino (%llx.%llx) "
2354 "tid %llu follows %llu\n",
2355 ceph_vinop(inode), cf->tid,
2359 ceph_put_cap_snap(capsnap);
2362 spin_lock(&ci->i_ceph_lock);
2366 void ceph_early_kick_flushing_caps(struct ceph_mds_client *mdsc,
2367 struct ceph_mds_session *session)
2369 struct ceph_inode_info *ci;
2370 struct ceph_cap *cap;
2371 u64 oldest_flush_tid;
2373 dout("early_kick_flushing_caps mds%d\n", session->s_mds);
2375 spin_lock(&mdsc->cap_dirty_lock);
2376 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2377 spin_unlock(&mdsc->cap_dirty_lock);
2379 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2380 spin_lock(&ci->i_ceph_lock);
2381 cap = ci->i_auth_cap;
2382 if (!(cap && cap->session == session)) {
2383 pr_err("%p auth cap %p not mds%d ???\n",
2384 &ci->vfs_inode, cap, session->s_mds);
2385 spin_unlock(&ci->i_ceph_lock);
2391 * if flushing caps were revoked, we re-send the cap flush
2392 * in client reconnect stage. This guarantees MDS * processes
2393 * the cap flush message before issuing the flushing caps to
2396 if ((cap->issued & ci->i_flushing_caps) !=
2397 ci->i_flushing_caps) {
2398 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2399 __kick_flushing_caps(mdsc, session, ci,
2402 ci->i_ceph_flags |= CEPH_I_KICK_FLUSH;
2405 spin_unlock(&ci->i_ceph_lock);
2409 void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
2410 struct ceph_mds_session *session)
2412 struct ceph_inode_info *ci;
2413 struct ceph_cap *cap;
2414 u64 oldest_flush_tid;
2416 dout("kick_flushing_caps mds%d\n", session->s_mds);
2418 spin_lock(&mdsc->cap_dirty_lock);
2419 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2420 spin_unlock(&mdsc->cap_dirty_lock);
2422 list_for_each_entry(ci, &session->s_cap_flushing, i_flushing_item) {
2423 spin_lock(&ci->i_ceph_lock);
2424 cap = ci->i_auth_cap;
2425 if (!(cap && cap->session == session)) {
2426 pr_err("%p auth cap %p not mds%d ???\n",
2427 &ci->vfs_inode, cap, session->s_mds);
2428 spin_unlock(&ci->i_ceph_lock);
2431 if (ci->i_ceph_flags & CEPH_I_KICK_FLUSH) {
2432 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2433 __kick_flushing_caps(mdsc, session, ci,
2436 spin_unlock(&ci->i_ceph_lock);
2440 static void kick_flushing_inode_caps(struct ceph_mds_client *mdsc,
2441 struct ceph_mds_session *session,
2442 struct inode *inode)
2443 __releases(ci->i_ceph_lock)
2445 struct ceph_inode_info *ci = ceph_inode(inode);
2446 struct ceph_cap *cap;
2448 cap = ci->i_auth_cap;
2449 dout("kick_flushing_inode_caps %p flushing %s\n", inode,
2450 ceph_cap_string(ci->i_flushing_caps));
2452 if (!list_empty(&ci->i_cap_flush_list)) {
2453 u64 oldest_flush_tid;
2454 spin_lock(&mdsc->cap_dirty_lock);
2455 list_move_tail(&ci->i_flushing_item,
2456 &cap->session->s_cap_flushing);
2457 oldest_flush_tid = __get_oldest_flush_tid(mdsc);
2458 spin_unlock(&mdsc->cap_dirty_lock);
2460 ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
2461 __kick_flushing_caps(mdsc, session, ci, oldest_flush_tid);
2462 spin_unlock(&ci->i_ceph_lock);
2464 spin_unlock(&ci->i_ceph_lock);
2470 * Take references to capabilities we hold, so that we don't release
2471 * them to the MDS prematurely.
2473 * Protected by i_ceph_lock.
2475 static void __take_cap_refs(struct ceph_inode_info *ci, int got,
2476 bool snap_rwsem_locked)
2478 if (got & CEPH_CAP_PIN)
2480 if (got & CEPH_CAP_FILE_RD)
2482 if (got & CEPH_CAP_FILE_CACHE)
2483 ci->i_rdcache_ref++;
2484 if (got & CEPH_CAP_FILE_WR) {
2485 if (ci->i_wr_ref == 0 && !ci->i_head_snapc) {
2486 BUG_ON(!snap_rwsem_locked);
2487 ci->i_head_snapc = ceph_get_snap_context(
2488 ci->i_snap_realm->cached_context);
2492 if (got & CEPH_CAP_FILE_BUFFER) {
2493 if (ci->i_wb_ref == 0)
2494 ihold(&ci->vfs_inode);
2496 dout("__take_cap_refs %p wb %d -> %d (?)\n",
2497 &ci->vfs_inode, ci->i_wb_ref-1, ci->i_wb_ref);
2502 * Try to grab cap references. Specify those refs we @want, and the
2503 * minimal set we @need. Also include the larger offset we are writing
2504 * to (when applicable), and check against max_size here as well.
2505 * Note that caller is responsible for ensuring max_size increases are
2506 * requested from the MDS.
2508 static int try_get_cap_refs(struct ceph_inode_info *ci, int need, int want,
2509 loff_t endoff, bool nonblock, int *got, int *err)
2511 struct inode *inode = &ci->vfs_inode;
2512 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
2514 int have, implemented;
2516 bool snap_rwsem_locked = false;
2518 dout("get_cap_refs %p need %s want %s\n", inode,
2519 ceph_cap_string(need), ceph_cap_string(want));
2522 spin_lock(&ci->i_ceph_lock);
2524 /* make sure file is actually open */
2525 file_wanted = __ceph_caps_file_wanted(ci);
2526 if ((file_wanted & need) != need) {
2527 dout("try_get_cap_refs need %s file_wanted %s, EBADF\n",
2528 ceph_cap_string(need), ceph_cap_string(file_wanted));
2534 /* finish pending truncate */
2535 while (ci->i_truncate_pending) {
2536 spin_unlock(&ci->i_ceph_lock);
2537 if (snap_rwsem_locked) {
2538 up_read(&mdsc->snap_rwsem);
2539 snap_rwsem_locked = false;
2541 __ceph_do_pending_vmtruncate(inode);
2542 spin_lock(&ci->i_ceph_lock);
2545 have = __ceph_caps_issued(ci, &implemented);
2547 if (have & need & CEPH_CAP_FILE_WR) {
2548 if (endoff >= 0 && endoff > (loff_t)ci->i_max_size) {
2549 dout("get_cap_refs %p endoff %llu > maxsize %llu\n",
2550 inode, endoff, ci->i_max_size);
2551 if (endoff > ci->i_requested_max_size) {
2558 * If a sync write is in progress, we must wait, so that we
2559 * can get a final snapshot value for size+mtime.
2561 if (__ceph_have_pending_cap_snap(ci)) {
2562 dout("get_cap_refs %p cap_snap_pending\n", inode);
2567 if ((have & need) == need) {
2569 * Look at (implemented & ~have & not) so that we keep waiting
2570 * on transition from wanted -> needed caps. This is needed
2571 * for WRBUFFER|WR -> WR to avoid a new WR sync write from
2572 * going before a prior buffered writeback happens.
2574 int not = want & ~(have & need);
2575 int revoking = implemented & ~have;
2576 dout("get_cap_refs %p have %s but not %s (revoking %s)\n",
2577 inode, ceph_cap_string(have), ceph_cap_string(not),
2578 ceph_cap_string(revoking));
2579 if ((revoking & not) == 0) {
2580 if (!snap_rwsem_locked &&
2581 !ci->i_head_snapc &&
2582 (need & CEPH_CAP_FILE_WR)) {
2583 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2585 * we can not call down_read() when
2586 * task isn't in TASK_RUNNING state
2594 spin_unlock(&ci->i_ceph_lock);
2595 down_read(&mdsc->snap_rwsem);
2596 snap_rwsem_locked = true;
2599 snap_rwsem_locked = true;
2601 *got = need | (have & want);
2602 if ((need & CEPH_CAP_FILE_RD) &&
2603 !(*got & CEPH_CAP_FILE_CACHE))
2604 ceph_disable_fscache_readpage(ci);
2605 __take_cap_refs(ci, *got, true);
2609 int session_readonly = false;
2610 if ((need & CEPH_CAP_FILE_WR) && ci->i_auth_cap) {
2611 struct ceph_mds_session *s = ci->i_auth_cap->session;
2612 spin_lock(&s->s_cap_lock);
2613 session_readonly = s->s_readonly;
2614 spin_unlock(&s->s_cap_lock);
2616 if (session_readonly) {
2617 dout("get_cap_refs %p needed %s but mds%d readonly\n",
2618 inode, ceph_cap_string(need), ci->i_auth_cap->mds);
2624 if (ci->i_ceph_flags & CEPH_I_CAP_DROPPED) {
2626 if (READ_ONCE(mdsc->fsc->mount_state) ==
2627 CEPH_MOUNT_SHUTDOWN) {
2628 dout("get_cap_refs %p forced umount\n", inode);
2633 mds_wanted = __ceph_caps_mds_wanted(ci, false);
2634 if (need & ~(mds_wanted & need)) {
2635 dout("get_cap_refs %p caps were dropped"
2636 " (session killed?)\n", inode);
2641 if (!(file_wanted & ~mds_wanted))
2642 ci->i_ceph_flags &= ~CEPH_I_CAP_DROPPED;
2645 dout("get_cap_refs %p have %s needed %s\n", inode,
2646 ceph_cap_string(have), ceph_cap_string(need));
2649 spin_unlock(&ci->i_ceph_lock);
2650 if (snap_rwsem_locked)
2651 up_read(&mdsc->snap_rwsem);
2653 dout("get_cap_refs %p ret %d got %s\n", inode,
2654 ret, ceph_cap_string(*got));
2659 * Check the offset we are writing up to against our current
2660 * max_size. If necessary, tell the MDS we want to write to
2663 static void check_max_size(struct inode *inode, loff_t endoff)
2665 struct ceph_inode_info *ci = ceph_inode(inode);
2668 /* do we need to explicitly request a larger max_size? */
2669 spin_lock(&ci->i_ceph_lock);
2670 if (endoff >= ci->i_max_size && endoff > ci->i_wanted_max_size) {
2671 dout("write %p at large endoff %llu, req max_size\n",
2673 ci->i_wanted_max_size = endoff;
2675 /* duplicate ceph_check_caps()'s logic */
2676 if (ci->i_auth_cap &&
2677 (ci->i_auth_cap->issued & CEPH_CAP_FILE_WR) &&
2678 ci->i_wanted_max_size > ci->i_max_size &&
2679 ci->i_wanted_max_size > ci->i_requested_max_size)
2681 spin_unlock(&ci->i_ceph_lock);
2683 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2686 int ceph_try_get_caps(struct ceph_inode_info *ci, int need, int want, int *got)
2690 BUG_ON(need & ~CEPH_CAP_FILE_RD);
2691 BUG_ON(want & ~(CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
2692 ret = ceph_pool_perm_check(ci, need);
2696 ret = try_get_cap_refs(ci, need, want, 0, true, got, &err);
2698 if (err == -EAGAIN) {
2700 } else if (err < 0) {
2708 * Wait for caps, and take cap references. If we can't get a WR cap
2709 * due to a small max_size, make sure we check_max_size (and possibly
2710 * ask the mds) so we don't get hung up indefinitely.
2712 int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
2713 loff_t endoff, int *got, struct page **pinned_page)
2715 int _got, ret, err = 0;
2717 ret = ceph_pool_perm_check(ci, need);
2723 check_max_size(&ci->vfs_inode, endoff);
2727 ret = try_get_cap_refs(ci, need, want, endoff,
2728 false, &_got, &err);
2735 DEFINE_WAIT_FUNC(wait, woken_wake_function);
2736 add_wait_queue(&ci->i_cap_wq, &wait);
2738 while (!try_get_cap_refs(ci, need, want, endoff,
2739 true, &_got, &err)) {
2740 if (signal_pending(current)) {
2744 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
2747 remove_wait_queue(&ci->i_cap_wq, &wait);
2755 if (err == -ESTALE) {
2756 /* session was killed, try renew caps */
2757 ret = ceph_renew_caps(&ci->vfs_inode);
2764 if (ci->i_inline_version != CEPH_INLINE_NONE &&
2765 (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
2766 i_size_read(&ci->vfs_inode) > 0) {
2768 find_get_page(ci->vfs_inode.i_mapping, 0);
2770 if (PageUptodate(page)) {
2771 *pinned_page = page;
2777 * drop cap refs first because getattr while
2778 * holding * caps refs can cause deadlock.
2780 ceph_put_cap_refs(ci, _got);
2784 * getattr request will bring inline data into
2787 ret = __ceph_do_getattr(&ci->vfs_inode, NULL,
2788 CEPH_STAT_CAP_INLINE_DATA,
2797 if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE))
2798 ceph_fscache_revalidate_cookie(ci);
2805 * Take cap refs. Caller must already know we hold at least one ref
2806 * on the caps in question or we don't know this is safe.
2808 void ceph_get_cap_refs(struct ceph_inode_info *ci, int caps)
2810 spin_lock(&ci->i_ceph_lock);
2811 __take_cap_refs(ci, caps, false);
2812 spin_unlock(&ci->i_ceph_lock);
2817 * drop cap_snap that is not associated with any snapshot.
2818 * we don't need to send FLUSHSNAP message for it.
2820 static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci,
2821 struct ceph_cap_snap *capsnap)
2823 if (!capsnap->need_flush &&
2824 !capsnap->writing && !capsnap->dirty_pages) {
2825 dout("dropping cap_snap %p follows %llu\n",
2826 capsnap, capsnap->follows);
2827 BUG_ON(capsnap->cap_flush.tid > 0);
2828 ceph_put_snap_context(capsnap->context);
2829 if (!list_is_last(&capsnap->ci_item, &ci->i_cap_snaps))
2830 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
2832 list_del(&capsnap->ci_item);
2833 ceph_put_cap_snap(capsnap);
2842 * If we released the last ref on any given cap, call ceph_check_caps
2843 * to release (or schedule a release).
2845 * If we are releasing a WR cap (from a sync write), finalize any affected
2846 * cap_snap, and wake up any waiters.
2848 void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
2850 struct inode *inode = &ci->vfs_inode;
2851 int last = 0, put = 0, flushsnaps = 0, wake = 0;
2853 spin_lock(&ci->i_ceph_lock);
2854 if (had & CEPH_CAP_PIN)
2856 if (had & CEPH_CAP_FILE_RD)
2857 if (--ci->i_rd_ref == 0)
2859 if (had & CEPH_CAP_FILE_CACHE)
2860 if (--ci->i_rdcache_ref == 0)
2862 if (had & CEPH_CAP_FILE_BUFFER) {
2863 if (--ci->i_wb_ref == 0) {
2867 dout("put_cap_refs %p wb %d -> %d (?)\n",
2868 inode, ci->i_wb_ref+1, ci->i_wb_ref);
2870 if (had & CEPH_CAP_FILE_WR)
2871 if (--ci->i_wr_ref == 0) {
2873 if (__ceph_have_pending_cap_snap(ci)) {
2874 struct ceph_cap_snap *capsnap =
2875 list_last_entry(&ci->i_cap_snaps,
2876 struct ceph_cap_snap,
2878 capsnap->writing = 0;
2879 if (ceph_try_drop_cap_snap(ci, capsnap))
2881 else if (__ceph_finish_cap_snap(ci, capsnap))
2885 if (ci->i_wrbuffer_ref_head == 0 &&
2886 ci->i_dirty_caps == 0 &&
2887 ci->i_flushing_caps == 0) {
2888 BUG_ON(!ci->i_head_snapc);
2889 ceph_put_snap_context(ci->i_head_snapc);
2890 ci->i_head_snapc = NULL;
2892 /* see comment in __ceph_remove_cap() */
2893 if (!__ceph_is_any_caps(ci) && ci->i_snap_realm)
2894 drop_inode_snap_realm(ci);
2896 spin_unlock(&ci->i_ceph_lock);
2898 dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
2899 last ? " last" : "", put ? " put" : "");
2901 if (last && !flushsnaps)
2902 ceph_check_caps(ci, 0, NULL);
2903 else if (flushsnaps)
2904 ceph_flush_snaps(ci, NULL);
2906 wake_up_all(&ci->i_cap_wq);
2912 * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
2913 * context. Adjust per-snap dirty page accounting as appropriate.
2914 * Once all dirty data for a cap_snap is flushed, flush snapped file
2915 * metadata back to the MDS. If we dropped the last ref, call
2918 void ceph_put_wrbuffer_cap_refs(struct ceph_inode_info *ci, int nr,
2919 struct ceph_snap_context *snapc)
2921 struct inode *inode = &ci->vfs_inode;
2922 struct ceph_cap_snap *capsnap = NULL;
2926 bool flush_snaps = false;
2927 bool complete_capsnap = false;
2929 spin_lock(&ci->i_ceph_lock);
2930 ci->i_wrbuffer_ref -= nr;
2931 if (ci->i_wrbuffer_ref == 0) {
2936 if (ci->i_head_snapc == snapc) {
2937 ci->i_wrbuffer_ref_head -= nr;
2938 if (ci->i_wrbuffer_ref_head == 0 &&
2939 ci->i_wr_ref == 0 &&
2940 ci->i_dirty_caps == 0 &&
2941 ci->i_flushing_caps == 0) {
2942 BUG_ON(!ci->i_head_snapc);
2943 ceph_put_snap_context(ci->i_head_snapc);
2944 ci->i_head_snapc = NULL;
2946 dout("put_wrbuffer_cap_refs on %p head %d/%d -> %d/%d %s\n",
2948 ci->i_wrbuffer_ref+nr, ci->i_wrbuffer_ref_head+nr,
2949 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
2950 last ? " LAST" : "");
2952 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
2953 if (capsnap->context == snapc) {
2959 capsnap->dirty_pages -= nr;
2960 if (capsnap->dirty_pages == 0) {
2961 complete_capsnap = true;
2962 if (!capsnap->writing) {
2963 if (ceph_try_drop_cap_snap(ci, capsnap)) {
2966 ci->i_ceph_flags |= CEPH_I_FLUSH_SNAPS;
2971 dout("put_wrbuffer_cap_refs on %p cap_snap %p "
2972 " snap %lld %d/%d -> %d/%d %s%s\n",
2973 inode, capsnap, capsnap->context->seq,
2974 ci->i_wrbuffer_ref+nr, capsnap->dirty_pages + nr,
2975 ci->i_wrbuffer_ref, capsnap->dirty_pages,
2976 last ? " (wrbuffer last)" : "",
2977 complete_capsnap ? " (complete capsnap)" : "");
2980 spin_unlock(&ci->i_ceph_lock);
2983 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
2984 } else if (flush_snaps) {
2985 ceph_flush_snaps(ci, NULL);
2987 if (complete_capsnap)
2988 wake_up_all(&ci->i_cap_wq);
2994 * Invalidate unlinked inode's aliases, so we can drop the inode ASAP.
2996 static void invalidate_aliases(struct inode *inode)
2998 struct dentry *dn, *prev = NULL;
3000 dout("invalidate_aliases inode %p\n", inode);
3001 d_prune_aliases(inode);
3003 * For non-directory inode, d_find_alias() only returns
3004 * hashed dentry. After calling d_invalidate(), the
3005 * dentry becomes unhashed.
3007 * For directory inode, d_find_alias() can return
3008 * unhashed dentry. But directory inode should have
3009 * one alias at most.
3011 while ((dn = d_find_alias(inode))) {
3026 * Handle a cap GRANT message from the MDS. (Note that a GRANT may
3027 * actually be a revocation if it specifies a smaller cap set.)
3029 * caller holds s_mutex and i_ceph_lock, we drop both.
3031 static void handle_cap_grant(struct ceph_mds_client *mdsc,
3032 struct inode *inode, struct ceph_mds_caps *grant,
3033 struct ceph_string **pns, u64 inline_version,
3034 void *inline_data, u32 inline_len,
3035 struct ceph_buffer *xattr_buf,
3036 struct ceph_mds_session *session,
3037 struct ceph_cap *cap, int issued)
3038 __releases(ci->i_ceph_lock)
3039 __releases(mdsc->snap_rwsem)
3041 struct ceph_inode_info *ci = ceph_inode(inode);
3042 int mds = session->s_mds;
3043 int seq = le32_to_cpu(grant->seq);
3044 int newcaps = le32_to_cpu(grant->caps);
3045 int used, wanted, dirty;
3046 u64 size = le64_to_cpu(grant->size);
3047 u64 max_size = le64_to_cpu(grant->max_size);
3048 struct timespec mtime, atime, ctime;
3051 bool writeback = false;
3052 bool queue_trunc = false;
3053 bool queue_invalidate = false;
3054 bool deleted_inode = false;
3055 bool fill_inline = false;
3057 dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
3058 inode, cap, mds, seq, ceph_cap_string(newcaps));
3059 dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
3064 * auth mds of the inode changed. we received the cap export message,
3065 * but still haven't received the cap import message. handle_cap_export
3066 * updated the new auth MDS' cap.
3068 * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing a message
3069 * that was sent before the cap import message. So don't remove caps.
3071 if (ceph_seq_cmp(seq, cap->seq) <= 0) {
3072 WARN_ON(cap != ci->i_auth_cap);
3073 WARN_ON(cap->cap_id != le64_to_cpu(grant->cap_id));
3075 newcaps |= cap->issued;
3079 * If CACHE is being revoked, and we have no dirty buffers,
3080 * try to invalidate (once). (If there are dirty buffers, we
3081 * will invalidate _after_ writeback.)
3083 if (!S_ISDIR(inode->i_mode) && /* don't invalidate readdir cache */
3084 ((cap->issued & ~newcaps) & CEPH_CAP_FILE_CACHE) &&
3085 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
3086 !(ci->i_wrbuffer_ref || ci->i_wb_ref)) {
3087 if (try_nonblocking_invalidate(inode)) {
3088 /* there were locked pages.. invalidate later
3089 in a separate thread. */
3090 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
3091 queue_invalidate = true;
3092 ci->i_rdcache_revoking = ci->i_rdcache_gen;
3097 /* side effects now are allowed */
3098 cap->cap_gen = session->s_cap_gen;
3101 __check_cap_issue(ci, cap, newcaps);
3103 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
3104 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
3105 inode->i_mode = le32_to_cpu(grant->mode);
3106 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(grant->uid));
3107 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(grant->gid));
3108 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
3109 from_kuid(&init_user_ns, inode->i_uid),
3110 from_kgid(&init_user_ns, inode->i_gid));
3113 if ((newcaps & CEPH_CAP_AUTH_SHARED) &&
3114 (issued & CEPH_CAP_LINK_EXCL) == 0) {
3115 set_nlink(inode, le32_to_cpu(grant->nlink));
3116 if (inode->i_nlink == 0 &&
3117 (newcaps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
3118 deleted_inode = true;
3121 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && grant->xattr_len) {
3122 int len = le32_to_cpu(grant->xattr_len);
3123 u64 version = le64_to_cpu(grant->xattr_version);
3125 if (version > ci->i_xattrs.version) {
3126 dout(" got new xattrs v%llu on %p len %d\n",
3127 version, inode, len);
3128 if (ci->i_xattrs.blob)
3129 ceph_buffer_put(ci->i_xattrs.blob);
3130 ci->i_xattrs.blob = ceph_buffer_get(xattr_buf);
3131 ci->i_xattrs.version = version;
3132 ceph_forget_all_cached_acls(inode);
3136 if (newcaps & CEPH_CAP_ANY_RD) {
3137 /* ctime/mtime/atime? */
3138 ceph_decode_timespec(&mtime, &grant->mtime);
3139 ceph_decode_timespec(&atime, &grant->atime);
3140 ceph_decode_timespec(&ctime, &grant->ctime);
3141 ceph_fill_file_time(inode, issued,
3142 le32_to_cpu(grant->time_warp_seq),
3143 &ctime, &mtime, &atime);
3146 if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
3147 /* file layout may have changed */
3148 s64 old_pool = ci->i_layout.pool_id;
3149 struct ceph_string *old_ns;
3151 ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout);
3152 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
3153 lockdep_is_held(&ci->i_ceph_lock));
3154 rcu_assign_pointer(ci->i_layout.pool_ns, *pns);
3156 if (ci->i_layout.pool_id != old_pool || *pns != old_ns)
3157 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
3161 /* size/truncate_seq? */
3162 queue_trunc = ceph_fill_file_size(inode, issued,
3163 le32_to_cpu(grant->truncate_seq),
3164 le64_to_cpu(grant->truncate_size),
3168 if (ci->i_auth_cap == cap && (newcaps & CEPH_CAP_ANY_FILE_WR)) {
3169 if (max_size != ci->i_max_size) {
3170 dout("max_size %lld -> %llu\n",
3171 ci->i_max_size, max_size);
3172 ci->i_max_size = max_size;
3173 if (max_size >= ci->i_wanted_max_size) {
3174 ci->i_wanted_max_size = 0; /* reset */
3175 ci->i_requested_max_size = 0;
3178 } else if (ci->i_wanted_max_size > ci->i_max_size &&
3179 ci->i_wanted_max_size > ci->i_requested_max_size) {
3180 /* CEPH_CAP_OP_IMPORT */
3185 /* check cap bits */
3186 wanted = __ceph_caps_wanted(ci);
3187 used = __ceph_caps_used(ci);
3188 dirty = __ceph_caps_dirty(ci);
3189 dout(" my wanted = %s, used = %s, dirty %s\n",
3190 ceph_cap_string(wanted),
3191 ceph_cap_string(used),
3192 ceph_cap_string(dirty));
3193 if (wanted != le32_to_cpu(grant->wanted)) {
3194 dout("mds wanted %s -> %s\n",
3195 ceph_cap_string(le32_to_cpu(grant->wanted)),
3196 ceph_cap_string(wanted));
3197 /* imported cap may not have correct mds_wanted */
3198 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT)
3202 /* revocation, grant, or no-op? */
3203 if (cap->issued & ~newcaps) {
3204 int revoking = cap->issued & ~newcaps;
3206 dout("revocation: %s -> %s (revoking %s)\n",
3207 ceph_cap_string(cap->issued),
3208 ceph_cap_string(newcaps),
3209 ceph_cap_string(revoking));
3210 if (revoking & used & CEPH_CAP_FILE_BUFFER)
3211 writeback = true; /* initiate writeback; will delay ack */
3212 else if (revoking == CEPH_CAP_FILE_CACHE &&
3213 (newcaps & CEPH_CAP_FILE_LAZYIO) == 0 &&
3215 ; /* do nothing yet, invalidation will be queued */
3216 else if (cap == ci->i_auth_cap)
3217 check_caps = 1; /* check auth cap only */
3219 check_caps = 2; /* check all caps */
3220 cap->issued = newcaps;
3221 cap->implemented |= newcaps;
3222 } else if (cap->issued == newcaps) {
3223 dout("caps unchanged: %s -> %s\n",
3224 ceph_cap_string(cap->issued), ceph_cap_string(newcaps));
3226 dout("grant: %s -> %s\n", ceph_cap_string(cap->issued),
3227 ceph_cap_string(newcaps));
3228 /* non-auth MDS is revoking the newly grant caps ? */
3229 if (cap == ci->i_auth_cap &&
3230 __ceph_caps_revoking_other(ci, cap, newcaps))
3233 cap->issued = newcaps;
3234 cap->implemented |= newcaps; /* add bits only, to
3235 * avoid stepping on a
3236 * pending revocation */
3239 BUG_ON(cap->issued & ~cap->implemented);
3241 if (inline_version > 0 && inline_version >= ci->i_inline_version) {
3242 ci->i_inline_version = inline_version;
3243 if (ci->i_inline_version != CEPH_INLINE_NONE &&
3244 (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
3248 if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
3249 if (newcaps & ~issued)
3251 kick_flushing_inode_caps(mdsc, session, inode);
3252 up_read(&mdsc->snap_rwsem);
3254 spin_unlock(&ci->i_ceph_lock);
3258 ceph_fill_inline_data(inode, NULL, inline_data, inline_len);
3261 ceph_queue_vmtruncate(inode);
3265 * queue inode for writeback: we can't actually call
3266 * filemap_write_and_wait, etc. from message handler
3269 ceph_queue_writeback(inode);
3270 if (queue_invalidate)
3271 ceph_queue_invalidate(inode);
3273 invalidate_aliases(inode);
3275 wake_up_all(&ci->i_cap_wq);
3277 if (check_caps == 1)
3278 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_AUTHONLY,
3280 else if (check_caps == 2)
3281 ceph_check_caps(ci, CHECK_CAPS_NODELAY, session);
3283 mutex_unlock(&session->s_mutex);
3287 * Handle FLUSH_ACK from MDS, indicating that metadata we sent to the
3288 * MDS has been safely committed.
3290 static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
3291 struct ceph_mds_caps *m,
3292 struct ceph_mds_session *session,
3293 struct ceph_cap *cap)
3294 __releases(ci->i_ceph_lock)
3296 struct ceph_inode_info *ci = ceph_inode(inode);
3297 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
3298 struct ceph_cap_flush *cf, *tmp_cf;
3299 LIST_HEAD(to_remove);
3300 unsigned seq = le32_to_cpu(m->seq);
3301 int dirty = le32_to_cpu(m->dirty);
3304 bool wake_ci = false;
3305 bool wake_mdsc = false;
3307 list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) {
3308 if (cf->tid == flush_tid)
3310 if (cf->caps == 0) /* capsnap */
3312 if (cf->tid <= flush_tid) {
3313 if (__finish_cap_flush(NULL, ci, cf))
3315 list_add_tail(&cf->i_list, &to_remove);
3317 cleaned &= ~cf->caps;
3323 dout("handle_cap_flush_ack inode %p mds%d seq %d on %s cleaned %s,"
3324 " flushing %s -> %s\n",
3325 inode, session->s_mds, seq, ceph_cap_string(dirty),
3326 ceph_cap_string(cleaned), ceph_cap_string(ci->i_flushing_caps),
3327 ceph_cap_string(ci->i_flushing_caps & ~cleaned));
3329 if (list_empty(&to_remove) && !cleaned)
3332 ci->i_flushing_caps &= ~cleaned;
3334 spin_lock(&mdsc->cap_dirty_lock);
3336 list_for_each_entry(cf, &to_remove, i_list) {
3337 if (__finish_cap_flush(mdsc, NULL, cf))
3341 if (ci->i_flushing_caps == 0) {
3342 if (list_empty(&ci->i_cap_flush_list)) {
3343 list_del_init(&ci->i_flushing_item);
3344 if (!list_empty(&session->s_cap_flushing)) {
3345 dout(" mds%d still flushing cap on %p\n",
3347 &list_first_entry(&session->s_cap_flushing,
3348 struct ceph_inode_info,
3349 i_flushing_item)->vfs_inode);
3352 mdsc->num_cap_flushing--;
3353 dout(" inode %p now !flushing\n", inode);
3355 if (ci->i_dirty_caps == 0) {
3356 dout(" inode %p now clean\n", inode);
3357 BUG_ON(!list_empty(&ci->i_dirty_item));
3359 if (ci->i_wr_ref == 0 &&
3360 ci->i_wrbuffer_ref_head == 0) {
3361 BUG_ON(!ci->i_head_snapc);
3362 ceph_put_snap_context(ci->i_head_snapc);
3363 ci->i_head_snapc = NULL;
3366 BUG_ON(list_empty(&ci->i_dirty_item));
3369 spin_unlock(&mdsc->cap_dirty_lock);
3372 spin_unlock(&ci->i_ceph_lock);
3374 while (!list_empty(&to_remove)) {
3375 cf = list_first_entry(&to_remove,
3376 struct ceph_cap_flush, i_list);
3377 list_del(&cf->i_list);
3378 ceph_free_cap_flush(cf);
3382 wake_up_all(&ci->i_cap_wq);
3384 wake_up_all(&mdsc->cap_flushing_wq);
3390 * Handle FLUSHSNAP_ACK. MDS has flushed snap data to disk and we can
3391 * throw away our cap_snap.
3393 * Caller hold s_mutex.
3395 static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
3396 struct ceph_mds_caps *m,
3397 struct ceph_mds_session *session)
3399 struct ceph_inode_info *ci = ceph_inode(inode);
3400 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
3401 u64 follows = le64_to_cpu(m->snap_follows);
3402 struct ceph_cap_snap *capsnap;
3403 bool flushed = false;
3404 bool wake_ci = false;
3405 bool wake_mdsc = false;
3407 dout("handle_cap_flushsnap_ack inode %p ci %p mds%d follows %lld\n",
3408 inode, ci, session->s_mds, follows);
3410 spin_lock(&ci->i_ceph_lock);
3411 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
3412 if (capsnap->follows == follows) {
3413 if (capsnap->cap_flush.tid != flush_tid) {
3414 dout(" cap_snap %p follows %lld tid %lld !="
3415 " %lld\n", capsnap, follows,
3416 flush_tid, capsnap->cap_flush.tid);
3422 dout(" skipping cap_snap %p follows %lld\n",
3423 capsnap, capsnap->follows);
3427 WARN_ON(capsnap->dirty_pages || capsnap->writing);
3428 dout(" removing %p cap_snap %p follows %lld\n",
3429 inode, capsnap, follows);
3430 list_del(&capsnap->ci_item);
3431 if (__finish_cap_flush(NULL, ci, &capsnap->cap_flush))
3434 spin_lock(&mdsc->cap_dirty_lock);
3436 if (list_empty(&ci->i_cap_flush_list))
3437 list_del_init(&ci->i_flushing_item);
3439 if (__finish_cap_flush(mdsc, NULL, &capsnap->cap_flush))
3442 spin_unlock(&mdsc->cap_dirty_lock);
3444 spin_unlock(&ci->i_ceph_lock);
3446 ceph_put_snap_context(capsnap->context);
3447 ceph_put_cap_snap(capsnap);
3449 wake_up_all(&ci->i_cap_wq);
3451 wake_up_all(&mdsc->cap_flushing_wq);
3457 * Handle TRUNC from MDS, indicating file truncation.
3459 * caller hold s_mutex.
3461 static void handle_cap_trunc(struct inode *inode,
3462 struct ceph_mds_caps *trunc,
3463 struct ceph_mds_session *session)
3464 __releases(ci->i_ceph_lock)
3466 struct ceph_inode_info *ci = ceph_inode(inode);
3467 int mds = session->s_mds;
3468 int seq = le32_to_cpu(trunc->seq);
3469 u32 truncate_seq = le32_to_cpu(trunc->truncate_seq);
3470 u64 truncate_size = le64_to_cpu(trunc->truncate_size);
3471 u64 size = le64_to_cpu(trunc->size);
3472 int implemented = 0;
3473 int dirty = __ceph_caps_dirty(ci);
3474 int issued = __ceph_caps_issued(ceph_inode(inode), &implemented);
3475 int queue_trunc = 0;
3477 issued |= implemented | dirty;
3479 dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
3480 inode, mds, seq, truncate_size, truncate_seq);
3481 queue_trunc = ceph_fill_file_size(inode, issued,
3482 truncate_seq, truncate_size, size);
3483 spin_unlock(&ci->i_ceph_lock);
3486 ceph_queue_vmtruncate(inode);
3490 * Handle EXPORT from MDS. Cap is being migrated _from_ this mds to a
3491 * different one. If we are the most recent migration we've seen (as
3492 * indicated by mseq), make note of the migrating cap bits for the
3493 * duration (until we see the corresponding IMPORT).
3495 * caller holds s_mutex
3497 static void handle_cap_export(struct inode *inode, struct ceph_mds_caps *ex,
3498 struct ceph_mds_cap_peer *ph,
3499 struct ceph_mds_session *session)
3501 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
3502 struct ceph_mds_session *tsession = NULL;
3503 struct ceph_cap *cap, *tcap, *new_cap = NULL;
3504 struct ceph_inode_info *ci = ceph_inode(inode);
3506 unsigned mseq = le32_to_cpu(ex->migrate_seq);
3507 unsigned t_seq, t_mseq;
3509 int mds = session->s_mds;
3512 t_cap_id = le64_to_cpu(ph->cap_id);
3513 t_seq = le32_to_cpu(ph->seq);
3514 t_mseq = le32_to_cpu(ph->mseq);
3515 target = le32_to_cpu(ph->mds);
3517 t_cap_id = t_seq = t_mseq = 0;
3521 dout("handle_cap_export inode %p ci %p mds%d mseq %d target %d\n",
3522 inode, ci, mds, mseq, target);
3524 spin_lock(&ci->i_ceph_lock);
3525 cap = __get_cap_for_mds(ci, mds);
3526 if (!cap || cap->cap_id != le64_to_cpu(ex->cap_id))
3530 __ceph_remove_cap(cap, false);
3531 if (!ci->i_auth_cap)
3532 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
3537 * now we know we haven't received the cap import message yet
3538 * because the exported cap still exist.
3541 issued = cap->issued;
3542 if (issued != cap->implemented)
3543 pr_err_ratelimited("handle_cap_export: issued != implemented: "
3544 "ino (%llx.%llx) mds%d seq %d mseq %d "
3545 "issued %s implemented %s\n",
3546 ceph_vinop(inode), mds, cap->seq, cap->mseq,
3547 ceph_cap_string(issued),
3548 ceph_cap_string(cap->implemented));
3551 tcap = __get_cap_for_mds(ci, target);
3553 /* already have caps from the target */
3554 if (tcap->cap_id == t_cap_id &&
3555 ceph_seq_cmp(tcap->seq, t_seq) < 0) {
3556 dout(" updating import cap %p mds%d\n", tcap, target);
3557 tcap->cap_id = t_cap_id;
3558 tcap->seq = t_seq - 1;
3559 tcap->issue_seq = t_seq - 1;
3560 tcap->mseq = t_mseq;
3561 tcap->issued |= issued;
3562 tcap->implemented |= issued;
3563 if (cap == ci->i_auth_cap)
3564 ci->i_auth_cap = tcap;
3566 if (!list_empty(&ci->i_cap_flush_list) &&
3567 ci->i_auth_cap == tcap) {
3568 spin_lock(&mdsc->cap_dirty_lock);
3569 list_move_tail(&ci->i_flushing_item,
3570 &tcap->session->s_cap_flushing);
3571 spin_unlock(&mdsc->cap_dirty_lock);
3574 __ceph_remove_cap(cap, false);
3576 } else if (tsession) {
3577 /* add placeholder for the export tagert */
3578 int flag = (cap == ci->i_auth_cap) ? CEPH_CAP_FLAG_AUTH : 0;
3580 ceph_add_cap(inode, tsession, t_cap_id, -1, issued, 0,
3581 t_seq - 1, t_mseq, (u64)-1, flag, &new_cap);
3583 if (!list_empty(&ci->i_cap_flush_list) &&
3584 ci->i_auth_cap == tcap) {
3585 spin_lock(&mdsc->cap_dirty_lock);
3586 list_move_tail(&ci->i_flushing_item,
3587 &tcap->session->s_cap_flushing);
3588 spin_unlock(&mdsc->cap_dirty_lock);
3591 __ceph_remove_cap(cap, false);
3595 spin_unlock(&ci->i_ceph_lock);
3596 mutex_unlock(&session->s_mutex);
3598 /* open target session */
3599 tsession = ceph_mdsc_open_export_target_session(mdsc, target);
3600 if (!IS_ERR(tsession)) {
3602 mutex_lock(&session->s_mutex);
3603 mutex_lock_nested(&tsession->s_mutex,
3604 SINGLE_DEPTH_NESTING);
3606 mutex_lock(&tsession->s_mutex);
3607 mutex_lock_nested(&session->s_mutex,
3608 SINGLE_DEPTH_NESTING);
3610 new_cap = ceph_get_cap(mdsc, NULL);
3619 spin_unlock(&ci->i_ceph_lock);
3620 mutex_unlock(&session->s_mutex);
3622 mutex_unlock(&tsession->s_mutex);
3623 ceph_put_mds_session(tsession);
3626 ceph_put_cap(mdsc, new_cap);
3630 * Handle cap IMPORT.
3632 * caller holds s_mutex. acquires i_ceph_lock
3634 static void handle_cap_import(struct ceph_mds_client *mdsc,
3635 struct inode *inode, struct ceph_mds_caps *im,
3636 struct ceph_mds_cap_peer *ph,
3637 struct ceph_mds_session *session,
3638 struct ceph_cap **target_cap, int *old_issued)
3639 __acquires(ci->i_ceph_lock)
3641 struct ceph_inode_info *ci = ceph_inode(inode);
3642 struct ceph_cap *cap, *ocap, *new_cap = NULL;
3643 int mds = session->s_mds;
3645 unsigned caps = le32_to_cpu(im->caps);
3646 unsigned wanted = le32_to_cpu(im->wanted);
3647 unsigned seq = le32_to_cpu(im->seq);
3648 unsigned mseq = le32_to_cpu(im->migrate_seq);
3649 u64 realmino = le64_to_cpu(im->realm);
3650 u64 cap_id = le64_to_cpu(im->cap_id);
3655 p_cap_id = le64_to_cpu(ph->cap_id);
3656 peer = le32_to_cpu(ph->mds);
3662 dout("handle_cap_import inode %p ci %p mds%d mseq %d peer %d\n",
3663 inode, ci, mds, mseq, peer);
3666 spin_lock(&ci->i_ceph_lock);
3667 cap = __get_cap_for_mds(ci, mds);
3670 spin_unlock(&ci->i_ceph_lock);
3671 new_cap = ceph_get_cap(mdsc, NULL);
3677 ceph_put_cap(mdsc, new_cap);
3682 __ceph_caps_issued(ci, &issued);
3683 issued |= __ceph_caps_dirty(ci);
3685 ceph_add_cap(inode, session, cap_id, -1, caps, wanted, seq, mseq,
3686 realmino, CEPH_CAP_FLAG_AUTH, &new_cap);
3688 ocap = peer >= 0 ? __get_cap_for_mds(ci, peer) : NULL;
3689 if (ocap && ocap->cap_id == p_cap_id) {
3690 dout(" remove export cap %p mds%d flags %d\n",
3691 ocap, peer, ph->flags);
3692 if ((ph->flags & CEPH_CAP_FLAG_AUTH) &&
3693 (ocap->seq != le32_to_cpu(ph->seq) ||
3694 ocap->mseq != le32_to_cpu(ph->mseq))) {
3695 pr_err_ratelimited("handle_cap_import: "
3696 "mismatched seq/mseq: ino (%llx.%llx) "
3697 "mds%d seq %d mseq %d importer mds%d "
3698 "has peer seq %d mseq %d\n",
3699 ceph_vinop(inode), peer, ocap->seq,
3700 ocap->mseq, mds, le32_to_cpu(ph->seq),
3701 le32_to_cpu(ph->mseq));
3703 __ceph_remove_cap(ocap, (ph->flags & CEPH_CAP_FLAG_RELEASE));
3706 /* make sure we re-request max_size, if necessary */
3707 ci->i_requested_max_size = 0;
3709 *old_issued = issued;
3714 * Handle a caps message from the MDS.
3716 * Identify the appropriate session, inode, and call the right handler
3717 * based on the cap op.
3719 void ceph_handle_caps(struct ceph_mds_session *session,
3720 struct ceph_msg *msg)
3722 struct ceph_mds_client *mdsc = session->s_mdsc;
3723 struct super_block *sb = mdsc->fsc->sb;
3724 struct inode *inode;
3725 struct ceph_inode_info *ci;
3726 struct ceph_cap *cap;
3727 struct ceph_mds_caps *h;
3728 struct ceph_mds_cap_peer *peer = NULL;
3729 struct ceph_snap_realm *realm = NULL;
3730 struct ceph_string *pool_ns = NULL;
3731 int mds = session->s_mds;
3734 struct ceph_vino vino;
3736 u64 inline_version = 0;
3737 void *inline_data = NULL;
3740 size_t snaptrace_len;
3743 dout("handle_caps from mds%d\n", mds);
3746 end = msg->front.iov_base + msg->front.iov_len;
3747 tid = le64_to_cpu(msg->hdr.tid);
3748 if (msg->front.iov_len < sizeof(*h))
3750 h = msg->front.iov_base;
3751 op = le32_to_cpu(h->op);
3752 vino.ino = le64_to_cpu(h->ino);
3753 vino.snap = CEPH_NOSNAP;
3754 seq = le32_to_cpu(h->seq);
3755 mseq = le32_to_cpu(h->migrate_seq);
3758 snaptrace_len = le32_to_cpu(h->snap_trace_len);
3759 p = snaptrace + snaptrace_len;
3761 if (le16_to_cpu(msg->hdr.version) >= 2) {
3763 ceph_decode_32_safe(&p, end, flock_len, bad);
3764 if (p + flock_len > end)
3769 if (le16_to_cpu(msg->hdr.version) >= 3) {
3770 if (op == CEPH_CAP_OP_IMPORT) {
3771 if (p + sizeof(*peer) > end)
3775 } else if (op == CEPH_CAP_OP_EXPORT) {
3776 /* recorded in unused fields */
3777 peer = (void *)&h->size;
3781 if (le16_to_cpu(msg->hdr.version) >= 4) {
3782 ceph_decode_64_safe(&p, end, inline_version, bad);
3783 ceph_decode_32_safe(&p, end, inline_len, bad);
3784 if (p + inline_len > end)
3790 if (le16_to_cpu(msg->hdr.version) >= 5) {
3791 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
3794 ceph_decode_32_safe(&p, end, epoch_barrier, bad);
3795 ceph_osdc_update_epoch_barrier(osdc, epoch_barrier);
3798 if (le16_to_cpu(msg->hdr.version) >= 8) {
3800 u32 caller_uid, caller_gid;
3804 ceph_decode_64_safe(&p, end, flush_tid, bad);
3806 ceph_decode_32_safe(&p, end, caller_uid, bad);
3807 ceph_decode_32_safe(&p, end, caller_gid, bad);
3809 ceph_decode_32_safe(&p, end, pool_ns_len, bad);
3810 if (pool_ns_len > 0) {
3811 ceph_decode_need(&p, end, pool_ns_len, bad);
3812 pool_ns = ceph_find_or_create_string(p, pool_ns_len);
3818 inode = ceph_find_inode(sb, vino);
3819 ci = ceph_inode(inode);
3820 dout(" op %s ino %llx.%llx inode %p\n", ceph_cap_op_name(op), vino.ino,
3823 mutex_lock(&session->s_mutex);
3825 dout(" mds%d seq %lld cap seq %u\n", session->s_mds, session->s_seq,
3829 dout(" i don't have ino %llx\n", vino.ino);
3831 if (op == CEPH_CAP_OP_IMPORT) {
3832 cap = ceph_get_cap(mdsc, NULL);
3833 cap->cap_ino = vino.ino;
3834 cap->queue_release = 1;
3835 cap->cap_id = le64_to_cpu(h->cap_id);
3838 cap->issue_seq = seq;
3839 spin_lock(&session->s_cap_lock);
3840 list_add_tail(&cap->session_caps,
3841 &session->s_cap_releases);
3842 session->s_num_cap_releases++;
3843 spin_unlock(&session->s_cap_lock);
3845 goto flush_cap_releases;
3848 /* these will work even if we don't have a cap yet */
3850 case CEPH_CAP_OP_FLUSHSNAP_ACK:
3851 handle_cap_flushsnap_ack(inode, tid, h, session);
3854 case CEPH_CAP_OP_EXPORT:
3855 handle_cap_export(inode, h, peer, session);
3858 case CEPH_CAP_OP_IMPORT:
3860 if (snaptrace_len) {
3861 down_write(&mdsc->snap_rwsem);
3862 ceph_update_snap_trace(mdsc, snaptrace,
3863 snaptrace + snaptrace_len,
3865 downgrade_write(&mdsc->snap_rwsem);
3867 down_read(&mdsc->snap_rwsem);
3869 handle_cap_import(mdsc, inode, h, peer, session,
3871 handle_cap_grant(mdsc, inode, h, &pool_ns,
3872 inline_version, inline_data, inline_len,
3873 msg->middle, session, cap, issued);
3875 ceph_put_snap_realm(mdsc, realm);
3879 /* the rest require a cap */
3880 spin_lock(&ci->i_ceph_lock);
3881 cap = __get_cap_for_mds(ceph_inode(inode), mds);
3883 dout(" no cap on %p ino %llx.%llx from mds%d\n",
3884 inode, ceph_ino(inode), ceph_snap(inode), mds);
3885 spin_unlock(&ci->i_ceph_lock);
3886 goto flush_cap_releases;
3889 /* note that each of these drops i_ceph_lock for us */
3891 case CEPH_CAP_OP_REVOKE:
3892 case CEPH_CAP_OP_GRANT:
3893 __ceph_caps_issued(ci, &issued);
3894 issued |= __ceph_caps_dirty(ci);
3895 handle_cap_grant(mdsc, inode, h, &pool_ns,
3896 inline_version, inline_data, inline_len,
3897 msg->middle, session, cap, issued);
3900 case CEPH_CAP_OP_FLUSH_ACK:
3901 handle_cap_flush_ack(inode, tid, h, session, cap);
3904 case CEPH_CAP_OP_TRUNC:
3905 handle_cap_trunc(inode, h, session);
3909 spin_unlock(&ci->i_ceph_lock);
3910 pr_err("ceph_handle_caps: unknown cap op %d %s\n", op,
3911 ceph_cap_op_name(op));
3918 * send any cap release message to try to move things
3919 * along for the mds (who clearly thinks we still have this
3922 ceph_send_cap_releases(mdsc, session);
3925 mutex_unlock(&session->s_mutex);
3928 ceph_put_string(pool_ns);
3932 pr_err("ceph_handle_caps: corrupt message\n");
3938 * Delayed work handler to process end of delayed cap release LRU list.
3940 void ceph_check_delayed_caps(struct ceph_mds_client *mdsc)
3942 struct inode *inode;
3943 struct ceph_inode_info *ci;
3944 int flags = CHECK_CAPS_NODELAY;
3946 dout("check_delayed_caps\n");
3948 spin_lock(&mdsc->cap_delay_lock);
3949 if (list_empty(&mdsc->cap_delay_list))
3951 ci = list_first_entry(&mdsc->cap_delay_list,
3952 struct ceph_inode_info,
3954 if ((ci->i_ceph_flags & CEPH_I_FLUSH) == 0 &&
3955 time_before(jiffies, ci->i_hold_caps_max))
3957 list_del_init(&ci->i_cap_delay_list);
3959 inode = igrab(&ci->vfs_inode);
3960 spin_unlock(&mdsc->cap_delay_lock);
3963 dout("check_delayed_caps on %p\n", inode);
3964 ceph_check_caps(ci, flags, NULL);
3968 spin_unlock(&mdsc->cap_delay_lock);
3972 * Flush all dirty caps to the mds
3974 void ceph_flush_dirty_caps(struct ceph_mds_client *mdsc)
3976 struct ceph_inode_info *ci;
3977 struct inode *inode;
3979 dout("flush_dirty_caps\n");
3980 spin_lock(&mdsc->cap_dirty_lock);
3981 while (!list_empty(&mdsc->cap_dirty)) {
3982 ci = list_first_entry(&mdsc->cap_dirty, struct ceph_inode_info,
3984 inode = &ci->vfs_inode;
3986 dout("flush_dirty_caps %p\n", inode);
3987 spin_unlock(&mdsc->cap_dirty_lock);
3988 ceph_check_caps(ci, CHECK_CAPS_NODELAY|CHECK_CAPS_FLUSH, NULL);
3990 spin_lock(&mdsc->cap_dirty_lock);
3992 spin_unlock(&mdsc->cap_dirty_lock);
3993 dout("flush_dirty_caps done\n");
3996 void __ceph_get_fmode(struct ceph_inode_info *ci, int fmode)
3999 int bits = (fmode << 1) | 1;
4000 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
4001 if (bits & (1 << i))
4002 ci->i_nr_by_mode[i]++;
4007 * Drop open file reference. If we were the last open file,
4008 * we may need to release capabilities to the MDS (or schedule
4009 * their delayed release).
4011 void ceph_put_fmode(struct ceph_inode_info *ci, int fmode)
4014 int bits = (fmode << 1) | 1;
4015 spin_lock(&ci->i_ceph_lock);
4016 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) {
4017 if (bits & (1 << i)) {
4018 BUG_ON(ci->i_nr_by_mode[i] == 0);
4019 if (--ci->i_nr_by_mode[i] == 0)
4023 dout("put_fmode %p fmode %d {%d,%d,%d,%d}\n",
4024 &ci->vfs_inode, fmode,
4025 ci->i_nr_by_mode[0], ci->i_nr_by_mode[1],
4026 ci->i_nr_by_mode[2], ci->i_nr_by_mode[3]);
4027 spin_unlock(&ci->i_ceph_lock);
4029 if (last && ci->i_vino.snap == CEPH_NOSNAP)
4030 ceph_check_caps(ci, 0, NULL);
4034 * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it
4035 * looks like the link count will hit 0, drop any other caps (other
4036 * than PIN) we don't specifically want (due to the file still being
4039 int ceph_drop_caps_for_unlink(struct inode *inode)
4041 struct ceph_inode_info *ci = ceph_inode(inode);
4042 int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
4044 spin_lock(&ci->i_ceph_lock);
4045 if (inode->i_nlink == 1) {
4046 drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
4048 ci->i_ceph_flags |= CEPH_I_NODELAY;
4049 if (__ceph_caps_dirty(ci)) {
4050 struct ceph_mds_client *mdsc =
4051 ceph_inode_to_client(inode)->mdsc;
4052 __cap_delay_requeue_front(mdsc, ci);
4055 spin_unlock(&ci->i_ceph_lock);
4060 * Helpers for embedding cap and dentry lease releases into mds
4063 * @force is used by dentry_release (below) to force inclusion of a
4064 * record for the directory inode, even when there aren't any caps to
4067 int ceph_encode_inode_release(void **p, struct inode *inode,
4068 int mds, int drop, int unless, int force)
4070 struct ceph_inode_info *ci = ceph_inode(inode);
4071 struct ceph_cap *cap;
4072 struct ceph_mds_request_release *rel = *p;
4076 spin_lock(&ci->i_ceph_lock);
4077 used = __ceph_caps_used(ci);
4078 dirty = __ceph_caps_dirty(ci);
4080 dout("encode_inode_release %p mds%d used|dirty %s drop %s unless %s\n",
4081 inode, mds, ceph_cap_string(used|dirty), ceph_cap_string(drop),
4082 ceph_cap_string(unless));
4084 /* only drop unused, clean caps */
4085 drop &= ~(used | dirty);
4087 cap = __get_cap_for_mds(ci, mds);
4088 if (cap && __cap_is_valid(cap)) {
4089 unless &= cap->issued;
4091 if (unless & CEPH_CAP_AUTH_EXCL)
4092 drop &= ~CEPH_CAP_AUTH_SHARED;
4093 if (unless & CEPH_CAP_LINK_EXCL)
4094 drop &= ~CEPH_CAP_LINK_SHARED;
4095 if (unless & CEPH_CAP_XATTR_EXCL)
4096 drop &= ~CEPH_CAP_XATTR_SHARED;
4097 if (unless & CEPH_CAP_FILE_EXCL)
4098 drop &= ~CEPH_CAP_FILE_SHARED;
4101 if (force || (cap->issued & drop)) {
4102 if (cap->issued & drop) {
4103 int wanted = __ceph_caps_wanted(ci);
4104 if ((ci->i_ceph_flags & CEPH_I_NODELAY) == 0)
4105 wanted |= cap->mds_wanted;
4106 dout("encode_inode_release %p cap %p "
4107 "%s -> %s, wanted %s -> %s\n", inode, cap,
4108 ceph_cap_string(cap->issued),
4109 ceph_cap_string(cap->issued & ~drop),
4110 ceph_cap_string(cap->mds_wanted),
4111 ceph_cap_string(wanted));
4113 cap->issued &= ~drop;
4114 cap->implemented &= ~drop;
4115 cap->mds_wanted = wanted;
4117 dout("encode_inode_release %p cap %p %s"
4118 " (force)\n", inode, cap,
4119 ceph_cap_string(cap->issued));
4122 rel->ino = cpu_to_le64(ceph_ino(inode));
4123 rel->cap_id = cpu_to_le64(cap->cap_id);
4124 rel->seq = cpu_to_le32(cap->seq);
4125 rel->issue_seq = cpu_to_le32(cap->issue_seq);
4126 rel->mseq = cpu_to_le32(cap->mseq);
4127 rel->caps = cpu_to_le32(cap->implemented);
4128 rel->wanted = cpu_to_le32(cap->mds_wanted);
4134 dout("encode_inode_release %p cap %p %s (noop)\n",
4135 inode, cap, ceph_cap_string(cap->issued));
4138 spin_unlock(&ci->i_ceph_lock);
4142 int ceph_encode_dentry_release(void **p, struct dentry *dentry,
4144 int mds, int drop, int unless)
4146 struct dentry *parent = NULL;
4147 struct ceph_mds_request_release *rel = *p;
4148 struct ceph_dentry_info *di = ceph_dentry(dentry);
4153 * force an record for the directory caps if we have a dentry lease.
4154 * this is racy (can't take i_ceph_lock and d_lock together), but it
4155 * doesn't have to be perfect; the mds will revoke anything we don't
4158 spin_lock(&dentry->d_lock);
4159 if (di->lease_session && di->lease_session->s_mds == mds)
4162 parent = dget(dentry->d_parent);
4163 dir = d_inode(parent);
4165 spin_unlock(&dentry->d_lock);
4167 ret = ceph_encode_inode_release(p, dir, mds, drop, unless, force);
4170 spin_lock(&dentry->d_lock);
4171 if (ret && di->lease_session && di->lease_session->s_mds == mds) {
4172 dout("encode_dentry_release %p mds%d seq %d\n",
4173 dentry, mds, (int)di->lease_seq);
4174 rel->dname_len = cpu_to_le32(dentry->d_name.len);
4175 memcpy(*p, dentry->d_name.name, dentry->d_name.len);
4176 *p += dentry->d_name.len;
4177 rel->dname_seq = cpu_to_le32(di->lease_seq);
4178 __ceph_mdsc_drop_dentry_lease(dentry);
4180 spin_unlock(&dentry->d_lock);