1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
5 #include <linux/wait.h>
6 #include <linux/slab.h>
8 #include <linux/sched.h>
9 #include <linux/debugfs.h>
10 #include <linux/seq_file.h>
11 #include <linux/ratelimit.h>
12 #include <linux/bits.h>
13 #include <linux/ktime.h>
14 #include <linux/bitmap.h>
17 #include "mds_client.h"
19 #include <linux/ceph/ceph_features.h>
20 #include <linux/ceph/messenger.h>
21 #include <linux/ceph/decode.h>
22 #include <linux/ceph/pagelist.h>
23 #include <linux/ceph/auth.h>
24 #include <linux/ceph/debugfs.h>
26 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
29 * A cluster of MDS (metadata server) daemons is responsible for
30 * managing the file system namespace (the directory hierarchy and
31 * inodes) and for coordinating shared access to storage. Metadata is
32 * partitioning hierarchically across a number of servers, and that
33 * partition varies over time as the cluster adjusts the distribution
34 * in order to balance load.
36 * The MDS client is primarily responsible to managing synchronous
37 * metadata requests for operations like open, unlink, and so forth.
38 * If there is a MDS failure, we find out about it when we (possibly
39 * request and) receive a new MDS map, and can resubmit affected
42 * For the most part, though, we take advantage of a lossless
43 * communications channel to the MDS, and do not need to worry about
44 * timing out or resubmitting requests.
46 * We maintain a stateful "session" with each MDS we interact with.
47 * Within each session, we sent periodic heartbeat messages to ensure
48 * any capabilities or leases we have been issues remain valid. If
49 * the session times out and goes stale, our leases and capabilities
50 * are no longer valid.
53 struct ceph_reconnect_state {
54 struct ceph_mds_session *session;
55 int nr_caps, nr_realms;
56 struct ceph_pagelist *pagelist;
61 static void __wake_requests(struct ceph_mds_client *mdsc,
62 struct list_head *head);
63 static void ceph_cap_release_work(struct work_struct *work);
64 static void ceph_cap_reclaim_work(struct work_struct *work);
66 static const struct ceph_connection_operations mds_con_ops;
73 static int parse_reply_info_quota(void **p, void *end,
74 struct ceph_mds_reply_info_in *info)
76 u8 struct_v, struct_compat;
79 ceph_decode_8_safe(p, end, struct_v, bad);
80 ceph_decode_8_safe(p, end, struct_compat, bad);
81 /* struct_v is expected to be >= 1. we only
82 * understand encoding with struct_compat == 1. */
83 if (!struct_v || struct_compat != 1)
85 ceph_decode_32_safe(p, end, struct_len, bad);
86 ceph_decode_need(p, end, struct_len, bad);
87 end = *p + struct_len;
88 ceph_decode_64_safe(p, end, info->max_bytes, bad);
89 ceph_decode_64_safe(p, end, info->max_files, bad);
97 * parse individual inode info
99 static int parse_reply_info_in(void **p, void *end,
100 struct ceph_mds_reply_info_in *info,
106 if (features == (u64)-1) {
109 ceph_decode_8_safe(p, end, struct_v, bad);
110 ceph_decode_8_safe(p, end, struct_compat, bad);
111 /* struct_v is expected to be >= 1. we only understand
112 * encoding with struct_compat == 1. */
113 if (!struct_v || struct_compat != 1)
115 ceph_decode_32_safe(p, end, struct_len, bad);
116 ceph_decode_need(p, end, struct_len, bad);
117 end = *p + struct_len;
120 ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
122 *p += sizeof(struct ceph_mds_reply_inode) +
123 sizeof(*info->in->fragtree.splits) *
124 le32_to_cpu(info->in->fragtree.nsplits);
126 ceph_decode_32_safe(p, end, info->symlink_len, bad);
127 ceph_decode_need(p, end, info->symlink_len, bad);
129 *p += info->symlink_len;
131 ceph_decode_copy_safe(p, end, &info->dir_layout,
132 sizeof(info->dir_layout), bad);
133 ceph_decode_32_safe(p, end, info->xattr_len, bad);
134 ceph_decode_need(p, end, info->xattr_len, bad);
135 info->xattr_data = *p;
136 *p += info->xattr_len;
138 if (features == (u64)-1) {
140 ceph_decode_64_safe(p, end, info->inline_version, bad);
141 ceph_decode_32_safe(p, end, info->inline_len, bad);
142 ceph_decode_need(p, end, info->inline_len, bad);
143 info->inline_data = *p;
144 *p += info->inline_len;
146 err = parse_reply_info_quota(p, end, info);
150 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
151 if (info->pool_ns_len > 0) {
152 ceph_decode_need(p, end, info->pool_ns_len, bad);
153 info->pool_ns_data = *p;
154 *p += info->pool_ns_len;
158 ceph_decode_need(p, end, sizeof(info->btime), bad);
159 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
161 /* change attribute */
162 ceph_decode_64_safe(p, end, info->change_attr, bad);
166 ceph_decode_32_safe(p, end, info->dir_pin, bad);
168 info->dir_pin = -ENODATA;
171 /* snapshot birth time, remains zero for v<=2 */
173 ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
174 ceph_decode_copy(p, &info->snap_btime,
175 sizeof(info->snap_btime));
177 memset(&info->snap_btime, 0, sizeof(info->snap_btime));
180 /* snapshot count, remains zero for v<=3 */
182 ceph_decode_64_safe(p, end, info->rsnaps, bad);
189 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
190 ceph_decode_64_safe(p, end, info->inline_version, bad);
191 ceph_decode_32_safe(p, end, info->inline_len, bad);
192 ceph_decode_need(p, end, info->inline_len, bad);
193 info->inline_data = *p;
194 *p += info->inline_len;
196 info->inline_version = CEPH_INLINE_NONE;
198 if (features & CEPH_FEATURE_MDS_QUOTA) {
199 err = parse_reply_info_quota(p, end, info);
207 info->pool_ns_len = 0;
208 info->pool_ns_data = NULL;
209 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
210 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
211 if (info->pool_ns_len > 0) {
212 ceph_decode_need(p, end, info->pool_ns_len, bad);
213 info->pool_ns_data = *p;
214 *p += info->pool_ns_len;
218 if (features & CEPH_FEATURE_FS_BTIME) {
219 ceph_decode_need(p, end, sizeof(info->btime), bad);
220 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
221 ceph_decode_64_safe(p, end, info->change_attr, bad);
224 info->dir_pin = -ENODATA;
225 /* info->snap_btime and info->rsnaps remain zero */
234 static int parse_reply_info_dir(void **p, void *end,
235 struct ceph_mds_reply_dirfrag **dirfrag,
238 if (features == (u64)-1) {
239 u8 struct_v, struct_compat;
241 ceph_decode_8_safe(p, end, struct_v, bad);
242 ceph_decode_8_safe(p, end, struct_compat, bad);
243 /* struct_v is expected to be >= 1. we only understand
244 * encoding whose struct_compat == 1. */
245 if (!struct_v || struct_compat != 1)
247 ceph_decode_32_safe(p, end, struct_len, bad);
248 ceph_decode_need(p, end, struct_len, bad);
249 end = *p + struct_len;
252 ceph_decode_need(p, end, sizeof(**dirfrag), bad);
254 *p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
255 if (unlikely(*p > end))
257 if (features == (u64)-1)
264 static int parse_reply_info_lease(void **p, void *end,
265 struct ceph_mds_reply_lease **lease,
268 if (features == (u64)-1) {
269 u8 struct_v, struct_compat;
271 ceph_decode_8_safe(p, end, struct_v, bad);
272 ceph_decode_8_safe(p, end, struct_compat, bad);
273 /* struct_v is expected to be >= 1. we only understand
274 * encoding whose struct_compat == 1. */
275 if (!struct_v || struct_compat != 1)
277 ceph_decode_32_safe(p, end, struct_len, bad);
278 ceph_decode_need(p, end, struct_len, bad);
279 end = *p + struct_len;
282 ceph_decode_need(p, end, sizeof(**lease), bad);
284 *p += sizeof(**lease);
285 if (features == (u64)-1)
293 * parse a normal reply, which may contain a (dir+)dentry and/or a
296 static int parse_reply_info_trace(void **p, void *end,
297 struct ceph_mds_reply_info_parsed *info,
302 if (info->head->is_dentry) {
303 err = parse_reply_info_in(p, end, &info->diri, features);
307 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
311 ceph_decode_32_safe(p, end, info->dname_len, bad);
312 ceph_decode_need(p, end, info->dname_len, bad);
314 *p += info->dname_len;
316 err = parse_reply_info_lease(p, end, &info->dlease, features);
321 if (info->head->is_target) {
322 err = parse_reply_info_in(p, end, &info->targeti, features);
327 if (unlikely(*p != end))
334 pr_err("problem parsing mds trace %d\n", err);
339 * parse readdir results
341 static int parse_reply_info_readdir(void **p, void *end,
342 struct ceph_mds_reply_info_parsed *info,
348 err = parse_reply_info_dir(p, end, &info->dir_dir, features);
352 ceph_decode_need(p, end, sizeof(num) + 2, bad);
353 num = ceph_decode_32(p);
355 u16 flags = ceph_decode_16(p);
356 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
357 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
358 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
359 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
364 BUG_ON(!info->dir_entries);
365 if ((unsigned long)(info->dir_entries + num) >
366 (unsigned long)info->dir_entries + info->dir_buf_size) {
367 pr_err("dir contents are larger than expected\n");
374 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
376 ceph_decode_32_safe(p, end, rde->name_len, bad);
377 ceph_decode_need(p, end, rde->name_len, bad);
380 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
383 err = parse_reply_info_lease(p, end, &rde->lease, features);
387 err = parse_reply_info_in(p, end, &rde->inode, features);
390 /* ceph_readdir_prepopulate() will update it */
397 /* Skip over any unrecognized fields */
404 pr_err("problem parsing dir contents %d\n", err);
409 * parse fcntl F_GETLK results
411 static int parse_reply_info_filelock(void **p, void *end,
412 struct ceph_mds_reply_info_parsed *info,
415 if (*p + sizeof(*info->filelock_reply) > end)
418 info->filelock_reply = *p;
420 /* Skip over any unrecognized fields */
428 #if BITS_PER_LONG == 64
430 #define DELEGATED_INO_AVAILABLE xa_mk_value(1)
432 static int ceph_parse_deleg_inos(void **p, void *end,
433 struct ceph_mds_session *s)
437 ceph_decode_32_safe(p, end, sets, bad);
438 dout("got %u sets of delegated inodes\n", sets);
442 ceph_decode_64_safe(p, end, start, bad);
443 ceph_decode_64_safe(p, end, len, bad);
445 /* Don't accept a delegation of system inodes */
446 if (start < CEPH_INO_SYSTEM_BASE) {
447 pr_warn_ratelimited("ceph: ignoring reserved inode range delegation (start=0x%llx len=0x%llx)\n",
452 int err = xa_insert(&s->s_delegated_inos, ino = start++,
453 DELEGATED_INO_AVAILABLE,
456 dout("added delegated inode 0x%llx\n",
458 } else if (err == -EBUSY) {
459 pr_warn("ceph: MDS delegated inode 0x%llx more than once.\n",
471 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
476 xa_for_each(&s->s_delegated_inos, ino, val) {
477 val = xa_erase(&s->s_delegated_inos, ino);
478 if (val == DELEGATED_INO_AVAILABLE)
484 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
486 return xa_insert(&s->s_delegated_inos, ino, DELEGATED_INO_AVAILABLE,
489 #else /* BITS_PER_LONG == 64 */
491 * FIXME: xarrays can't handle 64-bit indexes on a 32-bit arch. For now, just
492 * ignore delegated_inos on 32 bit arch. Maybe eventually add xarrays for top
495 static int ceph_parse_deleg_inos(void **p, void *end,
496 struct ceph_mds_session *s)
500 ceph_decode_32_safe(p, end, sets, bad);
502 ceph_decode_skip_n(p, end, sets * 2 * sizeof(__le64), bad);
508 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
513 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
517 #endif /* BITS_PER_LONG == 64 */
520 * parse create results
522 static int parse_reply_info_create(void **p, void *end,
523 struct ceph_mds_reply_info_parsed *info,
524 u64 features, struct ceph_mds_session *s)
528 if (features == (u64)-1 ||
529 (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
531 /* Malformed reply? */
532 info->has_create_ino = false;
533 } else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
534 info->has_create_ino = true;
535 /* struct_v, struct_compat, and len */
536 ceph_decode_skip_n(p, end, 2 + sizeof(u32), bad);
537 ceph_decode_64_safe(p, end, info->ino, bad);
538 ret = ceph_parse_deleg_inos(p, end, s);
543 ceph_decode_64_safe(p, end, info->ino, bad);
544 info->has_create_ino = true;
551 /* Skip over any unrecognized fields */
559 * parse extra results
561 static int parse_reply_info_extra(void **p, void *end,
562 struct ceph_mds_reply_info_parsed *info,
563 u64 features, struct ceph_mds_session *s)
565 u32 op = le32_to_cpu(info->head->op);
567 if (op == CEPH_MDS_OP_GETFILELOCK)
568 return parse_reply_info_filelock(p, end, info, features);
569 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
570 return parse_reply_info_readdir(p, end, info, features);
571 else if (op == CEPH_MDS_OP_CREATE)
572 return parse_reply_info_create(p, end, info, features, s);
578 * parse entire mds reply
580 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
581 struct ceph_mds_reply_info_parsed *info,
588 info->head = msg->front.iov_base;
589 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
590 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
593 ceph_decode_32_safe(&p, end, len, bad);
595 ceph_decode_need(&p, end, len, bad);
596 err = parse_reply_info_trace(&p, p+len, info, features);
602 ceph_decode_32_safe(&p, end, len, bad);
604 ceph_decode_need(&p, end, len, bad);
605 err = parse_reply_info_extra(&p, p+len, info, features, s);
611 ceph_decode_32_safe(&p, end, len, bad);
612 info->snapblob_len = len;
623 pr_err("mds parse_reply err %d\n", err);
627 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
629 if (!info->dir_entries)
631 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
638 const char *ceph_session_state_name(int s)
641 case CEPH_MDS_SESSION_NEW: return "new";
642 case CEPH_MDS_SESSION_OPENING: return "opening";
643 case CEPH_MDS_SESSION_OPEN: return "open";
644 case CEPH_MDS_SESSION_HUNG: return "hung";
645 case CEPH_MDS_SESSION_CLOSING: return "closing";
646 case CEPH_MDS_SESSION_CLOSED: return "closed";
647 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
648 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
649 case CEPH_MDS_SESSION_REJECTED: return "rejected";
650 default: return "???";
654 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
656 if (refcount_inc_not_zero(&s->s_ref))
661 void ceph_put_mds_session(struct ceph_mds_session *s)
663 if (IS_ERR_OR_NULL(s))
666 if (refcount_dec_and_test(&s->s_ref)) {
667 if (s->s_auth.authorizer)
668 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
669 WARN_ON(mutex_is_locked(&s->s_mutex));
670 xa_destroy(&s->s_delegated_inos);
676 * called under mdsc->mutex
678 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
681 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
683 return ceph_get_mds_session(mdsc->sessions[mds]);
686 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
688 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
694 static int __verify_registered_session(struct ceph_mds_client *mdsc,
695 struct ceph_mds_session *s)
697 if (s->s_mds >= mdsc->max_sessions ||
698 mdsc->sessions[s->s_mds] != s)
704 * create+register a new session for given mds.
705 * called under mdsc->mutex.
707 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
710 struct ceph_mds_session *s;
712 if (mds >= mdsc->mdsmap->possible_max_rank)
713 return ERR_PTR(-EINVAL);
715 s = kzalloc(sizeof(*s), GFP_NOFS);
717 return ERR_PTR(-ENOMEM);
719 if (mds >= mdsc->max_sessions) {
720 int newmax = 1 << get_count_order(mds + 1);
721 struct ceph_mds_session **sa;
723 dout("%s: realloc to %d\n", __func__, newmax);
724 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
727 if (mdsc->sessions) {
728 memcpy(sa, mdsc->sessions,
729 mdsc->max_sessions * sizeof(void *));
730 kfree(mdsc->sessions);
733 mdsc->max_sessions = newmax;
736 dout("%s: mds%d\n", __func__, mds);
739 s->s_state = CEPH_MDS_SESSION_NEW;
740 mutex_init(&s->s_mutex);
742 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
744 atomic_set(&s->s_cap_gen, 1);
745 s->s_cap_ttl = jiffies - 1;
747 spin_lock_init(&s->s_cap_lock);
748 INIT_LIST_HEAD(&s->s_caps);
749 refcount_set(&s->s_ref, 1);
750 INIT_LIST_HEAD(&s->s_waiting);
751 INIT_LIST_HEAD(&s->s_unsafe);
752 xa_init(&s->s_delegated_inos);
753 INIT_LIST_HEAD(&s->s_cap_releases);
754 INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
756 INIT_LIST_HEAD(&s->s_cap_dirty);
757 INIT_LIST_HEAD(&s->s_cap_flushing);
759 mdsc->sessions[mds] = s;
760 atomic_inc(&mdsc->num_sessions);
761 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
763 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
764 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
770 return ERR_PTR(-ENOMEM);
774 * called under mdsc->mutex
776 static void __unregister_session(struct ceph_mds_client *mdsc,
777 struct ceph_mds_session *s)
779 dout("__unregister_session mds%d %p\n", s->s_mds, s);
780 BUG_ON(mdsc->sessions[s->s_mds] != s);
781 mdsc->sessions[s->s_mds] = NULL;
782 ceph_con_close(&s->s_con);
783 ceph_put_mds_session(s);
784 atomic_dec(&mdsc->num_sessions);
788 * drop session refs in request.
790 * should be last request ref, or hold mdsc->mutex
792 static void put_request_session(struct ceph_mds_request *req)
794 if (req->r_session) {
795 ceph_put_mds_session(req->r_session);
796 req->r_session = NULL;
800 void ceph_mdsc_iterate_sessions(struct ceph_mds_client *mdsc,
801 void (*cb)(struct ceph_mds_session *),
806 mutex_lock(&mdsc->mutex);
807 for (mds = 0; mds < mdsc->max_sessions; ++mds) {
808 struct ceph_mds_session *s;
810 s = __ceph_lookup_mds_session(mdsc, mds);
814 if (check_state && !check_session_state(s)) {
815 ceph_put_mds_session(s);
819 mutex_unlock(&mdsc->mutex);
821 ceph_put_mds_session(s);
822 mutex_lock(&mdsc->mutex);
824 mutex_unlock(&mdsc->mutex);
827 void ceph_mdsc_release_request(struct kref *kref)
829 struct ceph_mds_request *req = container_of(kref,
830 struct ceph_mds_request,
832 ceph_mdsc_release_dir_caps_no_check(req);
833 destroy_reply_info(&req->r_reply_info);
835 ceph_msg_put(req->r_request);
837 ceph_msg_put(req->r_reply);
839 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
843 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
846 iput(req->r_target_inode);
849 if (req->r_old_dentry)
850 dput(req->r_old_dentry);
851 if (req->r_old_dentry_dir) {
853 * track (and drop pins for) r_old_dentry_dir
854 * separately, since r_old_dentry's d_parent may have
855 * changed between the dir mutex being dropped and
856 * this request being freed.
858 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
860 iput(req->r_old_dentry_dir);
864 put_cred(req->r_cred);
866 ceph_pagelist_release(req->r_pagelist);
867 put_request_session(req);
868 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
869 WARN_ON_ONCE(!list_empty(&req->r_wait));
870 kmem_cache_free(ceph_mds_request_cachep, req);
873 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
876 * lookup session, bump ref if found.
878 * called under mdsc->mutex.
880 static struct ceph_mds_request *
881 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
883 struct ceph_mds_request *req;
885 req = lookup_request(&mdsc->request_tree, tid);
887 ceph_mdsc_get_request(req);
893 * Register an in-flight request, and assign a tid. Link to directory
894 * are modifying (if any).
896 * Called under mdsc->mutex.
898 static void __register_request(struct ceph_mds_client *mdsc,
899 struct ceph_mds_request *req,
904 req->r_tid = ++mdsc->last_tid;
905 if (req->r_num_caps) {
906 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
909 pr_err("__register_request %p "
910 "failed to reserve caps: %d\n", req, ret);
911 /* set req->r_err to fail early from __do_request */
916 dout("__register_request %p tid %lld\n", req, req->r_tid);
917 ceph_mdsc_get_request(req);
918 insert_request(&mdsc->request_tree, req);
920 req->r_cred = get_current_cred();
922 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
923 mdsc->oldest_tid = req->r_tid;
926 struct ceph_inode_info *ci = ceph_inode(dir);
929 req->r_unsafe_dir = dir;
930 spin_lock(&ci->i_unsafe_lock);
931 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
932 spin_unlock(&ci->i_unsafe_lock);
936 static void __unregister_request(struct ceph_mds_client *mdsc,
937 struct ceph_mds_request *req)
939 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
941 /* Never leave an unregistered request on an unsafe list! */
942 list_del_init(&req->r_unsafe_item);
944 if (req->r_tid == mdsc->oldest_tid) {
945 struct rb_node *p = rb_next(&req->r_node);
946 mdsc->oldest_tid = 0;
948 struct ceph_mds_request *next_req =
949 rb_entry(p, struct ceph_mds_request, r_node);
950 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
951 mdsc->oldest_tid = next_req->r_tid;
958 erase_request(&mdsc->request_tree, req);
960 if (req->r_unsafe_dir) {
961 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
962 spin_lock(&ci->i_unsafe_lock);
963 list_del_init(&req->r_unsafe_dir_item);
964 spin_unlock(&ci->i_unsafe_lock);
966 if (req->r_target_inode &&
967 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
968 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
969 spin_lock(&ci->i_unsafe_lock);
970 list_del_init(&req->r_unsafe_target_item);
971 spin_unlock(&ci->i_unsafe_lock);
974 if (req->r_unsafe_dir) {
975 iput(req->r_unsafe_dir);
976 req->r_unsafe_dir = NULL;
979 complete_all(&req->r_safe_completion);
981 ceph_mdsc_put_request(req);
985 * Walk back up the dentry tree until we hit a dentry representing a
986 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
987 * when calling this) to ensure that the objects won't disappear while we're
988 * working with them. Once we hit a candidate dentry, we attempt to take a
989 * reference to it, and return that as the result.
991 static struct inode *get_nonsnap_parent(struct dentry *dentry)
993 struct inode *inode = NULL;
995 while (dentry && !IS_ROOT(dentry)) {
996 inode = d_inode_rcu(dentry);
997 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
999 dentry = dentry->d_parent;
1002 inode = igrab(inode);
1007 * Choose mds to send request to next. If there is a hint set in the
1008 * request (e.g., due to a prior forward hint from the mds), use that.
1009 * Otherwise, consult frag tree and/or caps to identify the
1010 * appropriate mds. If all else fails, choose randomly.
1012 * Called under mdsc->mutex.
1014 static int __choose_mds(struct ceph_mds_client *mdsc,
1015 struct ceph_mds_request *req,
1018 struct inode *inode;
1019 struct ceph_inode_info *ci;
1020 struct ceph_cap *cap;
1021 int mode = req->r_direct_mode;
1023 u32 hash = req->r_direct_hash;
1024 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1030 * is there a specific mds we should try? ignore hint if we have
1031 * no session and the mds is not up (active or recovering).
1033 if (req->r_resend_mds >= 0 &&
1034 (__have_session(mdsc, req->r_resend_mds) ||
1035 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1036 dout("%s using resend_mds mds%d\n", __func__,
1038 return req->r_resend_mds;
1041 if (mode == USE_RANDOM_MDS)
1046 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1047 inode = req->r_inode;
1050 /* req->r_dentry is non-null for LSSNAP request */
1052 inode = get_nonsnap_parent(req->r_dentry);
1054 dout("%s using snapdir's parent %p\n", __func__, inode);
1056 } else if (req->r_dentry) {
1057 /* ignore race with rename; old or new d_parent is okay */
1058 struct dentry *parent;
1062 parent = READ_ONCE(req->r_dentry->d_parent);
1063 dir = req->r_parent ? : d_inode_rcu(parent);
1065 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1066 /* not this fs or parent went negative */
1067 inode = d_inode(req->r_dentry);
1070 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1071 /* direct snapped/virtual snapdir requests
1072 * based on parent dir inode */
1073 inode = get_nonsnap_parent(parent);
1074 dout("%s using nonsnap parent %p\n", __func__, inode);
1077 inode = d_inode(req->r_dentry);
1078 if (!inode || mode == USE_AUTH_MDS) {
1081 hash = ceph_dentry_hash(dir, req->r_dentry);
1090 dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1094 ci = ceph_inode(inode);
1096 if (is_hash && S_ISDIR(inode->i_mode)) {
1097 struct ceph_inode_frag frag;
1100 ceph_choose_frag(ci, hash, &frag, &found);
1102 if (mode == USE_ANY_MDS && frag.ndist > 0) {
1105 /* choose a random replica */
1106 get_random_bytes(&r, 1);
1109 dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1110 __func__, inode, ceph_vinop(inode),
1111 frag.frag, mds, (int)r, frag.ndist);
1112 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1113 CEPH_MDS_STATE_ACTIVE &&
1114 !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1118 /* since this file/dir wasn't known to be
1119 * replicated, then we want to look for the
1120 * authoritative mds. */
1121 if (frag.mds >= 0) {
1122 /* choose auth mds */
1124 dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1125 __func__, inode, ceph_vinop(inode),
1127 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1128 CEPH_MDS_STATE_ACTIVE) {
1129 if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1134 mode = USE_AUTH_MDS;
1138 spin_lock(&ci->i_ceph_lock);
1140 if (mode == USE_AUTH_MDS)
1141 cap = ci->i_auth_cap;
1142 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1143 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1145 spin_unlock(&ci->i_ceph_lock);
1149 mds = cap->session->s_mds;
1150 dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1151 inode, ceph_vinop(inode), mds,
1152 cap == ci->i_auth_cap ? "auth " : "", cap);
1153 spin_unlock(&ci->i_ceph_lock);
1162 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1163 dout("%s chose random mds%d\n", __func__, mds);
1171 struct ceph_msg *ceph_create_session_msg(u32 op, u64 seq)
1173 struct ceph_msg *msg;
1174 struct ceph_mds_session_head *h;
1176 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1179 pr_err("ENOMEM creating session %s msg\n",
1180 ceph_session_op_name(op));
1183 h = msg->front.iov_base;
1184 h->op = cpu_to_le32(op);
1185 h->seq = cpu_to_le64(seq);
1190 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1191 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1192 static int encode_supported_features(void **p, void *end)
1194 static const size_t count = ARRAY_SIZE(feature_bits);
1198 size_t size = FEATURE_BYTES(count);
1200 if (WARN_ON_ONCE(*p + 4 + size > end))
1203 ceph_encode_32(p, size);
1204 memset(*p, 0, size);
1205 for (i = 0; i < count; i++)
1206 ((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
1209 if (WARN_ON_ONCE(*p + 4 > end))
1212 ceph_encode_32(p, 0);
1218 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1219 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1220 static int encode_metric_spec(void **p, void *end)
1222 static const size_t count = ARRAY_SIZE(metric_bits);
1225 if (WARN_ON_ONCE(*p + 2 > end))
1228 ceph_encode_8(p, 1); /* version */
1229 ceph_encode_8(p, 1); /* compat */
1233 size_t size = METRIC_BYTES(count);
1235 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1238 /* metric spec info length */
1239 ceph_encode_32(p, 4 + size);
1242 ceph_encode_32(p, size);
1243 memset(*p, 0, size);
1244 for (i = 0; i < count; i++)
1245 ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1248 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1251 /* metric spec info length */
1252 ceph_encode_32(p, 4);
1254 ceph_encode_32(p, 0);
1261 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1262 * to include additional client metadata fields.
1264 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1266 struct ceph_msg *msg;
1267 struct ceph_mds_session_head *h;
1269 int extra_bytes = 0;
1270 int metadata_key_count = 0;
1271 struct ceph_options *opt = mdsc->fsc->client->options;
1272 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1277 const char* metadata[][2] = {
1278 {"hostname", mdsc->nodename},
1279 {"kernel_version", init_utsname()->release},
1280 {"entity_id", opt->name ? : ""},
1281 {"root", fsopt->server_path ? : "/"},
1285 /* Calculate serialized length of metadata */
1286 extra_bytes = 4; /* map length */
1287 for (i = 0; metadata[i][0]; ++i) {
1288 extra_bytes += 8 + strlen(metadata[i][0]) +
1289 strlen(metadata[i][1]);
1290 metadata_key_count++;
1293 /* supported feature */
1295 count = ARRAY_SIZE(feature_bits);
1297 size = FEATURE_BYTES(count);
1298 extra_bytes += 4 + size;
1302 count = ARRAY_SIZE(metric_bits);
1304 size = METRIC_BYTES(count);
1305 extra_bytes += 2 + 4 + 4 + size;
1307 /* Allocate the message */
1308 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1311 pr_err("ENOMEM creating session open msg\n");
1312 return ERR_PTR(-ENOMEM);
1314 p = msg->front.iov_base;
1315 end = p + msg->front.iov_len;
1318 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1319 h->seq = cpu_to_le64(seq);
1322 * Serialize client metadata into waiting buffer space, using
1323 * the format that userspace expects for map<string, string>
1325 * ClientSession messages with metadata are v4
1327 msg->hdr.version = cpu_to_le16(4);
1328 msg->hdr.compat_version = cpu_to_le16(1);
1330 /* The write pointer, following the session_head structure */
1333 /* Number of entries in the map */
1334 ceph_encode_32(&p, metadata_key_count);
1336 /* Two length-prefixed strings for each entry in the map */
1337 for (i = 0; metadata[i][0]; ++i) {
1338 size_t const key_len = strlen(metadata[i][0]);
1339 size_t const val_len = strlen(metadata[i][1]);
1341 ceph_encode_32(&p, key_len);
1342 memcpy(p, metadata[i][0], key_len);
1344 ceph_encode_32(&p, val_len);
1345 memcpy(p, metadata[i][1], val_len);
1349 ret = encode_supported_features(&p, end);
1351 pr_err("encode_supported_features failed!\n");
1353 return ERR_PTR(ret);
1356 ret = encode_metric_spec(&p, end);
1358 pr_err("encode_metric_spec failed!\n");
1360 return ERR_PTR(ret);
1363 msg->front.iov_len = p - msg->front.iov_base;
1364 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1370 * send session open request.
1372 * called under mdsc->mutex
1374 static int __open_session(struct ceph_mds_client *mdsc,
1375 struct ceph_mds_session *session)
1377 struct ceph_msg *msg;
1379 int mds = session->s_mds;
1381 /* wait for mds to go active? */
1382 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1383 dout("open_session to mds%d (%s)\n", mds,
1384 ceph_mds_state_name(mstate));
1385 session->s_state = CEPH_MDS_SESSION_OPENING;
1386 session->s_renew_requested = jiffies;
1388 /* send connect message */
1389 msg = create_session_open_msg(mdsc, session->s_seq);
1391 return PTR_ERR(msg);
1392 ceph_con_send(&session->s_con, msg);
1397 * open sessions for any export targets for the given mds
1399 * called under mdsc->mutex
1401 static struct ceph_mds_session *
1402 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1404 struct ceph_mds_session *session;
1407 session = __ceph_lookup_mds_session(mdsc, target);
1409 session = register_session(mdsc, target);
1410 if (IS_ERR(session))
1413 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1414 session->s_state == CEPH_MDS_SESSION_CLOSING) {
1415 ret = __open_session(mdsc, session);
1417 return ERR_PTR(ret);
1423 struct ceph_mds_session *
1424 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1426 struct ceph_mds_session *session;
1428 dout("open_export_target_session to mds%d\n", target);
1430 mutex_lock(&mdsc->mutex);
1431 session = __open_export_target_session(mdsc, target);
1432 mutex_unlock(&mdsc->mutex);
1437 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1438 struct ceph_mds_session *session)
1440 struct ceph_mds_info *mi;
1441 struct ceph_mds_session *ts;
1442 int i, mds = session->s_mds;
1444 if (mds >= mdsc->mdsmap->possible_max_rank)
1447 mi = &mdsc->mdsmap->m_info[mds];
1448 dout("open_export_target_sessions for mds%d (%d targets)\n",
1449 session->s_mds, mi->num_export_targets);
1451 for (i = 0; i < mi->num_export_targets; i++) {
1452 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1453 ceph_put_mds_session(ts);
1457 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1458 struct ceph_mds_session *session)
1460 mutex_lock(&mdsc->mutex);
1461 __open_export_target_sessions(mdsc, session);
1462 mutex_unlock(&mdsc->mutex);
1469 static void detach_cap_releases(struct ceph_mds_session *session,
1470 struct list_head *target)
1472 lockdep_assert_held(&session->s_cap_lock);
1474 list_splice_init(&session->s_cap_releases, target);
1475 session->s_num_cap_releases = 0;
1476 dout("dispose_cap_releases mds%d\n", session->s_mds);
1479 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1480 struct list_head *dispose)
1482 while (!list_empty(dispose)) {
1483 struct ceph_cap *cap;
1484 /* zero out the in-progress message */
1485 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1486 list_del(&cap->session_caps);
1487 ceph_put_cap(mdsc, cap);
1491 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1492 struct ceph_mds_session *session)
1494 struct ceph_mds_request *req;
1496 struct ceph_inode_info *ci;
1498 dout("cleanup_session_requests mds%d\n", session->s_mds);
1499 mutex_lock(&mdsc->mutex);
1500 while (!list_empty(&session->s_unsafe)) {
1501 req = list_first_entry(&session->s_unsafe,
1502 struct ceph_mds_request, r_unsafe_item);
1503 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1505 if (req->r_target_inode) {
1506 /* dropping unsafe change of inode's attributes */
1507 ci = ceph_inode(req->r_target_inode);
1508 errseq_set(&ci->i_meta_err, -EIO);
1510 if (req->r_unsafe_dir) {
1511 /* dropping unsafe directory operation */
1512 ci = ceph_inode(req->r_unsafe_dir);
1513 errseq_set(&ci->i_meta_err, -EIO);
1515 __unregister_request(mdsc, req);
1517 /* zero r_attempts, so kick_requests() will re-send requests */
1518 p = rb_first(&mdsc->request_tree);
1520 req = rb_entry(p, struct ceph_mds_request, r_node);
1522 if (req->r_session &&
1523 req->r_session->s_mds == session->s_mds)
1524 req->r_attempts = 0;
1526 mutex_unlock(&mdsc->mutex);
1530 * Helper to safely iterate over all caps associated with a session, with
1531 * special care taken to handle a racing __ceph_remove_cap().
1533 * Caller must hold session s_mutex.
1535 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1536 int (*cb)(struct inode *, struct ceph_cap *,
1539 struct list_head *p;
1540 struct ceph_cap *cap;
1541 struct inode *inode, *last_inode = NULL;
1542 struct ceph_cap *old_cap = NULL;
1545 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1546 spin_lock(&session->s_cap_lock);
1547 p = session->s_caps.next;
1548 while (p != &session->s_caps) {
1549 cap = list_entry(p, struct ceph_cap, session_caps);
1550 inode = igrab(&cap->ci->vfs_inode);
1555 session->s_cap_iterator = cap;
1556 spin_unlock(&session->s_cap_lock);
1563 ceph_put_cap(session->s_mdsc, old_cap);
1567 ret = cb(inode, cap, arg);
1570 spin_lock(&session->s_cap_lock);
1573 dout("iterate_session_caps finishing cap %p removal\n",
1575 BUG_ON(cap->session != session);
1576 cap->session = NULL;
1577 list_del_init(&cap->session_caps);
1578 session->s_nr_caps--;
1579 atomic64_dec(&session->s_mdsc->metric.total_caps);
1580 if (cap->queue_release)
1581 __ceph_queue_cap_release(session, cap);
1583 old_cap = cap; /* put_cap it w/o locks held */
1590 session->s_cap_iterator = NULL;
1591 spin_unlock(&session->s_cap_lock);
1595 ceph_put_cap(session->s_mdsc, old_cap);
1600 static int remove_capsnaps(struct ceph_mds_client *mdsc, struct inode *inode)
1602 struct ceph_inode_info *ci = ceph_inode(inode);
1603 struct ceph_cap_snap *capsnap;
1604 int capsnap_release = 0;
1606 lockdep_assert_held(&ci->i_ceph_lock);
1608 dout("removing capsnaps, ci is %p, inode is %p\n", ci, inode);
1610 while (!list_empty(&ci->i_cap_snaps)) {
1611 capsnap = list_first_entry(&ci->i_cap_snaps,
1612 struct ceph_cap_snap, ci_item);
1613 __ceph_remove_capsnap(inode, capsnap, NULL, NULL);
1614 ceph_put_snap_context(capsnap->context);
1615 ceph_put_cap_snap(capsnap);
1618 wake_up_all(&ci->i_cap_wq);
1619 wake_up_all(&mdsc->cap_flushing_wq);
1620 return capsnap_release;
1623 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1626 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1627 struct ceph_mds_client *mdsc = fsc->mdsc;
1628 struct ceph_inode_info *ci = ceph_inode(inode);
1629 LIST_HEAD(to_remove);
1630 bool dirty_dropped = false;
1631 bool invalidate = false;
1632 int capsnap_release = 0;
1634 dout("removing cap %p, ci is %p, inode is %p\n",
1635 cap, ci, &ci->vfs_inode);
1636 spin_lock(&ci->i_ceph_lock);
1637 __ceph_remove_cap(cap, false);
1638 if (!ci->i_auth_cap) {
1639 struct ceph_cap_flush *cf;
1641 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1642 if (inode->i_data.nrpages > 0)
1644 if (ci->i_wrbuffer_ref > 0)
1645 mapping_set_error(&inode->i_data, -EIO);
1648 while (!list_empty(&ci->i_cap_flush_list)) {
1649 cf = list_first_entry(&ci->i_cap_flush_list,
1650 struct ceph_cap_flush, i_list);
1651 list_move(&cf->i_list, &to_remove);
1654 spin_lock(&mdsc->cap_dirty_lock);
1656 list_for_each_entry(cf, &to_remove, i_list)
1657 list_del_init(&cf->g_list);
1659 if (!list_empty(&ci->i_dirty_item)) {
1660 pr_warn_ratelimited(
1661 " dropping dirty %s state for %p %lld\n",
1662 ceph_cap_string(ci->i_dirty_caps),
1663 inode, ceph_ino(inode));
1664 ci->i_dirty_caps = 0;
1665 list_del_init(&ci->i_dirty_item);
1666 dirty_dropped = true;
1668 if (!list_empty(&ci->i_flushing_item)) {
1669 pr_warn_ratelimited(
1670 " dropping dirty+flushing %s state for %p %lld\n",
1671 ceph_cap_string(ci->i_flushing_caps),
1672 inode, ceph_ino(inode));
1673 ci->i_flushing_caps = 0;
1674 list_del_init(&ci->i_flushing_item);
1675 mdsc->num_cap_flushing--;
1676 dirty_dropped = true;
1678 spin_unlock(&mdsc->cap_dirty_lock);
1680 if (dirty_dropped) {
1681 errseq_set(&ci->i_meta_err, -EIO);
1683 if (ci->i_wrbuffer_ref_head == 0 &&
1684 ci->i_wr_ref == 0 &&
1685 ci->i_dirty_caps == 0 &&
1686 ci->i_flushing_caps == 0) {
1687 ceph_put_snap_context(ci->i_head_snapc);
1688 ci->i_head_snapc = NULL;
1692 if (atomic_read(&ci->i_filelock_ref) > 0) {
1693 /* make further file lock syscall return -EIO */
1694 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1695 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1696 inode, ceph_ino(inode));
1699 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1700 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1701 ci->i_prealloc_cap_flush = NULL;
1704 if (!list_empty(&ci->i_cap_snaps))
1705 capsnap_release = remove_capsnaps(mdsc, inode);
1707 spin_unlock(&ci->i_ceph_lock);
1708 while (!list_empty(&to_remove)) {
1709 struct ceph_cap_flush *cf;
1710 cf = list_first_entry(&to_remove,
1711 struct ceph_cap_flush, i_list);
1712 list_del_init(&cf->i_list);
1713 if (!cf->is_capsnap)
1714 ceph_free_cap_flush(cf);
1717 wake_up_all(&ci->i_cap_wq);
1719 ceph_queue_invalidate(inode);
1722 while (capsnap_release--)
1728 * caller must hold session s_mutex
1730 static void remove_session_caps(struct ceph_mds_session *session)
1732 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1733 struct super_block *sb = fsc->sb;
1736 dout("remove_session_caps on %p\n", session);
1737 ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1739 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1741 spin_lock(&session->s_cap_lock);
1742 if (session->s_nr_caps > 0) {
1743 struct inode *inode;
1744 struct ceph_cap *cap, *prev = NULL;
1745 struct ceph_vino vino;
1747 * iterate_session_caps() skips inodes that are being
1748 * deleted, we need to wait until deletions are complete.
1749 * __wait_on_freeing_inode() is designed for the job,
1750 * but it is not exported, so use lookup inode function
1753 while (!list_empty(&session->s_caps)) {
1754 cap = list_entry(session->s_caps.next,
1755 struct ceph_cap, session_caps);
1759 vino = cap->ci->i_vino;
1760 spin_unlock(&session->s_cap_lock);
1762 inode = ceph_find_inode(sb, vino);
1765 spin_lock(&session->s_cap_lock);
1769 // drop cap expires and unlock s_cap_lock
1770 detach_cap_releases(session, &dispose);
1772 BUG_ON(session->s_nr_caps > 0);
1773 BUG_ON(!list_empty(&session->s_cap_flushing));
1774 spin_unlock(&session->s_cap_lock);
1775 dispose_cap_releases(session->s_mdsc, &dispose);
1785 * wake up any threads waiting on this session's caps. if the cap is
1786 * old (didn't get renewed on the client reconnect), remove it now.
1788 * caller must hold s_mutex.
1790 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1793 struct ceph_inode_info *ci = ceph_inode(inode);
1794 unsigned long ev = (unsigned long)arg;
1796 if (ev == RECONNECT) {
1797 spin_lock(&ci->i_ceph_lock);
1798 ci->i_wanted_max_size = 0;
1799 ci->i_requested_max_size = 0;
1800 spin_unlock(&ci->i_ceph_lock);
1801 } else if (ev == RENEWCAPS) {
1802 if (cap->cap_gen < atomic_read(&cap->session->s_cap_gen)) {
1803 /* mds did not re-issue stale cap */
1804 spin_lock(&ci->i_ceph_lock);
1805 cap->issued = cap->implemented = CEPH_CAP_PIN;
1806 spin_unlock(&ci->i_ceph_lock);
1808 } else if (ev == FORCE_RO) {
1810 wake_up_all(&ci->i_cap_wq);
1814 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1816 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1817 ceph_iterate_session_caps(session, wake_up_session_cb,
1818 (void *)(unsigned long)ev);
1822 * Send periodic message to MDS renewing all currently held caps. The
1823 * ack will reset the expiration for all caps from this session.
1825 * caller holds s_mutex
1827 static int send_renew_caps(struct ceph_mds_client *mdsc,
1828 struct ceph_mds_session *session)
1830 struct ceph_msg *msg;
1833 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1834 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1835 pr_info("mds%d caps stale\n", session->s_mds);
1836 session->s_renew_requested = jiffies;
1838 /* do not try to renew caps until a recovering mds has reconnected
1839 * with its clients. */
1840 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1841 if (state < CEPH_MDS_STATE_RECONNECT) {
1842 dout("send_renew_caps ignoring mds%d (%s)\n",
1843 session->s_mds, ceph_mds_state_name(state));
1847 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1848 ceph_mds_state_name(state));
1849 msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1850 ++session->s_renew_seq);
1853 ceph_con_send(&session->s_con, msg);
1857 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1858 struct ceph_mds_session *session, u64 seq)
1860 struct ceph_msg *msg;
1862 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1863 session->s_mds, ceph_session_state_name(session->s_state), seq);
1864 msg = ceph_create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1867 ceph_con_send(&session->s_con, msg);
1873 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1875 * Called under session->s_mutex
1877 static void renewed_caps(struct ceph_mds_client *mdsc,
1878 struct ceph_mds_session *session, int is_renew)
1883 spin_lock(&session->s_cap_lock);
1884 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1886 session->s_cap_ttl = session->s_renew_requested +
1887 mdsc->mdsmap->m_session_timeout*HZ;
1890 if (time_before(jiffies, session->s_cap_ttl)) {
1891 pr_info("mds%d caps renewed\n", session->s_mds);
1894 pr_info("mds%d caps still stale\n", session->s_mds);
1897 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1898 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1899 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1900 spin_unlock(&session->s_cap_lock);
1903 wake_up_session_caps(session, RENEWCAPS);
1907 * send a session close request
1909 static int request_close_session(struct ceph_mds_session *session)
1911 struct ceph_msg *msg;
1913 dout("request_close_session mds%d state %s seq %lld\n",
1914 session->s_mds, ceph_session_state_name(session->s_state),
1916 msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_CLOSE,
1920 ceph_con_send(&session->s_con, msg);
1925 * Called with s_mutex held.
1927 static int __close_session(struct ceph_mds_client *mdsc,
1928 struct ceph_mds_session *session)
1930 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1932 session->s_state = CEPH_MDS_SESSION_CLOSING;
1933 return request_close_session(session);
1936 static bool drop_negative_children(struct dentry *dentry)
1938 struct dentry *child;
1939 bool all_negative = true;
1941 if (!d_is_dir(dentry))
1944 spin_lock(&dentry->d_lock);
1945 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1946 if (d_really_is_positive(child)) {
1947 all_negative = false;
1951 spin_unlock(&dentry->d_lock);
1954 shrink_dcache_parent(dentry);
1956 return all_negative;
1960 * Trim old(er) caps.
1962 * Because we can't cache an inode without one or more caps, we do
1963 * this indirectly: if a cap is unused, we prune its aliases, at which
1964 * point the inode will hopefully get dropped to.
1966 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1967 * memory pressure from the MDS, though, so it needn't be perfect.
1969 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1971 int *remaining = arg;
1972 struct ceph_inode_info *ci = ceph_inode(inode);
1973 int used, wanted, oissued, mine;
1975 if (*remaining <= 0)
1978 spin_lock(&ci->i_ceph_lock);
1979 mine = cap->issued | cap->implemented;
1980 used = __ceph_caps_used(ci);
1981 wanted = __ceph_caps_file_wanted(ci);
1982 oissued = __ceph_caps_issued_other(ci, cap);
1984 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1985 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1986 ceph_cap_string(used), ceph_cap_string(wanted));
1987 if (cap == ci->i_auth_cap) {
1988 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1989 !list_empty(&ci->i_cap_snaps))
1991 if ((used | wanted) & CEPH_CAP_ANY_WR)
1993 /* Note: it's possible that i_filelock_ref becomes non-zero
1994 * after dropping auth caps. It doesn't hurt because reply
1995 * of lock mds request will re-add auth caps. */
1996 if (atomic_read(&ci->i_filelock_ref) > 0)
1999 /* The inode has cached pages, but it's no longer used.
2000 * we can safely drop it */
2001 if (S_ISREG(inode->i_mode) &&
2002 wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
2003 !(oissued & CEPH_CAP_FILE_CACHE)) {
2007 if ((used | wanted) & ~oissued & mine)
2008 goto out; /* we need these caps */
2011 /* we aren't the only cap.. just remove us */
2012 ceph_remove_cap(cap, true);
2015 struct dentry *dentry;
2016 /* try dropping referring dentries */
2017 spin_unlock(&ci->i_ceph_lock);
2018 dentry = d_find_any_alias(inode);
2019 if (dentry && drop_negative_children(dentry)) {
2022 d_prune_aliases(inode);
2023 count = atomic_read(&inode->i_count);
2026 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
2035 spin_unlock(&ci->i_ceph_lock);
2040 * Trim session cap count down to some max number.
2042 int ceph_trim_caps(struct ceph_mds_client *mdsc,
2043 struct ceph_mds_session *session,
2046 int trim_caps = session->s_nr_caps - max_caps;
2048 dout("trim_caps mds%d start: %d / %d, trim %d\n",
2049 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
2050 if (trim_caps > 0) {
2051 int remaining = trim_caps;
2053 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2054 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
2055 session->s_mds, session->s_nr_caps, max_caps,
2056 trim_caps - remaining);
2059 ceph_flush_cap_releases(mdsc, session);
2063 static int check_caps_flush(struct ceph_mds_client *mdsc,
2068 spin_lock(&mdsc->cap_dirty_lock);
2069 if (!list_empty(&mdsc->cap_flush_list)) {
2070 struct ceph_cap_flush *cf =
2071 list_first_entry(&mdsc->cap_flush_list,
2072 struct ceph_cap_flush, g_list);
2073 if (cf->tid <= want_flush_tid) {
2074 dout("check_caps_flush still flushing tid "
2075 "%llu <= %llu\n", cf->tid, want_flush_tid);
2079 spin_unlock(&mdsc->cap_dirty_lock);
2084 * flush all dirty inode data to disk.
2086 * returns true if we've flushed through want_flush_tid
2088 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2091 dout("check_caps_flush want %llu\n", want_flush_tid);
2093 wait_event(mdsc->cap_flushing_wq,
2094 check_caps_flush(mdsc, want_flush_tid));
2096 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2100 * called under s_mutex
2102 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2103 struct ceph_mds_session *session)
2105 struct ceph_msg *msg = NULL;
2106 struct ceph_mds_cap_release *head;
2107 struct ceph_mds_cap_item *item;
2108 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2109 struct ceph_cap *cap;
2110 LIST_HEAD(tmp_list);
2111 int num_cap_releases;
2112 __le32 barrier, *cap_barrier;
2114 down_read(&osdc->lock);
2115 barrier = cpu_to_le32(osdc->epoch_barrier);
2116 up_read(&osdc->lock);
2118 spin_lock(&session->s_cap_lock);
2120 list_splice_init(&session->s_cap_releases, &tmp_list);
2121 num_cap_releases = session->s_num_cap_releases;
2122 session->s_num_cap_releases = 0;
2123 spin_unlock(&session->s_cap_lock);
2125 while (!list_empty(&tmp_list)) {
2127 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2128 PAGE_SIZE, GFP_NOFS, false);
2131 head = msg->front.iov_base;
2132 head->num = cpu_to_le32(0);
2133 msg->front.iov_len = sizeof(*head);
2135 msg->hdr.version = cpu_to_le16(2);
2136 msg->hdr.compat_version = cpu_to_le16(1);
2139 cap = list_first_entry(&tmp_list, struct ceph_cap,
2141 list_del(&cap->session_caps);
2144 head = msg->front.iov_base;
2145 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2147 item = msg->front.iov_base + msg->front.iov_len;
2148 item->ino = cpu_to_le64(cap->cap_ino);
2149 item->cap_id = cpu_to_le64(cap->cap_id);
2150 item->migrate_seq = cpu_to_le32(cap->mseq);
2151 item->seq = cpu_to_le32(cap->issue_seq);
2152 msg->front.iov_len += sizeof(*item);
2154 ceph_put_cap(mdsc, cap);
2156 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2157 // Append cap_barrier field
2158 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2159 *cap_barrier = barrier;
2160 msg->front.iov_len += sizeof(*cap_barrier);
2162 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2163 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2164 ceph_con_send(&session->s_con, msg);
2169 BUG_ON(num_cap_releases != 0);
2171 spin_lock(&session->s_cap_lock);
2172 if (!list_empty(&session->s_cap_releases))
2174 spin_unlock(&session->s_cap_lock);
2177 // Append cap_barrier field
2178 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2179 *cap_barrier = barrier;
2180 msg->front.iov_len += sizeof(*cap_barrier);
2182 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2183 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2184 ceph_con_send(&session->s_con, msg);
2188 pr_err("send_cap_releases mds%d, failed to allocate message\n",
2190 spin_lock(&session->s_cap_lock);
2191 list_splice(&tmp_list, &session->s_cap_releases);
2192 session->s_num_cap_releases += num_cap_releases;
2193 spin_unlock(&session->s_cap_lock);
2196 static void ceph_cap_release_work(struct work_struct *work)
2198 struct ceph_mds_session *session =
2199 container_of(work, struct ceph_mds_session, s_cap_release_work);
2201 mutex_lock(&session->s_mutex);
2202 if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2203 session->s_state == CEPH_MDS_SESSION_HUNG)
2204 ceph_send_cap_releases(session->s_mdsc, session);
2205 mutex_unlock(&session->s_mutex);
2206 ceph_put_mds_session(session);
2209 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2210 struct ceph_mds_session *session)
2215 ceph_get_mds_session(session);
2216 if (queue_work(mdsc->fsc->cap_wq,
2217 &session->s_cap_release_work)) {
2218 dout("cap release work queued\n");
2220 ceph_put_mds_session(session);
2221 dout("failed to queue cap release work\n");
2226 * caller holds session->s_cap_lock
2228 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2229 struct ceph_cap *cap)
2231 list_add_tail(&cap->session_caps, &session->s_cap_releases);
2232 session->s_num_cap_releases++;
2234 if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2235 ceph_flush_cap_releases(session->s_mdsc, session);
2238 static void ceph_cap_reclaim_work(struct work_struct *work)
2240 struct ceph_mds_client *mdsc =
2241 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2242 int ret = ceph_trim_dentries(mdsc);
2244 ceph_queue_cap_reclaim_work(mdsc);
2247 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2252 if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2253 dout("caps reclaim work queued\n");
2255 dout("failed to queue caps release work\n");
2259 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2264 val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2265 if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2266 atomic_set(&mdsc->cap_reclaim_pending, 0);
2267 ceph_queue_cap_reclaim_work(mdsc);
2275 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2278 struct ceph_inode_info *ci = ceph_inode(dir);
2279 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2280 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2281 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2282 unsigned int num_entries;
2285 spin_lock(&ci->i_ceph_lock);
2286 num_entries = ci->i_files + ci->i_subdirs;
2287 spin_unlock(&ci->i_ceph_lock);
2288 num_entries = max(num_entries, 1U);
2289 num_entries = min(num_entries, opt->max_readdir);
2291 order = get_order(size * num_entries);
2292 while (order >= 0) {
2293 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2296 if (rinfo->dir_entries)
2300 if (!rinfo->dir_entries)
2303 num_entries = (PAGE_SIZE << order) / size;
2304 num_entries = min(num_entries, opt->max_readdir);
2306 rinfo->dir_buf_size = PAGE_SIZE << order;
2307 req->r_num_caps = num_entries + 1;
2308 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2309 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2314 * Create an mds request.
2316 struct ceph_mds_request *
2317 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2319 struct ceph_mds_request *req;
2321 req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2323 return ERR_PTR(-ENOMEM);
2325 mutex_init(&req->r_fill_mutex);
2327 req->r_started = jiffies;
2328 req->r_start_latency = ktime_get();
2329 req->r_resend_mds = -1;
2330 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2331 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2333 kref_init(&req->r_kref);
2334 RB_CLEAR_NODE(&req->r_node);
2335 INIT_LIST_HEAD(&req->r_wait);
2336 init_completion(&req->r_completion);
2337 init_completion(&req->r_safe_completion);
2338 INIT_LIST_HEAD(&req->r_unsafe_item);
2340 ktime_get_coarse_real_ts64(&req->r_stamp);
2343 req->r_direct_mode = mode;
2348 * return oldest (lowest) request, tid in request tree, 0 if none.
2350 * called under mdsc->mutex.
2352 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2354 if (RB_EMPTY_ROOT(&mdsc->request_tree))
2356 return rb_entry(rb_first(&mdsc->request_tree),
2357 struct ceph_mds_request, r_node);
2360 static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2362 return mdsc->oldest_tid;
2366 * Build a dentry's path. Allocate on heap; caller must kfree. Based
2367 * on build_path_from_dentry in fs/cifs/dir.c.
2369 * If @stop_on_nosnap, generate path relative to the first non-snapped
2372 * Encode hidden .snap dirs as a double /, i.e.
2373 * foo/.snap/bar -> foo//bar
2375 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2378 struct dentry *temp;
2385 return ERR_PTR(-EINVAL);
2389 return ERR_PTR(-ENOMEM);
2394 seq = read_seqbegin(&rename_lock);
2398 struct inode *inode;
2400 spin_lock(&temp->d_lock);
2401 inode = d_inode(temp);
2402 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2403 dout("build_path path+%d: %p SNAPDIR\n",
2405 } else if (stop_on_nosnap && inode && dentry != temp &&
2406 ceph_snap(inode) == CEPH_NOSNAP) {
2407 spin_unlock(&temp->d_lock);
2408 pos++; /* get rid of any prepended '/' */
2411 pos -= temp->d_name.len;
2413 spin_unlock(&temp->d_lock);
2416 memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2418 spin_unlock(&temp->d_lock);
2419 temp = READ_ONCE(temp->d_parent);
2421 /* Are we at the root? */
2425 /* Are we out of buffer? */
2431 base = ceph_ino(d_inode(temp));
2434 if (read_seqretry(&rename_lock, seq))
2439 * A rename didn't occur, but somehow we didn't end up where
2440 * we thought we would. Throw a warning and try again.
2442 pr_warn("build_path did not end path lookup where "
2443 "expected, pos is %d\n", pos);
2448 *plen = PATH_MAX - 1 - pos;
2449 dout("build_path on %p %d built %llx '%.*s'\n",
2450 dentry, d_count(dentry), base, *plen, path + pos);
2454 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2455 const char **ppath, int *ppathlen, u64 *pino,
2456 bool *pfreepath, bool parent_locked)
2462 dir = d_inode_rcu(dentry->d_parent);
2463 if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2464 *pino = ceph_ino(dir);
2466 *ppath = dentry->d_name.name;
2467 *ppathlen = dentry->d_name.len;
2471 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2473 return PTR_ERR(path);
2479 static int build_inode_path(struct inode *inode,
2480 const char **ppath, int *ppathlen, u64 *pino,
2483 struct dentry *dentry;
2486 if (ceph_snap(inode) == CEPH_NOSNAP) {
2487 *pino = ceph_ino(inode);
2491 dentry = d_find_alias(inode);
2492 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2495 return PTR_ERR(path);
2502 * request arguments may be specified via an inode *, a dentry *, or
2503 * an explicit ino+path.
2505 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2506 struct inode *rdiri, const char *rpath,
2507 u64 rino, const char **ppath, int *pathlen,
2508 u64 *ino, bool *freepath, bool parent_locked)
2513 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2514 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2516 } else if (rdentry) {
2517 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2518 freepath, parent_locked);
2519 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2521 } else if (rpath || rino) {
2524 *pathlen = rpath ? strlen(rpath) : 0;
2525 dout(" path %.*s\n", *pathlen, rpath);
2531 static void encode_timestamp_and_gids(void **p,
2532 const struct ceph_mds_request *req)
2534 struct ceph_timespec ts;
2537 ceph_encode_timespec64(&ts, &req->r_stamp);
2538 ceph_encode_copy(p, &ts, sizeof(ts));
2541 ceph_encode_32(p, req->r_cred->group_info->ngroups);
2542 for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2543 ceph_encode_64(p, from_kgid(&init_user_ns,
2544 req->r_cred->group_info->gid[i]));
2548 * called under mdsc->mutex
2550 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2551 struct ceph_mds_request *req,
2552 bool drop_cap_releases)
2554 int mds = session->s_mds;
2555 struct ceph_mds_client *mdsc = session->s_mdsc;
2556 struct ceph_msg *msg;
2557 struct ceph_mds_request_head_old *head;
2558 const char *path1 = NULL;
2559 const char *path2 = NULL;
2560 u64 ino1 = 0, ino2 = 0;
2561 int pathlen1 = 0, pathlen2 = 0;
2562 bool freepath1 = false, freepath2 = false;
2567 bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
2569 ret = set_request_path_attr(req->r_inode, req->r_dentry,
2570 req->r_parent, req->r_path1, req->r_ino1.ino,
2571 &path1, &pathlen1, &ino1, &freepath1,
2572 test_bit(CEPH_MDS_R_PARENT_LOCKED,
2573 &req->r_req_flags));
2579 /* If r_old_dentry is set, then assume that its parent is locked */
2580 ret = set_request_path_attr(NULL, req->r_old_dentry,
2581 req->r_old_dentry_dir,
2582 req->r_path2, req->r_ino2.ino,
2583 &path2, &pathlen2, &ino2, &freepath2, true);
2589 len = legacy ? sizeof(*head) : sizeof(struct ceph_mds_request_head);
2590 len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2591 sizeof(struct ceph_timespec);
2592 len += sizeof(u32) + (sizeof(u64) * req->r_cred->group_info->ngroups);
2594 /* calculate (max) length for cap releases */
2595 len += sizeof(struct ceph_mds_request_release) *
2596 (!!req->r_inode_drop + !!req->r_dentry_drop +
2597 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2599 if (req->r_dentry_drop)
2601 if (req->r_old_dentry_drop)
2604 msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2606 msg = ERR_PTR(-ENOMEM);
2610 msg->hdr.tid = cpu_to_le64(req->r_tid);
2613 * The old ceph_mds_request_head didn't contain a version field, and
2614 * one was added when we moved the message version from 3->4.
2617 msg->hdr.version = cpu_to_le16(3);
2618 head = msg->front.iov_base;
2619 p = msg->front.iov_base + sizeof(*head);
2621 struct ceph_mds_request_head *new_head = msg->front.iov_base;
2623 msg->hdr.version = cpu_to_le16(4);
2624 new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
2625 head = (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2626 p = msg->front.iov_base + sizeof(*new_head);
2629 end = msg->front.iov_base + msg->front.iov_len;
2631 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2632 head->op = cpu_to_le32(req->r_op);
2633 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
2634 req->r_cred->fsuid));
2635 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
2636 req->r_cred->fsgid));
2637 head->ino = cpu_to_le64(req->r_deleg_ino);
2638 head->args = req->r_args;
2640 ceph_encode_filepath(&p, end, ino1, path1);
2641 ceph_encode_filepath(&p, end, ino2, path2);
2643 /* make note of release offset, in case we need to replay */
2644 req->r_request_release_offset = p - msg->front.iov_base;
2648 if (req->r_inode_drop)
2649 releases += ceph_encode_inode_release(&p,
2650 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2651 mds, req->r_inode_drop, req->r_inode_unless,
2652 req->r_op == CEPH_MDS_OP_READDIR);
2653 if (req->r_dentry_drop)
2654 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2655 req->r_parent, mds, req->r_dentry_drop,
2656 req->r_dentry_unless);
2657 if (req->r_old_dentry_drop)
2658 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2659 req->r_old_dentry_dir, mds,
2660 req->r_old_dentry_drop,
2661 req->r_old_dentry_unless);
2662 if (req->r_old_inode_drop)
2663 releases += ceph_encode_inode_release(&p,
2664 d_inode(req->r_old_dentry),
2665 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2667 if (drop_cap_releases) {
2669 p = msg->front.iov_base + req->r_request_release_offset;
2672 head->num_releases = cpu_to_le16(releases);
2674 encode_timestamp_and_gids(&p, req);
2676 if (WARN_ON_ONCE(p > end)) {
2678 msg = ERR_PTR(-ERANGE);
2682 msg->front.iov_len = p - msg->front.iov_base;
2683 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2685 if (req->r_pagelist) {
2686 struct ceph_pagelist *pagelist = req->r_pagelist;
2687 ceph_msg_data_add_pagelist(msg, pagelist);
2688 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2690 msg->hdr.data_len = 0;
2693 msg->hdr.data_off = cpu_to_le16(0);
2697 ceph_mdsc_free_path((char *)path2, pathlen2);
2700 ceph_mdsc_free_path((char *)path1, pathlen1);
2706 * called under mdsc->mutex if error, under no mutex if
2709 static void complete_request(struct ceph_mds_client *mdsc,
2710 struct ceph_mds_request *req)
2712 req->r_end_latency = ktime_get();
2714 if (req->r_callback)
2715 req->r_callback(mdsc, req);
2716 complete_all(&req->r_completion);
2719 static struct ceph_mds_request_head_old *
2720 find_old_request_head(void *p, u64 features)
2722 bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2723 struct ceph_mds_request_head *new_head;
2726 return (struct ceph_mds_request_head_old *)p;
2727 new_head = (struct ceph_mds_request_head *)p;
2728 return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2732 * called under mdsc->mutex
2734 static int __prepare_send_request(struct ceph_mds_session *session,
2735 struct ceph_mds_request *req,
2736 bool drop_cap_releases)
2738 int mds = session->s_mds;
2739 struct ceph_mds_client *mdsc = session->s_mdsc;
2740 struct ceph_mds_request_head_old *rhead;
2741 struct ceph_msg *msg;
2746 struct ceph_cap *cap =
2747 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2750 req->r_sent_on_mseq = cap->mseq;
2752 req->r_sent_on_mseq = -1;
2754 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2755 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2757 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2761 * Replay. Do not regenerate message (and rebuild
2762 * paths, etc.); just use the original message.
2763 * Rebuilding paths will break for renames because
2764 * d_move mangles the src name.
2766 msg = req->r_request;
2767 rhead = find_old_request_head(msg->front.iov_base,
2768 session->s_con.peer_features);
2770 flags = le32_to_cpu(rhead->flags);
2771 flags |= CEPH_MDS_FLAG_REPLAY;
2772 rhead->flags = cpu_to_le32(flags);
2774 if (req->r_target_inode)
2775 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2777 rhead->num_retry = req->r_attempts - 1;
2779 /* remove cap/dentry releases from message */
2780 rhead->num_releases = 0;
2782 p = msg->front.iov_base + req->r_request_release_offset;
2783 encode_timestamp_and_gids(&p, req);
2785 msg->front.iov_len = p - msg->front.iov_base;
2786 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2790 if (req->r_request) {
2791 ceph_msg_put(req->r_request);
2792 req->r_request = NULL;
2794 msg = create_request_message(session, req, drop_cap_releases);
2796 req->r_err = PTR_ERR(msg);
2797 return PTR_ERR(msg);
2799 req->r_request = msg;
2801 rhead = find_old_request_head(msg->front.iov_base,
2802 session->s_con.peer_features);
2803 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2804 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2805 flags |= CEPH_MDS_FLAG_REPLAY;
2806 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2807 flags |= CEPH_MDS_FLAG_ASYNC;
2809 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2810 rhead->flags = cpu_to_le32(flags);
2811 rhead->num_fwd = req->r_num_fwd;
2812 rhead->num_retry = req->r_attempts - 1;
2814 dout(" r_parent = %p\n", req->r_parent);
2819 * called under mdsc->mutex
2821 static int __send_request(struct ceph_mds_session *session,
2822 struct ceph_mds_request *req,
2823 bool drop_cap_releases)
2827 err = __prepare_send_request(session, req, drop_cap_releases);
2829 ceph_msg_get(req->r_request);
2830 ceph_con_send(&session->s_con, req->r_request);
2837 * send request, or put it on the appropriate wait list.
2839 static void __do_request(struct ceph_mds_client *mdsc,
2840 struct ceph_mds_request *req)
2842 struct ceph_mds_session *session = NULL;
2847 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2848 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2849 __unregister_request(mdsc, req);
2853 if (req->r_timeout &&
2854 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2855 dout("do_request timed out\n");
2859 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2860 dout("do_request forced umount\n");
2864 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2865 if (mdsc->mdsmap_err) {
2866 err = mdsc->mdsmap_err;
2867 dout("do_request mdsmap err %d\n", err);
2870 if (mdsc->mdsmap->m_epoch == 0) {
2871 dout("do_request no mdsmap, waiting for map\n");
2872 list_add(&req->r_wait, &mdsc->waiting_for_map);
2875 if (!(mdsc->fsc->mount_options->flags &
2876 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2877 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2878 err = -EHOSTUNREACH;
2883 put_request_session(req);
2885 mds = __choose_mds(mdsc, req, &random);
2887 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2888 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2892 dout("do_request no mds or not active, waiting for map\n");
2893 list_add(&req->r_wait, &mdsc->waiting_for_map);
2897 /* get, open session */
2898 session = __ceph_lookup_mds_session(mdsc, mds);
2900 session = register_session(mdsc, mds);
2901 if (IS_ERR(session)) {
2902 err = PTR_ERR(session);
2906 req->r_session = ceph_get_mds_session(session);
2908 dout("do_request mds%d session %p state %s\n", mds, session,
2909 ceph_session_state_name(session->s_state));
2910 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2911 session->s_state != CEPH_MDS_SESSION_HUNG) {
2913 * We cannot queue async requests since the caps and delegated
2914 * inodes are bound to the session. Just return -EJUKEBOX and
2915 * let the caller retry a sync request in that case.
2917 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2923 * If the session has been REJECTED, then return a hard error,
2924 * unless it's a CLEANRECOVER mount, in which case we'll queue
2925 * it to the mdsc queue.
2927 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2928 if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
2929 list_add(&req->r_wait, &mdsc->waiting_for_map);
2935 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2936 session->s_state == CEPH_MDS_SESSION_CLOSING) {
2937 err = __open_session(mdsc, session);
2940 /* retry the same mds later */
2942 req->r_resend_mds = mds;
2944 list_add(&req->r_wait, &session->s_waiting);
2949 req->r_resend_mds = -1; /* forget any previous mds hint */
2951 if (req->r_request_started == 0) /* note request start time */
2952 req->r_request_started = jiffies;
2954 err = __send_request(session, req, false);
2957 ceph_put_mds_session(session);
2960 dout("__do_request early error %d\n", err);
2962 complete_request(mdsc, req);
2963 __unregister_request(mdsc, req);
2969 * called under mdsc->mutex
2971 static void __wake_requests(struct ceph_mds_client *mdsc,
2972 struct list_head *head)
2974 struct ceph_mds_request *req;
2975 LIST_HEAD(tmp_list);
2977 list_splice_init(head, &tmp_list);
2979 while (!list_empty(&tmp_list)) {
2980 req = list_entry(tmp_list.next,
2981 struct ceph_mds_request, r_wait);
2982 list_del_init(&req->r_wait);
2983 dout(" wake request %p tid %llu\n", req, req->r_tid);
2984 __do_request(mdsc, req);
2989 * Wake up threads with requests pending for @mds, so that they can
2990 * resubmit their requests to a possibly different mds.
2992 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2994 struct ceph_mds_request *req;
2995 struct rb_node *p = rb_first(&mdsc->request_tree);
2997 dout("kick_requests mds%d\n", mds);
2999 req = rb_entry(p, struct ceph_mds_request, r_node);
3001 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3003 if (req->r_attempts > 0)
3004 continue; /* only new requests */
3005 if (req->r_session &&
3006 req->r_session->s_mds == mds) {
3007 dout(" kicking tid %llu\n", req->r_tid);
3008 list_del_init(&req->r_wait);
3009 __do_request(mdsc, req);
3014 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
3015 struct ceph_mds_request *req)
3019 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
3021 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3022 if (req->r_parent) {
3023 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
3024 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
3025 CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
3026 spin_lock(&ci->i_ceph_lock);
3027 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
3028 __ceph_touch_fmode(ci, mdsc, fmode);
3029 spin_unlock(&ci->i_ceph_lock);
3031 if (req->r_old_dentry_dir)
3032 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
3036 err = ceph_wait_on_async_create(req->r_inode);
3038 dout("%s: wait for async create returned: %d\n",
3044 if (!err && req->r_old_inode) {
3045 err = ceph_wait_on_async_create(req->r_old_inode);
3047 dout("%s: wait for async create returned: %d\n",
3053 dout("submit_request on %p for inode %p\n", req, dir);
3054 mutex_lock(&mdsc->mutex);
3055 __register_request(mdsc, req, dir);
3056 __do_request(mdsc, req);
3058 mutex_unlock(&mdsc->mutex);
3062 static int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3063 struct ceph_mds_request *req)
3068 dout("do_request waiting\n");
3069 if (!req->r_timeout && req->r_wait_for_completion) {
3070 err = req->r_wait_for_completion(mdsc, req);
3072 long timeleft = wait_for_completion_killable_timeout(
3074 ceph_timeout_jiffies(req->r_timeout));
3078 err = -ETIMEDOUT; /* timed out */
3080 err = timeleft; /* killed */
3082 dout("do_request waited, got %d\n", err);
3083 mutex_lock(&mdsc->mutex);
3085 /* only abort if we didn't race with a real reply */
3086 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3087 err = le32_to_cpu(req->r_reply_info.head->result);
3088 } else if (err < 0) {
3089 dout("aborted request %lld with %d\n", req->r_tid, err);
3092 * ensure we aren't running concurrently with
3093 * ceph_fill_trace or ceph_readdir_prepopulate, which
3094 * rely on locks (dir mutex) held by our caller.
3096 mutex_lock(&req->r_fill_mutex);
3098 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3099 mutex_unlock(&req->r_fill_mutex);
3101 if (req->r_parent &&
3102 (req->r_op & CEPH_MDS_OP_WRITE))
3103 ceph_invalidate_dir_request(req);
3108 mutex_unlock(&mdsc->mutex);
3113 * Synchrously perform an mds request. Take care of all of the
3114 * session setup, forwarding, retry details.
3116 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3118 struct ceph_mds_request *req)
3122 dout("do_request on %p\n", req);
3125 err = ceph_mdsc_submit_request(mdsc, dir, req);
3127 err = ceph_mdsc_wait_request(mdsc, req);
3128 dout("do_request %p done, result %d\n", req, err);
3133 * Invalidate dir's completeness, dentry lease state on an aborted MDS
3134 * namespace request.
3136 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3138 struct inode *dir = req->r_parent;
3139 struct inode *old_dir = req->r_old_dentry_dir;
3141 dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
3143 ceph_dir_clear_complete(dir);
3145 ceph_dir_clear_complete(old_dir);
3147 ceph_invalidate_dentry_lease(req->r_dentry);
3148 if (req->r_old_dentry)
3149 ceph_invalidate_dentry_lease(req->r_old_dentry);
3155 * We take the session mutex and parse and process the reply immediately.
3156 * This preserves the logical ordering of replies, capabilities, etc., sent
3157 * by the MDS as they are applied to our local cache.
3159 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3161 struct ceph_mds_client *mdsc = session->s_mdsc;
3162 struct ceph_mds_request *req;
3163 struct ceph_mds_reply_head *head = msg->front.iov_base;
3164 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
3165 struct ceph_snap_realm *realm;
3168 int mds = session->s_mds;
3170 if (msg->front.iov_len < sizeof(*head)) {
3171 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
3176 /* get request, session */
3177 tid = le64_to_cpu(msg->hdr.tid);
3178 mutex_lock(&mdsc->mutex);
3179 req = lookup_get_request(mdsc, tid);
3181 dout("handle_reply on unknown tid %llu\n", tid);
3182 mutex_unlock(&mdsc->mutex);
3185 dout("handle_reply %p\n", req);
3187 /* correct session? */
3188 if (req->r_session != session) {
3189 pr_err("mdsc_handle_reply got %llu on session mds%d"
3190 " not mds%d\n", tid, session->s_mds,
3191 req->r_session ? req->r_session->s_mds : -1);
3192 mutex_unlock(&mdsc->mutex);
3197 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3198 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3199 pr_warn("got a dup %s reply on %llu from mds%d\n",
3200 head->safe ? "safe" : "unsafe", tid, mds);
3201 mutex_unlock(&mdsc->mutex);
3204 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3205 pr_warn("got unsafe after safe on %llu from mds%d\n",
3207 mutex_unlock(&mdsc->mutex);
3211 result = le32_to_cpu(head->result);
3215 * if we're not talking to the authority, send to them
3216 * if the authority has changed while we weren't looking,
3217 * send to new authority
3218 * Otherwise we just have to return an ESTALE
3220 if (result == -ESTALE) {
3221 dout("got ESTALE on request %llu\n", req->r_tid);
3222 req->r_resend_mds = -1;
3223 if (req->r_direct_mode != USE_AUTH_MDS) {
3224 dout("not using auth, setting for that now\n");
3225 req->r_direct_mode = USE_AUTH_MDS;
3226 __do_request(mdsc, req);
3227 mutex_unlock(&mdsc->mutex);
3230 int mds = __choose_mds(mdsc, req, NULL);
3231 if (mds >= 0 && mds != req->r_session->s_mds) {
3232 dout("but auth changed, so resending\n");
3233 __do_request(mdsc, req);
3234 mutex_unlock(&mdsc->mutex);
3238 dout("have to return ESTALE on request %llu\n", req->r_tid);
3243 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3244 __unregister_request(mdsc, req);
3246 /* last request during umount? */
3247 if (mdsc->stopping && !__get_oldest_req(mdsc))
3248 complete_all(&mdsc->safe_umount_waiters);
3250 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3252 * We already handled the unsafe response, now do the
3253 * cleanup. No need to examine the response; the MDS
3254 * doesn't include any result info in the safe
3255 * response. And even if it did, there is nothing
3256 * useful we could do with a revised return value.
3258 dout("got safe reply %llu, mds%d\n", tid, mds);
3260 mutex_unlock(&mdsc->mutex);
3264 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3265 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3268 dout("handle_reply tid %lld result %d\n", tid, result);
3269 rinfo = &req->r_reply_info;
3270 if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3271 err = parse_reply_info(session, msg, rinfo, (u64)-1);
3273 err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3274 mutex_unlock(&mdsc->mutex);
3276 /* Must find target inode outside of mutexes to avoid deadlocks */
3277 if ((err >= 0) && rinfo->head->is_target) {
3279 struct ceph_vino tvino = {
3280 .ino = le64_to_cpu(rinfo->targeti.in->ino),
3281 .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3284 in = ceph_get_inode(mdsc->fsc->sb, tvino);
3287 mutex_lock(&session->s_mutex);
3290 req->r_target_inode = in;
3293 mutex_lock(&session->s_mutex);
3295 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3302 if (rinfo->snapblob_len) {
3303 down_write(&mdsc->snap_rwsem);
3304 ceph_update_snap_trace(mdsc, rinfo->snapblob,
3305 rinfo->snapblob + rinfo->snapblob_len,
3306 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3308 downgrade_write(&mdsc->snap_rwsem);
3310 down_read(&mdsc->snap_rwsem);
3313 /* insert trace into our cache */
3314 mutex_lock(&req->r_fill_mutex);
3315 current->journal_info = req;
3316 err = ceph_fill_trace(mdsc->fsc->sb, req);
3318 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3319 req->r_op == CEPH_MDS_OP_LSSNAP))
3320 ceph_readdir_prepopulate(req, req->r_session);
3322 current->journal_info = NULL;
3323 mutex_unlock(&req->r_fill_mutex);
3325 up_read(&mdsc->snap_rwsem);
3327 ceph_put_snap_realm(mdsc, realm);
3330 if (req->r_target_inode &&
3331 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3332 struct ceph_inode_info *ci =
3333 ceph_inode(req->r_target_inode);
3334 spin_lock(&ci->i_unsafe_lock);
3335 list_add_tail(&req->r_unsafe_target_item,
3336 &ci->i_unsafe_iops);
3337 spin_unlock(&ci->i_unsafe_lock);
3340 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3343 mutex_lock(&mdsc->mutex);
3344 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3348 req->r_reply = ceph_msg_get(msg);
3349 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3352 dout("reply arrived after request %lld was aborted\n", tid);
3354 mutex_unlock(&mdsc->mutex);
3356 mutex_unlock(&session->s_mutex);
3358 /* kick calling process */
3359 complete_request(mdsc, req);
3361 ceph_update_metadata_metrics(&mdsc->metric, req->r_start_latency,
3362 req->r_end_latency, err);
3364 ceph_mdsc_put_request(req);
3371 * handle mds notification that our request has been forwarded.
3373 static void handle_forward(struct ceph_mds_client *mdsc,
3374 struct ceph_mds_session *session,
3375 struct ceph_msg *msg)
3377 struct ceph_mds_request *req;
3378 u64 tid = le64_to_cpu(msg->hdr.tid);
3382 void *p = msg->front.iov_base;
3383 void *end = p + msg->front.iov_len;
3385 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3386 next_mds = ceph_decode_32(&p);
3387 fwd_seq = ceph_decode_32(&p);
3389 mutex_lock(&mdsc->mutex);
3390 req = lookup_get_request(mdsc, tid);
3392 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3393 goto out; /* dup reply? */
3396 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3397 dout("forward tid %llu aborted, unregistering\n", tid);
3398 __unregister_request(mdsc, req);
3399 } else if (fwd_seq <= req->r_num_fwd) {
3400 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3401 tid, next_mds, req->r_num_fwd, fwd_seq);
3403 /* resend. forward race not possible; mds would drop */
3404 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3406 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3407 req->r_attempts = 0;
3408 req->r_num_fwd = fwd_seq;
3409 req->r_resend_mds = next_mds;
3410 put_request_session(req);
3411 __do_request(mdsc, req);
3413 ceph_mdsc_put_request(req);
3415 mutex_unlock(&mdsc->mutex);
3419 pr_err("mdsc_handle_forward decode error err=%d\n", err);
3422 static int __decode_session_metadata(void **p, void *end,
3425 /* map<string,string> */
3428 ceph_decode_32_safe(p, end, n, bad);
3431 ceph_decode_32_safe(p, end, len, bad);
3432 ceph_decode_need(p, end, len, bad);
3433 err_str = !strncmp(*p, "error_string", len);
3435 ceph_decode_32_safe(p, end, len, bad);
3436 ceph_decode_need(p, end, len, bad);
3438 * Match "blocklisted (blacklisted)" from newer MDSes,
3439 * or "blacklisted" from older MDSes.
3441 if (err_str && strnstr(*p, "blacklisted", len))
3442 *blocklisted = true;
3451 * handle a mds session control message
3453 static void handle_session(struct ceph_mds_session *session,
3454 struct ceph_msg *msg)
3456 struct ceph_mds_client *mdsc = session->s_mdsc;
3457 int mds = session->s_mds;
3458 int msg_version = le16_to_cpu(msg->hdr.version);
3459 void *p = msg->front.iov_base;
3460 void *end = p + msg->front.iov_len;
3461 struct ceph_mds_session_head *h;
3463 u64 seq, features = 0;
3465 bool blocklisted = false;
3468 ceph_decode_need(&p, end, sizeof(*h), bad);
3472 op = le32_to_cpu(h->op);
3473 seq = le64_to_cpu(h->seq);
3475 if (msg_version >= 3) {
3477 /* version >= 2, metadata */
3478 if (__decode_session_metadata(&p, end, &blocklisted) < 0)
3480 /* version >= 3, feature bits */
3481 ceph_decode_32_safe(&p, end, len, bad);
3483 ceph_decode_64_safe(&p, end, features, bad);
3484 p += len - sizeof(features);
3488 mutex_lock(&mdsc->mutex);
3489 if (op == CEPH_SESSION_CLOSE) {
3490 ceph_get_mds_session(session);
3491 __unregister_session(mdsc, session);
3493 /* FIXME: this ttl calculation is generous */
3494 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3495 mutex_unlock(&mdsc->mutex);
3497 mutex_lock(&session->s_mutex);
3499 dout("handle_session mds%d %s %p state %s seq %llu\n",
3500 mds, ceph_session_op_name(op), session,
3501 ceph_session_state_name(session->s_state), seq);
3503 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3504 session->s_state = CEPH_MDS_SESSION_OPEN;
3505 pr_info("mds%d came back\n", session->s_mds);
3509 case CEPH_SESSION_OPEN:
3510 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3511 pr_info("mds%d reconnect success\n", session->s_mds);
3512 session->s_state = CEPH_MDS_SESSION_OPEN;
3513 session->s_features = features;
3514 renewed_caps(mdsc, session, 0);
3515 if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT, &session->s_features))
3516 metric_schedule_delayed(&mdsc->metric);
3519 __close_session(mdsc, session);
3522 case CEPH_SESSION_RENEWCAPS:
3523 if (session->s_renew_seq == seq)
3524 renewed_caps(mdsc, session, 1);
3527 case CEPH_SESSION_CLOSE:
3528 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3529 pr_info("mds%d reconnect denied\n", session->s_mds);
3530 session->s_state = CEPH_MDS_SESSION_CLOSED;
3531 cleanup_session_requests(mdsc, session);
3532 remove_session_caps(session);
3533 wake = 2; /* for good measure */
3534 wake_up_all(&mdsc->session_close_wq);
3537 case CEPH_SESSION_STALE:
3538 pr_info("mds%d caps went stale, renewing\n",
3540 atomic_inc(&session->s_cap_gen);
3541 session->s_cap_ttl = jiffies - 1;
3542 send_renew_caps(mdsc, session);
3545 case CEPH_SESSION_RECALL_STATE:
3546 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3549 case CEPH_SESSION_FLUSHMSG:
3550 send_flushmsg_ack(mdsc, session, seq);
3553 case CEPH_SESSION_FORCE_RO:
3554 dout("force_session_readonly %p\n", session);
3555 spin_lock(&session->s_cap_lock);
3556 session->s_readonly = true;
3557 spin_unlock(&session->s_cap_lock);
3558 wake_up_session_caps(session, FORCE_RO);
3561 case CEPH_SESSION_REJECT:
3562 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3563 pr_info("mds%d rejected session\n", session->s_mds);
3564 session->s_state = CEPH_MDS_SESSION_REJECTED;
3565 cleanup_session_requests(mdsc, session);
3566 remove_session_caps(session);
3568 mdsc->fsc->blocklisted = true;
3569 wake = 2; /* for good measure */
3573 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3577 mutex_unlock(&session->s_mutex);
3579 mutex_lock(&mdsc->mutex);
3580 __wake_requests(mdsc, &session->s_waiting);
3582 kick_requests(mdsc, mds);
3583 mutex_unlock(&mdsc->mutex);
3585 if (op == CEPH_SESSION_CLOSE)
3586 ceph_put_mds_session(session);
3590 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3591 (int)msg->front.iov_len);
3596 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3600 dcaps = xchg(&req->r_dir_caps, 0);
3602 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3603 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3607 void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
3611 dcaps = xchg(&req->r_dir_caps, 0);
3613 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3614 ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
3620 * called under session->mutex.
3622 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3623 struct ceph_mds_session *session)
3625 struct ceph_mds_request *req, *nreq;
3628 dout("replay_unsafe_requests mds%d\n", session->s_mds);
3630 mutex_lock(&mdsc->mutex);
3631 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3632 __send_request(session, req, true);
3635 * also re-send old requests when MDS enters reconnect stage. So that MDS
3636 * can process completed request in clientreplay stage.
3638 p = rb_first(&mdsc->request_tree);
3640 req = rb_entry(p, struct ceph_mds_request, r_node);
3642 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3644 if (req->r_attempts == 0)
3645 continue; /* only old requests */
3646 if (!req->r_session)
3648 if (req->r_session->s_mds != session->s_mds)
3651 ceph_mdsc_release_dir_caps_no_check(req);
3653 __send_request(session, req, true);
3655 mutex_unlock(&mdsc->mutex);
3658 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3660 struct ceph_msg *reply;
3661 struct ceph_pagelist *_pagelist;
3666 if (!recon_state->allow_multi)
3669 /* can't handle message that contains both caps and realm */
3670 BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3672 /* pre-allocate new pagelist */
3673 _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3677 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3681 /* placeholder for nr_caps */
3682 err = ceph_pagelist_encode_32(_pagelist, 0);
3686 if (recon_state->nr_caps) {
3687 /* currently encoding caps */
3688 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3692 /* placeholder for nr_realms (currently encoding relams) */
3693 err = ceph_pagelist_encode_32(_pagelist, 0);
3698 err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3702 page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3703 addr = kmap_atomic(page);
3704 if (recon_state->nr_caps) {
3705 /* currently encoding caps */
3706 *addr = cpu_to_le32(recon_state->nr_caps);
3708 /* currently encoding relams */
3709 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3711 kunmap_atomic(addr);
3713 reply->hdr.version = cpu_to_le16(5);
3714 reply->hdr.compat_version = cpu_to_le16(4);
3716 reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3717 ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3719 ceph_con_send(&recon_state->session->s_con, reply);
3720 ceph_pagelist_release(recon_state->pagelist);
3722 recon_state->pagelist = _pagelist;
3723 recon_state->nr_caps = 0;
3724 recon_state->nr_realms = 0;
3725 recon_state->msg_version = 5;
3728 ceph_msg_put(reply);
3730 ceph_pagelist_release(_pagelist);
3734 static struct dentry* d_find_primary(struct inode *inode)
3736 struct dentry *alias, *dn = NULL;
3738 if (hlist_empty(&inode->i_dentry))
3741 spin_lock(&inode->i_lock);
3742 if (hlist_empty(&inode->i_dentry))
3745 if (S_ISDIR(inode->i_mode)) {
3746 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
3747 if (!IS_ROOT(alias))
3752 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
3753 spin_lock(&alias->d_lock);
3754 if (!d_unhashed(alias) &&
3755 (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
3756 dn = dget_dlock(alias);
3758 spin_unlock(&alias->d_lock);
3763 spin_unlock(&inode->i_lock);
3768 * Encode information about a cap for a reconnect with the MDS.
3770 static int reconnect_caps_cb(struct inode *inode, struct ceph_cap *cap,
3774 struct ceph_mds_cap_reconnect v2;
3775 struct ceph_mds_cap_reconnect_v1 v1;
3777 struct ceph_inode_info *ci = cap->ci;
3778 struct ceph_reconnect_state *recon_state = arg;
3779 struct ceph_pagelist *pagelist = recon_state->pagelist;
3780 struct dentry *dentry;
3786 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3787 inode, ceph_vinop(inode), cap, cap->cap_id,
3788 ceph_cap_string(cap->issued));
3790 dentry = d_find_primary(inode);
3792 /* set pathbase to parent dir when msg_version >= 2 */
3793 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase,
3794 recon_state->msg_version >= 2);
3797 err = PTR_ERR(path);
3806 spin_lock(&ci->i_ceph_lock);
3807 cap->seq = 0; /* reset cap seq */
3808 cap->issue_seq = 0; /* and issue_seq */
3809 cap->mseq = 0; /* and migrate_seq */
3810 cap->cap_gen = atomic_read(&cap->session->s_cap_gen);
3812 /* These are lost when the session goes away */
3813 if (S_ISDIR(inode->i_mode)) {
3814 if (cap->issued & CEPH_CAP_DIR_CREATE) {
3815 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3816 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3818 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3821 if (recon_state->msg_version >= 2) {
3822 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3823 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3824 rec.v2.issued = cpu_to_le32(cap->issued);
3825 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3826 rec.v2.pathbase = cpu_to_le64(pathbase);
3827 rec.v2.flock_len = (__force __le32)
3828 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3830 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3831 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3832 rec.v1.issued = cpu_to_le32(cap->issued);
3833 rec.v1.size = cpu_to_le64(i_size_read(inode));
3834 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3835 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3836 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3837 rec.v1.pathbase = cpu_to_le64(pathbase);
3840 if (list_empty(&ci->i_cap_snaps)) {
3841 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3843 struct ceph_cap_snap *capsnap =
3844 list_first_entry(&ci->i_cap_snaps,
3845 struct ceph_cap_snap, ci_item);
3846 snap_follows = capsnap->follows;
3848 spin_unlock(&ci->i_ceph_lock);
3850 if (recon_state->msg_version >= 2) {
3851 int num_fcntl_locks, num_flock_locks;
3852 struct ceph_filelock *flocks = NULL;
3853 size_t struct_len, total_len = sizeof(u64);
3857 if (rec.v2.flock_len) {
3858 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3860 num_fcntl_locks = 0;
3861 num_flock_locks = 0;
3863 if (num_fcntl_locks + num_flock_locks > 0) {
3864 flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3865 sizeof(struct ceph_filelock),
3871 err = ceph_encode_locks_to_buffer(inode, flocks,
3886 if (recon_state->msg_version >= 3) {
3887 /* version, compat_version and struct_len */
3888 total_len += 2 * sizeof(u8) + sizeof(u32);
3892 * number of encoded locks is stable, so copy to pagelist
3894 struct_len = 2 * sizeof(u32) +
3895 (num_fcntl_locks + num_flock_locks) *
3896 sizeof(struct ceph_filelock);
3897 rec.v2.flock_len = cpu_to_le32(struct_len);
3899 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
3902 struct_len += sizeof(u64); /* snap_follows */
3904 total_len += struct_len;
3906 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
3907 err = send_reconnect_partial(recon_state);
3909 goto out_freeflocks;
3910 pagelist = recon_state->pagelist;
3913 err = ceph_pagelist_reserve(pagelist, total_len);
3915 goto out_freeflocks;
3917 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3918 if (recon_state->msg_version >= 3) {
3919 ceph_pagelist_encode_8(pagelist, struct_v);
3920 ceph_pagelist_encode_8(pagelist, 1);
3921 ceph_pagelist_encode_32(pagelist, struct_len);
3923 ceph_pagelist_encode_string(pagelist, path, pathlen);
3924 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3925 ceph_locks_to_pagelist(flocks, pagelist,
3926 num_fcntl_locks, num_flock_locks);
3928 ceph_pagelist_encode_64(pagelist, snap_follows);
3932 err = ceph_pagelist_reserve(pagelist,
3933 sizeof(u64) + sizeof(u32) +
3934 pathlen + sizeof(rec.v1));
3938 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3939 ceph_pagelist_encode_string(pagelist, path, pathlen);
3940 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3944 ceph_mdsc_free_path(path, pathlen);
3946 recon_state->nr_caps++;
3950 static int encode_snap_realms(struct ceph_mds_client *mdsc,
3951 struct ceph_reconnect_state *recon_state)
3954 struct ceph_pagelist *pagelist = recon_state->pagelist;
3957 if (recon_state->msg_version >= 4) {
3958 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
3964 * snaprealms. we provide mds with the ino, seq (version), and
3965 * parent for all of our realms. If the mds has any newer info,
3968 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3969 struct ceph_snap_realm *realm =
3970 rb_entry(p, struct ceph_snap_realm, node);
3971 struct ceph_mds_snaprealm_reconnect sr_rec;
3973 if (recon_state->msg_version >= 4) {
3974 size_t need = sizeof(u8) * 2 + sizeof(u32) +
3977 if (pagelist->length + need > RECONNECT_MAX_SIZE) {
3978 err = send_reconnect_partial(recon_state);
3981 pagelist = recon_state->pagelist;
3984 err = ceph_pagelist_reserve(pagelist, need);
3988 ceph_pagelist_encode_8(pagelist, 1);
3989 ceph_pagelist_encode_8(pagelist, 1);
3990 ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
3993 dout(" adding snap realm %llx seq %lld parent %llx\n",
3994 realm->ino, realm->seq, realm->parent_ino);
3995 sr_rec.ino = cpu_to_le64(realm->ino);
3996 sr_rec.seq = cpu_to_le64(realm->seq);
3997 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3999 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
4003 recon_state->nr_realms++;
4011 * If an MDS fails and recovers, clients need to reconnect in order to
4012 * reestablish shared state. This includes all caps issued through
4013 * this session _and_ the snap_realm hierarchy. Because it's not
4014 * clear which snap realms the mds cares about, we send everything we
4015 * know about.. that ensures we'll then get any new info the
4016 * recovering MDS might have.
4018 * This is a relatively heavyweight operation, but it's rare.
4020 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
4021 struct ceph_mds_session *session)
4023 struct ceph_msg *reply;
4024 int mds = session->s_mds;
4026 struct ceph_reconnect_state recon_state = {
4031 pr_info("mds%d reconnect start\n", mds);
4033 recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
4034 if (!recon_state.pagelist)
4035 goto fail_nopagelist;
4037 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4041 xa_destroy(&session->s_delegated_inos);
4043 mutex_lock(&session->s_mutex);
4044 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4047 dout("session %p state %s\n", session,
4048 ceph_session_state_name(session->s_state));
4050 atomic_inc(&session->s_cap_gen);
4052 spin_lock(&session->s_cap_lock);
4053 /* don't know if session is readonly */
4054 session->s_readonly = 0;
4056 * notify __ceph_remove_cap() that we are composing cap reconnect.
4057 * If a cap get released before being added to the cap reconnect,
4058 * __ceph_remove_cap() should skip queuing cap release.
4060 session->s_cap_reconnect = 1;
4061 /* drop old cap expires; we're about to reestablish that state */
4062 detach_cap_releases(session, &dispose);
4063 spin_unlock(&session->s_cap_lock);
4064 dispose_cap_releases(mdsc, &dispose);
4066 /* trim unused caps to reduce MDS's cache rejoin time */
4067 if (mdsc->fsc->sb->s_root)
4068 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4070 ceph_con_close(&session->s_con);
4071 ceph_con_open(&session->s_con,
4072 CEPH_ENTITY_TYPE_MDS, mds,
4073 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4075 /* replay unsafe requests */
4076 replay_unsafe_requests(mdsc, session);
4078 ceph_early_kick_flushing_caps(mdsc, session);
4080 down_read(&mdsc->snap_rwsem);
4082 /* placeholder for nr_caps */
4083 err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4087 if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4088 recon_state.msg_version = 3;
4089 recon_state.allow_multi = true;
4090 } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4091 recon_state.msg_version = 3;
4093 recon_state.msg_version = 2;
4095 /* trsaverse this session's caps */
4096 err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4098 spin_lock(&session->s_cap_lock);
4099 session->s_cap_reconnect = 0;
4100 spin_unlock(&session->s_cap_lock);
4105 /* check if all realms can be encoded into current message */
4106 if (mdsc->num_snap_realms) {
4108 recon_state.pagelist->length +
4109 mdsc->num_snap_realms *
4110 sizeof(struct ceph_mds_snaprealm_reconnect);
4111 if (recon_state.msg_version >= 4) {
4112 /* number of realms */
4113 total_len += sizeof(u32);
4114 /* version, compat_version and struct_len */
4115 total_len += mdsc->num_snap_realms *
4116 (2 * sizeof(u8) + sizeof(u32));
4118 if (total_len > RECONNECT_MAX_SIZE) {
4119 if (!recon_state.allow_multi) {
4123 if (recon_state.nr_caps) {
4124 err = send_reconnect_partial(&recon_state);
4128 recon_state.msg_version = 5;
4132 err = encode_snap_realms(mdsc, &recon_state);
4136 if (recon_state.msg_version >= 5) {
4137 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4142 if (recon_state.nr_caps || recon_state.nr_realms) {
4144 list_first_entry(&recon_state.pagelist->head,
4146 __le32 *addr = kmap_atomic(page);
4147 if (recon_state.nr_caps) {
4148 WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4149 *addr = cpu_to_le32(recon_state.nr_caps);
4150 } else if (recon_state.msg_version >= 4) {
4151 *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4153 kunmap_atomic(addr);
4156 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4157 if (recon_state.msg_version >= 4)
4158 reply->hdr.compat_version = cpu_to_le16(4);
4160 reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4161 ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
4163 ceph_con_send(&session->s_con, reply);
4165 mutex_unlock(&session->s_mutex);
4167 mutex_lock(&mdsc->mutex);
4168 __wake_requests(mdsc, &session->s_waiting);
4169 mutex_unlock(&mdsc->mutex);
4171 up_read(&mdsc->snap_rwsem);
4172 ceph_pagelist_release(recon_state.pagelist);
4176 ceph_msg_put(reply);
4177 up_read(&mdsc->snap_rwsem);
4178 mutex_unlock(&session->s_mutex);
4180 ceph_pagelist_release(recon_state.pagelist);
4182 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
4188 * compare old and new mdsmaps, kicking requests
4189 * and closing out old connections as necessary
4191 * called under mdsc->mutex.
4193 static void check_new_map(struct ceph_mds_client *mdsc,
4194 struct ceph_mdsmap *newmap,
4195 struct ceph_mdsmap *oldmap)
4198 int oldstate, newstate;
4199 struct ceph_mds_session *s;
4200 unsigned long targets[DIV_ROUND_UP(CEPH_MAX_MDS, sizeof(unsigned long))] = {0};
4202 dout("check_new_map new %u old %u\n",
4203 newmap->m_epoch, oldmap->m_epoch);
4205 if (newmap->m_info) {
4206 for (i = 0; i < newmap->possible_max_rank; i++) {
4207 for (j = 0; j < newmap->m_info[i].num_export_targets; j++)
4208 set_bit(newmap->m_info[i].export_targets[j], targets);
4212 for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4213 if (!mdsc->sessions[i])
4215 s = mdsc->sessions[i];
4216 oldstate = ceph_mdsmap_get_state(oldmap, i);
4217 newstate = ceph_mdsmap_get_state(newmap, i);
4219 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
4220 i, ceph_mds_state_name(oldstate),
4221 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4222 ceph_mds_state_name(newstate),
4223 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4224 ceph_session_state_name(s->s_state));
4226 if (i >= newmap->possible_max_rank) {
4227 /* force close session for stopped mds */
4228 ceph_get_mds_session(s);
4229 __unregister_session(mdsc, s);
4230 __wake_requests(mdsc, &s->s_waiting);
4231 mutex_unlock(&mdsc->mutex);
4233 mutex_lock(&s->s_mutex);
4234 cleanup_session_requests(mdsc, s);
4235 remove_session_caps(s);
4236 mutex_unlock(&s->s_mutex);
4238 ceph_put_mds_session(s);
4240 mutex_lock(&mdsc->mutex);
4241 kick_requests(mdsc, i);
4245 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4246 ceph_mdsmap_get_addr(newmap, i),
4247 sizeof(struct ceph_entity_addr))) {
4249 mutex_unlock(&mdsc->mutex);
4250 mutex_lock(&s->s_mutex);
4251 mutex_lock(&mdsc->mutex);
4252 ceph_con_close(&s->s_con);
4253 mutex_unlock(&s->s_mutex);
4254 s->s_state = CEPH_MDS_SESSION_RESTARTING;
4255 } else if (oldstate == newstate) {
4256 continue; /* nothing new with this mds */
4262 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4263 newstate >= CEPH_MDS_STATE_RECONNECT) {
4264 mutex_unlock(&mdsc->mutex);
4265 clear_bit(i, targets);
4266 send_mds_reconnect(mdsc, s);
4267 mutex_lock(&mdsc->mutex);
4271 * kick request on any mds that has gone active.
4273 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4274 newstate >= CEPH_MDS_STATE_ACTIVE) {
4275 if (oldstate != CEPH_MDS_STATE_CREATING &&
4276 oldstate != CEPH_MDS_STATE_STARTING)
4277 pr_info("mds%d recovery completed\n", s->s_mds);
4278 kick_requests(mdsc, i);
4279 mutex_unlock(&mdsc->mutex);
4280 mutex_lock(&s->s_mutex);
4281 mutex_lock(&mdsc->mutex);
4282 ceph_kick_flushing_caps(mdsc, s);
4283 mutex_unlock(&s->s_mutex);
4284 wake_up_session_caps(s, RECONNECT);
4289 * Only open and reconnect sessions that don't exist yet.
4291 for (i = 0; i < newmap->possible_max_rank; i++) {
4293 * In case the import MDS is crashed just after
4294 * the EImportStart journal is flushed, so when
4295 * a standby MDS takes over it and is replaying
4296 * the EImportStart journal the new MDS daemon
4297 * will wait the client to reconnect it, but the
4298 * client may never register/open the session yet.
4300 * Will try to reconnect that MDS daemon if the
4301 * rank number is in the export targets array and
4302 * is the up:reconnect state.
4304 newstate = ceph_mdsmap_get_state(newmap, i);
4305 if (!test_bit(i, targets) || newstate != CEPH_MDS_STATE_RECONNECT)
4309 * The session maybe registered and opened by some
4310 * requests which were choosing random MDSes during
4311 * the mdsc->mutex's unlock/lock gap below in rare
4312 * case. But the related MDS daemon will just queue
4313 * that requests and be still waiting for the client's
4314 * reconnection request in up:reconnect state.
4316 s = __ceph_lookup_mds_session(mdsc, i);
4318 s = __open_export_target_session(mdsc, i);
4321 pr_err("failed to open export target session, err %d\n",
4326 dout("send reconnect to export target mds.%d\n", i);
4327 mutex_unlock(&mdsc->mutex);
4328 send_mds_reconnect(mdsc, s);
4329 ceph_put_mds_session(s);
4330 mutex_lock(&mdsc->mutex);
4333 for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4334 s = mdsc->sessions[i];
4337 if (!ceph_mdsmap_is_laggy(newmap, i))
4339 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4340 s->s_state == CEPH_MDS_SESSION_HUNG ||
4341 s->s_state == CEPH_MDS_SESSION_CLOSING) {
4342 dout(" connecting to export targets of laggy mds%d\n",
4344 __open_export_target_sessions(mdsc, s);
4356 * caller must hold session s_mutex, dentry->d_lock
4358 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4360 struct ceph_dentry_info *di = ceph_dentry(dentry);
4362 ceph_put_mds_session(di->lease_session);
4363 di->lease_session = NULL;
4366 static void handle_lease(struct ceph_mds_client *mdsc,
4367 struct ceph_mds_session *session,
4368 struct ceph_msg *msg)
4370 struct super_block *sb = mdsc->fsc->sb;
4371 struct inode *inode;
4372 struct dentry *parent, *dentry;
4373 struct ceph_dentry_info *di;
4374 int mds = session->s_mds;
4375 struct ceph_mds_lease *h = msg->front.iov_base;
4377 struct ceph_vino vino;
4381 dout("handle_lease from mds%d\n", mds);
4384 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4386 vino.ino = le64_to_cpu(h->ino);
4387 vino.snap = CEPH_NOSNAP;
4388 seq = le32_to_cpu(h->seq);
4389 dname.len = get_unaligned_le32(h + 1);
4390 if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4392 dname.name = (void *)(h + 1) + sizeof(u32);
4395 inode = ceph_find_inode(sb, vino);
4396 dout("handle_lease %s, ino %llx %p %.*s\n",
4397 ceph_lease_op_name(h->action), vino.ino, inode,
4398 dname.len, dname.name);
4400 mutex_lock(&session->s_mutex);
4401 inc_session_sequence(session);
4404 dout("handle_lease no inode %llx\n", vino.ino);
4409 parent = d_find_alias(inode);
4411 dout("no parent dentry on inode %p\n", inode);
4413 goto release; /* hrm... */
4415 dname.hash = full_name_hash(parent, dname.name, dname.len);
4416 dentry = d_lookup(parent, &dname);
4421 spin_lock(&dentry->d_lock);
4422 di = ceph_dentry(dentry);
4423 switch (h->action) {
4424 case CEPH_MDS_LEASE_REVOKE:
4425 if (di->lease_session == session) {
4426 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4427 h->seq = cpu_to_le32(di->lease_seq);
4428 __ceph_mdsc_drop_dentry_lease(dentry);
4433 case CEPH_MDS_LEASE_RENEW:
4434 if (di->lease_session == session &&
4435 di->lease_gen == atomic_read(&session->s_cap_gen) &&
4436 di->lease_renew_from &&
4437 di->lease_renew_after == 0) {
4438 unsigned long duration =
4439 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4441 di->lease_seq = seq;
4442 di->time = di->lease_renew_from + duration;
4443 di->lease_renew_after = di->lease_renew_from +
4445 di->lease_renew_from = 0;
4449 spin_unlock(&dentry->d_lock);
4456 /* let's just reuse the same message */
4457 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4459 ceph_con_send(&session->s_con, msg);
4462 mutex_unlock(&session->s_mutex);
4467 pr_err("corrupt lease message\n");
4471 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4472 struct dentry *dentry, char action,
4475 struct ceph_msg *msg;
4476 struct ceph_mds_lease *lease;
4478 int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4480 dout("lease_send_msg identry %p %s to mds%d\n",
4481 dentry, ceph_lease_op_name(action), session->s_mds);
4483 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4486 lease = msg->front.iov_base;
4487 lease->action = action;
4488 lease->seq = cpu_to_le32(seq);
4490 spin_lock(&dentry->d_lock);
4491 dir = d_inode(dentry->d_parent);
4492 lease->ino = cpu_to_le64(ceph_ino(dir));
4493 lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4495 put_unaligned_le32(dentry->d_name.len, lease + 1);
4496 memcpy((void *)(lease + 1) + 4,
4497 dentry->d_name.name, dentry->d_name.len);
4498 spin_unlock(&dentry->d_lock);
4500 * if this is a preemptive lease RELEASE, no need to
4501 * flush request stream, since the actual request will
4504 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
4506 ceph_con_send(&session->s_con, msg);
4510 * lock unlock the session, to wait ongoing session activities
4512 static void lock_unlock_session(struct ceph_mds_session *s)
4514 mutex_lock(&s->s_mutex);
4515 mutex_unlock(&s->s_mutex);
4518 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4520 struct ceph_fs_client *fsc = mdsc->fsc;
4522 if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4525 if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4528 if (!READ_ONCE(fsc->blocklisted))
4531 pr_info("auto reconnect after blocklisted\n");
4532 ceph_force_reconnect(fsc->sb);
4535 bool check_session_state(struct ceph_mds_session *s)
4537 struct ceph_fs_client *fsc = s->s_mdsc->fsc;
4539 switch (s->s_state) {
4540 case CEPH_MDS_SESSION_OPEN:
4541 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4542 s->s_state = CEPH_MDS_SESSION_HUNG;
4543 pr_info("mds%d hung\n", s->s_mds);
4546 case CEPH_MDS_SESSION_CLOSING:
4547 /* Should never reach this when not force unmounting */
4548 WARN_ON_ONCE(s->s_ttl &&
4549 READ_ONCE(fsc->mount_state) != CEPH_MOUNT_SHUTDOWN);
4551 case CEPH_MDS_SESSION_NEW:
4552 case CEPH_MDS_SESSION_RESTARTING:
4553 case CEPH_MDS_SESSION_CLOSED:
4554 case CEPH_MDS_SESSION_REJECTED:
4562 * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
4563 * then we need to retransmit that request.
4565 void inc_session_sequence(struct ceph_mds_session *s)
4567 lockdep_assert_held(&s->s_mutex);
4571 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4574 dout("resending session close request for mds%d\n", s->s_mds);
4575 ret = request_close_session(s);
4577 pr_err("unable to close session to mds%d: %d\n",
4583 * delayed work -- periodically trim expired leases, renew caps with mds. If
4584 * the @delay parameter is set to 0 or if it's more than 5 secs, the default
4585 * workqueue delay value of 5 secs will be used.
4587 static void schedule_delayed(struct ceph_mds_client *mdsc, unsigned long delay)
4589 unsigned long max_delay = HZ * 5;
4591 /* 5 secs default delay */
4592 if (!delay || (delay > max_delay))
4594 schedule_delayed_work(&mdsc->delayed_work,
4595 round_jiffies_relative(delay));
4598 static void delayed_work(struct work_struct *work)
4600 struct ceph_mds_client *mdsc =
4601 container_of(work, struct ceph_mds_client, delayed_work.work);
4602 unsigned long delay;
4607 dout("mdsc delayed_work\n");
4612 mutex_lock(&mdsc->mutex);
4613 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4614 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4615 mdsc->last_renew_caps);
4617 mdsc->last_renew_caps = jiffies;
4619 for (i = 0; i < mdsc->max_sessions; i++) {
4620 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4624 if (!check_session_state(s)) {
4625 ceph_put_mds_session(s);
4628 mutex_unlock(&mdsc->mutex);
4630 mutex_lock(&s->s_mutex);
4632 send_renew_caps(mdsc, s);
4634 ceph_con_keepalive(&s->s_con);
4635 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4636 s->s_state == CEPH_MDS_SESSION_HUNG)
4637 ceph_send_cap_releases(mdsc, s);
4638 mutex_unlock(&s->s_mutex);
4639 ceph_put_mds_session(s);
4641 mutex_lock(&mdsc->mutex);
4643 mutex_unlock(&mdsc->mutex);
4645 delay = ceph_check_delayed_caps(mdsc);
4647 ceph_queue_cap_reclaim_work(mdsc);
4649 ceph_trim_snapid_map(mdsc);
4651 maybe_recover_session(mdsc);
4653 schedule_delayed(mdsc, delay);
4656 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4659 struct ceph_mds_client *mdsc;
4662 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4666 mutex_init(&mdsc->mutex);
4667 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4668 if (!mdsc->mdsmap) {
4673 init_completion(&mdsc->safe_umount_waiters);
4674 init_waitqueue_head(&mdsc->session_close_wq);
4675 INIT_LIST_HEAD(&mdsc->waiting_for_map);
4676 mdsc->quotarealms_inodes = RB_ROOT;
4677 mutex_init(&mdsc->quotarealms_inodes_mutex);
4678 init_rwsem(&mdsc->snap_rwsem);
4679 mdsc->snap_realms = RB_ROOT;
4680 INIT_LIST_HEAD(&mdsc->snap_empty);
4681 spin_lock_init(&mdsc->snap_empty_lock);
4682 mdsc->request_tree = RB_ROOT;
4683 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4684 mdsc->last_renew_caps = jiffies;
4685 INIT_LIST_HEAD(&mdsc->cap_delay_list);
4686 INIT_LIST_HEAD(&mdsc->cap_wait_list);
4687 spin_lock_init(&mdsc->cap_delay_lock);
4688 INIT_LIST_HEAD(&mdsc->snap_flush_list);
4689 spin_lock_init(&mdsc->snap_flush_lock);
4690 mdsc->last_cap_flush_tid = 1;
4691 INIT_LIST_HEAD(&mdsc->cap_flush_list);
4692 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4693 spin_lock_init(&mdsc->cap_dirty_lock);
4694 init_waitqueue_head(&mdsc->cap_flushing_wq);
4695 INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4696 err = ceph_metric_init(&mdsc->metric);
4700 spin_lock_init(&mdsc->dentry_list_lock);
4701 INIT_LIST_HEAD(&mdsc->dentry_leases);
4702 INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4704 ceph_caps_init(mdsc);
4705 ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4707 spin_lock_init(&mdsc->snapid_map_lock);
4708 mdsc->snapid_map_tree = RB_ROOT;
4709 INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4711 init_rwsem(&mdsc->pool_perm_rwsem);
4712 mdsc->pool_perm_tree = RB_ROOT;
4714 strscpy(mdsc->nodename, utsname()->nodename,
4715 sizeof(mdsc->nodename));
4721 kfree(mdsc->mdsmap);
4728 * Wait for safe replies on open mds requests. If we time out, drop
4729 * all requests from the tree to avoid dangling dentry refs.
4731 static void wait_requests(struct ceph_mds_client *mdsc)
4733 struct ceph_options *opts = mdsc->fsc->client->options;
4734 struct ceph_mds_request *req;
4736 mutex_lock(&mdsc->mutex);
4737 if (__get_oldest_req(mdsc)) {
4738 mutex_unlock(&mdsc->mutex);
4740 dout("wait_requests waiting for requests\n");
4741 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4742 ceph_timeout_jiffies(opts->mount_timeout));
4744 /* tear down remaining requests */
4745 mutex_lock(&mdsc->mutex);
4746 while ((req = __get_oldest_req(mdsc))) {
4747 dout("wait_requests timed out on tid %llu\n",
4749 list_del_init(&req->r_wait);
4750 __unregister_request(mdsc, req);
4753 mutex_unlock(&mdsc->mutex);
4754 dout("wait_requests done\n");
4757 void send_flush_mdlog(struct ceph_mds_session *s)
4759 struct ceph_msg *msg;
4762 * Pre-luminous MDS crashes when it sees an unknown session request
4764 if (!CEPH_HAVE_FEATURE(s->s_con.peer_features, SERVER_LUMINOUS))
4767 mutex_lock(&s->s_mutex);
4768 dout("request mdlog flush to mds%d (%s)s seq %lld\n", s->s_mds,
4769 ceph_session_state_name(s->s_state), s->s_seq);
4770 msg = ceph_create_session_msg(CEPH_SESSION_REQUEST_FLUSH_MDLOG,
4773 pr_err("failed to request mdlog flush to mds%d (%s) seq %lld\n",
4774 s->s_mds, ceph_session_state_name(s->s_state), s->s_seq);
4776 ceph_con_send(&s->s_con, msg);
4778 mutex_unlock(&s->s_mutex);
4782 * called before mount is ro, and before dentries are torn down.
4783 * (hmm, does this still race with new lookups?)
4785 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4787 dout("pre_umount\n");
4790 ceph_mdsc_iterate_sessions(mdsc, send_flush_mdlog, true);
4791 ceph_mdsc_iterate_sessions(mdsc, lock_unlock_session, false);
4792 ceph_flush_dirty_caps(mdsc);
4793 wait_requests(mdsc);
4796 * wait for reply handlers to drop their request refs and
4797 * their inode/dcache refs
4801 ceph_cleanup_quotarealms_inodes(mdsc);
4805 * wait for all write mds requests to flush.
4807 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
4809 struct ceph_mds_request *req = NULL, *nextreq;
4812 mutex_lock(&mdsc->mutex);
4813 dout("wait_unsafe_requests want %lld\n", want_tid);
4815 req = __get_oldest_req(mdsc);
4816 while (req && req->r_tid <= want_tid) {
4817 /* find next request */
4818 n = rb_next(&req->r_node);
4820 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4823 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4824 (req->r_op & CEPH_MDS_OP_WRITE)) {
4826 ceph_mdsc_get_request(req);
4828 ceph_mdsc_get_request(nextreq);
4829 mutex_unlock(&mdsc->mutex);
4830 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
4831 req->r_tid, want_tid);
4832 wait_for_completion(&req->r_safe_completion);
4833 mutex_lock(&mdsc->mutex);
4834 ceph_mdsc_put_request(req);
4836 break; /* next dne before, so we're done! */
4837 if (RB_EMPTY_NODE(&nextreq->r_node)) {
4838 /* next request was removed from tree */
4839 ceph_mdsc_put_request(nextreq);
4842 ceph_mdsc_put_request(nextreq); /* won't go away */
4846 mutex_unlock(&mdsc->mutex);
4847 dout("wait_unsafe_requests done\n");
4850 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4852 u64 want_tid, want_flush;
4854 if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
4858 mutex_lock(&mdsc->mutex);
4859 want_tid = mdsc->last_tid;
4860 mutex_unlock(&mdsc->mutex);
4862 ceph_flush_dirty_caps(mdsc);
4863 spin_lock(&mdsc->cap_dirty_lock);
4864 want_flush = mdsc->last_cap_flush_tid;
4865 if (!list_empty(&mdsc->cap_flush_list)) {
4866 struct ceph_cap_flush *cf =
4867 list_last_entry(&mdsc->cap_flush_list,
4868 struct ceph_cap_flush, g_list);
4871 spin_unlock(&mdsc->cap_dirty_lock);
4873 dout("sync want tid %lld flush_seq %lld\n",
4874 want_tid, want_flush);
4876 wait_unsafe_requests(mdsc, want_tid);
4877 wait_caps_flush(mdsc, want_flush);
4881 * true if all sessions are closed, or we force unmount
4883 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
4885 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4887 return atomic_read(&mdsc->num_sessions) <= skipped;
4891 * called after sb is ro.
4893 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
4895 struct ceph_options *opts = mdsc->fsc->client->options;
4896 struct ceph_mds_session *session;
4900 dout("close_sessions\n");
4902 /* close sessions */
4903 mutex_lock(&mdsc->mutex);
4904 for (i = 0; i < mdsc->max_sessions; i++) {
4905 session = __ceph_lookup_mds_session(mdsc, i);
4908 mutex_unlock(&mdsc->mutex);
4909 mutex_lock(&session->s_mutex);
4910 if (__close_session(mdsc, session) <= 0)
4912 mutex_unlock(&session->s_mutex);
4913 ceph_put_mds_session(session);
4914 mutex_lock(&mdsc->mutex);
4916 mutex_unlock(&mdsc->mutex);
4918 dout("waiting for sessions to close\n");
4919 wait_event_timeout(mdsc->session_close_wq,
4920 done_closing_sessions(mdsc, skipped),
4921 ceph_timeout_jiffies(opts->mount_timeout));
4923 /* tear down remaining sessions */
4924 mutex_lock(&mdsc->mutex);
4925 for (i = 0; i < mdsc->max_sessions; i++) {
4926 if (mdsc->sessions[i]) {
4927 session = ceph_get_mds_session(mdsc->sessions[i]);
4928 __unregister_session(mdsc, session);
4929 mutex_unlock(&mdsc->mutex);
4930 mutex_lock(&session->s_mutex);
4931 remove_session_caps(session);
4932 mutex_unlock(&session->s_mutex);
4933 ceph_put_mds_session(session);
4934 mutex_lock(&mdsc->mutex);
4937 WARN_ON(!list_empty(&mdsc->cap_delay_list));
4938 mutex_unlock(&mdsc->mutex);
4940 ceph_cleanup_snapid_map(mdsc);
4941 ceph_cleanup_empty_realms(mdsc);
4943 cancel_work_sync(&mdsc->cap_reclaim_work);
4944 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4949 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
4951 struct ceph_mds_session *session;
4954 dout("force umount\n");
4956 mutex_lock(&mdsc->mutex);
4957 for (mds = 0; mds < mdsc->max_sessions; mds++) {
4958 session = __ceph_lookup_mds_session(mdsc, mds);
4962 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
4963 __unregister_session(mdsc, session);
4964 __wake_requests(mdsc, &session->s_waiting);
4965 mutex_unlock(&mdsc->mutex);
4967 mutex_lock(&session->s_mutex);
4968 __close_session(mdsc, session);
4969 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
4970 cleanup_session_requests(mdsc, session);
4971 remove_session_caps(session);
4973 mutex_unlock(&session->s_mutex);
4974 ceph_put_mds_session(session);
4976 mutex_lock(&mdsc->mutex);
4977 kick_requests(mdsc, mds);
4979 __wake_requests(mdsc, &mdsc->waiting_for_map);
4980 mutex_unlock(&mdsc->mutex);
4983 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
4987 * Make sure the delayed work stopped before releasing
4990 * Because the cancel_delayed_work_sync() will only
4991 * guarantee that the work finishes executing. But the
4992 * delayed work will re-arm itself again after that.
4994 flush_delayed_work(&mdsc->delayed_work);
4997 ceph_mdsmap_destroy(mdsc->mdsmap);
4998 kfree(mdsc->sessions);
4999 ceph_caps_finalize(mdsc);
5000 ceph_pool_perm_destroy(mdsc);
5003 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
5005 struct ceph_mds_client *mdsc = fsc->mdsc;
5006 dout("mdsc_destroy %p\n", mdsc);
5011 /* flush out any connection work with references to us */
5014 ceph_mdsc_stop(mdsc);
5016 ceph_metric_destroy(&mdsc->metric);
5020 dout("mdsc_destroy %p done\n", mdsc);
5023 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5025 struct ceph_fs_client *fsc = mdsc->fsc;
5026 const char *mds_namespace = fsc->mount_options->mds_namespace;
5027 void *p = msg->front.iov_base;
5028 void *end = p + msg->front.iov_len;
5031 u32 mount_fscid = (u32)-1;
5034 ceph_decode_need(&p, end, sizeof(u32), bad);
5035 epoch = ceph_decode_32(&p);
5037 dout("handle_fsmap epoch %u\n", epoch);
5039 /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
5040 ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
5042 ceph_decode_32_safe(&p, end, num_fs, bad);
5043 while (num_fs-- > 0) {
5044 void *info_p, *info_end;
5048 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
5049 p += 2; // info_v, info_cv
5050 info_len = ceph_decode_32(&p);
5051 ceph_decode_need(&p, end, info_len, bad);
5053 info_end = p + info_len;
5056 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
5057 fscid = ceph_decode_32(&info_p);
5058 namelen = ceph_decode_32(&info_p);
5059 ceph_decode_need(&info_p, info_end, namelen, bad);
5061 if (mds_namespace &&
5062 strlen(mds_namespace) == namelen &&
5063 !strncmp(mds_namespace, (char *)info_p, namelen)) {
5064 mount_fscid = fscid;
5069 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
5070 if (mount_fscid != (u32)-1) {
5071 fsc->client->monc.fs_cluster_id = mount_fscid;
5072 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
5074 ceph_monc_renew_subs(&fsc->client->monc);
5082 pr_err("error decoding fsmap\n");
5084 mutex_lock(&mdsc->mutex);
5085 mdsc->mdsmap_err = err;
5086 __wake_requests(mdsc, &mdsc->waiting_for_map);
5087 mutex_unlock(&mdsc->mutex);
5091 * handle mds map update.
5093 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
5097 void *p = msg->front.iov_base;
5098 void *end = p + msg->front.iov_len;
5099 struct ceph_mdsmap *newmap, *oldmap;
5100 struct ceph_fsid fsid;
5103 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5104 ceph_decode_copy(&p, &fsid, sizeof(fsid));
5105 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
5107 epoch = ceph_decode_32(&p);
5108 maplen = ceph_decode_32(&p);
5109 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
5111 /* do we need it? */
5112 mutex_lock(&mdsc->mutex);
5113 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
5114 dout("handle_map epoch %u <= our %u\n",
5115 epoch, mdsc->mdsmap->m_epoch);
5116 mutex_unlock(&mdsc->mutex);
5120 newmap = ceph_mdsmap_decode(&p, end, ceph_msgr2(mdsc->fsc->client));
5121 if (IS_ERR(newmap)) {
5122 err = PTR_ERR(newmap);
5126 /* swap into place */
5128 oldmap = mdsc->mdsmap;
5129 mdsc->mdsmap = newmap;
5130 check_new_map(mdsc, newmap, oldmap);
5131 ceph_mdsmap_destroy(oldmap);
5133 mdsc->mdsmap = newmap; /* first mds map */
5135 mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5138 __wake_requests(mdsc, &mdsc->waiting_for_map);
5139 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5140 mdsc->mdsmap->m_epoch);
5142 mutex_unlock(&mdsc->mutex);
5143 schedule_delayed(mdsc, 0);
5147 mutex_unlock(&mdsc->mutex);
5149 pr_err("error decoding mdsmap %d\n", err);
5153 static struct ceph_connection *mds_get_con(struct ceph_connection *con)
5155 struct ceph_mds_session *s = con->private;
5157 if (ceph_get_mds_session(s))
5162 static void mds_put_con(struct ceph_connection *con)
5164 struct ceph_mds_session *s = con->private;
5166 ceph_put_mds_session(s);
5170 * if the client is unresponsive for long enough, the mds will kill
5171 * the session entirely.
5173 static void mds_peer_reset(struct ceph_connection *con)
5175 struct ceph_mds_session *s = con->private;
5176 struct ceph_mds_client *mdsc = s->s_mdsc;
5178 pr_warn("mds%d closed our session\n", s->s_mds);
5179 send_mds_reconnect(mdsc, s);
5182 static void mds_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5184 struct ceph_mds_session *s = con->private;
5185 struct ceph_mds_client *mdsc = s->s_mdsc;
5186 int type = le16_to_cpu(msg->hdr.type);
5188 mutex_lock(&mdsc->mutex);
5189 if (__verify_registered_session(mdsc, s) < 0) {
5190 mutex_unlock(&mdsc->mutex);
5193 mutex_unlock(&mdsc->mutex);
5196 case CEPH_MSG_MDS_MAP:
5197 ceph_mdsc_handle_mdsmap(mdsc, msg);
5199 case CEPH_MSG_FS_MAP_USER:
5200 ceph_mdsc_handle_fsmap(mdsc, msg);
5202 case CEPH_MSG_CLIENT_SESSION:
5203 handle_session(s, msg);
5205 case CEPH_MSG_CLIENT_REPLY:
5206 handle_reply(s, msg);
5208 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
5209 handle_forward(mdsc, s, msg);
5211 case CEPH_MSG_CLIENT_CAPS:
5212 ceph_handle_caps(s, msg);
5214 case CEPH_MSG_CLIENT_SNAP:
5215 ceph_handle_snap(mdsc, s, msg);
5217 case CEPH_MSG_CLIENT_LEASE:
5218 handle_lease(mdsc, s, msg);
5220 case CEPH_MSG_CLIENT_QUOTA:
5221 ceph_handle_quota(mdsc, s, msg);
5225 pr_err("received unknown message type %d %s\n", type,
5226 ceph_msg_type_name(type));
5237 * Note: returned pointer is the address of a structure that's
5238 * managed separately. Caller must *not* attempt to free it.
5240 static struct ceph_auth_handshake *
5241 mds_get_authorizer(struct ceph_connection *con, int *proto, int force_new)
5243 struct ceph_mds_session *s = con->private;
5244 struct ceph_mds_client *mdsc = s->s_mdsc;
5245 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5246 struct ceph_auth_handshake *auth = &s->s_auth;
5249 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5250 force_new, proto, NULL, NULL);
5252 return ERR_PTR(ret);
5257 static int mds_add_authorizer_challenge(struct ceph_connection *con,
5258 void *challenge_buf, int challenge_buf_len)
5260 struct ceph_mds_session *s = con->private;
5261 struct ceph_mds_client *mdsc = s->s_mdsc;
5262 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5264 return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5265 challenge_buf, challenge_buf_len);
5268 static int mds_verify_authorizer_reply(struct ceph_connection *con)
5270 struct ceph_mds_session *s = con->private;
5271 struct ceph_mds_client *mdsc = s->s_mdsc;
5272 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5273 struct ceph_auth_handshake *auth = &s->s_auth;
5275 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5276 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5277 NULL, NULL, NULL, NULL);
5280 static int mds_invalidate_authorizer(struct ceph_connection *con)
5282 struct ceph_mds_session *s = con->private;
5283 struct ceph_mds_client *mdsc = s->s_mdsc;
5284 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5286 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
5288 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
5291 static int mds_get_auth_request(struct ceph_connection *con,
5292 void *buf, int *buf_len,
5293 void **authorizer, int *authorizer_len)
5295 struct ceph_mds_session *s = con->private;
5296 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5297 struct ceph_auth_handshake *auth = &s->s_auth;
5300 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5305 *authorizer = auth->authorizer_buf;
5306 *authorizer_len = auth->authorizer_buf_len;
5310 static int mds_handle_auth_reply_more(struct ceph_connection *con,
5311 void *reply, int reply_len,
5312 void *buf, int *buf_len,
5313 void **authorizer, int *authorizer_len)
5315 struct ceph_mds_session *s = con->private;
5316 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5317 struct ceph_auth_handshake *auth = &s->s_auth;
5320 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5325 *authorizer = auth->authorizer_buf;
5326 *authorizer_len = auth->authorizer_buf_len;
5330 static int mds_handle_auth_done(struct ceph_connection *con,
5331 u64 global_id, void *reply, int reply_len,
5332 u8 *session_key, int *session_key_len,
5333 u8 *con_secret, int *con_secret_len)
5335 struct ceph_mds_session *s = con->private;
5336 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5337 struct ceph_auth_handshake *auth = &s->s_auth;
5339 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5340 session_key, session_key_len,
5341 con_secret, con_secret_len);
5344 static int mds_handle_auth_bad_method(struct ceph_connection *con,
5345 int used_proto, int result,
5346 const int *allowed_protos, int proto_cnt,
5347 const int *allowed_modes, int mode_cnt)
5349 struct ceph_mds_session *s = con->private;
5350 struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
5353 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
5355 allowed_protos, proto_cnt,
5356 allowed_modes, mode_cnt)) {
5357 ret = ceph_monc_validate_auth(monc);
5365 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
5366 struct ceph_msg_header *hdr, int *skip)
5368 struct ceph_msg *msg;
5369 int type = (int) le16_to_cpu(hdr->type);
5370 int front_len = (int) le32_to_cpu(hdr->front_len);
5376 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
5378 pr_err("unable to allocate msg type %d len %d\n",
5386 static int mds_sign_message(struct ceph_msg *msg)
5388 struct ceph_mds_session *s = msg->con->private;
5389 struct ceph_auth_handshake *auth = &s->s_auth;
5391 return ceph_auth_sign_message(auth, msg);
5394 static int mds_check_message_signature(struct ceph_msg *msg)
5396 struct ceph_mds_session *s = msg->con->private;
5397 struct ceph_auth_handshake *auth = &s->s_auth;
5399 return ceph_auth_check_message_signature(auth, msg);
5402 static const struct ceph_connection_operations mds_con_ops = {
5405 .alloc_msg = mds_alloc_msg,
5406 .dispatch = mds_dispatch,
5407 .peer_reset = mds_peer_reset,
5408 .get_authorizer = mds_get_authorizer,
5409 .add_authorizer_challenge = mds_add_authorizer_challenge,
5410 .verify_authorizer_reply = mds_verify_authorizer_reply,
5411 .invalidate_authorizer = mds_invalidate_authorizer,
5412 .sign_message = mds_sign_message,
5413 .check_message_signature = mds_check_message_signature,
5414 .get_auth_request = mds_get_auth_request,
5415 .handle_auth_reply_more = mds_handle_auth_reply_more,
5416 .handle_auth_done = mds_handle_auth_done,
5417 .handle_auth_bad_method = mds_handle_auth_bad_method,