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>
16 #include "mds_client.h"
18 #include <linux/ceph/ceph_features.h>
19 #include <linux/ceph/messenger.h>
20 #include <linux/ceph/decode.h>
21 #include <linux/ceph/pagelist.h>
22 #include <linux/ceph/auth.h>
23 #include <linux/ceph/debugfs.h>
25 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
28 * A cluster of MDS (metadata server) daemons is responsible for
29 * managing the file system namespace (the directory hierarchy and
30 * inodes) and for coordinating shared access to storage. Metadata is
31 * partitioning hierarchically across a number of servers, and that
32 * partition varies over time as the cluster adjusts the distribution
33 * in order to balance load.
35 * The MDS client is primarily responsible to managing synchronous
36 * metadata requests for operations like open, unlink, and so forth.
37 * If there is a MDS failure, we find out about it when we (possibly
38 * request and) receive a new MDS map, and can resubmit affected
41 * For the most part, though, we take advantage of a lossless
42 * communications channel to the MDS, and do not need to worry about
43 * timing out or resubmitting requests.
45 * We maintain a stateful "session" with each MDS we interact with.
46 * Within each session, we sent periodic heartbeat messages to ensure
47 * any capabilities or leases we have been issues remain valid. If
48 * the session times out and goes stale, our leases and capabilities
49 * are no longer valid.
52 struct ceph_reconnect_state {
53 struct ceph_mds_session *session;
54 int nr_caps, nr_realms;
55 struct ceph_pagelist *pagelist;
60 static void __wake_requests(struct ceph_mds_client *mdsc,
61 struct list_head *head);
62 static void ceph_cap_release_work(struct work_struct *work);
63 static void ceph_cap_reclaim_work(struct work_struct *work);
65 static const struct ceph_connection_operations mds_con_ops;
72 static int parse_reply_info_quota(void **p, void *end,
73 struct ceph_mds_reply_info_in *info)
75 u8 struct_v, struct_compat;
78 ceph_decode_8_safe(p, end, struct_v, bad);
79 ceph_decode_8_safe(p, end, struct_compat, bad);
80 /* struct_v is expected to be >= 1. we only
81 * understand encoding with struct_compat == 1. */
82 if (!struct_v || struct_compat != 1)
84 ceph_decode_32_safe(p, end, struct_len, bad);
85 ceph_decode_need(p, end, struct_len, bad);
86 end = *p + struct_len;
87 ceph_decode_64_safe(p, end, info->max_bytes, bad);
88 ceph_decode_64_safe(p, end, info->max_files, bad);
96 * parse individual inode info
98 static int parse_reply_info_in(void **p, void *end,
99 struct ceph_mds_reply_info_in *info,
105 if (features == (u64)-1) {
108 ceph_decode_8_safe(p, end, struct_v, bad);
109 ceph_decode_8_safe(p, end, struct_compat, bad);
110 /* struct_v is expected to be >= 1. we only understand
111 * encoding with struct_compat == 1. */
112 if (!struct_v || struct_compat != 1)
114 ceph_decode_32_safe(p, end, struct_len, bad);
115 ceph_decode_need(p, end, struct_len, bad);
116 end = *p + struct_len;
119 ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
121 *p += sizeof(struct ceph_mds_reply_inode) +
122 sizeof(*info->in->fragtree.splits) *
123 le32_to_cpu(info->in->fragtree.nsplits);
125 ceph_decode_32_safe(p, end, info->symlink_len, bad);
126 ceph_decode_need(p, end, info->symlink_len, bad);
128 *p += info->symlink_len;
130 ceph_decode_copy_safe(p, end, &info->dir_layout,
131 sizeof(info->dir_layout), bad);
132 ceph_decode_32_safe(p, end, info->xattr_len, bad);
133 ceph_decode_need(p, end, info->xattr_len, bad);
134 info->xattr_data = *p;
135 *p += info->xattr_len;
137 if (features == (u64)-1) {
139 ceph_decode_64_safe(p, end, info->inline_version, bad);
140 ceph_decode_32_safe(p, end, info->inline_len, bad);
141 ceph_decode_need(p, end, info->inline_len, bad);
142 info->inline_data = *p;
143 *p += info->inline_len;
145 err = parse_reply_info_quota(p, end, info);
149 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
150 if (info->pool_ns_len > 0) {
151 ceph_decode_need(p, end, info->pool_ns_len, bad);
152 info->pool_ns_data = *p;
153 *p += info->pool_ns_len;
157 ceph_decode_need(p, end, sizeof(info->btime), bad);
158 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
160 /* change attribute */
161 ceph_decode_64_safe(p, end, info->change_attr, bad);
165 ceph_decode_32_safe(p, end, info->dir_pin, bad);
167 info->dir_pin = -ENODATA;
170 /* snapshot birth time, remains zero for v<=2 */
172 ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
173 ceph_decode_copy(p, &info->snap_btime,
174 sizeof(info->snap_btime));
176 memset(&info->snap_btime, 0, sizeof(info->snap_btime));
181 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
182 ceph_decode_64_safe(p, end, info->inline_version, bad);
183 ceph_decode_32_safe(p, end, info->inline_len, bad);
184 ceph_decode_need(p, end, info->inline_len, bad);
185 info->inline_data = *p;
186 *p += info->inline_len;
188 info->inline_version = CEPH_INLINE_NONE;
190 if (features & CEPH_FEATURE_MDS_QUOTA) {
191 err = parse_reply_info_quota(p, end, info);
199 info->pool_ns_len = 0;
200 info->pool_ns_data = NULL;
201 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
202 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
203 if (info->pool_ns_len > 0) {
204 ceph_decode_need(p, end, info->pool_ns_len, bad);
205 info->pool_ns_data = *p;
206 *p += info->pool_ns_len;
210 if (features & CEPH_FEATURE_FS_BTIME) {
211 ceph_decode_need(p, end, sizeof(info->btime), bad);
212 ceph_decode_copy(p, &info->btime, sizeof(info->btime));
213 ceph_decode_64_safe(p, end, info->change_attr, bad);
216 info->dir_pin = -ENODATA;
217 /* info->snap_btime remains zero */
226 static int parse_reply_info_dir(void **p, void *end,
227 struct ceph_mds_reply_dirfrag **dirfrag,
230 if (features == (u64)-1) {
231 u8 struct_v, struct_compat;
233 ceph_decode_8_safe(p, end, struct_v, bad);
234 ceph_decode_8_safe(p, end, struct_compat, bad);
235 /* struct_v is expected to be >= 1. we only understand
236 * encoding whose struct_compat == 1. */
237 if (!struct_v || struct_compat != 1)
239 ceph_decode_32_safe(p, end, struct_len, bad);
240 ceph_decode_need(p, end, struct_len, bad);
241 end = *p + struct_len;
244 ceph_decode_need(p, end, sizeof(**dirfrag), bad);
246 *p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
247 if (unlikely(*p > end))
249 if (features == (u64)-1)
256 static int parse_reply_info_lease(void **p, void *end,
257 struct ceph_mds_reply_lease **lease,
260 if (features == (u64)-1) {
261 u8 struct_v, struct_compat;
263 ceph_decode_8_safe(p, end, struct_v, bad);
264 ceph_decode_8_safe(p, end, struct_compat, bad);
265 /* struct_v is expected to be >= 1. we only understand
266 * encoding whose struct_compat == 1. */
267 if (!struct_v || struct_compat != 1)
269 ceph_decode_32_safe(p, end, struct_len, bad);
270 ceph_decode_need(p, end, struct_len, bad);
271 end = *p + struct_len;
274 ceph_decode_need(p, end, sizeof(**lease), bad);
276 *p += sizeof(**lease);
277 if (features == (u64)-1)
285 * parse a normal reply, which may contain a (dir+)dentry and/or a
288 static int parse_reply_info_trace(void **p, void *end,
289 struct ceph_mds_reply_info_parsed *info,
294 if (info->head->is_dentry) {
295 err = parse_reply_info_in(p, end, &info->diri, features);
299 err = parse_reply_info_dir(p, end, &info->dirfrag, features);
303 ceph_decode_32_safe(p, end, info->dname_len, bad);
304 ceph_decode_need(p, end, info->dname_len, bad);
306 *p += info->dname_len;
308 err = parse_reply_info_lease(p, end, &info->dlease, features);
313 if (info->head->is_target) {
314 err = parse_reply_info_in(p, end, &info->targeti, features);
319 if (unlikely(*p != end))
326 pr_err("problem parsing mds trace %d\n", err);
331 * parse readdir results
333 static int parse_reply_info_readdir(void **p, void *end,
334 struct ceph_mds_reply_info_parsed *info,
340 err = parse_reply_info_dir(p, end, &info->dir_dir, features);
344 ceph_decode_need(p, end, sizeof(num) + 2, bad);
345 num = ceph_decode_32(p);
347 u16 flags = ceph_decode_16(p);
348 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
349 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
350 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
351 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
356 BUG_ON(!info->dir_entries);
357 if ((unsigned long)(info->dir_entries + num) >
358 (unsigned long)info->dir_entries + info->dir_buf_size) {
359 pr_err("dir contents are larger than expected\n");
366 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
368 ceph_decode_32_safe(p, end, rde->name_len, bad);
369 ceph_decode_need(p, end, rde->name_len, bad);
372 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
375 err = parse_reply_info_lease(p, end, &rde->lease, features);
379 err = parse_reply_info_in(p, end, &rde->inode, features);
382 /* ceph_readdir_prepopulate() will update it */
389 /* Skip over any unrecognized fields */
396 pr_err("problem parsing dir contents %d\n", err);
401 * parse fcntl F_GETLK results
403 static int parse_reply_info_filelock(void **p, void *end,
404 struct ceph_mds_reply_info_parsed *info,
407 if (*p + sizeof(*info->filelock_reply) > end)
410 info->filelock_reply = *p;
412 /* Skip over any unrecognized fields */
420 #if BITS_PER_LONG == 64
422 #define DELEGATED_INO_AVAILABLE xa_mk_value(1)
424 static int ceph_parse_deleg_inos(void **p, void *end,
425 struct ceph_mds_session *s)
429 ceph_decode_32_safe(p, end, sets, bad);
430 dout("got %u sets of delegated inodes\n", sets);
434 ceph_decode_64_safe(p, end, start, bad);
435 ceph_decode_64_safe(p, end, len, bad);
437 int err = xa_insert(&s->s_delegated_inos, ino = start++,
438 DELEGATED_INO_AVAILABLE,
441 dout("added delegated inode 0x%llx\n",
443 } else if (err == -EBUSY) {
444 pr_warn("ceph: MDS delegated inode 0x%llx more than once.\n",
456 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
461 xa_for_each(&s->s_delegated_inos, ino, val) {
462 val = xa_erase(&s->s_delegated_inos, ino);
463 if (val == DELEGATED_INO_AVAILABLE)
469 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
471 return xa_insert(&s->s_delegated_inos, ino, DELEGATED_INO_AVAILABLE,
474 #else /* BITS_PER_LONG == 64 */
476 * FIXME: xarrays can't handle 64-bit indexes on a 32-bit arch. For now, just
477 * ignore delegated_inos on 32 bit arch. Maybe eventually add xarrays for top
480 static int ceph_parse_deleg_inos(void **p, void *end,
481 struct ceph_mds_session *s)
485 ceph_decode_32_safe(p, end, sets, bad);
487 ceph_decode_skip_n(p, end, sets * 2 * sizeof(__le64), bad);
493 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
498 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
502 #endif /* BITS_PER_LONG == 64 */
505 * parse create results
507 static int parse_reply_info_create(void **p, void *end,
508 struct ceph_mds_reply_info_parsed *info,
509 u64 features, struct ceph_mds_session *s)
513 if (features == (u64)-1 ||
514 (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
516 /* Malformed reply? */
517 info->has_create_ino = false;
518 } else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
519 info->has_create_ino = true;
520 /* struct_v, struct_compat, and len */
521 ceph_decode_skip_n(p, end, 2 + sizeof(u32), bad);
522 ceph_decode_64_safe(p, end, info->ino, bad);
523 ret = ceph_parse_deleg_inos(p, end, s);
528 ceph_decode_64_safe(p, end, info->ino, bad);
529 info->has_create_ino = true;
536 /* Skip over any unrecognized fields */
544 * parse extra results
546 static int parse_reply_info_extra(void **p, void *end,
547 struct ceph_mds_reply_info_parsed *info,
548 u64 features, struct ceph_mds_session *s)
550 u32 op = le32_to_cpu(info->head->op);
552 if (op == CEPH_MDS_OP_GETFILELOCK)
553 return parse_reply_info_filelock(p, end, info, features);
554 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
555 return parse_reply_info_readdir(p, end, info, features);
556 else if (op == CEPH_MDS_OP_CREATE)
557 return parse_reply_info_create(p, end, info, features, s);
563 * parse entire mds reply
565 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
566 struct ceph_mds_reply_info_parsed *info,
573 info->head = msg->front.iov_base;
574 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
575 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
578 ceph_decode_32_safe(&p, end, len, bad);
580 ceph_decode_need(&p, end, len, bad);
581 err = parse_reply_info_trace(&p, p+len, info, features);
587 ceph_decode_32_safe(&p, end, len, bad);
589 ceph_decode_need(&p, end, len, bad);
590 err = parse_reply_info_extra(&p, p+len, info, features, s);
596 ceph_decode_32_safe(&p, end, len, bad);
597 info->snapblob_len = len;
608 pr_err("mds parse_reply err %d\n", err);
612 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
614 if (!info->dir_entries)
616 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
623 const char *ceph_session_state_name(int s)
626 case CEPH_MDS_SESSION_NEW: return "new";
627 case CEPH_MDS_SESSION_OPENING: return "opening";
628 case CEPH_MDS_SESSION_OPEN: return "open";
629 case CEPH_MDS_SESSION_HUNG: return "hung";
630 case CEPH_MDS_SESSION_CLOSING: return "closing";
631 case CEPH_MDS_SESSION_CLOSED: return "closed";
632 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
633 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
634 case CEPH_MDS_SESSION_REJECTED: return "rejected";
635 default: return "???";
639 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
641 if (refcount_inc_not_zero(&s->s_ref)) {
642 dout("mdsc get_session %p %d -> %d\n", s,
643 refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
646 dout("mdsc get_session %p 0 -- FAIL\n", s);
651 void ceph_put_mds_session(struct ceph_mds_session *s)
653 dout("mdsc put_session %p %d -> %d\n", s,
654 refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
655 if (refcount_dec_and_test(&s->s_ref)) {
656 if (s->s_auth.authorizer)
657 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
658 WARN_ON(mutex_is_locked(&s->s_mutex));
659 xa_destroy(&s->s_delegated_inos);
665 * called under mdsc->mutex
667 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
670 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
672 return ceph_get_mds_session(mdsc->sessions[mds]);
675 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
677 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
683 static int __verify_registered_session(struct ceph_mds_client *mdsc,
684 struct ceph_mds_session *s)
686 if (s->s_mds >= mdsc->max_sessions ||
687 mdsc->sessions[s->s_mds] != s)
693 * create+register a new session for given mds.
694 * called under mdsc->mutex.
696 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
699 struct ceph_mds_session *s;
701 if (mds >= mdsc->mdsmap->possible_max_rank)
702 return ERR_PTR(-EINVAL);
704 s = kzalloc(sizeof(*s), GFP_NOFS);
706 return ERR_PTR(-ENOMEM);
708 if (mds >= mdsc->max_sessions) {
709 int newmax = 1 << get_count_order(mds + 1);
710 struct ceph_mds_session **sa;
712 dout("%s: realloc to %d\n", __func__, newmax);
713 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
716 if (mdsc->sessions) {
717 memcpy(sa, mdsc->sessions,
718 mdsc->max_sessions * sizeof(void *));
719 kfree(mdsc->sessions);
722 mdsc->max_sessions = newmax;
725 dout("%s: mds%d\n", __func__, mds);
728 s->s_state = CEPH_MDS_SESSION_NEW;
731 mutex_init(&s->s_mutex);
733 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
735 spin_lock_init(&s->s_gen_ttl_lock);
737 s->s_cap_ttl = jiffies - 1;
739 spin_lock_init(&s->s_cap_lock);
740 s->s_renew_requested = 0;
742 INIT_LIST_HEAD(&s->s_caps);
744 refcount_set(&s->s_ref, 1);
745 INIT_LIST_HEAD(&s->s_waiting);
746 INIT_LIST_HEAD(&s->s_unsafe);
747 xa_init(&s->s_delegated_inos);
748 s->s_num_cap_releases = 0;
749 s->s_cap_reconnect = 0;
750 s->s_cap_iterator = NULL;
751 INIT_LIST_HEAD(&s->s_cap_releases);
752 INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
754 INIT_LIST_HEAD(&s->s_cap_dirty);
755 INIT_LIST_HEAD(&s->s_cap_flushing);
757 mdsc->sessions[mds] = s;
758 atomic_inc(&mdsc->num_sessions);
759 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
761 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
762 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
768 return ERR_PTR(-ENOMEM);
772 * called under mdsc->mutex
774 static void __unregister_session(struct ceph_mds_client *mdsc,
775 struct ceph_mds_session *s)
777 dout("__unregister_session mds%d %p\n", s->s_mds, s);
778 BUG_ON(mdsc->sessions[s->s_mds] != s);
779 mdsc->sessions[s->s_mds] = NULL;
780 ceph_con_close(&s->s_con);
781 ceph_put_mds_session(s);
782 atomic_dec(&mdsc->num_sessions);
786 * drop session refs in request.
788 * should be last request ref, or hold mdsc->mutex
790 static void put_request_session(struct ceph_mds_request *req)
792 if (req->r_session) {
793 ceph_put_mds_session(req->r_session);
794 req->r_session = NULL;
798 void ceph_mdsc_release_request(struct kref *kref)
800 struct ceph_mds_request *req = container_of(kref,
801 struct ceph_mds_request,
803 ceph_mdsc_release_dir_caps_no_check(req);
804 destroy_reply_info(&req->r_reply_info);
806 ceph_msg_put(req->r_request);
808 ceph_msg_put(req->r_reply);
810 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
811 /* avoid calling iput_final() in mds dispatch threads */
812 ceph_async_iput(req->r_inode);
815 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
816 ceph_async_iput(req->r_parent);
818 ceph_async_iput(req->r_target_inode);
821 if (req->r_old_dentry)
822 dput(req->r_old_dentry);
823 if (req->r_old_dentry_dir) {
825 * track (and drop pins for) r_old_dentry_dir
826 * separately, since r_old_dentry's d_parent may have
827 * changed between the dir mutex being dropped and
828 * this request being freed.
830 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
832 ceph_async_iput(req->r_old_dentry_dir);
836 put_cred(req->r_cred);
838 ceph_pagelist_release(req->r_pagelist);
839 put_request_session(req);
840 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
841 WARN_ON_ONCE(!list_empty(&req->r_wait));
842 kmem_cache_free(ceph_mds_request_cachep, req);
845 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
848 * lookup session, bump ref if found.
850 * called under mdsc->mutex.
852 static struct ceph_mds_request *
853 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
855 struct ceph_mds_request *req;
857 req = lookup_request(&mdsc->request_tree, tid);
859 ceph_mdsc_get_request(req);
865 * Register an in-flight request, and assign a tid. Link to directory
866 * are modifying (if any).
868 * Called under mdsc->mutex.
870 static void __register_request(struct ceph_mds_client *mdsc,
871 struct ceph_mds_request *req,
876 req->r_tid = ++mdsc->last_tid;
877 if (req->r_num_caps) {
878 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
881 pr_err("__register_request %p "
882 "failed to reserve caps: %d\n", req, ret);
883 /* set req->r_err to fail early from __do_request */
888 dout("__register_request %p tid %lld\n", req, req->r_tid);
889 ceph_mdsc_get_request(req);
890 insert_request(&mdsc->request_tree, req);
892 req->r_cred = get_current_cred();
894 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
895 mdsc->oldest_tid = req->r_tid;
898 struct ceph_inode_info *ci = ceph_inode(dir);
901 req->r_unsafe_dir = dir;
902 spin_lock(&ci->i_unsafe_lock);
903 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
904 spin_unlock(&ci->i_unsafe_lock);
908 static void __unregister_request(struct ceph_mds_client *mdsc,
909 struct ceph_mds_request *req)
911 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
913 /* Never leave an unregistered request on an unsafe list! */
914 list_del_init(&req->r_unsafe_item);
916 if (req->r_tid == mdsc->oldest_tid) {
917 struct rb_node *p = rb_next(&req->r_node);
918 mdsc->oldest_tid = 0;
920 struct ceph_mds_request *next_req =
921 rb_entry(p, struct ceph_mds_request, r_node);
922 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
923 mdsc->oldest_tid = next_req->r_tid;
930 erase_request(&mdsc->request_tree, req);
932 if (req->r_unsafe_dir) {
933 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
934 spin_lock(&ci->i_unsafe_lock);
935 list_del_init(&req->r_unsafe_dir_item);
936 spin_unlock(&ci->i_unsafe_lock);
938 if (req->r_target_inode &&
939 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
940 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
941 spin_lock(&ci->i_unsafe_lock);
942 list_del_init(&req->r_unsafe_target_item);
943 spin_unlock(&ci->i_unsafe_lock);
946 if (req->r_unsafe_dir) {
947 /* avoid calling iput_final() in mds dispatch threads */
948 ceph_async_iput(req->r_unsafe_dir);
949 req->r_unsafe_dir = NULL;
952 complete_all(&req->r_safe_completion);
954 ceph_mdsc_put_request(req);
958 * Walk back up the dentry tree until we hit a dentry representing a
959 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
960 * when calling this) to ensure that the objects won't disappear while we're
961 * working with them. Once we hit a candidate dentry, we attempt to take a
962 * reference to it, and return that as the result.
964 static struct inode *get_nonsnap_parent(struct dentry *dentry)
966 struct inode *inode = NULL;
968 while (dentry && !IS_ROOT(dentry)) {
969 inode = d_inode_rcu(dentry);
970 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
972 dentry = dentry->d_parent;
975 inode = igrab(inode);
980 * Choose mds to send request to next. If there is a hint set in the
981 * request (e.g., due to a prior forward hint from the mds), use that.
982 * Otherwise, consult frag tree and/or caps to identify the
983 * appropriate mds. If all else fails, choose randomly.
985 * Called under mdsc->mutex.
987 static int __choose_mds(struct ceph_mds_client *mdsc,
988 struct ceph_mds_request *req,
992 struct ceph_inode_info *ci;
993 struct ceph_cap *cap;
994 int mode = req->r_direct_mode;
996 u32 hash = req->r_direct_hash;
997 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
1003 * is there a specific mds we should try? ignore hint if we have
1004 * no session and the mds is not up (active or recovering).
1006 if (req->r_resend_mds >= 0 &&
1007 (__have_session(mdsc, req->r_resend_mds) ||
1008 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1009 dout("%s using resend_mds mds%d\n", __func__,
1011 return req->r_resend_mds;
1014 if (mode == USE_RANDOM_MDS)
1019 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1020 inode = req->r_inode;
1023 /* req->r_dentry is non-null for LSSNAP request */
1025 inode = get_nonsnap_parent(req->r_dentry);
1027 dout("%s using snapdir's parent %p\n", __func__, inode);
1029 } else if (req->r_dentry) {
1030 /* ignore race with rename; old or new d_parent is okay */
1031 struct dentry *parent;
1035 parent = READ_ONCE(req->r_dentry->d_parent);
1036 dir = req->r_parent ? : d_inode_rcu(parent);
1038 if (!dir || dir->i_sb != mdsc->fsc->sb) {
1039 /* not this fs or parent went negative */
1040 inode = d_inode(req->r_dentry);
1043 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
1044 /* direct snapped/virtual snapdir requests
1045 * based on parent dir inode */
1046 inode = get_nonsnap_parent(parent);
1047 dout("%s using nonsnap parent %p\n", __func__, inode);
1050 inode = d_inode(req->r_dentry);
1051 if (!inode || mode == USE_AUTH_MDS) {
1054 hash = ceph_dentry_hash(dir, req->r_dentry);
1063 dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1067 ci = ceph_inode(inode);
1069 if (is_hash && S_ISDIR(inode->i_mode)) {
1070 struct ceph_inode_frag frag;
1073 ceph_choose_frag(ci, hash, &frag, &found);
1075 if (mode == USE_ANY_MDS && frag.ndist > 0) {
1078 /* choose a random replica */
1079 get_random_bytes(&r, 1);
1082 dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1083 __func__, inode, ceph_vinop(inode),
1084 frag.frag, mds, (int)r, frag.ndist);
1085 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1086 CEPH_MDS_STATE_ACTIVE &&
1087 !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1091 /* since this file/dir wasn't known to be
1092 * replicated, then we want to look for the
1093 * authoritative mds. */
1094 if (frag.mds >= 0) {
1095 /* choose auth mds */
1097 dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1098 __func__, inode, ceph_vinop(inode),
1100 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1101 CEPH_MDS_STATE_ACTIVE) {
1102 if (!ceph_mdsmap_is_laggy(mdsc->mdsmap,
1107 mode = USE_AUTH_MDS;
1111 spin_lock(&ci->i_ceph_lock);
1113 if (mode == USE_AUTH_MDS)
1114 cap = ci->i_auth_cap;
1115 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1116 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1118 spin_unlock(&ci->i_ceph_lock);
1119 ceph_async_iput(inode);
1122 mds = cap->session->s_mds;
1123 dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1124 inode, ceph_vinop(inode), mds,
1125 cap == ci->i_auth_cap ? "auth " : "", cap);
1126 spin_unlock(&ci->i_ceph_lock);
1128 /* avoid calling iput_final() while holding mdsc->mutex or
1129 * in mds dispatch threads */
1130 ceph_async_iput(inode);
1137 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1138 dout("%s chose random mds%d\n", __func__, mds);
1146 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
1148 struct ceph_msg *msg;
1149 struct ceph_mds_session_head *h;
1151 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1154 pr_err("create_session_msg ENOMEM creating msg\n");
1157 h = msg->front.iov_base;
1158 h->op = cpu_to_le32(op);
1159 h->seq = cpu_to_le64(seq);
1164 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1165 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1166 static int encode_supported_features(void **p, void *end)
1168 static const size_t count = ARRAY_SIZE(feature_bits);
1172 size_t size = FEATURE_BYTES(count);
1174 if (WARN_ON_ONCE(*p + 4 + size > end))
1177 ceph_encode_32(p, size);
1178 memset(*p, 0, size);
1179 for (i = 0; i < count; i++)
1180 ((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
1183 if (WARN_ON_ONCE(*p + 4 > end))
1186 ceph_encode_32(p, 0);
1192 static const unsigned char metric_bits[] = CEPHFS_METRIC_SPEC_CLIENT_SUPPORTED;
1193 #define METRIC_BYTES(cnt) (DIV_ROUND_UP((size_t)metric_bits[cnt - 1] + 1, 64) * 8)
1194 static int encode_metric_spec(void **p, void *end)
1196 static const size_t count = ARRAY_SIZE(metric_bits);
1199 if (WARN_ON_ONCE(*p + 2 > end))
1202 ceph_encode_8(p, 1); /* version */
1203 ceph_encode_8(p, 1); /* compat */
1207 size_t size = METRIC_BYTES(count);
1209 if (WARN_ON_ONCE(*p + 4 + 4 + size > end))
1212 /* metric spec info length */
1213 ceph_encode_32(p, 4 + size);
1216 ceph_encode_32(p, size);
1217 memset(*p, 0, size);
1218 for (i = 0; i < count; i++)
1219 ((unsigned char *)(*p))[i / 8] |= BIT(metric_bits[i] % 8);
1222 if (WARN_ON_ONCE(*p + 4 + 4 > end))
1225 /* metric spec info length */
1226 ceph_encode_32(p, 4);
1228 ceph_encode_32(p, 0);
1235 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1236 * to include additional client metadata fields.
1238 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1240 struct ceph_msg *msg;
1241 struct ceph_mds_session_head *h;
1243 int extra_bytes = 0;
1244 int metadata_key_count = 0;
1245 struct ceph_options *opt = mdsc->fsc->client->options;
1246 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1251 const char* metadata[][2] = {
1252 {"hostname", mdsc->nodename},
1253 {"kernel_version", init_utsname()->release},
1254 {"entity_id", opt->name ? : ""},
1255 {"root", fsopt->server_path ? : "/"},
1259 /* Calculate serialized length of metadata */
1260 extra_bytes = 4; /* map length */
1261 for (i = 0; metadata[i][0]; ++i) {
1262 extra_bytes += 8 + strlen(metadata[i][0]) +
1263 strlen(metadata[i][1]);
1264 metadata_key_count++;
1267 /* supported feature */
1269 count = ARRAY_SIZE(feature_bits);
1271 size = FEATURE_BYTES(count);
1272 extra_bytes += 4 + size;
1276 count = ARRAY_SIZE(metric_bits);
1278 size = METRIC_BYTES(count);
1279 extra_bytes += 2 + 4 + 4 + size;
1281 /* Allocate the message */
1282 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1285 pr_err("create_session_msg ENOMEM creating msg\n");
1286 return ERR_PTR(-ENOMEM);
1288 p = msg->front.iov_base;
1289 end = p + msg->front.iov_len;
1292 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1293 h->seq = cpu_to_le64(seq);
1296 * Serialize client metadata into waiting buffer space, using
1297 * the format that userspace expects for map<string, string>
1299 * ClientSession messages with metadata are v4
1301 msg->hdr.version = cpu_to_le16(4);
1302 msg->hdr.compat_version = cpu_to_le16(1);
1304 /* The write pointer, following the session_head structure */
1307 /* Number of entries in the map */
1308 ceph_encode_32(&p, metadata_key_count);
1310 /* Two length-prefixed strings for each entry in the map */
1311 for (i = 0; metadata[i][0]; ++i) {
1312 size_t const key_len = strlen(metadata[i][0]);
1313 size_t const val_len = strlen(metadata[i][1]);
1315 ceph_encode_32(&p, key_len);
1316 memcpy(p, metadata[i][0], key_len);
1318 ceph_encode_32(&p, val_len);
1319 memcpy(p, metadata[i][1], val_len);
1323 ret = encode_supported_features(&p, end);
1325 pr_err("encode_supported_features failed!\n");
1327 return ERR_PTR(ret);
1330 ret = encode_metric_spec(&p, end);
1332 pr_err("encode_metric_spec failed!\n");
1334 return ERR_PTR(ret);
1337 msg->front.iov_len = p - msg->front.iov_base;
1338 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1344 * send session open request.
1346 * called under mdsc->mutex
1348 static int __open_session(struct ceph_mds_client *mdsc,
1349 struct ceph_mds_session *session)
1351 struct ceph_msg *msg;
1353 int mds = session->s_mds;
1355 /* wait for mds to go active? */
1356 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1357 dout("open_session to mds%d (%s)\n", mds,
1358 ceph_mds_state_name(mstate));
1359 session->s_state = CEPH_MDS_SESSION_OPENING;
1360 session->s_renew_requested = jiffies;
1362 /* send connect message */
1363 msg = create_session_open_msg(mdsc, session->s_seq);
1365 return PTR_ERR(msg);
1366 ceph_con_send(&session->s_con, msg);
1371 * open sessions for any export targets for the given mds
1373 * called under mdsc->mutex
1375 static struct ceph_mds_session *
1376 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1378 struct ceph_mds_session *session;
1381 session = __ceph_lookup_mds_session(mdsc, target);
1383 session = register_session(mdsc, target);
1384 if (IS_ERR(session))
1387 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1388 session->s_state == CEPH_MDS_SESSION_CLOSING) {
1389 ret = __open_session(mdsc, session);
1391 return ERR_PTR(ret);
1397 struct ceph_mds_session *
1398 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1400 struct ceph_mds_session *session;
1402 dout("open_export_target_session to mds%d\n", target);
1404 mutex_lock(&mdsc->mutex);
1405 session = __open_export_target_session(mdsc, target);
1406 mutex_unlock(&mdsc->mutex);
1411 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1412 struct ceph_mds_session *session)
1414 struct ceph_mds_info *mi;
1415 struct ceph_mds_session *ts;
1416 int i, mds = session->s_mds;
1418 if (mds >= mdsc->mdsmap->possible_max_rank)
1421 mi = &mdsc->mdsmap->m_info[mds];
1422 dout("open_export_target_sessions for mds%d (%d targets)\n",
1423 session->s_mds, mi->num_export_targets);
1425 for (i = 0; i < mi->num_export_targets; i++) {
1426 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1428 ceph_put_mds_session(ts);
1432 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1433 struct ceph_mds_session *session)
1435 mutex_lock(&mdsc->mutex);
1436 __open_export_target_sessions(mdsc, session);
1437 mutex_unlock(&mdsc->mutex);
1444 static void detach_cap_releases(struct ceph_mds_session *session,
1445 struct list_head *target)
1447 lockdep_assert_held(&session->s_cap_lock);
1449 list_splice_init(&session->s_cap_releases, target);
1450 session->s_num_cap_releases = 0;
1451 dout("dispose_cap_releases mds%d\n", session->s_mds);
1454 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1455 struct list_head *dispose)
1457 while (!list_empty(dispose)) {
1458 struct ceph_cap *cap;
1459 /* zero out the in-progress message */
1460 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1461 list_del(&cap->session_caps);
1462 ceph_put_cap(mdsc, cap);
1466 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1467 struct ceph_mds_session *session)
1469 struct ceph_mds_request *req;
1471 struct ceph_inode_info *ci;
1473 dout("cleanup_session_requests mds%d\n", session->s_mds);
1474 mutex_lock(&mdsc->mutex);
1475 while (!list_empty(&session->s_unsafe)) {
1476 req = list_first_entry(&session->s_unsafe,
1477 struct ceph_mds_request, r_unsafe_item);
1478 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1480 if (req->r_target_inode) {
1481 /* dropping unsafe change of inode's attributes */
1482 ci = ceph_inode(req->r_target_inode);
1483 errseq_set(&ci->i_meta_err, -EIO);
1485 if (req->r_unsafe_dir) {
1486 /* dropping unsafe directory operation */
1487 ci = ceph_inode(req->r_unsafe_dir);
1488 errseq_set(&ci->i_meta_err, -EIO);
1490 __unregister_request(mdsc, req);
1492 /* zero r_attempts, so kick_requests() will re-send requests */
1493 p = rb_first(&mdsc->request_tree);
1495 req = rb_entry(p, struct ceph_mds_request, r_node);
1497 if (req->r_session &&
1498 req->r_session->s_mds == session->s_mds)
1499 req->r_attempts = 0;
1501 mutex_unlock(&mdsc->mutex);
1505 * Helper to safely iterate over all caps associated with a session, with
1506 * special care taken to handle a racing __ceph_remove_cap().
1508 * Caller must hold session s_mutex.
1510 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1511 int (*cb)(struct inode *, struct ceph_cap *,
1514 struct list_head *p;
1515 struct ceph_cap *cap;
1516 struct inode *inode, *last_inode = NULL;
1517 struct ceph_cap *old_cap = NULL;
1520 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1521 spin_lock(&session->s_cap_lock);
1522 p = session->s_caps.next;
1523 while (p != &session->s_caps) {
1524 cap = list_entry(p, struct ceph_cap, session_caps);
1525 inode = igrab(&cap->ci->vfs_inode);
1530 session->s_cap_iterator = cap;
1531 spin_unlock(&session->s_cap_lock);
1534 /* avoid calling iput_final() while holding
1535 * s_mutex or in mds dispatch threads */
1536 ceph_async_iput(last_inode);
1540 ceph_put_cap(session->s_mdsc, old_cap);
1544 ret = cb(inode, cap, arg);
1547 spin_lock(&session->s_cap_lock);
1550 dout("iterate_session_caps finishing cap %p removal\n",
1552 BUG_ON(cap->session != session);
1553 cap->session = NULL;
1554 list_del_init(&cap->session_caps);
1555 session->s_nr_caps--;
1556 atomic64_dec(&session->s_mdsc->metric.total_caps);
1557 if (cap->queue_release)
1558 __ceph_queue_cap_release(session, cap);
1560 old_cap = cap; /* put_cap it w/o locks held */
1567 session->s_cap_iterator = NULL;
1568 spin_unlock(&session->s_cap_lock);
1570 ceph_async_iput(last_inode);
1572 ceph_put_cap(session->s_mdsc, old_cap);
1577 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1580 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1581 struct ceph_inode_info *ci = ceph_inode(inode);
1582 LIST_HEAD(to_remove);
1583 bool dirty_dropped = false;
1584 bool invalidate = false;
1586 dout("removing cap %p, ci is %p, inode is %p\n",
1587 cap, ci, &ci->vfs_inode);
1588 spin_lock(&ci->i_ceph_lock);
1589 __ceph_remove_cap(cap, false);
1590 if (!ci->i_auth_cap) {
1591 struct ceph_cap_flush *cf;
1592 struct ceph_mds_client *mdsc = fsc->mdsc;
1594 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1595 if (inode->i_data.nrpages > 0)
1597 if (ci->i_wrbuffer_ref > 0)
1598 mapping_set_error(&inode->i_data, -EIO);
1601 while (!list_empty(&ci->i_cap_flush_list)) {
1602 cf = list_first_entry(&ci->i_cap_flush_list,
1603 struct ceph_cap_flush, i_list);
1604 list_move(&cf->i_list, &to_remove);
1607 spin_lock(&mdsc->cap_dirty_lock);
1609 list_for_each_entry(cf, &to_remove, i_list)
1610 list_del(&cf->g_list);
1612 if (!list_empty(&ci->i_dirty_item)) {
1613 pr_warn_ratelimited(
1614 " dropping dirty %s state for %p %lld\n",
1615 ceph_cap_string(ci->i_dirty_caps),
1616 inode, ceph_ino(inode));
1617 ci->i_dirty_caps = 0;
1618 list_del_init(&ci->i_dirty_item);
1619 dirty_dropped = true;
1621 if (!list_empty(&ci->i_flushing_item)) {
1622 pr_warn_ratelimited(
1623 " dropping dirty+flushing %s state for %p %lld\n",
1624 ceph_cap_string(ci->i_flushing_caps),
1625 inode, ceph_ino(inode));
1626 ci->i_flushing_caps = 0;
1627 list_del_init(&ci->i_flushing_item);
1628 mdsc->num_cap_flushing--;
1629 dirty_dropped = true;
1631 spin_unlock(&mdsc->cap_dirty_lock);
1633 if (dirty_dropped) {
1634 errseq_set(&ci->i_meta_err, -EIO);
1636 if (ci->i_wrbuffer_ref_head == 0 &&
1637 ci->i_wr_ref == 0 &&
1638 ci->i_dirty_caps == 0 &&
1639 ci->i_flushing_caps == 0) {
1640 ceph_put_snap_context(ci->i_head_snapc);
1641 ci->i_head_snapc = NULL;
1645 if (atomic_read(&ci->i_filelock_ref) > 0) {
1646 /* make further file lock syscall return -EIO */
1647 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1648 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1649 inode, ceph_ino(inode));
1652 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1653 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1654 ci->i_prealloc_cap_flush = NULL;
1657 spin_unlock(&ci->i_ceph_lock);
1658 while (!list_empty(&to_remove)) {
1659 struct ceph_cap_flush *cf;
1660 cf = list_first_entry(&to_remove,
1661 struct ceph_cap_flush, i_list);
1662 list_del(&cf->i_list);
1663 ceph_free_cap_flush(cf);
1666 wake_up_all(&ci->i_cap_wq);
1668 ceph_queue_invalidate(inode);
1675 * caller must hold session s_mutex
1677 static void remove_session_caps(struct ceph_mds_session *session)
1679 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1680 struct super_block *sb = fsc->sb;
1683 dout("remove_session_caps on %p\n", session);
1684 ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1686 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1688 spin_lock(&session->s_cap_lock);
1689 if (session->s_nr_caps > 0) {
1690 struct inode *inode;
1691 struct ceph_cap *cap, *prev = NULL;
1692 struct ceph_vino vino;
1694 * iterate_session_caps() skips inodes that are being
1695 * deleted, we need to wait until deletions are complete.
1696 * __wait_on_freeing_inode() is designed for the job,
1697 * but it is not exported, so use lookup inode function
1700 while (!list_empty(&session->s_caps)) {
1701 cap = list_entry(session->s_caps.next,
1702 struct ceph_cap, session_caps);
1706 vino = cap->ci->i_vino;
1707 spin_unlock(&session->s_cap_lock);
1709 inode = ceph_find_inode(sb, vino);
1710 /* avoid calling iput_final() while holding s_mutex */
1711 ceph_async_iput(inode);
1713 spin_lock(&session->s_cap_lock);
1717 // drop cap expires and unlock s_cap_lock
1718 detach_cap_releases(session, &dispose);
1720 BUG_ON(session->s_nr_caps > 0);
1721 BUG_ON(!list_empty(&session->s_cap_flushing));
1722 spin_unlock(&session->s_cap_lock);
1723 dispose_cap_releases(session->s_mdsc, &dispose);
1733 * wake up any threads waiting on this session's caps. if the cap is
1734 * old (didn't get renewed on the client reconnect), remove it now.
1736 * caller must hold s_mutex.
1738 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1741 struct ceph_inode_info *ci = ceph_inode(inode);
1742 unsigned long ev = (unsigned long)arg;
1744 if (ev == RECONNECT) {
1745 spin_lock(&ci->i_ceph_lock);
1746 ci->i_wanted_max_size = 0;
1747 ci->i_requested_max_size = 0;
1748 spin_unlock(&ci->i_ceph_lock);
1749 } else if (ev == RENEWCAPS) {
1750 if (cap->cap_gen < cap->session->s_cap_gen) {
1751 /* mds did not re-issue stale cap */
1752 spin_lock(&ci->i_ceph_lock);
1753 cap->issued = cap->implemented = CEPH_CAP_PIN;
1754 spin_unlock(&ci->i_ceph_lock);
1756 } else if (ev == FORCE_RO) {
1758 wake_up_all(&ci->i_cap_wq);
1762 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1764 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1765 ceph_iterate_session_caps(session, wake_up_session_cb,
1766 (void *)(unsigned long)ev);
1770 * Send periodic message to MDS renewing all currently held caps. The
1771 * ack will reset the expiration for all caps from this session.
1773 * caller holds s_mutex
1775 static int send_renew_caps(struct ceph_mds_client *mdsc,
1776 struct ceph_mds_session *session)
1778 struct ceph_msg *msg;
1781 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1782 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1783 pr_info("mds%d caps stale\n", session->s_mds);
1784 session->s_renew_requested = jiffies;
1786 /* do not try to renew caps until a recovering mds has reconnected
1787 * with its clients. */
1788 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1789 if (state < CEPH_MDS_STATE_RECONNECT) {
1790 dout("send_renew_caps ignoring mds%d (%s)\n",
1791 session->s_mds, ceph_mds_state_name(state));
1795 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1796 ceph_mds_state_name(state));
1797 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1798 ++session->s_renew_seq);
1801 ceph_con_send(&session->s_con, msg);
1805 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1806 struct ceph_mds_session *session, u64 seq)
1808 struct ceph_msg *msg;
1810 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1811 session->s_mds, ceph_session_state_name(session->s_state), seq);
1812 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1815 ceph_con_send(&session->s_con, msg);
1821 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1823 * Called under session->s_mutex
1825 static void renewed_caps(struct ceph_mds_client *mdsc,
1826 struct ceph_mds_session *session, int is_renew)
1831 spin_lock(&session->s_cap_lock);
1832 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1834 session->s_cap_ttl = session->s_renew_requested +
1835 mdsc->mdsmap->m_session_timeout*HZ;
1838 if (time_before(jiffies, session->s_cap_ttl)) {
1839 pr_info("mds%d caps renewed\n", session->s_mds);
1842 pr_info("mds%d caps still stale\n", session->s_mds);
1845 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1846 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1847 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1848 spin_unlock(&session->s_cap_lock);
1851 wake_up_session_caps(session, RENEWCAPS);
1855 * send a session close request
1857 static int request_close_session(struct ceph_mds_session *session)
1859 struct ceph_msg *msg;
1861 dout("request_close_session mds%d state %s seq %lld\n",
1862 session->s_mds, ceph_session_state_name(session->s_state),
1864 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1867 ceph_con_send(&session->s_con, msg);
1872 * Called with s_mutex held.
1874 static int __close_session(struct ceph_mds_client *mdsc,
1875 struct ceph_mds_session *session)
1877 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1879 session->s_state = CEPH_MDS_SESSION_CLOSING;
1880 return request_close_session(session);
1883 static bool drop_negative_children(struct dentry *dentry)
1885 struct dentry *child;
1886 bool all_negative = true;
1888 if (!d_is_dir(dentry))
1891 spin_lock(&dentry->d_lock);
1892 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1893 if (d_really_is_positive(child)) {
1894 all_negative = false;
1898 spin_unlock(&dentry->d_lock);
1901 shrink_dcache_parent(dentry);
1903 return all_negative;
1907 * Trim old(er) caps.
1909 * Because we can't cache an inode without one or more caps, we do
1910 * this indirectly: if a cap is unused, we prune its aliases, at which
1911 * point the inode will hopefully get dropped to.
1913 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1914 * memory pressure from the MDS, though, so it needn't be perfect.
1916 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1918 int *remaining = arg;
1919 struct ceph_inode_info *ci = ceph_inode(inode);
1920 int used, wanted, oissued, mine;
1922 if (*remaining <= 0)
1925 spin_lock(&ci->i_ceph_lock);
1926 mine = cap->issued | cap->implemented;
1927 used = __ceph_caps_used(ci);
1928 wanted = __ceph_caps_file_wanted(ci);
1929 oissued = __ceph_caps_issued_other(ci, cap);
1931 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1932 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1933 ceph_cap_string(used), ceph_cap_string(wanted));
1934 if (cap == ci->i_auth_cap) {
1935 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1936 !list_empty(&ci->i_cap_snaps))
1938 if ((used | wanted) & CEPH_CAP_ANY_WR)
1940 /* Note: it's possible that i_filelock_ref becomes non-zero
1941 * after dropping auth caps. It doesn't hurt because reply
1942 * of lock mds request will re-add auth caps. */
1943 if (atomic_read(&ci->i_filelock_ref) > 0)
1946 /* The inode has cached pages, but it's no longer used.
1947 * we can safely drop it */
1948 if (S_ISREG(inode->i_mode) &&
1949 wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1950 !(oissued & CEPH_CAP_FILE_CACHE)) {
1954 if ((used | wanted) & ~oissued & mine)
1955 goto out; /* we need these caps */
1958 /* we aren't the only cap.. just remove us */
1959 __ceph_remove_cap(cap, true);
1962 struct dentry *dentry;
1963 /* try dropping referring dentries */
1964 spin_unlock(&ci->i_ceph_lock);
1965 dentry = d_find_any_alias(inode);
1966 if (dentry && drop_negative_children(dentry)) {
1969 d_prune_aliases(inode);
1970 count = atomic_read(&inode->i_count);
1973 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1982 spin_unlock(&ci->i_ceph_lock);
1987 * Trim session cap count down to some max number.
1989 int ceph_trim_caps(struct ceph_mds_client *mdsc,
1990 struct ceph_mds_session *session,
1993 int trim_caps = session->s_nr_caps - max_caps;
1995 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1996 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1997 if (trim_caps > 0) {
1998 int remaining = trim_caps;
2000 ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
2001 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
2002 session->s_mds, session->s_nr_caps, max_caps,
2003 trim_caps - remaining);
2006 ceph_flush_cap_releases(mdsc, session);
2010 static int check_caps_flush(struct ceph_mds_client *mdsc,
2015 spin_lock(&mdsc->cap_dirty_lock);
2016 if (!list_empty(&mdsc->cap_flush_list)) {
2017 struct ceph_cap_flush *cf =
2018 list_first_entry(&mdsc->cap_flush_list,
2019 struct ceph_cap_flush, g_list);
2020 if (cf->tid <= want_flush_tid) {
2021 dout("check_caps_flush still flushing tid "
2022 "%llu <= %llu\n", cf->tid, want_flush_tid);
2026 spin_unlock(&mdsc->cap_dirty_lock);
2031 * flush all dirty inode data to disk.
2033 * returns true if we've flushed through want_flush_tid
2035 static void wait_caps_flush(struct ceph_mds_client *mdsc,
2038 dout("check_caps_flush want %llu\n", want_flush_tid);
2040 wait_event(mdsc->cap_flushing_wq,
2041 check_caps_flush(mdsc, want_flush_tid));
2043 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2047 * called under s_mutex
2049 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
2050 struct ceph_mds_session *session)
2052 struct ceph_msg *msg = NULL;
2053 struct ceph_mds_cap_release *head;
2054 struct ceph_mds_cap_item *item;
2055 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
2056 struct ceph_cap *cap;
2057 LIST_HEAD(tmp_list);
2058 int num_cap_releases;
2059 __le32 barrier, *cap_barrier;
2061 down_read(&osdc->lock);
2062 barrier = cpu_to_le32(osdc->epoch_barrier);
2063 up_read(&osdc->lock);
2065 spin_lock(&session->s_cap_lock);
2067 list_splice_init(&session->s_cap_releases, &tmp_list);
2068 num_cap_releases = session->s_num_cap_releases;
2069 session->s_num_cap_releases = 0;
2070 spin_unlock(&session->s_cap_lock);
2072 while (!list_empty(&tmp_list)) {
2074 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2075 PAGE_SIZE, GFP_NOFS, false);
2078 head = msg->front.iov_base;
2079 head->num = cpu_to_le32(0);
2080 msg->front.iov_len = sizeof(*head);
2082 msg->hdr.version = cpu_to_le16(2);
2083 msg->hdr.compat_version = cpu_to_le16(1);
2086 cap = list_first_entry(&tmp_list, struct ceph_cap,
2088 list_del(&cap->session_caps);
2091 head = msg->front.iov_base;
2092 put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2094 item = msg->front.iov_base + msg->front.iov_len;
2095 item->ino = cpu_to_le64(cap->cap_ino);
2096 item->cap_id = cpu_to_le64(cap->cap_id);
2097 item->migrate_seq = cpu_to_le32(cap->mseq);
2098 item->seq = cpu_to_le32(cap->issue_seq);
2099 msg->front.iov_len += sizeof(*item);
2101 ceph_put_cap(mdsc, cap);
2103 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2104 // Append cap_barrier field
2105 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2106 *cap_barrier = barrier;
2107 msg->front.iov_len += sizeof(*cap_barrier);
2109 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2110 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2111 ceph_con_send(&session->s_con, msg);
2116 BUG_ON(num_cap_releases != 0);
2118 spin_lock(&session->s_cap_lock);
2119 if (!list_empty(&session->s_cap_releases))
2121 spin_unlock(&session->s_cap_lock);
2124 // Append cap_barrier field
2125 cap_barrier = msg->front.iov_base + msg->front.iov_len;
2126 *cap_barrier = barrier;
2127 msg->front.iov_len += sizeof(*cap_barrier);
2129 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2130 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2131 ceph_con_send(&session->s_con, msg);
2135 pr_err("send_cap_releases mds%d, failed to allocate message\n",
2137 spin_lock(&session->s_cap_lock);
2138 list_splice(&tmp_list, &session->s_cap_releases);
2139 session->s_num_cap_releases += num_cap_releases;
2140 spin_unlock(&session->s_cap_lock);
2143 static void ceph_cap_release_work(struct work_struct *work)
2145 struct ceph_mds_session *session =
2146 container_of(work, struct ceph_mds_session, s_cap_release_work);
2148 mutex_lock(&session->s_mutex);
2149 if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2150 session->s_state == CEPH_MDS_SESSION_HUNG)
2151 ceph_send_cap_releases(session->s_mdsc, session);
2152 mutex_unlock(&session->s_mutex);
2153 ceph_put_mds_session(session);
2156 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2157 struct ceph_mds_session *session)
2162 ceph_get_mds_session(session);
2163 if (queue_work(mdsc->fsc->cap_wq,
2164 &session->s_cap_release_work)) {
2165 dout("cap release work queued\n");
2167 ceph_put_mds_session(session);
2168 dout("failed to queue cap release work\n");
2173 * caller holds session->s_cap_lock
2175 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2176 struct ceph_cap *cap)
2178 list_add_tail(&cap->session_caps, &session->s_cap_releases);
2179 session->s_num_cap_releases++;
2181 if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2182 ceph_flush_cap_releases(session->s_mdsc, session);
2185 static void ceph_cap_reclaim_work(struct work_struct *work)
2187 struct ceph_mds_client *mdsc =
2188 container_of(work, struct ceph_mds_client, cap_reclaim_work);
2189 int ret = ceph_trim_dentries(mdsc);
2191 ceph_queue_cap_reclaim_work(mdsc);
2194 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2199 if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2200 dout("caps reclaim work queued\n");
2202 dout("failed to queue caps release work\n");
2206 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2211 val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2212 if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2213 atomic_set(&mdsc->cap_reclaim_pending, 0);
2214 ceph_queue_cap_reclaim_work(mdsc);
2222 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2225 struct ceph_inode_info *ci = ceph_inode(dir);
2226 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2227 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2228 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2229 unsigned int num_entries;
2232 spin_lock(&ci->i_ceph_lock);
2233 num_entries = ci->i_files + ci->i_subdirs;
2234 spin_unlock(&ci->i_ceph_lock);
2235 num_entries = max(num_entries, 1U);
2236 num_entries = min(num_entries, opt->max_readdir);
2238 order = get_order(size * num_entries);
2239 while (order >= 0) {
2240 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2243 if (rinfo->dir_entries)
2247 if (!rinfo->dir_entries)
2250 num_entries = (PAGE_SIZE << order) / size;
2251 num_entries = min(num_entries, opt->max_readdir);
2253 rinfo->dir_buf_size = PAGE_SIZE << order;
2254 req->r_num_caps = num_entries + 1;
2255 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2256 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2261 * Create an mds request.
2263 struct ceph_mds_request *
2264 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2266 struct ceph_mds_request *req;
2268 req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2270 return ERR_PTR(-ENOMEM);
2272 mutex_init(&req->r_fill_mutex);
2274 req->r_started = jiffies;
2275 req->r_start_latency = ktime_get();
2276 req->r_resend_mds = -1;
2277 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2278 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2280 kref_init(&req->r_kref);
2281 RB_CLEAR_NODE(&req->r_node);
2282 INIT_LIST_HEAD(&req->r_wait);
2283 init_completion(&req->r_completion);
2284 init_completion(&req->r_safe_completion);
2285 INIT_LIST_HEAD(&req->r_unsafe_item);
2287 ktime_get_coarse_real_ts64(&req->r_stamp);
2290 req->r_direct_mode = mode;
2295 * return oldest (lowest) request, tid in request tree, 0 if none.
2297 * called under mdsc->mutex.
2299 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2301 if (RB_EMPTY_ROOT(&mdsc->request_tree))
2303 return rb_entry(rb_first(&mdsc->request_tree),
2304 struct ceph_mds_request, r_node);
2307 static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2309 return mdsc->oldest_tid;
2313 * Build a dentry's path. Allocate on heap; caller must kfree. Based
2314 * on build_path_from_dentry in fs/cifs/dir.c.
2316 * If @stop_on_nosnap, generate path relative to the first non-snapped
2319 * Encode hidden .snap dirs as a double /, i.e.
2320 * foo/.snap/bar -> foo//bar
2322 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2325 struct dentry *temp;
2332 return ERR_PTR(-EINVAL);
2336 return ERR_PTR(-ENOMEM);
2341 seq = read_seqbegin(&rename_lock);
2345 struct inode *inode;
2347 spin_lock(&temp->d_lock);
2348 inode = d_inode(temp);
2349 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2350 dout("build_path path+%d: %p SNAPDIR\n",
2352 } else if (stop_on_nosnap && inode && dentry != temp &&
2353 ceph_snap(inode) == CEPH_NOSNAP) {
2354 spin_unlock(&temp->d_lock);
2355 pos++; /* get rid of any prepended '/' */
2358 pos -= temp->d_name.len;
2360 spin_unlock(&temp->d_lock);
2363 memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2365 spin_unlock(&temp->d_lock);
2366 temp = READ_ONCE(temp->d_parent);
2368 /* Are we at the root? */
2372 /* Are we out of buffer? */
2378 base = ceph_ino(d_inode(temp));
2381 if (read_seqretry(&rename_lock, seq))
2386 * A rename didn't occur, but somehow we didn't end up where
2387 * we thought we would. Throw a warning and try again.
2389 pr_warn("build_path did not end path lookup where "
2390 "expected, pos is %d\n", pos);
2395 *plen = PATH_MAX - 1 - pos;
2396 dout("build_path on %p %d built %llx '%.*s'\n",
2397 dentry, d_count(dentry), base, *plen, path + pos);
2401 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2402 const char **ppath, int *ppathlen, u64 *pino,
2403 bool *pfreepath, bool parent_locked)
2409 dir = d_inode_rcu(dentry->d_parent);
2410 if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2411 *pino = ceph_ino(dir);
2413 *ppath = dentry->d_name.name;
2414 *ppathlen = dentry->d_name.len;
2418 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2420 return PTR_ERR(path);
2426 static int build_inode_path(struct inode *inode,
2427 const char **ppath, int *ppathlen, u64 *pino,
2430 struct dentry *dentry;
2433 if (ceph_snap(inode) == CEPH_NOSNAP) {
2434 *pino = ceph_ino(inode);
2438 dentry = d_find_alias(inode);
2439 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2442 return PTR_ERR(path);
2449 * request arguments may be specified via an inode *, a dentry *, or
2450 * an explicit ino+path.
2452 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2453 struct inode *rdiri, const char *rpath,
2454 u64 rino, const char **ppath, int *pathlen,
2455 u64 *ino, bool *freepath, bool parent_locked)
2460 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2461 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2463 } else if (rdentry) {
2464 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2465 freepath, parent_locked);
2466 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2468 } else if (rpath || rino) {
2471 *pathlen = rpath ? strlen(rpath) : 0;
2472 dout(" path %.*s\n", *pathlen, rpath);
2479 * called under mdsc->mutex
2481 static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
2482 struct ceph_mds_request *req,
2483 bool drop_cap_releases)
2485 int mds = session->s_mds;
2486 struct ceph_mds_client *mdsc = session->s_mdsc;
2487 struct ceph_msg *msg;
2488 struct ceph_mds_request_head_old *head;
2489 const char *path1 = NULL;
2490 const char *path2 = NULL;
2491 u64 ino1 = 0, ino2 = 0;
2492 int pathlen1 = 0, pathlen2 = 0;
2493 bool freepath1 = false, freepath2 = false;
2498 bool legacy = !(session->s_con.peer_features & CEPH_FEATURE_FS_BTIME);
2500 ret = set_request_path_attr(req->r_inode, req->r_dentry,
2501 req->r_parent, req->r_path1, req->r_ino1.ino,
2502 &path1, &pathlen1, &ino1, &freepath1,
2503 test_bit(CEPH_MDS_R_PARENT_LOCKED,
2504 &req->r_req_flags));
2510 /* If r_old_dentry is set, then assume that its parent is locked */
2511 ret = set_request_path_attr(NULL, req->r_old_dentry,
2512 req->r_old_dentry_dir,
2513 req->r_path2, req->r_ino2.ino,
2514 &path2, &pathlen2, &ino2, &freepath2, true);
2522 len = sizeof(*head);
2524 /* New style: add gid_list and any later fields */
2525 len = sizeof(struct ceph_mds_request_head) + sizeof(u32) +
2526 (sizeof(u64) * req->r_cred->group_info->ngroups);
2529 len += pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2530 sizeof(struct ceph_timespec);
2532 /* calculate (max) length for cap releases */
2533 len += sizeof(struct ceph_mds_request_release) *
2534 (!!req->r_inode_drop + !!req->r_dentry_drop +
2535 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2537 if (req->r_dentry_drop)
2539 if (req->r_old_dentry_drop)
2542 msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2544 msg = ERR_PTR(-ENOMEM);
2548 msg->hdr.tid = cpu_to_le64(req->r_tid);
2551 * The old ceph_mds_request_header didn't contain a version field, and
2552 * one was added when we moved the message version from 3->4.
2555 msg->hdr.version = cpu_to_le16(3);
2556 head = msg->front.iov_base;
2557 p = msg->front.iov_base + sizeof(*head);
2559 struct ceph_mds_request_head *new_head = msg->front.iov_base;
2561 msg->hdr.version = cpu_to_le16(4);
2562 new_head->version = cpu_to_le16(CEPH_MDS_REQUEST_HEAD_VERSION);
2563 head = (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2564 p = msg->front.iov_base + sizeof(*new_head);
2567 end = msg->front.iov_base + msg->front.iov_len;
2569 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2570 head->op = cpu_to_le32(req->r_op);
2571 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns,
2572 req->r_cred->fsuid));
2573 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns,
2574 req->r_cred->fsgid));
2575 head->ino = cpu_to_le64(req->r_deleg_ino);
2576 head->args = req->r_args;
2578 ceph_encode_filepath(&p, end, ino1, path1);
2579 ceph_encode_filepath(&p, end, ino2, path2);
2581 /* make note of release offset, in case we need to replay */
2582 req->r_request_release_offset = p - msg->front.iov_base;
2586 if (req->r_inode_drop)
2587 releases += ceph_encode_inode_release(&p,
2588 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2589 mds, req->r_inode_drop, req->r_inode_unless,
2590 req->r_op == CEPH_MDS_OP_READDIR);
2591 if (req->r_dentry_drop)
2592 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2593 req->r_parent, mds, req->r_dentry_drop,
2594 req->r_dentry_unless);
2595 if (req->r_old_dentry_drop)
2596 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2597 req->r_old_dentry_dir, mds,
2598 req->r_old_dentry_drop,
2599 req->r_old_dentry_unless);
2600 if (req->r_old_inode_drop)
2601 releases += ceph_encode_inode_release(&p,
2602 d_inode(req->r_old_dentry),
2603 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2605 if (drop_cap_releases) {
2607 p = msg->front.iov_base + req->r_request_release_offset;
2610 head->num_releases = cpu_to_le16(releases);
2614 struct ceph_timespec ts;
2615 ceph_encode_timespec64(&ts, &req->r_stamp);
2616 ceph_encode_copy(&p, &ts, sizeof(ts));
2621 ceph_encode_32(&p, req->r_cred->group_info->ngroups);
2622 for (i = 0; i < req->r_cred->group_info->ngroups; i++)
2623 ceph_encode_64(&p, from_kgid(&init_user_ns,
2624 req->r_cred->group_info->gid[i]));
2627 if (WARN_ON_ONCE(p > end)) {
2629 msg = ERR_PTR(-ERANGE);
2633 msg->front.iov_len = p - msg->front.iov_base;
2634 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2636 if (req->r_pagelist) {
2637 struct ceph_pagelist *pagelist = req->r_pagelist;
2638 ceph_msg_data_add_pagelist(msg, pagelist);
2639 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2641 msg->hdr.data_len = 0;
2644 msg->hdr.data_off = cpu_to_le16(0);
2648 ceph_mdsc_free_path((char *)path2, pathlen2);
2651 ceph_mdsc_free_path((char *)path1, pathlen1);
2657 * called under mdsc->mutex if error, under no mutex if
2660 static void complete_request(struct ceph_mds_client *mdsc,
2661 struct ceph_mds_request *req)
2663 req->r_end_latency = ktime_get();
2665 if (req->r_callback)
2666 req->r_callback(mdsc, req);
2667 complete_all(&req->r_completion);
2670 static struct ceph_mds_request_head_old *
2671 find_old_request_head(void *p, u64 features)
2673 bool legacy = !(features & CEPH_FEATURE_FS_BTIME);
2674 struct ceph_mds_request_head *new_head;
2677 return (struct ceph_mds_request_head_old *)p;
2678 new_head = (struct ceph_mds_request_head *)p;
2679 return (struct ceph_mds_request_head_old *)&new_head->oldest_client_tid;
2683 * called under mdsc->mutex
2685 static int __prepare_send_request(struct ceph_mds_session *session,
2686 struct ceph_mds_request *req,
2687 bool drop_cap_releases)
2689 int mds = session->s_mds;
2690 struct ceph_mds_client *mdsc = session->s_mdsc;
2691 struct ceph_mds_request_head_old *rhead;
2692 struct ceph_msg *msg;
2697 struct ceph_cap *cap =
2698 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2701 req->r_sent_on_mseq = cap->mseq;
2703 req->r_sent_on_mseq = -1;
2705 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2706 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2708 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2712 * Replay. Do not regenerate message (and rebuild
2713 * paths, etc.); just use the original message.
2714 * Rebuilding paths will break for renames because
2715 * d_move mangles the src name.
2717 msg = req->r_request;
2718 rhead = find_old_request_head(msg->front.iov_base,
2719 session->s_con.peer_features);
2721 flags = le32_to_cpu(rhead->flags);
2722 flags |= CEPH_MDS_FLAG_REPLAY;
2723 rhead->flags = cpu_to_le32(flags);
2725 if (req->r_target_inode)
2726 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2728 rhead->num_retry = req->r_attempts - 1;
2730 /* remove cap/dentry releases from message */
2731 rhead->num_releases = 0;
2734 p = msg->front.iov_base + req->r_request_release_offset;
2736 struct ceph_timespec ts;
2737 ceph_encode_timespec64(&ts, &req->r_stamp);
2738 ceph_encode_copy(&p, &ts, sizeof(ts));
2741 msg->front.iov_len = p - msg->front.iov_base;
2742 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2746 if (req->r_request) {
2747 ceph_msg_put(req->r_request);
2748 req->r_request = NULL;
2750 msg = create_request_message(session, req, drop_cap_releases);
2752 req->r_err = PTR_ERR(msg);
2753 return PTR_ERR(msg);
2755 req->r_request = msg;
2757 rhead = find_old_request_head(msg->front.iov_base,
2758 session->s_con.peer_features);
2759 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2760 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2761 flags |= CEPH_MDS_FLAG_REPLAY;
2762 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2763 flags |= CEPH_MDS_FLAG_ASYNC;
2765 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2766 rhead->flags = cpu_to_le32(flags);
2767 rhead->num_fwd = req->r_num_fwd;
2768 rhead->num_retry = req->r_attempts - 1;
2770 dout(" r_parent = %p\n", req->r_parent);
2775 * called under mdsc->mutex
2777 static int __send_request(struct ceph_mds_session *session,
2778 struct ceph_mds_request *req,
2779 bool drop_cap_releases)
2783 err = __prepare_send_request(session, req, drop_cap_releases);
2785 ceph_msg_get(req->r_request);
2786 ceph_con_send(&session->s_con, req->r_request);
2793 * send request, or put it on the appropriate wait list.
2795 static void __do_request(struct ceph_mds_client *mdsc,
2796 struct ceph_mds_request *req)
2798 struct ceph_mds_session *session = NULL;
2803 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2804 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2805 __unregister_request(mdsc, req);
2809 if (req->r_timeout &&
2810 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2811 dout("do_request timed out\n");
2815 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2816 dout("do_request forced umount\n");
2820 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2821 if (mdsc->mdsmap_err) {
2822 err = mdsc->mdsmap_err;
2823 dout("do_request mdsmap err %d\n", err);
2826 if (mdsc->mdsmap->m_epoch == 0) {
2827 dout("do_request no mdsmap, waiting for map\n");
2828 list_add(&req->r_wait, &mdsc->waiting_for_map);
2831 if (!(mdsc->fsc->mount_options->flags &
2832 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2833 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2834 err = -EHOSTUNREACH;
2839 put_request_session(req);
2841 mds = __choose_mds(mdsc, req, &random);
2843 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2844 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2848 dout("do_request no mds or not active, waiting for map\n");
2849 list_add(&req->r_wait, &mdsc->waiting_for_map);
2853 /* get, open session */
2854 session = __ceph_lookup_mds_session(mdsc, mds);
2856 session = register_session(mdsc, mds);
2857 if (IS_ERR(session)) {
2858 err = PTR_ERR(session);
2862 req->r_session = ceph_get_mds_session(session);
2864 dout("do_request mds%d session %p state %s\n", mds, session,
2865 ceph_session_state_name(session->s_state));
2866 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2867 session->s_state != CEPH_MDS_SESSION_HUNG) {
2869 * We cannot queue async requests since the caps and delegated
2870 * inodes are bound to the session. Just return -EJUKEBOX and
2871 * let the caller retry a sync request in that case.
2873 if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2879 * If the session has been REJECTED, then return a hard error,
2880 * unless it's a CLEANRECOVER mount, in which case we'll queue
2881 * it to the mdsc queue.
2883 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2884 if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
2885 list_add(&req->r_wait, &mdsc->waiting_for_map);
2891 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2892 session->s_state == CEPH_MDS_SESSION_CLOSING) {
2893 err = __open_session(mdsc, session);
2896 /* retry the same mds later */
2898 req->r_resend_mds = mds;
2900 list_add(&req->r_wait, &session->s_waiting);
2905 req->r_resend_mds = -1; /* forget any previous mds hint */
2907 if (req->r_request_started == 0) /* note request start time */
2908 req->r_request_started = jiffies;
2910 err = __send_request(session, req, false);
2913 ceph_put_mds_session(session);
2916 dout("__do_request early error %d\n", err);
2918 complete_request(mdsc, req);
2919 __unregister_request(mdsc, req);
2925 * called under mdsc->mutex
2927 static void __wake_requests(struct ceph_mds_client *mdsc,
2928 struct list_head *head)
2930 struct ceph_mds_request *req;
2931 LIST_HEAD(tmp_list);
2933 list_splice_init(head, &tmp_list);
2935 while (!list_empty(&tmp_list)) {
2936 req = list_entry(tmp_list.next,
2937 struct ceph_mds_request, r_wait);
2938 list_del_init(&req->r_wait);
2939 dout(" wake request %p tid %llu\n", req, req->r_tid);
2940 __do_request(mdsc, req);
2945 * Wake up threads with requests pending for @mds, so that they can
2946 * resubmit their requests to a possibly different mds.
2948 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2950 struct ceph_mds_request *req;
2951 struct rb_node *p = rb_first(&mdsc->request_tree);
2953 dout("kick_requests mds%d\n", mds);
2955 req = rb_entry(p, struct ceph_mds_request, r_node);
2957 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2959 if (req->r_attempts > 0)
2960 continue; /* only new requests */
2961 if (req->r_session &&
2962 req->r_session->s_mds == mds) {
2963 dout(" kicking tid %llu\n", req->r_tid);
2964 list_del_init(&req->r_wait);
2965 __do_request(mdsc, req);
2970 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
2971 struct ceph_mds_request *req)
2975 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2977 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2978 if (req->r_parent) {
2979 struct ceph_inode_info *ci = ceph_inode(req->r_parent);
2980 int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
2981 CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
2982 spin_lock(&ci->i_ceph_lock);
2983 ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
2984 __ceph_touch_fmode(ci, mdsc, fmode);
2985 spin_unlock(&ci->i_ceph_lock);
2986 ihold(req->r_parent);
2988 if (req->r_old_dentry_dir)
2989 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2993 err = ceph_wait_on_async_create(req->r_inode);
2995 dout("%s: wait for async create returned: %d\n",
3001 if (!err && req->r_old_inode) {
3002 err = ceph_wait_on_async_create(req->r_old_inode);
3004 dout("%s: wait for async create returned: %d\n",
3010 dout("submit_request on %p for inode %p\n", req, dir);
3011 mutex_lock(&mdsc->mutex);
3012 __register_request(mdsc, req, dir);
3013 __do_request(mdsc, req);
3015 mutex_unlock(&mdsc->mutex);
3019 static int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
3020 struct ceph_mds_request *req)
3025 dout("do_request waiting\n");
3026 if (!req->r_timeout && req->r_wait_for_completion) {
3027 err = req->r_wait_for_completion(mdsc, req);
3029 long timeleft = wait_for_completion_killable_timeout(
3031 ceph_timeout_jiffies(req->r_timeout));
3035 err = -ETIMEDOUT; /* timed out */
3037 err = timeleft; /* killed */
3039 dout("do_request waited, got %d\n", err);
3040 mutex_lock(&mdsc->mutex);
3042 /* only abort if we didn't race with a real reply */
3043 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
3044 err = le32_to_cpu(req->r_reply_info.head->result);
3045 } else if (err < 0) {
3046 dout("aborted request %lld with %d\n", req->r_tid, err);
3049 * ensure we aren't running concurrently with
3050 * ceph_fill_trace or ceph_readdir_prepopulate, which
3051 * rely on locks (dir mutex) held by our caller.
3053 mutex_lock(&req->r_fill_mutex);
3055 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
3056 mutex_unlock(&req->r_fill_mutex);
3058 if (req->r_parent &&
3059 (req->r_op & CEPH_MDS_OP_WRITE))
3060 ceph_invalidate_dir_request(req);
3065 mutex_unlock(&mdsc->mutex);
3070 * Synchrously perform an mds request. Take care of all of the
3071 * session setup, forwarding, retry details.
3073 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
3075 struct ceph_mds_request *req)
3079 dout("do_request on %p\n", req);
3082 err = ceph_mdsc_submit_request(mdsc, dir, req);
3084 err = ceph_mdsc_wait_request(mdsc, req);
3085 dout("do_request %p done, result %d\n", req, err);
3090 * Invalidate dir's completeness, dentry lease state on an aborted MDS
3091 * namespace request.
3093 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
3095 struct inode *dir = req->r_parent;
3096 struct inode *old_dir = req->r_old_dentry_dir;
3098 dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
3100 ceph_dir_clear_complete(dir);
3102 ceph_dir_clear_complete(old_dir);
3104 ceph_invalidate_dentry_lease(req->r_dentry);
3105 if (req->r_old_dentry)
3106 ceph_invalidate_dentry_lease(req->r_old_dentry);
3112 * We take the session mutex and parse and process the reply immediately.
3113 * This preserves the logical ordering of replies, capabilities, etc., sent
3114 * by the MDS as they are applied to our local cache.
3116 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
3118 struct ceph_mds_client *mdsc = session->s_mdsc;
3119 struct ceph_mds_request *req;
3120 struct ceph_mds_reply_head *head = msg->front.iov_base;
3121 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
3122 struct ceph_snap_realm *realm;
3125 int mds = session->s_mds;
3127 if (msg->front.iov_len < sizeof(*head)) {
3128 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
3133 /* get request, session */
3134 tid = le64_to_cpu(msg->hdr.tid);
3135 mutex_lock(&mdsc->mutex);
3136 req = lookup_get_request(mdsc, tid);
3138 dout("handle_reply on unknown tid %llu\n", tid);
3139 mutex_unlock(&mdsc->mutex);
3142 dout("handle_reply %p\n", req);
3144 /* correct session? */
3145 if (req->r_session != session) {
3146 pr_err("mdsc_handle_reply got %llu on session mds%d"
3147 " not mds%d\n", tid, session->s_mds,
3148 req->r_session ? req->r_session->s_mds : -1);
3149 mutex_unlock(&mdsc->mutex);
3154 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3155 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3156 pr_warn("got a dup %s reply on %llu from mds%d\n",
3157 head->safe ? "safe" : "unsafe", tid, mds);
3158 mutex_unlock(&mdsc->mutex);
3161 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3162 pr_warn("got unsafe after safe on %llu from mds%d\n",
3164 mutex_unlock(&mdsc->mutex);
3168 result = le32_to_cpu(head->result);
3172 * if we're not talking to the authority, send to them
3173 * if the authority has changed while we weren't looking,
3174 * send to new authority
3175 * Otherwise we just have to return an ESTALE
3177 if (result == -ESTALE) {
3178 dout("got ESTALE on request %llu\n", req->r_tid);
3179 req->r_resend_mds = -1;
3180 if (req->r_direct_mode != USE_AUTH_MDS) {
3181 dout("not using auth, setting for that now\n");
3182 req->r_direct_mode = USE_AUTH_MDS;
3183 __do_request(mdsc, req);
3184 mutex_unlock(&mdsc->mutex);
3187 int mds = __choose_mds(mdsc, req, NULL);
3188 if (mds >= 0 && mds != req->r_session->s_mds) {
3189 dout("but auth changed, so resending\n");
3190 __do_request(mdsc, req);
3191 mutex_unlock(&mdsc->mutex);
3195 dout("have to return ESTALE on request %llu\n", req->r_tid);
3200 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3201 __unregister_request(mdsc, req);
3203 /* last request during umount? */
3204 if (mdsc->stopping && !__get_oldest_req(mdsc))
3205 complete_all(&mdsc->safe_umount_waiters);
3207 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3209 * We already handled the unsafe response, now do the
3210 * cleanup. No need to examine the response; the MDS
3211 * doesn't include any result info in the safe
3212 * response. And even if it did, there is nothing
3213 * useful we could do with a revised return value.
3215 dout("got safe reply %llu, mds%d\n", tid, mds);
3217 mutex_unlock(&mdsc->mutex);
3221 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3222 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3225 dout("handle_reply tid %lld result %d\n", tid, result);
3226 rinfo = &req->r_reply_info;
3227 if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3228 err = parse_reply_info(session, msg, rinfo, (u64)-1);
3230 err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3231 mutex_unlock(&mdsc->mutex);
3233 /* Must find target inode outside of mutexes to avoid deadlocks */
3234 if ((err >= 0) && rinfo->head->is_target) {
3236 struct ceph_vino tvino = {
3237 .ino = le64_to_cpu(rinfo->targeti.in->ino),
3238 .snap = le64_to_cpu(rinfo->targeti.in->snapid)
3241 in = ceph_get_inode(mdsc->fsc->sb, tvino);
3244 mutex_lock(&session->s_mutex);
3247 req->r_target_inode = in;
3250 mutex_lock(&session->s_mutex);
3252 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3259 if (rinfo->snapblob_len) {
3260 down_write(&mdsc->snap_rwsem);
3261 ceph_update_snap_trace(mdsc, rinfo->snapblob,
3262 rinfo->snapblob + rinfo->snapblob_len,
3263 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3265 downgrade_write(&mdsc->snap_rwsem);
3267 down_read(&mdsc->snap_rwsem);
3270 /* insert trace into our cache */
3271 mutex_lock(&req->r_fill_mutex);
3272 current->journal_info = req;
3273 err = ceph_fill_trace(mdsc->fsc->sb, req);
3275 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3276 req->r_op == CEPH_MDS_OP_LSSNAP))
3277 ceph_readdir_prepopulate(req, req->r_session);
3279 current->journal_info = NULL;
3280 mutex_unlock(&req->r_fill_mutex);
3282 up_read(&mdsc->snap_rwsem);
3284 ceph_put_snap_realm(mdsc, realm);
3287 if (req->r_target_inode &&
3288 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3289 struct ceph_inode_info *ci =
3290 ceph_inode(req->r_target_inode);
3291 spin_lock(&ci->i_unsafe_lock);
3292 list_add_tail(&req->r_unsafe_target_item,
3293 &ci->i_unsafe_iops);
3294 spin_unlock(&ci->i_unsafe_lock);
3297 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3300 mutex_lock(&mdsc->mutex);
3301 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3305 req->r_reply = ceph_msg_get(msg);
3306 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3309 dout("reply arrived after request %lld was aborted\n", tid);
3311 mutex_unlock(&mdsc->mutex);
3313 mutex_unlock(&session->s_mutex);
3315 /* kick calling process */
3316 complete_request(mdsc, req);
3318 ceph_update_metadata_latency(&mdsc->metric, req->r_start_latency,
3319 req->r_end_latency, err);
3321 ceph_mdsc_put_request(req);
3328 * handle mds notification that our request has been forwarded.
3330 static void handle_forward(struct ceph_mds_client *mdsc,
3331 struct ceph_mds_session *session,
3332 struct ceph_msg *msg)
3334 struct ceph_mds_request *req;
3335 u64 tid = le64_to_cpu(msg->hdr.tid);
3339 void *p = msg->front.iov_base;
3340 void *end = p + msg->front.iov_len;
3342 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3343 next_mds = ceph_decode_32(&p);
3344 fwd_seq = ceph_decode_32(&p);
3346 mutex_lock(&mdsc->mutex);
3347 req = lookup_get_request(mdsc, tid);
3349 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3350 goto out; /* dup reply? */
3353 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3354 dout("forward tid %llu aborted, unregistering\n", tid);
3355 __unregister_request(mdsc, req);
3356 } else if (fwd_seq <= req->r_num_fwd) {
3357 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3358 tid, next_mds, req->r_num_fwd, fwd_seq);
3360 /* resend. forward race not possible; mds would drop */
3361 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3363 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3364 req->r_attempts = 0;
3365 req->r_num_fwd = fwd_seq;
3366 req->r_resend_mds = next_mds;
3367 put_request_session(req);
3368 __do_request(mdsc, req);
3370 ceph_mdsc_put_request(req);
3372 mutex_unlock(&mdsc->mutex);
3376 pr_err("mdsc_handle_forward decode error err=%d\n", err);
3379 static int __decode_session_metadata(void **p, void *end,
3382 /* map<string,string> */
3385 ceph_decode_32_safe(p, end, n, bad);
3388 ceph_decode_32_safe(p, end, len, bad);
3389 ceph_decode_need(p, end, len, bad);
3390 err_str = !strncmp(*p, "error_string", len);
3392 ceph_decode_32_safe(p, end, len, bad);
3393 ceph_decode_need(p, end, len, bad);
3395 * Match "blocklisted (blacklisted)" from newer MDSes,
3396 * or "blacklisted" from older MDSes.
3398 if (err_str && strnstr(*p, "blacklisted", len))
3399 *blocklisted = true;
3408 * handle a mds session control message
3410 static void handle_session(struct ceph_mds_session *session,
3411 struct ceph_msg *msg)
3413 struct ceph_mds_client *mdsc = session->s_mdsc;
3414 int mds = session->s_mds;
3415 int msg_version = le16_to_cpu(msg->hdr.version);
3416 void *p = msg->front.iov_base;
3417 void *end = p + msg->front.iov_len;
3418 struct ceph_mds_session_head *h;
3420 u64 seq, features = 0;
3422 bool blocklisted = false;
3425 ceph_decode_need(&p, end, sizeof(*h), bad);
3429 op = le32_to_cpu(h->op);
3430 seq = le64_to_cpu(h->seq);
3432 if (msg_version >= 3) {
3434 /* version >= 2, metadata */
3435 if (__decode_session_metadata(&p, end, &blocklisted) < 0)
3437 /* version >= 3, feature bits */
3438 ceph_decode_32_safe(&p, end, len, bad);
3440 ceph_decode_64_safe(&p, end, features, bad);
3441 p += len - sizeof(features);
3445 mutex_lock(&mdsc->mutex);
3446 if (op == CEPH_SESSION_CLOSE) {
3447 ceph_get_mds_session(session);
3448 __unregister_session(mdsc, session);
3450 /* FIXME: this ttl calculation is generous */
3451 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3452 mutex_unlock(&mdsc->mutex);
3454 mutex_lock(&session->s_mutex);
3456 dout("handle_session mds%d %s %p state %s seq %llu\n",
3457 mds, ceph_session_op_name(op), session,
3458 ceph_session_state_name(session->s_state), seq);
3460 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3461 session->s_state = CEPH_MDS_SESSION_OPEN;
3462 pr_info("mds%d came back\n", session->s_mds);
3466 case CEPH_SESSION_OPEN:
3467 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3468 pr_info("mds%d reconnect success\n", session->s_mds);
3469 session->s_state = CEPH_MDS_SESSION_OPEN;
3470 session->s_features = features;
3471 renewed_caps(mdsc, session, 0);
3472 if (test_bit(CEPHFS_FEATURE_METRIC_COLLECT, &session->s_features))
3473 metric_schedule_delayed(&mdsc->metric);
3476 __close_session(mdsc, session);
3479 case CEPH_SESSION_RENEWCAPS:
3480 if (session->s_renew_seq == seq)
3481 renewed_caps(mdsc, session, 1);
3484 case CEPH_SESSION_CLOSE:
3485 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3486 pr_info("mds%d reconnect denied\n", session->s_mds);
3487 session->s_state = CEPH_MDS_SESSION_CLOSED;
3488 cleanup_session_requests(mdsc, session);
3489 remove_session_caps(session);
3490 wake = 2; /* for good measure */
3491 wake_up_all(&mdsc->session_close_wq);
3494 case CEPH_SESSION_STALE:
3495 pr_info("mds%d caps went stale, renewing\n",
3497 spin_lock(&session->s_gen_ttl_lock);
3498 session->s_cap_gen++;
3499 session->s_cap_ttl = jiffies - 1;
3500 spin_unlock(&session->s_gen_ttl_lock);
3501 send_renew_caps(mdsc, session);
3504 case CEPH_SESSION_RECALL_STATE:
3505 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3508 case CEPH_SESSION_FLUSHMSG:
3509 send_flushmsg_ack(mdsc, session, seq);
3512 case CEPH_SESSION_FORCE_RO:
3513 dout("force_session_readonly %p\n", session);
3514 spin_lock(&session->s_cap_lock);
3515 session->s_readonly = true;
3516 spin_unlock(&session->s_cap_lock);
3517 wake_up_session_caps(session, FORCE_RO);
3520 case CEPH_SESSION_REJECT:
3521 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3522 pr_info("mds%d rejected session\n", session->s_mds);
3523 session->s_state = CEPH_MDS_SESSION_REJECTED;
3524 cleanup_session_requests(mdsc, session);
3525 remove_session_caps(session);
3527 mdsc->fsc->blocklisted = true;
3528 wake = 2; /* for good measure */
3532 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3536 mutex_unlock(&session->s_mutex);
3538 mutex_lock(&mdsc->mutex);
3539 __wake_requests(mdsc, &session->s_waiting);
3541 kick_requests(mdsc, mds);
3542 mutex_unlock(&mdsc->mutex);
3544 if (op == CEPH_SESSION_CLOSE)
3545 ceph_put_mds_session(session);
3549 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3550 (int)msg->front.iov_len);
3555 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3559 dcaps = xchg(&req->r_dir_caps, 0);
3561 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3562 ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3566 void ceph_mdsc_release_dir_caps_no_check(struct ceph_mds_request *req)
3570 dcaps = xchg(&req->r_dir_caps, 0);
3572 dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3573 ceph_put_cap_refs_no_check_caps(ceph_inode(req->r_parent),
3579 * called under session->mutex.
3581 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3582 struct ceph_mds_session *session)
3584 struct ceph_mds_request *req, *nreq;
3587 dout("replay_unsafe_requests mds%d\n", session->s_mds);
3589 mutex_lock(&mdsc->mutex);
3590 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3591 __send_request(session, req, true);
3594 * also re-send old requests when MDS enters reconnect stage. So that MDS
3595 * can process completed request in clientreplay stage.
3597 p = rb_first(&mdsc->request_tree);
3599 req = rb_entry(p, struct ceph_mds_request, r_node);
3601 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3603 if (req->r_attempts == 0)
3604 continue; /* only old requests */
3605 if (!req->r_session)
3607 if (req->r_session->s_mds != session->s_mds)
3610 ceph_mdsc_release_dir_caps_no_check(req);
3612 __send_request(session, req, true);
3614 mutex_unlock(&mdsc->mutex);
3617 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3619 struct ceph_msg *reply;
3620 struct ceph_pagelist *_pagelist;
3625 if (!recon_state->allow_multi)
3628 /* can't handle message that contains both caps and realm */
3629 BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3631 /* pre-allocate new pagelist */
3632 _pagelist = ceph_pagelist_alloc(GFP_NOFS);
3636 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3640 /* placeholder for nr_caps */
3641 err = ceph_pagelist_encode_32(_pagelist, 0);
3645 if (recon_state->nr_caps) {
3646 /* currently encoding caps */
3647 err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3651 /* placeholder for nr_realms (currently encoding relams) */
3652 err = ceph_pagelist_encode_32(_pagelist, 0);
3657 err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3661 page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3662 addr = kmap_atomic(page);
3663 if (recon_state->nr_caps) {
3664 /* currently encoding caps */
3665 *addr = cpu_to_le32(recon_state->nr_caps);
3667 /* currently encoding relams */
3668 *(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3670 kunmap_atomic(addr);
3672 reply->hdr.version = cpu_to_le16(5);
3673 reply->hdr.compat_version = cpu_to_le16(4);
3675 reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3676 ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3678 ceph_con_send(&recon_state->session->s_con, reply);
3679 ceph_pagelist_release(recon_state->pagelist);
3681 recon_state->pagelist = _pagelist;
3682 recon_state->nr_caps = 0;
3683 recon_state->nr_realms = 0;
3684 recon_state->msg_version = 5;
3687 ceph_msg_put(reply);
3689 ceph_pagelist_release(_pagelist);
3693 static struct dentry* d_find_primary(struct inode *inode)
3695 struct dentry *alias, *dn = NULL;
3697 if (hlist_empty(&inode->i_dentry))
3700 spin_lock(&inode->i_lock);
3701 if (hlist_empty(&inode->i_dentry))
3704 if (S_ISDIR(inode->i_mode)) {
3705 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
3706 if (!IS_ROOT(alias))
3711 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
3712 spin_lock(&alias->d_lock);
3713 if (!d_unhashed(alias) &&
3714 (ceph_dentry(alias)->flags & CEPH_DENTRY_PRIMARY_LINK)) {
3715 dn = dget_dlock(alias);
3717 spin_unlock(&alias->d_lock);
3722 spin_unlock(&inode->i_lock);
3727 * Encode information about a cap for a reconnect with the MDS.
3729 static int reconnect_caps_cb(struct inode *inode, struct ceph_cap *cap,
3733 struct ceph_mds_cap_reconnect v2;
3734 struct ceph_mds_cap_reconnect_v1 v1;
3736 struct ceph_inode_info *ci = cap->ci;
3737 struct ceph_reconnect_state *recon_state = arg;
3738 struct ceph_pagelist *pagelist = recon_state->pagelist;
3739 struct dentry *dentry;
3745 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3746 inode, ceph_vinop(inode), cap, cap->cap_id,
3747 ceph_cap_string(cap->issued));
3749 dentry = d_find_primary(inode);
3751 /* set pathbase to parent dir when msg_version >= 2 */
3752 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase,
3753 recon_state->msg_version >= 2);
3756 err = PTR_ERR(path);
3765 spin_lock(&ci->i_ceph_lock);
3766 cap->seq = 0; /* reset cap seq */
3767 cap->issue_seq = 0; /* and issue_seq */
3768 cap->mseq = 0; /* and migrate_seq */
3769 cap->cap_gen = cap->session->s_cap_gen;
3771 /* These are lost when the session goes away */
3772 if (S_ISDIR(inode->i_mode)) {
3773 if (cap->issued & CEPH_CAP_DIR_CREATE) {
3774 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3775 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3777 cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3780 if (recon_state->msg_version >= 2) {
3781 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3782 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3783 rec.v2.issued = cpu_to_le32(cap->issued);
3784 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3785 rec.v2.pathbase = cpu_to_le64(pathbase);
3786 rec.v2.flock_len = (__force __le32)
3787 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3789 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3790 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3791 rec.v1.issued = cpu_to_le32(cap->issued);
3792 rec.v1.size = cpu_to_le64(inode->i_size);
3793 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3794 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3795 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3796 rec.v1.pathbase = cpu_to_le64(pathbase);
3799 if (list_empty(&ci->i_cap_snaps)) {
3800 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3802 struct ceph_cap_snap *capsnap =
3803 list_first_entry(&ci->i_cap_snaps,
3804 struct ceph_cap_snap, ci_item);
3805 snap_follows = capsnap->follows;
3807 spin_unlock(&ci->i_ceph_lock);
3809 if (recon_state->msg_version >= 2) {
3810 int num_fcntl_locks, num_flock_locks;
3811 struct ceph_filelock *flocks = NULL;
3812 size_t struct_len, total_len = sizeof(u64);
3816 if (rec.v2.flock_len) {
3817 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3819 num_fcntl_locks = 0;
3820 num_flock_locks = 0;
3822 if (num_fcntl_locks + num_flock_locks > 0) {
3823 flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3824 sizeof(struct ceph_filelock),
3830 err = ceph_encode_locks_to_buffer(inode, flocks,
3845 if (recon_state->msg_version >= 3) {
3846 /* version, compat_version and struct_len */
3847 total_len += 2 * sizeof(u8) + sizeof(u32);
3851 * number of encoded locks is stable, so copy to pagelist
3853 struct_len = 2 * sizeof(u32) +
3854 (num_fcntl_locks + num_flock_locks) *
3855 sizeof(struct ceph_filelock);
3856 rec.v2.flock_len = cpu_to_le32(struct_len);
3858 struct_len += sizeof(u32) + pathlen + sizeof(rec.v2);
3861 struct_len += sizeof(u64); /* snap_follows */
3863 total_len += struct_len;
3865 if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
3866 err = send_reconnect_partial(recon_state);
3868 goto out_freeflocks;
3869 pagelist = recon_state->pagelist;
3872 err = ceph_pagelist_reserve(pagelist, total_len);
3874 goto out_freeflocks;
3876 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3877 if (recon_state->msg_version >= 3) {
3878 ceph_pagelist_encode_8(pagelist, struct_v);
3879 ceph_pagelist_encode_8(pagelist, 1);
3880 ceph_pagelist_encode_32(pagelist, struct_len);
3882 ceph_pagelist_encode_string(pagelist, path, pathlen);
3883 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3884 ceph_locks_to_pagelist(flocks, pagelist,
3885 num_fcntl_locks, num_flock_locks);
3887 ceph_pagelist_encode_64(pagelist, snap_follows);
3891 err = ceph_pagelist_reserve(pagelist,
3892 sizeof(u64) + sizeof(u32) +
3893 pathlen + sizeof(rec.v1));
3897 ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3898 ceph_pagelist_encode_string(pagelist, path, pathlen);
3899 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3903 ceph_mdsc_free_path(path, pathlen);
3905 recon_state->nr_caps++;
3909 static int encode_snap_realms(struct ceph_mds_client *mdsc,
3910 struct ceph_reconnect_state *recon_state)
3913 struct ceph_pagelist *pagelist = recon_state->pagelist;
3916 if (recon_state->msg_version >= 4) {
3917 err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
3923 * snaprealms. we provide mds with the ino, seq (version), and
3924 * parent for all of our realms. If the mds has any newer info,
3927 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3928 struct ceph_snap_realm *realm =
3929 rb_entry(p, struct ceph_snap_realm, node);
3930 struct ceph_mds_snaprealm_reconnect sr_rec;
3932 if (recon_state->msg_version >= 4) {
3933 size_t need = sizeof(u8) * 2 + sizeof(u32) +
3936 if (pagelist->length + need > RECONNECT_MAX_SIZE) {
3937 err = send_reconnect_partial(recon_state);
3940 pagelist = recon_state->pagelist;
3943 err = ceph_pagelist_reserve(pagelist, need);
3947 ceph_pagelist_encode_8(pagelist, 1);
3948 ceph_pagelist_encode_8(pagelist, 1);
3949 ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
3952 dout(" adding snap realm %llx seq %lld parent %llx\n",
3953 realm->ino, realm->seq, realm->parent_ino);
3954 sr_rec.ino = cpu_to_le64(realm->ino);
3955 sr_rec.seq = cpu_to_le64(realm->seq);
3956 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3958 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3962 recon_state->nr_realms++;
3970 * If an MDS fails and recovers, clients need to reconnect in order to
3971 * reestablish shared state. This includes all caps issued through
3972 * this session _and_ the snap_realm hierarchy. Because it's not
3973 * clear which snap realms the mds cares about, we send everything we
3974 * know about.. that ensures we'll then get any new info the
3975 * recovering MDS might have.
3977 * This is a relatively heavyweight operation, but it's rare.
3979 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3980 struct ceph_mds_session *session)
3982 struct ceph_msg *reply;
3983 int mds = session->s_mds;
3985 struct ceph_reconnect_state recon_state = {
3990 pr_info("mds%d reconnect start\n", mds);
3992 recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
3993 if (!recon_state.pagelist)
3994 goto fail_nopagelist;
3996 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
4000 xa_destroy(&session->s_delegated_inos);
4002 mutex_lock(&session->s_mutex);
4003 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
4006 dout("session %p state %s\n", session,
4007 ceph_session_state_name(session->s_state));
4009 spin_lock(&session->s_gen_ttl_lock);
4010 session->s_cap_gen++;
4011 spin_unlock(&session->s_gen_ttl_lock);
4013 spin_lock(&session->s_cap_lock);
4014 /* don't know if session is readonly */
4015 session->s_readonly = 0;
4017 * notify __ceph_remove_cap() that we are composing cap reconnect.
4018 * If a cap get released before being added to the cap reconnect,
4019 * __ceph_remove_cap() should skip queuing cap release.
4021 session->s_cap_reconnect = 1;
4022 /* drop old cap expires; we're about to reestablish that state */
4023 detach_cap_releases(session, &dispose);
4024 spin_unlock(&session->s_cap_lock);
4025 dispose_cap_releases(mdsc, &dispose);
4027 /* trim unused caps to reduce MDS's cache rejoin time */
4028 if (mdsc->fsc->sb->s_root)
4029 shrink_dcache_parent(mdsc->fsc->sb->s_root);
4031 ceph_con_close(&session->s_con);
4032 ceph_con_open(&session->s_con,
4033 CEPH_ENTITY_TYPE_MDS, mds,
4034 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
4036 /* replay unsafe requests */
4037 replay_unsafe_requests(mdsc, session);
4039 ceph_early_kick_flushing_caps(mdsc, session);
4041 down_read(&mdsc->snap_rwsem);
4043 /* placeholder for nr_caps */
4044 err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
4048 if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
4049 recon_state.msg_version = 3;
4050 recon_state.allow_multi = true;
4051 } else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
4052 recon_state.msg_version = 3;
4054 recon_state.msg_version = 2;
4056 /* trsaverse this session's caps */
4057 err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
4059 spin_lock(&session->s_cap_lock);
4060 session->s_cap_reconnect = 0;
4061 spin_unlock(&session->s_cap_lock);
4066 /* check if all realms can be encoded into current message */
4067 if (mdsc->num_snap_realms) {
4069 recon_state.pagelist->length +
4070 mdsc->num_snap_realms *
4071 sizeof(struct ceph_mds_snaprealm_reconnect);
4072 if (recon_state.msg_version >= 4) {
4073 /* number of realms */
4074 total_len += sizeof(u32);
4075 /* version, compat_version and struct_len */
4076 total_len += mdsc->num_snap_realms *
4077 (2 * sizeof(u8) + sizeof(u32));
4079 if (total_len > RECONNECT_MAX_SIZE) {
4080 if (!recon_state.allow_multi) {
4084 if (recon_state.nr_caps) {
4085 err = send_reconnect_partial(&recon_state);
4089 recon_state.msg_version = 5;
4093 err = encode_snap_realms(mdsc, &recon_state);
4097 if (recon_state.msg_version >= 5) {
4098 err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
4103 if (recon_state.nr_caps || recon_state.nr_realms) {
4105 list_first_entry(&recon_state.pagelist->head,
4107 __le32 *addr = kmap_atomic(page);
4108 if (recon_state.nr_caps) {
4109 WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
4110 *addr = cpu_to_le32(recon_state.nr_caps);
4111 } else if (recon_state.msg_version >= 4) {
4112 *(addr + 1) = cpu_to_le32(recon_state.nr_realms);
4114 kunmap_atomic(addr);
4117 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
4118 if (recon_state.msg_version >= 4)
4119 reply->hdr.compat_version = cpu_to_le16(4);
4121 reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
4122 ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
4124 ceph_con_send(&session->s_con, reply);
4126 mutex_unlock(&session->s_mutex);
4128 mutex_lock(&mdsc->mutex);
4129 __wake_requests(mdsc, &session->s_waiting);
4130 mutex_unlock(&mdsc->mutex);
4132 up_read(&mdsc->snap_rwsem);
4133 ceph_pagelist_release(recon_state.pagelist);
4137 ceph_msg_put(reply);
4138 up_read(&mdsc->snap_rwsem);
4139 mutex_unlock(&session->s_mutex);
4141 ceph_pagelist_release(recon_state.pagelist);
4143 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
4149 * compare old and new mdsmaps, kicking requests
4150 * and closing out old connections as necessary
4152 * called under mdsc->mutex.
4154 static void check_new_map(struct ceph_mds_client *mdsc,
4155 struct ceph_mdsmap *newmap,
4156 struct ceph_mdsmap *oldmap)
4159 int oldstate, newstate;
4160 struct ceph_mds_session *s;
4162 dout("check_new_map new %u old %u\n",
4163 newmap->m_epoch, oldmap->m_epoch);
4165 for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4166 if (!mdsc->sessions[i])
4168 s = mdsc->sessions[i];
4169 oldstate = ceph_mdsmap_get_state(oldmap, i);
4170 newstate = ceph_mdsmap_get_state(newmap, i);
4172 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
4173 i, ceph_mds_state_name(oldstate),
4174 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
4175 ceph_mds_state_name(newstate),
4176 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
4177 ceph_session_state_name(s->s_state));
4179 if (i >= newmap->possible_max_rank) {
4180 /* force close session for stopped mds */
4181 ceph_get_mds_session(s);
4182 __unregister_session(mdsc, s);
4183 __wake_requests(mdsc, &s->s_waiting);
4184 mutex_unlock(&mdsc->mutex);
4186 mutex_lock(&s->s_mutex);
4187 cleanup_session_requests(mdsc, s);
4188 remove_session_caps(s);
4189 mutex_unlock(&s->s_mutex);
4191 ceph_put_mds_session(s);
4193 mutex_lock(&mdsc->mutex);
4194 kick_requests(mdsc, i);
4198 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
4199 ceph_mdsmap_get_addr(newmap, i),
4200 sizeof(struct ceph_entity_addr))) {
4202 mutex_unlock(&mdsc->mutex);
4203 mutex_lock(&s->s_mutex);
4204 mutex_lock(&mdsc->mutex);
4205 ceph_con_close(&s->s_con);
4206 mutex_unlock(&s->s_mutex);
4207 s->s_state = CEPH_MDS_SESSION_RESTARTING;
4208 } else if (oldstate == newstate) {
4209 continue; /* nothing new with this mds */
4215 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4216 newstate >= CEPH_MDS_STATE_RECONNECT) {
4217 mutex_unlock(&mdsc->mutex);
4218 send_mds_reconnect(mdsc, s);
4219 mutex_lock(&mdsc->mutex);
4223 * kick request on any mds that has gone active.
4225 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4226 newstate >= CEPH_MDS_STATE_ACTIVE) {
4227 if (oldstate != CEPH_MDS_STATE_CREATING &&
4228 oldstate != CEPH_MDS_STATE_STARTING)
4229 pr_info("mds%d recovery completed\n", s->s_mds);
4230 kick_requests(mdsc, i);
4231 mutex_unlock(&mdsc->mutex);
4232 mutex_lock(&s->s_mutex);
4233 mutex_lock(&mdsc->mutex);
4234 ceph_kick_flushing_caps(mdsc, s);
4235 mutex_unlock(&s->s_mutex);
4236 wake_up_session_caps(s, RECONNECT);
4240 for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4241 s = mdsc->sessions[i];
4244 if (!ceph_mdsmap_is_laggy(newmap, i))
4246 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4247 s->s_state == CEPH_MDS_SESSION_HUNG ||
4248 s->s_state == CEPH_MDS_SESSION_CLOSING) {
4249 dout(" connecting to export targets of laggy mds%d\n",
4251 __open_export_target_sessions(mdsc, s);
4263 * caller must hold session s_mutex, dentry->d_lock
4265 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4267 struct ceph_dentry_info *di = ceph_dentry(dentry);
4269 ceph_put_mds_session(di->lease_session);
4270 di->lease_session = NULL;
4273 static void handle_lease(struct ceph_mds_client *mdsc,
4274 struct ceph_mds_session *session,
4275 struct ceph_msg *msg)
4277 struct super_block *sb = mdsc->fsc->sb;
4278 struct inode *inode;
4279 struct dentry *parent, *dentry;
4280 struct ceph_dentry_info *di;
4281 int mds = session->s_mds;
4282 struct ceph_mds_lease *h = msg->front.iov_base;
4284 struct ceph_vino vino;
4288 dout("handle_lease from mds%d\n", mds);
4291 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4293 vino.ino = le64_to_cpu(h->ino);
4294 vino.snap = CEPH_NOSNAP;
4295 seq = le32_to_cpu(h->seq);
4296 dname.len = get_unaligned_le32(h + 1);
4297 if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4299 dname.name = (void *)(h + 1) + sizeof(u32);
4302 inode = ceph_find_inode(sb, vino);
4303 dout("handle_lease %s, ino %llx %p %.*s\n",
4304 ceph_lease_op_name(h->action), vino.ino, inode,
4305 dname.len, dname.name);
4307 mutex_lock(&session->s_mutex);
4308 inc_session_sequence(session);
4311 dout("handle_lease no inode %llx\n", vino.ino);
4316 parent = d_find_alias(inode);
4318 dout("no parent dentry on inode %p\n", inode);
4320 goto release; /* hrm... */
4322 dname.hash = full_name_hash(parent, dname.name, dname.len);
4323 dentry = d_lookup(parent, &dname);
4328 spin_lock(&dentry->d_lock);
4329 di = ceph_dentry(dentry);
4330 switch (h->action) {
4331 case CEPH_MDS_LEASE_REVOKE:
4332 if (di->lease_session == session) {
4333 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4334 h->seq = cpu_to_le32(di->lease_seq);
4335 __ceph_mdsc_drop_dentry_lease(dentry);
4340 case CEPH_MDS_LEASE_RENEW:
4341 if (di->lease_session == session &&
4342 di->lease_gen == session->s_cap_gen &&
4343 di->lease_renew_from &&
4344 di->lease_renew_after == 0) {
4345 unsigned long duration =
4346 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4348 di->lease_seq = seq;
4349 di->time = di->lease_renew_from + duration;
4350 di->lease_renew_after = di->lease_renew_from +
4352 di->lease_renew_from = 0;
4356 spin_unlock(&dentry->d_lock);
4363 /* let's just reuse the same message */
4364 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4366 ceph_con_send(&session->s_con, msg);
4369 mutex_unlock(&session->s_mutex);
4370 /* avoid calling iput_final() in mds dispatch threads */
4371 ceph_async_iput(inode);
4375 pr_err("corrupt lease message\n");
4379 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4380 struct dentry *dentry, char action,
4383 struct ceph_msg *msg;
4384 struct ceph_mds_lease *lease;
4386 int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4388 dout("lease_send_msg identry %p %s to mds%d\n",
4389 dentry, ceph_lease_op_name(action), session->s_mds);
4391 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4394 lease = msg->front.iov_base;
4395 lease->action = action;
4396 lease->seq = cpu_to_le32(seq);
4398 spin_lock(&dentry->d_lock);
4399 dir = d_inode(dentry->d_parent);
4400 lease->ino = cpu_to_le64(ceph_ino(dir));
4401 lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4403 put_unaligned_le32(dentry->d_name.len, lease + 1);
4404 memcpy((void *)(lease + 1) + 4,
4405 dentry->d_name.name, dentry->d_name.len);
4406 spin_unlock(&dentry->d_lock);
4408 * if this is a preemptive lease RELEASE, no need to
4409 * flush request stream, since the actual request will
4412 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
4414 ceph_con_send(&session->s_con, msg);
4418 * lock unlock sessions, to wait ongoing session activities
4420 static void lock_unlock_sessions(struct ceph_mds_client *mdsc)
4424 mutex_lock(&mdsc->mutex);
4425 for (i = 0; i < mdsc->max_sessions; i++) {
4426 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4429 mutex_unlock(&mdsc->mutex);
4430 mutex_lock(&s->s_mutex);
4431 mutex_unlock(&s->s_mutex);
4432 ceph_put_mds_session(s);
4433 mutex_lock(&mdsc->mutex);
4435 mutex_unlock(&mdsc->mutex);
4438 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4440 struct ceph_fs_client *fsc = mdsc->fsc;
4442 if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4445 if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4448 if (!READ_ONCE(fsc->blocklisted))
4451 pr_info("auto reconnect after blocklisted\n");
4452 ceph_force_reconnect(fsc->sb);
4455 bool check_session_state(struct ceph_mds_session *s)
4457 switch (s->s_state) {
4458 case CEPH_MDS_SESSION_OPEN:
4459 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4460 s->s_state = CEPH_MDS_SESSION_HUNG;
4461 pr_info("mds%d hung\n", s->s_mds);
4464 case CEPH_MDS_SESSION_CLOSING:
4465 /* Should never reach this when we're unmounting */
4468 case CEPH_MDS_SESSION_NEW:
4469 case CEPH_MDS_SESSION_RESTARTING:
4470 case CEPH_MDS_SESSION_CLOSED:
4471 case CEPH_MDS_SESSION_REJECTED:
4479 * If the sequence is incremented while we're waiting on a REQUEST_CLOSE reply,
4480 * then we need to retransmit that request.
4482 void inc_session_sequence(struct ceph_mds_session *s)
4484 lockdep_assert_held(&s->s_mutex);
4488 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4491 dout("resending session close request for mds%d\n", s->s_mds);
4492 ret = request_close_session(s);
4494 pr_err("unable to close session to mds%d: %d\n",
4500 * delayed work -- periodically trim expired leases, renew caps with mds
4502 static void schedule_delayed(struct ceph_mds_client *mdsc)
4505 unsigned hz = round_jiffies_relative(HZ * delay);
4506 schedule_delayed_work(&mdsc->delayed_work, hz);
4509 static void delayed_work(struct work_struct *work)
4512 struct ceph_mds_client *mdsc =
4513 container_of(work, struct ceph_mds_client, delayed_work.work);
4517 dout("mdsc delayed_work\n");
4522 mutex_lock(&mdsc->mutex);
4523 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4524 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4525 mdsc->last_renew_caps);
4527 mdsc->last_renew_caps = jiffies;
4529 for (i = 0; i < mdsc->max_sessions; i++) {
4530 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4534 if (!check_session_state(s)) {
4535 ceph_put_mds_session(s);
4538 mutex_unlock(&mdsc->mutex);
4540 mutex_lock(&s->s_mutex);
4542 send_renew_caps(mdsc, s);
4544 ceph_con_keepalive(&s->s_con);
4545 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4546 s->s_state == CEPH_MDS_SESSION_HUNG)
4547 ceph_send_cap_releases(mdsc, s);
4548 mutex_unlock(&s->s_mutex);
4549 ceph_put_mds_session(s);
4551 mutex_lock(&mdsc->mutex);
4553 mutex_unlock(&mdsc->mutex);
4555 ceph_check_delayed_caps(mdsc);
4557 ceph_queue_cap_reclaim_work(mdsc);
4559 ceph_trim_snapid_map(mdsc);
4561 maybe_recover_session(mdsc);
4563 schedule_delayed(mdsc);
4566 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4569 struct ceph_mds_client *mdsc;
4572 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4576 mutex_init(&mdsc->mutex);
4577 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4578 if (!mdsc->mdsmap) {
4583 init_completion(&mdsc->safe_umount_waiters);
4584 init_waitqueue_head(&mdsc->session_close_wq);
4585 INIT_LIST_HEAD(&mdsc->waiting_for_map);
4586 mdsc->sessions = NULL;
4587 atomic_set(&mdsc->num_sessions, 0);
4588 mdsc->max_sessions = 0;
4590 atomic64_set(&mdsc->quotarealms_count, 0);
4591 mdsc->quotarealms_inodes = RB_ROOT;
4592 mutex_init(&mdsc->quotarealms_inodes_mutex);
4593 mdsc->last_snap_seq = 0;
4594 init_rwsem(&mdsc->snap_rwsem);
4595 mdsc->snap_realms = RB_ROOT;
4596 INIT_LIST_HEAD(&mdsc->snap_empty);
4597 mdsc->num_snap_realms = 0;
4598 spin_lock_init(&mdsc->snap_empty_lock);
4600 mdsc->oldest_tid = 0;
4601 mdsc->request_tree = RB_ROOT;
4602 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4603 mdsc->last_renew_caps = jiffies;
4604 INIT_LIST_HEAD(&mdsc->cap_delay_list);
4605 INIT_LIST_HEAD(&mdsc->cap_wait_list);
4606 spin_lock_init(&mdsc->cap_delay_lock);
4607 INIT_LIST_HEAD(&mdsc->snap_flush_list);
4608 spin_lock_init(&mdsc->snap_flush_lock);
4609 mdsc->last_cap_flush_tid = 1;
4610 INIT_LIST_HEAD(&mdsc->cap_flush_list);
4611 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4612 mdsc->num_cap_flushing = 0;
4613 spin_lock_init(&mdsc->cap_dirty_lock);
4614 init_waitqueue_head(&mdsc->cap_flushing_wq);
4615 INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4616 atomic_set(&mdsc->cap_reclaim_pending, 0);
4617 err = ceph_metric_init(&mdsc->metric);
4621 spin_lock_init(&mdsc->dentry_list_lock);
4622 INIT_LIST_HEAD(&mdsc->dentry_leases);
4623 INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4625 ceph_caps_init(mdsc);
4626 ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4628 spin_lock_init(&mdsc->snapid_map_lock);
4629 mdsc->snapid_map_tree = RB_ROOT;
4630 INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4632 init_rwsem(&mdsc->pool_perm_rwsem);
4633 mdsc->pool_perm_tree = RB_ROOT;
4635 strscpy(mdsc->nodename, utsname()->nodename,
4636 sizeof(mdsc->nodename));
4642 kfree(mdsc->mdsmap);
4649 * Wait for safe replies on open mds requests. If we time out, drop
4650 * all requests from the tree to avoid dangling dentry refs.
4652 static void wait_requests(struct ceph_mds_client *mdsc)
4654 struct ceph_options *opts = mdsc->fsc->client->options;
4655 struct ceph_mds_request *req;
4657 mutex_lock(&mdsc->mutex);
4658 if (__get_oldest_req(mdsc)) {
4659 mutex_unlock(&mdsc->mutex);
4661 dout("wait_requests waiting for requests\n");
4662 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4663 ceph_timeout_jiffies(opts->mount_timeout));
4665 /* tear down remaining requests */
4666 mutex_lock(&mdsc->mutex);
4667 while ((req = __get_oldest_req(mdsc))) {
4668 dout("wait_requests timed out on tid %llu\n",
4670 list_del_init(&req->r_wait);
4671 __unregister_request(mdsc, req);
4674 mutex_unlock(&mdsc->mutex);
4675 dout("wait_requests done\n");
4679 * called before mount is ro, and before dentries are torn down.
4680 * (hmm, does this still race with new lookups?)
4682 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4684 dout("pre_umount\n");
4687 lock_unlock_sessions(mdsc);
4688 ceph_flush_dirty_caps(mdsc);
4689 wait_requests(mdsc);
4692 * wait for reply handlers to drop their request refs and
4693 * their inode/dcache refs
4697 ceph_cleanup_quotarealms_inodes(mdsc);
4701 * wait for all write mds requests to flush.
4703 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
4705 struct ceph_mds_request *req = NULL, *nextreq;
4708 mutex_lock(&mdsc->mutex);
4709 dout("wait_unsafe_requests want %lld\n", want_tid);
4711 req = __get_oldest_req(mdsc);
4712 while (req && req->r_tid <= want_tid) {
4713 /* find next request */
4714 n = rb_next(&req->r_node);
4716 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4719 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4720 (req->r_op & CEPH_MDS_OP_WRITE)) {
4722 ceph_mdsc_get_request(req);
4724 ceph_mdsc_get_request(nextreq);
4725 mutex_unlock(&mdsc->mutex);
4726 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
4727 req->r_tid, want_tid);
4728 wait_for_completion(&req->r_safe_completion);
4729 mutex_lock(&mdsc->mutex);
4730 ceph_mdsc_put_request(req);
4732 break; /* next dne before, so we're done! */
4733 if (RB_EMPTY_NODE(&nextreq->r_node)) {
4734 /* next request was removed from tree */
4735 ceph_mdsc_put_request(nextreq);
4738 ceph_mdsc_put_request(nextreq); /* won't go away */
4742 mutex_unlock(&mdsc->mutex);
4743 dout("wait_unsafe_requests done\n");
4746 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4748 u64 want_tid, want_flush;
4750 if (READ_ONCE(mdsc->fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN)
4754 mutex_lock(&mdsc->mutex);
4755 want_tid = mdsc->last_tid;
4756 mutex_unlock(&mdsc->mutex);
4758 ceph_flush_dirty_caps(mdsc);
4759 spin_lock(&mdsc->cap_dirty_lock);
4760 want_flush = mdsc->last_cap_flush_tid;
4761 if (!list_empty(&mdsc->cap_flush_list)) {
4762 struct ceph_cap_flush *cf =
4763 list_last_entry(&mdsc->cap_flush_list,
4764 struct ceph_cap_flush, g_list);
4767 spin_unlock(&mdsc->cap_dirty_lock);
4769 dout("sync want tid %lld flush_seq %lld\n",
4770 want_tid, want_flush);
4772 wait_unsafe_requests(mdsc, want_tid);
4773 wait_caps_flush(mdsc, want_flush);
4777 * true if all sessions are closed, or we force unmount
4779 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
4781 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4783 return atomic_read(&mdsc->num_sessions) <= skipped;
4787 * called after sb is ro.
4789 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
4791 struct ceph_options *opts = mdsc->fsc->client->options;
4792 struct ceph_mds_session *session;
4796 dout("close_sessions\n");
4798 /* close sessions */
4799 mutex_lock(&mdsc->mutex);
4800 for (i = 0; i < mdsc->max_sessions; i++) {
4801 session = __ceph_lookup_mds_session(mdsc, i);
4804 mutex_unlock(&mdsc->mutex);
4805 mutex_lock(&session->s_mutex);
4806 if (__close_session(mdsc, session) <= 0)
4808 mutex_unlock(&session->s_mutex);
4809 ceph_put_mds_session(session);
4810 mutex_lock(&mdsc->mutex);
4812 mutex_unlock(&mdsc->mutex);
4814 dout("waiting for sessions to close\n");
4815 wait_event_timeout(mdsc->session_close_wq,
4816 done_closing_sessions(mdsc, skipped),
4817 ceph_timeout_jiffies(opts->mount_timeout));
4819 /* tear down remaining sessions */
4820 mutex_lock(&mdsc->mutex);
4821 for (i = 0; i < mdsc->max_sessions; i++) {
4822 if (mdsc->sessions[i]) {
4823 session = ceph_get_mds_session(mdsc->sessions[i]);
4824 __unregister_session(mdsc, session);
4825 mutex_unlock(&mdsc->mutex);
4826 mutex_lock(&session->s_mutex);
4827 remove_session_caps(session);
4828 mutex_unlock(&session->s_mutex);
4829 ceph_put_mds_session(session);
4830 mutex_lock(&mdsc->mutex);
4833 WARN_ON(!list_empty(&mdsc->cap_delay_list));
4834 mutex_unlock(&mdsc->mutex);
4836 ceph_cleanup_snapid_map(mdsc);
4837 ceph_cleanup_empty_realms(mdsc);
4839 cancel_work_sync(&mdsc->cap_reclaim_work);
4840 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4845 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
4847 struct ceph_mds_session *session;
4850 dout("force umount\n");
4852 mutex_lock(&mdsc->mutex);
4853 for (mds = 0; mds < mdsc->max_sessions; mds++) {
4854 session = __ceph_lookup_mds_session(mdsc, mds);
4858 if (session->s_state == CEPH_MDS_SESSION_REJECTED)
4859 __unregister_session(mdsc, session);
4860 __wake_requests(mdsc, &session->s_waiting);
4861 mutex_unlock(&mdsc->mutex);
4863 mutex_lock(&session->s_mutex);
4864 __close_session(mdsc, session);
4865 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
4866 cleanup_session_requests(mdsc, session);
4867 remove_session_caps(session);
4869 mutex_unlock(&session->s_mutex);
4870 ceph_put_mds_session(session);
4872 mutex_lock(&mdsc->mutex);
4873 kick_requests(mdsc, mds);
4875 __wake_requests(mdsc, &mdsc->waiting_for_map);
4876 mutex_unlock(&mdsc->mutex);
4879 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
4883 * Make sure the delayed work stopped before releasing
4886 * Because the cancel_delayed_work_sync() will only
4887 * guarantee that the work finishes executing. But the
4888 * delayed work will re-arm itself again after that.
4890 flush_delayed_work(&mdsc->delayed_work);
4893 ceph_mdsmap_destroy(mdsc->mdsmap);
4894 kfree(mdsc->sessions);
4895 ceph_caps_finalize(mdsc);
4896 ceph_pool_perm_destroy(mdsc);
4899 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
4901 struct ceph_mds_client *mdsc = fsc->mdsc;
4902 dout("mdsc_destroy %p\n", mdsc);
4907 /* flush out any connection work with references to us */
4910 ceph_mdsc_stop(mdsc);
4912 ceph_metric_destroy(&mdsc->metric);
4914 flush_delayed_work(&mdsc->metric.delayed_work);
4917 dout("mdsc_destroy %p done\n", mdsc);
4920 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4922 struct ceph_fs_client *fsc = mdsc->fsc;
4923 const char *mds_namespace = fsc->mount_options->mds_namespace;
4924 void *p = msg->front.iov_base;
4925 void *end = p + msg->front.iov_len;
4928 u32 mount_fscid = (u32)-1;
4931 ceph_decode_need(&p, end, sizeof(u32), bad);
4932 epoch = ceph_decode_32(&p);
4934 dout("handle_fsmap epoch %u\n", epoch);
4936 /* struct_v, struct_cv, map_len, epoch, legacy_client_fscid */
4937 ceph_decode_skip_n(&p, end, 2 + sizeof(u32) * 3, bad);
4939 ceph_decode_32_safe(&p, end, num_fs, bad);
4940 while (num_fs-- > 0) {
4941 void *info_p, *info_end;
4945 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
4946 p += 2; // info_v, info_cv
4947 info_len = ceph_decode_32(&p);
4948 ceph_decode_need(&p, end, info_len, bad);
4950 info_end = p + info_len;
4953 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
4954 fscid = ceph_decode_32(&info_p);
4955 namelen = ceph_decode_32(&info_p);
4956 ceph_decode_need(&info_p, info_end, namelen, bad);
4958 if (mds_namespace &&
4959 strlen(mds_namespace) == namelen &&
4960 !strncmp(mds_namespace, (char *)info_p, namelen)) {
4961 mount_fscid = fscid;
4966 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
4967 if (mount_fscid != (u32)-1) {
4968 fsc->client->monc.fs_cluster_id = mount_fscid;
4969 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
4971 ceph_monc_renew_subs(&fsc->client->monc);
4979 pr_err("error decoding fsmap\n");
4981 mutex_lock(&mdsc->mutex);
4982 mdsc->mdsmap_err = err;
4983 __wake_requests(mdsc, &mdsc->waiting_for_map);
4984 mutex_unlock(&mdsc->mutex);
4988 * handle mds map update.
4990 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4994 void *p = msg->front.iov_base;
4995 void *end = p + msg->front.iov_len;
4996 struct ceph_mdsmap *newmap, *oldmap;
4997 struct ceph_fsid fsid;
5000 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
5001 ceph_decode_copy(&p, &fsid, sizeof(fsid));
5002 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
5004 epoch = ceph_decode_32(&p);
5005 maplen = ceph_decode_32(&p);
5006 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
5008 /* do we need it? */
5009 mutex_lock(&mdsc->mutex);
5010 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
5011 dout("handle_map epoch %u <= our %u\n",
5012 epoch, mdsc->mdsmap->m_epoch);
5013 mutex_unlock(&mdsc->mutex);
5017 newmap = ceph_mdsmap_decode(&p, end, ceph_msgr2(mdsc->fsc->client));
5018 if (IS_ERR(newmap)) {
5019 err = PTR_ERR(newmap);
5023 /* swap into place */
5025 oldmap = mdsc->mdsmap;
5026 mdsc->mdsmap = newmap;
5027 check_new_map(mdsc, newmap, oldmap);
5028 ceph_mdsmap_destroy(oldmap);
5030 mdsc->mdsmap = newmap; /* first mds map */
5032 mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
5035 __wake_requests(mdsc, &mdsc->waiting_for_map);
5036 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
5037 mdsc->mdsmap->m_epoch);
5039 mutex_unlock(&mdsc->mutex);
5040 schedule_delayed(mdsc);
5044 mutex_unlock(&mdsc->mutex);
5046 pr_err("error decoding mdsmap %d\n", err);
5050 static struct ceph_connection *con_get(struct ceph_connection *con)
5052 struct ceph_mds_session *s = con->private;
5054 if (ceph_get_mds_session(s))
5059 static void con_put(struct ceph_connection *con)
5061 struct ceph_mds_session *s = con->private;
5063 ceph_put_mds_session(s);
5067 * if the client is unresponsive for long enough, the mds will kill
5068 * the session entirely.
5070 static void peer_reset(struct ceph_connection *con)
5072 struct ceph_mds_session *s = con->private;
5073 struct ceph_mds_client *mdsc = s->s_mdsc;
5075 pr_warn("mds%d closed our session\n", s->s_mds);
5076 send_mds_reconnect(mdsc, s);
5079 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
5081 struct ceph_mds_session *s = con->private;
5082 struct ceph_mds_client *mdsc = s->s_mdsc;
5083 int type = le16_to_cpu(msg->hdr.type);
5085 mutex_lock(&mdsc->mutex);
5086 if (__verify_registered_session(mdsc, s) < 0) {
5087 mutex_unlock(&mdsc->mutex);
5090 mutex_unlock(&mdsc->mutex);
5093 case CEPH_MSG_MDS_MAP:
5094 ceph_mdsc_handle_mdsmap(mdsc, msg);
5096 case CEPH_MSG_FS_MAP_USER:
5097 ceph_mdsc_handle_fsmap(mdsc, msg);
5099 case CEPH_MSG_CLIENT_SESSION:
5100 handle_session(s, msg);
5102 case CEPH_MSG_CLIENT_REPLY:
5103 handle_reply(s, msg);
5105 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
5106 handle_forward(mdsc, s, msg);
5108 case CEPH_MSG_CLIENT_CAPS:
5109 ceph_handle_caps(s, msg);
5111 case CEPH_MSG_CLIENT_SNAP:
5112 ceph_handle_snap(mdsc, s, msg);
5114 case CEPH_MSG_CLIENT_LEASE:
5115 handle_lease(mdsc, s, msg);
5117 case CEPH_MSG_CLIENT_QUOTA:
5118 ceph_handle_quota(mdsc, s, msg);
5122 pr_err("received unknown message type %d %s\n", type,
5123 ceph_msg_type_name(type));
5134 * Note: returned pointer is the address of a structure that's
5135 * managed separately. Caller must *not* attempt to free it.
5137 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
5138 int *proto, int force_new)
5140 struct ceph_mds_session *s = con->private;
5141 struct ceph_mds_client *mdsc = s->s_mdsc;
5142 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5143 struct ceph_auth_handshake *auth = &s->s_auth;
5146 ret = __ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5147 force_new, proto, NULL, NULL);
5149 return ERR_PTR(ret);
5154 static int add_authorizer_challenge(struct ceph_connection *con,
5155 void *challenge_buf, int challenge_buf_len)
5157 struct ceph_mds_session *s = con->private;
5158 struct ceph_mds_client *mdsc = s->s_mdsc;
5159 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5161 return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
5162 challenge_buf, challenge_buf_len);
5165 static int verify_authorizer_reply(struct ceph_connection *con)
5167 struct ceph_mds_session *s = con->private;
5168 struct ceph_mds_client *mdsc = s->s_mdsc;
5169 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5170 struct ceph_auth_handshake *auth = &s->s_auth;
5172 return ceph_auth_verify_authorizer_reply(ac, auth->authorizer,
5173 auth->authorizer_reply_buf, auth->authorizer_reply_buf_len,
5174 NULL, NULL, NULL, NULL);
5177 static int invalidate_authorizer(struct ceph_connection *con)
5179 struct ceph_mds_session *s = con->private;
5180 struct ceph_mds_client *mdsc = s->s_mdsc;
5181 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
5183 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
5185 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
5188 static int mds_get_auth_request(struct ceph_connection *con,
5189 void *buf, int *buf_len,
5190 void **authorizer, int *authorizer_len)
5192 struct ceph_mds_session *s = con->private;
5193 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5194 struct ceph_auth_handshake *auth = &s->s_auth;
5197 ret = ceph_auth_get_authorizer(ac, auth, CEPH_ENTITY_TYPE_MDS,
5202 *authorizer = auth->authorizer_buf;
5203 *authorizer_len = auth->authorizer_buf_len;
5207 static int mds_handle_auth_reply_more(struct ceph_connection *con,
5208 void *reply, int reply_len,
5209 void *buf, int *buf_len,
5210 void **authorizer, int *authorizer_len)
5212 struct ceph_mds_session *s = con->private;
5213 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5214 struct ceph_auth_handshake *auth = &s->s_auth;
5217 ret = ceph_auth_handle_svc_reply_more(ac, auth, reply, reply_len,
5222 *authorizer = auth->authorizer_buf;
5223 *authorizer_len = auth->authorizer_buf_len;
5227 static int mds_handle_auth_done(struct ceph_connection *con,
5228 u64 global_id, void *reply, int reply_len,
5229 u8 *session_key, int *session_key_len,
5230 u8 *con_secret, int *con_secret_len)
5232 struct ceph_mds_session *s = con->private;
5233 struct ceph_auth_client *ac = s->s_mdsc->fsc->client->monc.auth;
5234 struct ceph_auth_handshake *auth = &s->s_auth;
5236 return ceph_auth_handle_svc_reply_done(ac, auth, reply, reply_len,
5237 session_key, session_key_len,
5238 con_secret, con_secret_len);
5241 static int mds_handle_auth_bad_method(struct ceph_connection *con,
5242 int used_proto, int result,
5243 const int *allowed_protos, int proto_cnt,
5244 const int *allowed_modes, int mode_cnt)
5246 struct ceph_mds_session *s = con->private;
5247 struct ceph_mon_client *monc = &s->s_mdsc->fsc->client->monc;
5250 if (ceph_auth_handle_bad_authorizer(monc->auth, CEPH_ENTITY_TYPE_MDS,
5252 allowed_protos, proto_cnt,
5253 allowed_modes, mode_cnt)) {
5254 ret = ceph_monc_validate_auth(monc);
5262 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
5263 struct ceph_msg_header *hdr, int *skip)
5265 struct ceph_msg *msg;
5266 int type = (int) le16_to_cpu(hdr->type);
5267 int front_len = (int) le32_to_cpu(hdr->front_len);
5273 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
5275 pr_err("unable to allocate msg type %d len %d\n",
5283 static int mds_sign_message(struct ceph_msg *msg)
5285 struct ceph_mds_session *s = msg->con->private;
5286 struct ceph_auth_handshake *auth = &s->s_auth;
5288 return ceph_auth_sign_message(auth, msg);
5291 static int mds_check_message_signature(struct ceph_msg *msg)
5293 struct ceph_mds_session *s = msg->con->private;
5294 struct ceph_auth_handshake *auth = &s->s_auth;
5296 return ceph_auth_check_message_signature(auth, msg);
5299 static const struct ceph_connection_operations mds_con_ops = {
5302 .dispatch = dispatch,
5303 .get_authorizer = get_authorizer,
5304 .add_authorizer_challenge = add_authorizer_challenge,
5305 .verify_authorizer_reply = verify_authorizer_reply,
5306 .invalidate_authorizer = invalidate_authorizer,
5307 .peer_reset = peer_reset,
5308 .alloc_msg = mds_alloc_msg,
5309 .sign_message = mds_sign_message,
5310 .check_message_signature = mds_check_message_signature,
5311 .get_auth_request = mds_get_auth_request,
5312 .handle_auth_reply_more = mds_handle_auth_reply_more,
5313 .handle_auth_done = mds_handle_auth_done,
5314 .handle_auth_bad_method = mds_handle_auth_bad_method,