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>
14 #include "mds_client.h"
16 #include <linux/ceph/ceph_features.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/pagelist.h>
20 #include <linux/ceph/auth.h>
21 #include <linux/ceph/debugfs.h>
24 * A cluster of MDS (metadata server) daemons is responsible for
25 * managing the file system namespace (the directory hierarchy and
26 * inodes) and for coordinating shared access to storage. Metadata is
27 * partitioning hierarchically across a number of servers, and that
28 * partition varies over time as the cluster adjusts the distribution
29 * in order to balance load.
31 * The MDS client is primarily responsible to managing synchronous
32 * metadata requests for operations like open, unlink, and so forth.
33 * If there is a MDS failure, we find out about it when we (possibly
34 * request and) receive a new MDS map, and can resubmit affected
37 * For the most part, though, we take advantage of a lossless
38 * communications channel to the MDS, and do not need to worry about
39 * timing out or resubmitting requests.
41 * We maintain a stateful "session" with each MDS we interact with.
42 * Within each session, we sent periodic heartbeat messages to ensure
43 * any capabilities or leases we have been issues remain valid. If
44 * the session times out and goes stale, our leases and capabilities
45 * are no longer valid.
48 struct ceph_reconnect_state {
50 struct ceph_pagelist *pagelist;
54 static void __wake_requests(struct ceph_mds_client *mdsc,
55 struct list_head *head);
57 static const struct ceph_connection_operations mds_con_ops;
65 * parse individual inode info
67 static int parse_reply_info_in(void **p, void *end,
68 struct ceph_mds_reply_info_in *info,
74 *p += sizeof(struct ceph_mds_reply_inode) +
75 sizeof(*info->in->fragtree.splits) *
76 le32_to_cpu(info->in->fragtree.nsplits);
78 ceph_decode_32_safe(p, end, info->symlink_len, bad);
79 ceph_decode_need(p, end, info->symlink_len, bad);
81 *p += info->symlink_len;
83 ceph_decode_copy_safe(p, end, &info->dir_layout,
84 sizeof(info->dir_layout), bad);
85 ceph_decode_32_safe(p, end, info->xattr_len, bad);
86 ceph_decode_need(p, end, info->xattr_len, bad);
87 info->xattr_data = *p;
88 *p += info->xattr_len;
90 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
91 ceph_decode_64_safe(p, end, info->inline_version, bad);
92 ceph_decode_32_safe(p, end, info->inline_len, bad);
93 ceph_decode_need(p, end, info->inline_len, bad);
94 info->inline_data = *p;
95 *p += info->inline_len;
97 info->inline_version = CEPH_INLINE_NONE;
99 if (features & CEPH_FEATURE_MDS_QUOTA) {
100 u8 struct_v, struct_compat;
104 * both struct_v and struct_compat are expected to be >= 1
106 ceph_decode_8_safe(p, end, struct_v, bad);
107 ceph_decode_8_safe(p, end, struct_compat, bad);
108 if (!struct_v || !struct_compat)
110 ceph_decode_32_safe(p, end, struct_len, bad);
111 ceph_decode_need(p, end, struct_len, bad);
112 ceph_decode_64_safe(p, end, info->max_bytes, bad);
113 ceph_decode_64_safe(p, end, info->max_files, bad);
119 info->pool_ns_len = 0;
120 info->pool_ns_data = NULL;
121 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
122 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
123 if (info->pool_ns_len > 0) {
124 ceph_decode_need(p, end, info->pool_ns_len, bad);
125 info->pool_ns_data = *p;
126 *p += info->pool_ns_len;
136 * parse a normal reply, which may contain a (dir+)dentry and/or a
139 static int parse_reply_info_trace(void **p, void *end,
140 struct ceph_mds_reply_info_parsed *info,
145 if (info->head->is_dentry) {
146 err = parse_reply_info_in(p, end, &info->diri, features);
150 if (unlikely(*p + sizeof(*info->dirfrag) > end))
153 *p += sizeof(*info->dirfrag) +
154 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
155 if (unlikely(*p > end))
158 ceph_decode_32_safe(p, end, info->dname_len, bad);
159 ceph_decode_need(p, end, info->dname_len, bad);
161 *p += info->dname_len;
163 *p += sizeof(*info->dlease);
166 if (info->head->is_target) {
167 err = parse_reply_info_in(p, end, &info->targeti, features);
172 if (unlikely(*p != end))
179 pr_err("problem parsing mds trace %d\n", err);
184 * parse readdir results
186 static int parse_reply_info_dir(void **p, void *end,
187 struct ceph_mds_reply_info_parsed *info,
194 if (*p + sizeof(*info->dir_dir) > end)
196 *p += sizeof(*info->dir_dir) +
197 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
201 ceph_decode_need(p, end, sizeof(num) + 2, bad);
202 num = ceph_decode_32(p);
204 u16 flags = ceph_decode_16(p);
205 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
206 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
207 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
208 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
213 BUG_ON(!info->dir_entries);
214 if ((unsigned long)(info->dir_entries + num) >
215 (unsigned long)info->dir_entries + info->dir_buf_size) {
216 pr_err("dir contents are larger than expected\n");
223 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
225 ceph_decode_need(p, end, sizeof(u32)*2, bad);
226 rde->name_len = ceph_decode_32(p);
227 ceph_decode_need(p, end, rde->name_len, bad);
230 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
232 *p += sizeof(struct ceph_mds_reply_lease);
235 err = parse_reply_info_in(p, end, &rde->inode, features);
238 /* ceph_readdir_prepopulate() will update it */
252 pr_err("problem parsing dir contents %d\n", err);
257 * parse fcntl F_GETLK results
259 static int parse_reply_info_filelock(void **p, void *end,
260 struct ceph_mds_reply_info_parsed *info,
263 if (*p + sizeof(*info->filelock_reply) > end)
266 info->filelock_reply = *p;
267 *p += sizeof(*info->filelock_reply);
269 if (unlikely(*p != end))
278 * parse create results
280 static int parse_reply_info_create(void **p, void *end,
281 struct ceph_mds_reply_info_parsed *info,
284 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
286 info->has_create_ino = false;
288 info->has_create_ino = true;
289 info->ino = ceph_decode_64(p);
293 if (unlikely(*p != end))
302 * parse extra results
304 static int parse_reply_info_extra(void **p, void *end,
305 struct ceph_mds_reply_info_parsed *info,
308 u32 op = le32_to_cpu(info->head->op);
310 if (op == CEPH_MDS_OP_GETFILELOCK)
311 return parse_reply_info_filelock(p, end, info, features);
312 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
313 return parse_reply_info_dir(p, end, info, features);
314 else if (op == CEPH_MDS_OP_CREATE)
315 return parse_reply_info_create(p, end, info, features);
321 * parse entire mds reply
323 static int parse_reply_info(struct ceph_msg *msg,
324 struct ceph_mds_reply_info_parsed *info,
331 info->head = msg->front.iov_base;
332 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
333 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
336 ceph_decode_32_safe(&p, end, len, bad);
338 ceph_decode_need(&p, end, len, bad);
339 err = parse_reply_info_trace(&p, p+len, info, features);
345 ceph_decode_32_safe(&p, end, len, bad);
347 ceph_decode_need(&p, end, len, bad);
348 err = parse_reply_info_extra(&p, p+len, info, features);
354 ceph_decode_32_safe(&p, end, len, bad);
355 info->snapblob_len = len;
366 pr_err("mds parse_reply err %d\n", err);
370 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
372 if (!info->dir_entries)
374 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
381 const char *ceph_session_state_name(int s)
384 case CEPH_MDS_SESSION_NEW: return "new";
385 case CEPH_MDS_SESSION_OPENING: return "opening";
386 case CEPH_MDS_SESSION_OPEN: return "open";
387 case CEPH_MDS_SESSION_HUNG: return "hung";
388 case CEPH_MDS_SESSION_CLOSING: return "closing";
389 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
390 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
391 case CEPH_MDS_SESSION_REJECTED: return "rejected";
392 default: return "???";
396 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
398 if (refcount_inc_not_zero(&s->s_ref)) {
399 dout("mdsc get_session %p %d -> %d\n", s,
400 refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
403 dout("mdsc get_session %p 0 -- FAIL\n", s);
408 void ceph_put_mds_session(struct ceph_mds_session *s)
410 dout("mdsc put_session %p %d -> %d\n", s,
411 refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
412 if (refcount_dec_and_test(&s->s_ref)) {
413 if (s->s_auth.authorizer)
414 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
420 * called under mdsc->mutex
422 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
425 struct ceph_mds_session *session;
427 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
429 session = mdsc->sessions[mds];
430 dout("lookup_mds_session %p %d\n", session,
431 refcount_read(&session->s_ref));
432 get_session(session);
436 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
438 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
444 static int __verify_registered_session(struct ceph_mds_client *mdsc,
445 struct ceph_mds_session *s)
447 if (s->s_mds >= mdsc->max_sessions ||
448 mdsc->sessions[s->s_mds] != s)
454 * create+register a new session for given mds.
455 * called under mdsc->mutex.
457 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
460 struct ceph_mds_session *s;
462 if (mds >= mdsc->mdsmap->m_num_mds)
463 return ERR_PTR(-EINVAL);
465 s = kzalloc(sizeof(*s), GFP_NOFS);
467 return ERR_PTR(-ENOMEM);
469 if (mds >= mdsc->max_sessions) {
470 int newmax = 1 << get_count_order(mds + 1);
471 struct ceph_mds_session **sa;
473 dout("%s: realloc to %d\n", __func__, newmax);
474 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
477 if (mdsc->sessions) {
478 memcpy(sa, mdsc->sessions,
479 mdsc->max_sessions * sizeof(void *));
480 kfree(mdsc->sessions);
483 mdsc->max_sessions = newmax;
486 dout("%s: mds%d\n", __func__, mds);
489 s->s_state = CEPH_MDS_SESSION_NEW;
492 mutex_init(&s->s_mutex);
494 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
496 spin_lock_init(&s->s_gen_ttl_lock);
498 s->s_cap_ttl = jiffies - 1;
500 spin_lock_init(&s->s_cap_lock);
501 s->s_renew_requested = 0;
503 INIT_LIST_HEAD(&s->s_caps);
506 refcount_set(&s->s_ref, 1);
507 INIT_LIST_HEAD(&s->s_waiting);
508 INIT_LIST_HEAD(&s->s_unsafe);
509 s->s_num_cap_releases = 0;
510 s->s_cap_reconnect = 0;
511 s->s_cap_iterator = NULL;
512 INIT_LIST_HEAD(&s->s_cap_releases);
513 INIT_LIST_HEAD(&s->s_cap_flushing);
515 mdsc->sessions[mds] = s;
516 atomic_inc(&mdsc->num_sessions);
517 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
519 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
520 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
526 return ERR_PTR(-ENOMEM);
530 * called under mdsc->mutex
532 static void __unregister_session(struct ceph_mds_client *mdsc,
533 struct ceph_mds_session *s)
535 dout("__unregister_session mds%d %p\n", s->s_mds, s);
536 BUG_ON(mdsc->sessions[s->s_mds] != s);
537 mdsc->sessions[s->s_mds] = NULL;
538 ceph_con_close(&s->s_con);
539 ceph_put_mds_session(s);
540 atomic_dec(&mdsc->num_sessions);
544 * drop session refs in request.
546 * should be last request ref, or hold mdsc->mutex
548 static void put_request_session(struct ceph_mds_request *req)
550 if (req->r_session) {
551 ceph_put_mds_session(req->r_session);
552 req->r_session = NULL;
556 void ceph_mdsc_release_request(struct kref *kref)
558 struct ceph_mds_request *req = container_of(kref,
559 struct ceph_mds_request,
561 destroy_reply_info(&req->r_reply_info);
563 ceph_msg_put(req->r_request);
565 ceph_msg_put(req->r_reply);
567 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
571 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
572 iput(req->r_target_inode);
575 if (req->r_old_dentry)
576 dput(req->r_old_dentry);
577 if (req->r_old_dentry_dir) {
579 * track (and drop pins for) r_old_dentry_dir
580 * separately, since r_old_dentry's d_parent may have
581 * changed between the dir mutex being dropped and
582 * this request being freed.
584 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
586 iput(req->r_old_dentry_dir);
591 ceph_pagelist_release(req->r_pagelist);
592 put_request_session(req);
593 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
597 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
600 * lookup session, bump ref if found.
602 * called under mdsc->mutex.
604 static struct ceph_mds_request *
605 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
607 struct ceph_mds_request *req;
609 req = lookup_request(&mdsc->request_tree, tid);
611 ceph_mdsc_get_request(req);
617 * Register an in-flight request, and assign a tid. Link to directory
618 * are modifying (if any).
620 * Called under mdsc->mutex.
622 static void __register_request(struct ceph_mds_client *mdsc,
623 struct ceph_mds_request *req,
628 req->r_tid = ++mdsc->last_tid;
629 if (req->r_num_caps) {
630 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
633 pr_err("__register_request %p "
634 "failed to reserve caps: %d\n", req, ret);
635 /* set req->r_err to fail early from __do_request */
640 dout("__register_request %p tid %lld\n", req, req->r_tid);
641 ceph_mdsc_get_request(req);
642 insert_request(&mdsc->request_tree, req);
644 req->r_uid = current_fsuid();
645 req->r_gid = current_fsgid();
647 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
648 mdsc->oldest_tid = req->r_tid;
652 req->r_unsafe_dir = dir;
656 static void __unregister_request(struct ceph_mds_client *mdsc,
657 struct ceph_mds_request *req)
659 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
661 /* Never leave an unregistered request on an unsafe list! */
662 list_del_init(&req->r_unsafe_item);
664 if (req->r_tid == mdsc->oldest_tid) {
665 struct rb_node *p = rb_next(&req->r_node);
666 mdsc->oldest_tid = 0;
668 struct ceph_mds_request *next_req =
669 rb_entry(p, struct ceph_mds_request, r_node);
670 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
671 mdsc->oldest_tid = next_req->r_tid;
678 erase_request(&mdsc->request_tree, req);
680 if (req->r_unsafe_dir &&
681 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
682 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
683 spin_lock(&ci->i_unsafe_lock);
684 list_del_init(&req->r_unsafe_dir_item);
685 spin_unlock(&ci->i_unsafe_lock);
687 if (req->r_target_inode &&
688 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
689 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
690 spin_lock(&ci->i_unsafe_lock);
691 list_del_init(&req->r_unsafe_target_item);
692 spin_unlock(&ci->i_unsafe_lock);
695 if (req->r_unsafe_dir) {
696 iput(req->r_unsafe_dir);
697 req->r_unsafe_dir = NULL;
700 complete_all(&req->r_safe_completion);
702 ceph_mdsc_put_request(req);
706 * Walk back up the dentry tree until we hit a dentry representing a
707 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
708 * when calling this) to ensure that the objects won't disappear while we're
709 * working with them. Once we hit a candidate dentry, we attempt to take a
710 * reference to it, and return that as the result.
712 static struct inode *get_nonsnap_parent(struct dentry *dentry)
714 struct inode *inode = NULL;
716 while (dentry && !IS_ROOT(dentry)) {
717 inode = d_inode_rcu(dentry);
718 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
720 dentry = dentry->d_parent;
723 inode = igrab(inode);
728 * Choose mds to send request to next. If there is a hint set in the
729 * request (e.g., due to a prior forward hint from the mds), use that.
730 * Otherwise, consult frag tree and/or caps to identify the
731 * appropriate mds. If all else fails, choose randomly.
733 * Called under mdsc->mutex.
735 static int __choose_mds(struct ceph_mds_client *mdsc,
736 struct ceph_mds_request *req)
739 struct ceph_inode_info *ci;
740 struct ceph_cap *cap;
741 int mode = req->r_direct_mode;
743 u32 hash = req->r_direct_hash;
744 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
747 * is there a specific mds we should try? ignore hint if we have
748 * no session and the mds is not up (active or recovering).
750 if (req->r_resend_mds >= 0 &&
751 (__have_session(mdsc, req->r_resend_mds) ||
752 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
753 dout("choose_mds using resend_mds mds%d\n",
755 return req->r_resend_mds;
758 if (mode == USE_RANDOM_MDS)
763 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
764 inode = req->r_inode;
767 /* req->r_dentry is non-null for LSSNAP request */
769 inode = get_nonsnap_parent(req->r_dentry);
771 dout("__choose_mds using snapdir's parent %p\n", inode);
773 } else if (req->r_dentry) {
774 /* ignore race with rename; old or new d_parent is okay */
775 struct dentry *parent;
779 parent = req->r_dentry->d_parent;
780 dir = req->r_parent ? : d_inode_rcu(parent);
782 if (!dir || dir->i_sb != mdsc->fsc->sb) {
783 /* not this fs or parent went negative */
784 inode = d_inode(req->r_dentry);
787 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
788 /* direct snapped/virtual snapdir requests
789 * based on parent dir inode */
790 inode = get_nonsnap_parent(parent);
791 dout("__choose_mds using nonsnap parent %p\n", inode);
794 inode = d_inode(req->r_dentry);
795 if (!inode || mode == USE_AUTH_MDS) {
798 hash = ceph_dentry_hash(dir, req->r_dentry);
807 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
811 ci = ceph_inode(inode);
813 if (is_hash && S_ISDIR(inode->i_mode)) {
814 struct ceph_inode_frag frag;
817 ceph_choose_frag(ci, hash, &frag, &found);
819 if (mode == USE_ANY_MDS && frag.ndist > 0) {
822 /* choose a random replica */
823 get_random_bytes(&r, 1);
826 dout("choose_mds %p %llx.%llx "
827 "frag %u mds%d (%d/%d)\n",
828 inode, ceph_vinop(inode),
831 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
832 CEPH_MDS_STATE_ACTIVE)
836 /* since this file/dir wasn't known to be
837 * replicated, then we want to look for the
838 * authoritative mds. */
841 /* choose auth mds */
843 dout("choose_mds %p %llx.%llx "
844 "frag %u mds%d (auth)\n",
845 inode, ceph_vinop(inode), frag.frag, mds);
846 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
847 CEPH_MDS_STATE_ACTIVE)
853 spin_lock(&ci->i_ceph_lock);
855 if (mode == USE_AUTH_MDS)
856 cap = ci->i_auth_cap;
857 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
858 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
860 spin_unlock(&ci->i_ceph_lock);
864 mds = cap->session->s_mds;
865 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
866 inode, ceph_vinop(inode), mds,
867 cap == ci->i_auth_cap ? "auth " : "", cap);
868 spin_unlock(&ci->i_ceph_lock);
874 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
875 dout("choose_mds chose random mds%d\n", mds);
883 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
885 struct ceph_msg *msg;
886 struct ceph_mds_session_head *h;
888 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
891 pr_err("create_session_msg ENOMEM creating msg\n");
894 h = msg->front.iov_base;
895 h->op = cpu_to_le32(op);
896 h->seq = cpu_to_le64(seq);
901 static void encode_supported_features(void **p, void *end)
903 static const unsigned char bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
904 static const size_t count = ARRAY_SIZE(bits);
908 size_t size = ((size_t)bits[count - 1] + 64) / 64 * 8;
910 BUG_ON(*p + 4 + size > end);
911 ceph_encode_32(p, size);
913 for (i = 0; i < count; i++)
914 ((unsigned char*)(*p))[i / 8] |= 1 << (bits[i] % 8);
917 BUG_ON(*p + 4 > end);
918 ceph_encode_32(p, 0);
923 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
924 * to include additional client metadata fields.
926 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
928 struct ceph_msg *msg;
929 struct ceph_mds_session_head *h;
932 int metadata_key_count = 0;
933 struct ceph_options *opt = mdsc->fsc->client->options;
934 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
937 const char* metadata[][2] = {
938 {"hostname", mdsc->nodename},
939 {"kernel_version", init_utsname()->release},
940 {"entity_id", opt->name ? : ""},
941 {"root", fsopt->server_path ? : "/"},
945 /* Calculate serialized length of metadata */
946 extra_bytes = 4; /* map length */
947 for (i = 0; metadata[i][0]; ++i) {
948 extra_bytes += 8 + strlen(metadata[i][0]) +
949 strlen(metadata[i][1]);
950 metadata_key_count++;
952 /* supported feature */
953 extra_bytes += 4 + 8;
955 /* Allocate the message */
956 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
959 pr_err("create_session_msg ENOMEM creating msg\n");
962 p = msg->front.iov_base;
963 end = p + msg->front.iov_len;
966 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
967 h->seq = cpu_to_le64(seq);
970 * Serialize client metadata into waiting buffer space, using
971 * the format that userspace expects for map<string, string>
973 * ClientSession messages with metadata are v2
975 msg->hdr.version = cpu_to_le16(3);
976 msg->hdr.compat_version = cpu_to_le16(1);
978 /* The write pointer, following the session_head structure */
981 /* Number of entries in the map */
982 ceph_encode_32(&p, metadata_key_count);
984 /* Two length-prefixed strings for each entry in the map */
985 for (i = 0; metadata[i][0]; ++i) {
986 size_t const key_len = strlen(metadata[i][0]);
987 size_t const val_len = strlen(metadata[i][1]);
989 ceph_encode_32(&p, key_len);
990 memcpy(p, metadata[i][0], key_len);
992 ceph_encode_32(&p, val_len);
993 memcpy(p, metadata[i][1], val_len);
997 encode_supported_features(&p, end);
998 msg->front.iov_len = p - msg->front.iov_base;
999 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1005 * send session open request.
1007 * called under mdsc->mutex
1009 static int __open_session(struct ceph_mds_client *mdsc,
1010 struct ceph_mds_session *session)
1012 struct ceph_msg *msg;
1014 int mds = session->s_mds;
1016 /* wait for mds to go active? */
1017 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1018 dout("open_session to mds%d (%s)\n", mds,
1019 ceph_mds_state_name(mstate));
1020 session->s_state = CEPH_MDS_SESSION_OPENING;
1021 session->s_renew_requested = jiffies;
1023 /* send connect message */
1024 msg = create_session_open_msg(mdsc, session->s_seq);
1027 ceph_con_send(&session->s_con, msg);
1032 * open sessions for any export targets for the given mds
1034 * called under mdsc->mutex
1036 static struct ceph_mds_session *
1037 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1039 struct ceph_mds_session *session;
1041 session = __ceph_lookup_mds_session(mdsc, target);
1043 session = register_session(mdsc, target);
1044 if (IS_ERR(session))
1047 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1048 session->s_state == CEPH_MDS_SESSION_CLOSING)
1049 __open_session(mdsc, session);
1054 struct ceph_mds_session *
1055 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1057 struct ceph_mds_session *session;
1059 dout("open_export_target_session to mds%d\n", target);
1061 mutex_lock(&mdsc->mutex);
1062 session = __open_export_target_session(mdsc, target);
1063 mutex_unlock(&mdsc->mutex);
1068 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1069 struct ceph_mds_session *session)
1071 struct ceph_mds_info *mi;
1072 struct ceph_mds_session *ts;
1073 int i, mds = session->s_mds;
1075 if (mds >= mdsc->mdsmap->m_num_mds)
1078 mi = &mdsc->mdsmap->m_info[mds];
1079 dout("open_export_target_sessions for mds%d (%d targets)\n",
1080 session->s_mds, mi->num_export_targets);
1082 for (i = 0; i < mi->num_export_targets; i++) {
1083 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1085 ceph_put_mds_session(ts);
1089 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1090 struct ceph_mds_session *session)
1092 mutex_lock(&mdsc->mutex);
1093 __open_export_target_sessions(mdsc, session);
1094 mutex_unlock(&mdsc->mutex);
1101 static void detach_cap_releases(struct ceph_mds_session *session,
1102 struct list_head *target)
1104 lockdep_assert_held(&session->s_cap_lock);
1106 list_splice_init(&session->s_cap_releases, target);
1107 session->s_num_cap_releases = 0;
1108 dout("dispose_cap_releases mds%d\n", session->s_mds);
1111 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1112 struct list_head *dispose)
1114 while (!list_empty(dispose)) {
1115 struct ceph_cap *cap;
1116 /* zero out the in-progress message */
1117 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1118 list_del(&cap->session_caps);
1119 ceph_put_cap(mdsc, cap);
1123 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1124 struct ceph_mds_session *session)
1126 struct ceph_mds_request *req;
1129 dout("cleanup_session_requests mds%d\n", session->s_mds);
1130 mutex_lock(&mdsc->mutex);
1131 while (!list_empty(&session->s_unsafe)) {
1132 req = list_first_entry(&session->s_unsafe,
1133 struct ceph_mds_request, r_unsafe_item);
1134 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1136 __unregister_request(mdsc, req);
1138 /* zero r_attempts, so kick_requests() will re-send requests */
1139 p = rb_first(&mdsc->request_tree);
1141 req = rb_entry(p, struct ceph_mds_request, r_node);
1143 if (req->r_session &&
1144 req->r_session->s_mds == session->s_mds)
1145 req->r_attempts = 0;
1147 mutex_unlock(&mdsc->mutex);
1151 * Helper to safely iterate over all caps associated with a session, with
1152 * special care taken to handle a racing __ceph_remove_cap().
1154 * Caller must hold session s_mutex.
1156 static int iterate_session_caps(struct ceph_mds_session *session,
1157 int (*cb)(struct inode *, struct ceph_cap *,
1160 struct list_head *p;
1161 struct ceph_cap *cap;
1162 struct inode *inode, *last_inode = NULL;
1163 struct ceph_cap *old_cap = NULL;
1166 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1167 spin_lock(&session->s_cap_lock);
1168 p = session->s_caps.next;
1169 while (p != &session->s_caps) {
1170 cap = list_entry(p, struct ceph_cap, session_caps);
1171 inode = igrab(&cap->ci->vfs_inode);
1176 session->s_cap_iterator = cap;
1177 spin_unlock(&session->s_cap_lock);
1184 ceph_put_cap(session->s_mdsc, old_cap);
1188 ret = cb(inode, cap, arg);
1191 spin_lock(&session->s_cap_lock);
1194 dout("iterate_session_caps finishing cap %p removal\n",
1196 BUG_ON(cap->session != session);
1197 cap->session = NULL;
1198 list_del_init(&cap->session_caps);
1199 session->s_nr_caps--;
1200 if (cap->queue_release) {
1201 list_add_tail(&cap->session_caps,
1202 &session->s_cap_releases);
1203 session->s_num_cap_releases++;
1205 old_cap = cap; /* put_cap it w/o locks held */
1213 session->s_cap_iterator = NULL;
1214 spin_unlock(&session->s_cap_lock);
1218 ceph_put_cap(session->s_mdsc, old_cap);
1223 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1226 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1227 struct ceph_inode_info *ci = ceph_inode(inode);
1228 LIST_HEAD(to_remove);
1230 bool invalidate = false;
1232 dout("removing cap %p, ci is %p, inode is %p\n",
1233 cap, ci, &ci->vfs_inode);
1234 spin_lock(&ci->i_ceph_lock);
1235 if (cap->mds_wanted | cap->issued)
1236 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1237 __ceph_remove_cap(cap, false);
1238 if (!ci->i_auth_cap) {
1239 struct ceph_cap_flush *cf;
1240 struct ceph_mds_client *mdsc = fsc->mdsc;
1242 if (ci->i_wrbuffer_ref > 0 &&
1243 READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
1246 while (!list_empty(&ci->i_cap_flush_list)) {
1247 cf = list_first_entry(&ci->i_cap_flush_list,
1248 struct ceph_cap_flush, i_list);
1249 list_move(&cf->i_list, &to_remove);
1252 spin_lock(&mdsc->cap_dirty_lock);
1254 list_for_each_entry(cf, &to_remove, i_list)
1255 list_del(&cf->g_list);
1257 if (!list_empty(&ci->i_dirty_item)) {
1258 pr_warn_ratelimited(
1259 " dropping dirty %s state for %p %lld\n",
1260 ceph_cap_string(ci->i_dirty_caps),
1261 inode, ceph_ino(inode));
1262 ci->i_dirty_caps = 0;
1263 list_del_init(&ci->i_dirty_item);
1266 if (!list_empty(&ci->i_flushing_item)) {
1267 pr_warn_ratelimited(
1268 " dropping dirty+flushing %s state for %p %lld\n",
1269 ceph_cap_string(ci->i_flushing_caps),
1270 inode, ceph_ino(inode));
1271 ci->i_flushing_caps = 0;
1272 list_del_init(&ci->i_flushing_item);
1273 mdsc->num_cap_flushing--;
1276 spin_unlock(&mdsc->cap_dirty_lock);
1278 if (atomic_read(&ci->i_filelock_ref) > 0) {
1279 /* make further file lock syscall return -EIO */
1280 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1281 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1282 inode, ceph_ino(inode));
1285 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1286 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1287 ci->i_prealloc_cap_flush = NULL;
1290 spin_unlock(&ci->i_ceph_lock);
1291 while (!list_empty(&to_remove)) {
1292 struct ceph_cap_flush *cf;
1293 cf = list_first_entry(&to_remove,
1294 struct ceph_cap_flush, i_list);
1295 list_del(&cf->i_list);
1296 ceph_free_cap_flush(cf);
1299 wake_up_all(&ci->i_cap_wq);
1301 ceph_queue_invalidate(inode);
1308 * caller must hold session s_mutex
1310 static void remove_session_caps(struct ceph_mds_session *session)
1312 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1313 struct super_block *sb = fsc->sb;
1316 dout("remove_session_caps on %p\n", session);
1317 iterate_session_caps(session, remove_session_caps_cb, fsc);
1319 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1321 spin_lock(&session->s_cap_lock);
1322 if (session->s_nr_caps > 0) {
1323 struct inode *inode;
1324 struct ceph_cap *cap, *prev = NULL;
1325 struct ceph_vino vino;
1327 * iterate_session_caps() skips inodes that are being
1328 * deleted, we need to wait until deletions are complete.
1329 * __wait_on_freeing_inode() is designed for the job,
1330 * but it is not exported, so use lookup inode function
1333 while (!list_empty(&session->s_caps)) {
1334 cap = list_entry(session->s_caps.next,
1335 struct ceph_cap, session_caps);
1339 vino = cap->ci->i_vino;
1340 spin_unlock(&session->s_cap_lock);
1342 inode = ceph_find_inode(sb, vino);
1345 spin_lock(&session->s_cap_lock);
1349 // drop cap expires and unlock s_cap_lock
1350 detach_cap_releases(session, &dispose);
1352 BUG_ON(session->s_nr_caps > 0);
1353 BUG_ON(!list_empty(&session->s_cap_flushing));
1354 spin_unlock(&session->s_cap_lock);
1355 dispose_cap_releases(session->s_mdsc, &dispose);
1365 * wake up any threads waiting on this session's caps. if the cap is
1366 * old (didn't get renewed on the client reconnect), remove it now.
1368 * caller must hold s_mutex.
1370 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1373 struct ceph_inode_info *ci = ceph_inode(inode);
1374 unsigned long ev = (unsigned long)arg;
1376 if (ev == RECONNECT) {
1377 spin_lock(&ci->i_ceph_lock);
1378 ci->i_wanted_max_size = 0;
1379 ci->i_requested_max_size = 0;
1380 spin_unlock(&ci->i_ceph_lock);
1381 } else if (ev == RENEWCAPS) {
1382 if (cap->cap_gen < cap->session->s_cap_gen) {
1383 /* mds did not re-issue stale cap */
1384 spin_lock(&ci->i_ceph_lock);
1385 cap->issued = cap->implemented = CEPH_CAP_PIN;
1386 /* make sure mds knows what we want */
1387 if (__ceph_caps_file_wanted(ci) & ~cap->mds_wanted)
1388 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1389 spin_unlock(&ci->i_ceph_lock);
1391 } else if (ev == FORCE_RO) {
1393 wake_up_all(&ci->i_cap_wq);
1397 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1399 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1400 iterate_session_caps(session, wake_up_session_cb,
1401 (void *)(unsigned long)ev);
1405 * Send periodic message to MDS renewing all currently held caps. The
1406 * ack will reset the expiration for all caps from this session.
1408 * caller holds s_mutex
1410 static int send_renew_caps(struct ceph_mds_client *mdsc,
1411 struct ceph_mds_session *session)
1413 struct ceph_msg *msg;
1416 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1417 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1418 pr_info("mds%d caps stale\n", session->s_mds);
1419 session->s_renew_requested = jiffies;
1421 /* do not try to renew caps until a recovering mds has reconnected
1422 * with its clients. */
1423 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1424 if (state < CEPH_MDS_STATE_RECONNECT) {
1425 dout("send_renew_caps ignoring mds%d (%s)\n",
1426 session->s_mds, ceph_mds_state_name(state));
1430 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1431 ceph_mds_state_name(state));
1432 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1433 ++session->s_renew_seq);
1436 ceph_con_send(&session->s_con, msg);
1440 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1441 struct ceph_mds_session *session, u64 seq)
1443 struct ceph_msg *msg;
1445 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1446 session->s_mds, ceph_session_state_name(session->s_state), seq);
1447 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1450 ceph_con_send(&session->s_con, msg);
1456 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1458 * Called under session->s_mutex
1460 static void renewed_caps(struct ceph_mds_client *mdsc,
1461 struct ceph_mds_session *session, int is_renew)
1466 spin_lock(&session->s_cap_lock);
1467 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1469 session->s_cap_ttl = session->s_renew_requested +
1470 mdsc->mdsmap->m_session_timeout*HZ;
1473 if (time_before(jiffies, session->s_cap_ttl)) {
1474 pr_info("mds%d caps renewed\n", session->s_mds);
1477 pr_info("mds%d caps still stale\n", session->s_mds);
1480 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1481 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1482 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1483 spin_unlock(&session->s_cap_lock);
1486 wake_up_session_caps(session, RENEWCAPS);
1490 * send a session close request
1492 static int request_close_session(struct ceph_mds_client *mdsc,
1493 struct ceph_mds_session *session)
1495 struct ceph_msg *msg;
1497 dout("request_close_session mds%d state %s seq %lld\n",
1498 session->s_mds, ceph_session_state_name(session->s_state),
1500 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1503 ceph_con_send(&session->s_con, msg);
1508 * Called with s_mutex held.
1510 static int __close_session(struct ceph_mds_client *mdsc,
1511 struct ceph_mds_session *session)
1513 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1515 session->s_state = CEPH_MDS_SESSION_CLOSING;
1516 return request_close_session(mdsc, session);
1519 static bool drop_negative_children(struct dentry *dentry)
1521 struct dentry *child;
1522 bool all_negative = true;
1524 if (!d_is_dir(dentry))
1527 spin_lock(&dentry->d_lock);
1528 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1529 if (d_really_is_positive(child)) {
1530 all_negative = false;
1534 spin_unlock(&dentry->d_lock);
1537 shrink_dcache_parent(dentry);
1539 return all_negative;
1543 * Trim old(er) caps.
1545 * Because we can't cache an inode without one or more caps, we do
1546 * this indirectly: if a cap is unused, we prune its aliases, at which
1547 * point the inode will hopefully get dropped to.
1549 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1550 * memory pressure from the MDS, though, so it needn't be perfect.
1552 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1554 struct ceph_mds_session *session = arg;
1555 struct ceph_inode_info *ci = ceph_inode(inode);
1556 int used, wanted, oissued, mine;
1558 if (session->s_trim_caps <= 0)
1561 spin_lock(&ci->i_ceph_lock);
1562 mine = cap->issued | cap->implemented;
1563 used = __ceph_caps_used(ci);
1564 wanted = __ceph_caps_file_wanted(ci);
1565 oissued = __ceph_caps_issued_other(ci, cap);
1567 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1568 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1569 ceph_cap_string(used), ceph_cap_string(wanted));
1570 if (cap == ci->i_auth_cap) {
1571 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1572 !list_empty(&ci->i_cap_snaps))
1574 if ((used | wanted) & CEPH_CAP_ANY_WR)
1576 /* Note: it's possible that i_filelock_ref becomes non-zero
1577 * after dropping auth caps. It doesn't hurt because reply
1578 * of lock mds request will re-add auth caps. */
1579 if (atomic_read(&ci->i_filelock_ref) > 0)
1582 /* The inode has cached pages, but it's no longer used.
1583 * we can safely drop it */
1584 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1585 !(oissued & CEPH_CAP_FILE_CACHE)) {
1589 if ((used | wanted) & ~oissued & mine)
1590 goto out; /* we need these caps */
1593 /* we aren't the only cap.. just remove us */
1594 __ceph_remove_cap(cap, true);
1595 session->s_trim_caps--;
1597 struct dentry *dentry;
1598 /* try dropping referring dentries */
1599 spin_unlock(&ci->i_ceph_lock);
1600 dentry = d_find_any_alias(inode);
1601 if (dentry && drop_negative_children(dentry)) {
1604 d_prune_aliases(inode);
1605 count = atomic_read(&inode->i_count);
1607 session->s_trim_caps--;
1608 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1617 spin_unlock(&ci->i_ceph_lock);
1622 * Trim session cap count down to some max number.
1624 int ceph_trim_caps(struct ceph_mds_client *mdsc,
1625 struct ceph_mds_session *session,
1628 int trim_caps = session->s_nr_caps - max_caps;
1630 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1631 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1632 if (trim_caps > 0) {
1633 session->s_trim_caps = trim_caps;
1634 iterate_session_caps(session, trim_caps_cb, session);
1635 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1636 session->s_mds, session->s_nr_caps, max_caps,
1637 trim_caps - session->s_trim_caps);
1638 session->s_trim_caps = 0;
1641 ceph_send_cap_releases(mdsc, session);
1645 static int check_caps_flush(struct ceph_mds_client *mdsc,
1650 spin_lock(&mdsc->cap_dirty_lock);
1651 if (!list_empty(&mdsc->cap_flush_list)) {
1652 struct ceph_cap_flush *cf =
1653 list_first_entry(&mdsc->cap_flush_list,
1654 struct ceph_cap_flush, g_list);
1655 if (cf->tid <= want_flush_tid) {
1656 dout("check_caps_flush still flushing tid "
1657 "%llu <= %llu\n", cf->tid, want_flush_tid);
1661 spin_unlock(&mdsc->cap_dirty_lock);
1666 * flush all dirty inode data to disk.
1668 * returns true if we've flushed through want_flush_tid
1670 static void wait_caps_flush(struct ceph_mds_client *mdsc,
1673 dout("check_caps_flush want %llu\n", want_flush_tid);
1675 wait_event(mdsc->cap_flushing_wq,
1676 check_caps_flush(mdsc, want_flush_tid));
1678 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
1682 * called under s_mutex
1684 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1685 struct ceph_mds_session *session)
1687 struct ceph_msg *msg = NULL;
1688 struct ceph_mds_cap_release *head;
1689 struct ceph_mds_cap_item *item;
1690 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
1691 struct ceph_cap *cap;
1692 LIST_HEAD(tmp_list);
1693 int num_cap_releases;
1694 __le32 barrier, *cap_barrier;
1696 down_read(&osdc->lock);
1697 barrier = cpu_to_le32(osdc->epoch_barrier);
1698 up_read(&osdc->lock);
1700 spin_lock(&session->s_cap_lock);
1702 list_splice_init(&session->s_cap_releases, &tmp_list);
1703 num_cap_releases = session->s_num_cap_releases;
1704 session->s_num_cap_releases = 0;
1705 spin_unlock(&session->s_cap_lock);
1707 while (!list_empty(&tmp_list)) {
1709 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1710 PAGE_SIZE, GFP_NOFS, false);
1713 head = msg->front.iov_base;
1714 head->num = cpu_to_le32(0);
1715 msg->front.iov_len = sizeof(*head);
1717 msg->hdr.version = cpu_to_le16(2);
1718 msg->hdr.compat_version = cpu_to_le16(1);
1721 cap = list_first_entry(&tmp_list, struct ceph_cap,
1723 list_del(&cap->session_caps);
1726 head = msg->front.iov_base;
1727 le32_add_cpu(&head->num, 1);
1728 item = msg->front.iov_base + msg->front.iov_len;
1729 item->ino = cpu_to_le64(cap->cap_ino);
1730 item->cap_id = cpu_to_le64(cap->cap_id);
1731 item->migrate_seq = cpu_to_le32(cap->mseq);
1732 item->seq = cpu_to_le32(cap->issue_seq);
1733 msg->front.iov_len += sizeof(*item);
1735 ceph_put_cap(mdsc, cap);
1737 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1738 // Append cap_barrier field
1739 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1740 *cap_barrier = barrier;
1741 msg->front.iov_len += sizeof(*cap_barrier);
1743 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1744 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1745 ceph_con_send(&session->s_con, msg);
1750 BUG_ON(num_cap_releases != 0);
1752 spin_lock(&session->s_cap_lock);
1753 if (!list_empty(&session->s_cap_releases))
1755 spin_unlock(&session->s_cap_lock);
1758 // Append cap_barrier field
1759 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1760 *cap_barrier = barrier;
1761 msg->front.iov_len += sizeof(*cap_barrier);
1763 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1764 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1765 ceph_con_send(&session->s_con, msg);
1769 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1771 spin_lock(&session->s_cap_lock);
1772 list_splice(&tmp_list, &session->s_cap_releases);
1773 session->s_num_cap_releases += num_cap_releases;
1774 spin_unlock(&session->s_cap_lock);
1781 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1784 struct ceph_inode_info *ci = ceph_inode(dir);
1785 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1786 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1787 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
1788 int order, num_entries;
1790 spin_lock(&ci->i_ceph_lock);
1791 num_entries = ci->i_files + ci->i_subdirs;
1792 spin_unlock(&ci->i_ceph_lock);
1793 num_entries = max(num_entries, 1);
1794 num_entries = min(num_entries, opt->max_readdir);
1796 order = get_order(size * num_entries);
1797 while (order >= 0) {
1798 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1801 if (rinfo->dir_entries)
1805 if (!rinfo->dir_entries)
1808 num_entries = (PAGE_SIZE << order) / size;
1809 num_entries = min(num_entries, opt->max_readdir);
1811 rinfo->dir_buf_size = PAGE_SIZE << order;
1812 req->r_num_caps = num_entries + 1;
1813 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1814 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1819 * Create an mds request.
1821 struct ceph_mds_request *
1822 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1824 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1825 struct timespec64 ts;
1828 return ERR_PTR(-ENOMEM);
1830 mutex_init(&req->r_fill_mutex);
1832 req->r_started = jiffies;
1833 req->r_resend_mds = -1;
1834 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1835 INIT_LIST_HEAD(&req->r_unsafe_target_item);
1837 kref_init(&req->r_kref);
1838 RB_CLEAR_NODE(&req->r_node);
1839 INIT_LIST_HEAD(&req->r_wait);
1840 init_completion(&req->r_completion);
1841 init_completion(&req->r_safe_completion);
1842 INIT_LIST_HEAD(&req->r_unsafe_item);
1844 ktime_get_coarse_real_ts64(&ts);
1845 req->r_stamp = timespec64_trunc(ts, mdsc->fsc->sb->s_time_gran);
1848 req->r_direct_mode = mode;
1853 * return oldest (lowest) request, tid in request tree, 0 if none.
1855 * called under mdsc->mutex.
1857 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1859 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1861 return rb_entry(rb_first(&mdsc->request_tree),
1862 struct ceph_mds_request, r_node);
1865 static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1867 return mdsc->oldest_tid;
1871 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1872 * on build_path_from_dentry in fs/cifs/dir.c.
1874 * If @stop_on_nosnap, generate path relative to the first non-snapped
1877 * Encode hidden .snap dirs as a double /, i.e.
1878 * foo/.snap/bar -> foo//bar
1880 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1883 struct dentry *temp;
1889 return ERR_PTR(-EINVAL);
1893 seq = read_seqbegin(&rename_lock);
1895 for (temp = dentry; !IS_ROOT(temp);) {
1896 struct inode *inode = d_inode(temp);
1897 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1898 len++; /* slash only */
1899 else if (stop_on_nosnap && inode &&
1900 ceph_snap(inode) == CEPH_NOSNAP)
1903 len += 1 + temp->d_name.len;
1904 temp = temp->d_parent;
1908 len--; /* no leading '/' */
1910 path = kmalloc(len+1, GFP_NOFS);
1912 return ERR_PTR(-ENOMEM);
1914 path[pos] = 0; /* trailing null */
1916 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1917 struct inode *inode;
1919 spin_lock(&temp->d_lock);
1920 inode = d_inode(temp);
1921 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1922 dout("build_path path+%d: %p SNAPDIR\n",
1924 } else if (stop_on_nosnap && inode &&
1925 ceph_snap(inode) == CEPH_NOSNAP) {
1926 spin_unlock(&temp->d_lock);
1929 pos -= temp->d_name.len;
1931 spin_unlock(&temp->d_lock);
1934 strncpy(path + pos, temp->d_name.name,
1937 spin_unlock(&temp->d_lock);
1940 temp = temp->d_parent;
1943 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1944 pr_err("build_path did not end path lookup where "
1945 "expected, namelen is %d, pos is %d\n", len, pos);
1946 /* presumably this is only possible if racing with a
1947 rename of one of the parent directories (we can not
1948 lock the dentries above us to prevent this, but
1949 retrying should be harmless) */
1954 *base = ceph_ino(d_inode(temp));
1956 dout("build_path on %p %d built %llx '%.*s'\n",
1957 dentry, d_count(dentry), *base, len, path);
1961 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
1962 const char **ppath, int *ppathlen, u64 *pino,
1969 dir = d_inode_rcu(dentry->d_parent);
1970 if (dir && ceph_snap(dir) == CEPH_NOSNAP) {
1971 *pino = ceph_ino(dir);
1973 *ppath = dentry->d_name.name;
1974 *ppathlen = dentry->d_name.len;
1978 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1980 return PTR_ERR(path);
1986 static int build_inode_path(struct inode *inode,
1987 const char **ppath, int *ppathlen, u64 *pino,
1990 struct dentry *dentry;
1993 if (ceph_snap(inode) == CEPH_NOSNAP) {
1994 *pino = ceph_ino(inode);
1998 dentry = d_find_alias(inode);
1999 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2002 return PTR_ERR(path);
2009 * request arguments may be specified via an inode *, a dentry *, or
2010 * an explicit ino+path.
2012 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2013 struct inode *rdiri, const char *rpath,
2014 u64 rino, const char **ppath, int *pathlen,
2015 u64 *ino, int *freepath)
2020 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2021 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2023 } else if (rdentry) {
2024 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2026 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2028 } else if (rpath || rino) {
2031 *pathlen = rpath ? strlen(rpath) : 0;
2032 dout(" path %.*s\n", *pathlen, rpath);
2039 * called under mdsc->mutex
2041 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
2042 struct ceph_mds_request *req,
2043 int mds, bool drop_cap_releases)
2045 struct ceph_msg *msg;
2046 struct ceph_mds_request_head *head;
2047 const char *path1 = NULL;
2048 const char *path2 = NULL;
2049 u64 ino1 = 0, ino2 = 0;
2050 int pathlen1 = 0, pathlen2 = 0;
2051 int freepath1 = 0, freepath2 = 0;
2057 ret = set_request_path_attr(req->r_inode, req->r_dentry,
2058 req->r_parent, req->r_path1, req->r_ino1.ino,
2059 &path1, &pathlen1, &ino1, &freepath1);
2065 ret = set_request_path_attr(NULL, req->r_old_dentry,
2066 req->r_old_dentry_dir,
2067 req->r_path2, req->r_ino2.ino,
2068 &path2, &pathlen2, &ino2, &freepath2);
2074 len = sizeof(*head) +
2075 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2076 sizeof(struct ceph_timespec);
2078 /* calculate (max) length for cap releases */
2079 len += sizeof(struct ceph_mds_request_release) *
2080 (!!req->r_inode_drop + !!req->r_dentry_drop +
2081 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2082 if (req->r_dentry_drop)
2083 len += req->r_dentry->d_name.len;
2084 if (req->r_old_dentry_drop)
2085 len += req->r_old_dentry->d_name.len;
2087 msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2089 msg = ERR_PTR(-ENOMEM);
2093 msg->hdr.version = cpu_to_le16(2);
2094 msg->hdr.tid = cpu_to_le64(req->r_tid);
2096 head = msg->front.iov_base;
2097 p = msg->front.iov_base + sizeof(*head);
2098 end = msg->front.iov_base + msg->front.iov_len;
2100 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2101 head->op = cpu_to_le32(req->r_op);
2102 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2103 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2104 head->args = req->r_args;
2106 ceph_encode_filepath(&p, end, ino1, path1);
2107 ceph_encode_filepath(&p, end, ino2, path2);
2109 /* make note of release offset, in case we need to replay */
2110 req->r_request_release_offset = p - msg->front.iov_base;
2114 if (req->r_inode_drop)
2115 releases += ceph_encode_inode_release(&p,
2116 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2117 mds, req->r_inode_drop, req->r_inode_unless, 0);
2118 if (req->r_dentry_drop)
2119 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2120 req->r_parent, mds, req->r_dentry_drop,
2121 req->r_dentry_unless);
2122 if (req->r_old_dentry_drop)
2123 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2124 req->r_old_dentry_dir, mds,
2125 req->r_old_dentry_drop,
2126 req->r_old_dentry_unless);
2127 if (req->r_old_inode_drop)
2128 releases += ceph_encode_inode_release(&p,
2129 d_inode(req->r_old_dentry),
2130 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2132 if (drop_cap_releases) {
2134 p = msg->front.iov_base + req->r_request_release_offset;
2137 head->num_releases = cpu_to_le16(releases);
2141 struct ceph_timespec ts;
2142 ceph_encode_timespec64(&ts, &req->r_stamp);
2143 ceph_encode_copy(&p, &ts, sizeof(ts));
2147 msg->front.iov_len = p - msg->front.iov_base;
2148 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2150 if (req->r_pagelist) {
2151 struct ceph_pagelist *pagelist = req->r_pagelist;
2152 ceph_msg_data_add_pagelist(msg, pagelist);
2153 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2155 msg->hdr.data_len = 0;
2158 msg->hdr.data_off = cpu_to_le16(0);
2162 kfree((char *)path2);
2165 kfree((char *)path1);
2171 * called under mdsc->mutex if error, under no mutex if
2174 static void complete_request(struct ceph_mds_client *mdsc,
2175 struct ceph_mds_request *req)
2177 if (req->r_callback)
2178 req->r_callback(mdsc, req);
2180 complete_all(&req->r_completion);
2184 * called under mdsc->mutex
2186 static int __prepare_send_request(struct ceph_mds_client *mdsc,
2187 struct ceph_mds_request *req,
2188 int mds, bool drop_cap_releases)
2190 struct ceph_mds_request_head *rhead;
2191 struct ceph_msg *msg;
2196 struct ceph_cap *cap =
2197 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2200 req->r_sent_on_mseq = cap->mseq;
2202 req->r_sent_on_mseq = -1;
2204 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2205 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2207 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2210 * Replay. Do not regenerate message (and rebuild
2211 * paths, etc.); just use the original message.
2212 * Rebuilding paths will break for renames because
2213 * d_move mangles the src name.
2215 msg = req->r_request;
2216 rhead = msg->front.iov_base;
2218 flags = le32_to_cpu(rhead->flags);
2219 flags |= CEPH_MDS_FLAG_REPLAY;
2220 rhead->flags = cpu_to_le32(flags);
2222 if (req->r_target_inode)
2223 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2225 rhead->num_retry = req->r_attempts - 1;
2227 /* remove cap/dentry releases from message */
2228 rhead->num_releases = 0;
2231 p = msg->front.iov_base + req->r_request_release_offset;
2233 struct ceph_timespec ts;
2234 ceph_encode_timespec64(&ts, &req->r_stamp);
2235 ceph_encode_copy(&p, &ts, sizeof(ts));
2238 msg->front.iov_len = p - msg->front.iov_base;
2239 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2243 if (req->r_request) {
2244 ceph_msg_put(req->r_request);
2245 req->r_request = NULL;
2247 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2249 req->r_err = PTR_ERR(msg);
2250 return PTR_ERR(msg);
2252 req->r_request = msg;
2254 rhead = msg->front.iov_base;
2255 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2256 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2257 flags |= CEPH_MDS_FLAG_REPLAY;
2259 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2260 rhead->flags = cpu_to_le32(flags);
2261 rhead->num_fwd = req->r_num_fwd;
2262 rhead->num_retry = req->r_attempts - 1;
2265 dout(" r_parent = %p\n", req->r_parent);
2270 * send request, or put it on the appropriate wait list.
2272 static void __do_request(struct ceph_mds_client *mdsc,
2273 struct ceph_mds_request *req)
2275 struct ceph_mds_session *session = NULL;
2279 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2280 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2281 __unregister_request(mdsc, req);
2285 if (req->r_timeout &&
2286 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2287 dout("do_request timed out\n");
2291 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2292 dout("do_request forced umount\n");
2296 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2297 if (mdsc->mdsmap_err) {
2298 err = mdsc->mdsmap_err;
2299 dout("do_request mdsmap err %d\n", err);
2302 if (mdsc->mdsmap->m_epoch == 0) {
2303 dout("do_request no mdsmap, waiting for map\n");
2304 list_add(&req->r_wait, &mdsc->waiting_for_map);
2307 if (!(mdsc->fsc->mount_options->flags &
2308 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2309 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2311 pr_info("probably no mds server is up\n");
2316 put_request_session(req);
2318 mds = __choose_mds(mdsc, req);
2320 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2321 dout("do_request no mds or not active, waiting for map\n");
2322 list_add(&req->r_wait, &mdsc->waiting_for_map);
2326 /* get, open session */
2327 session = __ceph_lookup_mds_session(mdsc, mds);
2329 session = register_session(mdsc, mds);
2330 if (IS_ERR(session)) {
2331 err = PTR_ERR(session);
2335 req->r_session = get_session(session);
2337 dout("do_request mds%d session %p state %s\n", mds, session,
2338 ceph_session_state_name(session->s_state));
2339 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2340 session->s_state != CEPH_MDS_SESSION_HUNG) {
2341 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2345 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2346 session->s_state == CEPH_MDS_SESSION_CLOSING)
2347 __open_session(mdsc, session);
2348 list_add(&req->r_wait, &session->s_waiting);
2353 req->r_resend_mds = -1; /* forget any previous mds hint */
2355 if (req->r_request_started == 0) /* note request start time */
2356 req->r_request_started = jiffies;
2358 err = __prepare_send_request(mdsc, req, mds, false);
2360 ceph_msg_get(req->r_request);
2361 ceph_con_send(&session->s_con, req->r_request);
2365 ceph_put_mds_session(session);
2368 dout("__do_request early error %d\n", err);
2370 complete_request(mdsc, req);
2371 __unregister_request(mdsc, req);
2377 * called under mdsc->mutex
2379 static void __wake_requests(struct ceph_mds_client *mdsc,
2380 struct list_head *head)
2382 struct ceph_mds_request *req;
2383 LIST_HEAD(tmp_list);
2385 list_splice_init(head, &tmp_list);
2387 while (!list_empty(&tmp_list)) {
2388 req = list_entry(tmp_list.next,
2389 struct ceph_mds_request, r_wait);
2390 list_del_init(&req->r_wait);
2391 dout(" wake request %p tid %llu\n", req, req->r_tid);
2392 __do_request(mdsc, req);
2397 * Wake up threads with requests pending for @mds, so that they can
2398 * resubmit their requests to a possibly different mds.
2400 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2402 struct ceph_mds_request *req;
2403 struct rb_node *p = rb_first(&mdsc->request_tree);
2405 dout("kick_requests mds%d\n", mds);
2407 req = rb_entry(p, struct ceph_mds_request, r_node);
2409 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2411 if (req->r_attempts > 0)
2412 continue; /* only new requests */
2413 if (req->r_session &&
2414 req->r_session->s_mds == mds) {
2415 dout(" kicking tid %llu\n", req->r_tid);
2416 list_del_init(&req->r_wait);
2417 __do_request(mdsc, req);
2422 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2423 struct ceph_mds_request *req)
2425 dout("submit_request on %p\n", req);
2426 mutex_lock(&mdsc->mutex);
2427 __register_request(mdsc, req, NULL);
2428 __do_request(mdsc, req);
2429 mutex_unlock(&mdsc->mutex);
2433 * Synchrously perform an mds request. Take care of all of the
2434 * session setup, forwarding, retry details.
2436 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2438 struct ceph_mds_request *req)
2442 dout("do_request on %p\n", req);
2444 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2446 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2448 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
2449 if (req->r_old_dentry_dir)
2450 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2454 mutex_lock(&mdsc->mutex);
2455 __register_request(mdsc, req, dir);
2456 __do_request(mdsc, req);
2464 mutex_unlock(&mdsc->mutex);
2465 dout("do_request waiting\n");
2466 if (!req->r_timeout && req->r_wait_for_completion) {
2467 err = req->r_wait_for_completion(mdsc, req);
2469 long timeleft = wait_for_completion_killable_timeout(
2471 ceph_timeout_jiffies(req->r_timeout));
2475 err = -EIO; /* timed out */
2477 err = timeleft; /* killed */
2479 dout("do_request waited, got %d\n", err);
2480 mutex_lock(&mdsc->mutex);
2482 /* only abort if we didn't race with a real reply */
2483 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2484 err = le32_to_cpu(req->r_reply_info.head->result);
2485 } else if (err < 0) {
2486 dout("aborted request %lld with %d\n", req->r_tid, err);
2489 * ensure we aren't running concurrently with
2490 * ceph_fill_trace or ceph_readdir_prepopulate, which
2491 * rely on locks (dir mutex) held by our caller.
2493 mutex_lock(&req->r_fill_mutex);
2495 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
2496 mutex_unlock(&req->r_fill_mutex);
2498 if (req->r_parent &&
2499 (req->r_op & CEPH_MDS_OP_WRITE))
2500 ceph_invalidate_dir_request(req);
2506 mutex_unlock(&mdsc->mutex);
2507 dout("do_request %p done, result %d\n", req, err);
2512 * Invalidate dir's completeness, dentry lease state on an aborted MDS
2513 * namespace request.
2515 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2517 struct inode *dir = req->r_parent;
2518 struct inode *old_dir = req->r_old_dentry_dir;
2520 dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
2522 ceph_dir_clear_complete(dir);
2524 ceph_dir_clear_complete(old_dir);
2526 ceph_invalidate_dentry_lease(req->r_dentry);
2527 if (req->r_old_dentry)
2528 ceph_invalidate_dentry_lease(req->r_old_dentry);
2534 * We take the session mutex and parse and process the reply immediately.
2535 * This preserves the logical ordering of replies, capabilities, etc., sent
2536 * by the MDS as they are applied to our local cache.
2538 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2540 struct ceph_mds_client *mdsc = session->s_mdsc;
2541 struct ceph_mds_request *req;
2542 struct ceph_mds_reply_head *head = msg->front.iov_base;
2543 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
2544 struct ceph_snap_realm *realm;
2547 int mds = session->s_mds;
2549 if (msg->front.iov_len < sizeof(*head)) {
2550 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2555 /* get request, session */
2556 tid = le64_to_cpu(msg->hdr.tid);
2557 mutex_lock(&mdsc->mutex);
2558 req = lookup_get_request(mdsc, tid);
2560 dout("handle_reply on unknown tid %llu\n", tid);
2561 mutex_unlock(&mdsc->mutex);
2564 dout("handle_reply %p\n", req);
2566 /* correct session? */
2567 if (req->r_session != session) {
2568 pr_err("mdsc_handle_reply got %llu on session mds%d"
2569 " not mds%d\n", tid, session->s_mds,
2570 req->r_session ? req->r_session->s_mds : -1);
2571 mutex_unlock(&mdsc->mutex);
2576 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2577 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
2578 pr_warn("got a dup %s reply on %llu from mds%d\n",
2579 head->safe ? "safe" : "unsafe", tid, mds);
2580 mutex_unlock(&mdsc->mutex);
2583 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
2584 pr_warn("got unsafe after safe on %llu from mds%d\n",
2586 mutex_unlock(&mdsc->mutex);
2590 result = le32_to_cpu(head->result);
2594 * if we're not talking to the authority, send to them
2595 * if the authority has changed while we weren't looking,
2596 * send to new authority
2597 * Otherwise we just have to return an ESTALE
2599 if (result == -ESTALE) {
2600 dout("got ESTALE on request %llu\n", req->r_tid);
2601 req->r_resend_mds = -1;
2602 if (req->r_direct_mode != USE_AUTH_MDS) {
2603 dout("not using auth, setting for that now\n");
2604 req->r_direct_mode = USE_AUTH_MDS;
2605 __do_request(mdsc, req);
2606 mutex_unlock(&mdsc->mutex);
2609 int mds = __choose_mds(mdsc, req);
2610 if (mds >= 0 && mds != req->r_session->s_mds) {
2611 dout("but auth changed, so resending\n");
2612 __do_request(mdsc, req);
2613 mutex_unlock(&mdsc->mutex);
2617 dout("have to return ESTALE on request %llu\n", req->r_tid);
2622 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2623 __unregister_request(mdsc, req);
2625 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2627 * We already handled the unsafe response, now do the
2628 * cleanup. No need to examine the response; the MDS
2629 * doesn't include any result info in the safe
2630 * response. And even if it did, there is nothing
2631 * useful we could do with a revised return value.
2633 dout("got safe reply %llu, mds%d\n", tid, mds);
2635 /* last unsafe request during umount? */
2636 if (mdsc->stopping && !__get_oldest_req(mdsc))
2637 complete_all(&mdsc->safe_umount_waiters);
2638 mutex_unlock(&mdsc->mutex);
2642 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2643 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2644 if (req->r_unsafe_dir) {
2645 struct ceph_inode_info *ci =
2646 ceph_inode(req->r_unsafe_dir);
2647 spin_lock(&ci->i_unsafe_lock);
2648 list_add_tail(&req->r_unsafe_dir_item,
2649 &ci->i_unsafe_dirops);
2650 spin_unlock(&ci->i_unsafe_lock);
2654 dout("handle_reply tid %lld result %d\n", tid, result);
2655 rinfo = &req->r_reply_info;
2656 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2657 mutex_unlock(&mdsc->mutex);
2659 mutex_lock(&session->s_mutex);
2661 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2668 if (rinfo->snapblob_len) {
2669 down_write(&mdsc->snap_rwsem);
2670 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2671 rinfo->snapblob + rinfo->snapblob_len,
2672 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2674 downgrade_write(&mdsc->snap_rwsem);
2676 down_read(&mdsc->snap_rwsem);
2679 /* insert trace into our cache */
2680 mutex_lock(&req->r_fill_mutex);
2681 current->journal_info = req;
2682 err = ceph_fill_trace(mdsc->fsc->sb, req);
2684 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2685 req->r_op == CEPH_MDS_OP_LSSNAP))
2686 ceph_readdir_prepopulate(req, req->r_session);
2687 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2689 current->journal_info = NULL;
2690 mutex_unlock(&req->r_fill_mutex);
2692 up_read(&mdsc->snap_rwsem);
2694 ceph_put_snap_realm(mdsc, realm);
2696 if (err == 0 && req->r_target_inode &&
2697 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2698 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2699 spin_lock(&ci->i_unsafe_lock);
2700 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2701 spin_unlock(&ci->i_unsafe_lock);
2704 mutex_lock(&mdsc->mutex);
2705 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2709 req->r_reply = ceph_msg_get(msg);
2710 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
2713 dout("reply arrived after request %lld was aborted\n", tid);
2715 mutex_unlock(&mdsc->mutex);
2717 mutex_unlock(&session->s_mutex);
2719 /* kick calling process */
2720 complete_request(mdsc, req);
2722 ceph_mdsc_put_request(req);
2729 * handle mds notification that our request has been forwarded.
2731 static void handle_forward(struct ceph_mds_client *mdsc,
2732 struct ceph_mds_session *session,
2733 struct ceph_msg *msg)
2735 struct ceph_mds_request *req;
2736 u64 tid = le64_to_cpu(msg->hdr.tid);
2740 void *p = msg->front.iov_base;
2741 void *end = p + msg->front.iov_len;
2743 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2744 next_mds = ceph_decode_32(&p);
2745 fwd_seq = ceph_decode_32(&p);
2747 mutex_lock(&mdsc->mutex);
2748 req = lookup_get_request(mdsc, tid);
2750 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2751 goto out; /* dup reply? */
2754 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2755 dout("forward tid %llu aborted, unregistering\n", tid);
2756 __unregister_request(mdsc, req);
2757 } else if (fwd_seq <= req->r_num_fwd) {
2758 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2759 tid, next_mds, req->r_num_fwd, fwd_seq);
2761 /* resend. forward race not possible; mds would drop */
2762 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2764 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
2765 req->r_attempts = 0;
2766 req->r_num_fwd = fwd_seq;
2767 req->r_resend_mds = next_mds;
2768 put_request_session(req);
2769 __do_request(mdsc, req);
2771 ceph_mdsc_put_request(req);
2773 mutex_unlock(&mdsc->mutex);
2777 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2781 * handle a mds session control message
2783 static void handle_session(struct ceph_mds_session *session,
2784 struct ceph_msg *msg)
2786 struct ceph_mds_client *mdsc = session->s_mdsc;
2789 int mds = session->s_mds;
2790 struct ceph_mds_session_head *h = msg->front.iov_base;
2794 if (msg->front.iov_len < sizeof(*h))
2796 op = le32_to_cpu(h->op);
2797 seq = le64_to_cpu(h->seq);
2799 mutex_lock(&mdsc->mutex);
2800 if (op == CEPH_SESSION_CLOSE) {
2801 get_session(session);
2802 __unregister_session(mdsc, session);
2804 /* FIXME: this ttl calculation is generous */
2805 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2806 mutex_unlock(&mdsc->mutex);
2808 mutex_lock(&session->s_mutex);
2810 dout("handle_session mds%d %s %p state %s seq %llu\n",
2811 mds, ceph_session_op_name(op), session,
2812 ceph_session_state_name(session->s_state), seq);
2814 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2815 session->s_state = CEPH_MDS_SESSION_OPEN;
2816 pr_info("mds%d came back\n", session->s_mds);
2820 case CEPH_SESSION_OPEN:
2821 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2822 pr_info("mds%d reconnect success\n", session->s_mds);
2823 session->s_state = CEPH_MDS_SESSION_OPEN;
2824 renewed_caps(mdsc, session, 0);
2827 __close_session(mdsc, session);
2830 case CEPH_SESSION_RENEWCAPS:
2831 if (session->s_renew_seq == seq)
2832 renewed_caps(mdsc, session, 1);
2835 case CEPH_SESSION_CLOSE:
2836 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2837 pr_info("mds%d reconnect denied\n", session->s_mds);
2838 cleanup_session_requests(mdsc, session);
2839 remove_session_caps(session);
2840 wake = 2; /* for good measure */
2841 wake_up_all(&mdsc->session_close_wq);
2844 case CEPH_SESSION_STALE:
2845 pr_info("mds%d caps went stale, renewing\n",
2847 spin_lock(&session->s_gen_ttl_lock);
2848 session->s_cap_gen++;
2849 session->s_cap_ttl = jiffies - 1;
2850 spin_unlock(&session->s_gen_ttl_lock);
2851 send_renew_caps(mdsc, session);
2854 case CEPH_SESSION_RECALL_STATE:
2855 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2858 case CEPH_SESSION_FLUSHMSG:
2859 send_flushmsg_ack(mdsc, session, seq);
2862 case CEPH_SESSION_FORCE_RO:
2863 dout("force_session_readonly %p\n", session);
2864 spin_lock(&session->s_cap_lock);
2865 session->s_readonly = true;
2866 spin_unlock(&session->s_cap_lock);
2867 wake_up_session_caps(session, FORCE_RO);
2870 case CEPH_SESSION_REJECT:
2871 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2872 pr_info("mds%d rejected session\n", session->s_mds);
2873 session->s_state = CEPH_MDS_SESSION_REJECTED;
2874 cleanup_session_requests(mdsc, session);
2875 remove_session_caps(session);
2876 wake = 2; /* for good measure */
2880 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2884 mutex_unlock(&session->s_mutex);
2886 mutex_lock(&mdsc->mutex);
2887 __wake_requests(mdsc, &session->s_waiting);
2889 kick_requests(mdsc, mds);
2890 mutex_unlock(&mdsc->mutex);
2892 if (op == CEPH_SESSION_CLOSE)
2893 ceph_put_mds_session(session);
2897 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2898 (int)msg->front.iov_len);
2905 * called under session->mutex.
2907 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2908 struct ceph_mds_session *session)
2910 struct ceph_mds_request *req, *nreq;
2914 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2916 mutex_lock(&mdsc->mutex);
2917 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2918 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2920 ceph_msg_get(req->r_request);
2921 ceph_con_send(&session->s_con, req->r_request);
2926 * also re-send old requests when MDS enters reconnect stage. So that MDS
2927 * can process completed request in clientreplay stage.
2929 p = rb_first(&mdsc->request_tree);
2931 req = rb_entry(p, struct ceph_mds_request, r_node);
2933 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2935 if (req->r_attempts == 0)
2936 continue; /* only old requests */
2937 if (req->r_session &&
2938 req->r_session->s_mds == session->s_mds) {
2939 err = __prepare_send_request(mdsc, req,
2940 session->s_mds, true);
2942 ceph_msg_get(req->r_request);
2943 ceph_con_send(&session->s_con, req->r_request);
2947 mutex_unlock(&mdsc->mutex);
2951 * Encode information about a cap for a reconnect with the MDS.
2953 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2957 struct ceph_mds_cap_reconnect v2;
2958 struct ceph_mds_cap_reconnect_v1 v1;
2960 struct ceph_inode_info *ci = cap->ci;
2961 struct ceph_reconnect_state *recon_state = arg;
2962 struct ceph_pagelist *pagelist = recon_state->pagelist;
2966 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2967 inode, ceph_vinop(inode), cap, cap->cap_id,
2968 ceph_cap_string(cap->issued));
2969 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2973 spin_lock(&ci->i_ceph_lock);
2974 cap->seq = 0; /* reset cap seq */
2975 cap->issue_seq = 0; /* and issue_seq */
2976 cap->mseq = 0; /* and migrate_seq */
2977 cap->cap_gen = cap->session->s_cap_gen;
2979 if (recon_state->msg_version >= 2) {
2980 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2981 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2982 rec.v2.issued = cpu_to_le32(cap->issued);
2983 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2984 rec.v2.pathbase = 0;
2985 rec.v2.flock_len = (__force __le32)
2986 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
2988 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2989 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2990 rec.v1.issued = cpu_to_le32(cap->issued);
2991 rec.v1.size = cpu_to_le64(inode->i_size);
2992 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
2993 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
2994 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2995 rec.v1.pathbase = 0;
2998 if (list_empty(&ci->i_cap_snaps)) {
2999 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3001 struct ceph_cap_snap *capsnap =
3002 list_first_entry(&ci->i_cap_snaps,
3003 struct ceph_cap_snap, ci_item);
3004 snap_follows = capsnap->follows;
3006 spin_unlock(&ci->i_ceph_lock);
3008 if (recon_state->msg_version >= 2) {
3009 int num_fcntl_locks, num_flock_locks;
3010 struct ceph_filelock *flocks = NULL;
3011 size_t struct_len, total_len = 0;
3015 if (rec.v2.flock_len) {
3016 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3018 num_fcntl_locks = 0;
3019 num_flock_locks = 0;
3021 if (num_fcntl_locks + num_flock_locks > 0) {
3022 flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3023 sizeof(struct ceph_filelock),
3029 err = ceph_encode_locks_to_buffer(inode, flocks,
3044 if (recon_state->msg_version >= 3) {
3045 /* version, compat_version and struct_len */
3046 total_len = 2 * sizeof(u8) + sizeof(u32);
3050 * number of encoded locks is stable, so copy to pagelist
3052 struct_len = 2 * sizeof(u32) +
3053 (num_fcntl_locks + num_flock_locks) *
3054 sizeof(struct ceph_filelock);
3055 rec.v2.flock_len = cpu_to_le32(struct_len);
3057 struct_len += sizeof(u32) + sizeof(rec.v2);
3060 struct_len += sizeof(u64); /* snap_follows */
3062 total_len += struct_len;
3063 err = ceph_pagelist_reserve(pagelist, total_len);
3069 if (recon_state->msg_version >= 3) {
3070 ceph_pagelist_encode_8(pagelist, struct_v);
3071 ceph_pagelist_encode_8(pagelist, 1);
3072 ceph_pagelist_encode_32(pagelist, struct_len);
3074 ceph_pagelist_encode_string(pagelist, NULL, 0);
3075 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3076 ceph_locks_to_pagelist(flocks, pagelist,
3077 num_fcntl_locks, num_flock_locks);
3079 ceph_pagelist_encode_64(pagelist, snap_follows);
3086 struct dentry *dentry;
3088 dentry = d_find_alias(inode);
3090 path = ceph_mdsc_build_path(dentry,
3091 &pathlen, &pathbase, 0);
3094 err = PTR_ERR(path);
3097 rec.v1.pathbase = cpu_to_le64(pathbase);
3100 err = ceph_pagelist_reserve(pagelist,
3101 pathlen + sizeof(u32) + sizeof(rec.v1));
3107 ceph_pagelist_encode_string(pagelist, path, pathlen);
3108 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3113 recon_state->nr_caps++;
3120 * If an MDS fails and recovers, clients need to reconnect in order to
3121 * reestablish shared state. This includes all caps issued through
3122 * this session _and_ the snap_realm hierarchy. Because it's not
3123 * clear which snap realms the mds cares about, we send everything we
3124 * know about.. that ensures we'll then get any new info the
3125 * recovering MDS might have.
3127 * This is a relatively heavyweight operation, but it's rare.
3129 * called with mdsc->mutex held.
3131 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3132 struct ceph_mds_session *session)
3134 struct ceph_msg *reply;
3136 int mds = session->s_mds;
3139 struct ceph_pagelist *pagelist;
3140 struct ceph_reconnect_state recon_state;
3143 pr_info("mds%d reconnect start\n", mds);
3145 pagelist = ceph_pagelist_alloc(GFP_NOFS);
3147 goto fail_nopagelist;
3149 reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3153 mutex_lock(&session->s_mutex);
3154 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3157 dout("session %p state %s\n", session,
3158 ceph_session_state_name(session->s_state));
3160 spin_lock(&session->s_gen_ttl_lock);
3161 session->s_cap_gen++;
3162 spin_unlock(&session->s_gen_ttl_lock);
3164 spin_lock(&session->s_cap_lock);
3165 /* don't know if session is readonly */
3166 session->s_readonly = 0;
3168 * notify __ceph_remove_cap() that we are composing cap reconnect.
3169 * If a cap get released before being added to the cap reconnect,
3170 * __ceph_remove_cap() should skip queuing cap release.
3172 session->s_cap_reconnect = 1;
3173 /* drop old cap expires; we're about to reestablish that state */
3174 detach_cap_releases(session, &dispose);
3175 spin_unlock(&session->s_cap_lock);
3176 dispose_cap_releases(mdsc, &dispose);
3178 /* trim unused caps to reduce MDS's cache rejoin time */
3179 if (mdsc->fsc->sb->s_root)
3180 shrink_dcache_parent(mdsc->fsc->sb->s_root);
3182 ceph_con_close(&session->s_con);
3183 ceph_con_open(&session->s_con,
3184 CEPH_ENTITY_TYPE_MDS, mds,
3185 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3187 /* replay unsafe requests */
3188 replay_unsafe_requests(mdsc, session);
3190 down_read(&mdsc->snap_rwsem);
3192 /* traverse this session's caps */
3193 s_nr_caps = session->s_nr_caps;
3194 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
3198 recon_state.nr_caps = 0;
3199 recon_state.pagelist = pagelist;
3200 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
3201 recon_state.msg_version = 3;
3203 recon_state.msg_version = 2;
3204 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
3208 spin_lock(&session->s_cap_lock);
3209 session->s_cap_reconnect = 0;
3210 spin_unlock(&session->s_cap_lock);
3213 * snaprealms. we provide mds with the ino, seq (version), and
3214 * parent for all of our realms. If the mds has any newer info,
3217 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3218 struct ceph_snap_realm *realm =
3219 rb_entry(p, struct ceph_snap_realm, node);
3220 struct ceph_mds_snaprealm_reconnect sr_rec;
3222 dout(" adding snap realm %llx seq %lld parent %llx\n",
3223 realm->ino, realm->seq, realm->parent_ino);
3224 sr_rec.ino = cpu_to_le64(realm->ino);
3225 sr_rec.seq = cpu_to_le64(realm->seq);
3226 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3227 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3232 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
3234 /* raced with cap release? */
3235 if (s_nr_caps != recon_state.nr_caps) {
3236 struct page *page = list_first_entry(&pagelist->head,
3238 __le32 *addr = kmap_atomic(page);
3239 *addr = cpu_to_le32(recon_state.nr_caps);
3240 kunmap_atomic(addr);
3243 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3244 ceph_msg_data_add_pagelist(reply, pagelist);
3246 ceph_early_kick_flushing_caps(mdsc, session);
3248 ceph_con_send(&session->s_con, reply);
3250 mutex_unlock(&session->s_mutex);
3252 mutex_lock(&mdsc->mutex);
3253 __wake_requests(mdsc, &session->s_waiting);
3254 mutex_unlock(&mdsc->mutex);
3256 up_read(&mdsc->snap_rwsem);
3257 ceph_pagelist_release(pagelist);
3261 ceph_msg_put(reply);
3262 up_read(&mdsc->snap_rwsem);
3263 mutex_unlock(&session->s_mutex);
3265 ceph_pagelist_release(pagelist);
3267 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
3273 * compare old and new mdsmaps, kicking requests
3274 * and closing out old connections as necessary
3276 * called under mdsc->mutex.
3278 static void check_new_map(struct ceph_mds_client *mdsc,
3279 struct ceph_mdsmap *newmap,
3280 struct ceph_mdsmap *oldmap)
3283 int oldstate, newstate;
3284 struct ceph_mds_session *s;
3286 dout("check_new_map new %u old %u\n",
3287 newmap->m_epoch, oldmap->m_epoch);
3289 for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
3290 if (!mdsc->sessions[i])
3292 s = mdsc->sessions[i];
3293 oldstate = ceph_mdsmap_get_state(oldmap, i);
3294 newstate = ceph_mdsmap_get_state(newmap, i);
3296 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
3297 i, ceph_mds_state_name(oldstate),
3298 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
3299 ceph_mds_state_name(newstate),
3300 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
3301 ceph_session_state_name(s->s_state));
3303 if (i >= newmap->m_num_mds ||
3304 memcmp(ceph_mdsmap_get_addr(oldmap, i),
3305 ceph_mdsmap_get_addr(newmap, i),
3306 sizeof(struct ceph_entity_addr))) {
3307 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3308 /* the session never opened, just close it
3311 __unregister_session(mdsc, s);
3312 __wake_requests(mdsc, &s->s_waiting);
3313 ceph_put_mds_session(s);
3314 } else if (i >= newmap->m_num_mds) {
3315 /* force close session for stopped mds */
3317 __unregister_session(mdsc, s);
3318 __wake_requests(mdsc, &s->s_waiting);
3319 kick_requests(mdsc, i);
3320 mutex_unlock(&mdsc->mutex);
3322 mutex_lock(&s->s_mutex);
3323 cleanup_session_requests(mdsc, s);
3324 remove_session_caps(s);
3325 mutex_unlock(&s->s_mutex);
3327 ceph_put_mds_session(s);
3329 mutex_lock(&mdsc->mutex);
3332 mutex_unlock(&mdsc->mutex);
3333 mutex_lock(&s->s_mutex);
3334 mutex_lock(&mdsc->mutex);
3335 ceph_con_close(&s->s_con);
3336 mutex_unlock(&s->s_mutex);
3337 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3339 } else if (oldstate == newstate) {
3340 continue; /* nothing new with this mds */
3346 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
3347 newstate >= CEPH_MDS_STATE_RECONNECT) {
3348 mutex_unlock(&mdsc->mutex);
3349 send_mds_reconnect(mdsc, s);
3350 mutex_lock(&mdsc->mutex);
3354 * kick request on any mds that has gone active.
3356 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3357 newstate >= CEPH_MDS_STATE_ACTIVE) {
3358 if (oldstate != CEPH_MDS_STATE_CREATING &&
3359 oldstate != CEPH_MDS_STATE_STARTING)
3360 pr_info("mds%d recovery completed\n", s->s_mds);
3361 kick_requests(mdsc, i);
3362 ceph_kick_flushing_caps(mdsc, s);
3363 wake_up_session_caps(s, RECONNECT);
3367 for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
3368 s = mdsc->sessions[i];
3371 if (!ceph_mdsmap_is_laggy(newmap, i))
3373 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3374 s->s_state == CEPH_MDS_SESSION_HUNG ||
3375 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3376 dout(" connecting to export targets of laggy mds%d\n",
3378 __open_export_target_sessions(mdsc, s);
3390 * caller must hold session s_mutex, dentry->d_lock
3392 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3394 struct ceph_dentry_info *di = ceph_dentry(dentry);
3396 ceph_put_mds_session(di->lease_session);
3397 di->lease_session = NULL;
3400 static void handle_lease(struct ceph_mds_client *mdsc,
3401 struct ceph_mds_session *session,
3402 struct ceph_msg *msg)
3404 struct super_block *sb = mdsc->fsc->sb;
3405 struct inode *inode;
3406 struct dentry *parent, *dentry;
3407 struct ceph_dentry_info *di;
3408 int mds = session->s_mds;
3409 struct ceph_mds_lease *h = msg->front.iov_base;
3411 struct ceph_vino vino;
3415 dout("handle_lease from mds%d\n", mds);
3418 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3420 vino.ino = le64_to_cpu(h->ino);
3421 vino.snap = CEPH_NOSNAP;
3422 seq = le32_to_cpu(h->seq);
3423 dname.len = get_unaligned_le32(h + 1);
3424 if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
3426 dname.name = (void *)(h + 1) + sizeof(u32);
3429 inode = ceph_find_inode(sb, vino);
3430 dout("handle_lease %s, ino %llx %p %.*s\n",
3431 ceph_lease_op_name(h->action), vino.ino, inode,
3432 dname.len, dname.name);
3434 mutex_lock(&session->s_mutex);
3438 dout("handle_lease no inode %llx\n", vino.ino);
3443 parent = d_find_alias(inode);
3445 dout("no parent dentry on inode %p\n", inode);
3447 goto release; /* hrm... */
3449 dname.hash = full_name_hash(parent, dname.name, dname.len);
3450 dentry = d_lookup(parent, &dname);
3455 spin_lock(&dentry->d_lock);
3456 di = ceph_dentry(dentry);
3457 switch (h->action) {
3458 case CEPH_MDS_LEASE_REVOKE:
3459 if (di->lease_session == session) {
3460 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3461 h->seq = cpu_to_le32(di->lease_seq);
3462 __ceph_mdsc_drop_dentry_lease(dentry);
3467 case CEPH_MDS_LEASE_RENEW:
3468 if (di->lease_session == session &&
3469 di->lease_gen == session->s_cap_gen &&
3470 di->lease_renew_from &&
3471 di->lease_renew_after == 0) {
3472 unsigned long duration =
3473 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
3475 di->lease_seq = seq;
3476 di->time = di->lease_renew_from + duration;
3477 di->lease_renew_after = di->lease_renew_from +
3479 di->lease_renew_from = 0;
3483 spin_unlock(&dentry->d_lock);
3490 /* let's just reuse the same message */
3491 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3493 ceph_con_send(&session->s_con, msg);
3497 mutex_unlock(&session->s_mutex);
3501 pr_err("corrupt lease message\n");
3505 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3506 struct inode *inode,
3507 struct dentry *dentry, char action,
3510 struct ceph_msg *msg;
3511 struct ceph_mds_lease *lease;
3512 int len = sizeof(*lease) + sizeof(u32);
3515 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3516 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3517 dnamelen = dentry->d_name.len;
3520 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
3523 lease = msg->front.iov_base;
3524 lease->action = action;
3525 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3526 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3527 lease->seq = cpu_to_le32(seq);
3528 put_unaligned_le32(dnamelen, lease + 1);
3529 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3532 * if this is a preemptive lease RELEASE, no need to
3533 * flush request stream, since the actual request will
3536 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3538 ceph_con_send(&session->s_con, msg);
3542 * lock unlock sessions, to wait ongoing session activities
3544 static void lock_unlock_sessions(struct ceph_mds_client *mdsc)
3548 mutex_lock(&mdsc->mutex);
3549 for (i = 0; i < mdsc->max_sessions; i++) {
3550 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3553 mutex_unlock(&mdsc->mutex);
3554 mutex_lock(&s->s_mutex);
3555 mutex_unlock(&s->s_mutex);
3556 ceph_put_mds_session(s);
3557 mutex_lock(&mdsc->mutex);
3559 mutex_unlock(&mdsc->mutex);
3565 * delayed work -- periodically trim expired leases, renew caps with mds
3567 static void schedule_delayed(struct ceph_mds_client *mdsc)
3570 unsigned hz = round_jiffies_relative(HZ * delay);
3571 schedule_delayed_work(&mdsc->delayed_work, hz);
3574 static void delayed_work(struct work_struct *work)
3577 struct ceph_mds_client *mdsc =
3578 container_of(work, struct ceph_mds_client, delayed_work.work);
3582 dout("mdsc delayed_work\n");
3583 ceph_check_delayed_caps(mdsc);
3585 mutex_lock(&mdsc->mutex);
3586 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3587 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3588 mdsc->last_renew_caps);
3590 mdsc->last_renew_caps = jiffies;
3592 for (i = 0; i < mdsc->max_sessions; i++) {
3593 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3596 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3597 dout("resending session close request for mds%d\n",
3599 request_close_session(mdsc, s);
3600 ceph_put_mds_session(s);
3603 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3604 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3605 s->s_state = CEPH_MDS_SESSION_HUNG;
3606 pr_info("mds%d hung\n", s->s_mds);
3609 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3610 /* this mds is failed or recovering, just wait */
3611 ceph_put_mds_session(s);
3614 mutex_unlock(&mdsc->mutex);
3616 mutex_lock(&s->s_mutex);
3618 send_renew_caps(mdsc, s);
3620 ceph_con_keepalive(&s->s_con);
3621 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3622 s->s_state == CEPH_MDS_SESSION_HUNG)
3623 ceph_send_cap_releases(mdsc, s);
3624 mutex_unlock(&s->s_mutex);
3625 ceph_put_mds_session(s);
3627 mutex_lock(&mdsc->mutex);
3629 mutex_unlock(&mdsc->mutex);
3631 schedule_delayed(mdsc);
3634 int ceph_mdsc_init(struct ceph_fs_client *fsc)
3637 struct ceph_mds_client *mdsc;
3639 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3643 mutex_init(&mdsc->mutex);
3644 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3645 if (!mdsc->mdsmap) {
3651 init_completion(&mdsc->safe_umount_waiters);
3652 init_waitqueue_head(&mdsc->session_close_wq);
3653 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3654 mdsc->sessions = NULL;
3655 atomic_set(&mdsc->num_sessions, 0);
3656 mdsc->max_sessions = 0;
3658 atomic64_set(&mdsc->quotarealms_count, 0);
3659 mdsc->last_snap_seq = 0;
3660 init_rwsem(&mdsc->snap_rwsem);
3661 mdsc->snap_realms = RB_ROOT;
3662 INIT_LIST_HEAD(&mdsc->snap_empty);
3663 spin_lock_init(&mdsc->snap_empty_lock);
3665 mdsc->oldest_tid = 0;
3666 mdsc->request_tree = RB_ROOT;
3667 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3668 mdsc->last_renew_caps = jiffies;
3669 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3670 spin_lock_init(&mdsc->cap_delay_lock);
3671 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3672 spin_lock_init(&mdsc->snap_flush_lock);
3673 mdsc->last_cap_flush_tid = 1;
3674 INIT_LIST_HEAD(&mdsc->cap_flush_list);
3675 INIT_LIST_HEAD(&mdsc->cap_dirty);
3676 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3677 mdsc->num_cap_flushing = 0;
3678 spin_lock_init(&mdsc->cap_dirty_lock);
3679 init_waitqueue_head(&mdsc->cap_flushing_wq);
3680 spin_lock_init(&mdsc->dentry_lru_lock);
3681 INIT_LIST_HEAD(&mdsc->dentry_lru);
3683 ceph_caps_init(mdsc);
3684 ceph_adjust_min_caps(mdsc, fsc->min_caps);
3686 init_rwsem(&mdsc->pool_perm_rwsem);
3687 mdsc->pool_perm_tree = RB_ROOT;
3689 strscpy(mdsc->nodename, utsname()->nodename,
3690 sizeof(mdsc->nodename));
3695 * Wait for safe replies on open mds requests. If we time out, drop
3696 * all requests from the tree to avoid dangling dentry refs.
3698 static void wait_requests(struct ceph_mds_client *mdsc)
3700 struct ceph_options *opts = mdsc->fsc->client->options;
3701 struct ceph_mds_request *req;
3703 mutex_lock(&mdsc->mutex);
3704 if (__get_oldest_req(mdsc)) {
3705 mutex_unlock(&mdsc->mutex);
3707 dout("wait_requests waiting for requests\n");
3708 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3709 ceph_timeout_jiffies(opts->mount_timeout));
3711 /* tear down remaining requests */
3712 mutex_lock(&mdsc->mutex);
3713 while ((req = __get_oldest_req(mdsc))) {
3714 dout("wait_requests timed out on tid %llu\n",
3716 __unregister_request(mdsc, req);
3719 mutex_unlock(&mdsc->mutex);
3720 dout("wait_requests done\n");
3724 * called before mount is ro, and before dentries are torn down.
3725 * (hmm, does this still race with new lookups?)
3727 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3729 dout("pre_umount\n");
3732 lock_unlock_sessions(mdsc);
3733 ceph_flush_dirty_caps(mdsc);
3734 wait_requests(mdsc);
3737 * wait for reply handlers to drop their request refs and
3738 * their inode/dcache refs
3744 * wait for all write mds requests to flush.
3746 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3748 struct ceph_mds_request *req = NULL, *nextreq;
3751 mutex_lock(&mdsc->mutex);
3752 dout("wait_unsafe_requests want %lld\n", want_tid);
3754 req = __get_oldest_req(mdsc);
3755 while (req && req->r_tid <= want_tid) {
3756 /* find next request */
3757 n = rb_next(&req->r_node);
3759 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3762 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3763 (req->r_op & CEPH_MDS_OP_WRITE)) {
3765 ceph_mdsc_get_request(req);
3767 ceph_mdsc_get_request(nextreq);
3768 mutex_unlock(&mdsc->mutex);
3769 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3770 req->r_tid, want_tid);
3771 wait_for_completion(&req->r_safe_completion);
3772 mutex_lock(&mdsc->mutex);
3773 ceph_mdsc_put_request(req);
3775 break; /* next dne before, so we're done! */
3776 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3777 /* next request was removed from tree */
3778 ceph_mdsc_put_request(nextreq);
3781 ceph_mdsc_put_request(nextreq); /* won't go away */
3785 mutex_unlock(&mdsc->mutex);
3786 dout("wait_unsafe_requests done\n");
3789 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3791 u64 want_tid, want_flush;
3793 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
3797 mutex_lock(&mdsc->mutex);
3798 want_tid = mdsc->last_tid;
3799 mutex_unlock(&mdsc->mutex);
3801 ceph_flush_dirty_caps(mdsc);
3802 spin_lock(&mdsc->cap_dirty_lock);
3803 want_flush = mdsc->last_cap_flush_tid;
3804 if (!list_empty(&mdsc->cap_flush_list)) {
3805 struct ceph_cap_flush *cf =
3806 list_last_entry(&mdsc->cap_flush_list,
3807 struct ceph_cap_flush, g_list);
3810 spin_unlock(&mdsc->cap_dirty_lock);
3812 dout("sync want tid %lld flush_seq %lld\n",
3813 want_tid, want_flush);
3815 wait_unsafe_requests(mdsc, want_tid);
3816 wait_caps_flush(mdsc, want_flush);
3820 * true if all sessions are closed, or we force unmount
3822 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
3824 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
3826 return atomic_read(&mdsc->num_sessions) <= skipped;
3830 * called after sb is ro.
3832 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3834 struct ceph_options *opts = mdsc->fsc->client->options;
3835 struct ceph_mds_session *session;
3839 dout("close_sessions\n");
3841 /* close sessions */
3842 mutex_lock(&mdsc->mutex);
3843 for (i = 0; i < mdsc->max_sessions; i++) {
3844 session = __ceph_lookup_mds_session(mdsc, i);
3847 mutex_unlock(&mdsc->mutex);
3848 mutex_lock(&session->s_mutex);
3849 if (__close_session(mdsc, session) <= 0)
3851 mutex_unlock(&session->s_mutex);
3852 ceph_put_mds_session(session);
3853 mutex_lock(&mdsc->mutex);
3855 mutex_unlock(&mdsc->mutex);
3857 dout("waiting for sessions to close\n");
3858 wait_event_timeout(mdsc->session_close_wq,
3859 done_closing_sessions(mdsc, skipped),
3860 ceph_timeout_jiffies(opts->mount_timeout));
3862 /* tear down remaining sessions */
3863 mutex_lock(&mdsc->mutex);
3864 for (i = 0; i < mdsc->max_sessions; i++) {
3865 if (mdsc->sessions[i]) {
3866 session = get_session(mdsc->sessions[i]);
3867 __unregister_session(mdsc, session);
3868 mutex_unlock(&mdsc->mutex);
3869 mutex_lock(&session->s_mutex);
3870 remove_session_caps(session);
3871 mutex_unlock(&session->s_mutex);
3872 ceph_put_mds_session(session);
3873 mutex_lock(&mdsc->mutex);
3876 WARN_ON(!list_empty(&mdsc->cap_delay_list));
3877 mutex_unlock(&mdsc->mutex);
3879 ceph_cleanup_empty_realms(mdsc);
3881 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3886 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3888 struct ceph_mds_session *session;
3891 dout("force umount\n");
3893 mutex_lock(&mdsc->mutex);
3894 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3895 session = __ceph_lookup_mds_session(mdsc, mds);
3898 mutex_unlock(&mdsc->mutex);
3899 mutex_lock(&session->s_mutex);
3900 __close_session(mdsc, session);
3901 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3902 cleanup_session_requests(mdsc, session);
3903 remove_session_caps(session);
3905 mutex_unlock(&session->s_mutex);
3906 ceph_put_mds_session(session);
3907 mutex_lock(&mdsc->mutex);
3908 kick_requests(mdsc, mds);
3910 __wake_requests(mdsc, &mdsc->waiting_for_map);
3911 mutex_unlock(&mdsc->mutex);
3914 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3917 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3919 ceph_mdsmap_destroy(mdsc->mdsmap);
3920 kfree(mdsc->sessions);
3921 ceph_caps_finalize(mdsc);
3922 ceph_pool_perm_destroy(mdsc);
3925 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3927 struct ceph_mds_client *mdsc = fsc->mdsc;
3928 dout("mdsc_destroy %p\n", mdsc);
3933 /* flush out any connection work with references to us */
3936 ceph_mdsc_stop(mdsc);
3940 dout("mdsc_destroy %p done\n", mdsc);
3943 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3945 struct ceph_fs_client *fsc = mdsc->fsc;
3946 const char *mds_namespace = fsc->mount_options->mds_namespace;
3947 void *p = msg->front.iov_base;
3948 void *end = p + msg->front.iov_len;
3952 u32 mount_fscid = (u32)-1;
3953 u8 struct_v, struct_cv;
3956 ceph_decode_need(&p, end, sizeof(u32), bad);
3957 epoch = ceph_decode_32(&p);
3959 dout("handle_fsmap epoch %u\n", epoch);
3961 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3962 struct_v = ceph_decode_8(&p);
3963 struct_cv = ceph_decode_8(&p);
3964 map_len = ceph_decode_32(&p);
3966 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3967 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3969 num_fs = ceph_decode_32(&p);
3970 while (num_fs-- > 0) {
3971 void *info_p, *info_end;
3976 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3977 info_v = ceph_decode_8(&p);
3978 info_cv = ceph_decode_8(&p);
3979 info_len = ceph_decode_32(&p);
3980 ceph_decode_need(&p, end, info_len, bad);
3982 info_end = p + info_len;
3985 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3986 fscid = ceph_decode_32(&info_p);
3987 namelen = ceph_decode_32(&info_p);
3988 ceph_decode_need(&info_p, info_end, namelen, bad);
3990 if (mds_namespace &&
3991 strlen(mds_namespace) == namelen &&
3992 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3993 mount_fscid = fscid;
3998 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3999 if (mount_fscid != (u32)-1) {
4000 fsc->client->monc.fs_cluster_id = mount_fscid;
4001 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
4003 ceph_monc_renew_subs(&fsc->client->monc);
4011 pr_err("error decoding fsmap\n");
4013 mutex_lock(&mdsc->mutex);
4014 mdsc->mdsmap_err = err;
4015 __wake_requests(mdsc, &mdsc->waiting_for_map);
4016 mutex_unlock(&mdsc->mutex);
4020 * handle mds map update.
4022 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4026 void *p = msg->front.iov_base;
4027 void *end = p + msg->front.iov_len;
4028 struct ceph_mdsmap *newmap, *oldmap;
4029 struct ceph_fsid fsid;
4032 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
4033 ceph_decode_copy(&p, &fsid, sizeof(fsid));
4034 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
4036 epoch = ceph_decode_32(&p);
4037 maplen = ceph_decode_32(&p);
4038 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
4040 /* do we need it? */
4041 mutex_lock(&mdsc->mutex);
4042 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
4043 dout("handle_map epoch %u <= our %u\n",
4044 epoch, mdsc->mdsmap->m_epoch);
4045 mutex_unlock(&mdsc->mutex);
4049 newmap = ceph_mdsmap_decode(&p, end);
4050 if (IS_ERR(newmap)) {
4051 err = PTR_ERR(newmap);
4055 /* swap into place */
4057 oldmap = mdsc->mdsmap;
4058 mdsc->mdsmap = newmap;
4059 check_new_map(mdsc, newmap, oldmap);
4060 ceph_mdsmap_destroy(oldmap);
4062 mdsc->mdsmap = newmap; /* first mds map */
4064 mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
4067 __wake_requests(mdsc, &mdsc->waiting_for_map);
4068 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
4069 mdsc->mdsmap->m_epoch);
4071 mutex_unlock(&mdsc->mutex);
4072 schedule_delayed(mdsc);
4076 mutex_unlock(&mdsc->mutex);
4078 pr_err("error decoding mdsmap %d\n", err);
4082 static struct ceph_connection *con_get(struct ceph_connection *con)
4084 struct ceph_mds_session *s = con->private;
4086 if (get_session(s)) {
4087 dout("mdsc con_get %p ok (%d)\n", s, refcount_read(&s->s_ref));
4090 dout("mdsc con_get %p FAIL\n", s);
4094 static void con_put(struct ceph_connection *con)
4096 struct ceph_mds_session *s = con->private;
4098 dout("mdsc con_put %p (%d)\n", s, refcount_read(&s->s_ref) - 1);
4099 ceph_put_mds_session(s);
4103 * if the client is unresponsive for long enough, the mds will kill
4104 * the session entirely.
4106 static void peer_reset(struct ceph_connection *con)
4108 struct ceph_mds_session *s = con->private;
4109 struct ceph_mds_client *mdsc = s->s_mdsc;
4111 pr_warn("mds%d closed our session\n", s->s_mds);
4112 send_mds_reconnect(mdsc, s);
4115 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4117 struct ceph_mds_session *s = con->private;
4118 struct ceph_mds_client *mdsc = s->s_mdsc;
4119 int type = le16_to_cpu(msg->hdr.type);
4121 mutex_lock(&mdsc->mutex);
4122 if (__verify_registered_session(mdsc, s) < 0) {
4123 mutex_unlock(&mdsc->mutex);
4126 mutex_unlock(&mdsc->mutex);
4129 case CEPH_MSG_MDS_MAP:
4130 ceph_mdsc_handle_mdsmap(mdsc, msg);
4132 case CEPH_MSG_FS_MAP_USER:
4133 ceph_mdsc_handle_fsmap(mdsc, msg);
4135 case CEPH_MSG_CLIENT_SESSION:
4136 handle_session(s, msg);
4138 case CEPH_MSG_CLIENT_REPLY:
4139 handle_reply(s, msg);
4141 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
4142 handle_forward(mdsc, s, msg);
4144 case CEPH_MSG_CLIENT_CAPS:
4145 ceph_handle_caps(s, msg);
4147 case CEPH_MSG_CLIENT_SNAP:
4148 ceph_handle_snap(mdsc, s, msg);
4150 case CEPH_MSG_CLIENT_LEASE:
4151 handle_lease(mdsc, s, msg);
4153 case CEPH_MSG_CLIENT_QUOTA:
4154 ceph_handle_quota(mdsc, s, msg);
4158 pr_err("received unknown message type %d %s\n", type,
4159 ceph_msg_type_name(type));
4170 * Note: returned pointer is the address of a structure that's
4171 * managed separately. Caller must *not* attempt to free it.
4173 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4174 int *proto, int force_new)
4176 struct ceph_mds_session *s = con->private;
4177 struct ceph_mds_client *mdsc = s->s_mdsc;
4178 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4179 struct ceph_auth_handshake *auth = &s->s_auth;
4181 if (force_new && auth->authorizer) {
4182 ceph_auth_destroy_authorizer(auth->authorizer);
4183 auth->authorizer = NULL;
4185 if (!auth->authorizer) {
4186 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4189 return ERR_PTR(ret);
4191 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4194 return ERR_PTR(ret);
4196 *proto = ac->protocol;
4201 static int add_authorizer_challenge(struct ceph_connection *con,
4202 void *challenge_buf, int challenge_buf_len)
4204 struct ceph_mds_session *s = con->private;
4205 struct ceph_mds_client *mdsc = s->s_mdsc;
4206 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4208 return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
4209 challenge_buf, challenge_buf_len);
4212 static int verify_authorizer_reply(struct ceph_connection *con)
4214 struct ceph_mds_session *s = con->private;
4215 struct ceph_mds_client *mdsc = s->s_mdsc;
4216 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4218 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4221 static int invalidate_authorizer(struct ceph_connection *con)
4223 struct ceph_mds_session *s = con->private;
4224 struct ceph_mds_client *mdsc = s->s_mdsc;
4225 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4227 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
4229 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
4232 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4233 struct ceph_msg_header *hdr, int *skip)
4235 struct ceph_msg *msg;
4236 int type = (int) le16_to_cpu(hdr->type);
4237 int front_len = (int) le32_to_cpu(hdr->front_len);
4243 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4245 pr_err("unable to allocate msg type %d len %d\n",
4253 static int mds_sign_message(struct ceph_msg *msg)
4255 struct ceph_mds_session *s = msg->con->private;
4256 struct ceph_auth_handshake *auth = &s->s_auth;
4258 return ceph_auth_sign_message(auth, msg);
4261 static int mds_check_message_signature(struct ceph_msg *msg)
4263 struct ceph_mds_session *s = msg->con->private;
4264 struct ceph_auth_handshake *auth = &s->s_auth;
4266 return ceph_auth_check_message_signature(auth, msg);
4269 static const struct ceph_connection_operations mds_con_ops = {
4272 .dispatch = dispatch,
4273 .get_authorizer = get_authorizer,
4274 .add_authorizer_challenge = add_authorizer_challenge,
4275 .verify_authorizer_reply = verify_authorizer_reply,
4276 .invalidate_authorizer = invalidate_authorizer,
4277 .peer_reset = peer_reset,
4278 .alloc_msg = mds_alloc_msg,
4279 .sign_message = mds_sign_message,
4280 .check_message_signature = mds_check_message_signature,