1 #include <linux/ceph/ceph_debug.h>
4 #include <linux/wait.h>
5 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/debugfs.h>
9 #include <linux/seq_file.h>
10 #include <linux/utsname.h>
13 #include "mds_client.h"
15 #include <linux/ceph/ceph_features.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/decode.h>
18 #include <linux/ceph/pagelist.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/debugfs.h>
23 * A cluster of MDS (metadata server) daemons is responsible for
24 * managing the file system namespace (the directory hierarchy and
25 * inodes) and for coordinating shared access to storage. Metadata is
26 * partitioning hierarchically across a number of servers, and that
27 * partition varies over time as the cluster adjusts the distribution
28 * in order to balance load.
30 * The MDS client is primarily responsible to managing synchronous
31 * metadata requests for operations like open, unlink, and so forth.
32 * If there is a MDS failure, we find out about it when we (possibly
33 * request and) receive a new MDS map, and can resubmit affected
36 * For the most part, though, we take advantage of a lossless
37 * communications channel to the MDS, and do not need to worry about
38 * timing out or resubmitting requests.
40 * We maintain a stateful "session" with each MDS we interact with.
41 * Within each session, we sent periodic heartbeat messages to ensure
42 * any capabilities or leases we have been issues remain valid. If
43 * the session times out and goes stale, our leases and capabilities
44 * are no longer valid.
47 struct ceph_reconnect_state
{
49 struct ceph_pagelist
*pagelist
;
53 static void __wake_requests(struct ceph_mds_client
*mdsc
,
54 struct list_head
*head
);
56 static const struct ceph_connection_operations mds_con_ops
;
64 * parse individual inode info
66 static int parse_reply_info_in(void **p
, void *end
,
67 struct ceph_mds_reply_info_in
*info
,
73 *p
+= sizeof(struct ceph_mds_reply_inode
) +
74 sizeof(*info
->in
->fragtree
.splits
) *
75 le32_to_cpu(info
->in
->fragtree
.nsplits
);
77 ceph_decode_32_safe(p
, end
, info
->symlink_len
, bad
);
78 ceph_decode_need(p
, end
, info
->symlink_len
, bad
);
80 *p
+= info
->symlink_len
;
82 if (features
& CEPH_FEATURE_DIRLAYOUTHASH
)
83 ceph_decode_copy_safe(p
, end
, &info
->dir_layout
,
84 sizeof(info
->dir_layout
), bad
);
86 memset(&info
->dir_layout
, 0, sizeof(info
->dir_layout
));
88 ceph_decode_32_safe(p
, end
, info
->xattr_len
, bad
);
89 ceph_decode_need(p
, end
, info
->xattr_len
, bad
);
90 info
->xattr_data
= *p
;
91 *p
+= info
->xattr_len
;
98 * parse a normal reply, which may contain a (dir+)dentry and/or a
101 static int parse_reply_info_trace(void **p
, void *end
,
102 struct ceph_mds_reply_info_parsed
*info
,
107 if (info
->head
->is_dentry
) {
108 err
= parse_reply_info_in(p
, end
, &info
->diri
, features
);
112 if (unlikely(*p
+ sizeof(*info
->dirfrag
) > end
))
115 *p
+= sizeof(*info
->dirfrag
) +
116 sizeof(u32
)*le32_to_cpu(info
->dirfrag
->ndist
);
117 if (unlikely(*p
> end
))
120 ceph_decode_32_safe(p
, end
, info
->dname_len
, bad
);
121 ceph_decode_need(p
, end
, info
->dname_len
, bad
);
123 *p
+= info
->dname_len
;
125 *p
+= sizeof(*info
->dlease
);
128 if (info
->head
->is_target
) {
129 err
= parse_reply_info_in(p
, end
, &info
->targeti
, features
);
134 if (unlikely(*p
!= end
))
141 pr_err("problem parsing mds trace %d\n", err
);
146 * parse readdir results
148 static int parse_reply_info_dir(void **p
, void *end
,
149 struct ceph_mds_reply_info_parsed
*info
,
156 if (*p
+ sizeof(*info
->dir_dir
) > end
)
158 *p
+= sizeof(*info
->dir_dir
) +
159 sizeof(u32
)*le32_to_cpu(info
->dir_dir
->ndist
);
163 ceph_decode_need(p
, end
, sizeof(num
) + 2, bad
);
164 num
= ceph_decode_32(p
);
165 info
->dir_end
= ceph_decode_8(p
);
166 info
->dir_complete
= ceph_decode_8(p
);
170 BUG_ON(!info
->dir_in
);
171 info
->dir_dname
= (void *)(info
->dir_in
+ num
);
172 info
->dir_dname_len
= (void *)(info
->dir_dname
+ num
);
173 info
->dir_dlease
= (void *)(info
->dir_dname_len
+ num
);
174 if ((unsigned long)(info
->dir_dlease
+ num
) >
175 (unsigned long)info
->dir_in
+ info
->dir_buf_size
) {
176 pr_err("dir contents are larger than expected\n");
184 ceph_decode_need(p
, end
, sizeof(u32
)*2, bad
);
185 info
->dir_dname_len
[i
] = ceph_decode_32(p
);
186 ceph_decode_need(p
, end
, info
->dir_dname_len
[i
], bad
);
187 info
->dir_dname
[i
] = *p
;
188 *p
+= info
->dir_dname_len
[i
];
189 dout("parsed dir dname '%.*s'\n", info
->dir_dname_len
[i
],
191 info
->dir_dlease
[i
] = *p
;
192 *p
+= sizeof(struct ceph_mds_reply_lease
);
195 err
= parse_reply_info_in(p
, end
, &info
->dir_in
[i
], features
);
210 pr_err("problem parsing dir contents %d\n", err
);
215 * parse fcntl F_GETLK results
217 static int parse_reply_info_filelock(void **p
, void *end
,
218 struct ceph_mds_reply_info_parsed
*info
,
221 if (*p
+ sizeof(*info
->filelock_reply
) > end
)
224 info
->filelock_reply
= *p
;
225 *p
+= sizeof(*info
->filelock_reply
);
227 if (unlikely(*p
!= end
))
236 * parse create results
238 static int parse_reply_info_create(void **p
, void *end
,
239 struct ceph_mds_reply_info_parsed
*info
,
242 if (features
& CEPH_FEATURE_REPLY_CREATE_INODE
) {
244 info
->has_create_ino
= false;
246 info
->has_create_ino
= true;
247 info
->ino
= ceph_decode_64(p
);
251 if (unlikely(*p
!= end
))
260 * parse extra results
262 static int parse_reply_info_extra(void **p
, void *end
,
263 struct ceph_mds_reply_info_parsed
*info
,
266 if (info
->head
->op
== CEPH_MDS_OP_GETFILELOCK
)
267 return parse_reply_info_filelock(p
, end
, info
, features
);
268 else if (info
->head
->op
== CEPH_MDS_OP_READDIR
||
269 info
->head
->op
== CEPH_MDS_OP_LSSNAP
)
270 return parse_reply_info_dir(p
, end
, info
, features
);
271 else if (info
->head
->op
== CEPH_MDS_OP_CREATE
)
272 return parse_reply_info_create(p
, end
, info
, features
);
278 * parse entire mds reply
280 static int parse_reply_info(struct ceph_msg
*msg
,
281 struct ceph_mds_reply_info_parsed
*info
,
288 info
->head
= msg
->front
.iov_base
;
289 p
= msg
->front
.iov_base
+ sizeof(struct ceph_mds_reply_head
);
290 end
= p
+ msg
->front
.iov_len
- sizeof(struct ceph_mds_reply_head
);
293 ceph_decode_32_safe(&p
, end
, len
, bad
);
295 ceph_decode_need(&p
, end
, len
, bad
);
296 err
= parse_reply_info_trace(&p
, p
+len
, info
, features
);
302 ceph_decode_32_safe(&p
, end
, len
, bad
);
304 ceph_decode_need(&p
, end
, len
, bad
);
305 err
= parse_reply_info_extra(&p
, p
+len
, info
, features
);
311 ceph_decode_32_safe(&p
, end
, len
, bad
);
312 info
->snapblob_len
= len
;
323 pr_err("mds parse_reply err %d\n", err
);
327 static void destroy_reply_info(struct ceph_mds_reply_info_parsed
*info
)
331 free_pages((unsigned long)info
->dir_in
, get_order(info
->dir_buf_size
));
338 static const char *session_state_name(int s
)
341 case CEPH_MDS_SESSION_NEW
: return "new";
342 case CEPH_MDS_SESSION_OPENING
: return "opening";
343 case CEPH_MDS_SESSION_OPEN
: return "open";
344 case CEPH_MDS_SESSION_HUNG
: return "hung";
345 case CEPH_MDS_SESSION_CLOSING
: return "closing";
346 case CEPH_MDS_SESSION_RESTARTING
: return "restarting";
347 case CEPH_MDS_SESSION_RECONNECTING
: return "reconnecting";
348 default: return "???";
352 static struct ceph_mds_session
*get_session(struct ceph_mds_session
*s
)
354 if (atomic_inc_not_zero(&s
->s_ref
)) {
355 dout("mdsc get_session %p %d -> %d\n", s
,
356 atomic_read(&s
->s_ref
)-1, atomic_read(&s
->s_ref
));
359 dout("mdsc get_session %p 0 -- FAIL", s
);
364 void ceph_put_mds_session(struct ceph_mds_session
*s
)
366 dout("mdsc put_session %p %d -> %d\n", s
,
367 atomic_read(&s
->s_ref
), atomic_read(&s
->s_ref
)-1);
368 if (atomic_dec_and_test(&s
->s_ref
)) {
369 if (s
->s_auth
.authorizer
)
370 ceph_auth_destroy_authorizer(
371 s
->s_mdsc
->fsc
->client
->monc
.auth
,
372 s
->s_auth
.authorizer
);
378 * called under mdsc->mutex
380 struct ceph_mds_session
*__ceph_lookup_mds_session(struct ceph_mds_client
*mdsc
,
383 struct ceph_mds_session
*session
;
385 if (mds
>= mdsc
->max_sessions
|| mdsc
->sessions
[mds
] == NULL
)
387 session
= mdsc
->sessions
[mds
];
388 dout("lookup_mds_session %p %d\n", session
,
389 atomic_read(&session
->s_ref
));
390 get_session(session
);
394 static bool __have_session(struct ceph_mds_client
*mdsc
, int mds
)
396 if (mds
>= mdsc
->max_sessions
)
398 return mdsc
->sessions
[mds
];
401 static int __verify_registered_session(struct ceph_mds_client
*mdsc
,
402 struct ceph_mds_session
*s
)
404 if (s
->s_mds
>= mdsc
->max_sessions
||
405 mdsc
->sessions
[s
->s_mds
] != s
)
411 * create+register a new session for given mds.
412 * called under mdsc->mutex.
414 static struct ceph_mds_session
*register_session(struct ceph_mds_client
*mdsc
,
417 struct ceph_mds_session
*s
;
419 if (mds
>= mdsc
->mdsmap
->m_max_mds
)
420 return ERR_PTR(-EINVAL
);
422 s
= kzalloc(sizeof(*s
), GFP_NOFS
);
424 return ERR_PTR(-ENOMEM
);
427 s
->s_state
= CEPH_MDS_SESSION_NEW
;
430 mutex_init(&s
->s_mutex
);
432 ceph_con_init(&s
->s_con
, s
, &mds_con_ops
, &mdsc
->fsc
->client
->msgr
);
434 spin_lock_init(&s
->s_gen_ttl_lock
);
436 s
->s_cap_ttl
= jiffies
- 1;
438 spin_lock_init(&s
->s_cap_lock
);
439 s
->s_renew_requested
= 0;
441 INIT_LIST_HEAD(&s
->s_caps
);
444 atomic_set(&s
->s_ref
, 1);
445 INIT_LIST_HEAD(&s
->s_waiting
);
446 INIT_LIST_HEAD(&s
->s_unsafe
);
447 s
->s_num_cap_releases
= 0;
448 s
->s_cap_reconnect
= 0;
449 s
->s_cap_iterator
= NULL
;
450 INIT_LIST_HEAD(&s
->s_cap_releases
);
451 INIT_LIST_HEAD(&s
->s_cap_releases_done
);
452 INIT_LIST_HEAD(&s
->s_cap_flushing
);
453 INIT_LIST_HEAD(&s
->s_cap_snaps_flushing
);
455 dout("register_session mds%d\n", mds
);
456 if (mds
>= mdsc
->max_sessions
) {
457 int newmax
= 1 << get_count_order(mds
+1);
458 struct ceph_mds_session
**sa
;
460 dout("register_session realloc to %d\n", newmax
);
461 sa
= kcalloc(newmax
, sizeof(void *), GFP_NOFS
);
464 if (mdsc
->sessions
) {
465 memcpy(sa
, mdsc
->sessions
,
466 mdsc
->max_sessions
* sizeof(void *));
467 kfree(mdsc
->sessions
);
470 mdsc
->max_sessions
= newmax
;
472 mdsc
->sessions
[mds
] = s
;
473 atomic_inc(&s
->s_ref
); /* one ref to sessions[], one to caller */
475 ceph_con_open(&s
->s_con
, CEPH_ENTITY_TYPE_MDS
, mds
,
476 ceph_mdsmap_get_addr(mdsc
->mdsmap
, mds
));
482 return ERR_PTR(-ENOMEM
);
486 * called under mdsc->mutex
488 static void __unregister_session(struct ceph_mds_client
*mdsc
,
489 struct ceph_mds_session
*s
)
491 dout("__unregister_session mds%d %p\n", s
->s_mds
, s
);
492 BUG_ON(mdsc
->sessions
[s
->s_mds
] != s
);
493 mdsc
->sessions
[s
->s_mds
] = NULL
;
494 ceph_con_close(&s
->s_con
);
495 ceph_put_mds_session(s
);
499 * drop session refs in request.
501 * should be last request ref, or hold mdsc->mutex
503 static void put_request_session(struct ceph_mds_request
*req
)
505 if (req
->r_session
) {
506 ceph_put_mds_session(req
->r_session
);
507 req
->r_session
= NULL
;
511 void ceph_mdsc_release_request(struct kref
*kref
)
513 struct ceph_mds_request
*req
= container_of(kref
,
514 struct ceph_mds_request
,
516 destroy_reply_info(&req
->r_reply_info
);
518 ceph_msg_put(req
->r_request
);
520 ceph_msg_put(req
->r_reply
);
522 ceph_put_cap_refs(ceph_inode(req
->r_inode
), CEPH_CAP_PIN
);
525 if (req
->r_locked_dir
)
526 ceph_put_cap_refs(ceph_inode(req
->r_locked_dir
), CEPH_CAP_PIN
);
527 if (req
->r_target_inode
)
528 iput(req
->r_target_inode
);
531 if (req
->r_old_dentry
)
532 dput(req
->r_old_dentry
);
533 if (req
->r_old_dentry_dir
) {
535 * track (and drop pins for) r_old_dentry_dir
536 * separately, since r_old_dentry's d_parent may have
537 * changed between the dir mutex being dropped and
538 * this request being freed.
540 ceph_put_cap_refs(ceph_inode(req
->r_old_dentry_dir
),
542 iput(req
->r_old_dentry_dir
);
546 put_request_session(req
);
547 ceph_unreserve_caps(req
->r_mdsc
, &req
->r_caps_reservation
);
552 * lookup session, bump ref if found.
554 * called under mdsc->mutex.
556 static struct ceph_mds_request
*__lookup_request(struct ceph_mds_client
*mdsc
,
559 struct ceph_mds_request
*req
;
560 struct rb_node
*n
= mdsc
->request_tree
.rb_node
;
563 req
= rb_entry(n
, struct ceph_mds_request
, r_node
);
564 if (tid
< req
->r_tid
)
566 else if (tid
> req
->r_tid
)
569 ceph_mdsc_get_request(req
);
576 static void __insert_request(struct ceph_mds_client
*mdsc
,
577 struct ceph_mds_request
*new)
579 struct rb_node
**p
= &mdsc
->request_tree
.rb_node
;
580 struct rb_node
*parent
= NULL
;
581 struct ceph_mds_request
*req
= NULL
;
585 req
= rb_entry(parent
, struct ceph_mds_request
, r_node
);
586 if (new->r_tid
< req
->r_tid
)
588 else if (new->r_tid
> req
->r_tid
)
594 rb_link_node(&new->r_node
, parent
, p
);
595 rb_insert_color(&new->r_node
, &mdsc
->request_tree
);
599 * Register an in-flight request, and assign a tid. Link to directory
600 * are modifying (if any).
602 * Called under mdsc->mutex.
604 static void __register_request(struct ceph_mds_client
*mdsc
,
605 struct ceph_mds_request
*req
,
608 req
->r_tid
= ++mdsc
->last_tid
;
610 ceph_reserve_caps(mdsc
, &req
->r_caps_reservation
,
612 dout("__register_request %p tid %lld\n", req
, req
->r_tid
);
613 ceph_mdsc_get_request(req
);
614 __insert_request(mdsc
, req
);
616 req
->r_uid
= current_fsuid();
617 req
->r_gid
= current_fsgid();
620 struct ceph_inode_info
*ci
= ceph_inode(dir
);
623 spin_lock(&ci
->i_unsafe_lock
);
624 req
->r_unsafe_dir
= dir
;
625 list_add_tail(&req
->r_unsafe_dir_item
, &ci
->i_unsafe_dirops
);
626 spin_unlock(&ci
->i_unsafe_lock
);
630 static void __unregister_request(struct ceph_mds_client
*mdsc
,
631 struct ceph_mds_request
*req
)
633 dout("__unregister_request %p tid %lld\n", req
, req
->r_tid
);
634 rb_erase(&req
->r_node
, &mdsc
->request_tree
);
635 RB_CLEAR_NODE(&req
->r_node
);
637 if (req
->r_unsafe_dir
) {
638 struct ceph_inode_info
*ci
= ceph_inode(req
->r_unsafe_dir
);
640 spin_lock(&ci
->i_unsafe_lock
);
641 list_del_init(&req
->r_unsafe_dir_item
);
642 spin_unlock(&ci
->i_unsafe_lock
);
644 iput(req
->r_unsafe_dir
);
645 req
->r_unsafe_dir
= NULL
;
648 complete_all(&req
->r_safe_completion
);
650 ceph_mdsc_put_request(req
);
654 * Choose mds to send request to next. If there is a hint set in the
655 * request (e.g., due to a prior forward hint from the mds), use that.
656 * Otherwise, consult frag tree and/or caps to identify the
657 * appropriate mds. If all else fails, choose randomly.
659 * Called under mdsc->mutex.
661 static struct dentry
*get_nonsnap_parent(struct dentry
*dentry
)
664 * we don't need to worry about protecting the d_parent access
665 * here because we never renaming inside the snapped namespace
666 * except to resplice to another snapdir, and either the old or new
667 * result is a valid result.
669 while (!IS_ROOT(dentry
) && ceph_snap(dentry
->d_inode
) != CEPH_NOSNAP
)
670 dentry
= dentry
->d_parent
;
674 static int __choose_mds(struct ceph_mds_client
*mdsc
,
675 struct ceph_mds_request
*req
)
678 struct ceph_inode_info
*ci
;
679 struct ceph_cap
*cap
;
680 int mode
= req
->r_direct_mode
;
682 u32 hash
= req
->r_direct_hash
;
683 bool is_hash
= req
->r_direct_is_hash
;
686 * is there a specific mds we should try? ignore hint if we have
687 * no session and the mds is not up (active or recovering).
689 if (req
->r_resend_mds
>= 0 &&
690 (__have_session(mdsc
, req
->r_resend_mds
) ||
691 ceph_mdsmap_get_state(mdsc
->mdsmap
, req
->r_resend_mds
) > 0)) {
692 dout("choose_mds using resend_mds mds%d\n",
694 return req
->r_resend_mds
;
697 if (mode
== USE_RANDOM_MDS
)
702 inode
= req
->r_inode
;
703 } else if (req
->r_dentry
) {
704 /* ignore race with rename; old or new d_parent is okay */
705 struct dentry
*parent
= req
->r_dentry
->d_parent
;
706 struct inode
*dir
= parent
->d_inode
;
708 if (dir
->i_sb
!= mdsc
->fsc
->sb
) {
710 inode
= req
->r_dentry
->d_inode
;
711 } else if (ceph_snap(dir
) != CEPH_NOSNAP
) {
712 /* direct snapped/virtual snapdir requests
713 * based on parent dir inode */
714 struct dentry
*dn
= get_nonsnap_parent(parent
);
716 dout("__choose_mds using nonsnap parent %p\n", inode
);
719 inode
= req
->r_dentry
->d_inode
;
720 if (!inode
|| mode
== USE_AUTH_MDS
) {
723 hash
= ceph_dentry_hash(dir
, req
->r_dentry
);
729 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode
, (int)is_hash
,
733 ci
= ceph_inode(inode
);
735 if (is_hash
&& S_ISDIR(inode
->i_mode
)) {
736 struct ceph_inode_frag frag
;
739 ceph_choose_frag(ci
, hash
, &frag
, &found
);
741 if (mode
== USE_ANY_MDS
&& frag
.ndist
> 0) {
744 /* choose a random replica */
745 get_random_bytes(&r
, 1);
748 dout("choose_mds %p %llx.%llx "
749 "frag %u mds%d (%d/%d)\n",
750 inode
, ceph_vinop(inode
),
753 if (ceph_mdsmap_get_state(mdsc
->mdsmap
, mds
) >=
754 CEPH_MDS_STATE_ACTIVE
)
758 /* since this file/dir wasn't known to be
759 * replicated, then we want to look for the
760 * authoritative mds. */
763 /* choose auth mds */
765 dout("choose_mds %p %llx.%llx "
766 "frag %u mds%d (auth)\n",
767 inode
, ceph_vinop(inode
), frag
.frag
, mds
);
768 if (ceph_mdsmap_get_state(mdsc
->mdsmap
, mds
) >=
769 CEPH_MDS_STATE_ACTIVE
)
775 spin_lock(&ci
->i_ceph_lock
);
777 if (mode
== USE_AUTH_MDS
)
778 cap
= ci
->i_auth_cap
;
779 if (!cap
&& !RB_EMPTY_ROOT(&ci
->i_caps
))
780 cap
= rb_entry(rb_first(&ci
->i_caps
), struct ceph_cap
, ci_node
);
782 spin_unlock(&ci
->i_ceph_lock
);
785 mds
= cap
->session
->s_mds
;
786 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
787 inode
, ceph_vinop(inode
), mds
,
788 cap
== ci
->i_auth_cap
? "auth " : "", cap
);
789 spin_unlock(&ci
->i_ceph_lock
);
793 mds
= ceph_mdsmap_get_random_mds(mdsc
->mdsmap
);
794 dout("choose_mds chose random mds%d\n", mds
);
802 static struct ceph_msg
*create_session_msg(u32 op
, u64 seq
)
804 struct ceph_msg
*msg
;
805 struct ceph_mds_session_head
*h
;
807 msg
= ceph_msg_new(CEPH_MSG_CLIENT_SESSION
, sizeof(*h
), GFP_NOFS
,
810 pr_err("create_session_msg ENOMEM creating msg\n");
813 h
= msg
->front
.iov_base
;
814 h
->op
= cpu_to_le32(op
);
815 h
->seq
= cpu_to_le64(seq
);
821 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
822 * to include additional client metadata fields.
824 static struct ceph_msg
*create_session_open_msg(struct ceph_mds_client
*mdsc
, u64 seq
)
826 struct ceph_msg
*msg
;
827 struct ceph_mds_session_head
*h
;
829 int metadata_bytes
= 0;
830 int metadata_key_count
= 0;
831 struct ceph_options
*opt
= mdsc
->fsc
->client
->options
;
834 const char* metadata
[3][2] = {
835 {"hostname", utsname()->nodename
},
836 {"entity_id", opt
->name
? opt
->name
: ""},
840 /* Calculate serialized length of metadata */
841 metadata_bytes
= 4; /* map length */
842 for (i
= 0; metadata
[i
][0] != NULL
; ++i
) {
843 metadata_bytes
+= 8 + strlen(metadata
[i
][0]) +
844 strlen(metadata
[i
][1]);
845 metadata_key_count
++;
848 /* Allocate the message */
849 msg
= ceph_msg_new(CEPH_MSG_CLIENT_SESSION
, sizeof(*h
) + metadata_bytes
,
852 pr_err("create_session_msg ENOMEM creating msg\n");
855 h
= msg
->front
.iov_base
;
856 h
->op
= cpu_to_le32(CEPH_SESSION_REQUEST_OPEN
);
857 h
->seq
= cpu_to_le64(seq
);
860 * Serialize client metadata into waiting buffer space, using
861 * the format that userspace expects for map<string, string>
863 msg
->hdr
.version
= 2; /* ClientSession messages with metadata are v2 */
865 /* The write pointer, following the session_head structure */
866 p
= msg
->front
.iov_base
+ sizeof(*h
);
868 /* Number of entries in the map */
869 ceph_encode_32(&p
, metadata_key_count
);
871 /* Two length-prefixed strings for each entry in the map */
872 for (i
= 0; metadata
[i
][0] != NULL
; ++i
) {
873 size_t const key_len
= strlen(metadata
[i
][0]);
874 size_t const val_len
= strlen(metadata
[i
][1]);
876 ceph_encode_32(&p
, key_len
);
877 memcpy(p
, metadata
[i
][0], key_len
);
879 ceph_encode_32(&p
, val_len
);
880 memcpy(p
, metadata
[i
][1], val_len
);
888 * send session open request.
890 * called under mdsc->mutex
892 static int __open_session(struct ceph_mds_client
*mdsc
,
893 struct ceph_mds_session
*session
)
895 struct ceph_msg
*msg
;
897 int mds
= session
->s_mds
;
899 /* wait for mds to go active? */
900 mstate
= ceph_mdsmap_get_state(mdsc
->mdsmap
, mds
);
901 dout("open_session to mds%d (%s)\n", mds
,
902 ceph_mds_state_name(mstate
));
903 session
->s_state
= CEPH_MDS_SESSION_OPENING
;
904 session
->s_renew_requested
= jiffies
;
906 /* send connect message */
907 msg
= create_session_open_msg(mdsc
, session
->s_seq
);
910 ceph_con_send(&session
->s_con
, msg
);
915 * open sessions for any export targets for the given mds
917 * called under mdsc->mutex
919 static struct ceph_mds_session
*
920 __open_export_target_session(struct ceph_mds_client
*mdsc
, int target
)
922 struct ceph_mds_session
*session
;
924 session
= __ceph_lookup_mds_session(mdsc
, target
);
926 session
= register_session(mdsc
, target
);
930 if (session
->s_state
== CEPH_MDS_SESSION_NEW
||
931 session
->s_state
== CEPH_MDS_SESSION_CLOSING
)
932 __open_session(mdsc
, session
);
937 struct ceph_mds_session
*
938 ceph_mdsc_open_export_target_session(struct ceph_mds_client
*mdsc
, int target
)
940 struct ceph_mds_session
*session
;
942 dout("open_export_target_session to mds%d\n", target
);
944 mutex_lock(&mdsc
->mutex
);
945 session
= __open_export_target_session(mdsc
, target
);
946 mutex_unlock(&mdsc
->mutex
);
951 static void __open_export_target_sessions(struct ceph_mds_client
*mdsc
,
952 struct ceph_mds_session
*session
)
954 struct ceph_mds_info
*mi
;
955 struct ceph_mds_session
*ts
;
956 int i
, mds
= session
->s_mds
;
958 if (mds
>= mdsc
->mdsmap
->m_max_mds
)
961 mi
= &mdsc
->mdsmap
->m_info
[mds
];
962 dout("open_export_target_sessions for mds%d (%d targets)\n",
963 session
->s_mds
, mi
->num_export_targets
);
965 for (i
= 0; i
< mi
->num_export_targets
; i
++) {
966 ts
= __open_export_target_session(mdsc
, mi
->export_targets
[i
]);
968 ceph_put_mds_session(ts
);
972 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client
*mdsc
,
973 struct ceph_mds_session
*session
)
975 mutex_lock(&mdsc
->mutex
);
976 __open_export_target_sessions(mdsc
, session
);
977 mutex_unlock(&mdsc
->mutex
);
985 * Free preallocated cap messages assigned to this session
987 static void cleanup_cap_releases(struct ceph_mds_session
*session
)
989 struct ceph_msg
*msg
;
991 spin_lock(&session
->s_cap_lock
);
992 while (!list_empty(&session
->s_cap_releases
)) {
993 msg
= list_first_entry(&session
->s_cap_releases
,
994 struct ceph_msg
, list_head
);
995 list_del_init(&msg
->list_head
);
998 while (!list_empty(&session
->s_cap_releases_done
)) {
999 msg
= list_first_entry(&session
->s_cap_releases_done
,
1000 struct ceph_msg
, list_head
);
1001 list_del_init(&msg
->list_head
);
1004 spin_unlock(&session
->s_cap_lock
);
1008 * Helper to safely iterate over all caps associated with a session, with
1009 * special care taken to handle a racing __ceph_remove_cap().
1011 * Caller must hold session s_mutex.
1013 static int iterate_session_caps(struct ceph_mds_session
*session
,
1014 int (*cb
)(struct inode
*, struct ceph_cap
*,
1017 struct list_head
*p
;
1018 struct ceph_cap
*cap
;
1019 struct inode
*inode
, *last_inode
= NULL
;
1020 struct ceph_cap
*old_cap
= NULL
;
1023 dout("iterate_session_caps %p mds%d\n", session
, session
->s_mds
);
1024 spin_lock(&session
->s_cap_lock
);
1025 p
= session
->s_caps
.next
;
1026 while (p
!= &session
->s_caps
) {
1027 cap
= list_entry(p
, struct ceph_cap
, session_caps
);
1028 inode
= igrab(&cap
->ci
->vfs_inode
);
1033 session
->s_cap_iterator
= cap
;
1034 spin_unlock(&session
->s_cap_lock
);
1041 ceph_put_cap(session
->s_mdsc
, old_cap
);
1045 ret
= cb(inode
, cap
, arg
);
1048 spin_lock(&session
->s_cap_lock
);
1050 if (cap
->ci
== NULL
) {
1051 dout("iterate_session_caps finishing cap %p removal\n",
1053 BUG_ON(cap
->session
!= session
);
1054 list_del_init(&cap
->session_caps
);
1055 session
->s_nr_caps
--;
1056 cap
->session
= NULL
;
1057 old_cap
= cap
; /* put_cap it w/o locks held */
1064 session
->s_cap_iterator
= NULL
;
1065 spin_unlock(&session
->s_cap_lock
);
1070 ceph_put_cap(session
->s_mdsc
, old_cap
);
1075 static int remove_session_caps_cb(struct inode
*inode
, struct ceph_cap
*cap
,
1078 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1081 dout("removing cap %p, ci is %p, inode is %p\n",
1082 cap
, ci
, &ci
->vfs_inode
);
1083 spin_lock(&ci
->i_ceph_lock
);
1084 __ceph_remove_cap(cap
, false);
1085 if (!__ceph_is_any_real_caps(ci
)) {
1086 struct ceph_mds_client
*mdsc
=
1087 ceph_sb_to_client(inode
->i_sb
)->mdsc
;
1089 spin_lock(&mdsc
->cap_dirty_lock
);
1090 if (!list_empty(&ci
->i_dirty_item
)) {
1091 pr_info(" dropping dirty %s state for %p %lld\n",
1092 ceph_cap_string(ci
->i_dirty_caps
),
1093 inode
, ceph_ino(inode
));
1094 ci
->i_dirty_caps
= 0;
1095 list_del_init(&ci
->i_dirty_item
);
1098 if (!list_empty(&ci
->i_flushing_item
)) {
1099 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1100 ceph_cap_string(ci
->i_flushing_caps
),
1101 inode
, ceph_ino(inode
));
1102 ci
->i_flushing_caps
= 0;
1103 list_del_init(&ci
->i_flushing_item
);
1104 mdsc
->num_cap_flushing
--;
1107 if (drop
&& ci
->i_wrbuffer_ref
) {
1108 pr_info(" dropping dirty data for %p %lld\n",
1109 inode
, ceph_ino(inode
));
1110 ci
->i_wrbuffer_ref
= 0;
1111 ci
->i_wrbuffer_ref_head
= 0;
1114 spin_unlock(&mdsc
->cap_dirty_lock
);
1116 spin_unlock(&ci
->i_ceph_lock
);
1123 * caller must hold session s_mutex
1125 static void remove_session_caps(struct ceph_mds_session
*session
)
1127 dout("remove_session_caps on %p\n", session
);
1128 iterate_session_caps(session
, remove_session_caps_cb
, NULL
);
1130 spin_lock(&session
->s_cap_lock
);
1131 if (session
->s_nr_caps
> 0) {
1132 struct super_block
*sb
= session
->s_mdsc
->fsc
->sb
;
1133 struct inode
*inode
;
1134 struct ceph_cap
*cap
, *prev
= NULL
;
1135 struct ceph_vino vino
;
1137 * iterate_session_caps() skips inodes that are being
1138 * deleted, we need to wait until deletions are complete.
1139 * __wait_on_freeing_inode() is designed for the job,
1140 * but it is not exported, so use lookup inode function
1143 while (!list_empty(&session
->s_caps
)) {
1144 cap
= list_entry(session
->s_caps
.next
,
1145 struct ceph_cap
, session_caps
);
1149 vino
= cap
->ci
->i_vino
;
1150 spin_unlock(&session
->s_cap_lock
);
1152 inode
= ceph_find_inode(sb
, vino
);
1155 spin_lock(&session
->s_cap_lock
);
1158 spin_unlock(&session
->s_cap_lock
);
1160 BUG_ON(session
->s_nr_caps
> 0);
1161 BUG_ON(!list_empty(&session
->s_cap_flushing
));
1162 cleanup_cap_releases(session
);
1166 * wake up any threads waiting on this session's caps. if the cap is
1167 * old (didn't get renewed on the client reconnect), remove it now.
1169 * caller must hold s_mutex.
1171 static int wake_up_session_cb(struct inode
*inode
, struct ceph_cap
*cap
,
1174 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1176 wake_up_all(&ci
->i_cap_wq
);
1178 spin_lock(&ci
->i_ceph_lock
);
1179 ci
->i_wanted_max_size
= 0;
1180 ci
->i_requested_max_size
= 0;
1181 spin_unlock(&ci
->i_ceph_lock
);
1186 static void wake_up_session_caps(struct ceph_mds_session
*session
,
1189 dout("wake_up_session_caps %p mds%d\n", session
, session
->s_mds
);
1190 iterate_session_caps(session
, wake_up_session_cb
,
1191 (void *)(unsigned long)reconnect
);
1195 * Send periodic message to MDS renewing all currently held caps. The
1196 * ack will reset the expiration for all caps from this session.
1198 * caller holds s_mutex
1200 static int send_renew_caps(struct ceph_mds_client
*mdsc
,
1201 struct ceph_mds_session
*session
)
1203 struct ceph_msg
*msg
;
1206 if (time_after_eq(jiffies
, session
->s_cap_ttl
) &&
1207 time_after_eq(session
->s_cap_ttl
, session
->s_renew_requested
))
1208 pr_info("mds%d caps stale\n", session
->s_mds
);
1209 session
->s_renew_requested
= jiffies
;
1211 /* do not try to renew caps until a recovering mds has reconnected
1212 * with its clients. */
1213 state
= ceph_mdsmap_get_state(mdsc
->mdsmap
, session
->s_mds
);
1214 if (state
< CEPH_MDS_STATE_RECONNECT
) {
1215 dout("send_renew_caps ignoring mds%d (%s)\n",
1216 session
->s_mds
, ceph_mds_state_name(state
));
1220 dout("send_renew_caps to mds%d (%s)\n", session
->s_mds
,
1221 ceph_mds_state_name(state
));
1222 msg
= create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS
,
1223 ++session
->s_renew_seq
);
1226 ceph_con_send(&session
->s_con
, msg
);
1230 static int send_flushmsg_ack(struct ceph_mds_client
*mdsc
,
1231 struct ceph_mds_session
*session
, u64 seq
)
1233 struct ceph_msg
*msg
;
1235 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1236 session
->s_mds
, session_state_name(session
->s_state
), seq
);
1237 msg
= create_session_msg(CEPH_SESSION_FLUSHMSG_ACK
, seq
);
1240 ceph_con_send(&session
->s_con
, msg
);
1246 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1248 * Called under session->s_mutex
1250 static void renewed_caps(struct ceph_mds_client
*mdsc
,
1251 struct ceph_mds_session
*session
, int is_renew
)
1256 spin_lock(&session
->s_cap_lock
);
1257 was_stale
= is_renew
&& time_after_eq(jiffies
, session
->s_cap_ttl
);
1259 session
->s_cap_ttl
= session
->s_renew_requested
+
1260 mdsc
->mdsmap
->m_session_timeout
*HZ
;
1263 if (time_before(jiffies
, session
->s_cap_ttl
)) {
1264 pr_info("mds%d caps renewed\n", session
->s_mds
);
1267 pr_info("mds%d caps still stale\n", session
->s_mds
);
1270 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1271 session
->s_mds
, session
->s_cap_ttl
, was_stale
? "stale" : "fresh",
1272 time_before(jiffies
, session
->s_cap_ttl
) ? "stale" : "fresh");
1273 spin_unlock(&session
->s_cap_lock
);
1276 wake_up_session_caps(session
, 0);
1280 * send a session close request
1282 static int request_close_session(struct ceph_mds_client
*mdsc
,
1283 struct ceph_mds_session
*session
)
1285 struct ceph_msg
*msg
;
1287 dout("request_close_session mds%d state %s seq %lld\n",
1288 session
->s_mds
, session_state_name(session
->s_state
),
1290 msg
= create_session_msg(CEPH_SESSION_REQUEST_CLOSE
, session
->s_seq
);
1293 ceph_con_send(&session
->s_con
, msg
);
1298 * Called with s_mutex held.
1300 static int __close_session(struct ceph_mds_client
*mdsc
,
1301 struct ceph_mds_session
*session
)
1303 if (session
->s_state
>= CEPH_MDS_SESSION_CLOSING
)
1305 session
->s_state
= CEPH_MDS_SESSION_CLOSING
;
1306 return request_close_session(mdsc
, session
);
1310 * Trim old(er) caps.
1312 * Because we can't cache an inode without one or more caps, we do
1313 * this indirectly: if a cap is unused, we prune its aliases, at which
1314 * point the inode will hopefully get dropped to.
1316 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1317 * memory pressure from the MDS, though, so it needn't be perfect.
1319 static int trim_caps_cb(struct inode
*inode
, struct ceph_cap
*cap
, void *arg
)
1321 struct ceph_mds_session
*session
= arg
;
1322 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1323 int used
, wanted
, oissued
, mine
;
1325 if (session
->s_trim_caps
<= 0)
1328 spin_lock(&ci
->i_ceph_lock
);
1329 mine
= cap
->issued
| cap
->implemented
;
1330 used
= __ceph_caps_used(ci
);
1331 wanted
= __ceph_caps_file_wanted(ci
);
1332 oissued
= __ceph_caps_issued_other(ci
, cap
);
1334 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1335 inode
, cap
, ceph_cap_string(mine
), ceph_cap_string(oissued
),
1336 ceph_cap_string(used
), ceph_cap_string(wanted
));
1337 if (cap
== ci
->i_auth_cap
) {
1338 if (ci
->i_dirty_caps
| ci
->i_flushing_caps
)
1340 if ((used
| wanted
) & CEPH_CAP_ANY_WR
)
1343 if ((used
| wanted
) & ~oissued
& mine
)
1344 goto out
; /* we need these caps */
1346 session
->s_trim_caps
--;
1348 /* we aren't the only cap.. just remove us */
1349 __ceph_remove_cap(cap
, true);
1351 /* try to drop referring dentries */
1352 spin_unlock(&ci
->i_ceph_lock
);
1353 d_prune_aliases(inode
);
1354 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1355 inode
, cap
, atomic_read(&inode
->i_count
));
1360 spin_unlock(&ci
->i_ceph_lock
);
1365 * Trim session cap count down to some max number.
1367 static int trim_caps(struct ceph_mds_client
*mdsc
,
1368 struct ceph_mds_session
*session
,
1371 int trim_caps
= session
->s_nr_caps
- max_caps
;
1373 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1374 session
->s_mds
, session
->s_nr_caps
, max_caps
, trim_caps
);
1375 if (trim_caps
> 0) {
1376 session
->s_trim_caps
= trim_caps
;
1377 iterate_session_caps(session
, trim_caps_cb
, session
);
1378 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1379 session
->s_mds
, session
->s_nr_caps
, max_caps
,
1380 trim_caps
- session
->s_trim_caps
);
1381 session
->s_trim_caps
= 0;
1384 ceph_add_cap_releases(mdsc
, session
);
1385 ceph_send_cap_releases(mdsc
, session
);
1390 * Allocate cap_release messages. If there is a partially full message
1391 * in the queue, try to allocate enough to cover it's remainder, so that
1392 * we can send it immediately.
1394 * Called under s_mutex.
1396 int ceph_add_cap_releases(struct ceph_mds_client
*mdsc
,
1397 struct ceph_mds_session
*session
)
1399 struct ceph_msg
*msg
, *partial
= NULL
;
1400 struct ceph_mds_cap_release
*head
;
1402 int extra
= mdsc
->fsc
->mount_options
->cap_release_safety
;
1405 dout("add_cap_releases %p mds%d extra %d\n", session
, session
->s_mds
,
1408 spin_lock(&session
->s_cap_lock
);
1410 if (!list_empty(&session
->s_cap_releases
)) {
1411 msg
= list_first_entry(&session
->s_cap_releases
,
1414 head
= msg
->front
.iov_base
;
1415 num
= le32_to_cpu(head
->num
);
1417 dout(" partial %p with (%d/%d)\n", msg
, num
,
1418 (int)CEPH_CAPS_PER_RELEASE
);
1419 extra
+= CEPH_CAPS_PER_RELEASE
- num
;
1423 while (session
->s_num_cap_releases
< session
->s_nr_caps
+ extra
) {
1424 spin_unlock(&session
->s_cap_lock
);
1425 msg
= ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE
, PAGE_CACHE_SIZE
,
1429 dout("add_cap_releases %p msg %p now %d\n", session
, msg
,
1430 (int)msg
->front
.iov_len
);
1431 head
= msg
->front
.iov_base
;
1432 head
->num
= cpu_to_le32(0);
1433 msg
->front
.iov_len
= sizeof(*head
);
1434 spin_lock(&session
->s_cap_lock
);
1435 list_add(&msg
->list_head
, &session
->s_cap_releases
);
1436 session
->s_num_cap_releases
+= CEPH_CAPS_PER_RELEASE
;
1440 head
= partial
->front
.iov_base
;
1441 num
= le32_to_cpu(head
->num
);
1442 dout(" queueing partial %p with %d/%d\n", partial
, num
,
1443 (int)CEPH_CAPS_PER_RELEASE
);
1444 list_move_tail(&partial
->list_head
,
1445 &session
->s_cap_releases_done
);
1446 session
->s_num_cap_releases
-= CEPH_CAPS_PER_RELEASE
- num
;
1449 spin_unlock(&session
->s_cap_lock
);
1455 * flush all dirty inode data to disk.
1457 * returns true if we've flushed through want_flush_seq
1459 static int check_cap_flush(struct ceph_mds_client
*mdsc
, u64 want_flush_seq
)
1463 dout("check_cap_flush want %lld\n", want_flush_seq
);
1464 mutex_lock(&mdsc
->mutex
);
1465 for (mds
= 0; ret
&& mds
< mdsc
->max_sessions
; mds
++) {
1466 struct ceph_mds_session
*session
= mdsc
->sessions
[mds
];
1470 get_session(session
);
1471 mutex_unlock(&mdsc
->mutex
);
1473 mutex_lock(&session
->s_mutex
);
1474 if (!list_empty(&session
->s_cap_flushing
)) {
1475 struct ceph_inode_info
*ci
=
1476 list_entry(session
->s_cap_flushing
.next
,
1477 struct ceph_inode_info
,
1479 struct inode
*inode
= &ci
->vfs_inode
;
1481 spin_lock(&ci
->i_ceph_lock
);
1482 if (ci
->i_cap_flush_seq
<= want_flush_seq
) {
1483 dout("check_cap_flush still flushing %p "
1484 "seq %lld <= %lld to mds%d\n", inode
,
1485 ci
->i_cap_flush_seq
, want_flush_seq
,
1489 spin_unlock(&ci
->i_ceph_lock
);
1491 mutex_unlock(&session
->s_mutex
);
1492 ceph_put_mds_session(session
);
1496 mutex_lock(&mdsc
->mutex
);
1499 mutex_unlock(&mdsc
->mutex
);
1500 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq
);
1505 * called under s_mutex
1507 void ceph_send_cap_releases(struct ceph_mds_client
*mdsc
,
1508 struct ceph_mds_session
*session
)
1510 struct ceph_msg
*msg
;
1512 dout("send_cap_releases mds%d\n", session
->s_mds
);
1513 spin_lock(&session
->s_cap_lock
);
1514 while (!list_empty(&session
->s_cap_releases_done
)) {
1515 msg
= list_first_entry(&session
->s_cap_releases_done
,
1516 struct ceph_msg
, list_head
);
1517 list_del_init(&msg
->list_head
);
1518 spin_unlock(&session
->s_cap_lock
);
1519 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
1520 dout("send_cap_releases mds%d %p\n", session
->s_mds
, msg
);
1521 ceph_con_send(&session
->s_con
, msg
);
1522 spin_lock(&session
->s_cap_lock
);
1524 spin_unlock(&session
->s_cap_lock
);
1527 static void discard_cap_releases(struct ceph_mds_client
*mdsc
,
1528 struct ceph_mds_session
*session
)
1530 struct ceph_msg
*msg
;
1531 struct ceph_mds_cap_release
*head
;
1534 dout("discard_cap_releases mds%d\n", session
->s_mds
);
1536 if (!list_empty(&session
->s_cap_releases
)) {
1537 /* zero out the in-progress message */
1538 msg
= list_first_entry(&session
->s_cap_releases
,
1539 struct ceph_msg
, list_head
);
1540 head
= msg
->front
.iov_base
;
1541 num
= le32_to_cpu(head
->num
);
1542 dout("discard_cap_releases mds%d %p %u\n",
1543 session
->s_mds
, msg
, num
);
1544 head
->num
= cpu_to_le32(0);
1545 msg
->front
.iov_len
= sizeof(*head
);
1546 session
->s_num_cap_releases
+= num
;
1549 /* requeue completed messages */
1550 while (!list_empty(&session
->s_cap_releases_done
)) {
1551 msg
= list_first_entry(&session
->s_cap_releases_done
,
1552 struct ceph_msg
, list_head
);
1553 list_del_init(&msg
->list_head
);
1555 head
= msg
->front
.iov_base
;
1556 num
= le32_to_cpu(head
->num
);
1557 dout("discard_cap_releases mds%d %p %u\n", session
->s_mds
, msg
,
1559 session
->s_num_cap_releases
+= num
;
1560 head
->num
= cpu_to_le32(0);
1561 msg
->front
.iov_len
= sizeof(*head
);
1562 list_add(&msg
->list_head
, &session
->s_cap_releases
);
1570 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request
*req
,
1573 struct ceph_inode_info
*ci
= ceph_inode(dir
);
1574 struct ceph_mds_reply_info_parsed
*rinfo
= &req
->r_reply_info
;
1575 struct ceph_mount_options
*opt
= req
->r_mdsc
->fsc
->mount_options
;
1576 size_t size
= sizeof(*rinfo
->dir_in
) + sizeof(*rinfo
->dir_dname_len
) +
1577 sizeof(*rinfo
->dir_dname
) + sizeof(*rinfo
->dir_dlease
);
1578 int order
, num_entries
;
1580 spin_lock(&ci
->i_ceph_lock
);
1581 num_entries
= ci
->i_files
+ ci
->i_subdirs
;
1582 spin_unlock(&ci
->i_ceph_lock
);
1583 num_entries
= max(num_entries
, 1);
1584 num_entries
= min(num_entries
, opt
->max_readdir
);
1586 order
= get_order(size
* num_entries
);
1587 while (order
>= 0) {
1588 rinfo
->dir_in
= (void*)__get_free_pages(GFP_NOFS
| __GFP_NOWARN
,
1597 num_entries
= (PAGE_SIZE
<< order
) / size
;
1598 num_entries
= min(num_entries
, opt
->max_readdir
);
1600 rinfo
->dir_buf_size
= PAGE_SIZE
<< order
;
1601 req
->r_num_caps
= num_entries
+ 1;
1602 req
->r_args
.readdir
.max_entries
= cpu_to_le32(num_entries
);
1603 req
->r_args
.readdir
.max_bytes
= cpu_to_le32(opt
->max_readdir_bytes
);
1608 * Create an mds request.
1610 struct ceph_mds_request
*
1611 ceph_mdsc_create_request(struct ceph_mds_client
*mdsc
, int op
, int mode
)
1613 struct ceph_mds_request
*req
= kzalloc(sizeof(*req
), GFP_NOFS
);
1616 return ERR_PTR(-ENOMEM
);
1618 mutex_init(&req
->r_fill_mutex
);
1620 req
->r_started
= jiffies
;
1621 req
->r_resend_mds
= -1;
1622 INIT_LIST_HEAD(&req
->r_unsafe_dir_item
);
1624 kref_init(&req
->r_kref
);
1625 INIT_LIST_HEAD(&req
->r_wait
);
1626 init_completion(&req
->r_completion
);
1627 init_completion(&req
->r_safe_completion
);
1628 INIT_LIST_HEAD(&req
->r_unsafe_item
);
1630 req
->r_stamp
= CURRENT_TIME
;
1633 req
->r_direct_mode
= mode
;
1638 * return oldest (lowest) request, tid in request tree, 0 if none.
1640 * called under mdsc->mutex.
1642 static struct ceph_mds_request
*__get_oldest_req(struct ceph_mds_client
*mdsc
)
1644 if (RB_EMPTY_ROOT(&mdsc
->request_tree
))
1646 return rb_entry(rb_first(&mdsc
->request_tree
),
1647 struct ceph_mds_request
, r_node
);
1650 static u64
__get_oldest_tid(struct ceph_mds_client
*mdsc
)
1652 struct ceph_mds_request
*req
= __get_oldest_req(mdsc
);
1660 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1661 * on build_path_from_dentry in fs/cifs/dir.c.
1663 * If @stop_on_nosnap, generate path relative to the first non-snapped
1666 * Encode hidden .snap dirs as a double /, i.e.
1667 * foo/.snap/bar -> foo//bar
1669 char *ceph_mdsc_build_path(struct dentry
*dentry
, int *plen
, u64
*base
,
1672 struct dentry
*temp
;
1678 return ERR_PTR(-EINVAL
);
1682 seq
= read_seqbegin(&rename_lock
);
1684 for (temp
= dentry
; !IS_ROOT(temp
);) {
1685 struct inode
*inode
= temp
->d_inode
;
1686 if (inode
&& ceph_snap(inode
) == CEPH_SNAPDIR
)
1687 len
++; /* slash only */
1688 else if (stop_on_nosnap
&& inode
&&
1689 ceph_snap(inode
) == CEPH_NOSNAP
)
1692 len
+= 1 + temp
->d_name
.len
;
1693 temp
= temp
->d_parent
;
1697 len
--; /* no leading '/' */
1699 path
= kmalloc(len
+1, GFP_NOFS
);
1701 return ERR_PTR(-ENOMEM
);
1703 path
[pos
] = 0; /* trailing null */
1705 for (temp
= dentry
; !IS_ROOT(temp
) && pos
!= 0; ) {
1706 struct inode
*inode
;
1708 spin_lock(&temp
->d_lock
);
1709 inode
= temp
->d_inode
;
1710 if (inode
&& ceph_snap(inode
) == CEPH_SNAPDIR
) {
1711 dout("build_path path+%d: %p SNAPDIR\n",
1713 } else if (stop_on_nosnap
&& inode
&&
1714 ceph_snap(inode
) == CEPH_NOSNAP
) {
1715 spin_unlock(&temp
->d_lock
);
1718 pos
-= temp
->d_name
.len
;
1720 spin_unlock(&temp
->d_lock
);
1723 strncpy(path
+ pos
, temp
->d_name
.name
,
1726 spin_unlock(&temp
->d_lock
);
1729 temp
= temp
->d_parent
;
1732 if (pos
!= 0 || read_seqretry(&rename_lock
, seq
)) {
1733 pr_err("build_path did not end path lookup where "
1734 "expected, namelen is %d, pos is %d\n", len
, pos
);
1735 /* presumably this is only possible if racing with a
1736 rename of one of the parent directories (we can not
1737 lock the dentries above us to prevent this, but
1738 retrying should be harmless) */
1743 *base
= ceph_ino(temp
->d_inode
);
1745 dout("build_path on %p %d built %llx '%.*s'\n",
1746 dentry
, d_count(dentry
), *base
, len
, path
);
1750 static int build_dentry_path(struct dentry
*dentry
,
1751 const char **ppath
, int *ppathlen
, u64
*pino
,
1756 if (ceph_snap(dentry
->d_parent
->d_inode
) == CEPH_NOSNAP
) {
1757 *pino
= ceph_ino(dentry
->d_parent
->d_inode
);
1758 *ppath
= dentry
->d_name
.name
;
1759 *ppathlen
= dentry
->d_name
.len
;
1762 path
= ceph_mdsc_build_path(dentry
, ppathlen
, pino
, 1);
1764 return PTR_ERR(path
);
1770 static int build_inode_path(struct inode
*inode
,
1771 const char **ppath
, int *ppathlen
, u64
*pino
,
1774 struct dentry
*dentry
;
1777 if (ceph_snap(inode
) == CEPH_NOSNAP
) {
1778 *pino
= ceph_ino(inode
);
1782 dentry
= d_find_alias(inode
);
1783 path
= ceph_mdsc_build_path(dentry
, ppathlen
, pino
, 1);
1786 return PTR_ERR(path
);
1793 * request arguments may be specified via an inode *, a dentry *, or
1794 * an explicit ino+path.
1796 static int set_request_path_attr(struct inode
*rinode
, struct dentry
*rdentry
,
1797 const char *rpath
, u64 rino
,
1798 const char **ppath
, int *pathlen
,
1799 u64
*ino
, int *freepath
)
1804 r
= build_inode_path(rinode
, ppath
, pathlen
, ino
, freepath
);
1805 dout(" inode %p %llx.%llx\n", rinode
, ceph_ino(rinode
),
1807 } else if (rdentry
) {
1808 r
= build_dentry_path(rdentry
, ppath
, pathlen
, ino
, freepath
);
1809 dout(" dentry %p %llx/%.*s\n", rdentry
, *ino
, *pathlen
,
1811 } else if (rpath
|| rino
) {
1814 *pathlen
= rpath
? strlen(rpath
) : 0;
1815 dout(" path %.*s\n", *pathlen
, rpath
);
1822 * called under mdsc->mutex
1824 static struct ceph_msg
*create_request_message(struct ceph_mds_client
*mdsc
,
1825 struct ceph_mds_request
*req
,
1828 struct ceph_msg
*msg
;
1829 struct ceph_mds_request_head
*head
;
1830 const char *path1
= NULL
;
1831 const char *path2
= NULL
;
1832 u64 ino1
= 0, ino2
= 0;
1833 int pathlen1
= 0, pathlen2
= 0;
1834 int freepath1
= 0, freepath2
= 0;
1840 ret
= set_request_path_attr(req
->r_inode
, req
->r_dentry
,
1841 req
->r_path1
, req
->r_ino1
.ino
,
1842 &path1
, &pathlen1
, &ino1
, &freepath1
);
1848 ret
= set_request_path_attr(NULL
, req
->r_old_dentry
,
1849 req
->r_path2
, req
->r_ino2
.ino
,
1850 &path2
, &pathlen2
, &ino2
, &freepath2
);
1856 len
= sizeof(*head
) +
1857 pathlen1
+ pathlen2
+ 2*(1 + sizeof(u32
) + sizeof(u64
)) +
1858 sizeof(struct timespec
);
1860 /* calculate (max) length for cap releases */
1861 len
+= sizeof(struct ceph_mds_request_release
) *
1862 (!!req
->r_inode_drop
+ !!req
->r_dentry_drop
+
1863 !!req
->r_old_inode_drop
+ !!req
->r_old_dentry_drop
);
1864 if (req
->r_dentry_drop
)
1865 len
+= req
->r_dentry
->d_name
.len
;
1866 if (req
->r_old_dentry_drop
)
1867 len
+= req
->r_old_dentry
->d_name
.len
;
1869 msg
= ceph_msg_new(CEPH_MSG_CLIENT_REQUEST
, len
, GFP_NOFS
, false);
1871 msg
= ERR_PTR(-ENOMEM
);
1875 msg
->hdr
.version
= 2;
1876 msg
->hdr
.tid
= cpu_to_le64(req
->r_tid
);
1878 head
= msg
->front
.iov_base
;
1879 p
= msg
->front
.iov_base
+ sizeof(*head
);
1880 end
= msg
->front
.iov_base
+ msg
->front
.iov_len
;
1882 head
->mdsmap_epoch
= cpu_to_le32(mdsc
->mdsmap
->m_epoch
);
1883 head
->op
= cpu_to_le32(req
->r_op
);
1884 head
->caller_uid
= cpu_to_le32(from_kuid(&init_user_ns
, req
->r_uid
));
1885 head
->caller_gid
= cpu_to_le32(from_kgid(&init_user_ns
, req
->r_gid
));
1886 head
->args
= req
->r_args
;
1888 ceph_encode_filepath(&p
, end
, ino1
, path1
);
1889 ceph_encode_filepath(&p
, end
, ino2
, path2
);
1891 /* make note of release offset, in case we need to replay */
1892 req
->r_request_release_offset
= p
- msg
->front
.iov_base
;
1896 if (req
->r_inode_drop
)
1897 releases
+= ceph_encode_inode_release(&p
,
1898 req
->r_inode
? req
->r_inode
: req
->r_dentry
->d_inode
,
1899 mds
, req
->r_inode_drop
, req
->r_inode_unless
, 0);
1900 if (req
->r_dentry_drop
)
1901 releases
+= ceph_encode_dentry_release(&p
, req
->r_dentry
,
1902 mds
, req
->r_dentry_drop
, req
->r_dentry_unless
);
1903 if (req
->r_old_dentry_drop
)
1904 releases
+= ceph_encode_dentry_release(&p
, req
->r_old_dentry
,
1905 mds
, req
->r_old_dentry_drop
, req
->r_old_dentry_unless
);
1906 if (req
->r_old_inode_drop
)
1907 releases
+= ceph_encode_inode_release(&p
,
1908 req
->r_old_dentry
->d_inode
,
1909 mds
, req
->r_old_inode_drop
, req
->r_old_inode_unless
, 0);
1910 head
->num_releases
= cpu_to_le16(releases
);
1913 ceph_encode_copy(&p
, &req
->r_stamp
, sizeof(req
->r_stamp
));
1916 msg
->front
.iov_len
= p
- msg
->front
.iov_base
;
1917 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
1919 if (req
->r_data_len
) {
1920 /* outbound data set only by ceph_sync_setxattr() */
1921 BUG_ON(!req
->r_pages
);
1922 ceph_msg_data_add_pages(msg
, req
->r_pages
, req
->r_data_len
, 0);
1925 msg
->hdr
.data_len
= cpu_to_le32(req
->r_data_len
);
1926 msg
->hdr
.data_off
= cpu_to_le16(0);
1930 kfree((char *)path2
);
1933 kfree((char *)path1
);
1939 * called under mdsc->mutex if error, under no mutex if
1942 static void complete_request(struct ceph_mds_client
*mdsc
,
1943 struct ceph_mds_request
*req
)
1945 if (req
->r_callback
)
1946 req
->r_callback(mdsc
, req
);
1948 complete_all(&req
->r_completion
);
1952 * called under mdsc->mutex
1954 static int __prepare_send_request(struct ceph_mds_client
*mdsc
,
1955 struct ceph_mds_request
*req
,
1958 struct ceph_mds_request_head
*rhead
;
1959 struct ceph_msg
*msg
;
1964 struct ceph_cap
*cap
=
1965 ceph_get_cap_for_mds(ceph_inode(req
->r_inode
), mds
);
1968 req
->r_sent_on_mseq
= cap
->mseq
;
1970 req
->r_sent_on_mseq
= -1;
1972 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req
,
1973 req
->r_tid
, ceph_mds_op_name(req
->r_op
), req
->r_attempts
);
1975 if (req
->r_got_unsafe
) {
1978 * Replay. Do not regenerate message (and rebuild
1979 * paths, etc.); just use the original message.
1980 * Rebuilding paths will break for renames because
1981 * d_move mangles the src name.
1983 msg
= req
->r_request
;
1984 rhead
= msg
->front
.iov_base
;
1986 flags
= le32_to_cpu(rhead
->flags
);
1987 flags
|= CEPH_MDS_FLAG_REPLAY
;
1988 rhead
->flags
= cpu_to_le32(flags
);
1990 if (req
->r_target_inode
)
1991 rhead
->ino
= cpu_to_le64(ceph_ino(req
->r_target_inode
));
1993 rhead
->num_retry
= req
->r_attempts
- 1;
1995 /* remove cap/dentry releases from message */
1996 rhead
->num_releases
= 0;
1999 p
= msg
->front
.iov_base
+ req
->r_request_release_offset
;
2000 ceph_encode_copy(&p
, &req
->r_stamp
, sizeof(req
->r_stamp
));
2002 msg
->front
.iov_len
= p
- msg
->front
.iov_base
;
2003 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
2007 if (req
->r_request
) {
2008 ceph_msg_put(req
->r_request
);
2009 req
->r_request
= NULL
;
2011 msg
= create_request_message(mdsc
, req
, mds
);
2013 req
->r_err
= PTR_ERR(msg
);
2014 complete_request(mdsc
, req
);
2015 return PTR_ERR(msg
);
2017 req
->r_request
= msg
;
2019 rhead
= msg
->front
.iov_base
;
2020 rhead
->oldest_client_tid
= cpu_to_le64(__get_oldest_tid(mdsc
));
2021 if (req
->r_got_unsafe
)
2022 flags
|= CEPH_MDS_FLAG_REPLAY
;
2023 if (req
->r_locked_dir
)
2024 flags
|= CEPH_MDS_FLAG_WANT_DENTRY
;
2025 rhead
->flags
= cpu_to_le32(flags
);
2026 rhead
->num_fwd
= req
->r_num_fwd
;
2027 rhead
->num_retry
= req
->r_attempts
- 1;
2030 dout(" r_locked_dir = %p\n", req
->r_locked_dir
);
2035 * send request, or put it on the appropriate wait list.
2037 static int __do_request(struct ceph_mds_client
*mdsc
,
2038 struct ceph_mds_request
*req
)
2040 struct ceph_mds_session
*session
= NULL
;
2044 if (req
->r_err
|| req
->r_got_result
) {
2046 __unregister_request(mdsc
, req
);
2050 if (req
->r_timeout
&&
2051 time_after_eq(jiffies
, req
->r_started
+ req
->r_timeout
)) {
2052 dout("do_request timed out\n");
2057 put_request_session(req
);
2059 mds
= __choose_mds(mdsc
, req
);
2061 ceph_mdsmap_get_state(mdsc
->mdsmap
, mds
) < CEPH_MDS_STATE_ACTIVE
) {
2062 dout("do_request no mds or not active, waiting for map\n");
2063 list_add(&req
->r_wait
, &mdsc
->waiting_for_map
);
2067 /* get, open session */
2068 session
= __ceph_lookup_mds_session(mdsc
, mds
);
2070 session
= register_session(mdsc
, mds
);
2071 if (IS_ERR(session
)) {
2072 err
= PTR_ERR(session
);
2076 req
->r_session
= get_session(session
);
2078 dout("do_request mds%d session %p state %s\n", mds
, session
,
2079 session_state_name(session
->s_state
));
2080 if (session
->s_state
!= CEPH_MDS_SESSION_OPEN
&&
2081 session
->s_state
!= CEPH_MDS_SESSION_HUNG
) {
2082 if (session
->s_state
== CEPH_MDS_SESSION_NEW
||
2083 session
->s_state
== CEPH_MDS_SESSION_CLOSING
)
2084 __open_session(mdsc
, session
);
2085 list_add(&req
->r_wait
, &session
->s_waiting
);
2090 req
->r_resend_mds
= -1; /* forget any previous mds hint */
2092 if (req
->r_request_started
== 0) /* note request start time */
2093 req
->r_request_started
= jiffies
;
2095 err
= __prepare_send_request(mdsc
, req
, mds
);
2097 ceph_msg_get(req
->r_request
);
2098 ceph_con_send(&session
->s_con
, req
->r_request
);
2102 ceph_put_mds_session(session
);
2108 complete_request(mdsc
, req
);
2113 * called under mdsc->mutex
2115 static void __wake_requests(struct ceph_mds_client
*mdsc
,
2116 struct list_head
*head
)
2118 struct ceph_mds_request
*req
;
2119 LIST_HEAD(tmp_list
);
2121 list_splice_init(head
, &tmp_list
);
2123 while (!list_empty(&tmp_list
)) {
2124 req
= list_entry(tmp_list
.next
,
2125 struct ceph_mds_request
, r_wait
);
2126 list_del_init(&req
->r_wait
);
2127 dout(" wake request %p tid %llu\n", req
, req
->r_tid
);
2128 __do_request(mdsc
, req
);
2133 * Wake up threads with requests pending for @mds, so that they can
2134 * resubmit their requests to a possibly different mds.
2136 static void kick_requests(struct ceph_mds_client
*mdsc
, int mds
)
2138 struct ceph_mds_request
*req
;
2139 struct rb_node
*p
= rb_first(&mdsc
->request_tree
);
2141 dout("kick_requests mds%d\n", mds
);
2143 req
= rb_entry(p
, struct ceph_mds_request
, r_node
);
2145 if (req
->r_got_unsafe
)
2147 if (req
->r_session
&&
2148 req
->r_session
->s_mds
== mds
) {
2149 dout(" kicking tid %llu\n", req
->r_tid
);
2150 list_del_init(&req
->r_wait
);
2151 __do_request(mdsc
, req
);
2156 void ceph_mdsc_submit_request(struct ceph_mds_client
*mdsc
,
2157 struct ceph_mds_request
*req
)
2159 dout("submit_request on %p\n", req
);
2160 mutex_lock(&mdsc
->mutex
);
2161 __register_request(mdsc
, req
, NULL
);
2162 __do_request(mdsc
, req
);
2163 mutex_unlock(&mdsc
->mutex
);
2167 * Synchrously perform an mds request. Take care of all of the
2168 * session setup, forwarding, retry details.
2170 int ceph_mdsc_do_request(struct ceph_mds_client
*mdsc
,
2172 struct ceph_mds_request
*req
)
2176 dout("do_request on %p\n", req
);
2178 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2180 ceph_get_cap_refs(ceph_inode(req
->r_inode
), CEPH_CAP_PIN
);
2181 if (req
->r_locked_dir
)
2182 ceph_get_cap_refs(ceph_inode(req
->r_locked_dir
), CEPH_CAP_PIN
);
2183 if (req
->r_old_dentry_dir
)
2184 ceph_get_cap_refs(ceph_inode(req
->r_old_dentry_dir
),
2188 mutex_lock(&mdsc
->mutex
);
2189 __register_request(mdsc
, req
, dir
);
2190 __do_request(mdsc
, req
);
2194 __unregister_request(mdsc
, req
);
2195 dout("do_request early error %d\n", err
);
2200 mutex_unlock(&mdsc
->mutex
);
2201 dout("do_request waiting\n");
2202 if (req
->r_timeout
) {
2203 err
= (long)wait_for_completion_killable_timeout(
2204 &req
->r_completion
, req
->r_timeout
);
2208 err
= wait_for_completion_killable(&req
->r_completion
);
2210 dout("do_request waited, got %d\n", err
);
2211 mutex_lock(&mdsc
->mutex
);
2213 /* only abort if we didn't race with a real reply */
2214 if (req
->r_got_result
) {
2215 err
= le32_to_cpu(req
->r_reply_info
.head
->result
);
2216 } else if (err
< 0) {
2217 dout("aborted request %lld with %d\n", req
->r_tid
, err
);
2220 * ensure we aren't running concurrently with
2221 * ceph_fill_trace or ceph_readdir_prepopulate, which
2222 * rely on locks (dir mutex) held by our caller.
2224 mutex_lock(&req
->r_fill_mutex
);
2226 req
->r_aborted
= true;
2227 mutex_unlock(&req
->r_fill_mutex
);
2229 if (req
->r_locked_dir
&&
2230 (req
->r_op
& CEPH_MDS_OP_WRITE
))
2231 ceph_invalidate_dir_request(req
);
2237 mutex_unlock(&mdsc
->mutex
);
2238 dout("do_request %p done, result %d\n", req
, err
);
2243 * Invalidate dir's completeness, dentry lease state on an aborted MDS
2244 * namespace request.
2246 void ceph_invalidate_dir_request(struct ceph_mds_request
*req
)
2248 struct inode
*inode
= req
->r_locked_dir
;
2250 dout("invalidate_dir_request %p (complete, lease(s))\n", inode
);
2252 ceph_dir_clear_complete(inode
);
2254 ceph_invalidate_dentry_lease(req
->r_dentry
);
2255 if (req
->r_old_dentry
)
2256 ceph_invalidate_dentry_lease(req
->r_old_dentry
);
2262 * We take the session mutex and parse and process the reply immediately.
2263 * This preserves the logical ordering of replies, capabilities, etc., sent
2264 * by the MDS as they are applied to our local cache.
2266 static void handle_reply(struct ceph_mds_session
*session
, struct ceph_msg
*msg
)
2268 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2269 struct ceph_mds_request
*req
;
2270 struct ceph_mds_reply_head
*head
= msg
->front
.iov_base
;
2271 struct ceph_mds_reply_info_parsed
*rinfo
; /* parsed reply info */
2274 int mds
= session
->s_mds
;
2276 if (msg
->front
.iov_len
< sizeof(*head
)) {
2277 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2282 /* get request, session */
2283 tid
= le64_to_cpu(msg
->hdr
.tid
);
2284 mutex_lock(&mdsc
->mutex
);
2285 req
= __lookup_request(mdsc
, tid
);
2287 dout("handle_reply on unknown tid %llu\n", tid
);
2288 mutex_unlock(&mdsc
->mutex
);
2291 dout("handle_reply %p\n", req
);
2293 /* correct session? */
2294 if (req
->r_session
!= session
) {
2295 pr_err("mdsc_handle_reply got %llu on session mds%d"
2296 " not mds%d\n", tid
, session
->s_mds
,
2297 req
->r_session
? req
->r_session
->s_mds
: -1);
2298 mutex_unlock(&mdsc
->mutex
);
2303 if ((req
->r_got_unsafe
&& !head
->safe
) ||
2304 (req
->r_got_safe
&& head
->safe
)) {
2305 pr_warn("got a dup %s reply on %llu from mds%d\n",
2306 head
->safe
? "safe" : "unsafe", tid
, mds
);
2307 mutex_unlock(&mdsc
->mutex
);
2310 if (req
->r_got_safe
&& !head
->safe
) {
2311 pr_warn("got unsafe after safe on %llu from mds%d\n",
2313 mutex_unlock(&mdsc
->mutex
);
2317 result
= le32_to_cpu(head
->result
);
2321 * if we're not talking to the authority, send to them
2322 * if the authority has changed while we weren't looking,
2323 * send to new authority
2324 * Otherwise we just have to return an ESTALE
2326 if (result
== -ESTALE
) {
2327 dout("got ESTALE on request %llu", req
->r_tid
);
2328 req
->r_resend_mds
= -1;
2329 if (req
->r_direct_mode
!= USE_AUTH_MDS
) {
2330 dout("not using auth, setting for that now");
2331 req
->r_direct_mode
= USE_AUTH_MDS
;
2332 __do_request(mdsc
, req
);
2333 mutex_unlock(&mdsc
->mutex
);
2336 int mds
= __choose_mds(mdsc
, req
);
2337 if (mds
>= 0 && mds
!= req
->r_session
->s_mds
) {
2338 dout("but auth changed, so resending");
2339 __do_request(mdsc
, req
);
2340 mutex_unlock(&mdsc
->mutex
);
2344 dout("have to return ESTALE on request %llu", req
->r_tid
);
2349 req
->r_got_safe
= true;
2350 __unregister_request(mdsc
, req
);
2352 if (req
->r_got_unsafe
) {
2354 * We already handled the unsafe response, now do the
2355 * cleanup. No need to examine the response; the MDS
2356 * doesn't include any result info in the safe
2357 * response. And even if it did, there is nothing
2358 * useful we could do with a revised return value.
2360 dout("got safe reply %llu, mds%d\n", tid
, mds
);
2361 list_del_init(&req
->r_unsafe_item
);
2363 /* last unsafe request during umount? */
2364 if (mdsc
->stopping
&& !__get_oldest_req(mdsc
))
2365 complete_all(&mdsc
->safe_umount_waiters
);
2366 mutex_unlock(&mdsc
->mutex
);
2370 req
->r_got_unsafe
= true;
2371 list_add_tail(&req
->r_unsafe_item
, &req
->r_session
->s_unsafe
);
2374 dout("handle_reply tid %lld result %d\n", tid
, result
);
2375 rinfo
= &req
->r_reply_info
;
2376 err
= parse_reply_info(msg
, rinfo
, session
->s_con
.peer_features
);
2377 mutex_unlock(&mdsc
->mutex
);
2379 mutex_lock(&session
->s_mutex
);
2381 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds
, tid
);
2387 if (rinfo
->snapblob_len
) {
2388 down_write(&mdsc
->snap_rwsem
);
2389 ceph_update_snap_trace(mdsc
, rinfo
->snapblob
,
2390 rinfo
->snapblob
+ rinfo
->snapblob_len
,
2391 le32_to_cpu(head
->op
) == CEPH_MDS_OP_RMSNAP
);
2392 downgrade_write(&mdsc
->snap_rwsem
);
2394 down_read(&mdsc
->snap_rwsem
);
2397 /* insert trace into our cache */
2398 mutex_lock(&req
->r_fill_mutex
);
2399 err
= ceph_fill_trace(mdsc
->fsc
->sb
, req
, req
->r_session
);
2401 if (result
== 0 && (req
->r_op
== CEPH_MDS_OP_READDIR
||
2402 req
->r_op
== CEPH_MDS_OP_LSSNAP
))
2403 ceph_readdir_prepopulate(req
, req
->r_session
);
2404 ceph_unreserve_caps(mdsc
, &req
->r_caps_reservation
);
2406 mutex_unlock(&req
->r_fill_mutex
);
2408 up_read(&mdsc
->snap_rwsem
);
2410 mutex_lock(&mdsc
->mutex
);
2411 if (!req
->r_aborted
) {
2417 req
->r_got_result
= true;
2420 dout("reply arrived after request %lld was aborted\n", tid
);
2422 mutex_unlock(&mdsc
->mutex
);
2424 ceph_add_cap_releases(mdsc
, req
->r_session
);
2425 mutex_unlock(&session
->s_mutex
);
2427 /* kick calling process */
2428 complete_request(mdsc
, req
);
2430 ceph_mdsc_put_request(req
);
2437 * handle mds notification that our request has been forwarded.
2439 static void handle_forward(struct ceph_mds_client
*mdsc
,
2440 struct ceph_mds_session
*session
,
2441 struct ceph_msg
*msg
)
2443 struct ceph_mds_request
*req
;
2444 u64 tid
= le64_to_cpu(msg
->hdr
.tid
);
2448 void *p
= msg
->front
.iov_base
;
2449 void *end
= p
+ msg
->front
.iov_len
;
2451 ceph_decode_need(&p
, end
, 2*sizeof(u32
), bad
);
2452 next_mds
= ceph_decode_32(&p
);
2453 fwd_seq
= ceph_decode_32(&p
);
2455 mutex_lock(&mdsc
->mutex
);
2456 req
= __lookup_request(mdsc
, tid
);
2458 dout("forward tid %llu to mds%d - req dne\n", tid
, next_mds
);
2459 goto out
; /* dup reply? */
2462 if (req
->r_aborted
) {
2463 dout("forward tid %llu aborted, unregistering\n", tid
);
2464 __unregister_request(mdsc
, req
);
2465 } else if (fwd_seq
<= req
->r_num_fwd
) {
2466 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2467 tid
, next_mds
, req
->r_num_fwd
, fwd_seq
);
2469 /* resend. forward race not possible; mds would drop */
2470 dout("forward tid %llu to mds%d (we resend)\n", tid
, next_mds
);
2472 BUG_ON(req
->r_got_result
);
2473 req
->r_num_fwd
= fwd_seq
;
2474 req
->r_resend_mds
= next_mds
;
2475 put_request_session(req
);
2476 __do_request(mdsc
, req
);
2478 ceph_mdsc_put_request(req
);
2480 mutex_unlock(&mdsc
->mutex
);
2484 pr_err("mdsc_handle_forward decode error err=%d\n", err
);
2488 * handle a mds session control message
2490 static void handle_session(struct ceph_mds_session
*session
,
2491 struct ceph_msg
*msg
)
2493 struct ceph_mds_client
*mdsc
= session
->s_mdsc
;
2496 int mds
= session
->s_mds
;
2497 struct ceph_mds_session_head
*h
= msg
->front
.iov_base
;
2501 if (msg
->front
.iov_len
!= sizeof(*h
))
2503 op
= le32_to_cpu(h
->op
);
2504 seq
= le64_to_cpu(h
->seq
);
2506 mutex_lock(&mdsc
->mutex
);
2507 if (op
== CEPH_SESSION_CLOSE
)
2508 __unregister_session(mdsc
, session
);
2509 /* FIXME: this ttl calculation is generous */
2510 session
->s_ttl
= jiffies
+ HZ
*mdsc
->mdsmap
->m_session_autoclose
;
2511 mutex_unlock(&mdsc
->mutex
);
2513 mutex_lock(&session
->s_mutex
);
2515 dout("handle_session mds%d %s %p state %s seq %llu\n",
2516 mds
, ceph_session_op_name(op
), session
,
2517 session_state_name(session
->s_state
), seq
);
2519 if (session
->s_state
== CEPH_MDS_SESSION_HUNG
) {
2520 session
->s_state
= CEPH_MDS_SESSION_OPEN
;
2521 pr_info("mds%d came back\n", session
->s_mds
);
2525 case CEPH_SESSION_OPEN
:
2526 if (session
->s_state
== CEPH_MDS_SESSION_RECONNECTING
)
2527 pr_info("mds%d reconnect success\n", session
->s_mds
);
2528 session
->s_state
= CEPH_MDS_SESSION_OPEN
;
2529 renewed_caps(mdsc
, session
, 0);
2532 __close_session(mdsc
, session
);
2535 case CEPH_SESSION_RENEWCAPS
:
2536 if (session
->s_renew_seq
== seq
)
2537 renewed_caps(mdsc
, session
, 1);
2540 case CEPH_SESSION_CLOSE
:
2541 if (session
->s_state
== CEPH_MDS_SESSION_RECONNECTING
)
2542 pr_info("mds%d reconnect denied\n", session
->s_mds
);
2543 remove_session_caps(session
);
2544 wake
= 2; /* for good measure */
2545 wake_up_all(&mdsc
->session_close_wq
);
2548 case CEPH_SESSION_STALE
:
2549 pr_info("mds%d caps went stale, renewing\n",
2551 spin_lock(&session
->s_gen_ttl_lock
);
2552 session
->s_cap_gen
++;
2553 session
->s_cap_ttl
= jiffies
- 1;
2554 spin_unlock(&session
->s_gen_ttl_lock
);
2555 send_renew_caps(mdsc
, session
);
2558 case CEPH_SESSION_RECALL_STATE
:
2559 trim_caps(mdsc
, session
, le32_to_cpu(h
->max_caps
));
2562 case CEPH_SESSION_FLUSHMSG
:
2563 send_flushmsg_ack(mdsc
, session
, seq
);
2567 pr_err("mdsc_handle_session bad op %d mds%d\n", op
, mds
);
2571 mutex_unlock(&session
->s_mutex
);
2573 mutex_lock(&mdsc
->mutex
);
2574 __wake_requests(mdsc
, &session
->s_waiting
);
2576 kick_requests(mdsc
, mds
);
2577 mutex_unlock(&mdsc
->mutex
);
2582 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds
,
2583 (int)msg
->front
.iov_len
);
2590 * called under session->mutex.
2592 static void replay_unsafe_requests(struct ceph_mds_client
*mdsc
,
2593 struct ceph_mds_session
*session
)
2595 struct ceph_mds_request
*req
, *nreq
;
2598 dout("replay_unsafe_requests mds%d\n", session
->s_mds
);
2600 mutex_lock(&mdsc
->mutex
);
2601 list_for_each_entry_safe(req
, nreq
, &session
->s_unsafe
, r_unsafe_item
) {
2602 err
= __prepare_send_request(mdsc
, req
, session
->s_mds
);
2604 ceph_msg_get(req
->r_request
);
2605 ceph_con_send(&session
->s_con
, req
->r_request
);
2608 mutex_unlock(&mdsc
->mutex
);
2612 * Encode information about a cap for a reconnect with the MDS.
2614 static int encode_caps_cb(struct inode
*inode
, struct ceph_cap
*cap
,
2618 struct ceph_mds_cap_reconnect v2
;
2619 struct ceph_mds_cap_reconnect_v1 v1
;
2622 struct ceph_inode_info
*ci
;
2623 struct ceph_reconnect_state
*recon_state
= arg
;
2624 struct ceph_pagelist
*pagelist
= recon_state
->pagelist
;
2628 struct dentry
*dentry
;
2632 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2633 inode
, ceph_vinop(inode
), cap
, cap
->cap_id
,
2634 ceph_cap_string(cap
->issued
));
2635 err
= ceph_pagelist_encode_64(pagelist
, ceph_ino(inode
));
2639 dentry
= d_find_alias(inode
);
2641 path
= ceph_mdsc_build_path(dentry
, &pathlen
, &pathbase
, 0);
2643 err
= PTR_ERR(path
);
2650 err
= ceph_pagelist_encode_string(pagelist
, path
, pathlen
);
2654 spin_lock(&ci
->i_ceph_lock
);
2655 cap
->seq
= 0; /* reset cap seq */
2656 cap
->issue_seq
= 0; /* and issue_seq */
2657 cap
->mseq
= 0; /* and migrate_seq */
2658 cap
->cap_gen
= cap
->session
->s_cap_gen
;
2660 if (recon_state
->flock
) {
2661 rec
.v2
.cap_id
= cpu_to_le64(cap
->cap_id
);
2662 rec
.v2
.wanted
= cpu_to_le32(__ceph_caps_wanted(ci
));
2663 rec
.v2
.issued
= cpu_to_le32(cap
->issued
);
2664 rec
.v2
.snaprealm
= cpu_to_le64(ci
->i_snap_realm
->ino
);
2665 rec
.v2
.pathbase
= cpu_to_le64(pathbase
);
2666 rec
.v2
.flock_len
= 0;
2667 reclen
= sizeof(rec
.v2
);
2669 rec
.v1
.cap_id
= cpu_to_le64(cap
->cap_id
);
2670 rec
.v1
.wanted
= cpu_to_le32(__ceph_caps_wanted(ci
));
2671 rec
.v1
.issued
= cpu_to_le32(cap
->issued
);
2672 rec
.v1
.size
= cpu_to_le64(inode
->i_size
);
2673 ceph_encode_timespec(&rec
.v1
.mtime
, &inode
->i_mtime
);
2674 ceph_encode_timespec(&rec
.v1
.atime
, &inode
->i_atime
);
2675 rec
.v1
.snaprealm
= cpu_to_le64(ci
->i_snap_realm
->ino
);
2676 rec
.v1
.pathbase
= cpu_to_le64(pathbase
);
2677 reclen
= sizeof(rec
.v1
);
2679 spin_unlock(&ci
->i_ceph_lock
);
2681 if (recon_state
->flock
) {
2682 int num_fcntl_locks
, num_flock_locks
;
2683 struct ceph_filelock
*flocks
;
2686 spin_lock(&inode
->i_lock
);
2687 ceph_count_locks(inode
, &num_fcntl_locks
, &num_flock_locks
);
2688 spin_unlock(&inode
->i_lock
);
2689 flocks
= kmalloc((num_fcntl_locks
+num_flock_locks
) *
2690 sizeof(struct ceph_filelock
), GFP_NOFS
);
2695 spin_lock(&inode
->i_lock
);
2696 err
= ceph_encode_locks_to_buffer(inode
, flocks
,
2699 spin_unlock(&inode
->i_lock
);
2707 * number of encoded locks is stable, so copy to pagelist
2709 rec
.v2
.flock_len
= cpu_to_le32(2*sizeof(u32
) +
2710 (num_fcntl_locks
+num_flock_locks
) *
2711 sizeof(struct ceph_filelock
));
2712 err
= ceph_pagelist_append(pagelist
, &rec
, reclen
);
2714 err
= ceph_locks_to_pagelist(flocks
, pagelist
,
2719 err
= ceph_pagelist_append(pagelist
, &rec
, reclen
);
2722 recon_state
->nr_caps
++;
2732 * If an MDS fails and recovers, clients need to reconnect in order to
2733 * reestablish shared state. This includes all caps issued through
2734 * this session _and_ the snap_realm hierarchy. Because it's not
2735 * clear which snap realms the mds cares about, we send everything we
2736 * know about.. that ensures we'll then get any new info the
2737 * recovering MDS might have.
2739 * This is a relatively heavyweight operation, but it's rare.
2741 * called with mdsc->mutex held.
2743 static void send_mds_reconnect(struct ceph_mds_client
*mdsc
,
2744 struct ceph_mds_session
*session
)
2746 struct ceph_msg
*reply
;
2748 int mds
= session
->s_mds
;
2751 struct ceph_pagelist
*pagelist
;
2752 struct ceph_reconnect_state recon_state
;
2754 pr_info("mds%d reconnect start\n", mds
);
2756 pagelist
= kmalloc(sizeof(*pagelist
), GFP_NOFS
);
2758 goto fail_nopagelist
;
2759 ceph_pagelist_init(pagelist
);
2761 reply
= ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT
, 0, GFP_NOFS
, false);
2765 mutex_lock(&session
->s_mutex
);
2766 session
->s_state
= CEPH_MDS_SESSION_RECONNECTING
;
2769 dout("session %p state %s\n", session
,
2770 session_state_name(session
->s_state
));
2772 spin_lock(&session
->s_gen_ttl_lock
);
2773 session
->s_cap_gen
++;
2774 spin_unlock(&session
->s_gen_ttl_lock
);
2776 spin_lock(&session
->s_cap_lock
);
2778 * notify __ceph_remove_cap() that we are composing cap reconnect.
2779 * If a cap get released before being added to the cap reconnect,
2780 * __ceph_remove_cap() should skip queuing cap release.
2782 session
->s_cap_reconnect
= 1;
2783 /* drop old cap expires; we're about to reestablish that state */
2784 discard_cap_releases(mdsc
, session
);
2785 spin_unlock(&session
->s_cap_lock
);
2787 /* trim unused caps to reduce MDS's cache rejoin time */
2788 shrink_dcache_parent(mdsc
->fsc
->sb
->s_root
);
2790 ceph_con_close(&session
->s_con
);
2791 ceph_con_open(&session
->s_con
,
2792 CEPH_ENTITY_TYPE_MDS
, mds
,
2793 ceph_mdsmap_get_addr(mdsc
->mdsmap
, mds
));
2795 /* replay unsafe requests */
2796 replay_unsafe_requests(mdsc
, session
);
2798 down_read(&mdsc
->snap_rwsem
);
2800 /* traverse this session's caps */
2801 s_nr_caps
= session
->s_nr_caps
;
2802 err
= ceph_pagelist_encode_32(pagelist
, s_nr_caps
);
2806 recon_state
.nr_caps
= 0;
2807 recon_state
.pagelist
= pagelist
;
2808 recon_state
.flock
= session
->s_con
.peer_features
& CEPH_FEATURE_FLOCK
;
2809 err
= iterate_session_caps(session
, encode_caps_cb
, &recon_state
);
2813 spin_lock(&session
->s_cap_lock
);
2814 session
->s_cap_reconnect
= 0;
2815 spin_unlock(&session
->s_cap_lock
);
2818 * snaprealms. we provide mds with the ino, seq (version), and
2819 * parent for all of our realms. If the mds has any newer info,
2822 for (p
= rb_first(&mdsc
->snap_realms
); p
; p
= rb_next(p
)) {
2823 struct ceph_snap_realm
*realm
=
2824 rb_entry(p
, struct ceph_snap_realm
, node
);
2825 struct ceph_mds_snaprealm_reconnect sr_rec
;
2827 dout(" adding snap realm %llx seq %lld parent %llx\n",
2828 realm
->ino
, realm
->seq
, realm
->parent_ino
);
2829 sr_rec
.ino
= cpu_to_le64(realm
->ino
);
2830 sr_rec
.seq
= cpu_to_le64(realm
->seq
);
2831 sr_rec
.parent
= cpu_to_le64(realm
->parent_ino
);
2832 err
= ceph_pagelist_append(pagelist
, &sr_rec
, sizeof(sr_rec
));
2837 if (recon_state
.flock
)
2838 reply
->hdr
.version
= cpu_to_le16(2);
2840 /* raced with cap release? */
2841 if (s_nr_caps
!= recon_state
.nr_caps
) {
2842 struct page
*page
= list_first_entry(&pagelist
->head
,
2844 __le32
*addr
= kmap_atomic(page
);
2845 *addr
= cpu_to_le32(recon_state
.nr_caps
);
2846 kunmap_atomic(addr
);
2849 reply
->hdr
.data_len
= cpu_to_le32(pagelist
->length
);
2850 ceph_msg_data_add_pagelist(reply
, pagelist
);
2851 ceph_con_send(&session
->s_con
, reply
);
2853 mutex_unlock(&session
->s_mutex
);
2855 mutex_lock(&mdsc
->mutex
);
2856 __wake_requests(mdsc
, &session
->s_waiting
);
2857 mutex_unlock(&mdsc
->mutex
);
2859 up_read(&mdsc
->snap_rwsem
);
2863 ceph_msg_put(reply
);
2864 up_read(&mdsc
->snap_rwsem
);
2865 mutex_unlock(&session
->s_mutex
);
2867 ceph_pagelist_release(pagelist
);
2869 pr_err("error %d preparing reconnect for mds%d\n", err
, mds
);
2875 * compare old and new mdsmaps, kicking requests
2876 * and closing out old connections as necessary
2878 * called under mdsc->mutex.
2880 static void check_new_map(struct ceph_mds_client
*mdsc
,
2881 struct ceph_mdsmap
*newmap
,
2882 struct ceph_mdsmap
*oldmap
)
2885 int oldstate
, newstate
;
2886 struct ceph_mds_session
*s
;
2888 dout("check_new_map new %u old %u\n",
2889 newmap
->m_epoch
, oldmap
->m_epoch
);
2891 for (i
= 0; i
< oldmap
->m_max_mds
&& i
< mdsc
->max_sessions
; i
++) {
2892 if (mdsc
->sessions
[i
] == NULL
)
2894 s
= mdsc
->sessions
[i
];
2895 oldstate
= ceph_mdsmap_get_state(oldmap
, i
);
2896 newstate
= ceph_mdsmap_get_state(newmap
, i
);
2898 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2899 i
, ceph_mds_state_name(oldstate
),
2900 ceph_mdsmap_is_laggy(oldmap
, i
) ? " (laggy)" : "",
2901 ceph_mds_state_name(newstate
),
2902 ceph_mdsmap_is_laggy(newmap
, i
) ? " (laggy)" : "",
2903 session_state_name(s
->s_state
));
2905 if (i
>= newmap
->m_max_mds
||
2906 memcmp(ceph_mdsmap_get_addr(oldmap
, i
),
2907 ceph_mdsmap_get_addr(newmap
, i
),
2908 sizeof(struct ceph_entity_addr
))) {
2909 if (s
->s_state
== CEPH_MDS_SESSION_OPENING
) {
2910 /* the session never opened, just close it
2912 __wake_requests(mdsc
, &s
->s_waiting
);
2913 __unregister_session(mdsc
, s
);
2916 mutex_unlock(&mdsc
->mutex
);
2917 mutex_lock(&s
->s_mutex
);
2918 mutex_lock(&mdsc
->mutex
);
2919 ceph_con_close(&s
->s_con
);
2920 mutex_unlock(&s
->s_mutex
);
2921 s
->s_state
= CEPH_MDS_SESSION_RESTARTING
;
2924 /* kick any requests waiting on the recovering mds */
2925 kick_requests(mdsc
, i
);
2926 } else if (oldstate
== newstate
) {
2927 continue; /* nothing new with this mds */
2933 if (s
->s_state
== CEPH_MDS_SESSION_RESTARTING
&&
2934 newstate
>= CEPH_MDS_STATE_RECONNECT
) {
2935 mutex_unlock(&mdsc
->mutex
);
2936 send_mds_reconnect(mdsc
, s
);
2937 mutex_lock(&mdsc
->mutex
);
2941 * kick request on any mds that has gone active.
2943 if (oldstate
< CEPH_MDS_STATE_ACTIVE
&&
2944 newstate
>= CEPH_MDS_STATE_ACTIVE
) {
2945 if (oldstate
!= CEPH_MDS_STATE_CREATING
&&
2946 oldstate
!= CEPH_MDS_STATE_STARTING
)
2947 pr_info("mds%d recovery completed\n", s
->s_mds
);
2948 kick_requests(mdsc
, i
);
2949 ceph_kick_flushing_caps(mdsc
, s
);
2950 wake_up_session_caps(s
, 1);
2954 for (i
= 0; i
< newmap
->m_max_mds
&& i
< mdsc
->max_sessions
; i
++) {
2955 s
= mdsc
->sessions
[i
];
2958 if (!ceph_mdsmap_is_laggy(newmap
, i
))
2960 if (s
->s_state
== CEPH_MDS_SESSION_OPEN
||
2961 s
->s_state
== CEPH_MDS_SESSION_HUNG
||
2962 s
->s_state
== CEPH_MDS_SESSION_CLOSING
) {
2963 dout(" connecting to export targets of laggy mds%d\n",
2965 __open_export_target_sessions(mdsc
, s
);
2977 * caller must hold session s_mutex, dentry->d_lock
2979 void __ceph_mdsc_drop_dentry_lease(struct dentry
*dentry
)
2981 struct ceph_dentry_info
*di
= ceph_dentry(dentry
);
2983 ceph_put_mds_session(di
->lease_session
);
2984 di
->lease_session
= NULL
;
2987 static void handle_lease(struct ceph_mds_client
*mdsc
,
2988 struct ceph_mds_session
*session
,
2989 struct ceph_msg
*msg
)
2991 struct super_block
*sb
= mdsc
->fsc
->sb
;
2992 struct inode
*inode
;
2993 struct dentry
*parent
, *dentry
;
2994 struct ceph_dentry_info
*di
;
2995 int mds
= session
->s_mds
;
2996 struct ceph_mds_lease
*h
= msg
->front
.iov_base
;
2998 struct ceph_vino vino
;
3002 dout("handle_lease from mds%d\n", mds
);
3005 if (msg
->front
.iov_len
< sizeof(*h
) + sizeof(u32
))
3007 vino
.ino
= le64_to_cpu(h
->ino
);
3008 vino
.snap
= CEPH_NOSNAP
;
3009 seq
= le32_to_cpu(h
->seq
);
3010 dname
.name
= (void *)h
+ sizeof(*h
) + sizeof(u32
);
3011 dname
.len
= msg
->front
.iov_len
- sizeof(*h
) - sizeof(u32
);
3012 if (dname
.len
!= get_unaligned_le32(h
+1))
3016 inode
= ceph_find_inode(sb
, vino
);
3017 dout("handle_lease %s, ino %llx %p %.*s\n",
3018 ceph_lease_op_name(h
->action
), vino
.ino
, inode
,
3019 dname
.len
, dname
.name
);
3021 mutex_lock(&session
->s_mutex
);
3024 if (inode
== NULL
) {
3025 dout("handle_lease no inode %llx\n", vino
.ino
);
3030 parent
= d_find_alias(inode
);
3032 dout("no parent dentry on inode %p\n", inode
);
3034 goto release
; /* hrm... */
3036 dname
.hash
= full_name_hash(dname
.name
, dname
.len
);
3037 dentry
= d_lookup(parent
, &dname
);
3042 spin_lock(&dentry
->d_lock
);
3043 di
= ceph_dentry(dentry
);
3044 switch (h
->action
) {
3045 case CEPH_MDS_LEASE_REVOKE
:
3046 if (di
->lease_session
== session
) {
3047 if (ceph_seq_cmp(di
->lease_seq
, seq
) > 0)
3048 h
->seq
= cpu_to_le32(di
->lease_seq
);
3049 __ceph_mdsc_drop_dentry_lease(dentry
);
3054 case CEPH_MDS_LEASE_RENEW
:
3055 if (di
->lease_session
== session
&&
3056 di
->lease_gen
== session
->s_cap_gen
&&
3057 di
->lease_renew_from
&&
3058 di
->lease_renew_after
== 0) {
3059 unsigned long duration
=
3060 le32_to_cpu(h
->duration_ms
) * HZ
/ 1000;
3062 di
->lease_seq
= seq
;
3063 dentry
->d_time
= di
->lease_renew_from
+ duration
;
3064 di
->lease_renew_after
= di
->lease_renew_from
+
3066 di
->lease_renew_from
= 0;
3070 spin_unlock(&dentry
->d_lock
);
3077 /* let's just reuse the same message */
3078 h
->action
= CEPH_MDS_LEASE_REVOKE_ACK
;
3080 ceph_con_send(&session
->s_con
, msg
);
3084 mutex_unlock(&session
->s_mutex
);
3088 pr_err("corrupt lease message\n");
3092 void ceph_mdsc_lease_send_msg(struct ceph_mds_session
*session
,
3093 struct inode
*inode
,
3094 struct dentry
*dentry
, char action
,
3097 struct ceph_msg
*msg
;
3098 struct ceph_mds_lease
*lease
;
3099 int len
= sizeof(*lease
) + sizeof(u32
);
3102 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3103 inode
, dentry
, ceph_lease_op_name(action
), session
->s_mds
);
3104 dnamelen
= dentry
->d_name
.len
;
3107 msg
= ceph_msg_new(CEPH_MSG_CLIENT_LEASE
, len
, GFP_NOFS
, false);
3110 lease
= msg
->front
.iov_base
;
3111 lease
->action
= action
;
3112 lease
->ino
= cpu_to_le64(ceph_vino(inode
).ino
);
3113 lease
->first
= lease
->last
= cpu_to_le64(ceph_vino(inode
).snap
);
3114 lease
->seq
= cpu_to_le32(seq
);
3115 put_unaligned_le32(dnamelen
, lease
+ 1);
3116 memcpy((void *)(lease
+ 1) + 4, dentry
->d_name
.name
, dnamelen
);
3119 * if this is a preemptive lease RELEASE, no need to
3120 * flush request stream, since the actual request will
3123 msg
->more_to_follow
= (action
== CEPH_MDS_LEASE_RELEASE
);
3125 ceph_con_send(&session
->s_con
, msg
);
3129 * Preemptively release a lease we expect to invalidate anyway.
3130 * Pass @inode always, @dentry is optional.
3132 void ceph_mdsc_lease_release(struct ceph_mds_client
*mdsc
, struct inode
*inode
,
3133 struct dentry
*dentry
)
3135 struct ceph_dentry_info
*di
;
3136 struct ceph_mds_session
*session
;
3139 BUG_ON(inode
== NULL
);
3140 BUG_ON(dentry
== NULL
);
3142 /* is dentry lease valid? */
3143 spin_lock(&dentry
->d_lock
);
3144 di
= ceph_dentry(dentry
);
3145 if (!di
|| !di
->lease_session
||
3146 di
->lease_session
->s_mds
< 0 ||
3147 di
->lease_gen
!= di
->lease_session
->s_cap_gen
||
3148 !time_before(jiffies
, dentry
->d_time
)) {
3149 dout("lease_release inode %p dentry %p -- "
3152 spin_unlock(&dentry
->d_lock
);
3156 /* we do have a lease on this dentry; note mds and seq */
3157 session
= ceph_get_mds_session(di
->lease_session
);
3158 seq
= di
->lease_seq
;
3159 __ceph_mdsc_drop_dentry_lease(dentry
);
3160 spin_unlock(&dentry
->d_lock
);
3162 dout("lease_release inode %p dentry %p to mds%d\n",
3163 inode
, dentry
, session
->s_mds
);
3164 ceph_mdsc_lease_send_msg(session
, inode
, dentry
,
3165 CEPH_MDS_LEASE_RELEASE
, seq
);
3166 ceph_put_mds_session(session
);
3170 * drop all leases (and dentry refs) in preparation for umount
3172 static void drop_leases(struct ceph_mds_client
*mdsc
)
3176 dout("drop_leases\n");
3177 mutex_lock(&mdsc
->mutex
);
3178 for (i
= 0; i
< mdsc
->max_sessions
; i
++) {
3179 struct ceph_mds_session
*s
= __ceph_lookup_mds_session(mdsc
, i
);
3182 mutex_unlock(&mdsc
->mutex
);
3183 mutex_lock(&s
->s_mutex
);
3184 mutex_unlock(&s
->s_mutex
);
3185 ceph_put_mds_session(s
);
3186 mutex_lock(&mdsc
->mutex
);
3188 mutex_unlock(&mdsc
->mutex
);
3194 * delayed work -- periodically trim expired leases, renew caps with mds
3196 static void schedule_delayed(struct ceph_mds_client
*mdsc
)
3199 unsigned hz
= round_jiffies_relative(HZ
* delay
);
3200 schedule_delayed_work(&mdsc
->delayed_work
, hz
);
3203 static void delayed_work(struct work_struct
*work
)
3206 struct ceph_mds_client
*mdsc
=
3207 container_of(work
, struct ceph_mds_client
, delayed_work
.work
);
3211 dout("mdsc delayed_work\n");
3212 ceph_check_delayed_caps(mdsc
);
3214 mutex_lock(&mdsc
->mutex
);
3215 renew_interval
= mdsc
->mdsmap
->m_session_timeout
>> 2;
3216 renew_caps
= time_after_eq(jiffies
, HZ
*renew_interval
+
3217 mdsc
->last_renew_caps
);
3219 mdsc
->last_renew_caps
= jiffies
;
3221 for (i
= 0; i
< mdsc
->max_sessions
; i
++) {
3222 struct ceph_mds_session
*s
= __ceph_lookup_mds_session(mdsc
, i
);
3225 if (s
->s_state
== CEPH_MDS_SESSION_CLOSING
) {
3226 dout("resending session close request for mds%d\n",
3228 request_close_session(mdsc
, s
);
3229 ceph_put_mds_session(s
);
3232 if (s
->s_ttl
&& time_after(jiffies
, s
->s_ttl
)) {
3233 if (s
->s_state
== CEPH_MDS_SESSION_OPEN
) {
3234 s
->s_state
= CEPH_MDS_SESSION_HUNG
;
3235 pr_info("mds%d hung\n", s
->s_mds
);
3238 if (s
->s_state
< CEPH_MDS_SESSION_OPEN
) {
3239 /* this mds is failed or recovering, just wait */
3240 ceph_put_mds_session(s
);
3243 mutex_unlock(&mdsc
->mutex
);
3245 mutex_lock(&s
->s_mutex
);
3247 send_renew_caps(mdsc
, s
);
3249 ceph_con_keepalive(&s
->s_con
);
3250 ceph_add_cap_releases(mdsc
, s
);
3251 if (s
->s_state
== CEPH_MDS_SESSION_OPEN
||
3252 s
->s_state
== CEPH_MDS_SESSION_HUNG
)
3253 ceph_send_cap_releases(mdsc
, s
);
3254 mutex_unlock(&s
->s_mutex
);
3255 ceph_put_mds_session(s
);
3257 mutex_lock(&mdsc
->mutex
);
3259 mutex_unlock(&mdsc
->mutex
);
3261 schedule_delayed(mdsc
);
3264 int ceph_mdsc_init(struct ceph_fs_client
*fsc
)
3267 struct ceph_mds_client
*mdsc
;
3269 mdsc
= kzalloc(sizeof(struct ceph_mds_client
), GFP_NOFS
);
3274 mutex_init(&mdsc
->mutex
);
3275 mdsc
->mdsmap
= kzalloc(sizeof(*mdsc
->mdsmap
), GFP_NOFS
);
3276 if (mdsc
->mdsmap
== NULL
) {
3281 init_completion(&mdsc
->safe_umount_waiters
);
3282 init_waitqueue_head(&mdsc
->session_close_wq
);
3283 INIT_LIST_HEAD(&mdsc
->waiting_for_map
);
3284 mdsc
->sessions
= NULL
;
3285 mdsc
->max_sessions
= 0;
3287 init_rwsem(&mdsc
->snap_rwsem
);
3288 mdsc
->snap_realms
= RB_ROOT
;
3289 INIT_LIST_HEAD(&mdsc
->snap_empty
);
3290 spin_lock_init(&mdsc
->snap_empty_lock
);
3292 mdsc
->request_tree
= RB_ROOT
;
3293 INIT_DELAYED_WORK(&mdsc
->delayed_work
, delayed_work
);
3294 mdsc
->last_renew_caps
= jiffies
;
3295 INIT_LIST_HEAD(&mdsc
->cap_delay_list
);
3296 spin_lock_init(&mdsc
->cap_delay_lock
);
3297 INIT_LIST_HEAD(&mdsc
->snap_flush_list
);
3298 spin_lock_init(&mdsc
->snap_flush_lock
);
3299 mdsc
->cap_flush_seq
= 0;
3300 INIT_LIST_HEAD(&mdsc
->cap_dirty
);
3301 INIT_LIST_HEAD(&mdsc
->cap_dirty_migrating
);
3302 mdsc
->num_cap_flushing
= 0;
3303 spin_lock_init(&mdsc
->cap_dirty_lock
);
3304 init_waitqueue_head(&mdsc
->cap_flushing_wq
);
3305 spin_lock_init(&mdsc
->dentry_lru_lock
);
3306 INIT_LIST_HEAD(&mdsc
->dentry_lru
);
3308 ceph_caps_init(mdsc
);
3309 ceph_adjust_min_caps(mdsc
, fsc
->min_caps
);
3315 * Wait for safe replies on open mds requests. If we time out, drop
3316 * all requests from the tree to avoid dangling dentry refs.
3318 static void wait_requests(struct ceph_mds_client
*mdsc
)
3320 struct ceph_mds_request
*req
;
3321 struct ceph_fs_client
*fsc
= mdsc
->fsc
;
3323 mutex_lock(&mdsc
->mutex
);
3324 if (__get_oldest_req(mdsc
)) {
3325 mutex_unlock(&mdsc
->mutex
);
3327 dout("wait_requests waiting for requests\n");
3328 wait_for_completion_timeout(&mdsc
->safe_umount_waiters
,
3329 fsc
->client
->options
->mount_timeout
* HZ
);
3331 /* tear down remaining requests */
3332 mutex_lock(&mdsc
->mutex
);
3333 while ((req
= __get_oldest_req(mdsc
))) {
3334 dout("wait_requests timed out on tid %llu\n",
3336 __unregister_request(mdsc
, req
);
3339 mutex_unlock(&mdsc
->mutex
);
3340 dout("wait_requests done\n");
3344 * called before mount is ro, and before dentries are torn down.
3345 * (hmm, does this still race with new lookups?)
3347 void ceph_mdsc_pre_umount(struct ceph_mds_client
*mdsc
)
3349 dout("pre_umount\n");
3353 ceph_flush_dirty_caps(mdsc
);
3354 wait_requests(mdsc
);
3357 * wait for reply handlers to drop their request refs and
3358 * their inode/dcache refs
3364 * wait for all write mds requests to flush.
3366 static void wait_unsafe_requests(struct ceph_mds_client
*mdsc
, u64 want_tid
)
3368 struct ceph_mds_request
*req
= NULL
, *nextreq
;
3371 mutex_lock(&mdsc
->mutex
);
3372 dout("wait_unsafe_requests want %lld\n", want_tid
);
3374 req
= __get_oldest_req(mdsc
);
3375 while (req
&& req
->r_tid
<= want_tid
) {
3376 /* find next request */
3377 n
= rb_next(&req
->r_node
);
3379 nextreq
= rb_entry(n
, struct ceph_mds_request
, r_node
);
3382 if ((req
->r_op
& CEPH_MDS_OP_WRITE
)) {
3384 ceph_mdsc_get_request(req
);
3386 ceph_mdsc_get_request(nextreq
);
3387 mutex_unlock(&mdsc
->mutex
);
3388 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3389 req
->r_tid
, want_tid
);
3390 wait_for_completion(&req
->r_safe_completion
);
3391 mutex_lock(&mdsc
->mutex
);
3392 ceph_mdsc_put_request(req
);
3394 break; /* next dne before, so we're done! */
3395 if (RB_EMPTY_NODE(&nextreq
->r_node
)) {
3396 /* next request was removed from tree */
3397 ceph_mdsc_put_request(nextreq
);
3400 ceph_mdsc_put_request(nextreq
); /* won't go away */
3404 mutex_unlock(&mdsc
->mutex
);
3405 dout("wait_unsafe_requests done\n");
3408 void ceph_mdsc_sync(struct ceph_mds_client
*mdsc
)
3410 u64 want_tid
, want_flush
;
3412 if (mdsc
->fsc
->mount_state
== CEPH_MOUNT_SHUTDOWN
)
3416 mutex_lock(&mdsc
->mutex
);
3417 want_tid
= mdsc
->last_tid
;
3418 want_flush
= mdsc
->cap_flush_seq
;
3419 mutex_unlock(&mdsc
->mutex
);
3420 dout("sync want tid %lld flush_seq %lld\n", want_tid
, want_flush
);
3422 ceph_flush_dirty_caps(mdsc
);
3424 wait_unsafe_requests(mdsc
, want_tid
);
3425 wait_event(mdsc
->cap_flushing_wq
, check_cap_flush(mdsc
, want_flush
));
3429 * true if all sessions are closed, or we force unmount
3431 static bool done_closing_sessions(struct ceph_mds_client
*mdsc
)
3435 if (mdsc
->fsc
->mount_state
== CEPH_MOUNT_SHUTDOWN
)
3438 mutex_lock(&mdsc
->mutex
);
3439 for (i
= 0; i
< mdsc
->max_sessions
; i
++)
3440 if (mdsc
->sessions
[i
])
3442 mutex_unlock(&mdsc
->mutex
);
3447 * called after sb is ro.
3449 void ceph_mdsc_close_sessions(struct ceph_mds_client
*mdsc
)
3451 struct ceph_mds_session
*session
;
3453 struct ceph_fs_client
*fsc
= mdsc
->fsc
;
3454 unsigned long timeout
= fsc
->client
->options
->mount_timeout
* HZ
;
3456 dout("close_sessions\n");
3458 /* close sessions */
3459 mutex_lock(&mdsc
->mutex
);
3460 for (i
= 0; i
< mdsc
->max_sessions
; i
++) {
3461 session
= __ceph_lookup_mds_session(mdsc
, i
);
3464 mutex_unlock(&mdsc
->mutex
);
3465 mutex_lock(&session
->s_mutex
);
3466 __close_session(mdsc
, session
);
3467 mutex_unlock(&session
->s_mutex
);
3468 ceph_put_mds_session(session
);
3469 mutex_lock(&mdsc
->mutex
);
3471 mutex_unlock(&mdsc
->mutex
);
3473 dout("waiting for sessions to close\n");
3474 wait_event_timeout(mdsc
->session_close_wq
, done_closing_sessions(mdsc
),
3477 /* tear down remaining sessions */
3478 mutex_lock(&mdsc
->mutex
);
3479 for (i
= 0; i
< mdsc
->max_sessions
; i
++) {
3480 if (mdsc
->sessions
[i
]) {
3481 session
= get_session(mdsc
->sessions
[i
]);
3482 __unregister_session(mdsc
, session
);
3483 mutex_unlock(&mdsc
->mutex
);
3484 mutex_lock(&session
->s_mutex
);
3485 remove_session_caps(session
);
3486 mutex_unlock(&session
->s_mutex
);
3487 ceph_put_mds_session(session
);
3488 mutex_lock(&mdsc
->mutex
);
3491 WARN_ON(!list_empty(&mdsc
->cap_delay_list
));
3492 mutex_unlock(&mdsc
->mutex
);
3494 ceph_cleanup_empty_realms(mdsc
);
3496 cancel_delayed_work_sync(&mdsc
->delayed_work
); /* cancel timer */
3501 static void ceph_mdsc_stop(struct ceph_mds_client
*mdsc
)
3504 cancel_delayed_work_sync(&mdsc
->delayed_work
); /* cancel timer */
3506 ceph_mdsmap_destroy(mdsc
->mdsmap
);
3507 kfree(mdsc
->sessions
);
3508 ceph_caps_finalize(mdsc
);
3511 void ceph_mdsc_destroy(struct ceph_fs_client
*fsc
)
3513 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
3515 dout("mdsc_destroy %p\n", mdsc
);
3516 ceph_mdsc_stop(mdsc
);
3518 /* flush out any connection work with references to us */
3523 dout("mdsc_destroy %p done\n", mdsc
);
3528 * handle mds map update.
3530 void ceph_mdsc_handle_map(struct ceph_mds_client
*mdsc
, struct ceph_msg
*msg
)
3534 void *p
= msg
->front
.iov_base
;
3535 void *end
= p
+ msg
->front
.iov_len
;
3536 struct ceph_mdsmap
*newmap
, *oldmap
;
3537 struct ceph_fsid fsid
;
3540 ceph_decode_need(&p
, end
, sizeof(fsid
)+2*sizeof(u32
), bad
);
3541 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
3542 if (ceph_check_fsid(mdsc
->fsc
->client
, &fsid
) < 0)
3544 epoch
= ceph_decode_32(&p
);
3545 maplen
= ceph_decode_32(&p
);
3546 dout("handle_map epoch %u len %d\n", epoch
, (int)maplen
);
3548 /* do we need it? */
3549 ceph_monc_got_mdsmap(&mdsc
->fsc
->client
->monc
, epoch
);
3550 mutex_lock(&mdsc
->mutex
);
3551 if (mdsc
->mdsmap
&& epoch
<= mdsc
->mdsmap
->m_epoch
) {
3552 dout("handle_map epoch %u <= our %u\n",
3553 epoch
, mdsc
->mdsmap
->m_epoch
);
3554 mutex_unlock(&mdsc
->mutex
);
3558 newmap
= ceph_mdsmap_decode(&p
, end
);
3559 if (IS_ERR(newmap
)) {
3560 err
= PTR_ERR(newmap
);
3564 /* swap into place */
3566 oldmap
= mdsc
->mdsmap
;
3567 mdsc
->mdsmap
= newmap
;
3568 check_new_map(mdsc
, newmap
, oldmap
);
3569 ceph_mdsmap_destroy(oldmap
);
3571 mdsc
->mdsmap
= newmap
; /* first mds map */
3573 mdsc
->fsc
->sb
->s_maxbytes
= mdsc
->mdsmap
->m_max_file_size
;
3575 __wake_requests(mdsc
, &mdsc
->waiting_for_map
);
3577 mutex_unlock(&mdsc
->mutex
);
3578 schedule_delayed(mdsc
);
3582 mutex_unlock(&mdsc
->mutex
);
3584 pr_err("error decoding mdsmap %d\n", err
);
3588 static struct ceph_connection
*con_get(struct ceph_connection
*con
)
3590 struct ceph_mds_session
*s
= con
->private;
3592 if (get_session(s
)) {
3593 dout("mdsc con_get %p ok (%d)\n", s
, atomic_read(&s
->s_ref
));
3596 dout("mdsc con_get %p FAIL\n", s
);
3600 static void con_put(struct ceph_connection
*con
)
3602 struct ceph_mds_session
*s
= con
->private;
3604 dout("mdsc con_put %p (%d)\n", s
, atomic_read(&s
->s_ref
) - 1);
3605 ceph_put_mds_session(s
);
3609 * if the client is unresponsive for long enough, the mds will kill
3610 * the session entirely.
3612 static void peer_reset(struct ceph_connection
*con
)
3614 struct ceph_mds_session
*s
= con
->private;
3615 struct ceph_mds_client
*mdsc
= s
->s_mdsc
;
3617 pr_warn("mds%d closed our session\n", s
->s_mds
);
3618 send_mds_reconnect(mdsc
, s
);
3621 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
3623 struct ceph_mds_session
*s
= con
->private;
3624 struct ceph_mds_client
*mdsc
= s
->s_mdsc
;
3625 int type
= le16_to_cpu(msg
->hdr
.type
);
3627 mutex_lock(&mdsc
->mutex
);
3628 if (__verify_registered_session(mdsc
, s
) < 0) {
3629 mutex_unlock(&mdsc
->mutex
);
3632 mutex_unlock(&mdsc
->mutex
);
3635 case CEPH_MSG_MDS_MAP
:
3636 ceph_mdsc_handle_map(mdsc
, msg
);
3638 case CEPH_MSG_CLIENT_SESSION
:
3639 handle_session(s
, msg
);
3641 case CEPH_MSG_CLIENT_REPLY
:
3642 handle_reply(s
, msg
);
3644 case CEPH_MSG_CLIENT_REQUEST_FORWARD
:
3645 handle_forward(mdsc
, s
, msg
);
3647 case CEPH_MSG_CLIENT_CAPS
:
3648 ceph_handle_caps(s
, msg
);
3650 case CEPH_MSG_CLIENT_SNAP
:
3651 ceph_handle_snap(mdsc
, s
, msg
);
3653 case CEPH_MSG_CLIENT_LEASE
:
3654 handle_lease(mdsc
, s
, msg
);
3658 pr_err("received unknown message type %d %s\n", type
,
3659 ceph_msg_type_name(type
));
3670 * Note: returned pointer is the address of a structure that's
3671 * managed separately. Caller must *not* attempt to free it.
3673 static struct ceph_auth_handshake
*get_authorizer(struct ceph_connection
*con
,
3674 int *proto
, int force_new
)
3676 struct ceph_mds_session
*s
= con
->private;
3677 struct ceph_mds_client
*mdsc
= s
->s_mdsc
;
3678 struct ceph_auth_client
*ac
= mdsc
->fsc
->client
->monc
.auth
;
3679 struct ceph_auth_handshake
*auth
= &s
->s_auth
;
3681 if (force_new
&& auth
->authorizer
) {
3682 ceph_auth_destroy_authorizer(ac
, auth
->authorizer
);
3683 auth
->authorizer
= NULL
;
3685 if (!auth
->authorizer
) {
3686 int ret
= ceph_auth_create_authorizer(ac
, CEPH_ENTITY_TYPE_MDS
,
3689 return ERR_PTR(ret
);
3691 int ret
= ceph_auth_update_authorizer(ac
, CEPH_ENTITY_TYPE_MDS
,
3694 return ERR_PTR(ret
);
3696 *proto
= ac
->protocol
;
3702 static int verify_authorizer_reply(struct ceph_connection
*con
, int len
)
3704 struct ceph_mds_session
*s
= con
->private;
3705 struct ceph_mds_client
*mdsc
= s
->s_mdsc
;
3706 struct ceph_auth_client
*ac
= mdsc
->fsc
->client
->monc
.auth
;
3708 return ceph_auth_verify_authorizer_reply(ac
, s
->s_auth
.authorizer
, len
);
3711 static int invalidate_authorizer(struct ceph_connection
*con
)
3713 struct ceph_mds_session
*s
= con
->private;
3714 struct ceph_mds_client
*mdsc
= s
->s_mdsc
;
3715 struct ceph_auth_client
*ac
= mdsc
->fsc
->client
->monc
.auth
;
3717 ceph_auth_invalidate_authorizer(ac
, CEPH_ENTITY_TYPE_MDS
);
3719 return ceph_monc_validate_auth(&mdsc
->fsc
->client
->monc
);
3722 static struct ceph_msg
*mds_alloc_msg(struct ceph_connection
*con
,
3723 struct ceph_msg_header
*hdr
, int *skip
)
3725 struct ceph_msg
*msg
;
3726 int type
= (int) le16_to_cpu(hdr
->type
);
3727 int front_len
= (int) le32_to_cpu(hdr
->front_len
);
3733 msg
= ceph_msg_new(type
, front_len
, GFP_NOFS
, false);
3735 pr_err("unable to allocate msg type %d len %d\n",
3743 static const struct ceph_connection_operations mds_con_ops
= {
3746 .dispatch
= dispatch
,
3747 .get_authorizer
= get_authorizer
,
3748 .verify_authorizer_reply
= verify_authorizer_reply
,
3749 .invalidate_authorizer
= invalidate_authorizer
,
3750 .peer_reset
= peer_reset
,
3751 .alloc_msg
= mds_alloc_msg
,