]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ceph/mds_client.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'linux_next' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
[net-next-2.6.git] / fs / ceph / mds_client.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
2f2dc053 2
496e5955 3#include <linux/fs.h>
2f2dc053 4#include <linux/wait.h>
5a0e3ad6 5#include <linux/slab.h>
2f2dc053 6#include <linux/sched.h>
3d14c5d2
YS		 7#include <linux/debugfs.h>
8#include <linux/seq_file.h>
496e5955 9#include <linux/smp_lock.h>
2f2dc053 10
2f2dc053 11#include "super.h"
3d14c5d2
YS		 12#include "mds_client.h"
13
14#include <linux/ceph/messenger.h>
15#include <linux/ceph/decode.h>
16#include <linux/ceph/pagelist.h>
17#include <linux/ceph/auth.h>
18#include <linux/ceph/debugfs.h>
2f2dc053
SW		 19
20/*
21 * A cluster of MDS (metadata server) daemons is responsible for
22 * managing the file system namespace (the directory hierarchy and
23 * inodes) and for coordinating shared access to storage. Metadata is
24 * partitioning hierarchically across a number of servers, and that
25 * partition varies over time as the cluster adjusts the distribution
26 * in order to balance load.
27 *
28 * The MDS client is primarily responsible to managing synchronous
29 * metadata requests for operations like open, unlink, and so forth.
30 * If there is a MDS failure, we find out about it when we (possibly
31 * request and) receive a new MDS map, and can resubmit affected
32 * requests.
33 *
34 * For the most part, though, we take advantage of a lossless
35 * communications channel to the MDS, and do not need to worry about
36 * timing out or resubmitting requests.
37 *
38 * We maintain a stateful "session" with each MDS we interact with.
39 * Within each session, we sent periodic heartbeat messages to ensure
40 * any capabilities or leases we have been issues remain valid. If
41 * the session times out and goes stale, our leases and capabilities
42 * are no longer valid.
43 */
44
20cb34ae
SW		 45struct ceph_reconnect_state {
46 struct ceph_pagelist *pagelist;
47 bool flock;
48};
49
2f2dc053
SW		 50static void __wake_requests(struct ceph_mds_client *mdsc,
51 struct list_head *head);
52
9e32789f 53static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 54
55
56/*
57 * mds reply parsing
58 */
59
60/*
61 * parse individual inode info
62 */
63static int parse_reply_info_in(void **p, void *end,
64 struct ceph_mds_reply_info_in *info)
65{
66 int err = -EIO;
67
68 info->in = *p;
69 *p += sizeof(struct ceph_mds_reply_inode) +
70 sizeof(*info->in->fragtree.splits) *
71 le32_to_cpu(info->in->fragtree.nsplits);
72
73 ceph_decode_32_safe(p, end, info->symlink_len, bad);
74 ceph_decode_need(p, end, info->symlink_len, bad);
75 info->symlink = *p;
76 *p += info->symlink_len;
77
78 ceph_decode_32_safe(p, end, info->xattr_len, bad);
79 ceph_decode_need(p, end, info->xattr_len, bad);
80 info->xattr_data = *p;
81 *p += info->xattr_len;
82 return 0;
83bad:
84 return err;
85}
86
87/*
88 * parse a normal reply, which may contain a (dir+)dentry and/or a
89 * target inode.
90 */
91static int parse_reply_info_trace(void **p, void *end,
92 struct ceph_mds_reply_info_parsed *info)
93{
94 int err;
95
96 if (info->head->is_dentry) {
97 err = parse_reply_info_in(p, end, &info->diri);
98 if (err < 0)
99 goto out_bad;
100
101 if (unlikely(*p + sizeof(*info->dirfrag) > end))
102 goto bad;
103 info->dirfrag = *p;
104 *p += sizeof(*info->dirfrag) +
105 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
106 if (unlikely(*p > end))
107 goto bad;
108
109 ceph_decode_32_safe(p, end, info->dname_len, bad);
110 ceph_decode_need(p, end, info->dname_len, bad);
111 info->dname = *p;
112 *p += info->dname_len;
113 info->dlease = *p;
114 *p += sizeof(*info->dlease);
115 }
116
117 if (info->head->is_target) {
118 err = parse_reply_info_in(p, end, &info->targeti);
119 if (err < 0)
120 goto out_bad;
121 }
122
123 if (unlikely(*p != end))
124 goto bad;
125 return 0;
126
127bad:
128 err = -EIO;
129out_bad:
130 pr_err("problem parsing mds trace %d\n", err);
131 return err;
132}
133
134/*
135 * parse readdir results
136 */
137static int parse_reply_info_dir(void **p, void *end,
138 struct ceph_mds_reply_info_parsed *info)
139{
140 u32 num, i = 0;
141 int err;
142
143 info->dir_dir = *p;
144 if (*p + sizeof(*info->dir_dir) > end)
145 goto bad;
146 *p += sizeof(*info->dir_dir) +
147 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
148 if (*p > end)
149 goto bad;
150
151 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea
SW		 152 num = ceph_decode_32(p);
153 info->dir_end = ceph_decode_8(p);
154 info->dir_complete = ceph_decode_8(p);
2f2dc053
SW		 155 if (num == 0)
156 goto done;
157
158 /* alloc large array */
159 info->dir_nr = num;
160 info->dir_in = kcalloc(num, sizeof(*info->dir_in) +
161 sizeof(*info->dir_dname) +
162 sizeof(*info->dir_dname_len) +
163 sizeof(*info->dir_dlease),
164 GFP_NOFS);
165 if (info->dir_in == NULL) {
166 err = -ENOMEM;
167 goto out_bad;
168 }
169 info->dir_dname = (void *)(info->dir_in + num);
170 info->dir_dname_len = (void *)(info->dir_dname + num);
171 info->dir_dlease = (void *)(info->dir_dname_len + num);
172
173 while (num) {
174 /* dentry */
175 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea 176 info->dir_dname_len[i] = ceph_decode_32(p);
2f2dc053
SW		 177 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
178 info->dir_dname[i] = *p;
179 *p += info->dir_dname_len[i];
180 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
181 info->dir_dname[i]);
182 info->dir_dlease[i] = *p;
183 *p += sizeof(struct ceph_mds_reply_lease);
184
185 /* inode */
186 err = parse_reply_info_in(p, end, &info->dir_in[i]);
187 if (err < 0)
188 goto out_bad;
189 i++;
190 num--;
191 }
192
193done:
194 if (*p != end)
195 goto bad;
196 return 0;
197
198bad:
199 err = -EIO;
200out_bad:
201 pr_err("problem parsing dir contents %d\n", err);
202 return err;
203}
204
205/*
206 * parse entire mds reply
207 */
208static int parse_reply_info(struct ceph_msg *msg,
209 struct ceph_mds_reply_info_parsed *info)
210{
211 void *p, *end;
212 u32 len;
213 int err;
214
215 info->head = msg->front.iov_base;
216 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
217 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
218
219 /* trace */
220 ceph_decode_32_safe(&p, end, len, bad);
221 if (len > 0) {
222 err = parse_reply_info_trace(&p, p+len, info);
223 if (err < 0)
224 goto out_bad;
225 }
226
227 /* dir content */
228 ceph_decode_32_safe(&p, end, len, bad);
229 if (len > 0) {
230 err = parse_reply_info_dir(&p, p+len, info);
231 if (err < 0)
232 goto out_bad;
233 }
234
235 /* snap blob */
236 ceph_decode_32_safe(&p, end, len, bad);
237 info->snapblob_len = len;
238 info->snapblob = p;
239 p += len;
240
241 if (p != end)
242 goto bad;
243 return 0;
244
245bad:
246 err = -EIO;
247out_bad:
248 pr_err("mds parse_reply err %d\n", err);
249 return err;
250}
251
252static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
253{
254 kfree(info->dir_in);
255}
256
257
258/*
259 * sessions
260 */
261static const char *session_state_name(int s)
262{
263 switch (s) {
264 case CEPH_MDS_SESSION_NEW: return "new";
265 case CEPH_MDS_SESSION_OPENING: return "opening";
266 case CEPH_MDS_SESSION_OPEN: return "open";
267 case CEPH_MDS_SESSION_HUNG: return "hung";
268 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 269 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053
SW		 270 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
271 default: return "???";
272 }
273}
274
275static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
276{
277 if (atomic_inc_not_zero(&s->s_ref)) {
278 dout("mdsc get_session %p %d -> %d\n", s,
279 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
280 return s;
281 } else {
282 dout("mdsc get_session %p 0 -- FAIL", s);
283 return NULL;
284 }
285}
286
287void ceph_put_mds_session(struct ceph_mds_session *s)
288{
289 dout("mdsc put_session %p %d -> %d\n", s,
290 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
4e7a5dcd
SW		 291 if (atomic_dec_and_test(&s->s_ref)) {
292 if (s->s_authorizer)
3d14c5d2
YS		 293 s->s_mdsc->fsc->client->monc.auth->ops->destroy_authorizer(
294 s->s_mdsc->fsc->client->monc.auth,
295 s->s_authorizer);
2f2dc053 296 kfree(s);
4e7a5dcd 297 }
2f2dc053
SW		 298}
299
300/*
301 * called under mdsc->mutex
302 */
303struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
304 int mds)
305{
306 struct ceph_mds_session *session;
307
308 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
309 return NULL;
310 session = mdsc->sessions[mds];
311 dout("lookup_mds_session %p %d\n", session,
312 atomic_read(&session->s_ref));
313 get_session(session);
314 return session;
315}
316
317static bool __have_session(struct ceph_mds_client *mdsc, int mds)
318{
319 if (mds >= mdsc->max_sessions)
320 return false;
321 return mdsc->sessions[mds];
322}
323
2600d2dd
SW		 324static int __verify_registered_session(struct ceph_mds_client *mdsc,
325 struct ceph_mds_session *s)
326{
327 if (s->s_mds >= mdsc->max_sessions ||
328 mdsc->sessions[s->s_mds] != s)
329 return -ENOENT;
330 return 0;
331}
332
2f2dc053
SW		 333/*
334 * create+register a new session for given mds.
335 * called under mdsc->mutex.
336 */
337static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
338 int mds)
339{
340 struct ceph_mds_session *s;
341
342 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 343 if (!s)
344 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 345 s->s_mdsc = mdsc;
346 s->s_mds = mds;
347 s->s_state = CEPH_MDS_SESSION_NEW;
348 s->s_ttl = 0;
349 s->s_seq = 0;
350 mutex_init(&s->s_mutex);
351
3d14c5d2 352 ceph_con_init(mdsc->fsc->client->msgr, &s->s_con);
2f2dc053
SW		 353 s->s_con.private = s;
354 s->s_con.ops = &mds_con_ops;
355 s->s_con.peer_name.type = CEPH_ENTITY_TYPE_MDS;
356 s->s_con.peer_name.num = cpu_to_le64(mds);
2f2dc053
SW		 357
358 spin_lock_init(&s->s_cap_lock);
359 s->s_cap_gen = 0;
360 s->s_cap_ttl = 0;
361 s->s_renew_requested = 0;
362 s->s_renew_seq = 0;
363 INIT_LIST_HEAD(&s->s_caps);
364 s->s_nr_caps = 0;
5dacf091 365 s->s_trim_caps = 0;
2f2dc053
SW		 366 atomic_set(&s->s_ref, 1);
367 INIT_LIST_HEAD(&s->s_waiting);
368 INIT_LIST_HEAD(&s->s_unsafe);
369 s->s_num_cap_releases = 0;
7c1332b8 370 s->s_cap_iterator = NULL;
2f2dc053
SW		 371 INIT_LIST_HEAD(&s->s_cap_releases);
372 INIT_LIST_HEAD(&s->s_cap_releases_done);
373 INIT_LIST_HEAD(&s->s_cap_flushing);
374 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
375
376 dout("register_session mds%d\n", mds);
377 if (mds >= mdsc->max_sessions) {
378 int newmax = 1 << get_count_order(mds+1);
379 struct ceph_mds_session **sa;
380
381 dout("register_session realloc to %d\n", newmax);
382 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
383 if (sa == NULL)
42ce56e5 384 goto fail_realloc;
2f2dc053
SW		 385 if (mdsc->sessions) {
386 memcpy(sa, mdsc->sessions,
387 mdsc->max_sessions * sizeof(void *));
388 kfree(mdsc->sessions);
389 }
390 mdsc->sessions = sa;
391 mdsc->max_sessions = newmax;
392 }
393 mdsc->sessions[mds] = s;
394 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5
SW		 395
396 ceph_con_open(&s->s_con, ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
397
2f2dc053 398 return s;
42ce56e5
SW		 399
400fail_realloc:
401 kfree(s);
402 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 403}
404
405/*
406 * called under mdsc->mutex
407 */
2600d2dd 408static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 409 struct ceph_mds_session *s)
2f2dc053 410{
2600d2dd
SW		 411 dout("__unregister_session mds%d %p\n", s->s_mds, s);
412 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 413 mdsc->sessions[s->s_mds] = NULL;
414 ceph_con_close(&s->s_con);
415 ceph_put_mds_session(s);
2f2dc053
SW		 416}
417
418/*
419 * drop session refs in request.
420 *
421 * should be last request ref, or hold mdsc->mutex
422 */
423static void put_request_session(struct ceph_mds_request *req)
424{
425 if (req->r_session) {
426 ceph_put_mds_session(req->r_session);
427 req->r_session = NULL;
428 }
429}
430
153c8e6b 431void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 432{
153c8e6b
SW		 433 struct ceph_mds_request *req = container_of(kref,
434 struct ceph_mds_request,
435 r_kref);
436 if (req->r_request)
437 ceph_msg_put(req->r_request);
438 if (req->r_reply) {
439 ceph_msg_put(req->r_reply);
440 destroy_reply_info(&req->r_reply_info);
441 }
442 if (req->r_inode) {
443 ceph_put_cap_refs(ceph_inode(req->r_inode),
444 CEPH_CAP_PIN);
445 iput(req->r_inode);
446 }
447 if (req->r_locked_dir)
448 ceph_put_cap_refs(ceph_inode(req->r_locked_dir),
449 CEPH_CAP_PIN);
450 if (req->r_target_inode)
451 iput(req->r_target_inode);
452 if (req->r_dentry)
453 dput(req->r_dentry);
454 if (req->r_old_dentry) {
455 ceph_put_cap_refs(
456 ceph_inode(req->r_old_dentry->d_parent->d_inode),
457 CEPH_CAP_PIN);
458 dput(req->r_old_dentry);
2f2dc053 459 }
153c8e6b
SW		 460 kfree(req->r_path1);
461 kfree(req->r_path2);
462 put_request_session(req);
37151668 463 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 464 kfree(req);
2f2dc053
SW		 465}
466
467/*
468 * lookup session, bump ref if found.
469 *
470 * called under mdsc->mutex.
471 */
472static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
473 u64 tid)
474{
475 struct ceph_mds_request *req;
44ca18f2
SW		 476 struct rb_node *n = mdsc->request_tree.rb_node;
477
478 while (n) {
479 req = rb_entry(n, struct ceph_mds_request, r_node);
480 if (tid < req->r_tid)
481 n = n->rb_left;
482 else if (tid > req->r_tid)
483 n = n->rb_right;
484 else {
485 ceph_mdsc_get_request(req);
486 return req;
487 }
488 }
489 return NULL;
490}
491
492static void __insert_request(struct ceph_mds_client *mdsc,
493 struct ceph_mds_request *new)
494{
495 struct rb_node **p = &mdsc->request_tree.rb_node;
496 struct rb_node *parent = NULL;
497 struct ceph_mds_request *req = NULL;
498
499 while (*p) {
500 parent = *p;
501 req = rb_entry(parent, struct ceph_mds_request, r_node);
502 if (new->r_tid < req->r_tid)
503 p = &(*p)->rb_left;
504 else if (new->r_tid > req->r_tid)
505 p = &(*p)->rb_right;
506 else
507 BUG();
508 }
509
510 rb_link_node(&new->r_node, parent, p);
511 rb_insert_color(&new->r_node, &mdsc->request_tree);
2f2dc053
SW		 512}
513
514/*
515 * Register an in-flight request, and assign a tid. Link to directory
516 * are modifying (if any).
517 *
518 * Called under mdsc->mutex.
519 */
520static void __register_request(struct ceph_mds_client *mdsc,
521 struct ceph_mds_request *req,
522 struct inode *dir)
523{
524 req->r_tid = ++mdsc->last_tid;
525 if (req->r_num_caps)
37151668
YS		 526 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
527 req->r_num_caps);
2f2dc053
SW		 528 dout("__register_request %p tid %lld\n", req, req->r_tid);
529 ceph_mdsc_get_request(req);
44ca18f2 530 __insert_request(mdsc, req);
2f2dc053
SW		 531
532 if (dir) {
533 struct ceph_inode_info *ci = ceph_inode(dir);
534
535 spin_lock(&ci->i_unsafe_lock);
536 req->r_unsafe_dir = dir;
537 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
538 spin_unlock(&ci->i_unsafe_lock);
539 }
540}
541
542static void __unregister_request(struct ceph_mds_client *mdsc,
543 struct ceph_mds_request *req)
544{
545 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
44ca18f2 546 rb_erase(&req->r_node, &mdsc->request_tree);
80fc7314 547 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 548
549 if (req->r_unsafe_dir) {
550 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
551
552 spin_lock(&ci->i_unsafe_lock);
553 list_del_init(&req->r_unsafe_dir_item);
554 spin_unlock(&ci->i_unsafe_lock);
555 }
94aa8ae1
SW		 556
557 ceph_mdsc_put_request(req);
2f2dc053
SW		 558}
559
560/*
561 * Choose mds to send request to next. If there is a hint set in the
562 * request (e.g., due to a prior forward hint from the mds), use that.
563 * Otherwise, consult frag tree and/or caps to identify the
564 * appropriate mds. If all else fails, choose randomly.
565 *
566 * Called under mdsc->mutex.
567 */
eb6bb1c5
SW		 568struct dentry *get_nonsnap_parent(struct dentry *dentry)
569{
570 while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
571 dentry = dentry->d_parent;
572 return dentry;
573}
574
2f2dc053
SW		 575static int __choose_mds(struct ceph_mds_client *mdsc,
576 struct ceph_mds_request *req)
577{
578 struct inode *inode;
579 struct ceph_inode_info *ci;
580 struct ceph_cap *cap;
581 int mode = req->r_direct_mode;
582 int mds = -1;
583 u32 hash = req->r_direct_hash;
584 bool is_hash = req->r_direct_is_hash;
585
586 /*
587 * is there a specific mds we should try? ignore hint if we have
588 * no session and the mds is not up (active or recovering).
589 */
590 if (req->r_resend_mds >= 0 &&
591 (__have_session(mdsc, req->r_resend_mds) ||
592 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
593 dout("choose_mds using resend_mds mds%d\n",
594 req->r_resend_mds);
595 return req->r_resend_mds;
596 }
597
598 if (mode == USE_RANDOM_MDS)
599 goto random;
600
601 inode = NULL;
602 if (req->r_inode) {
603 inode = req->r_inode;
604 } else if (req->r_dentry) {
eb6bb1c5
SW		 605 struct inode *dir = req->r_dentry->d_parent->d_inode;
606
3d14c5d2 607 if (dir->i_sb != mdsc->fsc->sb) {
eb6bb1c5
SW		 608 /* not this fs! */
609 inode = req->r_dentry->d_inode;
610 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
611 /* direct snapped/virtual snapdir requests
612 * based on parent dir inode */
613 struct dentry *dn =
614 get_nonsnap_parent(req->r_dentry->d_parent);
615 inode = dn->d_inode;
616 dout("__choose_mds using nonsnap parent %p\n", inode);
617 } else if (req->r_dentry->d_inode) {
618 /* dentry target */
2f2dc053
SW		 619 inode = req->r_dentry->d_inode;
620 } else {
eb6bb1c5
SW		 621 /* dir + name */
622 inode = dir;
2f2dc053
SW		 623 hash = req->r_dentry->d_name.hash;
624 is_hash = true;
625 }
626 }
eb6bb1c5 627
2f2dc053
SW		 628 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
629 (int)hash, mode);
630 if (!inode)
631 goto random;
632 ci = ceph_inode(inode);
633
634 if (is_hash && S_ISDIR(inode->i_mode)) {
635 struct ceph_inode_frag frag;
636 int found;
637
638 ceph_choose_frag(ci, hash, &frag, &found);
639 if (found) {
640 if (mode == USE_ANY_MDS && frag.ndist > 0) {
641 u8 r;
642
643 /* choose a random replica */
644 get_random_bytes(&r, 1);
645 r %= frag.ndist;
646 mds = frag.dist[r];
647 dout("choose_mds %p %llx.%llx "
648 "frag %u mds%d (%d/%d)\n",
649 inode, ceph_vinop(inode),
650 frag.frag, frag.mds,
651 (int)r, frag.ndist);
652 return mds;
653 }
654
655 /* since this file/dir wasn't known to be
656 * replicated, then we want to look for the
657 * authoritative mds. */
658 mode = USE_AUTH_MDS;
659 if (frag.mds >= 0) {
660 /* choose auth mds */
661 mds = frag.mds;
662 dout("choose_mds %p %llx.%llx "
663 "frag %u mds%d (auth)\n",
664 inode, ceph_vinop(inode), frag.frag, mds);
665 return mds;
666 }
667 }
668 }
669
670 spin_lock(&inode->i_lock);
671 cap = NULL;
672 if (mode == USE_AUTH_MDS)
673 cap = ci->i_auth_cap;
674 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
675 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
676 if (!cap) {
677 spin_unlock(&inode->i_lock);
678 goto random;
679 }
680 mds = cap->session->s_mds;
681 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
682 inode, ceph_vinop(inode), mds,
683 cap == ci->i_auth_cap ? "auth " : "", cap);
684 spin_unlock(&inode->i_lock);
685 return mds;
686
687random:
688 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
689 dout("choose_mds chose random mds%d\n", mds);
690 return mds;
691}
692
693
694/*
695 * session messages
696 */
697static struct ceph_msg *create_session_msg(u32 op, u64 seq)
698{
699 struct ceph_msg *msg;
700 struct ceph_mds_session_head *h;
701
34d23762 702 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS);
a79832f2 703 if (!msg) {
2f2dc053 704 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 705 return NULL;
2f2dc053
SW		 706 }
707 h = msg->front.iov_base;
708 h->op = cpu_to_le32(op);
709 h->seq = cpu_to_le64(seq);
710 return msg;
711}
712
713/*
714 * send session open request.
715 *
716 * called under mdsc->mutex
717 */
718static int __open_session(struct ceph_mds_client *mdsc,
719 struct ceph_mds_session *session)
720{
721 struct ceph_msg *msg;
722 int mstate;
723 int mds = session->s_mds;
2f2dc053
SW		 724
725 /* wait for mds to go active? */
726 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
727 dout("open_session to mds%d (%s)\n", mds,
728 ceph_mds_state_name(mstate));
729 session->s_state = CEPH_MDS_SESSION_OPENING;
730 session->s_renew_requested = jiffies;
731
732 /* send connect message */
733 msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
a79832f2
SW		 734 if (!msg)
735 return -ENOMEM;
2f2dc053 736 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 737 return 0;
738}
739
ed0552a1
SW		 740/*
741 * open sessions for any export targets for the given mds
742 *
743 * called under mdsc->mutex
744 */
745static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
746 struct ceph_mds_session *session)
747{
748 struct ceph_mds_info *mi;
749 struct ceph_mds_session *ts;
750 int i, mds = session->s_mds;
751 int target;
752
753 if (mds >= mdsc->mdsmap->m_max_mds)
754 return;
755 mi = &mdsc->mdsmap->m_info[mds];
756 dout("open_export_target_sessions for mds%d (%d targets)\n",
757 session->s_mds, mi->num_export_targets);
758
759 for (i = 0; i < mi->num_export_targets; i++) {
760 target = mi->export_targets[i];
761 ts = __ceph_lookup_mds_session(mdsc, target);
762 if (!ts) {
763 ts = register_session(mdsc, target);
764 if (IS_ERR(ts))
765 return;
766 }
767 if (session->s_state == CEPH_MDS_SESSION_NEW ||
768 session->s_state == CEPH_MDS_SESSION_CLOSING)
769 __open_session(mdsc, session);
770 else
771 dout(" mds%d target mds%d %p is %s\n", session->s_mds,
772 i, ts, session_state_name(ts->s_state));
773 ceph_put_mds_session(ts);
774 }
775}
776
154f42c2
SW		 777void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
778 struct ceph_mds_session *session)
779{
780 mutex_lock(&mdsc->mutex);
781 __open_export_target_sessions(mdsc, session);
782 mutex_unlock(&mdsc->mutex);
783}
784
2f2dc053
SW		 785/*
786 * session caps
787 */
788
789/*
790 * Free preallocated cap messages assigned to this session
791 */
792static void cleanup_cap_releases(struct ceph_mds_session *session)
793{
794 struct ceph_msg *msg;
795
796 spin_lock(&session->s_cap_lock);
797 while (!list_empty(&session->s_cap_releases)) {
798 msg = list_first_entry(&session->s_cap_releases,
799 struct ceph_msg, list_head);
800 list_del_init(&msg->list_head);
801 ceph_msg_put(msg);
802 }
803 while (!list_empty(&session->s_cap_releases_done)) {
804 msg = list_first_entry(&session->s_cap_releases_done,
805 struct ceph_msg, list_head);
806 list_del_init(&msg->list_head);
807 ceph_msg_put(msg);
808 }
809 spin_unlock(&session->s_cap_lock);
810}
811
812/*
f818a736
SW		 813 * Helper to safely iterate over all caps associated with a session, with
814 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 815 *
f818a736 816 * Caller must hold session s_mutex.
2f2dc053
SW		 817 */
818static int iterate_session_caps(struct ceph_mds_session *session,
819 int (*cb)(struct inode *, struct ceph_cap *,
820 void *), void *arg)
821{
7c1332b8
SW		 822 struct list_head *p;
823 struct ceph_cap *cap;
824 struct inode *inode, *last_inode = NULL;
825 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 826 int ret;
827
828 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
829 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 830 p = session->s_caps.next;
831 while (p != &session->s_caps) {
832 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 833 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 834 if (!inode) {
835 p = p->next;
2f2dc053 836 continue;
7c1332b8
SW		 837 }
838 session->s_cap_iterator = cap;
2f2dc053 839 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 840
841 if (last_inode) {
842 iput(last_inode);
843 last_inode = NULL;
844 }
845 if (old_cap) {
37151668 846 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 847 old_cap = NULL;
848 }
849
2f2dc053 850 ret = cb(inode, cap, arg);
7c1332b8
SW		 851 last_inode = inode;
852
2f2dc053 853 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 854 p = p->next;
855 if (cap->ci == NULL) {
856 dout("iterate_session_caps finishing cap %p removal\n",
857 cap);
858 BUG_ON(cap->session != session);
859 list_del_init(&cap->session_caps);
860 session->s_nr_caps--;
861 cap->session = NULL;
862 old_cap = cap; /* put_cap it w/o locks held */
863 }
5dacf091
SW		 864 if (ret < 0)
865 goto out;
2f2dc053 866 }
5dacf091
SW		 867 ret = 0;
868out:
7c1332b8 869 session->s_cap_iterator = NULL;
2f2dc053 870 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 871
872 if (last_inode)
873 iput(last_inode);
874 if (old_cap)
37151668 875 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 876
5dacf091 877 return ret;
2f2dc053
SW		 878}
879
880static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 881 void *arg)
2f2dc053
SW		 882{
883 struct ceph_inode_info *ci = ceph_inode(inode);
6c99f254
SW		 884 int drop = 0;
885
2f2dc053
SW		 886 dout("removing cap %p, ci is %p, inode is %p\n",
887 cap, ci, &ci->vfs_inode);
6c99f254
SW		 888 spin_lock(&inode->i_lock);
889 __ceph_remove_cap(cap);
890 if (!__ceph_is_any_real_caps(ci)) {
891 struct ceph_mds_client *mdsc =
3d14c5d2 892 ceph_sb_to_client(inode->i_sb)->mdsc;
6c99f254
SW		 893
894 spin_lock(&mdsc->cap_dirty_lock);
895 if (!list_empty(&ci->i_dirty_item)) {
896 pr_info(" dropping dirty %s state for %p %lld\n",
897 ceph_cap_string(ci->i_dirty_caps),
898 inode, ceph_ino(inode));
899 ci->i_dirty_caps = 0;
900 list_del_init(&ci->i_dirty_item);
901 drop = 1;
902 }
903 if (!list_empty(&ci->i_flushing_item)) {
904 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
905 ceph_cap_string(ci->i_flushing_caps),
906 inode, ceph_ino(inode));
907 ci->i_flushing_caps = 0;
908 list_del_init(&ci->i_flushing_item);
909 mdsc->num_cap_flushing--;
910 drop = 1;
911 }
912 if (drop && ci->i_wrbuffer_ref) {
913 pr_info(" dropping dirty data for %p %lld\n",
914 inode, ceph_ino(inode));
915 ci->i_wrbuffer_ref = 0;
916 ci->i_wrbuffer_ref_head = 0;
917 drop++;
918 }
919 spin_unlock(&mdsc->cap_dirty_lock);
920 }
921 spin_unlock(&inode->i_lock);
922 while (drop--)
923 iput(inode);
2f2dc053
SW		 924 return 0;
925}
926
927/*
928 * caller must hold session s_mutex
929 */
930static void remove_session_caps(struct ceph_mds_session *session)
931{
932 dout("remove_session_caps on %p\n", session);
933 iterate_session_caps(session, remove_session_caps_cb, NULL);
934 BUG_ON(session->s_nr_caps > 0);
6c99f254 935 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 936 cleanup_cap_releases(session);
937}
938
939/*
940 * wake up any threads waiting on this session's caps. if the cap is
941 * old (didn't get renewed on the client reconnect), remove it now.
942 *
943 * caller must hold s_mutex.
944 */
945static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
946 void *arg)
947{
0dc2570f
SW		 948 struct ceph_inode_info *ci = ceph_inode(inode);
949
03066f23 950 wake_up_all(&ci->i_cap_wq);
0dc2570f
SW		 951 if (arg) {
952 spin_lock(&inode->i_lock);
953 ci->i_wanted_max_size = 0;
954 ci->i_requested_max_size = 0;
955 spin_unlock(&inode->i_lock);
956 }
2f2dc053
SW		 957 return 0;
958}
959
0dc2570f
SW		 960static void wake_up_session_caps(struct ceph_mds_session *session,
961 int reconnect)
2f2dc053
SW		 962{
963 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 964 iterate_session_caps(session, wake_up_session_cb,
965 (void *)(unsigned long)reconnect);
2f2dc053
SW		 966}
967
968/*
969 * Send periodic message to MDS renewing all currently held caps. The
970 * ack will reset the expiration for all caps from this session.
971 *
972 * caller holds s_mutex
973 */
974static int send_renew_caps(struct ceph_mds_client *mdsc,
975 struct ceph_mds_session *session)
976{
977 struct ceph_msg *msg;
978 int state;
979
980 if (time_after_eq(jiffies, session->s_cap_ttl) &&
981 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
982 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 983 session->s_renew_requested = jiffies;
2f2dc053
SW		 984
985 /* do not try to renew caps until a recovering mds has reconnected
986 * with its clients. */
987 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
988 if (state < CEPH_MDS_STATE_RECONNECT) {
989 dout("send_renew_caps ignoring mds%d (%s)\n",
990 session->s_mds, ceph_mds_state_name(state));
991 return 0;
992 }
993
994 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
995 ceph_mds_state_name(state));
2f2dc053
SW		 996 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
997 ++session->s_renew_seq);
a79832f2
SW		 998 if (!msg)
999 return -ENOMEM;
2f2dc053
SW		 1000 ceph_con_send(&session->s_con, msg);
1001 return 0;
1002}
1003
1004/*
1005 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1006 *
1007 * Called under session->s_mutex
2f2dc053
SW		 1008 */
1009static void renewed_caps(struct ceph_mds_client *mdsc,
1010 struct ceph_mds_session *session, int is_renew)
1011{
1012 int was_stale;
1013 int wake = 0;
1014
1015 spin_lock(&session->s_cap_lock);
1016 was_stale = is_renew && (session->s_cap_ttl == 0 ||
1017 time_after_eq(jiffies, session->s_cap_ttl));
1018
1019 session->s_cap_ttl = session->s_renew_requested +
1020 mdsc->mdsmap->m_session_timeout*HZ;
1021
1022 if (was_stale) {
1023 if (time_before(jiffies, session->s_cap_ttl)) {
1024 pr_info("mds%d caps renewed\n", session->s_mds);
1025 wake = 1;
1026 } else {
1027 pr_info("mds%d caps still stale\n", session->s_mds);
1028 }
1029 }
1030 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1031 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1032 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1033 spin_unlock(&session->s_cap_lock);
1034
1035 if (wake)
0dc2570f 1036 wake_up_session_caps(session, 0);
2f2dc053
SW		 1037}
1038
1039/*
1040 * send a session close request
1041 */
1042static int request_close_session(struct ceph_mds_client *mdsc,
1043 struct ceph_mds_session *session)
1044{
1045 struct ceph_msg *msg;
2f2dc053
SW		 1046
1047 dout("request_close_session mds%d state %s seq %lld\n",
1048 session->s_mds, session_state_name(session->s_state),
1049 session->s_seq);
1050 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1051 if (!msg)
1052 return -ENOMEM;
1053 ceph_con_send(&session->s_con, msg);
1054 return 0;
2f2dc053
SW		 1055}
1056
1057/*
1058 * Called with s_mutex held.
1059 */
1060static int __close_session(struct ceph_mds_client *mdsc,
1061 struct ceph_mds_session *session)
1062{
1063 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1064 return 0;
1065 session->s_state = CEPH_MDS_SESSION_CLOSING;
1066 return request_close_session(mdsc, session);
1067}
1068
1069/*
1070 * Trim old(er) caps.
1071 *
1072 * Because we can't cache an inode without one or more caps, we do
1073 * this indirectly: if a cap is unused, we prune its aliases, at which
1074 * point the inode will hopefully get dropped to.
1075 *
1076 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1077 * memory pressure from the MDS, though, so it needn't be perfect.
1078 */
1079static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1080{
1081 struct ceph_mds_session *session = arg;
1082 struct ceph_inode_info *ci = ceph_inode(inode);
1083 int used, oissued, mine;
1084
1085 if (session->s_trim_caps <= 0)
1086 return -1;
1087
1088 spin_lock(&inode->i_lock);
1089 mine = cap->issued | cap->implemented;
1090 used = __ceph_caps_used(ci);
1091 oissued = __ceph_caps_issued_other(ci, cap);
1092
1093 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n",
1094 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1095 ceph_cap_string(used));
1096 if (ci->i_dirty_caps)
1097 goto out; /* dirty caps */
1098 if ((used & ~oissued) & mine)
1099 goto out; /* we need these caps */
1100
1101 session->s_trim_caps--;
1102 if (oissued) {
1103 /* we aren't the only cap.. just remove us */
7c1332b8 1104 __ceph_remove_cap(cap);
2f2dc053
SW		 1105 } else {
1106 /* try to drop referring dentries */
1107 spin_unlock(&inode->i_lock);
1108 d_prune_aliases(inode);
1109 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1110 inode, cap, atomic_read(&inode->i_count));
1111 return 0;
1112 }
1113
1114out:
1115 spin_unlock(&inode->i_lock);
1116 return 0;
1117}
1118
1119/*
1120 * Trim session cap count down to some max number.
1121 */
1122static int trim_caps(struct ceph_mds_client *mdsc,
1123 struct ceph_mds_session *session,
1124 int max_caps)
1125{
1126 int trim_caps = session->s_nr_caps - max_caps;
1127
1128 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1129 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1130 if (trim_caps > 0) {
1131 session->s_trim_caps = trim_caps;
1132 iterate_session_caps(session, trim_caps_cb, session);
1133 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1134 session->s_mds, session->s_nr_caps, max_caps,
1135 trim_caps - session->s_trim_caps);
5dacf091 1136 session->s_trim_caps = 0;
2f2dc053
SW		 1137 }
1138 return 0;
1139}
1140
1141/*
1142 * Allocate cap_release messages. If there is a partially full message
1143 * in the queue, try to allocate enough to cover it's remainder, so that
1144 * we can send it immediately.
1145 *
1146 * Called under s_mutex.
1147 */
2b2300d6 1148int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
ee6b272b 1149 struct ceph_mds_session *session)
2f2dc053 1150{
38e8883e 1151 struct ceph_msg *msg, *partial = NULL;
2f2dc053
SW		 1152 struct ceph_mds_cap_release *head;
1153 int err = -ENOMEM;
3d14c5d2 1154 int extra = mdsc->fsc->mount_options->cap_release_safety;
38e8883e 1155 int num;
2f2dc053 1156
38e8883e
SW		 1157 dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1158 extra);
2f2dc053
SW		 1159
1160 spin_lock(&session->s_cap_lock);
1161
1162 if (!list_empty(&session->s_cap_releases)) {
1163 msg = list_first_entry(&session->s_cap_releases,
1164 struct ceph_msg,
1165 list_head);
1166 head = msg->front.iov_base;
38e8883e
SW		 1167 num = le32_to_cpu(head->num);
1168 if (num) {
1169 dout(" partial %p with (%d/%d)\n", msg, num,
1170 (int)CEPH_CAPS_PER_RELEASE);
1171 extra += CEPH_CAPS_PER_RELEASE - num;
1172 partial = msg;
1173 }
2f2dc053 1174 }
2f2dc053
SW		 1175 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1176 spin_unlock(&session->s_cap_lock);
34d23762
YS		 1177 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1178 GFP_NOFS);
2f2dc053
SW		 1179 if (!msg)
1180 goto out_unlocked;
1181 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1182 (int)msg->front.iov_len);
1183 head = msg->front.iov_base;
1184 head->num = cpu_to_le32(0);
1185 msg->front.iov_len = sizeof(*head);
1186 spin_lock(&session->s_cap_lock);
1187 list_add(&msg->list_head, &session->s_cap_releases);
1188 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1189 }
1190
38e8883e
SW		 1191 if (partial) {
1192 head = partial->front.iov_base;
1193 num = le32_to_cpu(head->num);
1194 dout(" queueing partial %p with %d/%d\n", partial, num,
1195 (int)CEPH_CAPS_PER_RELEASE);
1196 list_move_tail(&partial->list_head,
1197 &session->s_cap_releases_done);
1198 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
2f2dc053
SW		 1199 }
1200 err = 0;
1201 spin_unlock(&session->s_cap_lock);
1202out_unlocked:
1203 return err;
1204}
1205
1206/*
1207 * flush all dirty inode data to disk.
1208 *
1209 * returns true if we've flushed through want_flush_seq
1210 */
1211static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1212{
1213 int mds, ret = 1;
1214
1215 dout("check_cap_flush want %lld\n", want_flush_seq);
1216 mutex_lock(&mdsc->mutex);
1217 for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1218 struct ceph_mds_session *session = mdsc->sessions[mds];
1219
1220 if (!session)
1221 continue;
1222 get_session(session);
1223 mutex_unlock(&mdsc->mutex);
1224
1225 mutex_lock(&session->s_mutex);
1226 if (!list_empty(&session->s_cap_flushing)) {
1227 struct ceph_inode_info *ci =
1228 list_entry(session->s_cap_flushing.next,
1229 struct ceph_inode_info,
1230 i_flushing_item);
1231 struct inode *inode = &ci->vfs_inode;
1232
1233 spin_lock(&inode->i_lock);
1234 if (ci->i_cap_flush_seq <= want_flush_seq) {
1235 dout("check_cap_flush still flushing %p "
1236 "seq %lld <= %lld to mds%d\n", inode,
1237 ci->i_cap_flush_seq, want_flush_seq,
1238 session->s_mds);
1239 ret = 0;
1240 }
1241 spin_unlock(&inode->i_lock);
1242 }
1243 mutex_unlock(&session->s_mutex);
1244 ceph_put_mds_session(session);
1245
1246 if (!ret)
1247 return ret;
1248 mutex_lock(&mdsc->mutex);
1249 }
1250
1251 mutex_unlock(&mdsc->mutex);
1252 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1253 return ret;
1254}
1255
1256/*
1257 * called under s_mutex
1258 */
3d7ded4d
SW		 1259void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1260 struct ceph_mds_session *session)
2f2dc053
SW		 1261{
1262 struct ceph_msg *msg;
1263
1264 dout("send_cap_releases mds%d\n", session->s_mds);
0f8605f2
SW		 1265 spin_lock(&session->s_cap_lock);
1266 while (!list_empty(&session->s_cap_releases_done)) {
2f2dc053
SW		 1267 msg = list_first_entry(&session->s_cap_releases_done,
1268 struct ceph_msg, list_head);
1269 list_del_init(&msg->list_head);
1270 spin_unlock(&session->s_cap_lock);
1271 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1272 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1273 ceph_con_send(&session->s_con, msg);
0f8605f2 1274 spin_lock(&session->s_cap_lock);
2f2dc053
SW		 1275 }
1276 spin_unlock(&session->s_cap_lock);
1277}
1278
e01a5946
SW		 1279static void discard_cap_releases(struct ceph_mds_client *mdsc,
1280 struct ceph_mds_session *session)
1281{
1282 struct ceph_msg *msg;
1283 struct ceph_mds_cap_release *head;
1284 unsigned num;
1285
1286 dout("discard_cap_releases mds%d\n", session->s_mds);
1287 spin_lock(&session->s_cap_lock);
1288
1289 /* zero out the in-progress message */
1290 msg = list_first_entry(&session->s_cap_releases,
1291 struct ceph_msg, list_head);
1292 head = msg->front.iov_base;
1293 num = le32_to_cpu(head->num);
1294 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg, num);
1295 head->num = cpu_to_le32(0);
1296 session->s_num_cap_releases += num;
1297
1298 /* requeue completed messages */
1299 while (!list_empty(&session->s_cap_releases_done)) {
1300 msg = list_first_entry(&session->s_cap_releases_done,
1301 struct ceph_msg, list_head);
1302 list_del_init(&msg->list_head);
1303
1304 head = msg->front.iov_base;
1305 num = le32_to_cpu(head->num);
1306 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1307 num);
1308 session->s_num_cap_releases += num;
1309 head->num = cpu_to_le32(0);
1310 msg->front.iov_len = sizeof(*head);
1311 list_add(&msg->list_head, &session->s_cap_releases);
1312 }
1313
1314 spin_unlock(&session->s_cap_lock);
1315}
1316
2f2dc053
SW		 1317/*
1318 * requests
1319 */
1320
1321/*
1322 * Create an mds request.
1323 */
1324struct ceph_mds_request *
1325ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1326{
1327 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1328
1329 if (!req)
1330 return ERR_PTR(-ENOMEM);
1331
b4556396 1332 mutex_init(&req->r_fill_mutex);
37151668 1333 req->r_mdsc = mdsc;
2f2dc053
SW		 1334 req->r_started = jiffies;
1335 req->r_resend_mds = -1;
1336 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1337 req->r_fmode = -1;
153c8e6b 1338 kref_init(&req->r_kref);
2f2dc053
SW		 1339 INIT_LIST_HEAD(&req->r_wait);
1340 init_completion(&req->r_completion);
1341 init_completion(&req->r_safe_completion);
1342 INIT_LIST_HEAD(&req->r_unsafe_item);
1343
1344 req->r_op = op;
1345 req->r_direct_mode = mode;
1346 return req;
1347}
1348
1349/*
44ca18f2 1350 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1351 *
1352 * called under mdsc->mutex.
1353 */
44ca18f2
SW		 1354static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1355{
1356 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1357 return NULL;
1358 return rb_entry(rb_first(&mdsc->request_tree),
1359 struct ceph_mds_request, r_node);
1360}
1361
2f2dc053
SW		 1362static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1363{
44ca18f2
SW		 1364 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1365
1366 if (req)
1367 return req->r_tid;
1368 return 0;
2f2dc053
SW		 1369}
1370
1371/*
1372 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1373 * on build_path_from_dentry in fs/cifs/dir.c.
1374 *
1375 * If @stop_on_nosnap, generate path relative to the first non-snapped
1376 * inode.
1377 *
1378 * Encode hidden .snap dirs as a double /, i.e.
1379 * foo/.snap/bar -> foo//bar
1380 */
1381char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1382 int stop_on_nosnap)
1383{
1384 struct dentry *temp;
1385 char *path;
1386 int len, pos;
1387
1388 if (dentry == NULL)
1389 return ERR_PTR(-EINVAL);
1390
1391retry:
1392 len = 0;
1393 for (temp = dentry; !IS_ROOT(temp);) {
1394 struct inode *inode = temp->d_inode;
1395 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1396 len++; /* slash only */
1397 else if (stop_on_nosnap && inode &&
1398 ceph_snap(inode) == CEPH_NOSNAP)
1399 break;
1400 else
1401 len += 1 + temp->d_name.len;
1402 temp = temp->d_parent;
1403 if (temp == NULL) {
6c99f254 1404 pr_err("build_path corrupt dentry %p\n", dentry);
2f2dc053
SW		 1405 return ERR_PTR(-EINVAL);
1406 }
1407 }
1408 if (len)
1409 len--; /* no leading '/' */
1410
1411 path = kmalloc(len+1, GFP_NOFS);
1412 if (path == NULL)
1413 return ERR_PTR(-ENOMEM);
1414 pos = len;
1415 path[pos] = 0; /* trailing null */
1416 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1417 struct inode *inode = temp->d_inode;
1418
1419 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1420 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1421 pos, temp);
1422 } else if (stop_on_nosnap && inode &&
1423 ceph_snap(inode) == CEPH_NOSNAP) {
1424 break;
1425 } else {
1426 pos -= temp->d_name.len;
1427 if (pos < 0)
1428 break;
1429 strncpy(path + pos, temp->d_name.name,
1430 temp->d_name.len);
2f2dc053
SW		 1431 }
1432 if (pos)
1433 path[--pos] = '/';
1434 temp = temp->d_parent;
1435 if (temp == NULL) {
104648ad 1436 pr_err("build_path corrupt dentry\n");
2f2dc053
SW		 1437 kfree(path);
1438 return ERR_PTR(-EINVAL);
1439 }
1440 }
1441 if (pos != 0) {
104648ad 1442 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1443 "expected, namelen is %d, pos is %d\n", len, pos);
1444 /* presumably this is only possible if racing with a
1445 rename of one of the parent directories (we can not
1446 lock the dentries above us to prevent this, but
1447 retrying should be harmless) */
1448 kfree(path);
1449 goto retry;
1450 }
1451
1452 *base = ceph_ino(temp->d_inode);
1453 *plen = len;
104648ad 1454 dout("build_path on %p %d built %llx '%.*s'\n",
2f2dc053
SW		 1455 dentry, atomic_read(&dentry->d_count), *base, len, path);
1456 return path;
1457}
1458
1459static int build_dentry_path(struct dentry *dentry,
1460 const char **ppath, int *ppathlen, u64 *pino,
1461 int *pfreepath)
1462{
1463 char *path;
1464
1465 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1466 *pino = ceph_ino(dentry->d_parent->d_inode);
1467 *ppath = dentry->d_name.name;
1468 *ppathlen = dentry->d_name.len;
1469 return 0;
1470 }
1471 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1472 if (IS_ERR(path))
1473 return PTR_ERR(path);
1474 *ppath = path;
1475 *pfreepath = 1;
1476 return 0;
1477}
1478
1479static int build_inode_path(struct inode *inode,
1480 const char **ppath, int *ppathlen, u64 *pino,
1481 int *pfreepath)
1482{
1483 struct dentry *dentry;
1484 char *path;
1485
1486 if (ceph_snap(inode) == CEPH_NOSNAP) {
1487 *pino = ceph_ino(inode);
1488 *ppathlen = 0;
1489 return 0;
1490 }
1491 dentry = d_find_alias(inode);
1492 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1493 dput(dentry);
1494 if (IS_ERR(path))
1495 return PTR_ERR(path);
1496 *ppath = path;
1497 *pfreepath = 1;
1498 return 0;
1499}
1500
1501/*
1502 * request arguments may be specified via an inode *, a dentry *, or
1503 * an explicit ino+path.
1504 */
1505static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1506 const char *rpath, u64 rino,
1507 const char **ppath, int *pathlen,
1508 u64 *ino, int *freepath)
1509{
1510 int r = 0;
1511
1512 if (rinode) {
1513 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1514 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1515 ceph_snap(rinode));
1516 } else if (rdentry) {
1517 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1518 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1519 *ppath);
1520 } else if (rpath) {
1521 *ino = rino;
1522 *ppath = rpath;
1523 *pathlen = strlen(rpath);
1524 dout(" path %.*s\n", *pathlen, rpath);
1525 }
1526
1527 return r;
1528}
1529
1530/*
1531 * called under mdsc->mutex
1532 */
1533static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1534 struct ceph_mds_request *req,
1535 int mds)
1536{
1537 struct ceph_msg *msg;
1538 struct ceph_mds_request_head *head;
1539 const char *path1 = NULL;
1540 const char *path2 = NULL;
1541 u64 ino1 = 0, ino2 = 0;
1542 int pathlen1 = 0, pathlen2 = 0;
1543 int freepath1 = 0, freepath2 = 0;
1544 int len;
1545 u16 releases;
1546 void *p, *end;
1547 int ret;
1548
1549 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1550 req->r_path1, req->r_ino1.ino,
1551 &path1, &pathlen1, &ino1, &freepath1);
1552 if (ret < 0) {
1553 msg = ERR_PTR(ret);
1554 goto out;
1555 }
1556
1557 ret = set_request_path_attr(NULL, req->r_old_dentry,
1558 req->r_path2, req->r_ino2.ino,
1559 &path2, &pathlen2, &ino2, &freepath2);
1560 if (ret < 0) {
1561 msg = ERR_PTR(ret);
1562 goto out_free1;
1563 }
1564
1565 len = sizeof(*head) +
ac8839d7 1566 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
2f2dc053
SW		 1567
1568 /* calculate (max) length for cap releases */
1569 len += sizeof(struct ceph_mds_request_release) *
1570 (!!req->r_inode_drop + !!req->r_dentry_drop +
1571 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1572 if (req->r_dentry_drop)
1573 len += req->r_dentry->d_name.len;
1574 if (req->r_old_dentry_drop)
1575 len += req->r_old_dentry->d_name.len;
1576
34d23762 1577 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS);
a79832f2
SW		 1578 if (!msg) {
1579 msg = ERR_PTR(-ENOMEM);
2f2dc053 1580 goto out_free2;
a79832f2 1581 }
2f2dc053 1582
6df058c0
SW		 1583 msg->hdr.tid = cpu_to_le64(req->r_tid);
1584
2f2dc053
SW		 1585 head = msg->front.iov_base;
1586 p = msg->front.iov_base + sizeof(*head);
1587 end = msg->front.iov_base + msg->front.iov_len;
1588
1589 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1590 head->op = cpu_to_le32(req->r_op);
1591 head->caller_uid = cpu_to_le32(current_fsuid());
1592 head->caller_gid = cpu_to_le32(current_fsgid());
1593 head->args = req->r_args;
1594
1595 ceph_encode_filepath(&p, end, ino1, path1);
1596 ceph_encode_filepath(&p, end, ino2, path2);
1597
e979cf50
SW		 1598 /* make note of release offset, in case we need to replay */
1599 req->r_request_release_offset = p - msg->front.iov_base;
1600
2f2dc053
SW		 1601 /* cap releases */
1602 releases = 0;
1603 if (req->r_inode_drop)
1604 releases += ceph_encode_inode_release(&p,
1605 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1606 mds, req->r_inode_drop, req->r_inode_unless, 0);
1607 if (req->r_dentry_drop)
1608 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1609 mds, req->r_dentry_drop, req->r_dentry_unless);
1610 if (req->r_old_dentry_drop)
1611 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1612 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1613 if (req->r_old_inode_drop)
1614 releases += ceph_encode_inode_release(&p,
1615 req->r_old_dentry->d_inode,
1616 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1617 head->num_releases = cpu_to_le16(releases);
1618
1619 BUG_ON(p > end);
1620 msg->front.iov_len = p - msg->front.iov_base;
1621 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1622
1623 msg->pages = req->r_pages;
1624 msg->nr_pages = req->r_num_pages;
1625 msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1626 msg->hdr.data_off = cpu_to_le16(0);
1627
1628out_free2:
1629 if (freepath2)
1630 kfree((char *)path2);
1631out_free1:
1632 if (freepath1)
1633 kfree((char *)path1);
1634out:
1635 return msg;
1636}
1637
1638/*
1639 * called under mdsc->mutex if error, under no mutex if
1640 * success.
1641 */
1642static void complete_request(struct ceph_mds_client *mdsc,
1643 struct ceph_mds_request *req)
1644{
1645 if (req->r_callback)
1646 req->r_callback(mdsc, req);
1647 else
03066f23 1648 complete_all(&req->r_completion);
2f2dc053
SW		 1649}
1650
1651/*
1652 * called under mdsc->mutex
1653 */
1654static int __prepare_send_request(struct ceph_mds_client *mdsc,
1655 struct ceph_mds_request *req,
1656 int mds)
1657{
1658 struct ceph_mds_request_head *rhead;
1659 struct ceph_msg *msg;
1660 int flags = 0;
1661
1662 req->r_mds = mds;
1663 req->r_attempts++;
e55b71f8
GF		 1664 if (req->r_inode) {
1665 struct ceph_cap *cap =
1666 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1667
1668 if (cap)
1669 req->r_sent_on_mseq = cap->mseq;
1670 else
1671 req->r_sent_on_mseq = -1;
1672 }
2f2dc053
SW		 1673 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1674 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1675
01a92f17
SW		 1676 if (req->r_got_unsafe) {
1677 /*
1678 * Replay. Do not regenerate message (and rebuild
1679 * paths, etc.); just use the original message.
1680 * Rebuilding paths will break for renames because
1681 * d_move mangles the src name.
1682 */
1683 msg = req->r_request;
1684 rhead = msg->front.iov_base;
1685
1686 flags = le32_to_cpu(rhead->flags);
1687 flags |= CEPH_MDS_FLAG_REPLAY;
1688 rhead->flags = cpu_to_le32(flags);
1689
1690 if (req->r_target_inode)
1691 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1692
1693 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 1694
1695 /* remove cap/dentry releases from message */
1696 rhead->num_releases = 0;
1697 msg->hdr.front_len = cpu_to_le32(req->r_request_release_offset);
1698 msg->front.iov_len = req->r_request_release_offset;
01a92f17
SW		 1699 return 0;
1700 }
1701
2f2dc053
SW		 1702 if (req->r_request) {
1703 ceph_msg_put(req->r_request);
1704 req->r_request = NULL;
1705 }
1706 msg = create_request_message(mdsc, req, mds);
1707 if (IS_ERR(msg)) {
e1518c7c 1708 req->r_err = PTR_ERR(msg);
2f2dc053 1709 complete_request(mdsc, req);
a79832f2 1710 return PTR_ERR(msg);
2f2dc053
SW		 1711 }
1712 req->r_request = msg;
1713
1714 rhead = msg->front.iov_base;
2f2dc053
SW		 1715 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1716 if (req->r_got_unsafe)
1717 flags |= CEPH_MDS_FLAG_REPLAY;
1718 if (req->r_locked_dir)
1719 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1720 rhead->flags = cpu_to_le32(flags);
1721 rhead->num_fwd = req->r_num_fwd;
1722 rhead->num_retry = req->r_attempts - 1;
01a92f17 1723 rhead->ino = 0;
2f2dc053
SW		 1724
1725 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 1726 return 0;
1727}
1728
1729/*
1730 * send request, or put it on the appropriate wait list.
1731 */
1732static int __do_request(struct ceph_mds_client *mdsc,
1733 struct ceph_mds_request *req)
1734{
1735 struct ceph_mds_session *session = NULL;
1736 int mds = -1;
1737 int err = -EAGAIN;
1738
e1518c7c 1739 if (req->r_err || req->r_got_result)
2f2dc053
SW		 1740 goto out;
1741
1742 if (req->r_timeout &&
1743 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1744 dout("do_request timed out\n");
1745 err = -EIO;
1746 goto finish;
1747 }
1748
1749 mds = __choose_mds(mdsc, req);
1750 if (mds < 0 ||
1751 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1752 dout("do_request no mds or not active, waiting for map\n");
1753 list_add(&req->r_wait, &mdsc->waiting_for_map);
1754 goto out;
1755 }
1756
1757 /* get, open session */
1758 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 1759 if (!session) {
2f2dc053 1760 session = register_session(mdsc, mds);
9c423956
SW		 1761 if (IS_ERR(session)) {
1762 err = PTR_ERR(session);
1763 goto finish;
1764 }
1765 }
2f2dc053
SW		 1766 dout("do_request mds%d session %p state %s\n", mds, session,
1767 session_state_name(session->s_state));
1768 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1769 session->s_state != CEPH_MDS_SESSION_HUNG) {
1770 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1771 session->s_state == CEPH_MDS_SESSION_CLOSING)
1772 __open_session(mdsc, session);
1773 list_add(&req->r_wait, &session->s_waiting);
1774 goto out_session;
1775 }
1776
1777 /* send request */
1778 req->r_session = get_session(session);
1779 req->r_resend_mds = -1; /* forget any previous mds hint */
1780
1781 if (req->r_request_started == 0) /* note request start time */
1782 req->r_request_started = jiffies;
1783
1784 err = __prepare_send_request(mdsc, req, mds);
1785 if (!err) {
1786 ceph_msg_get(req->r_request);
1787 ceph_con_send(&session->s_con, req->r_request);
1788 }
1789
1790out_session:
1791 ceph_put_mds_session(session);
1792out:
1793 return err;
1794
1795finish:
e1518c7c 1796 req->r_err = err;
2f2dc053
SW		 1797 complete_request(mdsc, req);
1798 goto out;
1799}
1800
1801/*
1802 * called under mdsc->mutex
1803 */
1804static void __wake_requests(struct ceph_mds_client *mdsc,
1805 struct list_head *head)
1806{
1807 struct ceph_mds_request *req, *nreq;
1808
1809 list_for_each_entry_safe(req, nreq, head, r_wait) {
1810 list_del_init(&req->r_wait);
1811 __do_request(mdsc, req);
1812 }
1813}
1814
1815/*
1816 * Wake up threads with requests pending for @mds, so that they can
29790f26 1817 * resubmit their requests to a possibly different mds.
2f2dc053 1818 */
29790f26 1819static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 1820{
44ca18f2
SW		 1821 struct ceph_mds_request *req;
1822 struct rb_node *p;
2f2dc053
SW		 1823
1824 dout("kick_requests mds%d\n", mds);
44ca18f2
SW		 1825 for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1826 req = rb_entry(p, struct ceph_mds_request, r_node);
1827 if (req->r_got_unsafe)
1828 continue;
1829 if (req->r_session &&
1830 req->r_session->s_mds == mds) {
1831 dout(" kicking tid %llu\n", req->r_tid);
1832 put_request_session(req);
1833 __do_request(mdsc, req);
2f2dc053
SW		 1834 }
1835 }
1836}
1837
1838void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1839 struct ceph_mds_request *req)
1840{
1841 dout("submit_request on %p\n", req);
1842 mutex_lock(&mdsc->mutex);
1843 __register_request(mdsc, req, NULL);
1844 __do_request(mdsc, req);
1845 mutex_unlock(&mdsc->mutex);
1846}
1847
1848/*
1849 * Synchrously perform an mds request. Take care of all of the
1850 * session setup, forwarding, retry details.
1851 */
1852int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
1853 struct inode *dir,
1854 struct ceph_mds_request *req)
1855{
1856 int err;
1857
1858 dout("do_request on %p\n", req);
1859
1860 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
1861 if (req->r_inode)
1862 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
1863 if (req->r_locked_dir)
1864 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
1865 if (req->r_old_dentry)
1866 ceph_get_cap_refs(
1867 ceph_inode(req->r_old_dentry->d_parent->d_inode),
1868 CEPH_CAP_PIN);
1869
1870 /* issue */
1871 mutex_lock(&mdsc->mutex);
1872 __register_request(mdsc, req, dir);
1873 __do_request(mdsc, req);
1874
e1518c7c
SW		 1875 if (req->r_err) {
1876 err = req->r_err;
1877 __unregister_request(mdsc, req);
1878 dout("do_request early error %d\n", err);
1879 goto out;
2f2dc053
SW		 1880 }
1881
e1518c7c
SW		 1882 /* wait */
1883 mutex_unlock(&mdsc->mutex);
1884 dout("do_request waiting\n");
1885 if (req->r_timeout) {
aa91647c 1886 err = (long)wait_for_completion_killable_timeout(
e1518c7c
SW		 1887 &req->r_completion, req->r_timeout);
1888 if (err == 0)
1889 err = -EIO;
1890 } else {
aa91647c 1891 err = wait_for_completion_killable(&req->r_completion);
e1518c7c
SW		 1892 }
1893 dout("do_request waited, got %d\n", err);
1894 mutex_lock(&mdsc->mutex);
5b1daecd 1895
e1518c7c
SW		 1896 /* only abort if we didn't race with a real reply */
1897 if (req->r_got_result) {
1898 err = le32_to_cpu(req->r_reply_info.head->result);
1899 } else if (err < 0) {
1900 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 1901
1902 /*
1903 * ensure we aren't running concurrently with
1904 * ceph_fill_trace or ceph_readdir_prepopulate, which
1905 * rely on locks (dir mutex) held by our caller.
1906 */
1907 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 1908 req->r_err = err;
1909 req->r_aborted = true;
b4556396 1910 mutex_unlock(&req->r_fill_mutex);
5b1daecd 1911
e1518c7c 1912 if (req->r_locked_dir &&
167c9e35
SW		 1913 (req->r_op & CEPH_MDS_OP_WRITE))
1914 ceph_invalidate_dir_request(req);
2f2dc053 1915 } else {
e1518c7c 1916 err = req->r_err;
2f2dc053 1917 }
2f2dc053 1918
e1518c7c
SW		 1919out:
1920 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 1921 dout("do_request %p done, result %d\n", req, err);
1922 return err;
1923}
1924
167c9e35
SW		 1925/*
1926 * Invalidate dir I_COMPLETE, dentry lease state on an aborted MDS
1927 * namespace request.
1928 */
1929void ceph_invalidate_dir_request(struct ceph_mds_request *req)
1930{
1931 struct inode *inode = req->r_locked_dir;
1932 struct ceph_inode_info *ci = ceph_inode(inode);
1933
1934 dout("invalidate_dir_request %p (I_COMPLETE, lease(s))\n", inode);
1935 spin_lock(&inode->i_lock);
1936 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1937 ci->i_release_count++;
1938 spin_unlock(&inode->i_lock);
1939
1940 if (req->r_dentry)
1941 ceph_invalidate_dentry_lease(req->r_dentry);
1942 if (req->r_old_dentry)
1943 ceph_invalidate_dentry_lease(req->r_old_dentry);
1944}
1945
2f2dc053
SW		 1946/*
1947 * Handle mds reply.
1948 *
1949 * We take the session mutex and parse and process the reply immediately.
1950 * This preserves the logical ordering of replies, capabilities, etc., sent
1951 * by the MDS as they are applied to our local cache.
1952 */
1953static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
1954{
1955 struct ceph_mds_client *mdsc = session->s_mdsc;
1956 struct ceph_mds_request *req;
1957 struct ceph_mds_reply_head *head = msg->front.iov_base;
1958 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
1959 u64 tid;
1960 int err, result;
2600d2dd 1961 int mds = session->s_mds;
2f2dc053 1962
2f2dc053
SW		 1963 if (msg->front.iov_len < sizeof(*head)) {
1964 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 1965 ceph_msg_dump(msg);
2f2dc053
SW		 1966 return;
1967 }
1968
1969 /* get request, session */
6df058c0 1970 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 1971 mutex_lock(&mdsc->mutex);
1972 req = __lookup_request(mdsc, tid);
1973 if (!req) {
1974 dout("handle_reply on unknown tid %llu\n", tid);
1975 mutex_unlock(&mdsc->mutex);
1976 return;
1977 }
1978 dout("handle_reply %p\n", req);
2f2dc053
SW		 1979
1980 /* correct session? */
d96d6049 1981 if (req->r_session != session) {
2f2dc053
SW		 1982 pr_err("mdsc_handle_reply got %llu on session mds%d"
1983 " not mds%d\n", tid, session->s_mds,
1984 req->r_session ? req->r_session->s_mds : -1);
1985 mutex_unlock(&mdsc->mutex);
1986 goto out;
1987 }
1988
1989 /* dup? */
1990 if ((req->r_got_unsafe && !head->safe) ||
1991 (req->r_got_safe && head->safe)) {
1992 pr_warning("got a dup %s reply on %llu from mds%d\n",
1993 head->safe ? "safe" : "unsafe", tid, mds);
1994 mutex_unlock(&mdsc->mutex);
1995 goto out;
1996 }
85792d0d
SW		 1997 if (req->r_got_safe && !head->safe) {
1998 pr_warning("got unsafe after safe on %llu from mds%d\n",
1999 tid, mds);
2000 mutex_unlock(&mdsc->mutex);
2001 goto out;
2002 }
2f2dc053
SW		 2003
2004 result = le32_to_cpu(head->result);
2005
2006 /*
e55b71f8
GF		 2007 * Handle an ESTALE
2008 * if we're not talking to the authority, send to them
2009 * if the authority has changed while we weren't looking,
2010 * send to new authority
2011 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2012 */
2013 if (result == -ESTALE) {
e55b71f8 2014 dout("got ESTALE on request %llu", req->r_tid);
213c99ee
SW		 2015 if (!req->r_inode) {
2016 /* do nothing; not an authority problem */
2017 } else if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2018 dout("not using auth, setting for that now");
2019 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2020 __do_request(mdsc, req);
2021 mutex_unlock(&mdsc->mutex);
2022 goto out;
e55b71f8
GF		 2023 } else {
2024 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
2025 struct ceph_cap *cap =
2026 ceph_get_cap_for_mds(ci, req->r_mds);;
2027
2028 dout("already using auth");
2029 if ((!cap || cap != ci->i_auth_cap) ||
2030 (cap->mseq != req->r_sent_on_mseq)) {
2031 dout("but cap changed, so resending");
2032 __do_request(mdsc, req);
2033 mutex_unlock(&mdsc->mutex);
2034 goto out;
2035 }
2f2dc053 2036 }
e55b71f8 2037 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2038 }
2039
e55b71f8 2040
2f2dc053
SW		 2041 if (head->safe) {
2042 req->r_got_safe = true;
2043 __unregister_request(mdsc, req);
03066f23 2044 complete_all(&req->r_safe_completion);
2f2dc053
SW		 2045
2046 if (req->r_got_unsafe) {
2047 /*
2048 * We already handled the unsafe response, now do the
2049 * cleanup. No need to examine the response; the MDS
2050 * doesn't include any result info in the safe
2051 * response. And even if it did, there is nothing
2052 * useful we could do with a revised return value.
2053 */
2054 dout("got safe reply %llu, mds%d\n", tid, mds);
2055 list_del_init(&req->r_unsafe_item);
2056
2057 /* last unsafe request during umount? */
44ca18f2 2058 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2059 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2060 mutex_unlock(&mdsc->mutex);
2061 goto out;
2062 }
e1518c7c 2063 } else {
2f2dc053
SW		 2064 req->r_got_unsafe = true;
2065 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2066 }
2067
2068 dout("handle_reply tid %lld result %d\n", tid, result);
2069 rinfo = &req->r_reply_info;
2070 err = parse_reply_info(msg, rinfo);
2071 mutex_unlock(&mdsc->mutex);
2072
2073 mutex_lock(&session->s_mutex);
2074 if (err < 0) {
2075 pr_err("mdsc_handle_reply got corrupt reply mds%d\n", mds);
9ec7cab1 2076 ceph_msg_dump(msg);
2f2dc053
SW		 2077 goto out_err;
2078 }
2079
2080 /* snap trace */
2081 if (rinfo->snapblob_len) {
2082 down_write(&mdsc->snap_rwsem);
2083 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2084 rinfo->snapblob + rinfo->snapblob_len,
2085 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2086 downgrade_write(&mdsc->snap_rwsem);
2087 } else {
2088 down_read(&mdsc->snap_rwsem);
2089 }
2090
2091 /* insert trace into our cache */
b4556396 2092 mutex_lock(&req->r_fill_mutex);
3d14c5d2 2093 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053
SW		 2094 if (err == 0) {
2095 if (result == 0 && rinfo->dir_nr)
2096 ceph_readdir_prepopulate(req, req->r_session);
37151668 2097 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2098 }
b4556396 2099 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2100
2101 up_read(&mdsc->snap_rwsem);
2102out_err:
e1518c7c
SW		 2103 mutex_lock(&mdsc->mutex);
2104 if (!req->r_aborted) {
2105 if (err) {
2106 req->r_err = err;
2107 } else {
2108 req->r_reply = msg;
2109 ceph_msg_get(msg);
2110 req->r_got_result = true;
2111 }
2f2dc053 2112 } else {
e1518c7c 2113 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2114 }
e1518c7c 2115 mutex_unlock(&mdsc->mutex);
2f2dc053 2116
ee6b272b 2117 ceph_add_cap_releases(mdsc, req->r_session);
2f2dc053
SW		 2118 mutex_unlock(&session->s_mutex);
2119
2120 /* kick calling process */
2121 complete_request(mdsc, req);
2122out:
2123 ceph_mdsc_put_request(req);
2124 return;
2125}
2126
2127
2128
2129/*
2130 * handle mds notification that our request has been forwarded.
2131 */
2600d2dd
SW		 2132static void handle_forward(struct ceph_mds_client *mdsc,
2133 struct ceph_mds_session *session,
2134 struct ceph_msg *msg)
2f2dc053
SW		 2135{
2136 struct ceph_mds_request *req;
a1ea787c 2137 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2138 u32 next_mds;
2139 u32 fwd_seq;
2f2dc053
SW		 2140 int err = -EINVAL;
2141 void *p = msg->front.iov_base;
2142 void *end = p + msg->front.iov_len;
2f2dc053 2143
a1ea787c 2144 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2145 next_mds = ceph_decode_32(&p);
2146 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2147
2148 mutex_lock(&mdsc->mutex);
2149 req = __lookup_request(mdsc, tid);
2150 if (!req) {
2a8e5e36 2151 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2152 goto out; /* dup reply? */
2153 }
2154
2a8e5e36
SW		 2155 if (req->r_aborted) {
2156 dout("forward tid %llu aborted, unregistering\n", tid);
2157 __unregister_request(mdsc, req);
2158 } else if (fwd_seq <= req->r_num_fwd) {
2159 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2160 tid, next_mds, req->r_num_fwd, fwd_seq);
2161 } else {
2162 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2163 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2164 BUG_ON(req->r_err);
2165 BUG_ON(req->r_got_result);
2f2dc053
SW		 2166 req->r_num_fwd = fwd_seq;
2167 req->r_resend_mds = next_mds;
2168 put_request_session(req);
2169 __do_request(mdsc, req);
2170 }
2171 ceph_mdsc_put_request(req);
2172out:
2173 mutex_unlock(&mdsc->mutex);
2174 return;
2175
2176bad:
2177 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2178}
2179
2180/*
2181 * handle a mds session control message
2182 */
2183static void handle_session(struct ceph_mds_session *session,
2184 struct ceph_msg *msg)
2185{
2186 struct ceph_mds_client *mdsc = session->s_mdsc;
2187 u32 op;
2188 u64 seq;
2600d2dd 2189 int mds = session->s_mds;
2f2dc053
SW		 2190 struct ceph_mds_session_head *h = msg->front.iov_base;
2191 int wake = 0;
2192
2f2dc053
SW		 2193 /* decode */
2194 if (msg->front.iov_len != sizeof(*h))
2195 goto bad;
2196 op = le32_to_cpu(h->op);
2197 seq = le64_to_cpu(h->seq);
2198
2199 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2200 if (op == CEPH_SESSION_CLOSE)
2201 __unregister_session(mdsc, session);
2f2dc053
SW		 2202 /* FIXME: this ttl calculation is generous */
2203 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2204 mutex_unlock(&mdsc->mutex);
2205
2206 mutex_lock(&session->s_mutex);
2207
2208 dout("handle_session mds%d %s %p state %s seq %llu\n",
2209 mds, ceph_session_op_name(op), session,
2210 session_state_name(session->s_state), seq);
2211
2212 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2213 session->s_state = CEPH_MDS_SESSION_OPEN;
2214 pr_info("mds%d came back\n", session->s_mds);
2215 }
2216
2217 switch (op) {
2218 case CEPH_SESSION_OPEN:
29790f26
SW		 2219 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2220 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2221 session->s_state = CEPH_MDS_SESSION_OPEN;
2222 renewed_caps(mdsc, session, 0);
2223 wake = 1;
2224 if (mdsc->stopping)
2225 __close_session(mdsc, session);
2226 break;
2227
2228 case CEPH_SESSION_RENEWCAPS:
2229 if (session->s_renew_seq == seq)
2230 renewed_caps(mdsc, session, 1);
2231 break;
2232
2233 case CEPH_SESSION_CLOSE:
29790f26
SW		 2234 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2235 pr_info("mds%d reconnect denied\n", session->s_mds);
2f2dc053
SW		 2236 remove_session_caps(session);
2237 wake = 1; /* for good measure */
f3c60c59 2238 wake_up_all(&mdsc->session_close_wq);
29790f26 2239 kick_requests(mdsc, mds);
2f2dc053
SW		 2240 break;
2241
2242 case CEPH_SESSION_STALE:
2243 pr_info("mds%d caps went stale, renewing\n",
2244 session->s_mds);
2245 spin_lock(&session->s_cap_lock);
2246 session->s_cap_gen++;
2247 session->s_cap_ttl = 0;
2248 spin_unlock(&session->s_cap_lock);
2249 send_renew_caps(mdsc, session);
2250 break;
2251
2252 case CEPH_SESSION_RECALL_STATE:
2253 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2254 break;
2255
2256 default:
2257 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2258 WARN_ON(1);
2259 }
2260
2261 mutex_unlock(&session->s_mutex);
2262 if (wake) {
2263 mutex_lock(&mdsc->mutex);
2264 __wake_requests(mdsc, &session->s_waiting);
2265 mutex_unlock(&mdsc->mutex);
2266 }
2267 return;
2268
2269bad:
2270 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2271 (int)msg->front.iov_len);
9ec7cab1 2272 ceph_msg_dump(msg);
2f2dc053
SW		 2273 return;
2274}
2275
2276
2277/*
2278 * called under session->mutex.
2279 */
2280static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2281 struct ceph_mds_session *session)
2282{
2283 struct ceph_mds_request *req, *nreq;
2284 int err;
2285
2286 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2287
2288 mutex_lock(&mdsc->mutex);
2289 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2290 err = __prepare_send_request(mdsc, req, session->s_mds);
2291 if (!err) {
2292 ceph_msg_get(req->r_request);
2293 ceph_con_send(&session->s_con, req->r_request);
2294 }
2295 }
2296 mutex_unlock(&mdsc->mutex);
2297}
2298
2299/*
2300 * Encode information about a cap for a reconnect with the MDS.
2301 */
2f2dc053
SW		 2302static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2303 void *arg)
2304{
20cb34ae
SW		 2305 union {
2306 struct ceph_mds_cap_reconnect v2;
2307 struct ceph_mds_cap_reconnect_v1 v1;
2308 } rec;
2309 size_t reclen;
2f2dc053 2310 struct ceph_inode_info *ci;
20cb34ae
SW		 2311 struct ceph_reconnect_state *recon_state = arg;
2312 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2313 char *path;
2314 int pathlen, err;
2315 u64 pathbase;
2316 struct dentry *dentry;
2317
2318 ci = cap->ci;
2319
2320 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2321 inode, ceph_vinop(inode), cap, cap->cap_id,
2322 ceph_cap_string(cap->issued));
93cea5be
SW		 2323 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2324 if (err)
2325 return err;
2f2dc053
SW		 2326
2327 dentry = d_find_alias(inode);
2328 if (dentry) {
2329 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2330 if (IS_ERR(path)) {
2331 err = PTR_ERR(path);
e072f8aa 2332 goto out_dput;
2f2dc053
SW		 2333 }
2334 } else {
2335 path = NULL;
2336 pathlen = 0;
2337 }
93cea5be
SW		 2338 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2339 if (err)
e072f8aa 2340 goto out_free;
2f2dc053 2341
2f2dc053
SW		 2342 spin_lock(&inode->i_lock);
2343 cap->seq = 0; /* reset cap seq */
2344 cap->issue_seq = 0; /* and issue_seq */
20cb34ae
SW		 2345
2346 if (recon_state->flock) {
2347 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2348 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2349 rec.v2.issued = cpu_to_le32(cap->issued);
2350 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2351 rec.v2.pathbase = cpu_to_le64(pathbase);
2352 rec.v2.flock_len = 0;
2353 reclen = sizeof(rec.v2);
2354 } else {
2355 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2356 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2357 rec.v1.issued = cpu_to_le32(cap->issued);
2358 rec.v1.size = cpu_to_le64(inode->i_size);
2359 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2360 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2361 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2362 rec.v1.pathbase = cpu_to_le64(pathbase);
2363 reclen = sizeof(rec.v1);
2364 }
2f2dc053
SW		 2365 spin_unlock(&inode->i_lock);
2366
40819f6f
GF		 2367 if (recon_state->flock) {
2368 int num_fcntl_locks, num_flock_locks;
fca4451a
GF		 2369 struct ceph_pagelist_cursor trunc_point;
2370
2371 ceph_pagelist_set_cursor(pagelist, &trunc_point);
2372 do {
496e5955 2373 lock_flocks();
fca4451a
GF		 2374 ceph_count_locks(inode, &num_fcntl_locks,
2375 &num_flock_locks);
2376 rec.v2.flock_len = (2*sizeof(u32) +
2377 (num_fcntl_locks+num_flock_locks) *
2378 sizeof(struct ceph_filelock));
496e5955 2379 unlock_flocks();
fca4451a
GF		 2380
2381 /* pre-alloc pagelist */
2382 ceph_pagelist_truncate(pagelist, &trunc_point);
2383 err = ceph_pagelist_append(pagelist, &rec, reclen);
2384 if (!err)
2385 err = ceph_pagelist_reserve(pagelist,
2386 rec.v2.flock_len);
2387
2388 /* encode locks */
2389 if (!err) {
496e5955 2390 lock_flocks();
fca4451a
GF		 2391 err = ceph_encode_locks(inode,
2392 pagelist,
2393 num_fcntl_locks,
2394 num_flock_locks);
496e5955 2395 unlock_flocks();
fca4451a
GF		 2396 }
2397 } while (err == -ENOSPC);
3612abbd
SW		 2398 } else {
2399 err = ceph_pagelist_append(pagelist, &rec, reclen);
40819f6f 2400 }
93cea5be 2401
e072f8aa 2402out_free:
2f2dc053 2403 kfree(path);
e072f8aa 2404out_dput:
2f2dc053 2405 dput(dentry);
93cea5be 2406 return err;
2f2dc053
SW		 2407}
2408
2409
2410/*
2411 * If an MDS fails and recovers, clients need to reconnect in order to
2412 * reestablish shared state. This includes all caps issued through
2413 * this session _and_ the snap_realm hierarchy. Because it's not
2414 * clear which snap realms the mds cares about, we send everything we
2415 * know about.. that ensures we'll then get any new info the
2416 * recovering MDS might have.
2417 *
2418 * This is a relatively heavyweight operation, but it's rare.
2419 *
2420 * called with mdsc->mutex held.
2421 */
34b6c855
SW		 2422static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2423 struct ceph_mds_session *session)
2f2dc053 2424{
2f2dc053 2425 struct ceph_msg *reply;
a105f00c 2426 struct rb_node *p;
34b6c855 2427 int mds = session->s_mds;
9abf82b8 2428 int err = -ENOMEM;
93cea5be 2429 struct ceph_pagelist *pagelist;
20cb34ae 2430 struct ceph_reconnect_state recon_state;
2f2dc053 2431
34b6c855 2432 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2433
93cea5be
SW		 2434 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2435 if (!pagelist)
2436 goto fail_nopagelist;
2437 ceph_pagelist_init(pagelist);
2438
34d23762 2439 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS);
a79832f2 2440 if (!reply)
93cea5be 2441 goto fail_nomsg;
93cea5be 2442
34b6c855
SW		 2443 mutex_lock(&session->s_mutex);
2444 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2445 session->s_seq = 0;
2f2dc053 2446
34b6c855
SW		 2447 ceph_con_open(&session->s_con,
2448 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2f2dc053 2449
34b6c855
SW		 2450 /* replay unsafe requests */
2451 replay_unsafe_requests(mdsc, session);
2f2dc053
SW		 2452
2453 down_read(&mdsc->snap_rwsem);
2454
2f2dc053
SW		 2455 dout("session %p state %s\n", session,
2456 session_state_name(session->s_state));
2457
e01a5946
SW		 2458 /* drop old cap expires; we're about to reestablish that state */
2459 discard_cap_releases(mdsc, session);
2460
2f2dc053 2461 /* traverse this session's caps */
93cea5be
SW		 2462 err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2463 if (err)
2464 goto fail;
20cb34ae
SW		 2465
2466 recon_state.pagelist = pagelist;
2467 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2468 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2469 if (err < 0)
9abf82b8 2470 goto fail;
2f2dc053
SW		 2471
2472 /*
2473 * snaprealms. we provide mds with the ino, seq (version), and
2474 * parent for all of our realms. If the mds has any newer info,
2475 * it will tell us.
2476 */
a105f00c
SW		 2477 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2478 struct ceph_snap_realm *realm =
2479 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 2480 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 2481
2482 dout(" adding snap realm %llx seq %lld parent %llx\n",
2483 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 2484 sr_rec.ino = cpu_to_le64(realm->ino);
2485 sr_rec.seq = cpu_to_le64(realm->seq);
2486 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2487 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2488 if (err)
2489 goto fail;
2f2dc053 2490 }
2f2dc053 2491
93cea5be 2492 reply->pagelist = pagelist;
20cb34ae
SW		 2493 if (recon_state.flock)
2494 reply->hdr.version = cpu_to_le16(2);
93cea5be
SW		 2495 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2496 reply->nr_pages = calc_pages_for(0, pagelist->length);
2f2dc053
SW		 2497 ceph_con_send(&session->s_con, reply);
2498
9abf82b8
SW		 2499 mutex_unlock(&session->s_mutex);
2500
2501 mutex_lock(&mdsc->mutex);
2502 __wake_requests(mdsc, &session->s_waiting);
2503 mutex_unlock(&mdsc->mutex);
2504
2f2dc053 2505 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 2506 return;
2507
93cea5be 2508fail:
2f2dc053 2509 ceph_msg_put(reply);
9abf82b8
SW		 2510 up_read(&mdsc->snap_rwsem);
2511 mutex_unlock(&session->s_mutex);
93cea5be
SW		 2512fail_nomsg:
2513 ceph_pagelist_release(pagelist);
2514 kfree(pagelist);
2515fail_nopagelist:
9abf82b8 2516 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 2517 return;
2f2dc053
SW		 2518}
2519
2520
2521/*
2522 * compare old and new mdsmaps, kicking requests
2523 * and closing out old connections as necessary
2524 *
2525 * called under mdsc->mutex.
2526 */
2527static void check_new_map(struct ceph_mds_client *mdsc,
2528 struct ceph_mdsmap *newmap,
2529 struct ceph_mdsmap *oldmap)
2530{
2531 int i;
2532 int oldstate, newstate;
2533 struct ceph_mds_session *s;
2534
2535 dout("check_new_map new %u old %u\n",
2536 newmap->m_epoch, oldmap->m_epoch);
2537
2538 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2539 if (mdsc->sessions[i] == NULL)
2540 continue;
2541 s = mdsc->sessions[i];
2542 oldstate = ceph_mdsmap_get_state(oldmap, i);
2543 newstate = ceph_mdsmap_get_state(newmap, i);
2544
0deb01c9 2545 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 2546 i, ceph_mds_state_name(oldstate),
0deb01c9 2547 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 2548 ceph_mds_state_name(newstate),
0deb01c9 2549 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
2f2dc053
SW		 2550 session_state_name(s->s_state));
2551
2552 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
2553 ceph_mdsmap_get_addr(newmap, i),
2554 sizeof(struct ceph_entity_addr))) {
2555 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2556 /* the session never opened, just close it
2557 * out now */
2558 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 2559 __unregister_session(mdsc, s);
2f2dc053
SW		 2560 } else {
2561 /* just close it */
2562 mutex_unlock(&mdsc->mutex);
2563 mutex_lock(&s->s_mutex);
2564 mutex_lock(&mdsc->mutex);
2565 ceph_con_close(&s->s_con);
2566 mutex_unlock(&s->s_mutex);
2567 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2568 }
2569
2570 /* kick any requests waiting on the recovering mds */
29790f26 2571 kick_requests(mdsc, i);
2f2dc053
SW		 2572 } else if (oldstate == newstate) {
2573 continue; /* nothing new with this mds */
2574 }
2575
2576 /*
2577 * send reconnect?
2578 */
2579 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 2580 newstate >= CEPH_MDS_STATE_RECONNECT) {
2581 mutex_unlock(&mdsc->mutex);
2582 send_mds_reconnect(mdsc, s);
2583 mutex_lock(&mdsc->mutex);
2584 }
2f2dc053
SW		 2585
2586 /*
29790f26 2587 * kick request on any mds that has gone active.
2f2dc053
SW		 2588 */
2589 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2590 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 2591 if (oldstate != CEPH_MDS_STATE_CREATING &&
2592 oldstate != CEPH_MDS_STATE_STARTING)
2593 pr_info("mds%d recovery completed\n", s->s_mds);
2594 kick_requests(mdsc, i);
2f2dc053 2595 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 2596 wake_up_session_caps(s, 1);
2f2dc053
SW		 2597 }
2598 }
cb170a22
SW		 2599
2600 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2601 s = mdsc->sessions[i];
2602 if (!s)
2603 continue;
2604 if (!ceph_mdsmap_is_laggy(newmap, i))
2605 continue;
2606 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2607 s->s_state == CEPH_MDS_SESSION_HUNG ||
2608 s->s_state == CEPH_MDS_SESSION_CLOSING) {
2609 dout(" connecting to export targets of laggy mds%d\n",
2610 i);
2611 __open_export_target_sessions(mdsc, s);
2612 }
2613 }
2f2dc053
SW		 2614}
2615
2616
2617
2618/*
2619 * leases
2620 */
2621
2622/*
2623 * caller must hold session s_mutex, dentry->d_lock
2624 */
2625void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2626{
2627 struct ceph_dentry_info *di = ceph_dentry(dentry);
2628
2629 ceph_put_mds_session(di->lease_session);
2630 di->lease_session = NULL;
2631}
2632
2600d2dd
SW		 2633static void handle_lease(struct ceph_mds_client *mdsc,
2634 struct ceph_mds_session *session,
2635 struct ceph_msg *msg)
2f2dc053 2636{
3d14c5d2 2637 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 2638 struct inode *inode;
2f2dc053
SW		 2639 struct ceph_inode_info *ci;
2640 struct dentry *parent, *dentry;
2641 struct ceph_dentry_info *di;
2600d2dd 2642 int mds = session->s_mds;
2f2dc053 2643 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 2644 u32 seq;
2f2dc053
SW		 2645 struct ceph_vino vino;
2646 int mask;
2647 struct qstr dname;
2648 int release = 0;
2649
2f2dc053
SW		 2650 dout("handle_lease from mds%d\n", mds);
2651
2652 /* decode */
2653 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2654 goto bad;
2655 vino.ino = le64_to_cpu(h->ino);
2656 vino.snap = CEPH_NOSNAP;
2657 mask = le16_to_cpu(h->mask);
1e5ea23d 2658 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 2659 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2660 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2661 if (dname.len != get_unaligned_le32(h+1))
2662 goto bad;
2663
2f2dc053
SW		 2664 mutex_lock(&session->s_mutex);
2665 session->s_seq++;
2666
2667 /* lookup inode */
2668 inode = ceph_find_inode(sb, vino);
1e5ea23d
SW		 2669 dout("handle_lease %s, mask %d, ino %llx %p %.*s\n",
2670 ceph_lease_op_name(h->action), mask, vino.ino, inode,
2671 dname.len, dname.name);
2f2dc053
SW		 2672 if (inode == NULL) {
2673 dout("handle_lease no inode %llx\n", vino.ino);
2674 goto release;
2675 }
2676 ci = ceph_inode(inode);
2677
2678 /* dentry */
2679 parent = d_find_alias(inode);
2680 if (!parent) {
2681 dout("no parent dentry on inode %p\n", inode);
2682 WARN_ON(1);
2683 goto release; /* hrm... */
2684 }
2685 dname.hash = full_name_hash(dname.name, dname.len);
2686 dentry = d_lookup(parent, &dname);
2687 dput(parent);
2688 if (!dentry)
2689 goto release;
2690
2691 spin_lock(&dentry->d_lock);
2692 di = ceph_dentry(dentry);
2693 switch (h->action) {
2694 case CEPH_MDS_LEASE_REVOKE:
2695 if (di && di->lease_session == session) {
1e5ea23d
SW		 2696 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2697 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 2698 __ceph_mdsc_drop_dentry_lease(dentry);
2699 }
2700 release = 1;
2701 break;
2702
2703 case CEPH_MDS_LEASE_RENEW:
2704 if (di && di->lease_session == session &&
2705 di->lease_gen == session->s_cap_gen &&
2706 di->lease_renew_from &&
2707 di->lease_renew_after == 0) {
2708 unsigned long duration =
2709 le32_to_cpu(h->duration_ms) * HZ / 1000;
2710
1e5ea23d 2711 di->lease_seq = seq;
2f2dc053
SW		 2712 dentry->d_time = di->lease_renew_from + duration;
2713 di->lease_renew_after = di->lease_renew_from +
2714 (duration >> 1);
2715 di->lease_renew_from = 0;
2716 }
2717 break;
2718 }
2719 spin_unlock(&dentry->d_lock);
2720 dput(dentry);
2721
2722 if (!release)
2723 goto out;
2724
2725release:
2726 /* let's just reuse the same message */
2727 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2728 ceph_msg_get(msg);
2729 ceph_con_send(&session->s_con, msg);
2730
2731out:
2732 iput(inode);
2733 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 2734 return;
2735
2736bad:
2737 pr_err("corrupt lease message\n");
9ec7cab1 2738 ceph_msg_dump(msg);
2f2dc053
SW		 2739}
2740
2741void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2742 struct inode *inode,
2743 struct dentry *dentry, char action,
2744 u32 seq)
2745{
2746 struct ceph_msg *msg;
2747 struct ceph_mds_lease *lease;
2748 int len = sizeof(*lease) + sizeof(u32);
2749 int dnamelen = 0;
2750
2751 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2752 inode, dentry, ceph_lease_op_name(action), session->s_mds);
2753 dnamelen = dentry->d_name.len;
2754 len += dnamelen;
2755
34d23762 2756 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS);
a79832f2 2757 if (!msg)
2f2dc053
SW		 2758 return;
2759 lease = msg->front.iov_base;
2760 lease->action = action;
dd1c9057 2761 lease->mask = cpu_to_le16(1);
2f2dc053
SW		 2762 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2763 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2764 lease->seq = cpu_to_le32(seq);
2765 put_unaligned_le32(dnamelen, lease + 1);
2766 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2767
2768 /*
2769 * if this is a preemptive lease RELEASE, no need to
2770 * flush request stream, since the actual request will
2771 * soon follow.
2772 */
2773 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2774
2775 ceph_con_send(&session->s_con, msg);
2776}
2777
2778/*
2779 * Preemptively release a lease we expect to invalidate anyway.
2780 * Pass @inode always, @dentry is optional.
2781 */
2782void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2783 struct dentry *dentry, int mask)
2784{
2785 struct ceph_dentry_info *di;
2786 struct ceph_mds_session *session;
2787 u32 seq;
2788
2789 BUG_ON(inode == NULL);
2790 BUG_ON(dentry == NULL);
dd1c9057 2791 BUG_ON(mask == 0);
2f2dc053
SW		 2792
2793 /* is dentry lease valid? */
2794 spin_lock(&dentry->d_lock);
2795 di = ceph_dentry(dentry);
2796 if (!di || !di->lease_session ||
2797 di->lease_session->s_mds < 0 ||
2798 di->lease_gen != di->lease_session->s_cap_gen ||
2799 !time_before(jiffies, dentry->d_time)) {
2800 dout("lease_release inode %p dentry %p -- "
2801 "no lease on %d\n",
2802 inode, dentry, mask);
2803 spin_unlock(&dentry->d_lock);
2804 return;
2805 }
2806
2807 /* we do have a lease on this dentry; note mds and seq */
2808 session = ceph_get_mds_session(di->lease_session);
2809 seq = di->lease_seq;
2810 __ceph_mdsc_drop_dentry_lease(dentry);
2811 spin_unlock(&dentry->d_lock);
2812
2813 dout("lease_release inode %p dentry %p mask %d to mds%d\n",
2814 inode, dentry, mask, session->s_mds);
2815 ceph_mdsc_lease_send_msg(session, inode, dentry,
2816 CEPH_MDS_LEASE_RELEASE, seq);
2817 ceph_put_mds_session(session);
2818}
2819
2820/*
2821 * drop all leases (and dentry refs) in preparation for umount
2822 */
2823static void drop_leases(struct ceph_mds_client *mdsc)
2824{
2825 int i;
2826
2827 dout("drop_leases\n");
2828 mutex_lock(&mdsc->mutex);
2829 for (i = 0; i < mdsc->max_sessions; i++) {
2830 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2831 if (!s)
2832 continue;
2833 mutex_unlock(&mdsc->mutex);
2834 mutex_lock(&s->s_mutex);
2835 mutex_unlock(&s->s_mutex);
2836 ceph_put_mds_session(s);
2837 mutex_lock(&mdsc->mutex);
2838 }
2839 mutex_unlock(&mdsc->mutex);
2840}
2841
2842
2843
2844/*
2845 * delayed work -- periodically trim expired leases, renew caps with mds
2846 */
2847static void schedule_delayed(struct ceph_mds_client *mdsc)
2848{
2849 int delay = 5;
2850 unsigned hz = round_jiffies_relative(HZ * delay);
2851 schedule_delayed_work(&mdsc->delayed_work, hz);
2852}
2853
2854static void delayed_work(struct work_struct *work)
2855{
2856 int i;
2857 struct ceph_mds_client *mdsc =
2858 container_of(work, struct ceph_mds_client, delayed_work.work);
2859 int renew_interval;
2860 int renew_caps;
2861
2862 dout("mdsc delayed_work\n");
afcdaea3 2863 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 2864
2865 mutex_lock(&mdsc->mutex);
2866 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
2867 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
2868 mdsc->last_renew_caps);
2869 if (renew_caps)
2870 mdsc->last_renew_caps = jiffies;
2871
2872 for (i = 0; i < mdsc->max_sessions; i++) {
2873 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2874 if (s == NULL)
2875 continue;
2876 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
2877 dout("resending session close request for mds%d\n",
2878 s->s_mds);
2879 request_close_session(mdsc, s);
2880 ceph_put_mds_session(s);
2881 continue;
2882 }
2883 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
2884 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
2885 s->s_state = CEPH_MDS_SESSION_HUNG;
2886 pr_info("mds%d hung\n", s->s_mds);
2887 }
2888 }
2889 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
2890 /* this mds is failed or recovering, just wait */
2891 ceph_put_mds_session(s);
2892 continue;
2893 }
2894 mutex_unlock(&mdsc->mutex);
2895
2896 mutex_lock(&s->s_mutex);
2897 if (renew_caps)
2898 send_renew_caps(mdsc, s);
2899 else
2900 ceph_con_keepalive(&s->s_con);
ee6b272b 2901 ceph_add_cap_releases(mdsc, s);
aab53dd9
SW		 2902 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2903 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 2904 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 2905 mutex_unlock(&s->s_mutex);
2906 ceph_put_mds_session(s);
2907
2908 mutex_lock(&mdsc->mutex);
2909 }
2910 mutex_unlock(&mdsc->mutex);
2911
2912 schedule_delayed(mdsc);
2913}
2914
3d14c5d2 2915int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 2916
2f2dc053 2917{
3d14c5d2
YS		 2918 struct ceph_mds_client *mdsc;
2919
2920 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
2921 if (!mdsc)
2922 return -ENOMEM;
2923 mdsc->fsc = fsc;
2924 fsc->mdsc = mdsc;
2f2dc053
SW		 2925 mutex_init(&mdsc->mutex);
2926 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
2d06eeb8
CR		 2927 if (mdsc->mdsmap == NULL)
2928 return -ENOMEM;
2929
2f2dc053 2930 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 2931 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 2932 INIT_LIST_HEAD(&mdsc->waiting_for_map);
2933 mdsc->sessions = NULL;
2934 mdsc->max_sessions = 0;
2935 mdsc->stopping = 0;
2936 init_rwsem(&mdsc->snap_rwsem);
a105f00c 2937 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 2938 INIT_LIST_HEAD(&mdsc->snap_empty);
2939 spin_lock_init(&mdsc->snap_empty_lock);
2940 mdsc->last_tid = 0;
44ca18f2 2941 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 2942 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
2943 mdsc->last_renew_caps = jiffies;
2944 INIT_LIST_HEAD(&mdsc->cap_delay_list);
2945 spin_lock_init(&mdsc->cap_delay_lock);
2946 INIT_LIST_HEAD(&mdsc->snap_flush_list);
2947 spin_lock_init(&mdsc->snap_flush_lock);
2948 mdsc->cap_flush_seq = 0;
2949 INIT_LIST_HEAD(&mdsc->cap_dirty);
2950 mdsc->num_cap_flushing = 0;
2951 spin_lock_init(&mdsc->cap_dirty_lock);
2952 init_waitqueue_head(&mdsc->cap_flushing_wq);
2953 spin_lock_init(&mdsc->dentry_lru_lock);
2954 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 2955
37151668 2956 ceph_caps_init(mdsc);
3d14c5d2 2957 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 2958
5f44f142 2959 return 0;
2f2dc053
SW		 2960}
2961
2962/*
2963 * Wait for safe replies on open mds requests. If we time out, drop
2964 * all requests from the tree to avoid dangling dentry refs.
2965 */
2966static void wait_requests(struct ceph_mds_client *mdsc)
2967{
2968 struct ceph_mds_request *req;
3d14c5d2 2969 struct ceph_fs_client *fsc = mdsc->fsc;
2f2dc053
SW		 2970
2971 mutex_lock(&mdsc->mutex);
44ca18f2 2972 if (__get_oldest_req(mdsc)) {
2f2dc053 2973 mutex_unlock(&mdsc->mutex);
44ca18f2 2974
2f2dc053
SW		 2975 dout("wait_requests waiting for requests\n");
2976 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3d14c5d2 2977 fsc->client->options->mount_timeout * HZ);
2f2dc053
SW		 2978
2979 /* tear down remaining requests */
44ca18f2
SW		 2980 mutex_lock(&mdsc->mutex);
2981 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 2982 dout("wait_requests timed out on tid %llu\n",
2983 req->r_tid);
44ca18f2 2984 __unregister_request(mdsc, req);
2f2dc053
SW		 2985 }
2986 }
2987 mutex_unlock(&mdsc->mutex);
2988 dout("wait_requests done\n");
2989}
2990
2991/*
2992 * called before mount is ro, and before dentries are torn down.
2993 * (hmm, does this still race with new lookups?)
2994 */
2995void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
2996{
2997 dout("pre_umount\n");
2998 mdsc->stopping = 1;
2999
3000 drop_leases(mdsc);
afcdaea3 3001 ceph_flush_dirty_caps(mdsc);
2f2dc053 3002 wait_requests(mdsc);
17c688c3
SW		 3003
3004 /*
3005 * wait for reply handlers to drop their request refs and
3006 * their inode/dcache refs
3007 */
3008 ceph_msgr_flush();
2f2dc053
SW		 3009}
3010
3011/*
3012 * wait for all write mds requests to flush.
3013 */
3014static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3015{
80fc7314 3016 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3017 struct rb_node *n;
2f2dc053
SW		 3018
3019 mutex_lock(&mdsc->mutex);
3020 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3021restart:
44ca18f2
SW		 3022 req = __get_oldest_req(mdsc);
3023 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3024 /* find next request */
3025 n = rb_next(&req->r_node);
3026 if (n)
3027 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3028 else
3029 nextreq = NULL;
44ca18f2
SW		 3030 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3031 /* write op */
3032 ceph_mdsc_get_request(req);
80fc7314
SW		 3033 if (nextreq)
3034 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3035 mutex_unlock(&mdsc->mutex);
3036 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3037 req->r_tid, want_tid);
3038 wait_for_completion(&req->r_safe_completion);
3039 mutex_lock(&mdsc->mutex);
44ca18f2 3040 ceph_mdsc_put_request(req);
80fc7314
SW		 3041 if (!nextreq)
3042 break; /* next dne before, so we're done! */
3043 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3044 /* next request was removed from tree */
3045 ceph_mdsc_put_request(nextreq);
3046 goto restart;
3047 }
3048 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3049 }
80fc7314 3050 req = nextreq;
2f2dc053
SW		 3051 }
3052 mutex_unlock(&mdsc->mutex);
3053 dout("wait_unsafe_requests done\n");
3054}
3055
3056void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3057{
3058 u64 want_tid, want_flush;
3059
3d14c5d2 3060 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3061 return;
3062
2f2dc053
SW		 3063 dout("sync\n");
3064 mutex_lock(&mdsc->mutex);
3065 want_tid = mdsc->last_tid;
3066 want_flush = mdsc->cap_flush_seq;
3067 mutex_unlock(&mdsc->mutex);
3068 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3069
afcdaea3 3070 ceph_flush_dirty_caps(mdsc);
2f2dc053
SW		 3071
3072 wait_unsafe_requests(mdsc, want_tid);
3073 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3074}
3075
f3c60c59
SW		 3076/*
3077 * true if all sessions are closed, or we force unmount
3078 */
3079bool done_closing_sessions(struct ceph_mds_client *mdsc)
3080{
3081 int i, n = 0;
3082
3d14c5d2 3083 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
f3c60c59
SW		 3084 return true;
3085
3086 mutex_lock(&mdsc->mutex);
3087 for (i = 0; i < mdsc->max_sessions; i++)
3088 if (mdsc->sessions[i])
3089 n++;
3090 mutex_unlock(&mdsc->mutex);
3091 return n == 0;
3092}
2f2dc053
SW		 3093
3094/*
3095 * called after sb is ro.
3096 */
3097void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3098{
3099 struct ceph_mds_session *session;
3100 int i;
3d14c5d2
YS		 3101 struct ceph_fs_client *fsc = mdsc->fsc;
3102 unsigned long timeout = fsc->client->options->mount_timeout * HZ;
2f2dc053
SW		 3103
3104 dout("close_sessions\n");
3105
2f2dc053 3106 /* close sessions */
f3c60c59
SW		 3107 mutex_lock(&mdsc->mutex);
3108 for (i = 0; i < mdsc->max_sessions; i++) {
3109 session = __ceph_lookup_mds_session(mdsc, i);
3110 if (!session)
3111 continue;
2f2dc053 3112 mutex_unlock(&mdsc->mutex);
f3c60c59
SW		 3113 mutex_lock(&session->s_mutex);
3114 __close_session(mdsc, session);
3115 mutex_unlock(&session->s_mutex);
3116 ceph_put_mds_session(session);
2f2dc053
SW		 3117 mutex_lock(&mdsc->mutex);
3118 }
f3c60c59
SW		 3119 mutex_unlock(&mdsc->mutex);
3120
3121 dout("waiting for sessions to close\n");
3122 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3123 timeout);
2f2dc053
SW		 3124
3125 /* tear down remaining sessions */
f3c60c59 3126 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3127 for (i = 0; i < mdsc->max_sessions; i++) {
3128 if (mdsc->sessions[i]) {
3129 session = get_session(mdsc->sessions[i]);
2600d2dd 3130 __unregister_session(mdsc, session);
2f2dc053
SW		 3131 mutex_unlock(&mdsc->mutex);
3132 mutex_lock(&session->s_mutex);
3133 remove_session_caps(session);
3134 mutex_unlock(&session->s_mutex);
3135 ceph_put_mds_session(session);
3136 mutex_lock(&mdsc->mutex);
3137 }
3138 }
2f2dc053 3139 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3140 mutex_unlock(&mdsc->mutex);
3141
3142 ceph_cleanup_empty_realms(mdsc);
3143
3144 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3145
3146 dout("stopped\n");
3147}
3148
3d14c5d2 3149static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3150{
3151 dout("stop\n");
3152 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3153 if (mdsc->mdsmap)
3154 ceph_mdsmap_destroy(mdsc->mdsmap);
3155 kfree(mdsc->sessions);
37151668 3156 ceph_caps_finalize(mdsc);
2f2dc053
SW		 3157}
3158
3d14c5d2
YS		 3159void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3160{
3161 struct ceph_mds_client *mdsc = fsc->mdsc;
3162
3163 ceph_mdsc_stop(mdsc);
3164 fsc->mdsc = NULL;
3165 kfree(mdsc);
3166}
3167
2f2dc053
SW		 3168
3169/*
3170 * handle mds map update.
3171 */
3172void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3173{
3174 u32 epoch;
3175 u32 maplen;
3176 void *p = msg->front.iov_base;
3177 void *end = p + msg->front.iov_len;
3178 struct ceph_mdsmap *newmap, *oldmap;
3179 struct ceph_fsid fsid;
3180 int err = -EINVAL;
3181
3182 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3183 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3184 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3185 return;
c89136ea
SW		 3186 epoch = ceph_decode_32(&p);
3187 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3188 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3189
3190 /* do we need it? */
3d14c5d2 3191 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
2f2dc053
SW		 3192 mutex_lock(&mdsc->mutex);
3193 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3194 dout("handle_map epoch %u <= our %u\n",
3195 epoch, mdsc->mdsmap->m_epoch);
3196 mutex_unlock(&mdsc->mutex);
3197 return;
3198 }
3199
3200 newmap = ceph_mdsmap_decode(&p, end);
3201 if (IS_ERR(newmap)) {
3202 err = PTR_ERR(newmap);
3203 goto bad_unlock;
3204 }
3205
3206 /* swap into place */
3207 if (mdsc->mdsmap) {
3208 oldmap = mdsc->mdsmap;
3209 mdsc->mdsmap = newmap;
3210 check_new_map(mdsc, newmap, oldmap);
3211 ceph_mdsmap_destroy(oldmap);
3212 } else {
3213 mdsc->mdsmap = newmap; /* first mds map */
3214 }
3d14c5d2 3215 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3216
3217 __wake_requests(mdsc, &mdsc->waiting_for_map);
3218
3219 mutex_unlock(&mdsc->mutex);
3220 schedule_delayed(mdsc);
3221 return;
3222
3223bad_unlock:
3224 mutex_unlock(&mdsc->mutex);
3225bad:
3226 pr_err("error decoding mdsmap %d\n", err);
3227 return;
3228}
3229
3230static struct ceph_connection *con_get(struct ceph_connection *con)
3231{
3232 struct ceph_mds_session *s = con->private;
3233
3234 if (get_session(s)) {
2600d2dd 3235 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3236 return con;
3237 }
3238 dout("mdsc con_get %p FAIL\n", s);
3239 return NULL;
3240}
3241
3242static void con_put(struct ceph_connection *con)
3243{
3244 struct ceph_mds_session *s = con->private;
3245
2f2dc053 3246 ceph_put_mds_session(s);
2600d2dd 3247 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3248}
3249
3250/*
3251 * if the client is unresponsive for long enough, the mds will kill
3252 * the session entirely.
3253 */
3254static void peer_reset(struct ceph_connection *con)
3255{
3256 struct ceph_mds_session *s = con->private;
7e70f0ed 3257 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3258
7e70f0ed
SW		 3259 pr_warning("mds%d closed our session\n", s->s_mds);
3260 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3261}
3262
3263static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3264{
3265 struct ceph_mds_session *s = con->private;
3266 struct ceph_mds_client *mdsc = s->s_mdsc;
3267 int type = le16_to_cpu(msg->hdr.type);
3268
2600d2dd
SW		 3269 mutex_lock(&mdsc->mutex);
3270 if (__verify_registered_session(mdsc, s) < 0) {
3271 mutex_unlock(&mdsc->mutex);
3272 goto out;
3273 }
3274 mutex_unlock(&mdsc->mutex);
3275
2f2dc053
SW		 3276 switch (type) {
3277 case CEPH_MSG_MDS_MAP:
3278 ceph_mdsc_handle_map(mdsc, msg);
3279 break;
3280 case CEPH_MSG_CLIENT_SESSION:
3281 handle_session(s, msg);
3282 break;
3283 case CEPH_MSG_CLIENT_REPLY:
3284 handle_reply(s, msg);
3285 break;
3286 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3287 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3288 break;
3289 case CEPH_MSG_CLIENT_CAPS:
3290 ceph_handle_caps(s, msg);
3291 break;
3292 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3293 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3294 break;
3295 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3296 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3297 break;
3298
3299 default:
3300 pr_err("received unknown message type %d %s\n", type,
3301 ceph_msg_type_name(type));
3302 }
2600d2dd 3303out:
2f2dc053
SW		 3304 ceph_msg_put(msg);
3305}
3306
4e7a5dcd
SW		 3307/*
3308 * authentication
3309 */
3310static int get_authorizer(struct ceph_connection *con,
3311 void **buf, int *len, int *proto,
3312 void **reply_buf, int *reply_len, int force_new)
3313{
3314 struct ceph_mds_session *s = con->private;
3315 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3316 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd
SW		 3317 int ret = 0;
3318
3319 if (force_new && s->s_authorizer) {
3320 ac->ops->destroy_authorizer(ac, s->s_authorizer);
3321 s->s_authorizer = NULL;
3322 }
3323 if (s->s_authorizer == NULL) {
3324 if (ac->ops->create_authorizer) {
3325 ret = ac->ops->create_authorizer(
3326 ac, CEPH_ENTITY_TYPE_MDS,
3327 &s->s_authorizer,
3328 &s->s_authorizer_buf,
3329 &s->s_authorizer_buf_len,
3330 &s->s_authorizer_reply_buf,
3331 &s->s_authorizer_reply_buf_len);
3332 if (ret)
3333 return ret;
3334 }
3335 }
3336
3337 *proto = ac->protocol;
3338 *buf = s->s_authorizer_buf;
3339 *len = s->s_authorizer_buf_len;
3340 *reply_buf = s->s_authorizer_reply_buf;
3341 *reply_len = s->s_authorizer_reply_buf_len;
3342 return 0;
3343}
3344
3345
3346static int verify_authorizer_reply(struct ceph_connection *con, int len)
3347{
3348 struct ceph_mds_session *s = con->private;
3349 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3350 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd
SW		 3351
3352 return ac->ops->verify_authorizer_reply(ac, s->s_authorizer, len);
3353}
3354
9bd2e6f8
SW		 3355static int invalidate_authorizer(struct ceph_connection *con)
3356{
3357 struct ceph_mds_session *s = con->private;
3358 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3359 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8
SW		 3360
3361 if (ac->ops->invalidate_authorizer)
3362 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3363
3d14c5d2 3364 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3365}
3366
9e32789f 3367static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 3368 .get = con_get,
3369 .put = con_put,
3370 .dispatch = dispatch,
4e7a5dcd
SW		 3371 .get_authorizer = get_authorizer,
3372 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3373 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 3374 .peer_reset = peer_reset,
2f2dc053
SW		 3375};
3376
2f2dc053 3377/* eof */
Clone of davem's net-next-2.6 tree