?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.18.129
[linux/fpc-iii.git] / fs / ceph / mds_client.c
blob6f29455c03fe69e53f4ded43bad84304fbc79a2f
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/fs.h>
4 #include <linux/wait.h>
5 #include <linux/slab.h>
6 #include <linux/gfp.h>
7 #include <linux/sched.h>
8 #include <linux/debugfs.h>
9 #include <linux/seq_file.h>
10 #include <linux/utsname.h>
11
12 #include "super.h"
13 #include "mds_client.h"
14
15 #include <linux/ceph/ceph_features.h>
16 #include <linux/ceph/messenger.h>
17 #include <linux/ceph/decode.h>
18 #include <linux/ceph/pagelist.h>
19 #include <linux/ceph/auth.h>
20 #include <linux/ceph/debugfs.h>
21
22 /*
23 * A cluster of MDS (metadata server) daemons is responsible for
24 * managing the file system namespace (the directory hierarchy and
25 * inodes) and for coordinating shared access to storage. Metadata is
26 * partitioning hierarchically across a number of servers, and that
27 * partition varies over time as the cluster adjusts the distribution
28 * in order to balance load.
29 *
30 * The MDS client is primarily responsible to managing synchronous
31 * metadata requests for operations like open, unlink, and so forth.
32 * If there is a MDS failure, we find out about it when we (possibly
33 * request and) receive a new MDS map, and can resubmit affected
34 * requests.
35 *
36 * For the most part, though, we take advantage of a lossless
37 * communications channel to the MDS, and do not need to worry about
38 * timing out or resubmitting requests.
39 *
40 * We maintain a stateful "session" with each MDS we interact with.
41 * Within each session, we sent periodic heartbeat messages to ensure
42 * any capabilities or leases we have been issues remain valid. If
43 * the session times out and goes stale, our leases and capabilities
44 * are no longer valid.
45 */
46
47 struct ceph_reconnect_state {
48 int nr_caps;
49 struct ceph_pagelist *pagelist;
50 bool flock;
51 };
52
53 static void __wake_requests(struct ceph_mds_client *mdsc,
54 struct list_head *head);
55
56 static const struct ceph_connection_operations mds_con_ops;
57
58
59 /*
60 * mds reply parsing
61 */
62
63 /*
64 * parse individual inode info
65 */
66 static int parse_reply_info_in(void **p, void *end,
67 struct ceph_mds_reply_info_in *info,
68 u64 features)
69 {
70 int err = -EIO;
71
72 info->in = *p;
73 *p += sizeof(struct ceph_mds_reply_inode) +
74 sizeof(*info->in->fragtree.splits) *
75 le32_to_cpu(info->in->fragtree.nsplits);
76
77 ceph_decode_32_safe(p, end, info->symlink_len, bad);
78 ceph_decode_need(p, end, info->symlink_len, bad);
79 info->symlink = *p;
80 *p += info->symlink_len;
81
82 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
83 ceph_decode_copy_safe(p, end, &info->dir_layout,
84 sizeof(info->dir_layout), bad);
85 else
86 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
87
88 ceph_decode_32_safe(p, end, info->xattr_len, bad);
89 ceph_decode_need(p, end, info->xattr_len, bad);
90 info->xattr_data = *p;
91 *p += info->xattr_len;
92 return 0;
93 bad:
94 return err;
95 }
96
97 /*
98 * parse a normal reply, which may contain a (dir+)dentry and/or a
99 * target inode.
100 */
101 static int parse_reply_info_trace(void **p, void *end,
102 struct ceph_mds_reply_info_parsed *info,
103 u64 features)
104 {
105 int err;
106
107 if (info->head->is_dentry) {
108 err = parse_reply_info_in(p, end, &info->diri, features);
109 if (err < 0)
110 goto out_bad;
111
112 if (unlikely(*p + sizeof(*info->dirfrag) > end))
113 goto bad;
114 info->dirfrag = *p;
115 *p += sizeof(*info->dirfrag) +
116 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
117 if (unlikely(*p > end))
118 goto bad;
119
120 ceph_decode_32_safe(p, end, info->dname_len, bad);
121 ceph_decode_need(p, end, info->dname_len, bad);
122 info->dname = *p;
123 *p += info->dname_len;
124 info->dlease = *p;
125 *p += sizeof(*info->dlease);
126 }
127
128 if (info->head->is_target) {
129 err = parse_reply_info_in(p, end, &info->targeti, features);
130 if (err < 0)
131 goto out_bad;
132 }
133
134 if (unlikely(*p != end))
135 goto bad;
136 return 0;
137
138 bad:
139 err = -EIO;
140 out_bad:
141 pr_err("problem parsing mds trace %d\n", err);
142 return err;
143 }
144
145 /*
146 * parse readdir results
147 */
148 static int parse_reply_info_dir(void **p, void *end,
149 struct ceph_mds_reply_info_parsed *info,
150 u64 features)
151 {
152 u32 num, i = 0;
153 int err;
154
155 info->dir_dir = *p;
156 if (*p + sizeof(*info->dir_dir) > end)
157 goto bad;
158 *p += sizeof(*info->dir_dir) +
159 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
160 if (*p > end)
161 goto bad;
162
163 ceph_decode_need(p, end, sizeof(num) + 2, bad);
164 num = ceph_decode_32(p);
165 info->dir_end = ceph_decode_8(p);
166 info->dir_complete = ceph_decode_8(p);
167 if (num == 0)
168 goto done;
169
170 BUG_ON(!info->dir_in);
171 info->dir_dname = (void *)(info->dir_in + num);
172 info->dir_dname_len = (void *)(info->dir_dname + num);
173 info->dir_dlease = (void *)(info->dir_dname_len + num);
174 if ((unsigned long)(info->dir_dlease + num) >
175 (unsigned long)info->dir_in + info->dir_buf_size) {
176 pr_err("dir contents are larger than expected\n");
177 WARN_ON(1);
178 goto bad;
179 }
180
181 info->dir_nr = num;
182 while (num) {
183 /* dentry */
184 ceph_decode_need(p, end, sizeof(u32)*2, bad);
185 info->dir_dname_len[i] = ceph_decode_32(p);
186 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
187 info->dir_dname[i] = *p;
188 *p += info->dir_dname_len[i];
189 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
190 info->dir_dname[i]);
191 info->dir_dlease[i] = *p;
192 *p += sizeof(struct ceph_mds_reply_lease);
193
194 /* inode */
195 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
196 if (err < 0)
197 goto out_bad;
198 i++;
199 num--;
200 }
201
202 done:
203 if (*p != end)
204 goto bad;
205 return 0;
206
207 bad:
208 err = -EIO;
209 out_bad:
210 pr_err("problem parsing dir contents %d\n", err);
211 return err;
212 }
213
214 /*
215 * parse fcntl F_GETLK results
216 */
217 static int parse_reply_info_filelock(void **p, void *end,
218 struct ceph_mds_reply_info_parsed *info,
219 u64 features)
220 {
221 if (*p + sizeof(*info->filelock_reply) > end)
222 goto bad;
223
224 info->filelock_reply = *p;
225 *p += sizeof(*info->filelock_reply);
226
227 if (unlikely(*p != end))
228 goto bad;
229 return 0;
230
231 bad:
232 return -EIO;
233 }
234
235 /*
236 * parse create results
237 */
238 static int parse_reply_info_create(void **p, void *end,
239 struct ceph_mds_reply_info_parsed *info,
240 u64 features)
241 {
242 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
243 if (*p == end) {
244 info->has_create_ino = false;
245 } else {
246 info->has_create_ino = true;
247 info->ino = ceph_decode_64(p);
248 }
249 }
250
251 if (unlikely(*p != end))
252 goto bad;
253 return 0;
254
255 bad:
256 return -EIO;
257 }
258
259 /*
260 * parse extra results
261 */
262 static int parse_reply_info_extra(void **p, void *end,
263 struct ceph_mds_reply_info_parsed *info,
264 u64 features)
265 {
266 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
267 return parse_reply_info_filelock(p, end, info, features);
268 else if (info->head->op == CEPH_MDS_OP_READDIR ||
269 info->head->op == CEPH_MDS_OP_LSSNAP)
270 return parse_reply_info_dir(p, end, info, features);
271 else if (info->head->op == CEPH_MDS_OP_CREATE)
272 return parse_reply_info_create(p, end, info, features);
273 else
274 return -EIO;
275 }
276
277 /*
278 * parse entire mds reply
279 */
280 static int parse_reply_info(struct ceph_msg *msg,
281 struct ceph_mds_reply_info_parsed *info,
282 u64 features)
283 {
284 void *p, *end;
285 u32 len;
286 int err;
287
288 info->head = msg->front.iov_base;
289 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
290 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
291
292 /* trace */
293 ceph_decode_32_safe(&p, end, len, bad);
294 if (len > 0) {
295 ceph_decode_need(&p, end, len, bad);
296 err = parse_reply_info_trace(&p, p+len, info, features);
297 if (err < 0)
298 goto out_bad;
299 }
300
301 /* extra */
302 ceph_decode_32_safe(&p, end, len, bad);
303 if (len > 0) {
304 ceph_decode_need(&p, end, len, bad);
305 err = parse_reply_info_extra(&p, p+len, info, features);
306 if (err < 0)
307 goto out_bad;
308 }
309
310 /* snap blob */
311 ceph_decode_32_safe(&p, end, len, bad);
312 info->snapblob_len = len;
313 info->snapblob = p;
314 p += len;
315
316 if (p != end)
317 goto bad;
318 return 0;
319
320 bad:
321 err = -EIO;
322 out_bad:
323 pr_err("mds parse_reply err %d\n", err);
324 return err;
325 }
326
327 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
328 {
329 if (!info->dir_in)
330 return;
331 free_pages((unsigned long)info->dir_in, get_order(info->dir_buf_size));
332 }
333
334
335 /*
336 * sessions
337 */
338 const char *ceph_session_state_name(int s)
339 {
340 switch (s) {
341 case CEPH_MDS_SESSION_NEW: return "new";
342 case CEPH_MDS_SESSION_OPENING: return "opening";
343 case CEPH_MDS_SESSION_OPEN: return "open";
344 case CEPH_MDS_SESSION_HUNG: return "hung";
345 case CEPH_MDS_SESSION_CLOSING: return "closing";
346 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
347 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
348 default: return "???";
349 }
350 }
351
352 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
353 {
354 if (atomic_inc_not_zero(&s->s_ref)) {
355 dout("mdsc get_session %p %d -> %d\n", s,
356 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
357 return s;
358 } else {
359 dout("mdsc get_session %p 0 -- FAIL", s);
360 return NULL;
361 }
362 }
363
364 void ceph_put_mds_session(struct ceph_mds_session *s)
365 {
366 dout("mdsc put_session %p %d -> %d\n", s,
367 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
368 if (atomic_dec_and_test(&s->s_ref)) {
369 if (s->s_auth.authorizer)
370 ceph_auth_destroy_authorizer(
371 s->s_mdsc->fsc->client->monc.auth,
372 s->s_auth.authorizer);
373 kfree(s);
374 }
375 }
376
377 /*
378 * called under mdsc->mutex
379 */
380 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
381 int mds)
382 {
383 struct ceph_mds_session *session;
384
385 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
386 return NULL;
387 session = mdsc->sessions[mds];
388 dout("lookup_mds_session %p %d\n", session,
389 atomic_read(&session->s_ref));
390 get_session(session);
391 return session;
392 }
393
394 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
395 {
396 if (mds >= mdsc->max_sessions)
397 return false;
398 return mdsc->sessions[mds];
399 }
400
401 static int __verify_registered_session(struct ceph_mds_client *mdsc,
402 struct ceph_mds_session *s)
403 {
404 if (s->s_mds >= mdsc->max_sessions ||
405 mdsc->sessions[s->s_mds] != s)
406 return -ENOENT;
407 return 0;
408 }
409
410 /*
411 * create+register a new session for given mds.
412 * called under mdsc->mutex.
413 */
414 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
415 int mds)
416 {
417 struct ceph_mds_session *s;
418
419 if (mds >= mdsc->mdsmap->m_max_mds)
420 return ERR_PTR(-EINVAL);
421
422 s = kzalloc(sizeof(*s), GFP_NOFS);
423 if (!s)
424 return ERR_PTR(-ENOMEM);
425 s->s_mdsc = mdsc;
426 s->s_mds = mds;
427 s->s_state = CEPH_MDS_SESSION_NEW;
428 s->s_ttl = 0;
429 s->s_seq = 0;
430 mutex_init(&s->s_mutex);
431
432 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
433
434 spin_lock_init(&s->s_gen_ttl_lock);
435 s->s_cap_gen = 0;
436 s->s_cap_ttl = jiffies - 1;
437
438 spin_lock_init(&s->s_cap_lock);
439 s->s_renew_requested = 0;
440 s->s_renew_seq = 0;
441 INIT_LIST_HEAD(&s->s_caps);
442 s->s_nr_caps = 0;
443 s->s_trim_caps = 0;
444 atomic_set(&s->s_ref, 1);
445 INIT_LIST_HEAD(&s->s_waiting);
446 INIT_LIST_HEAD(&s->s_unsafe);
447 s->s_num_cap_releases = 0;
448 s->s_cap_reconnect = 0;
449 s->s_cap_iterator = NULL;
450 INIT_LIST_HEAD(&s->s_cap_releases);
451 INIT_LIST_HEAD(&s->s_cap_releases_done);
452 INIT_LIST_HEAD(&s->s_cap_flushing);
453 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
454
455 dout("register_session mds%d\n", mds);
456 if (mds >= mdsc->max_sessions) {
457 int newmax = 1 << get_count_order(mds+1);
458 struct ceph_mds_session **sa;
459
460 dout("register_session realloc to %d\n", newmax);
461 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
462 if (sa == NULL)
463 goto fail_realloc;
464 if (mdsc->sessions) {
465 memcpy(sa, mdsc->sessions,
466 mdsc->max_sessions * sizeof(void *));
467 kfree(mdsc->sessions);
468 }
469 mdsc->sessions = sa;
470 mdsc->max_sessions = newmax;
471 }
472 mdsc->sessions[mds] = s;
473 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
474
475 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
476 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
477
478 return s;
479
480 fail_realloc:
481 kfree(s);
482 return ERR_PTR(-ENOMEM);
483 }
484
485 /*
486 * called under mdsc->mutex
487 */
488 static void __unregister_session(struct ceph_mds_client *mdsc,
489 struct ceph_mds_session *s)
490 {
491 dout("__unregister_session mds%d %p\n", s->s_mds, s);
492 BUG_ON(mdsc->sessions[s->s_mds] != s);
493 mdsc->sessions[s->s_mds] = NULL;
494 ceph_con_close(&s->s_con);
495 ceph_put_mds_session(s);
496 }
497
498 /*
499 * drop session refs in request.
500 *
501 * should be last request ref, or hold mdsc->mutex
502 */
503 static void put_request_session(struct ceph_mds_request *req)
504 {
505 if (req->r_session) {
506 ceph_put_mds_session(req->r_session);
507 req->r_session = NULL;
508 }
509 }
510
511 void ceph_mdsc_release_request(struct kref *kref)
512 {
513 struct ceph_mds_request *req = container_of(kref,
514 struct ceph_mds_request,
515 r_kref);
516 destroy_reply_info(&req->r_reply_info);
517 if (req->r_request)
518 ceph_msg_put(req->r_request);
519 if (req->r_reply)
520 ceph_msg_put(req->r_reply);
521 if (req->r_inode) {
522 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
523 iput(req->r_inode);
524 }
525 if (req->r_locked_dir)
526 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
527 if (req->r_target_inode)
528 iput(req->r_target_inode);
529 if (req->r_dentry)
530 dput(req->r_dentry);
531 if (req->r_old_dentry)
532 dput(req->r_old_dentry);
533 if (req->r_old_dentry_dir) {
534 /*
535 * track (and drop pins for) r_old_dentry_dir
536 * separately, since r_old_dentry's d_parent may have
537 * changed between the dir mutex being dropped and
538 * this request being freed.
539 */
540 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
541 CEPH_CAP_PIN);
542 iput(req->r_old_dentry_dir);
543 }
544 kfree(req->r_path1);
545 kfree(req->r_path2);
546 if (req->r_pagelist)
547 ceph_pagelist_release(req->r_pagelist);
548 put_request_session(req);
549 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
550 kfree(req);
551 }
552
553 /*
554 * lookup session, bump ref if found.
555 *
556 * called under mdsc->mutex.
557 */
558 static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
559 u64 tid)
560 {
561 struct ceph_mds_request *req;
562 struct rb_node *n = mdsc->request_tree.rb_node;
563
564 while (n) {
565 req = rb_entry(n, struct ceph_mds_request, r_node);
566 if (tid < req->r_tid)
567 n = n->rb_left;
568 else if (tid > req->r_tid)
569 n = n->rb_right;
570 else {
571 ceph_mdsc_get_request(req);
572 return req;
573 }
574 }
575 return NULL;
576 }
577
578 static void __insert_request(struct ceph_mds_client *mdsc,
579 struct ceph_mds_request *new)
580 {
581 struct rb_node **p = &mdsc->request_tree.rb_node;
582 struct rb_node *parent = NULL;
583 struct ceph_mds_request *req = NULL;
584
585 while (*p) {
586 parent = *p;
587 req = rb_entry(parent, struct ceph_mds_request, r_node);
588 if (new->r_tid < req->r_tid)
589 p = &(*p)->rb_left;
590 else if (new->r_tid > req->r_tid)
591 p = &(*p)->rb_right;
592 else
593 BUG();
594 }
595
596 rb_link_node(&new->r_node, parent, p);
597 rb_insert_color(&new->r_node, &mdsc->request_tree);
598 }
599
600 /*
601 * Register an in-flight request, and assign a tid. Link to directory
602 * are modifying (if any).
603 *
604 * Called under mdsc->mutex.
605 */
606 static void __register_request(struct ceph_mds_client *mdsc,
607 struct ceph_mds_request *req,
608 struct inode *dir)
609 {
610 req->r_tid = ++mdsc->last_tid;
611 if (req->r_num_caps)
612 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
613 req->r_num_caps);
614 dout("__register_request %p tid %lld\n", req, req->r_tid);
615 ceph_mdsc_get_request(req);
616 __insert_request(mdsc, req);
617
618 req->r_uid = current_fsuid();
619 req->r_gid = current_fsgid();
620
621 if (dir) {
622 struct ceph_inode_info *ci = ceph_inode(dir);
623
624 ihold(dir);
625 spin_lock(&ci->i_unsafe_lock);
626 req->r_unsafe_dir = dir;
627 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
628 spin_unlock(&ci->i_unsafe_lock);
629 }
630 }
631
632 static void __unregister_request(struct ceph_mds_client *mdsc,
633 struct ceph_mds_request *req)
634 {
635 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
636 rb_erase(&req->r_node, &mdsc->request_tree);
637 RB_CLEAR_NODE(&req->r_node);
638
639 if (req->r_unsafe_dir) {
640 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
641
642 spin_lock(&ci->i_unsafe_lock);
643 list_del_init(&req->r_unsafe_dir_item);
644 spin_unlock(&ci->i_unsafe_lock);
645
646 iput(req->r_unsafe_dir);
647 req->r_unsafe_dir = NULL;
648 }
649
650 complete_all(&req->r_safe_completion);
651
652 ceph_mdsc_put_request(req);
653 }
654
655 /*
656 * Choose mds to send request to next. If there is a hint set in the
657 * request (e.g., due to a prior forward hint from the mds), use that.
658 * Otherwise, consult frag tree and/or caps to identify the
659 * appropriate mds. If all else fails, choose randomly.
660 *
661 * Called under mdsc->mutex.
662 */
663 static struct dentry *get_nonsnap_parent(struct dentry *dentry)
664 {
665 /*
666 * we don't need to worry about protecting the d_parent access
667 * here because we never renaming inside the snapped namespace
668 * except to resplice to another snapdir, and either the old or new
669 * result is a valid result.
670 */
671 while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
672 dentry = dentry->d_parent;
673 return dentry;
674 }
675
676 static int __choose_mds(struct ceph_mds_client *mdsc,
677 struct ceph_mds_request *req)
678 {
679 struct inode *inode;
680 struct ceph_inode_info *ci;
681 struct ceph_cap *cap;
682 int mode = req->r_direct_mode;
683 int mds = -1;
684 u32 hash = req->r_direct_hash;
685 bool is_hash = req->r_direct_is_hash;
686
687 /*
688 * is there a specific mds we should try? ignore hint if we have
689 * no session and the mds is not up (active or recovering).
690 */
691 if (req->r_resend_mds >= 0 &&
692 (__have_session(mdsc, req->r_resend_mds) ||
693 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
694 dout("choose_mds using resend_mds mds%d\n",
695 req->r_resend_mds);
696 return req->r_resend_mds;
697 }
698
699 if (mode == USE_RANDOM_MDS)
700 goto random;
701
702 inode = NULL;
703 if (req->r_inode) {
704 inode = req->r_inode;
705 } else if (req->r_dentry) {
706 /* ignore race with rename; old or new d_parent is okay */
707 struct dentry *parent = req->r_dentry->d_parent;
708 struct inode *dir = parent->d_inode;
709
710 if (dir->i_sb != mdsc->fsc->sb) {
711 /* not this fs! */
712 inode = req->r_dentry->d_inode;
713 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
714 /* direct snapped/virtual snapdir requests
715 * based on parent dir inode */
716 struct dentry *dn = get_nonsnap_parent(parent);
717 inode = dn->d_inode;
718 dout("__choose_mds using nonsnap parent %p\n", inode);
719 } else {
720 /* dentry target */
721 inode = req->r_dentry->d_inode;
722 if (!inode || mode == USE_AUTH_MDS) {
723 /* dir + name */
724 inode = dir;
725 hash = ceph_dentry_hash(dir, req->r_dentry);
726 is_hash = true;
727 }
728 }
729 }
730
731 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
732 (int)hash, mode);
733 if (!inode)
734 goto random;
735 ci = ceph_inode(inode);
736
737 if (is_hash && S_ISDIR(inode->i_mode)) {
738 struct ceph_inode_frag frag;
739 int found;
740
741 ceph_choose_frag(ci, hash, &frag, &found);
742 if (found) {
743 if (mode == USE_ANY_MDS && frag.ndist > 0) {
744 u8 r;
745
746 /* choose a random replica */
747 get_random_bytes(&r, 1);
748 r %= frag.ndist;
749 mds = frag.dist[r];
750 dout("choose_mds %p %llx.%llx "
751 "frag %u mds%d (%d/%d)\n",
752 inode, ceph_vinop(inode),
753 frag.frag, mds,
754 (int)r, frag.ndist);
755 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
756 CEPH_MDS_STATE_ACTIVE)
757 return mds;
758 }
759
760 /* since this file/dir wasn't known to be
761 * replicated, then we want to look for the
762 * authoritative mds. */
763 mode = USE_AUTH_MDS;
764 if (frag.mds >= 0) {
765 /* choose auth mds */
766 mds = frag.mds;
767 dout("choose_mds %p %llx.%llx "
768 "frag %u mds%d (auth)\n",
769 inode, ceph_vinop(inode), frag.frag, mds);
770 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
771 CEPH_MDS_STATE_ACTIVE)
772 return mds;
773 }
774 }
775 }
776
777 spin_lock(&ci->i_ceph_lock);
778 cap = NULL;
779 if (mode == USE_AUTH_MDS)
780 cap = ci->i_auth_cap;
781 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
782 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
783 if (!cap) {
784 spin_unlock(&ci->i_ceph_lock);
785 goto random;
786 }
787 mds = cap->session->s_mds;
788 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
789 inode, ceph_vinop(inode), mds,
790 cap == ci->i_auth_cap ? "auth " : "", cap);
791 spin_unlock(&ci->i_ceph_lock);
792 return mds;
793
794 random:
795 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
796 dout("choose_mds chose random mds%d\n", mds);
797 return mds;
798 }
799
800
801 /*
802 * session messages
803 */
804 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
805 {
806 struct ceph_msg *msg;
807 struct ceph_mds_session_head *h;
808
809 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
810 false);
811 if (!msg) {
812 pr_err("create_session_msg ENOMEM creating msg\n");
813 return NULL;
814 }
815 h = msg->front.iov_base;
816 h->op = cpu_to_le32(op);
817 h->seq = cpu_to_le64(seq);
818
819 return msg;
820 }
821
822 /*
823 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
824 * to include additional client metadata fields.
825 */
826 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
827 {
828 struct ceph_msg *msg;
829 struct ceph_mds_session_head *h;
830 int i = -1;
831 int metadata_bytes = 0;
832 int metadata_key_count = 0;
833 struct ceph_options *opt = mdsc->fsc->client->options;
834 void *p;
835
836 const char* metadata[3][2] = {
837 {"hostname", utsname()->nodename},
838 {"entity_id", opt->name ? opt->name : ""},
839 {NULL, NULL}
840 };
841
842 /* Calculate serialized length of metadata */
843 metadata_bytes = 4; /* map length */
844 for (i = 0; metadata[i][0] != NULL; ++i) {
845 metadata_bytes += 8 + strlen(metadata[i][0]) +
846 strlen(metadata[i][1]);
847 metadata_key_count++;
848 }
849
850 /* Allocate the message */
851 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
852 GFP_NOFS, false);
853 if (!msg) {
854 pr_err("create_session_msg ENOMEM creating msg\n");
855 return NULL;
856 }
857 h = msg->front.iov_base;
858 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
859 h->seq = cpu_to_le64(seq);
860
861 /*
862 * Serialize client metadata into waiting buffer space, using
863 * the format that userspace expects for map<string, string>
864 */
865 msg->hdr.version = 2; /* ClientSession messages with metadata are v2 */
866
867 /* The write pointer, following the session_head structure */
868 p = msg->front.iov_base + sizeof(*h);
869
870 /* Number of entries in the map */
871 ceph_encode_32(&p, metadata_key_count);
872
873 /* Two length-prefixed strings for each entry in the map */
874 for (i = 0; metadata[i][0] != NULL; ++i) {
875 size_t const key_len = strlen(metadata[i][0]);
876 size_t const val_len = strlen(metadata[i][1]);
877
878 ceph_encode_32(&p, key_len);
879 memcpy(p, metadata[i][0], key_len);
880 p += key_len;
881 ceph_encode_32(&p, val_len);
882 memcpy(p, metadata[i][1], val_len);
883 p += val_len;
884 }
885
886 return msg;
887 }
888
889 /*
890 * send session open request.
891 *
892 * called under mdsc->mutex
893 */
894 static int __open_session(struct ceph_mds_client *mdsc,
895 struct ceph_mds_session *session)
896 {
897 struct ceph_msg *msg;
898 int mstate;
899 int mds = session->s_mds;
900
901 /* wait for mds to go active? */
902 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
903 dout("open_session to mds%d (%s)\n", mds,
904 ceph_mds_state_name(mstate));
905 session->s_state = CEPH_MDS_SESSION_OPENING;
906 session->s_renew_requested = jiffies;
907
908 /* send connect message */
909 msg = create_session_open_msg(mdsc, session->s_seq);
910 if (!msg)
911 return -ENOMEM;
912 ceph_con_send(&session->s_con, msg);
913 return 0;
914 }
915
916 /*
917 * open sessions for any export targets for the given mds
918 *
919 * called under mdsc->mutex
920 */
921 static struct ceph_mds_session *
922 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
923 {
924 struct ceph_mds_session *session;
925
926 session = __ceph_lookup_mds_session(mdsc, target);
927 if (!session) {
928 session = register_session(mdsc, target);
929 if (IS_ERR(session))
930 return session;
931 }
932 if (session->s_state == CEPH_MDS_SESSION_NEW ||
933 session->s_state == CEPH_MDS_SESSION_CLOSING)
934 __open_session(mdsc, session);
935
936 return session;
937 }
938
939 struct ceph_mds_session *
940 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
941 {
942 struct ceph_mds_session *session;
943
944 dout("open_export_target_session to mds%d\n", target);
945
946 mutex_lock(&mdsc->mutex);
947 session = __open_export_target_session(mdsc, target);
948 mutex_unlock(&mdsc->mutex);
949
950 return session;
951 }
952
953 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
954 struct ceph_mds_session *session)
955 {
956 struct ceph_mds_info *mi;
957 struct ceph_mds_session *ts;
958 int i, mds = session->s_mds;
959
960 if (mds >= mdsc->mdsmap->m_max_mds)
961 return;
962
963 mi = &mdsc->mdsmap->m_info[mds];
964 dout("open_export_target_sessions for mds%d (%d targets)\n",
965 session->s_mds, mi->num_export_targets);
966
967 for (i = 0; i < mi->num_export_targets; i++) {
968 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
969 if (!IS_ERR(ts))
970 ceph_put_mds_session(ts);
971 }
972 }
973
974 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
975 struct ceph_mds_session *session)
976 {
977 mutex_lock(&mdsc->mutex);
978 __open_export_target_sessions(mdsc, session);
979 mutex_unlock(&mdsc->mutex);
980 }
981
982 /*
983 * session caps
984 */
985
986 /*
987 * Free preallocated cap messages assigned to this session
988 */
989 static void cleanup_cap_releases(struct ceph_mds_session *session)
990 {
991 struct ceph_msg *msg;
992
993 spin_lock(&session->s_cap_lock);
994 while (!list_empty(&session->s_cap_releases)) {
995 msg = list_first_entry(&session->s_cap_releases,
996 struct ceph_msg, list_head);
997 list_del_init(&msg->list_head);
998 ceph_msg_put(msg);
999 }
1000 while (!list_empty(&session->s_cap_releases_done)) {
1001 msg = list_first_entry(&session->s_cap_releases_done,
1002 struct ceph_msg, list_head);
1003 list_del_init(&msg->list_head);
1004 ceph_msg_put(msg);
1005 }
1006 spin_unlock(&session->s_cap_lock);
1007 }
1008
1009 /*
1010 * Helper to safely iterate over all caps associated with a session, with
1011 * special care taken to handle a racing __ceph_remove_cap().
1012 *
1013 * Caller must hold session s_mutex.
1014 */
1015 static int iterate_session_caps(struct ceph_mds_session *session,
1016 int (*cb)(struct inode *, struct ceph_cap *,
1017 void *), void *arg)
1018 {
1019 struct list_head *p;
1020 struct ceph_cap *cap;
1021 struct inode *inode, *last_inode = NULL;
1022 struct ceph_cap *old_cap = NULL;
1023 int ret;
1024
1025 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1026 spin_lock(&session->s_cap_lock);
1027 p = session->s_caps.next;
1028 while (p != &session->s_caps) {
1029 cap = list_entry(p, struct ceph_cap, session_caps);
1030 inode = igrab(&cap->ci->vfs_inode);
1031 if (!inode) {
1032 p = p->next;
1033 continue;
1034 }
1035 session->s_cap_iterator = cap;
1036 spin_unlock(&session->s_cap_lock);
1037
1038 if (last_inode) {
1039 iput(last_inode);
1040 last_inode = NULL;
1041 }
1042 if (old_cap) {
1043 ceph_put_cap(session->s_mdsc, old_cap);
1044 old_cap = NULL;
1045 }
1046
1047 ret = cb(inode, cap, arg);
1048 last_inode = inode;
1049
1050 spin_lock(&session->s_cap_lock);
1051 p = p->next;
1052 if (cap->ci == NULL) {
1053 dout("iterate_session_caps finishing cap %p removal\n",
1054 cap);
1055 BUG_ON(cap->session != session);
1056 list_del_init(&cap->session_caps);
1057 session->s_nr_caps--;
1058 cap->session = NULL;
1059 old_cap = cap; /* put_cap it w/o locks held */
1060 }
1061 if (ret < 0)
1062 goto out;
1063 }
1064 ret = 0;
1065 out:
1066 session->s_cap_iterator = NULL;
1067 spin_unlock(&session->s_cap_lock);
1068
1069 if (last_inode)
1070 iput(last_inode);
1071 if (old_cap)
1072 ceph_put_cap(session->s_mdsc, old_cap);
1073
1074 return ret;
1075 }
1076
1077 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1078 void *arg)
1079 {
1080 struct ceph_inode_info *ci = ceph_inode(inode);
1081 int drop = 0;
1082
1083 dout("removing cap %p, ci is %p, inode is %p\n",
1084 cap, ci, &ci->vfs_inode);
1085 spin_lock(&ci->i_ceph_lock);
1086 __ceph_remove_cap(cap, false);
1087 if (!__ceph_is_any_real_caps(ci)) {
1088 struct ceph_mds_client *mdsc =
1089 ceph_sb_to_client(inode->i_sb)->mdsc;
1090
1091 spin_lock(&mdsc->cap_dirty_lock);
1092 if (!list_empty(&ci->i_dirty_item)) {
1093 pr_info(" dropping dirty %s state for %p %lld\n",
1094 ceph_cap_string(ci->i_dirty_caps),
1095 inode, ceph_ino(inode));
1096 ci->i_dirty_caps = 0;
1097 list_del_init(&ci->i_dirty_item);
1098 drop = 1;
1099 }
1100 if (!list_empty(&ci->i_flushing_item)) {
1101 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1102 ceph_cap_string(ci->i_flushing_caps),
1103 inode, ceph_ino(inode));
1104 ci->i_flushing_caps = 0;
1105 list_del_init(&ci->i_flushing_item);
1106 mdsc->num_cap_flushing--;
1107 drop = 1;
1108 }
1109 if (drop && ci->i_wrbuffer_ref) {
1110 pr_info(" dropping dirty data for %p %lld\n",
1111 inode, ceph_ino(inode));
1112 ci->i_wrbuffer_ref = 0;
1113 ci->i_wrbuffer_ref_head = 0;
1114 drop++;
1115 }
1116 spin_unlock(&mdsc->cap_dirty_lock);
1117 }
1118 spin_unlock(&ci->i_ceph_lock);
1119 while (drop--)
1120 iput(inode);
1121 return 0;
1122 }
1123
1124 /*
1125 * caller must hold session s_mutex
1126 */
1127 static void remove_session_caps(struct ceph_mds_session *session)
1128 {
1129 dout("remove_session_caps on %p\n", session);
1130 iterate_session_caps(session, remove_session_caps_cb, NULL);
1131
1132 spin_lock(&session->s_cap_lock);
1133 if (session->s_nr_caps > 0) {
1134 struct super_block *sb = session->s_mdsc->fsc->sb;
1135 struct inode *inode;
1136 struct ceph_cap *cap, *prev = NULL;
1137 struct ceph_vino vino;
1138 /*
1139 * iterate_session_caps() skips inodes that are being
1140 * deleted, we need to wait until deletions are complete.
1141 * __wait_on_freeing_inode() is designed for the job,
1142 * but it is not exported, so use lookup inode function
1143 * to access it.
1144 */
1145 while (!list_empty(&session->s_caps)) {
1146 cap = list_entry(session->s_caps.next,
1147 struct ceph_cap, session_caps);
1148 if (cap == prev)
1149 break;
1150 prev = cap;
1151 vino = cap->ci->i_vino;
1152 spin_unlock(&session->s_cap_lock);
1153
1154 inode = ceph_find_inode(sb, vino);
1155 iput(inode);
1156
1157 spin_lock(&session->s_cap_lock);
1158 }
1159 }
1160 spin_unlock(&session->s_cap_lock);
1161
1162 BUG_ON(session->s_nr_caps > 0);
1163 BUG_ON(!list_empty(&session->s_cap_flushing));
1164 cleanup_cap_releases(session);
1165 }
1166
1167 /*
1168 * wake up any threads waiting on this session's caps. if the cap is
1169 * old (didn't get renewed on the client reconnect), remove it now.
1170 *
1171 * caller must hold s_mutex.
1172 */
1173 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1174 void *arg)
1175 {
1176 struct ceph_inode_info *ci = ceph_inode(inode);
1177
1178 wake_up_all(&ci->i_cap_wq);
1179 if (arg) {
1180 spin_lock(&ci->i_ceph_lock);
1181 ci->i_wanted_max_size = 0;
1182 ci->i_requested_max_size = 0;
1183 spin_unlock(&ci->i_ceph_lock);
1184 }
1185 return 0;
1186 }
1187
1188 static void wake_up_session_caps(struct ceph_mds_session *session,
1189 int reconnect)
1190 {
1191 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1192 iterate_session_caps(session, wake_up_session_cb,
1193 (void *)(unsigned long)reconnect);
1194 }
1195
1196 /*
1197 * Send periodic message to MDS renewing all currently held caps. The
1198 * ack will reset the expiration for all caps from this session.
1199 *
1200 * caller holds s_mutex
1201 */
1202 static int send_renew_caps(struct ceph_mds_client *mdsc,
1203 struct ceph_mds_session *session)
1204 {
1205 struct ceph_msg *msg;
1206 int state;
1207
1208 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1209 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1210 pr_info("mds%d caps stale\n", session->s_mds);
1211 session->s_renew_requested = jiffies;
1212
1213 /* do not try to renew caps until a recovering mds has reconnected
1214 * with its clients. */
1215 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1216 if (state < CEPH_MDS_STATE_RECONNECT) {
1217 dout("send_renew_caps ignoring mds%d (%s)\n",
1218 session->s_mds, ceph_mds_state_name(state));
1219 return 0;
1220 }
1221
1222 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1223 ceph_mds_state_name(state));
1224 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1225 ++session->s_renew_seq);
1226 if (!msg)
1227 return -ENOMEM;
1228 ceph_con_send(&session->s_con, msg);
1229 return 0;
1230 }
1231
1232 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1233 struct ceph_mds_session *session, u64 seq)
1234 {
1235 struct ceph_msg *msg;
1236
1237 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1238 session->s_mds, ceph_session_state_name(session->s_state), seq);
1239 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1240 if (!msg)
1241 return -ENOMEM;
1242 ceph_con_send(&session->s_con, msg);
1243 return 0;
1244 }
1245
1246
1247 /*
1248 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1249 *
1250 * Called under session->s_mutex
1251 */
1252 static void renewed_caps(struct ceph_mds_client *mdsc,
1253 struct ceph_mds_session *session, int is_renew)
1254 {
1255 int was_stale;
1256 int wake = 0;
1257
1258 spin_lock(&session->s_cap_lock);
1259 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1260
1261 session->s_cap_ttl = session->s_renew_requested +
1262 mdsc->mdsmap->m_session_timeout*HZ;
1263
1264 if (was_stale) {
1265 if (time_before(jiffies, session->s_cap_ttl)) {
1266 pr_info("mds%d caps renewed\n", session->s_mds);
1267 wake = 1;
1268 } else {
1269 pr_info("mds%d caps still stale\n", session->s_mds);
1270 }
1271 }
1272 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1273 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1274 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1275 spin_unlock(&session->s_cap_lock);
1276
1277 if (wake)
1278 wake_up_session_caps(session, 0);
1279 }
1280
1281 /*
1282 * send a session close request
1283 */
1284 static int request_close_session(struct ceph_mds_client *mdsc,
1285 struct ceph_mds_session *session)
1286 {
1287 struct ceph_msg *msg;
1288
1289 dout("request_close_session mds%d state %s seq %lld\n",
1290 session->s_mds, ceph_session_state_name(session->s_state),
1291 session->s_seq);
1292 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1293 if (!msg)
1294 return -ENOMEM;
1295 ceph_con_send(&session->s_con, msg);
1296 return 0;
1297 }
1298
1299 /*
1300 * Called with s_mutex held.
1301 */
1302 static int __close_session(struct ceph_mds_client *mdsc,
1303 struct ceph_mds_session *session)
1304 {
1305 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1306 return 0;
1307 session->s_state = CEPH_MDS_SESSION_CLOSING;
1308 return request_close_session(mdsc, session);
1309 }
1310
1311 /*
1312 * Trim old(er) caps.
1313 *
1314 * Because we can't cache an inode without one or more caps, we do
1315 * this indirectly: if a cap is unused, we prune its aliases, at which
1316 * point the inode will hopefully get dropped to.
1317 *
1318 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1319 * memory pressure from the MDS, though, so it needn't be perfect.
1320 */
1321 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1322 {
1323 struct ceph_mds_session *session = arg;
1324 struct ceph_inode_info *ci = ceph_inode(inode);
1325 int used, wanted, oissued, mine;
1326
1327 if (session->s_trim_caps <= 0)
1328 return -1;
1329
1330 spin_lock(&ci->i_ceph_lock);
1331 mine = cap->issued | cap->implemented;
1332 used = __ceph_caps_used(ci);
1333 wanted = __ceph_caps_file_wanted(ci);
1334 oissued = __ceph_caps_issued_other(ci, cap);
1335
1336 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1337 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1338 ceph_cap_string(used), ceph_cap_string(wanted));
1339 if (cap == ci->i_auth_cap) {
1340 if (ci->i_dirty_caps | ci->i_flushing_caps)
1341 goto out;
1342 if ((used | wanted) & CEPH_CAP_ANY_WR)
1343 goto out;
1344 }
1345 if ((used | wanted) & ~oissued & mine)
1346 goto out; /* we need these caps */
1347
1348 session->s_trim_caps--;
1349 if (oissued) {
1350 /* we aren't the only cap.. just remove us */
1351 __ceph_remove_cap(cap, true);
1352 } else {
1353 /* try to drop referring dentries */
1354 spin_unlock(&ci->i_ceph_lock);
1355 d_prune_aliases(inode);
1356 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1357 inode, cap, atomic_read(&inode->i_count));
1358 return 0;
1359 }
1360
1361 out:
1362 spin_unlock(&ci->i_ceph_lock);
1363 return 0;
1364 }
1365
1366 /*
1367 * Trim session cap count down to some max number.
1368 */
1369 static int trim_caps(struct ceph_mds_client *mdsc,
1370 struct ceph_mds_session *session,
1371 int max_caps)
1372 {
1373 int trim_caps = session->s_nr_caps - max_caps;
1374
1375 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1376 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1377 if (trim_caps > 0) {
1378 session->s_trim_caps = trim_caps;
1379 iterate_session_caps(session, trim_caps_cb, session);
1380 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1381 session->s_mds, session->s_nr_caps, max_caps,
1382 trim_caps - session->s_trim_caps);
1383 session->s_trim_caps = 0;
1384 }
1385
1386 ceph_add_cap_releases(mdsc, session);
1387 ceph_send_cap_releases(mdsc, session);
1388 return 0;
1389 }
1390
1391 /*
1392 * Allocate cap_release messages. If there is a partially full message
1393 * in the queue, try to allocate enough to cover it's remainder, so that
1394 * we can send it immediately.
1395 *
1396 * Called under s_mutex.
1397 */
1398 int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
1399 struct ceph_mds_session *session)
1400 {
1401 struct ceph_msg *msg, *partial = NULL;
1402 struct ceph_mds_cap_release *head;
1403 int err = -ENOMEM;
1404 int extra = mdsc->fsc->mount_options->cap_release_safety;
1405 int num;
1406
1407 dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1408 extra);
1409
1410 spin_lock(&session->s_cap_lock);
1411
1412 if (!list_empty(&session->s_cap_releases)) {
1413 msg = list_first_entry(&session->s_cap_releases,
1414 struct ceph_msg,
1415 list_head);
1416 head = msg->front.iov_base;
1417 num = le32_to_cpu(head->num);
1418 if (num) {
1419 dout(" partial %p with (%d/%d)\n", msg, num,
1420 (int)CEPH_CAPS_PER_RELEASE);
1421 extra += CEPH_CAPS_PER_RELEASE - num;
1422 partial = msg;
1423 }
1424 }
1425 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1426 spin_unlock(&session->s_cap_lock);
1427 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1428 GFP_NOFS, false);
1429 if (!msg)
1430 goto out_unlocked;
1431 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1432 (int)msg->front.iov_len);
1433 head = msg->front.iov_base;
1434 head->num = cpu_to_le32(0);
1435 msg->front.iov_len = sizeof(*head);
1436 spin_lock(&session->s_cap_lock);
1437 list_add(&msg->list_head, &session->s_cap_releases);
1438 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1439 }
1440
1441 if (partial) {
1442 head = partial->front.iov_base;
1443 num = le32_to_cpu(head->num);
1444 dout(" queueing partial %p with %d/%d\n", partial, num,
1445 (int)CEPH_CAPS_PER_RELEASE);
1446 list_move_tail(&partial->list_head,
1447 &session->s_cap_releases_done);
1448 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
1449 }
1450 err = 0;
1451 spin_unlock(&session->s_cap_lock);
1452 out_unlocked:
1453 return err;
1454 }
1455
1456 /*
1457 * flush all dirty inode data to disk.
1458 *
1459 * returns true if we've flushed through want_flush_seq
1460 */
1461 static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1462 {
1463 int mds, ret = 1;
1464
1465 dout("check_cap_flush want %lld\n", want_flush_seq);
1466 mutex_lock(&mdsc->mutex);
1467 for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1468 struct ceph_mds_session *session = mdsc->sessions[mds];
1469
1470 if (!session)
1471 continue;
1472 get_session(session);
1473 mutex_unlock(&mdsc->mutex);
1474
1475 mutex_lock(&session->s_mutex);
1476 if (!list_empty(&session->s_cap_flushing)) {
1477 struct ceph_inode_info *ci =
1478 list_entry(session->s_cap_flushing.next,
1479 struct ceph_inode_info,
1480 i_flushing_item);
1481 struct inode *inode = &ci->vfs_inode;
1482
1483 spin_lock(&ci->i_ceph_lock);
1484 if (ci->i_cap_flush_seq <= want_flush_seq) {
1485 dout("check_cap_flush still flushing %p "
1486 "seq %lld <= %lld to mds%d\n", inode,
1487 ci->i_cap_flush_seq, want_flush_seq,
1488 session->s_mds);
1489 ret = 0;
1490 }
1491 spin_unlock(&ci->i_ceph_lock);
1492 }
1493 mutex_unlock(&session->s_mutex);
1494 ceph_put_mds_session(session);
1495
1496 if (!ret)
1497 return ret;
1498 mutex_lock(&mdsc->mutex);
1499 }
1500
1501 mutex_unlock(&mdsc->mutex);
1502 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1503 return ret;
1504 }
1505
1506 /*
1507 * called under s_mutex
1508 */
1509 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1510 struct ceph_mds_session *session)
1511 {
1512 struct ceph_msg *msg;
1513
1514 dout("send_cap_releases mds%d\n", session->s_mds);
1515 spin_lock(&session->s_cap_lock);
1516 while (!list_empty(&session->s_cap_releases_done)) {
1517 msg = list_first_entry(&session->s_cap_releases_done,
1518 struct ceph_msg, list_head);
1519 list_del_init(&msg->list_head);
1520 spin_unlock(&session->s_cap_lock);
1521 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1522 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1523 ceph_con_send(&session->s_con, msg);
1524 spin_lock(&session->s_cap_lock);
1525 }
1526 spin_unlock(&session->s_cap_lock);
1527 }
1528
1529 static void discard_cap_releases(struct ceph_mds_client *mdsc,
1530 struct ceph_mds_session *session)
1531 {
1532 struct ceph_msg *msg;
1533 struct ceph_mds_cap_release *head;
1534 unsigned num;
1535
1536 dout("discard_cap_releases mds%d\n", session->s_mds);
1537
1538 if (!list_empty(&session->s_cap_releases)) {
1539 /* zero out the in-progress message */
1540 msg = list_first_entry(&session->s_cap_releases,
1541 struct ceph_msg, list_head);
1542 head = msg->front.iov_base;
1543 num = le32_to_cpu(head->num);
1544 dout("discard_cap_releases mds%d %p %u\n",
1545 session->s_mds, msg, num);
1546 head->num = cpu_to_le32(0);
1547 msg->front.iov_len = sizeof(*head);
1548 session->s_num_cap_releases += num;
1549 }
1550
1551 /* requeue completed messages */
1552 while (!list_empty(&session->s_cap_releases_done)) {
1553 msg = list_first_entry(&session->s_cap_releases_done,
1554 struct ceph_msg, list_head);
1555 list_del_init(&msg->list_head);
1556
1557 head = msg->front.iov_base;
1558 num = le32_to_cpu(head->num);
1559 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1560 num);
1561 session->s_num_cap_releases += num;
1562 head->num = cpu_to_le32(0);
1563 msg->front.iov_len = sizeof(*head);
1564 list_add(&msg->list_head, &session->s_cap_releases);
1565 }
1566 }
1567
1568 /*
1569 * requests
1570 */
1571
1572 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1573 struct inode *dir)
1574 {
1575 struct ceph_inode_info *ci = ceph_inode(dir);
1576 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1577 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1578 size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1579 sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1580 int order, num_entries;
1581
1582 spin_lock(&ci->i_ceph_lock);
1583 num_entries = ci->i_files + ci->i_subdirs;
1584 spin_unlock(&ci->i_ceph_lock);
1585 num_entries = max(num_entries, 1);
1586 num_entries = min(num_entries, opt->max_readdir);
1587
1588 order = get_order(size * num_entries);
1589 while (order >= 0) {
1590 rinfo->dir_in = (void*)__get_free_pages(GFP_NOFS | __GFP_NOWARN,
1591 order);
1592 if (rinfo->dir_in)
1593 break;
1594 order--;
1595 }
1596 if (!rinfo->dir_in)
1597 return -ENOMEM;
1598
1599 num_entries = (PAGE_SIZE << order) / size;
1600 num_entries = min(num_entries, opt->max_readdir);
1601
1602 rinfo->dir_buf_size = PAGE_SIZE << order;
1603 req->r_num_caps = num_entries + 1;
1604 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1605 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1606 return 0;
1607 }
1608
1609 /*
1610 * Create an mds request.
1611 */
1612 struct ceph_mds_request *
1613 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1614 {
1615 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1616
1617 if (!req)
1618 return ERR_PTR(-ENOMEM);
1619
1620 mutex_init(&req->r_fill_mutex);
1621 req->r_mdsc = mdsc;
1622 req->r_started = jiffies;
1623 req->r_resend_mds = -1;
1624 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1625 req->r_fmode = -1;
1626 kref_init(&req->r_kref);
1627 INIT_LIST_HEAD(&req->r_wait);
1628 init_completion(&req->r_completion);
1629 init_completion(&req->r_safe_completion);
1630 INIT_LIST_HEAD(&req->r_unsafe_item);
1631
1632 req->r_stamp = CURRENT_TIME;
1633
1634 req->r_op = op;
1635 req->r_direct_mode = mode;
1636 return req;
1637 }
1638
1639 /*
1640 * return oldest (lowest) request, tid in request tree, 0 if none.
1641 *
1642 * called under mdsc->mutex.
1643 */
1644 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1645 {
1646 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1647 return NULL;
1648 return rb_entry(rb_first(&mdsc->request_tree),
1649 struct ceph_mds_request, r_node);
1650 }
1651
1652 static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1653 {
1654 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1655
1656 if (req)
1657 return req->r_tid;
1658 return 0;
1659 }
1660
1661 /*
1662 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1663 * on build_path_from_dentry in fs/cifs/dir.c.
1664 *
1665 * If @stop_on_nosnap, generate path relative to the first non-snapped
1666 * inode.
1667 *
1668 * Encode hidden .snap dirs as a double /, i.e.
1669 * foo/.snap/bar -> foo//bar
1670 */
1671 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1672 int stop_on_nosnap)
1673 {
1674 struct dentry *temp;
1675 char *path;
1676 int len, pos;
1677 unsigned seq;
1678
1679 if (dentry == NULL)
1680 return ERR_PTR(-EINVAL);
1681
1682 retry:
1683 len = 0;
1684 seq = read_seqbegin(&rename_lock);
1685 rcu_read_lock();
1686 for (temp = dentry; !IS_ROOT(temp);) {
1687 struct inode *inode = temp->d_inode;
1688 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1689 len++; /* slash only */
1690 else if (stop_on_nosnap && inode &&
1691 ceph_snap(inode) == CEPH_NOSNAP)
1692 break;
1693 else
1694 len += 1 + temp->d_name.len;
1695 temp = temp->d_parent;
1696 }
1697 rcu_read_unlock();
1698 if (len)
1699 len--; /* no leading '/' */
1700
1701 path = kmalloc(len+1, GFP_NOFS);
1702 if (path == NULL)
1703 return ERR_PTR(-ENOMEM);
1704 pos = len;
1705 path[pos] = 0; /* trailing null */
1706 rcu_read_lock();
1707 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1708 struct inode *inode;
1709
1710 spin_lock(&temp->d_lock);
1711 inode = temp->d_inode;
1712 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1713 dout("build_path path+%d: %p SNAPDIR\n",
1714 pos, temp);
1715 } else if (stop_on_nosnap && inode &&
1716 ceph_snap(inode) == CEPH_NOSNAP) {
1717 spin_unlock(&temp->d_lock);
1718 break;
1719 } else {
1720 pos -= temp->d_name.len;
1721 if (pos < 0) {
1722 spin_unlock(&temp->d_lock);
1723 break;
1724 }
1725 strncpy(path + pos, temp->d_name.name,
1726 temp->d_name.len);
1727 }
1728 spin_unlock(&temp->d_lock);
1729 if (pos)
1730 path[--pos] = '/';
1731 temp = temp->d_parent;
1732 }
1733 rcu_read_unlock();
1734 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1735 pr_err("build_path did not end path lookup where "
1736 "expected, namelen is %d, pos is %d\n", len, pos);
1737 /* presumably this is only possible if racing with a
1738 rename of one of the parent directories (we can not
1739 lock the dentries above us to prevent this, but
1740 retrying should be harmless) */
1741 kfree(path);
1742 goto retry;
1743 }
1744
1745 *base = ceph_ino(temp->d_inode);
1746 *plen = len;
1747 dout("build_path on %p %d built %llx '%.*s'\n",
1748 dentry, d_count(dentry), *base, len, path);
1749 return path;
1750 }
1751
1752 static int build_dentry_path(struct dentry *dentry,
1753 const char **ppath, int *ppathlen, u64 *pino,
1754 int *pfreepath)
1755 {
1756 char *path;
1757
1758 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1759 *pino = ceph_ino(dentry->d_parent->d_inode);
1760 *ppath = dentry->d_name.name;
1761 *ppathlen = dentry->d_name.len;
1762 return 0;
1763 }
1764 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1765 if (IS_ERR(path))
1766 return PTR_ERR(path);
1767 *ppath = path;
1768 *pfreepath = 1;
1769 return 0;
1770 }
1771
1772 static int build_inode_path(struct inode *inode,
1773 const char **ppath, int *ppathlen, u64 *pino,
1774 int *pfreepath)
1775 {
1776 struct dentry *dentry;
1777 char *path;
1778
1779 if (ceph_snap(inode) == CEPH_NOSNAP) {
1780 *pino = ceph_ino(inode);
1781 *ppathlen = 0;
1782 return 0;
1783 }
1784 dentry = d_find_alias(inode);
1785 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1786 dput(dentry);
1787 if (IS_ERR(path))
1788 return PTR_ERR(path);
1789 *ppath = path;
1790 *pfreepath = 1;
1791 return 0;
1792 }
1793
1794 /*
1795 * request arguments may be specified via an inode *, a dentry *, or
1796 * an explicit ino+path.
1797 */
1798 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1799 const char *rpath, u64 rino,
1800 const char **ppath, int *pathlen,
1801 u64 *ino, int *freepath)
1802 {
1803 int r = 0;
1804
1805 if (rinode) {
1806 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1807 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1808 ceph_snap(rinode));
1809 } else if (rdentry) {
1810 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1811 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1812 *ppath);
1813 } else if (rpath || rino) {
1814 *ino = rino;
1815 *ppath = rpath;
1816 *pathlen = rpath ? strlen(rpath) : 0;
1817 dout(" path %.*s\n", *pathlen, rpath);
1818 }
1819
1820 return r;
1821 }
1822
1823 /*
1824 * called under mdsc->mutex
1825 */
1826 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1827 struct ceph_mds_request *req,
1828 int mds)
1829 {
1830 struct ceph_msg *msg;
1831 struct ceph_mds_request_head *head;
1832 const char *path1 = NULL;
1833 const char *path2 = NULL;
1834 u64 ino1 = 0, ino2 = 0;
1835 int pathlen1 = 0, pathlen2 = 0;
1836 int freepath1 = 0, freepath2 = 0;
1837 int len;
1838 u16 releases;
1839 void *p, *end;
1840 int ret;
1841
1842 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1843 req->r_path1, req->r_ino1.ino,
1844 &path1, &pathlen1, &ino1, &freepath1);
1845 if (ret < 0) {
1846 msg = ERR_PTR(ret);
1847 goto out;
1848 }
1849
1850 ret = set_request_path_attr(NULL, req->r_old_dentry,
1851 req->r_path2, req->r_ino2.ino,
1852 &path2, &pathlen2, &ino2, &freepath2);
1853 if (ret < 0) {
1854 msg = ERR_PTR(ret);
1855 goto out_free1;
1856 }
1857
1858 len = sizeof(*head) +
1859 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
1860 sizeof(struct ceph_timespec);
1861
1862 /* calculate (max) length for cap releases */
1863 len += sizeof(struct ceph_mds_request_release) *
1864 (!!req->r_inode_drop + !!req->r_dentry_drop +
1865 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1866 if (req->r_dentry_drop)
1867 len += req->r_dentry->d_name.len;
1868 if (req->r_old_dentry_drop)
1869 len += req->r_old_dentry->d_name.len;
1870
1871 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
1872 if (!msg) {
1873 msg = ERR_PTR(-ENOMEM);
1874 goto out_free2;
1875 }
1876
1877 msg->hdr.version = 2;
1878 msg->hdr.tid = cpu_to_le64(req->r_tid);
1879
1880 head = msg->front.iov_base;
1881 p = msg->front.iov_base + sizeof(*head);
1882 end = msg->front.iov_base + msg->front.iov_len;
1883
1884 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1885 head->op = cpu_to_le32(req->r_op);
1886 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1887 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
1888 head->args = req->r_args;
1889
1890 ceph_encode_filepath(&p, end, ino1, path1);
1891 ceph_encode_filepath(&p, end, ino2, path2);
1892
1893 /* make note of release offset, in case we need to replay */
1894 req->r_request_release_offset = p - msg->front.iov_base;
1895
1896 /* cap releases */
1897 releases = 0;
1898 if (req->r_inode_drop)
1899 releases += ceph_encode_inode_release(&p,
1900 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1901 mds, req->r_inode_drop, req->r_inode_unless, 0);
1902 if (req->r_dentry_drop)
1903 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1904 mds, req->r_dentry_drop, req->r_dentry_unless);
1905 if (req->r_old_dentry_drop)
1906 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1907 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1908 if (req->r_old_inode_drop)
1909 releases += ceph_encode_inode_release(&p,
1910 req->r_old_dentry->d_inode,
1911 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1912 head->num_releases = cpu_to_le16(releases);
1913
1914 /* time stamp */
1915 ceph_encode_copy(&p, &req->r_stamp, sizeof(req->r_stamp));
1916
1917 BUG_ON(p > end);
1918 msg->front.iov_len = p - msg->front.iov_base;
1919 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1920
1921 if (req->r_pagelist) {
1922 struct ceph_pagelist *pagelist = req->r_pagelist;
1923 atomic_inc(&pagelist->refcnt);
1924 ceph_msg_data_add_pagelist(msg, pagelist);
1925 msg->hdr.data_len = cpu_to_le32(pagelist->length);
1926 } else {
1927 msg->hdr.data_len = 0;
1928 }
1929
1930 msg->hdr.data_off = cpu_to_le16(0);
1931
1932 out_free2:
1933 if (freepath2)
1934 kfree((char *)path2);
1935 out_free1:
1936 if (freepath1)
1937 kfree((char *)path1);
1938 out:
1939 return msg;
1940 }
1941
1942 /*
1943 * called under mdsc->mutex if error, under no mutex if
1944 * success.
1945 */
1946 static void complete_request(struct ceph_mds_client *mdsc,
1947 struct ceph_mds_request *req)
1948 {
1949 if (req->r_callback)
1950 req->r_callback(mdsc, req);
1951 else
1952 complete_all(&req->r_completion);
1953 }
1954
1955 /*
1956 * called under mdsc->mutex
1957 */
1958 static int __prepare_send_request(struct ceph_mds_client *mdsc,
1959 struct ceph_mds_request *req,
1960 int mds)
1961 {
1962 struct ceph_mds_request_head *rhead;
1963 struct ceph_msg *msg;
1964 int flags = 0;
1965
1966 req->r_attempts++;
1967 if (req->r_inode) {
1968 struct ceph_cap *cap =
1969 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1970
1971 if (cap)
1972 req->r_sent_on_mseq = cap->mseq;
1973 else
1974 req->r_sent_on_mseq = -1;
1975 }
1976 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1977 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1978
1979 if (req->r_got_unsafe) {
1980 void *p;
1981 /*
1982 * Replay. Do not regenerate message (and rebuild
1983 * paths, etc.); just use the original message.
1984 * Rebuilding paths will break for renames because
1985 * d_move mangles the src name.
1986 */
1987 msg = req->r_request;
1988 rhead = msg->front.iov_base;
1989
1990 flags = le32_to_cpu(rhead->flags);
1991 flags |= CEPH_MDS_FLAG_REPLAY;
1992 rhead->flags = cpu_to_le32(flags);
1993
1994 if (req->r_target_inode)
1995 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1996
1997 rhead->num_retry = req->r_attempts - 1;
1998
1999 /* remove cap/dentry releases from message */
2000 rhead->num_releases = 0;
2001
2002 /* time stamp */
2003 p = msg->front.iov_base + req->r_request_release_offset;
2004 ceph_encode_copy(&p, &req->r_stamp, sizeof(req->r_stamp));
2005
2006 msg->front.iov_len = p - msg->front.iov_base;
2007 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2008 return 0;
2009 }
2010
2011 if (req->r_request) {
2012 ceph_msg_put(req->r_request);
2013 req->r_request = NULL;
2014 }
2015 msg = create_request_message(mdsc, req, mds);
2016 if (IS_ERR(msg)) {
2017 req->r_err = PTR_ERR(msg);
2018 complete_request(mdsc, req);
2019 return PTR_ERR(msg);
2020 }
2021 req->r_request = msg;
2022
2023 rhead = msg->front.iov_base;
2024 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2025 if (req->r_got_unsafe)
2026 flags |= CEPH_MDS_FLAG_REPLAY;
2027 if (req->r_locked_dir)
2028 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2029 rhead->flags = cpu_to_le32(flags);
2030 rhead->num_fwd = req->r_num_fwd;
2031 rhead->num_retry = req->r_attempts - 1;
2032 rhead->ino = 0;
2033
2034 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2035 return 0;
2036 }
2037
2038 /*
2039 * send request, or put it on the appropriate wait list.
2040 */
2041 static int __do_request(struct ceph_mds_client *mdsc,
2042 struct ceph_mds_request *req)
2043 {
2044 struct ceph_mds_session *session = NULL;
2045 int mds = -1;
2046 int err = -EAGAIN;
2047
2048 if (req->r_err || req->r_got_result) {
2049 if (req->r_aborted)
2050 __unregister_request(mdsc, req);
2051 goto out;
2052 }
2053
2054 if (req->r_timeout &&
2055 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2056 dout("do_request timed out\n");
2057 err = -EIO;
2058 goto finish;
2059 }
2060
2061 put_request_session(req);
2062
2063 mds = __choose_mds(mdsc, req);
2064 if (mds < 0 ||
2065 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2066 dout("do_request no mds or not active, waiting for map\n");
2067 list_add(&req->r_wait, &mdsc->waiting_for_map);
2068 goto out;
2069 }
2070
2071 /* get, open session */
2072 session = __ceph_lookup_mds_session(mdsc, mds);
2073 if (!session) {
2074 session = register_session(mdsc, mds);
2075 if (IS_ERR(session)) {
2076 err = PTR_ERR(session);
2077 goto finish;
2078 }
2079 }
2080 req->r_session = get_session(session);
2081
2082 dout("do_request mds%d session %p state %s\n", mds, session,
2083 ceph_session_state_name(session->s_state));
2084 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2085 session->s_state != CEPH_MDS_SESSION_HUNG) {
2086 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2087 session->s_state == CEPH_MDS_SESSION_CLOSING)
2088 __open_session(mdsc, session);
2089 list_add(&req->r_wait, &session->s_waiting);
2090 goto out_session;
2091 }
2092
2093 /* send request */
2094 req->r_resend_mds = -1; /* forget any previous mds hint */
2095
2096 if (req->r_request_started == 0) /* note request start time */
2097 req->r_request_started = jiffies;
2098
2099 err = __prepare_send_request(mdsc, req, mds);
2100 if (!err) {
2101 ceph_msg_get(req->r_request);
2102 ceph_con_send(&session->s_con, req->r_request);
2103 }
2104
2105 out_session:
2106 ceph_put_mds_session(session);
2107 out:
2108 return err;
2109
2110 finish:
2111 req->r_err = err;
2112 complete_request(mdsc, req);
2113 goto out;
2114 }
2115
2116 /*
2117 * called under mdsc->mutex
2118 */
2119 static void __wake_requests(struct ceph_mds_client *mdsc,
2120 struct list_head *head)
2121 {
2122 struct ceph_mds_request *req;
2123 LIST_HEAD(tmp_list);
2124
2125 list_splice_init(head, &tmp_list);
2126
2127 while (!list_empty(&tmp_list)) {
2128 req = list_entry(tmp_list.next,
2129 struct ceph_mds_request, r_wait);
2130 list_del_init(&req->r_wait);
2131 dout(" wake request %p tid %llu\n", req, req->r_tid);
2132 __do_request(mdsc, req);
2133 }
2134 }
2135
2136 /*
2137 * Wake up threads with requests pending for @mds, so that they can
2138 * resubmit their requests to a possibly different mds.
2139 */
2140 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2141 {
2142 struct ceph_mds_request *req;
2143 struct rb_node *p = rb_first(&mdsc->request_tree);
2144
2145 dout("kick_requests mds%d\n", mds);
2146 while (p) {
2147 req = rb_entry(p, struct ceph_mds_request, r_node);
2148 p = rb_next(p);
2149 if (req->r_got_unsafe)
2150 continue;
2151 if (req->r_session &&
2152 req->r_session->s_mds == mds) {
2153 dout(" kicking tid %llu\n", req->r_tid);
2154 list_del_init(&req->r_wait);
2155 __do_request(mdsc, req);
2156 }
2157 }
2158 }
2159
2160 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2161 struct ceph_mds_request *req)
2162 {
2163 dout("submit_request on %p\n", req);
2164 mutex_lock(&mdsc->mutex);
2165 __register_request(mdsc, req, NULL);
2166 __do_request(mdsc, req);
2167 mutex_unlock(&mdsc->mutex);
2168 }
2169
2170 /*
2171 * Synchrously perform an mds request. Take care of all of the
2172 * session setup, forwarding, retry details.
2173 */
2174 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2175 struct inode *dir,
2176 struct ceph_mds_request *req)
2177 {
2178 int err;
2179
2180 dout("do_request on %p\n", req);
2181
2182 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2183 if (req->r_inode)
2184 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2185 if (req->r_locked_dir)
2186 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
2187 if (req->r_old_dentry_dir)
2188 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2189 CEPH_CAP_PIN);
2190
2191 /* issue */
2192 mutex_lock(&mdsc->mutex);
2193 __register_request(mdsc, req, dir);
2194 __do_request(mdsc, req);
2195
2196 if (req->r_err) {
2197 err = req->r_err;
2198 __unregister_request(mdsc, req);
2199 dout("do_request early error %d\n", err);
2200 goto out;
2201 }
2202
2203 /* wait */
2204 mutex_unlock(&mdsc->mutex);
2205 dout("do_request waiting\n");
2206 if (req->r_timeout) {
2207 err = (long)wait_for_completion_killable_timeout(
2208 &req->r_completion, req->r_timeout);
2209 if (err == 0)
2210 err = -EIO;
2211 } else {
2212 err = wait_for_completion_killable(&req->r_completion);
2213 }
2214 dout("do_request waited, got %d\n", err);
2215 mutex_lock(&mdsc->mutex);
2216
2217 /* only abort if we didn't race with a real reply */
2218 if (req->r_got_result) {
2219 err = le32_to_cpu(req->r_reply_info.head->result);
2220 } else if (err < 0) {
2221 dout("aborted request %lld with %d\n", req->r_tid, err);
2222
2223 /*
2224 * ensure we aren't running concurrently with
2225 * ceph_fill_trace or ceph_readdir_prepopulate, which
2226 * rely on locks (dir mutex) held by our caller.
2227 */
2228 mutex_lock(&req->r_fill_mutex);
2229 req->r_err = err;
2230 req->r_aborted = true;
2231 mutex_unlock(&req->r_fill_mutex);
2232
2233 if (req->r_locked_dir &&
2234 (req->r_op & CEPH_MDS_OP_WRITE))
2235 ceph_invalidate_dir_request(req);
2236 } else {
2237 err = req->r_err;
2238 }
2239
2240 out:
2241 mutex_unlock(&mdsc->mutex);
2242 dout("do_request %p done, result %d\n", req, err);
2243 return err;
2244 }
2245
2246 /*
2247 * Invalidate dir's completeness, dentry lease state on an aborted MDS
2248 * namespace request.
2249 */
2250 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2251 {
2252 struct inode *inode = req->r_locked_dir;
2253
2254 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
2255
2256 ceph_dir_clear_complete(inode);
2257 if (req->r_dentry)
2258 ceph_invalidate_dentry_lease(req->r_dentry);
2259 if (req->r_old_dentry)
2260 ceph_invalidate_dentry_lease(req->r_old_dentry);
2261 }
2262
2263 /*
2264 * Handle mds reply.
2265 *
2266 * We take the session mutex and parse and process the reply immediately.
2267 * This preserves the logical ordering of replies, capabilities, etc., sent
2268 * by the MDS as they are applied to our local cache.
2269 */
2270 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2271 {
2272 struct ceph_mds_client *mdsc = session->s_mdsc;
2273 struct ceph_mds_request *req;
2274 struct ceph_mds_reply_head *head = msg->front.iov_base;
2275 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
2276 u64 tid;
2277 int err, result;
2278 int mds = session->s_mds;
2279
2280 if (msg->front.iov_len < sizeof(*head)) {
2281 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2282 ceph_msg_dump(msg);
2283 return;
2284 }
2285
2286 /* get request, session */
2287 tid = le64_to_cpu(msg->hdr.tid);
2288 mutex_lock(&mdsc->mutex);
2289 req = __lookup_request(mdsc, tid);
2290 if (!req) {
2291 dout("handle_reply on unknown tid %llu\n", tid);
2292 mutex_unlock(&mdsc->mutex);
2293 return;
2294 }
2295 dout("handle_reply %p\n", req);
2296
2297 /* correct session? */
2298 if (req->r_session != session) {
2299 pr_err("mdsc_handle_reply got %llu on session mds%d"
2300 " not mds%d\n", tid, session->s_mds,
2301 req->r_session ? req->r_session->s_mds : -1);
2302 mutex_unlock(&mdsc->mutex);
2303 goto out;
2304 }
2305
2306 /* dup? */
2307 if ((req->r_got_unsafe && !head->safe) ||
2308 (req->r_got_safe && head->safe)) {
2309 pr_warn("got a dup %s reply on %llu from mds%d\n",
2310 head->safe ? "safe" : "unsafe", tid, mds);
2311 mutex_unlock(&mdsc->mutex);
2312 goto out;
2313 }
2314 if (req->r_got_safe && !head->safe) {
2315 pr_warn("got unsafe after safe on %llu from mds%d\n",
2316 tid, mds);
2317 mutex_unlock(&mdsc->mutex);
2318 goto out;
2319 }
2320
2321 result = le32_to_cpu(head->result);
2322
2323 /*
2324 * Handle an ESTALE
2325 * if we're not talking to the authority, send to them
2326 * if the authority has changed while we weren't looking,
2327 * send to new authority
2328 * Otherwise we just have to return an ESTALE
2329 */
2330 if (result == -ESTALE) {
2331 dout("got ESTALE on request %llu", req->r_tid);
2332 req->r_resend_mds = -1;
2333 if (req->r_direct_mode != USE_AUTH_MDS) {
2334 dout("not using auth, setting for that now");
2335 req->r_direct_mode = USE_AUTH_MDS;
2336 __do_request(mdsc, req);
2337 mutex_unlock(&mdsc->mutex);
2338 goto out;
2339 } else {
2340 int mds = __choose_mds(mdsc, req);
2341 if (mds >= 0 && mds != req->r_session->s_mds) {
2342 dout("but auth changed, so resending");
2343 __do_request(mdsc, req);
2344 mutex_unlock(&mdsc->mutex);
2345 goto out;
2346 }
2347 }
2348 dout("have to return ESTALE on request %llu", req->r_tid);
2349 }
2350
2351
2352 if (head->safe) {
2353 req->r_got_safe = true;
2354 __unregister_request(mdsc, req);
2355
2356 if (req->r_got_unsafe) {
2357 /*
2358 * We already handled the unsafe response, now do the
2359 * cleanup. No need to examine the response; the MDS
2360 * doesn't include any result info in the safe
2361 * response. And even if it did, there is nothing
2362 * useful we could do with a revised return value.
2363 */
2364 dout("got safe reply %llu, mds%d\n", tid, mds);
2365 list_del_init(&req->r_unsafe_item);
2366
2367 /* last unsafe request during umount? */
2368 if (mdsc->stopping && !__get_oldest_req(mdsc))
2369 complete_all(&mdsc->safe_umount_waiters);
2370 mutex_unlock(&mdsc->mutex);
2371 goto out;
2372 }
2373 } else {
2374 req->r_got_unsafe = true;
2375 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2376 }
2377
2378 dout("handle_reply tid %lld result %d\n", tid, result);
2379 rinfo = &req->r_reply_info;
2380 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2381 mutex_unlock(&mdsc->mutex);
2382
2383 mutex_lock(&session->s_mutex);
2384 if (err < 0) {
2385 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2386 ceph_msg_dump(msg);
2387 goto out_err;
2388 }
2389
2390 /* snap trace */
2391 if (rinfo->snapblob_len) {
2392 down_write(&mdsc->snap_rwsem);
2393 ceph_update_snap_trace(mdsc, rinfo->snapblob,
2394 rinfo->snapblob + rinfo->snapblob_len,
2395 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
2396 downgrade_write(&mdsc->snap_rwsem);
2397 } else {
2398 down_read(&mdsc->snap_rwsem);
2399 }
2400
2401 /* insert trace into our cache */
2402 mutex_lock(&req->r_fill_mutex);
2403 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2404 if (err == 0) {
2405 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2406 req->r_op == CEPH_MDS_OP_LSSNAP))
2407 ceph_readdir_prepopulate(req, req->r_session);
2408 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2409 }
2410 mutex_unlock(&req->r_fill_mutex);
2411
2412 up_read(&mdsc->snap_rwsem);
2413 out_err:
2414 mutex_lock(&mdsc->mutex);
2415 if (!req->r_aborted) {
2416 if (err) {
2417 req->r_err = err;
2418 } else {
2419 req->r_reply = msg;
2420 ceph_msg_get(msg);
2421 req->r_got_result = true;
2422 }
2423 } else {
2424 dout("reply arrived after request %lld was aborted\n", tid);
2425 }
2426 mutex_unlock(&mdsc->mutex);
2427
2428 ceph_add_cap_releases(mdsc, req->r_session);
2429 mutex_unlock(&session->s_mutex);
2430
2431 /* kick calling process */
2432 complete_request(mdsc, req);
2433 out:
2434 ceph_mdsc_put_request(req);
2435 return;
2436 }
2437
2438
2439
2440 /*
2441 * handle mds notification that our request has been forwarded.
2442 */
2443 static void handle_forward(struct ceph_mds_client *mdsc,
2444 struct ceph_mds_session *session,
2445 struct ceph_msg *msg)
2446 {
2447 struct ceph_mds_request *req;
2448 u64 tid = le64_to_cpu(msg->hdr.tid);
2449 u32 next_mds;
2450 u32 fwd_seq;
2451 int err = -EINVAL;
2452 void *p = msg->front.iov_base;
2453 void *end = p + msg->front.iov_len;
2454
2455 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2456 next_mds = ceph_decode_32(&p);
2457 fwd_seq = ceph_decode_32(&p);
2458
2459 mutex_lock(&mdsc->mutex);
2460 req = __lookup_request(mdsc, tid);
2461 if (!req) {
2462 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2463 goto out; /* dup reply? */
2464 }
2465
2466 if (req->r_aborted) {
2467 dout("forward tid %llu aborted, unregistering\n", tid);
2468 __unregister_request(mdsc, req);
2469 } else if (fwd_seq <= req->r_num_fwd) {
2470 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2471 tid, next_mds, req->r_num_fwd, fwd_seq);
2472 } else {
2473 /* resend. forward race not possible; mds would drop */
2474 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2475 BUG_ON(req->r_err);
2476 BUG_ON(req->r_got_result);
2477 req->r_num_fwd = fwd_seq;
2478 req->r_resend_mds = next_mds;
2479 put_request_session(req);
2480 __do_request(mdsc, req);
2481 }
2482 ceph_mdsc_put_request(req);
2483 out:
2484 mutex_unlock(&mdsc->mutex);
2485 return;
2486
2487 bad:
2488 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2489 }
2490
2491 /*
2492 * handle a mds session control message
2493 */
2494 static void handle_session(struct ceph_mds_session *session,
2495 struct ceph_msg *msg)
2496 {
2497 struct ceph_mds_client *mdsc = session->s_mdsc;
2498 u32 op;
2499 u64 seq;
2500 int mds = session->s_mds;
2501 struct ceph_mds_session_head *h = msg->front.iov_base;
2502 int wake = 0;
2503
2504 /* decode */
2505 if (msg->front.iov_len != sizeof(*h))
2506 goto bad;
2507 op = le32_to_cpu(h->op);
2508 seq = le64_to_cpu(h->seq);
2509
2510 mutex_lock(&mdsc->mutex);
2511 if (op == CEPH_SESSION_CLOSE)
2512 __unregister_session(mdsc, session);
2513 /* FIXME: this ttl calculation is generous */
2514 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2515 mutex_unlock(&mdsc->mutex);
2516
2517 mutex_lock(&session->s_mutex);
2518
2519 dout("handle_session mds%d %s %p state %s seq %llu\n",
2520 mds, ceph_session_op_name(op), session,
2521 ceph_session_state_name(session->s_state), seq);
2522
2523 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2524 session->s_state = CEPH_MDS_SESSION_OPEN;
2525 pr_info("mds%d came back\n", session->s_mds);
2526 }
2527
2528 switch (op) {
2529 case CEPH_SESSION_OPEN:
2530 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2531 pr_info("mds%d reconnect success\n", session->s_mds);
2532 session->s_state = CEPH_MDS_SESSION_OPEN;
2533 renewed_caps(mdsc, session, 0);
2534 wake = 1;
2535 if (mdsc->stopping)
2536 __close_session(mdsc, session);
2537 break;
2538
2539 case CEPH_SESSION_RENEWCAPS:
2540 if (session->s_renew_seq == seq)
2541 renewed_caps(mdsc, session, 1);
2542 break;
2543
2544 case CEPH_SESSION_CLOSE:
2545 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2546 pr_info("mds%d reconnect denied\n", session->s_mds);
2547 remove_session_caps(session);
2548 wake = 2; /* for good measure */
2549 wake_up_all(&mdsc->session_close_wq);
2550 break;
2551
2552 case CEPH_SESSION_STALE:
2553 pr_info("mds%d caps went stale, renewing\n",
2554 session->s_mds);
2555 spin_lock(&session->s_gen_ttl_lock);
2556 session->s_cap_gen++;
2557 session->s_cap_ttl = jiffies - 1;
2558 spin_unlock(&session->s_gen_ttl_lock);
2559 send_renew_caps(mdsc, session);
2560 break;
2561
2562 case CEPH_SESSION_RECALL_STATE:
2563 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2564 break;
2565
2566 case CEPH_SESSION_FLUSHMSG:
2567 send_flushmsg_ack(mdsc, session, seq);
2568 break;
2569
2570 default:
2571 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2572 WARN_ON(1);
2573 }
2574
2575 mutex_unlock(&session->s_mutex);
2576 if (wake) {
2577 mutex_lock(&mdsc->mutex);
2578 __wake_requests(mdsc, &session->s_waiting);
2579 if (wake == 2)
2580 kick_requests(mdsc, mds);
2581 mutex_unlock(&mdsc->mutex);
2582 }
2583 return;
2584
2585 bad:
2586 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2587 (int)msg->front.iov_len);
2588 ceph_msg_dump(msg);
2589 return;
2590 }
2591
2592
2593 /*
2594 * called under session->mutex.
2595 */
2596 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2597 struct ceph_mds_session *session)
2598 {
2599 struct ceph_mds_request *req, *nreq;
2600 int err;
2601
2602 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2603
2604 mutex_lock(&mdsc->mutex);
2605 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2606 err = __prepare_send_request(mdsc, req, session->s_mds);
2607 if (!err) {
2608 ceph_msg_get(req->r_request);
2609 ceph_con_send(&session->s_con, req->r_request);
2610 }
2611 }
2612 mutex_unlock(&mdsc->mutex);
2613 }
2614
2615 /*
2616 * Encode information about a cap for a reconnect with the MDS.
2617 */
2618 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2619 void *arg)
2620 {
2621 union {
2622 struct ceph_mds_cap_reconnect v2;
2623 struct ceph_mds_cap_reconnect_v1 v1;
2624 } rec;
2625 size_t reclen;
2626 struct ceph_inode_info *ci;
2627 struct ceph_reconnect_state *recon_state = arg;
2628 struct ceph_pagelist *pagelist = recon_state->pagelist;
2629 char *path;
2630 int pathlen, err;
2631 u64 pathbase;
2632 struct dentry *dentry;
2633
2634 ci = cap->ci;
2635
2636 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2637 inode, ceph_vinop(inode), cap, cap->cap_id,
2638 ceph_cap_string(cap->issued));
2639 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2640 if (err)
2641 return err;
2642
2643 dentry = d_find_alias(inode);
2644 if (dentry) {
2645 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2646 if (IS_ERR(path)) {
2647 err = PTR_ERR(path);
2648 goto out_dput;
2649 }
2650 } else {
2651 path = NULL;
2652 pathlen = 0;
2653 }
2654 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2655 if (err)
2656 goto out_free;
2657
2658 spin_lock(&ci->i_ceph_lock);
2659 cap->seq = 0; /* reset cap seq */
2660 cap->issue_seq = 0; /* and issue_seq */
2661 cap->mseq = 0; /* and migrate_seq */
2662 cap->cap_gen = cap->session->s_cap_gen;
2663
2664 if (recon_state->flock) {
2665 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2666 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2667 rec.v2.issued = cpu_to_le32(cap->issued);
2668 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2669 rec.v2.pathbase = cpu_to_le64(pathbase);
2670 rec.v2.flock_len = 0;
2671 reclen = sizeof(rec.v2);
2672 } else {
2673 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2674 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2675 rec.v1.issued = cpu_to_le32(cap->issued);
2676 rec.v1.size = cpu_to_le64(inode->i_size);
2677 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2678 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2679 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2680 rec.v1.pathbase = cpu_to_le64(pathbase);
2681 reclen = sizeof(rec.v1);
2682 }
2683 spin_unlock(&ci->i_ceph_lock);
2684
2685 if (recon_state->flock) {
2686 int num_fcntl_locks, num_flock_locks;
2687 struct ceph_filelock *flocks;
2688
2689 encode_again:
2690 spin_lock(&inode->i_lock);
2691 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2692 spin_unlock(&inode->i_lock);
2693 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2694 sizeof(struct ceph_filelock), GFP_NOFS);
2695 if (!flocks) {
2696 err = -ENOMEM;
2697 goto out_free;
2698 }
2699 spin_lock(&inode->i_lock);
2700 err = ceph_encode_locks_to_buffer(inode, flocks,
2701 num_fcntl_locks,
2702 num_flock_locks);
2703 spin_unlock(&inode->i_lock);
2704 if (err) {
2705 kfree(flocks);
2706 if (err == -ENOSPC)
2707 goto encode_again;
2708 goto out_free;
2709 }
2710 /*
2711 * number of encoded locks is stable, so copy to pagelist
2712 */
2713 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2714 (num_fcntl_locks+num_flock_locks) *
2715 sizeof(struct ceph_filelock));
2716 err = ceph_pagelist_append(pagelist, &rec, reclen);
2717 if (!err)
2718 err = ceph_locks_to_pagelist(flocks, pagelist,
2719 num_fcntl_locks,
2720 num_flock_locks);
2721 kfree(flocks);
2722 } else {
2723 err = ceph_pagelist_append(pagelist, &rec, reclen);
2724 }
2725
2726 recon_state->nr_caps++;
2727 out_free:
2728 kfree(path);
2729 out_dput:
2730 dput(dentry);
2731 return err;
2732 }
2733
2734
2735 /*
2736 * If an MDS fails and recovers, clients need to reconnect in order to
2737 * reestablish shared state. This includes all caps issued through
2738 * this session _and_ the snap_realm hierarchy. Because it's not
2739 * clear which snap realms the mds cares about, we send everything we
2740 * know about.. that ensures we'll then get any new info the
2741 * recovering MDS might have.
2742 *
2743 * This is a relatively heavyweight operation, but it's rare.
2744 *
2745 * called with mdsc->mutex held.
2746 */
2747 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2748 struct ceph_mds_session *session)
2749 {
2750 struct ceph_msg *reply;
2751 struct rb_node *p;
2752 int mds = session->s_mds;
2753 int err = -ENOMEM;
2754 int s_nr_caps;
2755 struct ceph_pagelist *pagelist;
2756 struct ceph_reconnect_state recon_state;
2757
2758 pr_info("mds%d reconnect start\n", mds);
2759
2760 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2761 if (!pagelist)
2762 goto fail_nopagelist;
2763 ceph_pagelist_init(pagelist);
2764
2765 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
2766 if (!reply)
2767 goto fail_nomsg;
2768
2769 mutex_lock(&session->s_mutex);
2770 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2771 session->s_seq = 0;
2772
2773 dout("session %p state %s\n", session,
2774 ceph_session_state_name(session->s_state));
2775
2776 spin_lock(&session->s_gen_ttl_lock);
2777 session->s_cap_gen++;
2778 spin_unlock(&session->s_gen_ttl_lock);
2779
2780 spin_lock(&session->s_cap_lock);
2781 /*
2782 * notify __ceph_remove_cap() that we are composing cap reconnect.
2783 * If a cap get released before being added to the cap reconnect,
2784 * __ceph_remove_cap() should skip queuing cap release.
2785 */
2786 session->s_cap_reconnect = 1;
2787 /* drop old cap expires; we're about to reestablish that state */
2788 discard_cap_releases(mdsc, session);
2789 spin_unlock(&session->s_cap_lock);
2790
2791 /* trim unused caps to reduce MDS's cache rejoin time */
2792 shrink_dcache_parent(mdsc->fsc->sb->s_root);
2793
2794 ceph_con_close(&session->s_con);
2795 ceph_con_open(&session->s_con,
2796 CEPH_ENTITY_TYPE_MDS, mds,
2797 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2798
2799 /* replay unsafe requests */
2800 replay_unsafe_requests(mdsc, session);
2801
2802 down_read(&mdsc->snap_rwsem);
2803
2804 /* traverse this session's caps */
2805 s_nr_caps = session->s_nr_caps;
2806 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
2807 if (err)
2808 goto fail;
2809
2810 recon_state.nr_caps = 0;
2811 recon_state.pagelist = pagelist;
2812 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2813 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2814 if (err < 0)
2815 goto fail;
2816
2817 spin_lock(&session->s_cap_lock);
2818 session->s_cap_reconnect = 0;
2819 spin_unlock(&session->s_cap_lock);
2820
2821 /*
2822 * snaprealms. we provide mds with the ino, seq (version), and
2823 * parent for all of our realms. If the mds has any newer info,
2824 * it will tell us.
2825 */
2826 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2827 struct ceph_snap_realm *realm =
2828 rb_entry(p, struct ceph_snap_realm, node);
2829 struct ceph_mds_snaprealm_reconnect sr_rec;
2830
2831 dout(" adding snap realm %llx seq %lld parent %llx\n",
2832 realm->ino, realm->seq, realm->parent_ino);
2833 sr_rec.ino = cpu_to_le64(realm->ino);
2834 sr_rec.seq = cpu_to_le64(realm->seq);
2835 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2836 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2837 if (err)
2838 goto fail;
2839 }
2840
2841 if (recon_state.flock)
2842 reply->hdr.version = cpu_to_le16(2);
2843
2844 /* raced with cap release? */
2845 if (s_nr_caps != recon_state.nr_caps) {
2846 struct page *page = list_first_entry(&pagelist->head,
2847 struct page, lru);
2848 __le32 *addr = kmap_atomic(page);
2849 *addr = cpu_to_le32(recon_state.nr_caps);
2850 kunmap_atomic(addr);
2851 }
2852
2853 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2854 ceph_msg_data_add_pagelist(reply, pagelist);
2855 ceph_con_send(&session->s_con, reply);
2856
2857 mutex_unlock(&session->s_mutex);
2858
2859 mutex_lock(&mdsc->mutex);
2860 __wake_requests(mdsc, &session->s_waiting);
2861 mutex_unlock(&mdsc->mutex);
2862
2863 up_read(&mdsc->snap_rwsem);
2864 return;
2865
2866 fail:
2867 ceph_msg_put(reply);
2868 up_read(&mdsc->snap_rwsem);
2869 mutex_unlock(&session->s_mutex);
2870 fail_nomsg:
2871 ceph_pagelist_release(pagelist);
2872 fail_nopagelist:
2873 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
2874 return;
2875 }
2876
2877
2878 /*
2879 * compare old and new mdsmaps, kicking requests
2880 * and closing out old connections as necessary
2881 *
2882 * called under mdsc->mutex.
2883 */
2884 static void check_new_map(struct ceph_mds_client *mdsc,
2885 struct ceph_mdsmap *newmap,
2886 struct ceph_mdsmap *oldmap)
2887 {
2888 int i;
2889 int oldstate, newstate;
2890 struct ceph_mds_session *s;
2891
2892 dout("check_new_map new %u old %u\n",
2893 newmap->m_epoch, oldmap->m_epoch);
2894
2895 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2896 if (mdsc->sessions[i] == NULL)
2897 continue;
2898 s = mdsc->sessions[i];
2899 oldstate = ceph_mdsmap_get_state(oldmap, i);
2900 newstate = ceph_mdsmap_get_state(newmap, i);
2901
2902 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2903 i, ceph_mds_state_name(oldstate),
2904 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2905 ceph_mds_state_name(newstate),
2906 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
2907 ceph_session_state_name(s->s_state));
2908
2909 if (i >= newmap->m_max_mds ||
2910 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2911 ceph_mdsmap_get_addr(newmap, i),
2912 sizeof(struct ceph_entity_addr))) {
2913 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2914 /* the session never opened, just close it
2915 * out now */
2916 __wake_requests(mdsc, &s->s_waiting);
2917 __unregister_session(mdsc, s);
2918 } else {
2919 /* just close it */
2920 mutex_unlock(&mdsc->mutex);
2921 mutex_lock(&s->s_mutex);
2922 mutex_lock(&mdsc->mutex);
2923 ceph_con_close(&s->s_con);
2924 mutex_unlock(&s->s_mutex);
2925 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2926 }
2927
2928 /* kick any requests waiting on the recovering mds */
2929 kick_requests(mdsc, i);
2930 } else if (oldstate == newstate) {
2931 continue; /* nothing new with this mds */
2932 }
2933
2934 /*
2935 * send reconnect?
2936 */
2937 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
2938 newstate >= CEPH_MDS_STATE_RECONNECT) {
2939 mutex_unlock(&mdsc->mutex);
2940 send_mds_reconnect(mdsc, s);
2941 mutex_lock(&mdsc->mutex);
2942 }
2943
2944 /*
2945 * kick request on any mds that has gone active.
2946 */
2947 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2948 newstate >= CEPH_MDS_STATE_ACTIVE) {
2949 if (oldstate != CEPH_MDS_STATE_CREATING &&
2950 oldstate != CEPH_MDS_STATE_STARTING)
2951 pr_info("mds%d recovery completed\n", s->s_mds);
2952 kick_requests(mdsc, i);
2953 ceph_kick_flushing_caps(mdsc, s);
2954 wake_up_session_caps(s, 1);
2955 }
2956 }
2957
2958 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
2959 s = mdsc->sessions[i];
2960 if (!s)
2961 continue;
2962 if (!ceph_mdsmap_is_laggy(newmap, i))
2963 continue;
2964 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2965 s->s_state == CEPH_MDS_SESSION_HUNG ||
2966 s->s_state == CEPH_MDS_SESSION_CLOSING) {
2967 dout(" connecting to export targets of laggy mds%d\n",
2968 i);
2969 __open_export_target_sessions(mdsc, s);
2970 }
2971 }
2972 }
2973
2974
2975
2976 /*
2977 * leases
2978 */
2979
2980 /*
2981 * caller must hold session s_mutex, dentry->d_lock
2982 */
2983 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2984 {
2985 struct ceph_dentry_info *di = ceph_dentry(dentry);
2986
2987 ceph_put_mds_session(di->lease_session);
2988 di->lease_session = NULL;
2989 }
2990
2991 static void handle_lease(struct ceph_mds_client *mdsc,
2992 struct ceph_mds_session *session,
2993 struct ceph_msg *msg)
2994 {
2995 struct super_block *sb = mdsc->fsc->sb;
2996 struct inode *inode;
2997 struct dentry *parent, *dentry;
2998 struct ceph_dentry_info *di;
2999 int mds = session->s_mds;
3000 struct ceph_mds_lease *h = msg->front.iov_base;
3001 u32 seq;
3002 struct ceph_vino vino;
3003 struct qstr dname;
3004 int release = 0;
3005
3006 dout("handle_lease from mds%d\n", mds);
3007
3008 /* decode */
3009 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3010 goto bad;
3011 vino.ino = le64_to_cpu(h->ino);
3012 vino.snap = CEPH_NOSNAP;
3013 seq = le32_to_cpu(h->seq);
3014 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3015 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3016 if (dname.len != get_unaligned_le32(h+1))
3017 goto bad;
3018
3019 /* lookup inode */
3020 inode = ceph_find_inode(sb, vino);
3021 dout("handle_lease %s, ino %llx %p %.*s\n",
3022 ceph_lease_op_name(h->action), vino.ino, inode,
3023 dname.len, dname.name);
3024
3025 mutex_lock(&session->s_mutex);
3026 session->s_seq++;
3027
3028 if (inode == NULL) {
3029 dout("handle_lease no inode %llx\n", vino.ino);
3030 goto release;
3031 }
3032
3033 /* dentry */
3034 parent = d_find_alias(inode);
3035 if (!parent) {
3036 dout("no parent dentry on inode %p\n", inode);
3037 WARN_ON(1);
3038 goto release; /* hrm... */
3039 }
3040 dname.hash = full_name_hash(dname.name, dname.len);
3041 dentry = d_lookup(parent, &dname);
3042 dput(parent);
3043 if (!dentry)
3044 goto release;
3045
3046 spin_lock(&dentry->d_lock);
3047 di = ceph_dentry(dentry);
3048 switch (h->action) {
3049 case CEPH_MDS_LEASE_REVOKE:
3050 if (di->lease_session == session) {
3051 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3052 h->seq = cpu_to_le32(di->lease_seq);
3053 __ceph_mdsc_drop_dentry_lease(dentry);
3054 }
3055 release = 1;
3056 break;
3057
3058 case CEPH_MDS_LEASE_RENEW:
3059 if (di->lease_session == session &&
3060 di->lease_gen == session->s_cap_gen &&
3061 di->lease_renew_from &&
3062 di->lease_renew_after == 0) {
3063 unsigned long duration =
3064 le32_to_cpu(h->duration_ms) * HZ / 1000;
3065
3066 di->lease_seq = seq;
3067 dentry->d_time = di->lease_renew_from + duration;
3068 di->lease_renew_after = di->lease_renew_from +
3069 (duration >> 1);
3070 di->lease_renew_from = 0;
3071 }
3072 break;
3073 }
3074 spin_unlock(&dentry->d_lock);
3075 dput(dentry);
3076
3077 if (!release)
3078 goto out;
3079
3080 release:
3081 /* let's just reuse the same message */
3082 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3083 ceph_msg_get(msg);
3084 ceph_con_send(&session->s_con, msg);
3085
3086 out:
3087 iput(inode);
3088 mutex_unlock(&session->s_mutex);
3089 return;
3090
3091 bad:
3092 pr_err("corrupt lease message\n");
3093 ceph_msg_dump(msg);
3094 }
3095
3096 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3097 struct inode *inode,
3098 struct dentry *dentry, char action,
3099 u32 seq)
3100 {
3101 struct ceph_msg *msg;
3102 struct ceph_mds_lease *lease;
3103 int len = sizeof(*lease) + sizeof(u32);
3104 int dnamelen = 0;
3105
3106 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3107 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3108 dnamelen = dentry->d_name.len;
3109 len += dnamelen;
3110
3111 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
3112 if (!msg)
3113 return;
3114 lease = msg->front.iov_base;
3115 lease->action = action;
3116 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3117 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3118 lease->seq = cpu_to_le32(seq);
3119 put_unaligned_le32(dnamelen, lease + 1);
3120 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3121
3122 /*
3123 * if this is a preemptive lease RELEASE, no need to
3124 * flush request stream, since the actual request will
3125 * soon follow.
3126 */
3127 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3128
3129 ceph_con_send(&session->s_con, msg);
3130 }
3131
3132 /*
3133 * Preemptively release a lease we expect to invalidate anyway.
3134 * Pass @inode always, @dentry is optional.
3135 */
3136 void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
3137 struct dentry *dentry)
3138 {
3139 struct ceph_dentry_info *di;
3140 struct ceph_mds_session *session;
3141 u32 seq;
3142
3143 BUG_ON(inode == NULL);
3144 BUG_ON(dentry == NULL);
3145
3146 /* is dentry lease valid? */
3147 spin_lock(&dentry->d_lock);
3148 di = ceph_dentry(dentry);
3149 if (!di || !di->lease_session ||
3150 di->lease_session->s_mds < 0 ||
3151 di->lease_gen != di->lease_session->s_cap_gen ||
3152 !time_before(jiffies, dentry->d_time)) {
3153 dout("lease_release inode %p dentry %p -- "
3154 "no lease\n",
3155 inode, dentry);
3156 spin_unlock(&dentry->d_lock);
3157 return;
3158 }
3159
3160 /* we do have a lease on this dentry; note mds and seq */
3161 session = ceph_get_mds_session(di->lease_session);
3162 seq = di->lease_seq;
3163 __ceph_mdsc_drop_dentry_lease(dentry);
3164 spin_unlock(&dentry->d_lock);
3165
3166 dout("lease_release inode %p dentry %p to mds%d\n",
3167 inode, dentry, session->s_mds);
3168 ceph_mdsc_lease_send_msg(session, inode, dentry,
3169 CEPH_MDS_LEASE_RELEASE, seq);
3170 ceph_put_mds_session(session);
3171 }
3172
3173 /*
3174 * drop all leases (and dentry refs) in preparation for umount
3175 */
3176 static void drop_leases(struct ceph_mds_client *mdsc)
3177 {
3178 int i;
3179
3180 dout("drop_leases\n");
3181 mutex_lock(&mdsc->mutex);
3182 for (i = 0; i < mdsc->max_sessions; i++) {
3183 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3184 if (!s)
3185 continue;
3186 mutex_unlock(&mdsc->mutex);
3187 mutex_lock(&s->s_mutex);
3188 mutex_unlock(&s->s_mutex);
3189 ceph_put_mds_session(s);
3190 mutex_lock(&mdsc->mutex);
3191 }
3192 mutex_unlock(&mdsc->mutex);
3193 }
3194
3195
3196
3197 /*
3198 * delayed work -- periodically trim expired leases, renew caps with mds
3199 */
3200 static void schedule_delayed(struct ceph_mds_client *mdsc)
3201 {
3202 int delay = 5;
3203 unsigned hz = round_jiffies_relative(HZ * delay);
3204 schedule_delayed_work(&mdsc->delayed_work, hz);
3205 }
3206
3207 static void delayed_work(struct work_struct *work)
3208 {
3209 int i;
3210 struct ceph_mds_client *mdsc =
3211 container_of(work, struct ceph_mds_client, delayed_work.work);
3212 int renew_interval;
3213 int renew_caps;
3214
3215 dout("mdsc delayed_work\n");
3216 ceph_check_delayed_caps(mdsc);
3217
3218 mutex_lock(&mdsc->mutex);
3219 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3220 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3221 mdsc->last_renew_caps);
3222 if (renew_caps)
3223 mdsc->last_renew_caps = jiffies;
3224
3225 for (i = 0; i < mdsc->max_sessions; i++) {
3226 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3227 if (s == NULL)
3228 continue;
3229 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3230 dout("resending session close request for mds%d\n",
3231 s->s_mds);
3232 request_close_session(mdsc, s);
3233 ceph_put_mds_session(s);
3234 continue;
3235 }
3236 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3237 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3238 s->s_state = CEPH_MDS_SESSION_HUNG;
3239 pr_info("mds%d hung\n", s->s_mds);
3240 }
3241 }
3242 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3243 /* this mds is failed or recovering, just wait */
3244 ceph_put_mds_session(s);
3245 continue;
3246 }
3247 mutex_unlock(&mdsc->mutex);
3248
3249 mutex_lock(&s->s_mutex);
3250 if (renew_caps)
3251 send_renew_caps(mdsc, s);
3252 else
3253 ceph_con_keepalive(&s->s_con);
3254 ceph_add_cap_releases(mdsc, s);
3255 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3256 s->s_state == CEPH_MDS_SESSION_HUNG)
3257 ceph_send_cap_releases(mdsc, s);
3258 mutex_unlock(&s->s_mutex);
3259 ceph_put_mds_session(s);
3260
3261 mutex_lock(&mdsc->mutex);
3262 }
3263 mutex_unlock(&mdsc->mutex);
3264
3265 schedule_delayed(mdsc);
3266 }
3267
3268 int ceph_mdsc_init(struct ceph_fs_client *fsc)
3269
3270 {
3271 struct ceph_mds_client *mdsc;
3272
3273 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3274 if (!mdsc)
3275 return -ENOMEM;
3276 mdsc->fsc = fsc;
3277 fsc->mdsc = mdsc;
3278 mutex_init(&mdsc->mutex);
3279 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3280 if (mdsc->mdsmap == NULL) {
3281 kfree(mdsc);
3282 return -ENOMEM;
3283 }
3284
3285 init_completion(&mdsc->safe_umount_waiters);
3286 init_waitqueue_head(&mdsc->session_close_wq);
3287 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3288 mdsc->sessions = NULL;
3289 mdsc->max_sessions = 0;
3290 mdsc->stopping = 0;
3291 init_rwsem(&mdsc->snap_rwsem);
3292 mdsc->snap_realms = RB_ROOT;
3293 INIT_LIST_HEAD(&mdsc->snap_empty);
3294 spin_lock_init(&mdsc->snap_empty_lock);
3295 mdsc->last_tid = 0;
3296 mdsc->request_tree = RB_ROOT;
3297 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3298 mdsc->last_renew_caps = jiffies;
3299 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3300 spin_lock_init(&mdsc->cap_delay_lock);
3301 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3302 spin_lock_init(&mdsc->snap_flush_lock);
3303 mdsc->cap_flush_seq = 0;
3304 INIT_LIST_HEAD(&mdsc->cap_dirty);
3305 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3306 mdsc->num_cap_flushing = 0;
3307 spin_lock_init(&mdsc->cap_dirty_lock);
3308 init_waitqueue_head(&mdsc->cap_flushing_wq);
3309 spin_lock_init(&mdsc->dentry_lru_lock);
3310 INIT_LIST_HEAD(&mdsc->dentry_lru);
3311
3312 ceph_caps_init(mdsc);
3313 ceph_adjust_min_caps(mdsc, fsc->min_caps);
3314
3315 return 0;
3316 }
3317
3318 /*
3319 * Wait for safe replies on open mds requests. If we time out, drop
3320 * all requests from the tree to avoid dangling dentry refs.
3321 */
3322 static void wait_requests(struct ceph_mds_client *mdsc)
3323 {
3324 struct ceph_mds_request *req;
3325 struct ceph_fs_client *fsc = mdsc->fsc;
3326
3327 mutex_lock(&mdsc->mutex);
3328 if (__get_oldest_req(mdsc)) {
3329 mutex_unlock(&mdsc->mutex);
3330
3331 dout("wait_requests waiting for requests\n");
3332 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3333 fsc->client->options->mount_timeout * HZ);
3334
3335 /* tear down remaining requests */
3336 mutex_lock(&mdsc->mutex);
3337 while ((req = __get_oldest_req(mdsc))) {
3338 dout("wait_requests timed out on tid %llu\n",
3339 req->r_tid);
3340 __unregister_request(mdsc, req);
3341 }
3342 }
3343 mutex_unlock(&mdsc->mutex);
3344 dout("wait_requests done\n");
3345 }
3346
3347 /*
3348 * called before mount is ro, and before dentries are torn down.
3349 * (hmm, does this still race with new lookups?)
3350 */
3351 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3352 {
3353 dout("pre_umount\n");
3354 mdsc->stopping = 1;
3355
3356 drop_leases(mdsc);
3357 ceph_flush_dirty_caps(mdsc);
3358 wait_requests(mdsc);
3359
3360 /*
3361 * wait for reply handlers to drop their request refs and
3362 * their inode/dcache refs
3363 */
3364 ceph_msgr_flush();
3365 }
3366
3367 /*
3368 * wait for all write mds requests to flush.
3369 */
3370 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3371 {
3372 struct ceph_mds_request *req = NULL, *nextreq;
3373 struct rb_node *n;
3374
3375 mutex_lock(&mdsc->mutex);
3376 dout("wait_unsafe_requests want %lld\n", want_tid);
3377 restart:
3378 req = __get_oldest_req(mdsc);
3379 while (req && req->r_tid <= want_tid) {
3380 /* find next request */
3381 n = rb_next(&req->r_node);
3382 if (n)
3383 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3384 else
3385 nextreq = NULL;
3386 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3387 /* write op */
3388 ceph_mdsc_get_request(req);
3389 if (nextreq)
3390 ceph_mdsc_get_request(nextreq);
3391 mutex_unlock(&mdsc->mutex);
3392 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3393 req->r_tid, want_tid);
3394 wait_for_completion(&req->r_safe_completion);
3395 mutex_lock(&mdsc->mutex);
3396 ceph_mdsc_put_request(req);
3397 if (!nextreq)
3398 break; /* next dne before, so we're done! */
3399 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3400 /* next request was removed from tree */
3401 ceph_mdsc_put_request(nextreq);
3402 goto restart;
3403 }
3404 ceph_mdsc_put_request(nextreq); /* won't go away */
3405 }
3406 req = nextreq;
3407 }
3408 mutex_unlock(&mdsc->mutex);
3409 dout("wait_unsafe_requests done\n");
3410 }
3411
3412 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3413 {
3414 u64 want_tid, want_flush;
3415
3416 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3417 return;
3418
3419 dout("sync\n");
3420 mutex_lock(&mdsc->mutex);
3421 want_tid = mdsc->last_tid;
3422 want_flush = mdsc->cap_flush_seq;
3423 mutex_unlock(&mdsc->mutex);
3424 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
3425
3426 ceph_flush_dirty_caps(mdsc);
3427
3428 wait_unsafe_requests(mdsc, want_tid);
3429 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
3430 }
3431
3432 /*
3433 * true if all sessions are closed, or we force unmount
3434 */
3435 static bool done_closing_sessions(struct ceph_mds_client *mdsc)
3436 {
3437 int i, n = 0;
3438
3439 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3440 return true;
3441
3442 mutex_lock(&mdsc->mutex);
3443 for (i = 0; i < mdsc->max_sessions; i++)
3444 if (mdsc->sessions[i])
3445 n++;
3446 mutex_unlock(&mdsc->mutex);
3447 return n == 0;
3448 }
3449
3450 /*
3451 * called after sb is ro.
3452 */
3453 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3454 {
3455 struct ceph_mds_session *session;
3456 int i;
3457 struct ceph_fs_client *fsc = mdsc->fsc;
3458 unsigned long timeout = fsc->client->options->mount_timeout * HZ;
3459
3460 dout("close_sessions\n");
3461
3462 /* close sessions */
3463 mutex_lock(&mdsc->mutex);
3464 for (i = 0; i < mdsc->max_sessions; i++) {
3465 session = __ceph_lookup_mds_session(mdsc, i);
3466 if (!session)
3467 continue;
3468 mutex_unlock(&mdsc->mutex);
3469 mutex_lock(&session->s_mutex);
3470 __close_session(mdsc, session);
3471 mutex_unlock(&session->s_mutex);
3472 ceph_put_mds_session(session);
3473 mutex_lock(&mdsc->mutex);
3474 }
3475 mutex_unlock(&mdsc->mutex);
3476
3477 dout("waiting for sessions to close\n");
3478 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3479 timeout);
3480
3481 /* tear down remaining sessions */
3482 mutex_lock(&mdsc->mutex);
3483 for (i = 0; i < mdsc->max_sessions; i++) {
3484 if (mdsc->sessions[i]) {
3485 session = get_session(mdsc->sessions[i]);
3486 __unregister_session(mdsc, session);
3487 mutex_unlock(&mdsc->mutex);
3488 mutex_lock(&session->s_mutex);
3489 remove_session_caps(session);
3490 mutex_unlock(&session->s_mutex);
3491 ceph_put_mds_session(session);
3492 mutex_lock(&mdsc->mutex);
3493 }
3494 }
3495 WARN_ON(!list_empty(&mdsc->cap_delay_list));
3496 mutex_unlock(&mdsc->mutex);
3497
3498 ceph_cleanup_empty_realms(mdsc);
3499
3500 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3501
3502 dout("stopped\n");
3503 }
3504
3505 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3506 {
3507 dout("stop\n");
3508 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3509 if (mdsc->mdsmap)
3510 ceph_mdsmap_destroy(mdsc->mdsmap);
3511 kfree(mdsc->sessions);
3512 ceph_caps_finalize(mdsc);
3513 }
3514
3515 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3516 {
3517 struct ceph_mds_client *mdsc = fsc->mdsc;
3518
3519 dout("mdsc_destroy %p\n", mdsc);
3520 ceph_mdsc_stop(mdsc);
3521
3522 /* flush out any connection work with references to us */
3523 ceph_msgr_flush();
3524
3525 fsc->mdsc = NULL;
3526 kfree(mdsc);
3527 dout("mdsc_destroy %p done\n", mdsc);
3528 }
3529
3530
3531 /*
3532 * handle mds map update.
3533 */
3534 void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3535 {
3536 u32 epoch;
3537 u32 maplen;
3538 void *p = msg->front.iov_base;
3539 void *end = p + msg->front.iov_len;
3540 struct ceph_mdsmap *newmap, *oldmap;
3541 struct ceph_fsid fsid;
3542 int err = -EINVAL;
3543
3544 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3545 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3546 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
3547 return;
3548 epoch = ceph_decode_32(&p);
3549 maplen = ceph_decode_32(&p);
3550 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3551
3552 /* do we need it? */
3553 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
3554 mutex_lock(&mdsc->mutex);
3555 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3556 dout("handle_map epoch %u <= our %u\n",
3557 epoch, mdsc->mdsmap->m_epoch);
3558 mutex_unlock(&mdsc->mutex);
3559 return;
3560 }
3561
3562 newmap = ceph_mdsmap_decode(&p, end);
3563 if (IS_ERR(newmap)) {
3564 err = PTR_ERR(newmap);
3565 goto bad_unlock;
3566 }
3567
3568 /* swap into place */
3569 if (mdsc->mdsmap) {
3570 oldmap = mdsc->mdsmap;
3571 mdsc->mdsmap = newmap;
3572 check_new_map(mdsc, newmap, oldmap);
3573 ceph_mdsmap_destroy(oldmap);
3574 } else {
3575 mdsc->mdsmap = newmap; /* first mds map */
3576 }
3577 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3578
3579 __wake_requests(mdsc, &mdsc->waiting_for_map);
3580
3581 mutex_unlock(&mdsc->mutex);
3582 schedule_delayed(mdsc);
3583 return;
3584
3585 bad_unlock:
3586 mutex_unlock(&mdsc->mutex);
3587 bad:
3588 pr_err("error decoding mdsmap %d\n", err);
3589 return;
3590 }
3591
3592 static struct ceph_connection *con_get(struct ceph_connection *con)
3593 {
3594 struct ceph_mds_session *s = con->private;
3595
3596 if (get_session(s)) {
3597 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
3598 return con;
3599 }
3600 dout("mdsc con_get %p FAIL\n", s);
3601 return NULL;
3602 }
3603
3604 static void con_put(struct ceph_connection *con)
3605 {
3606 struct ceph_mds_session *s = con->private;
3607
3608 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
3609 ceph_put_mds_session(s);
3610 }
3611
3612 /*
3613 * if the client is unresponsive for long enough, the mds will kill
3614 * the session entirely.
3615 */
3616 static void peer_reset(struct ceph_connection *con)
3617 {
3618 struct ceph_mds_session *s = con->private;
3619 struct ceph_mds_client *mdsc = s->s_mdsc;
3620
3621 pr_warn("mds%d closed our session\n", s->s_mds);
3622 send_mds_reconnect(mdsc, s);
3623 }
3624
3625 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3626 {
3627 struct ceph_mds_session *s = con->private;
3628 struct ceph_mds_client *mdsc = s->s_mdsc;
3629 int type = le16_to_cpu(msg->hdr.type);
3630
3631 mutex_lock(&mdsc->mutex);
3632 if (__verify_registered_session(mdsc, s) < 0) {
3633 mutex_unlock(&mdsc->mutex);
3634 goto out;
3635 }
3636 mutex_unlock(&mdsc->mutex);
3637
3638 switch (type) {
3639 case CEPH_MSG_MDS_MAP:
3640 ceph_mdsc_handle_map(mdsc, msg);
3641 break;
3642 case CEPH_MSG_CLIENT_SESSION:
3643 handle_session(s, msg);
3644 break;
3645 case CEPH_MSG_CLIENT_REPLY:
3646 handle_reply(s, msg);
3647 break;
3648 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
3649 handle_forward(mdsc, s, msg);
3650 break;
3651 case CEPH_MSG_CLIENT_CAPS:
3652 ceph_handle_caps(s, msg);
3653 break;
3654 case CEPH_MSG_CLIENT_SNAP:
3655 ceph_handle_snap(mdsc, s, msg);
3656 break;
3657 case CEPH_MSG_CLIENT_LEASE:
3658 handle_lease(mdsc, s, msg);
3659 break;
3660
3661 default:
3662 pr_err("received unknown message type %d %s\n", type,
3663 ceph_msg_type_name(type));
3664 }
3665 out:
3666 ceph_msg_put(msg);
3667 }
3668
3669 /*
3670 * authentication
3671 */
3672
3673 /*
3674 * Note: returned pointer is the address of a structure that's
3675 * managed separately. Caller must *not* attempt to free it.
3676 */
3677 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
3678 int *proto, int force_new)
3679 {
3680 struct ceph_mds_session *s = con->private;
3681 struct ceph_mds_client *mdsc = s->s_mdsc;
3682 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3683 struct ceph_auth_handshake *auth = &s->s_auth;
3684
3685 if (force_new && auth->authorizer) {
3686 ceph_auth_destroy_authorizer(ac, auth->authorizer);
3687 auth->authorizer = NULL;
3688 }
3689 if (!auth->authorizer) {
3690 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3691 auth);
3692 if (ret)
3693 return ERR_PTR(ret);
3694 } else {
3695 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3696 auth);
3697 if (ret)
3698 return ERR_PTR(ret);
3699 }
3700 *proto = ac->protocol;
3701
3702 return auth;
3703 }
3704
3705
3706 static int verify_authorizer_reply(struct ceph_connection *con, int len)
3707 {
3708 struct ceph_mds_session *s = con->private;
3709 struct ceph_mds_client *mdsc = s->s_mdsc;
3710 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3711
3712 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
3713 }
3714
3715 static int invalidate_authorizer(struct ceph_connection *con)
3716 {
3717 struct ceph_mds_session *s = con->private;
3718 struct ceph_mds_client *mdsc = s->s_mdsc;
3719 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3720
3721 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3722
3723 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
3724 }
3725
3726 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3727 struct ceph_msg_header *hdr, int *skip)
3728 {
3729 struct ceph_msg *msg;
3730 int type = (int) le16_to_cpu(hdr->type);
3731 int front_len = (int) le32_to_cpu(hdr->front_len);
3732
3733 if (con->in_msg)
3734 return con->in_msg;
3735
3736 *skip = 0;
3737 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3738 if (!msg) {
3739 pr_err("unable to allocate msg type %d len %d\n",
3740 type, front_len);
3741 return NULL;
3742 }
3743
3744 return msg;
3745 }
3746
3747 static const struct ceph_connection_operations mds_con_ops = {
3748 .get = con_get,
3749 .put = con_put,
3750 .dispatch = dispatch,
3751 .get_authorizer = get_authorizer,
3752 .verify_authorizer_reply = verify_authorizer_reply,
3753 .invalidate_authorizer = invalidate_authorizer,
3754 .peer_reset = peer_reset,
3755 .alloc_msg = mds_alloc_msg,
3756 };
3757
3758 /* eof */
Linux with added FPC-III support