blob: 1766947fc07ab0892027fd5a3fc042e552310275 [file] [log] [blame]
Sage Weil2f2dc052009-10-06 11:31:09 -07001#include "ceph_debug.h"
2
3#include <linux/wait.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09004#include <linux/slab.h>
Sage Weil2f2dc052009-10-06 11:31:09 -07005#include <linux/sched.h>
6
7#include "mds_client.h"
8#include "mon_client.h"
9#include "super.h"
10#include "messenger.h"
11#include "decode.h"
Sage Weil4e7a5dc2009-11-18 16:19:57 -080012#include "auth.h"
Sage Weil93cea5b2009-12-23 12:21:51 -080013#include "pagelist.h"
Sage Weil2f2dc052009-10-06 11:31:09 -070014
15/*
16 * A cluster of MDS (metadata server) daemons is responsible for
17 * managing the file system namespace (the directory hierarchy and
18 * inodes) and for coordinating shared access to storage. Metadata is
19 * partitioning hierarchically across a number of servers, and that
20 * partition varies over time as the cluster adjusts the distribution
21 * in order to balance load.
22 *
23 * The MDS client is primarily responsible to managing synchronous
24 * metadata requests for operations like open, unlink, and so forth.
25 * If there is a MDS failure, we find out about it when we (possibly
26 * request and) receive a new MDS map, and can resubmit affected
27 * requests.
28 *
29 * For the most part, though, we take advantage of a lossless
30 * communications channel to the MDS, and do not need to worry about
31 * timing out or resubmitting requests.
32 *
33 * We maintain a stateful "session" with each MDS we interact with.
34 * Within each session, we sent periodic heartbeat messages to ensure
35 * any capabilities or leases we have been issues remain valid. If
36 * the session times out and goes stale, our leases and capabilities
37 * are no longer valid.
38 */
39
40static void __wake_requests(struct ceph_mds_client *mdsc,
41 struct list_head *head);
42
Tobias Klauser9e327892010-05-20 10:40:19 +020043static const struct ceph_connection_operations mds_con_ops;
Sage Weil2f2dc052009-10-06 11:31:09 -070044
45
46/*
47 * mds reply parsing
48 */
49
50/*
51 * parse individual inode info
52 */
53static int parse_reply_info_in(void **p, void *end,
54 struct ceph_mds_reply_info_in *info)
55{
56 int err = -EIO;
57
58 info->in = *p;
59 *p += sizeof(struct ceph_mds_reply_inode) +
60 sizeof(*info->in->fragtree.splits) *
61 le32_to_cpu(info->in->fragtree.nsplits);
62
63 ceph_decode_32_safe(p, end, info->symlink_len, bad);
64 ceph_decode_need(p, end, info->symlink_len, bad);
65 info->symlink = *p;
66 *p += info->symlink_len;
67
68 ceph_decode_32_safe(p, end, info->xattr_len, bad);
69 ceph_decode_need(p, end, info->xattr_len, bad);
70 info->xattr_data = *p;
71 *p += info->xattr_len;
72 return 0;
73bad:
74 return err;
75}
76
77/*
78 * parse a normal reply, which may contain a (dir+)dentry and/or a
79 * target inode.
80 */
81static int parse_reply_info_trace(void **p, void *end,
82 struct ceph_mds_reply_info_parsed *info)
83{
84 int err;
85
86 if (info->head->is_dentry) {
87 err = parse_reply_info_in(p, end, &info->diri);
88 if (err < 0)
89 goto out_bad;
90
91 if (unlikely(*p + sizeof(*info->dirfrag) > end))
92 goto bad;
93 info->dirfrag = *p;
94 *p += sizeof(*info->dirfrag) +
95 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
96 if (unlikely(*p > end))
97 goto bad;
98
99 ceph_decode_32_safe(p, end, info->dname_len, bad);
100 ceph_decode_need(p, end, info->dname_len, bad);
101 info->dname = *p;
102 *p += info->dname_len;
103 info->dlease = *p;
104 *p += sizeof(*info->dlease);
105 }
106
107 if (info->head->is_target) {
108 err = parse_reply_info_in(p, end, &info->targeti);
109 if (err < 0)
110 goto out_bad;
111 }
112
113 if (unlikely(*p != end))
114 goto bad;
115 return 0;
116
117bad:
118 err = -EIO;
119out_bad:
120 pr_err("problem parsing mds trace %d\n", err);
121 return err;
122}
123
124/*
125 * parse readdir results
126 */
127static int parse_reply_info_dir(void **p, void *end,
128 struct ceph_mds_reply_info_parsed *info)
129{
130 u32 num, i = 0;
131 int err;
132
133 info->dir_dir = *p;
134 if (*p + sizeof(*info->dir_dir) > end)
135 goto bad;
136 *p += sizeof(*info->dir_dir) +
137 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
138 if (*p > end)
139 goto bad;
140
141 ceph_decode_need(p, end, sizeof(num) + 2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700142 num = ceph_decode_32(p);
143 info->dir_end = ceph_decode_8(p);
144 info->dir_complete = ceph_decode_8(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700145 if (num == 0)
146 goto done;
147
148 /* alloc large array */
149 info->dir_nr = num;
150 info->dir_in = kcalloc(num, sizeof(*info->dir_in) +
151 sizeof(*info->dir_dname) +
152 sizeof(*info->dir_dname_len) +
153 sizeof(*info->dir_dlease),
154 GFP_NOFS);
155 if (info->dir_in == NULL) {
156 err = -ENOMEM;
157 goto out_bad;
158 }
159 info->dir_dname = (void *)(info->dir_in + num);
160 info->dir_dname_len = (void *)(info->dir_dname + num);
161 info->dir_dlease = (void *)(info->dir_dname_len + num);
162
163 while (num) {
164 /* dentry */
165 ceph_decode_need(p, end, sizeof(u32)*2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700166 info->dir_dname_len[i] = ceph_decode_32(p);
Sage Weil2f2dc052009-10-06 11:31:09 -0700167 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
168 info->dir_dname[i] = *p;
169 *p += info->dir_dname_len[i];
170 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
171 info->dir_dname[i]);
172 info->dir_dlease[i] = *p;
173 *p += sizeof(struct ceph_mds_reply_lease);
174
175 /* inode */
176 err = parse_reply_info_in(p, end, &info->dir_in[i]);
177 if (err < 0)
178 goto out_bad;
179 i++;
180 num--;
181 }
182
183done:
184 if (*p != end)
185 goto bad;
186 return 0;
187
188bad:
189 err = -EIO;
190out_bad:
191 pr_err("problem parsing dir contents %d\n", err);
192 return err;
193}
194
195/*
196 * parse entire mds reply
197 */
198static int parse_reply_info(struct ceph_msg *msg,
199 struct ceph_mds_reply_info_parsed *info)
200{
201 void *p, *end;
202 u32 len;
203 int err;
204
205 info->head = msg->front.iov_base;
206 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
207 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
208
209 /* trace */
210 ceph_decode_32_safe(&p, end, len, bad);
211 if (len > 0) {
212 err = parse_reply_info_trace(&p, p+len, info);
213 if (err < 0)
214 goto out_bad;
215 }
216
217 /* dir content */
218 ceph_decode_32_safe(&p, end, len, bad);
219 if (len > 0) {
220 err = parse_reply_info_dir(&p, p+len, info);
221 if (err < 0)
222 goto out_bad;
223 }
224
225 /* snap blob */
226 ceph_decode_32_safe(&p, end, len, bad);
227 info->snapblob_len = len;
228 info->snapblob = p;
229 p += len;
230
231 if (p != end)
232 goto bad;
233 return 0;
234
235bad:
236 err = -EIO;
237out_bad:
238 pr_err("mds parse_reply err %d\n", err);
239 return err;
240}
241
242static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
243{
244 kfree(info->dir_in);
245}
246
247
248/*
249 * sessions
250 */
251static const char *session_state_name(int s)
252{
253 switch (s) {
254 case CEPH_MDS_SESSION_NEW: return "new";
255 case CEPH_MDS_SESSION_OPENING: return "opening";
256 case CEPH_MDS_SESSION_OPEN: return "open";
257 case CEPH_MDS_SESSION_HUNG: return "hung";
258 case CEPH_MDS_SESSION_CLOSING: return "closing";
Sage Weil44ca18f2010-02-15 12:08:46 -0800259 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
Sage Weil2f2dc052009-10-06 11:31:09 -0700260 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
261 default: return "???";
262 }
263}
264
265static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
266{
267 if (atomic_inc_not_zero(&s->s_ref)) {
268 dout("mdsc get_session %p %d -> %d\n", s,
269 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
270 return s;
271 } else {
272 dout("mdsc get_session %p 0 -- FAIL", s);
273 return NULL;
274 }
275}
276
277void ceph_put_mds_session(struct ceph_mds_session *s)
278{
279 dout("mdsc put_session %p %d -> %d\n", s,
280 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800281 if (atomic_dec_and_test(&s->s_ref)) {
282 if (s->s_authorizer)
283 s->s_mdsc->client->monc.auth->ops->destroy_authorizer(
284 s->s_mdsc->client->monc.auth, s->s_authorizer);
Sage Weil2f2dc052009-10-06 11:31:09 -0700285 kfree(s);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800286 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700287}
288
289/*
290 * called under mdsc->mutex
291 */
292struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
293 int mds)
294{
295 struct ceph_mds_session *session;
296
297 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
298 return NULL;
299 session = mdsc->sessions[mds];
300 dout("lookup_mds_session %p %d\n", session,
301 atomic_read(&session->s_ref));
302 get_session(session);
303 return session;
304}
305
306static bool __have_session(struct ceph_mds_client *mdsc, int mds)
307{
308 if (mds >= mdsc->max_sessions)
309 return false;
310 return mdsc->sessions[mds];
311}
312
Sage Weil2600d2d2010-02-22 15:12:16 -0800313static int __verify_registered_session(struct ceph_mds_client *mdsc,
314 struct ceph_mds_session *s)
315{
316 if (s->s_mds >= mdsc->max_sessions ||
317 mdsc->sessions[s->s_mds] != s)
318 return -ENOENT;
319 return 0;
320}
321
Sage Weil2f2dc052009-10-06 11:31:09 -0700322/*
323 * create+register a new session for given mds.
324 * called under mdsc->mutex.
325 */
326static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
327 int mds)
328{
329 struct ceph_mds_session *s;
330
331 s = kzalloc(sizeof(*s), GFP_NOFS);
Dan Carpenter4736b002010-03-20 15:30:16 +0300332 if (!s)
333 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700334 s->s_mdsc = mdsc;
335 s->s_mds = mds;
336 s->s_state = CEPH_MDS_SESSION_NEW;
337 s->s_ttl = 0;
338 s->s_seq = 0;
339 mutex_init(&s->s_mutex);
340
341 ceph_con_init(mdsc->client->msgr, &s->s_con);
342 s->s_con.private = s;
343 s->s_con.ops = &mds_con_ops;
344 s->s_con.peer_name.type = CEPH_ENTITY_TYPE_MDS;
345 s->s_con.peer_name.num = cpu_to_le64(mds);
Sage Weil2f2dc052009-10-06 11:31:09 -0700346
347 spin_lock_init(&s->s_cap_lock);
348 s->s_cap_gen = 0;
349 s->s_cap_ttl = 0;
350 s->s_renew_requested = 0;
351 s->s_renew_seq = 0;
352 INIT_LIST_HEAD(&s->s_caps);
353 s->s_nr_caps = 0;
Sage Weil5dacf092009-12-21 20:40:34 -0800354 s->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -0700355 atomic_set(&s->s_ref, 1);
356 INIT_LIST_HEAD(&s->s_waiting);
357 INIT_LIST_HEAD(&s->s_unsafe);
358 s->s_num_cap_releases = 0;
Sage Weil7c1332b2010-02-16 11:39:45 -0800359 s->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700360 INIT_LIST_HEAD(&s->s_cap_releases);
361 INIT_LIST_HEAD(&s->s_cap_releases_done);
362 INIT_LIST_HEAD(&s->s_cap_flushing);
363 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
364
365 dout("register_session mds%d\n", mds);
366 if (mds >= mdsc->max_sessions) {
367 int newmax = 1 << get_count_order(mds+1);
368 struct ceph_mds_session **sa;
369
370 dout("register_session realloc to %d\n", newmax);
371 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
372 if (sa == NULL)
Sage Weil42ce56e2009-11-18 11:22:36 -0800373 goto fail_realloc;
Sage Weil2f2dc052009-10-06 11:31:09 -0700374 if (mdsc->sessions) {
375 memcpy(sa, mdsc->sessions,
376 mdsc->max_sessions * sizeof(void *));
377 kfree(mdsc->sessions);
378 }
379 mdsc->sessions = sa;
380 mdsc->max_sessions = newmax;
381 }
382 mdsc->sessions[mds] = s;
383 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
Sage Weil42ce56e2009-11-18 11:22:36 -0800384
385 ceph_con_open(&s->s_con, ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
386
Sage Weil2f2dc052009-10-06 11:31:09 -0700387 return s;
Sage Weil42ce56e2009-11-18 11:22:36 -0800388
389fail_realloc:
390 kfree(s);
391 return ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -0700392}
393
394/*
395 * called under mdsc->mutex
396 */
Sage Weil2600d2d2010-02-22 15:12:16 -0800397static void __unregister_session(struct ceph_mds_client *mdsc,
Sage Weil42ce56e2009-11-18 11:22:36 -0800398 struct ceph_mds_session *s)
Sage Weil2f2dc052009-10-06 11:31:09 -0700399{
Sage Weil2600d2d2010-02-22 15:12:16 -0800400 dout("__unregister_session mds%d %p\n", s->s_mds, s);
401 BUG_ON(mdsc->sessions[s->s_mds] != s);
Sage Weil42ce56e2009-11-18 11:22:36 -0800402 mdsc->sessions[s->s_mds] = NULL;
403 ceph_con_close(&s->s_con);
404 ceph_put_mds_session(s);
Sage Weil2f2dc052009-10-06 11:31:09 -0700405}
406
407/*
408 * drop session refs in request.
409 *
410 * should be last request ref, or hold mdsc->mutex
411 */
412static void put_request_session(struct ceph_mds_request *req)
413{
414 if (req->r_session) {
415 ceph_put_mds_session(req->r_session);
416 req->r_session = NULL;
417 }
418}
419
Sage Weil153c8e62009-12-07 12:31:09 -0800420void ceph_mdsc_release_request(struct kref *kref)
Sage Weil2f2dc052009-10-06 11:31:09 -0700421{
Sage Weil153c8e62009-12-07 12:31:09 -0800422 struct ceph_mds_request *req = container_of(kref,
423 struct ceph_mds_request,
424 r_kref);
425 if (req->r_request)
426 ceph_msg_put(req->r_request);
427 if (req->r_reply) {
428 ceph_msg_put(req->r_reply);
429 destroy_reply_info(&req->r_reply_info);
Sage Weil2f2dc052009-10-06 11:31:09 -0700430 }
Sage Weil153c8e62009-12-07 12:31:09 -0800431 if (req->r_inode) {
432 ceph_put_cap_refs(ceph_inode(req->r_inode),
433 CEPH_CAP_PIN);
434 iput(req->r_inode);
435 }
436 if (req->r_locked_dir)
437 ceph_put_cap_refs(ceph_inode(req->r_locked_dir),
438 CEPH_CAP_PIN);
439 if (req->r_target_inode)
440 iput(req->r_target_inode);
441 if (req->r_dentry)
442 dput(req->r_dentry);
443 if (req->r_old_dentry) {
444 ceph_put_cap_refs(
445 ceph_inode(req->r_old_dentry->d_parent->d_inode),
446 CEPH_CAP_PIN);
447 dput(req->r_old_dentry);
448 }
449 kfree(req->r_path1);
450 kfree(req->r_path2);
451 put_request_session(req);
452 ceph_unreserve_caps(&req->r_caps_reservation);
453 kfree(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700454}
455
456/*
457 * lookup session, bump ref if found.
458 *
459 * called under mdsc->mutex.
460 */
461static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
462 u64 tid)
463{
464 struct ceph_mds_request *req;
Sage Weil44ca18f2010-02-15 12:08:46 -0800465 struct rb_node *n = mdsc->request_tree.rb_node;
466
467 while (n) {
468 req = rb_entry(n, struct ceph_mds_request, r_node);
469 if (tid < req->r_tid)
470 n = n->rb_left;
471 else if (tid > req->r_tid)
472 n = n->rb_right;
473 else {
474 ceph_mdsc_get_request(req);
475 return req;
476 }
477 }
478 return NULL;
479}
480
481static void __insert_request(struct ceph_mds_client *mdsc,
482 struct ceph_mds_request *new)
483{
484 struct rb_node **p = &mdsc->request_tree.rb_node;
485 struct rb_node *parent = NULL;
486 struct ceph_mds_request *req = NULL;
487
488 while (*p) {
489 parent = *p;
490 req = rb_entry(parent, struct ceph_mds_request, r_node);
491 if (new->r_tid < req->r_tid)
492 p = &(*p)->rb_left;
493 else if (new->r_tid > req->r_tid)
494 p = &(*p)->rb_right;
495 else
496 BUG();
497 }
498
499 rb_link_node(&new->r_node, parent, p);
500 rb_insert_color(&new->r_node, &mdsc->request_tree);
Sage Weil2f2dc052009-10-06 11:31:09 -0700501}
502
503/*
504 * Register an in-flight request, and assign a tid. Link to directory
505 * are modifying (if any).
506 *
507 * Called under mdsc->mutex.
508 */
509static void __register_request(struct ceph_mds_client *mdsc,
510 struct ceph_mds_request *req,
511 struct inode *dir)
512{
513 req->r_tid = ++mdsc->last_tid;
514 if (req->r_num_caps)
515 ceph_reserve_caps(&req->r_caps_reservation, req->r_num_caps);
516 dout("__register_request %p tid %lld\n", req, req->r_tid);
517 ceph_mdsc_get_request(req);
Sage Weil44ca18f2010-02-15 12:08:46 -0800518 __insert_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700519
520 if (dir) {
521 struct ceph_inode_info *ci = ceph_inode(dir);
522
523 spin_lock(&ci->i_unsafe_lock);
524 req->r_unsafe_dir = dir;
525 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
526 spin_unlock(&ci->i_unsafe_lock);
527 }
528}
529
530static void __unregister_request(struct ceph_mds_client *mdsc,
531 struct ceph_mds_request *req)
532{
533 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -0800534 rb_erase(&req->r_node, &mdsc->request_tree);
Sage Weil80fc7312010-03-16 15:28:54 -0700535 RB_CLEAR_NODE(&req->r_node);
Sage Weil2f2dc052009-10-06 11:31:09 -0700536
537 if (req->r_unsafe_dir) {
538 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
539
540 spin_lock(&ci->i_unsafe_lock);
541 list_del_init(&req->r_unsafe_dir_item);
542 spin_unlock(&ci->i_unsafe_lock);
543 }
Sage Weil94aa8ae2010-03-28 21:22:50 -0700544
545 ceph_mdsc_put_request(req);
Sage Weil2f2dc052009-10-06 11:31:09 -0700546}
547
548/*
549 * Choose mds to send request to next. If there is a hint set in the
550 * request (e.g., due to a prior forward hint from the mds), use that.
551 * Otherwise, consult frag tree and/or caps to identify the
552 * appropriate mds. If all else fails, choose randomly.
553 *
554 * Called under mdsc->mutex.
555 */
556static int __choose_mds(struct ceph_mds_client *mdsc,
557 struct ceph_mds_request *req)
558{
559 struct inode *inode;
560 struct ceph_inode_info *ci;
561 struct ceph_cap *cap;
562 int mode = req->r_direct_mode;
563 int mds = -1;
564 u32 hash = req->r_direct_hash;
565 bool is_hash = req->r_direct_is_hash;
566
567 /*
568 * is there a specific mds we should try? ignore hint if we have
569 * no session and the mds is not up (active or recovering).
570 */
571 if (req->r_resend_mds >= 0 &&
572 (__have_session(mdsc, req->r_resend_mds) ||
573 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
574 dout("choose_mds using resend_mds mds%d\n",
575 req->r_resend_mds);
576 return req->r_resend_mds;
577 }
578
579 if (mode == USE_RANDOM_MDS)
580 goto random;
581
582 inode = NULL;
583 if (req->r_inode) {
584 inode = req->r_inode;
585 } else if (req->r_dentry) {
586 if (req->r_dentry->d_inode) {
587 inode = req->r_dentry->d_inode;
588 } else {
589 inode = req->r_dentry->d_parent->d_inode;
590 hash = req->r_dentry->d_name.hash;
591 is_hash = true;
592 }
593 }
594 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
595 (int)hash, mode);
596 if (!inode)
597 goto random;
598 ci = ceph_inode(inode);
599
600 if (is_hash && S_ISDIR(inode->i_mode)) {
601 struct ceph_inode_frag frag;
602 int found;
603
604 ceph_choose_frag(ci, hash, &frag, &found);
605 if (found) {
606 if (mode == USE_ANY_MDS && frag.ndist > 0) {
607 u8 r;
608
609 /* choose a random replica */
610 get_random_bytes(&r, 1);
611 r %= frag.ndist;
612 mds = frag.dist[r];
613 dout("choose_mds %p %llx.%llx "
614 "frag %u mds%d (%d/%d)\n",
615 inode, ceph_vinop(inode),
616 frag.frag, frag.mds,
617 (int)r, frag.ndist);
618 return mds;
619 }
620
621 /* since this file/dir wasn't known to be
622 * replicated, then we want to look for the
623 * authoritative mds. */
624 mode = USE_AUTH_MDS;
625 if (frag.mds >= 0) {
626 /* choose auth mds */
627 mds = frag.mds;
628 dout("choose_mds %p %llx.%llx "
629 "frag %u mds%d (auth)\n",
630 inode, ceph_vinop(inode), frag.frag, mds);
631 return mds;
632 }
633 }
634 }
635
636 spin_lock(&inode->i_lock);
637 cap = NULL;
638 if (mode == USE_AUTH_MDS)
639 cap = ci->i_auth_cap;
640 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
641 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
642 if (!cap) {
643 spin_unlock(&inode->i_lock);
644 goto random;
645 }
646 mds = cap->session->s_mds;
647 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
648 inode, ceph_vinop(inode), mds,
649 cap == ci->i_auth_cap ? "auth " : "", cap);
650 spin_unlock(&inode->i_lock);
651 return mds;
652
653random:
654 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
655 dout("choose_mds chose random mds%d\n", mds);
656 return mds;
657}
658
659
660/*
661 * session messages
662 */
663static struct ceph_msg *create_session_msg(u32 op, u64 seq)
664{
665 struct ceph_msg *msg;
666 struct ceph_mds_session_head *h;
667
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700668 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -0700669 if (!msg) {
Sage Weil2f2dc052009-10-06 11:31:09 -0700670 pr_err("create_session_msg ENOMEM creating msg\n");
Sage Weila79832f2010-04-01 16:06:19 -0700671 return NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700672 }
673 h = msg->front.iov_base;
674 h->op = cpu_to_le32(op);
675 h->seq = cpu_to_le64(seq);
676 return msg;
677}
678
679/*
680 * send session open request.
681 *
682 * called under mdsc->mutex
683 */
684static int __open_session(struct ceph_mds_client *mdsc,
685 struct ceph_mds_session *session)
686{
687 struct ceph_msg *msg;
688 int mstate;
689 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -0700690
691 /* wait for mds to go active? */
692 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
693 dout("open_session to mds%d (%s)\n", mds,
694 ceph_mds_state_name(mstate));
695 session->s_state = CEPH_MDS_SESSION_OPENING;
696 session->s_renew_requested = jiffies;
697
698 /* send connect message */
699 msg = create_session_msg(CEPH_SESSION_REQUEST_OPEN, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -0700700 if (!msg)
701 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -0700702 ceph_con_send(&session->s_con, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -0700703 return 0;
704}
705
706/*
707 * session caps
708 */
709
710/*
711 * Free preallocated cap messages assigned to this session
712 */
713static void cleanup_cap_releases(struct ceph_mds_session *session)
714{
715 struct ceph_msg *msg;
716
717 spin_lock(&session->s_cap_lock);
718 while (!list_empty(&session->s_cap_releases)) {
719 msg = list_first_entry(&session->s_cap_releases,
720 struct ceph_msg, list_head);
721 list_del_init(&msg->list_head);
722 ceph_msg_put(msg);
723 }
724 while (!list_empty(&session->s_cap_releases_done)) {
725 msg = list_first_entry(&session->s_cap_releases_done,
726 struct ceph_msg, list_head);
727 list_del_init(&msg->list_head);
728 ceph_msg_put(msg);
729 }
730 spin_unlock(&session->s_cap_lock);
731}
732
733/*
Sage Weilf818a732010-05-11 20:56:31 -0700734 * Helper to safely iterate over all caps associated with a session, with
735 * special care taken to handle a racing __ceph_remove_cap().
Sage Weil2f2dc052009-10-06 11:31:09 -0700736 *
Sage Weilf818a732010-05-11 20:56:31 -0700737 * Caller must hold session s_mutex.
Sage Weil2f2dc052009-10-06 11:31:09 -0700738 */
739static int iterate_session_caps(struct ceph_mds_session *session,
740 int (*cb)(struct inode *, struct ceph_cap *,
741 void *), void *arg)
742{
Sage Weil7c1332b2010-02-16 11:39:45 -0800743 struct list_head *p;
744 struct ceph_cap *cap;
745 struct inode *inode, *last_inode = NULL;
746 struct ceph_cap *old_cap = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700747 int ret;
748
749 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
750 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800751 p = session->s_caps.next;
752 while (p != &session->s_caps) {
753 cap = list_entry(p, struct ceph_cap, session_caps);
Sage Weil2f2dc052009-10-06 11:31:09 -0700754 inode = igrab(&cap->ci->vfs_inode);
Sage Weil7c1332b2010-02-16 11:39:45 -0800755 if (!inode) {
756 p = p->next;
Sage Weil2f2dc052009-10-06 11:31:09 -0700757 continue;
Sage Weil7c1332b2010-02-16 11:39:45 -0800758 }
759 session->s_cap_iterator = cap;
Sage Weil2f2dc052009-10-06 11:31:09 -0700760 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800761
762 if (last_inode) {
763 iput(last_inode);
764 last_inode = NULL;
765 }
766 if (old_cap) {
767 ceph_put_cap(old_cap);
768 old_cap = NULL;
769 }
770
Sage Weil2f2dc052009-10-06 11:31:09 -0700771 ret = cb(inode, cap, arg);
Sage Weil7c1332b2010-02-16 11:39:45 -0800772 last_inode = inode;
773
Sage Weil2f2dc052009-10-06 11:31:09 -0700774 spin_lock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800775 p = p->next;
776 if (cap->ci == NULL) {
777 dout("iterate_session_caps finishing cap %p removal\n",
778 cap);
779 BUG_ON(cap->session != session);
780 list_del_init(&cap->session_caps);
781 session->s_nr_caps--;
782 cap->session = NULL;
783 old_cap = cap; /* put_cap it w/o locks held */
784 }
Sage Weil5dacf092009-12-21 20:40:34 -0800785 if (ret < 0)
786 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -0700787 }
Sage Weil5dacf092009-12-21 20:40:34 -0800788 ret = 0;
789out:
Sage Weil7c1332b2010-02-16 11:39:45 -0800790 session->s_cap_iterator = NULL;
Sage Weil2f2dc052009-10-06 11:31:09 -0700791 spin_unlock(&session->s_cap_lock);
Sage Weil7c1332b2010-02-16 11:39:45 -0800792
793 if (last_inode)
794 iput(last_inode);
795 if (old_cap)
796 ceph_put_cap(old_cap);
797
Sage Weil5dacf092009-12-21 20:40:34 -0800798 return ret;
Sage Weil2f2dc052009-10-06 11:31:09 -0700799}
800
801static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
Sage Weil6c99f252010-05-10 16:12:25 -0700802 void *arg)
Sage Weil2f2dc052009-10-06 11:31:09 -0700803{
804 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weil6c99f252010-05-10 16:12:25 -0700805 int drop = 0;
806
Sage Weil2f2dc052009-10-06 11:31:09 -0700807 dout("removing cap %p, ci is %p, inode is %p\n",
808 cap, ci, &ci->vfs_inode);
Sage Weil6c99f252010-05-10 16:12:25 -0700809 spin_lock(&inode->i_lock);
810 __ceph_remove_cap(cap);
811 if (!__ceph_is_any_real_caps(ci)) {
812 struct ceph_mds_client *mdsc =
813 &ceph_sb_to_client(inode->i_sb)->mdsc;
814
815 spin_lock(&mdsc->cap_dirty_lock);
816 if (!list_empty(&ci->i_dirty_item)) {
817 pr_info(" dropping dirty %s state for %p %lld\n",
818 ceph_cap_string(ci->i_dirty_caps),
819 inode, ceph_ino(inode));
820 ci->i_dirty_caps = 0;
821 list_del_init(&ci->i_dirty_item);
822 drop = 1;
823 }
824 if (!list_empty(&ci->i_flushing_item)) {
825 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
826 ceph_cap_string(ci->i_flushing_caps),
827 inode, ceph_ino(inode));
828 ci->i_flushing_caps = 0;
829 list_del_init(&ci->i_flushing_item);
830 mdsc->num_cap_flushing--;
831 drop = 1;
832 }
833 if (drop && ci->i_wrbuffer_ref) {
834 pr_info(" dropping dirty data for %p %lld\n",
835 inode, ceph_ino(inode));
836 ci->i_wrbuffer_ref = 0;
837 ci->i_wrbuffer_ref_head = 0;
838 drop++;
839 }
840 spin_unlock(&mdsc->cap_dirty_lock);
841 }
842 spin_unlock(&inode->i_lock);
843 while (drop--)
844 iput(inode);
Sage Weil2f2dc052009-10-06 11:31:09 -0700845 return 0;
846}
847
848/*
849 * caller must hold session s_mutex
850 */
851static void remove_session_caps(struct ceph_mds_session *session)
852{
853 dout("remove_session_caps on %p\n", session);
854 iterate_session_caps(session, remove_session_caps_cb, NULL);
855 BUG_ON(session->s_nr_caps > 0);
Sage Weil6c99f252010-05-10 16:12:25 -0700856 BUG_ON(!list_empty(&session->s_cap_flushing));
Sage Weil2f2dc052009-10-06 11:31:09 -0700857 cleanup_cap_releases(session);
858}
859
860/*
861 * wake up any threads waiting on this session's caps. if the cap is
862 * old (didn't get renewed on the client reconnect), remove it now.
863 *
864 * caller must hold s_mutex.
865 */
866static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
867 void *arg)
868{
Sage Weil0dc25702009-11-20 13:43:45 -0800869 struct ceph_inode_info *ci = ceph_inode(inode);
870
871 wake_up(&ci->i_cap_wq);
872 if (arg) {
873 spin_lock(&inode->i_lock);
874 ci->i_wanted_max_size = 0;
875 ci->i_requested_max_size = 0;
876 spin_unlock(&inode->i_lock);
877 }
Sage Weil2f2dc052009-10-06 11:31:09 -0700878 return 0;
879}
880
Sage Weil0dc25702009-11-20 13:43:45 -0800881static void wake_up_session_caps(struct ceph_mds_session *session,
882 int reconnect)
Sage Weil2f2dc052009-10-06 11:31:09 -0700883{
884 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
Sage Weil0dc25702009-11-20 13:43:45 -0800885 iterate_session_caps(session, wake_up_session_cb,
886 (void *)(unsigned long)reconnect);
Sage Weil2f2dc052009-10-06 11:31:09 -0700887}
888
889/*
890 * Send periodic message to MDS renewing all currently held caps. The
891 * ack will reset the expiration for all caps from this session.
892 *
893 * caller holds s_mutex
894 */
895static int send_renew_caps(struct ceph_mds_client *mdsc,
896 struct ceph_mds_session *session)
897{
898 struct ceph_msg *msg;
899 int state;
900
901 if (time_after_eq(jiffies, session->s_cap_ttl) &&
902 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
903 pr_info("mds%d caps stale\n", session->s_mds);
Sage Weile4cb4cb2010-03-18 13:43:09 -0700904 session->s_renew_requested = jiffies;
Sage Weil2f2dc052009-10-06 11:31:09 -0700905
906 /* do not try to renew caps until a recovering mds has reconnected
907 * with its clients. */
908 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
909 if (state < CEPH_MDS_STATE_RECONNECT) {
910 dout("send_renew_caps ignoring mds%d (%s)\n",
911 session->s_mds, ceph_mds_state_name(state));
912 return 0;
913 }
914
915 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
916 ceph_mds_state_name(state));
Sage Weil2f2dc052009-10-06 11:31:09 -0700917 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
918 ++session->s_renew_seq);
Sage Weila79832f2010-04-01 16:06:19 -0700919 if (!msg)
920 return -ENOMEM;
Sage Weil2f2dc052009-10-06 11:31:09 -0700921 ceph_con_send(&session->s_con, msg);
922 return 0;
923}
924
925/*
926 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
Sage Weil0dc25702009-11-20 13:43:45 -0800927 *
928 * Called under session->s_mutex
Sage Weil2f2dc052009-10-06 11:31:09 -0700929 */
930static void renewed_caps(struct ceph_mds_client *mdsc,
931 struct ceph_mds_session *session, int is_renew)
932{
933 int was_stale;
934 int wake = 0;
935
936 spin_lock(&session->s_cap_lock);
937 was_stale = is_renew && (session->s_cap_ttl == 0 ||
938 time_after_eq(jiffies, session->s_cap_ttl));
939
940 session->s_cap_ttl = session->s_renew_requested +
941 mdsc->mdsmap->m_session_timeout*HZ;
942
943 if (was_stale) {
944 if (time_before(jiffies, session->s_cap_ttl)) {
945 pr_info("mds%d caps renewed\n", session->s_mds);
946 wake = 1;
947 } else {
948 pr_info("mds%d caps still stale\n", session->s_mds);
949 }
950 }
951 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
952 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
953 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
954 spin_unlock(&session->s_cap_lock);
955
956 if (wake)
Sage Weil0dc25702009-11-20 13:43:45 -0800957 wake_up_session_caps(session, 0);
Sage Weil2f2dc052009-10-06 11:31:09 -0700958}
959
960/*
961 * send a session close request
962 */
963static int request_close_session(struct ceph_mds_client *mdsc,
964 struct ceph_mds_session *session)
965{
966 struct ceph_msg *msg;
Sage Weil2f2dc052009-10-06 11:31:09 -0700967
968 dout("request_close_session mds%d state %s seq %lld\n",
969 session->s_mds, session_state_name(session->s_state),
970 session->s_seq);
971 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
Sage Weila79832f2010-04-01 16:06:19 -0700972 if (!msg)
973 return -ENOMEM;
974 ceph_con_send(&session->s_con, msg);
975 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -0700976}
977
978/*
979 * Called with s_mutex held.
980 */
981static int __close_session(struct ceph_mds_client *mdsc,
982 struct ceph_mds_session *session)
983{
984 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
985 return 0;
986 session->s_state = CEPH_MDS_SESSION_CLOSING;
987 return request_close_session(mdsc, session);
988}
989
990/*
991 * Trim old(er) caps.
992 *
993 * Because we can't cache an inode without one or more caps, we do
994 * this indirectly: if a cap is unused, we prune its aliases, at which
995 * point the inode will hopefully get dropped to.
996 *
997 * Yes, this is a bit sloppy. Our only real goal here is to respond to
998 * memory pressure from the MDS, though, so it needn't be perfect.
999 */
1000static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1001{
1002 struct ceph_mds_session *session = arg;
1003 struct ceph_inode_info *ci = ceph_inode(inode);
1004 int used, oissued, mine;
1005
1006 if (session->s_trim_caps <= 0)
1007 return -1;
1008
1009 spin_lock(&inode->i_lock);
1010 mine = cap->issued | cap->implemented;
1011 used = __ceph_caps_used(ci);
1012 oissued = __ceph_caps_issued_other(ci, cap);
1013
1014 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s\n",
1015 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1016 ceph_cap_string(used));
1017 if (ci->i_dirty_caps)
1018 goto out; /* dirty caps */
1019 if ((used & ~oissued) & mine)
1020 goto out; /* we need these caps */
1021
1022 session->s_trim_caps--;
1023 if (oissued) {
1024 /* we aren't the only cap.. just remove us */
Sage Weil7c1332b2010-02-16 11:39:45 -08001025 __ceph_remove_cap(cap);
Sage Weil2f2dc052009-10-06 11:31:09 -07001026 } else {
1027 /* try to drop referring dentries */
1028 spin_unlock(&inode->i_lock);
1029 d_prune_aliases(inode);
1030 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1031 inode, cap, atomic_read(&inode->i_count));
1032 return 0;
1033 }
1034
1035out:
1036 spin_unlock(&inode->i_lock);
1037 return 0;
1038}
1039
1040/*
1041 * Trim session cap count down to some max number.
1042 */
1043static int trim_caps(struct ceph_mds_client *mdsc,
1044 struct ceph_mds_session *session,
1045 int max_caps)
1046{
1047 int trim_caps = session->s_nr_caps - max_caps;
1048
1049 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1050 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1051 if (trim_caps > 0) {
1052 session->s_trim_caps = trim_caps;
1053 iterate_session_caps(session, trim_caps_cb, session);
1054 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1055 session->s_mds, session->s_nr_caps, max_caps,
1056 trim_caps - session->s_trim_caps);
Sage Weil5dacf092009-12-21 20:40:34 -08001057 session->s_trim_caps = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001058 }
1059 return 0;
1060}
1061
1062/*
1063 * Allocate cap_release messages. If there is a partially full message
1064 * in the queue, try to allocate enough to cover it's remainder, so that
1065 * we can send it immediately.
1066 *
1067 * Called under s_mutex.
1068 */
Sage Weil2b2300d2010-06-09 16:52:04 -07001069int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
1070 struct ceph_mds_session *session,
1071 int extra)
Sage Weil2f2dc052009-10-06 11:31:09 -07001072{
1073 struct ceph_msg *msg;
1074 struct ceph_mds_cap_release *head;
1075 int err = -ENOMEM;
1076
1077 if (extra < 0)
Sage Weil6b805182009-10-27 11:50:50 -07001078 extra = mdsc->client->mount_args->cap_release_safety;
Sage Weil2f2dc052009-10-06 11:31:09 -07001079
1080 spin_lock(&session->s_cap_lock);
1081
1082 if (!list_empty(&session->s_cap_releases)) {
1083 msg = list_first_entry(&session->s_cap_releases,
1084 struct ceph_msg,
1085 list_head);
1086 head = msg->front.iov_base;
1087 extra += CEPH_CAPS_PER_RELEASE - le32_to_cpu(head->num);
1088 }
1089
1090 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1091 spin_unlock(&session->s_cap_lock);
Yehuda Sadeh34d23762010-04-06 14:33:58 -07001092 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1093 GFP_NOFS);
Sage Weil2f2dc052009-10-06 11:31:09 -07001094 if (!msg)
1095 goto out_unlocked;
1096 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1097 (int)msg->front.iov_len);
1098 head = msg->front.iov_base;
1099 head->num = cpu_to_le32(0);
1100 msg->front.iov_len = sizeof(*head);
1101 spin_lock(&session->s_cap_lock);
1102 list_add(&msg->list_head, &session->s_cap_releases);
1103 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1104 }
1105
1106 if (!list_empty(&session->s_cap_releases)) {
1107 msg = list_first_entry(&session->s_cap_releases,
1108 struct ceph_msg,
1109 list_head);
1110 head = msg->front.iov_base;
1111 if (head->num) {
1112 dout(" queueing non-full %p (%d)\n", msg,
1113 le32_to_cpu(head->num));
1114 list_move_tail(&msg->list_head,
1115 &session->s_cap_releases_done);
1116 session->s_num_cap_releases -=
1117 CEPH_CAPS_PER_RELEASE - le32_to_cpu(head->num);
1118 }
1119 }
1120 err = 0;
1121 spin_unlock(&session->s_cap_lock);
1122out_unlocked:
1123 return err;
1124}
1125
1126/*
1127 * flush all dirty inode data to disk.
1128 *
1129 * returns true if we've flushed through want_flush_seq
1130 */
1131static int check_cap_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
1132{
1133 int mds, ret = 1;
1134
1135 dout("check_cap_flush want %lld\n", want_flush_seq);
1136 mutex_lock(&mdsc->mutex);
1137 for (mds = 0; ret && mds < mdsc->max_sessions; mds++) {
1138 struct ceph_mds_session *session = mdsc->sessions[mds];
1139
1140 if (!session)
1141 continue;
1142 get_session(session);
1143 mutex_unlock(&mdsc->mutex);
1144
1145 mutex_lock(&session->s_mutex);
1146 if (!list_empty(&session->s_cap_flushing)) {
1147 struct ceph_inode_info *ci =
1148 list_entry(session->s_cap_flushing.next,
1149 struct ceph_inode_info,
1150 i_flushing_item);
1151 struct inode *inode = &ci->vfs_inode;
1152
1153 spin_lock(&inode->i_lock);
1154 if (ci->i_cap_flush_seq <= want_flush_seq) {
1155 dout("check_cap_flush still flushing %p "
1156 "seq %lld <= %lld to mds%d\n", inode,
1157 ci->i_cap_flush_seq, want_flush_seq,
1158 session->s_mds);
1159 ret = 0;
1160 }
1161 spin_unlock(&inode->i_lock);
1162 }
1163 mutex_unlock(&session->s_mutex);
1164 ceph_put_mds_session(session);
1165
1166 if (!ret)
1167 return ret;
1168 mutex_lock(&mdsc->mutex);
1169 }
1170
1171 mutex_unlock(&mdsc->mutex);
1172 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
1173 return ret;
1174}
1175
1176/*
1177 * called under s_mutex
1178 */
Sage Weil3d7ded42010-06-09 16:47:10 -07001179void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1180 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07001181{
1182 struct ceph_msg *msg;
1183
1184 dout("send_cap_releases mds%d\n", session->s_mds);
Sage Weil0f8605f2010-05-05 15:51:35 -07001185 spin_lock(&session->s_cap_lock);
1186 while (!list_empty(&session->s_cap_releases_done)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001187 msg = list_first_entry(&session->s_cap_releases_done,
1188 struct ceph_msg, list_head);
1189 list_del_init(&msg->list_head);
1190 spin_unlock(&session->s_cap_lock);
1191 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1192 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1193 ceph_con_send(&session->s_con, msg);
Sage Weil0f8605f2010-05-05 15:51:35 -07001194 spin_lock(&session->s_cap_lock);
Sage Weil2f2dc052009-10-06 11:31:09 -07001195 }
1196 spin_unlock(&session->s_cap_lock);
1197}
1198
Sage Weile01a5942010-05-10 15:36:44 -07001199static void discard_cap_releases(struct ceph_mds_client *mdsc,
1200 struct ceph_mds_session *session)
1201{
1202 struct ceph_msg *msg;
1203 struct ceph_mds_cap_release *head;
1204 unsigned num;
1205
1206 dout("discard_cap_releases mds%d\n", session->s_mds);
1207 spin_lock(&session->s_cap_lock);
1208
1209 /* zero out the in-progress message */
1210 msg = list_first_entry(&session->s_cap_releases,
1211 struct ceph_msg, list_head);
1212 head = msg->front.iov_base;
1213 num = le32_to_cpu(head->num);
1214 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg, num);
1215 head->num = cpu_to_le32(0);
1216 session->s_num_cap_releases += num;
1217
1218 /* requeue completed messages */
1219 while (!list_empty(&session->s_cap_releases_done)) {
1220 msg = list_first_entry(&session->s_cap_releases_done,
1221 struct ceph_msg, list_head);
1222 list_del_init(&msg->list_head);
1223
1224 head = msg->front.iov_base;
1225 num = le32_to_cpu(head->num);
1226 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1227 num);
1228 session->s_num_cap_releases += num;
1229 head->num = cpu_to_le32(0);
1230 msg->front.iov_len = sizeof(*head);
1231 list_add(&msg->list_head, &session->s_cap_releases);
1232 }
1233
1234 spin_unlock(&session->s_cap_lock);
1235}
1236
Sage Weil2f2dc052009-10-06 11:31:09 -07001237/*
1238 * requests
1239 */
1240
1241/*
1242 * Create an mds request.
1243 */
1244struct ceph_mds_request *
1245ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1246{
1247 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1248
1249 if (!req)
1250 return ERR_PTR(-ENOMEM);
1251
Sage Weilb4556392010-05-13 12:01:13 -07001252 mutex_init(&req->r_fill_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07001253 req->r_started = jiffies;
1254 req->r_resend_mds = -1;
1255 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1256 req->r_fmode = -1;
Sage Weil153c8e62009-12-07 12:31:09 -08001257 kref_init(&req->r_kref);
Sage Weil2f2dc052009-10-06 11:31:09 -07001258 INIT_LIST_HEAD(&req->r_wait);
1259 init_completion(&req->r_completion);
1260 init_completion(&req->r_safe_completion);
1261 INIT_LIST_HEAD(&req->r_unsafe_item);
1262
1263 req->r_op = op;
1264 req->r_direct_mode = mode;
1265 return req;
1266}
1267
1268/*
Sage Weil44ca18f2010-02-15 12:08:46 -08001269 * return oldest (lowest) request, tid in request tree, 0 if none.
Sage Weil2f2dc052009-10-06 11:31:09 -07001270 *
1271 * called under mdsc->mutex.
1272 */
Sage Weil44ca18f2010-02-15 12:08:46 -08001273static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1274{
1275 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1276 return NULL;
1277 return rb_entry(rb_first(&mdsc->request_tree),
1278 struct ceph_mds_request, r_node);
1279}
1280
Sage Weil2f2dc052009-10-06 11:31:09 -07001281static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1282{
Sage Weil44ca18f2010-02-15 12:08:46 -08001283 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1284
1285 if (req)
1286 return req->r_tid;
1287 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07001288}
1289
1290/*
1291 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1292 * on build_path_from_dentry in fs/cifs/dir.c.
1293 *
1294 * If @stop_on_nosnap, generate path relative to the first non-snapped
1295 * inode.
1296 *
1297 * Encode hidden .snap dirs as a double /, i.e.
1298 * foo/.snap/bar -> foo//bar
1299 */
1300char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1301 int stop_on_nosnap)
1302{
1303 struct dentry *temp;
1304 char *path;
1305 int len, pos;
1306
1307 if (dentry == NULL)
1308 return ERR_PTR(-EINVAL);
1309
1310retry:
1311 len = 0;
1312 for (temp = dentry; !IS_ROOT(temp);) {
1313 struct inode *inode = temp->d_inode;
1314 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1315 len++; /* slash only */
1316 else if (stop_on_nosnap && inode &&
1317 ceph_snap(inode) == CEPH_NOSNAP)
1318 break;
1319 else
1320 len += 1 + temp->d_name.len;
1321 temp = temp->d_parent;
1322 if (temp == NULL) {
Sage Weil6c99f252010-05-10 16:12:25 -07001323 pr_err("build_path corrupt dentry %p\n", dentry);
Sage Weil2f2dc052009-10-06 11:31:09 -07001324 return ERR_PTR(-EINVAL);
1325 }
1326 }
1327 if (len)
1328 len--; /* no leading '/' */
1329
1330 path = kmalloc(len+1, GFP_NOFS);
1331 if (path == NULL)
1332 return ERR_PTR(-ENOMEM);
1333 pos = len;
1334 path[pos] = 0; /* trailing null */
1335 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1336 struct inode *inode = temp->d_inode;
1337
1338 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weil104648a2010-03-18 10:14:30 -07001339 dout("build_path path+%d: %p SNAPDIR\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07001340 pos, temp);
1341 } else if (stop_on_nosnap && inode &&
1342 ceph_snap(inode) == CEPH_NOSNAP) {
1343 break;
1344 } else {
1345 pos -= temp->d_name.len;
1346 if (pos < 0)
1347 break;
1348 strncpy(path + pos, temp->d_name.name,
1349 temp->d_name.len);
Sage Weil2f2dc052009-10-06 11:31:09 -07001350 }
1351 if (pos)
1352 path[--pos] = '/';
1353 temp = temp->d_parent;
1354 if (temp == NULL) {
Sage Weil104648a2010-03-18 10:14:30 -07001355 pr_err("build_path corrupt dentry\n");
Sage Weil2f2dc052009-10-06 11:31:09 -07001356 kfree(path);
1357 return ERR_PTR(-EINVAL);
1358 }
1359 }
1360 if (pos != 0) {
Sage Weil104648a2010-03-18 10:14:30 -07001361 pr_err("build_path did not end path lookup where "
Sage Weil2f2dc052009-10-06 11:31:09 -07001362 "expected, namelen is %d, pos is %d\n", len, pos);
1363 /* presumably this is only possible if racing with a
1364 rename of one of the parent directories (we can not
1365 lock the dentries above us to prevent this, but
1366 retrying should be harmless) */
1367 kfree(path);
1368 goto retry;
1369 }
1370
1371 *base = ceph_ino(temp->d_inode);
1372 *plen = len;
Sage Weil104648a2010-03-18 10:14:30 -07001373 dout("build_path on %p %d built %llx '%.*s'\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07001374 dentry, atomic_read(&dentry->d_count), *base, len, path);
1375 return path;
1376}
1377
1378static int build_dentry_path(struct dentry *dentry,
1379 const char **ppath, int *ppathlen, u64 *pino,
1380 int *pfreepath)
1381{
1382 char *path;
1383
1384 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1385 *pino = ceph_ino(dentry->d_parent->d_inode);
1386 *ppath = dentry->d_name.name;
1387 *ppathlen = dentry->d_name.len;
1388 return 0;
1389 }
1390 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1391 if (IS_ERR(path))
1392 return PTR_ERR(path);
1393 *ppath = path;
1394 *pfreepath = 1;
1395 return 0;
1396}
1397
1398static int build_inode_path(struct inode *inode,
1399 const char **ppath, int *ppathlen, u64 *pino,
1400 int *pfreepath)
1401{
1402 struct dentry *dentry;
1403 char *path;
1404
1405 if (ceph_snap(inode) == CEPH_NOSNAP) {
1406 *pino = ceph_ino(inode);
1407 *ppathlen = 0;
1408 return 0;
1409 }
1410 dentry = d_find_alias(inode);
1411 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1412 dput(dentry);
1413 if (IS_ERR(path))
1414 return PTR_ERR(path);
1415 *ppath = path;
1416 *pfreepath = 1;
1417 return 0;
1418}
1419
1420/*
1421 * request arguments may be specified via an inode *, a dentry *, or
1422 * an explicit ino+path.
1423 */
1424static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1425 const char *rpath, u64 rino,
1426 const char **ppath, int *pathlen,
1427 u64 *ino, int *freepath)
1428{
1429 int r = 0;
1430
1431 if (rinode) {
1432 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1433 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1434 ceph_snap(rinode));
1435 } else if (rdentry) {
1436 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1437 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1438 *ppath);
1439 } else if (rpath) {
1440 *ino = rino;
1441 *ppath = rpath;
1442 *pathlen = strlen(rpath);
1443 dout(" path %.*s\n", *pathlen, rpath);
1444 }
1445
1446 return r;
1447}
1448
1449/*
1450 * called under mdsc->mutex
1451 */
1452static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1453 struct ceph_mds_request *req,
1454 int mds)
1455{
1456 struct ceph_msg *msg;
1457 struct ceph_mds_request_head *head;
1458 const char *path1 = NULL;
1459 const char *path2 = NULL;
1460 u64 ino1 = 0, ino2 = 0;
1461 int pathlen1 = 0, pathlen2 = 0;
1462 int freepath1 = 0, freepath2 = 0;
1463 int len;
1464 u16 releases;
1465 void *p, *end;
1466 int ret;
1467
1468 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1469 req->r_path1, req->r_ino1.ino,
1470 &path1, &pathlen1, &ino1, &freepath1);
1471 if (ret < 0) {
1472 msg = ERR_PTR(ret);
1473 goto out;
1474 }
1475
1476 ret = set_request_path_attr(NULL, req->r_old_dentry,
1477 req->r_path2, req->r_ino2.ino,
1478 &path2, &pathlen2, &ino2, &freepath2);
1479 if (ret < 0) {
1480 msg = ERR_PTR(ret);
1481 goto out_free1;
1482 }
1483
1484 len = sizeof(*head) +
Sage Weilac8839d2010-01-27 14:28:10 -08001485 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
Sage Weil2f2dc052009-10-06 11:31:09 -07001486
1487 /* calculate (max) length for cap releases */
1488 len += sizeof(struct ceph_mds_request_release) *
1489 (!!req->r_inode_drop + !!req->r_dentry_drop +
1490 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1491 if (req->r_dentry_drop)
1492 len += req->r_dentry->d_name.len;
1493 if (req->r_old_dentry_drop)
1494 len += req->r_old_dentry->d_name.len;
1495
Yehuda Sadeh34d23762010-04-06 14:33:58 -07001496 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -07001497 if (!msg) {
1498 msg = ERR_PTR(-ENOMEM);
Sage Weil2f2dc052009-10-06 11:31:09 -07001499 goto out_free2;
Sage Weila79832f2010-04-01 16:06:19 -07001500 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001501
Sage Weil6df058c2009-12-22 11:24:33 -08001502 msg->hdr.tid = cpu_to_le64(req->r_tid);
1503
Sage Weil2f2dc052009-10-06 11:31:09 -07001504 head = msg->front.iov_base;
1505 p = msg->front.iov_base + sizeof(*head);
1506 end = msg->front.iov_base + msg->front.iov_len;
1507
1508 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1509 head->op = cpu_to_le32(req->r_op);
1510 head->caller_uid = cpu_to_le32(current_fsuid());
1511 head->caller_gid = cpu_to_le32(current_fsgid());
1512 head->args = req->r_args;
1513
1514 ceph_encode_filepath(&p, end, ino1, path1);
1515 ceph_encode_filepath(&p, end, ino2, path2);
1516
1517 /* cap releases */
1518 releases = 0;
1519 if (req->r_inode_drop)
1520 releases += ceph_encode_inode_release(&p,
1521 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1522 mds, req->r_inode_drop, req->r_inode_unless, 0);
1523 if (req->r_dentry_drop)
1524 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1525 mds, req->r_dentry_drop, req->r_dentry_unless);
1526 if (req->r_old_dentry_drop)
1527 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1528 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1529 if (req->r_old_inode_drop)
1530 releases += ceph_encode_inode_release(&p,
1531 req->r_old_dentry->d_inode,
1532 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1533 head->num_releases = cpu_to_le16(releases);
1534
1535 BUG_ON(p > end);
1536 msg->front.iov_len = p - msg->front.iov_base;
1537 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1538
1539 msg->pages = req->r_pages;
1540 msg->nr_pages = req->r_num_pages;
1541 msg->hdr.data_len = cpu_to_le32(req->r_data_len);
1542 msg->hdr.data_off = cpu_to_le16(0);
1543
1544out_free2:
1545 if (freepath2)
1546 kfree((char *)path2);
1547out_free1:
1548 if (freepath1)
1549 kfree((char *)path1);
1550out:
1551 return msg;
1552}
1553
1554/*
1555 * called under mdsc->mutex if error, under no mutex if
1556 * success.
1557 */
1558static void complete_request(struct ceph_mds_client *mdsc,
1559 struct ceph_mds_request *req)
1560{
1561 if (req->r_callback)
1562 req->r_callback(mdsc, req);
1563 else
1564 complete(&req->r_completion);
1565}
1566
1567/*
1568 * called under mdsc->mutex
1569 */
1570static int __prepare_send_request(struct ceph_mds_client *mdsc,
1571 struct ceph_mds_request *req,
1572 int mds)
1573{
1574 struct ceph_mds_request_head *rhead;
1575 struct ceph_msg *msg;
1576 int flags = 0;
1577
1578 req->r_mds = mds;
1579 req->r_attempts++;
1580 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
1581 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
1582
1583 if (req->r_request) {
1584 ceph_msg_put(req->r_request);
1585 req->r_request = NULL;
1586 }
1587 msg = create_request_message(mdsc, req, mds);
1588 if (IS_ERR(msg)) {
Sage Weile1518c72010-05-13 11:19:06 -07001589 req->r_err = PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001590 complete_request(mdsc, req);
Sage Weila79832f2010-04-01 16:06:19 -07001591 return PTR_ERR(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001592 }
1593 req->r_request = msg;
1594
1595 rhead = msg->front.iov_base;
Sage Weil2f2dc052009-10-06 11:31:09 -07001596 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1597 if (req->r_got_unsafe)
1598 flags |= CEPH_MDS_FLAG_REPLAY;
1599 if (req->r_locked_dir)
1600 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1601 rhead->flags = cpu_to_le32(flags);
1602 rhead->num_fwd = req->r_num_fwd;
1603 rhead->num_retry = req->r_attempts - 1;
1604
1605 dout(" r_locked_dir = %p\n", req->r_locked_dir);
1606
1607 if (req->r_target_inode && req->r_got_unsafe)
1608 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1609 else
1610 rhead->ino = 0;
1611 return 0;
1612}
1613
1614/*
1615 * send request, or put it on the appropriate wait list.
1616 */
1617static int __do_request(struct ceph_mds_client *mdsc,
1618 struct ceph_mds_request *req)
1619{
1620 struct ceph_mds_session *session = NULL;
1621 int mds = -1;
1622 int err = -EAGAIN;
1623
Sage Weile1518c72010-05-13 11:19:06 -07001624 if (req->r_err || req->r_got_result)
Sage Weil2f2dc052009-10-06 11:31:09 -07001625 goto out;
1626
1627 if (req->r_timeout &&
1628 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1629 dout("do_request timed out\n");
1630 err = -EIO;
1631 goto finish;
1632 }
1633
1634 mds = __choose_mds(mdsc, req);
1635 if (mds < 0 ||
1636 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1637 dout("do_request no mds or not active, waiting for map\n");
1638 list_add(&req->r_wait, &mdsc->waiting_for_map);
1639 goto out;
1640 }
1641
1642 /* get, open session */
1643 session = __ceph_lookup_mds_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001644 if (!session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001645 session = register_session(mdsc, mds);
Sage Weil9c4239562010-03-20 20:43:28 -07001646 if (IS_ERR(session)) {
1647 err = PTR_ERR(session);
1648 goto finish;
1649 }
1650 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001651 dout("do_request mds%d session %p state %s\n", mds, session,
1652 session_state_name(session->s_state));
1653 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1654 session->s_state != CEPH_MDS_SESSION_HUNG) {
1655 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1656 session->s_state == CEPH_MDS_SESSION_CLOSING)
1657 __open_session(mdsc, session);
1658 list_add(&req->r_wait, &session->s_waiting);
1659 goto out_session;
1660 }
1661
1662 /* send request */
1663 req->r_session = get_session(session);
1664 req->r_resend_mds = -1; /* forget any previous mds hint */
1665
1666 if (req->r_request_started == 0) /* note request start time */
1667 req->r_request_started = jiffies;
1668
1669 err = __prepare_send_request(mdsc, req, mds);
1670 if (!err) {
1671 ceph_msg_get(req->r_request);
1672 ceph_con_send(&session->s_con, req->r_request);
1673 }
1674
1675out_session:
1676 ceph_put_mds_session(session);
1677out:
1678 return err;
1679
1680finish:
Sage Weile1518c72010-05-13 11:19:06 -07001681 req->r_err = err;
Sage Weil2f2dc052009-10-06 11:31:09 -07001682 complete_request(mdsc, req);
1683 goto out;
1684}
1685
1686/*
1687 * called under mdsc->mutex
1688 */
1689static void __wake_requests(struct ceph_mds_client *mdsc,
1690 struct list_head *head)
1691{
1692 struct ceph_mds_request *req, *nreq;
1693
1694 list_for_each_entry_safe(req, nreq, head, r_wait) {
1695 list_del_init(&req->r_wait);
1696 __do_request(mdsc, req);
1697 }
1698}
1699
1700/*
1701 * Wake up threads with requests pending for @mds, so that they can
Sage Weil29790f22010-03-18 14:45:05 -07001702 * resubmit their requests to a possibly different mds.
Sage Weil2f2dc052009-10-06 11:31:09 -07001703 */
Sage Weil29790f22010-03-18 14:45:05 -07001704static void kick_requests(struct ceph_mds_client *mdsc, int mds)
Sage Weil2f2dc052009-10-06 11:31:09 -07001705{
Sage Weil44ca18f2010-02-15 12:08:46 -08001706 struct ceph_mds_request *req;
1707 struct rb_node *p;
Sage Weil2f2dc052009-10-06 11:31:09 -07001708
1709 dout("kick_requests mds%d\n", mds);
Sage Weil44ca18f2010-02-15 12:08:46 -08001710 for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1711 req = rb_entry(p, struct ceph_mds_request, r_node);
1712 if (req->r_got_unsafe)
1713 continue;
1714 if (req->r_session &&
1715 req->r_session->s_mds == mds) {
1716 dout(" kicking tid %llu\n", req->r_tid);
1717 put_request_session(req);
1718 __do_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07001719 }
1720 }
1721}
1722
1723void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1724 struct ceph_mds_request *req)
1725{
1726 dout("submit_request on %p\n", req);
1727 mutex_lock(&mdsc->mutex);
1728 __register_request(mdsc, req, NULL);
1729 __do_request(mdsc, req);
1730 mutex_unlock(&mdsc->mutex);
1731}
1732
1733/*
1734 * Synchrously perform an mds request. Take care of all of the
1735 * session setup, forwarding, retry details.
1736 */
1737int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
1738 struct inode *dir,
1739 struct ceph_mds_request *req)
1740{
1741 int err;
1742
1743 dout("do_request on %p\n", req);
1744
1745 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
1746 if (req->r_inode)
1747 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
1748 if (req->r_locked_dir)
1749 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
1750 if (req->r_old_dentry)
1751 ceph_get_cap_refs(
1752 ceph_inode(req->r_old_dentry->d_parent->d_inode),
1753 CEPH_CAP_PIN);
1754
1755 /* issue */
1756 mutex_lock(&mdsc->mutex);
1757 __register_request(mdsc, req, dir);
1758 __do_request(mdsc, req);
1759
Sage Weile1518c72010-05-13 11:19:06 -07001760 if (req->r_err) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001761 err = req->r_err;
Sage Weile1518c72010-05-13 11:19:06 -07001762 __unregister_request(mdsc, req);
1763 dout("do_request early error %d\n", err);
1764 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -07001765 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001766
Sage Weile1518c72010-05-13 11:19:06 -07001767 /* wait */
1768 mutex_unlock(&mdsc->mutex);
1769 dout("do_request waiting\n");
1770 if (req->r_timeout) {
Sage Weilaa916472010-05-24 11:15:51 -07001771 err = (long)wait_for_completion_killable_timeout(
Sage Weile1518c72010-05-13 11:19:06 -07001772 &req->r_completion, req->r_timeout);
1773 if (err == 0)
1774 err = -EIO;
1775 } else {
Sage Weilaa916472010-05-24 11:15:51 -07001776 err = wait_for_completion_killable(&req->r_completion);
Sage Weile1518c72010-05-13 11:19:06 -07001777 }
1778 dout("do_request waited, got %d\n", err);
1779 mutex_lock(&mdsc->mutex);
1780
1781 /* only abort if we didn't race with a real reply */
1782 if (req->r_got_result) {
1783 err = le32_to_cpu(req->r_reply_info.head->result);
1784 } else if (err < 0) {
1785 dout("aborted request %lld with %d\n", req->r_tid, err);
Sage Weilb4556392010-05-13 12:01:13 -07001786
1787 /*
1788 * ensure we aren't running concurrently with
1789 * ceph_fill_trace or ceph_readdir_prepopulate, which
1790 * rely on locks (dir mutex) held by our caller.
1791 */
1792 mutex_lock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07001793 req->r_err = err;
1794 req->r_aborted = true;
Sage Weilb4556392010-05-13 12:01:13 -07001795 mutex_unlock(&req->r_fill_mutex);
Sage Weile1518c72010-05-13 11:19:06 -07001796
1797 if (req->r_locked_dir &&
Sage Weil167c9e32010-05-14 10:02:57 -07001798 (req->r_op & CEPH_MDS_OP_WRITE))
1799 ceph_invalidate_dir_request(req);
Sage Weile1518c72010-05-13 11:19:06 -07001800 } else {
1801 err = req->r_err;
1802 }
1803
1804out:
1805 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07001806 dout("do_request %p done, result %d\n", req, err);
1807 return err;
1808}
1809
1810/*
Sage Weil167c9e32010-05-14 10:02:57 -07001811 * Invalidate dir I_COMPLETE, dentry lease state on an aborted MDS
1812 * namespace request.
1813 */
1814void ceph_invalidate_dir_request(struct ceph_mds_request *req)
1815{
1816 struct inode *inode = req->r_locked_dir;
1817 struct ceph_inode_info *ci = ceph_inode(inode);
1818
1819 dout("invalidate_dir_request %p (I_COMPLETE, lease(s))\n", inode);
1820 spin_lock(&inode->i_lock);
1821 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1822 ci->i_release_count++;
1823 spin_unlock(&inode->i_lock);
1824
1825 if (req->r_dentry)
1826 ceph_invalidate_dentry_lease(req->r_dentry);
1827 if (req->r_old_dentry)
1828 ceph_invalidate_dentry_lease(req->r_old_dentry);
1829}
1830
1831/*
Sage Weil2f2dc052009-10-06 11:31:09 -07001832 * Handle mds reply.
1833 *
1834 * We take the session mutex and parse and process the reply immediately.
1835 * This preserves the logical ordering of replies, capabilities, etc., sent
1836 * by the MDS as they are applied to our local cache.
1837 */
1838static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
1839{
1840 struct ceph_mds_client *mdsc = session->s_mdsc;
1841 struct ceph_mds_request *req;
1842 struct ceph_mds_reply_head *head = msg->front.iov_base;
1843 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
1844 u64 tid;
1845 int err, result;
Sage Weil2600d2d2010-02-22 15:12:16 -08001846 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07001847
Sage Weil2f2dc052009-10-06 11:31:09 -07001848 if (msg->front.iov_len < sizeof(*head)) {
1849 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08001850 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001851 return;
1852 }
1853
1854 /* get request, session */
Sage Weil6df058c2009-12-22 11:24:33 -08001855 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07001856 mutex_lock(&mdsc->mutex);
1857 req = __lookup_request(mdsc, tid);
1858 if (!req) {
1859 dout("handle_reply on unknown tid %llu\n", tid);
1860 mutex_unlock(&mdsc->mutex);
1861 return;
1862 }
1863 dout("handle_reply %p\n", req);
Sage Weil2f2dc052009-10-06 11:31:09 -07001864
1865 /* correct session? */
Sage Weild96d6042010-03-20 20:50:58 -07001866 if (req->r_session != session) {
Sage Weil2f2dc052009-10-06 11:31:09 -07001867 pr_err("mdsc_handle_reply got %llu on session mds%d"
1868 " not mds%d\n", tid, session->s_mds,
1869 req->r_session ? req->r_session->s_mds : -1);
1870 mutex_unlock(&mdsc->mutex);
1871 goto out;
1872 }
1873
1874 /* dup? */
1875 if ((req->r_got_unsafe && !head->safe) ||
1876 (req->r_got_safe && head->safe)) {
1877 pr_warning("got a dup %s reply on %llu from mds%d\n",
1878 head->safe ? "safe" : "unsafe", tid, mds);
1879 mutex_unlock(&mdsc->mutex);
1880 goto out;
1881 }
Sage Weil85792d0d2010-05-13 09:06:02 -07001882 if (req->r_got_safe && !head->safe) {
1883 pr_warning("got unsafe after safe on %llu from mds%d\n",
1884 tid, mds);
1885 mutex_unlock(&mdsc->mutex);
1886 goto out;
1887 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001888
1889 result = le32_to_cpu(head->result);
1890
1891 /*
1892 * Tolerate 2 consecutive ESTALEs from the same mds.
1893 * FIXME: we should be looking at the cap migrate_seq.
1894 */
1895 if (result == -ESTALE) {
1896 req->r_direct_mode = USE_AUTH_MDS;
1897 req->r_num_stale++;
1898 if (req->r_num_stale <= 2) {
1899 __do_request(mdsc, req);
1900 mutex_unlock(&mdsc->mutex);
1901 goto out;
1902 }
1903 } else {
1904 req->r_num_stale = 0;
1905 }
1906
1907 if (head->safe) {
1908 req->r_got_safe = true;
1909 __unregister_request(mdsc, req);
1910 complete(&req->r_safe_completion);
1911
1912 if (req->r_got_unsafe) {
1913 /*
1914 * We already handled the unsafe response, now do the
1915 * cleanup. No need to examine the response; the MDS
1916 * doesn't include any result info in the safe
1917 * response. And even if it did, there is nothing
1918 * useful we could do with a revised return value.
1919 */
1920 dout("got safe reply %llu, mds%d\n", tid, mds);
1921 list_del_init(&req->r_unsafe_item);
1922
1923 /* last unsafe request during umount? */
Sage Weil44ca18f2010-02-15 12:08:46 -08001924 if (mdsc->stopping && !__get_oldest_req(mdsc))
Sage Weil2f2dc052009-10-06 11:31:09 -07001925 complete(&mdsc->safe_umount_waiters);
1926 mutex_unlock(&mdsc->mutex);
1927 goto out;
1928 }
Sage Weile1518c72010-05-13 11:19:06 -07001929 } else {
Sage Weil2f2dc052009-10-06 11:31:09 -07001930 req->r_got_unsafe = true;
1931 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
1932 }
1933
1934 dout("handle_reply tid %lld result %d\n", tid, result);
1935 rinfo = &req->r_reply_info;
1936 err = parse_reply_info(msg, rinfo);
1937 mutex_unlock(&mdsc->mutex);
1938
1939 mutex_lock(&session->s_mutex);
1940 if (err < 0) {
1941 pr_err("mdsc_handle_reply got corrupt reply mds%d\n", mds);
Sage Weil9ec7cab2009-12-14 15:13:47 -08001942 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07001943 goto out_err;
1944 }
1945
1946 /* snap trace */
1947 if (rinfo->snapblob_len) {
1948 down_write(&mdsc->snap_rwsem);
1949 ceph_update_snap_trace(mdsc, rinfo->snapblob,
1950 rinfo->snapblob + rinfo->snapblob_len,
1951 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
1952 downgrade_write(&mdsc->snap_rwsem);
1953 } else {
1954 down_read(&mdsc->snap_rwsem);
1955 }
1956
1957 /* insert trace into our cache */
Sage Weilb4556392010-05-13 12:01:13 -07001958 mutex_lock(&req->r_fill_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07001959 err = ceph_fill_trace(mdsc->client->sb, req, req->r_session);
1960 if (err == 0) {
1961 if (result == 0 && rinfo->dir_nr)
1962 ceph_readdir_prepopulate(req, req->r_session);
1963 ceph_unreserve_caps(&req->r_caps_reservation);
1964 }
Sage Weilb4556392010-05-13 12:01:13 -07001965 mutex_unlock(&req->r_fill_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07001966
1967 up_read(&mdsc->snap_rwsem);
1968out_err:
Sage Weile1518c72010-05-13 11:19:06 -07001969 mutex_lock(&mdsc->mutex);
1970 if (!req->r_aborted) {
1971 if (err) {
1972 req->r_err = err;
1973 } else {
1974 req->r_reply = msg;
1975 ceph_msg_get(msg);
1976 req->r_got_result = true;
1977 }
Sage Weil2f2dc052009-10-06 11:31:09 -07001978 } else {
Sage Weile1518c72010-05-13 11:19:06 -07001979 dout("reply arrived after request %lld was aborted\n", tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07001980 }
Sage Weile1518c72010-05-13 11:19:06 -07001981 mutex_unlock(&mdsc->mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07001982
Sage Weil2b2300d2010-06-09 16:52:04 -07001983 ceph_add_cap_releases(mdsc, req->r_session, -1);
Sage Weil2f2dc052009-10-06 11:31:09 -07001984 mutex_unlock(&session->s_mutex);
1985
1986 /* kick calling process */
1987 complete_request(mdsc, req);
1988out:
1989 ceph_mdsc_put_request(req);
1990 return;
1991}
1992
1993
1994
1995/*
1996 * handle mds notification that our request has been forwarded.
1997 */
Sage Weil2600d2d2010-02-22 15:12:16 -08001998static void handle_forward(struct ceph_mds_client *mdsc,
1999 struct ceph_mds_session *session,
2000 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002001{
2002 struct ceph_mds_request *req;
Sage Weila1ea7872010-02-23 14:02:44 -08002003 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weil2f2dc052009-10-06 11:31:09 -07002004 u32 next_mds;
2005 u32 fwd_seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002006 int err = -EINVAL;
2007 void *p = msg->front.iov_base;
2008 void *end = p + msg->front.iov_len;
Sage Weil2f2dc052009-10-06 11:31:09 -07002009
Sage Weila1ea7872010-02-23 14:02:44 -08002010 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -07002011 next_mds = ceph_decode_32(&p);
2012 fwd_seq = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07002013
2014 mutex_lock(&mdsc->mutex);
2015 req = __lookup_request(mdsc, tid);
2016 if (!req) {
Sage Weil2a8e5e32010-05-28 16:43:16 -07002017 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002018 goto out; /* dup reply? */
2019 }
2020
Sage Weil2a8e5e32010-05-28 16:43:16 -07002021 if (req->r_aborted) {
2022 dout("forward tid %llu aborted, unregistering\n", tid);
2023 __unregister_request(mdsc, req);
2024 } else if (fwd_seq <= req->r_num_fwd) {
2025 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
Sage Weil2f2dc052009-10-06 11:31:09 -07002026 tid, next_mds, req->r_num_fwd, fwd_seq);
2027 } else {
2028 /* resend. forward race not possible; mds would drop */
Sage Weil2a8e5e32010-05-28 16:43:16 -07002029 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2030 BUG_ON(req->r_err);
2031 BUG_ON(req->r_got_result);
Sage Weil2f2dc052009-10-06 11:31:09 -07002032 req->r_num_fwd = fwd_seq;
2033 req->r_resend_mds = next_mds;
2034 put_request_session(req);
2035 __do_request(mdsc, req);
2036 }
2037 ceph_mdsc_put_request(req);
2038out:
2039 mutex_unlock(&mdsc->mutex);
2040 return;
2041
2042bad:
2043 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2044}
2045
2046/*
2047 * handle a mds session control message
2048 */
2049static void handle_session(struct ceph_mds_session *session,
2050 struct ceph_msg *msg)
2051{
2052 struct ceph_mds_client *mdsc = session->s_mdsc;
2053 u32 op;
2054 u64 seq;
Sage Weil2600d2d2010-02-22 15:12:16 -08002055 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002056 struct ceph_mds_session_head *h = msg->front.iov_base;
2057 int wake = 0;
2058
Sage Weil2f2dc052009-10-06 11:31:09 -07002059 /* decode */
2060 if (msg->front.iov_len != sizeof(*h))
2061 goto bad;
2062 op = le32_to_cpu(h->op);
2063 seq = le64_to_cpu(h->seq);
2064
2065 mutex_lock(&mdsc->mutex);
Sage Weil2600d2d2010-02-22 15:12:16 -08002066 if (op == CEPH_SESSION_CLOSE)
2067 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002068 /* FIXME: this ttl calculation is generous */
2069 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2070 mutex_unlock(&mdsc->mutex);
2071
2072 mutex_lock(&session->s_mutex);
2073
2074 dout("handle_session mds%d %s %p state %s seq %llu\n",
2075 mds, ceph_session_op_name(op), session,
2076 session_state_name(session->s_state), seq);
2077
2078 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2079 session->s_state = CEPH_MDS_SESSION_OPEN;
2080 pr_info("mds%d came back\n", session->s_mds);
2081 }
2082
2083 switch (op) {
2084 case CEPH_SESSION_OPEN:
Sage Weil29790f22010-03-18 14:45:05 -07002085 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2086 pr_info("mds%d reconnect success\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002087 session->s_state = CEPH_MDS_SESSION_OPEN;
2088 renewed_caps(mdsc, session, 0);
2089 wake = 1;
2090 if (mdsc->stopping)
2091 __close_session(mdsc, session);
2092 break;
2093
2094 case CEPH_SESSION_RENEWCAPS:
2095 if (session->s_renew_seq == seq)
2096 renewed_caps(mdsc, session, 1);
2097 break;
2098
2099 case CEPH_SESSION_CLOSE:
Sage Weil29790f22010-03-18 14:45:05 -07002100 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2101 pr_info("mds%d reconnect denied\n", session->s_mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002102 remove_session_caps(session);
2103 wake = 1; /* for good measure */
2104 complete(&mdsc->session_close_waiters);
Sage Weil29790f22010-03-18 14:45:05 -07002105 kick_requests(mdsc, mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002106 break;
2107
2108 case CEPH_SESSION_STALE:
2109 pr_info("mds%d caps went stale, renewing\n",
2110 session->s_mds);
2111 spin_lock(&session->s_cap_lock);
2112 session->s_cap_gen++;
2113 session->s_cap_ttl = 0;
2114 spin_unlock(&session->s_cap_lock);
2115 send_renew_caps(mdsc, session);
2116 break;
2117
2118 case CEPH_SESSION_RECALL_STATE:
2119 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2120 break;
2121
2122 default:
2123 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2124 WARN_ON(1);
2125 }
2126
2127 mutex_unlock(&session->s_mutex);
2128 if (wake) {
2129 mutex_lock(&mdsc->mutex);
2130 __wake_requests(mdsc, &session->s_waiting);
2131 mutex_unlock(&mdsc->mutex);
2132 }
2133 return;
2134
2135bad:
2136 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2137 (int)msg->front.iov_len);
Sage Weil9ec7cab2009-12-14 15:13:47 -08002138 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002139 return;
2140}
2141
2142
2143/*
2144 * called under session->mutex.
2145 */
2146static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2147 struct ceph_mds_session *session)
2148{
2149 struct ceph_mds_request *req, *nreq;
2150 int err;
2151
2152 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2153
2154 mutex_lock(&mdsc->mutex);
2155 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2156 err = __prepare_send_request(mdsc, req, session->s_mds);
2157 if (!err) {
2158 ceph_msg_get(req->r_request);
2159 ceph_con_send(&session->s_con, req->r_request);
2160 }
2161 }
2162 mutex_unlock(&mdsc->mutex);
2163}
2164
2165/*
2166 * Encode information about a cap for a reconnect with the MDS.
2167 */
Sage Weil2f2dc052009-10-06 11:31:09 -07002168static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2169 void *arg)
2170{
Sage Weil93cea5b2009-12-23 12:21:51 -08002171 struct ceph_mds_cap_reconnect rec;
Sage Weil2f2dc052009-10-06 11:31:09 -07002172 struct ceph_inode_info *ci;
Sage Weil93cea5b2009-12-23 12:21:51 -08002173 struct ceph_pagelist *pagelist = arg;
Sage Weil2f2dc052009-10-06 11:31:09 -07002174 char *path;
2175 int pathlen, err;
2176 u64 pathbase;
2177 struct dentry *dentry;
2178
2179 ci = cap->ci;
2180
2181 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2182 inode, ceph_vinop(inode), cap, cap->cap_id,
2183 ceph_cap_string(cap->issued));
Sage Weil93cea5b2009-12-23 12:21:51 -08002184 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2185 if (err)
2186 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002187
2188 dentry = d_find_alias(inode);
2189 if (dentry) {
2190 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2191 if (IS_ERR(path)) {
2192 err = PTR_ERR(path);
2193 BUG_ON(err);
2194 }
2195 } else {
2196 path = NULL;
2197 pathlen = 0;
2198 }
Sage Weil93cea5b2009-12-23 12:21:51 -08002199 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2200 if (err)
2201 goto out;
Sage Weil2f2dc052009-10-06 11:31:09 -07002202
Sage Weil2f2dc052009-10-06 11:31:09 -07002203 spin_lock(&inode->i_lock);
2204 cap->seq = 0; /* reset cap seq */
2205 cap->issue_seq = 0; /* and issue_seq */
Sage Weil93cea5b2009-12-23 12:21:51 -08002206 rec.cap_id = cpu_to_le64(cap->cap_id);
2207 rec.pathbase = cpu_to_le64(pathbase);
2208 rec.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2209 rec.issued = cpu_to_le32(cap->issued);
2210 rec.size = cpu_to_le64(inode->i_size);
2211 ceph_encode_timespec(&rec.mtime, &inode->i_mtime);
2212 ceph_encode_timespec(&rec.atime, &inode->i_atime);
2213 rec.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
Sage Weil2f2dc052009-10-06 11:31:09 -07002214 spin_unlock(&inode->i_lock);
2215
Sage Weil93cea5b2009-12-23 12:21:51 -08002216 err = ceph_pagelist_append(pagelist, &rec, sizeof(rec));
2217
2218out:
Sage Weil2f2dc052009-10-06 11:31:09 -07002219 kfree(path);
2220 dput(dentry);
Sage Weil93cea5b2009-12-23 12:21:51 -08002221 return err;
Sage Weil2f2dc052009-10-06 11:31:09 -07002222}
2223
2224
2225/*
2226 * If an MDS fails and recovers, clients need to reconnect in order to
2227 * reestablish shared state. This includes all caps issued through
2228 * this session _and_ the snap_realm hierarchy. Because it's not
2229 * clear which snap realms the mds cares about, we send everything we
2230 * know about.. that ensures we'll then get any new info the
2231 * recovering MDS might have.
2232 *
2233 * This is a relatively heavyweight operation, but it's rare.
2234 *
2235 * called with mdsc->mutex held.
2236 */
Sage Weil34b6c852010-05-10 16:31:25 -07002237static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2238 struct ceph_mds_session *session)
Sage Weil2f2dc052009-10-06 11:31:09 -07002239{
Sage Weil2f2dc052009-10-06 11:31:09 -07002240 struct ceph_msg *reply;
Sage Weila105f002010-02-15 14:37:55 -08002241 struct rb_node *p;
Sage Weil34b6c852010-05-10 16:31:25 -07002242 int mds = session->s_mds;
Sage Weil9abf82b2010-05-10 21:58:38 -07002243 int err = -ENOMEM;
Sage Weil93cea5b2009-12-23 12:21:51 -08002244 struct ceph_pagelist *pagelist;
Sage Weil2f2dc052009-10-06 11:31:09 -07002245
Sage Weil34b6c852010-05-10 16:31:25 -07002246 pr_info("mds%d reconnect start\n", mds);
Sage Weil2f2dc052009-10-06 11:31:09 -07002247
Sage Weil93cea5b2009-12-23 12:21:51 -08002248 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2249 if (!pagelist)
2250 goto fail_nopagelist;
2251 ceph_pagelist_init(pagelist);
2252
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002253 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -07002254 if (!reply)
Sage Weil93cea5b2009-12-23 12:21:51 -08002255 goto fail_nomsg;
Sage Weil93cea5b2009-12-23 12:21:51 -08002256
Sage Weil34b6c852010-05-10 16:31:25 -07002257 mutex_lock(&session->s_mutex);
2258 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2259 session->s_seq = 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07002260
Sage Weil34b6c852010-05-10 16:31:25 -07002261 ceph_con_open(&session->s_con,
2262 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
Sage Weil2f2dc052009-10-06 11:31:09 -07002263
Sage Weil34b6c852010-05-10 16:31:25 -07002264 /* replay unsafe requests */
2265 replay_unsafe_requests(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002266
2267 down_read(&mdsc->snap_rwsem);
2268
Sage Weil2f2dc052009-10-06 11:31:09 -07002269 dout("session %p state %s\n", session,
2270 session_state_name(session->s_state));
2271
Sage Weile01a5942010-05-10 15:36:44 -07002272 /* drop old cap expires; we're about to reestablish that state */
2273 discard_cap_releases(mdsc, session);
2274
Sage Weil2f2dc052009-10-06 11:31:09 -07002275 /* traverse this session's caps */
Sage Weil93cea5b2009-12-23 12:21:51 -08002276 err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2277 if (err)
2278 goto fail;
2279 err = iterate_session_caps(session, encode_caps_cb, pagelist);
Sage Weil2f2dc052009-10-06 11:31:09 -07002280 if (err < 0)
Sage Weil9abf82b2010-05-10 21:58:38 -07002281 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002282
2283 /*
2284 * snaprealms. we provide mds with the ino, seq (version), and
2285 * parent for all of our realms. If the mds has any newer info,
2286 * it will tell us.
2287 */
Sage Weila105f002010-02-15 14:37:55 -08002288 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2289 struct ceph_snap_realm *realm =
2290 rb_entry(p, struct ceph_snap_realm, node);
Sage Weil93cea5b2009-12-23 12:21:51 -08002291 struct ceph_mds_snaprealm_reconnect sr_rec;
Sage Weil2f2dc052009-10-06 11:31:09 -07002292
2293 dout(" adding snap realm %llx seq %lld parent %llx\n",
2294 realm->ino, realm->seq, realm->parent_ino);
Sage Weil93cea5b2009-12-23 12:21:51 -08002295 sr_rec.ino = cpu_to_le64(realm->ino);
2296 sr_rec.seq = cpu_to_le64(realm->seq);
2297 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2298 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2299 if (err)
2300 goto fail;
Sage Weil2f2dc052009-10-06 11:31:09 -07002301 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002302
Sage Weil93cea5b2009-12-23 12:21:51 -08002303 reply->pagelist = pagelist;
2304 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2305 reply->nr_pages = calc_pages_for(0, pagelist->length);
Sage Weil2f2dc052009-10-06 11:31:09 -07002306 ceph_con_send(&session->s_con, reply);
2307
Sage Weil9abf82b2010-05-10 21:58:38 -07002308 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002309
Sage Weil9abf82b2010-05-10 21:58:38 -07002310 mutex_lock(&mdsc->mutex);
2311 __wake_requests(mdsc, &session->s_waiting);
2312 mutex_unlock(&mdsc->mutex);
2313
Sage Weil2f2dc052009-10-06 11:31:09 -07002314 up_read(&mdsc->snap_rwsem);
Sage Weil2f2dc052009-10-06 11:31:09 -07002315 return;
2316
Sage Weil93cea5b2009-12-23 12:21:51 -08002317fail:
Sage Weil2f2dc052009-10-06 11:31:09 -07002318 ceph_msg_put(reply);
Sage Weil9abf82b2010-05-10 21:58:38 -07002319 up_read(&mdsc->snap_rwsem);
2320 mutex_unlock(&session->s_mutex);
Sage Weil93cea5b2009-12-23 12:21:51 -08002321fail_nomsg:
2322 ceph_pagelist_release(pagelist);
2323 kfree(pagelist);
2324fail_nopagelist:
Sage Weil9abf82b2010-05-10 21:58:38 -07002325 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
Sage Weil9abf82b2010-05-10 21:58:38 -07002326 return;
Sage Weil2f2dc052009-10-06 11:31:09 -07002327}
2328
2329
2330/*
2331 * compare old and new mdsmaps, kicking requests
2332 * and closing out old connections as necessary
2333 *
2334 * called under mdsc->mutex.
2335 */
2336static void check_new_map(struct ceph_mds_client *mdsc,
2337 struct ceph_mdsmap *newmap,
2338 struct ceph_mdsmap *oldmap)
2339{
2340 int i;
2341 int oldstate, newstate;
2342 struct ceph_mds_session *s;
2343
2344 dout("check_new_map new %u old %u\n",
2345 newmap->m_epoch, oldmap->m_epoch);
2346
2347 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2348 if (mdsc->sessions[i] == NULL)
2349 continue;
2350 s = mdsc->sessions[i];
2351 oldstate = ceph_mdsmap_get_state(oldmap, i);
2352 newstate = ceph_mdsmap_get_state(newmap, i);
2353
2354 dout("check_new_map mds%d state %s -> %s (session %s)\n",
2355 i, ceph_mds_state_name(oldstate),
2356 ceph_mds_state_name(newstate),
2357 session_state_name(s->s_state));
2358
2359 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
2360 ceph_mdsmap_get_addr(newmap, i),
2361 sizeof(struct ceph_entity_addr))) {
2362 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2363 /* the session never opened, just close it
2364 * out now */
2365 __wake_requests(mdsc, &s->s_waiting);
Sage Weil2600d2d2010-02-22 15:12:16 -08002366 __unregister_session(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07002367 } else {
2368 /* just close it */
2369 mutex_unlock(&mdsc->mutex);
2370 mutex_lock(&s->s_mutex);
2371 mutex_lock(&mdsc->mutex);
2372 ceph_con_close(&s->s_con);
2373 mutex_unlock(&s->s_mutex);
2374 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2375 }
2376
2377 /* kick any requests waiting on the recovering mds */
Sage Weil29790f22010-03-18 14:45:05 -07002378 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002379 } else if (oldstate == newstate) {
2380 continue; /* nothing new with this mds */
2381 }
2382
2383 /*
2384 * send reconnect?
2385 */
2386 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
Sage Weil34b6c852010-05-10 16:31:25 -07002387 newstate >= CEPH_MDS_STATE_RECONNECT) {
2388 mutex_unlock(&mdsc->mutex);
2389 send_mds_reconnect(mdsc, s);
2390 mutex_lock(&mdsc->mutex);
2391 }
Sage Weil2f2dc052009-10-06 11:31:09 -07002392
2393 /*
Sage Weil29790f22010-03-18 14:45:05 -07002394 * kick request on any mds that has gone active.
Sage Weil2f2dc052009-10-06 11:31:09 -07002395 */
2396 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2397 newstate >= CEPH_MDS_STATE_ACTIVE) {
Sage Weil29790f22010-03-18 14:45:05 -07002398 if (oldstate != CEPH_MDS_STATE_CREATING &&
2399 oldstate != CEPH_MDS_STATE_STARTING)
2400 pr_info("mds%d recovery completed\n", s->s_mds);
2401 kick_requests(mdsc, i);
Sage Weil2f2dc052009-10-06 11:31:09 -07002402 ceph_kick_flushing_caps(mdsc, s);
Sage Weil0dc25702009-11-20 13:43:45 -08002403 wake_up_session_caps(s, 1);
Sage Weil2f2dc052009-10-06 11:31:09 -07002404 }
2405 }
2406}
2407
2408
2409
2410/*
2411 * leases
2412 */
2413
2414/*
2415 * caller must hold session s_mutex, dentry->d_lock
2416 */
2417void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2418{
2419 struct ceph_dentry_info *di = ceph_dentry(dentry);
2420
2421 ceph_put_mds_session(di->lease_session);
2422 di->lease_session = NULL;
2423}
2424
Sage Weil2600d2d2010-02-22 15:12:16 -08002425static void handle_lease(struct ceph_mds_client *mdsc,
2426 struct ceph_mds_session *session,
2427 struct ceph_msg *msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002428{
2429 struct super_block *sb = mdsc->client->sb;
2430 struct inode *inode;
Sage Weil2f2dc052009-10-06 11:31:09 -07002431 struct ceph_inode_info *ci;
2432 struct dentry *parent, *dentry;
2433 struct ceph_dentry_info *di;
Sage Weil2600d2d2010-02-22 15:12:16 -08002434 int mds = session->s_mds;
Sage Weil2f2dc052009-10-06 11:31:09 -07002435 struct ceph_mds_lease *h = msg->front.iov_base;
Sage Weil1e5ea232010-06-04 10:05:40 -07002436 u32 seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002437 struct ceph_vino vino;
2438 int mask;
2439 struct qstr dname;
2440 int release = 0;
2441
Sage Weil2f2dc052009-10-06 11:31:09 -07002442 dout("handle_lease from mds%d\n", mds);
2443
2444 /* decode */
2445 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2446 goto bad;
2447 vino.ino = le64_to_cpu(h->ino);
2448 vino.snap = CEPH_NOSNAP;
2449 mask = le16_to_cpu(h->mask);
Sage Weil1e5ea232010-06-04 10:05:40 -07002450 seq = le32_to_cpu(h->seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002451 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2452 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2453 if (dname.len != get_unaligned_le32(h+1))
2454 goto bad;
2455
Sage Weil2f2dc052009-10-06 11:31:09 -07002456 mutex_lock(&session->s_mutex);
2457 session->s_seq++;
2458
2459 /* lookup inode */
2460 inode = ceph_find_inode(sb, vino);
Sage Weil1e5ea232010-06-04 10:05:40 -07002461 dout("handle_lease %s, mask %d, ino %llx %p %.*s\n",
2462 ceph_lease_op_name(h->action), mask, vino.ino, inode,
2463 dname.len, dname.name);
Sage Weil2f2dc052009-10-06 11:31:09 -07002464 if (inode == NULL) {
2465 dout("handle_lease no inode %llx\n", vino.ino);
2466 goto release;
2467 }
2468 ci = ceph_inode(inode);
2469
2470 /* dentry */
2471 parent = d_find_alias(inode);
2472 if (!parent) {
2473 dout("no parent dentry on inode %p\n", inode);
2474 WARN_ON(1);
2475 goto release; /* hrm... */
2476 }
2477 dname.hash = full_name_hash(dname.name, dname.len);
2478 dentry = d_lookup(parent, &dname);
2479 dput(parent);
2480 if (!dentry)
2481 goto release;
2482
2483 spin_lock(&dentry->d_lock);
2484 di = ceph_dentry(dentry);
2485 switch (h->action) {
2486 case CEPH_MDS_LEASE_REVOKE:
2487 if (di && di->lease_session == session) {
Sage Weil1e5ea232010-06-04 10:05:40 -07002488 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2489 h->seq = cpu_to_le32(di->lease_seq);
Sage Weil2f2dc052009-10-06 11:31:09 -07002490 __ceph_mdsc_drop_dentry_lease(dentry);
2491 }
2492 release = 1;
2493 break;
2494
2495 case CEPH_MDS_LEASE_RENEW:
2496 if (di && di->lease_session == session &&
2497 di->lease_gen == session->s_cap_gen &&
2498 di->lease_renew_from &&
2499 di->lease_renew_after == 0) {
2500 unsigned long duration =
2501 le32_to_cpu(h->duration_ms) * HZ / 1000;
2502
Sage Weil1e5ea232010-06-04 10:05:40 -07002503 di->lease_seq = seq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002504 dentry->d_time = di->lease_renew_from + duration;
2505 di->lease_renew_after = di->lease_renew_from +
2506 (duration >> 1);
2507 di->lease_renew_from = 0;
2508 }
2509 break;
2510 }
2511 spin_unlock(&dentry->d_lock);
2512 dput(dentry);
2513
2514 if (!release)
2515 goto out;
2516
2517release:
2518 /* let's just reuse the same message */
2519 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2520 ceph_msg_get(msg);
2521 ceph_con_send(&session->s_con, msg);
2522
2523out:
2524 iput(inode);
2525 mutex_unlock(&session->s_mutex);
Sage Weil2f2dc052009-10-06 11:31:09 -07002526 return;
2527
2528bad:
2529 pr_err("corrupt lease message\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -08002530 ceph_msg_dump(msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07002531}
2532
2533void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2534 struct inode *inode,
2535 struct dentry *dentry, char action,
2536 u32 seq)
2537{
2538 struct ceph_msg *msg;
2539 struct ceph_mds_lease *lease;
2540 int len = sizeof(*lease) + sizeof(u32);
2541 int dnamelen = 0;
2542
2543 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2544 inode, dentry, ceph_lease_op_name(action), session->s_mds);
2545 dnamelen = dentry->d_name.len;
2546 len += dnamelen;
2547
Yehuda Sadeh34d23762010-04-06 14:33:58 -07002548 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS);
Sage Weila79832f2010-04-01 16:06:19 -07002549 if (!msg)
Sage Weil2f2dc052009-10-06 11:31:09 -07002550 return;
2551 lease = msg->front.iov_base;
2552 lease->action = action;
Sage Weildd1c9052010-05-25 16:45:25 -07002553 lease->mask = cpu_to_le16(1);
Sage Weil2f2dc052009-10-06 11:31:09 -07002554 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2555 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2556 lease->seq = cpu_to_le32(seq);
2557 put_unaligned_le32(dnamelen, lease + 1);
2558 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2559
2560 /*
2561 * if this is a preemptive lease RELEASE, no need to
2562 * flush request stream, since the actual request will
2563 * soon follow.
2564 */
2565 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2566
2567 ceph_con_send(&session->s_con, msg);
2568}
2569
2570/*
2571 * Preemptively release a lease we expect to invalidate anyway.
2572 * Pass @inode always, @dentry is optional.
2573 */
2574void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2575 struct dentry *dentry, int mask)
2576{
2577 struct ceph_dentry_info *di;
2578 struct ceph_mds_session *session;
2579 u32 seq;
2580
2581 BUG_ON(inode == NULL);
2582 BUG_ON(dentry == NULL);
Sage Weildd1c9052010-05-25 16:45:25 -07002583 BUG_ON(mask == 0);
Sage Weil2f2dc052009-10-06 11:31:09 -07002584
2585 /* is dentry lease valid? */
2586 spin_lock(&dentry->d_lock);
2587 di = ceph_dentry(dentry);
2588 if (!di || !di->lease_session ||
2589 di->lease_session->s_mds < 0 ||
2590 di->lease_gen != di->lease_session->s_cap_gen ||
2591 !time_before(jiffies, dentry->d_time)) {
2592 dout("lease_release inode %p dentry %p -- "
2593 "no lease on %d\n",
2594 inode, dentry, mask);
2595 spin_unlock(&dentry->d_lock);
2596 return;
2597 }
2598
2599 /* we do have a lease on this dentry; note mds and seq */
2600 session = ceph_get_mds_session(di->lease_session);
2601 seq = di->lease_seq;
2602 __ceph_mdsc_drop_dentry_lease(dentry);
2603 spin_unlock(&dentry->d_lock);
2604
2605 dout("lease_release inode %p dentry %p mask %d to mds%d\n",
2606 inode, dentry, mask, session->s_mds);
2607 ceph_mdsc_lease_send_msg(session, inode, dentry,
2608 CEPH_MDS_LEASE_RELEASE, seq);
2609 ceph_put_mds_session(session);
2610}
2611
2612/*
2613 * drop all leases (and dentry refs) in preparation for umount
2614 */
2615static void drop_leases(struct ceph_mds_client *mdsc)
2616{
2617 int i;
2618
2619 dout("drop_leases\n");
2620 mutex_lock(&mdsc->mutex);
2621 for (i = 0; i < mdsc->max_sessions; i++) {
2622 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2623 if (!s)
2624 continue;
2625 mutex_unlock(&mdsc->mutex);
2626 mutex_lock(&s->s_mutex);
2627 mutex_unlock(&s->s_mutex);
2628 ceph_put_mds_session(s);
2629 mutex_lock(&mdsc->mutex);
2630 }
2631 mutex_unlock(&mdsc->mutex);
2632}
2633
2634
2635
2636/*
2637 * delayed work -- periodically trim expired leases, renew caps with mds
2638 */
2639static void schedule_delayed(struct ceph_mds_client *mdsc)
2640{
2641 int delay = 5;
2642 unsigned hz = round_jiffies_relative(HZ * delay);
2643 schedule_delayed_work(&mdsc->delayed_work, hz);
2644}
2645
2646static void delayed_work(struct work_struct *work)
2647{
2648 int i;
2649 struct ceph_mds_client *mdsc =
2650 container_of(work, struct ceph_mds_client, delayed_work.work);
2651 int renew_interval;
2652 int renew_caps;
2653
2654 dout("mdsc delayed_work\n");
Sage Weilafcdaea2009-10-14 14:27:38 -07002655 ceph_check_delayed_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07002656
2657 mutex_lock(&mdsc->mutex);
2658 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
2659 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
2660 mdsc->last_renew_caps);
2661 if (renew_caps)
2662 mdsc->last_renew_caps = jiffies;
2663
2664 for (i = 0; i < mdsc->max_sessions; i++) {
2665 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2666 if (s == NULL)
2667 continue;
2668 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
2669 dout("resending session close request for mds%d\n",
2670 s->s_mds);
2671 request_close_session(mdsc, s);
2672 ceph_put_mds_session(s);
2673 continue;
2674 }
2675 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
2676 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
2677 s->s_state = CEPH_MDS_SESSION_HUNG;
2678 pr_info("mds%d hung\n", s->s_mds);
2679 }
2680 }
2681 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
2682 /* this mds is failed or recovering, just wait */
2683 ceph_put_mds_session(s);
2684 continue;
2685 }
2686 mutex_unlock(&mdsc->mutex);
2687
2688 mutex_lock(&s->s_mutex);
2689 if (renew_caps)
2690 send_renew_caps(mdsc, s);
2691 else
2692 ceph_con_keepalive(&s->s_con);
Sage Weil2b2300d2010-06-09 16:52:04 -07002693 ceph_add_cap_releases(mdsc, s, -1);
Sage Weilaab53dd2010-03-17 16:30:21 -07002694 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2695 s->s_state == CEPH_MDS_SESSION_HUNG)
Sage Weil3d7ded42010-06-09 16:47:10 -07002696 ceph_send_cap_releases(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07002697 mutex_unlock(&s->s_mutex);
2698 ceph_put_mds_session(s);
2699
2700 mutex_lock(&mdsc->mutex);
2701 }
2702 mutex_unlock(&mdsc->mutex);
2703
2704 schedule_delayed(mdsc);
2705}
2706
2707
Sage Weil5f44f142009-11-18 14:52:18 -08002708int ceph_mdsc_init(struct ceph_mds_client *mdsc, struct ceph_client *client)
Sage Weil2f2dc052009-10-06 11:31:09 -07002709{
2710 mdsc->client = client;
2711 mutex_init(&mdsc->mutex);
2712 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08002713 if (mdsc->mdsmap == NULL)
2714 return -ENOMEM;
2715
Sage Weil2f2dc052009-10-06 11:31:09 -07002716 init_completion(&mdsc->safe_umount_waiters);
2717 init_completion(&mdsc->session_close_waiters);
2718 INIT_LIST_HEAD(&mdsc->waiting_for_map);
2719 mdsc->sessions = NULL;
2720 mdsc->max_sessions = 0;
2721 mdsc->stopping = 0;
2722 init_rwsem(&mdsc->snap_rwsem);
Sage Weila105f002010-02-15 14:37:55 -08002723 mdsc->snap_realms = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07002724 INIT_LIST_HEAD(&mdsc->snap_empty);
2725 spin_lock_init(&mdsc->snap_empty_lock);
2726 mdsc->last_tid = 0;
Sage Weil44ca18f2010-02-15 12:08:46 -08002727 mdsc->request_tree = RB_ROOT;
Sage Weil2f2dc052009-10-06 11:31:09 -07002728 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
2729 mdsc->last_renew_caps = jiffies;
2730 INIT_LIST_HEAD(&mdsc->cap_delay_list);
2731 spin_lock_init(&mdsc->cap_delay_lock);
2732 INIT_LIST_HEAD(&mdsc->snap_flush_list);
2733 spin_lock_init(&mdsc->snap_flush_lock);
2734 mdsc->cap_flush_seq = 0;
2735 INIT_LIST_HEAD(&mdsc->cap_dirty);
2736 mdsc->num_cap_flushing = 0;
2737 spin_lock_init(&mdsc->cap_dirty_lock);
2738 init_waitqueue_head(&mdsc->cap_flushing_wq);
2739 spin_lock_init(&mdsc->dentry_lru_lock);
2740 INIT_LIST_HEAD(&mdsc->dentry_lru);
Cheng Renquan2d06eeb2010-03-26 18:04:40 +08002741
Sage Weil5f44f142009-11-18 14:52:18 -08002742 return 0;
Sage Weil2f2dc052009-10-06 11:31:09 -07002743}
2744
2745/*
2746 * Wait for safe replies on open mds requests. If we time out, drop
2747 * all requests from the tree to avoid dangling dentry refs.
2748 */
2749static void wait_requests(struct ceph_mds_client *mdsc)
2750{
2751 struct ceph_mds_request *req;
2752 struct ceph_client *client = mdsc->client;
2753
2754 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08002755 if (__get_oldest_req(mdsc)) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002756 mutex_unlock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08002757
Sage Weil2f2dc052009-10-06 11:31:09 -07002758 dout("wait_requests waiting for requests\n");
2759 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
Sage Weil6b805182009-10-27 11:50:50 -07002760 client->mount_args->mount_timeout * HZ);
Sage Weil2f2dc052009-10-06 11:31:09 -07002761
2762 /* tear down remaining requests */
Sage Weil44ca18f2010-02-15 12:08:46 -08002763 mutex_lock(&mdsc->mutex);
2764 while ((req = __get_oldest_req(mdsc))) {
Sage Weil2f2dc052009-10-06 11:31:09 -07002765 dout("wait_requests timed out on tid %llu\n",
2766 req->r_tid);
Sage Weil44ca18f2010-02-15 12:08:46 -08002767 __unregister_request(mdsc, req);
Sage Weil2f2dc052009-10-06 11:31:09 -07002768 }
2769 }
2770 mutex_unlock(&mdsc->mutex);
2771 dout("wait_requests done\n");
2772}
2773
2774/*
2775 * called before mount is ro, and before dentries are torn down.
2776 * (hmm, does this still race with new lookups?)
2777 */
2778void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
2779{
2780 dout("pre_umount\n");
2781 mdsc->stopping = 1;
2782
2783 drop_leases(mdsc);
Sage Weilafcdaea2009-10-14 14:27:38 -07002784 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07002785 wait_requests(mdsc);
2786}
2787
2788/*
2789 * wait for all write mds requests to flush.
2790 */
2791static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
2792{
Sage Weil80fc7312010-03-16 15:28:54 -07002793 struct ceph_mds_request *req = NULL, *nextreq;
Sage Weil44ca18f2010-02-15 12:08:46 -08002794 struct rb_node *n;
Sage Weil2f2dc052009-10-06 11:31:09 -07002795
2796 mutex_lock(&mdsc->mutex);
2797 dout("wait_unsafe_requests want %lld\n", want_tid);
Sage Weil80fc7312010-03-16 15:28:54 -07002798restart:
Sage Weil44ca18f2010-02-15 12:08:46 -08002799 req = __get_oldest_req(mdsc);
2800 while (req && req->r_tid <= want_tid) {
Sage Weil80fc7312010-03-16 15:28:54 -07002801 /* find next request */
2802 n = rb_next(&req->r_node);
2803 if (n)
2804 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
2805 else
2806 nextreq = NULL;
Sage Weil44ca18f2010-02-15 12:08:46 -08002807 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
2808 /* write op */
2809 ceph_mdsc_get_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07002810 if (nextreq)
2811 ceph_mdsc_get_request(nextreq);
Sage Weil44ca18f2010-02-15 12:08:46 -08002812 mutex_unlock(&mdsc->mutex);
2813 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
2814 req->r_tid, want_tid);
2815 wait_for_completion(&req->r_safe_completion);
2816 mutex_lock(&mdsc->mutex);
Sage Weil44ca18f2010-02-15 12:08:46 -08002817 ceph_mdsc_put_request(req);
Sage Weil80fc7312010-03-16 15:28:54 -07002818 if (!nextreq)
2819 break; /* next dne before, so we're done! */
2820 if (RB_EMPTY_NODE(&nextreq->r_node)) {
2821 /* next request was removed from tree */
2822 ceph_mdsc_put_request(nextreq);
2823 goto restart;
2824 }
2825 ceph_mdsc_put_request(nextreq); /* won't go away */
Sage Weil44ca18f2010-02-15 12:08:46 -08002826 }
Sage Weil80fc7312010-03-16 15:28:54 -07002827 req = nextreq;
Sage Weil2f2dc052009-10-06 11:31:09 -07002828 }
2829 mutex_unlock(&mdsc->mutex);
2830 dout("wait_unsafe_requests done\n");
2831}
2832
2833void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
2834{
2835 u64 want_tid, want_flush;
2836
Sage Weil56b7cf92010-05-03 15:22:00 -07002837 if (mdsc->client->mount_state == CEPH_MOUNT_SHUTDOWN)
2838 return;
2839
Sage Weil2f2dc052009-10-06 11:31:09 -07002840 dout("sync\n");
2841 mutex_lock(&mdsc->mutex);
2842 want_tid = mdsc->last_tid;
2843 want_flush = mdsc->cap_flush_seq;
2844 mutex_unlock(&mdsc->mutex);
2845 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
2846
Sage Weilafcdaea2009-10-14 14:27:38 -07002847 ceph_flush_dirty_caps(mdsc);
Sage Weil2f2dc052009-10-06 11:31:09 -07002848
2849 wait_unsafe_requests(mdsc, want_tid);
2850 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
2851}
2852
2853
2854/*
2855 * called after sb is ro.
2856 */
2857void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
2858{
2859 struct ceph_mds_session *session;
2860 int i;
2861 int n;
2862 struct ceph_client *client = mdsc->client;
Sage Weil6b805182009-10-27 11:50:50 -07002863 unsigned long started, timeout = client->mount_args->mount_timeout * HZ;
Sage Weil2f2dc052009-10-06 11:31:09 -07002864
2865 dout("close_sessions\n");
2866
2867 mutex_lock(&mdsc->mutex);
2868
2869 /* close sessions */
2870 started = jiffies;
2871 while (time_before(jiffies, started + timeout)) {
2872 dout("closing sessions\n");
2873 n = 0;
2874 for (i = 0; i < mdsc->max_sessions; i++) {
2875 session = __ceph_lookup_mds_session(mdsc, i);
2876 if (!session)
2877 continue;
2878 mutex_unlock(&mdsc->mutex);
2879 mutex_lock(&session->s_mutex);
2880 __close_session(mdsc, session);
2881 mutex_unlock(&session->s_mutex);
2882 ceph_put_mds_session(session);
2883 mutex_lock(&mdsc->mutex);
2884 n++;
2885 }
2886 if (n == 0)
2887 break;
2888
2889 if (client->mount_state == CEPH_MOUNT_SHUTDOWN)
2890 break;
2891
2892 dout("waiting for sessions to close\n");
2893 mutex_unlock(&mdsc->mutex);
2894 wait_for_completion_timeout(&mdsc->session_close_waiters,
2895 timeout);
2896 mutex_lock(&mdsc->mutex);
2897 }
2898
2899 /* tear down remaining sessions */
2900 for (i = 0; i < mdsc->max_sessions; i++) {
2901 if (mdsc->sessions[i]) {
2902 session = get_session(mdsc->sessions[i]);
Sage Weil2600d2d2010-02-22 15:12:16 -08002903 __unregister_session(mdsc, session);
Sage Weil2f2dc052009-10-06 11:31:09 -07002904 mutex_unlock(&mdsc->mutex);
2905 mutex_lock(&session->s_mutex);
2906 remove_session_caps(session);
2907 mutex_unlock(&session->s_mutex);
2908 ceph_put_mds_session(session);
2909 mutex_lock(&mdsc->mutex);
2910 }
2911 }
2912
2913 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2914
2915 mutex_unlock(&mdsc->mutex);
2916
2917 ceph_cleanup_empty_realms(mdsc);
2918
2919 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
2920
2921 dout("stopped\n");
2922}
2923
2924void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2925{
2926 dout("stop\n");
2927 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
2928 if (mdsc->mdsmap)
2929 ceph_mdsmap_destroy(mdsc->mdsmap);
2930 kfree(mdsc->sessions);
2931}
2932
2933
2934/*
2935 * handle mds map update.
2936 */
2937void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2938{
2939 u32 epoch;
2940 u32 maplen;
2941 void *p = msg->front.iov_base;
2942 void *end = p + msg->front.iov_len;
2943 struct ceph_mdsmap *newmap, *oldmap;
2944 struct ceph_fsid fsid;
2945 int err = -EINVAL;
2946
2947 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
2948 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weil07433042009-11-18 16:50:41 -08002949 if (ceph_check_fsid(mdsc->client, &fsid) < 0)
2950 return;
Sage Weilc89136e2009-10-14 09:59:09 -07002951 epoch = ceph_decode_32(&p);
2952 maplen = ceph_decode_32(&p);
Sage Weil2f2dc052009-10-06 11:31:09 -07002953 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
2954
2955 /* do we need it? */
2956 ceph_monc_got_mdsmap(&mdsc->client->monc, epoch);
2957 mutex_lock(&mdsc->mutex);
2958 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
2959 dout("handle_map epoch %u <= our %u\n",
2960 epoch, mdsc->mdsmap->m_epoch);
2961 mutex_unlock(&mdsc->mutex);
2962 return;
2963 }
2964
2965 newmap = ceph_mdsmap_decode(&p, end);
2966 if (IS_ERR(newmap)) {
2967 err = PTR_ERR(newmap);
2968 goto bad_unlock;
2969 }
2970
2971 /* swap into place */
2972 if (mdsc->mdsmap) {
2973 oldmap = mdsc->mdsmap;
2974 mdsc->mdsmap = newmap;
2975 check_new_map(mdsc, newmap, oldmap);
2976 ceph_mdsmap_destroy(oldmap);
2977 } else {
2978 mdsc->mdsmap = newmap; /* first mds map */
2979 }
2980 mdsc->client->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2981
2982 __wake_requests(mdsc, &mdsc->waiting_for_map);
2983
2984 mutex_unlock(&mdsc->mutex);
2985 schedule_delayed(mdsc);
2986 return;
2987
2988bad_unlock:
2989 mutex_unlock(&mdsc->mutex);
2990bad:
2991 pr_err("error decoding mdsmap %d\n", err);
2992 return;
2993}
2994
2995static struct ceph_connection *con_get(struct ceph_connection *con)
2996{
2997 struct ceph_mds_session *s = con->private;
2998
2999 if (get_session(s)) {
Sage Weil2600d2d2010-02-22 15:12:16 -08003000 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
Sage Weil2f2dc052009-10-06 11:31:09 -07003001 return con;
3002 }
3003 dout("mdsc con_get %p FAIL\n", s);
3004 return NULL;
3005}
3006
3007static void con_put(struct ceph_connection *con)
3008{
3009 struct ceph_mds_session *s = con->private;
3010
Sage Weil2f2dc052009-10-06 11:31:09 -07003011 ceph_put_mds_session(s);
Sage Weil2600d2d2010-02-22 15:12:16 -08003012 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref));
Sage Weil2f2dc052009-10-06 11:31:09 -07003013}
3014
3015/*
3016 * if the client is unresponsive for long enough, the mds will kill
3017 * the session entirely.
3018 */
3019static void peer_reset(struct ceph_connection *con)
3020{
3021 struct ceph_mds_session *s = con->private;
Sage Weil7e70f0e2010-03-18 13:59:12 -07003022 struct ceph_mds_client *mdsc = s->s_mdsc;
Sage Weil2f2dc052009-10-06 11:31:09 -07003023
Sage Weil7e70f0e2010-03-18 13:59:12 -07003024 pr_warning("mds%d closed our session\n", s->s_mds);
3025 send_mds_reconnect(mdsc, s);
Sage Weil2f2dc052009-10-06 11:31:09 -07003026}
3027
3028static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3029{
3030 struct ceph_mds_session *s = con->private;
3031 struct ceph_mds_client *mdsc = s->s_mdsc;
3032 int type = le16_to_cpu(msg->hdr.type);
3033
Sage Weil2600d2d2010-02-22 15:12:16 -08003034 mutex_lock(&mdsc->mutex);
3035 if (__verify_registered_session(mdsc, s) < 0) {
3036 mutex_unlock(&mdsc->mutex);
3037 goto out;
3038 }
3039 mutex_unlock(&mdsc->mutex);
3040
Sage Weil2f2dc052009-10-06 11:31:09 -07003041 switch (type) {
3042 case CEPH_MSG_MDS_MAP:
3043 ceph_mdsc_handle_map(mdsc, msg);
3044 break;
3045 case CEPH_MSG_CLIENT_SESSION:
3046 handle_session(s, msg);
3047 break;
3048 case CEPH_MSG_CLIENT_REPLY:
3049 handle_reply(s, msg);
3050 break;
3051 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
Sage Weil2600d2d2010-02-22 15:12:16 -08003052 handle_forward(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003053 break;
3054 case CEPH_MSG_CLIENT_CAPS:
3055 ceph_handle_caps(s, msg);
3056 break;
3057 case CEPH_MSG_CLIENT_SNAP:
Sage Weil2600d2d2010-02-22 15:12:16 -08003058 ceph_handle_snap(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003059 break;
3060 case CEPH_MSG_CLIENT_LEASE:
Sage Weil2600d2d2010-02-22 15:12:16 -08003061 handle_lease(mdsc, s, msg);
Sage Weil2f2dc052009-10-06 11:31:09 -07003062 break;
3063
3064 default:
3065 pr_err("received unknown message type %d %s\n", type,
3066 ceph_msg_type_name(type));
3067 }
Sage Weil2600d2d2010-02-22 15:12:16 -08003068out:
Sage Weil2f2dc052009-10-06 11:31:09 -07003069 ceph_msg_put(msg);
3070}
3071
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003072/*
3073 * authentication
3074 */
3075static int get_authorizer(struct ceph_connection *con,
3076 void **buf, int *len, int *proto,
3077 void **reply_buf, int *reply_len, int force_new)
3078{
3079 struct ceph_mds_session *s = con->private;
3080 struct ceph_mds_client *mdsc = s->s_mdsc;
3081 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3082 int ret = 0;
3083
3084 if (force_new && s->s_authorizer) {
3085 ac->ops->destroy_authorizer(ac, s->s_authorizer);
3086 s->s_authorizer = NULL;
3087 }
3088 if (s->s_authorizer == NULL) {
3089 if (ac->ops->create_authorizer) {
3090 ret = ac->ops->create_authorizer(
3091 ac, CEPH_ENTITY_TYPE_MDS,
3092 &s->s_authorizer,
3093 &s->s_authorizer_buf,
3094 &s->s_authorizer_buf_len,
3095 &s->s_authorizer_reply_buf,
3096 &s->s_authorizer_reply_buf_len);
3097 if (ret)
3098 return ret;
3099 }
3100 }
3101
3102 *proto = ac->protocol;
3103 *buf = s->s_authorizer_buf;
3104 *len = s->s_authorizer_buf_len;
3105 *reply_buf = s->s_authorizer_reply_buf;
3106 *reply_len = s->s_authorizer_reply_buf_len;
3107 return 0;
3108}
3109
3110
3111static int verify_authorizer_reply(struct ceph_connection *con, int len)
3112{
3113 struct ceph_mds_session *s = con->private;
3114 struct ceph_mds_client *mdsc = s->s_mdsc;
3115 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3116
3117 return ac->ops->verify_authorizer_reply(ac, s->s_authorizer, len);
3118}
3119
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003120static int invalidate_authorizer(struct ceph_connection *con)
3121{
3122 struct ceph_mds_session *s = con->private;
3123 struct ceph_mds_client *mdsc = s->s_mdsc;
3124 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3125
3126 if (ac->ops->invalidate_authorizer)
3127 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3128
3129 return ceph_monc_validate_auth(&mdsc->client->monc);
3130}
3131
Tobias Klauser9e327892010-05-20 10:40:19 +02003132static const struct ceph_connection_operations mds_con_ops = {
Sage Weil2f2dc052009-10-06 11:31:09 -07003133 .get = con_get,
3134 .put = con_put,
3135 .dispatch = dispatch,
Sage Weil4e7a5dc2009-11-18 16:19:57 -08003136 .get_authorizer = get_authorizer,
3137 .verify_authorizer_reply = verify_authorizer_reply,
Sage Weil9bd2e6f2010-02-02 16:21:06 -08003138 .invalidate_authorizer = invalidate_authorizer,
Sage Weil2f2dc052009-10-06 11:31:09 -07003139 .peer_reset = peer_reset,
Sage Weil2f2dc052009-10-06 11:31:09 -07003140};
3141
3142
3143
3144
3145/* eof */