]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ceph/mds_client.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: reuse request message when replaying against recovering mds
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / mds_client.c
CommitLineData
2f2dc053
SW		 1#include "ceph_debug.h"
2
3#include <linux/wait.h>
5a0e3ad6 4#include <linux/slab.h>
2f2dc053
SW		 5#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"
4e7a5dcd 12#include "auth.h"
93cea5be 13#include "pagelist.h"
2f2dc053
SW		 14
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37 * are no longer valid.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
9e32789f 43static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 44
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);
c89136ea
SW		 142 num = ceph_decode_32(p);
143 info->dir_end = ceph_decode_8(p);
144 info->dir_complete = ceph_decode_8(p);
2f2dc053
SW		 145 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);
c89136ea 166 info->dir_dname_len[i] = ceph_decode_32(p);
2f2dc053
SW		 167 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";
44ca18f2 259 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053
SW		 260 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);
4e7a5dcd
SW		 281 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);
2f2dc053 285 kfree(s);
4e7a5dcd 286 }
2f2dc053
SW		 287}
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
2600d2dd
SW		 313static 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
2f2dc053
SW		 322/*
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);
4736b009
DC		 332 if (!s)
333 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 334 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);
2f2dc053
SW		 346
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;
5dacf091 354 s->s_trim_caps = 0;
2f2dc053
SW		 355 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;
7c1332b8 359 s->s_cap_iterator = NULL;
2f2dc053
SW		 360 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)
42ce56e5 373 goto fail_realloc;
2f2dc053
SW		 374 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 */
42ce56e5
SW		 384
385 ceph_con_open(&s->s_con, ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
386
2f2dc053 387 return s;
42ce56e5
SW		 388
389fail_realloc:
390 kfree(s);
391 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 392}
393
394/*
395 * called under mdsc->mutex
396 */
2600d2dd 397static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 398 struct ceph_mds_session *s)
2f2dc053 399{
2600d2dd
SW		 400 dout("__unregister_session mds%d %p\n", s->s_mds, s);
401 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 402 mdsc->sessions[s->s_mds] = NULL;
403 ceph_con_close(&s->s_con);
404 ceph_put_mds_session(s);
2f2dc053
SW		 405}
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
153c8e6b 420void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 421{
153c8e6b
SW		 422 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);
430 }
431 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);
2f2dc053 448 }
153c8e6b
SW		 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);
2f2dc053
SW		 454}
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;
44ca18f2
SW		 465 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);
2f2dc053
SW		 501}
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);
44ca18f2 518 __insert_request(mdsc, req);
2f2dc053
SW		 519
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);
44ca18f2 534 rb_erase(&req->r_node, &mdsc->request_tree);
80fc7314 535 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 536
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 }
94aa8ae1
SW		 544
545 ceph_mdsc_put_request(req);
2f2dc053
SW		 546}
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
34d23762 668 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS);
a79832f2 669 if (!msg) {
2f2dc053 670 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 671 return NULL;
2f2dc053
SW		 672 }
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;
2f2dc053
SW		 690
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);
a79832f2
SW		 700 if (!msg)
701 return -ENOMEM;
2f2dc053 702 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 703 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/*
f818a736
SW		 734 * Helper to safely iterate over all caps associated with a session, with
735 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 736 *
f818a736 737 * Caller must hold session s_mutex.
2f2dc053
SW		 738 */
739static int iterate_session_caps(struct ceph_mds_session *session,
740 int (*cb)(struct inode *, struct ceph_cap *,
741 void *), void *arg)
742{
7c1332b8
SW		 743 struct list_head *p;
744 struct ceph_cap *cap;
745 struct inode *inode, *last_inode = NULL;
746 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 747 int ret;
748
749 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
750 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 751 p = session->s_caps.next;
752 while (p != &session->s_caps) {
753 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 754 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 755 if (!inode) {
756 p = p->next;
2f2dc053 757 continue;
7c1332b8
SW		 758 }
759 session->s_cap_iterator = cap;
2f2dc053 760 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 761
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
2f2dc053 771 ret = cb(inode, cap, arg);
7c1332b8
SW		 772 last_inode = inode;
773
2f2dc053 774 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 775 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 }
5dacf091
SW		 785 if (ret < 0)
786 goto out;
2f2dc053 787 }
5dacf091
SW		 788 ret = 0;
789out:
7c1332b8 790 session->s_cap_iterator = NULL;
2f2dc053 791 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 792
793 if (last_inode)
794 iput(last_inode);
795 if (old_cap)
796 ceph_put_cap(old_cap);
797
5dacf091 798 return ret;
2f2dc053
SW		 799}
800
801static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 802 void *arg)
2f2dc053
SW		 803{
804 struct ceph_inode_info *ci = ceph_inode(inode);
6c99f254
SW		 805 int drop = 0;
806
2f2dc053
SW		 807 dout("removing cap %p, ci is %p, inode is %p\n",
808 cap, ci, &ci->vfs_inode);
6c99f254
SW		 809 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);
2f2dc053
SW		 845 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);
6c99f254 856 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 857 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{
0dc2570f
SW		 869 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 }
2f2dc053
SW		 878 return 0;
879}
880
0dc2570f
SW		 881static void wake_up_session_caps(struct ceph_mds_session *session,
882 int reconnect)
2f2dc053
SW		 883{
884 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 885 iterate_session_caps(session, wake_up_session_cb,
886 (void *)(unsigned long)reconnect);
2f2dc053
SW		 887}
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);
e4cb4cb8 904 session->s_renew_requested = jiffies;
2f2dc053
SW		 905
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));
2f2dc053
SW		 917 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
918 ++session->s_renew_seq);
a79832f2
SW		 919 if (!msg)
920 return -ENOMEM;
2f2dc053
SW		 921 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?).
0dc2570f
SW		 927 *
928 * Called under session->s_mutex
2f2dc053
SW		 929 */
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)
0dc2570f 957 wake_up_session_caps(session, 0);
2f2dc053
SW		 958}
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;
2f2dc053
SW		 967
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);
a79832f2
SW		 972 if (!msg)
973 return -ENOMEM;
974 ceph_con_send(&session->s_con, msg);
975 return 0;
2f2dc053
SW		 976}
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 */
7c1332b8 1025 __ceph_remove_cap(cap);
2f2dc053
SW		 1026 } 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);
5dacf091 1057 session->s_trim_caps = 0;
2f2dc053
SW		 1058 }
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 */
2b2300d6
SW		 1069int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
1070 struct ceph_mds_session *session,
1071 int extra)
2f2dc053
SW		 1072{
1073 struct ceph_msg *msg;
1074 struct ceph_mds_cap_release *head;
1075 int err = -ENOMEM;
1076
1077 if (extra < 0)
6b805185 1078 extra = mdsc->client->mount_args->cap_release_safety;
2f2dc053
SW		 1079
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);
34d23762
YS		 1092 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
1093 GFP_NOFS);
2f2dc053
SW		 1094 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 */
3d7ded4d
SW		 1179void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1180 struct ceph_mds_session *session)
2f2dc053
SW		 1181{
1182 struct ceph_msg *msg;
1183
1184 dout("send_cap_releases mds%d\n", session->s_mds);
0f8605f2
SW		 1185 spin_lock(&session->s_cap_lock);
1186 while (!list_empty(&session->s_cap_releases_done)) {
2f2dc053
SW		 1187 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);
0f8605f2 1194 spin_lock(&session->s_cap_lock);
2f2dc053
SW		 1195 }
1196 spin_unlock(&session->s_cap_lock);
1197}
1198
e01a5946
SW		 1199static 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
2f2dc053
SW		 1237/*
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
b4556396 1252 mutex_init(&req->r_fill_mutex);
2f2dc053
SW		 1253 req->r_started = jiffies;
1254 req->r_resend_mds = -1;
1255 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1256 req->r_fmode = -1;
153c8e6b 1257 kref_init(&req->r_kref);
2f2dc053
SW		 1258 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/*
44ca18f2 1269 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1270 *
1271 * called under mdsc->mutex.
1272 */
44ca18f2
SW		 1273static 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
2f2dc053
SW		 1281static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1282{
44ca18f2
SW		 1283 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1284
1285 if (req)
1286 return req->r_tid;
1287 return 0;
2f2dc053
SW		 1288}
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) {
6c99f254 1323 pr_err("build_path corrupt dentry %p\n", dentry);
2f2dc053
SW		 1324 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) {
104648ad 1339 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1340 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);
2f2dc053
SW		 1350 }
1351 if (pos)
1352 path[--pos] = '/';
1353 temp = temp->d_parent;
1354 if (temp == NULL) {
104648ad 1355 pr_err("build_path corrupt dentry\n");
2f2dc053
SW		 1356 kfree(path);
1357 return ERR_PTR(-EINVAL);
1358 }
1359 }
1360 if (pos != 0) {
104648ad 1361 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1362 "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;
104648ad 1373 dout("build_path on %p %d built %llx '%.*s'\n",
2f2dc053
SW		 1374 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) +
ac8839d7 1485 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64));
2f2dc053
SW		 1486
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
34d23762 1496 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS);
a79832f2
SW		 1497 if (!msg) {
1498 msg = ERR_PTR(-ENOMEM);
2f2dc053 1499 goto out_free2;
a79832f2 1500 }
2f2dc053 1501
6df058c0
SW		 1502 msg->hdr.tid = cpu_to_le64(req->r_tid);
1503
2f2dc053
SW		 1504 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
01a92f17
SW		 1583 if (req->r_got_unsafe) {
1584 /*
1585 * Replay. Do not regenerate message (and rebuild
1586 * paths, etc.); just use the original message.
1587 * Rebuilding paths will break for renames because
1588 * d_move mangles the src name.
1589 */
1590 msg = req->r_request;
1591 rhead = msg->front.iov_base;
1592
1593 flags = le32_to_cpu(rhead->flags);
1594 flags |= CEPH_MDS_FLAG_REPLAY;
1595 rhead->flags = cpu_to_le32(flags);
1596
1597 if (req->r_target_inode)
1598 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
1599
1600 rhead->num_retry = req->r_attempts - 1;
1601 return 0;
1602 }
1603
2f2dc053
SW		 1604 if (req->r_request) {
1605 ceph_msg_put(req->r_request);
1606 req->r_request = NULL;
1607 }
1608 msg = create_request_message(mdsc, req, mds);
1609 if (IS_ERR(msg)) {
e1518c7c 1610 req->r_err = PTR_ERR(msg);
2f2dc053 1611 complete_request(mdsc, req);
a79832f2 1612 return PTR_ERR(msg);
2f2dc053
SW		 1613 }
1614 req->r_request = msg;
1615
1616 rhead = msg->front.iov_base;
2f2dc053
SW		 1617 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
1618 if (req->r_got_unsafe)
1619 flags |= CEPH_MDS_FLAG_REPLAY;
1620 if (req->r_locked_dir)
1621 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
1622 rhead->flags = cpu_to_le32(flags);
1623 rhead->num_fwd = req->r_num_fwd;
1624 rhead->num_retry = req->r_attempts - 1;
01a92f17 1625 rhead->ino = 0;
2f2dc053
SW		 1626
1627 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 1628 return 0;
1629}
1630
1631/*
1632 * send request, or put it on the appropriate wait list.
1633 */
1634static int __do_request(struct ceph_mds_client *mdsc,
1635 struct ceph_mds_request *req)
1636{
1637 struct ceph_mds_session *session = NULL;
1638 int mds = -1;
1639 int err = -EAGAIN;
1640
e1518c7c 1641 if (req->r_err || req->r_got_result)
2f2dc053
SW		 1642 goto out;
1643
1644 if (req->r_timeout &&
1645 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
1646 dout("do_request timed out\n");
1647 err = -EIO;
1648 goto finish;
1649 }
1650
1651 mds = __choose_mds(mdsc, req);
1652 if (mds < 0 ||
1653 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
1654 dout("do_request no mds or not active, waiting for map\n");
1655 list_add(&req->r_wait, &mdsc->waiting_for_map);
1656 goto out;
1657 }
1658
1659 /* get, open session */
1660 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 1661 if (!session) {
2f2dc053 1662 session = register_session(mdsc, mds);
9c423956
SW		 1663 if (IS_ERR(session)) {
1664 err = PTR_ERR(session);
1665 goto finish;
1666 }
1667 }
2f2dc053
SW		 1668 dout("do_request mds%d session %p state %s\n", mds, session,
1669 session_state_name(session->s_state));
1670 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
1671 session->s_state != CEPH_MDS_SESSION_HUNG) {
1672 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1673 session->s_state == CEPH_MDS_SESSION_CLOSING)
1674 __open_session(mdsc, session);
1675 list_add(&req->r_wait, &session->s_waiting);
1676 goto out_session;
1677 }
1678
1679 /* send request */
1680 req->r_session = get_session(session);
1681 req->r_resend_mds = -1; /* forget any previous mds hint */
1682
1683 if (req->r_request_started == 0) /* note request start time */
1684 req->r_request_started = jiffies;
1685
1686 err = __prepare_send_request(mdsc, req, mds);
1687 if (!err) {
1688 ceph_msg_get(req->r_request);
1689 ceph_con_send(&session->s_con, req->r_request);
1690 }
1691
1692out_session:
1693 ceph_put_mds_session(session);
1694out:
1695 return err;
1696
1697finish:
e1518c7c 1698 req->r_err = err;
2f2dc053
SW		 1699 complete_request(mdsc, req);
1700 goto out;
1701}
1702
1703/*
1704 * called under mdsc->mutex
1705 */
1706static void __wake_requests(struct ceph_mds_client *mdsc,
1707 struct list_head *head)
1708{
1709 struct ceph_mds_request *req, *nreq;
1710
1711 list_for_each_entry_safe(req, nreq, head, r_wait) {
1712 list_del_init(&req->r_wait);
1713 __do_request(mdsc, req);
1714 }
1715}
1716
1717/*
1718 * Wake up threads with requests pending for @mds, so that they can
29790f26 1719 * resubmit their requests to a possibly different mds.
2f2dc053 1720 */
29790f26 1721static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 1722{
44ca18f2
SW		 1723 struct ceph_mds_request *req;
1724 struct rb_node *p;
2f2dc053
SW		 1725
1726 dout("kick_requests mds%d\n", mds);
44ca18f2
SW		 1727 for (p = rb_first(&mdsc->request_tree); p; p = rb_next(p)) {
1728 req = rb_entry(p, struct ceph_mds_request, r_node);
1729 if (req->r_got_unsafe)
1730 continue;
1731 if (req->r_session &&
1732 req->r_session->s_mds == mds) {
1733 dout(" kicking tid %llu\n", req->r_tid);
1734 put_request_session(req);
1735 __do_request(mdsc, req);
2f2dc053
SW		 1736 }
1737 }
1738}
1739
1740void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
1741 struct ceph_mds_request *req)
1742{
1743 dout("submit_request on %p\n", req);
1744 mutex_lock(&mdsc->mutex);
1745 __register_request(mdsc, req, NULL);
1746 __do_request(mdsc, req);
1747 mutex_unlock(&mdsc->mutex);
1748}
1749
1750/*
1751 * Synchrously perform an mds request. Take care of all of the
1752 * session setup, forwarding, retry details.
1753 */
1754int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
1755 struct inode *dir,
1756 struct ceph_mds_request *req)
1757{
1758 int err;
1759
1760 dout("do_request on %p\n", req);
1761
1762 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
1763 if (req->r_inode)
1764 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
1765 if (req->r_locked_dir)
1766 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
1767 if (req->r_old_dentry)
1768 ceph_get_cap_refs(
1769 ceph_inode(req->r_old_dentry->d_parent->d_inode),
1770 CEPH_CAP_PIN);
1771
1772 /* issue */
1773 mutex_lock(&mdsc->mutex);
1774 __register_request(mdsc, req, dir);
1775 __do_request(mdsc, req);
1776
e1518c7c
SW		 1777 if (req->r_err) {
1778 err = req->r_err;
1779 __unregister_request(mdsc, req);
1780 dout("do_request early error %d\n", err);
1781 goto out;
2f2dc053
SW		 1782 }
1783
e1518c7c
SW		 1784 /* wait */
1785 mutex_unlock(&mdsc->mutex);
1786 dout("do_request waiting\n");
1787 if (req->r_timeout) {
aa91647c 1788 err = (long)wait_for_completion_killable_timeout(
e1518c7c
SW		 1789 &req->r_completion, req->r_timeout);
1790 if (err == 0)
1791 err = -EIO;
1792 } else {
aa91647c 1793 err = wait_for_completion_killable(&req->r_completion);
e1518c7c
SW		 1794 }
1795 dout("do_request waited, got %d\n", err);
1796 mutex_lock(&mdsc->mutex);
5b1daecd 1797
e1518c7c
SW		 1798 /* only abort if we didn't race with a real reply */
1799 if (req->r_got_result) {
1800 err = le32_to_cpu(req->r_reply_info.head->result);
1801 } else if (err < 0) {
1802 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 1803
1804 /*
1805 * ensure we aren't running concurrently with
1806 * ceph_fill_trace or ceph_readdir_prepopulate, which
1807 * rely on locks (dir mutex) held by our caller.
1808 */
1809 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 1810 req->r_err = err;
1811 req->r_aborted = true;
b4556396 1812 mutex_unlock(&req->r_fill_mutex);
5b1daecd 1813
e1518c7c 1814 if (req->r_locked_dir &&
167c9e35
SW		 1815 (req->r_op & CEPH_MDS_OP_WRITE))
1816 ceph_invalidate_dir_request(req);
2f2dc053 1817 } else {
e1518c7c 1818 err = req->r_err;
2f2dc053 1819 }
2f2dc053 1820
e1518c7c
SW		 1821out:
1822 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 1823 dout("do_request %p done, result %d\n", req, err);
1824 return err;
1825}
1826
167c9e35
SW		 1827/*
1828 * Invalidate dir I_COMPLETE, dentry lease state on an aborted MDS
1829 * namespace request.
1830 */
1831void ceph_invalidate_dir_request(struct ceph_mds_request *req)
1832{
1833 struct inode *inode = req->r_locked_dir;
1834 struct ceph_inode_info *ci = ceph_inode(inode);
1835
1836 dout("invalidate_dir_request %p (I_COMPLETE, lease(s))\n", inode);
1837 spin_lock(&inode->i_lock);
1838 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1839 ci->i_release_count++;
1840 spin_unlock(&inode->i_lock);
1841
1842 if (req->r_dentry)
1843 ceph_invalidate_dentry_lease(req->r_dentry);
1844 if (req->r_old_dentry)
1845 ceph_invalidate_dentry_lease(req->r_old_dentry);
1846}
1847
2f2dc053
SW		 1848/*
1849 * Handle mds reply.
1850 *
1851 * We take the session mutex and parse and process the reply immediately.
1852 * This preserves the logical ordering of replies, capabilities, etc., sent
1853 * by the MDS as they are applied to our local cache.
1854 */
1855static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
1856{
1857 struct ceph_mds_client *mdsc = session->s_mdsc;
1858 struct ceph_mds_request *req;
1859 struct ceph_mds_reply_head *head = msg->front.iov_base;
1860 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
1861 u64 tid;
1862 int err, result;
2600d2dd 1863 int mds = session->s_mds;
2f2dc053 1864
2f2dc053
SW		 1865 if (msg->front.iov_len < sizeof(*head)) {
1866 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 1867 ceph_msg_dump(msg);
2f2dc053
SW		 1868 return;
1869 }
1870
1871 /* get request, session */
6df058c0 1872 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 1873 mutex_lock(&mdsc->mutex);
1874 req = __lookup_request(mdsc, tid);
1875 if (!req) {
1876 dout("handle_reply on unknown tid %llu\n", tid);
1877 mutex_unlock(&mdsc->mutex);
1878 return;
1879 }
1880 dout("handle_reply %p\n", req);
2f2dc053
SW		 1881
1882 /* correct session? */
d96d6049 1883 if (req->r_session != session) {
2f2dc053
SW		 1884 pr_err("mdsc_handle_reply got %llu on session mds%d"
1885 " not mds%d\n", tid, session->s_mds,
1886 req->r_session ? req->r_session->s_mds : -1);
1887 mutex_unlock(&mdsc->mutex);
1888 goto out;
1889 }
1890
1891 /* dup? */
1892 if ((req->r_got_unsafe && !head->safe) ||
1893 (req->r_got_safe && head->safe)) {
1894 pr_warning("got a dup %s reply on %llu from mds%d\n",
1895 head->safe ? "safe" : "unsafe", tid, mds);
1896 mutex_unlock(&mdsc->mutex);
1897 goto out;
1898 }
85792d0d
SW		 1899 if (req->r_got_safe && !head->safe) {
1900 pr_warning("got unsafe after safe on %llu from mds%d\n",
1901 tid, mds);
1902 mutex_unlock(&mdsc->mutex);
1903 goto out;
1904 }
2f2dc053
SW		 1905
1906 result = le32_to_cpu(head->result);
1907
1908 /*
1909 * Tolerate 2 consecutive ESTALEs from the same mds.
1910 * FIXME: we should be looking at the cap migrate_seq.
1911 */
1912 if (result == -ESTALE) {
1913 req->r_direct_mode = USE_AUTH_MDS;
1914 req->r_num_stale++;
1915 if (req->r_num_stale <= 2) {
1916 __do_request(mdsc, req);
1917 mutex_unlock(&mdsc->mutex);
1918 goto out;
1919 }
1920 } else {
1921 req->r_num_stale = 0;
1922 }
1923
1924 if (head->safe) {
1925 req->r_got_safe = true;
1926 __unregister_request(mdsc, req);
1927 complete(&req->r_safe_completion);
1928
1929 if (req->r_got_unsafe) {
1930 /*
1931 * We already handled the unsafe response, now do the
1932 * cleanup. No need to examine the response; the MDS
1933 * doesn't include any result info in the safe
1934 * response. And even if it did, there is nothing
1935 * useful we could do with a revised return value.
1936 */
1937 dout("got safe reply %llu, mds%d\n", tid, mds);
1938 list_del_init(&req->r_unsafe_item);
1939
1940 /* last unsafe request during umount? */
44ca18f2 1941 if (mdsc->stopping && !__get_oldest_req(mdsc))
2f2dc053
SW		 1942 complete(&mdsc->safe_umount_waiters);
1943 mutex_unlock(&mdsc->mutex);
1944 goto out;
1945 }
e1518c7c 1946 } else {
2f2dc053
SW		 1947 req->r_got_unsafe = true;
1948 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
1949 }
1950
1951 dout("handle_reply tid %lld result %d\n", tid, result);
1952 rinfo = &req->r_reply_info;
1953 err = parse_reply_info(msg, rinfo);
1954 mutex_unlock(&mdsc->mutex);
1955
1956 mutex_lock(&session->s_mutex);
1957 if (err < 0) {
1958 pr_err("mdsc_handle_reply got corrupt reply mds%d\n", mds);
9ec7cab1 1959 ceph_msg_dump(msg);
2f2dc053
SW		 1960 goto out_err;
1961 }
1962
1963 /* snap trace */
1964 if (rinfo->snapblob_len) {
1965 down_write(&mdsc->snap_rwsem);
1966 ceph_update_snap_trace(mdsc, rinfo->snapblob,
1967 rinfo->snapblob + rinfo->snapblob_len,
1968 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP);
1969 downgrade_write(&mdsc->snap_rwsem);
1970 } else {
1971 down_read(&mdsc->snap_rwsem);
1972 }
1973
1974 /* insert trace into our cache */
b4556396 1975 mutex_lock(&req->r_fill_mutex);
2f2dc053
SW		 1976 err = ceph_fill_trace(mdsc->client->sb, req, req->r_session);
1977 if (err == 0) {
1978 if (result == 0 && rinfo->dir_nr)
1979 ceph_readdir_prepopulate(req, req->r_session);
1980 ceph_unreserve_caps(&req->r_caps_reservation);
1981 }
b4556396 1982 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 1983
1984 up_read(&mdsc->snap_rwsem);
1985out_err:
e1518c7c
SW		 1986 mutex_lock(&mdsc->mutex);
1987 if (!req->r_aborted) {
1988 if (err) {
1989 req->r_err = err;
1990 } else {
1991 req->r_reply = msg;
1992 ceph_msg_get(msg);
1993 req->r_got_result = true;
1994 }
2f2dc053 1995 } else {
e1518c7c 1996 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 1997 }
e1518c7c 1998 mutex_unlock(&mdsc->mutex);
2f2dc053 1999
2b2300d6 2000 ceph_add_cap_releases(mdsc, req->r_session, -1);
2f2dc053
SW		 2001 mutex_unlock(&session->s_mutex);
2002
2003 /* kick calling process */
2004 complete_request(mdsc, req);
2005out:
2006 ceph_mdsc_put_request(req);
2007 return;
2008}
2009
2010
2011
2012/*
2013 * handle mds notification that our request has been forwarded.
2014 */
2600d2dd
SW		 2015static void handle_forward(struct ceph_mds_client *mdsc,
2016 struct ceph_mds_session *session,
2017 struct ceph_msg *msg)
2f2dc053
SW		 2018{
2019 struct ceph_mds_request *req;
a1ea787c 2020 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2021 u32 next_mds;
2022 u32 fwd_seq;
2f2dc053
SW		 2023 int err = -EINVAL;
2024 void *p = msg->front.iov_base;
2025 void *end = p + msg->front.iov_len;
2f2dc053 2026
a1ea787c 2027 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2028 next_mds = ceph_decode_32(&p);
2029 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2030
2031 mutex_lock(&mdsc->mutex);
2032 req = __lookup_request(mdsc, tid);
2033 if (!req) {
2a8e5e36 2034 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2035 goto out; /* dup reply? */
2036 }
2037
2a8e5e36
SW		 2038 if (req->r_aborted) {
2039 dout("forward tid %llu aborted, unregistering\n", tid);
2040 __unregister_request(mdsc, req);
2041 } else if (fwd_seq <= req->r_num_fwd) {
2042 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2043 tid, next_mds, req->r_num_fwd, fwd_seq);
2044 } else {
2045 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2046 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2047 BUG_ON(req->r_err);
2048 BUG_ON(req->r_got_result);
2f2dc053
SW		 2049 req->r_num_fwd = fwd_seq;
2050 req->r_resend_mds = next_mds;
2051 put_request_session(req);
2052 __do_request(mdsc, req);
2053 }
2054 ceph_mdsc_put_request(req);
2055out:
2056 mutex_unlock(&mdsc->mutex);
2057 return;
2058
2059bad:
2060 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2061}
2062
2063/*
2064 * handle a mds session control message
2065 */
2066static void handle_session(struct ceph_mds_session *session,
2067 struct ceph_msg *msg)
2068{
2069 struct ceph_mds_client *mdsc = session->s_mdsc;
2070 u32 op;
2071 u64 seq;
2600d2dd 2072 int mds = session->s_mds;
2f2dc053
SW		 2073 struct ceph_mds_session_head *h = msg->front.iov_base;
2074 int wake = 0;
2075
2f2dc053
SW		 2076 /* decode */
2077 if (msg->front.iov_len != sizeof(*h))
2078 goto bad;
2079 op = le32_to_cpu(h->op);
2080 seq = le64_to_cpu(h->seq);
2081
2082 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2083 if (op == CEPH_SESSION_CLOSE)
2084 __unregister_session(mdsc, session);
2f2dc053
SW		 2085 /* FIXME: this ttl calculation is generous */
2086 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2087 mutex_unlock(&mdsc->mutex);
2088
2089 mutex_lock(&session->s_mutex);
2090
2091 dout("handle_session mds%d %s %p state %s seq %llu\n",
2092 mds, ceph_session_op_name(op), session,
2093 session_state_name(session->s_state), seq);
2094
2095 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2096 session->s_state = CEPH_MDS_SESSION_OPEN;
2097 pr_info("mds%d came back\n", session->s_mds);
2098 }
2099
2100 switch (op) {
2101 case CEPH_SESSION_OPEN:
29790f26
SW		 2102 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2103 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2104 session->s_state = CEPH_MDS_SESSION_OPEN;
2105 renewed_caps(mdsc, session, 0);
2106 wake = 1;
2107 if (mdsc->stopping)
2108 __close_session(mdsc, session);
2109 break;
2110
2111 case CEPH_SESSION_RENEWCAPS:
2112 if (session->s_renew_seq == seq)
2113 renewed_caps(mdsc, session, 1);
2114 break;
2115
2116 case CEPH_SESSION_CLOSE:
29790f26
SW		 2117 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2118 pr_info("mds%d reconnect denied\n", session->s_mds);
2f2dc053
SW		 2119 remove_session_caps(session);
2120 wake = 1; /* for good measure */
2121 complete(&mdsc->session_close_waiters);
29790f26 2122 kick_requests(mdsc, mds);
2f2dc053
SW		 2123 break;
2124
2125 case CEPH_SESSION_STALE:
2126 pr_info("mds%d caps went stale, renewing\n",
2127 session->s_mds);
2128 spin_lock(&session->s_cap_lock);
2129 session->s_cap_gen++;
2130 session->s_cap_ttl = 0;
2131 spin_unlock(&session->s_cap_lock);
2132 send_renew_caps(mdsc, session);
2133 break;
2134
2135 case CEPH_SESSION_RECALL_STATE:
2136 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2137 break;
2138
2139 default:
2140 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2141 WARN_ON(1);
2142 }
2143
2144 mutex_unlock(&session->s_mutex);
2145 if (wake) {
2146 mutex_lock(&mdsc->mutex);
2147 __wake_requests(mdsc, &session->s_waiting);
2148 mutex_unlock(&mdsc->mutex);
2149 }
2150 return;
2151
2152bad:
2153 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2154 (int)msg->front.iov_len);
9ec7cab1 2155 ceph_msg_dump(msg);
2f2dc053
SW		 2156 return;
2157}
2158
2159
2160/*
2161 * called under session->mutex.
2162 */
2163static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2164 struct ceph_mds_session *session)
2165{
2166 struct ceph_mds_request *req, *nreq;
2167 int err;
2168
2169 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2170
2171 mutex_lock(&mdsc->mutex);
2172 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2173 err = __prepare_send_request(mdsc, req, session->s_mds);
2174 if (!err) {
2175 ceph_msg_get(req->r_request);
2176 ceph_con_send(&session->s_con, req->r_request);
2177 }
2178 }
2179 mutex_unlock(&mdsc->mutex);
2180}
2181
2182/*
2183 * Encode information about a cap for a reconnect with the MDS.
2184 */
2f2dc053
SW		 2185static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2186 void *arg)
2187{
93cea5be 2188 struct ceph_mds_cap_reconnect rec;
2f2dc053 2189 struct ceph_inode_info *ci;
93cea5be 2190 struct ceph_pagelist *pagelist = arg;
2f2dc053
SW		 2191 char *path;
2192 int pathlen, err;
2193 u64 pathbase;
2194 struct dentry *dentry;
2195
2196 ci = cap->ci;
2197
2198 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2199 inode, ceph_vinop(inode), cap, cap->cap_id,
2200 ceph_cap_string(cap->issued));
93cea5be
SW		 2201 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2202 if (err)
2203 return err;
2f2dc053
SW		 2204
2205 dentry = d_find_alias(inode);
2206 if (dentry) {
2207 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2208 if (IS_ERR(path)) {
2209 err = PTR_ERR(path);
2210 BUG_ON(err);
2211 }
2212 } else {
2213 path = NULL;
2214 pathlen = 0;
2215 }
93cea5be
SW		 2216 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2217 if (err)
2218 goto out;
2f2dc053 2219
2f2dc053
SW		 2220 spin_lock(&inode->i_lock);
2221 cap->seq = 0; /* reset cap seq */
2222 cap->issue_seq = 0; /* and issue_seq */
93cea5be
SW		 2223 rec.cap_id = cpu_to_le64(cap->cap_id);
2224 rec.pathbase = cpu_to_le64(pathbase);
2225 rec.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2226 rec.issued = cpu_to_le32(cap->issued);
2227 rec.size = cpu_to_le64(inode->i_size);
2228 ceph_encode_timespec(&rec.mtime, &inode->i_mtime);
2229 ceph_encode_timespec(&rec.atime, &inode->i_atime);
2230 rec.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2f2dc053
SW		 2231 spin_unlock(&inode->i_lock);
2232
93cea5be
SW		 2233 err = ceph_pagelist_append(pagelist, &rec, sizeof(rec));
2234
2235out:
2f2dc053
SW		 2236 kfree(path);
2237 dput(dentry);
93cea5be 2238 return err;
2f2dc053
SW		 2239}
2240
2241
2242/*
2243 * If an MDS fails and recovers, clients need to reconnect in order to
2244 * reestablish shared state. This includes all caps issued through
2245 * this session _and_ the snap_realm hierarchy. Because it's not
2246 * clear which snap realms the mds cares about, we send everything we
2247 * know about.. that ensures we'll then get any new info the
2248 * recovering MDS might have.
2249 *
2250 * This is a relatively heavyweight operation, but it's rare.
2251 *
2252 * called with mdsc->mutex held.
2253 */
34b6c855
SW		 2254static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2255 struct ceph_mds_session *session)
2f2dc053 2256{
2f2dc053 2257 struct ceph_msg *reply;
a105f00c 2258 struct rb_node *p;
34b6c855 2259 int mds = session->s_mds;
9abf82b8 2260 int err = -ENOMEM;
93cea5be 2261 struct ceph_pagelist *pagelist;
2f2dc053 2262
34b6c855 2263 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2264
93cea5be
SW		 2265 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2266 if (!pagelist)
2267 goto fail_nopagelist;
2268 ceph_pagelist_init(pagelist);
2269
34d23762 2270 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS);
a79832f2 2271 if (!reply)
93cea5be 2272 goto fail_nomsg;
93cea5be 2273
34b6c855
SW		 2274 mutex_lock(&session->s_mutex);
2275 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2276 session->s_seq = 0;
2f2dc053 2277
34b6c855
SW		 2278 ceph_con_open(&session->s_con,
2279 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2f2dc053 2280
34b6c855
SW		 2281 /* replay unsafe requests */
2282 replay_unsafe_requests(mdsc, session);
2f2dc053
SW		 2283
2284 down_read(&mdsc->snap_rwsem);
2285
2f2dc053
SW		 2286 dout("session %p state %s\n", session,
2287 session_state_name(session->s_state));
2288
e01a5946
SW		 2289 /* drop old cap expires; we're about to reestablish that state */
2290 discard_cap_releases(mdsc, session);
2291
2f2dc053 2292 /* traverse this session's caps */
93cea5be
SW		 2293 err = ceph_pagelist_encode_32(pagelist, session->s_nr_caps);
2294 if (err)
2295 goto fail;
2296 err = iterate_session_caps(session, encode_caps_cb, pagelist);
2f2dc053 2297 if (err < 0)
9abf82b8 2298 goto fail;
2f2dc053
SW		 2299
2300 /*
2301 * snaprealms. we provide mds with the ino, seq (version), and
2302 * parent for all of our realms. If the mds has any newer info,
2303 * it will tell us.
2304 */
a105f00c
SW		 2305 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2306 struct ceph_snap_realm *realm =
2307 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 2308 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 2309
2310 dout(" adding snap realm %llx seq %lld parent %llx\n",
2311 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 2312 sr_rec.ino = cpu_to_le64(realm->ino);
2313 sr_rec.seq = cpu_to_le64(realm->seq);
2314 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2315 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2316 if (err)
2317 goto fail;
2f2dc053 2318 }
2f2dc053 2319
93cea5be
SW		 2320 reply->pagelist = pagelist;
2321 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2322 reply->nr_pages = calc_pages_for(0, pagelist->length);
2f2dc053
SW		 2323 ceph_con_send(&session->s_con, reply);
2324
9abf82b8
SW		 2325 mutex_unlock(&session->s_mutex);
2326
2327 mutex_lock(&mdsc->mutex);
2328 __wake_requests(mdsc, &session->s_waiting);
2329 mutex_unlock(&mdsc->mutex);
2330
2f2dc053 2331 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 2332 return;
2333
93cea5be 2334fail:
2f2dc053 2335 ceph_msg_put(reply);
9abf82b8
SW		 2336 up_read(&mdsc->snap_rwsem);
2337 mutex_unlock(&session->s_mutex);
93cea5be
SW		 2338fail_nomsg:
2339 ceph_pagelist_release(pagelist);
2340 kfree(pagelist);
2341fail_nopagelist:
9abf82b8 2342 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 2343 return;
2f2dc053
SW		 2344}
2345
2346
2347/*
2348 * compare old and new mdsmaps, kicking requests
2349 * and closing out old connections as necessary
2350 *
2351 * called under mdsc->mutex.
2352 */
2353static void check_new_map(struct ceph_mds_client *mdsc,
2354 struct ceph_mdsmap *newmap,
2355 struct ceph_mdsmap *oldmap)
2356{
2357 int i;
2358 int oldstate, newstate;
2359 struct ceph_mds_session *s;
2360
2361 dout("check_new_map new %u old %u\n",
2362 newmap->m_epoch, oldmap->m_epoch);
2363
2364 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2365 if (mdsc->sessions[i] == NULL)
2366 continue;
2367 s = mdsc->sessions[i];
2368 oldstate = ceph_mdsmap_get_state(oldmap, i);
2369 newstate = ceph_mdsmap_get_state(newmap, i);
2370
2371 dout("check_new_map mds%d state %s -> %s (session %s)\n",
2372 i, ceph_mds_state_name(oldstate),
2373 ceph_mds_state_name(newstate),
2374 session_state_name(s->s_state));
2375
2376 if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
2377 ceph_mdsmap_get_addr(newmap, i),
2378 sizeof(struct ceph_entity_addr))) {
2379 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2380 /* the session never opened, just close it
2381 * out now */
2382 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 2383 __unregister_session(mdsc, s);
2f2dc053
SW		 2384 } else {
2385 /* just close it */
2386 mutex_unlock(&mdsc->mutex);
2387 mutex_lock(&s->s_mutex);
2388 mutex_lock(&mdsc->mutex);
2389 ceph_con_close(&s->s_con);
2390 mutex_unlock(&s->s_mutex);
2391 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2392 }
2393
2394 /* kick any requests waiting on the recovering mds */
29790f26 2395 kick_requests(mdsc, i);
2f2dc053
SW		 2396 } else if (oldstate == newstate) {
2397 continue; /* nothing new with this mds */
2398 }
2399
2400 /*
2401 * send reconnect?
2402 */
2403 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 2404 newstate >= CEPH_MDS_STATE_RECONNECT) {
2405 mutex_unlock(&mdsc->mutex);
2406 send_mds_reconnect(mdsc, s);
2407 mutex_lock(&mdsc->mutex);
2408 }
2f2dc053
SW		 2409
2410 /*
29790f26 2411 * kick request on any mds that has gone active.
2f2dc053
SW		 2412 */
2413 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2414 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 2415 if (oldstate != CEPH_MDS_STATE_CREATING &&
2416 oldstate != CEPH_MDS_STATE_STARTING)
2417 pr_info("mds%d recovery completed\n", s->s_mds);
2418 kick_requests(mdsc, i);
2f2dc053 2419 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 2420 wake_up_session_caps(s, 1);
2f2dc053
SW		 2421 }
2422 }
2423}
2424
2425
2426
2427/*
2428 * leases
2429 */
2430
2431/*
2432 * caller must hold session s_mutex, dentry->d_lock
2433 */
2434void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
2435{
2436 struct ceph_dentry_info *di = ceph_dentry(dentry);
2437
2438 ceph_put_mds_session(di->lease_session);
2439 di->lease_session = NULL;
2440}
2441
2600d2dd
SW		 2442static void handle_lease(struct ceph_mds_client *mdsc,
2443 struct ceph_mds_session *session,
2444 struct ceph_msg *msg)
2f2dc053
SW		 2445{
2446 struct super_block *sb = mdsc->client->sb;
2447 struct inode *inode;
2f2dc053
SW		 2448 struct ceph_inode_info *ci;
2449 struct dentry *parent, *dentry;
2450 struct ceph_dentry_info *di;
2600d2dd 2451 int mds = session->s_mds;
2f2dc053 2452 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 2453 u32 seq;
2f2dc053
SW		 2454 struct ceph_vino vino;
2455 int mask;
2456 struct qstr dname;
2457 int release = 0;
2458
2f2dc053
SW		 2459 dout("handle_lease from mds%d\n", mds);
2460
2461 /* decode */
2462 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
2463 goto bad;
2464 vino.ino = le64_to_cpu(h->ino);
2465 vino.snap = CEPH_NOSNAP;
2466 mask = le16_to_cpu(h->mask);
1e5ea23d 2467 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 2468 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
2469 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
2470 if (dname.len != get_unaligned_le32(h+1))
2471 goto bad;
2472
2f2dc053
SW		 2473 mutex_lock(&session->s_mutex);
2474 session->s_seq++;
2475
2476 /* lookup inode */
2477 inode = ceph_find_inode(sb, vino);
1e5ea23d
SW		 2478 dout("handle_lease %s, mask %d, ino %llx %p %.*s\n",
2479 ceph_lease_op_name(h->action), mask, vino.ino, inode,
2480 dname.len, dname.name);
2f2dc053
SW		 2481 if (inode == NULL) {
2482 dout("handle_lease no inode %llx\n", vino.ino);
2483 goto release;
2484 }
2485 ci = ceph_inode(inode);
2486
2487 /* dentry */
2488 parent = d_find_alias(inode);
2489 if (!parent) {
2490 dout("no parent dentry on inode %p\n", inode);
2491 WARN_ON(1);
2492 goto release; /* hrm... */
2493 }
2494 dname.hash = full_name_hash(dname.name, dname.len);
2495 dentry = d_lookup(parent, &dname);
2496 dput(parent);
2497 if (!dentry)
2498 goto release;
2499
2500 spin_lock(&dentry->d_lock);
2501 di = ceph_dentry(dentry);
2502 switch (h->action) {
2503 case CEPH_MDS_LEASE_REVOKE:
2504 if (di && di->lease_session == session) {
1e5ea23d
SW		 2505 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
2506 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 2507 __ceph_mdsc_drop_dentry_lease(dentry);
2508 }
2509 release = 1;
2510 break;
2511
2512 case CEPH_MDS_LEASE_RENEW:
2513 if (di && di->lease_session == session &&
2514 di->lease_gen == session->s_cap_gen &&
2515 di->lease_renew_from &&
2516 di->lease_renew_after == 0) {
2517 unsigned long duration =
2518 le32_to_cpu(h->duration_ms) * HZ / 1000;
2519
1e5ea23d 2520 di->lease_seq = seq;
2f2dc053
SW		 2521 dentry->d_time = di->lease_renew_from + duration;
2522 di->lease_renew_after = di->lease_renew_from +
2523 (duration >> 1);
2524 di->lease_renew_from = 0;
2525 }
2526 break;
2527 }
2528 spin_unlock(&dentry->d_lock);
2529 dput(dentry);
2530
2531 if (!release)
2532 goto out;
2533
2534release:
2535 /* let's just reuse the same message */
2536 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
2537 ceph_msg_get(msg);
2538 ceph_con_send(&session->s_con, msg);
2539
2540out:
2541 iput(inode);
2542 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 2543 return;
2544
2545bad:
2546 pr_err("corrupt lease message\n");
9ec7cab1 2547 ceph_msg_dump(msg);
2f2dc053
SW		 2548}
2549
2550void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
2551 struct inode *inode,
2552 struct dentry *dentry, char action,
2553 u32 seq)
2554{
2555 struct ceph_msg *msg;
2556 struct ceph_mds_lease *lease;
2557 int len = sizeof(*lease) + sizeof(u32);
2558 int dnamelen = 0;
2559
2560 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
2561 inode, dentry, ceph_lease_op_name(action), session->s_mds);
2562 dnamelen = dentry->d_name.len;
2563 len += dnamelen;
2564
34d23762 2565 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS);
a79832f2 2566 if (!msg)
2f2dc053
SW		 2567 return;
2568 lease = msg->front.iov_base;
2569 lease->action = action;
dd1c9057 2570 lease->mask = cpu_to_le16(1);
2f2dc053
SW		 2571 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
2572 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
2573 lease->seq = cpu_to_le32(seq);
2574 put_unaligned_le32(dnamelen, lease + 1);
2575 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
2576
2577 /*
2578 * if this is a preemptive lease RELEASE, no need to
2579 * flush request stream, since the actual request will
2580 * soon follow.
2581 */
2582 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
2583
2584 ceph_con_send(&session->s_con, msg);
2585}
2586
2587/*
2588 * Preemptively release a lease we expect to invalidate anyway.
2589 * Pass @inode always, @dentry is optional.
2590 */
2591void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2592 struct dentry *dentry, int mask)
2593{
2594 struct ceph_dentry_info *di;
2595 struct ceph_mds_session *session;
2596 u32 seq;
2597
2598 BUG_ON(inode == NULL);
2599 BUG_ON(dentry == NULL);
dd1c9057 2600 BUG_ON(mask == 0);
2f2dc053
SW		 2601
2602 /* is dentry lease valid? */
2603 spin_lock(&dentry->d_lock);
2604 di = ceph_dentry(dentry);
2605 if (!di || !di->lease_session ||
2606 di->lease_session->s_mds < 0 ||
2607 di->lease_gen != di->lease_session->s_cap_gen ||
2608 !time_before(jiffies, dentry->d_time)) {
2609 dout("lease_release inode %p dentry %p -- "
2610 "no lease on %d\n",
2611 inode, dentry, mask);
2612 spin_unlock(&dentry->d_lock);
2613 return;
2614 }
2615
2616 /* we do have a lease on this dentry; note mds and seq */
2617 session = ceph_get_mds_session(di->lease_session);
2618 seq = di->lease_seq;
2619 __ceph_mdsc_drop_dentry_lease(dentry);
2620 spin_unlock(&dentry->d_lock);
2621
2622 dout("lease_release inode %p dentry %p mask %d to mds%d\n",
2623 inode, dentry, mask, session->s_mds);
2624 ceph_mdsc_lease_send_msg(session, inode, dentry,
2625 CEPH_MDS_LEASE_RELEASE, seq);
2626 ceph_put_mds_session(session);
2627}
2628
2629/*
2630 * drop all leases (and dentry refs) in preparation for umount
2631 */
2632static void drop_leases(struct ceph_mds_client *mdsc)
2633{
2634 int i;
2635
2636 dout("drop_leases\n");
2637 mutex_lock(&mdsc->mutex);
2638 for (i = 0; i < mdsc->max_sessions; i++) {
2639 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2640 if (!s)
2641 continue;
2642 mutex_unlock(&mdsc->mutex);
2643 mutex_lock(&s->s_mutex);
2644 mutex_unlock(&s->s_mutex);
2645 ceph_put_mds_session(s);
2646 mutex_lock(&mdsc->mutex);
2647 }
2648 mutex_unlock(&mdsc->mutex);
2649}
2650
2651
2652
2653/*
2654 * delayed work -- periodically trim expired leases, renew caps with mds
2655 */
2656static void schedule_delayed(struct ceph_mds_client *mdsc)
2657{
2658 int delay = 5;
2659 unsigned hz = round_jiffies_relative(HZ * delay);
2660 schedule_delayed_work(&mdsc->delayed_work, hz);
2661}
2662
2663static void delayed_work(struct work_struct *work)
2664{
2665 int i;
2666 struct ceph_mds_client *mdsc =
2667 container_of(work, struct ceph_mds_client, delayed_work.work);
2668 int renew_interval;
2669 int renew_caps;
2670
2671 dout("mdsc delayed_work\n");
afcdaea3 2672 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 2673
2674 mutex_lock(&mdsc->mutex);
2675 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
2676 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
2677 mdsc->last_renew_caps);
2678 if (renew_caps)
2679 mdsc->last_renew_caps = jiffies;
2680
2681 for (i = 0; i < mdsc->max_sessions; i++) {
2682 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
2683 if (s == NULL)
2684 continue;
2685 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
2686 dout("resending session close request for mds%d\n",
2687 s->s_mds);
2688 request_close_session(mdsc, s);
2689 ceph_put_mds_session(s);
2690 continue;
2691 }
2692 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
2693 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
2694 s->s_state = CEPH_MDS_SESSION_HUNG;
2695 pr_info("mds%d hung\n", s->s_mds);
2696 }
2697 }
2698 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
2699 /* this mds is failed or recovering, just wait */
2700 ceph_put_mds_session(s);
2701 continue;
2702 }
2703 mutex_unlock(&mdsc->mutex);
2704
2705 mutex_lock(&s->s_mutex);
2706 if (renew_caps)
2707 send_renew_caps(mdsc, s);
2708 else
2709 ceph_con_keepalive(&s->s_con);
2b2300d6 2710 ceph_add_cap_releases(mdsc, s, -1);
aab53dd9
SW		 2711 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
2712 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 2713 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 2714 mutex_unlock(&s->s_mutex);
2715 ceph_put_mds_session(s);
2716
2717 mutex_lock(&mdsc->mutex);
2718 }
2719 mutex_unlock(&mdsc->mutex);
2720
2721 schedule_delayed(mdsc);
2722}
2723
2724
5f44f142 2725int ceph_mdsc_init(struct ceph_mds_client *mdsc, struct ceph_client *client)
2f2dc053
SW		 2726{
2727 mdsc->client = client;
2728 mutex_init(&mdsc->mutex);
2729 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
2d06eeb8
CR		 2730 if (mdsc->mdsmap == NULL)
2731 return -ENOMEM;
2732
2f2dc053
SW		 2733 init_completion(&mdsc->safe_umount_waiters);
2734 init_completion(&mdsc->session_close_waiters);
2735 INIT_LIST_HEAD(&mdsc->waiting_for_map);
2736 mdsc->sessions = NULL;
2737 mdsc->max_sessions = 0;
2738 mdsc->stopping = 0;
2739 init_rwsem(&mdsc->snap_rwsem);
a105f00c 2740 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 2741 INIT_LIST_HEAD(&mdsc->snap_empty);
2742 spin_lock_init(&mdsc->snap_empty_lock);
2743 mdsc->last_tid = 0;
44ca18f2 2744 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 2745 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
2746 mdsc->last_renew_caps = jiffies;
2747 INIT_LIST_HEAD(&mdsc->cap_delay_list);
2748 spin_lock_init(&mdsc->cap_delay_lock);
2749 INIT_LIST_HEAD(&mdsc->snap_flush_list);
2750 spin_lock_init(&mdsc->snap_flush_lock);
2751 mdsc->cap_flush_seq = 0;
2752 INIT_LIST_HEAD(&mdsc->cap_dirty);
2753 mdsc->num_cap_flushing = 0;
2754 spin_lock_init(&mdsc->cap_dirty_lock);
2755 init_waitqueue_head(&mdsc->cap_flushing_wq);
2756 spin_lock_init(&mdsc->dentry_lru_lock);
2757 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 2758
5f44f142 2759 return 0;
2f2dc053
SW		 2760}
2761
2762/*
2763 * Wait for safe replies on open mds requests. If we time out, drop
2764 * all requests from the tree to avoid dangling dentry refs.
2765 */
2766static void wait_requests(struct ceph_mds_client *mdsc)
2767{
2768 struct ceph_mds_request *req;
2769 struct ceph_client *client = mdsc->client;
2770
2771 mutex_lock(&mdsc->mutex);
44ca18f2 2772 if (__get_oldest_req(mdsc)) {
2f2dc053 2773 mutex_unlock(&mdsc->mutex);
44ca18f2 2774
2f2dc053
SW		 2775 dout("wait_requests waiting for requests\n");
2776 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
6b805185 2777 client->mount_args->mount_timeout * HZ);
2f2dc053
SW		 2778
2779 /* tear down remaining requests */
44ca18f2
SW		 2780 mutex_lock(&mdsc->mutex);
2781 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 2782 dout("wait_requests timed out on tid %llu\n",
2783 req->r_tid);
44ca18f2 2784 __unregister_request(mdsc, req);
2f2dc053
SW		 2785 }
2786 }
2787 mutex_unlock(&mdsc->mutex);
2788 dout("wait_requests done\n");
2789}
2790
2791/*
2792 * called before mount is ro, and before dentries are torn down.
2793 * (hmm, does this still race with new lookups?)
2794 */
2795void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
2796{
2797 dout("pre_umount\n");
2798 mdsc->stopping = 1;
2799
2800 drop_leases(mdsc);
afcdaea3 2801 ceph_flush_dirty_caps(mdsc);
2f2dc053 2802 wait_requests(mdsc);
17c688c3
SW		 2803
2804 /*
2805 * wait for reply handlers to drop their request refs and
2806 * their inode/dcache refs
2807 */
2808 ceph_msgr_flush();
2f2dc053
SW		 2809}
2810
2811/*
2812 * wait for all write mds requests to flush.
2813 */
2814static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
2815{
80fc7314 2816 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 2817 struct rb_node *n;
2f2dc053
SW		 2818
2819 mutex_lock(&mdsc->mutex);
2820 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 2821restart:
44ca18f2
SW		 2822 req = __get_oldest_req(mdsc);
2823 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 2824 /* find next request */
2825 n = rb_next(&req->r_node);
2826 if (n)
2827 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
2828 else
2829 nextreq = NULL;
44ca18f2
SW		 2830 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
2831 /* write op */
2832 ceph_mdsc_get_request(req);
80fc7314
SW		 2833 if (nextreq)
2834 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 2835 mutex_unlock(&mdsc->mutex);
2836 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
2837 req->r_tid, want_tid);
2838 wait_for_completion(&req->r_safe_completion);
2839 mutex_lock(&mdsc->mutex);
44ca18f2 2840 ceph_mdsc_put_request(req);
80fc7314
SW		 2841 if (!nextreq)
2842 break; /* next dne before, so we're done! */
2843 if (RB_EMPTY_NODE(&nextreq->r_node)) {
2844 /* next request was removed from tree */
2845 ceph_mdsc_put_request(nextreq);
2846 goto restart;
2847 }
2848 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 2849 }
80fc7314 2850 req = nextreq;
2f2dc053
SW		 2851 }
2852 mutex_unlock(&mdsc->mutex);
2853 dout("wait_unsafe_requests done\n");
2854}
2855
2856void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
2857{
2858 u64 want_tid, want_flush;
2859
56b7cf95
SW		 2860 if (mdsc->client->mount_state == CEPH_MOUNT_SHUTDOWN)
2861 return;
2862
2f2dc053
SW		 2863 dout("sync\n");
2864 mutex_lock(&mdsc->mutex);
2865 want_tid = mdsc->last_tid;
2866 want_flush = mdsc->cap_flush_seq;
2867 mutex_unlock(&mdsc->mutex);
2868 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
2869
afcdaea3 2870 ceph_flush_dirty_caps(mdsc);
2f2dc053
SW		 2871
2872 wait_unsafe_requests(mdsc, want_tid);
2873 wait_event(mdsc->cap_flushing_wq, check_cap_flush(mdsc, want_flush));
2874}
2875
2876
2877/*
2878 * called after sb is ro.
2879 */
2880void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
2881{
2882 struct ceph_mds_session *session;
2883 int i;
2884 int n;
2885 struct ceph_client *client = mdsc->client;
6b805185 2886 unsigned long started, timeout = client->mount_args->mount_timeout * HZ;
2f2dc053
SW		 2887
2888 dout("close_sessions\n");
2889
2890 mutex_lock(&mdsc->mutex);
2891
2892 /* close sessions */
2893 started = jiffies;
2894 while (time_before(jiffies, started + timeout)) {
2895 dout("closing sessions\n");
2896 n = 0;
2897 for (i = 0; i < mdsc->max_sessions; i++) {
2898 session = __ceph_lookup_mds_session(mdsc, i);
2899 if (!session)
2900 continue;
2901 mutex_unlock(&mdsc->mutex);
2902 mutex_lock(&session->s_mutex);
2903 __close_session(mdsc, session);
2904 mutex_unlock(&session->s_mutex);
2905 ceph_put_mds_session(session);
2906 mutex_lock(&mdsc->mutex);
2907 n++;
2908 }
2909 if (n == 0)
2910 break;
2911
2912 if (client->mount_state == CEPH_MOUNT_SHUTDOWN)
2913 break;
2914
2915 dout("waiting for sessions to close\n");
2916 mutex_unlock(&mdsc->mutex);
2917 wait_for_completion_timeout(&mdsc->session_close_waiters,
2918 timeout);
2919 mutex_lock(&mdsc->mutex);
2920 }
2921
2922 /* tear down remaining sessions */
2923 for (i = 0; i < mdsc->max_sessions; i++) {
2924 if (mdsc->sessions[i]) {
2925 session = get_session(mdsc->sessions[i]);
2600d2dd 2926 __unregister_session(mdsc, session);
2f2dc053
SW		 2927 mutex_unlock(&mdsc->mutex);
2928 mutex_lock(&session->s_mutex);
2929 remove_session_caps(session);
2930 mutex_unlock(&session->s_mutex);
2931 ceph_put_mds_session(session);
2932 mutex_lock(&mdsc->mutex);
2933 }
2934 }
2935
2936 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2937
2938 mutex_unlock(&mdsc->mutex);
2939
2940 ceph_cleanup_empty_realms(mdsc);
2941
2942 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
2943
2944 dout("stopped\n");
2945}
2946
2947void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2948{
2949 dout("stop\n");
2950 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
2951 if (mdsc->mdsmap)
2952 ceph_mdsmap_destroy(mdsc->mdsmap);
2953 kfree(mdsc->sessions);
2954}
2955
2956
2957/*
2958 * handle mds map update.
2959 */
2960void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2961{
2962 u32 epoch;
2963 u32 maplen;
2964 void *p = msg->front.iov_base;
2965 void *end = p + msg->front.iov_len;
2966 struct ceph_mdsmap *newmap, *oldmap;
2967 struct ceph_fsid fsid;
2968 int err = -EINVAL;
2969
2970 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
2971 ceph_decode_copy(&p, &fsid, sizeof(fsid));
0743304d
SW		 2972 if (ceph_check_fsid(mdsc->client, &fsid) < 0)
2973 return;
c89136ea
SW		 2974 epoch = ceph_decode_32(&p);
2975 maplen = ceph_decode_32(&p);
2f2dc053
SW		 2976 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
2977
2978 /* do we need it? */
2979 ceph_monc_got_mdsmap(&mdsc->client->monc, epoch);
2980 mutex_lock(&mdsc->mutex);
2981 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
2982 dout("handle_map epoch %u <= our %u\n",
2983 epoch, mdsc->mdsmap->m_epoch);
2984 mutex_unlock(&mdsc->mutex);
2985 return;
2986 }
2987
2988 newmap = ceph_mdsmap_decode(&p, end);
2989 if (IS_ERR(newmap)) {
2990 err = PTR_ERR(newmap);
2991 goto bad_unlock;
2992 }
2993
2994 /* swap into place */
2995 if (mdsc->mdsmap) {
2996 oldmap = mdsc->mdsmap;
2997 mdsc->mdsmap = newmap;
2998 check_new_map(mdsc, newmap, oldmap);
2999 ceph_mdsmap_destroy(oldmap);
3000 } else {
3001 mdsc->mdsmap = newmap; /* first mds map */
3002 }
3003 mdsc->client->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3004
3005 __wake_requests(mdsc, &mdsc->waiting_for_map);
3006
3007 mutex_unlock(&mdsc->mutex);
3008 schedule_delayed(mdsc);
3009 return;
3010
3011bad_unlock:
3012 mutex_unlock(&mdsc->mutex);
3013bad:
3014 pr_err("error decoding mdsmap %d\n", err);
3015 return;
3016}
3017
3018static struct ceph_connection *con_get(struct ceph_connection *con)
3019{
3020 struct ceph_mds_session *s = con->private;
3021
3022 if (get_session(s)) {
2600d2dd 3023 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3024 return con;
3025 }
3026 dout("mdsc con_get %p FAIL\n", s);
3027 return NULL;
3028}
3029
3030static void con_put(struct ceph_connection *con)
3031{
3032 struct ceph_mds_session *s = con->private;
3033
2f2dc053 3034 ceph_put_mds_session(s);
2600d2dd 3035 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3036}
3037
3038/*
3039 * if the client is unresponsive for long enough, the mds will kill
3040 * the session entirely.
3041 */
3042static void peer_reset(struct ceph_connection *con)
3043{
3044 struct ceph_mds_session *s = con->private;
7e70f0ed 3045 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3046
7e70f0ed
SW		 3047 pr_warning("mds%d closed our session\n", s->s_mds);
3048 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3049}
3050
3051static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3052{
3053 struct ceph_mds_session *s = con->private;
3054 struct ceph_mds_client *mdsc = s->s_mdsc;
3055 int type = le16_to_cpu(msg->hdr.type);
3056
2600d2dd
SW		 3057 mutex_lock(&mdsc->mutex);
3058 if (__verify_registered_session(mdsc, s) < 0) {
3059 mutex_unlock(&mdsc->mutex);
3060 goto out;
3061 }
3062 mutex_unlock(&mdsc->mutex);
3063
2f2dc053
SW		 3064 switch (type) {
3065 case CEPH_MSG_MDS_MAP:
3066 ceph_mdsc_handle_map(mdsc, msg);
3067 break;
3068 case CEPH_MSG_CLIENT_SESSION:
3069 handle_session(s, msg);
3070 break;
3071 case CEPH_MSG_CLIENT_REPLY:
3072 handle_reply(s, msg);
3073 break;
3074 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3075 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3076 break;
3077 case CEPH_MSG_CLIENT_CAPS:
3078 ceph_handle_caps(s, msg);
3079 break;
3080 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3081 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3082 break;
3083 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3084 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3085 break;
3086
3087 default:
3088 pr_err("received unknown message type %d %s\n", type,
3089 ceph_msg_type_name(type));
3090 }
2600d2dd 3091out:
2f2dc053
SW		 3092 ceph_msg_put(msg);
3093}
3094
4e7a5dcd
SW		 3095/*
3096 * authentication
3097 */
3098static int get_authorizer(struct ceph_connection *con,
3099 void **buf, int *len, int *proto,
3100 void **reply_buf, int *reply_len, int force_new)
3101{
3102 struct ceph_mds_session *s = con->private;
3103 struct ceph_mds_client *mdsc = s->s_mdsc;
3104 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3105 int ret = 0;
3106
3107 if (force_new && s->s_authorizer) {
3108 ac->ops->destroy_authorizer(ac, s->s_authorizer);
3109 s->s_authorizer = NULL;
3110 }
3111 if (s->s_authorizer == NULL) {
3112 if (ac->ops->create_authorizer) {
3113 ret = ac->ops->create_authorizer(
3114 ac, CEPH_ENTITY_TYPE_MDS,
3115 &s->s_authorizer,
3116 &s->s_authorizer_buf,
3117 &s->s_authorizer_buf_len,
3118 &s->s_authorizer_reply_buf,
3119 &s->s_authorizer_reply_buf_len);
3120 if (ret)
3121 return ret;
3122 }
3123 }
3124
3125 *proto = ac->protocol;
3126 *buf = s->s_authorizer_buf;
3127 *len = s->s_authorizer_buf_len;
3128 *reply_buf = s->s_authorizer_reply_buf;
3129 *reply_len = s->s_authorizer_reply_buf_len;
3130 return 0;
3131}
3132
3133
3134static int verify_authorizer_reply(struct ceph_connection *con, int len)
3135{
3136 struct ceph_mds_session *s = con->private;
3137 struct ceph_mds_client *mdsc = s->s_mdsc;
3138 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3139
3140 return ac->ops->verify_authorizer_reply(ac, s->s_authorizer, len);
3141}
3142
9bd2e6f8
SW		 3143static int invalidate_authorizer(struct ceph_connection *con)
3144{
3145 struct ceph_mds_session *s = con->private;
3146 struct ceph_mds_client *mdsc = s->s_mdsc;
3147 struct ceph_auth_client *ac = mdsc->client->monc.auth;
3148
3149 if (ac->ops->invalidate_authorizer)
3150 ac->ops->invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3151
3152 return ceph_monc_validate_auth(&mdsc->client->monc);
3153}
3154
9e32789f 3155static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 3156 .get = con_get,
3157 .put = con_put,
3158 .dispatch = dispatch,
4e7a5dcd
SW		 3159 .get_authorizer = get_authorizer,
3160 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3161 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 3162 .peer_reset = peer_reset,
2f2dc053
SW		 3163};
3164
3165
3166
3167
3168/* eof */
ubuntu-bionic-kernel mirror