projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'drm-next-4.10' of git://people.freedesktop.org/~agd5f/linux into drm...
[linux-2.6-block.git] / fs / ceph / mds_client.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
2f2dc053 2
496e5955 3#include <linux/fs.h>
2f2dc053 4#include <linux/wait.h>
5a0e3ad6 5#include <linux/slab.h>
54008399 6#include <linux/gfp.h>
2f2dc053 7#include <linux/sched.h>
3d14c5d2
YS		 8#include <linux/debugfs.h>
9#include <linux/seq_file.h>
dbd0c8bf 10#include <linux/utsname.h>
3e0708b9 11#include <linux/ratelimit.h>
2f2dc053 12
2f2dc053 13#include "super.h"
3d14c5d2
YS		 14#include "mds_client.h"
15
1fe60e51 16#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/pagelist.h>
20#include <linux/ceph/auth.h>
21#include <linux/ceph/debugfs.h>
2f2dc053
SW		 22
23/*
24 * A cluster of MDS (metadata server) daemons is responsible for
25 * managing the file system namespace (the directory hierarchy and
26 * inodes) and for coordinating shared access to storage. Metadata is
27 * partitioning hierarchically across a number of servers, and that
28 * partition varies over time as the cluster adjusts the distribution
29 * in order to balance load.
30 *
31 * The MDS client is primarily responsible to managing synchronous
32 * metadata requests for operations like open, unlink, and so forth.
33 * If there is a MDS failure, we find out about it when we (possibly
34 * request and) receive a new MDS map, and can resubmit affected
35 * requests.
36 *
37 * For the most part, though, we take advantage of a lossless
38 * communications channel to the MDS, and do not need to worry about
39 * timing out or resubmitting requests.
40 *
41 * We maintain a stateful "session" with each MDS we interact with.
42 * Within each session, we sent periodic heartbeat messages to ensure
43 * any capabilities or leases we have been issues remain valid. If
44 * the session times out and goes stale, our leases and capabilities
45 * are no longer valid.
46 */
47
20cb34ae 48struct ceph_reconnect_state {
44c99757 49 int nr_caps;
20cb34ae 50 struct ceph_pagelist *pagelist;
121f22a1 51 unsigned msg_version;
20cb34ae
SW		 52};
53
2f2dc053
SW		 54static void __wake_requests(struct ceph_mds_client *mdsc,
55 struct list_head *head);
56
9e32789f 57static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 58
59
60/*
61 * mds reply parsing
62 */
63
64/*
65 * parse individual inode info
66 */
67static int parse_reply_info_in(void **p, void *end,
14303d20 68 struct ceph_mds_reply_info_in *info,
12b4629a 69 u64 features)
2f2dc053
SW		 70{
71 int err = -EIO;
72
73 info->in = *p;
74 *p += sizeof(struct ceph_mds_reply_inode) +
75 sizeof(*info->in->fragtree.splits) *
76 le32_to_cpu(info->in->fragtree.nsplits);
77
78 ceph_decode_32_safe(p, end, info->symlink_len, bad);
79 ceph_decode_need(p, end, info->symlink_len, bad);
80 info->symlink = *p;
81 *p += info->symlink_len;
82
14303d20
SW		 83 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
84 ceph_decode_copy_safe(p, end, &info->dir_layout,
85 sizeof(info->dir_layout), bad);
86 else
87 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
88
2f2dc053
SW		 89 ceph_decode_32_safe(p, end, info->xattr_len, bad);
90 ceph_decode_need(p, end, info->xattr_len, bad);
91 info->xattr_data = *p;
92 *p += info->xattr_len;
fb01d1f8
YZ		 93
94 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
95 ceph_decode_64_safe(p, end, info->inline_version, bad);
96 ceph_decode_32_safe(p, end, info->inline_len, bad);
97 ceph_decode_need(p, end, info->inline_len, bad);
98 info->inline_data = *p;
99 *p += info->inline_len;
100 } else
101 info->inline_version = CEPH_INLINE_NONE;
102
779fe0fb
YZ		 103 info->pool_ns_len = 0;
104 info->pool_ns_data = NULL;
5ea5c5e0
YZ		 105 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
106 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
779fe0fb
YZ		 107 if (info->pool_ns_len > 0) {
108 ceph_decode_need(p, end, info->pool_ns_len, bad);
109 info->pool_ns_data = *p;
110 *p += info->pool_ns_len;
111 }
5ea5c5e0
YZ		 112 }
113
2f2dc053
SW		 114 return 0;
115bad:
116 return err;
117}
118
119/*
120 * parse a normal reply, which may contain a (dir+)dentry and/or a
121 * target inode.
122 */
123static int parse_reply_info_trace(void **p, void *end,
14303d20 124 struct ceph_mds_reply_info_parsed *info,
12b4629a 125 u64 features)
2f2dc053
SW		 126{
127 int err;
128
129 if (info->head->is_dentry) {
14303d20 130 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 131 if (err < 0)
132 goto out_bad;
133
134 if (unlikely(*p + sizeof(*info->dirfrag) > end))
135 goto bad;
136 info->dirfrag = *p;
137 *p += sizeof(*info->dirfrag) +
138 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
139 if (unlikely(*p > end))
140 goto bad;
141
142 ceph_decode_32_safe(p, end, info->dname_len, bad);
143 ceph_decode_need(p, end, info->dname_len, bad);
144 info->dname = *p;
145 *p += info->dname_len;
146 info->dlease = *p;
147 *p += sizeof(*info->dlease);
148 }
149
150 if (info->head->is_target) {
14303d20 151 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 152 if (err < 0)
153 goto out_bad;
154 }
155
156 if (unlikely(*p != end))
157 goto bad;
158 return 0;
159
160bad:
161 err = -EIO;
162out_bad:
163 pr_err("problem parsing mds trace %d\n", err);
164 return err;
165}
166
167/*
168 * parse readdir results
169 */
170static int parse_reply_info_dir(void **p, void *end,
14303d20 171 struct ceph_mds_reply_info_parsed *info,
12b4629a 172 u64 features)
2f2dc053
SW		 173{
174 u32 num, i = 0;
175 int err;
176
177 info->dir_dir = *p;
178 if (*p + sizeof(*info->dir_dir) > end)
179 goto bad;
180 *p += sizeof(*info->dir_dir) +
181 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
182 if (*p > end)
183 goto bad;
184
185 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea 186 num = ceph_decode_32(p);
956d39d6
YZ		 187 {
188 u16 flags = ceph_decode_16(p);
189 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
190 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
f3c4ebe6 191 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
956d39d6 192 }
2f2dc053
SW		 193 if (num == 0)
194 goto done;
195
2a5beea3
YZ		 196 BUG_ON(!info->dir_entries);
197 if ((unsigned long)(info->dir_entries + num) >
198 (unsigned long)info->dir_entries + info->dir_buf_size) {
54008399
YZ		 199 pr_err("dir contents are larger than expected\n");
200 WARN_ON(1);
201 goto bad;
202 }
2f2dc053 203
54008399 204 info->dir_nr = num;
2f2dc053 205 while (num) {
2a5beea3 206 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
2f2dc053
SW		 207 /* dentry */
208 ceph_decode_need(p, end, sizeof(u32)*2, bad);
2a5beea3
YZ		 209 rde->name_len = ceph_decode_32(p);
210 ceph_decode_need(p, end, rde->name_len, bad);
211 rde->name = *p;
212 *p += rde->name_len;
213 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
214 rde->lease = *p;
2f2dc053
SW		 215 *p += sizeof(struct ceph_mds_reply_lease);
216
217 /* inode */
2a5beea3 218 err = parse_reply_info_in(p, end, &rde->inode, features);
2f2dc053
SW		 219 if (err < 0)
220 goto out_bad;
8974eebd
YZ		 221 /* ceph_readdir_prepopulate() will update it */
222 rde->offset = 0;
2f2dc053
SW		 223 i++;
224 num--;
225 }
226
227done:
228 if (*p != end)
229 goto bad;
230 return 0;
231
232bad:
233 err = -EIO;
234out_bad:
235 pr_err("problem parsing dir contents %d\n", err);
236 return err;
237}
238
25933abd
HS		 239/*
240 * parse fcntl F_GETLK results
241 */
242static int parse_reply_info_filelock(void **p, void *end,
14303d20 243 struct ceph_mds_reply_info_parsed *info,
12b4629a 244 u64 features)
25933abd
HS		 245{
246 if (*p + sizeof(*info->filelock_reply) > end)
247 goto bad;
248
249 info->filelock_reply = *p;
250 *p += sizeof(*info->filelock_reply);
251
252 if (unlikely(*p != end))
253 goto bad;
254 return 0;
255
256bad:
257 return -EIO;
258}
259
6e8575fa
SL		 260/*
261 * parse create results
262 */
263static int parse_reply_info_create(void **p, void *end,
264 struct ceph_mds_reply_info_parsed *info,
12b4629a 265 u64 features)
6e8575fa
SL		 266{
267 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
268 if (*p == end) {
269 info->has_create_ino = false;
270 } else {
271 info->has_create_ino = true;
272 info->ino = ceph_decode_64(p);
273 }
274 }
275
276 if (unlikely(*p != end))
277 goto bad;
278 return 0;
279
280bad:
281 return -EIO;
282}
283
25933abd
HS		 284/*
285 * parse extra results
286 */
287static int parse_reply_info_extra(void **p, void *end,
14303d20 288 struct ceph_mds_reply_info_parsed *info,
12b4629a 289 u64 features)
25933abd
HS		 290{
291 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
14303d20 292 return parse_reply_info_filelock(p, end, info, features);
8a034497
YZ		 293 else if (info->head->op == CEPH_MDS_OP_READDIR ||
294 info->head->op == CEPH_MDS_OP_LSSNAP)
14303d20 295 return parse_reply_info_dir(p, end, info, features);
6e8575fa
SL		 296 else if (info->head->op == CEPH_MDS_OP_CREATE)
297 return parse_reply_info_create(p, end, info, features);
298 else
299 return -EIO;
25933abd
HS		 300}
301
2f2dc053
SW		 302/*
303 * parse entire mds reply
304 */
305static int parse_reply_info(struct ceph_msg *msg,
14303d20 306 struct ceph_mds_reply_info_parsed *info,
12b4629a 307 u64 features)
2f2dc053
SW		 308{
309 void *p, *end;
310 u32 len;
311 int err;
312
313 info->head = msg->front.iov_base;
314 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
315 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
316
317 /* trace */
318 ceph_decode_32_safe(&p, end, len, bad);
319 if (len > 0) {
32852a81 320 ceph_decode_need(&p, end, len, bad);
14303d20 321 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 322 if (err < 0)
323 goto out_bad;
324 }
325
25933abd 326 /* extra */
2f2dc053
SW		 327 ceph_decode_32_safe(&p, end, len, bad);
328 if (len > 0) {
32852a81 329 ceph_decode_need(&p, end, len, bad);
14303d20 330 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 331 if (err < 0)
332 goto out_bad;
333 }
334
335 /* snap blob */
336 ceph_decode_32_safe(&p, end, len, bad);
337 info->snapblob_len = len;
338 info->snapblob = p;
339 p += len;
340
341 if (p != end)
342 goto bad;
343 return 0;
344
345bad:
346 err = -EIO;
347out_bad:
348 pr_err("mds parse_reply err %d\n", err);
349 return err;
350}
351
352static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
353{
2a5beea3 354 if (!info->dir_entries)
54008399 355 return;
2a5beea3 356 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
2f2dc053
SW		 357}
358
359
360/*
361 * sessions
362 */
a687ecaf 363const char *ceph_session_state_name(int s)
2f2dc053
SW		 364{
365 switch (s) {
366 case CEPH_MDS_SESSION_NEW: return "new";
367 case CEPH_MDS_SESSION_OPENING: return "opening";
368 case CEPH_MDS_SESSION_OPEN: return "open";
369 case CEPH_MDS_SESSION_HUNG: return "hung";
370 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 371 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053 372 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
fcff415c 373 case CEPH_MDS_SESSION_REJECTED: return "rejected";
2f2dc053
SW		 374 default: return "???";
375 }
376}
377
378static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
379{
380 if (atomic_inc_not_zero(&s->s_ref)) {
381 dout("mdsc get_session %p %d -> %d\n", s,
382 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
383 return s;
384 } else {
385 dout("mdsc get_session %p 0 -- FAIL", s);
386 return NULL;
387 }
388}
389
390void ceph_put_mds_session(struct ceph_mds_session *s)
391{
392 dout("mdsc put_session %p %d -> %d\n", s,
393 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
4e7a5dcd 394 if (atomic_dec_and_test(&s->s_ref)) {
6c4a1915 395 if (s->s_auth.authorizer)
6c1ea260 396 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
2f2dc053 397 kfree(s);
4e7a5dcd 398 }
2f2dc053
SW		 399}
400
401/*
402 * called under mdsc->mutex
403 */
404struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
405 int mds)
406{
407 struct ceph_mds_session *session;
408
409 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
410 return NULL;
411 session = mdsc->sessions[mds];
412 dout("lookup_mds_session %p %d\n", session,
413 atomic_read(&session->s_ref));
414 get_session(session);
415 return session;
416}
417
418static bool __have_session(struct ceph_mds_client *mdsc, int mds)
419{
420 if (mds >= mdsc->max_sessions)
421 return false;
422 return mdsc->sessions[mds];
423}
424
2600d2dd
SW		 425static int __verify_registered_session(struct ceph_mds_client *mdsc,
426 struct ceph_mds_session *s)
427{
428 if (s->s_mds >= mdsc->max_sessions ||
429 mdsc->sessions[s->s_mds] != s)
430 return -ENOENT;
431 return 0;
432}
433
2f2dc053
SW		 434/*
435 * create+register a new session for given mds.
436 * called under mdsc->mutex.
437 */
438static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
439 int mds)
440{
441 struct ceph_mds_session *s;
442
c338c07c
NY		 443 if (mds >= mdsc->mdsmap->m_max_mds)
444 return ERR_PTR(-EINVAL);
445
2f2dc053 446 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 447 if (!s)
448 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 449 s->s_mdsc = mdsc;
450 s->s_mds = mds;
451 s->s_state = CEPH_MDS_SESSION_NEW;
452 s->s_ttl = 0;
453 s->s_seq = 0;
454 mutex_init(&s->s_mutex);
455
b7a9e5dd 456 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 457
d8fb02ab 458 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 459 s->s_cap_gen = 0;
1ce208a6 460 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 461
462 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 463 s->s_renew_requested = 0;
464 s->s_renew_seq = 0;
465 INIT_LIST_HEAD(&s->s_caps);
466 s->s_nr_caps = 0;
5dacf091 467 s->s_trim_caps = 0;
2f2dc053
SW		 468 atomic_set(&s->s_ref, 1);
469 INIT_LIST_HEAD(&s->s_waiting);
470 INIT_LIST_HEAD(&s->s_unsafe);
471 s->s_num_cap_releases = 0;
99a9c273 472 s->s_cap_reconnect = 0;
7c1332b8 473 s->s_cap_iterator = NULL;
2f2dc053 474 INIT_LIST_HEAD(&s->s_cap_releases);
2f2dc053 475 INIT_LIST_HEAD(&s->s_cap_flushing);
2f2dc053
SW		 476
477 dout("register_session mds%d\n", mds);
478 if (mds >= mdsc->max_sessions) {
479 int newmax = 1 << get_count_order(mds+1);
480 struct ceph_mds_session **sa;
481
482 dout("register_session realloc to %d\n", newmax);
483 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
484 if (sa == NULL)
42ce56e5 485 goto fail_realloc;
2f2dc053
SW		 486 if (mdsc->sessions) {
487 memcpy(sa, mdsc->sessions,
488 mdsc->max_sessions * sizeof(void *));
489 kfree(mdsc->sessions);
490 }
491 mdsc->sessions = sa;
492 mdsc->max_sessions = newmax;
493 }
494 mdsc->sessions[mds] = s;
86d8f67b 495 atomic_inc(&mdsc->num_sessions);
2f2dc053 496 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 497
b7a9e5dd
SW		 498 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
499 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 500
2f2dc053 501 return s;
42ce56e5
SW		 502
503fail_realloc:
504 kfree(s);
505 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 506}
507
508/*
509 * called under mdsc->mutex
510 */
2600d2dd 511static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 512 struct ceph_mds_session *s)
2f2dc053 513{
2600d2dd
SW		 514 dout("__unregister_session mds%d %p\n", s->s_mds, s);
515 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 516 mdsc->sessions[s->s_mds] = NULL;
517 ceph_con_close(&s->s_con);
518 ceph_put_mds_session(s);
86d8f67b 519 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 520}
521
522/*
523 * drop session refs in request.
524 *
525 * should be last request ref, or hold mdsc->mutex
526 */
527static void put_request_session(struct ceph_mds_request *req)
528{
529 if (req->r_session) {
530 ceph_put_mds_session(req->r_session);
531 req->r_session = NULL;
532 }
533}
534
153c8e6b 535void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 536{
153c8e6b
SW		 537 struct ceph_mds_request *req = container_of(kref,
538 struct ceph_mds_request,
539 r_kref);
54008399 540 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 541 if (req->r_request)
542 ceph_msg_put(req->r_request);
54008399 543 if (req->r_reply)
153c8e6b 544 ceph_msg_put(req->r_reply);
153c8e6b 545 if (req->r_inode) {
41b02e1f 546 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 547 iput(req->r_inode);
548 }
549 if (req->r_locked_dir)
41b02e1f 550 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
e96a650a 551 iput(req->r_target_inode);
153c8e6b
SW		 552 if (req->r_dentry)
553 dput(req->r_dentry);
844d87c3
SW		 554 if (req->r_old_dentry)
555 dput(req->r_old_dentry);
556 if (req->r_old_dentry_dir) {
41b02e1f
SW		 557 /*
558 * track (and drop pins for) r_old_dentry_dir
559 * separately, since r_old_dentry's d_parent may have
560 * changed between the dir mutex being dropped and
561 * this request being freed.
562 */
563 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
564 CEPH_CAP_PIN);
41b02e1f 565 iput(req->r_old_dentry_dir);
2f2dc053 566 }
153c8e6b
SW		 567 kfree(req->r_path1);
568 kfree(req->r_path2);
25e6bae3
YZ		 569 if (req->r_pagelist)
570 ceph_pagelist_release(req->r_pagelist);
153c8e6b 571 put_request_session(req);
37151668 572 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 573 kfree(req);
2f2dc053
SW		 574}
575
fcd00b68
ID		 576DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
577
2f2dc053
SW		 578/*
579 * lookup session, bump ref if found.
580 *
581 * called under mdsc->mutex.
582 */
fcd00b68
ID		 583static struct ceph_mds_request *
584lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
2f2dc053
SW		 585{
586 struct ceph_mds_request *req;
44ca18f2 587
fcd00b68
ID		 588 req = lookup_request(&mdsc->request_tree, tid);
589 if (req)
590 ceph_mdsc_get_request(req);
44ca18f2 591
fcd00b68 592 return req;
2f2dc053
SW		 593}
594
595/*
596 * Register an in-flight request, and assign a tid. Link to directory
597 * are modifying (if any).
598 *
599 * Called under mdsc->mutex.
600 */
601static void __register_request(struct ceph_mds_client *mdsc,
602 struct ceph_mds_request *req,
603 struct inode *dir)
604{
605 req->r_tid = ++mdsc->last_tid;
606 if (req->r_num_caps)
37151668
YS		 607 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
608 req->r_num_caps);
2f2dc053
SW		 609 dout("__register_request %p tid %lld\n", req, req->r_tid);
610 ceph_mdsc_get_request(req);
fcd00b68 611 insert_request(&mdsc->request_tree, req);
2f2dc053 612
cb4276cc
SW		 613 req->r_uid = current_fsuid();
614 req->r_gid = current_fsgid();
615
e8a7b8b1
YZ		 616 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
617 mdsc->oldest_tid = req->r_tid;
618
2f2dc053 619 if (dir) {
3b663780 620 ihold(dir);
2f2dc053 621 req->r_unsafe_dir = dir;
2f2dc053
SW		 622 }
623}
624
625static void __unregister_request(struct ceph_mds_client *mdsc,
626 struct ceph_mds_request *req)
627{
628 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1
YZ		 629
630 if (req->r_tid == mdsc->oldest_tid) {
631 struct rb_node *p = rb_next(&req->r_node);
632 mdsc->oldest_tid = 0;
633 while (p) {
634 struct ceph_mds_request *next_req =
635 rb_entry(p, struct ceph_mds_request, r_node);
636 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
637 mdsc->oldest_tid = next_req->r_tid;
638 break;
639 }
640 p = rb_next(p);
641 }
642 }
643
fcd00b68 644 erase_request(&mdsc->request_tree, req);
2f2dc053 645
4c06ace8 646 if (req->r_unsafe_dir && req->r_got_unsafe) {
2f2dc053 647 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 648 spin_lock(&ci->i_unsafe_lock);
649 list_del_init(&req->r_unsafe_dir_item);
650 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 651 }
68cd5b4b
YZ		 652 if (req->r_target_inode && req->r_got_unsafe) {
653 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
654 spin_lock(&ci->i_unsafe_lock);
655 list_del_init(&req->r_unsafe_target_item);
656 spin_unlock(&ci->i_unsafe_lock);
657 }
3b663780 658
4c06ace8 659 if (req->r_unsafe_dir) {
3b663780
SW		 660 iput(req->r_unsafe_dir);
661 req->r_unsafe_dir = NULL;
2f2dc053 662 }
94aa8ae1 663
fc55d2c9
YZ		 664 complete_all(&req->r_safe_completion);
665
94aa8ae1 666 ceph_mdsc_put_request(req);
2f2dc053
SW		 667}
668
669/*
670 * Choose mds to send request to next. If there is a hint set in the
671 * request (e.g., due to a prior forward hint from the mds), use that.
672 * Otherwise, consult frag tree and/or caps to identify the
673 * appropriate mds. If all else fails, choose randomly.
674 *
675 * Called under mdsc->mutex.
676 */
7fd7d101 677static struct dentry *get_nonsnap_parent(struct dentry *dentry)
eb6bb1c5 678{
d79698da
SW		 679 /*
680 * we don't need to worry about protecting the d_parent access
681 * here because we never renaming inside the snapped namespace
682 * except to resplice to another snapdir, and either the old or new
683 * result is a valid result.
684 */
2b0143b5 685 while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
eb6bb1c5
SW		 686 dentry = dentry->d_parent;
687 return dentry;
688}
689
2f2dc053
SW		 690static int __choose_mds(struct ceph_mds_client *mdsc,
691 struct ceph_mds_request *req)
692{
693 struct inode *inode;
694 struct ceph_inode_info *ci;
695 struct ceph_cap *cap;
696 int mode = req->r_direct_mode;
697 int mds = -1;
698 u32 hash = req->r_direct_hash;
699 bool is_hash = req->r_direct_is_hash;
700
701 /*
702 * is there a specific mds we should try? ignore hint if we have
703 * no session and the mds is not up (active or recovering).
704 */
705 if (req->r_resend_mds >= 0 &&
706 (__have_session(mdsc, req->r_resend_mds) ||
707 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
708 dout("choose_mds using resend_mds mds%d\n",
709 req->r_resend_mds);
710 return req->r_resend_mds;
711 }
712
713 if (mode == USE_RANDOM_MDS)
714 goto random;
715
716 inode = NULL;
717 if (req->r_inode) {
718 inode = req->r_inode;
719 } else if (req->r_dentry) {
d79698da
SW		 720 /* ignore race with rename; old or new d_parent is okay */
721 struct dentry *parent = req->r_dentry->d_parent;
2b0143b5 722 struct inode *dir = d_inode(parent);
eb6bb1c5 723
3d14c5d2 724 if (dir->i_sb != mdsc->fsc->sb) {
eb6bb1c5 725 /* not this fs! */
2b0143b5 726 inode = d_inode(req->r_dentry);
eb6bb1c5
SW		 727 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
728 /* direct snapped/virtual snapdir requests
729 * based on parent dir inode */
d79698da 730 struct dentry *dn = get_nonsnap_parent(parent);
2b0143b5 731 inode = d_inode(dn);
eb6bb1c5 732 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 733 } else {
eb6bb1c5 734 /* dentry target */
2b0143b5 735 inode = d_inode(req->r_dentry);
ca18bede
YZ		 736 if (!inode || mode == USE_AUTH_MDS) {
737 /* dir + name */
738 inode = dir;
739 hash = ceph_dentry_hash(dir, req->r_dentry);
740 is_hash = true;
741 }
2f2dc053
SW		 742 }
743 }
eb6bb1c5 744
2f2dc053
SW		 745 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
746 (int)hash, mode);
747 if (!inode)
748 goto random;
749 ci = ceph_inode(inode);
750
751 if (is_hash && S_ISDIR(inode->i_mode)) {
752 struct ceph_inode_frag frag;
753 int found;
754
755 ceph_choose_frag(ci, hash, &frag, &found);
756 if (found) {
757 if (mode == USE_ANY_MDS && frag.ndist > 0) {
758 u8 r;
759
760 /* choose a random replica */
761 get_random_bytes(&r, 1);
762 r %= frag.ndist;
763 mds = frag.dist[r];
764 dout("choose_mds %p %llx.%llx "
765 "frag %u mds%d (%d/%d)\n",
766 inode, ceph_vinop(inode),
d66bbd44 767 frag.frag, mds,
2f2dc053 768 (int)r, frag.ndist);
d66bbd44
SW		 769 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
770 CEPH_MDS_STATE_ACTIVE)
771 return mds;
2f2dc053
SW		 772 }
773
774 /* since this file/dir wasn't known to be
775 * replicated, then we want to look for the
776 * authoritative mds. */
777 mode = USE_AUTH_MDS;
778 if (frag.mds >= 0) {
779 /* choose auth mds */
780 mds = frag.mds;
781 dout("choose_mds %p %llx.%llx "
782 "frag %u mds%d (auth)\n",
783 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 784 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
785 CEPH_MDS_STATE_ACTIVE)
786 return mds;
2f2dc053
SW		 787 }
788 }
789 }
790
be655596 791 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 792 cap = NULL;
793 if (mode == USE_AUTH_MDS)
794 cap = ci->i_auth_cap;
795 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
796 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
797 if (!cap) {
be655596 798 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 799 goto random;
800 }
801 mds = cap->session->s_mds;
802 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
803 inode, ceph_vinop(inode), mds,
804 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 805 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 806 return mds;
807
808random:
809 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
810 dout("choose_mds chose random mds%d\n", mds);
811 return mds;
812}
813
814
815/*
816 * session messages
817 */
818static struct ceph_msg *create_session_msg(u32 op, u64 seq)
819{
820 struct ceph_msg *msg;
821 struct ceph_mds_session_head *h;
822
b61c2763
SW		 823 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
824 false);
a79832f2 825 if (!msg) {
2f2dc053 826 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 827 return NULL;
2f2dc053
SW		 828 }
829 h = msg->front.iov_base;
830 h->op = cpu_to_le32(op);
831 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 832
833 return msg;
834}
835
836/*
837 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
838 * to include additional client metadata fields.
839 */
840static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
841{
842 struct ceph_msg *msg;
843 struct ceph_mds_session_head *h;
844 int i = -1;
845 int metadata_bytes = 0;
846 int metadata_key_count = 0;
847 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 848 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
dbd0c8bf
JS		 849 void *p;
850
a6a5ce4f 851 const char* metadata[][2] = {
dbd0c8bf 852 {"hostname", utsname()->nodename},
a6a5ce4f 853 {"kernel_version", utsname()->release},
3f384954
YZ		 854 {"entity_id", opt->name ? : ""},
855 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 856 {NULL, NULL}
857 };
858
859 /* Calculate serialized length of metadata */
860 metadata_bytes = 4; /* map length */
861 for (i = 0; metadata[i][0] != NULL; ++i) {
862 metadata_bytes += 8 + strlen(metadata[i][0]) +
863 strlen(metadata[i][1]);
864 metadata_key_count++;
865 }
866
867 /* Allocate the message */
868 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
869 GFP_NOFS, false);
870 if (!msg) {
871 pr_err("create_session_msg ENOMEM creating msg\n");
872 return NULL;
873 }
874 h = msg->front.iov_base;
875 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
876 h->seq = cpu_to_le64(seq);
877
878 /*
879 * Serialize client metadata into waiting buffer space, using
880 * the format that userspace expects for map<string, string>
7cfa0313
JS		 881 *
882 * ClientSession messages with metadata are v2
dbd0c8bf 883 */
7cfa0313
JS		 884 msg->hdr.version = cpu_to_le16(2);
885 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 886
887 /* The write pointer, following the session_head structure */
888 p = msg->front.iov_base + sizeof(*h);
889
890 /* Number of entries in the map */
891 ceph_encode_32(&p, metadata_key_count);
892
893 /* Two length-prefixed strings for each entry in the map */
894 for (i = 0; metadata[i][0] != NULL; ++i) {
895 size_t const key_len = strlen(metadata[i][0]);
896 size_t const val_len = strlen(metadata[i][1]);
897
898 ceph_encode_32(&p, key_len);
899 memcpy(p, metadata[i][0], key_len);
900 p += key_len;
901 ceph_encode_32(&p, val_len);
902 memcpy(p, metadata[i][1], val_len);
903 p += val_len;
904 }
905
2f2dc053
SW		 906 return msg;
907}
908
909/*
910 * send session open request.
911 *
912 * called under mdsc->mutex
913 */
914static int __open_session(struct ceph_mds_client *mdsc,
915 struct ceph_mds_session *session)
916{
917 struct ceph_msg *msg;
918 int mstate;
919 int mds = session->s_mds;
2f2dc053
SW		 920
921 /* wait for mds to go active? */
922 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
923 dout("open_session to mds%d (%s)\n", mds,
924 ceph_mds_state_name(mstate));
925 session->s_state = CEPH_MDS_SESSION_OPENING;
926 session->s_renew_requested = jiffies;
927
928 /* send connect message */
dbd0c8bf 929 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 930 if (!msg)
931 return -ENOMEM;
2f2dc053 932 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 933 return 0;
934}
935
ed0552a1
SW		 936/*
937 * open sessions for any export targets for the given mds
938 *
939 * called under mdsc->mutex
940 */
5d72d13c
YZ		 941static struct ceph_mds_session *
942__open_export_target_session(struct ceph_mds_client *mdsc, int target)
943{
944 struct ceph_mds_session *session;
945
946 session = __ceph_lookup_mds_session(mdsc, target);
947 if (!session) {
948 session = register_session(mdsc, target);
949 if (IS_ERR(session))
950 return session;
951 }
952 if (session->s_state == CEPH_MDS_SESSION_NEW ||
953 session->s_state == CEPH_MDS_SESSION_CLOSING)
954 __open_session(mdsc, session);
955
956 return session;
957}
958
959struct ceph_mds_session *
960ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
961{
962 struct ceph_mds_session *session;
963
964 dout("open_export_target_session to mds%d\n", target);
965
966 mutex_lock(&mdsc->mutex);
967 session = __open_export_target_session(mdsc, target);
968 mutex_unlock(&mdsc->mutex);
969
970 return session;
971}
972
ed0552a1
SW		 973static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
974 struct ceph_mds_session *session)
975{
976 struct ceph_mds_info *mi;
977 struct ceph_mds_session *ts;
978 int i, mds = session->s_mds;
ed0552a1
SW		 979
980 if (mds >= mdsc->mdsmap->m_max_mds)
981 return;
5d72d13c 982
ed0552a1
SW		 983 mi = &mdsc->mdsmap->m_info[mds];
984 dout("open_export_target_sessions for mds%d (%d targets)\n",
985 session->s_mds, mi->num_export_targets);
986
987 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 988 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
989 if (!IS_ERR(ts))
990 ceph_put_mds_session(ts);
ed0552a1
SW		 991 }
992}
993
154f42c2
SW		 994void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
995 struct ceph_mds_session *session)
996{
997 mutex_lock(&mdsc->mutex);
998 __open_export_target_sessions(mdsc, session);
999 mutex_unlock(&mdsc->mutex);
1000}
1001
2f2dc053
SW		 1002/*
1003 * session caps
1004 */
1005
745a8e3b
YZ		 1006/* caller holds s_cap_lock, we drop it */
1007static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1008 struct ceph_mds_session *session)
1009 __releases(session->s_cap_lock)
2f2dc053 1010{
745a8e3b
YZ		 1011 LIST_HEAD(tmp_list);
1012 list_splice_init(&session->s_cap_releases, &tmp_list);
1013 session->s_num_cap_releases = 0;
1014 spin_unlock(&session->s_cap_lock);
2f2dc053 1015
745a8e3b
YZ		 1016 dout("cleanup_cap_releases mds%d\n", session->s_mds);
1017 while (!list_empty(&tmp_list)) {
1018 struct ceph_cap *cap;
1019 /* zero out the in-progress message */
1020 cap = list_first_entry(&tmp_list,
1021 struct ceph_cap, session_caps);
1022 list_del(&cap->session_caps);
1023 ceph_put_cap(mdsc, cap);
2f2dc053 1024 }
2f2dc053
SW		 1025}
1026
1c841a96
YZ		 1027static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1028 struct ceph_mds_session *session)
1029{
1030 struct ceph_mds_request *req;
1031 struct rb_node *p;
1032
1033 dout("cleanup_session_requests mds%d\n", session->s_mds);
1034 mutex_lock(&mdsc->mutex);
1035 while (!list_empty(&session->s_unsafe)) {
1036 req = list_first_entry(&session->s_unsafe,
1037 struct ceph_mds_request, r_unsafe_item);
1038 list_del_init(&req->r_unsafe_item);
3e0708b9
YZ		 1039 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1040 req->r_tid);
1c841a96
YZ		 1041 __unregister_request(mdsc, req);
1042 }
1043 /* zero r_attempts, so kick_requests() will re-send requests */
1044 p = rb_first(&mdsc->request_tree);
1045 while (p) {
1046 req = rb_entry(p, struct ceph_mds_request, r_node);
1047 p = rb_next(p);
1048 if (req->r_session &&
1049 req->r_session->s_mds == session->s_mds)
1050 req->r_attempts = 0;
1051 }
1052 mutex_unlock(&mdsc->mutex);
1053}
1054
2f2dc053 1055/*
f818a736
SW		 1056 * Helper to safely iterate over all caps associated with a session, with
1057 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1058 *
f818a736 1059 * Caller must hold session s_mutex.
2f2dc053
SW		 1060 */
1061static int iterate_session_caps(struct ceph_mds_session *session,
1062 int (*cb)(struct inode *, struct ceph_cap *,
1063 void *), void *arg)
1064{
7c1332b8
SW		 1065 struct list_head *p;
1066 struct ceph_cap *cap;
1067 struct inode *inode, *last_inode = NULL;
1068 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1069 int ret;
1070
1071 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1072 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1073 p = session->s_caps.next;
1074 while (p != &session->s_caps) {
1075 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1076 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1077 if (!inode) {
1078 p = p->next;
2f2dc053 1079 continue;
7c1332b8
SW		 1080 }
1081 session->s_cap_iterator = cap;
2f2dc053 1082 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1083
1084 if (last_inode) {
1085 iput(last_inode);
1086 last_inode = NULL;
1087 }
1088 if (old_cap) {
37151668 1089 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1090 old_cap = NULL;
1091 }
1092
2f2dc053 1093 ret = cb(inode, cap, arg);
7c1332b8
SW		 1094 last_inode = inode;
1095
2f2dc053 1096 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1097 p = p->next;
1098 if (cap->ci == NULL) {
1099 dout("iterate_session_caps finishing cap %p removal\n",
1100 cap);
1101 BUG_ON(cap->session != session);
745a8e3b 1102 cap->session = NULL;
7c1332b8
SW		 1103 list_del_init(&cap->session_caps);
1104 session->s_nr_caps--;
745a8e3b
YZ		 1105 if (cap->queue_release) {
1106 list_add_tail(&cap->session_caps,
1107 &session->s_cap_releases);
1108 session->s_num_cap_releases++;
1109 } else {
1110 old_cap = cap; /* put_cap it w/o locks held */
1111 }
7c1332b8 1112 }
5dacf091
SW		 1113 if (ret < 0)
1114 goto out;
2f2dc053 1115 }
5dacf091
SW		 1116 ret = 0;
1117out:
7c1332b8 1118 session->s_cap_iterator = NULL;
2f2dc053 1119 spin_unlock(&session->s_cap_lock);
7c1332b8 1120
e96a650a 1121 iput(last_inode);
7c1332b8 1122 if (old_cap)
37151668 1123 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1124
5dacf091 1125 return ret;
2f2dc053
SW		 1126}
1127
1128static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1129 void *arg)
2f2dc053 1130{
6c93df5d 1131 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
2f2dc053 1132 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1133 LIST_HEAD(to_remove);
6c93df5d
YZ		 1134 bool drop = false;
1135 bool invalidate = false;
6c99f254 1136
2f2dc053
SW		 1137 dout("removing cap %p, ci is %p, inode is %p\n",
1138 cap, ci, &ci->vfs_inode);
be655596 1139 spin_lock(&ci->i_ceph_lock);
a096b09a 1140 __ceph_remove_cap(cap, false);
571ade33 1141 if (!ci->i_auth_cap) {
553adfd9 1142 struct ceph_cap_flush *cf;
6c93df5d 1143 struct ceph_mds_client *mdsc = fsc->mdsc;
6c99f254 1144
77310320
YZ		 1145 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1146
6c93df5d
YZ		 1147 if (ci->i_wrbuffer_ref > 0 &&
1148 ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
1149 invalidate = true;
1150
e4500b5e
YZ		 1151 while (!list_empty(&ci->i_cap_flush_list)) {
1152 cf = list_first_entry(&ci->i_cap_flush_list,
1153 struct ceph_cap_flush, i_list);
8cdcc07d 1154 list_move(&cf->i_list, &to_remove);
553adfd9
YZ		 1155 }
1156
6c99f254 1157 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1158
e4500b5e
YZ		 1159 list_for_each_entry(cf, &to_remove, i_list)
1160 list_del(&cf->g_list);
8310b089 1161
6c99f254 1162 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1163 pr_warn_ratelimited(
1164 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1165 ceph_cap_string(ci->i_dirty_caps),
1166 inode, ceph_ino(inode));
1167 ci->i_dirty_caps = 0;
1168 list_del_init(&ci->i_dirty_item);
6c93df5d 1169 drop = true;
6c99f254
SW		 1170 }
1171 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1172 pr_warn_ratelimited(
1173 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1174 ceph_cap_string(ci->i_flushing_caps),
1175 inode, ceph_ino(inode));
1176 ci->i_flushing_caps = 0;
1177 list_del_init(&ci->i_flushing_item);
1178 mdsc->num_cap_flushing--;
6c93df5d 1179 drop = true;
6c99f254 1180 }
6c99f254 1181 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1182
f66fd9f0 1183 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1184 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1185 ci->i_prealloc_cap_flush = NULL;
1186 }
6c99f254 1187 }
be655596 1188 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1189 while (!list_empty(&to_remove)) {
1190 struct ceph_cap_flush *cf;
1191 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1192 struct ceph_cap_flush, i_list);
1193 list_del(&cf->i_list);
f66fd9f0 1194 ceph_free_cap_flush(cf);
553adfd9 1195 }
77310320
YZ		 1196
1197 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1198 if (invalidate)
1199 ceph_queue_invalidate(inode);
77310320 1200 if (drop)
6c99f254 1201 iput(inode);
2f2dc053
SW		 1202 return 0;
1203}
1204
1205/*
1206 * caller must hold session s_mutex
1207 */
1208static void remove_session_caps(struct ceph_mds_session *session)
1209{
6c93df5d
YZ		 1210 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1211 struct super_block *sb = fsc->sb;
2f2dc053 1212 dout("remove_session_caps on %p\n", session);
6c93df5d 1213 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1214
c8799fc4
YZ		 1215 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1216
6f60f889
YZ		 1217 spin_lock(&session->s_cap_lock);
1218 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1219 struct inode *inode;
1220 struct ceph_cap *cap, *prev = NULL;
1221 struct ceph_vino vino;
1222 /*
1223 * iterate_session_caps() skips inodes that are being
1224 * deleted, we need to wait until deletions are complete.
1225 * __wait_on_freeing_inode() is designed for the job,
1226 * but it is not exported, so use lookup inode function
1227 * to access it.
1228 */
1229 while (!list_empty(&session->s_caps)) {
1230 cap = list_entry(session->s_caps.next,
1231 struct ceph_cap, session_caps);
1232 if (cap == prev)
1233 break;
1234 prev = cap;
1235 vino = cap->ci->i_vino;
1236 spin_unlock(&session->s_cap_lock);
1237
ed284c49 1238 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1239 iput(inode);
1240
1241 spin_lock(&session->s_cap_lock);
1242 }
1243 }
745a8e3b
YZ		 1244
1245 // drop cap expires and unlock s_cap_lock
1246 cleanup_cap_releases(session->s_mdsc, session);
6f60f889 1247
2f2dc053 1248 BUG_ON(session->s_nr_caps > 0);
6c99f254 1249 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1250}
1251
1252/*
1253 * wake up any threads waiting on this session's caps. if the cap is
1254 * old (didn't get renewed on the client reconnect), remove it now.
1255 *
1256 * caller must hold s_mutex.
1257 */
1258static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1259 void *arg)
1260{
0dc2570f
SW		 1261 struct ceph_inode_info *ci = ceph_inode(inode);
1262
0dc2570f 1263 if (arg) {
be655596 1264 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1265 ci->i_wanted_max_size = 0;
1266 ci->i_requested_max_size = 0;
be655596 1267 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1268 }
e5360309 1269 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1270 return 0;
1271}
1272
0dc2570f
SW		 1273static void wake_up_session_caps(struct ceph_mds_session *session,
1274 int reconnect)
2f2dc053
SW		 1275{
1276 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1277 iterate_session_caps(session, wake_up_session_cb,
1278 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1279}
1280
1281/*
1282 * Send periodic message to MDS renewing all currently held caps. The
1283 * ack will reset the expiration for all caps from this session.
1284 *
1285 * caller holds s_mutex
1286 */
1287static int send_renew_caps(struct ceph_mds_client *mdsc,
1288 struct ceph_mds_session *session)
1289{
1290 struct ceph_msg *msg;
1291 int state;
1292
1293 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1294 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1295 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1296 session->s_renew_requested = jiffies;
2f2dc053
SW		 1297
1298 /* do not try to renew caps until a recovering mds has reconnected
1299 * with its clients. */
1300 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1301 if (state < CEPH_MDS_STATE_RECONNECT) {
1302 dout("send_renew_caps ignoring mds%d (%s)\n",
1303 session->s_mds, ceph_mds_state_name(state));
1304 return 0;
1305 }
1306
1307 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1308 ceph_mds_state_name(state));
2f2dc053
SW		 1309 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1310 ++session->s_renew_seq);
a79832f2
SW		 1311 if (!msg)
1312 return -ENOMEM;
2f2dc053
SW		 1313 ceph_con_send(&session->s_con, msg);
1314 return 0;
1315}
1316
186e4f7a
YZ		 1317static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1318 struct ceph_mds_session *session, u64 seq)
1319{
1320 struct ceph_msg *msg;
1321
1322 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1323 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1324 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1325 if (!msg)
1326 return -ENOMEM;
1327 ceph_con_send(&session->s_con, msg);
1328 return 0;
1329}
1330
1331
2f2dc053
SW		 1332/*
1333 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1334 *
1335 * Called under session->s_mutex
2f2dc053
SW		 1336 */
1337static void renewed_caps(struct ceph_mds_client *mdsc,
1338 struct ceph_mds_session *session, int is_renew)
1339{
1340 int was_stale;
1341 int wake = 0;
1342
1343 spin_lock(&session->s_cap_lock);
1ce208a6 1344 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1345
1346 session->s_cap_ttl = session->s_renew_requested +
1347 mdsc->mdsmap->m_session_timeout*HZ;
1348
1349 if (was_stale) {
1350 if (time_before(jiffies, session->s_cap_ttl)) {
1351 pr_info("mds%d caps renewed\n", session->s_mds);
1352 wake = 1;
1353 } else {
1354 pr_info("mds%d caps still stale\n", session->s_mds);
1355 }
1356 }
1357 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1358 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1359 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1360 spin_unlock(&session->s_cap_lock);
1361
1362 if (wake)
0dc2570f 1363 wake_up_session_caps(session, 0);
2f2dc053
SW		 1364}
1365
1366/*
1367 * send a session close request
1368 */
1369static int request_close_session(struct ceph_mds_client *mdsc,
1370 struct ceph_mds_session *session)
1371{
1372 struct ceph_msg *msg;
2f2dc053
SW		 1373
1374 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1375 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1376 session->s_seq);
1377 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1378 if (!msg)
1379 return -ENOMEM;
1380 ceph_con_send(&session->s_con, msg);
fcff415c 1381 return 1;
2f2dc053
SW		 1382}
1383
1384/*
1385 * Called with s_mutex held.
1386 */
1387static int __close_session(struct ceph_mds_client *mdsc,
1388 struct ceph_mds_session *session)
1389{
1390 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1391 return 0;
1392 session->s_state = CEPH_MDS_SESSION_CLOSING;
1393 return request_close_session(mdsc, session);
1394}
1395
1396/*
1397 * Trim old(er) caps.
1398 *
1399 * Because we can't cache an inode without one or more caps, we do
1400 * this indirectly: if a cap is unused, we prune its aliases, at which
1401 * point the inode will hopefully get dropped to.
1402 *
1403 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1404 * memory pressure from the MDS, though, so it needn't be perfect.
1405 */
1406static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1407{
1408 struct ceph_mds_session *session = arg;
1409 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1410 int used, wanted, oissued, mine;
2f2dc053
SW		 1411
1412 if (session->s_trim_caps <= 0)
1413 return -1;
1414
be655596 1415 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1416 mine = cap->issued | cap->implemented;
1417 used = __ceph_caps_used(ci);
979abfdd 1418 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1419 oissued = __ceph_caps_issued_other(ci, cap);
1420
979abfdd 1421 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1422 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1423 ceph_cap_string(used), ceph_cap_string(wanted));
1424 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1425 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1426 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1427 goto out;
1428 if ((used | wanted) & CEPH_CAP_ANY_WR)
1429 goto out;
1430 }
5e804ac4
YZ		 1431 /* The inode has cached pages, but it's no longer used.
1432 * we can safely drop it */
1433 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1434 !(oissued & CEPH_CAP_FILE_CACHE)) {
1435 used = 0;
1436 oissued = 0;
1437 }
979abfdd 1438 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1439 goto out; /* we need these caps */
1440
1441 session->s_trim_caps--;
1442 if (oissued) {
1443 /* we aren't the only cap.. just remove us */
a096b09a 1444 __ceph_remove_cap(cap, true);
2f2dc053 1445 } else {
5e804ac4 1446 /* try dropping referring dentries */
be655596 1447 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1448 d_prune_aliases(inode);
1449 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1450 inode, cap, atomic_read(&inode->i_count));
1451 return 0;
1452 }
1453
1454out:
be655596 1455 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1456 return 0;
1457}
1458
1459/*
1460 * Trim session cap count down to some max number.
1461 */
1462static int trim_caps(struct ceph_mds_client *mdsc,
1463 struct ceph_mds_session *session,
1464 int max_caps)
1465{
1466 int trim_caps = session->s_nr_caps - max_caps;
1467
1468 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1469 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1470 if (trim_caps > 0) {
1471 session->s_trim_caps = trim_caps;
1472 iterate_session_caps(session, trim_caps_cb, session);
1473 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1474 session->s_mds, session->s_nr_caps, max_caps,
1475 trim_caps - session->s_trim_caps);
5dacf091 1476 session->s_trim_caps = 0;
2f2dc053 1477 }
a56371d9 1478
a56371d9 1479 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1480 return 0;
1481}
1482
8310b089
YZ		 1483static int check_caps_flush(struct ceph_mds_client *mdsc,
1484 u64 want_flush_tid)
1485{
8310b089
YZ		 1486 int ret = 1;
1487
1488 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1489 if (!list_empty(&mdsc->cap_flush_list)) {
1490 struct ceph_cap_flush *cf =
1491 list_first_entry(&mdsc->cap_flush_list,
1492 struct ceph_cap_flush, g_list);
1493 if (cf->tid <= want_flush_tid) {
1494 dout("check_caps_flush still flushing tid "
1495 "%llu <= %llu\n", cf->tid, want_flush_tid);
1496 ret = 0;
1497 }
8310b089
YZ		 1498 }
1499 spin_unlock(&mdsc->cap_dirty_lock);
1500 return ret;
d3383a8e
YZ		 1501}
1502
2f2dc053
SW		 1503/*
1504 * flush all dirty inode data to disk.
1505 *
8310b089 1506 * returns true if we've flushed through want_flush_tid
2f2dc053 1507 */
affbc19a 1508static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1509 u64 want_flush_tid)
2f2dc053 1510{
0e294387 1511 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1512
1513 wait_event(mdsc->cap_flushing_wq,
1514 check_caps_flush(mdsc, want_flush_tid));
1515
1516 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1517}
1518
1519/*
1520 * called under s_mutex
1521 */
3d7ded4d
SW		 1522void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1523 struct ceph_mds_session *session)
2f2dc053 1524{
745a8e3b
YZ		 1525 struct ceph_msg *msg = NULL;
1526 struct ceph_mds_cap_release *head;
1527 struct ceph_mds_cap_item *item;
1528 struct ceph_cap *cap;
1529 LIST_HEAD(tmp_list);
1530 int num_cap_releases;
2f2dc053 1531
0f8605f2 1532 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1533again:
1534 list_splice_init(&session->s_cap_releases, &tmp_list);
1535 num_cap_releases = session->s_num_cap_releases;
1536 session->s_num_cap_releases = 0;
2f2dc053 1537 spin_unlock(&session->s_cap_lock);
e01a5946 1538
745a8e3b
YZ		 1539 while (!list_empty(&tmp_list)) {
1540 if (!msg) {
1541 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1542 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1543 if (!msg)
1544 goto out_err;
1545 head = msg->front.iov_base;
1546 head->num = cpu_to_le32(0);
1547 msg->front.iov_len = sizeof(*head);
1548 }
1549 cap = list_first_entry(&tmp_list, struct ceph_cap,
1550 session_caps);
1551 list_del(&cap->session_caps);
1552 num_cap_releases--;
e01a5946 1553
00bd8edb 1554 head = msg->front.iov_base;
745a8e3b
YZ		 1555 le32_add_cpu(&head->num, 1);
1556 item = msg->front.iov_base + msg->front.iov_len;
1557 item->ino = cpu_to_le64(cap->cap_ino);
1558 item->cap_id = cpu_to_le64(cap->cap_id);
1559 item->migrate_seq = cpu_to_le32(cap->mseq);
1560 item->seq = cpu_to_le32(cap->issue_seq);
1561 msg->front.iov_len += sizeof(*item);
1562
1563 ceph_put_cap(mdsc, cap);
1564
1565 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1566 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1567 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1568 ceph_con_send(&session->s_con, msg);
1569 msg = NULL;
1570 }
00bd8edb 1571 }
e01a5946 1572
745a8e3b 1573 BUG_ON(num_cap_releases != 0);
e01a5946 1574
745a8e3b
YZ		 1575 spin_lock(&session->s_cap_lock);
1576 if (!list_empty(&session->s_cap_releases))
1577 goto again;
1578 spin_unlock(&session->s_cap_lock);
1579
1580 if (msg) {
1581 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1582 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1583 ceph_con_send(&session->s_con, msg);
e01a5946 1584 }
745a8e3b
YZ		 1585 return;
1586out_err:
1587 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1588 session->s_mds);
1589 spin_lock(&session->s_cap_lock);
1590 list_splice(&tmp_list, &session->s_cap_releases);
1591 session->s_num_cap_releases += num_cap_releases;
1592 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1593}
1594
2f2dc053
SW		 1595/*
1596 * requests
1597 */
1598
54008399
YZ		 1599int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1600 struct inode *dir)
1601{
1602 struct ceph_inode_info *ci = ceph_inode(dir);
1603 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1604 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1605 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1606 int order, num_entries;
1607
1608 spin_lock(&ci->i_ceph_lock);
1609 num_entries = ci->i_files + ci->i_subdirs;
1610 spin_unlock(&ci->i_ceph_lock);
1611 num_entries = max(num_entries, 1);
1612 num_entries = min(num_entries, opt->max_readdir);
1613
1614 order = get_order(size * num_entries);
1615 while (order >= 0) {
2a5beea3
YZ		 1616 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1617 __GFP_NOWARN,
1618 order);
1619 if (rinfo->dir_entries)
54008399
YZ		 1620 break;
1621 order--;
1622 }
2a5beea3 1623 if (!rinfo->dir_entries)
54008399
YZ		 1624 return -ENOMEM;
1625
1626 num_entries = (PAGE_SIZE << order) / size;
1627 num_entries = min(num_entries, opt->max_readdir);
1628
1629 rinfo->dir_buf_size = PAGE_SIZE << order;
1630 req->r_num_caps = num_entries + 1;
1631 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1632 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1633 return 0;
1634}
1635
2f2dc053
SW		 1636/*
1637 * Create an mds request.
1638 */
1639struct ceph_mds_request *
1640ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1641{
1642 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1643
1644 if (!req)
1645 return ERR_PTR(-ENOMEM);
1646
b4556396 1647 mutex_init(&req->r_fill_mutex);
37151668 1648 req->r_mdsc = mdsc;
2f2dc053
SW		 1649 req->r_started = jiffies;
1650 req->r_resend_mds = -1;
1651 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1652 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1653 req->r_fmode = -1;
153c8e6b 1654 kref_init(&req->r_kref);
fcd00b68 1655 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1656 INIT_LIST_HEAD(&req->r_wait);
1657 init_completion(&req->r_completion);
1658 init_completion(&req->r_safe_completion);
1659 INIT_LIST_HEAD(&req->r_unsafe_item);
1660
8bbd4714 1661 req->r_stamp = current_fs_time(mdsc->fsc->sb);
b8e69066 1662
2f2dc053
SW		 1663 req->r_op = op;
1664 req->r_direct_mode = mode;
1665 return req;
1666}
1667
1668/*
44ca18f2 1669 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1670 *
1671 * called under mdsc->mutex.
1672 */
44ca18f2
SW		 1673static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1674{
1675 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1676 return NULL;
1677 return rb_entry(rb_first(&mdsc->request_tree),
1678 struct ceph_mds_request, r_node);
1679}
1680
e8a7b8b1 1681static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1682{
e8a7b8b1 1683 return mdsc->oldest_tid;
2f2dc053
SW		 1684}
1685
1686/*
1687 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1688 * on build_path_from_dentry in fs/cifs/dir.c.
1689 *
1690 * If @stop_on_nosnap, generate path relative to the first non-snapped
1691 * inode.
1692 *
1693 * Encode hidden .snap dirs as a double /, i.e.
1694 * foo/.snap/bar -> foo//bar
1695 */
1696char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1697 int stop_on_nosnap)
1698{
1699 struct dentry *temp;
1700 char *path;
1701 int len, pos;
1b71fe2e 1702 unsigned seq;
2f2dc053
SW		 1703
1704 if (dentry == NULL)
1705 return ERR_PTR(-EINVAL);
1706
1707retry:
1708 len = 0;
1b71fe2e
AV		 1709 seq = read_seqbegin(&rename_lock);
1710 rcu_read_lock();
2f2dc053 1711 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1712 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1713 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1714 len++; /* slash only */
1715 else if (stop_on_nosnap && inode &&
1716 ceph_snap(inode) == CEPH_NOSNAP)
1717 break;
1718 else
1719 len += 1 + temp->d_name.len;
1720 temp = temp->d_parent;
2f2dc053 1721 }
1b71fe2e 1722 rcu_read_unlock();
2f2dc053
SW		 1723 if (len)
1724 len--; /* no leading '/' */
1725
1726 path = kmalloc(len+1, GFP_NOFS);
1727 if (path == NULL)
1728 return ERR_PTR(-ENOMEM);
1729 pos = len;
1730 path[pos] = 0; /* trailing null */
1b71fe2e 1731 rcu_read_lock();
2f2dc053 1732 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1733 struct inode *inode;
2f2dc053 1734
1b71fe2e 1735 spin_lock(&temp->d_lock);
2b0143b5 1736 inode = d_inode(temp);
2f2dc053 1737 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1738 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1739 pos, temp);
1740 } else if (stop_on_nosnap && inode &&
1741 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1742 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1743 break;
1744 } else {
1745 pos -= temp->d_name.len;
1b71fe2e
AV		 1746 if (pos < 0) {
1747 spin_unlock(&temp->d_lock);
2f2dc053 1748 break;
1b71fe2e 1749 }
2f2dc053
SW		 1750 strncpy(path + pos, temp->d_name.name,
1751 temp->d_name.len);
2f2dc053 1752 }
1b71fe2e 1753 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1754 if (pos)
1755 path[--pos] = '/';
1756 temp = temp->d_parent;
2f2dc053 1757 }
1b71fe2e
AV		 1758 rcu_read_unlock();
1759 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1760 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1761 "expected, namelen is %d, pos is %d\n", len, pos);
1762 /* presumably this is only possible if racing with a
1763 rename of one of the parent directories (we can not
1764 lock the dentries above us to prevent this, but
1765 retrying should be harmless) */
1766 kfree(path);
1767 goto retry;
1768 }
1769
2b0143b5 1770 *base = ceph_ino(d_inode(temp));
2f2dc053 1771 *plen = len;
104648ad 1772 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1773 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1774 return path;
1775}
1776
1777static int build_dentry_path(struct dentry *dentry,
1778 const char **ppath, int *ppathlen, u64 *pino,
1779 int *pfreepath)
1780{
1781 char *path;
1782
2b0143b5
DH		 1783 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1784 *pino = ceph_ino(d_inode(dentry->d_parent));
2f2dc053
SW		 1785 *ppath = dentry->d_name.name;
1786 *ppathlen = dentry->d_name.len;
1787 return 0;
1788 }
1789 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1790 if (IS_ERR(path))
1791 return PTR_ERR(path);
1792 *ppath = path;
1793 *pfreepath = 1;
1794 return 0;
1795}
1796
1797static int build_inode_path(struct inode *inode,
1798 const char **ppath, int *ppathlen, u64 *pino,
1799 int *pfreepath)
1800{
1801 struct dentry *dentry;
1802 char *path;
1803
1804 if (ceph_snap(inode) == CEPH_NOSNAP) {
1805 *pino = ceph_ino(inode);
1806 *ppathlen = 0;
1807 return 0;
1808 }
1809 dentry = d_find_alias(inode);
1810 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1811 dput(dentry);
1812 if (IS_ERR(path))
1813 return PTR_ERR(path);
1814 *ppath = path;
1815 *pfreepath = 1;
1816 return 0;
1817}
1818
1819/*
1820 * request arguments may be specified via an inode *, a dentry *, or
1821 * an explicit ino+path.
1822 */
1823static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1824 const char *rpath, u64 rino,
1825 const char **ppath, int *pathlen,
1826 u64 *ino, int *freepath)
1827{
1828 int r = 0;
1829
1830 if (rinode) {
1831 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1832 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1833 ceph_snap(rinode));
1834 } else if (rdentry) {
1835 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1836 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1837 *ppath);
795858db 1838 } else if (rpath || rino) {
2f2dc053
SW		 1839 *ino = rino;
1840 *ppath = rpath;
b000056a 1841 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1842 dout(" path %.*s\n", *pathlen, rpath);
1843 }
1844
1845 return r;
1846}
1847
1848/*
1849 * called under mdsc->mutex
1850 */
1851static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1852 struct ceph_mds_request *req,
6e6f0923 1853 int mds, bool drop_cap_releases)
2f2dc053
SW		 1854{
1855 struct ceph_msg *msg;
1856 struct ceph_mds_request_head *head;
1857 const char *path1 = NULL;
1858 const char *path2 = NULL;
1859 u64 ino1 = 0, ino2 = 0;
1860 int pathlen1 = 0, pathlen2 = 0;
1861 int freepath1 = 0, freepath2 = 0;
1862 int len;
1863 u16 releases;
1864 void *p, *end;
1865 int ret;
1866
1867 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1868 req->r_path1, req->r_ino1.ino,
1869 &path1, &pathlen1, &ino1, &freepath1);
1870 if (ret < 0) {
1871 msg = ERR_PTR(ret);
1872 goto out;
1873 }
1874
1875 ret = set_request_path_attr(NULL, req->r_old_dentry,
1876 req->r_path2, req->r_ino2.ino,
1877 &path2, &pathlen2, &ino2, &freepath2);
1878 if (ret < 0) {
1879 msg = ERR_PTR(ret);
1880 goto out_free1;
1881 }
1882
1883 len = sizeof(*head) +
b8e69066 1884 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 1885 sizeof(struct ceph_timespec);
2f2dc053
SW		 1886
1887 /* calculate (max) length for cap releases */
1888 len += sizeof(struct ceph_mds_request_release) *
1889 (!!req->r_inode_drop + !!req->r_dentry_drop +
1890 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1891 if (req->r_dentry_drop)
1892 len += req->r_dentry->d_name.len;
1893 if (req->r_old_dentry_drop)
1894 len += req->r_old_dentry->d_name.len;
1895
b61c2763 1896 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1897 if (!msg) {
1898 msg = ERR_PTR(-ENOMEM);
2f2dc053 1899 goto out_free2;
a79832f2 1900 }
2f2dc053 1901
7cfa0313 1902 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1903 msg->hdr.tid = cpu_to_le64(req->r_tid);
1904
2f2dc053
SW		 1905 head = msg->front.iov_base;
1906 p = msg->front.iov_base + sizeof(*head);
1907 end = msg->front.iov_base + msg->front.iov_len;
1908
1909 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1910 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1911 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1912 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1913 head->args = req->r_args;
1914
1915 ceph_encode_filepath(&p, end, ino1, path1);
1916 ceph_encode_filepath(&p, end, ino2, path2);
1917
e979cf50
SW		 1918 /* make note of release offset, in case we need to replay */
1919 req->r_request_release_offset = p - msg->front.iov_base;
1920
2f2dc053
SW		 1921 /* cap releases */
1922 releases = 0;
1923 if (req->r_inode_drop)
1924 releases += ceph_encode_inode_release(&p,
2b0143b5 1925 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 1926 mds, req->r_inode_drop, req->r_inode_unless, 0);
1927 if (req->r_dentry_drop)
1928 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1929 mds, req->r_dentry_drop, req->r_dentry_unless);
1930 if (req->r_old_dentry_drop)
1931 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1932 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1933 if (req->r_old_inode_drop)
1934 releases += ceph_encode_inode_release(&p,
2b0143b5 1935 d_inode(req->r_old_dentry),
2f2dc053 1936 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 1937
1938 if (drop_cap_releases) {
1939 releases = 0;
1940 p = msg->front.iov_base + req->r_request_release_offset;
1941 }
1942
2f2dc053
SW		 1943 head->num_releases = cpu_to_le16(releases);
1944
b8e69066 1945 /* time stamp */
1f041a89
YZ		 1946 {
1947 struct ceph_timespec ts;
1948 ceph_encode_timespec(&ts, &req->r_stamp);
1949 ceph_encode_copy(&p, &ts, sizeof(ts));
1950 }
b8e69066 1951
2f2dc053
SW		 1952 BUG_ON(p > end);
1953 msg->front.iov_len = p - msg->front.iov_base;
1954 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1955
25e6bae3
YZ		 1956 if (req->r_pagelist) {
1957 struct ceph_pagelist *pagelist = req->r_pagelist;
1958 atomic_inc(&pagelist->refcnt);
1959 ceph_msg_data_add_pagelist(msg, pagelist);
1960 msg->hdr.data_len = cpu_to_le32(pagelist->length);
1961 } else {
1962 msg->hdr.data_len = 0;
ebf18f47 1963 }
02afca6c 1964
2f2dc053
SW		 1965 msg->hdr.data_off = cpu_to_le16(0);
1966
1967out_free2:
1968 if (freepath2)
1969 kfree((char *)path2);
1970out_free1:
1971 if (freepath1)
1972 kfree((char *)path1);
1973out:
1974 return msg;
1975}
1976
1977/*
1978 * called under mdsc->mutex if error, under no mutex if
1979 * success.
1980 */
1981static void complete_request(struct ceph_mds_client *mdsc,
1982 struct ceph_mds_request *req)
1983{
1984 if (req->r_callback)
1985 req->r_callback(mdsc, req);
1986 else
03066f23 1987 complete_all(&req->r_completion);
2f2dc053
SW		 1988}
1989
1990/*
1991 * called under mdsc->mutex
1992 */
1993static int __prepare_send_request(struct ceph_mds_client *mdsc,
1994 struct ceph_mds_request *req,
6e6f0923 1995 int mds, bool drop_cap_releases)
2f2dc053
SW		 1996{
1997 struct ceph_mds_request_head *rhead;
1998 struct ceph_msg *msg;
1999 int flags = 0;
2000
2f2dc053 2001 req->r_attempts++;
e55b71f8
GF		 2002 if (req->r_inode) {
2003 struct ceph_cap *cap =
2004 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2005
2006 if (cap)
2007 req->r_sent_on_mseq = cap->mseq;
2008 else
2009 req->r_sent_on_mseq = -1;
2010 }
2f2dc053
SW		 2011 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2012 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2013
01a92f17 2014 if (req->r_got_unsafe) {
c5c9a0bf 2015 void *p;
01a92f17
SW		 2016 /*
2017 * Replay. Do not regenerate message (and rebuild
2018 * paths, etc.); just use the original message.
2019 * Rebuilding paths will break for renames because
2020 * d_move mangles the src name.
2021 */
2022 msg = req->r_request;
2023 rhead = msg->front.iov_base;
2024
2025 flags = le32_to_cpu(rhead->flags);
2026 flags |= CEPH_MDS_FLAG_REPLAY;
2027 rhead->flags = cpu_to_le32(flags);
2028
2029 if (req->r_target_inode)
2030 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2031
2032 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2033
2034 /* remove cap/dentry releases from message */
2035 rhead->num_releases = 0;
c5c9a0bf
YZ		 2036
2037 /* time stamp */
2038 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2039 {
2040 struct ceph_timespec ts;
2041 ceph_encode_timespec(&ts, &req->r_stamp);
2042 ceph_encode_copy(&p, &ts, sizeof(ts));
2043 }
c5c9a0bf
YZ		 2044
2045 msg->front.iov_len = p - msg->front.iov_base;
2046 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2047 return 0;
2048 }
2049
2f2dc053
SW		 2050 if (req->r_request) {
2051 ceph_msg_put(req->r_request);
2052 req->r_request = NULL;
2053 }
6e6f0923 2054 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2055 if (IS_ERR(msg)) {
e1518c7c 2056 req->r_err = PTR_ERR(msg);
a79832f2 2057 return PTR_ERR(msg);
2f2dc053
SW		 2058 }
2059 req->r_request = msg;
2060
2061 rhead = msg->front.iov_base;
2f2dc053
SW		 2062 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2063 if (req->r_got_unsafe)
2064 flags |= CEPH_MDS_FLAG_REPLAY;
2065 if (req->r_locked_dir)
2066 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2067 rhead->flags = cpu_to_le32(flags);
2068 rhead->num_fwd = req->r_num_fwd;
2069 rhead->num_retry = req->r_attempts - 1;
01a92f17 2070 rhead->ino = 0;
2f2dc053
SW		 2071
2072 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 2073 return 0;
2074}
2075
2076/*
2077 * send request, or put it on the appropriate wait list.
2078 */
2079static int __do_request(struct ceph_mds_client *mdsc,
2080 struct ceph_mds_request *req)
2081{
2082 struct ceph_mds_session *session = NULL;
2083 int mds = -1;
48fec5d0 2084 int err = 0;
2f2dc053 2085
eb1b8af3
YZ		 2086 if (req->r_err || req->r_got_result) {
2087 if (req->r_aborted)
2088 __unregister_request(mdsc, req);
2f2dc053 2089 goto out;
eb1b8af3 2090 }
2f2dc053
SW		 2091
2092 if (req->r_timeout &&
2093 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2094 dout("do_request timed out\n");
2095 err = -EIO;
2096 goto finish;
2097 }
48fec5d0
YZ		 2098 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2099 dout("do_request forced umount\n");
2100 err = -EIO;
2101 goto finish;
2102 }
e9e427f0
YZ		 2103 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2104 if (mdsc->mdsmap_err) {
2105 err = mdsc->mdsmap_err;
2106 dout("do_request mdsmap err %d\n", err);
2107 goto finish;
2108 }
2109 if (!(mdsc->fsc->mount_options->flags &
2110 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2111 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2112 err = -ENOENT;
2113 pr_info("probably no mds server is up\n");
2114 goto finish;
2115 }
2116 }
2f2dc053 2117
dc69e2e9
SW		 2118 put_request_session(req);
2119
2f2dc053
SW		 2120 mds = __choose_mds(mdsc, req);
2121 if (mds < 0 ||
2122 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2123 dout("do_request no mds or not active, waiting for map\n");
2124 list_add(&req->r_wait, &mdsc->waiting_for_map);
2125 goto out;
2126 }
2127
2128 /* get, open session */
2129 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2130 if (!session) {
2f2dc053 2131 session = register_session(mdsc, mds);
9c423956
SW		 2132 if (IS_ERR(session)) {
2133 err = PTR_ERR(session);
2134 goto finish;
2135 }
2136 }
dc69e2e9
SW		 2137 req->r_session = get_session(session);
2138
2f2dc053 2139 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2140 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2141 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2142 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2143 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2144 err = -EACCES;
2145 goto out_session;
2146 }
2f2dc053
SW		 2147 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2148 session->s_state == CEPH_MDS_SESSION_CLOSING)
2149 __open_session(mdsc, session);
2150 list_add(&req->r_wait, &session->s_waiting);
2151 goto out_session;
2152 }
2153
2154 /* send request */
2f2dc053
SW		 2155 req->r_resend_mds = -1; /* forget any previous mds hint */
2156
2157 if (req->r_request_started == 0) /* note request start time */
2158 req->r_request_started = jiffies;
2159
6e6f0923 2160 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2161 if (!err) {
2162 ceph_msg_get(req->r_request);
2163 ceph_con_send(&session->s_con, req->r_request);
2164 }
2165
2166out_session:
2167 ceph_put_mds_session(session);
48fec5d0
YZ		 2168finish:
2169 if (err) {
2170 dout("__do_request early error %d\n", err);
2171 req->r_err = err;
2172 complete_request(mdsc, req);
2173 __unregister_request(mdsc, req);
2174 }
2f2dc053
SW		 2175out:
2176 return err;
2f2dc053
SW		 2177}
2178
2179/*
2180 * called under mdsc->mutex
2181 */
2182static void __wake_requests(struct ceph_mds_client *mdsc,
2183 struct list_head *head)
2184{
ed75ec2c
YZ		 2185 struct ceph_mds_request *req;
2186 LIST_HEAD(tmp_list);
2187
2188 list_splice_init(head, &tmp_list);
2f2dc053 2189
ed75ec2c
YZ		 2190 while (!list_empty(&tmp_list)) {
2191 req = list_entry(tmp_list.next,
2192 struct ceph_mds_request, r_wait);
2f2dc053 2193 list_del_init(&req->r_wait);
7971bd92 2194 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2195 __do_request(mdsc, req);
2196 }
2197}
2198
2199/*
2200 * Wake up threads with requests pending for @mds, so that they can
29790f26 2201 * resubmit their requests to a possibly different mds.
2f2dc053 2202 */
29790f26 2203static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2204{
44ca18f2 2205 struct ceph_mds_request *req;
282c1052 2206 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2207
2208 dout("kick_requests mds%d\n", mds);
282c1052 2209 while (p) {
44ca18f2 2210 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2211 p = rb_next(p);
44ca18f2
SW		 2212 if (req->r_got_unsafe)
2213 continue;
3de22be6
YZ		 2214 if (req->r_attempts > 0)
2215 continue; /* only new requests */
44ca18f2
SW		 2216 if (req->r_session &&
2217 req->r_session->s_mds == mds) {
2218 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2219 list_del_init(&req->r_wait);
44ca18f2 2220 __do_request(mdsc, req);
2f2dc053
SW		 2221 }
2222 }
2223}
2224
2225void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2226 struct ceph_mds_request *req)
2227{
2228 dout("submit_request on %p\n", req);
2229 mutex_lock(&mdsc->mutex);
2230 __register_request(mdsc, req, NULL);
2231 __do_request(mdsc, req);
2232 mutex_unlock(&mdsc->mutex);
2233}
2234
2235/*
2236 * Synchrously perform an mds request. Take care of all of the
2237 * session setup, forwarding, retry details.
2238 */
2239int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2240 struct inode *dir,
2241 struct ceph_mds_request *req)
2242{
2243 int err;
2244
2245 dout("do_request on %p\n", req);
2246
2247 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2248 if (req->r_inode)
2249 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2250 if (req->r_locked_dir)
2251 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
844d87c3 2252 if (req->r_old_dentry_dir)
41b02e1f
SW		 2253 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2254 CEPH_CAP_PIN);
2f2dc053
SW		 2255
2256 /* issue */
2257 mutex_lock(&mdsc->mutex);
2258 __register_request(mdsc, req, dir);
2259 __do_request(mdsc, req);
2260
e1518c7c
SW		 2261 if (req->r_err) {
2262 err = req->r_err;
e1518c7c 2263 goto out;
2f2dc053
SW		 2264 }
2265
e1518c7c
SW		 2266 /* wait */
2267 mutex_unlock(&mdsc->mutex);
2268 dout("do_request waiting\n");
5be73034 2269 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2270 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2271 } else {
5be73034
ID		 2272 long timeleft = wait_for_completion_killable_timeout(
2273 &req->r_completion,
2274 ceph_timeout_jiffies(req->r_timeout));
2275 if (timeleft > 0)
2276 err = 0;
2277 else if (!timeleft)
2278 err = -EIO; /* timed out */
2279 else
2280 err = timeleft; /* killed */
e1518c7c
SW		 2281 }
2282 dout("do_request waited, got %d\n", err);
2283 mutex_lock(&mdsc->mutex);
5b1daecd 2284
e1518c7c
SW		 2285 /* only abort if we didn't race with a real reply */
2286 if (req->r_got_result) {
2287 err = le32_to_cpu(req->r_reply_info.head->result);
2288 } else if (err < 0) {
2289 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2290
2291 /*
2292 * ensure we aren't running concurrently with
2293 * ceph_fill_trace or ceph_readdir_prepopulate, which
2294 * rely on locks (dir mutex) held by our caller.
2295 */
2296 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 2297 req->r_err = err;
2298 req->r_aborted = true;
b4556396 2299 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2300
e1518c7c 2301 if (req->r_locked_dir &&
167c9e35
SW		 2302 (req->r_op & CEPH_MDS_OP_WRITE))
2303 ceph_invalidate_dir_request(req);
2f2dc053 2304 } else {
e1518c7c 2305 err = req->r_err;
2f2dc053 2306 }
2f2dc053 2307
e1518c7c
SW		 2308out:
2309 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2310 dout("do_request %p done, result %d\n", req, err);
2311 return err;
2312}
2313
167c9e35 2314/*
2f276c51 2315 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2316 * namespace request.
2317 */
2318void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2319{
2320 struct inode *inode = req->r_locked_dir;
167c9e35 2321
2f276c51 2322 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2323
2f276c51 2324 ceph_dir_clear_complete(inode);
167c9e35
SW		 2325 if (req->r_dentry)
2326 ceph_invalidate_dentry_lease(req->r_dentry);
2327 if (req->r_old_dentry)
2328 ceph_invalidate_dentry_lease(req->r_old_dentry);
2329}
2330
2f2dc053
SW		 2331/*
2332 * Handle mds reply.
2333 *
2334 * We take the session mutex and parse and process the reply immediately.
2335 * This preserves the logical ordering of replies, capabilities, etc., sent
2336 * by the MDS as they are applied to our local cache.
2337 */
2338static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2339{
2340 struct ceph_mds_client *mdsc = session->s_mdsc;
2341 struct ceph_mds_request *req;
2342 struct ceph_mds_reply_head *head = msg->front.iov_base;
2343 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2344 struct ceph_snap_realm *realm;
2f2dc053
SW		 2345 u64 tid;
2346 int err, result;
2600d2dd 2347 int mds = session->s_mds;
2f2dc053 2348
2f2dc053
SW		 2349 if (msg->front.iov_len < sizeof(*head)) {
2350 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2351 ceph_msg_dump(msg);
2f2dc053
SW		 2352 return;
2353 }
2354
2355 /* get request, session */
6df058c0 2356 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2357 mutex_lock(&mdsc->mutex);
fcd00b68 2358 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2359 if (!req) {
2360 dout("handle_reply on unknown tid %llu\n", tid);
2361 mutex_unlock(&mdsc->mutex);
2362 return;
2363 }
2364 dout("handle_reply %p\n", req);
2f2dc053
SW		 2365
2366 /* correct session? */
d96d6049 2367 if (req->r_session != session) {
2f2dc053
SW		 2368 pr_err("mdsc_handle_reply got %llu on session mds%d"
2369 " not mds%d\n", tid, session->s_mds,
2370 req->r_session ? req->r_session->s_mds : -1);
2371 mutex_unlock(&mdsc->mutex);
2372 goto out;
2373 }
2374
2375 /* dup? */
2376 if ((req->r_got_unsafe && !head->safe) ||
2377 (req->r_got_safe && head->safe)) {
f3ae1b97 2378 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2379 head->safe ? "safe" : "unsafe", tid, mds);
2380 mutex_unlock(&mdsc->mutex);
2381 goto out;
2382 }
1550d34e 2383 if (req->r_got_safe) {
f3ae1b97 2384 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2385 tid, mds);
2386 mutex_unlock(&mdsc->mutex);
2387 goto out;
2388 }
2f2dc053
SW		 2389
2390 result = le32_to_cpu(head->result);
2391
2392 /*
e55b71f8
GF		 2393 * Handle an ESTALE
2394 * if we're not talking to the authority, send to them
2395 * if the authority has changed while we weren't looking,
2396 * send to new authority
2397 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2398 */
2399 if (result == -ESTALE) {
e55b71f8 2400 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2401 req->r_resend_mds = -1;
ca18bede 2402 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2403 dout("not using auth, setting for that now");
2404 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2405 __do_request(mdsc, req);
2406 mutex_unlock(&mdsc->mutex);
2407 goto out;
e55b71f8 2408 } else {
ca18bede
YZ		 2409 int mds = __choose_mds(mdsc, req);
2410 if (mds >= 0 && mds != req->r_session->s_mds) {
2411 dout("but auth changed, so resending");
e55b71f8
GF		 2412 __do_request(mdsc, req);
2413 mutex_unlock(&mdsc->mutex);
2414 goto out;
2415 }
2f2dc053 2416 }
e55b71f8 2417 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2418 }
2419
e55b71f8 2420
2f2dc053
SW		 2421 if (head->safe) {
2422 req->r_got_safe = true;
2423 __unregister_request(mdsc, req);
2f2dc053
SW		 2424
2425 if (req->r_got_unsafe) {
2426 /*
2427 * We already handled the unsafe response, now do the
2428 * cleanup. No need to examine the response; the MDS
2429 * doesn't include any result info in the safe
2430 * response. And even if it did, there is nothing
2431 * useful we could do with a revised return value.
2432 */
2433 dout("got safe reply %llu, mds%d\n", tid, mds);
2434 list_del_init(&req->r_unsafe_item);
2435
2436 /* last unsafe request during umount? */
44ca18f2 2437 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2438 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2439 mutex_unlock(&mdsc->mutex);
2440 goto out;
2441 }
e1518c7c 2442 } else {
2f2dc053
SW		 2443 req->r_got_unsafe = true;
2444 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2445 if (req->r_unsafe_dir) {
2446 struct ceph_inode_info *ci =
2447 ceph_inode(req->r_unsafe_dir);
2448 spin_lock(&ci->i_unsafe_lock);
2449 list_add_tail(&req->r_unsafe_dir_item,
2450 &ci->i_unsafe_dirops);
2451 spin_unlock(&ci->i_unsafe_lock);
2452 }
2f2dc053
SW		 2453 }
2454
2455 dout("handle_reply tid %lld result %d\n", tid, result);
2456 rinfo = &req->r_reply_info;
14303d20 2457 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2458 mutex_unlock(&mdsc->mutex);
2459
2460 mutex_lock(&session->s_mutex);
2461 if (err < 0) {
25933abd 2462 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2463 ceph_msg_dump(msg);
2f2dc053
SW		 2464 goto out_err;
2465 }
2466
2467 /* snap trace */
982d6011 2468 realm = NULL;
2f2dc053
SW		 2469 if (rinfo->snapblob_len) {
2470 down_write(&mdsc->snap_rwsem);
2471 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2472 rinfo->snapblob + rinfo->snapblob_len,
2473 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2474 &realm);
2f2dc053
SW		 2475 downgrade_write(&mdsc->snap_rwsem);
2476 } else {
2477 down_read(&mdsc->snap_rwsem);
2478 }
2479
2480 /* insert trace into our cache */
b4556396 2481 mutex_lock(&req->r_fill_mutex);
315f2408 2482 current->journal_info = req;
3d14c5d2 2483 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053 2484 if (err == 0) {
6e8575fa 2485 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2486 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2487 ceph_readdir_prepopulate(req, req->r_session);
37151668 2488 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2489 }
315f2408 2490 current->journal_info = NULL;
b4556396 2491 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2492
2493 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2494 if (realm)
2495 ceph_put_snap_realm(mdsc, realm);
68cd5b4b
YZ		 2496
2497 if (err == 0 && req->r_got_unsafe && req->r_target_inode) {
2498 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2499 spin_lock(&ci->i_unsafe_lock);
2500 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2501 spin_unlock(&ci->i_unsafe_lock);
2502 }
2f2dc053 2503out_err:
e1518c7c
SW		 2504 mutex_lock(&mdsc->mutex);
2505 if (!req->r_aborted) {
2506 if (err) {
2507 req->r_err = err;
2508 } else {
5fdb1389 2509 req->r_reply = ceph_msg_get(msg);
e1518c7c
SW		 2510 req->r_got_result = true;
2511 }
2f2dc053 2512 } else {
e1518c7c 2513 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2514 }
e1518c7c 2515 mutex_unlock(&mdsc->mutex);
2f2dc053 2516
2f2dc053
SW		 2517 mutex_unlock(&session->s_mutex);
2518
2519 /* kick calling process */
2520 complete_request(mdsc, req);
2521out:
2522 ceph_mdsc_put_request(req);
2523 return;
2524}
2525
2526
2527
2528/*
2529 * handle mds notification that our request has been forwarded.
2530 */
2600d2dd
SW		 2531static void handle_forward(struct ceph_mds_client *mdsc,
2532 struct ceph_mds_session *session,
2533 struct ceph_msg *msg)
2f2dc053
SW		 2534{
2535 struct ceph_mds_request *req;
a1ea787c 2536 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2537 u32 next_mds;
2538 u32 fwd_seq;
2f2dc053
SW		 2539 int err = -EINVAL;
2540 void *p = msg->front.iov_base;
2541 void *end = p + msg->front.iov_len;
2f2dc053 2542
a1ea787c 2543 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2544 next_mds = ceph_decode_32(&p);
2545 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2546
2547 mutex_lock(&mdsc->mutex);
fcd00b68 2548 req = lookup_get_request(mdsc, tid);
2f2dc053 2549 if (!req) {
2a8e5e36 2550 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2551 goto out; /* dup reply? */
2552 }
2553
2a8e5e36
SW		 2554 if (req->r_aborted) {
2555 dout("forward tid %llu aborted, unregistering\n", tid);
2556 __unregister_request(mdsc, req);
2557 } else if (fwd_seq <= req->r_num_fwd) {
2558 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2559 tid, next_mds, req->r_num_fwd, fwd_seq);
2560 } else {
2561 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2562 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2563 BUG_ON(req->r_err);
2564 BUG_ON(req->r_got_result);
3de22be6 2565 req->r_attempts = 0;
2f2dc053
SW		 2566 req->r_num_fwd = fwd_seq;
2567 req->r_resend_mds = next_mds;
2568 put_request_session(req);
2569 __do_request(mdsc, req);
2570 }
2571 ceph_mdsc_put_request(req);
2572out:
2573 mutex_unlock(&mdsc->mutex);
2574 return;
2575
2576bad:
2577 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2578}
2579
2580/*
2581 * handle a mds session control message
2582 */
2583static void handle_session(struct ceph_mds_session *session,
2584 struct ceph_msg *msg)
2585{
2586 struct ceph_mds_client *mdsc = session->s_mdsc;
2587 u32 op;
2588 u64 seq;
2600d2dd 2589 int mds = session->s_mds;
2f2dc053
SW		 2590 struct ceph_mds_session_head *h = msg->front.iov_base;
2591 int wake = 0;
2592
2f2dc053
SW		 2593 /* decode */
2594 if (msg->front.iov_len != sizeof(*h))
2595 goto bad;
2596 op = le32_to_cpu(h->op);
2597 seq = le64_to_cpu(h->seq);
2598
2599 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2600 if (op == CEPH_SESSION_CLOSE)
2601 __unregister_session(mdsc, session);
2f2dc053
SW		 2602 /* FIXME: this ttl calculation is generous */
2603 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2604 mutex_unlock(&mdsc->mutex);
2605
2606 mutex_lock(&session->s_mutex);
2607
2608 dout("handle_session mds%d %s %p state %s seq %llu\n",
2609 mds, ceph_session_op_name(op), session,
a687ecaf 2610 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2611
2612 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2613 session->s_state = CEPH_MDS_SESSION_OPEN;
2614 pr_info("mds%d came back\n", session->s_mds);
2615 }
2616
2617 switch (op) {
2618 case CEPH_SESSION_OPEN:
29790f26
SW		 2619 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2620 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2621 session->s_state = CEPH_MDS_SESSION_OPEN;
2622 renewed_caps(mdsc, session, 0);
2623 wake = 1;
2624 if (mdsc->stopping)
2625 __close_session(mdsc, session);
2626 break;
2627
2628 case CEPH_SESSION_RENEWCAPS:
2629 if (session->s_renew_seq == seq)
2630 renewed_caps(mdsc, session, 1);
2631 break;
2632
2633 case CEPH_SESSION_CLOSE:
29790f26
SW		 2634 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2635 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2636 cleanup_session_requests(mdsc, session);
2f2dc053 2637 remove_session_caps(session);
656e4382 2638 wake = 2; /* for good measure */
f3c60c59 2639 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2640 break;
2641
2642 case CEPH_SESSION_STALE:
2643 pr_info("mds%d caps went stale, renewing\n",
2644 session->s_mds);
d8fb02ab 2645 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2646 session->s_cap_gen++;
1ce208a6 2647 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2648 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2649 send_renew_caps(mdsc, session);
2650 break;
2651
2652 case CEPH_SESSION_RECALL_STATE:
2653 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2654 break;
2655
186e4f7a
YZ		 2656 case CEPH_SESSION_FLUSHMSG:
2657 send_flushmsg_ack(mdsc, session, seq);
2658 break;
2659
03f4fcb0
YZ		 2660 case CEPH_SESSION_FORCE_RO:
2661 dout("force_session_readonly %p\n", session);
2662 spin_lock(&session->s_cap_lock);
2663 session->s_readonly = true;
2664 spin_unlock(&session->s_cap_lock);
2665 wake_up_session_caps(session, 0);
2666 break;
2667
fcff415c
YZ		 2668 case CEPH_SESSION_REJECT:
2669 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2670 pr_info("mds%d rejected session\n", session->s_mds);
2671 session->s_state = CEPH_MDS_SESSION_REJECTED;
2672 cleanup_session_requests(mdsc, session);
2673 remove_session_caps(session);
2674 wake = 2; /* for good measure */
2675 break;
2676
2f2dc053
SW		 2677 default:
2678 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2679 WARN_ON(1);
2680 }
2681
2682 mutex_unlock(&session->s_mutex);
2683 if (wake) {
2684 mutex_lock(&mdsc->mutex);
2685 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2686 if (wake == 2)
2687 kick_requests(mdsc, mds);
2f2dc053
SW		 2688 mutex_unlock(&mdsc->mutex);
2689 }
2690 return;
2691
2692bad:
2693 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2694 (int)msg->front.iov_len);
9ec7cab1 2695 ceph_msg_dump(msg);
2f2dc053
SW		 2696 return;
2697}
2698
2699
2700/*
2701 * called under session->mutex.
2702 */
2703static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2704 struct ceph_mds_session *session)
2705{
2706 struct ceph_mds_request *req, *nreq;
3de22be6 2707 struct rb_node *p;
2f2dc053
SW		 2708 int err;
2709
2710 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2711
2712 mutex_lock(&mdsc->mutex);
2713 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2714 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2715 if (!err) {
2716 ceph_msg_get(req->r_request);
2717 ceph_con_send(&session->s_con, req->r_request);
2718 }
2719 }
3de22be6
YZ		 2720
2721 /*
2722 * also re-send old requests when MDS enters reconnect stage. So that MDS
2723 * can process completed request in clientreplay stage.
2724 */
2725 p = rb_first(&mdsc->request_tree);
2726 while (p) {
2727 req = rb_entry(p, struct ceph_mds_request, r_node);
2728 p = rb_next(p);
2729 if (req->r_got_unsafe)
2730 continue;
2731 if (req->r_attempts == 0)
2732 continue; /* only old requests */
2733 if (req->r_session &&
2734 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2735 err = __prepare_send_request(mdsc, req,
2736 session->s_mds, true);
3de22be6
YZ		 2737 if (!err) {
2738 ceph_msg_get(req->r_request);
2739 ceph_con_send(&session->s_con, req->r_request);
2740 }
2741 }
2742 }
2f2dc053
SW		 2743 mutex_unlock(&mdsc->mutex);
2744}
2745
2746/*
2747 * Encode information about a cap for a reconnect with the MDS.
2748 */
2f2dc053
SW		 2749static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2750 void *arg)
2751{
20cb34ae
SW		 2752 union {
2753 struct ceph_mds_cap_reconnect v2;
2754 struct ceph_mds_cap_reconnect_v1 v1;
2755 } rec;
2f2dc053 2756 struct ceph_inode_info *ci;
20cb34ae
SW		 2757 struct ceph_reconnect_state *recon_state = arg;
2758 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2759 char *path;
2760 int pathlen, err;
2761 u64 pathbase;
3469ed0d 2762 u64 snap_follows;
2f2dc053
SW		 2763 struct dentry *dentry;
2764
2765 ci = cap->ci;
2766
2767 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2768 inode, ceph_vinop(inode), cap, cap->cap_id,
2769 ceph_cap_string(cap->issued));
93cea5be
SW		 2770 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2771 if (err)
2772 return err;
2f2dc053
SW		 2773
2774 dentry = d_find_alias(inode);
2775 if (dentry) {
2776 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2777 if (IS_ERR(path)) {
2778 err = PTR_ERR(path);
e072f8aa 2779 goto out_dput;
2f2dc053
SW		 2780 }
2781 } else {
2782 path = NULL;
2783 pathlen = 0;
4eacd4cb 2784 pathbase = 0;
2f2dc053 2785 }
2f2dc053 2786
be655596 2787 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2788 cap->seq = 0; /* reset cap seq */
2789 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2790 cap->mseq = 0; /* and migrate_seq */
99a9c273 2791 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2792
121f22a1 2793 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2794 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2795 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2796 rec.v2.issued = cpu_to_le32(cap->issued);
2797 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2798 rec.v2.pathbase = cpu_to_le64(pathbase);
2799 rec.v2.flock_len = 0;
20cb34ae
SW		 2800 } else {
2801 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2802 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2803 rec.v1.issued = cpu_to_le32(cap->issued);
2804 rec.v1.size = cpu_to_le64(inode->i_size);
2805 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2806 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2807 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2808 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2809 }
3469ed0d
YZ		 2810
2811 if (list_empty(&ci->i_cap_snaps)) {
2812 snap_follows = 0;
2813 } else {
2814 struct ceph_cap_snap *capsnap =
2815 list_first_entry(&ci->i_cap_snaps,
2816 struct ceph_cap_snap, ci_item);
2817 snap_follows = capsnap->follows;
20cb34ae 2818 }
be655596 2819 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2820
121f22a1 2821 if (recon_state->msg_version >= 2) {
40819f6f 2822 int num_fcntl_locks, num_flock_locks;
39be95e9 2823 struct ceph_filelock *flocks;
121f22a1
YZ		 2824 size_t struct_len, total_len = 0;
2825 u8 struct_v = 0;
39be95e9
JS		 2826
2827encode_again:
39be95e9 2828 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
39be95e9
JS		 2829 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2830 sizeof(struct ceph_filelock), GFP_NOFS);
2831 if (!flocks) {
2832 err = -ENOMEM;
2833 goto out_free;
2834 }
39be95e9
JS		 2835 err = ceph_encode_locks_to_buffer(inode, flocks,
2836 num_fcntl_locks,
2837 num_flock_locks);
39be95e9
JS		 2838 if (err) {
2839 kfree(flocks);
2840 if (err == -ENOSPC)
2841 goto encode_again;
2842 goto out_free;
2843 }
121f22a1
YZ		 2844
2845 if (recon_state->msg_version >= 3) {
2846 /* version, compat_version and struct_len */
2847 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 2848 struct_v = 2;
121f22a1 2849 }
39be95e9
JS		 2850 /*
2851 * number of encoded locks is stable, so copy to pagelist
2852 */
121f22a1
YZ		 2853 struct_len = 2 * sizeof(u32) +
2854 (num_fcntl_locks + num_flock_locks) *
2855 sizeof(struct ceph_filelock);
2856 rec.v2.flock_len = cpu_to_le32(struct_len);
2857
2858 struct_len += sizeof(rec.v2);
2859 struct_len += sizeof(u32) + pathlen;
2860
3469ed0d
YZ		 2861 if (struct_v >= 2)
2862 struct_len += sizeof(u64); /* snap_follows */
2863
121f22a1
YZ		 2864 total_len += struct_len;
2865 err = ceph_pagelist_reserve(pagelist, total_len);
2866
2867 if (!err) {
2868 if (recon_state->msg_version >= 3) {
2869 ceph_pagelist_encode_8(pagelist, struct_v);
2870 ceph_pagelist_encode_8(pagelist, 1);
2871 ceph_pagelist_encode_32(pagelist, struct_len);
2872 }
2873 ceph_pagelist_encode_string(pagelist, path, pathlen);
2874 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
2875 ceph_locks_to_pagelist(flocks, pagelist,
2876 num_fcntl_locks,
2877 num_flock_locks);
3469ed0d
YZ		 2878 if (struct_v >= 2)
2879 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 2880 }
39be95e9 2881 kfree(flocks);
3612abbd 2882 } else {
121f22a1
YZ		 2883 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
2884 err = ceph_pagelist_reserve(pagelist, size);
2885 if (!err) {
2886 ceph_pagelist_encode_string(pagelist, path, pathlen);
2887 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
2888 }
40819f6f 2889 }
44c99757
YZ		 2890
2891 recon_state->nr_caps++;
e072f8aa 2892out_free:
2f2dc053 2893 kfree(path);
e072f8aa 2894out_dput:
2f2dc053 2895 dput(dentry);
93cea5be 2896 return err;
2f2dc053
SW		 2897}
2898
2899
2900/*
2901 * If an MDS fails and recovers, clients need to reconnect in order to
2902 * reestablish shared state. This includes all caps issued through
2903 * this session _and_ the snap_realm hierarchy. Because it's not
2904 * clear which snap realms the mds cares about, we send everything we
2905 * know about.. that ensures we'll then get any new info the
2906 * recovering MDS might have.
2907 *
2908 * This is a relatively heavyweight operation, but it's rare.
2909 *
2910 * called with mdsc->mutex held.
2911 */
34b6c855
SW		 2912static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2913 struct ceph_mds_session *session)
2f2dc053 2914{
2f2dc053 2915 struct ceph_msg *reply;
a105f00c 2916 struct rb_node *p;
34b6c855 2917 int mds = session->s_mds;
9abf82b8 2918 int err = -ENOMEM;
44c99757 2919 int s_nr_caps;
93cea5be 2920 struct ceph_pagelist *pagelist;
20cb34ae 2921 struct ceph_reconnect_state recon_state;
2f2dc053 2922
34b6c855 2923 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2924
93cea5be
SW		 2925 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2926 if (!pagelist)
2927 goto fail_nopagelist;
2928 ceph_pagelist_init(pagelist);
2929
b61c2763 2930 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 2931 if (!reply)
93cea5be 2932 goto fail_nomsg;
93cea5be 2933
34b6c855
SW		 2934 mutex_lock(&session->s_mutex);
2935 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2936 session->s_seq = 0;
2f2dc053 2937
2f2dc053 2938 dout("session %p state %s\n", session,
a687ecaf 2939 ceph_session_state_name(session->s_state));
2f2dc053 2940
99a9c273
YZ		 2941 spin_lock(&session->s_gen_ttl_lock);
2942 session->s_cap_gen++;
2943 spin_unlock(&session->s_gen_ttl_lock);
2944
2945 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 2946 /* don't know if session is readonly */
2947 session->s_readonly = 0;
99a9c273
YZ		 2948 /*
2949 * notify __ceph_remove_cap() that we are composing cap reconnect.
2950 * If a cap get released before being added to the cap reconnect,
2951 * __ceph_remove_cap() should skip queuing cap release.
2952 */
2953 session->s_cap_reconnect = 1;
e01a5946 2954 /* drop old cap expires; we're about to reestablish that state */
745a8e3b 2955 cleanup_cap_releases(mdsc, session);
e01a5946 2956
5d23371f 2957 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 2958 if (mdsc->fsc->sb->s_root)
2959 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 2960
2961 ceph_con_close(&session->s_con);
2962 ceph_con_open(&session->s_con,
2963 CEPH_ENTITY_TYPE_MDS, mds,
2964 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2965
2966 /* replay unsafe requests */
2967 replay_unsafe_requests(mdsc, session);
2968
2969 down_read(&mdsc->snap_rwsem);
2970
2f2dc053 2971 /* traverse this session's caps */
44c99757
YZ		 2972 s_nr_caps = session->s_nr_caps;
2973 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 2974 if (err)
2975 goto fail;
20cb34ae 2976
44c99757 2977 recon_state.nr_caps = 0;
20cb34ae 2978 recon_state.pagelist = pagelist;
121f22a1
YZ		 2979 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
2980 recon_state.msg_version = 3;
2981 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
2982 recon_state.msg_version = 2;
2983 else
2984 recon_state.msg_version = 1;
20cb34ae 2985 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2986 if (err < 0)
9abf82b8 2987 goto fail;
2f2dc053 2988
99a9c273
YZ		 2989 spin_lock(&session->s_cap_lock);
2990 session->s_cap_reconnect = 0;
2991 spin_unlock(&session->s_cap_lock);
2992
2f2dc053
SW		 2993 /*
2994 * snaprealms. we provide mds with the ino, seq (version), and
2995 * parent for all of our realms. If the mds has any newer info,
2996 * it will tell us.
2997 */
a105f00c
SW		 2998 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2999 struct ceph_snap_realm *realm =
3000 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3001 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3002
3003 dout(" adding snap realm %llx seq %lld parent %llx\n",
3004 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3005 sr_rec.ino = cpu_to_le64(realm->ino);
3006 sr_rec.seq = cpu_to_le64(realm->seq);
3007 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3008 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3009 if (err)
3010 goto fail;
2f2dc053 3011 }
2f2dc053 3012
121f22a1 3013 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3014
3015 /* raced with cap release? */
3016 if (s_nr_caps != recon_state.nr_caps) {
3017 struct page *page = list_first_entry(&pagelist->head,
3018 struct page, lru);
3019 __le32 *addr = kmap_atomic(page);
3020 *addr = cpu_to_le32(recon_state.nr_caps);
3021 kunmap_atomic(addr);
ebf18f47 3022 }
44c99757
YZ		 3023
3024 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3025 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3026
3027 ceph_early_kick_flushing_caps(mdsc, session);
3028
2f2dc053
SW		 3029 ceph_con_send(&session->s_con, reply);
3030
9abf82b8
SW		 3031 mutex_unlock(&session->s_mutex);
3032
3033 mutex_lock(&mdsc->mutex);
3034 __wake_requests(mdsc, &session->s_waiting);
3035 mutex_unlock(&mdsc->mutex);
3036
2f2dc053 3037 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3038 return;
3039
93cea5be 3040fail:
2f2dc053 3041 ceph_msg_put(reply);
9abf82b8
SW		 3042 up_read(&mdsc->snap_rwsem);
3043 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3044fail_nomsg:
3045 ceph_pagelist_release(pagelist);
93cea5be 3046fail_nopagelist:
9abf82b8 3047 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3048 return;
2f2dc053
SW		 3049}
3050
3051
3052/*
3053 * compare old and new mdsmaps, kicking requests
3054 * and closing out old connections as necessary
3055 *
3056 * called under mdsc->mutex.
3057 */
3058static void check_new_map(struct ceph_mds_client *mdsc,
3059 struct ceph_mdsmap *newmap,
3060 struct ceph_mdsmap *oldmap)
3061{
3062 int i;
3063 int oldstate, newstate;
3064 struct ceph_mds_session *s;
3065
3066 dout("check_new_map new %u old %u\n",
3067 newmap->m_epoch, oldmap->m_epoch);
3068
3069 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
3070 if (mdsc->sessions[i] == NULL)
3071 continue;
3072 s = mdsc->sessions[i];
3073 oldstate = ceph_mdsmap_get_state(oldmap, i);
3074 newstate = ceph_mdsmap_get_state(newmap, i);
3075
0deb01c9 3076 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3077 i, ceph_mds_state_name(oldstate),
0deb01c9 3078 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3079 ceph_mds_state_name(newstate),
0deb01c9 3080 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3081 ceph_session_state_name(s->s_state));
2f2dc053 3082
3e8f43a0
YZ		 3083 if (i >= newmap->m_max_mds ||
3084 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3085 ceph_mdsmap_get_addr(newmap, i),
3086 sizeof(struct ceph_entity_addr))) {
3087 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3088 /* the session never opened, just close it
3089 * out now */
3090 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 3091 __unregister_session(mdsc, s);
2f2dc053
SW		 3092 } else {
3093 /* just close it */
3094 mutex_unlock(&mdsc->mutex);
3095 mutex_lock(&s->s_mutex);
3096 mutex_lock(&mdsc->mutex);
3097 ceph_con_close(&s->s_con);
3098 mutex_unlock(&s->s_mutex);
3099 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3100 }
2f2dc053
SW		 3101 } else if (oldstate == newstate) {
3102 continue; /* nothing new with this mds */
3103 }
3104
3105 /*
3106 * send reconnect?
3107 */
3108 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3109 newstate >= CEPH_MDS_STATE_RECONNECT) {
3110 mutex_unlock(&mdsc->mutex);
3111 send_mds_reconnect(mdsc, s);
3112 mutex_lock(&mdsc->mutex);
3113 }
2f2dc053
SW		 3114
3115 /*
29790f26 3116 * kick request on any mds that has gone active.
2f2dc053
SW		 3117 */
3118 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3119 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3120 if (oldstate != CEPH_MDS_STATE_CREATING &&
3121 oldstate != CEPH_MDS_STATE_STARTING)
3122 pr_info("mds%d recovery completed\n", s->s_mds);
3123 kick_requests(mdsc, i);
2f2dc053 3124 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3125 wake_up_session_caps(s, 1);
2f2dc053
SW		 3126 }
3127 }
cb170a22
SW		 3128
3129 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3130 s = mdsc->sessions[i];
3131 if (!s)
3132 continue;
3133 if (!ceph_mdsmap_is_laggy(newmap, i))
3134 continue;
3135 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3136 s->s_state == CEPH_MDS_SESSION_HUNG ||
3137 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3138 dout(" connecting to export targets of laggy mds%d\n",
3139 i);
3140 __open_export_target_sessions(mdsc, s);
3141 }
3142 }
2f2dc053
SW		 3143}
3144
3145
3146
3147/*
3148 * leases
3149 */
3150
3151/*
3152 * caller must hold session s_mutex, dentry->d_lock
3153 */
3154void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3155{
3156 struct ceph_dentry_info *di = ceph_dentry(dentry);
3157
3158 ceph_put_mds_session(di->lease_session);
3159 di->lease_session = NULL;
3160}
3161
2600d2dd
SW		 3162static void handle_lease(struct ceph_mds_client *mdsc,
3163 struct ceph_mds_session *session,
3164 struct ceph_msg *msg)
2f2dc053 3165{
3d14c5d2 3166 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3167 struct inode *inode;
2f2dc053
SW		 3168 struct dentry *parent, *dentry;
3169 struct ceph_dentry_info *di;
2600d2dd 3170 int mds = session->s_mds;
2f2dc053 3171 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3172 u32 seq;
2f2dc053 3173 struct ceph_vino vino;
2f2dc053
SW		 3174 struct qstr dname;
3175 int release = 0;
3176
2f2dc053
SW		 3177 dout("handle_lease from mds%d\n", mds);
3178
3179 /* decode */
3180 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3181 goto bad;
3182 vino.ino = le64_to_cpu(h->ino);
3183 vino.snap = CEPH_NOSNAP;
1e5ea23d 3184 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3185 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3186 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3187 if (dname.len != get_unaligned_le32(h+1))
3188 goto bad;
3189
2f2dc053
SW		 3190 /* lookup inode */
3191 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3192 dout("handle_lease %s, ino %llx %p %.*s\n",
3193 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3194 dname.len, dname.name);
6cd3bcad
YZ		 3195
3196 mutex_lock(&session->s_mutex);
3197 session->s_seq++;
3198
2f2dc053
SW		 3199 if (inode == NULL) {
3200 dout("handle_lease no inode %llx\n", vino.ino);
3201 goto release;
3202 }
2f2dc053
SW		 3203
3204 /* dentry */
3205 parent = d_find_alias(inode);
3206 if (!parent) {
3207 dout("no parent dentry on inode %p\n", inode);
3208 WARN_ON(1);
3209 goto release; /* hrm... */
3210 }
8387ff25 3211 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3212 dentry = d_lookup(parent, &dname);
3213 dput(parent);
3214 if (!dentry)
3215 goto release;
3216
3217 spin_lock(&dentry->d_lock);
3218 di = ceph_dentry(dentry);
3219 switch (h->action) {
3220 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3221 if (di->lease_session == session) {
1e5ea23d
SW		 3222 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3223 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3224 __ceph_mdsc_drop_dentry_lease(dentry);
3225 }
3226 release = 1;
3227 break;
3228
3229 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3230 if (di->lease_session == session &&
2f2dc053
SW		 3231 di->lease_gen == session->s_cap_gen &&
3232 di->lease_renew_from &&
3233 di->lease_renew_after == 0) {
3234 unsigned long duration =
3563dbdd 3235 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3236
1e5ea23d 3237 di->lease_seq = seq;
9b16f03c 3238 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3239 di->lease_renew_after = di->lease_renew_from +
3240 (duration >> 1);
3241 di->lease_renew_from = 0;
3242 }
3243 break;
3244 }
3245 spin_unlock(&dentry->d_lock);
3246 dput(dentry);
3247
3248 if (!release)
3249 goto out;
3250
3251release:
3252 /* let's just reuse the same message */
3253 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3254 ceph_msg_get(msg);
3255 ceph_con_send(&session->s_con, msg);
3256
3257out:
3258 iput(inode);
3259 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3260 return;
3261
3262bad:
3263 pr_err("corrupt lease message\n");
9ec7cab1 3264 ceph_msg_dump(msg);
2f2dc053
SW		 3265}
3266
3267void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3268 struct inode *inode,
3269 struct dentry *dentry, char action,
3270 u32 seq)
3271{
3272 struct ceph_msg *msg;
3273 struct ceph_mds_lease *lease;
3274 int len = sizeof(*lease) + sizeof(u32);
3275 int dnamelen = 0;
3276
3277 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3278 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3279 dnamelen = dentry->d_name.len;
3280 len += dnamelen;
3281
b61c2763 3282 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3283 if (!msg)
2f2dc053
SW		 3284 return;
3285 lease = msg->front.iov_base;
3286 lease->action = action;
2f2dc053
SW		 3287 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3288 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3289 lease->seq = cpu_to_le32(seq);
3290 put_unaligned_le32(dnamelen, lease + 1);
3291 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3292
3293 /*
3294 * if this is a preemptive lease RELEASE, no need to
3295 * flush request stream, since the actual request will
3296 * soon follow.
3297 */
3298 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3299
3300 ceph_con_send(&session->s_con, msg);
3301}
3302
2f2dc053
SW		 3303/*
3304 * drop all leases (and dentry refs) in preparation for umount
3305 */
3306static void drop_leases(struct ceph_mds_client *mdsc)
3307{
3308 int i;
3309
3310 dout("drop_leases\n");
3311 mutex_lock(&mdsc->mutex);
3312 for (i = 0; i < mdsc->max_sessions; i++) {
3313 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3314 if (!s)
3315 continue;
3316 mutex_unlock(&mdsc->mutex);
3317 mutex_lock(&s->s_mutex);
3318 mutex_unlock(&s->s_mutex);
3319 ceph_put_mds_session(s);
3320 mutex_lock(&mdsc->mutex);
3321 }
3322 mutex_unlock(&mdsc->mutex);
3323}
3324
3325
3326
3327/*
3328 * delayed work -- periodically trim expired leases, renew caps with mds
3329 */
3330static void schedule_delayed(struct ceph_mds_client *mdsc)
3331{
3332 int delay = 5;
3333 unsigned hz = round_jiffies_relative(HZ * delay);
3334 schedule_delayed_work(&mdsc->delayed_work, hz);
3335}
3336
3337static void delayed_work(struct work_struct *work)
3338{
3339 int i;
3340 struct ceph_mds_client *mdsc =
3341 container_of(work, struct ceph_mds_client, delayed_work.work);
3342 int renew_interval;
3343 int renew_caps;
3344
3345 dout("mdsc delayed_work\n");
afcdaea3 3346 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3347
3348 mutex_lock(&mdsc->mutex);
3349 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3350 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3351 mdsc->last_renew_caps);
3352 if (renew_caps)
3353 mdsc->last_renew_caps = jiffies;
3354
3355 for (i = 0; i < mdsc->max_sessions; i++) {
3356 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3357 if (s == NULL)
3358 continue;
3359 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3360 dout("resending session close request for mds%d\n",
3361 s->s_mds);
3362 request_close_session(mdsc, s);
3363 ceph_put_mds_session(s);
3364 continue;
3365 }
3366 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3367 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3368 s->s_state = CEPH_MDS_SESSION_HUNG;
3369 pr_info("mds%d hung\n", s->s_mds);
3370 }
3371 }
3372 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3373 /* this mds is failed or recovering, just wait */
3374 ceph_put_mds_session(s);
3375 continue;
3376 }
3377 mutex_unlock(&mdsc->mutex);
3378
3379 mutex_lock(&s->s_mutex);
3380 if (renew_caps)
3381 send_renew_caps(mdsc, s);
3382 else
3383 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3384 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3385 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3386 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3387 mutex_unlock(&s->s_mutex);
3388 ceph_put_mds_session(s);
3389
3390 mutex_lock(&mdsc->mutex);
3391 }
3392 mutex_unlock(&mdsc->mutex);
3393
3394 schedule_delayed(mdsc);
3395}
3396
3d14c5d2 3397int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3398
2f2dc053 3399{
3d14c5d2
YS		 3400 struct ceph_mds_client *mdsc;
3401
3402 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3403 if (!mdsc)
3404 return -ENOMEM;
3405 mdsc->fsc = fsc;
3406 fsc->mdsc = mdsc;
2f2dc053
SW		 3407 mutex_init(&mdsc->mutex);
3408 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3409 if (mdsc->mdsmap == NULL) {
3410 kfree(mdsc);
2d06eeb8 3411 return -ENOMEM;
fb3101b6 3412 }
2d06eeb8 3413
2f2dc053 3414 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3415 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3416 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3417 mdsc->sessions = NULL;
86d8f67b 3418 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3419 mdsc->max_sessions = 0;
3420 mdsc->stopping = 0;
affbc19a 3421 mdsc->last_snap_seq = 0;
2f2dc053 3422 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3423 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3424 INIT_LIST_HEAD(&mdsc->snap_empty);
3425 spin_lock_init(&mdsc->snap_empty_lock);
3426 mdsc->last_tid = 0;
e8a7b8b1 3427 mdsc->oldest_tid = 0;
44ca18f2 3428 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3429 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3430 mdsc->last_renew_caps = jiffies;
3431 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3432 spin_lock_init(&mdsc->cap_delay_lock);
3433 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3434 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3435 mdsc->last_cap_flush_tid = 1;
e4500b5e 3436 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3437 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3438 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3439 mdsc->num_cap_flushing = 0;
3440 spin_lock_init(&mdsc->cap_dirty_lock);
3441 init_waitqueue_head(&mdsc->cap_flushing_wq);
3442 spin_lock_init(&mdsc->dentry_lru_lock);
3443 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3444
37151668 3445 ceph_caps_init(mdsc);
3d14c5d2 3446 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3447
10183a69
YZ		 3448 init_rwsem(&mdsc->pool_perm_rwsem);
3449 mdsc->pool_perm_tree = RB_ROOT;
3450
5f44f142 3451 return 0;
2f2dc053
SW		 3452}
3453
3454/*
3455 * Wait for safe replies on open mds requests. If we time out, drop
3456 * all requests from the tree to avoid dangling dentry refs.
3457 */
3458static void wait_requests(struct ceph_mds_client *mdsc)
3459{
a319bf56 3460 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3461 struct ceph_mds_request *req;
2f2dc053
SW		 3462
3463 mutex_lock(&mdsc->mutex);
44ca18f2 3464 if (__get_oldest_req(mdsc)) {
2f2dc053 3465 mutex_unlock(&mdsc->mutex);
44ca18f2 3466
2f2dc053
SW		 3467 dout("wait_requests waiting for requests\n");
3468 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3469 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3470
3471 /* tear down remaining requests */
44ca18f2
SW		 3472 mutex_lock(&mdsc->mutex);
3473 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3474 dout("wait_requests timed out on tid %llu\n",
3475 req->r_tid);
44ca18f2 3476 __unregister_request(mdsc, req);
2f2dc053
SW		 3477 }
3478 }
3479 mutex_unlock(&mdsc->mutex);
3480 dout("wait_requests done\n");
3481}
3482
3483/*
3484 * called before mount is ro, and before dentries are torn down.
3485 * (hmm, does this still race with new lookups?)
3486 */
3487void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3488{
3489 dout("pre_umount\n");
3490 mdsc->stopping = 1;
3491
3492 drop_leases(mdsc);
afcdaea3 3493 ceph_flush_dirty_caps(mdsc);
2f2dc053 3494 wait_requests(mdsc);
17c688c3
SW		 3495
3496 /*
3497 * wait for reply handlers to drop their request refs and
3498 * their inode/dcache refs
3499 */
3500 ceph_msgr_flush();
2f2dc053
SW		 3501}
3502
3503/*
3504 * wait for all write mds requests to flush.
3505 */
3506static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3507{
80fc7314 3508 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3509 struct rb_node *n;
2f2dc053
SW		 3510
3511 mutex_lock(&mdsc->mutex);
3512 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3513restart:
44ca18f2
SW		 3514 req = __get_oldest_req(mdsc);
3515 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3516 /* find next request */
3517 n = rb_next(&req->r_node);
3518 if (n)
3519 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3520 else
3521 nextreq = NULL;
e8a7b8b1
YZ		 3522 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3523 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3524 /* write op */
3525 ceph_mdsc_get_request(req);
80fc7314
SW		 3526 if (nextreq)
3527 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3528 mutex_unlock(&mdsc->mutex);
3529 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3530 req->r_tid, want_tid);
3531 wait_for_completion(&req->r_safe_completion);
3532 mutex_lock(&mdsc->mutex);
44ca18f2 3533 ceph_mdsc_put_request(req);
80fc7314
SW		 3534 if (!nextreq)
3535 break; /* next dne before, so we're done! */
3536 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3537 /* next request was removed from tree */
3538 ceph_mdsc_put_request(nextreq);
3539 goto restart;
3540 }
3541 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3542 }
80fc7314 3543 req = nextreq;
2f2dc053
SW		 3544 }
3545 mutex_unlock(&mdsc->mutex);
3546 dout("wait_unsafe_requests done\n");
3547}
3548
3549void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3550{
0e294387 3551 u64 want_tid, want_flush;
2f2dc053 3552
48fec5d0 3553 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3554 return;
3555
2f2dc053
SW		 3556 dout("sync\n");
3557 mutex_lock(&mdsc->mutex);
3558 want_tid = mdsc->last_tid;
2f2dc053 3559 mutex_unlock(&mdsc->mutex);
2f2dc053 3560
afcdaea3 3561 ceph_flush_dirty_caps(mdsc);
d3383a8e 3562 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3563 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3564 if (!list_empty(&mdsc->cap_flush_list)) {
3565 struct ceph_cap_flush *cf =
3566 list_last_entry(&mdsc->cap_flush_list,
3567 struct ceph_cap_flush, g_list);
3568 cf->wake = true;
3569 }
d3383a8e
YZ		 3570 spin_unlock(&mdsc->cap_dirty_lock);
3571
0e294387
YZ		 3572 dout("sync want tid %lld flush_seq %lld\n",
3573 want_tid, want_flush);
2f2dc053
SW		 3574
3575 wait_unsafe_requests(mdsc, want_tid);
0e294387 3576 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3577}
3578
f3c60c59
SW		 3579/*
3580 * true if all sessions are closed, or we force unmount
3581 */
fcff415c 3582static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3583{
48fec5d0 3584 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3585 return true;
fcff415c 3586 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3587}
2f2dc053
SW		 3588
3589/*
3590 * called after sb is ro.
3591 */
3592void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3593{
a319bf56 3594 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3595 struct ceph_mds_session *session;
3596 int i;
fcff415c 3597 int skipped = 0;
2f2dc053
SW		 3598
3599 dout("close_sessions\n");
3600
2f2dc053 3601 /* close sessions */
f3c60c59
SW		 3602 mutex_lock(&mdsc->mutex);
3603 for (i = 0; i < mdsc->max_sessions; i++) {
3604 session = __ceph_lookup_mds_session(mdsc, i);
3605 if (!session)
3606 continue;
2f2dc053 3607 mutex_unlock(&mdsc->mutex);
f3c60c59 3608 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3609 if (__close_session(mdsc, session) <= 0)
3610 skipped++;
f3c60c59
SW		 3611 mutex_unlock(&session->s_mutex);
3612 ceph_put_mds_session(session);
2f2dc053
SW		 3613 mutex_lock(&mdsc->mutex);
3614 }
f3c60c59
SW		 3615 mutex_unlock(&mdsc->mutex);
3616
3617 dout("waiting for sessions to close\n");
fcff415c
YZ		 3618 wait_event_timeout(mdsc->session_close_wq,
3619 done_closing_sessions(mdsc, skipped),
a319bf56 3620 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3621
3622 /* tear down remaining sessions */
f3c60c59 3623 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3624 for (i = 0; i < mdsc->max_sessions; i++) {
3625 if (mdsc->sessions[i]) {
3626 session = get_session(mdsc->sessions[i]);
2600d2dd 3627 __unregister_session(mdsc, session);
2f2dc053
SW		 3628 mutex_unlock(&mdsc->mutex);
3629 mutex_lock(&session->s_mutex);
3630 remove_session_caps(session);
3631 mutex_unlock(&session->s_mutex);
3632 ceph_put_mds_session(session);
3633 mutex_lock(&mdsc->mutex);
3634 }
3635 }
2f2dc053 3636 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3637 mutex_unlock(&mdsc->mutex);
3638
3639 ceph_cleanup_empty_realms(mdsc);
3640
3641 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3642
3643 dout("stopped\n");
3644}
3645
48fec5d0
YZ		 3646void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3647{
3648 struct ceph_mds_session *session;
3649 int mds;
3650
3651 dout("force umount\n");
3652
3653 mutex_lock(&mdsc->mutex);
3654 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3655 session = __ceph_lookup_mds_session(mdsc, mds);
3656 if (!session)
3657 continue;
3658 mutex_unlock(&mdsc->mutex);
3659 mutex_lock(&session->s_mutex);
3660 __close_session(mdsc, session);
3661 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3662 cleanup_session_requests(mdsc, session);
3663 remove_session_caps(session);
3664 }
3665 mutex_unlock(&session->s_mutex);
3666 ceph_put_mds_session(session);
3667 mutex_lock(&mdsc->mutex);
3668 kick_requests(mdsc, mds);
3669 }
3670 __wake_requests(mdsc, &mdsc->waiting_for_map);
3671 mutex_unlock(&mdsc->mutex);
3672}
3673
3d14c5d2 3674static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3675{
3676 dout("stop\n");
3677 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3678 if (mdsc->mdsmap)
3679 ceph_mdsmap_destroy(mdsc->mdsmap);
3680 kfree(mdsc->sessions);
37151668 3681 ceph_caps_finalize(mdsc);
10183a69 3682 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3683}
3684
3d14c5d2
YS		 3685void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3686{
3687 struct ceph_mds_client *mdsc = fsc->mdsc;
3688
ef550f6f 3689 dout("mdsc_destroy %p\n", mdsc);
3d14c5d2 3690 ceph_mdsc_stop(mdsc);
ef550f6f
SW		 3691
3692 /* flush out any connection work with references to us */
3693 ceph_msgr_flush();
3694
3d14c5d2
YS		 3695 fsc->mdsc = NULL;
3696 kfree(mdsc);
ef550f6f 3697 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3698}
3699
430afbad
YZ		 3700void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3701{
3702 struct ceph_fs_client *fsc = mdsc->fsc;
3703 const char *mds_namespace = fsc->mount_options->mds_namespace;
3704 void *p = msg->front.iov_base;
3705 void *end = p + msg->front.iov_len;
3706 u32 epoch;
3707 u32 map_len;
3708 u32 num_fs;
3709 u32 mount_fscid = (u32)-1;
3710 u8 struct_v, struct_cv;
3711 int err = -EINVAL;
3712
3713 ceph_decode_need(&p, end, sizeof(u32), bad);
3714 epoch = ceph_decode_32(&p);
3715
3716 dout("handle_fsmap epoch %u\n", epoch);
3717
3718 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3719 struct_v = ceph_decode_8(&p);
3720 struct_cv = ceph_decode_8(&p);
3721 map_len = ceph_decode_32(&p);
3722
3723 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3724 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3725
3726 num_fs = ceph_decode_32(&p);
3727 while (num_fs-- > 0) {
3728 void *info_p, *info_end;
3729 u32 info_len;
3730 u8 info_v, info_cv;
3731 u32 fscid, namelen;
3732
3733 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3734 info_v = ceph_decode_8(&p);
3735 info_cv = ceph_decode_8(&p);
3736 info_len = ceph_decode_32(&p);
3737 ceph_decode_need(&p, end, info_len, bad);
3738 info_p = p;
3739 info_end = p + info_len;
3740 p = info_end;
3741
3742 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3743 fscid = ceph_decode_32(&info_p);
3744 namelen = ceph_decode_32(&info_p);
3745 ceph_decode_need(&info_p, info_end, namelen, bad);
3746
3747 if (mds_namespace &&
3748 strlen(mds_namespace) == namelen &&
3749 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3750 mount_fscid = fscid;
3751 break;
3752 }
3753 }
3754
3755 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3756 if (mount_fscid != (u32)-1) {
3757 fsc->client->monc.fs_cluster_id = mount_fscid;
3758 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
3759 0, true);
3760 ceph_monc_renew_subs(&fsc->client->monc);
3761 } else {
3762 err = -ENOENT;
3763 goto err_out;
3764 }
3765 return;
3766bad:
3767 pr_err("error decoding fsmap\n");
3768err_out:
3769 mutex_lock(&mdsc->mutex);
3770 mdsc->mdsmap_err = -ENOENT;
3771 __wake_requests(mdsc, &mdsc->waiting_for_map);
3772 mutex_unlock(&mdsc->mutex);
3773 return;
3774}
2f2dc053
SW		 3775
3776/*
3777 * handle mds map update.
3778 */
430afbad 3779void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053
SW		 3780{
3781 u32 epoch;
3782 u32 maplen;
3783 void *p = msg->front.iov_base;
3784 void *end = p + msg->front.iov_len;
3785 struct ceph_mdsmap *newmap, *oldmap;
3786 struct ceph_fsid fsid;
3787 int err = -EINVAL;
3788
3789 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3790 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3791 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3792 return;
c89136ea
SW		 3793 epoch = ceph_decode_32(&p);
3794 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3795 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3796
3797 /* do we need it? */
2f2dc053
SW		 3798 mutex_lock(&mdsc->mutex);
3799 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3800 dout("handle_map epoch %u <= our %u\n",
3801 epoch, mdsc->mdsmap->m_epoch);
3802 mutex_unlock(&mdsc->mutex);
3803 return;
3804 }
3805
3806 newmap = ceph_mdsmap_decode(&p, end);
3807 if (IS_ERR(newmap)) {
3808 err = PTR_ERR(newmap);
3809 goto bad_unlock;
3810 }
3811
3812 /* swap into place */
3813 if (mdsc->mdsmap) {
3814 oldmap = mdsc->mdsmap;
3815 mdsc->mdsmap = newmap;
3816 check_new_map(mdsc, newmap, oldmap);
3817 ceph_mdsmap_destroy(oldmap);
3818 } else {
3819 mdsc->mdsmap = newmap; /* first mds map */
3820 }
3d14c5d2 3821 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3822
3823 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 3824 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3825 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 3826
3827 mutex_unlock(&mdsc->mutex);
3828 schedule_delayed(mdsc);
3829 return;
3830
3831bad_unlock:
3832 mutex_unlock(&mdsc->mutex);
3833bad:
3834 pr_err("error decoding mdsmap %d\n", err);
3835 return;
3836}
3837
3838static struct ceph_connection *con_get(struct ceph_connection *con)
3839{
3840 struct ceph_mds_session *s = con->private;
3841
3842 if (get_session(s)) {
2600d2dd 3843 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3844 return con;
3845 }
3846 dout("mdsc con_get %p FAIL\n", s);
3847 return NULL;
3848}
3849
3850static void con_put(struct ceph_connection *con)
3851{
3852 struct ceph_mds_session *s = con->private;
3853
7d8e18a6 3854 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
2f2dc053
SW		 3855 ceph_put_mds_session(s);
3856}
3857
3858/*
3859 * if the client is unresponsive for long enough, the mds will kill
3860 * the session entirely.
3861 */
3862static void peer_reset(struct ceph_connection *con)
3863{
3864 struct ceph_mds_session *s = con->private;
7e70f0ed 3865 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3866
f3ae1b97 3867 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3868 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3869}
3870
3871static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3872{
3873 struct ceph_mds_session *s = con->private;
3874 struct ceph_mds_client *mdsc = s->s_mdsc;
3875 int type = le16_to_cpu(msg->hdr.type);
3876
2600d2dd
SW		 3877 mutex_lock(&mdsc->mutex);
3878 if (__verify_registered_session(mdsc, s) < 0) {
3879 mutex_unlock(&mdsc->mutex);
3880 goto out;
3881 }
3882 mutex_unlock(&mdsc->mutex);
3883
2f2dc053
SW		 3884 switch (type) {
3885 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 3886 ceph_mdsc_handle_mdsmap(mdsc, msg);
3887 break;
3888 case CEPH_MSG_FS_MAP_USER:
3889 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 3890 break;
3891 case CEPH_MSG_CLIENT_SESSION:
3892 handle_session(s, msg);
3893 break;
3894 case CEPH_MSG_CLIENT_REPLY:
3895 handle_reply(s, msg);
3896 break;
3897 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3898 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3899 break;
3900 case CEPH_MSG_CLIENT_CAPS:
3901 ceph_handle_caps(s, msg);
3902 break;
3903 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3904 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3905 break;
3906 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3907 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3908 break;
3909
3910 default:
3911 pr_err("received unknown message type %d %s\n", type,
3912 ceph_msg_type_name(type));
3913 }
2600d2dd 3914out:
2f2dc053
SW		 3915 ceph_msg_put(msg);
3916}
3917
4e7a5dcd
SW		 3918/*
3919 * authentication
3920 */
a3530df3
AE		 3921
3922/*
3923 * Note: returned pointer is the address of a structure that's
3924 * managed separately. Caller must *not* attempt to free it.
3925 */
3926static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3927 int *proto, int force_new)
4e7a5dcd
SW		 3928{
3929 struct ceph_mds_session *s = con->private;
3930 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3931 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 3932 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 3933
74f1869f 3934 if (force_new && auth->authorizer) {
6c1ea260 3935 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 3936 auth->authorizer = NULL;
4e7a5dcd 3937 }
27859f97
SW		 3938 if (!auth->authorizer) {
3939 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3940 auth);
0bed9b5c
SW		 3941 if (ret)
3942 return ERR_PTR(ret);
27859f97
SW		 3943 } else {
3944 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3945 auth);
a255651d 3946 if (ret)
a3530df3 3947 return ERR_PTR(ret);
4e7a5dcd 3948 }
4e7a5dcd 3949 *proto = ac->protocol;
74f1869f 3950
a3530df3 3951 return auth;
4e7a5dcd
SW		 3952}
3953
3954
0dde5848 3955static int verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 3956{
3957 struct ceph_mds_session *s = con->private;
3958 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3959 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 3960
0dde5848 3961 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4e7a5dcd
SW		 3962}
3963
9bd2e6f8
SW		 3964static int invalidate_authorizer(struct ceph_connection *con)
3965{
3966 struct ceph_mds_session *s = con->private;
3967 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3968 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 3969
27859f97 3970 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 3971
3d14c5d2 3972 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3973}
3974
53ded495
AE		 3975static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3976 struct ceph_msg_header *hdr, int *skip)
3977{
3978 struct ceph_msg *msg;
3979 int type = (int) le16_to_cpu(hdr->type);
3980 int front_len = (int) le32_to_cpu(hdr->front_len);
3981
3982 if (con->in_msg)
3983 return con->in_msg;
3984
3985 *skip = 0;
3986 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3987 if (!msg) {
3988 pr_err("unable to allocate msg type %d len %d\n",
3989 type, front_len);
3990 return NULL;
3991 }
53ded495
AE		 3992
3993 return msg;
3994}
3995
79dbd1ba 3996static int mds_sign_message(struct ceph_msg *msg)
33d07337 3997{
79dbd1ba 3998 struct ceph_mds_session *s = msg->con->private;
33d07337 3999 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4000
33d07337
YZ		 4001 return ceph_auth_sign_message(auth, msg);
4002}
4003
79dbd1ba 4004static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 4005{
79dbd1ba 4006 struct ceph_mds_session *s = msg->con->private;
33d07337 4007 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4008
33d07337
YZ		 4009 return ceph_auth_check_message_signature(auth, msg);
4010}
4011
9e32789f 4012static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4013 .get = con_get,
4014 .put = con_put,
4015 .dispatch = dispatch,
4e7a5dcd
SW		 4016 .get_authorizer = get_authorizer,
4017 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4018 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4019 .peer_reset = peer_reset,
53ded495 4020 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4021 .sign_message = mds_sign_message,
4022 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4023};
4024
2f2dc053 4025/* eof */
Linux 6.x block layer and io_uring tree(s)