]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - fs/ceph/mds_client.c
projects / mirror_ubuntu-hirsute-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: handle CEPH_SESSION_REJECT message
[mirror_ubuntu-hirsute-kernel.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45 * are no longer valid.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);
1154 list_del(&cf->i_list);
1155 list_add(&cf->i_list, &to_remove);
553adfd9
YZ		 1156 }
1157
6c99f254 1158 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1159
e4500b5e
YZ		 1160 list_for_each_entry(cf, &to_remove, i_list)
1161 list_del(&cf->g_list);
8310b089 1162
6c99f254 1163 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1164 pr_warn_ratelimited(
1165 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1166 ceph_cap_string(ci->i_dirty_caps),
1167 inode, ceph_ino(inode));
1168 ci->i_dirty_caps = 0;
1169 list_del_init(&ci->i_dirty_item);
6c93df5d 1170 drop = true;
6c99f254
SW		 1171 }
1172 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1173 pr_warn_ratelimited(
1174 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1175 ceph_cap_string(ci->i_flushing_caps),
1176 inode, ceph_ino(inode));
1177 ci->i_flushing_caps = 0;
1178 list_del_init(&ci->i_flushing_item);
1179 mdsc->num_cap_flushing--;
6c93df5d 1180 drop = true;
6c99f254 1181 }
6c99f254 1182 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1183
f66fd9f0 1184 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1185 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1186 ci->i_prealloc_cap_flush = NULL;
1187 }
6c99f254 1188 }
be655596 1189 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1190 while (!list_empty(&to_remove)) {
1191 struct ceph_cap_flush *cf;
1192 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1193 struct ceph_cap_flush, i_list);
1194 list_del(&cf->i_list);
f66fd9f0 1195 ceph_free_cap_flush(cf);
553adfd9 1196 }
77310320
YZ		 1197
1198 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1199 if (invalidate)
1200 ceph_queue_invalidate(inode);
77310320 1201 if (drop)
6c99f254 1202 iput(inode);
2f2dc053
SW		 1203 return 0;
1204}
1205
1206/*
1207 * caller must hold session s_mutex
1208 */
1209static void remove_session_caps(struct ceph_mds_session *session)
1210{
6c93df5d
YZ		 1211 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1212 struct super_block *sb = fsc->sb;
2f2dc053 1213 dout("remove_session_caps on %p\n", session);
6c93df5d 1214 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1215
c8799fc4
YZ		 1216 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1217
6f60f889
YZ		 1218 spin_lock(&session->s_cap_lock);
1219 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1220 struct inode *inode;
1221 struct ceph_cap *cap, *prev = NULL;
1222 struct ceph_vino vino;
1223 /*
1224 * iterate_session_caps() skips inodes that are being
1225 * deleted, we need to wait until deletions are complete.
1226 * __wait_on_freeing_inode() is designed for the job,
1227 * but it is not exported, so use lookup inode function
1228 * to access it.
1229 */
1230 while (!list_empty(&session->s_caps)) {
1231 cap = list_entry(session->s_caps.next,
1232 struct ceph_cap, session_caps);
1233 if (cap == prev)
1234 break;
1235 prev = cap;
1236 vino = cap->ci->i_vino;
1237 spin_unlock(&session->s_cap_lock);
1238
ed284c49 1239 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1240 iput(inode);
1241
1242 spin_lock(&session->s_cap_lock);
1243 }
1244 }
745a8e3b
YZ		 1245
1246 // drop cap expires and unlock s_cap_lock
1247 cleanup_cap_releases(session->s_mdsc, session);
6f60f889 1248
2f2dc053 1249 BUG_ON(session->s_nr_caps > 0);
6c99f254 1250 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1251}
1252
1253/*
1254 * wake up any threads waiting on this session's caps. if the cap is
1255 * old (didn't get renewed on the client reconnect), remove it now.
1256 *
1257 * caller must hold s_mutex.
1258 */
1259static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1260 void *arg)
1261{
0dc2570f
SW		 1262 struct ceph_inode_info *ci = ceph_inode(inode);
1263
0dc2570f 1264 if (arg) {
be655596 1265 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1266 ci->i_wanted_max_size = 0;
1267 ci->i_requested_max_size = 0;
be655596 1268 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1269 }
e5360309 1270 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1271 return 0;
1272}
1273
0dc2570f
SW		 1274static void wake_up_session_caps(struct ceph_mds_session *session,
1275 int reconnect)
2f2dc053
SW		 1276{
1277 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1278 iterate_session_caps(session, wake_up_session_cb,
1279 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1280}
1281
1282/*
1283 * Send periodic message to MDS renewing all currently held caps. The
1284 * ack will reset the expiration for all caps from this session.
1285 *
1286 * caller holds s_mutex
1287 */
1288static int send_renew_caps(struct ceph_mds_client *mdsc,
1289 struct ceph_mds_session *session)
1290{
1291 struct ceph_msg *msg;
1292 int state;
1293
1294 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1295 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1296 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1297 session->s_renew_requested = jiffies;
2f2dc053
SW		 1298
1299 /* do not try to renew caps until a recovering mds has reconnected
1300 * with its clients. */
1301 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1302 if (state < CEPH_MDS_STATE_RECONNECT) {
1303 dout("send_renew_caps ignoring mds%d (%s)\n",
1304 session->s_mds, ceph_mds_state_name(state));
1305 return 0;
1306 }
1307
1308 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1309 ceph_mds_state_name(state));
2f2dc053
SW		 1310 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1311 ++session->s_renew_seq);
a79832f2
SW		 1312 if (!msg)
1313 return -ENOMEM;
2f2dc053
SW		 1314 ceph_con_send(&session->s_con, msg);
1315 return 0;
1316}
1317
186e4f7a
YZ		 1318static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1319 struct ceph_mds_session *session, u64 seq)
1320{
1321 struct ceph_msg *msg;
1322
1323 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1324 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1325 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1326 if (!msg)
1327 return -ENOMEM;
1328 ceph_con_send(&session->s_con, msg);
1329 return 0;
1330}
1331
1332
2f2dc053
SW		 1333/*
1334 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1335 *
1336 * Called under session->s_mutex
2f2dc053
SW		 1337 */
1338static void renewed_caps(struct ceph_mds_client *mdsc,
1339 struct ceph_mds_session *session, int is_renew)
1340{
1341 int was_stale;
1342 int wake = 0;
1343
1344 spin_lock(&session->s_cap_lock);
1ce208a6 1345 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1346
1347 session->s_cap_ttl = session->s_renew_requested +
1348 mdsc->mdsmap->m_session_timeout*HZ;
1349
1350 if (was_stale) {
1351 if (time_before(jiffies, session->s_cap_ttl)) {
1352 pr_info("mds%d caps renewed\n", session->s_mds);
1353 wake = 1;
1354 } else {
1355 pr_info("mds%d caps still stale\n", session->s_mds);
1356 }
1357 }
1358 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1359 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1360 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1361 spin_unlock(&session->s_cap_lock);
1362
1363 if (wake)
0dc2570f 1364 wake_up_session_caps(session, 0);
2f2dc053
SW		 1365}
1366
1367/*
1368 * send a session close request
1369 */
1370static int request_close_session(struct ceph_mds_client *mdsc,
1371 struct ceph_mds_session *session)
1372{
1373 struct ceph_msg *msg;
2f2dc053
SW		 1374
1375 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1376 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1377 session->s_seq);
1378 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1379 if (!msg)
1380 return -ENOMEM;
1381 ceph_con_send(&session->s_con, msg);
fcff415c 1382 return 1;
2f2dc053
SW		 1383}
1384
1385/*
1386 * Called with s_mutex held.
1387 */
1388static int __close_session(struct ceph_mds_client *mdsc,
1389 struct ceph_mds_session *session)
1390{
1391 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1392 return 0;
1393 session->s_state = CEPH_MDS_SESSION_CLOSING;
1394 return request_close_session(mdsc, session);
1395}
1396
1397/*
1398 * Trim old(er) caps.
1399 *
1400 * Because we can't cache an inode without one or more caps, we do
1401 * this indirectly: if a cap is unused, we prune its aliases, at which
1402 * point the inode will hopefully get dropped to.
1403 *
1404 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1405 * memory pressure from the MDS, though, so it needn't be perfect.
1406 */
1407static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1408{
1409 struct ceph_mds_session *session = arg;
1410 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1411 int used, wanted, oissued, mine;
2f2dc053
SW		 1412
1413 if (session->s_trim_caps <= 0)
1414 return -1;
1415
be655596 1416 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1417 mine = cap->issued | cap->implemented;
1418 used = __ceph_caps_used(ci);
979abfdd 1419 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1420 oissued = __ceph_caps_issued_other(ci, cap);
1421
979abfdd 1422 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1423 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1424 ceph_cap_string(used), ceph_cap_string(wanted));
1425 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1426 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1427 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1428 goto out;
1429 if ((used | wanted) & CEPH_CAP_ANY_WR)
1430 goto out;
1431 }
5e804ac4
YZ		 1432 /* The inode has cached pages, but it's no longer used.
1433 * we can safely drop it */
1434 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1435 !(oissued & CEPH_CAP_FILE_CACHE)) {
1436 used = 0;
1437 oissued = 0;
1438 }
979abfdd 1439 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1440 goto out; /* we need these caps */
1441
1442 session->s_trim_caps--;
1443 if (oissued) {
1444 /* we aren't the only cap.. just remove us */
a096b09a 1445 __ceph_remove_cap(cap, true);
2f2dc053 1446 } else {
5e804ac4 1447 /* try dropping referring dentries */
be655596 1448 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1449 d_prune_aliases(inode);
1450 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1451 inode, cap, atomic_read(&inode->i_count));
1452 return 0;
1453 }
1454
1455out:
be655596 1456 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1457 return 0;
1458}
1459
1460/*
1461 * Trim session cap count down to some max number.
1462 */
1463static int trim_caps(struct ceph_mds_client *mdsc,
1464 struct ceph_mds_session *session,
1465 int max_caps)
1466{
1467 int trim_caps = session->s_nr_caps - max_caps;
1468
1469 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1470 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1471 if (trim_caps > 0) {
1472 session->s_trim_caps = trim_caps;
1473 iterate_session_caps(session, trim_caps_cb, session);
1474 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1475 session->s_mds, session->s_nr_caps, max_caps,
1476 trim_caps - session->s_trim_caps);
5dacf091 1477 session->s_trim_caps = 0;
2f2dc053 1478 }
a56371d9 1479
a56371d9 1480 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1481 return 0;
1482}
1483
8310b089
YZ		 1484static int check_caps_flush(struct ceph_mds_client *mdsc,
1485 u64 want_flush_tid)
1486{
8310b089
YZ		 1487 int ret = 1;
1488
1489 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1490 if (!list_empty(&mdsc->cap_flush_list)) {
1491 struct ceph_cap_flush *cf =
1492 list_first_entry(&mdsc->cap_flush_list,
1493 struct ceph_cap_flush, g_list);
1494 if (cf->tid <= want_flush_tid) {
1495 dout("check_caps_flush still flushing tid "
1496 "%llu <= %llu\n", cf->tid, want_flush_tid);
1497 ret = 0;
1498 }
8310b089
YZ		 1499 }
1500 spin_unlock(&mdsc->cap_dirty_lock);
1501 return ret;
d3383a8e
YZ		 1502}
1503
2f2dc053
SW		 1504/*
1505 * flush all dirty inode data to disk.
1506 *
8310b089 1507 * returns true if we've flushed through want_flush_tid
2f2dc053 1508 */
affbc19a 1509static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1510 u64 want_flush_tid)
2f2dc053 1511{
0e294387 1512 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1513
1514 wait_event(mdsc->cap_flushing_wq,
1515 check_caps_flush(mdsc, want_flush_tid));
1516
1517 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1518}
1519
1520/*
1521 * called under s_mutex
1522 */
3d7ded4d
SW		 1523void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1524 struct ceph_mds_session *session)
2f2dc053 1525{
745a8e3b
YZ		 1526 struct ceph_msg *msg = NULL;
1527 struct ceph_mds_cap_release *head;
1528 struct ceph_mds_cap_item *item;
1529 struct ceph_cap *cap;
1530 LIST_HEAD(tmp_list);
1531 int num_cap_releases;
2f2dc053 1532
0f8605f2 1533 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1534again:
1535 list_splice_init(&session->s_cap_releases, &tmp_list);
1536 num_cap_releases = session->s_num_cap_releases;
1537 session->s_num_cap_releases = 0;
2f2dc053 1538 spin_unlock(&session->s_cap_lock);
e01a5946 1539
745a8e3b
YZ		 1540 while (!list_empty(&tmp_list)) {
1541 if (!msg) {
1542 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1543 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1544 if (!msg)
1545 goto out_err;
1546 head = msg->front.iov_base;
1547 head->num = cpu_to_le32(0);
1548 msg->front.iov_len = sizeof(*head);
1549 }
1550 cap = list_first_entry(&tmp_list, struct ceph_cap,
1551 session_caps);
1552 list_del(&cap->session_caps);
1553 num_cap_releases--;
e01a5946 1554
00bd8edb 1555 head = msg->front.iov_base;
745a8e3b
YZ		 1556 le32_add_cpu(&head->num, 1);
1557 item = msg->front.iov_base + msg->front.iov_len;
1558 item->ino = cpu_to_le64(cap->cap_ino);
1559 item->cap_id = cpu_to_le64(cap->cap_id);
1560 item->migrate_seq = cpu_to_le32(cap->mseq);
1561 item->seq = cpu_to_le32(cap->issue_seq);
1562 msg->front.iov_len += sizeof(*item);
1563
1564 ceph_put_cap(mdsc, cap);
1565
1566 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1567 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1568 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1569 ceph_con_send(&session->s_con, msg);
1570 msg = NULL;
1571 }
00bd8edb 1572 }
e01a5946 1573
745a8e3b 1574 BUG_ON(num_cap_releases != 0);
e01a5946 1575
745a8e3b
YZ		 1576 spin_lock(&session->s_cap_lock);
1577 if (!list_empty(&session->s_cap_releases))
1578 goto again;
1579 spin_unlock(&session->s_cap_lock);
1580
1581 if (msg) {
1582 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1583 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1584 ceph_con_send(&session->s_con, msg);
e01a5946 1585 }
745a8e3b
YZ		 1586 return;
1587out_err:
1588 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1589 session->s_mds);
1590 spin_lock(&session->s_cap_lock);
1591 list_splice(&tmp_list, &session->s_cap_releases);
1592 session->s_num_cap_releases += num_cap_releases;
1593 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1594}
1595
2f2dc053
SW		 1596/*
1597 * requests
1598 */
1599
54008399
YZ		 1600int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1601 struct inode *dir)
1602{
1603 struct ceph_inode_info *ci = ceph_inode(dir);
1604 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1605 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1606 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1607 int order, num_entries;
1608
1609 spin_lock(&ci->i_ceph_lock);
1610 num_entries = ci->i_files + ci->i_subdirs;
1611 spin_unlock(&ci->i_ceph_lock);
1612 num_entries = max(num_entries, 1);
1613 num_entries = min(num_entries, opt->max_readdir);
1614
1615 order = get_order(size * num_entries);
1616 while (order >= 0) {
2a5beea3
YZ		 1617 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1618 __GFP_NOWARN,
1619 order);
1620 if (rinfo->dir_entries)
54008399
YZ		 1621 break;
1622 order--;
1623 }
2a5beea3 1624 if (!rinfo->dir_entries)
54008399
YZ		 1625 return -ENOMEM;
1626
1627 num_entries = (PAGE_SIZE << order) / size;
1628 num_entries = min(num_entries, opt->max_readdir);
1629
1630 rinfo->dir_buf_size = PAGE_SIZE << order;
1631 req->r_num_caps = num_entries + 1;
1632 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1633 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1634 return 0;
1635}
1636
2f2dc053
SW		 1637/*
1638 * Create an mds request.
1639 */
1640struct ceph_mds_request *
1641ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1642{
1643 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1644
1645 if (!req)
1646 return ERR_PTR(-ENOMEM);
1647
b4556396 1648 mutex_init(&req->r_fill_mutex);
37151668 1649 req->r_mdsc = mdsc;
2f2dc053
SW		 1650 req->r_started = jiffies;
1651 req->r_resend_mds = -1;
1652 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1653 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1654 req->r_fmode = -1;
153c8e6b 1655 kref_init(&req->r_kref);
fcd00b68 1656 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1657 INIT_LIST_HEAD(&req->r_wait);
1658 init_completion(&req->r_completion);
1659 init_completion(&req->r_safe_completion);
1660 INIT_LIST_HEAD(&req->r_unsafe_item);
1661
8bbd4714 1662 req->r_stamp = current_fs_time(mdsc->fsc->sb);
b8e69066 1663
2f2dc053
SW		 1664 req->r_op = op;
1665 req->r_direct_mode = mode;
1666 return req;
1667}
1668
1669/*
44ca18f2 1670 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1671 *
1672 * called under mdsc->mutex.
1673 */
44ca18f2
SW		 1674static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1675{
1676 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1677 return NULL;
1678 return rb_entry(rb_first(&mdsc->request_tree),
1679 struct ceph_mds_request, r_node);
1680}
1681
e8a7b8b1 1682static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1683{
e8a7b8b1 1684 return mdsc->oldest_tid;
2f2dc053
SW		 1685}
1686
1687/*
1688 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1689 * on build_path_from_dentry in fs/cifs/dir.c.
1690 *
1691 * If @stop_on_nosnap, generate path relative to the first non-snapped
1692 * inode.
1693 *
1694 * Encode hidden .snap dirs as a double /, i.e.
1695 * foo/.snap/bar -> foo//bar
1696 */
1697char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1698 int stop_on_nosnap)
1699{
1700 struct dentry *temp;
1701 char *path;
1702 int len, pos;
1b71fe2e 1703 unsigned seq;
2f2dc053
SW		 1704
1705 if (dentry == NULL)
1706 return ERR_PTR(-EINVAL);
1707
1708retry:
1709 len = 0;
1b71fe2e
AV		 1710 seq = read_seqbegin(&rename_lock);
1711 rcu_read_lock();
2f2dc053 1712 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1713 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1714 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1715 len++; /* slash only */
1716 else if (stop_on_nosnap && inode &&
1717 ceph_snap(inode) == CEPH_NOSNAP)
1718 break;
1719 else
1720 len += 1 + temp->d_name.len;
1721 temp = temp->d_parent;
2f2dc053 1722 }
1b71fe2e 1723 rcu_read_unlock();
2f2dc053
SW		 1724 if (len)
1725 len--; /* no leading '/' */
1726
1727 path = kmalloc(len+1, GFP_NOFS);
1728 if (path == NULL)
1729 return ERR_PTR(-ENOMEM);
1730 pos = len;
1731 path[pos] = 0; /* trailing null */
1b71fe2e 1732 rcu_read_lock();
2f2dc053 1733 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1734 struct inode *inode;
2f2dc053 1735
1b71fe2e 1736 spin_lock(&temp->d_lock);
2b0143b5 1737 inode = d_inode(temp);
2f2dc053 1738 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1739 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1740 pos, temp);
1741 } else if (stop_on_nosnap && inode &&
1742 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1743 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1744 break;
1745 } else {
1746 pos -= temp->d_name.len;
1b71fe2e
AV		 1747 if (pos < 0) {
1748 spin_unlock(&temp->d_lock);
2f2dc053 1749 break;
1b71fe2e 1750 }
2f2dc053
SW		 1751 strncpy(path + pos, temp->d_name.name,
1752 temp->d_name.len);
2f2dc053 1753 }
1b71fe2e 1754 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1755 if (pos)
1756 path[--pos] = '/';
1757 temp = temp->d_parent;
2f2dc053 1758 }
1b71fe2e
AV		 1759 rcu_read_unlock();
1760 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1761 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1762 "expected, namelen is %d, pos is %d\n", len, pos);
1763 /* presumably this is only possible if racing with a
1764 rename of one of the parent directories (we can not
1765 lock the dentries above us to prevent this, but
1766 retrying should be harmless) */
1767 kfree(path);
1768 goto retry;
1769 }
1770
2b0143b5 1771 *base = ceph_ino(d_inode(temp));
2f2dc053 1772 *plen = len;
104648ad 1773 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1774 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1775 return path;
1776}
1777
1778static int build_dentry_path(struct dentry *dentry,
1779 const char **ppath, int *ppathlen, u64 *pino,
1780 int *pfreepath)
1781{
1782 char *path;
1783
2b0143b5
DH		 1784 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1785 *pino = ceph_ino(d_inode(dentry->d_parent));
2f2dc053
SW		 1786 *ppath = dentry->d_name.name;
1787 *ppathlen = dentry->d_name.len;
1788 return 0;
1789 }
1790 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1791 if (IS_ERR(path))
1792 return PTR_ERR(path);
1793 *ppath = path;
1794 *pfreepath = 1;
1795 return 0;
1796}
1797
1798static int build_inode_path(struct inode *inode,
1799 const char **ppath, int *ppathlen, u64 *pino,
1800 int *pfreepath)
1801{
1802 struct dentry *dentry;
1803 char *path;
1804
1805 if (ceph_snap(inode) == CEPH_NOSNAP) {
1806 *pino = ceph_ino(inode);
1807 *ppathlen = 0;
1808 return 0;
1809 }
1810 dentry = d_find_alias(inode);
1811 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1812 dput(dentry);
1813 if (IS_ERR(path))
1814 return PTR_ERR(path);
1815 *ppath = path;
1816 *pfreepath = 1;
1817 return 0;
1818}
1819
1820/*
1821 * request arguments may be specified via an inode *, a dentry *, or
1822 * an explicit ino+path.
1823 */
1824static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1825 const char *rpath, u64 rino,
1826 const char **ppath, int *pathlen,
1827 u64 *ino, int *freepath)
1828{
1829 int r = 0;
1830
1831 if (rinode) {
1832 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1833 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1834 ceph_snap(rinode));
1835 } else if (rdentry) {
1836 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1837 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1838 *ppath);
795858db 1839 } else if (rpath || rino) {
2f2dc053
SW		 1840 *ino = rino;
1841 *ppath = rpath;
b000056a 1842 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1843 dout(" path %.*s\n", *pathlen, rpath);
1844 }
1845
1846 return r;
1847}
1848
1849/*
1850 * called under mdsc->mutex
1851 */
1852static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1853 struct ceph_mds_request *req,
6e6f0923 1854 int mds, bool drop_cap_releases)
2f2dc053
SW		 1855{
1856 struct ceph_msg *msg;
1857 struct ceph_mds_request_head *head;
1858 const char *path1 = NULL;
1859 const char *path2 = NULL;
1860 u64 ino1 = 0, ino2 = 0;
1861 int pathlen1 = 0, pathlen2 = 0;
1862 int freepath1 = 0, freepath2 = 0;
1863 int len;
1864 u16 releases;
1865 void *p, *end;
1866 int ret;
1867
1868 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1869 req->r_path1, req->r_ino1.ino,
1870 &path1, &pathlen1, &ino1, &freepath1);
1871 if (ret < 0) {
1872 msg = ERR_PTR(ret);
1873 goto out;
1874 }
1875
1876 ret = set_request_path_attr(NULL, req->r_old_dentry,
1877 req->r_path2, req->r_ino2.ino,
1878 &path2, &pathlen2, &ino2, &freepath2);
1879 if (ret < 0) {
1880 msg = ERR_PTR(ret);
1881 goto out_free1;
1882 }
1883
1884 len = sizeof(*head) +
b8e69066 1885 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 1886 sizeof(struct ceph_timespec);
2f2dc053
SW		 1887
1888 /* calculate (max) length for cap releases */
1889 len += sizeof(struct ceph_mds_request_release) *
1890 (!!req->r_inode_drop + !!req->r_dentry_drop +
1891 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1892 if (req->r_dentry_drop)
1893 len += req->r_dentry->d_name.len;
1894 if (req->r_old_dentry_drop)
1895 len += req->r_old_dentry->d_name.len;
1896
b61c2763 1897 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1898 if (!msg) {
1899 msg = ERR_PTR(-ENOMEM);
2f2dc053 1900 goto out_free2;
a79832f2 1901 }
2f2dc053 1902
7cfa0313 1903 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1904 msg->hdr.tid = cpu_to_le64(req->r_tid);
1905
2f2dc053
SW		 1906 head = msg->front.iov_base;
1907 p = msg->front.iov_base + sizeof(*head);
1908 end = msg->front.iov_base + msg->front.iov_len;
1909
1910 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1911 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1912 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1913 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1914 head->args = req->r_args;
1915
1916 ceph_encode_filepath(&p, end, ino1, path1);
1917 ceph_encode_filepath(&p, end, ino2, path2);
1918
e979cf50
SW		 1919 /* make note of release offset, in case we need to replay */
1920 req->r_request_release_offset = p - msg->front.iov_base;
1921
2f2dc053
SW		 1922 /* cap releases */
1923 releases = 0;
1924 if (req->r_inode_drop)
1925 releases += ceph_encode_inode_release(&p,
2b0143b5 1926 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 1927 mds, req->r_inode_drop, req->r_inode_unless, 0);
1928 if (req->r_dentry_drop)
1929 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1930 mds, req->r_dentry_drop, req->r_dentry_unless);
1931 if (req->r_old_dentry_drop)
1932 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1933 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1934 if (req->r_old_inode_drop)
1935 releases += ceph_encode_inode_release(&p,
2b0143b5 1936 d_inode(req->r_old_dentry),
2f2dc053 1937 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 1938
1939 if (drop_cap_releases) {
1940 releases = 0;
1941 p = msg->front.iov_base + req->r_request_release_offset;
1942 }
1943
2f2dc053
SW		 1944 head->num_releases = cpu_to_le16(releases);
1945
b8e69066 1946 /* time stamp */
1f041a89
YZ		 1947 {
1948 struct ceph_timespec ts;
1949 ceph_encode_timespec(&ts, &req->r_stamp);
1950 ceph_encode_copy(&p, &ts, sizeof(ts));
1951 }
b8e69066 1952
2f2dc053
SW		 1953 BUG_ON(p > end);
1954 msg->front.iov_len = p - msg->front.iov_base;
1955 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1956
25e6bae3
YZ		 1957 if (req->r_pagelist) {
1958 struct ceph_pagelist *pagelist = req->r_pagelist;
1959 atomic_inc(&pagelist->refcnt);
1960 ceph_msg_data_add_pagelist(msg, pagelist);
1961 msg->hdr.data_len = cpu_to_le32(pagelist->length);
1962 } else {
1963 msg->hdr.data_len = 0;
ebf18f47 1964 }
02afca6c 1965
2f2dc053
SW		 1966 msg->hdr.data_off = cpu_to_le16(0);
1967
1968out_free2:
1969 if (freepath2)
1970 kfree((char *)path2);
1971out_free1:
1972 if (freepath1)
1973 kfree((char *)path1);
1974out:
1975 return msg;
1976}
1977
1978/*
1979 * called under mdsc->mutex if error, under no mutex if
1980 * success.
1981 */
1982static void complete_request(struct ceph_mds_client *mdsc,
1983 struct ceph_mds_request *req)
1984{
1985 if (req->r_callback)
1986 req->r_callback(mdsc, req);
1987 else
03066f23 1988 complete_all(&req->r_completion);
2f2dc053
SW		 1989}
1990
1991/*
1992 * called under mdsc->mutex
1993 */
1994static int __prepare_send_request(struct ceph_mds_client *mdsc,
1995 struct ceph_mds_request *req,
6e6f0923 1996 int mds, bool drop_cap_releases)
2f2dc053
SW		 1997{
1998 struct ceph_mds_request_head *rhead;
1999 struct ceph_msg *msg;
2000 int flags = 0;
2001
2f2dc053 2002 req->r_attempts++;
e55b71f8
GF		 2003 if (req->r_inode) {
2004 struct ceph_cap *cap =
2005 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2006
2007 if (cap)
2008 req->r_sent_on_mseq = cap->mseq;
2009 else
2010 req->r_sent_on_mseq = -1;
2011 }
2f2dc053
SW		 2012 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2013 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2014
01a92f17 2015 if (req->r_got_unsafe) {
c5c9a0bf 2016 void *p;
01a92f17
SW		 2017 /*
2018 * Replay. Do not regenerate message (and rebuild
2019 * paths, etc.); just use the original message.
2020 * Rebuilding paths will break for renames because
2021 * d_move mangles the src name.
2022 */
2023 msg = req->r_request;
2024 rhead = msg->front.iov_base;
2025
2026 flags = le32_to_cpu(rhead->flags);
2027 flags |= CEPH_MDS_FLAG_REPLAY;
2028 rhead->flags = cpu_to_le32(flags);
2029
2030 if (req->r_target_inode)
2031 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2032
2033 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2034
2035 /* remove cap/dentry releases from message */
2036 rhead->num_releases = 0;
c5c9a0bf
YZ		 2037
2038 /* time stamp */
2039 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2040 {
2041 struct ceph_timespec ts;
2042 ceph_encode_timespec(&ts, &req->r_stamp);
2043 ceph_encode_copy(&p, &ts, sizeof(ts));
2044 }
c5c9a0bf
YZ		 2045
2046 msg->front.iov_len = p - msg->front.iov_base;
2047 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2048 return 0;
2049 }
2050
2f2dc053
SW		 2051 if (req->r_request) {
2052 ceph_msg_put(req->r_request);
2053 req->r_request = NULL;
2054 }
6e6f0923 2055 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2056 if (IS_ERR(msg)) {
e1518c7c 2057 req->r_err = PTR_ERR(msg);
a79832f2 2058 return PTR_ERR(msg);
2f2dc053
SW		 2059 }
2060 req->r_request = msg;
2061
2062 rhead = msg->front.iov_base;
2f2dc053
SW		 2063 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2064 if (req->r_got_unsafe)
2065 flags |= CEPH_MDS_FLAG_REPLAY;
2066 if (req->r_locked_dir)
2067 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2068 rhead->flags = cpu_to_le32(flags);
2069 rhead->num_fwd = req->r_num_fwd;
2070 rhead->num_retry = req->r_attempts - 1;
01a92f17 2071 rhead->ino = 0;
2f2dc053
SW		 2072
2073 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 2074 return 0;
2075}
2076
2077/*
2078 * send request, or put it on the appropriate wait list.
2079 */
2080static int __do_request(struct ceph_mds_client *mdsc,
2081 struct ceph_mds_request *req)
2082{
2083 struct ceph_mds_session *session = NULL;
2084 int mds = -1;
48fec5d0 2085 int err = 0;
2f2dc053 2086
eb1b8af3
YZ		 2087 if (req->r_err || req->r_got_result) {
2088 if (req->r_aborted)
2089 __unregister_request(mdsc, req);
2f2dc053 2090 goto out;
eb1b8af3 2091 }
2f2dc053
SW		 2092
2093 if (req->r_timeout &&
2094 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2095 dout("do_request timed out\n");
2096 err = -EIO;
2097 goto finish;
2098 }
48fec5d0
YZ		 2099 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2100 dout("do_request forced umount\n");
2101 err = -EIO;
2102 goto finish;
2103 }
2f2dc053 2104
dc69e2e9
SW		 2105 put_request_session(req);
2106
2f2dc053
SW		 2107 mds = __choose_mds(mdsc, req);
2108 if (mds < 0 ||
2109 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
430afbad
YZ		 2110 if (mdsc->mdsmap_err) {
2111 err = mdsc->mdsmap_err;
2112 dout("do_request mdsmap err %d\n", err);
2113 goto finish;
2114 }
2f2dc053
SW		 2115 dout("do_request no mds or not active, waiting for map\n");
2116 list_add(&req->r_wait, &mdsc->waiting_for_map);
2117 goto out;
2118 }
2119
2120 /* get, open session */
2121 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2122 if (!session) {
2f2dc053 2123 session = register_session(mdsc, mds);
9c423956
SW		 2124 if (IS_ERR(session)) {
2125 err = PTR_ERR(session);
2126 goto finish;
2127 }
2128 }
dc69e2e9
SW		 2129 req->r_session = get_session(session);
2130
2f2dc053 2131 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2132 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2133 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2134 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2135 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2136 err = -EACCES;
2137 goto out_session;
2138 }
2f2dc053
SW		 2139 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2140 session->s_state == CEPH_MDS_SESSION_CLOSING)
2141 __open_session(mdsc, session);
2142 list_add(&req->r_wait, &session->s_waiting);
2143 goto out_session;
2144 }
2145
2146 /* send request */
2f2dc053
SW		 2147 req->r_resend_mds = -1; /* forget any previous mds hint */
2148
2149 if (req->r_request_started == 0) /* note request start time */
2150 req->r_request_started = jiffies;
2151
6e6f0923 2152 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2153 if (!err) {
2154 ceph_msg_get(req->r_request);
2155 ceph_con_send(&session->s_con, req->r_request);
2156 }
2157
2158out_session:
2159 ceph_put_mds_session(session);
48fec5d0
YZ		 2160finish:
2161 if (err) {
2162 dout("__do_request early error %d\n", err);
2163 req->r_err = err;
2164 complete_request(mdsc, req);
2165 __unregister_request(mdsc, req);
2166 }
2f2dc053
SW		 2167out:
2168 return err;
2f2dc053
SW		 2169}
2170
2171/*
2172 * called under mdsc->mutex
2173 */
2174static void __wake_requests(struct ceph_mds_client *mdsc,
2175 struct list_head *head)
2176{
ed75ec2c
YZ		 2177 struct ceph_mds_request *req;
2178 LIST_HEAD(tmp_list);
2179
2180 list_splice_init(head, &tmp_list);
2f2dc053 2181
ed75ec2c
YZ		 2182 while (!list_empty(&tmp_list)) {
2183 req = list_entry(tmp_list.next,
2184 struct ceph_mds_request, r_wait);
2f2dc053 2185 list_del_init(&req->r_wait);
7971bd92 2186 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2187 __do_request(mdsc, req);
2188 }
2189}
2190
2191/*
2192 * Wake up threads with requests pending for @mds, so that they can
29790f26 2193 * resubmit their requests to a possibly different mds.
2f2dc053 2194 */
29790f26 2195static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2196{
44ca18f2 2197 struct ceph_mds_request *req;
282c1052 2198 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2199
2200 dout("kick_requests mds%d\n", mds);
282c1052 2201 while (p) {
44ca18f2 2202 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2203 p = rb_next(p);
44ca18f2
SW		 2204 if (req->r_got_unsafe)
2205 continue;
3de22be6
YZ		 2206 if (req->r_attempts > 0)
2207 continue; /* only new requests */
44ca18f2
SW		 2208 if (req->r_session &&
2209 req->r_session->s_mds == mds) {
2210 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2211 list_del_init(&req->r_wait);
44ca18f2 2212 __do_request(mdsc, req);
2f2dc053
SW		 2213 }
2214 }
2215}
2216
2217void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2218 struct ceph_mds_request *req)
2219{
2220 dout("submit_request on %p\n", req);
2221 mutex_lock(&mdsc->mutex);
2222 __register_request(mdsc, req, NULL);
2223 __do_request(mdsc, req);
2224 mutex_unlock(&mdsc->mutex);
2225}
2226
2227/*
2228 * Synchrously perform an mds request. Take care of all of the
2229 * session setup, forwarding, retry details.
2230 */
2231int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2232 struct inode *dir,
2233 struct ceph_mds_request *req)
2234{
2235 int err;
2236
2237 dout("do_request on %p\n", req);
2238
2239 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2240 if (req->r_inode)
2241 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2242 if (req->r_locked_dir)
2243 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
844d87c3 2244 if (req->r_old_dentry_dir)
41b02e1f
SW		 2245 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2246 CEPH_CAP_PIN);
2f2dc053
SW		 2247
2248 /* issue */
2249 mutex_lock(&mdsc->mutex);
2250 __register_request(mdsc, req, dir);
2251 __do_request(mdsc, req);
2252
e1518c7c
SW		 2253 if (req->r_err) {
2254 err = req->r_err;
e1518c7c 2255 goto out;
2f2dc053
SW		 2256 }
2257
e1518c7c
SW		 2258 /* wait */
2259 mutex_unlock(&mdsc->mutex);
2260 dout("do_request waiting\n");
5be73034 2261 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2262 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2263 } else {
5be73034
ID		 2264 long timeleft = wait_for_completion_killable_timeout(
2265 &req->r_completion,
2266 ceph_timeout_jiffies(req->r_timeout));
2267 if (timeleft > 0)
2268 err = 0;
2269 else if (!timeleft)
2270 err = -EIO; /* timed out */
2271 else
2272 err = timeleft; /* killed */
e1518c7c
SW		 2273 }
2274 dout("do_request waited, got %d\n", err);
2275 mutex_lock(&mdsc->mutex);
5b1daecd 2276
e1518c7c
SW		 2277 /* only abort if we didn't race with a real reply */
2278 if (req->r_got_result) {
2279 err = le32_to_cpu(req->r_reply_info.head->result);
2280 } else if (err < 0) {
2281 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2282
2283 /*
2284 * ensure we aren't running concurrently with
2285 * ceph_fill_trace or ceph_readdir_prepopulate, which
2286 * rely on locks (dir mutex) held by our caller.
2287 */
2288 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 2289 req->r_err = err;
2290 req->r_aborted = true;
b4556396 2291 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2292
e1518c7c 2293 if (req->r_locked_dir &&
167c9e35
SW		 2294 (req->r_op & CEPH_MDS_OP_WRITE))
2295 ceph_invalidate_dir_request(req);
2f2dc053 2296 } else {
e1518c7c 2297 err = req->r_err;
2f2dc053 2298 }
2f2dc053 2299
e1518c7c
SW		 2300out:
2301 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2302 dout("do_request %p done, result %d\n", req, err);
2303 return err;
2304}
2305
167c9e35 2306/*
2f276c51 2307 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2308 * namespace request.
2309 */
2310void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2311{
2312 struct inode *inode = req->r_locked_dir;
167c9e35 2313
2f276c51 2314 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2315
2f276c51 2316 ceph_dir_clear_complete(inode);
167c9e35
SW		 2317 if (req->r_dentry)
2318 ceph_invalidate_dentry_lease(req->r_dentry);
2319 if (req->r_old_dentry)
2320 ceph_invalidate_dentry_lease(req->r_old_dentry);
2321}
2322
2f2dc053
SW		 2323/*
2324 * Handle mds reply.
2325 *
2326 * We take the session mutex and parse and process the reply immediately.
2327 * This preserves the logical ordering of replies, capabilities, etc., sent
2328 * by the MDS as they are applied to our local cache.
2329 */
2330static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2331{
2332 struct ceph_mds_client *mdsc = session->s_mdsc;
2333 struct ceph_mds_request *req;
2334 struct ceph_mds_reply_head *head = msg->front.iov_base;
2335 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2336 struct ceph_snap_realm *realm;
2f2dc053
SW		 2337 u64 tid;
2338 int err, result;
2600d2dd 2339 int mds = session->s_mds;
2f2dc053 2340
2f2dc053
SW		 2341 if (msg->front.iov_len < sizeof(*head)) {
2342 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2343 ceph_msg_dump(msg);
2f2dc053
SW		 2344 return;
2345 }
2346
2347 /* get request, session */
6df058c0 2348 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2349 mutex_lock(&mdsc->mutex);
fcd00b68 2350 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2351 if (!req) {
2352 dout("handle_reply on unknown tid %llu\n", tid);
2353 mutex_unlock(&mdsc->mutex);
2354 return;
2355 }
2356 dout("handle_reply %p\n", req);
2f2dc053
SW		 2357
2358 /* correct session? */
d96d6049 2359 if (req->r_session != session) {
2f2dc053
SW		 2360 pr_err("mdsc_handle_reply got %llu on session mds%d"
2361 " not mds%d\n", tid, session->s_mds,
2362 req->r_session ? req->r_session->s_mds : -1);
2363 mutex_unlock(&mdsc->mutex);
2364 goto out;
2365 }
2366
2367 /* dup? */
2368 if ((req->r_got_unsafe && !head->safe) ||
2369 (req->r_got_safe && head->safe)) {
f3ae1b97 2370 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2371 head->safe ? "safe" : "unsafe", tid, mds);
2372 mutex_unlock(&mdsc->mutex);
2373 goto out;
2374 }
1550d34e 2375 if (req->r_got_safe) {
f3ae1b97 2376 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2377 tid, mds);
2378 mutex_unlock(&mdsc->mutex);
2379 goto out;
2380 }
2f2dc053
SW		 2381
2382 result = le32_to_cpu(head->result);
2383
2384 /*
e55b71f8
GF		 2385 * Handle an ESTALE
2386 * if we're not talking to the authority, send to them
2387 * if the authority has changed while we weren't looking,
2388 * send to new authority
2389 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2390 */
2391 if (result == -ESTALE) {
e55b71f8 2392 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2393 req->r_resend_mds = -1;
ca18bede 2394 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2395 dout("not using auth, setting for that now");
2396 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2397 __do_request(mdsc, req);
2398 mutex_unlock(&mdsc->mutex);
2399 goto out;
e55b71f8 2400 } else {
ca18bede
YZ		 2401 int mds = __choose_mds(mdsc, req);
2402 if (mds >= 0 && mds != req->r_session->s_mds) {
2403 dout("but auth changed, so resending");
e55b71f8
GF		 2404 __do_request(mdsc, req);
2405 mutex_unlock(&mdsc->mutex);
2406 goto out;
2407 }
2f2dc053 2408 }
e55b71f8 2409 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2410 }
2411
e55b71f8 2412
2f2dc053
SW		 2413 if (head->safe) {
2414 req->r_got_safe = true;
2415 __unregister_request(mdsc, req);
2f2dc053
SW		 2416
2417 if (req->r_got_unsafe) {
2418 /*
2419 * We already handled the unsafe response, now do the
2420 * cleanup. No need to examine the response; the MDS
2421 * doesn't include any result info in the safe
2422 * response. And even if it did, there is nothing
2423 * useful we could do with a revised return value.
2424 */
2425 dout("got safe reply %llu, mds%d\n", tid, mds);
2426 list_del_init(&req->r_unsafe_item);
2427
2428 /* last unsafe request during umount? */
44ca18f2 2429 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2430 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2431 mutex_unlock(&mdsc->mutex);
2432 goto out;
2433 }
e1518c7c 2434 } else {
2f2dc053
SW		 2435 req->r_got_unsafe = true;
2436 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2437 if (req->r_unsafe_dir) {
2438 struct ceph_inode_info *ci =
2439 ceph_inode(req->r_unsafe_dir);
2440 spin_lock(&ci->i_unsafe_lock);
2441 list_add_tail(&req->r_unsafe_dir_item,
2442 &ci->i_unsafe_dirops);
2443 spin_unlock(&ci->i_unsafe_lock);
2444 }
2f2dc053
SW		 2445 }
2446
2447 dout("handle_reply tid %lld result %d\n", tid, result);
2448 rinfo = &req->r_reply_info;
14303d20 2449 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2450 mutex_unlock(&mdsc->mutex);
2451
2452 mutex_lock(&session->s_mutex);
2453 if (err < 0) {
25933abd 2454 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2455 ceph_msg_dump(msg);
2f2dc053
SW		 2456 goto out_err;
2457 }
2458
2459 /* snap trace */
982d6011 2460 realm = NULL;
2f2dc053
SW		 2461 if (rinfo->snapblob_len) {
2462 down_write(&mdsc->snap_rwsem);
2463 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2464 rinfo->snapblob + rinfo->snapblob_len,
2465 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2466 &realm);
2f2dc053
SW		 2467 downgrade_write(&mdsc->snap_rwsem);
2468 } else {
2469 down_read(&mdsc->snap_rwsem);
2470 }
2471
2472 /* insert trace into our cache */
b4556396 2473 mutex_lock(&req->r_fill_mutex);
315f2408 2474 current->journal_info = req;
3d14c5d2 2475 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053 2476 if (err == 0) {
6e8575fa 2477 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2478 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2479 ceph_readdir_prepopulate(req, req->r_session);
37151668 2480 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2481 }
315f2408 2482 current->journal_info = NULL;
b4556396 2483 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2484
2485 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2486 if (realm)
2487 ceph_put_snap_realm(mdsc, realm);
68cd5b4b
YZ		 2488
2489 if (err == 0 && req->r_got_unsafe && req->r_target_inode) {
2490 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2491 spin_lock(&ci->i_unsafe_lock);
2492 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2493 spin_unlock(&ci->i_unsafe_lock);
2494 }
2f2dc053 2495out_err:
e1518c7c
SW		 2496 mutex_lock(&mdsc->mutex);
2497 if (!req->r_aborted) {
2498 if (err) {
2499 req->r_err = err;
2500 } else {
5fdb1389 2501 req->r_reply = ceph_msg_get(msg);
e1518c7c
SW		 2502 req->r_got_result = true;
2503 }
2f2dc053 2504 } else {
e1518c7c 2505 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2506 }
e1518c7c 2507 mutex_unlock(&mdsc->mutex);
2f2dc053 2508
2f2dc053
SW		 2509 mutex_unlock(&session->s_mutex);
2510
2511 /* kick calling process */
2512 complete_request(mdsc, req);
2513out:
2514 ceph_mdsc_put_request(req);
2515 return;
2516}
2517
2518
2519
2520/*
2521 * handle mds notification that our request has been forwarded.
2522 */
2600d2dd
SW		 2523static void handle_forward(struct ceph_mds_client *mdsc,
2524 struct ceph_mds_session *session,
2525 struct ceph_msg *msg)
2f2dc053
SW		 2526{
2527 struct ceph_mds_request *req;
a1ea787c 2528 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2529 u32 next_mds;
2530 u32 fwd_seq;
2f2dc053
SW		 2531 int err = -EINVAL;
2532 void *p = msg->front.iov_base;
2533 void *end = p + msg->front.iov_len;
2f2dc053 2534
a1ea787c 2535 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2536 next_mds = ceph_decode_32(&p);
2537 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2538
2539 mutex_lock(&mdsc->mutex);
fcd00b68 2540 req = lookup_get_request(mdsc, tid);
2f2dc053 2541 if (!req) {
2a8e5e36 2542 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2543 goto out; /* dup reply? */
2544 }
2545
2a8e5e36
SW		 2546 if (req->r_aborted) {
2547 dout("forward tid %llu aborted, unregistering\n", tid);
2548 __unregister_request(mdsc, req);
2549 } else if (fwd_seq <= req->r_num_fwd) {
2550 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2551 tid, next_mds, req->r_num_fwd, fwd_seq);
2552 } else {
2553 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2554 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2555 BUG_ON(req->r_err);
2556 BUG_ON(req->r_got_result);
3de22be6 2557 req->r_attempts = 0;
2f2dc053
SW		 2558 req->r_num_fwd = fwd_seq;
2559 req->r_resend_mds = next_mds;
2560 put_request_session(req);
2561 __do_request(mdsc, req);
2562 }
2563 ceph_mdsc_put_request(req);
2564out:
2565 mutex_unlock(&mdsc->mutex);
2566 return;
2567
2568bad:
2569 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2570}
2571
2572/*
2573 * handle a mds session control message
2574 */
2575static void handle_session(struct ceph_mds_session *session,
2576 struct ceph_msg *msg)
2577{
2578 struct ceph_mds_client *mdsc = session->s_mdsc;
2579 u32 op;
2580 u64 seq;
2600d2dd 2581 int mds = session->s_mds;
2f2dc053
SW		 2582 struct ceph_mds_session_head *h = msg->front.iov_base;
2583 int wake = 0;
2584
2f2dc053
SW		 2585 /* decode */
2586 if (msg->front.iov_len != sizeof(*h))
2587 goto bad;
2588 op = le32_to_cpu(h->op);
2589 seq = le64_to_cpu(h->seq);
2590
2591 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2592 if (op == CEPH_SESSION_CLOSE)
2593 __unregister_session(mdsc, session);
2f2dc053
SW		 2594 /* FIXME: this ttl calculation is generous */
2595 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2596 mutex_unlock(&mdsc->mutex);
2597
2598 mutex_lock(&session->s_mutex);
2599
2600 dout("handle_session mds%d %s %p state %s seq %llu\n",
2601 mds, ceph_session_op_name(op), session,
a687ecaf 2602 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2603
2604 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2605 session->s_state = CEPH_MDS_SESSION_OPEN;
2606 pr_info("mds%d came back\n", session->s_mds);
2607 }
2608
2609 switch (op) {
2610 case CEPH_SESSION_OPEN:
29790f26
SW		 2611 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2612 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2613 session->s_state = CEPH_MDS_SESSION_OPEN;
2614 renewed_caps(mdsc, session, 0);
2615 wake = 1;
2616 if (mdsc->stopping)
2617 __close_session(mdsc, session);
2618 break;
2619
2620 case CEPH_SESSION_RENEWCAPS:
2621 if (session->s_renew_seq == seq)
2622 renewed_caps(mdsc, session, 1);
2623 break;
2624
2625 case CEPH_SESSION_CLOSE:
29790f26
SW		 2626 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2627 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2628 cleanup_session_requests(mdsc, session);
2f2dc053 2629 remove_session_caps(session);
656e4382 2630 wake = 2; /* for good measure */
f3c60c59 2631 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2632 break;
2633
2634 case CEPH_SESSION_STALE:
2635 pr_info("mds%d caps went stale, renewing\n",
2636 session->s_mds);
d8fb02ab 2637 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2638 session->s_cap_gen++;
1ce208a6 2639 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2640 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2641 send_renew_caps(mdsc, session);
2642 break;
2643
2644 case CEPH_SESSION_RECALL_STATE:
2645 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2646 break;
2647
186e4f7a
YZ		 2648 case CEPH_SESSION_FLUSHMSG:
2649 send_flushmsg_ack(mdsc, session, seq);
2650 break;
2651
03f4fcb0
YZ		 2652 case CEPH_SESSION_FORCE_RO:
2653 dout("force_session_readonly %p\n", session);
2654 spin_lock(&session->s_cap_lock);
2655 session->s_readonly = true;
2656 spin_unlock(&session->s_cap_lock);
2657 wake_up_session_caps(session, 0);
2658 break;
2659
fcff415c
YZ		 2660 case CEPH_SESSION_REJECT:
2661 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2662 pr_info("mds%d rejected session\n", session->s_mds);
2663 session->s_state = CEPH_MDS_SESSION_REJECTED;
2664 cleanup_session_requests(mdsc, session);
2665 remove_session_caps(session);
2666 wake = 2; /* for good measure */
2667 break;
2668
2f2dc053
SW		 2669 default:
2670 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2671 WARN_ON(1);
2672 }
2673
2674 mutex_unlock(&session->s_mutex);
2675 if (wake) {
2676 mutex_lock(&mdsc->mutex);
2677 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2678 if (wake == 2)
2679 kick_requests(mdsc, mds);
2f2dc053
SW		 2680 mutex_unlock(&mdsc->mutex);
2681 }
2682 return;
2683
2684bad:
2685 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2686 (int)msg->front.iov_len);
9ec7cab1 2687 ceph_msg_dump(msg);
2f2dc053
SW		 2688 return;
2689}
2690
2691
2692/*
2693 * called under session->mutex.
2694 */
2695static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2696 struct ceph_mds_session *session)
2697{
2698 struct ceph_mds_request *req, *nreq;
3de22be6 2699 struct rb_node *p;
2f2dc053
SW		 2700 int err;
2701
2702 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2703
2704 mutex_lock(&mdsc->mutex);
2705 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2706 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2707 if (!err) {
2708 ceph_msg_get(req->r_request);
2709 ceph_con_send(&session->s_con, req->r_request);
2710 }
2711 }
3de22be6
YZ		 2712
2713 /*
2714 * also re-send old requests when MDS enters reconnect stage. So that MDS
2715 * can process completed request in clientreplay stage.
2716 */
2717 p = rb_first(&mdsc->request_tree);
2718 while (p) {
2719 req = rb_entry(p, struct ceph_mds_request, r_node);
2720 p = rb_next(p);
2721 if (req->r_got_unsafe)
2722 continue;
2723 if (req->r_attempts == 0)
2724 continue; /* only old requests */
2725 if (req->r_session &&
2726 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2727 err = __prepare_send_request(mdsc, req,
2728 session->s_mds, true);
3de22be6
YZ		 2729 if (!err) {
2730 ceph_msg_get(req->r_request);
2731 ceph_con_send(&session->s_con, req->r_request);
2732 }
2733 }
2734 }
2f2dc053
SW		 2735 mutex_unlock(&mdsc->mutex);
2736}
2737
2738/*
2739 * Encode information about a cap for a reconnect with the MDS.
2740 */
2f2dc053
SW		 2741static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2742 void *arg)
2743{
20cb34ae
SW		 2744 union {
2745 struct ceph_mds_cap_reconnect v2;
2746 struct ceph_mds_cap_reconnect_v1 v1;
2747 } rec;
2f2dc053 2748 struct ceph_inode_info *ci;
20cb34ae
SW		 2749 struct ceph_reconnect_state *recon_state = arg;
2750 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2751 char *path;
2752 int pathlen, err;
2753 u64 pathbase;
3469ed0d 2754 u64 snap_follows;
2f2dc053
SW		 2755 struct dentry *dentry;
2756
2757 ci = cap->ci;
2758
2759 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2760 inode, ceph_vinop(inode), cap, cap->cap_id,
2761 ceph_cap_string(cap->issued));
93cea5be
SW		 2762 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2763 if (err)
2764 return err;
2f2dc053
SW		 2765
2766 dentry = d_find_alias(inode);
2767 if (dentry) {
2768 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2769 if (IS_ERR(path)) {
2770 err = PTR_ERR(path);
e072f8aa 2771 goto out_dput;
2f2dc053
SW		 2772 }
2773 } else {
2774 path = NULL;
2775 pathlen = 0;
4eacd4cb 2776 pathbase = 0;
2f2dc053 2777 }
2f2dc053 2778
be655596 2779 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2780 cap->seq = 0; /* reset cap seq */
2781 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2782 cap->mseq = 0; /* and migrate_seq */
99a9c273 2783 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2784
121f22a1 2785 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2786 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2787 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2788 rec.v2.issued = cpu_to_le32(cap->issued);
2789 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2790 rec.v2.pathbase = cpu_to_le64(pathbase);
2791 rec.v2.flock_len = 0;
20cb34ae
SW		 2792 } else {
2793 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2794 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2795 rec.v1.issued = cpu_to_le32(cap->issued);
2796 rec.v1.size = cpu_to_le64(inode->i_size);
2797 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2798 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2799 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2800 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2801 }
3469ed0d
YZ		 2802
2803 if (list_empty(&ci->i_cap_snaps)) {
2804 snap_follows = 0;
2805 } else {
2806 struct ceph_cap_snap *capsnap =
2807 list_first_entry(&ci->i_cap_snaps,
2808 struct ceph_cap_snap, ci_item);
2809 snap_follows = capsnap->follows;
20cb34ae 2810 }
be655596 2811 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2812
121f22a1 2813 if (recon_state->msg_version >= 2) {
40819f6f 2814 int num_fcntl_locks, num_flock_locks;
39be95e9 2815 struct ceph_filelock *flocks;
121f22a1
YZ		 2816 size_t struct_len, total_len = 0;
2817 u8 struct_v = 0;
39be95e9
JS		 2818
2819encode_again:
39be95e9 2820 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
39be95e9
JS		 2821 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2822 sizeof(struct ceph_filelock), GFP_NOFS);
2823 if (!flocks) {
2824 err = -ENOMEM;
2825 goto out_free;
2826 }
39be95e9
JS		 2827 err = ceph_encode_locks_to_buffer(inode, flocks,
2828 num_fcntl_locks,
2829 num_flock_locks);
39be95e9
JS		 2830 if (err) {
2831 kfree(flocks);
2832 if (err == -ENOSPC)
2833 goto encode_again;
2834 goto out_free;
2835 }
121f22a1
YZ		 2836
2837 if (recon_state->msg_version >= 3) {
2838 /* version, compat_version and struct_len */
2839 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 2840 struct_v = 2;
121f22a1 2841 }
39be95e9
JS		 2842 /*
2843 * number of encoded locks is stable, so copy to pagelist
2844 */
121f22a1
YZ		 2845 struct_len = 2 * sizeof(u32) +
2846 (num_fcntl_locks + num_flock_locks) *
2847 sizeof(struct ceph_filelock);
2848 rec.v2.flock_len = cpu_to_le32(struct_len);
2849
2850 struct_len += sizeof(rec.v2);
2851 struct_len += sizeof(u32) + pathlen;
2852
3469ed0d
YZ		 2853 if (struct_v >= 2)
2854 struct_len += sizeof(u64); /* snap_follows */
2855
121f22a1
YZ		 2856 total_len += struct_len;
2857 err = ceph_pagelist_reserve(pagelist, total_len);
2858
2859 if (!err) {
2860 if (recon_state->msg_version >= 3) {
2861 ceph_pagelist_encode_8(pagelist, struct_v);
2862 ceph_pagelist_encode_8(pagelist, 1);
2863 ceph_pagelist_encode_32(pagelist, struct_len);
2864 }
2865 ceph_pagelist_encode_string(pagelist, path, pathlen);
2866 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
2867 ceph_locks_to_pagelist(flocks, pagelist,
2868 num_fcntl_locks,
2869 num_flock_locks);
3469ed0d
YZ		 2870 if (struct_v >= 2)
2871 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 2872 }
39be95e9 2873 kfree(flocks);
3612abbd 2874 } else {
121f22a1
YZ		 2875 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
2876 err = ceph_pagelist_reserve(pagelist, size);
2877 if (!err) {
2878 ceph_pagelist_encode_string(pagelist, path, pathlen);
2879 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
2880 }
40819f6f 2881 }
44c99757
YZ		 2882
2883 recon_state->nr_caps++;
e072f8aa 2884out_free:
2f2dc053 2885 kfree(path);
e072f8aa 2886out_dput:
2f2dc053 2887 dput(dentry);
93cea5be 2888 return err;
2f2dc053
SW		 2889}
2890
2891
2892/*
2893 * If an MDS fails and recovers, clients need to reconnect in order to
2894 * reestablish shared state. This includes all caps issued through
2895 * this session _and_ the snap_realm hierarchy. Because it's not
2896 * clear which snap realms the mds cares about, we send everything we
2897 * know about.. that ensures we'll then get any new info the
2898 * recovering MDS might have.
2899 *
2900 * This is a relatively heavyweight operation, but it's rare.
2901 *
2902 * called with mdsc->mutex held.
2903 */
34b6c855
SW		 2904static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2905 struct ceph_mds_session *session)
2f2dc053 2906{
2f2dc053 2907 struct ceph_msg *reply;
a105f00c 2908 struct rb_node *p;
34b6c855 2909 int mds = session->s_mds;
9abf82b8 2910 int err = -ENOMEM;
44c99757 2911 int s_nr_caps;
93cea5be 2912 struct ceph_pagelist *pagelist;
20cb34ae 2913 struct ceph_reconnect_state recon_state;
2f2dc053 2914
34b6c855 2915 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2916
93cea5be
SW		 2917 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2918 if (!pagelist)
2919 goto fail_nopagelist;
2920 ceph_pagelist_init(pagelist);
2921
b61c2763 2922 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 2923 if (!reply)
93cea5be 2924 goto fail_nomsg;
93cea5be 2925
34b6c855
SW		 2926 mutex_lock(&session->s_mutex);
2927 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2928 session->s_seq = 0;
2f2dc053 2929
2f2dc053 2930 dout("session %p state %s\n", session,
a687ecaf 2931 ceph_session_state_name(session->s_state));
2f2dc053 2932
99a9c273
YZ		 2933 spin_lock(&session->s_gen_ttl_lock);
2934 session->s_cap_gen++;
2935 spin_unlock(&session->s_gen_ttl_lock);
2936
2937 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 2938 /* don't know if session is readonly */
2939 session->s_readonly = 0;
99a9c273
YZ		 2940 /*
2941 * notify __ceph_remove_cap() that we are composing cap reconnect.
2942 * If a cap get released before being added to the cap reconnect,
2943 * __ceph_remove_cap() should skip queuing cap release.
2944 */
2945 session->s_cap_reconnect = 1;
e01a5946 2946 /* drop old cap expires; we're about to reestablish that state */
745a8e3b 2947 cleanup_cap_releases(mdsc, session);
e01a5946 2948
5d23371f 2949 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 2950 if (mdsc->fsc->sb->s_root)
2951 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 2952
2953 ceph_con_close(&session->s_con);
2954 ceph_con_open(&session->s_con,
2955 CEPH_ENTITY_TYPE_MDS, mds,
2956 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2957
2958 /* replay unsafe requests */
2959 replay_unsafe_requests(mdsc, session);
2960
2961 down_read(&mdsc->snap_rwsem);
2962
2f2dc053 2963 /* traverse this session's caps */
44c99757
YZ		 2964 s_nr_caps = session->s_nr_caps;
2965 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 2966 if (err)
2967 goto fail;
20cb34ae 2968
44c99757 2969 recon_state.nr_caps = 0;
20cb34ae 2970 recon_state.pagelist = pagelist;
121f22a1
YZ		 2971 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
2972 recon_state.msg_version = 3;
2973 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
2974 recon_state.msg_version = 2;
2975 else
2976 recon_state.msg_version = 1;
20cb34ae 2977 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2978 if (err < 0)
9abf82b8 2979 goto fail;
2f2dc053 2980
99a9c273
YZ		 2981 spin_lock(&session->s_cap_lock);
2982 session->s_cap_reconnect = 0;
2983 spin_unlock(&session->s_cap_lock);
2984
2f2dc053
SW		 2985 /*
2986 * snaprealms. we provide mds with the ino, seq (version), and
2987 * parent for all of our realms. If the mds has any newer info,
2988 * it will tell us.
2989 */
a105f00c
SW		 2990 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2991 struct ceph_snap_realm *realm =
2992 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 2993 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 2994
2995 dout(" adding snap realm %llx seq %lld parent %llx\n",
2996 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 2997 sr_rec.ino = cpu_to_le64(realm->ino);
2998 sr_rec.seq = cpu_to_le64(realm->seq);
2999 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3000 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3001 if (err)
3002 goto fail;
2f2dc053 3003 }
2f2dc053 3004
121f22a1 3005 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3006
3007 /* raced with cap release? */
3008 if (s_nr_caps != recon_state.nr_caps) {
3009 struct page *page = list_first_entry(&pagelist->head,
3010 struct page, lru);
3011 __le32 *addr = kmap_atomic(page);
3012 *addr = cpu_to_le32(recon_state.nr_caps);
3013 kunmap_atomic(addr);
ebf18f47 3014 }
44c99757
YZ		 3015
3016 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3017 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3018
3019 ceph_early_kick_flushing_caps(mdsc, session);
3020
2f2dc053
SW		 3021 ceph_con_send(&session->s_con, reply);
3022
9abf82b8
SW		 3023 mutex_unlock(&session->s_mutex);
3024
3025 mutex_lock(&mdsc->mutex);
3026 __wake_requests(mdsc, &session->s_waiting);
3027 mutex_unlock(&mdsc->mutex);
3028
2f2dc053 3029 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3030 return;
3031
93cea5be 3032fail:
2f2dc053 3033 ceph_msg_put(reply);
9abf82b8
SW		 3034 up_read(&mdsc->snap_rwsem);
3035 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3036fail_nomsg:
3037 ceph_pagelist_release(pagelist);
93cea5be 3038fail_nopagelist:
9abf82b8 3039 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3040 return;
2f2dc053
SW		 3041}
3042
3043
3044/*
3045 * compare old and new mdsmaps, kicking requests
3046 * and closing out old connections as necessary
3047 *
3048 * called under mdsc->mutex.
3049 */
3050static void check_new_map(struct ceph_mds_client *mdsc,
3051 struct ceph_mdsmap *newmap,
3052 struct ceph_mdsmap *oldmap)
3053{
3054 int i;
3055 int oldstate, newstate;
3056 struct ceph_mds_session *s;
3057
3058 dout("check_new_map new %u old %u\n",
3059 newmap->m_epoch, oldmap->m_epoch);
3060
3061 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
3062 if (mdsc->sessions[i] == NULL)
3063 continue;
3064 s = mdsc->sessions[i];
3065 oldstate = ceph_mdsmap_get_state(oldmap, i);
3066 newstate = ceph_mdsmap_get_state(newmap, i);
3067
0deb01c9 3068 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3069 i, ceph_mds_state_name(oldstate),
0deb01c9 3070 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3071 ceph_mds_state_name(newstate),
0deb01c9 3072 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3073 ceph_session_state_name(s->s_state));
2f2dc053 3074
3e8f43a0
YZ		 3075 if (i >= newmap->m_max_mds ||
3076 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3077 ceph_mdsmap_get_addr(newmap, i),
3078 sizeof(struct ceph_entity_addr))) {
3079 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3080 /* the session never opened, just close it
3081 * out now */
3082 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 3083 __unregister_session(mdsc, s);
2f2dc053
SW		 3084 } else {
3085 /* just close it */
3086 mutex_unlock(&mdsc->mutex);
3087 mutex_lock(&s->s_mutex);
3088 mutex_lock(&mdsc->mutex);
3089 ceph_con_close(&s->s_con);
3090 mutex_unlock(&s->s_mutex);
3091 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3092 }
2f2dc053
SW		 3093 } else if (oldstate == newstate) {
3094 continue; /* nothing new with this mds */
3095 }
3096
3097 /*
3098 * send reconnect?
3099 */
3100 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3101 newstate >= CEPH_MDS_STATE_RECONNECT) {
3102 mutex_unlock(&mdsc->mutex);
3103 send_mds_reconnect(mdsc, s);
3104 mutex_lock(&mdsc->mutex);
3105 }
2f2dc053
SW		 3106
3107 /*
29790f26 3108 * kick request on any mds that has gone active.
2f2dc053
SW		 3109 */
3110 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3111 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3112 if (oldstate != CEPH_MDS_STATE_CREATING &&
3113 oldstate != CEPH_MDS_STATE_STARTING)
3114 pr_info("mds%d recovery completed\n", s->s_mds);
3115 kick_requests(mdsc, i);
2f2dc053 3116 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3117 wake_up_session_caps(s, 1);
2f2dc053
SW		 3118 }
3119 }
cb170a22
SW		 3120
3121 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3122 s = mdsc->sessions[i];
3123 if (!s)
3124 continue;
3125 if (!ceph_mdsmap_is_laggy(newmap, i))
3126 continue;
3127 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3128 s->s_state == CEPH_MDS_SESSION_HUNG ||
3129 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3130 dout(" connecting to export targets of laggy mds%d\n",
3131 i);
3132 __open_export_target_sessions(mdsc, s);
3133 }
3134 }
2f2dc053
SW		 3135}
3136
3137
3138
3139/*
3140 * leases
3141 */
3142
3143/*
3144 * caller must hold session s_mutex, dentry->d_lock
3145 */
3146void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3147{
3148 struct ceph_dentry_info *di = ceph_dentry(dentry);
3149
3150 ceph_put_mds_session(di->lease_session);
3151 di->lease_session = NULL;
3152}
3153
2600d2dd
SW		 3154static void handle_lease(struct ceph_mds_client *mdsc,
3155 struct ceph_mds_session *session,
3156 struct ceph_msg *msg)
2f2dc053 3157{
3d14c5d2 3158 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3159 struct inode *inode;
2f2dc053
SW		 3160 struct dentry *parent, *dentry;
3161 struct ceph_dentry_info *di;
2600d2dd 3162 int mds = session->s_mds;
2f2dc053 3163 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3164 u32 seq;
2f2dc053 3165 struct ceph_vino vino;
2f2dc053
SW		 3166 struct qstr dname;
3167 int release = 0;
3168
2f2dc053
SW		 3169 dout("handle_lease from mds%d\n", mds);
3170
3171 /* decode */
3172 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3173 goto bad;
3174 vino.ino = le64_to_cpu(h->ino);
3175 vino.snap = CEPH_NOSNAP;
1e5ea23d 3176 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3177 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3178 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3179 if (dname.len != get_unaligned_le32(h+1))
3180 goto bad;
3181
2f2dc053
SW		 3182 /* lookup inode */
3183 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3184 dout("handle_lease %s, ino %llx %p %.*s\n",
3185 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3186 dname.len, dname.name);
6cd3bcad
YZ		 3187
3188 mutex_lock(&session->s_mutex);
3189 session->s_seq++;
3190
2f2dc053
SW		 3191 if (inode == NULL) {
3192 dout("handle_lease no inode %llx\n", vino.ino);
3193 goto release;
3194 }
2f2dc053
SW		 3195
3196 /* dentry */
3197 parent = d_find_alias(inode);
3198 if (!parent) {
3199 dout("no parent dentry on inode %p\n", inode);
3200 WARN_ON(1);
3201 goto release; /* hrm... */
3202 }
8387ff25 3203 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3204 dentry = d_lookup(parent, &dname);
3205 dput(parent);
3206 if (!dentry)
3207 goto release;
3208
3209 spin_lock(&dentry->d_lock);
3210 di = ceph_dentry(dentry);
3211 switch (h->action) {
3212 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3213 if (di->lease_session == session) {
1e5ea23d
SW		 3214 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3215 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3216 __ceph_mdsc_drop_dentry_lease(dentry);
3217 }
3218 release = 1;
3219 break;
3220
3221 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3222 if (di->lease_session == session &&
2f2dc053
SW		 3223 di->lease_gen == session->s_cap_gen &&
3224 di->lease_renew_from &&
3225 di->lease_renew_after == 0) {
3226 unsigned long duration =
3563dbdd 3227 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3228
1e5ea23d 3229 di->lease_seq = seq;
9b16f03c 3230 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3231 di->lease_renew_after = di->lease_renew_from +
3232 (duration >> 1);
3233 di->lease_renew_from = 0;
3234 }
3235 break;
3236 }
3237 spin_unlock(&dentry->d_lock);
3238 dput(dentry);
3239
3240 if (!release)
3241 goto out;
3242
3243release:
3244 /* let's just reuse the same message */
3245 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3246 ceph_msg_get(msg);
3247 ceph_con_send(&session->s_con, msg);
3248
3249out:
3250 iput(inode);
3251 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3252 return;
3253
3254bad:
3255 pr_err("corrupt lease message\n");
9ec7cab1 3256 ceph_msg_dump(msg);
2f2dc053
SW		 3257}
3258
3259void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3260 struct inode *inode,
3261 struct dentry *dentry, char action,
3262 u32 seq)
3263{
3264 struct ceph_msg *msg;
3265 struct ceph_mds_lease *lease;
3266 int len = sizeof(*lease) + sizeof(u32);
3267 int dnamelen = 0;
3268
3269 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3270 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3271 dnamelen = dentry->d_name.len;
3272 len += dnamelen;
3273
b61c2763 3274 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3275 if (!msg)
2f2dc053
SW		 3276 return;
3277 lease = msg->front.iov_base;
3278 lease->action = action;
2f2dc053
SW		 3279 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3280 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3281 lease->seq = cpu_to_le32(seq);
3282 put_unaligned_le32(dnamelen, lease + 1);
3283 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3284
3285 /*
3286 * if this is a preemptive lease RELEASE, no need to
3287 * flush request stream, since the actual request will
3288 * soon follow.
3289 */
3290 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3291
3292 ceph_con_send(&session->s_con, msg);
3293}
3294
2f2dc053
SW		 3295/*
3296 * drop all leases (and dentry refs) in preparation for umount
3297 */
3298static void drop_leases(struct ceph_mds_client *mdsc)
3299{
3300 int i;
3301
3302 dout("drop_leases\n");
3303 mutex_lock(&mdsc->mutex);
3304 for (i = 0; i < mdsc->max_sessions; i++) {
3305 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3306 if (!s)
3307 continue;
3308 mutex_unlock(&mdsc->mutex);
3309 mutex_lock(&s->s_mutex);
3310 mutex_unlock(&s->s_mutex);
3311 ceph_put_mds_session(s);
3312 mutex_lock(&mdsc->mutex);
3313 }
3314 mutex_unlock(&mdsc->mutex);
3315}
3316
3317
3318
3319/*
3320 * delayed work -- periodically trim expired leases, renew caps with mds
3321 */
3322static void schedule_delayed(struct ceph_mds_client *mdsc)
3323{
3324 int delay = 5;
3325 unsigned hz = round_jiffies_relative(HZ * delay);
3326 schedule_delayed_work(&mdsc->delayed_work, hz);
3327}
3328
3329static void delayed_work(struct work_struct *work)
3330{
3331 int i;
3332 struct ceph_mds_client *mdsc =
3333 container_of(work, struct ceph_mds_client, delayed_work.work);
3334 int renew_interval;
3335 int renew_caps;
3336
3337 dout("mdsc delayed_work\n");
afcdaea3 3338 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3339
3340 mutex_lock(&mdsc->mutex);
3341 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3342 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3343 mdsc->last_renew_caps);
3344 if (renew_caps)
3345 mdsc->last_renew_caps = jiffies;
3346
3347 for (i = 0; i < mdsc->max_sessions; i++) {
3348 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3349 if (s == NULL)
3350 continue;
3351 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3352 dout("resending session close request for mds%d\n",
3353 s->s_mds);
3354 request_close_session(mdsc, s);
3355 ceph_put_mds_session(s);
3356 continue;
3357 }
3358 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3359 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3360 s->s_state = CEPH_MDS_SESSION_HUNG;
3361 pr_info("mds%d hung\n", s->s_mds);
3362 }
3363 }
3364 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3365 /* this mds is failed or recovering, just wait */
3366 ceph_put_mds_session(s);
3367 continue;
3368 }
3369 mutex_unlock(&mdsc->mutex);
3370
3371 mutex_lock(&s->s_mutex);
3372 if (renew_caps)
3373 send_renew_caps(mdsc, s);
3374 else
3375 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3376 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3377 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3378 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3379 mutex_unlock(&s->s_mutex);
3380 ceph_put_mds_session(s);
3381
3382 mutex_lock(&mdsc->mutex);
3383 }
3384 mutex_unlock(&mdsc->mutex);
3385
3386 schedule_delayed(mdsc);
3387}
3388
3d14c5d2 3389int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3390
2f2dc053 3391{
3d14c5d2
YS		 3392 struct ceph_mds_client *mdsc;
3393
3394 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3395 if (!mdsc)
3396 return -ENOMEM;
3397 mdsc->fsc = fsc;
3398 fsc->mdsc = mdsc;
2f2dc053
SW		 3399 mutex_init(&mdsc->mutex);
3400 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3401 if (mdsc->mdsmap == NULL) {
3402 kfree(mdsc);
2d06eeb8 3403 return -ENOMEM;
fb3101b6 3404 }
2d06eeb8 3405
2f2dc053 3406 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3407 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3408 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3409 mdsc->sessions = NULL;
86d8f67b 3410 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3411 mdsc->max_sessions = 0;
3412 mdsc->stopping = 0;
affbc19a 3413 mdsc->last_snap_seq = 0;
2f2dc053 3414 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3415 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3416 INIT_LIST_HEAD(&mdsc->snap_empty);
3417 spin_lock_init(&mdsc->snap_empty_lock);
3418 mdsc->last_tid = 0;
e8a7b8b1 3419 mdsc->oldest_tid = 0;
44ca18f2 3420 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3421 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3422 mdsc->last_renew_caps = jiffies;
3423 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3424 spin_lock_init(&mdsc->cap_delay_lock);
3425 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3426 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3427 mdsc->last_cap_flush_tid = 1;
e4500b5e 3428 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3429 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3430 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3431 mdsc->num_cap_flushing = 0;
3432 spin_lock_init(&mdsc->cap_dirty_lock);
3433 init_waitqueue_head(&mdsc->cap_flushing_wq);
3434 spin_lock_init(&mdsc->dentry_lru_lock);
3435 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3436
37151668 3437 ceph_caps_init(mdsc);
3d14c5d2 3438 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3439
10183a69
YZ		 3440 init_rwsem(&mdsc->pool_perm_rwsem);
3441 mdsc->pool_perm_tree = RB_ROOT;
3442
5f44f142 3443 return 0;
2f2dc053
SW		 3444}
3445
3446/*
3447 * Wait for safe replies on open mds requests. If we time out, drop
3448 * all requests from the tree to avoid dangling dentry refs.
3449 */
3450static void wait_requests(struct ceph_mds_client *mdsc)
3451{
a319bf56 3452 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3453 struct ceph_mds_request *req;
2f2dc053
SW		 3454
3455 mutex_lock(&mdsc->mutex);
44ca18f2 3456 if (__get_oldest_req(mdsc)) {
2f2dc053 3457 mutex_unlock(&mdsc->mutex);
44ca18f2 3458
2f2dc053
SW		 3459 dout("wait_requests waiting for requests\n");
3460 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3461 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3462
3463 /* tear down remaining requests */
44ca18f2
SW		 3464 mutex_lock(&mdsc->mutex);
3465 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3466 dout("wait_requests timed out on tid %llu\n",
3467 req->r_tid);
44ca18f2 3468 __unregister_request(mdsc, req);
2f2dc053
SW		 3469 }
3470 }
3471 mutex_unlock(&mdsc->mutex);
3472 dout("wait_requests done\n");
3473}
3474
3475/*
3476 * called before mount is ro, and before dentries are torn down.
3477 * (hmm, does this still race with new lookups?)
3478 */
3479void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3480{
3481 dout("pre_umount\n");
3482 mdsc->stopping = 1;
3483
3484 drop_leases(mdsc);
afcdaea3 3485 ceph_flush_dirty_caps(mdsc);
2f2dc053 3486 wait_requests(mdsc);
17c688c3
SW		 3487
3488 /*
3489 * wait for reply handlers to drop their request refs and
3490 * their inode/dcache refs
3491 */
3492 ceph_msgr_flush();
2f2dc053
SW		 3493}
3494
3495/*
3496 * wait for all write mds requests to flush.
3497 */
3498static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3499{
80fc7314 3500 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3501 struct rb_node *n;
2f2dc053
SW		 3502
3503 mutex_lock(&mdsc->mutex);
3504 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3505restart:
44ca18f2
SW		 3506 req = __get_oldest_req(mdsc);
3507 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3508 /* find next request */
3509 n = rb_next(&req->r_node);
3510 if (n)
3511 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3512 else
3513 nextreq = NULL;
e8a7b8b1
YZ		 3514 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3515 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3516 /* write op */
3517 ceph_mdsc_get_request(req);
80fc7314
SW		 3518 if (nextreq)
3519 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3520 mutex_unlock(&mdsc->mutex);
3521 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3522 req->r_tid, want_tid);
3523 wait_for_completion(&req->r_safe_completion);
3524 mutex_lock(&mdsc->mutex);
44ca18f2 3525 ceph_mdsc_put_request(req);
80fc7314
SW		 3526 if (!nextreq)
3527 break; /* next dne before, so we're done! */
3528 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3529 /* next request was removed from tree */
3530 ceph_mdsc_put_request(nextreq);
3531 goto restart;
3532 }
3533 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3534 }
80fc7314 3535 req = nextreq;
2f2dc053
SW		 3536 }
3537 mutex_unlock(&mdsc->mutex);
3538 dout("wait_unsafe_requests done\n");
3539}
3540
3541void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3542{
0e294387 3543 u64 want_tid, want_flush;
2f2dc053 3544
48fec5d0 3545 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3546 return;
3547
2f2dc053
SW		 3548 dout("sync\n");
3549 mutex_lock(&mdsc->mutex);
3550 want_tid = mdsc->last_tid;
2f2dc053 3551 mutex_unlock(&mdsc->mutex);
2f2dc053 3552
afcdaea3 3553 ceph_flush_dirty_caps(mdsc);
d3383a8e 3554 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3555 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3556 if (!list_empty(&mdsc->cap_flush_list)) {
3557 struct ceph_cap_flush *cf =
3558 list_last_entry(&mdsc->cap_flush_list,
3559 struct ceph_cap_flush, g_list);
3560 cf->wake = true;
3561 }
d3383a8e
YZ		 3562 spin_unlock(&mdsc->cap_dirty_lock);
3563
0e294387
YZ		 3564 dout("sync want tid %lld flush_seq %lld\n",
3565 want_tid, want_flush);
2f2dc053
SW		 3566
3567 wait_unsafe_requests(mdsc, want_tid);
0e294387 3568 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3569}
3570
f3c60c59
SW		 3571/*
3572 * true if all sessions are closed, or we force unmount
3573 */
fcff415c 3574static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3575{
48fec5d0 3576 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3577 return true;
fcff415c 3578 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3579}
2f2dc053
SW		 3580
3581/*
3582 * called after sb is ro.
3583 */
3584void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3585{
a319bf56 3586 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3587 struct ceph_mds_session *session;
3588 int i;
fcff415c 3589 int skipped = 0;
2f2dc053
SW		 3590
3591 dout("close_sessions\n");
3592
2f2dc053 3593 /* close sessions */
f3c60c59
SW		 3594 mutex_lock(&mdsc->mutex);
3595 for (i = 0; i < mdsc->max_sessions; i++) {
3596 session = __ceph_lookup_mds_session(mdsc, i);
3597 if (!session)
3598 continue;
2f2dc053 3599 mutex_unlock(&mdsc->mutex);
f3c60c59 3600 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3601 if (__close_session(mdsc, session) <= 0)
3602 skipped++;
f3c60c59
SW		 3603 mutex_unlock(&session->s_mutex);
3604 ceph_put_mds_session(session);
2f2dc053
SW		 3605 mutex_lock(&mdsc->mutex);
3606 }
f3c60c59
SW		 3607 mutex_unlock(&mdsc->mutex);
3608
3609 dout("waiting for sessions to close\n");
fcff415c
YZ		 3610 wait_event_timeout(mdsc->session_close_wq,
3611 done_closing_sessions(mdsc, skipped),
a319bf56 3612 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3613
3614 /* tear down remaining sessions */
f3c60c59 3615 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3616 for (i = 0; i < mdsc->max_sessions; i++) {
3617 if (mdsc->sessions[i]) {
3618 session = get_session(mdsc->sessions[i]);
2600d2dd 3619 __unregister_session(mdsc, session);
2f2dc053
SW		 3620 mutex_unlock(&mdsc->mutex);
3621 mutex_lock(&session->s_mutex);
3622 remove_session_caps(session);
3623 mutex_unlock(&session->s_mutex);
3624 ceph_put_mds_session(session);
3625 mutex_lock(&mdsc->mutex);
3626 }
3627 }
2f2dc053 3628 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3629 mutex_unlock(&mdsc->mutex);
3630
3631 ceph_cleanup_empty_realms(mdsc);
3632
3633 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3634
3635 dout("stopped\n");
3636}
3637
48fec5d0
YZ		 3638void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3639{
3640 struct ceph_mds_session *session;
3641 int mds;
3642
3643 dout("force umount\n");
3644
3645 mutex_lock(&mdsc->mutex);
3646 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3647 session = __ceph_lookup_mds_session(mdsc, mds);
3648 if (!session)
3649 continue;
3650 mutex_unlock(&mdsc->mutex);
3651 mutex_lock(&session->s_mutex);
3652 __close_session(mdsc, session);
3653 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3654 cleanup_session_requests(mdsc, session);
3655 remove_session_caps(session);
3656 }
3657 mutex_unlock(&session->s_mutex);
3658 ceph_put_mds_session(session);
3659 mutex_lock(&mdsc->mutex);
3660 kick_requests(mdsc, mds);
3661 }
3662 __wake_requests(mdsc, &mdsc->waiting_for_map);
3663 mutex_unlock(&mdsc->mutex);
3664}
3665
3d14c5d2 3666static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3667{
3668 dout("stop\n");
3669 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3670 if (mdsc->mdsmap)
3671 ceph_mdsmap_destroy(mdsc->mdsmap);
3672 kfree(mdsc->sessions);
37151668 3673 ceph_caps_finalize(mdsc);
10183a69 3674 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3675}
3676
3d14c5d2
YS		 3677void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3678{
3679 struct ceph_mds_client *mdsc = fsc->mdsc;
3680
ef550f6f 3681 dout("mdsc_destroy %p\n", mdsc);
3d14c5d2 3682 ceph_mdsc_stop(mdsc);
ef550f6f
SW		 3683
3684 /* flush out any connection work with references to us */
3685 ceph_msgr_flush();
3686
3d14c5d2
YS		 3687 fsc->mdsc = NULL;
3688 kfree(mdsc);
ef550f6f 3689 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3690}
3691
430afbad
YZ		 3692void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3693{
3694 struct ceph_fs_client *fsc = mdsc->fsc;
3695 const char *mds_namespace = fsc->mount_options->mds_namespace;
3696 void *p = msg->front.iov_base;
3697 void *end = p + msg->front.iov_len;
3698 u32 epoch;
3699 u32 map_len;
3700 u32 num_fs;
3701 u32 mount_fscid = (u32)-1;
3702 u8 struct_v, struct_cv;
3703 int err = -EINVAL;
3704
3705 ceph_decode_need(&p, end, sizeof(u32), bad);
3706 epoch = ceph_decode_32(&p);
3707
3708 dout("handle_fsmap epoch %u\n", epoch);
3709
3710 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3711 struct_v = ceph_decode_8(&p);
3712 struct_cv = ceph_decode_8(&p);
3713 map_len = ceph_decode_32(&p);
3714
3715 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3716 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3717
3718 num_fs = ceph_decode_32(&p);
3719 while (num_fs-- > 0) {
3720 void *info_p, *info_end;
3721 u32 info_len;
3722 u8 info_v, info_cv;
3723 u32 fscid, namelen;
3724
3725 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3726 info_v = ceph_decode_8(&p);
3727 info_cv = ceph_decode_8(&p);
3728 info_len = ceph_decode_32(&p);
3729 ceph_decode_need(&p, end, info_len, bad);
3730 info_p = p;
3731 info_end = p + info_len;
3732 p = info_end;
3733
3734 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3735 fscid = ceph_decode_32(&info_p);
3736 namelen = ceph_decode_32(&info_p);
3737 ceph_decode_need(&info_p, info_end, namelen, bad);
3738
3739 if (mds_namespace &&
3740 strlen(mds_namespace) == namelen &&
3741 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3742 mount_fscid = fscid;
3743 break;
3744 }
3745 }
3746
3747 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3748 if (mount_fscid != (u32)-1) {
3749 fsc->client->monc.fs_cluster_id = mount_fscid;
3750 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
3751 0, true);
3752 ceph_monc_renew_subs(&fsc->client->monc);
3753 } else {
3754 err = -ENOENT;
3755 goto err_out;
3756 }
3757 return;
3758bad:
3759 pr_err("error decoding fsmap\n");
3760err_out:
3761 mutex_lock(&mdsc->mutex);
3762 mdsc->mdsmap_err = -ENOENT;
3763 __wake_requests(mdsc, &mdsc->waiting_for_map);
3764 mutex_unlock(&mdsc->mutex);
3765 return;
3766}
2f2dc053
SW		 3767
3768/*
3769 * handle mds map update.
3770 */
430afbad 3771void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053
SW		 3772{
3773 u32 epoch;
3774 u32 maplen;
3775 void *p = msg->front.iov_base;
3776 void *end = p + msg->front.iov_len;
3777 struct ceph_mdsmap *newmap, *oldmap;
3778 struct ceph_fsid fsid;
3779 int err = -EINVAL;
3780
3781 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3782 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3783 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3784 return;
c89136ea
SW		 3785 epoch = ceph_decode_32(&p);
3786 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3787 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3788
3789 /* do we need it? */
2f2dc053
SW		 3790 mutex_lock(&mdsc->mutex);
3791 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3792 dout("handle_map epoch %u <= our %u\n",
3793 epoch, mdsc->mdsmap->m_epoch);
3794 mutex_unlock(&mdsc->mutex);
3795 return;
3796 }
3797
3798 newmap = ceph_mdsmap_decode(&p, end);
3799 if (IS_ERR(newmap)) {
3800 err = PTR_ERR(newmap);
3801 goto bad_unlock;
3802 }
3803
3804 /* swap into place */
3805 if (mdsc->mdsmap) {
3806 oldmap = mdsc->mdsmap;
3807 mdsc->mdsmap = newmap;
3808 check_new_map(mdsc, newmap, oldmap);
3809 ceph_mdsmap_destroy(oldmap);
3810 } else {
3811 mdsc->mdsmap = newmap; /* first mds map */
3812 }
3d14c5d2 3813 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3814
3815 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 3816 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3817 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 3818
3819 mutex_unlock(&mdsc->mutex);
3820 schedule_delayed(mdsc);
3821 return;
3822
3823bad_unlock:
3824 mutex_unlock(&mdsc->mutex);
3825bad:
3826 pr_err("error decoding mdsmap %d\n", err);
3827 return;
3828}
3829
3830static struct ceph_connection *con_get(struct ceph_connection *con)
3831{
3832 struct ceph_mds_session *s = con->private;
3833
3834 if (get_session(s)) {
2600d2dd 3835 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3836 return con;
3837 }
3838 dout("mdsc con_get %p FAIL\n", s);
3839 return NULL;
3840}
3841
3842static void con_put(struct ceph_connection *con)
3843{
3844 struct ceph_mds_session *s = con->private;
3845
7d8e18a6 3846 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
2f2dc053
SW		 3847 ceph_put_mds_session(s);
3848}
3849
3850/*
3851 * if the client is unresponsive for long enough, the mds will kill
3852 * the session entirely.
3853 */
3854static void peer_reset(struct ceph_connection *con)
3855{
3856 struct ceph_mds_session *s = con->private;
7e70f0ed 3857 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3858
f3ae1b97 3859 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3860 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3861}
3862
3863static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3864{
3865 struct ceph_mds_session *s = con->private;
3866 struct ceph_mds_client *mdsc = s->s_mdsc;
3867 int type = le16_to_cpu(msg->hdr.type);
3868
2600d2dd
SW		 3869 mutex_lock(&mdsc->mutex);
3870 if (__verify_registered_session(mdsc, s) < 0) {
3871 mutex_unlock(&mdsc->mutex);
3872 goto out;
3873 }
3874 mutex_unlock(&mdsc->mutex);
3875
2f2dc053
SW		 3876 switch (type) {
3877 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 3878 ceph_mdsc_handle_mdsmap(mdsc, msg);
3879 break;
3880 case CEPH_MSG_FS_MAP_USER:
3881 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 3882 break;
3883 case CEPH_MSG_CLIENT_SESSION:
3884 handle_session(s, msg);
3885 break;
3886 case CEPH_MSG_CLIENT_REPLY:
3887 handle_reply(s, msg);
3888 break;
3889 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3890 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3891 break;
3892 case CEPH_MSG_CLIENT_CAPS:
3893 ceph_handle_caps(s, msg);
3894 break;
3895 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3896 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3897 break;
3898 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3899 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3900 break;
3901
3902 default:
3903 pr_err("received unknown message type %d %s\n", type,
3904 ceph_msg_type_name(type));
3905 }
2600d2dd 3906out:
2f2dc053
SW		 3907 ceph_msg_put(msg);
3908}
3909
4e7a5dcd
SW		 3910/*
3911 * authentication
3912 */
a3530df3
AE		 3913
3914/*
3915 * Note: returned pointer is the address of a structure that's
3916 * managed separately. Caller must *not* attempt to free it.
3917 */
3918static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3919 int *proto, int force_new)
4e7a5dcd
SW		 3920{
3921 struct ceph_mds_session *s = con->private;
3922 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3923 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 3924 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 3925
74f1869f 3926 if (force_new && auth->authorizer) {
6c1ea260 3927 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 3928 auth->authorizer = NULL;
4e7a5dcd 3929 }
27859f97
SW		 3930 if (!auth->authorizer) {
3931 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3932 auth);
0bed9b5c
SW		 3933 if (ret)
3934 return ERR_PTR(ret);
27859f97
SW		 3935 } else {
3936 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3937 auth);
a255651d 3938 if (ret)
a3530df3 3939 return ERR_PTR(ret);
4e7a5dcd 3940 }
4e7a5dcd 3941 *proto = ac->protocol;
74f1869f 3942
a3530df3 3943 return auth;
4e7a5dcd
SW		 3944}
3945
3946
3947static int verify_authorizer_reply(struct ceph_connection *con, int len)
3948{
3949 struct ceph_mds_session *s = con->private;
3950 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3951 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 3952
27859f97 3953 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
4e7a5dcd
SW		 3954}
3955
9bd2e6f8
SW		 3956static int invalidate_authorizer(struct ceph_connection *con)
3957{
3958 struct ceph_mds_session *s = con->private;
3959 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3960 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 3961
27859f97 3962 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 3963
3d14c5d2 3964 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3965}
3966
53ded495
AE		 3967static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3968 struct ceph_msg_header *hdr, int *skip)
3969{
3970 struct ceph_msg *msg;
3971 int type = (int) le16_to_cpu(hdr->type);
3972 int front_len = (int) le32_to_cpu(hdr->front_len);
3973
3974 if (con->in_msg)
3975 return con->in_msg;
3976
3977 *skip = 0;
3978 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3979 if (!msg) {
3980 pr_err("unable to allocate msg type %d len %d\n",
3981 type, front_len);
3982 return NULL;
3983 }
53ded495
AE		 3984
3985 return msg;
3986}
3987
79dbd1ba 3988static int mds_sign_message(struct ceph_msg *msg)
33d07337 3989{
79dbd1ba 3990 struct ceph_mds_session *s = msg->con->private;
33d07337 3991 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 3992
33d07337
YZ		 3993 return ceph_auth_sign_message(auth, msg);
3994}
3995
79dbd1ba 3996static int mds_check_message_signature(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_check_message_signature(auth, msg);
4002}
4003
9e32789f 4004static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4005 .get = con_get,
4006 .put = con_put,
4007 .dispatch = dispatch,
4e7a5dcd
SW		 4008 .get_authorizer = get_authorizer,
4009 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4010 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4011 .peer_reset = peer_reset,
53ded495 4012 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4013 .sign_message = mds_sign_message,
4014 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4015};
4016
2f2dc053 4017/* eof */
Ubuntu Hirsute Linux kernel mirror