projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: scattered page writeback
[deliverable/linux.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45 * are no longer valid.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
SW		 50 struct ceph_pagelist *pagelist;
51 bool flock;
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
5ea5c5e0
YZ		 103 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
104 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
105 ceph_decode_need(p, end, info->pool_ns_len, bad);
106 *p += info->pool_ns_len;
107 } else {
108 info->pool_ns_len = 0;
109 }
110
2f2dc053
SW		 111 return 0;
112bad:
113 return err;
114}
115
116/*
117 * parse a normal reply, which may contain a (dir+)dentry and/or a
118 * target inode.
119 */
120static int parse_reply_info_trace(void **p, void *end,
14303d20 121 struct ceph_mds_reply_info_parsed *info,
12b4629a 122 u64 features)
2f2dc053
SW		 123{
124 int err;
125
126 if (info->head->is_dentry) {
14303d20 127 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 128 if (err < 0)
129 goto out_bad;
130
131 if (unlikely(*p + sizeof(*info->dirfrag) > end))
132 goto bad;
133 info->dirfrag = *p;
134 *p += sizeof(*info->dirfrag) +
135 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
136 if (unlikely(*p > end))
137 goto bad;
138
139 ceph_decode_32_safe(p, end, info->dname_len, bad);
140 ceph_decode_need(p, end, info->dname_len, bad);
141 info->dname = *p;
142 *p += info->dname_len;
143 info->dlease = *p;
144 *p += sizeof(*info->dlease);
145 }
146
147 if (info->head->is_target) {
14303d20 148 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 149 if (err < 0)
150 goto out_bad;
151 }
152
153 if (unlikely(*p != end))
154 goto bad;
155 return 0;
156
157bad:
158 err = -EIO;
159out_bad:
160 pr_err("problem parsing mds trace %d\n", err);
161 return err;
162}
163
164/*
165 * parse readdir results
166 */
167static int parse_reply_info_dir(void **p, void *end,
14303d20 168 struct ceph_mds_reply_info_parsed *info,
12b4629a 169 u64 features)
2f2dc053
SW		 170{
171 u32 num, i = 0;
172 int err;
173
174 info->dir_dir = *p;
175 if (*p + sizeof(*info->dir_dir) > end)
176 goto bad;
177 *p += sizeof(*info->dir_dir) +
178 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
179 if (*p > end)
180 goto bad;
181
182 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea
SW		 183 num = ceph_decode_32(p);
184 info->dir_end = ceph_decode_8(p);
185 info->dir_complete = ceph_decode_8(p);
2f2dc053
SW		 186 if (num == 0)
187 goto done;
188
54008399 189 BUG_ON(!info->dir_in);
2f2dc053
SW		 190 info->dir_dname = (void *)(info->dir_in + num);
191 info->dir_dname_len = (void *)(info->dir_dname + num);
192 info->dir_dlease = (void *)(info->dir_dname_len + num);
54008399
YZ		 193 if ((unsigned long)(info->dir_dlease + num) >
194 (unsigned long)info->dir_in + info->dir_buf_size) {
195 pr_err("dir contents are larger than expected\n");
196 WARN_ON(1);
197 goto bad;
198 }
2f2dc053 199
54008399 200 info->dir_nr = num;
2f2dc053
SW		 201 while (num) {
202 /* dentry */
203 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea 204 info->dir_dname_len[i] = ceph_decode_32(p);
2f2dc053
SW		 205 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
206 info->dir_dname[i] = *p;
207 *p += info->dir_dname_len[i];
208 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
209 info->dir_dname[i]);
210 info->dir_dlease[i] = *p;
211 *p += sizeof(struct ceph_mds_reply_lease);
212
213 /* inode */
14303d20 214 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
2f2dc053
SW		 215 if (err < 0)
216 goto out_bad;
217 i++;
218 num--;
219 }
220
221done:
222 if (*p != end)
223 goto bad;
224 return 0;
225
226bad:
227 err = -EIO;
228out_bad:
229 pr_err("problem parsing dir contents %d\n", err);
230 return err;
231}
232
25933abd
HS		 233/*
234 * parse fcntl F_GETLK results
235 */
236static int parse_reply_info_filelock(void **p, void *end,
14303d20 237 struct ceph_mds_reply_info_parsed *info,
12b4629a 238 u64 features)
25933abd
HS		 239{
240 if (*p + sizeof(*info->filelock_reply) > end)
241 goto bad;
242
243 info->filelock_reply = *p;
244 *p += sizeof(*info->filelock_reply);
245
246 if (unlikely(*p != end))
247 goto bad;
248 return 0;
249
250bad:
251 return -EIO;
252}
253
6e8575fa
SL		 254/*
255 * parse create results
256 */
257static int parse_reply_info_create(void **p, void *end,
258 struct ceph_mds_reply_info_parsed *info,
12b4629a 259 u64 features)
6e8575fa
SL		 260{
261 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
262 if (*p == end) {
263 info->has_create_ino = false;
264 } else {
265 info->has_create_ino = true;
266 info->ino = ceph_decode_64(p);
267 }
268 }
269
270 if (unlikely(*p != end))
271 goto bad;
272 return 0;
273
274bad:
275 return -EIO;
276}
277
25933abd
HS		 278/*
279 * parse extra results
280 */
281static int parse_reply_info_extra(void **p, void *end,
14303d20 282 struct ceph_mds_reply_info_parsed *info,
12b4629a 283 u64 features)
25933abd
HS		 284{
285 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
14303d20 286 return parse_reply_info_filelock(p, end, info, features);
8a034497
YZ		 287 else if (info->head->op == CEPH_MDS_OP_READDIR ||
288 info->head->op == CEPH_MDS_OP_LSSNAP)
14303d20 289 return parse_reply_info_dir(p, end, info, features);
6e8575fa
SL		 290 else if (info->head->op == CEPH_MDS_OP_CREATE)
291 return parse_reply_info_create(p, end, info, features);
292 else
293 return -EIO;
25933abd
HS		 294}
295
2f2dc053
SW		 296/*
297 * parse entire mds reply
298 */
299static int parse_reply_info(struct ceph_msg *msg,
14303d20 300 struct ceph_mds_reply_info_parsed *info,
12b4629a 301 u64 features)
2f2dc053
SW		 302{
303 void *p, *end;
304 u32 len;
305 int err;
306
307 info->head = msg->front.iov_base;
308 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
309 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
310
311 /* trace */
312 ceph_decode_32_safe(&p, end, len, bad);
313 if (len > 0) {
32852a81 314 ceph_decode_need(&p, end, len, bad);
14303d20 315 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 316 if (err < 0)
317 goto out_bad;
318 }
319
25933abd 320 /* extra */
2f2dc053
SW		 321 ceph_decode_32_safe(&p, end, len, bad);
322 if (len > 0) {
32852a81 323 ceph_decode_need(&p, end, len, bad);
14303d20 324 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 325 if (err < 0)
326 goto out_bad;
327 }
328
329 /* snap blob */
330 ceph_decode_32_safe(&p, end, len, bad);
331 info->snapblob_len = len;
332 info->snapblob = p;
333 p += len;
334
335 if (p != end)
336 goto bad;
337 return 0;
338
339bad:
340 err = -EIO;
341out_bad:
342 pr_err("mds parse_reply err %d\n", err);
343 return err;
344}
345
346static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
347{
54008399
YZ		 348 if (!info->dir_in)
349 return;
350 free_pages((unsigned long)info->dir_in, get_order(info->dir_buf_size));
2f2dc053
SW		 351}
352
353
354/*
355 * sessions
356 */
a687ecaf 357const char *ceph_session_state_name(int s)
2f2dc053
SW		 358{
359 switch (s) {
360 case CEPH_MDS_SESSION_NEW: return "new";
361 case CEPH_MDS_SESSION_OPENING: return "opening";
362 case CEPH_MDS_SESSION_OPEN: return "open";
363 case CEPH_MDS_SESSION_HUNG: return "hung";
364 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 365 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053
SW		 366 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
367 default: return "???";
368 }
369}
370
371static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
372{
373 if (atomic_inc_not_zero(&s->s_ref)) {
374 dout("mdsc get_session %p %d -> %d\n", s,
375 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
376 return s;
377 } else {
378 dout("mdsc get_session %p 0 -- FAIL", s);
379 return NULL;
380 }
381}
382
383void ceph_put_mds_session(struct ceph_mds_session *s)
384{
385 dout("mdsc put_session %p %d -> %d\n", s,
386 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
4e7a5dcd 387 if (atomic_dec_and_test(&s->s_ref)) {
6c4a1915 388 if (s->s_auth.authorizer)
27859f97
SW		 389 ceph_auth_destroy_authorizer(
390 s->s_mdsc->fsc->client->monc.auth,
391 s->s_auth.authorizer);
2f2dc053 392 kfree(s);
4e7a5dcd 393 }
2f2dc053
SW		 394}
395
396/*
397 * called under mdsc->mutex
398 */
399struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
400 int mds)
401{
402 struct ceph_mds_session *session;
403
404 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
405 return NULL;
406 session = mdsc->sessions[mds];
407 dout("lookup_mds_session %p %d\n", session,
408 atomic_read(&session->s_ref));
409 get_session(session);
410 return session;
411}
412
413static bool __have_session(struct ceph_mds_client *mdsc, int mds)
414{
415 if (mds >= mdsc->max_sessions)
416 return false;
417 return mdsc->sessions[mds];
418}
419
2600d2dd
SW		 420static int __verify_registered_session(struct ceph_mds_client *mdsc,
421 struct ceph_mds_session *s)
422{
423 if (s->s_mds >= mdsc->max_sessions ||
424 mdsc->sessions[s->s_mds] != s)
425 return -ENOENT;
426 return 0;
427}
428
2f2dc053
SW		 429/*
430 * create+register a new session for given mds.
431 * called under mdsc->mutex.
432 */
433static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
434 int mds)
435{
436 struct ceph_mds_session *s;
437
c338c07c
NY		 438 if (mds >= mdsc->mdsmap->m_max_mds)
439 return ERR_PTR(-EINVAL);
440
2f2dc053 441 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 442 if (!s)
443 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 444 s->s_mdsc = mdsc;
445 s->s_mds = mds;
446 s->s_state = CEPH_MDS_SESSION_NEW;
447 s->s_ttl = 0;
448 s->s_seq = 0;
449 mutex_init(&s->s_mutex);
450
b7a9e5dd 451 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 452
d8fb02ab 453 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 454 s->s_cap_gen = 0;
1ce208a6 455 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 456
457 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 458 s->s_renew_requested = 0;
459 s->s_renew_seq = 0;
460 INIT_LIST_HEAD(&s->s_caps);
461 s->s_nr_caps = 0;
5dacf091 462 s->s_trim_caps = 0;
2f2dc053
SW		 463 atomic_set(&s->s_ref, 1);
464 INIT_LIST_HEAD(&s->s_waiting);
465 INIT_LIST_HEAD(&s->s_unsafe);
466 s->s_num_cap_releases = 0;
99a9c273 467 s->s_cap_reconnect = 0;
7c1332b8 468 s->s_cap_iterator = NULL;
2f2dc053 469 INIT_LIST_HEAD(&s->s_cap_releases);
2f2dc053
SW		 470 INIT_LIST_HEAD(&s->s_cap_flushing);
471 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
472
473 dout("register_session mds%d\n", mds);
474 if (mds >= mdsc->max_sessions) {
475 int newmax = 1 << get_count_order(mds+1);
476 struct ceph_mds_session **sa;
477
478 dout("register_session realloc to %d\n", newmax);
479 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
480 if (sa == NULL)
42ce56e5 481 goto fail_realloc;
2f2dc053
SW		 482 if (mdsc->sessions) {
483 memcpy(sa, mdsc->sessions,
484 mdsc->max_sessions * sizeof(void *));
485 kfree(mdsc->sessions);
486 }
487 mdsc->sessions = sa;
488 mdsc->max_sessions = newmax;
489 }
490 mdsc->sessions[mds] = s;
86d8f67b 491 atomic_inc(&mdsc->num_sessions);
2f2dc053 492 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 493
b7a9e5dd
SW		 494 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
495 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 496
2f2dc053 497 return s;
42ce56e5
SW		 498
499fail_realloc:
500 kfree(s);
501 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 502}
503
504/*
505 * called under mdsc->mutex
506 */
2600d2dd 507static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 508 struct ceph_mds_session *s)
2f2dc053 509{
2600d2dd
SW		 510 dout("__unregister_session mds%d %p\n", s->s_mds, s);
511 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 512 mdsc->sessions[s->s_mds] = NULL;
513 ceph_con_close(&s->s_con);
514 ceph_put_mds_session(s);
86d8f67b 515 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 516}
517
518/*
519 * drop session refs in request.
520 *
521 * should be last request ref, or hold mdsc->mutex
522 */
523static void put_request_session(struct ceph_mds_request *req)
524{
525 if (req->r_session) {
526 ceph_put_mds_session(req->r_session);
527 req->r_session = NULL;
528 }
529}
530
153c8e6b 531void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 532{
153c8e6b
SW		 533 struct ceph_mds_request *req = container_of(kref,
534 struct ceph_mds_request,
535 r_kref);
54008399 536 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 537 if (req->r_request)
538 ceph_msg_put(req->r_request);
54008399 539 if (req->r_reply)
153c8e6b 540 ceph_msg_put(req->r_reply);
153c8e6b 541 if (req->r_inode) {
41b02e1f 542 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 543 iput(req->r_inode);
544 }
545 if (req->r_locked_dir)
41b02e1f 546 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
e96a650a 547 iput(req->r_target_inode);
153c8e6b
SW		 548 if (req->r_dentry)
549 dput(req->r_dentry);
844d87c3
SW		 550 if (req->r_old_dentry)
551 dput(req->r_old_dentry);
552 if (req->r_old_dentry_dir) {
41b02e1f
SW		 553 /*
554 * track (and drop pins for) r_old_dentry_dir
555 * separately, since r_old_dentry's d_parent may have
556 * changed between the dir mutex being dropped and
557 * this request being freed.
558 */
559 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
560 CEPH_CAP_PIN);
41b02e1f 561 iput(req->r_old_dentry_dir);
2f2dc053 562 }
153c8e6b
SW		 563 kfree(req->r_path1);
564 kfree(req->r_path2);
25e6bae3
YZ		 565 if (req->r_pagelist)
566 ceph_pagelist_release(req->r_pagelist);
153c8e6b 567 put_request_session(req);
37151668 568 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 569 kfree(req);
2f2dc053
SW		 570}
571
572/*
573 * lookup session, bump ref if found.
574 *
575 * called under mdsc->mutex.
576 */
577static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
578 u64 tid)
579{
580 struct ceph_mds_request *req;
44ca18f2
SW		 581 struct rb_node *n = mdsc->request_tree.rb_node;
582
583 while (n) {
584 req = rb_entry(n, struct ceph_mds_request, r_node);
585 if (tid < req->r_tid)
586 n = n->rb_left;
587 else if (tid > req->r_tid)
588 n = n->rb_right;
589 else {
590 ceph_mdsc_get_request(req);
591 return req;
592 }
593 }
594 return NULL;
595}
596
597static void __insert_request(struct ceph_mds_client *mdsc,
598 struct ceph_mds_request *new)
599{
600 struct rb_node **p = &mdsc->request_tree.rb_node;
601 struct rb_node *parent = NULL;
602 struct ceph_mds_request *req = NULL;
603
604 while (*p) {
605 parent = *p;
606 req = rb_entry(parent, struct ceph_mds_request, r_node);
607 if (new->r_tid < req->r_tid)
608 p = &(*p)->rb_left;
609 else if (new->r_tid > req->r_tid)
610 p = &(*p)->rb_right;
611 else
612 BUG();
613 }
614
615 rb_link_node(&new->r_node, parent, p);
616 rb_insert_color(&new->r_node, &mdsc->request_tree);
2f2dc053
SW		 617}
618
619/*
620 * Register an in-flight request, and assign a tid. Link to directory
621 * are modifying (if any).
622 *
623 * Called under mdsc->mutex.
624 */
625static void __register_request(struct ceph_mds_client *mdsc,
626 struct ceph_mds_request *req,
627 struct inode *dir)
628{
629 req->r_tid = ++mdsc->last_tid;
630 if (req->r_num_caps)
37151668
YS		 631 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
632 req->r_num_caps);
2f2dc053
SW		 633 dout("__register_request %p tid %lld\n", req, req->r_tid);
634 ceph_mdsc_get_request(req);
44ca18f2 635 __insert_request(mdsc, req);
2f2dc053 636
cb4276cc
SW		 637 req->r_uid = current_fsuid();
638 req->r_gid = current_fsgid();
639
e8a7b8b1
YZ		 640 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
641 mdsc->oldest_tid = req->r_tid;
642
2f2dc053 643 if (dir) {
3b663780 644 ihold(dir);
2f2dc053 645 req->r_unsafe_dir = dir;
2f2dc053
SW		 646 }
647}
648
649static void __unregister_request(struct ceph_mds_client *mdsc,
650 struct ceph_mds_request *req)
651{
652 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1
YZ		 653
654 if (req->r_tid == mdsc->oldest_tid) {
655 struct rb_node *p = rb_next(&req->r_node);
656 mdsc->oldest_tid = 0;
657 while (p) {
658 struct ceph_mds_request *next_req =
659 rb_entry(p, struct ceph_mds_request, r_node);
660 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
661 mdsc->oldest_tid = next_req->r_tid;
662 break;
663 }
664 p = rb_next(p);
665 }
666 }
667
44ca18f2 668 rb_erase(&req->r_node, &mdsc->request_tree);
80fc7314 669 RB_CLEAR_NODE(&req->r_node);
2f2dc053 670
4c06ace8 671 if (req->r_unsafe_dir && req->r_got_unsafe) {
2f2dc053 672 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 673 spin_lock(&ci->i_unsafe_lock);
674 list_del_init(&req->r_unsafe_dir_item);
675 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 676 }
68cd5b4b
YZ		 677 if (req->r_target_inode && req->r_got_unsafe) {
678 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
679 spin_lock(&ci->i_unsafe_lock);
680 list_del_init(&req->r_unsafe_target_item);
681 spin_unlock(&ci->i_unsafe_lock);
682 }
3b663780 683
4c06ace8 684 if (req->r_unsafe_dir) {
3b663780
SW		 685 iput(req->r_unsafe_dir);
686 req->r_unsafe_dir = NULL;
2f2dc053 687 }
94aa8ae1 688
fc55d2c9
YZ		 689 complete_all(&req->r_safe_completion);
690
94aa8ae1 691 ceph_mdsc_put_request(req);
2f2dc053
SW		 692}
693
694/*
695 * Choose mds to send request to next. If there is a hint set in the
696 * request (e.g., due to a prior forward hint from the mds), use that.
697 * Otherwise, consult frag tree and/or caps to identify the
698 * appropriate mds. If all else fails, choose randomly.
699 *
700 * Called under mdsc->mutex.
701 */
7fd7d101 702static struct dentry *get_nonsnap_parent(struct dentry *dentry)
eb6bb1c5 703{
d79698da
SW		 704 /*
705 * we don't need to worry about protecting the d_parent access
706 * here because we never renaming inside the snapped namespace
707 * except to resplice to another snapdir, and either the old or new
708 * result is a valid result.
709 */
2b0143b5 710 while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
eb6bb1c5
SW		 711 dentry = dentry->d_parent;
712 return dentry;
713}
714
2f2dc053
SW		 715static int __choose_mds(struct ceph_mds_client *mdsc,
716 struct ceph_mds_request *req)
717{
718 struct inode *inode;
719 struct ceph_inode_info *ci;
720 struct ceph_cap *cap;
721 int mode = req->r_direct_mode;
722 int mds = -1;
723 u32 hash = req->r_direct_hash;
724 bool is_hash = req->r_direct_is_hash;
725
726 /*
727 * is there a specific mds we should try? ignore hint if we have
728 * no session and the mds is not up (active or recovering).
729 */
730 if (req->r_resend_mds >= 0 &&
731 (__have_session(mdsc, req->r_resend_mds) ||
732 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
733 dout("choose_mds using resend_mds mds%d\n",
734 req->r_resend_mds);
735 return req->r_resend_mds;
736 }
737
738 if (mode == USE_RANDOM_MDS)
739 goto random;
740
741 inode = NULL;
742 if (req->r_inode) {
743 inode = req->r_inode;
744 } else if (req->r_dentry) {
d79698da
SW		 745 /* ignore race with rename; old or new d_parent is okay */
746 struct dentry *parent = req->r_dentry->d_parent;
2b0143b5 747 struct inode *dir = d_inode(parent);
eb6bb1c5 748
3d14c5d2 749 if (dir->i_sb != mdsc->fsc->sb) {
eb6bb1c5 750 /* not this fs! */
2b0143b5 751 inode = d_inode(req->r_dentry);
eb6bb1c5
SW		 752 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
753 /* direct snapped/virtual snapdir requests
754 * based on parent dir inode */
d79698da 755 struct dentry *dn = get_nonsnap_parent(parent);
2b0143b5 756 inode = d_inode(dn);
eb6bb1c5 757 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 758 } else {
eb6bb1c5 759 /* dentry target */
2b0143b5 760 inode = d_inode(req->r_dentry);
ca18bede
YZ		 761 if (!inode || mode == USE_AUTH_MDS) {
762 /* dir + name */
763 inode = dir;
764 hash = ceph_dentry_hash(dir, req->r_dentry);
765 is_hash = true;
766 }
2f2dc053
SW		 767 }
768 }
eb6bb1c5 769
2f2dc053
SW		 770 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
771 (int)hash, mode);
772 if (!inode)
773 goto random;
774 ci = ceph_inode(inode);
775
776 if (is_hash && S_ISDIR(inode->i_mode)) {
777 struct ceph_inode_frag frag;
778 int found;
779
780 ceph_choose_frag(ci, hash, &frag, &found);
781 if (found) {
782 if (mode == USE_ANY_MDS && frag.ndist > 0) {
783 u8 r;
784
785 /* choose a random replica */
786 get_random_bytes(&r, 1);
787 r %= frag.ndist;
788 mds = frag.dist[r];
789 dout("choose_mds %p %llx.%llx "
790 "frag %u mds%d (%d/%d)\n",
791 inode, ceph_vinop(inode),
d66bbd44 792 frag.frag, mds,
2f2dc053 793 (int)r, frag.ndist);
d66bbd44
SW		 794 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
795 CEPH_MDS_STATE_ACTIVE)
796 return mds;
2f2dc053
SW		 797 }
798
799 /* since this file/dir wasn't known to be
800 * replicated, then we want to look for the
801 * authoritative mds. */
802 mode = USE_AUTH_MDS;
803 if (frag.mds >= 0) {
804 /* choose auth mds */
805 mds = frag.mds;
806 dout("choose_mds %p %llx.%llx "
807 "frag %u mds%d (auth)\n",
808 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 809 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
810 CEPH_MDS_STATE_ACTIVE)
811 return mds;
2f2dc053
SW		 812 }
813 }
814 }
815
be655596 816 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 817 cap = NULL;
818 if (mode == USE_AUTH_MDS)
819 cap = ci->i_auth_cap;
820 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
821 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
822 if (!cap) {
be655596 823 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 824 goto random;
825 }
826 mds = cap->session->s_mds;
827 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
828 inode, ceph_vinop(inode), mds,
829 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 830 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 831 return mds;
832
833random:
834 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
835 dout("choose_mds chose random mds%d\n", mds);
836 return mds;
837}
838
839
840/*
841 * session messages
842 */
843static struct ceph_msg *create_session_msg(u32 op, u64 seq)
844{
845 struct ceph_msg *msg;
846 struct ceph_mds_session_head *h;
847
b61c2763
SW		 848 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
849 false);
a79832f2 850 if (!msg) {
2f2dc053 851 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 852 return NULL;
2f2dc053
SW		 853 }
854 h = msg->front.iov_base;
855 h->op = cpu_to_le32(op);
856 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 857
858 return msg;
859}
860
861/*
862 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
863 * to include additional client metadata fields.
864 */
865static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
866{
867 struct ceph_msg *msg;
868 struct ceph_mds_session_head *h;
869 int i = -1;
870 int metadata_bytes = 0;
871 int metadata_key_count = 0;
872 struct ceph_options *opt = mdsc->fsc->client->options;
873 void *p;
874
a6a5ce4f 875 const char* metadata[][2] = {
dbd0c8bf 876 {"hostname", utsname()->nodename},
a6a5ce4f 877 {"kernel_version", utsname()->release},
dbd0c8bf
JS		 878 {"entity_id", opt->name ? opt->name : ""},
879 {NULL, NULL}
880 };
881
882 /* Calculate serialized length of metadata */
883 metadata_bytes = 4; /* map length */
884 for (i = 0; metadata[i][0] != NULL; ++i) {
885 metadata_bytes += 8 + strlen(metadata[i][0]) +
886 strlen(metadata[i][1]);
887 metadata_key_count++;
888 }
889
890 /* Allocate the message */
891 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
892 GFP_NOFS, false);
893 if (!msg) {
894 pr_err("create_session_msg ENOMEM creating msg\n");
895 return NULL;
896 }
897 h = msg->front.iov_base;
898 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
899 h->seq = cpu_to_le64(seq);
900
901 /*
902 * Serialize client metadata into waiting buffer space, using
903 * the format that userspace expects for map<string, string>
7cfa0313
JS		 904 *
905 * ClientSession messages with metadata are v2
dbd0c8bf 906 */
7cfa0313
JS		 907 msg->hdr.version = cpu_to_le16(2);
908 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 909
910 /* The write pointer, following the session_head structure */
911 p = msg->front.iov_base + sizeof(*h);
912
913 /* Number of entries in the map */
914 ceph_encode_32(&p, metadata_key_count);
915
916 /* Two length-prefixed strings for each entry in the map */
917 for (i = 0; metadata[i][0] != NULL; ++i) {
918 size_t const key_len = strlen(metadata[i][0]);
919 size_t const val_len = strlen(metadata[i][1]);
920
921 ceph_encode_32(&p, key_len);
922 memcpy(p, metadata[i][0], key_len);
923 p += key_len;
924 ceph_encode_32(&p, val_len);
925 memcpy(p, metadata[i][1], val_len);
926 p += val_len;
927 }
928
2f2dc053
SW		 929 return msg;
930}
931
932/*
933 * send session open request.
934 *
935 * called under mdsc->mutex
936 */
937static int __open_session(struct ceph_mds_client *mdsc,
938 struct ceph_mds_session *session)
939{
940 struct ceph_msg *msg;
941 int mstate;
942 int mds = session->s_mds;
2f2dc053
SW		 943
944 /* wait for mds to go active? */
945 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
946 dout("open_session to mds%d (%s)\n", mds,
947 ceph_mds_state_name(mstate));
948 session->s_state = CEPH_MDS_SESSION_OPENING;
949 session->s_renew_requested = jiffies;
950
951 /* send connect message */
dbd0c8bf 952 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 953 if (!msg)
954 return -ENOMEM;
2f2dc053 955 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 956 return 0;
957}
958
ed0552a1
SW		 959/*
960 * open sessions for any export targets for the given mds
961 *
962 * called under mdsc->mutex
963 */
5d72d13c
YZ		 964static struct ceph_mds_session *
965__open_export_target_session(struct ceph_mds_client *mdsc, int target)
966{
967 struct ceph_mds_session *session;
968
969 session = __ceph_lookup_mds_session(mdsc, target);
970 if (!session) {
971 session = register_session(mdsc, target);
972 if (IS_ERR(session))
973 return session;
974 }
975 if (session->s_state == CEPH_MDS_SESSION_NEW ||
976 session->s_state == CEPH_MDS_SESSION_CLOSING)
977 __open_session(mdsc, session);
978
979 return session;
980}
981
982struct ceph_mds_session *
983ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
984{
985 struct ceph_mds_session *session;
986
987 dout("open_export_target_session to mds%d\n", target);
988
989 mutex_lock(&mdsc->mutex);
990 session = __open_export_target_session(mdsc, target);
991 mutex_unlock(&mdsc->mutex);
992
993 return session;
994}
995
ed0552a1
SW		 996static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
997 struct ceph_mds_session *session)
998{
999 struct ceph_mds_info *mi;
1000 struct ceph_mds_session *ts;
1001 int i, mds = session->s_mds;
ed0552a1
SW		 1002
1003 if (mds >= mdsc->mdsmap->m_max_mds)
1004 return;
5d72d13c 1005
ed0552a1
SW		 1006 mi = &mdsc->mdsmap->m_info[mds];
1007 dout("open_export_target_sessions for mds%d (%d targets)\n",
1008 session->s_mds, mi->num_export_targets);
1009
1010 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1011 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1012 if (!IS_ERR(ts))
1013 ceph_put_mds_session(ts);
ed0552a1
SW		 1014 }
1015}
1016
154f42c2
SW		 1017void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1018 struct ceph_mds_session *session)
1019{
1020 mutex_lock(&mdsc->mutex);
1021 __open_export_target_sessions(mdsc, session);
1022 mutex_unlock(&mdsc->mutex);
1023}
1024
2f2dc053
SW		 1025/*
1026 * session caps
1027 */
1028
745a8e3b
YZ		 1029/* caller holds s_cap_lock, we drop it */
1030static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1031 struct ceph_mds_session *session)
1032 __releases(session->s_cap_lock)
2f2dc053 1033{
745a8e3b
YZ		 1034 LIST_HEAD(tmp_list);
1035 list_splice_init(&session->s_cap_releases, &tmp_list);
1036 session->s_num_cap_releases = 0;
1037 spin_unlock(&session->s_cap_lock);
2f2dc053 1038
745a8e3b
YZ		 1039 dout("cleanup_cap_releases mds%d\n", session->s_mds);
1040 while (!list_empty(&tmp_list)) {
1041 struct ceph_cap *cap;
1042 /* zero out the in-progress message */
1043 cap = list_first_entry(&tmp_list,
1044 struct ceph_cap, session_caps);
1045 list_del(&cap->session_caps);
1046 ceph_put_cap(mdsc, cap);
2f2dc053 1047 }
2f2dc053
SW		 1048}
1049
1c841a96
YZ		 1050static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1051 struct ceph_mds_session *session)
1052{
1053 struct ceph_mds_request *req;
1054 struct rb_node *p;
1055
1056 dout("cleanup_session_requests mds%d\n", session->s_mds);
1057 mutex_lock(&mdsc->mutex);
1058 while (!list_empty(&session->s_unsafe)) {
1059 req = list_first_entry(&session->s_unsafe,
1060 struct ceph_mds_request, r_unsafe_item);
1061 list_del_init(&req->r_unsafe_item);
3e0708b9
YZ		 1062 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1063 req->r_tid);
1c841a96
YZ		 1064 __unregister_request(mdsc, req);
1065 }
1066 /* zero r_attempts, so kick_requests() will re-send requests */
1067 p = rb_first(&mdsc->request_tree);
1068 while (p) {
1069 req = rb_entry(p, struct ceph_mds_request, r_node);
1070 p = rb_next(p);
1071 if (req->r_session &&
1072 req->r_session->s_mds == session->s_mds)
1073 req->r_attempts = 0;
1074 }
1075 mutex_unlock(&mdsc->mutex);
1076}
1077
2f2dc053 1078/*
f818a736
SW		 1079 * Helper to safely iterate over all caps associated with a session, with
1080 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1081 *
f818a736 1082 * Caller must hold session s_mutex.
2f2dc053
SW		 1083 */
1084static int iterate_session_caps(struct ceph_mds_session *session,
1085 int (*cb)(struct inode *, struct ceph_cap *,
1086 void *), void *arg)
1087{
7c1332b8
SW		 1088 struct list_head *p;
1089 struct ceph_cap *cap;
1090 struct inode *inode, *last_inode = NULL;
1091 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1092 int ret;
1093
1094 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1095 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1096 p = session->s_caps.next;
1097 while (p != &session->s_caps) {
1098 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1099 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1100 if (!inode) {
1101 p = p->next;
2f2dc053 1102 continue;
7c1332b8
SW		 1103 }
1104 session->s_cap_iterator = cap;
2f2dc053 1105 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1106
1107 if (last_inode) {
1108 iput(last_inode);
1109 last_inode = NULL;
1110 }
1111 if (old_cap) {
37151668 1112 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1113 old_cap = NULL;
1114 }
1115
2f2dc053 1116 ret = cb(inode, cap, arg);
7c1332b8
SW		 1117 last_inode = inode;
1118
2f2dc053 1119 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1120 p = p->next;
1121 if (cap->ci == NULL) {
1122 dout("iterate_session_caps finishing cap %p removal\n",
1123 cap);
1124 BUG_ON(cap->session != session);
745a8e3b 1125 cap->session = NULL;
7c1332b8
SW		 1126 list_del_init(&cap->session_caps);
1127 session->s_nr_caps--;
745a8e3b
YZ		 1128 if (cap->queue_release) {
1129 list_add_tail(&cap->session_caps,
1130 &session->s_cap_releases);
1131 session->s_num_cap_releases++;
1132 } else {
1133 old_cap = cap; /* put_cap it w/o locks held */
1134 }
7c1332b8 1135 }
5dacf091
SW		 1136 if (ret < 0)
1137 goto out;
2f2dc053 1138 }
5dacf091
SW		 1139 ret = 0;
1140out:
7c1332b8 1141 session->s_cap_iterator = NULL;
2f2dc053 1142 spin_unlock(&session->s_cap_lock);
7c1332b8 1143
e96a650a 1144 iput(last_inode);
7c1332b8 1145 if (old_cap)
37151668 1146 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1147
5dacf091 1148 return ret;
2f2dc053
SW		 1149}
1150
1151static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1152 void *arg)
2f2dc053
SW		 1153{
1154 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1155 LIST_HEAD(to_remove);
6c99f254
SW		 1156 int drop = 0;
1157
2f2dc053
SW		 1158 dout("removing cap %p, ci is %p, inode is %p\n",
1159 cap, ci, &ci->vfs_inode);
be655596 1160 spin_lock(&ci->i_ceph_lock);
a096b09a 1161 __ceph_remove_cap(cap, false);
571ade33 1162 if (!ci->i_auth_cap) {
553adfd9 1163 struct ceph_cap_flush *cf;
6c99f254 1164 struct ceph_mds_client *mdsc =
3d14c5d2 1165 ceph_sb_to_client(inode->i_sb)->mdsc;
6c99f254 1166
553adfd9
YZ		 1167 while (true) {
1168 struct rb_node *n = rb_first(&ci->i_cap_flush_tree);
1169 if (!n)
1170 break;
1171 cf = rb_entry(n, struct ceph_cap_flush, i_node);
1172 rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
1173 list_add(&cf->list, &to_remove);
1174 }
1175
6c99f254 1176 spin_lock(&mdsc->cap_dirty_lock);
8310b089
YZ		 1177
1178 list_for_each_entry(cf, &to_remove, list)
1179 rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
1180
6c99f254 1181 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1182 pr_warn_ratelimited(
1183 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1184 ceph_cap_string(ci->i_dirty_caps),
1185 inode, ceph_ino(inode));
1186 ci->i_dirty_caps = 0;
1187 list_del_init(&ci->i_dirty_item);
1188 drop = 1;
1189 }
1190 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1191 pr_warn_ratelimited(
1192 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1193 ceph_cap_string(ci->i_flushing_caps),
1194 inode, ceph_ino(inode));
1195 ci->i_flushing_caps = 0;
1196 list_del_init(&ci->i_flushing_item);
1197 mdsc->num_cap_flushing--;
1198 drop = 1;
1199 }
6c99f254 1200 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1201
f66fd9f0
YZ		 1202 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1203 list_add(&ci->i_prealloc_cap_flush->list, &to_remove);
1204 ci->i_prealloc_cap_flush = NULL;
1205 }
6c99f254 1206 }
be655596 1207 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1208 while (!list_empty(&to_remove)) {
1209 struct ceph_cap_flush *cf;
1210 cf = list_first_entry(&to_remove,
1211 struct ceph_cap_flush, list);
1212 list_del(&cf->list);
f66fd9f0 1213 ceph_free_cap_flush(cf);
553adfd9 1214 }
6c99f254
SW		 1215 while (drop--)
1216 iput(inode);
2f2dc053
SW		 1217 return 0;
1218}
1219
1220/*
1221 * caller must hold session s_mutex
1222 */
1223static void remove_session_caps(struct ceph_mds_session *session)
1224{
1225 dout("remove_session_caps on %p\n", session);
1226 iterate_session_caps(session, remove_session_caps_cb, NULL);
6f60f889
YZ		 1227
1228 spin_lock(&session->s_cap_lock);
1229 if (session->s_nr_caps > 0) {
1230 struct super_block *sb = session->s_mdsc->fsc->sb;
1231 struct inode *inode;
1232 struct ceph_cap *cap, *prev = NULL;
1233 struct ceph_vino vino;
1234 /*
1235 * iterate_session_caps() skips inodes that are being
1236 * deleted, we need to wait until deletions are complete.
1237 * __wait_on_freeing_inode() is designed for the job,
1238 * but it is not exported, so use lookup inode function
1239 * to access it.
1240 */
1241 while (!list_empty(&session->s_caps)) {
1242 cap = list_entry(session->s_caps.next,
1243 struct ceph_cap, session_caps);
1244 if (cap == prev)
1245 break;
1246 prev = cap;
1247 vino = cap->ci->i_vino;
1248 spin_unlock(&session->s_cap_lock);
1249
ed284c49 1250 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1251 iput(inode);
1252
1253 spin_lock(&session->s_cap_lock);
1254 }
1255 }
745a8e3b
YZ		 1256
1257 // drop cap expires and unlock s_cap_lock
1258 cleanup_cap_releases(session->s_mdsc, session);
6f60f889 1259
2f2dc053 1260 BUG_ON(session->s_nr_caps > 0);
6c99f254 1261 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1262}
1263
1264/*
1265 * wake up any threads waiting on this session's caps. if the cap is
1266 * old (didn't get renewed on the client reconnect), remove it now.
1267 *
1268 * caller must hold s_mutex.
1269 */
1270static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1271 void *arg)
1272{
0dc2570f
SW		 1273 struct ceph_inode_info *ci = ceph_inode(inode);
1274
03066f23 1275 wake_up_all(&ci->i_cap_wq);
0dc2570f 1276 if (arg) {
be655596 1277 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1278 ci->i_wanted_max_size = 0;
1279 ci->i_requested_max_size = 0;
be655596 1280 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1281 }
2f2dc053
SW		 1282 return 0;
1283}
1284
0dc2570f
SW		 1285static void wake_up_session_caps(struct ceph_mds_session *session,
1286 int reconnect)
2f2dc053
SW		 1287{
1288 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1289 iterate_session_caps(session, wake_up_session_cb,
1290 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1291}
1292
1293/*
1294 * Send periodic message to MDS renewing all currently held caps. The
1295 * ack will reset the expiration for all caps from this session.
1296 *
1297 * caller holds s_mutex
1298 */
1299static int send_renew_caps(struct ceph_mds_client *mdsc,
1300 struct ceph_mds_session *session)
1301{
1302 struct ceph_msg *msg;
1303 int state;
1304
1305 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1306 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1307 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1308 session->s_renew_requested = jiffies;
2f2dc053
SW		 1309
1310 /* do not try to renew caps until a recovering mds has reconnected
1311 * with its clients. */
1312 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1313 if (state < CEPH_MDS_STATE_RECONNECT) {
1314 dout("send_renew_caps ignoring mds%d (%s)\n",
1315 session->s_mds, ceph_mds_state_name(state));
1316 return 0;
1317 }
1318
1319 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1320 ceph_mds_state_name(state));
2f2dc053
SW		 1321 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1322 ++session->s_renew_seq);
a79832f2
SW		 1323 if (!msg)
1324 return -ENOMEM;
2f2dc053
SW		 1325 ceph_con_send(&session->s_con, msg);
1326 return 0;
1327}
1328
186e4f7a
YZ		 1329static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1330 struct ceph_mds_session *session, u64 seq)
1331{
1332 struct ceph_msg *msg;
1333
1334 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1335 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1336 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1337 if (!msg)
1338 return -ENOMEM;
1339 ceph_con_send(&session->s_con, msg);
1340 return 0;
1341}
1342
1343
2f2dc053
SW		 1344/*
1345 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1346 *
1347 * Called under session->s_mutex
2f2dc053
SW		 1348 */
1349static void renewed_caps(struct ceph_mds_client *mdsc,
1350 struct ceph_mds_session *session, int is_renew)
1351{
1352 int was_stale;
1353 int wake = 0;
1354
1355 spin_lock(&session->s_cap_lock);
1ce208a6 1356 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1357
1358 session->s_cap_ttl = session->s_renew_requested +
1359 mdsc->mdsmap->m_session_timeout*HZ;
1360
1361 if (was_stale) {
1362 if (time_before(jiffies, session->s_cap_ttl)) {
1363 pr_info("mds%d caps renewed\n", session->s_mds);
1364 wake = 1;
1365 } else {
1366 pr_info("mds%d caps still stale\n", session->s_mds);
1367 }
1368 }
1369 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1370 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1371 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1372 spin_unlock(&session->s_cap_lock);
1373
1374 if (wake)
0dc2570f 1375 wake_up_session_caps(session, 0);
2f2dc053
SW		 1376}
1377
1378/*
1379 * send a session close request
1380 */
1381static int request_close_session(struct ceph_mds_client *mdsc,
1382 struct ceph_mds_session *session)
1383{
1384 struct ceph_msg *msg;
2f2dc053
SW		 1385
1386 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1387 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1388 session->s_seq);
1389 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1390 if (!msg)
1391 return -ENOMEM;
1392 ceph_con_send(&session->s_con, msg);
1393 return 0;
2f2dc053
SW		 1394}
1395
1396/*
1397 * Called with s_mutex held.
1398 */
1399static int __close_session(struct ceph_mds_client *mdsc,
1400 struct ceph_mds_session *session)
1401{
1402 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1403 return 0;
1404 session->s_state = CEPH_MDS_SESSION_CLOSING;
1405 return request_close_session(mdsc, session);
1406}
1407
1408/*
1409 * Trim old(er) caps.
1410 *
1411 * Because we can't cache an inode without one or more caps, we do
1412 * this indirectly: if a cap is unused, we prune its aliases, at which
1413 * point the inode will hopefully get dropped to.
1414 *
1415 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1416 * memory pressure from the MDS, though, so it needn't be perfect.
1417 */
1418static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1419{
1420 struct ceph_mds_session *session = arg;
1421 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1422 int used, wanted, oissued, mine;
2f2dc053
SW		 1423
1424 if (session->s_trim_caps <= 0)
1425 return -1;
1426
be655596 1427 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1428 mine = cap->issued | cap->implemented;
1429 used = __ceph_caps_used(ci);
979abfdd 1430 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1431 oissued = __ceph_caps_issued_other(ci, cap);
1432
979abfdd 1433 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1434 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1435 ceph_cap_string(used), ceph_cap_string(wanted));
1436 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1437 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1438 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1439 goto out;
1440 if ((used | wanted) & CEPH_CAP_ANY_WR)
1441 goto out;
1442 }
5e804ac4
YZ		 1443 /* The inode has cached pages, but it's no longer used.
1444 * we can safely drop it */
1445 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1446 !(oissued & CEPH_CAP_FILE_CACHE)) {
1447 used = 0;
1448 oissued = 0;
1449 }
979abfdd 1450 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1451 goto out; /* we need these caps */
1452
1453 session->s_trim_caps--;
1454 if (oissued) {
1455 /* we aren't the only cap.. just remove us */
a096b09a 1456 __ceph_remove_cap(cap, true);
2f2dc053 1457 } else {
5e804ac4 1458 /* try dropping referring dentries */
be655596 1459 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1460 d_prune_aliases(inode);
1461 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1462 inode, cap, atomic_read(&inode->i_count));
1463 return 0;
1464 }
1465
1466out:
be655596 1467 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1468 return 0;
1469}
1470
1471/*
1472 * Trim session cap count down to some max number.
1473 */
1474static int trim_caps(struct ceph_mds_client *mdsc,
1475 struct ceph_mds_session *session,
1476 int max_caps)
1477{
1478 int trim_caps = session->s_nr_caps - max_caps;
1479
1480 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1481 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1482 if (trim_caps > 0) {
1483 session->s_trim_caps = trim_caps;
1484 iterate_session_caps(session, trim_caps_cb, session);
1485 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1486 session->s_mds, session->s_nr_caps, max_caps,
1487 trim_caps - session->s_trim_caps);
5dacf091 1488 session->s_trim_caps = 0;
2f2dc053 1489 }
a56371d9 1490
a56371d9 1491 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1492 return 0;
1493}
1494
8310b089
YZ		 1495static int check_capsnap_flush(struct ceph_inode_info *ci,
1496 u64 want_snap_seq)
d3383a8e 1497{
8310b089 1498 int ret = 1;
d3383a8e 1499 spin_lock(&ci->i_ceph_lock);
affbc19a
YZ		 1500 if (want_snap_seq > 0 && !list_empty(&ci->i_cap_snaps)) {
1501 struct ceph_cap_snap *capsnap =
1502 list_first_entry(&ci->i_cap_snaps,
1503 struct ceph_cap_snap, ci_item);
8310b089 1504 ret = capsnap->follows >= want_snap_seq;
affbc19a 1505 }
d3383a8e 1506 spin_unlock(&ci->i_ceph_lock);
8310b089
YZ		 1507 return ret;
1508}
1509
1510static int check_caps_flush(struct ceph_mds_client *mdsc,
1511 u64 want_flush_tid)
1512{
1513 struct rb_node *n;
1514 struct ceph_cap_flush *cf;
1515 int ret = 1;
1516
1517 spin_lock(&mdsc->cap_dirty_lock);
1518 n = rb_first(&mdsc->cap_flush_tree);
1519 cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
1520 if (cf && cf->tid <= want_flush_tid) {
1521 dout("check_caps_flush still flushing tid %llu <= %llu\n",
1522 cf->tid, want_flush_tid);
1523 ret = 0;
1524 }
1525 spin_unlock(&mdsc->cap_dirty_lock);
1526 return ret;
d3383a8e
YZ		 1527}
1528
2f2dc053
SW		 1529/*
1530 * flush all dirty inode data to disk.
1531 *
8310b089 1532 * returns true if we've flushed through want_flush_tid
2f2dc053 1533 */
affbc19a 1534static void wait_caps_flush(struct ceph_mds_client *mdsc,
8310b089 1535 u64 want_flush_tid, u64 want_snap_seq)
2f2dc053 1536{
d3383a8e 1537 int mds;
2f2dc053 1538
8310b089
YZ		 1539 dout("check_caps_flush want %llu snap want %llu\n",
1540 want_flush_tid, want_snap_seq);
2f2dc053 1541 mutex_lock(&mdsc->mutex);
affbc19a 1542 for (mds = 0; mds < mdsc->max_sessions; ) {
2f2dc053 1543 struct ceph_mds_session *session = mdsc->sessions[mds];
8310b089 1544 struct inode *inode = NULL;
2f2dc053 1545
affbc19a
YZ		 1546 if (!session) {
1547 mds++;
2f2dc053 1548 continue;
affbc19a 1549 }
2f2dc053
SW		 1550 get_session(session);
1551 mutex_unlock(&mdsc->mutex);
1552
1553 mutex_lock(&session->s_mutex);
affbc19a
YZ		 1554 if (!list_empty(&session->s_cap_snaps_flushing)) {
1555 struct ceph_cap_snap *capsnap =
1556 list_first_entry(&session->s_cap_snaps_flushing,
1557 struct ceph_cap_snap,
1558 flushing_item);
1559 struct ceph_inode_info *ci = capsnap->ci;
8310b089 1560 if (!check_capsnap_flush(ci, want_snap_seq)) {
affbc19a
YZ		 1561 dout("check_cap_flush still flushing snap %p "
1562 "follows %lld <= %lld to mds%d\n",
1563 &ci->vfs_inode, capsnap->follows,
1564 want_snap_seq, mds);
8310b089 1565 inode = igrab(&ci->vfs_inode);
2f2dc053 1566 }
2f2dc053
SW		 1567 }
1568 mutex_unlock(&session->s_mutex);
1569 ceph_put_mds_session(session);
1570
8310b089 1571 if (inode) {
affbc19a 1572 wait_event(mdsc->cap_flushing_wq,
8310b089
YZ		 1573 check_capsnap_flush(ceph_inode(inode),
1574 want_snap_seq));
1575 iput(inode);
1576 } else {
affbc19a 1577 mds++;
8310b089 1578 }
affbc19a 1579
2f2dc053
SW		 1580 mutex_lock(&mdsc->mutex);
1581 }
2f2dc053 1582 mutex_unlock(&mdsc->mutex);
8310b089
YZ		 1583
1584 wait_event(mdsc->cap_flushing_wq,
1585 check_caps_flush(mdsc, want_flush_tid));
1586
1587 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1588}
1589
1590/*
1591 * called under s_mutex
1592 */
3d7ded4d
SW		 1593void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1594 struct ceph_mds_session *session)
2f2dc053 1595{
745a8e3b
YZ		 1596 struct ceph_msg *msg = NULL;
1597 struct ceph_mds_cap_release *head;
1598 struct ceph_mds_cap_item *item;
1599 struct ceph_cap *cap;
1600 LIST_HEAD(tmp_list);
1601 int num_cap_releases;
2f2dc053 1602
0f8605f2 1603 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1604again:
1605 list_splice_init(&session->s_cap_releases, &tmp_list);
1606 num_cap_releases = session->s_num_cap_releases;
1607 session->s_num_cap_releases = 0;
2f2dc053 1608 spin_unlock(&session->s_cap_lock);
e01a5946 1609
745a8e3b
YZ		 1610 while (!list_empty(&tmp_list)) {
1611 if (!msg) {
1612 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1613 PAGE_CACHE_SIZE, GFP_NOFS, false);
1614 if (!msg)
1615 goto out_err;
1616 head = msg->front.iov_base;
1617 head->num = cpu_to_le32(0);
1618 msg->front.iov_len = sizeof(*head);
1619 }
1620 cap = list_first_entry(&tmp_list, struct ceph_cap,
1621 session_caps);
1622 list_del(&cap->session_caps);
1623 num_cap_releases--;
e01a5946 1624
00bd8edb 1625 head = msg->front.iov_base;
745a8e3b
YZ		 1626 le32_add_cpu(&head->num, 1);
1627 item = msg->front.iov_base + msg->front.iov_len;
1628 item->ino = cpu_to_le64(cap->cap_ino);
1629 item->cap_id = cpu_to_le64(cap->cap_id);
1630 item->migrate_seq = cpu_to_le32(cap->mseq);
1631 item->seq = cpu_to_le32(cap->issue_seq);
1632 msg->front.iov_len += sizeof(*item);
1633
1634 ceph_put_cap(mdsc, cap);
1635
1636 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1637 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1638 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1639 ceph_con_send(&session->s_con, msg);
1640 msg = NULL;
1641 }
00bd8edb 1642 }
e01a5946 1643
745a8e3b 1644 BUG_ON(num_cap_releases != 0);
e01a5946 1645
745a8e3b
YZ		 1646 spin_lock(&session->s_cap_lock);
1647 if (!list_empty(&session->s_cap_releases))
1648 goto again;
1649 spin_unlock(&session->s_cap_lock);
1650
1651 if (msg) {
1652 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1653 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1654 ceph_con_send(&session->s_con, msg);
e01a5946 1655 }
745a8e3b
YZ		 1656 return;
1657out_err:
1658 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1659 session->s_mds);
1660 spin_lock(&session->s_cap_lock);
1661 list_splice(&tmp_list, &session->s_cap_releases);
1662 session->s_num_cap_releases += num_cap_releases;
1663 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1664}
1665
2f2dc053
SW		 1666/*
1667 * requests
1668 */
1669
54008399
YZ		 1670int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1671 struct inode *dir)
1672{
1673 struct ceph_inode_info *ci = ceph_inode(dir);
1674 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1675 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1676 size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1677 sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1678 int order, num_entries;
1679
1680 spin_lock(&ci->i_ceph_lock);
1681 num_entries = ci->i_files + ci->i_subdirs;
1682 spin_unlock(&ci->i_ceph_lock);
1683 num_entries = max(num_entries, 1);
1684 num_entries = min(num_entries, opt->max_readdir);
1685
1686 order = get_order(size * num_entries);
1687 while (order >= 0) {
687265e5
YZ		 1688 rinfo->dir_in = (void*)__get_free_pages(GFP_KERNEL |
1689 __GFP_NOWARN,
54008399
YZ		 1690 order);
1691 if (rinfo->dir_in)
1692 break;
1693 order--;
1694 }
1695 if (!rinfo->dir_in)
1696 return -ENOMEM;
1697
1698 num_entries = (PAGE_SIZE << order) / size;
1699 num_entries = min(num_entries, opt->max_readdir);
1700
1701 rinfo->dir_buf_size = PAGE_SIZE << order;
1702 req->r_num_caps = num_entries + 1;
1703 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1704 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1705 return 0;
1706}
1707
2f2dc053
SW		 1708/*
1709 * Create an mds request.
1710 */
1711struct ceph_mds_request *
1712ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1713{
1714 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1715
1716 if (!req)
1717 return ERR_PTR(-ENOMEM);
1718
b4556396 1719 mutex_init(&req->r_fill_mutex);
37151668 1720 req->r_mdsc = mdsc;
2f2dc053
SW		 1721 req->r_started = jiffies;
1722 req->r_resend_mds = -1;
1723 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1724 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1725 req->r_fmode = -1;
153c8e6b 1726 kref_init(&req->r_kref);
2f2dc053
SW		 1727 INIT_LIST_HEAD(&req->r_wait);
1728 init_completion(&req->r_completion);
1729 init_completion(&req->r_safe_completion);
1730 INIT_LIST_HEAD(&req->r_unsafe_item);
1731
b8e69066
SW		 1732 req->r_stamp = CURRENT_TIME;
1733
2f2dc053
SW		 1734 req->r_op = op;
1735 req->r_direct_mode = mode;
1736 return req;
1737}
1738
1739/*
44ca18f2 1740 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1741 *
1742 * called under mdsc->mutex.
1743 */
44ca18f2
SW		 1744static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1745{
1746 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1747 return NULL;
1748 return rb_entry(rb_first(&mdsc->request_tree),
1749 struct ceph_mds_request, r_node);
1750}
1751
e8a7b8b1 1752static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1753{
e8a7b8b1 1754 return mdsc->oldest_tid;
2f2dc053
SW		 1755}
1756
1757/*
1758 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1759 * on build_path_from_dentry in fs/cifs/dir.c.
1760 *
1761 * If @stop_on_nosnap, generate path relative to the first non-snapped
1762 * inode.
1763 *
1764 * Encode hidden .snap dirs as a double /, i.e.
1765 * foo/.snap/bar -> foo//bar
1766 */
1767char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1768 int stop_on_nosnap)
1769{
1770 struct dentry *temp;
1771 char *path;
1772 int len, pos;
1b71fe2e 1773 unsigned seq;
2f2dc053
SW		 1774
1775 if (dentry == NULL)
1776 return ERR_PTR(-EINVAL);
1777
1778retry:
1779 len = 0;
1b71fe2e
AV		 1780 seq = read_seqbegin(&rename_lock);
1781 rcu_read_lock();
2f2dc053 1782 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1783 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1784 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1785 len++; /* slash only */
1786 else if (stop_on_nosnap && inode &&
1787 ceph_snap(inode) == CEPH_NOSNAP)
1788 break;
1789 else
1790 len += 1 + temp->d_name.len;
1791 temp = temp->d_parent;
2f2dc053 1792 }
1b71fe2e 1793 rcu_read_unlock();
2f2dc053
SW		 1794 if (len)
1795 len--; /* no leading '/' */
1796
1797 path = kmalloc(len+1, GFP_NOFS);
1798 if (path == NULL)
1799 return ERR_PTR(-ENOMEM);
1800 pos = len;
1801 path[pos] = 0; /* trailing null */
1b71fe2e 1802 rcu_read_lock();
2f2dc053 1803 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1804 struct inode *inode;
2f2dc053 1805
1b71fe2e 1806 spin_lock(&temp->d_lock);
2b0143b5 1807 inode = d_inode(temp);
2f2dc053 1808 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1809 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1810 pos, temp);
1811 } else if (stop_on_nosnap && inode &&
1812 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1813 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1814 break;
1815 } else {
1816 pos -= temp->d_name.len;
1b71fe2e
AV		 1817 if (pos < 0) {
1818 spin_unlock(&temp->d_lock);
2f2dc053 1819 break;
1b71fe2e 1820 }
2f2dc053
SW		 1821 strncpy(path + pos, temp->d_name.name,
1822 temp->d_name.len);
2f2dc053 1823 }
1b71fe2e 1824 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1825 if (pos)
1826 path[--pos] = '/';
1827 temp = temp->d_parent;
2f2dc053 1828 }
1b71fe2e
AV		 1829 rcu_read_unlock();
1830 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1831 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1832 "expected, namelen is %d, pos is %d\n", len, pos);
1833 /* presumably this is only possible if racing with a
1834 rename of one of the parent directories (we can not
1835 lock the dentries above us to prevent this, but
1836 retrying should be harmless) */
1837 kfree(path);
1838 goto retry;
1839 }
1840
2b0143b5 1841 *base = ceph_ino(d_inode(temp));
2f2dc053 1842 *plen = len;
104648ad 1843 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1844 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1845 return path;
1846}
1847
1848static int build_dentry_path(struct dentry *dentry,
1849 const char **ppath, int *ppathlen, u64 *pino,
1850 int *pfreepath)
1851{
1852 char *path;
1853
2b0143b5
DH		 1854 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1855 *pino = ceph_ino(d_inode(dentry->d_parent));
2f2dc053
SW		 1856 *ppath = dentry->d_name.name;
1857 *ppathlen = dentry->d_name.len;
1858 return 0;
1859 }
1860 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1861 if (IS_ERR(path))
1862 return PTR_ERR(path);
1863 *ppath = path;
1864 *pfreepath = 1;
1865 return 0;
1866}
1867
1868static int build_inode_path(struct inode *inode,
1869 const char **ppath, int *ppathlen, u64 *pino,
1870 int *pfreepath)
1871{
1872 struct dentry *dentry;
1873 char *path;
1874
1875 if (ceph_snap(inode) == CEPH_NOSNAP) {
1876 *pino = ceph_ino(inode);
1877 *ppathlen = 0;
1878 return 0;
1879 }
1880 dentry = d_find_alias(inode);
1881 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1882 dput(dentry);
1883 if (IS_ERR(path))
1884 return PTR_ERR(path);
1885 *ppath = path;
1886 *pfreepath = 1;
1887 return 0;
1888}
1889
1890/*
1891 * request arguments may be specified via an inode *, a dentry *, or
1892 * an explicit ino+path.
1893 */
1894static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1895 const char *rpath, u64 rino,
1896 const char **ppath, int *pathlen,
1897 u64 *ino, int *freepath)
1898{
1899 int r = 0;
1900
1901 if (rinode) {
1902 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1903 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1904 ceph_snap(rinode));
1905 } else if (rdentry) {
1906 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1907 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1908 *ppath);
795858db 1909 } else if (rpath || rino) {
2f2dc053
SW		 1910 *ino = rino;
1911 *ppath = rpath;
b000056a 1912 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1913 dout(" path %.*s\n", *pathlen, rpath);
1914 }
1915
1916 return r;
1917}
1918
1919/*
1920 * called under mdsc->mutex
1921 */
1922static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1923 struct ceph_mds_request *req,
6e6f0923 1924 int mds, bool drop_cap_releases)
2f2dc053
SW		 1925{
1926 struct ceph_msg *msg;
1927 struct ceph_mds_request_head *head;
1928 const char *path1 = NULL;
1929 const char *path2 = NULL;
1930 u64 ino1 = 0, ino2 = 0;
1931 int pathlen1 = 0, pathlen2 = 0;
1932 int freepath1 = 0, freepath2 = 0;
1933 int len;
1934 u16 releases;
1935 void *p, *end;
1936 int ret;
1937
1938 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1939 req->r_path1, req->r_ino1.ino,
1940 &path1, &pathlen1, &ino1, &freepath1);
1941 if (ret < 0) {
1942 msg = ERR_PTR(ret);
1943 goto out;
1944 }
1945
1946 ret = set_request_path_attr(NULL, req->r_old_dentry,
1947 req->r_path2, req->r_ino2.ino,
1948 &path2, &pathlen2, &ino2, &freepath2);
1949 if (ret < 0) {
1950 msg = ERR_PTR(ret);
1951 goto out_free1;
1952 }
1953
1954 len = sizeof(*head) +
b8e69066 1955 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 1956 sizeof(struct ceph_timespec);
2f2dc053
SW		 1957
1958 /* calculate (max) length for cap releases */
1959 len += sizeof(struct ceph_mds_request_release) *
1960 (!!req->r_inode_drop + !!req->r_dentry_drop +
1961 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1962 if (req->r_dentry_drop)
1963 len += req->r_dentry->d_name.len;
1964 if (req->r_old_dentry_drop)
1965 len += req->r_old_dentry->d_name.len;
1966
b61c2763 1967 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1968 if (!msg) {
1969 msg = ERR_PTR(-ENOMEM);
2f2dc053 1970 goto out_free2;
a79832f2 1971 }
2f2dc053 1972
7cfa0313 1973 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1974 msg->hdr.tid = cpu_to_le64(req->r_tid);
1975
2f2dc053
SW		 1976 head = msg->front.iov_base;
1977 p = msg->front.iov_base + sizeof(*head);
1978 end = msg->front.iov_base + msg->front.iov_len;
1979
1980 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1981 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1982 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1983 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1984 head->args = req->r_args;
1985
1986 ceph_encode_filepath(&p, end, ino1, path1);
1987 ceph_encode_filepath(&p, end, ino2, path2);
1988
e979cf50
SW		 1989 /* make note of release offset, in case we need to replay */
1990 req->r_request_release_offset = p - msg->front.iov_base;
1991
2f2dc053
SW		 1992 /* cap releases */
1993 releases = 0;
1994 if (req->r_inode_drop)
1995 releases += ceph_encode_inode_release(&p,
2b0143b5 1996 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 1997 mds, req->r_inode_drop, req->r_inode_unless, 0);
1998 if (req->r_dentry_drop)
1999 releases += ceph_encode_dentry_release(&p, req->r_dentry,
2000 mds, req->r_dentry_drop, req->r_dentry_unless);
2001 if (req->r_old_dentry_drop)
2002 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2003 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
2004 if (req->r_old_inode_drop)
2005 releases += ceph_encode_inode_release(&p,
2b0143b5 2006 d_inode(req->r_old_dentry),
2f2dc053 2007 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 2008
2009 if (drop_cap_releases) {
2010 releases = 0;
2011 p = msg->front.iov_base + req->r_request_release_offset;
2012 }
2013
2f2dc053
SW		 2014 head->num_releases = cpu_to_le16(releases);
2015
b8e69066 2016 /* time stamp */
1f041a89
YZ		 2017 {
2018 struct ceph_timespec ts;
2019 ceph_encode_timespec(&ts, &req->r_stamp);
2020 ceph_encode_copy(&p, &ts, sizeof(ts));
2021 }
b8e69066 2022
2f2dc053
SW		 2023 BUG_ON(p > end);
2024 msg->front.iov_len = p - msg->front.iov_base;
2025 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2026
25e6bae3
YZ		 2027 if (req->r_pagelist) {
2028 struct ceph_pagelist *pagelist = req->r_pagelist;
2029 atomic_inc(&pagelist->refcnt);
2030 ceph_msg_data_add_pagelist(msg, pagelist);
2031 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2032 } else {
2033 msg->hdr.data_len = 0;
ebf18f47 2034 }
02afca6c 2035
2f2dc053
SW		 2036 msg->hdr.data_off = cpu_to_le16(0);
2037
2038out_free2:
2039 if (freepath2)
2040 kfree((char *)path2);
2041out_free1:
2042 if (freepath1)
2043 kfree((char *)path1);
2044out:
2045 return msg;
2046}
2047
2048/*
2049 * called under mdsc->mutex if error, under no mutex if
2050 * success.
2051 */
2052static void complete_request(struct ceph_mds_client *mdsc,
2053 struct ceph_mds_request *req)
2054{
2055 if (req->r_callback)
2056 req->r_callback(mdsc, req);
2057 else
03066f23 2058 complete_all(&req->r_completion);
2f2dc053
SW		 2059}
2060
2061/*
2062 * called under mdsc->mutex
2063 */
2064static int __prepare_send_request(struct ceph_mds_client *mdsc,
2065 struct ceph_mds_request *req,
6e6f0923 2066 int mds, bool drop_cap_releases)
2f2dc053
SW		 2067{
2068 struct ceph_mds_request_head *rhead;
2069 struct ceph_msg *msg;
2070 int flags = 0;
2071
2f2dc053 2072 req->r_attempts++;
e55b71f8
GF		 2073 if (req->r_inode) {
2074 struct ceph_cap *cap =
2075 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2076
2077 if (cap)
2078 req->r_sent_on_mseq = cap->mseq;
2079 else
2080 req->r_sent_on_mseq = -1;
2081 }
2f2dc053
SW		 2082 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2083 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2084
01a92f17 2085 if (req->r_got_unsafe) {
c5c9a0bf 2086 void *p;
01a92f17
SW		 2087 /*
2088 * Replay. Do not regenerate message (and rebuild
2089 * paths, etc.); just use the original message.
2090 * Rebuilding paths will break for renames because
2091 * d_move mangles the src name.
2092 */
2093 msg = req->r_request;
2094 rhead = msg->front.iov_base;
2095
2096 flags = le32_to_cpu(rhead->flags);
2097 flags |= CEPH_MDS_FLAG_REPLAY;
2098 rhead->flags = cpu_to_le32(flags);
2099
2100 if (req->r_target_inode)
2101 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2102
2103 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2104
2105 /* remove cap/dentry releases from message */
2106 rhead->num_releases = 0;
c5c9a0bf
YZ		 2107
2108 /* time stamp */
2109 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2110 {
2111 struct ceph_timespec ts;
2112 ceph_encode_timespec(&ts, &req->r_stamp);
2113 ceph_encode_copy(&p, &ts, sizeof(ts));
2114 }
c5c9a0bf
YZ		 2115
2116 msg->front.iov_len = p - msg->front.iov_base;
2117 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2118 return 0;
2119 }
2120
2f2dc053
SW		 2121 if (req->r_request) {
2122 ceph_msg_put(req->r_request);
2123 req->r_request = NULL;
2124 }
6e6f0923 2125 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2126 if (IS_ERR(msg)) {
e1518c7c 2127 req->r_err = PTR_ERR(msg);
a79832f2 2128 return PTR_ERR(msg);
2f2dc053
SW		 2129 }
2130 req->r_request = msg;
2131
2132 rhead = msg->front.iov_base;
2f2dc053
SW		 2133 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2134 if (req->r_got_unsafe)
2135 flags |= CEPH_MDS_FLAG_REPLAY;
2136 if (req->r_locked_dir)
2137 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2138 rhead->flags = cpu_to_le32(flags);
2139 rhead->num_fwd = req->r_num_fwd;
2140 rhead->num_retry = req->r_attempts - 1;
01a92f17 2141 rhead->ino = 0;
2f2dc053
SW		 2142
2143 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 2144 return 0;
2145}
2146
2147/*
2148 * send request, or put it on the appropriate wait list.
2149 */
2150static int __do_request(struct ceph_mds_client *mdsc,
2151 struct ceph_mds_request *req)
2152{
2153 struct ceph_mds_session *session = NULL;
2154 int mds = -1;
48fec5d0 2155 int err = 0;
2f2dc053 2156
eb1b8af3
YZ		 2157 if (req->r_err || req->r_got_result) {
2158 if (req->r_aborted)
2159 __unregister_request(mdsc, req);
2f2dc053 2160 goto out;
eb1b8af3 2161 }
2f2dc053
SW		 2162
2163 if (req->r_timeout &&
2164 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2165 dout("do_request timed out\n");
2166 err = -EIO;
2167 goto finish;
2168 }
48fec5d0
YZ		 2169 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2170 dout("do_request forced umount\n");
2171 err = -EIO;
2172 goto finish;
2173 }
2f2dc053 2174
dc69e2e9
SW		 2175 put_request_session(req);
2176
2f2dc053
SW		 2177 mds = __choose_mds(mdsc, req);
2178 if (mds < 0 ||
2179 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2180 dout("do_request no mds or not active, waiting for map\n");
2181 list_add(&req->r_wait, &mdsc->waiting_for_map);
2182 goto out;
2183 }
2184
2185 /* get, open session */
2186 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2187 if (!session) {
2f2dc053 2188 session = register_session(mdsc, mds);
9c423956
SW		 2189 if (IS_ERR(session)) {
2190 err = PTR_ERR(session);
2191 goto finish;
2192 }
2193 }
dc69e2e9
SW		 2194 req->r_session = get_session(session);
2195
2f2dc053 2196 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2197 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2198 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2199 session->s_state != CEPH_MDS_SESSION_HUNG) {
2200 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2201 session->s_state == CEPH_MDS_SESSION_CLOSING)
2202 __open_session(mdsc, session);
2203 list_add(&req->r_wait, &session->s_waiting);
2204 goto out_session;
2205 }
2206
2207 /* send request */
2f2dc053
SW		 2208 req->r_resend_mds = -1; /* forget any previous mds hint */
2209
2210 if (req->r_request_started == 0) /* note request start time */
2211 req->r_request_started = jiffies;
2212
6e6f0923 2213 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2214 if (!err) {
2215 ceph_msg_get(req->r_request);
2216 ceph_con_send(&session->s_con, req->r_request);
2217 }
2218
2219out_session:
2220 ceph_put_mds_session(session);
48fec5d0
YZ		 2221finish:
2222 if (err) {
2223 dout("__do_request early error %d\n", err);
2224 req->r_err = err;
2225 complete_request(mdsc, req);
2226 __unregister_request(mdsc, req);
2227 }
2f2dc053
SW		 2228out:
2229 return err;
2f2dc053
SW		 2230}
2231
2232/*
2233 * called under mdsc->mutex
2234 */
2235static void __wake_requests(struct ceph_mds_client *mdsc,
2236 struct list_head *head)
2237{
ed75ec2c
YZ		 2238 struct ceph_mds_request *req;
2239 LIST_HEAD(tmp_list);
2240
2241 list_splice_init(head, &tmp_list);
2f2dc053 2242
ed75ec2c
YZ		 2243 while (!list_empty(&tmp_list)) {
2244 req = list_entry(tmp_list.next,
2245 struct ceph_mds_request, r_wait);
2f2dc053 2246 list_del_init(&req->r_wait);
7971bd92 2247 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2248 __do_request(mdsc, req);
2249 }
2250}
2251
2252/*
2253 * Wake up threads with requests pending for @mds, so that they can
29790f26 2254 * resubmit their requests to a possibly different mds.
2f2dc053 2255 */
29790f26 2256static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2257{
44ca18f2 2258 struct ceph_mds_request *req;
282c1052 2259 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2260
2261 dout("kick_requests mds%d\n", mds);
282c1052 2262 while (p) {
44ca18f2 2263 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2264 p = rb_next(p);
44ca18f2
SW		 2265 if (req->r_got_unsafe)
2266 continue;
3de22be6
YZ		 2267 if (req->r_attempts > 0)
2268 continue; /* only new requests */
44ca18f2
SW		 2269 if (req->r_session &&
2270 req->r_session->s_mds == mds) {
2271 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2272 list_del_init(&req->r_wait);
44ca18f2 2273 __do_request(mdsc, req);
2f2dc053
SW		 2274 }
2275 }
2276}
2277
2278void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2279 struct ceph_mds_request *req)
2280{
2281 dout("submit_request on %p\n", req);
2282 mutex_lock(&mdsc->mutex);
2283 __register_request(mdsc, req, NULL);
2284 __do_request(mdsc, req);
2285 mutex_unlock(&mdsc->mutex);
2286}
2287
2288/*
2289 * Synchrously perform an mds request. Take care of all of the
2290 * session setup, forwarding, retry details.
2291 */
2292int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2293 struct inode *dir,
2294 struct ceph_mds_request *req)
2295{
2296 int err;
2297
2298 dout("do_request on %p\n", req);
2299
2300 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2301 if (req->r_inode)
2302 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2303 if (req->r_locked_dir)
2304 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
844d87c3 2305 if (req->r_old_dentry_dir)
41b02e1f
SW		 2306 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2307 CEPH_CAP_PIN);
2f2dc053 2308
5ea5c5e0
YZ		 2309 /* deny access to directories with pool_ns layouts */
2310 if (req->r_inode && S_ISDIR(req->r_inode->i_mode) &&
2311 ceph_inode(req->r_inode)->i_pool_ns_len)
2312 return -EIO;
2313 if (req->r_locked_dir &&
2314 ceph_inode(req->r_locked_dir)->i_pool_ns_len)
2315 return -EIO;
2316
2f2dc053
SW		 2317 /* issue */
2318 mutex_lock(&mdsc->mutex);
2319 __register_request(mdsc, req, dir);
2320 __do_request(mdsc, req);
2321
e1518c7c
SW		 2322 if (req->r_err) {
2323 err = req->r_err;
e1518c7c 2324 goto out;
2f2dc053
SW		 2325 }
2326
e1518c7c
SW		 2327 /* wait */
2328 mutex_unlock(&mdsc->mutex);
2329 dout("do_request waiting\n");
5be73034 2330 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2331 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2332 } else {
5be73034
ID		 2333 long timeleft = wait_for_completion_killable_timeout(
2334 &req->r_completion,
2335 ceph_timeout_jiffies(req->r_timeout));
2336 if (timeleft > 0)
2337 err = 0;
2338 else if (!timeleft)
2339 err = -EIO; /* timed out */
2340 else
2341 err = timeleft; /* killed */
e1518c7c
SW		 2342 }
2343 dout("do_request waited, got %d\n", err);
2344 mutex_lock(&mdsc->mutex);
5b1daecd 2345
e1518c7c
SW		 2346 /* only abort if we didn't race with a real reply */
2347 if (req->r_got_result) {
2348 err = le32_to_cpu(req->r_reply_info.head->result);
2349 } else if (err < 0) {
2350 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2351
2352 /*
2353 * ensure we aren't running concurrently with
2354 * ceph_fill_trace or ceph_readdir_prepopulate, which
2355 * rely on locks (dir mutex) held by our caller.
2356 */
2357 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 2358 req->r_err = err;
2359 req->r_aborted = true;
b4556396 2360 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2361
e1518c7c 2362 if (req->r_locked_dir &&
167c9e35
SW		 2363 (req->r_op & CEPH_MDS_OP_WRITE))
2364 ceph_invalidate_dir_request(req);
2f2dc053 2365 } else {
e1518c7c 2366 err = req->r_err;
2f2dc053 2367 }
2f2dc053 2368
e1518c7c
SW		 2369out:
2370 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2371 dout("do_request %p done, result %d\n", req, err);
2372 return err;
2373}
2374
167c9e35 2375/*
2f276c51 2376 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2377 * namespace request.
2378 */
2379void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2380{
2381 struct inode *inode = req->r_locked_dir;
167c9e35 2382
2f276c51 2383 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2384
2f276c51 2385 ceph_dir_clear_complete(inode);
167c9e35
SW		 2386 if (req->r_dentry)
2387 ceph_invalidate_dentry_lease(req->r_dentry);
2388 if (req->r_old_dentry)
2389 ceph_invalidate_dentry_lease(req->r_old_dentry);
2390}
2391
2f2dc053
SW		 2392/*
2393 * Handle mds reply.
2394 *
2395 * We take the session mutex and parse and process the reply immediately.
2396 * This preserves the logical ordering of replies, capabilities, etc., sent
2397 * by the MDS as they are applied to our local cache.
2398 */
2399static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2400{
2401 struct ceph_mds_client *mdsc = session->s_mdsc;
2402 struct ceph_mds_request *req;
2403 struct ceph_mds_reply_head *head = msg->front.iov_base;
2404 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2405 struct ceph_snap_realm *realm;
2f2dc053
SW		 2406 u64 tid;
2407 int err, result;
2600d2dd 2408 int mds = session->s_mds;
2f2dc053 2409
2f2dc053
SW		 2410 if (msg->front.iov_len < sizeof(*head)) {
2411 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2412 ceph_msg_dump(msg);
2f2dc053
SW		 2413 return;
2414 }
2415
2416 /* get request, session */
6df058c0 2417 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2418 mutex_lock(&mdsc->mutex);
2419 req = __lookup_request(mdsc, tid);
2420 if (!req) {
2421 dout("handle_reply on unknown tid %llu\n", tid);
2422 mutex_unlock(&mdsc->mutex);
2423 return;
2424 }
2425 dout("handle_reply %p\n", req);
2f2dc053
SW		 2426
2427 /* correct session? */
d96d6049 2428 if (req->r_session != session) {
2f2dc053
SW		 2429 pr_err("mdsc_handle_reply got %llu on session mds%d"
2430 " not mds%d\n", tid, session->s_mds,
2431 req->r_session ? req->r_session->s_mds : -1);
2432 mutex_unlock(&mdsc->mutex);
2433 goto out;
2434 }
2435
2436 /* dup? */
2437 if ((req->r_got_unsafe && !head->safe) ||
2438 (req->r_got_safe && head->safe)) {
f3ae1b97 2439 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2440 head->safe ? "safe" : "unsafe", tid, mds);
2441 mutex_unlock(&mdsc->mutex);
2442 goto out;
2443 }
1550d34e 2444 if (req->r_got_safe) {
f3ae1b97 2445 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2446 tid, mds);
2447 mutex_unlock(&mdsc->mutex);
2448 goto out;
2449 }
2f2dc053
SW		 2450
2451 result = le32_to_cpu(head->result);
2452
2453 /*
e55b71f8
GF		 2454 * Handle an ESTALE
2455 * if we're not talking to the authority, send to them
2456 * if the authority has changed while we weren't looking,
2457 * send to new authority
2458 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2459 */
2460 if (result == -ESTALE) {
e55b71f8 2461 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2462 req->r_resend_mds = -1;
ca18bede 2463 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2464 dout("not using auth, setting for that now");
2465 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2466 __do_request(mdsc, req);
2467 mutex_unlock(&mdsc->mutex);
2468 goto out;
e55b71f8 2469 } else {
ca18bede
YZ		 2470 int mds = __choose_mds(mdsc, req);
2471 if (mds >= 0 && mds != req->r_session->s_mds) {
2472 dout("but auth changed, so resending");
e55b71f8
GF		 2473 __do_request(mdsc, req);
2474 mutex_unlock(&mdsc->mutex);
2475 goto out;
2476 }
2f2dc053 2477 }
e55b71f8 2478 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2479 }
2480
e55b71f8 2481
2f2dc053
SW		 2482 if (head->safe) {
2483 req->r_got_safe = true;
2484 __unregister_request(mdsc, req);
2f2dc053
SW		 2485
2486 if (req->r_got_unsafe) {
2487 /*
2488 * We already handled the unsafe response, now do the
2489 * cleanup. No need to examine the response; the MDS
2490 * doesn't include any result info in the safe
2491 * response. And even if it did, there is nothing
2492 * useful we could do with a revised return value.
2493 */
2494 dout("got safe reply %llu, mds%d\n", tid, mds);
2495 list_del_init(&req->r_unsafe_item);
2496
2497 /* last unsafe request during umount? */
44ca18f2 2498 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2499 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2500 mutex_unlock(&mdsc->mutex);
2501 goto out;
2502 }
e1518c7c 2503 } else {
2f2dc053
SW		 2504 req->r_got_unsafe = true;
2505 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2506 if (req->r_unsafe_dir) {
2507 struct ceph_inode_info *ci =
2508 ceph_inode(req->r_unsafe_dir);
2509 spin_lock(&ci->i_unsafe_lock);
2510 list_add_tail(&req->r_unsafe_dir_item,
2511 &ci->i_unsafe_dirops);
2512 spin_unlock(&ci->i_unsafe_lock);
2513 }
2f2dc053
SW		 2514 }
2515
2516 dout("handle_reply tid %lld result %d\n", tid, result);
2517 rinfo = &req->r_reply_info;
14303d20 2518 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2519 mutex_unlock(&mdsc->mutex);
2520
2521 mutex_lock(&session->s_mutex);
2522 if (err < 0) {
25933abd 2523 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2524 ceph_msg_dump(msg);
2f2dc053
SW		 2525 goto out_err;
2526 }
2527
2528 /* snap trace */
982d6011 2529 realm = NULL;
2f2dc053
SW		 2530 if (rinfo->snapblob_len) {
2531 down_write(&mdsc->snap_rwsem);
2532 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2533 rinfo->snapblob + rinfo->snapblob_len,
2534 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2535 &realm);
2f2dc053
SW		 2536 downgrade_write(&mdsc->snap_rwsem);
2537 } else {
2538 down_read(&mdsc->snap_rwsem);
2539 }
2540
2541 /* insert trace into our cache */
b4556396 2542 mutex_lock(&req->r_fill_mutex);
3d14c5d2 2543 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053 2544 if (err == 0) {
6e8575fa 2545 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2546 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2547 ceph_readdir_prepopulate(req, req->r_session);
37151668 2548 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2549 }
b4556396 2550 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2551
2552 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2553 if (realm)
2554 ceph_put_snap_realm(mdsc, realm);
68cd5b4b
YZ		 2555
2556 if (err == 0 && req->r_got_unsafe && req->r_target_inode) {
2557 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2558 spin_lock(&ci->i_unsafe_lock);
2559 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2560 spin_unlock(&ci->i_unsafe_lock);
2561 }
2f2dc053 2562out_err:
e1518c7c
SW		 2563 mutex_lock(&mdsc->mutex);
2564 if (!req->r_aborted) {
2565 if (err) {
2566 req->r_err = err;
2567 } else {
5fdb1389 2568 req->r_reply = ceph_msg_get(msg);
e1518c7c
SW		 2569 req->r_got_result = true;
2570 }
2f2dc053 2571 } else {
e1518c7c 2572 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2573 }
e1518c7c 2574 mutex_unlock(&mdsc->mutex);
2f2dc053 2575
2f2dc053
SW		 2576 mutex_unlock(&session->s_mutex);
2577
2578 /* kick calling process */
2579 complete_request(mdsc, req);
2580out:
2581 ceph_mdsc_put_request(req);
2582 return;
2583}
2584
2585
2586
2587/*
2588 * handle mds notification that our request has been forwarded.
2589 */
2600d2dd
SW		 2590static void handle_forward(struct ceph_mds_client *mdsc,
2591 struct ceph_mds_session *session,
2592 struct ceph_msg *msg)
2f2dc053
SW		 2593{
2594 struct ceph_mds_request *req;
a1ea787c 2595 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2596 u32 next_mds;
2597 u32 fwd_seq;
2f2dc053
SW		 2598 int err = -EINVAL;
2599 void *p = msg->front.iov_base;
2600 void *end = p + msg->front.iov_len;
2f2dc053 2601
a1ea787c 2602 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2603 next_mds = ceph_decode_32(&p);
2604 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2605
2606 mutex_lock(&mdsc->mutex);
2607 req = __lookup_request(mdsc, tid);
2608 if (!req) {
2a8e5e36 2609 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2610 goto out; /* dup reply? */
2611 }
2612
2a8e5e36
SW		 2613 if (req->r_aborted) {
2614 dout("forward tid %llu aborted, unregistering\n", tid);
2615 __unregister_request(mdsc, req);
2616 } else if (fwd_seq <= req->r_num_fwd) {
2617 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2618 tid, next_mds, req->r_num_fwd, fwd_seq);
2619 } else {
2620 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2621 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2622 BUG_ON(req->r_err);
2623 BUG_ON(req->r_got_result);
3de22be6 2624 req->r_attempts = 0;
2f2dc053
SW		 2625 req->r_num_fwd = fwd_seq;
2626 req->r_resend_mds = next_mds;
2627 put_request_session(req);
2628 __do_request(mdsc, req);
2629 }
2630 ceph_mdsc_put_request(req);
2631out:
2632 mutex_unlock(&mdsc->mutex);
2633 return;
2634
2635bad:
2636 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2637}
2638
2639/*
2640 * handle a mds session control message
2641 */
2642static void handle_session(struct ceph_mds_session *session,
2643 struct ceph_msg *msg)
2644{
2645 struct ceph_mds_client *mdsc = session->s_mdsc;
2646 u32 op;
2647 u64 seq;
2600d2dd 2648 int mds = session->s_mds;
2f2dc053
SW		 2649 struct ceph_mds_session_head *h = msg->front.iov_base;
2650 int wake = 0;
2651
2f2dc053
SW		 2652 /* decode */
2653 if (msg->front.iov_len != sizeof(*h))
2654 goto bad;
2655 op = le32_to_cpu(h->op);
2656 seq = le64_to_cpu(h->seq);
2657
2658 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2659 if (op == CEPH_SESSION_CLOSE)
2660 __unregister_session(mdsc, session);
2f2dc053
SW		 2661 /* FIXME: this ttl calculation is generous */
2662 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2663 mutex_unlock(&mdsc->mutex);
2664
2665 mutex_lock(&session->s_mutex);
2666
2667 dout("handle_session mds%d %s %p state %s seq %llu\n",
2668 mds, ceph_session_op_name(op), session,
a687ecaf 2669 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2670
2671 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2672 session->s_state = CEPH_MDS_SESSION_OPEN;
2673 pr_info("mds%d came back\n", session->s_mds);
2674 }
2675
2676 switch (op) {
2677 case CEPH_SESSION_OPEN:
29790f26
SW		 2678 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2679 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2680 session->s_state = CEPH_MDS_SESSION_OPEN;
2681 renewed_caps(mdsc, session, 0);
2682 wake = 1;
2683 if (mdsc->stopping)
2684 __close_session(mdsc, session);
2685 break;
2686
2687 case CEPH_SESSION_RENEWCAPS:
2688 if (session->s_renew_seq == seq)
2689 renewed_caps(mdsc, session, 1);
2690 break;
2691
2692 case CEPH_SESSION_CLOSE:
29790f26
SW		 2693 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2694 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2695 cleanup_session_requests(mdsc, session);
2f2dc053 2696 remove_session_caps(session);
656e4382 2697 wake = 2; /* for good measure */
f3c60c59 2698 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2699 break;
2700
2701 case CEPH_SESSION_STALE:
2702 pr_info("mds%d caps went stale, renewing\n",
2703 session->s_mds);
d8fb02ab 2704 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2705 session->s_cap_gen++;
1ce208a6 2706 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2707 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2708 send_renew_caps(mdsc, session);
2709 break;
2710
2711 case CEPH_SESSION_RECALL_STATE:
2712 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2713 break;
2714
186e4f7a
YZ		 2715 case CEPH_SESSION_FLUSHMSG:
2716 send_flushmsg_ack(mdsc, session, seq);
2717 break;
2718
03f4fcb0
YZ		 2719 case CEPH_SESSION_FORCE_RO:
2720 dout("force_session_readonly %p\n", session);
2721 spin_lock(&session->s_cap_lock);
2722 session->s_readonly = true;
2723 spin_unlock(&session->s_cap_lock);
2724 wake_up_session_caps(session, 0);
2725 break;
2726
2f2dc053
SW		 2727 default:
2728 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2729 WARN_ON(1);
2730 }
2731
2732 mutex_unlock(&session->s_mutex);
2733 if (wake) {
2734 mutex_lock(&mdsc->mutex);
2735 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2736 if (wake == 2)
2737 kick_requests(mdsc, mds);
2f2dc053
SW		 2738 mutex_unlock(&mdsc->mutex);
2739 }
2740 return;
2741
2742bad:
2743 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2744 (int)msg->front.iov_len);
9ec7cab1 2745 ceph_msg_dump(msg);
2f2dc053
SW		 2746 return;
2747}
2748
2749
2750/*
2751 * called under session->mutex.
2752 */
2753static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2754 struct ceph_mds_session *session)
2755{
2756 struct ceph_mds_request *req, *nreq;
3de22be6 2757 struct rb_node *p;
2f2dc053
SW		 2758 int err;
2759
2760 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2761
2762 mutex_lock(&mdsc->mutex);
2763 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2764 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2765 if (!err) {
2766 ceph_msg_get(req->r_request);
2767 ceph_con_send(&session->s_con, req->r_request);
2768 }
2769 }
3de22be6
YZ		 2770
2771 /*
2772 * also re-send old requests when MDS enters reconnect stage. So that MDS
2773 * can process completed request in clientreplay stage.
2774 */
2775 p = rb_first(&mdsc->request_tree);
2776 while (p) {
2777 req = rb_entry(p, struct ceph_mds_request, r_node);
2778 p = rb_next(p);
2779 if (req->r_got_unsafe)
2780 continue;
2781 if (req->r_attempts == 0)
2782 continue; /* only old requests */
2783 if (req->r_session &&
2784 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2785 err = __prepare_send_request(mdsc, req,
2786 session->s_mds, true);
3de22be6
YZ		 2787 if (!err) {
2788 ceph_msg_get(req->r_request);
2789 ceph_con_send(&session->s_con, req->r_request);
2790 }
2791 }
2792 }
2f2dc053
SW		 2793 mutex_unlock(&mdsc->mutex);
2794}
2795
2796/*
2797 * Encode information about a cap for a reconnect with the MDS.
2798 */
2f2dc053
SW		 2799static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2800 void *arg)
2801{
20cb34ae
SW		 2802 union {
2803 struct ceph_mds_cap_reconnect v2;
2804 struct ceph_mds_cap_reconnect_v1 v1;
2805 } rec;
2806 size_t reclen;
2f2dc053 2807 struct ceph_inode_info *ci;
20cb34ae
SW		 2808 struct ceph_reconnect_state *recon_state = arg;
2809 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2810 char *path;
2811 int pathlen, err;
2812 u64 pathbase;
2813 struct dentry *dentry;
2814
2815 ci = cap->ci;
2816
2817 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2818 inode, ceph_vinop(inode), cap, cap->cap_id,
2819 ceph_cap_string(cap->issued));
93cea5be
SW		 2820 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2821 if (err)
2822 return err;
2f2dc053
SW		 2823
2824 dentry = d_find_alias(inode);
2825 if (dentry) {
2826 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2827 if (IS_ERR(path)) {
2828 err = PTR_ERR(path);
e072f8aa 2829 goto out_dput;
2f2dc053
SW		 2830 }
2831 } else {
2832 path = NULL;
2833 pathlen = 0;
2834 }
93cea5be
SW		 2835 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2836 if (err)
e072f8aa 2837 goto out_free;
2f2dc053 2838
be655596 2839 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2840 cap->seq = 0; /* reset cap seq */
2841 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2842 cap->mseq = 0; /* and migrate_seq */
99a9c273 2843 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae
SW		 2844
2845 if (recon_state->flock) {
2846 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2847 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2848 rec.v2.issued = cpu_to_le32(cap->issued);
2849 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2850 rec.v2.pathbase = cpu_to_le64(pathbase);
2851 rec.v2.flock_len = 0;
2852 reclen = sizeof(rec.v2);
2853 } else {
2854 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2855 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2856 rec.v1.issued = cpu_to_le32(cap->issued);
2857 rec.v1.size = cpu_to_le64(inode->i_size);
2858 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2859 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2860 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2861 rec.v1.pathbase = cpu_to_le64(pathbase);
2862 reclen = sizeof(rec.v1);
2863 }
be655596 2864 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2865
40819f6f
GF		 2866 if (recon_state->flock) {
2867 int num_fcntl_locks, num_flock_locks;
39be95e9
JS		 2868 struct ceph_filelock *flocks;
2869
2870encode_again:
39be95e9 2871 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
39be95e9
JS		 2872 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2873 sizeof(struct ceph_filelock), GFP_NOFS);
2874 if (!flocks) {
2875 err = -ENOMEM;
2876 goto out_free;
2877 }
39be95e9
JS		 2878 err = ceph_encode_locks_to_buffer(inode, flocks,
2879 num_fcntl_locks,
2880 num_flock_locks);
39be95e9
JS		 2881 if (err) {
2882 kfree(flocks);
2883 if (err == -ENOSPC)
2884 goto encode_again;
2885 goto out_free;
2886 }
2887 /*
2888 * number of encoded locks is stable, so copy to pagelist
2889 */
2890 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2891 (num_fcntl_locks+num_flock_locks) *
2892 sizeof(struct ceph_filelock));
2893 err = ceph_pagelist_append(pagelist, &rec, reclen);
2894 if (!err)
2895 err = ceph_locks_to_pagelist(flocks, pagelist,
2896 num_fcntl_locks,
2897 num_flock_locks);
2898 kfree(flocks);
3612abbd
SW		 2899 } else {
2900 err = ceph_pagelist_append(pagelist, &rec, reclen);
40819f6f 2901 }
44c99757
YZ		 2902
2903 recon_state->nr_caps++;
e072f8aa 2904out_free:
2f2dc053 2905 kfree(path);
e072f8aa 2906out_dput:
2f2dc053 2907 dput(dentry);
93cea5be 2908 return err;
2f2dc053
SW		 2909}
2910
2911
2912/*
2913 * If an MDS fails and recovers, clients need to reconnect in order to
2914 * reestablish shared state. This includes all caps issued through
2915 * this session _and_ the snap_realm hierarchy. Because it's not
2916 * clear which snap realms the mds cares about, we send everything we
2917 * know about.. that ensures we'll then get any new info the
2918 * recovering MDS might have.
2919 *
2920 * This is a relatively heavyweight operation, but it's rare.
2921 *
2922 * called with mdsc->mutex held.
2923 */
34b6c855
SW		 2924static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2925 struct ceph_mds_session *session)
2f2dc053 2926{
2f2dc053 2927 struct ceph_msg *reply;
a105f00c 2928 struct rb_node *p;
34b6c855 2929 int mds = session->s_mds;
9abf82b8 2930 int err = -ENOMEM;
44c99757 2931 int s_nr_caps;
93cea5be 2932 struct ceph_pagelist *pagelist;
20cb34ae 2933 struct ceph_reconnect_state recon_state;
2f2dc053 2934
34b6c855 2935 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2936
93cea5be
SW		 2937 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2938 if (!pagelist)
2939 goto fail_nopagelist;
2940 ceph_pagelist_init(pagelist);
2941
b61c2763 2942 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 2943 if (!reply)
93cea5be 2944 goto fail_nomsg;
93cea5be 2945
34b6c855
SW		 2946 mutex_lock(&session->s_mutex);
2947 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2948 session->s_seq = 0;
2f2dc053 2949
2f2dc053 2950 dout("session %p state %s\n", session,
a687ecaf 2951 ceph_session_state_name(session->s_state));
2f2dc053 2952
99a9c273
YZ		 2953 spin_lock(&session->s_gen_ttl_lock);
2954 session->s_cap_gen++;
2955 spin_unlock(&session->s_gen_ttl_lock);
2956
2957 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 2958 /* don't know if session is readonly */
2959 session->s_readonly = 0;
99a9c273
YZ		 2960 /*
2961 * notify __ceph_remove_cap() that we are composing cap reconnect.
2962 * If a cap get released before being added to the cap reconnect,
2963 * __ceph_remove_cap() should skip queuing cap release.
2964 */
2965 session->s_cap_reconnect = 1;
e01a5946 2966 /* drop old cap expires; we're about to reestablish that state */
745a8e3b 2967 cleanup_cap_releases(mdsc, session);
e01a5946 2968
5d23371f 2969 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 2970 if (mdsc->fsc->sb->s_root)
2971 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 2972
2973 ceph_con_close(&session->s_con);
2974 ceph_con_open(&session->s_con,
2975 CEPH_ENTITY_TYPE_MDS, mds,
2976 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2977
2978 /* replay unsafe requests */
2979 replay_unsafe_requests(mdsc, session);
2980
2981 down_read(&mdsc->snap_rwsem);
2982
2f2dc053 2983 /* traverse this session's caps */
44c99757
YZ		 2984 s_nr_caps = session->s_nr_caps;
2985 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 2986 if (err)
2987 goto fail;
20cb34ae 2988
44c99757 2989 recon_state.nr_caps = 0;
20cb34ae
SW		 2990 recon_state.pagelist = pagelist;
2991 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2992 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2993 if (err < 0)
9abf82b8 2994 goto fail;
2f2dc053 2995
99a9c273
YZ		 2996 spin_lock(&session->s_cap_lock);
2997 session->s_cap_reconnect = 0;
2998 spin_unlock(&session->s_cap_lock);
2999
2f2dc053
SW		 3000 /*
3001 * snaprealms. we provide mds with the ino, seq (version), and
3002 * parent for all of our realms. If the mds has any newer info,
3003 * it will tell us.
3004 */
a105f00c
SW		 3005 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3006 struct ceph_snap_realm *realm =
3007 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3008 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3009
3010 dout(" adding snap realm %llx seq %lld parent %llx\n",
3011 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3012 sr_rec.ino = cpu_to_le64(realm->ino);
3013 sr_rec.seq = cpu_to_le64(realm->seq);
3014 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3015 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3016 if (err)
3017 goto fail;
2f2dc053 3018 }
2f2dc053 3019
20cb34ae
SW		 3020 if (recon_state.flock)
3021 reply->hdr.version = cpu_to_le16(2);
44c99757
YZ		 3022
3023 /* raced with cap release? */
3024 if (s_nr_caps != recon_state.nr_caps) {
3025 struct page *page = list_first_entry(&pagelist->head,
3026 struct page, lru);
3027 __le32 *addr = kmap_atomic(page);
3028 *addr = cpu_to_le32(recon_state.nr_caps);
3029 kunmap_atomic(addr);
ebf18f47 3030 }
44c99757
YZ		 3031
3032 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3033 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3034
3035 ceph_early_kick_flushing_caps(mdsc, session);
3036
2f2dc053
SW		 3037 ceph_con_send(&session->s_con, reply);
3038
9abf82b8
SW		 3039 mutex_unlock(&session->s_mutex);
3040
3041 mutex_lock(&mdsc->mutex);
3042 __wake_requests(mdsc, &session->s_waiting);
3043 mutex_unlock(&mdsc->mutex);
3044
2f2dc053 3045 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3046 return;
3047
93cea5be 3048fail:
2f2dc053 3049 ceph_msg_put(reply);
9abf82b8
SW		 3050 up_read(&mdsc->snap_rwsem);
3051 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3052fail_nomsg:
3053 ceph_pagelist_release(pagelist);
93cea5be 3054fail_nopagelist:
9abf82b8 3055 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3056 return;
2f2dc053
SW		 3057}
3058
3059
3060/*
3061 * compare old and new mdsmaps, kicking requests
3062 * and closing out old connections as necessary
3063 *
3064 * called under mdsc->mutex.
3065 */
3066static void check_new_map(struct ceph_mds_client *mdsc,
3067 struct ceph_mdsmap *newmap,
3068 struct ceph_mdsmap *oldmap)
3069{
3070 int i;
3071 int oldstate, newstate;
3072 struct ceph_mds_session *s;
3073
3074 dout("check_new_map new %u old %u\n",
3075 newmap->m_epoch, oldmap->m_epoch);
3076
3077 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
3078 if (mdsc->sessions[i] == NULL)
3079 continue;
3080 s = mdsc->sessions[i];
3081 oldstate = ceph_mdsmap_get_state(oldmap, i);
3082 newstate = ceph_mdsmap_get_state(newmap, i);
3083
0deb01c9 3084 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3085 i, ceph_mds_state_name(oldstate),
0deb01c9 3086 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3087 ceph_mds_state_name(newstate),
0deb01c9 3088 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3089 ceph_session_state_name(s->s_state));
2f2dc053 3090
3e8f43a0
YZ		 3091 if (i >= newmap->m_max_mds ||
3092 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3093 ceph_mdsmap_get_addr(newmap, i),
3094 sizeof(struct ceph_entity_addr))) {
3095 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3096 /* the session never opened, just close it
3097 * out now */
3098 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 3099 __unregister_session(mdsc, s);
2f2dc053
SW		 3100 } else {
3101 /* just close it */
3102 mutex_unlock(&mdsc->mutex);
3103 mutex_lock(&s->s_mutex);
3104 mutex_lock(&mdsc->mutex);
3105 ceph_con_close(&s->s_con);
3106 mutex_unlock(&s->s_mutex);
3107 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3108 }
2f2dc053
SW		 3109 } else if (oldstate == newstate) {
3110 continue; /* nothing new with this mds */
3111 }
3112
3113 /*
3114 * send reconnect?
3115 */
3116 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3117 newstate >= CEPH_MDS_STATE_RECONNECT) {
3118 mutex_unlock(&mdsc->mutex);
3119 send_mds_reconnect(mdsc, s);
3120 mutex_lock(&mdsc->mutex);
3121 }
2f2dc053
SW		 3122
3123 /*
29790f26 3124 * kick request on any mds that has gone active.
2f2dc053
SW		 3125 */
3126 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3127 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3128 if (oldstate != CEPH_MDS_STATE_CREATING &&
3129 oldstate != CEPH_MDS_STATE_STARTING)
3130 pr_info("mds%d recovery completed\n", s->s_mds);
3131 kick_requests(mdsc, i);
2f2dc053 3132 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3133 wake_up_session_caps(s, 1);
2f2dc053
SW		 3134 }
3135 }
cb170a22
SW		 3136
3137 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3138 s = mdsc->sessions[i];
3139 if (!s)
3140 continue;
3141 if (!ceph_mdsmap_is_laggy(newmap, i))
3142 continue;
3143 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3144 s->s_state == CEPH_MDS_SESSION_HUNG ||
3145 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3146 dout(" connecting to export targets of laggy mds%d\n",
3147 i);
3148 __open_export_target_sessions(mdsc, s);
3149 }
3150 }
2f2dc053
SW		 3151}
3152
3153
3154
3155/*
3156 * leases
3157 */
3158
3159/*
3160 * caller must hold session s_mutex, dentry->d_lock
3161 */
3162void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3163{
3164 struct ceph_dentry_info *di = ceph_dentry(dentry);
3165
3166 ceph_put_mds_session(di->lease_session);
3167 di->lease_session = NULL;
3168}
3169
2600d2dd
SW		 3170static void handle_lease(struct ceph_mds_client *mdsc,
3171 struct ceph_mds_session *session,
3172 struct ceph_msg *msg)
2f2dc053 3173{
3d14c5d2 3174 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3175 struct inode *inode;
2f2dc053
SW		 3176 struct dentry *parent, *dentry;
3177 struct ceph_dentry_info *di;
2600d2dd 3178 int mds = session->s_mds;
2f2dc053 3179 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3180 u32 seq;
2f2dc053 3181 struct ceph_vino vino;
2f2dc053
SW		 3182 struct qstr dname;
3183 int release = 0;
3184
2f2dc053
SW		 3185 dout("handle_lease from mds%d\n", mds);
3186
3187 /* decode */
3188 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3189 goto bad;
3190 vino.ino = le64_to_cpu(h->ino);
3191 vino.snap = CEPH_NOSNAP;
1e5ea23d 3192 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3193 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3194 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3195 if (dname.len != get_unaligned_le32(h+1))
3196 goto bad;
3197
2f2dc053
SW		 3198 /* lookup inode */
3199 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3200 dout("handle_lease %s, ino %llx %p %.*s\n",
3201 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3202 dname.len, dname.name);
6cd3bcad
YZ		 3203
3204 mutex_lock(&session->s_mutex);
3205 session->s_seq++;
3206
2f2dc053
SW		 3207 if (inode == NULL) {
3208 dout("handle_lease no inode %llx\n", vino.ino);
3209 goto release;
3210 }
2f2dc053
SW		 3211
3212 /* dentry */
3213 parent = d_find_alias(inode);
3214 if (!parent) {
3215 dout("no parent dentry on inode %p\n", inode);
3216 WARN_ON(1);
3217 goto release; /* hrm... */
3218 }
3219 dname.hash = full_name_hash(dname.name, dname.len);
3220 dentry = d_lookup(parent, &dname);
3221 dput(parent);
3222 if (!dentry)
3223 goto release;
3224
3225 spin_lock(&dentry->d_lock);
3226 di = ceph_dentry(dentry);
3227 switch (h->action) {
3228 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3229 if (di->lease_session == session) {
1e5ea23d
SW		 3230 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3231 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3232 __ceph_mdsc_drop_dentry_lease(dentry);
3233 }
3234 release = 1;
3235 break;
3236
3237 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3238 if (di->lease_session == session &&
2f2dc053
SW		 3239 di->lease_gen == session->s_cap_gen &&
3240 di->lease_renew_from &&
3241 di->lease_renew_after == 0) {
3242 unsigned long duration =
3563dbdd 3243 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3244
1e5ea23d 3245 di->lease_seq = seq;
2f2dc053
SW		 3246 dentry->d_time = di->lease_renew_from + duration;
3247 di->lease_renew_after = di->lease_renew_from +
3248 (duration >> 1);
3249 di->lease_renew_from = 0;
3250 }
3251 break;
3252 }
3253 spin_unlock(&dentry->d_lock);
3254 dput(dentry);
3255
3256 if (!release)
3257 goto out;
3258
3259release:
3260 /* let's just reuse the same message */
3261 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3262 ceph_msg_get(msg);
3263 ceph_con_send(&session->s_con, msg);
3264
3265out:
3266 iput(inode);
3267 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3268 return;
3269
3270bad:
3271 pr_err("corrupt lease message\n");
9ec7cab1 3272 ceph_msg_dump(msg);
2f2dc053
SW		 3273}
3274
3275void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3276 struct inode *inode,
3277 struct dentry *dentry, char action,
3278 u32 seq)
3279{
3280 struct ceph_msg *msg;
3281 struct ceph_mds_lease *lease;
3282 int len = sizeof(*lease) + sizeof(u32);
3283 int dnamelen = 0;
3284
3285 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3286 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3287 dnamelen = dentry->d_name.len;
3288 len += dnamelen;
3289
b61c2763 3290 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3291 if (!msg)
2f2dc053
SW		 3292 return;
3293 lease = msg->front.iov_base;
3294 lease->action = action;
2f2dc053
SW		 3295 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3296 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3297 lease->seq = cpu_to_le32(seq);
3298 put_unaligned_le32(dnamelen, lease + 1);
3299 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3300
3301 /*
3302 * if this is a preemptive lease RELEASE, no need to
3303 * flush request stream, since the actual request will
3304 * soon follow.
3305 */
3306 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3307
3308 ceph_con_send(&session->s_con, msg);
3309}
3310
3311/*
3312 * Preemptively release a lease we expect to invalidate anyway.
3313 * Pass @inode always, @dentry is optional.
3314 */
3315void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2f90b852 3316 struct dentry *dentry)
2f2dc053
SW		 3317{
3318 struct ceph_dentry_info *di;
3319 struct ceph_mds_session *session;
3320 u32 seq;
3321
3322 BUG_ON(inode == NULL);
3323 BUG_ON(dentry == NULL);
2f2dc053
SW		 3324
3325 /* is dentry lease valid? */
3326 spin_lock(&dentry->d_lock);
3327 di = ceph_dentry(dentry);
3328 if (!di || !di->lease_session ||
3329 di->lease_session->s_mds < 0 ||
3330 di->lease_gen != di->lease_session->s_cap_gen ||
3331 !time_before(jiffies, dentry->d_time)) {
3332 dout("lease_release inode %p dentry %p -- "
2f90b852
SW		 3333 "no lease\n",
3334 inode, dentry);
2f2dc053
SW		 3335 spin_unlock(&dentry->d_lock);
3336 return;
3337 }
3338
3339 /* we do have a lease on this dentry; note mds and seq */
3340 session = ceph_get_mds_session(di->lease_session);
3341 seq = di->lease_seq;
3342 __ceph_mdsc_drop_dentry_lease(dentry);
3343 spin_unlock(&dentry->d_lock);
3344
2f90b852
SW		 3345 dout("lease_release inode %p dentry %p to mds%d\n",
3346 inode, dentry, session->s_mds);
2f2dc053
SW		 3347 ceph_mdsc_lease_send_msg(session, inode, dentry,
3348 CEPH_MDS_LEASE_RELEASE, seq);
3349 ceph_put_mds_session(session);
3350}
3351
3352/*
3353 * drop all leases (and dentry refs) in preparation for umount
3354 */
3355static void drop_leases(struct ceph_mds_client *mdsc)
3356{
3357 int i;
3358
3359 dout("drop_leases\n");
3360 mutex_lock(&mdsc->mutex);
3361 for (i = 0; i < mdsc->max_sessions; i++) {
3362 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3363 if (!s)
3364 continue;
3365 mutex_unlock(&mdsc->mutex);
3366 mutex_lock(&s->s_mutex);
3367 mutex_unlock(&s->s_mutex);
3368 ceph_put_mds_session(s);
3369 mutex_lock(&mdsc->mutex);
3370 }
3371 mutex_unlock(&mdsc->mutex);
3372}
3373
3374
3375
3376/*
3377 * delayed work -- periodically trim expired leases, renew caps with mds
3378 */
3379static void schedule_delayed(struct ceph_mds_client *mdsc)
3380{
3381 int delay = 5;
3382 unsigned hz = round_jiffies_relative(HZ * delay);
3383 schedule_delayed_work(&mdsc->delayed_work, hz);
3384}
3385
3386static void delayed_work(struct work_struct *work)
3387{
3388 int i;
3389 struct ceph_mds_client *mdsc =
3390 container_of(work, struct ceph_mds_client, delayed_work.work);
3391 int renew_interval;
3392 int renew_caps;
3393
3394 dout("mdsc delayed_work\n");
afcdaea3 3395 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3396
3397 mutex_lock(&mdsc->mutex);
3398 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3399 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3400 mdsc->last_renew_caps);
3401 if (renew_caps)
3402 mdsc->last_renew_caps = jiffies;
3403
3404 for (i = 0; i < mdsc->max_sessions; i++) {
3405 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3406 if (s == NULL)
3407 continue;
3408 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3409 dout("resending session close request for mds%d\n",
3410 s->s_mds);
3411 request_close_session(mdsc, s);
3412 ceph_put_mds_session(s);
3413 continue;
3414 }
3415 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3416 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3417 s->s_state = CEPH_MDS_SESSION_HUNG;
3418 pr_info("mds%d hung\n", s->s_mds);
3419 }
3420 }
3421 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3422 /* this mds is failed or recovering, just wait */
3423 ceph_put_mds_session(s);
3424 continue;
3425 }
3426 mutex_unlock(&mdsc->mutex);
3427
3428 mutex_lock(&s->s_mutex);
3429 if (renew_caps)
3430 send_renew_caps(mdsc, s);
3431 else
3432 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3433 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3434 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3435 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3436 mutex_unlock(&s->s_mutex);
3437 ceph_put_mds_session(s);
3438
3439 mutex_lock(&mdsc->mutex);
3440 }
3441 mutex_unlock(&mdsc->mutex);
3442
3443 schedule_delayed(mdsc);
3444}
3445
3d14c5d2 3446int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3447
2f2dc053 3448{
3d14c5d2
YS		 3449 struct ceph_mds_client *mdsc;
3450
3451 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3452 if (!mdsc)
3453 return -ENOMEM;
3454 mdsc->fsc = fsc;
3455 fsc->mdsc = mdsc;
2f2dc053
SW		 3456 mutex_init(&mdsc->mutex);
3457 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3458 if (mdsc->mdsmap == NULL) {
3459 kfree(mdsc);
2d06eeb8 3460 return -ENOMEM;
fb3101b6 3461 }
2d06eeb8 3462
2f2dc053 3463 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3464 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3465 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3466 mdsc->sessions = NULL;
86d8f67b 3467 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3468 mdsc->max_sessions = 0;
3469 mdsc->stopping = 0;
affbc19a 3470 mdsc->last_snap_seq = 0;
2f2dc053 3471 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3472 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3473 INIT_LIST_HEAD(&mdsc->snap_empty);
3474 spin_lock_init(&mdsc->snap_empty_lock);
3475 mdsc->last_tid = 0;
e8a7b8b1 3476 mdsc->oldest_tid = 0;
44ca18f2 3477 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3478 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3479 mdsc->last_renew_caps = jiffies;
3480 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3481 spin_lock_init(&mdsc->cap_delay_lock);
3482 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3483 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3484 mdsc->last_cap_flush_tid = 1;
8310b089 3485 mdsc->cap_flush_tree = RB_ROOT;
2f2dc053 3486 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3487 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3488 mdsc->num_cap_flushing = 0;
3489 spin_lock_init(&mdsc->cap_dirty_lock);
3490 init_waitqueue_head(&mdsc->cap_flushing_wq);
3491 spin_lock_init(&mdsc->dentry_lru_lock);
3492 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3493
37151668 3494 ceph_caps_init(mdsc);
3d14c5d2 3495 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3496
10183a69
YZ		 3497 init_rwsem(&mdsc->pool_perm_rwsem);
3498 mdsc->pool_perm_tree = RB_ROOT;
3499
5f44f142 3500 return 0;
2f2dc053
SW		 3501}
3502
3503/*
3504 * Wait for safe replies on open mds requests. If we time out, drop
3505 * all requests from the tree to avoid dangling dentry refs.
3506 */
3507static void wait_requests(struct ceph_mds_client *mdsc)
3508{
a319bf56 3509 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3510 struct ceph_mds_request *req;
2f2dc053
SW		 3511
3512 mutex_lock(&mdsc->mutex);
44ca18f2 3513 if (__get_oldest_req(mdsc)) {
2f2dc053 3514 mutex_unlock(&mdsc->mutex);
44ca18f2 3515
2f2dc053
SW		 3516 dout("wait_requests waiting for requests\n");
3517 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3518 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3519
3520 /* tear down remaining requests */
44ca18f2
SW		 3521 mutex_lock(&mdsc->mutex);
3522 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3523 dout("wait_requests timed out on tid %llu\n",
3524 req->r_tid);
44ca18f2 3525 __unregister_request(mdsc, req);
2f2dc053
SW		 3526 }
3527 }
3528 mutex_unlock(&mdsc->mutex);
3529 dout("wait_requests done\n");
3530}
3531
3532/*
3533 * called before mount is ro, and before dentries are torn down.
3534 * (hmm, does this still race with new lookups?)
3535 */
3536void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3537{
3538 dout("pre_umount\n");
3539 mdsc->stopping = 1;
3540
3541 drop_leases(mdsc);
afcdaea3 3542 ceph_flush_dirty_caps(mdsc);
2f2dc053 3543 wait_requests(mdsc);
17c688c3
SW		 3544
3545 /*
3546 * wait for reply handlers to drop their request refs and
3547 * their inode/dcache refs
3548 */
3549 ceph_msgr_flush();
2f2dc053
SW		 3550}
3551
3552/*
3553 * wait for all write mds requests to flush.
3554 */
3555static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3556{
80fc7314 3557 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3558 struct rb_node *n;
2f2dc053
SW		 3559
3560 mutex_lock(&mdsc->mutex);
3561 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3562restart:
44ca18f2
SW		 3563 req = __get_oldest_req(mdsc);
3564 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3565 /* find next request */
3566 n = rb_next(&req->r_node);
3567 if (n)
3568 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3569 else
3570 nextreq = NULL;
e8a7b8b1
YZ		 3571 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3572 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3573 /* write op */
3574 ceph_mdsc_get_request(req);
80fc7314
SW		 3575 if (nextreq)
3576 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3577 mutex_unlock(&mdsc->mutex);
3578 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3579 req->r_tid, want_tid);
3580 wait_for_completion(&req->r_safe_completion);
3581 mutex_lock(&mdsc->mutex);
44ca18f2 3582 ceph_mdsc_put_request(req);
80fc7314
SW		 3583 if (!nextreq)
3584 break; /* next dne before, so we're done! */
3585 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3586 /* next request was removed from tree */
3587 ceph_mdsc_put_request(nextreq);
3588 goto restart;
3589 }
3590 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3591 }
80fc7314 3592 req = nextreq;
2f2dc053
SW		 3593 }
3594 mutex_unlock(&mdsc->mutex);
3595 dout("wait_unsafe_requests done\n");
3596}
3597
3598void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3599{
affbc19a 3600 u64 want_tid, want_flush, want_snap;
2f2dc053 3601
48fec5d0 3602 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3603 return;
3604
2f2dc053
SW		 3605 dout("sync\n");
3606 mutex_lock(&mdsc->mutex);
3607 want_tid = mdsc->last_tid;
2f2dc053 3608 mutex_unlock(&mdsc->mutex);
2f2dc053 3609
afcdaea3 3610 ceph_flush_dirty_caps(mdsc);
d3383a8e 3611 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3612 want_flush = mdsc->last_cap_flush_tid;
d3383a8e
YZ		 3613 spin_unlock(&mdsc->cap_dirty_lock);
3614
affbc19a
YZ		 3615 down_read(&mdsc->snap_rwsem);
3616 want_snap = mdsc->last_snap_seq;
3617 up_read(&mdsc->snap_rwsem);
3618
3619 dout("sync want tid %lld flush_seq %lld snap_seq %lld\n",
3620 want_tid, want_flush, want_snap);
2f2dc053
SW		 3621
3622 wait_unsafe_requests(mdsc, want_tid);
affbc19a 3623 wait_caps_flush(mdsc, want_flush, want_snap);
2f2dc053
SW		 3624}
3625
f3c60c59
SW		 3626/*
3627 * true if all sessions are closed, or we force unmount
3628 */
7fd7d101 3629static bool done_closing_sessions(struct ceph_mds_client *mdsc)
f3c60c59 3630{
48fec5d0 3631 if (ACCESS_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3632 return true;
86d8f67b 3633 return atomic_read(&mdsc->num_sessions) == 0;
f3c60c59 3634}
2f2dc053
SW		 3635
3636/*
3637 * called after sb is ro.
3638 */
3639void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3640{
a319bf56 3641 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3642 struct ceph_mds_session *session;
3643 int i;
2f2dc053
SW		 3644
3645 dout("close_sessions\n");
3646
2f2dc053 3647 /* close sessions */
f3c60c59
SW		 3648 mutex_lock(&mdsc->mutex);
3649 for (i = 0; i < mdsc->max_sessions; i++) {
3650 session = __ceph_lookup_mds_session(mdsc, i);
3651 if (!session)
3652 continue;
2f2dc053 3653 mutex_unlock(&mdsc->mutex);
f3c60c59
SW		 3654 mutex_lock(&session->s_mutex);
3655 __close_session(mdsc, session);
3656 mutex_unlock(&session->s_mutex);
3657 ceph_put_mds_session(session);
2f2dc053
SW		 3658 mutex_lock(&mdsc->mutex);
3659 }
f3c60c59
SW		 3660 mutex_unlock(&mdsc->mutex);
3661
3662 dout("waiting for sessions to close\n");
3663 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
a319bf56 3664 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3665
3666 /* tear down remaining sessions */
f3c60c59 3667 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3668 for (i = 0; i < mdsc->max_sessions; i++) {
3669 if (mdsc->sessions[i]) {
3670 session = get_session(mdsc->sessions[i]);
2600d2dd 3671 __unregister_session(mdsc, session);
2f2dc053
SW		 3672 mutex_unlock(&mdsc->mutex);
3673 mutex_lock(&session->s_mutex);
3674 remove_session_caps(session);
3675 mutex_unlock(&session->s_mutex);
3676 ceph_put_mds_session(session);
3677 mutex_lock(&mdsc->mutex);
3678 }
3679 }
2f2dc053 3680 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3681 mutex_unlock(&mdsc->mutex);
3682
3683 ceph_cleanup_empty_realms(mdsc);
3684
3685 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3686
3687 dout("stopped\n");
3688}
3689
48fec5d0
YZ		 3690void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3691{
3692 struct ceph_mds_session *session;
3693 int mds;
3694
3695 dout("force umount\n");
3696
3697 mutex_lock(&mdsc->mutex);
3698 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3699 session = __ceph_lookup_mds_session(mdsc, mds);
3700 if (!session)
3701 continue;
3702 mutex_unlock(&mdsc->mutex);
3703 mutex_lock(&session->s_mutex);
3704 __close_session(mdsc, session);
3705 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3706 cleanup_session_requests(mdsc, session);
3707 remove_session_caps(session);
3708 }
3709 mutex_unlock(&session->s_mutex);
3710 ceph_put_mds_session(session);
3711 mutex_lock(&mdsc->mutex);
3712 kick_requests(mdsc, mds);
3713 }
3714 __wake_requests(mdsc, &mdsc->waiting_for_map);
3715 mutex_unlock(&mdsc->mutex);
3716}
3717
3d14c5d2 3718static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3719{
3720 dout("stop\n");
3721 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3722 if (mdsc->mdsmap)
3723 ceph_mdsmap_destroy(mdsc->mdsmap);
3724 kfree(mdsc->sessions);
37151668 3725 ceph_caps_finalize(mdsc);
10183a69 3726 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3727}
3728
3d14c5d2
YS		 3729void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3730{
3731 struct ceph_mds_client *mdsc = fsc->mdsc;
3732
ef550f6f 3733 dout("mdsc_destroy %p\n", mdsc);
3d14c5d2 3734 ceph_mdsc_stop(mdsc);
ef550f6f
SW		 3735
3736 /* flush out any connection work with references to us */
3737 ceph_msgr_flush();
3738
3d14c5d2
YS		 3739 fsc->mdsc = NULL;
3740 kfree(mdsc);
ef550f6f 3741 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3742}
3743
2f2dc053
SW		 3744
3745/*
3746 * handle mds map update.
3747 */
3748void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3749{
3750 u32 epoch;
3751 u32 maplen;
3752 void *p = msg->front.iov_base;
3753 void *end = p + msg->front.iov_len;
3754 struct ceph_mdsmap *newmap, *oldmap;
3755 struct ceph_fsid fsid;
3756 int err = -EINVAL;
3757
3758 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3759 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3760 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3761 return;
c89136ea
SW		 3762 epoch = ceph_decode_32(&p);
3763 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3764 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3765
3766 /* do we need it? */
2f2dc053
SW		 3767 mutex_lock(&mdsc->mutex);
3768 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3769 dout("handle_map epoch %u <= our %u\n",
3770 epoch, mdsc->mdsmap->m_epoch);
3771 mutex_unlock(&mdsc->mutex);
3772 return;
3773 }
3774
3775 newmap = ceph_mdsmap_decode(&p, end);
3776 if (IS_ERR(newmap)) {
3777 err = PTR_ERR(newmap);
3778 goto bad_unlock;
3779 }
3780
3781 /* swap into place */
3782 if (mdsc->mdsmap) {
3783 oldmap = mdsc->mdsmap;
3784 mdsc->mdsmap = newmap;
3785 check_new_map(mdsc, newmap, oldmap);
3786 ceph_mdsmap_destroy(oldmap);
3787 } else {
3788 mdsc->mdsmap = newmap; /* first mds map */
3789 }
3d14c5d2 3790 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3791
3792 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 3793 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3794 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 3795
3796 mutex_unlock(&mdsc->mutex);
3797 schedule_delayed(mdsc);
3798 return;
3799
3800bad_unlock:
3801 mutex_unlock(&mdsc->mutex);
3802bad:
3803 pr_err("error decoding mdsmap %d\n", err);
3804 return;
3805}
3806
3807static struct ceph_connection *con_get(struct ceph_connection *con)
3808{
3809 struct ceph_mds_session *s = con->private;
3810
3811 if (get_session(s)) {
2600d2dd 3812 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3813 return con;
3814 }
3815 dout("mdsc con_get %p FAIL\n", s);
3816 return NULL;
3817}
3818
3819static void con_put(struct ceph_connection *con)
3820{
3821 struct ceph_mds_session *s = con->private;
3822
7d8e18a6 3823 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
2f2dc053
SW		 3824 ceph_put_mds_session(s);
3825}
3826
3827/*
3828 * if the client is unresponsive for long enough, the mds will kill
3829 * the session entirely.
3830 */
3831static void peer_reset(struct ceph_connection *con)
3832{
3833 struct ceph_mds_session *s = con->private;
7e70f0ed 3834 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3835
f3ae1b97 3836 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3837 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3838}
3839
3840static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3841{
3842 struct ceph_mds_session *s = con->private;
3843 struct ceph_mds_client *mdsc = s->s_mdsc;
3844 int type = le16_to_cpu(msg->hdr.type);
3845
2600d2dd
SW		 3846 mutex_lock(&mdsc->mutex);
3847 if (__verify_registered_session(mdsc, s) < 0) {
3848 mutex_unlock(&mdsc->mutex);
3849 goto out;
3850 }
3851 mutex_unlock(&mdsc->mutex);
3852
2f2dc053
SW		 3853 switch (type) {
3854 case CEPH_MSG_MDS_MAP:
3855 ceph_mdsc_handle_map(mdsc, msg);
3856 break;
3857 case CEPH_MSG_CLIENT_SESSION:
3858 handle_session(s, msg);
3859 break;
3860 case CEPH_MSG_CLIENT_REPLY:
3861 handle_reply(s, msg);
3862 break;
3863 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3864 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3865 break;
3866 case CEPH_MSG_CLIENT_CAPS:
3867 ceph_handle_caps(s, msg);
3868 break;
3869 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3870 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3871 break;
3872 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3873 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3874 break;
3875
3876 default:
3877 pr_err("received unknown message type %d %s\n", type,
3878 ceph_msg_type_name(type));
3879 }
2600d2dd 3880out:
2f2dc053
SW		 3881 ceph_msg_put(msg);
3882}
3883
4e7a5dcd
SW		 3884/*
3885 * authentication
3886 */
a3530df3
AE		 3887
3888/*
3889 * Note: returned pointer is the address of a structure that's
3890 * managed separately. Caller must *not* attempt to free it.
3891 */
3892static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3893 int *proto, int force_new)
4e7a5dcd
SW		 3894{
3895 struct ceph_mds_session *s = con->private;
3896 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3897 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 3898 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 3899
74f1869f 3900 if (force_new && auth->authorizer) {
27859f97 3901 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f 3902 auth->authorizer = NULL;
4e7a5dcd 3903 }
27859f97
SW		 3904 if (!auth->authorizer) {
3905 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3906 auth);
0bed9b5c
SW		 3907 if (ret)
3908 return ERR_PTR(ret);
27859f97
SW		 3909 } else {
3910 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3911 auth);
a255651d 3912 if (ret)
a3530df3 3913 return ERR_PTR(ret);
4e7a5dcd 3914 }
4e7a5dcd 3915 *proto = ac->protocol;
74f1869f 3916
a3530df3 3917 return auth;
4e7a5dcd
SW		 3918}
3919
3920
3921static int verify_authorizer_reply(struct ceph_connection *con, int len)
3922{
3923 struct ceph_mds_session *s = con->private;
3924 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3925 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 3926
27859f97 3927 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
4e7a5dcd
SW		 3928}
3929
9bd2e6f8
SW		 3930static int invalidate_authorizer(struct ceph_connection *con)
3931{
3932 struct ceph_mds_session *s = con->private;
3933 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3934 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 3935
27859f97 3936 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 3937
3d14c5d2 3938 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3939}
3940
53ded495
AE		 3941static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3942 struct ceph_msg_header *hdr, int *skip)
3943{
3944 struct ceph_msg *msg;
3945 int type = (int) le16_to_cpu(hdr->type);
3946 int front_len = (int) le32_to_cpu(hdr->front_len);
3947
3948 if (con->in_msg)
3949 return con->in_msg;
3950
3951 *skip = 0;
3952 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3953 if (!msg) {
3954 pr_err("unable to allocate msg type %d len %d\n",
3955 type, front_len);
3956 return NULL;
3957 }
53ded495
AE		 3958
3959 return msg;
3960}
3961
79dbd1ba 3962static int mds_sign_message(struct ceph_msg *msg)
33d07337 3963{
79dbd1ba 3964 struct ceph_mds_session *s = msg->con->private;
33d07337 3965 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 3966
33d07337
YZ		 3967 return ceph_auth_sign_message(auth, msg);
3968}
3969
79dbd1ba 3970static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 3971{
79dbd1ba 3972 struct ceph_mds_session *s = msg->con->private;
33d07337 3973 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 3974
33d07337
YZ		 3975 return ceph_auth_check_message_signature(auth, msg);
3976}
3977
9e32789f 3978static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 3979 .get = con_get,
3980 .put = con_put,
3981 .dispatch = dispatch,
4e7a5dcd
SW		 3982 .get_authorizer = get_authorizer,
3983 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3984 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 3985 .peer_reset = peer_reset,
53ded495 3986 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 3987 .sign_message = mds_sign_message,
3988 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 3989};
3990
2f2dc053 3991/* eof */
Linux kernel - Deliverables
This page took 0.545413 seconds and 5 git commands to generate.