]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/ceph/mds_client.c
projects / mirror_ubuntu-eoan-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: don't use CEPH_OSD_FLAG_ORDERSNAP
[mirror_ubuntu-eoan-kernel.git] / fs / ceph / mds_client.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
2f2dc053 2
496e5955 3#include <linux/fs.h>
2f2dc053 4#include <linux/wait.h>
5a0e3ad6 5#include <linux/slab.h>
54008399 6#include <linux/gfp.h>
2f2dc053 7#include <linux/sched.h>
3d14c5d2
YS		 8#include <linux/debugfs.h>
9#include <linux/seq_file.h>
dbd0c8bf 10#include <linux/utsname.h>
3e0708b9 11#include <linux/ratelimit.h>
2f2dc053 12
2f2dc053 13#include "super.h"
3d14c5d2
YS		 14#include "mds_client.h"
15
1fe60e51 16#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/pagelist.h>
20#include <linux/ceph/auth.h>
21#include <linux/ceph/debugfs.h>
2f2dc053
SW		 22
23/*
24 * A cluster of MDS (metadata server) daemons is responsible for
25 * managing the file system namespace (the directory hierarchy and
26 * inodes) and for coordinating shared access to storage. Metadata is
27 * partitioning hierarchically across a number of servers, and that
28 * partition varies over time as the cluster adjusts the distribution
29 * in order to balance load.
30 *
31 * The MDS client is primarily responsible to managing synchronous
32 * metadata requests for operations like open, unlink, and so forth.
33 * If there is a MDS failure, we find out about it when we (possibly
34 * request and) receive a new MDS map, and can resubmit affected
35 * requests.
36 *
37 * For the most part, though, we take advantage of a lossless
38 * communications channel to the MDS, and do not need to worry about
39 * timing out or resubmitting requests.
40 *
41 * We maintain a stateful "session" with each MDS we interact with.
42 * Within each session, we sent periodic heartbeat messages to ensure
43 * any capabilities or leases we have been issues remain valid. If
44 * the session times out and goes stale, our leases and capabilities
45 * are no longer valid.
46 */
47
20cb34ae 48struct ceph_reconnect_state {
44c99757 49 int nr_caps;
20cb34ae 50 struct ceph_pagelist *pagelist;
121f22a1 51 unsigned msg_version;
20cb34ae
SW		 52};
53
2f2dc053
SW		 54static void __wake_requests(struct ceph_mds_client *mdsc,
55 struct list_head *head);
56
9e32789f 57static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 58
59
60/*
61 * mds reply parsing
62 */
63
64/*
65 * parse individual inode info
66 */
67static int parse_reply_info_in(void **p, void *end,
14303d20 68 struct ceph_mds_reply_info_in *info,
12b4629a 69 u64 features)
2f2dc053
SW		 70{
71 int err = -EIO;
72
73 info->in = *p;
74 *p += sizeof(struct ceph_mds_reply_inode) +
75 sizeof(*info->in->fragtree.splits) *
76 le32_to_cpu(info->in->fragtree.nsplits);
77
78 ceph_decode_32_safe(p, end, info->symlink_len, bad);
79 ceph_decode_need(p, end, info->symlink_len, bad);
80 info->symlink = *p;
81 *p += info->symlink_len;
82
14303d20
SW		 83 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
84 ceph_decode_copy_safe(p, end, &info->dir_layout,
85 sizeof(info->dir_layout), bad);
86 else
87 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
88
2f2dc053
SW		 89 ceph_decode_32_safe(p, end, info->xattr_len, bad);
90 ceph_decode_need(p, end, info->xattr_len, bad);
91 info->xattr_data = *p;
92 *p += info->xattr_len;
fb01d1f8
YZ		 93
94 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
95 ceph_decode_64_safe(p, end, info->inline_version, bad);
96 ceph_decode_32_safe(p, end, info->inline_len, bad);
97 ceph_decode_need(p, end, info->inline_len, bad);
98 info->inline_data = *p;
99 *p += info->inline_len;
100 } else
101 info->inline_version = CEPH_INLINE_NONE;
102
779fe0fb
YZ		 103 info->pool_ns_len = 0;
104 info->pool_ns_data = NULL;
5ea5c5e0
YZ		 105 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
106 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
779fe0fb
YZ		 107 if (info->pool_ns_len > 0) {
108 ceph_decode_need(p, end, info->pool_ns_len, bad);
109 info->pool_ns_data = *p;
110 *p += info->pool_ns_len;
111 }
5ea5c5e0
YZ		 112 }
113
2f2dc053
SW		 114 return 0;
115bad:
116 return err;
117}
118
119/*
120 * parse a normal reply, which may contain a (dir+)dentry and/or a
121 * target inode.
122 */
123static int parse_reply_info_trace(void **p, void *end,
14303d20 124 struct ceph_mds_reply_info_parsed *info,
12b4629a 125 u64 features)
2f2dc053
SW		 126{
127 int err;
128
129 if (info->head->is_dentry) {
14303d20 130 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 131 if (err < 0)
132 goto out_bad;
133
134 if (unlikely(*p + sizeof(*info->dirfrag) > end))
135 goto bad;
136 info->dirfrag = *p;
137 *p += sizeof(*info->dirfrag) +
138 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
139 if (unlikely(*p > end))
140 goto bad;
141
142 ceph_decode_32_safe(p, end, info->dname_len, bad);
143 ceph_decode_need(p, end, info->dname_len, bad);
144 info->dname = *p;
145 *p += info->dname_len;
146 info->dlease = *p;
147 *p += sizeof(*info->dlease);
148 }
149
150 if (info->head->is_target) {
14303d20 151 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 152 if (err < 0)
153 goto out_bad;
154 }
155
156 if (unlikely(*p != end))
157 goto bad;
158 return 0;
159
160bad:
161 err = -EIO;
162out_bad:
163 pr_err("problem parsing mds trace %d\n", err);
164 return err;
165}
166
167/*
168 * parse readdir results
169 */
170static int parse_reply_info_dir(void **p, void *end,
14303d20 171 struct ceph_mds_reply_info_parsed *info,
12b4629a 172 u64 features)
2f2dc053
SW		 173{
174 u32 num, i = 0;
175 int err;
176
177 info->dir_dir = *p;
178 if (*p + sizeof(*info->dir_dir) > end)
179 goto bad;
180 *p += sizeof(*info->dir_dir) +
181 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
182 if (*p > end)
183 goto bad;
184
185 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea 186 num = ceph_decode_32(p);
956d39d6
YZ		 187 {
188 u16 flags = ceph_decode_16(p);
189 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
190 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
f3c4ebe6 191 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
79162547 192 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
956d39d6 193 }
2f2dc053
SW		 194 if (num == 0)
195 goto done;
196
2a5beea3
YZ		 197 BUG_ON(!info->dir_entries);
198 if ((unsigned long)(info->dir_entries + num) >
199 (unsigned long)info->dir_entries + info->dir_buf_size) {
54008399
YZ		 200 pr_err("dir contents are larger than expected\n");
201 WARN_ON(1);
202 goto bad;
203 }
2f2dc053 204
54008399 205 info->dir_nr = num;
2f2dc053 206 while (num) {
2a5beea3 207 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
2f2dc053
SW		 208 /* dentry */
209 ceph_decode_need(p, end, sizeof(u32)*2, bad);
2a5beea3
YZ		 210 rde->name_len = ceph_decode_32(p);
211 ceph_decode_need(p, end, rde->name_len, bad);
212 rde->name = *p;
213 *p += rde->name_len;
214 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
215 rde->lease = *p;
2f2dc053
SW		 216 *p += sizeof(struct ceph_mds_reply_lease);
217
218 /* inode */
2a5beea3 219 err = parse_reply_info_in(p, end, &rde->inode, features);
2f2dc053
SW		 220 if (err < 0)
221 goto out_bad;
8974eebd
YZ		 222 /* ceph_readdir_prepopulate() will update it */
223 rde->offset = 0;
2f2dc053
SW		 224 i++;
225 num--;
226 }
227
228done:
229 if (*p != end)
230 goto bad;
231 return 0;
232
233bad:
234 err = -EIO;
235out_bad:
236 pr_err("problem parsing dir contents %d\n", err);
237 return err;
238}
239
25933abd
HS		 240/*
241 * parse fcntl F_GETLK results
242 */
243static int parse_reply_info_filelock(void **p, void *end,
14303d20 244 struct ceph_mds_reply_info_parsed *info,
12b4629a 245 u64 features)
25933abd
HS		 246{
247 if (*p + sizeof(*info->filelock_reply) > end)
248 goto bad;
249
250 info->filelock_reply = *p;
251 *p += sizeof(*info->filelock_reply);
252
253 if (unlikely(*p != end))
254 goto bad;
255 return 0;
256
257bad:
258 return -EIO;
259}
260
6e8575fa
SL		 261/*
262 * parse create results
263 */
264static int parse_reply_info_create(void **p, void *end,
265 struct ceph_mds_reply_info_parsed *info,
12b4629a 266 u64 features)
6e8575fa
SL		 267{
268 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
269 if (*p == end) {
270 info->has_create_ino = false;
271 } else {
272 info->has_create_ino = true;
273 info->ino = ceph_decode_64(p);
274 }
275 }
276
277 if (unlikely(*p != end))
278 goto bad;
279 return 0;
280
281bad:
282 return -EIO;
283}
284
25933abd
HS		 285/*
286 * parse extra results
287 */
288static int parse_reply_info_extra(void **p, void *end,
14303d20 289 struct ceph_mds_reply_info_parsed *info,
12b4629a 290 u64 features)
25933abd 291{
6df8c9d8
JL		 292 u32 op = le32_to_cpu(info->head->op);
293
294 if (op == CEPH_MDS_OP_GETFILELOCK)
14303d20 295 return parse_reply_info_filelock(p, end, info, features);
6df8c9d8 296 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
14303d20 297 return parse_reply_info_dir(p, end, info, features);
6df8c9d8 298 else if (op == CEPH_MDS_OP_CREATE)
6e8575fa
SL		 299 return parse_reply_info_create(p, end, info, features);
300 else
301 return -EIO;
25933abd
HS		 302}
303
2f2dc053
SW		 304/*
305 * parse entire mds reply
306 */
307static int parse_reply_info(struct ceph_msg *msg,
14303d20 308 struct ceph_mds_reply_info_parsed *info,
12b4629a 309 u64 features)
2f2dc053
SW		 310{
311 void *p, *end;
312 u32 len;
313 int err;
314
315 info->head = msg->front.iov_base;
316 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
317 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
318
319 /* trace */
320 ceph_decode_32_safe(&p, end, len, bad);
321 if (len > 0) {
32852a81 322 ceph_decode_need(&p, end, len, bad);
14303d20 323 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 324 if (err < 0)
325 goto out_bad;
326 }
327
25933abd 328 /* extra */
2f2dc053
SW		 329 ceph_decode_32_safe(&p, end, len, bad);
330 if (len > 0) {
32852a81 331 ceph_decode_need(&p, end, len, bad);
14303d20 332 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 333 if (err < 0)
334 goto out_bad;
335 }
336
337 /* snap blob */
338 ceph_decode_32_safe(&p, end, len, bad);
339 info->snapblob_len = len;
340 info->snapblob = p;
341 p += len;
342
343 if (p != end)
344 goto bad;
345 return 0;
346
347bad:
348 err = -EIO;
349out_bad:
350 pr_err("mds parse_reply err %d\n", err);
351 return err;
352}
353
354static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
355{
2a5beea3 356 if (!info->dir_entries)
54008399 357 return;
2a5beea3 358 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
2f2dc053
SW		 359}
360
361
362/*
363 * sessions
364 */
a687ecaf 365const char *ceph_session_state_name(int s)
2f2dc053
SW		 366{
367 switch (s) {
368 case CEPH_MDS_SESSION_NEW: return "new";
369 case CEPH_MDS_SESSION_OPENING: return "opening";
370 case CEPH_MDS_SESSION_OPEN: return "open";
371 case CEPH_MDS_SESSION_HUNG: return "hung";
372 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 373 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053 374 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
fcff415c 375 case CEPH_MDS_SESSION_REJECTED: return "rejected";
2f2dc053
SW		 376 default: return "???";
377 }
378}
379
380static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
381{
3997c01d 382 if (refcount_inc_not_zero(&s->s_ref)) {
2f2dc053 383 dout("mdsc get_session %p %d -> %d\n", s,
3997c01d 384 refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
2f2dc053
SW		 385 return s;
386 } else {
387 dout("mdsc get_session %p 0 -- FAIL", s);
388 return NULL;
389 }
390}
391
392void ceph_put_mds_session(struct ceph_mds_session *s)
393{
394 dout("mdsc put_session %p %d -> %d\n", s,
3997c01d
ER		 395 refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
396 if (refcount_dec_and_test(&s->s_ref)) {
6c4a1915 397 if (s->s_auth.authorizer)
6c1ea260 398 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
2f2dc053 399 kfree(s);
4e7a5dcd 400 }
2f2dc053
SW		 401}
402
403/*
404 * called under mdsc->mutex
405 */
406struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
407 int mds)
408{
409 struct ceph_mds_session *session;
410
411 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
412 return NULL;
413 session = mdsc->sessions[mds];
414 dout("lookup_mds_session %p %d\n", session,
3997c01d 415 refcount_read(&session->s_ref));
2f2dc053
SW		 416 get_session(session);
417 return session;
418}
419
420static bool __have_session(struct ceph_mds_client *mdsc, int mds)
421{
422 if (mds >= mdsc->max_sessions)
423 return false;
424 return mdsc->sessions[mds];
425}
426
2600d2dd
SW		 427static int __verify_registered_session(struct ceph_mds_client *mdsc,
428 struct ceph_mds_session *s)
429{
430 if (s->s_mds >= mdsc->max_sessions ||
431 mdsc->sessions[s->s_mds] != s)
432 return -ENOENT;
433 return 0;
434}
435
2f2dc053
SW		 436/*
437 * create+register a new session for given mds.
438 * called under mdsc->mutex.
439 */
440static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
441 int mds)
442{
443 struct ceph_mds_session *s;
444
76201b63 445 if (mds >= mdsc->mdsmap->m_num_mds)
c338c07c
NY		 446 return ERR_PTR(-EINVAL);
447
2f2dc053 448 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 449 if (!s)
450 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 451 s->s_mdsc = mdsc;
452 s->s_mds = mds;
453 s->s_state = CEPH_MDS_SESSION_NEW;
454 s->s_ttl = 0;
455 s->s_seq = 0;
456 mutex_init(&s->s_mutex);
457
b7a9e5dd 458 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 459
d8fb02ab 460 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 461 s->s_cap_gen = 0;
1ce208a6 462 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 463
464 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 465 s->s_renew_requested = 0;
466 s->s_renew_seq = 0;
467 INIT_LIST_HEAD(&s->s_caps);
468 s->s_nr_caps = 0;
5dacf091 469 s->s_trim_caps = 0;
3997c01d 470 refcount_set(&s->s_ref, 1);
2f2dc053
SW		 471 INIT_LIST_HEAD(&s->s_waiting);
472 INIT_LIST_HEAD(&s->s_unsafe);
473 s->s_num_cap_releases = 0;
99a9c273 474 s->s_cap_reconnect = 0;
7c1332b8 475 s->s_cap_iterator = NULL;
2f2dc053 476 INIT_LIST_HEAD(&s->s_cap_releases);
2f2dc053 477 INIT_LIST_HEAD(&s->s_cap_flushing);
2f2dc053
SW		 478
479 dout("register_session mds%d\n", mds);
480 if (mds >= mdsc->max_sessions) {
481 int newmax = 1 << get_count_order(mds+1);
482 struct ceph_mds_session **sa;
483
484 dout("register_session realloc to %d\n", newmax);
485 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
486 if (sa == NULL)
42ce56e5 487 goto fail_realloc;
2f2dc053
SW		 488 if (mdsc->sessions) {
489 memcpy(sa, mdsc->sessions,
490 mdsc->max_sessions * sizeof(void *));
491 kfree(mdsc->sessions);
492 }
493 mdsc->sessions = sa;
494 mdsc->max_sessions = newmax;
495 }
496 mdsc->sessions[mds] = s;
86d8f67b 497 atomic_inc(&mdsc->num_sessions);
3997c01d 498 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 499
b7a9e5dd
SW		 500 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
501 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 502
2f2dc053 503 return s;
42ce56e5
SW		 504
505fail_realloc:
506 kfree(s);
507 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 508}
509
510/*
511 * called under mdsc->mutex
512 */
2600d2dd 513static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 514 struct ceph_mds_session *s)
2f2dc053 515{
2600d2dd
SW		 516 dout("__unregister_session mds%d %p\n", s->s_mds, s);
517 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 518 mdsc->sessions[s->s_mds] = NULL;
519 ceph_con_close(&s->s_con);
520 ceph_put_mds_session(s);
86d8f67b 521 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 522}
523
524/*
525 * drop session refs in request.
526 *
527 * should be last request ref, or hold mdsc->mutex
528 */
529static void put_request_session(struct ceph_mds_request *req)
530{
531 if (req->r_session) {
532 ceph_put_mds_session(req->r_session);
533 req->r_session = NULL;
534 }
535}
536
153c8e6b 537void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 538{
153c8e6b
SW		 539 struct ceph_mds_request *req = container_of(kref,
540 struct ceph_mds_request,
541 r_kref);
54008399 542 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 543 if (req->r_request)
544 ceph_msg_put(req->r_request);
54008399 545 if (req->r_reply)
153c8e6b 546 ceph_msg_put(req->r_reply);
153c8e6b 547 if (req->r_inode) {
41b02e1f 548 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 549 iput(req->r_inode);
550 }
3dd69aab
JL		 551 if (req->r_parent)
552 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
e96a650a 553 iput(req->r_target_inode);
153c8e6b
SW		 554 if (req->r_dentry)
555 dput(req->r_dentry);
844d87c3
SW		 556 if (req->r_old_dentry)
557 dput(req->r_old_dentry);
558 if (req->r_old_dentry_dir) {
41b02e1f
SW		 559 /*
560 * track (and drop pins for) r_old_dentry_dir
561 * separately, since r_old_dentry's d_parent may have
562 * changed between the dir mutex being dropped and
563 * this request being freed.
564 */
565 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
566 CEPH_CAP_PIN);
41b02e1f 567 iput(req->r_old_dentry_dir);
2f2dc053 568 }
153c8e6b
SW		 569 kfree(req->r_path1);
570 kfree(req->r_path2);
25e6bae3
YZ		 571 if (req->r_pagelist)
572 ceph_pagelist_release(req->r_pagelist);
153c8e6b 573 put_request_session(req);
37151668 574 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 575 kfree(req);
2f2dc053
SW		 576}
577
fcd00b68
ID		 578DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
579
2f2dc053
SW		 580/*
581 * lookup session, bump ref if found.
582 *
583 * called under mdsc->mutex.
584 */
fcd00b68
ID		 585static struct ceph_mds_request *
586lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
2f2dc053
SW		 587{
588 struct ceph_mds_request *req;
44ca18f2 589
fcd00b68
ID		 590 req = lookup_request(&mdsc->request_tree, tid);
591 if (req)
592 ceph_mdsc_get_request(req);
44ca18f2 593
fcd00b68 594 return req;
2f2dc053
SW		 595}
596
597/*
598 * Register an in-flight request, and assign a tid. Link to directory
599 * are modifying (if any).
600 *
601 * Called under mdsc->mutex.
602 */
603static void __register_request(struct ceph_mds_client *mdsc,
604 struct ceph_mds_request *req,
605 struct inode *dir)
606{
607 req->r_tid = ++mdsc->last_tid;
608 if (req->r_num_caps)
37151668
YS		 609 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
610 req->r_num_caps);
2f2dc053
SW		 611 dout("__register_request %p tid %lld\n", req, req->r_tid);
612 ceph_mdsc_get_request(req);
fcd00b68 613 insert_request(&mdsc->request_tree, req);
2f2dc053 614
cb4276cc
SW		 615 req->r_uid = current_fsuid();
616 req->r_gid = current_fsgid();
617
e8a7b8b1
YZ		 618 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
619 mdsc->oldest_tid = req->r_tid;
620
2f2dc053 621 if (dir) {
3b663780 622 ihold(dir);
2f2dc053 623 req->r_unsafe_dir = dir;
2f2dc053
SW		 624 }
625}
626
627static void __unregister_request(struct ceph_mds_client *mdsc,
628 struct ceph_mds_request *req)
629{
630 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1 631
df963ea8
JL		 632 /* Never leave an unregistered request on an unsafe list! */
633 list_del_init(&req->r_unsafe_item);
634
e8a7b8b1
YZ		 635 if (req->r_tid == mdsc->oldest_tid) {
636 struct rb_node *p = rb_next(&req->r_node);
637 mdsc->oldest_tid = 0;
638 while (p) {
639 struct ceph_mds_request *next_req =
640 rb_entry(p, struct ceph_mds_request, r_node);
641 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
642 mdsc->oldest_tid = next_req->r_tid;
643 break;
644 }
645 p = rb_next(p);
646 }
647 }
648
fcd00b68 649 erase_request(&mdsc->request_tree, req);
2f2dc053 650
bc2de10d
JL		 651 if (req->r_unsafe_dir &&
652 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053 653 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 654 spin_lock(&ci->i_unsafe_lock);
655 list_del_init(&req->r_unsafe_dir_item);
656 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 657 }
bc2de10d
JL		 658 if (req->r_target_inode &&
659 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 660 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
661 spin_lock(&ci->i_unsafe_lock);
662 list_del_init(&req->r_unsafe_target_item);
663 spin_unlock(&ci->i_unsafe_lock);
664 }
3b663780 665
4c06ace8 666 if (req->r_unsafe_dir) {
3b663780
SW		 667 iput(req->r_unsafe_dir);
668 req->r_unsafe_dir = NULL;
2f2dc053 669 }
94aa8ae1 670
fc55d2c9
YZ		 671 complete_all(&req->r_safe_completion);
672
94aa8ae1 673 ceph_mdsc_put_request(req);
2f2dc053
SW		 674}
675
30c71233
JL		 676/*
677 * Walk back up the dentry tree until we hit a dentry representing a
678 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
679 * when calling this) to ensure that the objects won't disappear while we're
680 * working with them. Once we hit a candidate dentry, we attempt to take a
681 * reference to it, and return that as the result.
682 */
f1075480
DC		 683static struct inode *get_nonsnap_parent(struct dentry *dentry)
684{
685 struct inode *inode = NULL;
30c71233
JL		 686
687 while (dentry && !IS_ROOT(dentry)) {
688 inode = d_inode_rcu(dentry);
689 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
690 break;
691 dentry = dentry->d_parent;
692 }
693 if (inode)
694 inode = igrab(inode);
695 return inode;
696}
697
2f2dc053
SW		 698/*
699 * Choose mds to send request to next. If there is a hint set in the
700 * request (e.g., due to a prior forward hint from the mds), use that.
701 * Otherwise, consult frag tree and/or caps to identify the
702 * appropriate mds. If all else fails, choose randomly.
703 *
704 * Called under mdsc->mutex.
705 */
706static int __choose_mds(struct ceph_mds_client *mdsc,
707 struct ceph_mds_request *req)
708{
709 struct inode *inode;
710 struct ceph_inode_info *ci;
711 struct ceph_cap *cap;
712 int mode = req->r_direct_mode;
713 int mds = -1;
714 u32 hash = req->r_direct_hash;
bc2de10d 715 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
2f2dc053
SW		 716
717 /*
718 * is there a specific mds we should try? ignore hint if we have
719 * no session and the mds is not up (active or recovering).
720 */
721 if (req->r_resend_mds >= 0 &&
722 (__have_session(mdsc, req->r_resend_mds) ||
723 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
724 dout("choose_mds using resend_mds mds%d\n",
725 req->r_resend_mds);
726 return req->r_resend_mds;
727 }
728
729 if (mode == USE_RANDOM_MDS)
730 goto random;
731
732 inode = NULL;
733 if (req->r_inode) {
5d37ca14
YZ		 734 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
735 inode = req->r_inode;
736 ihold(inode);
737 } else {
738 /* req->r_dentry is non-null for LSSNAP request.
739 * fall-thru */
740 WARN_ON_ONCE(!req->r_dentry);
741 }
742 }
743 if (!inode && req->r_dentry) {
d79698da 744 /* ignore race with rename; old or new d_parent is okay */
30c71233
JL		 745 struct dentry *parent;
746 struct inode *dir;
747
748 rcu_read_lock();
749 parent = req->r_dentry->d_parent;
3dd69aab 750 dir = req->r_parent ? : d_inode_rcu(parent);
eb6bb1c5 751
30c71233
JL		 752 if (!dir || dir->i_sb != mdsc->fsc->sb) {
753 /* not this fs or parent went negative */
2b0143b5 754 inode = d_inode(req->r_dentry);
30c71233
JL		 755 if (inode)
756 ihold(inode);
eb6bb1c5
SW		 757 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
758 /* direct snapped/virtual snapdir requests
759 * based on parent dir inode */
30c71233 760 inode = get_nonsnap_parent(parent);
eb6bb1c5 761 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 762 } else {
eb6bb1c5 763 /* dentry target */
2b0143b5 764 inode = d_inode(req->r_dentry);
ca18bede
YZ		 765 if (!inode || mode == USE_AUTH_MDS) {
766 /* dir + name */
30c71233 767 inode = igrab(dir);
ca18bede
YZ		 768 hash = ceph_dentry_hash(dir, req->r_dentry);
769 is_hash = true;
30c71233
JL		 770 } else {
771 ihold(inode);
ca18bede 772 }
2f2dc053 773 }
30c71233 774 rcu_read_unlock();
2f2dc053 775 }
eb6bb1c5 776
2f2dc053
SW		 777 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
778 (int)hash, mode);
779 if (!inode)
780 goto random;
781 ci = ceph_inode(inode);
782
783 if (is_hash && S_ISDIR(inode->i_mode)) {
784 struct ceph_inode_frag frag;
785 int found;
786
787 ceph_choose_frag(ci, hash, &frag, &found);
788 if (found) {
789 if (mode == USE_ANY_MDS && frag.ndist > 0) {
790 u8 r;
791
792 /* choose a random replica */
793 get_random_bytes(&r, 1);
794 r %= frag.ndist;
795 mds = frag.dist[r];
796 dout("choose_mds %p %llx.%llx "
797 "frag %u mds%d (%d/%d)\n",
798 inode, ceph_vinop(inode),
d66bbd44 799 frag.frag, mds,
2f2dc053 800 (int)r, frag.ndist);
d66bbd44
SW		 801 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
802 CEPH_MDS_STATE_ACTIVE)
30c71233 803 goto out;
2f2dc053
SW		 804 }
805
806 /* since this file/dir wasn't known to be
807 * replicated, then we want to look for the
808 * authoritative mds. */
809 mode = USE_AUTH_MDS;
810 if (frag.mds >= 0) {
811 /* choose auth mds */
812 mds = frag.mds;
813 dout("choose_mds %p %llx.%llx "
814 "frag %u mds%d (auth)\n",
815 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 816 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
817 CEPH_MDS_STATE_ACTIVE)
30c71233 818 goto out;
2f2dc053
SW		 819 }
820 }
821 }
822
be655596 823 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 824 cap = NULL;
825 if (mode == USE_AUTH_MDS)
826 cap = ci->i_auth_cap;
827 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
828 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
829 if (!cap) {
be655596 830 spin_unlock(&ci->i_ceph_lock);
30c71233 831 iput(inode);
2f2dc053
SW		 832 goto random;
833 }
834 mds = cap->session->s_mds;
835 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
836 inode, ceph_vinop(inode), mds,
837 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 838 spin_unlock(&ci->i_ceph_lock);
30c71233
JL		 839out:
840 iput(inode);
2f2dc053
SW		 841 return mds;
842
843random:
844 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
845 dout("choose_mds chose random mds%d\n", mds);
846 return mds;
847}
848
849
850/*
851 * session messages
852 */
853static struct ceph_msg *create_session_msg(u32 op, u64 seq)
854{
855 struct ceph_msg *msg;
856 struct ceph_mds_session_head *h;
857
b61c2763
SW		 858 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
859 false);
a79832f2 860 if (!msg) {
2f2dc053 861 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 862 return NULL;
2f2dc053
SW		 863 }
864 h = msg->front.iov_base;
865 h->op = cpu_to_le32(op);
866 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 867
868 return msg;
869}
870
871/*
872 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
873 * to include additional client metadata fields.
874 */
875static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
876{
877 struct ceph_msg *msg;
878 struct ceph_mds_session_head *h;
879 int i = -1;
880 int metadata_bytes = 0;
881 int metadata_key_count = 0;
882 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 883 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
dbd0c8bf
JS		 884 void *p;
885
a6a5ce4f 886 const char* metadata[][2] = {
dbd0c8bf 887 {"hostname", utsname()->nodename},
a6a5ce4f 888 {"kernel_version", utsname()->release},
3f384954
YZ		 889 {"entity_id", opt->name ? : ""},
890 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 891 {NULL, NULL}
892 };
893
894 /* Calculate serialized length of metadata */
895 metadata_bytes = 4; /* map length */
896 for (i = 0; metadata[i][0] != NULL; ++i) {
897 metadata_bytes += 8 + strlen(metadata[i][0]) +
898 strlen(metadata[i][1]);
899 metadata_key_count++;
900 }
901
902 /* Allocate the message */
903 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
904 GFP_NOFS, false);
905 if (!msg) {
906 pr_err("create_session_msg ENOMEM creating msg\n");
907 return NULL;
908 }
909 h = msg->front.iov_base;
910 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
911 h->seq = cpu_to_le64(seq);
912
913 /*
914 * Serialize client metadata into waiting buffer space, using
915 * the format that userspace expects for map<string, string>
7cfa0313
JS		 916 *
917 * ClientSession messages with metadata are v2
dbd0c8bf 918 */
7cfa0313
JS		 919 msg->hdr.version = cpu_to_le16(2);
920 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 921
922 /* The write pointer, following the session_head structure */
923 p = msg->front.iov_base + sizeof(*h);
924
925 /* Number of entries in the map */
926 ceph_encode_32(&p, metadata_key_count);
927
928 /* Two length-prefixed strings for each entry in the map */
929 for (i = 0; metadata[i][0] != NULL; ++i) {
930 size_t const key_len = strlen(metadata[i][0]);
931 size_t const val_len = strlen(metadata[i][1]);
932
933 ceph_encode_32(&p, key_len);
934 memcpy(p, metadata[i][0], key_len);
935 p += key_len;
936 ceph_encode_32(&p, val_len);
937 memcpy(p, metadata[i][1], val_len);
938 p += val_len;
939 }
940
2f2dc053
SW		 941 return msg;
942}
943
944/*
945 * send session open request.
946 *
947 * called under mdsc->mutex
948 */
949static int __open_session(struct ceph_mds_client *mdsc,
950 struct ceph_mds_session *session)
951{
952 struct ceph_msg *msg;
953 int mstate;
954 int mds = session->s_mds;
2f2dc053
SW		 955
956 /* wait for mds to go active? */
957 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
958 dout("open_session to mds%d (%s)\n", mds,
959 ceph_mds_state_name(mstate));
960 session->s_state = CEPH_MDS_SESSION_OPENING;
961 session->s_renew_requested = jiffies;
962
963 /* send connect message */
dbd0c8bf 964 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 965 if (!msg)
966 return -ENOMEM;
2f2dc053 967 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 968 return 0;
969}
970
ed0552a1
SW		 971/*
972 * open sessions for any export targets for the given mds
973 *
974 * called under mdsc->mutex
975 */
5d72d13c
YZ		 976static struct ceph_mds_session *
977__open_export_target_session(struct ceph_mds_client *mdsc, int target)
978{
979 struct ceph_mds_session *session;
980
981 session = __ceph_lookup_mds_session(mdsc, target);
982 if (!session) {
983 session = register_session(mdsc, target);
984 if (IS_ERR(session))
985 return session;
986 }
987 if (session->s_state == CEPH_MDS_SESSION_NEW ||
988 session->s_state == CEPH_MDS_SESSION_CLOSING)
989 __open_session(mdsc, session);
990
991 return session;
992}
993
994struct ceph_mds_session *
995ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
996{
997 struct ceph_mds_session *session;
998
999 dout("open_export_target_session to mds%d\n", target);
1000
1001 mutex_lock(&mdsc->mutex);
1002 session = __open_export_target_session(mdsc, target);
1003 mutex_unlock(&mdsc->mutex);
1004
1005 return session;
1006}
1007
ed0552a1
SW		 1008static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1009 struct ceph_mds_session *session)
1010{
1011 struct ceph_mds_info *mi;
1012 struct ceph_mds_session *ts;
1013 int i, mds = session->s_mds;
ed0552a1 1014
76201b63 1015 if (mds >= mdsc->mdsmap->m_num_mds)
ed0552a1 1016 return;
5d72d13c 1017
ed0552a1
SW		 1018 mi = &mdsc->mdsmap->m_info[mds];
1019 dout("open_export_target_sessions for mds%d (%d targets)\n",
1020 session->s_mds, mi->num_export_targets);
1021
1022 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1023 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1024 if (!IS_ERR(ts))
1025 ceph_put_mds_session(ts);
ed0552a1
SW		 1026 }
1027}
1028
154f42c2
SW		 1029void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1030 struct ceph_mds_session *session)
1031{
1032 mutex_lock(&mdsc->mutex);
1033 __open_export_target_sessions(mdsc, session);
1034 mutex_unlock(&mdsc->mutex);
1035}
1036
2f2dc053
SW		 1037/*
1038 * session caps
1039 */
1040
745a8e3b
YZ		 1041/* caller holds s_cap_lock, we drop it */
1042static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1043 struct ceph_mds_session *session)
1044 __releases(session->s_cap_lock)
2f2dc053 1045{
745a8e3b
YZ		 1046 LIST_HEAD(tmp_list);
1047 list_splice_init(&session->s_cap_releases, &tmp_list);
1048 session->s_num_cap_releases = 0;
1049 spin_unlock(&session->s_cap_lock);
2f2dc053 1050
745a8e3b
YZ		 1051 dout("cleanup_cap_releases mds%d\n", session->s_mds);
1052 while (!list_empty(&tmp_list)) {
1053 struct ceph_cap *cap;
1054 /* zero out the in-progress message */
1055 cap = list_first_entry(&tmp_list,
1056 struct ceph_cap, session_caps);
1057 list_del(&cap->session_caps);
1058 ceph_put_cap(mdsc, cap);
2f2dc053 1059 }
2f2dc053
SW		 1060}
1061
1c841a96
YZ		 1062static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1063 struct ceph_mds_session *session)
1064{
1065 struct ceph_mds_request *req;
1066 struct rb_node *p;
1067
1068 dout("cleanup_session_requests mds%d\n", session->s_mds);
1069 mutex_lock(&mdsc->mutex);
1070 while (!list_empty(&session->s_unsafe)) {
1071 req = list_first_entry(&session->s_unsafe,
1072 struct ceph_mds_request, r_unsafe_item);
3e0708b9
YZ		 1073 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1074 req->r_tid);
1c841a96
YZ		 1075 __unregister_request(mdsc, req);
1076 }
1077 /* zero r_attempts, so kick_requests() will re-send requests */
1078 p = rb_first(&mdsc->request_tree);
1079 while (p) {
1080 req = rb_entry(p, struct ceph_mds_request, r_node);
1081 p = rb_next(p);
1082 if (req->r_session &&
1083 req->r_session->s_mds == session->s_mds)
1084 req->r_attempts = 0;
1085 }
1086 mutex_unlock(&mdsc->mutex);
1087}
1088
2f2dc053 1089/*
f818a736
SW		 1090 * Helper to safely iterate over all caps associated with a session, with
1091 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1092 *
f818a736 1093 * Caller must hold session s_mutex.
2f2dc053
SW		 1094 */
1095static int iterate_session_caps(struct ceph_mds_session *session,
1096 int (*cb)(struct inode *, struct ceph_cap *,
1097 void *), void *arg)
1098{
7c1332b8
SW		 1099 struct list_head *p;
1100 struct ceph_cap *cap;
1101 struct inode *inode, *last_inode = NULL;
1102 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1103 int ret;
1104
1105 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1106 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1107 p = session->s_caps.next;
1108 while (p != &session->s_caps) {
1109 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1110 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1111 if (!inode) {
1112 p = p->next;
2f2dc053 1113 continue;
7c1332b8
SW		 1114 }
1115 session->s_cap_iterator = cap;
2f2dc053 1116 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1117
1118 if (last_inode) {
1119 iput(last_inode);
1120 last_inode = NULL;
1121 }
1122 if (old_cap) {
37151668 1123 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1124 old_cap = NULL;
1125 }
1126
2f2dc053 1127 ret = cb(inode, cap, arg);
7c1332b8
SW		 1128 last_inode = inode;
1129
2f2dc053 1130 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1131 p = p->next;
1132 if (cap->ci == NULL) {
1133 dout("iterate_session_caps finishing cap %p removal\n",
1134 cap);
1135 BUG_ON(cap->session != session);
745a8e3b 1136 cap->session = NULL;
7c1332b8
SW		 1137 list_del_init(&cap->session_caps);
1138 session->s_nr_caps--;
745a8e3b
YZ		 1139 if (cap->queue_release) {
1140 list_add_tail(&cap->session_caps,
1141 &session->s_cap_releases);
1142 session->s_num_cap_releases++;
1143 } else {
1144 old_cap = cap; /* put_cap it w/o locks held */
1145 }
7c1332b8 1146 }
5dacf091
SW		 1147 if (ret < 0)
1148 goto out;
2f2dc053 1149 }
5dacf091
SW		 1150 ret = 0;
1151out:
7c1332b8 1152 session->s_cap_iterator = NULL;
2f2dc053 1153 spin_unlock(&session->s_cap_lock);
7c1332b8 1154
e96a650a 1155 iput(last_inode);
7c1332b8 1156 if (old_cap)
37151668 1157 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1158
5dacf091 1159 return ret;
2f2dc053
SW		 1160}
1161
1162static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1163 void *arg)
2f2dc053 1164{
6c93df5d 1165 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
2f2dc053 1166 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1167 LIST_HEAD(to_remove);
6c93df5d
YZ		 1168 bool drop = false;
1169 bool invalidate = false;
6c99f254 1170
2f2dc053
SW		 1171 dout("removing cap %p, ci is %p, inode is %p\n",
1172 cap, ci, &ci->vfs_inode);
be655596 1173 spin_lock(&ci->i_ceph_lock);
a096b09a 1174 __ceph_remove_cap(cap, false);
571ade33 1175 if (!ci->i_auth_cap) {
553adfd9 1176 struct ceph_cap_flush *cf;
6c93df5d 1177 struct ceph_mds_client *mdsc = fsc->mdsc;
6c99f254 1178
77310320
YZ		 1179 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1180
6c93df5d 1181 if (ci->i_wrbuffer_ref > 0 &&
52953d55 1182 READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
6c93df5d
YZ		 1183 invalidate = true;
1184
e4500b5e
YZ		 1185 while (!list_empty(&ci->i_cap_flush_list)) {
1186 cf = list_first_entry(&ci->i_cap_flush_list,
1187 struct ceph_cap_flush, i_list);
8cdcc07d 1188 list_move(&cf->i_list, &to_remove);
553adfd9
YZ		 1189 }
1190
6c99f254 1191 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1192
e4500b5e
YZ		 1193 list_for_each_entry(cf, &to_remove, i_list)
1194 list_del(&cf->g_list);
8310b089 1195
6c99f254 1196 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1197 pr_warn_ratelimited(
1198 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1199 ceph_cap_string(ci->i_dirty_caps),
1200 inode, ceph_ino(inode));
1201 ci->i_dirty_caps = 0;
1202 list_del_init(&ci->i_dirty_item);
6c93df5d 1203 drop = true;
6c99f254
SW		 1204 }
1205 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1206 pr_warn_ratelimited(
1207 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1208 ceph_cap_string(ci->i_flushing_caps),
1209 inode, ceph_ino(inode));
1210 ci->i_flushing_caps = 0;
1211 list_del_init(&ci->i_flushing_item);
1212 mdsc->num_cap_flushing--;
6c93df5d 1213 drop = true;
6c99f254 1214 }
6c99f254 1215 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1216
f66fd9f0 1217 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1218 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1219 ci->i_prealloc_cap_flush = NULL;
1220 }
6c99f254 1221 }
be655596 1222 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1223 while (!list_empty(&to_remove)) {
1224 struct ceph_cap_flush *cf;
1225 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1226 struct ceph_cap_flush, i_list);
1227 list_del(&cf->i_list);
f66fd9f0 1228 ceph_free_cap_flush(cf);
553adfd9 1229 }
77310320
YZ		 1230
1231 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1232 if (invalidate)
1233 ceph_queue_invalidate(inode);
77310320 1234 if (drop)
6c99f254 1235 iput(inode);
2f2dc053
SW		 1236 return 0;
1237}
1238
1239/*
1240 * caller must hold session s_mutex
1241 */
1242static void remove_session_caps(struct ceph_mds_session *session)
1243{
6c93df5d
YZ		 1244 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1245 struct super_block *sb = fsc->sb;
2f2dc053 1246 dout("remove_session_caps on %p\n", session);
6c93df5d 1247 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1248
c8799fc4
YZ		 1249 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1250
6f60f889
YZ		 1251 spin_lock(&session->s_cap_lock);
1252 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1253 struct inode *inode;
1254 struct ceph_cap *cap, *prev = NULL;
1255 struct ceph_vino vino;
1256 /*
1257 * iterate_session_caps() skips inodes that are being
1258 * deleted, we need to wait until deletions are complete.
1259 * __wait_on_freeing_inode() is designed for the job,
1260 * but it is not exported, so use lookup inode function
1261 * to access it.
1262 */
1263 while (!list_empty(&session->s_caps)) {
1264 cap = list_entry(session->s_caps.next,
1265 struct ceph_cap, session_caps);
1266 if (cap == prev)
1267 break;
1268 prev = cap;
1269 vino = cap->ci->i_vino;
1270 spin_unlock(&session->s_cap_lock);
1271
ed284c49 1272 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1273 iput(inode);
1274
1275 spin_lock(&session->s_cap_lock);
1276 }
1277 }
745a8e3b
YZ		 1278
1279 // drop cap expires and unlock s_cap_lock
1280 cleanup_cap_releases(session->s_mdsc, session);
6f60f889 1281
2f2dc053 1282 BUG_ON(session->s_nr_caps > 0);
6c99f254 1283 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1284}
1285
1286/*
1287 * wake up any threads waiting on this session's caps. if the cap is
1288 * old (didn't get renewed on the client reconnect), remove it now.
1289 *
1290 * caller must hold s_mutex.
1291 */
1292static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1293 void *arg)
1294{
0dc2570f
SW		 1295 struct ceph_inode_info *ci = ceph_inode(inode);
1296
0dc2570f 1297 if (arg) {
be655596 1298 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1299 ci->i_wanted_max_size = 0;
1300 ci->i_requested_max_size = 0;
be655596 1301 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1302 }
e5360309 1303 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1304 return 0;
1305}
1306
0dc2570f
SW		 1307static void wake_up_session_caps(struct ceph_mds_session *session,
1308 int reconnect)
2f2dc053
SW		 1309{
1310 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1311 iterate_session_caps(session, wake_up_session_cb,
1312 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1313}
1314
1315/*
1316 * Send periodic message to MDS renewing all currently held caps. The
1317 * ack will reset the expiration for all caps from this session.
1318 *
1319 * caller holds s_mutex
1320 */
1321static int send_renew_caps(struct ceph_mds_client *mdsc,
1322 struct ceph_mds_session *session)
1323{
1324 struct ceph_msg *msg;
1325 int state;
1326
1327 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1328 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1329 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1330 session->s_renew_requested = jiffies;
2f2dc053
SW		 1331
1332 /* do not try to renew caps until a recovering mds has reconnected
1333 * with its clients. */
1334 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1335 if (state < CEPH_MDS_STATE_RECONNECT) {
1336 dout("send_renew_caps ignoring mds%d (%s)\n",
1337 session->s_mds, ceph_mds_state_name(state));
1338 return 0;
1339 }
1340
1341 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1342 ceph_mds_state_name(state));
2f2dc053
SW		 1343 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1344 ++session->s_renew_seq);
a79832f2
SW		 1345 if (!msg)
1346 return -ENOMEM;
2f2dc053
SW		 1347 ceph_con_send(&session->s_con, msg);
1348 return 0;
1349}
1350
186e4f7a
YZ		 1351static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1352 struct ceph_mds_session *session, u64 seq)
1353{
1354 struct ceph_msg *msg;
1355
1356 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1357 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1358 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1359 if (!msg)
1360 return -ENOMEM;
1361 ceph_con_send(&session->s_con, msg);
1362 return 0;
1363}
1364
1365
2f2dc053
SW		 1366/*
1367 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1368 *
1369 * Called under session->s_mutex
2f2dc053
SW		 1370 */
1371static void renewed_caps(struct ceph_mds_client *mdsc,
1372 struct ceph_mds_session *session, int is_renew)
1373{
1374 int was_stale;
1375 int wake = 0;
1376
1377 spin_lock(&session->s_cap_lock);
1ce208a6 1378 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1379
1380 session->s_cap_ttl = session->s_renew_requested +
1381 mdsc->mdsmap->m_session_timeout*HZ;
1382
1383 if (was_stale) {
1384 if (time_before(jiffies, session->s_cap_ttl)) {
1385 pr_info("mds%d caps renewed\n", session->s_mds);
1386 wake = 1;
1387 } else {
1388 pr_info("mds%d caps still stale\n", session->s_mds);
1389 }
1390 }
1391 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1392 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1393 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1394 spin_unlock(&session->s_cap_lock);
1395
1396 if (wake)
0dc2570f 1397 wake_up_session_caps(session, 0);
2f2dc053
SW		 1398}
1399
1400/*
1401 * send a session close request
1402 */
1403static int request_close_session(struct ceph_mds_client *mdsc,
1404 struct ceph_mds_session *session)
1405{
1406 struct ceph_msg *msg;
2f2dc053
SW		 1407
1408 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1409 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1410 session->s_seq);
1411 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1412 if (!msg)
1413 return -ENOMEM;
1414 ceph_con_send(&session->s_con, msg);
fcff415c 1415 return 1;
2f2dc053
SW		 1416}
1417
1418/*
1419 * Called with s_mutex held.
1420 */
1421static int __close_session(struct ceph_mds_client *mdsc,
1422 struct ceph_mds_session *session)
1423{
1424 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1425 return 0;
1426 session->s_state = CEPH_MDS_SESSION_CLOSING;
1427 return request_close_session(mdsc, session);
1428}
1429
1430/*
1431 * Trim old(er) caps.
1432 *
1433 * Because we can't cache an inode without one or more caps, we do
1434 * this indirectly: if a cap is unused, we prune its aliases, at which
1435 * point the inode will hopefully get dropped to.
1436 *
1437 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1438 * memory pressure from the MDS, though, so it needn't be perfect.
1439 */
1440static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1441{
1442 struct ceph_mds_session *session = arg;
1443 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1444 int used, wanted, oissued, mine;
2f2dc053
SW		 1445
1446 if (session->s_trim_caps <= 0)
1447 return -1;
1448
be655596 1449 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1450 mine = cap->issued | cap->implemented;
1451 used = __ceph_caps_used(ci);
979abfdd 1452 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1453 oissued = __ceph_caps_issued_other(ci, cap);
1454
979abfdd 1455 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1456 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1457 ceph_cap_string(used), ceph_cap_string(wanted));
1458 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1459 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1460 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1461 goto out;
1462 if ((used | wanted) & CEPH_CAP_ANY_WR)
1463 goto out;
1464 }
5e804ac4
YZ		 1465 /* The inode has cached pages, but it's no longer used.
1466 * we can safely drop it */
1467 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1468 !(oissued & CEPH_CAP_FILE_CACHE)) {
1469 used = 0;
1470 oissued = 0;
1471 }
979abfdd 1472 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1473 goto out; /* we need these caps */
1474
1475 session->s_trim_caps--;
1476 if (oissued) {
1477 /* we aren't the only cap.. just remove us */
a096b09a 1478 __ceph_remove_cap(cap, true);
2f2dc053 1479 } else {
5e804ac4 1480 /* try dropping referring dentries */
be655596 1481 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1482 d_prune_aliases(inode);
1483 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1484 inode, cap, atomic_read(&inode->i_count));
1485 return 0;
1486 }
1487
1488out:
be655596 1489 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1490 return 0;
1491}
1492
1493/*
1494 * Trim session cap count down to some max number.
1495 */
1496static int trim_caps(struct ceph_mds_client *mdsc,
1497 struct ceph_mds_session *session,
1498 int max_caps)
1499{
1500 int trim_caps = session->s_nr_caps - max_caps;
1501
1502 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1503 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1504 if (trim_caps > 0) {
1505 session->s_trim_caps = trim_caps;
1506 iterate_session_caps(session, trim_caps_cb, session);
1507 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1508 session->s_mds, session->s_nr_caps, max_caps,
1509 trim_caps - session->s_trim_caps);
5dacf091 1510 session->s_trim_caps = 0;
2f2dc053 1511 }
a56371d9 1512
a56371d9 1513 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1514 return 0;
1515}
1516
8310b089
YZ		 1517static int check_caps_flush(struct ceph_mds_client *mdsc,
1518 u64 want_flush_tid)
1519{
8310b089
YZ		 1520 int ret = 1;
1521
1522 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1523 if (!list_empty(&mdsc->cap_flush_list)) {
1524 struct ceph_cap_flush *cf =
1525 list_first_entry(&mdsc->cap_flush_list,
1526 struct ceph_cap_flush, g_list);
1527 if (cf->tid <= want_flush_tid) {
1528 dout("check_caps_flush still flushing tid "
1529 "%llu <= %llu\n", cf->tid, want_flush_tid);
1530 ret = 0;
1531 }
8310b089
YZ		 1532 }
1533 spin_unlock(&mdsc->cap_dirty_lock);
1534 return ret;
d3383a8e
YZ		 1535}
1536
2f2dc053
SW		 1537/*
1538 * flush all dirty inode data to disk.
1539 *
8310b089 1540 * returns true if we've flushed through want_flush_tid
2f2dc053 1541 */
affbc19a 1542static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1543 u64 want_flush_tid)
2f2dc053 1544{
0e294387 1545 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1546
1547 wait_event(mdsc->cap_flushing_wq,
1548 check_caps_flush(mdsc, want_flush_tid));
1549
1550 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1551}
1552
1553/*
1554 * called under s_mutex
1555 */
3d7ded4d
SW		 1556void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1557 struct ceph_mds_session *session)
2f2dc053 1558{
745a8e3b
YZ		 1559 struct ceph_msg *msg = NULL;
1560 struct ceph_mds_cap_release *head;
1561 struct ceph_mds_cap_item *item;
92475f05 1562 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
745a8e3b
YZ		 1563 struct ceph_cap *cap;
1564 LIST_HEAD(tmp_list);
1565 int num_cap_releases;
92475f05
JL		 1566 __le32 barrier, *cap_barrier;
1567
1568 down_read(&osdc->lock);
1569 barrier = cpu_to_le32(osdc->epoch_barrier);
1570 up_read(&osdc->lock);
2f2dc053 1571
0f8605f2 1572 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1573again:
1574 list_splice_init(&session->s_cap_releases, &tmp_list);
1575 num_cap_releases = session->s_num_cap_releases;
1576 session->s_num_cap_releases = 0;
2f2dc053 1577 spin_unlock(&session->s_cap_lock);
e01a5946 1578
745a8e3b
YZ		 1579 while (!list_empty(&tmp_list)) {
1580 if (!msg) {
1581 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1582 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1583 if (!msg)
1584 goto out_err;
1585 head = msg->front.iov_base;
1586 head->num = cpu_to_le32(0);
1587 msg->front.iov_len = sizeof(*head);
92475f05
JL		 1588
1589 msg->hdr.version = cpu_to_le16(2);
1590 msg->hdr.compat_version = cpu_to_le16(1);
745a8e3b 1591 }
92475f05 1592
745a8e3b
YZ		 1593 cap = list_first_entry(&tmp_list, struct ceph_cap,
1594 session_caps);
1595 list_del(&cap->session_caps);
1596 num_cap_releases--;
e01a5946 1597
00bd8edb 1598 head = msg->front.iov_base;
745a8e3b
YZ		 1599 le32_add_cpu(&head->num, 1);
1600 item = msg->front.iov_base + msg->front.iov_len;
1601 item->ino = cpu_to_le64(cap->cap_ino);
1602 item->cap_id = cpu_to_le64(cap->cap_id);
1603 item->migrate_seq = cpu_to_le32(cap->mseq);
1604 item->seq = cpu_to_le32(cap->issue_seq);
1605 msg->front.iov_len += sizeof(*item);
1606
1607 ceph_put_cap(mdsc, cap);
1608
1609 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
92475f05
JL		 1610 // Append cap_barrier field
1611 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1612 *cap_barrier = barrier;
1613 msg->front.iov_len += sizeof(*cap_barrier);
1614
745a8e3b
YZ		 1615 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1616 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1617 ceph_con_send(&session->s_con, msg);
1618 msg = NULL;
1619 }
00bd8edb 1620 }
e01a5946 1621
745a8e3b 1622 BUG_ON(num_cap_releases != 0);
e01a5946 1623
745a8e3b
YZ		 1624 spin_lock(&session->s_cap_lock);
1625 if (!list_empty(&session->s_cap_releases))
1626 goto again;
1627 spin_unlock(&session->s_cap_lock);
1628
1629 if (msg) {
92475f05
JL		 1630 // Append cap_barrier field
1631 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1632 *cap_barrier = barrier;
1633 msg->front.iov_len += sizeof(*cap_barrier);
1634
745a8e3b
YZ		 1635 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1636 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1637 ceph_con_send(&session->s_con, msg);
e01a5946 1638 }
745a8e3b
YZ		 1639 return;
1640out_err:
1641 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1642 session->s_mds);
1643 spin_lock(&session->s_cap_lock);
1644 list_splice(&tmp_list, &session->s_cap_releases);
1645 session->s_num_cap_releases += num_cap_releases;
1646 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1647}
1648
2f2dc053
SW		 1649/*
1650 * requests
1651 */
1652
54008399
YZ		 1653int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1654 struct inode *dir)
1655{
1656 struct ceph_inode_info *ci = ceph_inode(dir);
1657 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1658 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1659 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1660 int order, num_entries;
1661
1662 spin_lock(&ci->i_ceph_lock);
1663 num_entries = ci->i_files + ci->i_subdirs;
1664 spin_unlock(&ci->i_ceph_lock);
1665 num_entries = max(num_entries, 1);
1666 num_entries = min(num_entries, opt->max_readdir);
1667
1668 order = get_order(size * num_entries);
1669 while (order >= 0) {
2a5beea3
YZ		 1670 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1671 __GFP_NOWARN,
1672 order);
1673 if (rinfo->dir_entries)
54008399
YZ		 1674 break;
1675 order--;
1676 }
2a5beea3 1677 if (!rinfo->dir_entries)
54008399
YZ		 1678 return -ENOMEM;
1679
1680 num_entries = (PAGE_SIZE << order) / size;
1681 num_entries = min(num_entries, opt->max_readdir);
1682
1683 rinfo->dir_buf_size = PAGE_SIZE << order;
1684 req->r_num_caps = num_entries + 1;
1685 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1686 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1687 return 0;
1688}
1689
2f2dc053
SW		 1690/*
1691 * Create an mds request.
1692 */
1693struct ceph_mds_request *
1694ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1695{
1696 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1697
1698 if (!req)
1699 return ERR_PTR(-ENOMEM);
1700
b4556396 1701 mutex_init(&req->r_fill_mutex);
37151668 1702 req->r_mdsc = mdsc;
2f2dc053
SW		 1703 req->r_started = jiffies;
1704 req->r_resend_mds = -1;
1705 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1706 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1707 req->r_fmode = -1;
153c8e6b 1708 kref_init(&req->r_kref);
fcd00b68 1709 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1710 INIT_LIST_HEAD(&req->r_wait);
1711 init_completion(&req->r_completion);
1712 init_completion(&req->r_safe_completion);
1713 INIT_LIST_HEAD(&req->r_unsafe_item);
1714
56199016 1715 req->r_stamp = timespec_trunc(current_kernel_time(), mdsc->fsc->sb->s_time_gran);
b8e69066 1716
2f2dc053
SW		 1717 req->r_op = op;
1718 req->r_direct_mode = mode;
1719 return req;
1720}
1721
1722/*
44ca18f2 1723 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1724 *
1725 * called under mdsc->mutex.
1726 */
44ca18f2
SW		 1727static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1728{
1729 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1730 return NULL;
1731 return rb_entry(rb_first(&mdsc->request_tree),
1732 struct ceph_mds_request, r_node);
1733}
1734
e8a7b8b1 1735static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1736{
e8a7b8b1 1737 return mdsc->oldest_tid;
2f2dc053
SW		 1738}
1739
1740/*
1741 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1742 * on build_path_from_dentry in fs/cifs/dir.c.
1743 *
1744 * If @stop_on_nosnap, generate path relative to the first non-snapped
1745 * inode.
1746 *
1747 * Encode hidden .snap dirs as a double /, i.e.
1748 * foo/.snap/bar -> foo//bar
1749 */
1750char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1751 int stop_on_nosnap)
1752{
1753 struct dentry *temp;
1754 char *path;
1755 int len, pos;
1b71fe2e 1756 unsigned seq;
2f2dc053
SW		 1757
1758 if (dentry == NULL)
1759 return ERR_PTR(-EINVAL);
1760
1761retry:
1762 len = 0;
1b71fe2e
AV		 1763 seq = read_seqbegin(&rename_lock);
1764 rcu_read_lock();
2f2dc053 1765 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1766 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1767 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1768 len++; /* slash only */
1769 else if (stop_on_nosnap && inode &&
1770 ceph_snap(inode) == CEPH_NOSNAP)
1771 break;
1772 else
1773 len += 1 + temp->d_name.len;
1774 temp = temp->d_parent;
2f2dc053 1775 }
1b71fe2e 1776 rcu_read_unlock();
2f2dc053
SW		 1777 if (len)
1778 len--; /* no leading '/' */
1779
1780 path = kmalloc(len+1, GFP_NOFS);
1781 if (path == NULL)
1782 return ERR_PTR(-ENOMEM);
1783 pos = len;
1784 path[pos] = 0; /* trailing null */
1b71fe2e 1785 rcu_read_lock();
2f2dc053 1786 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1787 struct inode *inode;
2f2dc053 1788
1b71fe2e 1789 spin_lock(&temp->d_lock);
2b0143b5 1790 inode = d_inode(temp);
2f2dc053 1791 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1792 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1793 pos, temp);
1794 } else if (stop_on_nosnap && inode &&
1795 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1796 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1797 break;
1798 } else {
1799 pos -= temp->d_name.len;
1b71fe2e
AV		 1800 if (pos < 0) {
1801 spin_unlock(&temp->d_lock);
2f2dc053 1802 break;
1b71fe2e 1803 }
2f2dc053
SW		 1804 strncpy(path + pos, temp->d_name.name,
1805 temp->d_name.len);
2f2dc053 1806 }
1b71fe2e 1807 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1808 if (pos)
1809 path[--pos] = '/';
1810 temp = temp->d_parent;
2f2dc053 1811 }
1b71fe2e
AV		 1812 rcu_read_unlock();
1813 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1814 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1815 "expected, namelen is %d, pos is %d\n", len, pos);
1816 /* presumably this is only possible if racing with a
1817 rename of one of the parent directories (we can not
1818 lock the dentries above us to prevent this, but
1819 retrying should be harmless) */
1820 kfree(path);
1821 goto retry;
1822 }
1823
2b0143b5 1824 *base = ceph_ino(d_inode(temp));
2f2dc053 1825 *plen = len;
104648ad 1826 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1827 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1828 return path;
1829}
1830
fd36a717 1831static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2f2dc053
SW		 1832 const char **ppath, int *ppathlen, u64 *pino,
1833 int *pfreepath)
1834{
1835 char *path;
1836
c6b0b656 1837 rcu_read_lock();
fd36a717
JL		 1838 if (!dir)
1839 dir = d_inode_rcu(dentry->d_parent);
c6b0b656
JL		 1840 if (dir && ceph_snap(dir) == CEPH_NOSNAP) {
1841 *pino = ceph_ino(dir);
1842 rcu_read_unlock();
2f2dc053
SW		 1843 *ppath = dentry->d_name.name;
1844 *ppathlen = dentry->d_name.len;
1845 return 0;
1846 }
c6b0b656 1847 rcu_read_unlock();
2f2dc053
SW		 1848 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1849 if (IS_ERR(path))
1850 return PTR_ERR(path);
1851 *ppath = path;
1852 *pfreepath = 1;
1853 return 0;
1854}
1855
1856static int build_inode_path(struct inode *inode,
1857 const char **ppath, int *ppathlen, u64 *pino,
1858 int *pfreepath)
1859{
1860 struct dentry *dentry;
1861 char *path;
1862
1863 if (ceph_snap(inode) == CEPH_NOSNAP) {
1864 *pino = ceph_ino(inode);
1865 *ppathlen = 0;
1866 return 0;
1867 }
1868 dentry = d_find_alias(inode);
1869 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1870 dput(dentry);
1871 if (IS_ERR(path))
1872 return PTR_ERR(path);
1873 *ppath = path;
1874 *pfreepath = 1;
1875 return 0;
1876}
1877
1878/*
1879 * request arguments may be specified via an inode *, a dentry *, or
1880 * an explicit ino+path.
1881 */
1882static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
fd36a717
JL		 1883 struct inode *rdiri, const char *rpath,
1884 u64 rino, const char **ppath, int *pathlen,
2f2dc053
SW		 1885 u64 *ino, int *freepath)
1886{
1887 int r = 0;
1888
1889 if (rinode) {
1890 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1891 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1892 ceph_snap(rinode));
1893 } else if (rdentry) {
fd36a717
JL		 1894 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
1895 freepath);
2f2dc053
SW		 1896 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1897 *ppath);
795858db 1898 } else if (rpath || rino) {
2f2dc053
SW		 1899 *ino = rino;
1900 *ppath = rpath;
b000056a 1901 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1902 dout(" path %.*s\n", *pathlen, rpath);
1903 }
1904
1905 return r;
1906}
1907
1908/*
1909 * called under mdsc->mutex
1910 */
1911static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1912 struct ceph_mds_request *req,
6e6f0923 1913 int mds, bool drop_cap_releases)
2f2dc053
SW		 1914{
1915 struct ceph_msg *msg;
1916 struct ceph_mds_request_head *head;
1917 const char *path1 = NULL;
1918 const char *path2 = NULL;
1919 u64 ino1 = 0, ino2 = 0;
1920 int pathlen1 = 0, pathlen2 = 0;
1921 int freepath1 = 0, freepath2 = 0;
1922 int len;
1923 u16 releases;
1924 void *p, *end;
1925 int ret;
1926
1927 ret = set_request_path_attr(req->r_inode, req->r_dentry,
3dd69aab 1928 req->r_parent, req->r_path1, req->r_ino1.ino,
2f2dc053
SW		 1929 &path1, &pathlen1, &ino1, &freepath1);
1930 if (ret < 0) {
1931 msg = ERR_PTR(ret);
1932 goto out;
1933 }
1934
1935 ret = set_request_path_attr(NULL, req->r_old_dentry,
fd36a717 1936 req->r_old_dentry_dir,
2f2dc053
SW		 1937 req->r_path2, req->r_ino2.ino,
1938 &path2, &pathlen2, &ino2, &freepath2);
1939 if (ret < 0) {
1940 msg = ERR_PTR(ret);
1941 goto out_free1;
1942 }
1943
1944 len = sizeof(*head) +
b8e69066 1945 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 1946 sizeof(struct ceph_timespec);
2f2dc053
SW		 1947
1948 /* calculate (max) length for cap releases */
1949 len += sizeof(struct ceph_mds_request_release) *
1950 (!!req->r_inode_drop + !!req->r_dentry_drop +
1951 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1952 if (req->r_dentry_drop)
1953 len += req->r_dentry->d_name.len;
1954 if (req->r_old_dentry_drop)
1955 len += req->r_old_dentry->d_name.len;
1956
b61c2763 1957 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1958 if (!msg) {
1959 msg = ERR_PTR(-ENOMEM);
2f2dc053 1960 goto out_free2;
a79832f2 1961 }
2f2dc053 1962
7cfa0313 1963 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1964 msg->hdr.tid = cpu_to_le64(req->r_tid);
1965
2f2dc053
SW		 1966 head = msg->front.iov_base;
1967 p = msg->front.iov_base + sizeof(*head);
1968 end = msg->front.iov_base + msg->front.iov_len;
1969
1970 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1971 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1972 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1973 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1974 head->args = req->r_args;
1975
1976 ceph_encode_filepath(&p, end, ino1, path1);
1977 ceph_encode_filepath(&p, end, ino2, path2);
1978
e979cf50
SW		 1979 /* make note of release offset, in case we need to replay */
1980 req->r_request_release_offset = p - msg->front.iov_base;
1981
2f2dc053
SW		 1982 /* cap releases */
1983 releases = 0;
1984 if (req->r_inode_drop)
1985 releases += ceph_encode_inode_release(&p,
2b0143b5 1986 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 1987 mds, req->r_inode_drop, req->r_inode_unless, 0);
1988 if (req->r_dentry_drop)
1989 releases += ceph_encode_dentry_release(&p, req->r_dentry,
3dd69aab 1990 req->r_parent, mds, req->r_dentry_drop,
ca6c8ae0 1991 req->r_dentry_unless);
2f2dc053
SW		 1992 if (req->r_old_dentry_drop)
1993 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
ca6c8ae0
JL		 1994 req->r_old_dentry_dir, mds,
1995 req->r_old_dentry_drop,
1996 req->r_old_dentry_unless);
2f2dc053
SW		 1997 if (req->r_old_inode_drop)
1998 releases += ceph_encode_inode_release(&p,
2b0143b5 1999 d_inode(req->r_old_dentry),
2f2dc053 2000 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 2001
2002 if (drop_cap_releases) {
2003 releases = 0;
2004 p = msg->front.iov_base + req->r_request_release_offset;
2005 }
2006
2f2dc053
SW		 2007 head->num_releases = cpu_to_le16(releases);
2008
b8e69066 2009 /* time stamp */
1f041a89
YZ		 2010 {
2011 struct ceph_timespec ts;
2012 ceph_encode_timespec(&ts, &req->r_stamp);
2013 ceph_encode_copy(&p, &ts, sizeof(ts));
2014 }
b8e69066 2015
2f2dc053
SW		 2016 BUG_ON(p > end);
2017 msg->front.iov_len = p - msg->front.iov_base;
2018 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2019
25e6bae3
YZ		 2020 if (req->r_pagelist) {
2021 struct ceph_pagelist *pagelist = req->r_pagelist;
0e1a5ee6 2022 refcount_inc(&pagelist->refcnt);
25e6bae3
YZ		 2023 ceph_msg_data_add_pagelist(msg, pagelist);
2024 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2025 } else {
2026 msg->hdr.data_len = 0;
ebf18f47 2027 }
02afca6c 2028
2f2dc053
SW		 2029 msg->hdr.data_off = cpu_to_le16(0);
2030
2031out_free2:
2032 if (freepath2)
2033 kfree((char *)path2);
2034out_free1:
2035 if (freepath1)
2036 kfree((char *)path1);
2037out:
2038 return msg;
2039}
2040
2041/*
2042 * called under mdsc->mutex if error, under no mutex if
2043 * success.
2044 */
2045static void complete_request(struct ceph_mds_client *mdsc,
2046 struct ceph_mds_request *req)
2047{
2048 if (req->r_callback)
2049 req->r_callback(mdsc, req);
2050 else
03066f23 2051 complete_all(&req->r_completion);
2f2dc053
SW		 2052}
2053
2054/*
2055 * called under mdsc->mutex
2056 */
2057static int __prepare_send_request(struct ceph_mds_client *mdsc,
2058 struct ceph_mds_request *req,
6e6f0923 2059 int mds, bool drop_cap_releases)
2f2dc053
SW		 2060{
2061 struct ceph_mds_request_head *rhead;
2062 struct ceph_msg *msg;
2063 int flags = 0;
2064
2f2dc053 2065 req->r_attempts++;
e55b71f8
GF		 2066 if (req->r_inode) {
2067 struct ceph_cap *cap =
2068 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2069
2070 if (cap)
2071 req->r_sent_on_mseq = cap->mseq;
2072 else
2073 req->r_sent_on_mseq = -1;
2074 }
2f2dc053
SW		 2075 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2076 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2077
bc2de10d 2078 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
c5c9a0bf 2079 void *p;
01a92f17
SW		 2080 /*
2081 * Replay. Do not regenerate message (and rebuild
2082 * paths, etc.); just use the original message.
2083 * Rebuilding paths will break for renames because
2084 * d_move mangles the src name.
2085 */
2086 msg = req->r_request;
2087 rhead = msg->front.iov_base;
2088
2089 flags = le32_to_cpu(rhead->flags);
2090 flags |= CEPH_MDS_FLAG_REPLAY;
2091 rhead->flags = cpu_to_le32(flags);
2092
2093 if (req->r_target_inode)
2094 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2095
2096 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2097
2098 /* remove cap/dentry releases from message */
2099 rhead->num_releases = 0;
c5c9a0bf
YZ		 2100
2101 /* time stamp */
2102 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2103 {
2104 struct ceph_timespec ts;
2105 ceph_encode_timespec(&ts, &req->r_stamp);
2106 ceph_encode_copy(&p, &ts, sizeof(ts));
2107 }
c5c9a0bf
YZ		 2108
2109 msg->front.iov_len = p - msg->front.iov_base;
2110 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2111 return 0;
2112 }
2113
2f2dc053
SW		 2114 if (req->r_request) {
2115 ceph_msg_put(req->r_request);
2116 req->r_request = NULL;
2117 }
6e6f0923 2118 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2119 if (IS_ERR(msg)) {
e1518c7c 2120 req->r_err = PTR_ERR(msg);
a79832f2 2121 return PTR_ERR(msg);
2f2dc053
SW		 2122 }
2123 req->r_request = msg;
2124
2125 rhead = msg->front.iov_base;
2f2dc053 2126 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
bc2de10d 2127 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2f2dc053 2128 flags |= CEPH_MDS_FLAG_REPLAY;
3dd69aab 2129 if (req->r_parent)
2f2dc053
SW		 2130 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2131 rhead->flags = cpu_to_le32(flags);
2132 rhead->num_fwd = req->r_num_fwd;
2133 rhead->num_retry = req->r_attempts - 1;
01a92f17 2134 rhead->ino = 0;
2f2dc053 2135
3dd69aab 2136 dout(" r_parent = %p\n", req->r_parent);
2f2dc053
SW		 2137 return 0;
2138}
2139
2140/*
2141 * send request, or put it on the appropriate wait list.
2142 */
2143static int __do_request(struct ceph_mds_client *mdsc,
2144 struct ceph_mds_request *req)
2145{
2146 struct ceph_mds_session *session = NULL;
2147 int mds = -1;
48fec5d0 2148 int err = 0;
2f2dc053 2149
bc2de10d
JL		 2150 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2151 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
eb1b8af3 2152 __unregister_request(mdsc, req);
2f2dc053 2153 goto out;
eb1b8af3 2154 }
2f2dc053
SW		 2155
2156 if (req->r_timeout &&
2157 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2158 dout("do_request timed out\n");
2159 err = -EIO;
2160 goto finish;
2161 }
52953d55 2162 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
48fec5d0
YZ		 2163 dout("do_request forced umount\n");
2164 err = -EIO;
2165 goto finish;
2166 }
52953d55 2167 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
e9e427f0
YZ		 2168 if (mdsc->mdsmap_err) {
2169 err = mdsc->mdsmap_err;
2170 dout("do_request mdsmap err %d\n", err);
2171 goto finish;
2172 }
cc8e8342
YZ		 2173 if (mdsc->mdsmap->m_epoch == 0) {
2174 dout("do_request no mdsmap, waiting for map\n");
2175 list_add(&req->r_wait, &mdsc->waiting_for_map);
2176 goto finish;
2177 }
e9e427f0
YZ		 2178 if (!(mdsc->fsc->mount_options->flags &
2179 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2180 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2181 err = -ENOENT;
2182 pr_info("probably no mds server is up\n");
2183 goto finish;
2184 }
2185 }
2f2dc053 2186
dc69e2e9
SW		 2187 put_request_session(req);
2188
2f2dc053
SW		 2189 mds = __choose_mds(mdsc, req);
2190 if (mds < 0 ||
2191 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2192 dout("do_request no mds or not active, waiting for map\n");
2193 list_add(&req->r_wait, &mdsc->waiting_for_map);
2194 goto out;
2195 }
2196
2197 /* get, open session */
2198 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2199 if (!session) {
2f2dc053 2200 session = register_session(mdsc, mds);
9c423956
SW		 2201 if (IS_ERR(session)) {
2202 err = PTR_ERR(session);
2203 goto finish;
2204 }
2205 }
dc69e2e9
SW		 2206 req->r_session = get_session(session);
2207
2f2dc053 2208 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2209 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2210 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2211 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2212 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2213 err = -EACCES;
2214 goto out_session;
2215 }
2f2dc053
SW		 2216 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2217 session->s_state == CEPH_MDS_SESSION_CLOSING)
2218 __open_session(mdsc, session);
2219 list_add(&req->r_wait, &session->s_waiting);
2220 goto out_session;
2221 }
2222
2223 /* send request */
2f2dc053
SW		 2224 req->r_resend_mds = -1; /* forget any previous mds hint */
2225
2226 if (req->r_request_started == 0) /* note request start time */
2227 req->r_request_started = jiffies;
2228
6e6f0923 2229 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2230 if (!err) {
2231 ceph_msg_get(req->r_request);
2232 ceph_con_send(&session->s_con, req->r_request);
2233 }
2234
2235out_session:
2236 ceph_put_mds_session(session);
48fec5d0
YZ		 2237finish:
2238 if (err) {
2239 dout("__do_request early error %d\n", err);
2240 req->r_err = err;
2241 complete_request(mdsc, req);
2242 __unregister_request(mdsc, req);
2243 }
2f2dc053
SW		 2244out:
2245 return err;
2f2dc053
SW		 2246}
2247
2248/*
2249 * called under mdsc->mutex
2250 */
2251static void __wake_requests(struct ceph_mds_client *mdsc,
2252 struct list_head *head)
2253{
ed75ec2c
YZ		 2254 struct ceph_mds_request *req;
2255 LIST_HEAD(tmp_list);
2256
2257 list_splice_init(head, &tmp_list);
2f2dc053 2258
ed75ec2c
YZ		 2259 while (!list_empty(&tmp_list)) {
2260 req = list_entry(tmp_list.next,
2261 struct ceph_mds_request, r_wait);
2f2dc053 2262 list_del_init(&req->r_wait);
7971bd92 2263 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2264 __do_request(mdsc, req);
2265 }
2266}
2267
2268/*
2269 * Wake up threads with requests pending for @mds, so that they can
29790f26 2270 * resubmit their requests to a possibly different mds.
2f2dc053 2271 */
29790f26 2272static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2273{
44ca18f2 2274 struct ceph_mds_request *req;
282c1052 2275 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2276
2277 dout("kick_requests mds%d\n", mds);
282c1052 2278 while (p) {
44ca18f2 2279 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2280 p = rb_next(p);
bc2de10d 2281 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
44ca18f2 2282 continue;
3de22be6
YZ		 2283 if (req->r_attempts > 0)
2284 continue; /* only new requests */
44ca18f2
SW		 2285 if (req->r_session &&
2286 req->r_session->s_mds == mds) {
2287 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2288 list_del_init(&req->r_wait);
44ca18f2 2289 __do_request(mdsc, req);
2f2dc053
SW		 2290 }
2291 }
2292}
2293
2294void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2295 struct ceph_mds_request *req)
2296{
2297 dout("submit_request on %p\n", req);
2298 mutex_lock(&mdsc->mutex);
2299 __register_request(mdsc, req, NULL);
2300 __do_request(mdsc, req);
2301 mutex_unlock(&mdsc->mutex);
2302}
2303
2304/*
2305 * Synchrously perform an mds request. Take care of all of the
2306 * session setup, forwarding, retry details.
2307 */
2308int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2309 struct inode *dir,
2310 struct ceph_mds_request *req)
2311{
2312 int err;
2313
2314 dout("do_request on %p\n", req);
2315
3dd69aab 2316 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2f2dc053
SW		 2317 if (req->r_inode)
2318 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3dd69aab
JL		 2319 if (req->r_parent)
2320 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
844d87c3 2321 if (req->r_old_dentry_dir)
41b02e1f
SW		 2322 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2323 CEPH_CAP_PIN);
2f2dc053
SW		 2324
2325 /* issue */
2326 mutex_lock(&mdsc->mutex);
2327 __register_request(mdsc, req, dir);
2328 __do_request(mdsc, req);
2329
e1518c7c
SW		 2330 if (req->r_err) {
2331 err = req->r_err;
e1518c7c 2332 goto out;
2f2dc053
SW		 2333 }
2334
e1518c7c
SW		 2335 /* wait */
2336 mutex_unlock(&mdsc->mutex);
2337 dout("do_request waiting\n");
5be73034 2338 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2339 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2340 } else {
5be73034
ID		 2341 long timeleft = wait_for_completion_killable_timeout(
2342 &req->r_completion,
2343 ceph_timeout_jiffies(req->r_timeout));
2344 if (timeleft > 0)
2345 err = 0;
2346 else if (!timeleft)
2347 err = -EIO; /* timed out */
2348 else
2349 err = timeleft; /* killed */
e1518c7c
SW		 2350 }
2351 dout("do_request waited, got %d\n", err);
2352 mutex_lock(&mdsc->mutex);
5b1daecd 2353
e1518c7c 2354 /* only abort if we didn't race with a real reply */
bc2de10d 2355 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
e1518c7c
SW		 2356 err = le32_to_cpu(req->r_reply_info.head->result);
2357 } else if (err < 0) {
2358 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2359
2360 /*
2361 * ensure we aren't running concurrently with
2362 * ceph_fill_trace or ceph_readdir_prepopulate, which
2363 * rely on locks (dir mutex) held by our caller.
2364 */
2365 mutex_lock(&req->r_fill_mutex);
e1518c7c 2366 req->r_err = err;
bc2de10d 2367 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
b4556396 2368 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2369
3dd69aab 2370 if (req->r_parent &&
167c9e35
SW		 2371 (req->r_op & CEPH_MDS_OP_WRITE))
2372 ceph_invalidate_dir_request(req);
2f2dc053 2373 } else {
e1518c7c 2374 err = req->r_err;
2f2dc053 2375 }
2f2dc053 2376
e1518c7c
SW		 2377out:
2378 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2379 dout("do_request %p done, result %d\n", req, err);
2380 return err;
2381}
2382
167c9e35 2383/*
2f276c51 2384 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2385 * namespace request.
2386 */
2387void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2388{
3dd69aab 2389 struct inode *inode = req->r_parent;
167c9e35 2390
2f276c51 2391 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2392
2f276c51 2393 ceph_dir_clear_complete(inode);
167c9e35
SW		 2394 if (req->r_dentry)
2395 ceph_invalidate_dentry_lease(req->r_dentry);
2396 if (req->r_old_dentry)
2397 ceph_invalidate_dentry_lease(req->r_old_dentry);
2398}
2399
2f2dc053
SW		 2400/*
2401 * Handle mds reply.
2402 *
2403 * We take the session mutex and parse and process the reply immediately.
2404 * This preserves the logical ordering of replies, capabilities, etc., sent
2405 * by the MDS as they are applied to our local cache.
2406 */
2407static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2408{
2409 struct ceph_mds_client *mdsc = session->s_mdsc;
2410 struct ceph_mds_request *req;
2411 struct ceph_mds_reply_head *head = msg->front.iov_base;
2412 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2413 struct ceph_snap_realm *realm;
2f2dc053
SW		 2414 u64 tid;
2415 int err, result;
2600d2dd 2416 int mds = session->s_mds;
2f2dc053 2417
2f2dc053
SW		 2418 if (msg->front.iov_len < sizeof(*head)) {
2419 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2420 ceph_msg_dump(msg);
2f2dc053
SW		 2421 return;
2422 }
2423
2424 /* get request, session */
6df058c0 2425 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2426 mutex_lock(&mdsc->mutex);
fcd00b68 2427 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2428 if (!req) {
2429 dout("handle_reply on unknown tid %llu\n", tid);
2430 mutex_unlock(&mdsc->mutex);
2431 return;
2432 }
2433 dout("handle_reply %p\n", req);
2f2dc053
SW		 2434
2435 /* correct session? */
d96d6049 2436 if (req->r_session != session) {
2f2dc053
SW		 2437 pr_err("mdsc_handle_reply got %llu on session mds%d"
2438 " not mds%d\n", tid, session->s_mds,
2439 req->r_session ? req->r_session->s_mds : -1);
2440 mutex_unlock(&mdsc->mutex);
2441 goto out;
2442 }
2443
2444 /* dup? */
bc2de10d
JL		 2445 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2446 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
f3ae1b97 2447 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2448 head->safe ? "safe" : "unsafe", tid, mds);
2449 mutex_unlock(&mdsc->mutex);
2450 goto out;
2451 }
bc2de10d 2452 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
f3ae1b97 2453 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2454 tid, mds);
2455 mutex_unlock(&mdsc->mutex);
2456 goto out;
2457 }
2f2dc053
SW		 2458
2459 result = le32_to_cpu(head->result);
2460
2461 /*
e55b71f8
GF		 2462 * Handle an ESTALE
2463 * if we're not talking to the authority, send to them
2464 * if the authority has changed while we weren't looking,
2465 * send to new authority
2466 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2467 */
2468 if (result == -ESTALE) {
e55b71f8 2469 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2470 req->r_resend_mds = -1;
ca18bede 2471 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2472 dout("not using auth, setting for that now");
2473 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2474 __do_request(mdsc, req);
2475 mutex_unlock(&mdsc->mutex);
2476 goto out;
e55b71f8 2477 } else {
ca18bede
YZ		 2478 int mds = __choose_mds(mdsc, req);
2479 if (mds >= 0 && mds != req->r_session->s_mds) {
2480 dout("but auth changed, so resending");
e55b71f8
GF		 2481 __do_request(mdsc, req);
2482 mutex_unlock(&mdsc->mutex);
2483 goto out;
2484 }
2f2dc053 2485 }
e55b71f8 2486 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2487 }
2488
e55b71f8 2489
2f2dc053 2490 if (head->safe) {
bc2de10d 2491 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2f2dc053 2492 __unregister_request(mdsc, req);
2f2dc053 2493
bc2de10d 2494 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053
SW		 2495 /*
2496 * We already handled the unsafe response, now do the
2497 * cleanup. No need to examine the response; the MDS
2498 * doesn't include any result info in the safe
2499 * response. And even if it did, there is nothing
2500 * useful we could do with a revised return value.
2501 */
2502 dout("got safe reply %llu, mds%d\n", tid, mds);
2f2dc053
SW		 2503
2504 /* last unsafe request during umount? */
44ca18f2 2505 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2506 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2507 mutex_unlock(&mdsc->mutex);
2508 goto out;
2509 }
e1518c7c 2510 } else {
bc2de10d 2511 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2f2dc053 2512 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2513 if (req->r_unsafe_dir) {
2514 struct ceph_inode_info *ci =
2515 ceph_inode(req->r_unsafe_dir);
2516 spin_lock(&ci->i_unsafe_lock);
2517 list_add_tail(&req->r_unsafe_dir_item,
2518 &ci->i_unsafe_dirops);
2519 spin_unlock(&ci->i_unsafe_lock);
2520 }
2f2dc053
SW		 2521 }
2522
2523 dout("handle_reply tid %lld result %d\n", tid, result);
2524 rinfo = &req->r_reply_info;
14303d20 2525 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2526 mutex_unlock(&mdsc->mutex);
2527
2528 mutex_lock(&session->s_mutex);
2529 if (err < 0) {
25933abd 2530 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2531 ceph_msg_dump(msg);
2f2dc053
SW		 2532 goto out_err;
2533 }
2534
2535 /* snap trace */
982d6011 2536 realm = NULL;
2f2dc053
SW		 2537 if (rinfo->snapblob_len) {
2538 down_write(&mdsc->snap_rwsem);
2539 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2540 rinfo->snapblob + rinfo->snapblob_len,
2541 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2542 &realm);
2f2dc053
SW		 2543 downgrade_write(&mdsc->snap_rwsem);
2544 } else {
2545 down_read(&mdsc->snap_rwsem);
2546 }
2547
2548 /* insert trace into our cache */
b4556396 2549 mutex_lock(&req->r_fill_mutex);
315f2408 2550 current->journal_info = req;
f5a03b08 2551 err = ceph_fill_trace(mdsc->fsc->sb, req);
2f2dc053 2552 if (err == 0) {
6e8575fa 2553 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2554 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2555 ceph_readdir_prepopulate(req, req->r_session);
37151668 2556 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2557 }
315f2408 2558 current->journal_info = NULL;
b4556396 2559 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2560
2561 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2562 if (realm)
2563 ceph_put_snap_realm(mdsc, realm);
68cd5b4b 2564
bc2de10d
JL		 2565 if (err == 0 && req->r_target_inode &&
2566 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 2567 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2568 spin_lock(&ci->i_unsafe_lock);
2569 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2570 spin_unlock(&ci->i_unsafe_lock);
2571 }
2f2dc053 2572out_err:
e1518c7c 2573 mutex_lock(&mdsc->mutex);
bc2de10d 2574 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
e1518c7c
SW		 2575 if (err) {
2576 req->r_err = err;
2577 } else {
5fdb1389 2578 req->r_reply = ceph_msg_get(msg);
bc2de10d 2579 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
e1518c7c 2580 }
2f2dc053 2581 } else {
e1518c7c 2582 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2583 }
e1518c7c 2584 mutex_unlock(&mdsc->mutex);
2f2dc053 2585
2f2dc053
SW		 2586 mutex_unlock(&session->s_mutex);
2587
2588 /* kick calling process */
2589 complete_request(mdsc, req);
2590out:
2591 ceph_mdsc_put_request(req);
2592 return;
2593}
2594
2595
2596
2597/*
2598 * handle mds notification that our request has been forwarded.
2599 */
2600d2dd
SW		 2600static void handle_forward(struct ceph_mds_client *mdsc,
2601 struct ceph_mds_session *session,
2602 struct ceph_msg *msg)
2f2dc053
SW		 2603{
2604 struct ceph_mds_request *req;
a1ea787c 2605 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2606 u32 next_mds;
2607 u32 fwd_seq;
2f2dc053
SW		 2608 int err = -EINVAL;
2609 void *p = msg->front.iov_base;
2610 void *end = p + msg->front.iov_len;
2f2dc053 2611
a1ea787c 2612 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2613 next_mds = ceph_decode_32(&p);
2614 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2615
2616 mutex_lock(&mdsc->mutex);
fcd00b68 2617 req = lookup_get_request(mdsc, tid);
2f2dc053 2618 if (!req) {
2a8e5e36 2619 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2620 goto out; /* dup reply? */
2621 }
2622
bc2de10d 2623 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2a8e5e36
SW		 2624 dout("forward tid %llu aborted, unregistering\n", tid);
2625 __unregister_request(mdsc, req);
2626 } else if (fwd_seq <= req->r_num_fwd) {
2627 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2628 tid, next_mds, req->r_num_fwd, fwd_seq);
2629 } else {
2630 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2631 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2632 BUG_ON(req->r_err);
bc2de10d 2633 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3de22be6 2634 req->r_attempts = 0;
2f2dc053
SW		 2635 req->r_num_fwd = fwd_seq;
2636 req->r_resend_mds = next_mds;
2637 put_request_session(req);
2638 __do_request(mdsc, req);
2639 }
2640 ceph_mdsc_put_request(req);
2641out:
2642 mutex_unlock(&mdsc->mutex);
2643 return;
2644
2645bad:
2646 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2647}
2648
2649/*
2650 * handle a mds session control message
2651 */
2652static void handle_session(struct ceph_mds_session *session,
2653 struct ceph_msg *msg)
2654{
2655 struct ceph_mds_client *mdsc = session->s_mdsc;
2656 u32 op;
2657 u64 seq;
2600d2dd 2658 int mds = session->s_mds;
2f2dc053
SW		 2659 struct ceph_mds_session_head *h = msg->front.iov_base;
2660 int wake = 0;
2661
2f2dc053
SW		 2662 /* decode */
2663 if (msg->front.iov_len != sizeof(*h))
2664 goto bad;
2665 op = le32_to_cpu(h->op);
2666 seq = le64_to_cpu(h->seq);
2667
2668 mutex_lock(&mdsc->mutex);
0a07fc8c
YZ		 2669 if (op == CEPH_SESSION_CLOSE) {
2670 get_session(session);
2600d2dd 2671 __unregister_session(mdsc, session);
0a07fc8c 2672 }
2f2dc053
SW		 2673 /* FIXME: this ttl calculation is generous */
2674 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2675 mutex_unlock(&mdsc->mutex);
2676
2677 mutex_lock(&session->s_mutex);
2678
2679 dout("handle_session mds%d %s %p state %s seq %llu\n",
2680 mds, ceph_session_op_name(op), session,
a687ecaf 2681 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2682
2683 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2684 session->s_state = CEPH_MDS_SESSION_OPEN;
2685 pr_info("mds%d came back\n", session->s_mds);
2686 }
2687
2688 switch (op) {
2689 case CEPH_SESSION_OPEN:
29790f26
SW		 2690 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2691 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2692 session->s_state = CEPH_MDS_SESSION_OPEN;
2693 renewed_caps(mdsc, session, 0);
2694 wake = 1;
2695 if (mdsc->stopping)
2696 __close_session(mdsc, session);
2697 break;
2698
2699 case CEPH_SESSION_RENEWCAPS:
2700 if (session->s_renew_seq == seq)
2701 renewed_caps(mdsc, session, 1);
2702 break;
2703
2704 case CEPH_SESSION_CLOSE:
29790f26
SW		 2705 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2706 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2707 cleanup_session_requests(mdsc, session);
2f2dc053 2708 remove_session_caps(session);
656e4382 2709 wake = 2; /* for good measure */
f3c60c59 2710 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2711 break;
2712
2713 case CEPH_SESSION_STALE:
2714 pr_info("mds%d caps went stale, renewing\n",
2715 session->s_mds);
d8fb02ab 2716 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2717 session->s_cap_gen++;
1ce208a6 2718 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2719 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2720 send_renew_caps(mdsc, session);
2721 break;
2722
2723 case CEPH_SESSION_RECALL_STATE:
2724 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2725 break;
2726
186e4f7a
YZ		 2727 case CEPH_SESSION_FLUSHMSG:
2728 send_flushmsg_ack(mdsc, session, seq);
2729 break;
2730
03f4fcb0
YZ		 2731 case CEPH_SESSION_FORCE_RO:
2732 dout("force_session_readonly %p\n", session);
2733 spin_lock(&session->s_cap_lock);
2734 session->s_readonly = true;
2735 spin_unlock(&session->s_cap_lock);
2736 wake_up_session_caps(session, 0);
2737 break;
2738
fcff415c
YZ		 2739 case CEPH_SESSION_REJECT:
2740 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2741 pr_info("mds%d rejected session\n", session->s_mds);
2742 session->s_state = CEPH_MDS_SESSION_REJECTED;
2743 cleanup_session_requests(mdsc, session);
2744 remove_session_caps(session);
2745 wake = 2; /* for good measure */
2746 break;
2747
2f2dc053
SW		 2748 default:
2749 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2750 WARN_ON(1);
2751 }
2752
2753 mutex_unlock(&session->s_mutex);
2754 if (wake) {
2755 mutex_lock(&mdsc->mutex);
2756 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2757 if (wake == 2)
2758 kick_requests(mdsc, mds);
2f2dc053
SW		 2759 mutex_unlock(&mdsc->mutex);
2760 }
0a07fc8c
YZ		 2761 if (op == CEPH_SESSION_CLOSE)
2762 ceph_put_mds_session(session);
2f2dc053
SW		 2763 return;
2764
2765bad:
2766 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2767 (int)msg->front.iov_len);
9ec7cab1 2768 ceph_msg_dump(msg);
2f2dc053
SW		 2769 return;
2770}
2771
2772
2773/*
2774 * called under session->mutex.
2775 */
2776static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2777 struct ceph_mds_session *session)
2778{
2779 struct ceph_mds_request *req, *nreq;
3de22be6 2780 struct rb_node *p;
2f2dc053
SW		 2781 int err;
2782
2783 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2784
2785 mutex_lock(&mdsc->mutex);
2786 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2787 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2788 if (!err) {
2789 ceph_msg_get(req->r_request);
2790 ceph_con_send(&session->s_con, req->r_request);
2791 }
2792 }
3de22be6
YZ		 2793
2794 /*
2795 * also re-send old requests when MDS enters reconnect stage. So that MDS
2796 * can process completed request in clientreplay stage.
2797 */
2798 p = rb_first(&mdsc->request_tree);
2799 while (p) {
2800 req = rb_entry(p, struct ceph_mds_request, r_node);
2801 p = rb_next(p);
bc2de10d 2802 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3de22be6
YZ		 2803 continue;
2804 if (req->r_attempts == 0)
2805 continue; /* only old requests */
2806 if (req->r_session &&
2807 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2808 err = __prepare_send_request(mdsc, req,
2809 session->s_mds, true);
3de22be6
YZ		 2810 if (!err) {
2811 ceph_msg_get(req->r_request);
2812 ceph_con_send(&session->s_con, req->r_request);
2813 }
2814 }
2815 }
2f2dc053
SW		 2816 mutex_unlock(&mdsc->mutex);
2817}
2818
2819/*
2820 * Encode information about a cap for a reconnect with the MDS.
2821 */
2f2dc053
SW		 2822static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2823 void *arg)
2824{
20cb34ae
SW		 2825 union {
2826 struct ceph_mds_cap_reconnect v2;
2827 struct ceph_mds_cap_reconnect_v1 v1;
2828 } rec;
2f2dc053 2829 struct ceph_inode_info *ci;
20cb34ae
SW		 2830 struct ceph_reconnect_state *recon_state = arg;
2831 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2832 char *path;
2833 int pathlen, err;
2834 u64 pathbase;
3469ed0d 2835 u64 snap_follows;
2f2dc053
SW		 2836 struct dentry *dentry;
2837
2838 ci = cap->ci;
2839
2840 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2841 inode, ceph_vinop(inode), cap, cap->cap_id,
2842 ceph_cap_string(cap->issued));
93cea5be
SW		 2843 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2844 if (err)
2845 return err;
2f2dc053
SW		 2846
2847 dentry = d_find_alias(inode);
2848 if (dentry) {
2849 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2850 if (IS_ERR(path)) {
2851 err = PTR_ERR(path);
e072f8aa 2852 goto out_dput;
2f2dc053
SW		 2853 }
2854 } else {
2855 path = NULL;
2856 pathlen = 0;
4eacd4cb 2857 pathbase = 0;
2f2dc053 2858 }
2f2dc053 2859
be655596 2860 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2861 cap->seq = 0; /* reset cap seq */
2862 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2863 cap->mseq = 0; /* and migrate_seq */
99a9c273 2864 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2865
121f22a1 2866 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2867 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2868 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2869 rec.v2.issued = cpu_to_le32(cap->issued);
2870 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2871 rec.v2.pathbase = cpu_to_le64(pathbase);
2872 rec.v2.flock_len = 0;
20cb34ae
SW		 2873 } else {
2874 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2875 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2876 rec.v1.issued = cpu_to_le32(cap->issued);
2877 rec.v1.size = cpu_to_le64(inode->i_size);
2878 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2879 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2880 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2881 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2882 }
3469ed0d
YZ		 2883
2884 if (list_empty(&ci->i_cap_snaps)) {
2885 snap_follows = 0;
2886 } else {
2887 struct ceph_cap_snap *capsnap =
2888 list_first_entry(&ci->i_cap_snaps,
2889 struct ceph_cap_snap, ci_item);
2890 snap_follows = capsnap->follows;
20cb34ae 2891 }
be655596 2892 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2893
121f22a1 2894 if (recon_state->msg_version >= 2) {
40819f6f 2895 int num_fcntl_locks, num_flock_locks;
39be95e9 2896 struct ceph_filelock *flocks;
121f22a1
YZ		 2897 size_t struct_len, total_len = 0;
2898 u8 struct_v = 0;
39be95e9
JS		 2899
2900encode_again:
39be95e9 2901 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
39be95e9
JS		 2902 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2903 sizeof(struct ceph_filelock), GFP_NOFS);
2904 if (!flocks) {
2905 err = -ENOMEM;
2906 goto out_free;
2907 }
39be95e9
JS		 2908 err = ceph_encode_locks_to_buffer(inode, flocks,
2909 num_fcntl_locks,
2910 num_flock_locks);
39be95e9
JS		 2911 if (err) {
2912 kfree(flocks);
2913 if (err == -ENOSPC)
2914 goto encode_again;
2915 goto out_free;
2916 }
121f22a1
YZ		 2917
2918 if (recon_state->msg_version >= 3) {
2919 /* version, compat_version and struct_len */
2920 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 2921 struct_v = 2;
121f22a1 2922 }
39be95e9
JS		 2923 /*
2924 * number of encoded locks is stable, so copy to pagelist
2925 */
121f22a1
YZ		 2926 struct_len = 2 * sizeof(u32) +
2927 (num_fcntl_locks + num_flock_locks) *
2928 sizeof(struct ceph_filelock);
2929 rec.v2.flock_len = cpu_to_le32(struct_len);
2930
2931 struct_len += sizeof(rec.v2);
2932 struct_len += sizeof(u32) + pathlen;
2933
3469ed0d
YZ		 2934 if (struct_v >= 2)
2935 struct_len += sizeof(u64); /* snap_follows */
2936
121f22a1
YZ		 2937 total_len += struct_len;
2938 err = ceph_pagelist_reserve(pagelist, total_len);
2939
2940 if (!err) {
2941 if (recon_state->msg_version >= 3) {
2942 ceph_pagelist_encode_8(pagelist, struct_v);
2943 ceph_pagelist_encode_8(pagelist, 1);
2944 ceph_pagelist_encode_32(pagelist, struct_len);
2945 }
2946 ceph_pagelist_encode_string(pagelist, path, pathlen);
2947 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
2948 ceph_locks_to_pagelist(flocks, pagelist,
2949 num_fcntl_locks,
2950 num_flock_locks);
3469ed0d
YZ		 2951 if (struct_v >= 2)
2952 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 2953 }
39be95e9 2954 kfree(flocks);
3612abbd 2955 } else {
121f22a1
YZ		 2956 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
2957 err = ceph_pagelist_reserve(pagelist, size);
2958 if (!err) {
2959 ceph_pagelist_encode_string(pagelist, path, pathlen);
2960 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
2961 }
40819f6f 2962 }
44c99757
YZ		 2963
2964 recon_state->nr_caps++;
e072f8aa 2965out_free:
2f2dc053 2966 kfree(path);
e072f8aa 2967out_dput:
2f2dc053 2968 dput(dentry);
93cea5be 2969 return err;
2f2dc053
SW		 2970}
2971
2972
2973/*
2974 * If an MDS fails and recovers, clients need to reconnect in order to
2975 * reestablish shared state. This includes all caps issued through
2976 * this session _and_ the snap_realm hierarchy. Because it's not
2977 * clear which snap realms the mds cares about, we send everything we
2978 * know about.. that ensures we'll then get any new info the
2979 * recovering MDS might have.
2980 *
2981 * This is a relatively heavyweight operation, but it's rare.
2982 *
2983 * called with mdsc->mutex held.
2984 */
34b6c855
SW		 2985static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2986 struct ceph_mds_session *session)
2f2dc053 2987{
2f2dc053 2988 struct ceph_msg *reply;
a105f00c 2989 struct rb_node *p;
34b6c855 2990 int mds = session->s_mds;
9abf82b8 2991 int err = -ENOMEM;
44c99757 2992 int s_nr_caps;
93cea5be 2993 struct ceph_pagelist *pagelist;
20cb34ae 2994 struct ceph_reconnect_state recon_state;
2f2dc053 2995
34b6c855 2996 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2997
93cea5be
SW		 2998 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2999 if (!pagelist)
3000 goto fail_nopagelist;
3001 ceph_pagelist_init(pagelist);
3002
b61c2763 3003 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 3004 if (!reply)
93cea5be 3005 goto fail_nomsg;
93cea5be 3006
34b6c855
SW		 3007 mutex_lock(&session->s_mutex);
3008 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3009 session->s_seq = 0;
2f2dc053 3010
2f2dc053 3011 dout("session %p state %s\n", session,
a687ecaf 3012 ceph_session_state_name(session->s_state));
2f2dc053 3013
99a9c273
YZ		 3014 spin_lock(&session->s_gen_ttl_lock);
3015 session->s_cap_gen++;
3016 spin_unlock(&session->s_gen_ttl_lock);
3017
3018 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 3019 /* don't know if session is readonly */
3020 session->s_readonly = 0;
99a9c273
YZ		 3021 /*
3022 * notify __ceph_remove_cap() that we are composing cap reconnect.
3023 * If a cap get released before being added to the cap reconnect,
3024 * __ceph_remove_cap() should skip queuing cap release.
3025 */
3026 session->s_cap_reconnect = 1;
e01a5946 3027 /* drop old cap expires; we're about to reestablish that state */
745a8e3b 3028 cleanup_cap_releases(mdsc, session);
e01a5946 3029
5d23371f 3030 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 3031 if (mdsc->fsc->sb->s_root)
3032 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 3033
3034 ceph_con_close(&session->s_con);
3035 ceph_con_open(&session->s_con,
3036 CEPH_ENTITY_TYPE_MDS, mds,
3037 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3038
3039 /* replay unsafe requests */
3040 replay_unsafe_requests(mdsc, session);
3041
3042 down_read(&mdsc->snap_rwsem);
3043
2f2dc053 3044 /* traverse this session's caps */
44c99757
YZ		 3045 s_nr_caps = session->s_nr_caps;
3046 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 3047 if (err)
3048 goto fail;
20cb34ae 3049
44c99757 3050 recon_state.nr_caps = 0;
20cb34ae 3051 recon_state.pagelist = pagelist;
121f22a1
YZ		 3052 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
3053 recon_state.msg_version = 3;
3054 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
3055 recon_state.msg_version = 2;
3056 else
3057 recon_state.msg_version = 1;
20cb34ae 3058 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 3059 if (err < 0)
9abf82b8 3060 goto fail;
2f2dc053 3061
99a9c273
YZ		 3062 spin_lock(&session->s_cap_lock);
3063 session->s_cap_reconnect = 0;
3064 spin_unlock(&session->s_cap_lock);
3065
2f2dc053
SW		 3066 /*
3067 * snaprealms. we provide mds with the ino, seq (version), and
3068 * parent for all of our realms. If the mds has any newer info,
3069 * it will tell us.
3070 */
a105f00c
SW		 3071 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3072 struct ceph_snap_realm *realm =
3073 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3074 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3075
3076 dout(" adding snap realm %llx seq %lld parent %llx\n",
3077 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3078 sr_rec.ino = cpu_to_le64(realm->ino);
3079 sr_rec.seq = cpu_to_le64(realm->seq);
3080 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3081 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3082 if (err)
3083 goto fail;
2f2dc053 3084 }
2f2dc053 3085
121f22a1 3086 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3087
3088 /* raced with cap release? */
3089 if (s_nr_caps != recon_state.nr_caps) {
3090 struct page *page = list_first_entry(&pagelist->head,
3091 struct page, lru);
3092 __le32 *addr = kmap_atomic(page);
3093 *addr = cpu_to_le32(recon_state.nr_caps);
3094 kunmap_atomic(addr);
ebf18f47 3095 }
44c99757
YZ		 3096
3097 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3098 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3099
3100 ceph_early_kick_flushing_caps(mdsc, session);
3101
2f2dc053
SW		 3102 ceph_con_send(&session->s_con, reply);
3103
9abf82b8
SW		 3104 mutex_unlock(&session->s_mutex);
3105
3106 mutex_lock(&mdsc->mutex);
3107 __wake_requests(mdsc, &session->s_waiting);
3108 mutex_unlock(&mdsc->mutex);
3109
2f2dc053 3110 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3111 return;
3112
93cea5be 3113fail:
2f2dc053 3114 ceph_msg_put(reply);
9abf82b8
SW		 3115 up_read(&mdsc->snap_rwsem);
3116 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3117fail_nomsg:
3118 ceph_pagelist_release(pagelist);
93cea5be 3119fail_nopagelist:
9abf82b8 3120 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3121 return;
2f2dc053
SW		 3122}
3123
3124
3125/*
3126 * compare old and new mdsmaps, kicking requests
3127 * and closing out old connections as necessary
3128 *
3129 * called under mdsc->mutex.
3130 */
3131static void check_new_map(struct ceph_mds_client *mdsc,
3132 struct ceph_mdsmap *newmap,
3133 struct ceph_mdsmap *oldmap)
3134{
3135 int i;
3136 int oldstate, newstate;
3137 struct ceph_mds_session *s;
3138
3139 dout("check_new_map new %u old %u\n",
3140 newmap->m_epoch, oldmap->m_epoch);
3141
76201b63 3142 for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
2f2dc053
SW		 3143 if (mdsc->sessions[i] == NULL)
3144 continue;
3145 s = mdsc->sessions[i];
3146 oldstate = ceph_mdsmap_get_state(oldmap, i);
3147 newstate = ceph_mdsmap_get_state(newmap, i);
3148
0deb01c9 3149 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3150 i, ceph_mds_state_name(oldstate),
0deb01c9 3151 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3152 ceph_mds_state_name(newstate),
0deb01c9 3153 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3154 ceph_session_state_name(s->s_state));
2f2dc053 3155
76201b63 3156 if (i >= newmap->m_num_mds ||
3e8f43a0 3157 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3158 ceph_mdsmap_get_addr(newmap, i),
3159 sizeof(struct ceph_entity_addr))) {
3160 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3161 /* the session never opened, just close it
3162 * out now */
0a07fc8c 3163 get_session(s);
2600d2dd 3164 __unregister_session(mdsc, s);
2f2dc053 3165 __wake_requests(mdsc, &s->s_waiting);
0a07fc8c 3166 ceph_put_mds_session(s);
2827528d
YZ		 3167 } else if (i >= newmap->m_num_mds) {
3168 /* force close session for stopped mds */
3169 get_session(s);
2600d2dd 3170 __unregister_session(mdsc, s);
2827528d
YZ		 3171 __wake_requests(mdsc, &s->s_waiting);
3172 kick_requests(mdsc, i);
3173 mutex_unlock(&mdsc->mutex);
3174
3175 mutex_lock(&s->s_mutex);
3176 cleanup_session_requests(mdsc, s);
3177 remove_session_caps(s);
3178 mutex_unlock(&s->s_mutex);
3179
3180 ceph_put_mds_session(s);
3181
3182 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3183 } else {
3184 /* just close it */
3185 mutex_unlock(&mdsc->mutex);
3186 mutex_lock(&s->s_mutex);
3187 mutex_lock(&mdsc->mutex);
3188 ceph_con_close(&s->s_con);
3189 mutex_unlock(&s->s_mutex);
3190 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3191 }
2f2dc053
SW		 3192 } else if (oldstate == newstate) {
3193 continue; /* nothing new with this mds */
3194 }
3195
3196 /*
3197 * send reconnect?
3198 */
3199 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3200 newstate >= CEPH_MDS_STATE_RECONNECT) {
3201 mutex_unlock(&mdsc->mutex);
3202 send_mds_reconnect(mdsc, s);
3203 mutex_lock(&mdsc->mutex);
3204 }
2f2dc053
SW		 3205
3206 /*
29790f26 3207 * kick request on any mds that has gone active.
2f2dc053
SW		 3208 */
3209 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3210 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3211 if (oldstate != CEPH_MDS_STATE_CREATING &&
3212 oldstate != CEPH_MDS_STATE_STARTING)
3213 pr_info("mds%d recovery completed\n", s->s_mds);
3214 kick_requests(mdsc, i);
2f2dc053 3215 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3216 wake_up_session_caps(s, 1);
2f2dc053
SW		 3217 }
3218 }
cb170a22 3219
76201b63 3220 for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
cb170a22
SW		 3221 s = mdsc->sessions[i];
3222 if (!s)
3223 continue;
3224 if (!ceph_mdsmap_is_laggy(newmap, i))
3225 continue;
3226 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3227 s->s_state == CEPH_MDS_SESSION_HUNG ||
3228 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3229 dout(" connecting to export targets of laggy mds%d\n",
3230 i);
3231 __open_export_target_sessions(mdsc, s);
3232 }
3233 }
2f2dc053
SW		 3234}
3235
3236
3237
3238/*
3239 * leases
3240 */
3241
3242/*
3243 * caller must hold session s_mutex, dentry->d_lock
3244 */
3245void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3246{
3247 struct ceph_dentry_info *di = ceph_dentry(dentry);
3248
3249 ceph_put_mds_session(di->lease_session);
3250 di->lease_session = NULL;
3251}
3252
2600d2dd
SW		 3253static void handle_lease(struct ceph_mds_client *mdsc,
3254 struct ceph_mds_session *session,
3255 struct ceph_msg *msg)
2f2dc053 3256{
3d14c5d2 3257 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3258 struct inode *inode;
2f2dc053
SW		 3259 struct dentry *parent, *dentry;
3260 struct ceph_dentry_info *di;
2600d2dd 3261 int mds = session->s_mds;
2f2dc053 3262 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3263 u32 seq;
2f2dc053 3264 struct ceph_vino vino;
2f2dc053
SW		 3265 struct qstr dname;
3266 int release = 0;
3267
2f2dc053
SW		 3268 dout("handle_lease from mds%d\n", mds);
3269
3270 /* decode */
3271 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3272 goto bad;
3273 vino.ino = le64_to_cpu(h->ino);
3274 vino.snap = CEPH_NOSNAP;
1e5ea23d 3275 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3276 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3277 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3278 if (dname.len != get_unaligned_le32(h+1))
3279 goto bad;
3280
2f2dc053
SW		 3281 /* lookup inode */
3282 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3283 dout("handle_lease %s, ino %llx %p %.*s\n",
3284 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3285 dname.len, dname.name);
6cd3bcad
YZ		 3286
3287 mutex_lock(&session->s_mutex);
3288 session->s_seq++;
3289
2f2dc053
SW		 3290 if (inode == NULL) {
3291 dout("handle_lease no inode %llx\n", vino.ino);
3292 goto release;
3293 }
2f2dc053
SW		 3294
3295 /* dentry */
3296 parent = d_find_alias(inode);
3297 if (!parent) {
3298 dout("no parent dentry on inode %p\n", inode);
3299 WARN_ON(1);
3300 goto release; /* hrm... */
3301 }
8387ff25 3302 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3303 dentry = d_lookup(parent, &dname);
3304 dput(parent);
3305 if (!dentry)
3306 goto release;
3307
3308 spin_lock(&dentry->d_lock);
3309 di = ceph_dentry(dentry);
3310 switch (h->action) {
3311 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3312 if (di->lease_session == session) {
1e5ea23d
SW		 3313 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3314 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3315 __ceph_mdsc_drop_dentry_lease(dentry);
3316 }
3317 release = 1;
3318 break;
3319
3320 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3321 if (di->lease_session == session &&
2f2dc053
SW		 3322 di->lease_gen == session->s_cap_gen &&
3323 di->lease_renew_from &&
3324 di->lease_renew_after == 0) {
3325 unsigned long duration =
3563dbdd 3326 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3327
1e5ea23d 3328 di->lease_seq = seq;
9b16f03c 3329 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3330 di->lease_renew_after = di->lease_renew_from +
3331 (duration >> 1);
3332 di->lease_renew_from = 0;
3333 }
3334 break;
3335 }
3336 spin_unlock(&dentry->d_lock);
3337 dput(dentry);
3338
3339 if (!release)
3340 goto out;
3341
3342release:
3343 /* let's just reuse the same message */
3344 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3345 ceph_msg_get(msg);
3346 ceph_con_send(&session->s_con, msg);
3347
3348out:
3349 iput(inode);
3350 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3351 return;
3352
3353bad:
3354 pr_err("corrupt lease message\n");
9ec7cab1 3355 ceph_msg_dump(msg);
2f2dc053
SW		 3356}
3357
3358void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3359 struct inode *inode,
3360 struct dentry *dentry, char action,
3361 u32 seq)
3362{
3363 struct ceph_msg *msg;
3364 struct ceph_mds_lease *lease;
3365 int len = sizeof(*lease) + sizeof(u32);
3366 int dnamelen = 0;
3367
3368 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3369 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3370 dnamelen = dentry->d_name.len;
3371 len += dnamelen;
3372
b61c2763 3373 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3374 if (!msg)
2f2dc053
SW		 3375 return;
3376 lease = msg->front.iov_base;
3377 lease->action = action;
2f2dc053
SW		 3378 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3379 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3380 lease->seq = cpu_to_le32(seq);
3381 put_unaligned_le32(dnamelen, lease + 1);
3382 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3383
3384 /*
3385 * if this is a preemptive lease RELEASE, no need to
3386 * flush request stream, since the actual request will
3387 * soon follow.
3388 */
3389 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3390
3391 ceph_con_send(&session->s_con, msg);
3392}
3393
2f2dc053
SW		 3394/*
3395 * drop all leases (and dentry refs) in preparation for umount
3396 */
3397static void drop_leases(struct ceph_mds_client *mdsc)
3398{
3399 int i;
3400
3401 dout("drop_leases\n");
3402 mutex_lock(&mdsc->mutex);
3403 for (i = 0; i < mdsc->max_sessions; i++) {
3404 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3405 if (!s)
3406 continue;
3407 mutex_unlock(&mdsc->mutex);
3408 mutex_lock(&s->s_mutex);
3409 mutex_unlock(&s->s_mutex);
3410 ceph_put_mds_session(s);
3411 mutex_lock(&mdsc->mutex);
3412 }
3413 mutex_unlock(&mdsc->mutex);
3414}
3415
3416
3417
3418/*
3419 * delayed work -- periodically trim expired leases, renew caps with mds
3420 */
3421static void schedule_delayed(struct ceph_mds_client *mdsc)
3422{
3423 int delay = 5;
3424 unsigned hz = round_jiffies_relative(HZ * delay);
3425 schedule_delayed_work(&mdsc->delayed_work, hz);
3426}
3427
3428static void delayed_work(struct work_struct *work)
3429{
3430 int i;
3431 struct ceph_mds_client *mdsc =
3432 container_of(work, struct ceph_mds_client, delayed_work.work);
3433 int renew_interval;
3434 int renew_caps;
3435
3436 dout("mdsc delayed_work\n");
afcdaea3 3437 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3438
3439 mutex_lock(&mdsc->mutex);
3440 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3441 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3442 mdsc->last_renew_caps);
3443 if (renew_caps)
3444 mdsc->last_renew_caps = jiffies;
3445
3446 for (i = 0; i < mdsc->max_sessions; i++) {
3447 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3448 if (s == NULL)
3449 continue;
3450 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3451 dout("resending session close request for mds%d\n",
3452 s->s_mds);
3453 request_close_session(mdsc, s);
3454 ceph_put_mds_session(s);
3455 continue;
3456 }
3457 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3458 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3459 s->s_state = CEPH_MDS_SESSION_HUNG;
3460 pr_info("mds%d hung\n", s->s_mds);
3461 }
3462 }
3463 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3464 /* this mds is failed or recovering, just wait */
3465 ceph_put_mds_session(s);
3466 continue;
3467 }
3468 mutex_unlock(&mdsc->mutex);
3469
3470 mutex_lock(&s->s_mutex);
3471 if (renew_caps)
3472 send_renew_caps(mdsc, s);
3473 else
3474 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3475 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3476 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3477 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3478 mutex_unlock(&s->s_mutex);
3479 ceph_put_mds_session(s);
3480
3481 mutex_lock(&mdsc->mutex);
3482 }
3483 mutex_unlock(&mdsc->mutex);
3484
3485 schedule_delayed(mdsc);
3486}
3487
3d14c5d2 3488int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3489
2f2dc053 3490{
3d14c5d2
YS		 3491 struct ceph_mds_client *mdsc;
3492
3493 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3494 if (!mdsc)
3495 return -ENOMEM;
3496 mdsc->fsc = fsc;
3497 fsc->mdsc = mdsc;
2f2dc053
SW		 3498 mutex_init(&mdsc->mutex);
3499 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3500 if (mdsc->mdsmap == NULL) {
3501 kfree(mdsc);
2d06eeb8 3502 return -ENOMEM;
fb3101b6 3503 }
2d06eeb8 3504
2f2dc053 3505 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3506 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3507 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3508 mdsc->sessions = NULL;
86d8f67b 3509 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3510 mdsc->max_sessions = 0;
3511 mdsc->stopping = 0;
affbc19a 3512 mdsc->last_snap_seq = 0;
2f2dc053 3513 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3514 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3515 INIT_LIST_HEAD(&mdsc->snap_empty);
3516 spin_lock_init(&mdsc->snap_empty_lock);
3517 mdsc->last_tid = 0;
e8a7b8b1 3518 mdsc->oldest_tid = 0;
44ca18f2 3519 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3520 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3521 mdsc->last_renew_caps = jiffies;
3522 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3523 spin_lock_init(&mdsc->cap_delay_lock);
3524 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3525 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3526 mdsc->last_cap_flush_tid = 1;
e4500b5e 3527 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3528 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3529 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3530 mdsc->num_cap_flushing = 0;
3531 spin_lock_init(&mdsc->cap_dirty_lock);
3532 init_waitqueue_head(&mdsc->cap_flushing_wq);
3533 spin_lock_init(&mdsc->dentry_lru_lock);
3534 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3535
37151668 3536 ceph_caps_init(mdsc);
3d14c5d2 3537 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3538
10183a69
YZ		 3539 init_rwsem(&mdsc->pool_perm_rwsem);
3540 mdsc->pool_perm_tree = RB_ROOT;
3541
5f44f142 3542 return 0;
2f2dc053
SW		 3543}
3544
3545/*
3546 * Wait for safe replies on open mds requests. If we time out, drop
3547 * all requests from the tree to avoid dangling dentry refs.
3548 */
3549static void wait_requests(struct ceph_mds_client *mdsc)
3550{
a319bf56 3551 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3552 struct ceph_mds_request *req;
2f2dc053
SW		 3553
3554 mutex_lock(&mdsc->mutex);
44ca18f2 3555 if (__get_oldest_req(mdsc)) {
2f2dc053 3556 mutex_unlock(&mdsc->mutex);
44ca18f2 3557
2f2dc053
SW		 3558 dout("wait_requests waiting for requests\n");
3559 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3560 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3561
3562 /* tear down remaining requests */
44ca18f2
SW		 3563 mutex_lock(&mdsc->mutex);
3564 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3565 dout("wait_requests timed out on tid %llu\n",
3566 req->r_tid);
44ca18f2 3567 __unregister_request(mdsc, req);
2f2dc053
SW		 3568 }
3569 }
3570 mutex_unlock(&mdsc->mutex);
3571 dout("wait_requests done\n");
3572}
3573
3574/*
3575 * called before mount is ro, and before dentries are torn down.
3576 * (hmm, does this still race with new lookups?)
3577 */
3578void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3579{
3580 dout("pre_umount\n");
3581 mdsc->stopping = 1;
3582
3583 drop_leases(mdsc);
afcdaea3 3584 ceph_flush_dirty_caps(mdsc);
2f2dc053 3585 wait_requests(mdsc);
17c688c3
SW		 3586
3587 /*
3588 * wait for reply handlers to drop their request refs and
3589 * their inode/dcache refs
3590 */
3591 ceph_msgr_flush();
2f2dc053
SW		 3592}
3593
3594/*
3595 * wait for all write mds requests to flush.
3596 */
3597static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3598{
80fc7314 3599 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3600 struct rb_node *n;
2f2dc053
SW		 3601
3602 mutex_lock(&mdsc->mutex);
3603 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3604restart:
44ca18f2
SW		 3605 req = __get_oldest_req(mdsc);
3606 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3607 /* find next request */
3608 n = rb_next(&req->r_node);
3609 if (n)
3610 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3611 else
3612 nextreq = NULL;
e8a7b8b1
YZ		 3613 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3614 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3615 /* write op */
3616 ceph_mdsc_get_request(req);
80fc7314
SW		 3617 if (nextreq)
3618 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3619 mutex_unlock(&mdsc->mutex);
3620 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3621 req->r_tid, want_tid);
3622 wait_for_completion(&req->r_safe_completion);
3623 mutex_lock(&mdsc->mutex);
44ca18f2 3624 ceph_mdsc_put_request(req);
80fc7314
SW		 3625 if (!nextreq)
3626 break; /* next dne before, so we're done! */
3627 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3628 /* next request was removed from tree */
3629 ceph_mdsc_put_request(nextreq);
3630 goto restart;
3631 }
3632 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3633 }
80fc7314 3634 req = nextreq;
2f2dc053
SW		 3635 }
3636 mutex_unlock(&mdsc->mutex);
3637 dout("wait_unsafe_requests done\n");
3638}
3639
3640void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3641{
0e294387 3642 u64 want_tid, want_flush;
2f2dc053 3643
52953d55 3644 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3645 return;
3646
2f2dc053
SW		 3647 dout("sync\n");
3648 mutex_lock(&mdsc->mutex);
3649 want_tid = mdsc->last_tid;
2f2dc053 3650 mutex_unlock(&mdsc->mutex);
2f2dc053 3651
afcdaea3 3652 ceph_flush_dirty_caps(mdsc);
d3383a8e 3653 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3654 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3655 if (!list_empty(&mdsc->cap_flush_list)) {
3656 struct ceph_cap_flush *cf =
3657 list_last_entry(&mdsc->cap_flush_list,
3658 struct ceph_cap_flush, g_list);
3659 cf->wake = true;
3660 }
d3383a8e
YZ		 3661 spin_unlock(&mdsc->cap_dirty_lock);
3662
0e294387
YZ		 3663 dout("sync want tid %lld flush_seq %lld\n",
3664 want_tid, want_flush);
2f2dc053
SW		 3665
3666 wait_unsafe_requests(mdsc, want_tid);
0e294387 3667 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3668}
3669
f3c60c59
SW		 3670/*
3671 * true if all sessions are closed, or we force unmount
3672 */
fcff415c 3673static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3674{
52953d55 3675 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3676 return true;
fcff415c 3677 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3678}
2f2dc053
SW		 3679
3680/*
3681 * called after sb is ro.
3682 */
3683void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3684{
a319bf56 3685 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3686 struct ceph_mds_session *session;
3687 int i;
fcff415c 3688 int skipped = 0;
2f2dc053
SW		 3689
3690 dout("close_sessions\n");
3691
2f2dc053 3692 /* close sessions */
f3c60c59
SW		 3693 mutex_lock(&mdsc->mutex);
3694 for (i = 0; i < mdsc->max_sessions; i++) {
3695 session = __ceph_lookup_mds_session(mdsc, i);
3696 if (!session)
3697 continue;
2f2dc053 3698 mutex_unlock(&mdsc->mutex);
f3c60c59 3699 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3700 if (__close_session(mdsc, session) <= 0)
3701 skipped++;
f3c60c59
SW		 3702 mutex_unlock(&session->s_mutex);
3703 ceph_put_mds_session(session);
2f2dc053
SW		 3704 mutex_lock(&mdsc->mutex);
3705 }
f3c60c59
SW		 3706 mutex_unlock(&mdsc->mutex);
3707
3708 dout("waiting for sessions to close\n");
fcff415c
YZ		 3709 wait_event_timeout(mdsc->session_close_wq,
3710 done_closing_sessions(mdsc, skipped),
a319bf56 3711 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3712
3713 /* tear down remaining sessions */
f3c60c59 3714 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3715 for (i = 0; i < mdsc->max_sessions; i++) {
3716 if (mdsc->sessions[i]) {
3717 session = get_session(mdsc->sessions[i]);
2600d2dd 3718 __unregister_session(mdsc, session);
2f2dc053
SW		 3719 mutex_unlock(&mdsc->mutex);
3720 mutex_lock(&session->s_mutex);
3721 remove_session_caps(session);
3722 mutex_unlock(&session->s_mutex);
3723 ceph_put_mds_session(session);
3724 mutex_lock(&mdsc->mutex);
3725 }
3726 }
2f2dc053 3727 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3728 mutex_unlock(&mdsc->mutex);
3729
3730 ceph_cleanup_empty_realms(mdsc);
3731
3732 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3733
3734 dout("stopped\n");
3735}
3736
48fec5d0
YZ		 3737void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3738{
3739 struct ceph_mds_session *session;
3740 int mds;
3741
3742 dout("force umount\n");
3743
3744 mutex_lock(&mdsc->mutex);
3745 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3746 session = __ceph_lookup_mds_session(mdsc, mds);
3747 if (!session)
3748 continue;
3749 mutex_unlock(&mdsc->mutex);
3750 mutex_lock(&session->s_mutex);
3751 __close_session(mdsc, session);
3752 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3753 cleanup_session_requests(mdsc, session);
3754 remove_session_caps(session);
3755 }
3756 mutex_unlock(&session->s_mutex);
3757 ceph_put_mds_session(session);
3758 mutex_lock(&mdsc->mutex);
3759 kick_requests(mdsc, mds);
3760 }
3761 __wake_requests(mdsc, &mdsc->waiting_for_map);
3762 mutex_unlock(&mdsc->mutex);
3763}
3764
3d14c5d2 3765static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3766{
3767 dout("stop\n");
3768 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3769 if (mdsc->mdsmap)
3770 ceph_mdsmap_destroy(mdsc->mdsmap);
3771 kfree(mdsc->sessions);
37151668 3772 ceph_caps_finalize(mdsc);
10183a69 3773 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3774}
3775
3d14c5d2
YS		 3776void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3777{
3778 struct ceph_mds_client *mdsc = fsc->mdsc;
ef550f6f 3779 dout("mdsc_destroy %p\n", mdsc);
ef550f6f
SW		 3780
3781 /* flush out any connection work with references to us */
3782 ceph_msgr_flush();
3783
62a65f36
YZ		 3784 ceph_mdsc_stop(mdsc);
3785
3d14c5d2
YS		 3786 fsc->mdsc = NULL;
3787 kfree(mdsc);
ef550f6f 3788 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3789}
3790
430afbad
YZ		 3791void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3792{
3793 struct ceph_fs_client *fsc = mdsc->fsc;
3794 const char *mds_namespace = fsc->mount_options->mds_namespace;
3795 void *p = msg->front.iov_base;
3796 void *end = p + msg->front.iov_len;
3797 u32 epoch;
3798 u32 map_len;
3799 u32 num_fs;
3800 u32 mount_fscid = (u32)-1;
3801 u8 struct_v, struct_cv;
3802 int err = -EINVAL;
3803
3804 ceph_decode_need(&p, end, sizeof(u32), bad);
3805 epoch = ceph_decode_32(&p);
3806
3807 dout("handle_fsmap epoch %u\n", epoch);
3808
3809 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3810 struct_v = ceph_decode_8(&p);
3811 struct_cv = ceph_decode_8(&p);
3812 map_len = ceph_decode_32(&p);
3813
3814 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3815 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3816
3817 num_fs = ceph_decode_32(&p);
3818 while (num_fs-- > 0) {
3819 void *info_p, *info_end;
3820 u32 info_len;
3821 u8 info_v, info_cv;
3822 u32 fscid, namelen;
3823
3824 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3825 info_v = ceph_decode_8(&p);
3826 info_cv = ceph_decode_8(&p);
3827 info_len = ceph_decode_32(&p);
3828 ceph_decode_need(&p, end, info_len, bad);
3829 info_p = p;
3830 info_end = p + info_len;
3831 p = info_end;
3832
3833 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3834 fscid = ceph_decode_32(&info_p);
3835 namelen = ceph_decode_32(&info_p);
3836 ceph_decode_need(&info_p, info_end, namelen, bad);
3837
3838 if (mds_namespace &&
3839 strlen(mds_namespace) == namelen &&
3840 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3841 mount_fscid = fscid;
3842 break;
3843 }
3844 }
3845
3846 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3847 if (mount_fscid != (u32)-1) {
3848 fsc->client->monc.fs_cluster_id = mount_fscid;
3849 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
3850 0, true);
3851 ceph_monc_renew_subs(&fsc->client->monc);
3852 } else {
3853 err = -ENOENT;
3854 goto err_out;
3855 }
3856 return;
3857bad:
3858 pr_err("error decoding fsmap\n");
3859err_out:
3860 mutex_lock(&mdsc->mutex);
3861 mdsc->mdsmap_err = -ENOENT;
3862 __wake_requests(mdsc, &mdsc->waiting_for_map);
3863 mutex_unlock(&mdsc->mutex);
3864 return;
3865}
2f2dc053
SW		 3866
3867/*
3868 * handle mds map update.
3869 */
430afbad 3870void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053
SW		 3871{
3872 u32 epoch;
3873 u32 maplen;
3874 void *p = msg->front.iov_base;
3875 void *end = p + msg->front.iov_len;
3876 struct ceph_mdsmap *newmap, *oldmap;
3877 struct ceph_fsid fsid;
3878 int err = -EINVAL;
3879
3880 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3881 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3882 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3883 return;
c89136ea
SW		 3884 epoch = ceph_decode_32(&p);
3885 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3886 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3887
3888 /* do we need it? */
2f2dc053
SW		 3889 mutex_lock(&mdsc->mutex);
3890 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3891 dout("handle_map epoch %u <= our %u\n",
3892 epoch, mdsc->mdsmap->m_epoch);
3893 mutex_unlock(&mdsc->mutex);
3894 return;
3895 }
3896
3897 newmap = ceph_mdsmap_decode(&p, end);
3898 if (IS_ERR(newmap)) {
3899 err = PTR_ERR(newmap);
3900 goto bad_unlock;
3901 }
3902
3903 /* swap into place */
3904 if (mdsc->mdsmap) {
3905 oldmap = mdsc->mdsmap;
3906 mdsc->mdsmap = newmap;
3907 check_new_map(mdsc, newmap, oldmap);
3908 ceph_mdsmap_destroy(oldmap);
3909 } else {
3910 mdsc->mdsmap = newmap; /* first mds map */
3911 }
3d14c5d2 3912 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3913
3914 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 3915 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3916 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 3917
3918 mutex_unlock(&mdsc->mutex);
3919 schedule_delayed(mdsc);
3920 return;
3921
3922bad_unlock:
3923 mutex_unlock(&mdsc->mutex);
3924bad:
3925 pr_err("error decoding mdsmap %d\n", err);
3926 return;
3927}
3928
3929static struct ceph_connection *con_get(struct ceph_connection *con)
3930{
3931 struct ceph_mds_session *s = con->private;
3932
3933 if (get_session(s)) {
3997c01d 3934 dout("mdsc con_get %p ok (%d)\n", s, refcount_read(&s->s_ref));
2f2dc053
SW		 3935 return con;
3936 }
3937 dout("mdsc con_get %p FAIL\n", s);
3938 return NULL;
3939}
3940
3941static void con_put(struct ceph_connection *con)
3942{
3943 struct ceph_mds_session *s = con->private;
3944
3997c01d 3945 dout("mdsc con_put %p (%d)\n", s, refcount_read(&s->s_ref) - 1);
2f2dc053
SW		 3946 ceph_put_mds_session(s);
3947}
3948
3949/*
3950 * if the client is unresponsive for long enough, the mds will kill
3951 * the session entirely.
3952 */
3953static void peer_reset(struct ceph_connection *con)
3954{
3955 struct ceph_mds_session *s = con->private;
7e70f0ed 3956 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3957
f3ae1b97 3958 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3959 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3960}
3961
3962static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3963{
3964 struct ceph_mds_session *s = con->private;
3965 struct ceph_mds_client *mdsc = s->s_mdsc;
3966 int type = le16_to_cpu(msg->hdr.type);
3967
2600d2dd
SW		 3968 mutex_lock(&mdsc->mutex);
3969 if (__verify_registered_session(mdsc, s) < 0) {
3970 mutex_unlock(&mdsc->mutex);
3971 goto out;
3972 }
3973 mutex_unlock(&mdsc->mutex);
3974
2f2dc053
SW		 3975 switch (type) {
3976 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 3977 ceph_mdsc_handle_mdsmap(mdsc, msg);
3978 break;
3979 case CEPH_MSG_FS_MAP_USER:
3980 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 3981 break;
3982 case CEPH_MSG_CLIENT_SESSION:
3983 handle_session(s, msg);
3984 break;
3985 case CEPH_MSG_CLIENT_REPLY:
3986 handle_reply(s, msg);
3987 break;
3988 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3989 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3990 break;
3991 case CEPH_MSG_CLIENT_CAPS:
3992 ceph_handle_caps(s, msg);
3993 break;
3994 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3995 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3996 break;
3997 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3998 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3999 break;
4000
4001 default:
4002 pr_err("received unknown message type %d %s\n", type,
4003 ceph_msg_type_name(type));
4004 }
2600d2dd 4005out:
2f2dc053
SW		 4006 ceph_msg_put(msg);
4007}
4008
4e7a5dcd
SW		 4009/*
4010 * authentication
4011 */
a3530df3
AE		 4012
4013/*
4014 * Note: returned pointer is the address of a structure that's
4015 * managed separately. Caller must *not* attempt to free it.
4016 */
4017static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 4018 int *proto, int force_new)
4e7a5dcd
SW		 4019{
4020 struct ceph_mds_session *s = con->private;
4021 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4022 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 4023 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 4024
74f1869f 4025 if (force_new && auth->authorizer) {
6c1ea260 4026 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 4027 auth->authorizer = NULL;
4e7a5dcd 4028 }
27859f97
SW		 4029 if (!auth->authorizer) {
4030 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4031 auth);
0bed9b5c
SW		 4032 if (ret)
4033 return ERR_PTR(ret);
27859f97
SW		 4034 } else {
4035 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4036 auth);
a255651d 4037 if (ret)
a3530df3 4038 return ERR_PTR(ret);
4e7a5dcd 4039 }
4e7a5dcd 4040 *proto = ac->protocol;
74f1869f 4041
a3530df3 4042 return auth;
4e7a5dcd
SW		 4043}
4044
4045
0dde5848 4046static int verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 4047{
4048 struct ceph_mds_session *s = con->private;
4049 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4050 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 4051
0dde5848 4052 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4e7a5dcd
SW		 4053}
4054
9bd2e6f8
SW		 4055static int invalidate_authorizer(struct ceph_connection *con)
4056{
4057 struct ceph_mds_session *s = con->private;
4058 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4059 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 4060
27859f97 4061 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 4062
3d14c5d2 4063 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 4064}
4065
53ded495
AE		 4066static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4067 struct ceph_msg_header *hdr, int *skip)
4068{
4069 struct ceph_msg *msg;
4070 int type = (int) le16_to_cpu(hdr->type);
4071 int front_len = (int) le32_to_cpu(hdr->front_len);
4072
4073 if (con->in_msg)
4074 return con->in_msg;
4075
4076 *skip = 0;
4077 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4078 if (!msg) {
4079 pr_err("unable to allocate msg type %d len %d\n",
4080 type, front_len);
4081 return NULL;
4082 }
53ded495
AE		 4083
4084 return msg;
4085}
4086
79dbd1ba 4087static int mds_sign_message(struct ceph_msg *msg)
33d07337 4088{
79dbd1ba 4089 struct ceph_mds_session *s = msg->con->private;
33d07337 4090 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4091
33d07337
YZ		 4092 return ceph_auth_sign_message(auth, msg);
4093}
4094
79dbd1ba 4095static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 4096{
79dbd1ba 4097 struct ceph_mds_session *s = msg->con->private;
33d07337 4098 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4099
33d07337
YZ		 4100 return ceph_auth_check_message_signature(auth, msg);
4101}
4102
9e32789f 4103static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4104 .get = con_get,
4105 .put = con_put,
4106 .dispatch = dispatch,
4e7a5dcd
SW		 4107 .get_authorizer = get_authorizer,
4108 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4109 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4110 .peer_reset = peer_reset,
53ded495 4111 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4112 .sign_message = mds_sign_message,
4113 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4114};
4115
2f2dc053 4116/* eof */
Ubuntu Eoan Linux kernel mirror