]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ceph/mds_client.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: flush pending works before shutdown super
[mirror_ubuntu-bionic-kernel.git] / fs / ceph / mds_client.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
2f2dc053 3
496e5955 4#include <linux/fs.h>
2f2dc053 5#include <linux/wait.h>
5a0e3ad6 6#include <linux/slab.h>
54008399 7#include <linux/gfp.h>
2f2dc053 8#include <linux/sched.h>
3d14c5d2
YS		 9#include <linux/debugfs.h>
10#include <linux/seq_file.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45 * are no longer valid.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
d37b1d99 411 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053
SW		 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);
d37b1d99 486 if (!sa)
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 {
38f340cc
YZ		 738 /* req->r_dentry is non-null for LSSNAP request */
739 rcu_read_lock();
740 inode = get_nonsnap_parent(req->r_dentry);
741 rcu_read_unlock();
742 dout("__choose_mds using snapdir's parent %p\n", inode);
5d37ca14 743 }
38f340cc 744 } else if (req->r_dentry) {
d79698da 745 /* ignore race with rename; old or new d_parent is okay */
30c71233
JL		 746 struct dentry *parent;
747 struct inode *dir;
748
749 rcu_read_lock();
750 parent = req->r_dentry->d_parent;
3dd69aab 751 dir = req->r_parent ? : d_inode_rcu(parent);
eb6bb1c5 752
30c71233
JL		 753 if (!dir || dir->i_sb != mdsc->fsc->sb) {
754 /* not this fs or parent went negative */
2b0143b5 755 inode = d_inode(req->r_dentry);
30c71233
JL		 756 if (inode)
757 ihold(inode);
eb6bb1c5
SW		 758 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
759 /* direct snapped/virtual snapdir requests
760 * based on parent dir inode */
30c71233 761 inode = get_nonsnap_parent(parent);
eb6bb1c5 762 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 763 } else {
eb6bb1c5 764 /* dentry target */
2b0143b5 765 inode = d_inode(req->r_dentry);
ca18bede
YZ		 766 if (!inode || mode == USE_AUTH_MDS) {
767 /* dir + name */
30c71233 768 inode = igrab(dir);
ca18bede
YZ		 769 hash = ceph_dentry_hash(dir, req->r_dentry);
770 is_hash = true;
30c71233
JL		 771 } else {
772 ihold(inode);
ca18bede 773 }
2f2dc053 774 }
30c71233 775 rcu_read_unlock();
2f2dc053 776 }
eb6bb1c5 777
2f2dc053
SW		 778 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
779 (int)hash, mode);
780 if (!inode)
781 goto random;
782 ci = ceph_inode(inode);
783
784 if (is_hash && S_ISDIR(inode->i_mode)) {
785 struct ceph_inode_frag frag;
786 int found;
787
788 ceph_choose_frag(ci, hash, &frag, &found);
789 if (found) {
790 if (mode == USE_ANY_MDS && frag.ndist > 0) {
791 u8 r;
792
793 /* choose a random replica */
794 get_random_bytes(&r, 1);
795 r %= frag.ndist;
796 mds = frag.dist[r];
797 dout("choose_mds %p %llx.%llx "
798 "frag %u mds%d (%d/%d)\n",
799 inode, ceph_vinop(inode),
d66bbd44 800 frag.frag, mds,
2f2dc053 801 (int)r, frag.ndist);
d66bbd44
SW		 802 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
803 CEPH_MDS_STATE_ACTIVE)
30c71233 804 goto out;
2f2dc053
SW		 805 }
806
807 /* since this file/dir wasn't known to be
808 * replicated, then we want to look for the
809 * authoritative mds. */
810 mode = USE_AUTH_MDS;
811 if (frag.mds >= 0) {
812 /* choose auth mds */
813 mds = frag.mds;
814 dout("choose_mds %p %llx.%llx "
815 "frag %u mds%d (auth)\n",
816 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 817 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
818 CEPH_MDS_STATE_ACTIVE)
30c71233 819 goto out;
2f2dc053
SW		 820 }
821 }
822 }
823
be655596 824 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 825 cap = NULL;
826 if (mode == USE_AUTH_MDS)
827 cap = ci->i_auth_cap;
828 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
829 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
830 if (!cap) {
be655596 831 spin_unlock(&ci->i_ceph_lock);
30c71233 832 iput(inode);
2f2dc053
SW		 833 goto random;
834 }
835 mds = cap->session->s_mds;
836 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
837 inode, ceph_vinop(inode), mds,
838 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 839 spin_unlock(&ci->i_ceph_lock);
30c71233
JL		 840out:
841 iput(inode);
2f2dc053
SW		 842 return mds;
843
844random:
845 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
846 dout("choose_mds chose random mds%d\n", mds);
847 return mds;
848}
849
850
851/*
852 * session messages
853 */
854static struct ceph_msg *create_session_msg(u32 op, u64 seq)
855{
856 struct ceph_msg *msg;
857 struct ceph_mds_session_head *h;
858
b61c2763
SW		 859 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
860 false);
a79832f2 861 if (!msg) {
2f2dc053 862 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 863 return NULL;
2f2dc053
SW		 864 }
865 h = msg->front.iov_base;
866 h->op = cpu_to_le32(op);
867 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 868
869 return msg;
870}
871
872/*
873 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
874 * to include additional client metadata fields.
875 */
876static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
877{
878 struct ceph_msg *msg;
879 struct ceph_mds_session_head *h;
880 int i = -1;
881 int metadata_bytes = 0;
882 int metadata_key_count = 0;
883 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 884 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
dbd0c8bf
JS		 885 void *p;
886
a6a5ce4f 887 const char* metadata[][2] = {
717e6f28
YZ		 888 {"hostname", mdsc->nodename},
889 {"kernel_version", init_utsname()->release},
3f384954
YZ		 890 {"entity_id", opt->name ? : ""},
891 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 892 {NULL, NULL}
893 };
894
895 /* Calculate serialized length of metadata */
896 metadata_bytes = 4; /* map length */
d37b1d99 897 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 898 metadata_bytes += 8 + strlen(metadata[i][0]) +
899 strlen(metadata[i][1]);
900 metadata_key_count++;
901 }
902
903 /* Allocate the message */
904 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
905 GFP_NOFS, false);
906 if (!msg) {
907 pr_err("create_session_msg ENOMEM creating msg\n");
908 return NULL;
909 }
910 h = msg->front.iov_base;
911 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
912 h->seq = cpu_to_le64(seq);
913
914 /*
915 * Serialize client metadata into waiting buffer space, using
916 * the format that userspace expects for map<string, string>
7cfa0313
JS		 917 *
918 * ClientSession messages with metadata are v2
dbd0c8bf 919 */
7cfa0313
JS		 920 msg->hdr.version = cpu_to_le16(2);
921 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 922
923 /* The write pointer, following the session_head structure */
924 p = msg->front.iov_base + sizeof(*h);
925
926 /* Number of entries in the map */
927 ceph_encode_32(&p, metadata_key_count);
928
929 /* Two length-prefixed strings for each entry in the map */
d37b1d99 930 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 931 size_t const key_len = strlen(metadata[i][0]);
932 size_t const val_len = strlen(metadata[i][1]);
933
934 ceph_encode_32(&p, key_len);
935 memcpy(p, metadata[i][0], key_len);
936 p += key_len;
937 ceph_encode_32(&p, val_len);
938 memcpy(p, metadata[i][1], val_len);
939 p += val_len;
940 }
941
2f2dc053
SW		 942 return msg;
943}
944
945/*
946 * send session open request.
947 *
948 * called under mdsc->mutex
949 */
950static int __open_session(struct ceph_mds_client *mdsc,
951 struct ceph_mds_session *session)
952{
953 struct ceph_msg *msg;
954 int mstate;
955 int mds = session->s_mds;
2f2dc053
SW		 956
957 /* wait for mds to go active? */
958 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
959 dout("open_session to mds%d (%s)\n", mds,
960 ceph_mds_state_name(mstate));
961 session->s_state = CEPH_MDS_SESSION_OPENING;
962 session->s_renew_requested = jiffies;
963
964 /* send connect message */
dbd0c8bf 965 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 966 if (!msg)
967 return -ENOMEM;
2f2dc053 968 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 969 return 0;
970}
971
ed0552a1
SW		 972/*
973 * open sessions for any export targets for the given mds
974 *
975 * called under mdsc->mutex
976 */
5d72d13c
YZ		 977static struct ceph_mds_session *
978__open_export_target_session(struct ceph_mds_client *mdsc, int target)
979{
980 struct ceph_mds_session *session;
981
982 session = __ceph_lookup_mds_session(mdsc, target);
983 if (!session) {
984 session = register_session(mdsc, target);
985 if (IS_ERR(session))
986 return session;
987 }
988 if (session->s_state == CEPH_MDS_SESSION_NEW ||
989 session->s_state == CEPH_MDS_SESSION_CLOSING)
990 __open_session(mdsc, session);
991
992 return session;
993}
994
995struct ceph_mds_session *
996ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
997{
998 struct ceph_mds_session *session;
999
1000 dout("open_export_target_session to mds%d\n", target);
1001
1002 mutex_lock(&mdsc->mutex);
1003 session = __open_export_target_session(mdsc, target);
1004 mutex_unlock(&mdsc->mutex);
1005
1006 return session;
1007}
1008
ed0552a1
SW		 1009static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1010 struct ceph_mds_session *session)
1011{
1012 struct ceph_mds_info *mi;
1013 struct ceph_mds_session *ts;
1014 int i, mds = session->s_mds;
ed0552a1 1015
76201b63 1016 if (mds >= mdsc->mdsmap->m_num_mds)
ed0552a1 1017 return;
5d72d13c 1018
ed0552a1
SW		 1019 mi = &mdsc->mdsmap->m_info[mds];
1020 dout("open_export_target_sessions for mds%d (%d targets)\n",
1021 session->s_mds, mi->num_export_targets);
1022
1023 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1024 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1025 if (!IS_ERR(ts))
1026 ceph_put_mds_session(ts);
ed0552a1
SW		 1027 }
1028}
1029
154f42c2
SW		 1030void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1031 struct ceph_mds_session *session)
1032{
1033 mutex_lock(&mdsc->mutex);
1034 __open_export_target_sessions(mdsc, session);
1035 mutex_unlock(&mdsc->mutex);
1036}
1037
2f2dc053
SW		 1038/*
1039 * session caps
1040 */
1041
c8a96a31
JL		 1042static void detach_cap_releases(struct ceph_mds_session *session,
1043 struct list_head *target)
2f2dc053 1044{
c8a96a31
JL		 1045 lockdep_assert_held(&session->s_cap_lock);
1046
1047 list_splice_init(&session->s_cap_releases, target);
745a8e3b 1048 session->s_num_cap_releases = 0;
c8a96a31
JL		 1049 dout("dispose_cap_releases mds%d\n", session->s_mds);
1050}
2f2dc053 1051
c8a96a31
JL		 1052static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1053 struct list_head *dispose)
1054{
1055 while (!list_empty(dispose)) {
745a8e3b
YZ		 1056 struct ceph_cap *cap;
1057 /* zero out the in-progress message */
c8a96a31 1058 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
745a8e3b
YZ		 1059 list_del(&cap->session_caps);
1060 ceph_put_cap(mdsc, cap);
2f2dc053 1061 }
2f2dc053
SW		 1062}
1063
1c841a96
YZ		 1064static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1065 struct ceph_mds_session *session)
1066{
1067 struct ceph_mds_request *req;
1068 struct rb_node *p;
1069
1070 dout("cleanup_session_requests mds%d\n", session->s_mds);
1071 mutex_lock(&mdsc->mutex);
1072 while (!list_empty(&session->s_unsafe)) {
1073 req = list_first_entry(&session->s_unsafe,
1074 struct ceph_mds_request, r_unsafe_item);
3e0708b9
YZ		 1075 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1076 req->r_tid);
1c841a96
YZ		 1077 __unregister_request(mdsc, req);
1078 }
1079 /* zero r_attempts, so kick_requests() will re-send requests */
1080 p = rb_first(&mdsc->request_tree);
1081 while (p) {
1082 req = rb_entry(p, struct ceph_mds_request, r_node);
1083 p = rb_next(p);
1084 if (req->r_session &&
1085 req->r_session->s_mds == session->s_mds)
1086 req->r_attempts = 0;
1087 }
1088 mutex_unlock(&mdsc->mutex);
1089}
1090
2f2dc053 1091/*
f818a736
SW		 1092 * Helper to safely iterate over all caps associated with a session, with
1093 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1094 *
f818a736 1095 * Caller must hold session s_mutex.
2f2dc053
SW		 1096 */
1097static int iterate_session_caps(struct ceph_mds_session *session,
1098 int (*cb)(struct inode *, struct ceph_cap *,
1099 void *), void *arg)
1100{
7c1332b8
SW		 1101 struct list_head *p;
1102 struct ceph_cap *cap;
1103 struct inode *inode, *last_inode = NULL;
1104 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1105 int ret;
1106
1107 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1108 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1109 p = session->s_caps.next;
1110 while (p != &session->s_caps) {
1111 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1112 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1113 if (!inode) {
1114 p = p->next;
2f2dc053 1115 continue;
7c1332b8
SW		 1116 }
1117 session->s_cap_iterator = cap;
2f2dc053 1118 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1119
1120 if (last_inode) {
1121 iput(last_inode);
1122 last_inode = NULL;
1123 }
1124 if (old_cap) {
37151668 1125 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1126 old_cap = NULL;
1127 }
1128
2f2dc053 1129 ret = cb(inode, cap, arg);
7c1332b8
SW		 1130 last_inode = inode;
1131
2f2dc053 1132 spin_lock(&session->s_cap_lock);
7c1332b8 1133 p = p->next;
d37b1d99 1134 if (!cap->ci) {
7c1332b8
SW		 1135 dout("iterate_session_caps finishing cap %p removal\n",
1136 cap);
1137 BUG_ON(cap->session != session);
745a8e3b 1138 cap->session = NULL;
7c1332b8
SW		 1139 list_del_init(&cap->session_caps);
1140 session->s_nr_caps--;
745a8e3b
YZ		 1141 if (cap->queue_release) {
1142 list_add_tail(&cap->session_caps,
1143 &session->s_cap_releases);
1144 session->s_num_cap_releases++;
1145 } else {
1146 old_cap = cap; /* put_cap it w/o locks held */
1147 }
7c1332b8 1148 }
5dacf091
SW		 1149 if (ret < 0)
1150 goto out;
2f2dc053 1151 }
5dacf091
SW		 1152 ret = 0;
1153out:
7c1332b8 1154 session->s_cap_iterator = NULL;
2f2dc053 1155 spin_unlock(&session->s_cap_lock);
7c1332b8 1156
e96a650a 1157 iput(last_inode);
7c1332b8 1158 if (old_cap)
37151668 1159 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1160
5dacf091 1161 return ret;
2f2dc053
SW		 1162}
1163
1164static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1165 void *arg)
2f2dc053 1166{
6c93df5d 1167 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
2f2dc053 1168 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1169 LIST_HEAD(to_remove);
6c93df5d
YZ		 1170 bool drop = false;
1171 bool invalidate = false;
6c99f254 1172
2f2dc053
SW		 1173 dout("removing cap %p, ci is %p, inode is %p\n",
1174 cap, ci, &ci->vfs_inode);
be655596 1175 spin_lock(&ci->i_ceph_lock);
a096b09a 1176 __ceph_remove_cap(cap, false);
571ade33 1177 if (!ci->i_auth_cap) {
553adfd9 1178 struct ceph_cap_flush *cf;
6c93df5d 1179 struct ceph_mds_client *mdsc = fsc->mdsc;
6c99f254 1180
77310320
YZ		 1181 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1182
6c93df5d 1183 if (ci->i_wrbuffer_ref > 0 &&
52953d55 1184 READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
6c93df5d
YZ		 1185 invalidate = true;
1186
e4500b5e
YZ		 1187 while (!list_empty(&ci->i_cap_flush_list)) {
1188 cf = list_first_entry(&ci->i_cap_flush_list,
1189 struct ceph_cap_flush, i_list);
8cdcc07d 1190 list_move(&cf->i_list, &to_remove);
553adfd9
YZ		 1191 }
1192
6c99f254 1193 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1194
e4500b5e
YZ		 1195 list_for_each_entry(cf, &to_remove, i_list)
1196 list_del(&cf->g_list);
8310b089 1197
6c99f254 1198 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1199 pr_warn_ratelimited(
1200 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1201 ceph_cap_string(ci->i_dirty_caps),
1202 inode, ceph_ino(inode));
1203 ci->i_dirty_caps = 0;
1204 list_del_init(&ci->i_dirty_item);
6c93df5d 1205 drop = true;
6c99f254
SW		 1206 }
1207 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1208 pr_warn_ratelimited(
1209 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1210 ceph_cap_string(ci->i_flushing_caps),
1211 inode, ceph_ino(inode));
1212 ci->i_flushing_caps = 0;
1213 list_del_init(&ci->i_flushing_item);
1214 mdsc->num_cap_flushing--;
6c93df5d 1215 drop = true;
6c99f254 1216 }
6c99f254 1217 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1218
b3f8d68f
YZ		 1219 if (atomic_read(&ci->i_filelock_ref) > 0) {
1220 /* make further file lock syscall return -EIO */
1221 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1222 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1223 inode, ceph_ino(inode));
1224 }
1225
f66fd9f0 1226 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1227 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1228 ci->i_prealloc_cap_flush = NULL;
1229 }
6c99f254 1230 }
be655596 1231 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1232 while (!list_empty(&to_remove)) {
1233 struct ceph_cap_flush *cf;
1234 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1235 struct ceph_cap_flush, i_list);
1236 list_del(&cf->i_list);
f66fd9f0 1237 ceph_free_cap_flush(cf);
553adfd9 1238 }
77310320
YZ		 1239
1240 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1241 if (invalidate)
1242 ceph_queue_invalidate(inode);
77310320 1243 if (drop)
6c99f254 1244 iput(inode);
2f2dc053
SW		 1245 return 0;
1246}
1247
1248/*
1249 * caller must hold session s_mutex
1250 */
1251static void remove_session_caps(struct ceph_mds_session *session)
1252{
6c93df5d
YZ		 1253 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1254 struct super_block *sb = fsc->sb;
c8a96a31
JL		 1255 LIST_HEAD(dispose);
1256
2f2dc053 1257 dout("remove_session_caps on %p\n", session);
6c93df5d 1258 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1259
c8799fc4
YZ		 1260 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1261
6f60f889
YZ		 1262 spin_lock(&session->s_cap_lock);
1263 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1264 struct inode *inode;
1265 struct ceph_cap *cap, *prev = NULL;
1266 struct ceph_vino vino;
1267 /*
1268 * iterate_session_caps() skips inodes that are being
1269 * deleted, we need to wait until deletions are complete.
1270 * __wait_on_freeing_inode() is designed for the job,
1271 * but it is not exported, so use lookup inode function
1272 * to access it.
1273 */
1274 while (!list_empty(&session->s_caps)) {
1275 cap = list_entry(session->s_caps.next,
1276 struct ceph_cap, session_caps);
1277 if (cap == prev)
1278 break;
1279 prev = cap;
1280 vino = cap->ci->i_vino;
1281 spin_unlock(&session->s_cap_lock);
1282
ed284c49 1283 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1284 iput(inode);
1285
1286 spin_lock(&session->s_cap_lock);
1287 }
1288 }
745a8e3b
YZ		 1289
1290 // drop cap expires and unlock s_cap_lock
c8a96a31 1291 detach_cap_releases(session, &dispose);
6f60f889 1292
2f2dc053 1293 BUG_ON(session->s_nr_caps > 0);
6c99f254 1294 BUG_ON(!list_empty(&session->s_cap_flushing));
c8a96a31
JL		 1295 spin_unlock(&session->s_cap_lock);
1296 dispose_cap_releases(session->s_mdsc, &dispose);
2f2dc053
SW		 1297}
1298
1299/*
1300 * wake up any threads waiting on this session's caps. if the cap is
1301 * old (didn't get renewed on the client reconnect), remove it now.
1302 *
1303 * caller must hold s_mutex.
1304 */
1305static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1306 void *arg)
1307{
0dc2570f
SW		 1308 struct ceph_inode_info *ci = ceph_inode(inode);
1309
0dc2570f 1310 if (arg) {
be655596 1311 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1312 ci->i_wanted_max_size = 0;
1313 ci->i_requested_max_size = 0;
be655596 1314 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1315 }
e5360309 1316 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1317 return 0;
1318}
1319
0dc2570f
SW		 1320static void wake_up_session_caps(struct ceph_mds_session *session,
1321 int reconnect)
2f2dc053
SW		 1322{
1323 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1324 iterate_session_caps(session, wake_up_session_cb,
1325 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1326}
1327
1328/*
1329 * Send periodic message to MDS renewing all currently held caps. The
1330 * ack will reset the expiration for all caps from this session.
1331 *
1332 * caller holds s_mutex
1333 */
1334static int send_renew_caps(struct ceph_mds_client *mdsc,
1335 struct ceph_mds_session *session)
1336{
1337 struct ceph_msg *msg;
1338 int state;
1339
1340 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1341 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1342 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1343 session->s_renew_requested = jiffies;
2f2dc053
SW		 1344
1345 /* do not try to renew caps until a recovering mds has reconnected
1346 * with its clients. */
1347 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1348 if (state < CEPH_MDS_STATE_RECONNECT) {
1349 dout("send_renew_caps ignoring mds%d (%s)\n",
1350 session->s_mds, ceph_mds_state_name(state));
1351 return 0;
1352 }
1353
1354 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1355 ceph_mds_state_name(state));
2f2dc053
SW		 1356 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1357 ++session->s_renew_seq);
a79832f2
SW		 1358 if (!msg)
1359 return -ENOMEM;
2f2dc053
SW		 1360 ceph_con_send(&session->s_con, msg);
1361 return 0;
1362}
1363
186e4f7a
YZ		 1364static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1365 struct ceph_mds_session *session, u64 seq)
1366{
1367 struct ceph_msg *msg;
1368
1369 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1370 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1371 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1372 if (!msg)
1373 return -ENOMEM;
1374 ceph_con_send(&session->s_con, msg);
1375 return 0;
1376}
1377
1378
2f2dc053
SW		 1379/*
1380 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1381 *
1382 * Called under session->s_mutex
2f2dc053
SW		 1383 */
1384static void renewed_caps(struct ceph_mds_client *mdsc,
1385 struct ceph_mds_session *session, int is_renew)
1386{
1387 int was_stale;
1388 int wake = 0;
1389
1390 spin_lock(&session->s_cap_lock);
1ce208a6 1391 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1392
1393 session->s_cap_ttl = session->s_renew_requested +
1394 mdsc->mdsmap->m_session_timeout*HZ;
1395
1396 if (was_stale) {
1397 if (time_before(jiffies, session->s_cap_ttl)) {
1398 pr_info("mds%d caps renewed\n", session->s_mds);
1399 wake = 1;
1400 } else {
1401 pr_info("mds%d caps still stale\n", session->s_mds);
1402 }
1403 }
1404 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1405 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1406 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1407 spin_unlock(&session->s_cap_lock);
1408
1409 if (wake)
0dc2570f 1410 wake_up_session_caps(session, 0);
2f2dc053
SW		 1411}
1412
1413/*
1414 * send a session close request
1415 */
1416static int request_close_session(struct ceph_mds_client *mdsc,
1417 struct ceph_mds_session *session)
1418{
1419 struct ceph_msg *msg;
2f2dc053
SW		 1420
1421 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1422 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1423 session->s_seq);
1424 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1425 if (!msg)
1426 return -ENOMEM;
1427 ceph_con_send(&session->s_con, msg);
fcff415c 1428 return 1;
2f2dc053
SW		 1429}
1430
1431/*
1432 * Called with s_mutex held.
1433 */
1434static int __close_session(struct ceph_mds_client *mdsc,
1435 struct ceph_mds_session *session)
1436{
1437 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1438 return 0;
1439 session->s_state = CEPH_MDS_SESSION_CLOSING;
1440 return request_close_session(mdsc, session);
1441}
1442
040d7860
YZ		 1443static bool drop_negative_children(struct dentry *dentry)
1444{
1445 struct dentry *child;
1446 bool all_negative = true;
1447
1448 if (!d_is_dir(dentry))
1449 goto out;
1450
1451 spin_lock(&dentry->d_lock);
1452 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1453 if (d_really_is_positive(child)) {
1454 all_negative = false;
1455 break;
1456 }
1457 }
1458 spin_unlock(&dentry->d_lock);
1459
1460 if (all_negative)
1461 shrink_dcache_parent(dentry);
1462out:
1463 return all_negative;
1464}
1465
2f2dc053
SW		 1466/*
1467 * Trim old(er) caps.
1468 *
1469 * Because we can't cache an inode without one or more caps, we do
1470 * this indirectly: if a cap is unused, we prune its aliases, at which
1471 * point the inode will hopefully get dropped to.
1472 *
1473 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1474 * memory pressure from the MDS, though, so it needn't be perfect.
1475 */
1476static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1477{
1478 struct ceph_mds_session *session = arg;
1479 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1480 int used, wanted, oissued, mine;
2f2dc053
SW		 1481
1482 if (session->s_trim_caps <= 0)
1483 return -1;
1484
be655596 1485 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1486 mine = cap->issued | cap->implemented;
1487 used = __ceph_caps_used(ci);
979abfdd 1488 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1489 oissued = __ceph_caps_issued_other(ci, cap);
1490
979abfdd 1491 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1492 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1493 ceph_cap_string(used), ceph_cap_string(wanted));
1494 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1495 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1496 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1497 goto out;
1498 if ((used | wanted) & CEPH_CAP_ANY_WR)
1499 goto out;
89aa5930
YZ		 1500 /* Note: it's possible that i_filelock_ref becomes non-zero
1501 * after dropping auth caps. It doesn't hurt because reply
1502 * of lock mds request will re-add auth caps. */
1503 if (atomic_read(&ci->i_filelock_ref) > 0)
1504 goto out;
979abfdd 1505 }
5e804ac4
YZ		 1506 /* The inode has cached pages, but it's no longer used.
1507 * we can safely drop it */
1508 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1509 !(oissued & CEPH_CAP_FILE_CACHE)) {
1510 used = 0;
1511 oissued = 0;
1512 }
979abfdd 1513 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1514 goto out; /* we need these caps */
1515
2f2dc053
SW		 1516 if (oissued) {
1517 /* we aren't the only cap.. just remove us */
a096b09a 1518 __ceph_remove_cap(cap, true);
040d7860 1519 session->s_trim_caps--;
2f2dc053 1520 } else {
040d7860 1521 struct dentry *dentry;
5e804ac4 1522 /* try dropping referring dentries */
be655596 1523 spin_unlock(&ci->i_ceph_lock);
040d7860
YZ		 1524 dentry = d_find_any_alias(inode);
1525 if (dentry && drop_negative_children(dentry)) {
1526 int count;
1527 dput(dentry);
1528 d_prune_aliases(inode);
1529 count = atomic_read(&inode->i_count);
1530 if (count == 1)
1531 session->s_trim_caps--;
1532 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1533 inode, cap, count);
1534 } else {
1535 dput(dentry);
1536 }
2f2dc053
SW		 1537 return 0;
1538 }
1539
1540out:
be655596 1541 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1542 return 0;
1543}
1544
1545/*
1546 * Trim session cap count down to some max number.
1547 */
1548static int trim_caps(struct ceph_mds_client *mdsc,
1549 struct ceph_mds_session *session,
1550 int max_caps)
1551{
1552 int trim_caps = session->s_nr_caps - max_caps;
1553
1554 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1555 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1556 if (trim_caps > 0) {
1557 session->s_trim_caps = trim_caps;
1558 iterate_session_caps(session, trim_caps_cb, session);
1559 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1560 session->s_mds, session->s_nr_caps, max_caps,
1561 trim_caps - session->s_trim_caps);
5dacf091 1562 session->s_trim_caps = 0;
2f2dc053 1563 }
a56371d9 1564
a56371d9 1565 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1566 return 0;
1567}
1568
8310b089
YZ		 1569static int check_caps_flush(struct ceph_mds_client *mdsc,
1570 u64 want_flush_tid)
1571{
8310b089
YZ		 1572 int ret = 1;
1573
1574 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1575 if (!list_empty(&mdsc->cap_flush_list)) {
1576 struct ceph_cap_flush *cf =
1577 list_first_entry(&mdsc->cap_flush_list,
1578 struct ceph_cap_flush, g_list);
1579 if (cf->tid <= want_flush_tid) {
1580 dout("check_caps_flush still flushing tid "
1581 "%llu <= %llu\n", cf->tid, want_flush_tid);
1582 ret = 0;
1583 }
8310b089
YZ		 1584 }
1585 spin_unlock(&mdsc->cap_dirty_lock);
1586 return ret;
d3383a8e
YZ		 1587}
1588
2f2dc053
SW		 1589/*
1590 * flush all dirty inode data to disk.
1591 *
8310b089 1592 * returns true if we've flushed through want_flush_tid
2f2dc053 1593 */
affbc19a 1594static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1595 u64 want_flush_tid)
2f2dc053 1596{
0e294387 1597 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1598
1599 wait_event(mdsc->cap_flushing_wq,
1600 check_caps_flush(mdsc, want_flush_tid));
1601
1602 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1603}
1604
1605/*
1606 * called under s_mutex
1607 */
3d7ded4d
SW		 1608void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1609 struct ceph_mds_session *session)
2f2dc053 1610{
745a8e3b
YZ		 1611 struct ceph_msg *msg = NULL;
1612 struct ceph_mds_cap_release *head;
1613 struct ceph_mds_cap_item *item;
92475f05 1614 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
745a8e3b
YZ		 1615 struct ceph_cap *cap;
1616 LIST_HEAD(tmp_list);
1617 int num_cap_releases;
92475f05
JL		 1618 __le32 barrier, *cap_barrier;
1619
1620 down_read(&osdc->lock);
1621 barrier = cpu_to_le32(osdc->epoch_barrier);
1622 up_read(&osdc->lock);
2f2dc053 1623
0f8605f2 1624 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1625again:
1626 list_splice_init(&session->s_cap_releases, &tmp_list);
1627 num_cap_releases = session->s_num_cap_releases;
1628 session->s_num_cap_releases = 0;
2f2dc053 1629 spin_unlock(&session->s_cap_lock);
e01a5946 1630
745a8e3b
YZ		 1631 while (!list_empty(&tmp_list)) {
1632 if (!msg) {
1633 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1634 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1635 if (!msg)
1636 goto out_err;
1637 head = msg->front.iov_base;
1638 head->num = cpu_to_le32(0);
1639 msg->front.iov_len = sizeof(*head);
92475f05
JL		 1640
1641 msg->hdr.version = cpu_to_le16(2);
1642 msg->hdr.compat_version = cpu_to_le16(1);
745a8e3b 1643 }
92475f05 1644
745a8e3b
YZ		 1645 cap = list_first_entry(&tmp_list, struct ceph_cap,
1646 session_caps);
1647 list_del(&cap->session_caps);
1648 num_cap_releases--;
e01a5946 1649
00bd8edb 1650 head = msg->front.iov_base;
745a8e3b
YZ		 1651 le32_add_cpu(&head->num, 1);
1652 item = msg->front.iov_base + msg->front.iov_len;
1653 item->ino = cpu_to_le64(cap->cap_ino);
1654 item->cap_id = cpu_to_le64(cap->cap_id);
1655 item->migrate_seq = cpu_to_le32(cap->mseq);
1656 item->seq = cpu_to_le32(cap->issue_seq);
1657 msg->front.iov_len += sizeof(*item);
1658
1659 ceph_put_cap(mdsc, cap);
1660
1661 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
92475f05
JL		 1662 // Append cap_barrier field
1663 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1664 *cap_barrier = barrier;
1665 msg->front.iov_len += sizeof(*cap_barrier);
1666
745a8e3b
YZ		 1667 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1668 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1669 ceph_con_send(&session->s_con, msg);
1670 msg = NULL;
1671 }
00bd8edb 1672 }
e01a5946 1673
745a8e3b 1674 BUG_ON(num_cap_releases != 0);
e01a5946 1675
745a8e3b
YZ		 1676 spin_lock(&session->s_cap_lock);
1677 if (!list_empty(&session->s_cap_releases))
1678 goto again;
1679 spin_unlock(&session->s_cap_lock);
1680
1681 if (msg) {
92475f05
JL		 1682 // Append cap_barrier field
1683 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1684 *cap_barrier = barrier;
1685 msg->front.iov_len += sizeof(*cap_barrier);
1686
745a8e3b
YZ		 1687 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1688 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1689 ceph_con_send(&session->s_con, msg);
e01a5946 1690 }
745a8e3b
YZ		 1691 return;
1692out_err:
1693 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1694 session->s_mds);
1695 spin_lock(&session->s_cap_lock);
1696 list_splice(&tmp_list, &session->s_cap_releases);
1697 session->s_num_cap_releases += num_cap_releases;
1698 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1699}
1700
2f2dc053
SW		 1701/*
1702 * requests
1703 */
1704
54008399
YZ		 1705int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1706 struct inode *dir)
1707{
1708 struct ceph_inode_info *ci = ceph_inode(dir);
1709 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1710 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1711 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1712 int order, num_entries;
1713
1714 spin_lock(&ci->i_ceph_lock);
1715 num_entries = ci->i_files + ci->i_subdirs;
1716 spin_unlock(&ci->i_ceph_lock);
1717 num_entries = max(num_entries, 1);
1718 num_entries = min(num_entries, opt->max_readdir);
1719
1720 order = get_order(size * num_entries);
1721 while (order >= 0) {
2a5beea3
YZ		 1722 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1723 __GFP_NOWARN,
1724 order);
1725 if (rinfo->dir_entries)
54008399
YZ		 1726 break;
1727 order--;
1728 }
2a5beea3 1729 if (!rinfo->dir_entries)
54008399
YZ		 1730 return -ENOMEM;
1731
1732 num_entries = (PAGE_SIZE << order) / size;
1733 num_entries = min(num_entries, opt->max_readdir);
1734
1735 rinfo->dir_buf_size = PAGE_SIZE << order;
1736 req->r_num_caps = num_entries + 1;
1737 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1738 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1739 return 0;
1740}
1741
2f2dc053
SW		 1742/*
1743 * Create an mds request.
1744 */
1745struct ceph_mds_request *
1746ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1747{
1748 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1749
1750 if (!req)
1751 return ERR_PTR(-ENOMEM);
1752
b4556396 1753 mutex_init(&req->r_fill_mutex);
37151668 1754 req->r_mdsc = mdsc;
2f2dc053
SW		 1755 req->r_started = jiffies;
1756 req->r_resend_mds = -1;
1757 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1758 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1759 req->r_fmode = -1;
153c8e6b 1760 kref_init(&req->r_kref);
fcd00b68 1761 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1762 INIT_LIST_HEAD(&req->r_wait);
1763 init_completion(&req->r_completion);
1764 init_completion(&req->r_safe_completion);
1765 INIT_LIST_HEAD(&req->r_unsafe_item);
1766
56199016 1767 req->r_stamp = timespec_trunc(current_kernel_time(), mdsc->fsc->sb->s_time_gran);
b8e69066 1768
2f2dc053
SW		 1769 req->r_op = op;
1770 req->r_direct_mode = mode;
1771 return req;
1772}
1773
1774/*
44ca18f2 1775 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1776 *
1777 * called under mdsc->mutex.
1778 */
44ca18f2
SW		 1779static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1780{
1781 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1782 return NULL;
1783 return rb_entry(rb_first(&mdsc->request_tree),
1784 struct ceph_mds_request, r_node);
1785}
1786
e8a7b8b1 1787static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1788{
e8a7b8b1 1789 return mdsc->oldest_tid;
2f2dc053
SW		 1790}
1791
1792/*
1793 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1794 * on build_path_from_dentry in fs/cifs/dir.c.
1795 *
1796 * If @stop_on_nosnap, generate path relative to the first non-snapped
1797 * inode.
1798 *
1799 * Encode hidden .snap dirs as a double /, i.e.
1800 * foo/.snap/bar -> foo//bar
1801 */
1802char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1803 int stop_on_nosnap)
1804{
1805 struct dentry *temp;
1806 char *path;
1807 int len, pos;
1b71fe2e 1808 unsigned seq;
2f2dc053 1809
d37b1d99 1810 if (!dentry)
2f2dc053
SW		 1811 return ERR_PTR(-EINVAL);
1812
1813retry:
1814 len = 0;
1b71fe2e
AV		 1815 seq = read_seqbegin(&rename_lock);
1816 rcu_read_lock();
2f2dc053 1817 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1818 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1819 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1820 len++; /* slash only */
1821 else if (stop_on_nosnap && inode &&
1822 ceph_snap(inode) == CEPH_NOSNAP)
1823 break;
1824 else
1825 len += 1 + temp->d_name.len;
1826 temp = temp->d_parent;
2f2dc053 1827 }
1b71fe2e 1828 rcu_read_unlock();
2f2dc053
SW		 1829 if (len)
1830 len--; /* no leading '/' */
1831
1832 path = kmalloc(len+1, GFP_NOFS);
d37b1d99 1833 if (!path)
2f2dc053
SW		 1834 return ERR_PTR(-ENOMEM);
1835 pos = len;
1836 path[pos] = 0; /* trailing null */
1b71fe2e 1837 rcu_read_lock();
2f2dc053 1838 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1839 struct inode *inode;
2f2dc053 1840
1b71fe2e 1841 spin_lock(&temp->d_lock);
2b0143b5 1842 inode = d_inode(temp);
2f2dc053 1843 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1844 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1845 pos, temp);
1846 } else if (stop_on_nosnap && inode &&
1847 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1848 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1849 break;
1850 } else {
1851 pos -= temp->d_name.len;
1b71fe2e
AV		 1852 if (pos < 0) {
1853 spin_unlock(&temp->d_lock);
2f2dc053 1854 break;
1b71fe2e 1855 }
2f2dc053
SW		 1856 strncpy(path + pos, temp->d_name.name,
1857 temp->d_name.len);
2f2dc053 1858 }
1b71fe2e 1859 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1860 if (pos)
1861 path[--pos] = '/';
1862 temp = temp->d_parent;
2f2dc053 1863 }
1b71fe2e
AV		 1864 rcu_read_unlock();
1865 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1866 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1867 "expected, namelen is %d, pos is %d\n", len, pos);
1868 /* presumably this is only possible if racing with a
1869 rename of one of the parent directories (we can not
1870 lock the dentries above us to prevent this, but
1871 retrying should be harmless) */
1872 kfree(path);
1873 goto retry;
1874 }
1875
2b0143b5 1876 *base = ceph_ino(d_inode(temp));
2f2dc053 1877 *plen = len;
104648ad 1878 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1879 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1880 return path;
1881}
1882
fd36a717 1883static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2f2dc053
SW		 1884 const char **ppath, int *ppathlen, u64 *pino,
1885 int *pfreepath)
1886{
1887 char *path;
1888
c6b0b656 1889 rcu_read_lock();
fd36a717
JL		 1890 if (!dir)
1891 dir = d_inode_rcu(dentry->d_parent);
c6b0b656
JL		 1892 if (dir && ceph_snap(dir) == CEPH_NOSNAP) {
1893 *pino = ceph_ino(dir);
1894 rcu_read_unlock();
2f2dc053
SW		 1895 *ppath = dentry->d_name.name;
1896 *ppathlen = dentry->d_name.len;
1897 return 0;
1898 }
c6b0b656 1899 rcu_read_unlock();
2f2dc053
SW		 1900 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1901 if (IS_ERR(path))
1902 return PTR_ERR(path);
1903 *ppath = path;
1904 *pfreepath = 1;
1905 return 0;
1906}
1907
1908static int build_inode_path(struct inode *inode,
1909 const char **ppath, int *ppathlen, u64 *pino,
1910 int *pfreepath)
1911{
1912 struct dentry *dentry;
1913 char *path;
1914
1915 if (ceph_snap(inode) == CEPH_NOSNAP) {
1916 *pino = ceph_ino(inode);
1917 *ppathlen = 0;
1918 return 0;
1919 }
1920 dentry = d_find_alias(inode);
1921 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1922 dput(dentry);
1923 if (IS_ERR(path))
1924 return PTR_ERR(path);
1925 *ppath = path;
1926 *pfreepath = 1;
1927 return 0;
1928}
1929
1930/*
1931 * request arguments may be specified via an inode *, a dentry *, or
1932 * an explicit ino+path.
1933 */
1934static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
fd36a717
JL		 1935 struct inode *rdiri, const char *rpath,
1936 u64 rino, const char **ppath, int *pathlen,
2f2dc053
SW		 1937 u64 *ino, int *freepath)
1938{
1939 int r = 0;
1940
1941 if (rinode) {
1942 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1943 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1944 ceph_snap(rinode));
1945 } else if (rdentry) {
fd36a717
JL		 1946 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
1947 freepath);
2f2dc053
SW		 1948 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1949 *ppath);
795858db 1950 } else if (rpath || rino) {
2f2dc053
SW		 1951 *ino = rino;
1952 *ppath = rpath;
b000056a 1953 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1954 dout(" path %.*s\n", *pathlen, rpath);
1955 }
1956
1957 return r;
1958}
1959
1960/*
1961 * called under mdsc->mutex
1962 */
1963static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1964 struct ceph_mds_request *req,
6e6f0923 1965 int mds, bool drop_cap_releases)
2f2dc053
SW		 1966{
1967 struct ceph_msg *msg;
1968 struct ceph_mds_request_head *head;
1969 const char *path1 = NULL;
1970 const char *path2 = NULL;
1971 u64 ino1 = 0, ino2 = 0;
1972 int pathlen1 = 0, pathlen2 = 0;
1973 int freepath1 = 0, freepath2 = 0;
1974 int len;
1975 u16 releases;
1976 void *p, *end;
1977 int ret;
1978
1979 ret = set_request_path_attr(req->r_inode, req->r_dentry,
3dd69aab 1980 req->r_parent, req->r_path1, req->r_ino1.ino,
2f2dc053
SW		 1981 &path1, &pathlen1, &ino1, &freepath1);
1982 if (ret < 0) {
1983 msg = ERR_PTR(ret);
1984 goto out;
1985 }
1986
1987 ret = set_request_path_attr(NULL, req->r_old_dentry,
fd36a717 1988 req->r_old_dentry_dir,
2f2dc053
SW		 1989 req->r_path2, req->r_ino2.ino,
1990 &path2, &pathlen2, &ino2, &freepath2);
1991 if (ret < 0) {
1992 msg = ERR_PTR(ret);
1993 goto out_free1;
1994 }
1995
1996 len = sizeof(*head) +
b8e69066 1997 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 1998 sizeof(struct ceph_timespec);
2f2dc053
SW		 1999
2000 /* calculate (max) length for cap releases */
2001 len += sizeof(struct ceph_mds_request_release) *
2002 (!!req->r_inode_drop + !!req->r_dentry_drop +
2003 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2004 if (req->r_dentry_drop)
2005 len += req->r_dentry->d_name.len;
2006 if (req->r_old_dentry_drop)
2007 len += req->r_old_dentry->d_name.len;
2008
b61c2763 2009 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 2010 if (!msg) {
2011 msg = ERR_PTR(-ENOMEM);
2f2dc053 2012 goto out_free2;
a79832f2 2013 }
2f2dc053 2014
7cfa0313 2015 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 2016 msg->hdr.tid = cpu_to_le64(req->r_tid);
2017
2f2dc053
SW		 2018 head = msg->front.iov_base;
2019 p = msg->front.iov_base + sizeof(*head);
2020 end = msg->front.iov_base + msg->front.iov_len;
2021
2022 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2023 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 2024 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2025 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 2026 head->args = req->r_args;
2027
2028 ceph_encode_filepath(&p, end, ino1, path1);
2029 ceph_encode_filepath(&p, end, ino2, path2);
2030
e979cf50
SW		 2031 /* make note of release offset, in case we need to replay */
2032 req->r_request_release_offset = p - msg->front.iov_base;
2033
2f2dc053
SW		 2034 /* cap releases */
2035 releases = 0;
2036 if (req->r_inode_drop)
2037 releases += ceph_encode_inode_release(&p,
2b0143b5 2038 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 2039 mds, req->r_inode_drop, req->r_inode_unless, 0);
2040 if (req->r_dentry_drop)
2041 releases += ceph_encode_dentry_release(&p, req->r_dentry,
3dd69aab 2042 req->r_parent, mds, req->r_dentry_drop,
ca6c8ae0 2043 req->r_dentry_unless);
2f2dc053
SW		 2044 if (req->r_old_dentry_drop)
2045 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
ca6c8ae0
JL		 2046 req->r_old_dentry_dir, mds,
2047 req->r_old_dentry_drop,
2048 req->r_old_dentry_unless);
2f2dc053
SW		 2049 if (req->r_old_inode_drop)
2050 releases += ceph_encode_inode_release(&p,
2b0143b5 2051 d_inode(req->r_old_dentry),
2f2dc053 2052 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 2053
2054 if (drop_cap_releases) {
2055 releases = 0;
2056 p = msg->front.iov_base + req->r_request_release_offset;
2057 }
2058
2f2dc053
SW		 2059 head->num_releases = cpu_to_le16(releases);
2060
b8e69066 2061 /* time stamp */
1f041a89
YZ		 2062 {
2063 struct ceph_timespec ts;
2064 ceph_encode_timespec(&ts, &req->r_stamp);
2065 ceph_encode_copy(&p, &ts, sizeof(ts));
2066 }
b8e69066 2067
2f2dc053
SW		 2068 BUG_ON(p > end);
2069 msg->front.iov_len = p - msg->front.iov_base;
2070 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2071
25e6bae3
YZ		 2072 if (req->r_pagelist) {
2073 struct ceph_pagelist *pagelist = req->r_pagelist;
0e1a5ee6 2074 refcount_inc(&pagelist->refcnt);
25e6bae3
YZ		 2075 ceph_msg_data_add_pagelist(msg, pagelist);
2076 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2077 } else {
2078 msg->hdr.data_len = 0;
ebf18f47 2079 }
02afca6c 2080
2f2dc053
SW		 2081 msg->hdr.data_off = cpu_to_le16(0);
2082
2083out_free2:
2084 if (freepath2)
2085 kfree((char *)path2);
2086out_free1:
2087 if (freepath1)
2088 kfree((char *)path1);
2089out:
2090 return msg;
2091}
2092
2093/*
2094 * called under mdsc->mutex if error, under no mutex if
2095 * success.
2096 */
2097static void complete_request(struct ceph_mds_client *mdsc,
2098 struct ceph_mds_request *req)
2099{
2100 if (req->r_callback)
2101 req->r_callback(mdsc, req);
2102 else
03066f23 2103 complete_all(&req->r_completion);
2f2dc053
SW		 2104}
2105
2106/*
2107 * called under mdsc->mutex
2108 */
2109static int __prepare_send_request(struct ceph_mds_client *mdsc,
2110 struct ceph_mds_request *req,
6e6f0923 2111 int mds, bool drop_cap_releases)
2f2dc053
SW		 2112{
2113 struct ceph_mds_request_head *rhead;
2114 struct ceph_msg *msg;
2115 int flags = 0;
2116
2f2dc053 2117 req->r_attempts++;
e55b71f8
GF		 2118 if (req->r_inode) {
2119 struct ceph_cap *cap =
2120 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2121
2122 if (cap)
2123 req->r_sent_on_mseq = cap->mseq;
2124 else
2125 req->r_sent_on_mseq = -1;
2126 }
2f2dc053
SW		 2127 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2128 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2129
bc2de10d 2130 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
c5c9a0bf 2131 void *p;
01a92f17
SW		 2132 /*
2133 * Replay. Do not regenerate message (and rebuild
2134 * paths, etc.); just use the original message.
2135 * Rebuilding paths will break for renames because
2136 * d_move mangles the src name.
2137 */
2138 msg = req->r_request;
2139 rhead = msg->front.iov_base;
2140
2141 flags = le32_to_cpu(rhead->flags);
2142 flags |= CEPH_MDS_FLAG_REPLAY;
2143 rhead->flags = cpu_to_le32(flags);
2144
2145 if (req->r_target_inode)
2146 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2147
2148 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2149
2150 /* remove cap/dentry releases from message */
2151 rhead->num_releases = 0;
c5c9a0bf
YZ		 2152
2153 /* time stamp */
2154 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2155 {
2156 struct ceph_timespec ts;
2157 ceph_encode_timespec(&ts, &req->r_stamp);
2158 ceph_encode_copy(&p, &ts, sizeof(ts));
2159 }
c5c9a0bf
YZ		 2160
2161 msg->front.iov_len = p - msg->front.iov_base;
2162 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2163 return 0;
2164 }
2165
2f2dc053
SW		 2166 if (req->r_request) {
2167 ceph_msg_put(req->r_request);
2168 req->r_request = NULL;
2169 }
6e6f0923 2170 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2171 if (IS_ERR(msg)) {
e1518c7c 2172 req->r_err = PTR_ERR(msg);
a79832f2 2173 return PTR_ERR(msg);
2f2dc053
SW		 2174 }
2175 req->r_request = msg;
2176
2177 rhead = msg->front.iov_base;
2f2dc053 2178 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
bc2de10d 2179 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2f2dc053 2180 flags |= CEPH_MDS_FLAG_REPLAY;
3dd69aab 2181 if (req->r_parent)
2f2dc053
SW		 2182 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2183 rhead->flags = cpu_to_le32(flags);
2184 rhead->num_fwd = req->r_num_fwd;
2185 rhead->num_retry = req->r_attempts - 1;
01a92f17 2186 rhead->ino = 0;
2f2dc053 2187
3dd69aab 2188 dout(" r_parent = %p\n", req->r_parent);
2f2dc053
SW		 2189 return 0;
2190}
2191
2192/*
2193 * send request, or put it on the appropriate wait list.
2194 */
2195static int __do_request(struct ceph_mds_client *mdsc,
2196 struct ceph_mds_request *req)
2197{
2198 struct ceph_mds_session *session = NULL;
2199 int mds = -1;
48fec5d0 2200 int err = 0;
2f2dc053 2201
bc2de10d
JL		 2202 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2203 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
eb1b8af3 2204 __unregister_request(mdsc, req);
2f2dc053 2205 goto out;
eb1b8af3 2206 }
2f2dc053
SW		 2207
2208 if (req->r_timeout &&
2209 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2210 dout("do_request timed out\n");
2211 err = -EIO;
2212 goto finish;
2213 }
52953d55 2214 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
48fec5d0
YZ		 2215 dout("do_request forced umount\n");
2216 err = -EIO;
2217 goto finish;
2218 }
52953d55 2219 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
e9e427f0
YZ		 2220 if (mdsc->mdsmap_err) {
2221 err = mdsc->mdsmap_err;
2222 dout("do_request mdsmap err %d\n", err);
2223 goto finish;
2224 }
cc8e8342
YZ		 2225 if (mdsc->mdsmap->m_epoch == 0) {
2226 dout("do_request no mdsmap, waiting for map\n");
2227 list_add(&req->r_wait, &mdsc->waiting_for_map);
2228 goto finish;
2229 }
e9e427f0
YZ		 2230 if (!(mdsc->fsc->mount_options->flags &
2231 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2232 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2233 err = -ENOENT;
2234 pr_info("probably no mds server is up\n");
2235 goto finish;
2236 }
2237 }
2f2dc053 2238
dc69e2e9
SW		 2239 put_request_session(req);
2240
2f2dc053
SW		 2241 mds = __choose_mds(mdsc, req);
2242 if (mds < 0 ||
2243 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2244 dout("do_request no mds or not active, waiting for map\n");
2245 list_add(&req->r_wait, &mdsc->waiting_for_map);
2246 goto out;
2247 }
2248
2249 /* get, open session */
2250 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2251 if (!session) {
2f2dc053 2252 session = register_session(mdsc, mds);
9c423956
SW		 2253 if (IS_ERR(session)) {
2254 err = PTR_ERR(session);
2255 goto finish;
2256 }
2257 }
dc69e2e9
SW		 2258 req->r_session = get_session(session);
2259
2f2dc053 2260 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2261 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2262 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2263 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2264 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2265 err = -EACCES;
2266 goto out_session;
2267 }
2f2dc053
SW		 2268 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2269 session->s_state == CEPH_MDS_SESSION_CLOSING)
2270 __open_session(mdsc, session);
2271 list_add(&req->r_wait, &session->s_waiting);
2272 goto out_session;
2273 }
2274
2275 /* send request */
2f2dc053
SW		 2276 req->r_resend_mds = -1; /* forget any previous mds hint */
2277
2278 if (req->r_request_started == 0) /* note request start time */
2279 req->r_request_started = jiffies;
2280
6e6f0923 2281 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2282 if (!err) {
2283 ceph_msg_get(req->r_request);
2284 ceph_con_send(&session->s_con, req->r_request);
2285 }
2286
2287out_session:
2288 ceph_put_mds_session(session);
48fec5d0
YZ		 2289finish:
2290 if (err) {
2291 dout("__do_request early error %d\n", err);
2292 req->r_err = err;
2293 complete_request(mdsc, req);
2294 __unregister_request(mdsc, req);
2295 }
2f2dc053
SW		 2296out:
2297 return err;
2f2dc053
SW		 2298}
2299
2300/*
2301 * called under mdsc->mutex
2302 */
2303static void __wake_requests(struct ceph_mds_client *mdsc,
2304 struct list_head *head)
2305{
ed75ec2c
YZ		 2306 struct ceph_mds_request *req;
2307 LIST_HEAD(tmp_list);
2308
2309 list_splice_init(head, &tmp_list);
2f2dc053 2310
ed75ec2c
YZ		 2311 while (!list_empty(&tmp_list)) {
2312 req = list_entry(tmp_list.next,
2313 struct ceph_mds_request, r_wait);
2f2dc053 2314 list_del_init(&req->r_wait);
7971bd92 2315 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2316 __do_request(mdsc, req);
2317 }
2318}
2319
2320/*
2321 * Wake up threads with requests pending for @mds, so that they can
29790f26 2322 * resubmit their requests to a possibly different mds.
2f2dc053 2323 */
29790f26 2324static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2325{
44ca18f2 2326 struct ceph_mds_request *req;
282c1052 2327 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2328
2329 dout("kick_requests mds%d\n", mds);
282c1052 2330 while (p) {
44ca18f2 2331 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2332 p = rb_next(p);
bc2de10d 2333 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
44ca18f2 2334 continue;
3de22be6
YZ		 2335 if (req->r_attempts > 0)
2336 continue; /* only new requests */
44ca18f2
SW		 2337 if (req->r_session &&
2338 req->r_session->s_mds == mds) {
2339 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2340 list_del_init(&req->r_wait);
44ca18f2 2341 __do_request(mdsc, req);
2f2dc053
SW		 2342 }
2343 }
2344}
2345
2346void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2347 struct ceph_mds_request *req)
2348{
2349 dout("submit_request on %p\n", req);
2350 mutex_lock(&mdsc->mutex);
2351 __register_request(mdsc, req, NULL);
2352 __do_request(mdsc, req);
2353 mutex_unlock(&mdsc->mutex);
2354}
2355
2356/*
2357 * Synchrously perform an mds request. Take care of all of the
2358 * session setup, forwarding, retry details.
2359 */
2360int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2361 struct inode *dir,
2362 struct ceph_mds_request *req)
2363{
2364 int err;
2365
2366 dout("do_request on %p\n", req);
2367
3dd69aab 2368 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2f2dc053
SW		 2369 if (req->r_inode)
2370 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3dd69aab
JL		 2371 if (req->r_parent)
2372 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
844d87c3 2373 if (req->r_old_dentry_dir)
41b02e1f
SW		 2374 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2375 CEPH_CAP_PIN);
2f2dc053
SW		 2376
2377 /* issue */
2378 mutex_lock(&mdsc->mutex);
2379 __register_request(mdsc, req, dir);
2380 __do_request(mdsc, req);
2381
e1518c7c
SW		 2382 if (req->r_err) {
2383 err = req->r_err;
e1518c7c 2384 goto out;
2f2dc053
SW		 2385 }
2386
e1518c7c
SW		 2387 /* wait */
2388 mutex_unlock(&mdsc->mutex);
2389 dout("do_request waiting\n");
5be73034 2390 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2391 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2392 } else {
5be73034
ID		 2393 long timeleft = wait_for_completion_killable_timeout(
2394 &req->r_completion,
2395 ceph_timeout_jiffies(req->r_timeout));
2396 if (timeleft > 0)
2397 err = 0;
2398 else if (!timeleft)
2399 err = -EIO; /* timed out */
2400 else
2401 err = timeleft; /* killed */
e1518c7c
SW		 2402 }
2403 dout("do_request waited, got %d\n", err);
2404 mutex_lock(&mdsc->mutex);
5b1daecd 2405
e1518c7c 2406 /* only abort if we didn't race with a real reply */
bc2de10d 2407 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
e1518c7c
SW		 2408 err = le32_to_cpu(req->r_reply_info.head->result);
2409 } else if (err < 0) {
2410 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2411
2412 /*
2413 * ensure we aren't running concurrently with
2414 * ceph_fill_trace or ceph_readdir_prepopulate, which
2415 * rely on locks (dir mutex) held by our caller.
2416 */
2417 mutex_lock(&req->r_fill_mutex);
e1518c7c 2418 req->r_err = err;
bc2de10d 2419 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
b4556396 2420 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2421
3dd69aab 2422 if (req->r_parent &&
167c9e35
SW		 2423 (req->r_op & CEPH_MDS_OP_WRITE))
2424 ceph_invalidate_dir_request(req);
2f2dc053 2425 } else {
e1518c7c 2426 err = req->r_err;
2f2dc053 2427 }
2f2dc053 2428
e1518c7c
SW		 2429out:
2430 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2431 dout("do_request %p done, result %d\n", req, err);
2432 return err;
2433}
2434
167c9e35 2435/*
2f276c51 2436 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2437 * namespace request.
2438 */
2439void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2440{
3dd69aab 2441 struct inode *inode = req->r_parent;
167c9e35 2442
2f276c51 2443 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2444
2f276c51 2445 ceph_dir_clear_complete(inode);
167c9e35
SW		 2446 if (req->r_dentry)
2447 ceph_invalidate_dentry_lease(req->r_dentry);
2448 if (req->r_old_dentry)
2449 ceph_invalidate_dentry_lease(req->r_old_dentry);
2450}
2451
2f2dc053
SW		 2452/*
2453 * Handle mds reply.
2454 *
2455 * We take the session mutex and parse and process the reply immediately.
2456 * This preserves the logical ordering of replies, capabilities, etc., sent
2457 * by the MDS as they are applied to our local cache.
2458 */
2459static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2460{
2461 struct ceph_mds_client *mdsc = session->s_mdsc;
2462 struct ceph_mds_request *req;
2463 struct ceph_mds_reply_head *head = msg->front.iov_base;
2464 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2465 struct ceph_snap_realm *realm;
2f2dc053
SW		 2466 u64 tid;
2467 int err, result;
2600d2dd 2468 int mds = session->s_mds;
2f2dc053 2469
2f2dc053
SW		 2470 if (msg->front.iov_len < sizeof(*head)) {
2471 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2472 ceph_msg_dump(msg);
2f2dc053
SW		 2473 return;
2474 }
2475
2476 /* get request, session */
6df058c0 2477 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2478 mutex_lock(&mdsc->mutex);
fcd00b68 2479 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2480 if (!req) {
2481 dout("handle_reply on unknown tid %llu\n", tid);
2482 mutex_unlock(&mdsc->mutex);
2483 return;
2484 }
2485 dout("handle_reply %p\n", req);
2f2dc053
SW		 2486
2487 /* correct session? */
d96d6049 2488 if (req->r_session != session) {
2f2dc053
SW		 2489 pr_err("mdsc_handle_reply got %llu on session mds%d"
2490 " not mds%d\n", tid, session->s_mds,
2491 req->r_session ? req->r_session->s_mds : -1);
2492 mutex_unlock(&mdsc->mutex);
2493 goto out;
2494 }
2495
2496 /* dup? */
bc2de10d
JL		 2497 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2498 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
f3ae1b97 2499 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2500 head->safe ? "safe" : "unsafe", tid, mds);
2501 mutex_unlock(&mdsc->mutex);
2502 goto out;
2503 }
bc2de10d 2504 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
f3ae1b97 2505 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2506 tid, mds);
2507 mutex_unlock(&mdsc->mutex);
2508 goto out;
2509 }
2f2dc053
SW		 2510
2511 result = le32_to_cpu(head->result);
2512
2513 /*
e55b71f8
GF		 2514 * Handle an ESTALE
2515 * if we're not talking to the authority, send to them
2516 * if the authority has changed while we weren't looking,
2517 * send to new authority
2518 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2519 */
2520 if (result == -ESTALE) {
e55b71f8 2521 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2522 req->r_resend_mds = -1;
ca18bede 2523 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2524 dout("not using auth, setting for that now");
2525 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2526 __do_request(mdsc, req);
2527 mutex_unlock(&mdsc->mutex);
2528 goto out;
e55b71f8 2529 } else {
ca18bede
YZ		 2530 int mds = __choose_mds(mdsc, req);
2531 if (mds >= 0 && mds != req->r_session->s_mds) {
2532 dout("but auth changed, so resending");
e55b71f8
GF		 2533 __do_request(mdsc, req);
2534 mutex_unlock(&mdsc->mutex);
2535 goto out;
2536 }
2f2dc053 2537 }
e55b71f8 2538 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2539 }
2540
e55b71f8 2541
2f2dc053 2542 if (head->safe) {
bc2de10d 2543 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2f2dc053 2544 __unregister_request(mdsc, req);
2f2dc053 2545
bc2de10d 2546 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053
SW		 2547 /*
2548 * We already handled the unsafe response, now do the
2549 * cleanup. No need to examine the response; the MDS
2550 * doesn't include any result info in the safe
2551 * response. And even if it did, there is nothing
2552 * useful we could do with a revised return value.
2553 */
2554 dout("got safe reply %llu, mds%d\n", tid, mds);
2f2dc053
SW		 2555
2556 /* last unsafe request during umount? */
44ca18f2 2557 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2558 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2559 mutex_unlock(&mdsc->mutex);
2560 goto out;
2561 }
e1518c7c 2562 } else {
bc2de10d 2563 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2f2dc053 2564 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2565 if (req->r_unsafe_dir) {
2566 struct ceph_inode_info *ci =
2567 ceph_inode(req->r_unsafe_dir);
2568 spin_lock(&ci->i_unsafe_lock);
2569 list_add_tail(&req->r_unsafe_dir_item,
2570 &ci->i_unsafe_dirops);
2571 spin_unlock(&ci->i_unsafe_lock);
2572 }
2f2dc053
SW		 2573 }
2574
2575 dout("handle_reply tid %lld result %d\n", tid, result);
2576 rinfo = &req->r_reply_info;
14303d20 2577 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2578 mutex_unlock(&mdsc->mutex);
2579
2580 mutex_lock(&session->s_mutex);
2581 if (err < 0) {
25933abd 2582 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2583 ceph_msg_dump(msg);
2f2dc053
SW		 2584 goto out_err;
2585 }
2586
2587 /* snap trace */
982d6011 2588 realm = NULL;
2f2dc053
SW		 2589 if (rinfo->snapblob_len) {
2590 down_write(&mdsc->snap_rwsem);
2591 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2592 rinfo->snapblob + rinfo->snapblob_len,
2593 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2594 &realm);
2f2dc053
SW		 2595 downgrade_write(&mdsc->snap_rwsem);
2596 } else {
2597 down_read(&mdsc->snap_rwsem);
2598 }
2599
2600 /* insert trace into our cache */
b4556396 2601 mutex_lock(&req->r_fill_mutex);
315f2408 2602 current->journal_info = req;
f5a03b08 2603 err = ceph_fill_trace(mdsc->fsc->sb, req);
2f2dc053 2604 if (err == 0) {
6e8575fa 2605 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2606 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2607 ceph_readdir_prepopulate(req, req->r_session);
37151668 2608 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2609 }
315f2408 2610 current->journal_info = NULL;
b4556396 2611 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2612
2613 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2614 if (realm)
2615 ceph_put_snap_realm(mdsc, realm);
68cd5b4b 2616
bc2de10d
JL		 2617 if (err == 0 && req->r_target_inode &&
2618 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 2619 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2620 spin_lock(&ci->i_unsafe_lock);
2621 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2622 spin_unlock(&ci->i_unsafe_lock);
2623 }
2f2dc053 2624out_err:
e1518c7c 2625 mutex_lock(&mdsc->mutex);
bc2de10d 2626 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
e1518c7c
SW		 2627 if (err) {
2628 req->r_err = err;
2629 } else {
5fdb1389 2630 req->r_reply = ceph_msg_get(msg);
bc2de10d 2631 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
e1518c7c 2632 }
2f2dc053 2633 } else {
e1518c7c 2634 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2635 }
e1518c7c 2636 mutex_unlock(&mdsc->mutex);
2f2dc053 2637
2f2dc053
SW		 2638 mutex_unlock(&session->s_mutex);
2639
2640 /* kick calling process */
2641 complete_request(mdsc, req);
2642out:
2643 ceph_mdsc_put_request(req);
2644 return;
2645}
2646
2647
2648
2649/*
2650 * handle mds notification that our request has been forwarded.
2651 */
2600d2dd
SW		 2652static void handle_forward(struct ceph_mds_client *mdsc,
2653 struct ceph_mds_session *session,
2654 struct ceph_msg *msg)
2f2dc053
SW		 2655{
2656 struct ceph_mds_request *req;
a1ea787c 2657 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2658 u32 next_mds;
2659 u32 fwd_seq;
2f2dc053
SW		 2660 int err = -EINVAL;
2661 void *p = msg->front.iov_base;
2662 void *end = p + msg->front.iov_len;
2f2dc053 2663
a1ea787c 2664 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2665 next_mds = ceph_decode_32(&p);
2666 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2667
2668 mutex_lock(&mdsc->mutex);
fcd00b68 2669 req = lookup_get_request(mdsc, tid);
2f2dc053 2670 if (!req) {
2a8e5e36 2671 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2672 goto out; /* dup reply? */
2673 }
2674
bc2de10d 2675 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2a8e5e36
SW		 2676 dout("forward tid %llu aborted, unregistering\n", tid);
2677 __unregister_request(mdsc, req);
2678 } else if (fwd_seq <= req->r_num_fwd) {
2679 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2680 tid, next_mds, req->r_num_fwd, fwd_seq);
2681 } else {
2682 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2683 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2684 BUG_ON(req->r_err);
bc2de10d 2685 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3de22be6 2686 req->r_attempts = 0;
2f2dc053
SW		 2687 req->r_num_fwd = fwd_seq;
2688 req->r_resend_mds = next_mds;
2689 put_request_session(req);
2690 __do_request(mdsc, req);
2691 }
2692 ceph_mdsc_put_request(req);
2693out:
2694 mutex_unlock(&mdsc->mutex);
2695 return;
2696
2697bad:
2698 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2699}
2700
2701/*
2702 * handle a mds session control message
2703 */
2704static void handle_session(struct ceph_mds_session *session,
2705 struct ceph_msg *msg)
2706{
2707 struct ceph_mds_client *mdsc = session->s_mdsc;
2708 u32 op;
2709 u64 seq;
2600d2dd 2710 int mds = session->s_mds;
2f2dc053
SW		 2711 struct ceph_mds_session_head *h = msg->front.iov_base;
2712 int wake = 0;
2713
2f2dc053
SW		 2714 /* decode */
2715 if (msg->front.iov_len != sizeof(*h))
2716 goto bad;
2717 op = le32_to_cpu(h->op);
2718 seq = le64_to_cpu(h->seq);
2719
2720 mutex_lock(&mdsc->mutex);
0a07fc8c
YZ		 2721 if (op == CEPH_SESSION_CLOSE) {
2722 get_session(session);
2600d2dd 2723 __unregister_session(mdsc, session);
0a07fc8c 2724 }
2f2dc053
SW		 2725 /* FIXME: this ttl calculation is generous */
2726 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2727 mutex_unlock(&mdsc->mutex);
2728
2729 mutex_lock(&session->s_mutex);
2730
2731 dout("handle_session mds%d %s %p state %s seq %llu\n",
2732 mds, ceph_session_op_name(op), session,
a687ecaf 2733 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2734
2735 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2736 session->s_state = CEPH_MDS_SESSION_OPEN;
2737 pr_info("mds%d came back\n", session->s_mds);
2738 }
2739
2740 switch (op) {
2741 case CEPH_SESSION_OPEN:
29790f26
SW		 2742 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2743 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2744 session->s_state = CEPH_MDS_SESSION_OPEN;
2745 renewed_caps(mdsc, session, 0);
2746 wake = 1;
2747 if (mdsc->stopping)
2748 __close_session(mdsc, session);
2749 break;
2750
2751 case CEPH_SESSION_RENEWCAPS:
2752 if (session->s_renew_seq == seq)
2753 renewed_caps(mdsc, session, 1);
2754 break;
2755
2756 case CEPH_SESSION_CLOSE:
29790f26
SW		 2757 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2758 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2759 cleanup_session_requests(mdsc, session);
2f2dc053 2760 remove_session_caps(session);
656e4382 2761 wake = 2; /* for good measure */
f3c60c59 2762 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2763 break;
2764
2765 case CEPH_SESSION_STALE:
2766 pr_info("mds%d caps went stale, renewing\n",
2767 session->s_mds);
d8fb02ab 2768 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2769 session->s_cap_gen++;
1ce208a6 2770 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2771 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2772 send_renew_caps(mdsc, session);
2773 break;
2774
2775 case CEPH_SESSION_RECALL_STATE:
2776 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2777 break;
2778
186e4f7a
YZ		 2779 case CEPH_SESSION_FLUSHMSG:
2780 send_flushmsg_ack(mdsc, session, seq);
2781 break;
2782
03f4fcb0
YZ		 2783 case CEPH_SESSION_FORCE_RO:
2784 dout("force_session_readonly %p\n", session);
2785 spin_lock(&session->s_cap_lock);
2786 session->s_readonly = true;
2787 spin_unlock(&session->s_cap_lock);
2788 wake_up_session_caps(session, 0);
2789 break;
2790
fcff415c
YZ		 2791 case CEPH_SESSION_REJECT:
2792 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2793 pr_info("mds%d rejected session\n", session->s_mds);
2794 session->s_state = CEPH_MDS_SESSION_REJECTED;
2795 cleanup_session_requests(mdsc, session);
2796 remove_session_caps(session);
2797 wake = 2; /* for good measure */
2798 break;
2799
2f2dc053
SW		 2800 default:
2801 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2802 WARN_ON(1);
2803 }
2804
2805 mutex_unlock(&session->s_mutex);
2806 if (wake) {
2807 mutex_lock(&mdsc->mutex);
2808 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2809 if (wake == 2)
2810 kick_requests(mdsc, mds);
2f2dc053
SW		 2811 mutex_unlock(&mdsc->mutex);
2812 }
0a07fc8c
YZ		 2813 if (op == CEPH_SESSION_CLOSE)
2814 ceph_put_mds_session(session);
2f2dc053
SW		 2815 return;
2816
2817bad:
2818 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2819 (int)msg->front.iov_len);
9ec7cab1 2820 ceph_msg_dump(msg);
2f2dc053
SW		 2821 return;
2822}
2823
2824
2825/*
2826 * called under session->mutex.
2827 */
2828static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2829 struct ceph_mds_session *session)
2830{
2831 struct ceph_mds_request *req, *nreq;
3de22be6 2832 struct rb_node *p;
2f2dc053
SW		 2833 int err;
2834
2835 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2836
2837 mutex_lock(&mdsc->mutex);
2838 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2839 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2840 if (!err) {
2841 ceph_msg_get(req->r_request);
2842 ceph_con_send(&session->s_con, req->r_request);
2843 }
2844 }
3de22be6
YZ		 2845
2846 /*
2847 * also re-send old requests when MDS enters reconnect stage. So that MDS
2848 * can process completed request in clientreplay stage.
2849 */
2850 p = rb_first(&mdsc->request_tree);
2851 while (p) {
2852 req = rb_entry(p, struct ceph_mds_request, r_node);
2853 p = rb_next(p);
bc2de10d 2854 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3de22be6
YZ		 2855 continue;
2856 if (req->r_attempts == 0)
2857 continue; /* only old requests */
2858 if (req->r_session &&
2859 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2860 err = __prepare_send_request(mdsc, req,
2861 session->s_mds, true);
3de22be6
YZ		 2862 if (!err) {
2863 ceph_msg_get(req->r_request);
2864 ceph_con_send(&session->s_con, req->r_request);
2865 }
2866 }
2867 }
2f2dc053
SW		 2868 mutex_unlock(&mdsc->mutex);
2869}
2870
2871/*
2872 * Encode information about a cap for a reconnect with the MDS.
2873 */
2f2dc053
SW		 2874static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2875 void *arg)
2876{
20cb34ae
SW		 2877 union {
2878 struct ceph_mds_cap_reconnect v2;
2879 struct ceph_mds_cap_reconnect_v1 v1;
2880 } rec;
b3f8d68f 2881 struct ceph_inode_info *ci = cap->ci;
20cb34ae
SW		 2882 struct ceph_reconnect_state *recon_state = arg;
2883 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2884 char *path;
2885 int pathlen, err;
2886 u64 pathbase;
3469ed0d 2887 u64 snap_follows;
2f2dc053
SW		 2888 struct dentry *dentry;
2889
2f2dc053
SW		 2890 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2891 inode, ceph_vinop(inode), cap, cap->cap_id,
2892 ceph_cap_string(cap->issued));
93cea5be
SW		 2893 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2894 if (err)
2895 return err;
2f2dc053
SW		 2896
2897 dentry = d_find_alias(inode);
2898 if (dentry) {
2899 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2900 if (IS_ERR(path)) {
2901 err = PTR_ERR(path);
e072f8aa 2902 goto out_dput;
2f2dc053
SW		 2903 }
2904 } else {
2905 path = NULL;
2906 pathlen = 0;
4eacd4cb 2907 pathbase = 0;
2f2dc053 2908 }
2f2dc053 2909
be655596 2910 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2911 cap->seq = 0; /* reset cap seq */
2912 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2913 cap->mseq = 0; /* and migrate_seq */
99a9c273 2914 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2915
121f22a1 2916 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2917 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2918 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2919 rec.v2.issued = cpu_to_le32(cap->issued);
2920 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2921 rec.v2.pathbase = cpu_to_le64(pathbase);
ec1dff25
JL		 2922 rec.v2.flock_len = (__force __le32)
2923 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
20cb34ae
SW		 2924 } else {
2925 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2926 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2927 rec.v1.issued = cpu_to_le32(cap->issued);
2928 rec.v1.size = cpu_to_le64(inode->i_size);
2929 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2930 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2931 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2932 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2933 }
3469ed0d
YZ		 2934
2935 if (list_empty(&ci->i_cap_snaps)) {
92776fd2 2936 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3469ed0d
YZ		 2937 } else {
2938 struct ceph_cap_snap *capsnap =
2939 list_first_entry(&ci->i_cap_snaps,
2940 struct ceph_cap_snap, ci_item);
2941 snap_follows = capsnap->follows;
20cb34ae 2942 }
be655596 2943 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2944
121f22a1 2945 if (recon_state->msg_version >= 2) {
40819f6f 2946 int num_fcntl_locks, num_flock_locks;
4deb14a2 2947 struct ceph_filelock *flocks = NULL;
121f22a1
YZ		 2948 size_t struct_len, total_len = 0;
2949 u8 struct_v = 0;
39be95e9
JS		 2950
2951encode_again:
b3f8d68f
YZ		 2952 if (rec.v2.flock_len) {
2953 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2954 } else {
2955 num_fcntl_locks = 0;
2956 num_flock_locks = 0;
2957 }
4deb14a2
YZ		 2958 if (num_fcntl_locks + num_flock_locks > 0) {
2959 flocks = kmalloc((num_fcntl_locks + num_flock_locks) *
2960 sizeof(struct ceph_filelock), GFP_NOFS);
2961 if (!flocks) {
2962 err = -ENOMEM;
2963 goto out_free;
2964 }
2965 err = ceph_encode_locks_to_buffer(inode, flocks,
2966 num_fcntl_locks,
2967 num_flock_locks);
2968 if (err) {
2969 kfree(flocks);
2970 flocks = NULL;
2971 if (err == -ENOSPC)
2972 goto encode_again;
2973 goto out_free;
2974 }
2975 } else {
39be95e9 2976 kfree(flocks);
4deb14a2 2977 flocks = NULL;
39be95e9 2978 }
121f22a1
YZ		 2979
2980 if (recon_state->msg_version >= 3) {
2981 /* version, compat_version and struct_len */
2982 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 2983 struct_v = 2;
121f22a1 2984 }
39be95e9
JS		 2985 /*
2986 * number of encoded locks is stable, so copy to pagelist
2987 */
121f22a1
YZ		 2988 struct_len = 2 * sizeof(u32) +
2989 (num_fcntl_locks + num_flock_locks) *
2990 sizeof(struct ceph_filelock);
2991 rec.v2.flock_len = cpu_to_le32(struct_len);
2992
2993 struct_len += sizeof(rec.v2);
2994 struct_len += sizeof(u32) + pathlen;
2995
3469ed0d
YZ		 2996 if (struct_v >= 2)
2997 struct_len += sizeof(u64); /* snap_follows */
2998
121f22a1
YZ		 2999 total_len += struct_len;
3000 err = ceph_pagelist_reserve(pagelist, total_len);
3001
3002 if (!err) {
3003 if (recon_state->msg_version >= 3) {
3004 ceph_pagelist_encode_8(pagelist, struct_v);
3005 ceph_pagelist_encode_8(pagelist, 1);
3006 ceph_pagelist_encode_32(pagelist, struct_len);
3007 }
3008 ceph_pagelist_encode_string(pagelist, path, pathlen);
3009 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3010 ceph_locks_to_pagelist(flocks, pagelist,
3011 num_fcntl_locks,
3012 num_flock_locks);
3469ed0d
YZ		 3013 if (struct_v >= 2)
3014 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 3015 }
39be95e9 3016 kfree(flocks);
3612abbd 3017 } else {
121f22a1
YZ		 3018 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
3019 err = ceph_pagelist_reserve(pagelist, size);
3020 if (!err) {
3021 ceph_pagelist_encode_string(pagelist, path, pathlen);
3022 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3023 }
40819f6f 3024 }
44c99757
YZ		 3025
3026 recon_state->nr_caps++;
e072f8aa 3027out_free:
2f2dc053 3028 kfree(path);
e072f8aa 3029out_dput:
2f2dc053 3030 dput(dentry);
93cea5be 3031 return err;
2f2dc053
SW		 3032}
3033
3034
3035/*
3036 * If an MDS fails and recovers, clients need to reconnect in order to
3037 * reestablish shared state. This includes all caps issued through
3038 * this session _and_ the snap_realm hierarchy. Because it's not
3039 * clear which snap realms the mds cares about, we send everything we
3040 * know about.. that ensures we'll then get any new info the
3041 * recovering MDS might have.
3042 *
3043 * This is a relatively heavyweight operation, but it's rare.
3044 *
3045 * called with mdsc->mutex held.
3046 */
34b6c855
SW		 3047static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3048 struct ceph_mds_session *session)
2f2dc053 3049{
2f2dc053 3050 struct ceph_msg *reply;
a105f00c 3051 struct rb_node *p;
34b6c855 3052 int mds = session->s_mds;
9abf82b8 3053 int err = -ENOMEM;
44c99757 3054 int s_nr_caps;
93cea5be 3055 struct ceph_pagelist *pagelist;
20cb34ae 3056 struct ceph_reconnect_state recon_state;
c8a96a31 3057 LIST_HEAD(dispose);
2f2dc053 3058
34b6c855 3059 pr_info("mds%d reconnect start\n", mds);
2f2dc053 3060
93cea5be
SW		 3061 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
3062 if (!pagelist)
3063 goto fail_nopagelist;
3064 ceph_pagelist_init(pagelist);
3065
b61c2763 3066 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 3067 if (!reply)
93cea5be 3068 goto fail_nomsg;
93cea5be 3069
34b6c855
SW		 3070 mutex_lock(&session->s_mutex);
3071 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3072 session->s_seq = 0;
2f2dc053 3073
2f2dc053 3074 dout("session %p state %s\n", session,
a687ecaf 3075 ceph_session_state_name(session->s_state));
2f2dc053 3076
99a9c273
YZ		 3077 spin_lock(&session->s_gen_ttl_lock);
3078 session->s_cap_gen++;
3079 spin_unlock(&session->s_gen_ttl_lock);
3080
3081 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 3082 /* don't know if session is readonly */
3083 session->s_readonly = 0;
99a9c273
YZ		 3084 /*
3085 * notify __ceph_remove_cap() that we are composing cap reconnect.
3086 * If a cap get released before being added to the cap reconnect,
3087 * __ceph_remove_cap() should skip queuing cap release.
3088 */
3089 session->s_cap_reconnect = 1;
e01a5946 3090 /* drop old cap expires; we're about to reestablish that state */
c8a96a31
JL		 3091 detach_cap_releases(session, &dispose);
3092 spin_unlock(&session->s_cap_lock);
3093 dispose_cap_releases(mdsc, &dispose);
e01a5946 3094
5d23371f 3095 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 3096 if (mdsc->fsc->sb->s_root)
3097 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 3098
3099 ceph_con_close(&session->s_con);
3100 ceph_con_open(&session->s_con,
3101 CEPH_ENTITY_TYPE_MDS, mds,
3102 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3103
3104 /* replay unsafe requests */
3105 replay_unsafe_requests(mdsc, session);
3106
3107 down_read(&mdsc->snap_rwsem);
3108
2f2dc053 3109 /* traverse this session's caps */
44c99757
YZ		 3110 s_nr_caps = session->s_nr_caps;
3111 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 3112 if (err)
3113 goto fail;
20cb34ae 3114
44c99757 3115 recon_state.nr_caps = 0;
20cb34ae 3116 recon_state.pagelist = pagelist;
121f22a1
YZ		 3117 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
3118 recon_state.msg_version = 3;
3119 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
3120 recon_state.msg_version = 2;
3121 else
3122 recon_state.msg_version = 1;
20cb34ae 3123 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 3124 if (err < 0)
9abf82b8 3125 goto fail;
2f2dc053 3126
99a9c273
YZ		 3127 spin_lock(&session->s_cap_lock);
3128 session->s_cap_reconnect = 0;
3129 spin_unlock(&session->s_cap_lock);
3130
2f2dc053
SW		 3131 /*
3132 * snaprealms. we provide mds with the ino, seq (version), and
3133 * parent for all of our realms. If the mds has any newer info,
3134 * it will tell us.
3135 */
a105f00c
SW		 3136 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3137 struct ceph_snap_realm *realm =
3138 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3139 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3140
3141 dout(" adding snap realm %llx seq %lld parent %llx\n",
3142 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3143 sr_rec.ino = cpu_to_le64(realm->ino);
3144 sr_rec.seq = cpu_to_le64(realm->seq);
3145 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3146 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3147 if (err)
3148 goto fail;
2f2dc053 3149 }
2f2dc053 3150
121f22a1 3151 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3152
3153 /* raced with cap release? */
3154 if (s_nr_caps != recon_state.nr_caps) {
3155 struct page *page = list_first_entry(&pagelist->head,
3156 struct page, lru);
3157 __le32 *addr = kmap_atomic(page);
3158 *addr = cpu_to_le32(recon_state.nr_caps);
3159 kunmap_atomic(addr);
ebf18f47 3160 }
44c99757
YZ		 3161
3162 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3163 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3164
3165 ceph_early_kick_flushing_caps(mdsc, session);
3166
2f2dc053
SW		 3167 ceph_con_send(&session->s_con, reply);
3168
9abf82b8
SW		 3169 mutex_unlock(&session->s_mutex);
3170
3171 mutex_lock(&mdsc->mutex);
3172 __wake_requests(mdsc, &session->s_waiting);
3173 mutex_unlock(&mdsc->mutex);
3174
2f2dc053 3175 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3176 return;
3177
93cea5be 3178fail:
2f2dc053 3179 ceph_msg_put(reply);
9abf82b8
SW		 3180 up_read(&mdsc->snap_rwsem);
3181 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3182fail_nomsg:
3183 ceph_pagelist_release(pagelist);
93cea5be 3184fail_nopagelist:
9abf82b8 3185 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3186 return;
2f2dc053
SW		 3187}
3188
3189
3190/*
3191 * compare old and new mdsmaps, kicking requests
3192 * and closing out old connections as necessary
3193 *
3194 * called under mdsc->mutex.
3195 */
3196static void check_new_map(struct ceph_mds_client *mdsc,
3197 struct ceph_mdsmap *newmap,
3198 struct ceph_mdsmap *oldmap)
3199{
3200 int i;
3201 int oldstate, newstate;
3202 struct ceph_mds_session *s;
3203
3204 dout("check_new_map new %u old %u\n",
3205 newmap->m_epoch, oldmap->m_epoch);
3206
76201b63 3207 for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
d37b1d99 3208 if (!mdsc->sessions[i])
2f2dc053
SW		 3209 continue;
3210 s = mdsc->sessions[i];
3211 oldstate = ceph_mdsmap_get_state(oldmap, i);
3212 newstate = ceph_mdsmap_get_state(newmap, i);
3213
0deb01c9 3214 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3215 i, ceph_mds_state_name(oldstate),
0deb01c9 3216 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3217 ceph_mds_state_name(newstate),
0deb01c9 3218 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3219 ceph_session_state_name(s->s_state));
2f2dc053 3220
76201b63 3221 if (i >= newmap->m_num_mds ||
3e8f43a0 3222 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3223 ceph_mdsmap_get_addr(newmap, i),
3224 sizeof(struct ceph_entity_addr))) {
3225 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3226 /* the session never opened, just close it
3227 * out now */
0a07fc8c 3228 get_session(s);
2600d2dd 3229 __unregister_session(mdsc, s);
2f2dc053 3230 __wake_requests(mdsc, &s->s_waiting);
0a07fc8c 3231 ceph_put_mds_session(s);
2827528d
YZ		 3232 } else if (i >= newmap->m_num_mds) {
3233 /* force close session for stopped mds */
3234 get_session(s);
2600d2dd 3235 __unregister_session(mdsc, s);
2827528d
YZ		 3236 __wake_requests(mdsc, &s->s_waiting);
3237 kick_requests(mdsc, i);
3238 mutex_unlock(&mdsc->mutex);
3239
3240 mutex_lock(&s->s_mutex);
3241 cleanup_session_requests(mdsc, s);
3242 remove_session_caps(s);
3243 mutex_unlock(&s->s_mutex);
3244
3245 ceph_put_mds_session(s);
3246
3247 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3248 } else {
3249 /* just close it */
3250 mutex_unlock(&mdsc->mutex);
3251 mutex_lock(&s->s_mutex);
3252 mutex_lock(&mdsc->mutex);
3253 ceph_con_close(&s->s_con);
3254 mutex_unlock(&s->s_mutex);
3255 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3256 }
2f2dc053
SW		 3257 } else if (oldstate == newstate) {
3258 continue; /* nothing new with this mds */
3259 }
3260
3261 /*
3262 * send reconnect?
3263 */
3264 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3265 newstate >= CEPH_MDS_STATE_RECONNECT) {
3266 mutex_unlock(&mdsc->mutex);
3267 send_mds_reconnect(mdsc, s);
3268 mutex_lock(&mdsc->mutex);
3269 }
2f2dc053
SW		 3270
3271 /*
29790f26 3272 * kick request on any mds that has gone active.
2f2dc053
SW		 3273 */
3274 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3275 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3276 if (oldstate != CEPH_MDS_STATE_CREATING &&
3277 oldstate != CEPH_MDS_STATE_STARTING)
3278 pr_info("mds%d recovery completed\n", s->s_mds);
3279 kick_requests(mdsc, i);
2f2dc053 3280 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3281 wake_up_session_caps(s, 1);
2f2dc053
SW		 3282 }
3283 }
cb170a22 3284
76201b63 3285 for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
cb170a22
SW		 3286 s = mdsc->sessions[i];
3287 if (!s)
3288 continue;
3289 if (!ceph_mdsmap_is_laggy(newmap, i))
3290 continue;
3291 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3292 s->s_state == CEPH_MDS_SESSION_HUNG ||
3293 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3294 dout(" connecting to export targets of laggy mds%d\n",
3295 i);
3296 __open_export_target_sessions(mdsc, s);
3297 }
3298 }
2f2dc053
SW		 3299}
3300
3301
3302
3303/*
3304 * leases
3305 */
3306
3307/*
3308 * caller must hold session s_mutex, dentry->d_lock
3309 */
3310void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3311{
3312 struct ceph_dentry_info *di = ceph_dentry(dentry);
3313
3314 ceph_put_mds_session(di->lease_session);
3315 di->lease_session = NULL;
3316}
3317
2600d2dd
SW		 3318static void handle_lease(struct ceph_mds_client *mdsc,
3319 struct ceph_mds_session *session,
3320 struct ceph_msg *msg)
2f2dc053 3321{
3d14c5d2 3322 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3323 struct inode *inode;
2f2dc053
SW		 3324 struct dentry *parent, *dentry;
3325 struct ceph_dentry_info *di;
2600d2dd 3326 int mds = session->s_mds;
2f2dc053 3327 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3328 u32 seq;
2f2dc053 3329 struct ceph_vino vino;
2f2dc053
SW		 3330 struct qstr dname;
3331 int release = 0;
3332
2f2dc053
SW		 3333 dout("handle_lease from mds%d\n", mds);
3334
3335 /* decode */
3336 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3337 goto bad;
3338 vino.ino = le64_to_cpu(h->ino);
3339 vino.snap = CEPH_NOSNAP;
1e5ea23d 3340 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3341 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3342 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3343 if (dname.len != get_unaligned_le32(h+1))
3344 goto bad;
3345
2f2dc053
SW		 3346 /* lookup inode */
3347 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3348 dout("handle_lease %s, ino %llx %p %.*s\n",
3349 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3350 dname.len, dname.name);
6cd3bcad
YZ		 3351
3352 mutex_lock(&session->s_mutex);
3353 session->s_seq++;
3354
d37b1d99 3355 if (!inode) {
2f2dc053
SW		 3356 dout("handle_lease no inode %llx\n", vino.ino);
3357 goto release;
3358 }
2f2dc053
SW		 3359
3360 /* dentry */
3361 parent = d_find_alias(inode);
3362 if (!parent) {
3363 dout("no parent dentry on inode %p\n", inode);
3364 WARN_ON(1);
3365 goto release; /* hrm... */
3366 }
8387ff25 3367 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3368 dentry = d_lookup(parent, &dname);
3369 dput(parent);
3370 if (!dentry)
3371 goto release;
3372
3373 spin_lock(&dentry->d_lock);
3374 di = ceph_dentry(dentry);
3375 switch (h->action) {
3376 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3377 if (di->lease_session == session) {
1e5ea23d
SW		 3378 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3379 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3380 __ceph_mdsc_drop_dentry_lease(dentry);
3381 }
3382 release = 1;
3383 break;
3384
3385 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3386 if (di->lease_session == session &&
2f2dc053
SW		 3387 di->lease_gen == session->s_cap_gen &&
3388 di->lease_renew_from &&
3389 di->lease_renew_after == 0) {
3390 unsigned long duration =
3563dbdd 3391 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3392
1e5ea23d 3393 di->lease_seq = seq;
9b16f03c 3394 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3395 di->lease_renew_after = di->lease_renew_from +
3396 (duration >> 1);
3397 di->lease_renew_from = 0;
3398 }
3399 break;
3400 }
3401 spin_unlock(&dentry->d_lock);
3402 dput(dentry);
3403
3404 if (!release)
3405 goto out;
3406
3407release:
3408 /* let's just reuse the same message */
3409 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3410 ceph_msg_get(msg);
3411 ceph_con_send(&session->s_con, msg);
3412
3413out:
3414 iput(inode);
3415 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3416 return;
3417
3418bad:
3419 pr_err("corrupt lease message\n");
9ec7cab1 3420 ceph_msg_dump(msg);
2f2dc053
SW		 3421}
3422
3423void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3424 struct inode *inode,
3425 struct dentry *dentry, char action,
3426 u32 seq)
3427{
3428 struct ceph_msg *msg;
3429 struct ceph_mds_lease *lease;
3430 int len = sizeof(*lease) + sizeof(u32);
3431 int dnamelen = 0;
3432
3433 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3434 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3435 dnamelen = dentry->d_name.len;
3436 len += dnamelen;
3437
b61c2763 3438 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3439 if (!msg)
2f2dc053
SW		 3440 return;
3441 lease = msg->front.iov_base;
3442 lease->action = action;
2f2dc053
SW		 3443 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3444 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3445 lease->seq = cpu_to_le32(seq);
3446 put_unaligned_le32(dnamelen, lease + 1);
3447 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3448
3449 /*
3450 * if this is a preemptive lease RELEASE, no need to
3451 * flush request stream, since the actual request will
3452 * soon follow.
3453 */
3454 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3455
3456 ceph_con_send(&session->s_con, msg);
3457}
3458
2f2dc053
SW		 3459/*
3460 * drop all leases (and dentry refs) in preparation for umount
3461 */
3462static void drop_leases(struct ceph_mds_client *mdsc)
3463{
3464 int i;
3465
3466 dout("drop_leases\n");
3467 mutex_lock(&mdsc->mutex);
3468 for (i = 0; i < mdsc->max_sessions; i++) {
3469 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3470 if (!s)
3471 continue;
3472 mutex_unlock(&mdsc->mutex);
3473 mutex_lock(&s->s_mutex);
3474 mutex_unlock(&s->s_mutex);
3475 ceph_put_mds_session(s);
3476 mutex_lock(&mdsc->mutex);
3477 }
3478 mutex_unlock(&mdsc->mutex);
3479}
3480
3481
3482
3483/*
3484 * delayed work -- periodically trim expired leases, renew caps with mds
3485 */
3486static void schedule_delayed(struct ceph_mds_client *mdsc)
3487{
3488 int delay = 5;
3489 unsigned hz = round_jiffies_relative(HZ * delay);
3490 schedule_delayed_work(&mdsc->delayed_work, hz);
3491}
3492
3493static void delayed_work(struct work_struct *work)
3494{
3495 int i;
3496 struct ceph_mds_client *mdsc =
3497 container_of(work, struct ceph_mds_client, delayed_work.work);
3498 int renew_interval;
3499 int renew_caps;
3500
3501 dout("mdsc delayed_work\n");
afcdaea3 3502 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3503
3504 mutex_lock(&mdsc->mutex);
3505 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3506 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3507 mdsc->last_renew_caps);
3508 if (renew_caps)
3509 mdsc->last_renew_caps = jiffies;
3510
3511 for (i = 0; i < mdsc->max_sessions; i++) {
3512 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
d37b1d99 3513 if (!s)
2f2dc053
SW		 3514 continue;
3515 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3516 dout("resending session close request for mds%d\n",
3517 s->s_mds);
3518 request_close_session(mdsc, s);
3519 ceph_put_mds_session(s);
3520 continue;
3521 }
3522 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3523 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3524 s->s_state = CEPH_MDS_SESSION_HUNG;
3525 pr_info("mds%d hung\n", s->s_mds);
3526 }
3527 }
3528 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3529 /* this mds is failed or recovering, just wait */
3530 ceph_put_mds_session(s);
3531 continue;
3532 }
3533 mutex_unlock(&mdsc->mutex);
3534
3535 mutex_lock(&s->s_mutex);
3536 if (renew_caps)
3537 send_renew_caps(mdsc, s);
3538 else
3539 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3540 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3541 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3542 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3543 mutex_unlock(&s->s_mutex);
3544 ceph_put_mds_session(s);
3545
3546 mutex_lock(&mdsc->mutex);
3547 }
3548 mutex_unlock(&mdsc->mutex);
3549
3550 schedule_delayed(mdsc);
3551}
3552
3d14c5d2 3553int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3554
2f2dc053 3555{
3d14c5d2
YS		 3556 struct ceph_mds_client *mdsc;
3557
3558 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3559 if (!mdsc)
3560 return -ENOMEM;
3561 mdsc->fsc = fsc;
3562 fsc->mdsc = mdsc;
2f2dc053
SW		 3563 mutex_init(&mdsc->mutex);
3564 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
d37b1d99 3565 if (!mdsc->mdsmap) {
fb3101b6 3566 kfree(mdsc);
2d06eeb8 3567 return -ENOMEM;
fb3101b6 3568 }
2d06eeb8 3569
2f2dc053 3570 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3571 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3572 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3573 mdsc->sessions = NULL;
86d8f67b 3574 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3575 mdsc->max_sessions = 0;
3576 mdsc->stopping = 0;
affbc19a 3577 mdsc->last_snap_seq = 0;
2f2dc053 3578 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3579 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3580 INIT_LIST_HEAD(&mdsc->snap_empty);
3581 spin_lock_init(&mdsc->snap_empty_lock);
3582 mdsc->last_tid = 0;
e8a7b8b1 3583 mdsc->oldest_tid = 0;
44ca18f2 3584 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3585 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3586 mdsc->last_renew_caps = jiffies;
3587 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3588 spin_lock_init(&mdsc->cap_delay_lock);
3589 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3590 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3591 mdsc->last_cap_flush_tid = 1;
e4500b5e 3592 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3593 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3594 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3595 mdsc->num_cap_flushing = 0;
3596 spin_lock_init(&mdsc->cap_dirty_lock);
3597 init_waitqueue_head(&mdsc->cap_flushing_wq);
3598 spin_lock_init(&mdsc->dentry_lru_lock);
3599 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3600
37151668 3601 ceph_caps_init(mdsc);
3d14c5d2 3602 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3603
10183a69
YZ		 3604 init_rwsem(&mdsc->pool_perm_rwsem);
3605 mdsc->pool_perm_tree = RB_ROOT;
3606
717e6f28
YZ		 3607 strncpy(mdsc->nodename, utsname()->nodename,
3608 sizeof(mdsc->nodename) - 1);
5f44f142 3609 return 0;
2f2dc053
SW		 3610}
3611
3612/*
3613 * Wait for safe replies on open mds requests. If we time out, drop
3614 * all requests from the tree to avoid dangling dentry refs.
3615 */
3616static void wait_requests(struct ceph_mds_client *mdsc)
3617{
a319bf56 3618 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3619 struct ceph_mds_request *req;
2f2dc053
SW		 3620
3621 mutex_lock(&mdsc->mutex);
44ca18f2 3622 if (__get_oldest_req(mdsc)) {
2f2dc053 3623 mutex_unlock(&mdsc->mutex);
44ca18f2 3624
2f2dc053
SW		 3625 dout("wait_requests waiting for requests\n");
3626 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3627 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3628
3629 /* tear down remaining requests */
44ca18f2
SW		 3630 mutex_lock(&mdsc->mutex);
3631 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3632 dout("wait_requests timed out on tid %llu\n",
3633 req->r_tid);
44ca18f2 3634 __unregister_request(mdsc, req);
2f2dc053
SW		 3635 }
3636 }
3637 mutex_unlock(&mdsc->mutex);
3638 dout("wait_requests done\n");
3639}
3640
3641/*
3642 * called before mount is ro, and before dentries are torn down.
3643 * (hmm, does this still race with new lookups?)
3644 */
3645void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3646{
3647 dout("pre_umount\n");
3648 mdsc->stopping = 1;
3649
3650 drop_leases(mdsc);
afcdaea3 3651 ceph_flush_dirty_caps(mdsc);
2f2dc053 3652 wait_requests(mdsc);
17c688c3
SW		 3653
3654 /*
3655 * wait for reply handlers to drop their request refs and
3656 * their inode/dcache refs
3657 */
3658 ceph_msgr_flush();
2f2dc053
SW		 3659}
3660
3661/*
3662 * wait for all write mds requests to flush.
3663 */
3664static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3665{
80fc7314 3666 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3667 struct rb_node *n;
2f2dc053
SW		 3668
3669 mutex_lock(&mdsc->mutex);
3670 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3671restart:
44ca18f2
SW		 3672 req = __get_oldest_req(mdsc);
3673 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3674 /* find next request */
3675 n = rb_next(&req->r_node);
3676 if (n)
3677 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3678 else
3679 nextreq = NULL;
e8a7b8b1
YZ		 3680 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3681 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3682 /* write op */
3683 ceph_mdsc_get_request(req);
80fc7314
SW		 3684 if (nextreq)
3685 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3686 mutex_unlock(&mdsc->mutex);
3687 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3688 req->r_tid, want_tid);
3689 wait_for_completion(&req->r_safe_completion);
3690 mutex_lock(&mdsc->mutex);
44ca18f2 3691 ceph_mdsc_put_request(req);
80fc7314
SW		 3692 if (!nextreq)
3693 break; /* next dne before, so we're done! */
3694 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3695 /* next request was removed from tree */
3696 ceph_mdsc_put_request(nextreq);
3697 goto restart;
3698 }
3699 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3700 }
80fc7314 3701 req = nextreq;
2f2dc053
SW		 3702 }
3703 mutex_unlock(&mdsc->mutex);
3704 dout("wait_unsafe_requests done\n");
3705}
3706
3707void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3708{
0e294387 3709 u64 want_tid, want_flush;
2f2dc053 3710
52953d55 3711 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3712 return;
3713
2f2dc053
SW		 3714 dout("sync\n");
3715 mutex_lock(&mdsc->mutex);
3716 want_tid = mdsc->last_tid;
2f2dc053 3717 mutex_unlock(&mdsc->mutex);
2f2dc053 3718
afcdaea3 3719 ceph_flush_dirty_caps(mdsc);
d3383a8e 3720 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3721 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3722 if (!list_empty(&mdsc->cap_flush_list)) {
3723 struct ceph_cap_flush *cf =
3724 list_last_entry(&mdsc->cap_flush_list,
3725 struct ceph_cap_flush, g_list);
3726 cf->wake = true;
3727 }
d3383a8e
YZ		 3728 spin_unlock(&mdsc->cap_dirty_lock);
3729
0e294387
YZ		 3730 dout("sync want tid %lld flush_seq %lld\n",
3731 want_tid, want_flush);
2f2dc053
SW		 3732
3733 wait_unsafe_requests(mdsc, want_tid);
0e294387 3734 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3735}
3736
f3c60c59
SW		 3737/*
3738 * true if all sessions are closed, or we force unmount
3739 */
fcff415c 3740static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3741{
52953d55 3742 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3743 return true;
fcff415c 3744 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3745}
2f2dc053
SW		 3746
3747/*
3748 * called after sb is ro.
3749 */
3750void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3751{
a319bf56 3752 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3753 struct ceph_mds_session *session;
3754 int i;
fcff415c 3755 int skipped = 0;
2f2dc053
SW		 3756
3757 dout("close_sessions\n");
3758
2f2dc053 3759 /* close sessions */
f3c60c59
SW		 3760 mutex_lock(&mdsc->mutex);
3761 for (i = 0; i < mdsc->max_sessions; i++) {
3762 session = __ceph_lookup_mds_session(mdsc, i);
3763 if (!session)
3764 continue;
2f2dc053 3765 mutex_unlock(&mdsc->mutex);
f3c60c59 3766 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3767 if (__close_session(mdsc, session) <= 0)
3768 skipped++;
f3c60c59
SW		 3769 mutex_unlock(&session->s_mutex);
3770 ceph_put_mds_session(session);
2f2dc053
SW		 3771 mutex_lock(&mdsc->mutex);
3772 }
f3c60c59
SW		 3773 mutex_unlock(&mdsc->mutex);
3774
3775 dout("waiting for sessions to close\n");
fcff415c
YZ		 3776 wait_event_timeout(mdsc->session_close_wq,
3777 done_closing_sessions(mdsc, skipped),
a319bf56 3778 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3779
3780 /* tear down remaining sessions */
f3c60c59 3781 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3782 for (i = 0; i < mdsc->max_sessions; i++) {
3783 if (mdsc->sessions[i]) {
3784 session = get_session(mdsc->sessions[i]);
2600d2dd 3785 __unregister_session(mdsc, session);
2f2dc053
SW		 3786 mutex_unlock(&mdsc->mutex);
3787 mutex_lock(&session->s_mutex);
3788 remove_session_caps(session);
3789 mutex_unlock(&session->s_mutex);
3790 ceph_put_mds_session(session);
3791 mutex_lock(&mdsc->mutex);
3792 }
3793 }
2f2dc053 3794 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3795 mutex_unlock(&mdsc->mutex);
3796
3797 ceph_cleanup_empty_realms(mdsc);
3798
3799 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3800
3801 dout("stopped\n");
3802}
3803
48fec5d0
YZ		 3804void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3805{
3806 struct ceph_mds_session *session;
3807 int mds;
3808
3809 dout("force umount\n");
3810
3811 mutex_lock(&mdsc->mutex);
3812 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3813 session = __ceph_lookup_mds_session(mdsc, mds);
3814 if (!session)
3815 continue;
3816 mutex_unlock(&mdsc->mutex);
3817 mutex_lock(&session->s_mutex);
3818 __close_session(mdsc, session);
3819 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3820 cleanup_session_requests(mdsc, session);
3821 remove_session_caps(session);
3822 }
3823 mutex_unlock(&session->s_mutex);
3824 ceph_put_mds_session(session);
3825 mutex_lock(&mdsc->mutex);
3826 kick_requests(mdsc, mds);
3827 }
3828 __wake_requests(mdsc, &mdsc->waiting_for_map);
3829 mutex_unlock(&mdsc->mutex);
3830}
3831
3d14c5d2 3832static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3833{
3834 dout("stop\n");
3835 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3836 if (mdsc->mdsmap)
3837 ceph_mdsmap_destroy(mdsc->mdsmap);
3838 kfree(mdsc->sessions);
37151668 3839 ceph_caps_finalize(mdsc);
10183a69 3840 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3841}
3842
3d14c5d2
YS		 3843void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3844{
3845 struct ceph_mds_client *mdsc = fsc->mdsc;
ef550f6f 3846 dout("mdsc_destroy %p\n", mdsc);
ef550f6f
SW		 3847
3848 /* flush out any connection work with references to us */
3849 ceph_msgr_flush();
3850
62a65f36
YZ		 3851 ceph_mdsc_stop(mdsc);
3852
3d14c5d2
YS		 3853 fsc->mdsc = NULL;
3854 kfree(mdsc);
ef550f6f 3855 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3856}
3857
430afbad
YZ		 3858void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3859{
3860 struct ceph_fs_client *fsc = mdsc->fsc;
3861 const char *mds_namespace = fsc->mount_options->mds_namespace;
3862 void *p = msg->front.iov_base;
3863 void *end = p + msg->front.iov_len;
3864 u32 epoch;
3865 u32 map_len;
3866 u32 num_fs;
3867 u32 mount_fscid = (u32)-1;
3868 u8 struct_v, struct_cv;
3869 int err = -EINVAL;
3870
3871 ceph_decode_need(&p, end, sizeof(u32), bad);
3872 epoch = ceph_decode_32(&p);
3873
3874 dout("handle_fsmap epoch %u\n", epoch);
3875
3876 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3877 struct_v = ceph_decode_8(&p);
3878 struct_cv = ceph_decode_8(&p);
3879 map_len = ceph_decode_32(&p);
3880
3881 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3882 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3883
3884 num_fs = ceph_decode_32(&p);
3885 while (num_fs-- > 0) {
3886 void *info_p, *info_end;
3887 u32 info_len;
3888 u8 info_v, info_cv;
3889 u32 fscid, namelen;
3890
3891 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3892 info_v = ceph_decode_8(&p);
3893 info_cv = ceph_decode_8(&p);
3894 info_len = ceph_decode_32(&p);
3895 ceph_decode_need(&p, end, info_len, bad);
3896 info_p = p;
3897 info_end = p + info_len;
3898 p = info_end;
3899
3900 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3901 fscid = ceph_decode_32(&info_p);
3902 namelen = ceph_decode_32(&info_p);
3903 ceph_decode_need(&info_p, info_end, namelen, bad);
3904
3905 if (mds_namespace &&
3906 strlen(mds_namespace) == namelen &&
3907 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3908 mount_fscid = fscid;
3909 break;
3910 }
3911 }
3912
3913 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3914 if (mount_fscid != (u32)-1) {
3915 fsc->client->monc.fs_cluster_id = mount_fscid;
3916 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
3917 0, true);
3918 ceph_monc_renew_subs(&fsc->client->monc);
3919 } else {
3920 err = -ENOENT;
3921 goto err_out;
3922 }
3923 return;
76bd6ec4 3924
430afbad
YZ		 3925bad:
3926 pr_err("error decoding fsmap\n");
3927err_out:
3928 mutex_lock(&mdsc->mutex);
76bd6ec4 3929 mdsc->mdsmap_err = err;
430afbad
YZ		 3930 __wake_requests(mdsc, &mdsc->waiting_for_map);
3931 mutex_unlock(&mdsc->mutex);
430afbad 3932}
2f2dc053
SW		 3933
3934/*
3935 * handle mds map update.
3936 */
430afbad 3937void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053
SW		 3938{
3939 u32 epoch;
3940 u32 maplen;
3941 void *p = msg->front.iov_base;
3942 void *end = p + msg->front.iov_len;
3943 struct ceph_mdsmap *newmap, *oldmap;
3944 struct ceph_fsid fsid;
3945 int err = -EINVAL;
3946
3947 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3948 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3949 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3950 return;
c89136ea
SW		 3951 epoch = ceph_decode_32(&p);
3952 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3953 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3954
3955 /* do we need it? */
2f2dc053
SW		 3956 mutex_lock(&mdsc->mutex);
3957 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3958 dout("handle_map epoch %u <= our %u\n",
3959 epoch, mdsc->mdsmap->m_epoch);
3960 mutex_unlock(&mdsc->mutex);
3961 return;
3962 }
3963
3964 newmap = ceph_mdsmap_decode(&p, end);
3965 if (IS_ERR(newmap)) {
3966 err = PTR_ERR(newmap);
3967 goto bad_unlock;
3968 }
3969
3970 /* swap into place */
3971 if (mdsc->mdsmap) {
3972 oldmap = mdsc->mdsmap;
3973 mdsc->mdsmap = newmap;
3974 check_new_map(mdsc, newmap, oldmap);
3975 ceph_mdsmap_destroy(oldmap);
3976 } else {
3977 mdsc->mdsmap = newmap; /* first mds map */
3978 }
3d14c5d2 3979 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3980
3981 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 3982 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
3983 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 3984
3985 mutex_unlock(&mdsc->mutex);
3986 schedule_delayed(mdsc);
3987 return;
3988
3989bad_unlock:
3990 mutex_unlock(&mdsc->mutex);
3991bad:
3992 pr_err("error decoding mdsmap %d\n", err);
3993 return;
3994}
3995
3996static struct ceph_connection *con_get(struct ceph_connection *con)
3997{
3998 struct ceph_mds_session *s = con->private;
3999
4000 if (get_session(s)) {
3997c01d 4001 dout("mdsc con_get %p ok (%d)\n", s, refcount_read(&s->s_ref));
2f2dc053
SW		 4002 return con;
4003 }
4004 dout("mdsc con_get %p FAIL\n", s);
4005 return NULL;
4006}
4007
4008static void con_put(struct ceph_connection *con)
4009{
4010 struct ceph_mds_session *s = con->private;
4011
3997c01d 4012 dout("mdsc con_put %p (%d)\n", s, refcount_read(&s->s_ref) - 1);
2f2dc053
SW		 4013 ceph_put_mds_session(s);
4014}
4015
4016/*
4017 * if the client is unresponsive for long enough, the mds will kill
4018 * the session entirely.
4019 */
4020static void peer_reset(struct ceph_connection *con)
4021{
4022 struct ceph_mds_session *s = con->private;
7e70f0ed 4023 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 4024
f3ae1b97 4025 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 4026 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 4027}
4028
4029static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4030{
4031 struct ceph_mds_session *s = con->private;
4032 struct ceph_mds_client *mdsc = s->s_mdsc;
4033 int type = le16_to_cpu(msg->hdr.type);
4034
2600d2dd
SW		 4035 mutex_lock(&mdsc->mutex);
4036 if (__verify_registered_session(mdsc, s) < 0) {
4037 mutex_unlock(&mdsc->mutex);
4038 goto out;
4039 }
4040 mutex_unlock(&mdsc->mutex);
4041
2f2dc053
SW		 4042 switch (type) {
4043 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 4044 ceph_mdsc_handle_mdsmap(mdsc, msg);
4045 break;
4046 case CEPH_MSG_FS_MAP_USER:
4047 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 4048 break;
4049 case CEPH_MSG_CLIENT_SESSION:
4050 handle_session(s, msg);
4051 break;
4052 case CEPH_MSG_CLIENT_REPLY:
4053 handle_reply(s, msg);
4054 break;
4055 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 4056 handle_forward(mdsc, s, msg);
2f2dc053
SW		 4057 break;
4058 case CEPH_MSG_CLIENT_CAPS:
4059 ceph_handle_caps(s, msg);
4060 break;
4061 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 4062 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 4063 break;
4064 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 4065 handle_lease(mdsc, s, msg);
2f2dc053
SW		 4066 break;
4067
4068 default:
4069 pr_err("received unknown message type %d %s\n", type,
4070 ceph_msg_type_name(type));
4071 }
2600d2dd 4072out:
2f2dc053
SW		 4073 ceph_msg_put(msg);
4074}
4075
4e7a5dcd
SW		 4076/*
4077 * authentication
4078 */
a3530df3
AE		 4079
4080/*
4081 * Note: returned pointer is the address of a structure that's
4082 * managed separately. Caller must *not* attempt to free it.
4083 */
4084static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 4085 int *proto, int force_new)
4e7a5dcd
SW		 4086{
4087 struct ceph_mds_session *s = con->private;
4088 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4089 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 4090 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 4091
74f1869f 4092 if (force_new && auth->authorizer) {
6c1ea260 4093 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 4094 auth->authorizer = NULL;
4e7a5dcd 4095 }
27859f97
SW		 4096 if (!auth->authorizer) {
4097 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4098 auth);
0bed9b5c
SW		 4099 if (ret)
4100 return ERR_PTR(ret);
27859f97
SW		 4101 } else {
4102 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4103 auth);
a255651d 4104 if (ret)
a3530df3 4105 return ERR_PTR(ret);
4e7a5dcd 4106 }
4e7a5dcd 4107 *proto = ac->protocol;
74f1869f 4108
a3530df3 4109 return auth;
4e7a5dcd
SW		 4110}
4111
4112
0dde5848 4113static int verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 4114{
4115 struct ceph_mds_session *s = con->private;
4116 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4117 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 4118
0dde5848 4119 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4e7a5dcd
SW		 4120}
4121
9bd2e6f8
SW		 4122static int invalidate_authorizer(struct ceph_connection *con)
4123{
4124 struct ceph_mds_session *s = con->private;
4125 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4126 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 4127
27859f97 4128 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 4129
3d14c5d2 4130 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 4131}
4132
53ded495
AE		 4133static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4134 struct ceph_msg_header *hdr, int *skip)
4135{
4136 struct ceph_msg *msg;
4137 int type = (int) le16_to_cpu(hdr->type);
4138 int front_len = (int) le32_to_cpu(hdr->front_len);
4139
4140 if (con->in_msg)
4141 return con->in_msg;
4142
4143 *skip = 0;
4144 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4145 if (!msg) {
4146 pr_err("unable to allocate msg type %d len %d\n",
4147 type, front_len);
4148 return NULL;
4149 }
53ded495
AE		 4150
4151 return msg;
4152}
4153
79dbd1ba 4154static int mds_sign_message(struct ceph_msg *msg)
33d07337 4155{
79dbd1ba 4156 struct ceph_mds_session *s = msg->con->private;
33d07337 4157 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4158
33d07337
YZ		 4159 return ceph_auth_sign_message(auth, msg);
4160}
4161
79dbd1ba 4162static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 4163{
79dbd1ba 4164 struct ceph_mds_session *s = msg->con->private;
33d07337 4165 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4166
33d07337
YZ		 4167 return ceph_auth_check_message_signature(auth, msg);
4168}
4169
9e32789f 4170static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4171 .get = con_get,
4172 .put = con_put,
4173 .dispatch = dispatch,
4e7a5dcd
SW		 4174 .get_authorizer = get_authorizer,
4175 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4176 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4177 .peer_reset = peer_reset,
53ded495 4178 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4179 .sign_message = mds_sign_message,
4180 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4181};
4182
2f2dc053 4183/* eof */
ubuntu-bionic-kernel mirror