]> 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: send cap releases more aggressively
[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45 * are no longer valid.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);
1de82206 56static void ceph_cap_release_work(struct work_struct *work);
2f2dc053 57
9e32789f 58static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 59
60
61/*
62 * mds reply parsing
63 */
64
65/*
66 * parse individual inode info
67 */
68static int parse_reply_info_in(void **p, void *end,
14303d20 69 struct ceph_mds_reply_info_in *info,
12b4629a 70 u64 features)
2f2dc053
SW		 71{
72 int err = -EIO;
73
74 info->in = *p;
75 *p += sizeof(struct ceph_mds_reply_inode) +
76 sizeof(*info->in->fragtree.splits) *
77 le32_to_cpu(info->in->fragtree.nsplits);
78
79 ceph_decode_32_safe(p, end, info->symlink_len, bad);
80 ceph_decode_need(p, end, info->symlink_len, bad);
81 info->symlink = *p;
82 *p += info->symlink_len;
83
14303d20
SW		 84 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
85 ceph_decode_copy_safe(p, end, &info->dir_layout,
86 sizeof(info->dir_layout), bad);
87 else
88 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
89
2f2dc053
SW		 90 ceph_decode_32_safe(p, end, info->xattr_len, bad);
91 ceph_decode_need(p, end, info->xattr_len, bad);
92 info->xattr_data = *p;
93 *p += info->xattr_len;
fb01d1f8
YZ		 94
95 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
96 ceph_decode_64_safe(p, end, info->inline_version, bad);
97 ceph_decode_32_safe(p, end, info->inline_len, bad);
98 ceph_decode_need(p, end, info->inline_len, bad);
99 info->inline_data = *p;
100 *p += info->inline_len;
101 } else
102 info->inline_version = CEPH_INLINE_NONE;
103
779fe0fb
YZ		 104 info->pool_ns_len = 0;
105 info->pool_ns_data = NULL;
5ea5c5e0
YZ		 106 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
107 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
779fe0fb
YZ		 108 if (info->pool_ns_len > 0) {
109 ceph_decode_need(p, end, info->pool_ns_len, bad);
110 info->pool_ns_data = *p;
111 *p += info->pool_ns_len;
112 }
5ea5c5e0
YZ		 113 }
114
2f2dc053
SW		 115 return 0;
116bad:
117 return err;
118}
119
120/*
121 * parse a normal reply, which may contain a (dir+)dentry and/or a
122 * target inode.
123 */
124static int parse_reply_info_trace(void **p, void *end,
14303d20 125 struct ceph_mds_reply_info_parsed *info,
12b4629a 126 u64 features)
2f2dc053
SW		 127{
128 int err;
129
130 if (info->head->is_dentry) {
14303d20 131 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 132 if (err < 0)
133 goto out_bad;
134
135 if (unlikely(*p + sizeof(*info->dirfrag) > end))
136 goto bad;
137 info->dirfrag = *p;
138 *p += sizeof(*info->dirfrag) +
139 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
140 if (unlikely(*p > end))
141 goto bad;
142
143 ceph_decode_32_safe(p, end, info->dname_len, bad);
144 ceph_decode_need(p, end, info->dname_len, bad);
145 info->dname = *p;
146 *p += info->dname_len;
147 info->dlease = *p;
148 *p += sizeof(*info->dlease);
149 }
150
151 if (info->head->is_target) {
14303d20 152 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 153 if (err < 0)
154 goto out_bad;
155 }
156
157 if (unlikely(*p != end))
158 goto bad;
159 return 0;
160
161bad:
162 err = -EIO;
163out_bad:
164 pr_err("problem parsing mds trace %d\n", err);
165 return err;
166}
167
168/*
169 * parse readdir results
170 */
171static int parse_reply_info_dir(void **p, void *end,
14303d20 172 struct ceph_mds_reply_info_parsed *info,
12b4629a 173 u64 features)
2f2dc053
SW		 174{
175 u32 num, i = 0;
176 int err;
177
178 info->dir_dir = *p;
179 if (*p + sizeof(*info->dir_dir) > end)
180 goto bad;
181 *p += sizeof(*info->dir_dir) +
182 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
183 if (*p > end)
184 goto bad;
185
186 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea 187 num = ceph_decode_32(p);
956d39d6
YZ		 188 {
189 u16 flags = ceph_decode_16(p);
190 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
191 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
f3c4ebe6 192 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
79162547 193 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
956d39d6 194 }
2f2dc053
SW		 195 if (num == 0)
196 goto done;
197
2a5beea3
YZ		 198 BUG_ON(!info->dir_entries);
199 if ((unsigned long)(info->dir_entries + num) >
200 (unsigned long)info->dir_entries + info->dir_buf_size) {
54008399
YZ		 201 pr_err("dir contents are larger than expected\n");
202 WARN_ON(1);
203 goto bad;
204 }
2f2dc053 205
54008399 206 info->dir_nr = num;
2f2dc053 207 while (num) {
2a5beea3 208 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
2f2dc053
SW		 209 /* dentry */
210 ceph_decode_need(p, end, sizeof(u32)*2, bad);
2a5beea3
YZ		 211 rde->name_len = ceph_decode_32(p);
212 ceph_decode_need(p, end, rde->name_len, bad);
213 rde->name = *p;
214 *p += rde->name_len;
215 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
216 rde->lease = *p;
2f2dc053
SW		 217 *p += sizeof(struct ceph_mds_reply_lease);
218
219 /* inode */
2a5beea3 220 err = parse_reply_info_in(p, end, &rde->inode, features);
2f2dc053
SW		 221 if (err < 0)
222 goto out_bad;
8974eebd
YZ		 223 /* ceph_readdir_prepopulate() will update it */
224 rde->offset = 0;
2f2dc053
SW		 225 i++;
226 num--;
227 }
228
229done:
230 if (*p != end)
231 goto bad;
232 return 0;
233
234bad:
235 err = -EIO;
236out_bad:
237 pr_err("problem parsing dir contents %d\n", err);
238 return err;
239}
240
25933abd
HS		 241/*
242 * parse fcntl F_GETLK results
243 */
244static int parse_reply_info_filelock(void **p, void *end,
14303d20 245 struct ceph_mds_reply_info_parsed *info,
12b4629a 246 u64 features)
25933abd
HS		 247{
248 if (*p + sizeof(*info->filelock_reply) > end)
249 goto bad;
250
251 info->filelock_reply = *p;
252 *p += sizeof(*info->filelock_reply);
253
254 if (unlikely(*p != end))
255 goto bad;
256 return 0;
257
258bad:
259 return -EIO;
260}
261
6e8575fa
SL		 262/*
263 * parse create results
264 */
265static int parse_reply_info_create(void **p, void *end,
266 struct ceph_mds_reply_info_parsed *info,
12b4629a 267 u64 features)
6e8575fa
SL		 268{
269 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
270 if (*p == end) {
271 info->has_create_ino = false;
272 } else {
273 info->has_create_ino = true;
274 info->ino = ceph_decode_64(p);
275 }
276 }
277
278 if (unlikely(*p != end))
279 goto bad;
280 return 0;
281
282bad:
283 return -EIO;
284}
285
25933abd
HS		 286/*
287 * parse extra results
288 */
289static int parse_reply_info_extra(void **p, void *end,
14303d20 290 struct ceph_mds_reply_info_parsed *info,
12b4629a 291 u64 features)
25933abd 292{
6df8c9d8
JL		 293 u32 op = le32_to_cpu(info->head->op);
294
295 if (op == CEPH_MDS_OP_GETFILELOCK)
14303d20 296 return parse_reply_info_filelock(p, end, info, features);
6df8c9d8 297 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
14303d20 298 return parse_reply_info_dir(p, end, info, features);
6df8c9d8 299 else if (op == CEPH_MDS_OP_CREATE)
6e8575fa
SL		 300 return parse_reply_info_create(p, end, info, features);
301 else
302 return -EIO;
25933abd
HS		 303}
304
2f2dc053
SW		 305/*
306 * parse entire mds reply
307 */
308static int parse_reply_info(struct ceph_msg *msg,
14303d20 309 struct ceph_mds_reply_info_parsed *info,
12b4629a 310 u64 features)
2f2dc053
SW		 311{
312 void *p, *end;
313 u32 len;
314 int err;
315
316 info->head = msg->front.iov_base;
317 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
318 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
319
320 /* trace */
321 ceph_decode_32_safe(&p, end, len, bad);
322 if (len > 0) {
32852a81 323 ceph_decode_need(&p, end, len, bad);
14303d20 324 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 325 if (err < 0)
326 goto out_bad;
327 }
328
25933abd 329 /* extra */
2f2dc053
SW		 330 ceph_decode_32_safe(&p, end, len, bad);
331 if (len > 0) {
32852a81 332 ceph_decode_need(&p, end, len, bad);
14303d20 333 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 334 if (err < 0)
335 goto out_bad;
336 }
337
338 /* snap blob */
339 ceph_decode_32_safe(&p, end, len, bad);
340 info->snapblob_len = len;
341 info->snapblob = p;
342 p += len;
343
344 if (p != end)
345 goto bad;
346 return 0;
347
348bad:
349 err = -EIO;
350out_bad:
351 pr_err("mds parse_reply err %d\n", err);
352 return err;
353}
354
355static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
356{
2a5beea3 357 if (!info->dir_entries)
54008399 358 return;
2a5beea3 359 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
2f2dc053
SW		 360}
361
362
363/*
364 * sessions
365 */
a687ecaf 366const char *ceph_session_state_name(int s)
2f2dc053
SW		 367{
368 switch (s) {
369 case CEPH_MDS_SESSION_NEW: return "new";
370 case CEPH_MDS_SESSION_OPENING: return "opening";
371 case CEPH_MDS_SESSION_OPEN: return "open";
372 case CEPH_MDS_SESSION_HUNG: return "hung";
373 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 374 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053 375 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
fcff415c 376 case CEPH_MDS_SESSION_REJECTED: return "rejected";
2f2dc053
SW		 377 default: return "???";
378 }
379}
380
381static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
382{
3997c01d 383 if (refcount_inc_not_zero(&s->s_ref)) {
2f2dc053 384 dout("mdsc get_session %p %d -> %d\n", s,
3997c01d 385 refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
2f2dc053
SW		 386 return s;
387 } else {
388 dout("mdsc get_session %p 0 -- FAIL", s);
389 return NULL;
390 }
391}
392
393void ceph_put_mds_session(struct ceph_mds_session *s)
394{
395 dout("mdsc put_session %p %d -> %d\n", s,
3997c01d
ER		 396 refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
397 if (refcount_dec_and_test(&s->s_ref)) {
6c4a1915 398 if (s->s_auth.authorizer)
6c1ea260 399 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
2f2dc053 400 kfree(s);
4e7a5dcd 401 }
2f2dc053
SW		 402}
403
404/*
405 * called under mdsc->mutex
406 */
407struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
408 int mds)
409{
410 struct ceph_mds_session *session;
411
d37b1d99 412 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053
SW		 413 return NULL;
414 session = mdsc->sessions[mds];
415 dout("lookup_mds_session %p %d\n", session,
3997c01d 416 refcount_read(&session->s_ref));
2f2dc053
SW		 417 get_session(session);
418 return session;
419}
420
421static bool __have_session(struct ceph_mds_client *mdsc, int mds)
422{
423 if (mds >= mdsc->max_sessions)
424 return false;
425 return mdsc->sessions[mds];
426}
427
2600d2dd
SW		 428static int __verify_registered_session(struct ceph_mds_client *mdsc,
429 struct ceph_mds_session *s)
430{
431 if (s->s_mds >= mdsc->max_sessions ||
432 mdsc->sessions[s->s_mds] != s)
433 return -ENOENT;
434 return 0;
435}
436
2f2dc053
SW		 437/*
438 * create+register a new session for given mds.
439 * called under mdsc->mutex.
440 */
441static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
442 int mds)
443{
444 struct ceph_mds_session *s;
445
76201b63 446 if (mds >= mdsc->mdsmap->m_num_mds)
c338c07c
NY		 447 return ERR_PTR(-EINVAL);
448
2f2dc053 449 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 450 if (!s)
451 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 452 s->s_mdsc = mdsc;
453 s->s_mds = mds;
454 s->s_state = CEPH_MDS_SESSION_NEW;
455 s->s_ttl = 0;
456 s->s_seq = 0;
457 mutex_init(&s->s_mutex);
458
b7a9e5dd 459 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 460
d8fb02ab 461 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 462 s->s_cap_gen = 0;
1ce208a6 463 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 464
465 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 466 s->s_renew_requested = 0;
467 s->s_renew_seq = 0;
468 INIT_LIST_HEAD(&s->s_caps);
469 s->s_nr_caps = 0;
5dacf091 470 s->s_trim_caps = 0;
3997c01d 471 refcount_set(&s->s_ref, 1);
2f2dc053
SW		 472 INIT_LIST_HEAD(&s->s_waiting);
473 INIT_LIST_HEAD(&s->s_unsafe);
474 s->s_num_cap_releases = 0;
99a9c273 475 s->s_cap_reconnect = 0;
7c1332b8 476 s->s_cap_iterator = NULL;
2f2dc053 477 INIT_LIST_HEAD(&s->s_cap_releases);
1de82206
YZ		 478 INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
479
2f2dc053 480 INIT_LIST_HEAD(&s->s_cap_flushing);
2f2dc053
SW		 481
482 dout("register_session mds%d\n", mds);
483 if (mds >= mdsc->max_sessions) {
484 int newmax = 1 << get_count_order(mds+1);
485 struct ceph_mds_session **sa;
486
487 dout("register_session realloc to %d\n", newmax);
488 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
d37b1d99 489 if (!sa)
42ce56e5 490 goto fail_realloc;
2f2dc053
SW		 491 if (mdsc->sessions) {
492 memcpy(sa, mdsc->sessions,
493 mdsc->max_sessions * sizeof(void *));
494 kfree(mdsc->sessions);
495 }
496 mdsc->sessions = sa;
497 mdsc->max_sessions = newmax;
498 }
499 mdsc->sessions[mds] = s;
86d8f67b 500 atomic_inc(&mdsc->num_sessions);
3997c01d 501 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 502
b7a9e5dd
SW		 503 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
504 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 505
2f2dc053 506 return s;
42ce56e5
SW		 507
508fail_realloc:
509 kfree(s);
510 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 511}
512
513/*
514 * called under mdsc->mutex
515 */
2600d2dd 516static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 517 struct ceph_mds_session *s)
2f2dc053 518{
2600d2dd
SW		 519 dout("__unregister_session mds%d %p\n", s->s_mds, s);
520 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5 521 mdsc->sessions[s->s_mds] = NULL;
1de82206 522 s->s_state = 0;
42ce56e5
SW		 523 ceph_con_close(&s->s_con);
524 ceph_put_mds_session(s);
86d8f67b 525 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 526}
527
528/*
529 * drop session refs in request.
530 *
531 * should be last request ref, or hold mdsc->mutex
532 */
533static void put_request_session(struct ceph_mds_request *req)
534{
535 if (req->r_session) {
536 ceph_put_mds_session(req->r_session);
537 req->r_session = NULL;
538 }
539}
540
153c8e6b 541void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 542{
153c8e6b
SW		 543 struct ceph_mds_request *req = container_of(kref,
544 struct ceph_mds_request,
545 r_kref);
54008399 546 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 547 if (req->r_request)
548 ceph_msg_put(req->r_request);
54008399 549 if (req->r_reply)
153c8e6b 550 ceph_msg_put(req->r_reply);
153c8e6b 551 if (req->r_inode) {
41b02e1f 552 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 553 iput(req->r_inode);
554 }
3dd69aab
JL		 555 if (req->r_parent)
556 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
e96a650a 557 iput(req->r_target_inode);
153c8e6b
SW		 558 if (req->r_dentry)
559 dput(req->r_dentry);
844d87c3
SW		 560 if (req->r_old_dentry)
561 dput(req->r_old_dentry);
562 if (req->r_old_dentry_dir) {
41b02e1f
SW		 563 /*
564 * track (and drop pins for) r_old_dentry_dir
565 * separately, since r_old_dentry's d_parent may have
566 * changed between the dir mutex being dropped and
567 * this request being freed.
568 */
569 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
570 CEPH_CAP_PIN);
41b02e1f 571 iput(req->r_old_dentry_dir);
2f2dc053 572 }
153c8e6b
SW		 573 kfree(req->r_path1);
574 kfree(req->r_path2);
25e6bae3
YZ		 575 if (req->r_pagelist)
576 ceph_pagelist_release(req->r_pagelist);
153c8e6b 577 put_request_session(req);
37151668 578 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 579 kfree(req);
2f2dc053
SW		 580}
581
fcd00b68
ID		 582DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
583
2f2dc053
SW		 584/*
585 * lookup session, bump ref if found.
586 *
587 * called under mdsc->mutex.
588 */
fcd00b68
ID		 589static struct ceph_mds_request *
590lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
2f2dc053
SW		 591{
592 struct ceph_mds_request *req;
44ca18f2 593
fcd00b68
ID		 594 req = lookup_request(&mdsc->request_tree, tid);
595 if (req)
596 ceph_mdsc_get_request(req);
44ca18f2 597
fcd00b68 598 return req;
2f2dc053
SW		 599}
600
601/*
602 * Register an in-flight request, and assign a tid. Link to directory
603 * are modifying (if any).
604 *
605 * Called under mdsc->mutex.
606 */
607static void __register_request(struct ceph_mds_client *mdsc,
608 struct ceph_mds_request *req,
609 struct inode *dir)
610{
611 req->r_tid = ++mdsc->last_tid;
612 if (req->r_num_caps)
37151668
YS		 613 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
614 req->r_num_caps);
2f2dc053
SW		 615 dout("__register_request %p tid %lld\n", req, req->r_tid);
616 ceph_mdsc_get_request(req);
fcd00b68 617 insert_request(&mdsc->request_tree, req);
2f2dc053 618
cb4276cc
SW		 619 req->r_uid = current_fsuid();
620 req->r_gid = current_fsgid();
621
e8a7b8b1
YZ		 622 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
623 mdsc->oldest_tid = req->r_tid;
624
2f2dc053 625 if (dir) {
3b663780 626 ihold(dir);
2f2dc053 627 req->r_unsafe_dir = dir;
2f2dc053
SW		 628 }
629}
630
631static void __unregister_request(struct ceph_mds_client *mdsc,
632 struct ceph_mds_request *req)
633{
634 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1 635
df963ea8
JL		 636 /* Never leave an unregistered request on an unsafe list! */
637 list_del_init(&req->r_unsafe_item);
638
e8a7b8b1
YZ		 639 if (req->r_tid == mdsc->oldest_tid) {
640 struct rb_node *p = rb_next(&req->r_node);
641 mdsc->oldest_tid = 0;
642 while (p) {
643 struct ceph_mds_request *next_req =
644 rb_entry(p, struct ceph_mds_request, r_node);
645 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
646 mdsc->oldest_tid = next_req->r_tid;
647 break;
648 }
649 p = rb_next(p);
650 }
651 }
652
fcd00b68 653 erase_request(&mdsc->request_tree, req);
2f2dc053 654
bc2de10d
JL		 655 if (req->r_unsafe_dir &&
656 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053 657 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 658 spin_lock(&ci->i_unsafe_lock);
659 list_del_init(&req->r_unsafe_dir_item);
660 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 661 }
bc2de10d
JL		 662 if (req->r_target_inode &&
663 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 664 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
665 spin_lock(&ci->i_unsafe_lock);
666 list_del_init(&req->r_unsafe_target_item);
667 spin_unlock(&ci->i_unsafe_lock);
668 }
3b663780 669
4c06ace8 670 if (req->r_unsafe_dir) {
3b663780
SW		 671 iput(req->r_unsafe_dir);
672 req->r_unsafe_dir = NULL;
2f2dc053 673 }
94aa8ae1 674
fc55d2c9
YZ		 675 complete_all(&req->r_safe_completion);
676
94aa8ae1 677 ceph_mdsc_put_request(req);
2f2dc053
SW		 678}
679
30c71233
JL		 680/*
681 * Walk back up the dentry tree until we hit a dentry representing a
682 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
683 * when calling this) to ensure that the objects won't disappear while we're
684 * working with them. Once we hit a candidate dentry, we attempt to take a
685 * reference to it, and return that as the result.
686 */
f1075480
DC		 687static struct inode *get_nonsnap_parent(struct dentry *dentry)
688{
689 struct inode *inode = NULL;
30c71233
JL		 690
691 while (dentry && !IS_ROOT(dentry)) {
692 inode = d_inode_rcu(dentry);
693 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
694 break;
695 dentry = dentry->d_parent;
696 }
697 if (inode)
698 inode = igrab(inode);
699 return inode;
700}
701
2f2dc053
SW		 702/*
703 * Choose mds to send request to next. If there is a hint set in the
704 * request (e.g., due to a prior forward hint from the mds), use that.
705 * Otherwise, consult frag tree and/or caps to identify the
706 * appropriate mds. If all else fails, choose randomly.
707 *
708 * Called under mdsc->mutex.
709 */
710static int __choose_mds(struct ceph_mds_client *mdsc,
711 struct ceph_mds_request *req)
712{
713 struct inode *inode;
714 struct ceph_inode_info *ci;
715 struct ceph_cap *cap;
716 int mode = req->r_direct_mode;
717 int mds = -1;
718 u32 hash = req->r_direct_hash;
bc2de10d 719 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
2f2dc053
SW		 720
721 /*
722 * is there a specific mds we should try? ignore hint if we have
723 * no session and the mds is not up (active or recovering).
724 */
725 if (req->r_resend_mds >= 0 &&
726 (__have_session(mdsc, req->r_resend_mds) ||
727 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
728 dout("choose_mds using resend_mds mds%d\n",
729 req->r_resend_mds);
730 return req->r_resend_mds;
731 }
732
733 if (mode == USE_RANDOM_MDS)
734 goto random;
735
736 inode = NULL;
737 if (req->r_inode) {
5d37ca14
YZ		 738 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
739 inode = req->r_inode;
740 ihold(inode);
741 } else {
38f340cc
YZ		 742 /* req->r_dentry is non-null for LSSNAP request */
743 rcu_read_lock();
744 inode = get_nonsnap_parent(req->r_dentry);
745 rcu_read_unlock();
746 dout("__choose_mds using snapdir's parent %p\n", inode);
5d37ca14 747 }
38f340cc 748 } else if (req->r_dentry) {
d79698da 749 /* ignore race with rename; old or new d_parent is okay */
30c71233
JL		 750 struct dentry *parent;
751 struct inode *dir;
752
753 rcu_read_lock();
754 parent = req->r_dentry->d_parent;
3dd69aab 755 dir = req->r_parent ? : d_inode_rcu(parent);
eb6bb1c5 756
30c71233
JL		 757 if (!dir || dir->i_sb != mdsc->fsc->sb) {
758 /* not this fs or parent went negative */
2b0143b5 759 inode = d_inode(req->r_dentry);
30c71233
JL		 760 if (inode)
761 ihold(inode);
eb6bb1c5
SW		 762 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
763 /* direct snapped/virtual snapdir requests
764 * based on parent dir inode */
30c71233 765 inode = get_nonsnap_parent(parent);
eb6bb1c5 766 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 767 } else {
eb6bb1c5 768 /* dentry target */
2b0143b5 769 inode = d_inode(req->r_dentry);
ca18bede
YZ		 770 if (!inode || mode == USE_AUTH_MDS) {
771 /* dir + name */
30c71233 772 inode = igrab(dir);
ca18bede
YZ		 773 hash = ceph_dentry_hash(dir, req->r_dentry);
774 is_hash = true;
30c71233
JL		 775 } else {
776 ihold(inode);
ca18bede 777 }
2f2dc053 778 }
30c71233 779 rcu_read_unlock();
2f2dc053 780 }
eb6bb1c5 781
2f2dc053
SW		 782 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
783 (int)hash, mode);
784 if (!inode)
785 goto random;
786 ci = ceph_inode(inode);
787
788 if (is_hash && S_ISDIR(inode->i_mode)) {
789 struct ceph_inode_frag frag;
790 int found;
791
792 ceph_choose_frag(ci, hash, &frag, &found);
793 if (found) {
794 if (mode == USE_ANY_MDS && frag.ndist > 0) {
795 u8 r;
796
797 /* choose a random replica */
798 get_random_bytes(&r, 1);
799 r %= frag.ndist;
800 mds = frag.dist[r];
801 dout("choose_mds %p %llx.%llx "
802 "frag %u mds%d (%d/%d)\n",
803 inode, ceph_vinop(inode),
d66bbd44 804 frag.frag, mds,
2f2dc053 805 (int)r, frag.ndist);
d66bbd44
SW		 806 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
807 CEPH_MDS_STATE_ACTIVE)
30c71233 808 goto out;
2f2dc053
SW		 809 }
810
811 /* since this file/dir wasn't known to be
812 * replicated, then we want to look for the
813 * authoritative mds. */
814 mode = USE_AUTH_MDS;
815 if (frag.mds >= 0) {
816 /* choose auth mds */
817 mds = frag.mds;
818 dout("choose_mds %p %llx.%llx "
819 "frag %u mds%d (auth)\n",
820 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 821 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
822 CEPH_MDS_STATE_ACTIVE)
30c71233 823 goto out;
2f2dc053
SW		 824 }
825 }
826 }
827
be655596 828 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 829 cap = NULL;
830 if (mode == USE_AUTH_MDS)
831 cap = ci->i_auth_cap;
832 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
833 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
834 if (!cap) {
be655596 835 spin_unlock(&ci->i_ceph_lock);
30c71233 836 iput(inode);
2f2dc053
SW		 837 goto random;
838 }
839 mds = cap->session->s_mds;
840 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
841 inode, ceph_vinop(inode), mds,
842 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 843 spin_unlock(&ci->i_ceph_lock);
30c71233
JL		 844out:
845 iput(inode);
2f2dc053
SW		 846 return mds;
847
848random:
849 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
850 dout("choose_mds chose random mds%d\n", mds);
851 return mds;
852}
853
854
855/*
856 * session messages
857 */
858static struct ceph_msg *create_session_msg(u32 op, u64 seq)
859{
860 struct ceph_msg *msg;
861 struct ceph_mds_session_head *h;
862
b61c2763
SW		 863 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
864 false);
a79832f2 865 if (!msg) {
2f2dc053 866 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 867 return NULL;
2f2dc053
SW		 868 }
869 h = msg->front.iov_base;
870 h->op = cpu_to_le32(op);
871 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 872
873 return msg;
874}
875
876/*
877 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
878 * to include additional client metadata fields.
879 */
880static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
881{
882 struct ceph_msg *msg;
883 struct ceph_mds_session_head *h;
884 int i = -1;
885 int metadata_bytes = 0;
886 int metadata_key_count = 0;
887 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 888 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
dbd0c8bf
JS		 889 void *p;
890
a6a5ce4f 891 const char* metadata[][2] = {
717e6f28
YZ		 892 {"hostname", mdsc->nodename},
893 {"kernel_version", init_utsname()->release},
3f384954
YZ		 894 {"entity_id", opt->name ? : ""},
895 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 896 {NULL, NULL}
897 };
898
899 /* Calculate serialized length of metadata */
900 metadata_bytes = 4; /* map length */
d37b1d99 901 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 902 metadata_bytes += 8 + strlen(metadata[i][0]) +
903 strlen(metadata[i][1]);
904 metadata_key_count++;
905 }
906
907 /* Allocate the message */
908 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
909 GFP_NOFS, false);
910 if (!msg) {
911 pr_err("create_session_msg ENOMEM creating msg\n");
912 return NULL;
913 }
914 h = msg->front.iov_base;
915 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
916 h->seq = cpu_to_le64(seq);
917
918 /*
919 * Serialize client metadata into waiting buffer space, using
920 * the format that userspace expects for map<string, string>
7cfa0313
JS		 921 *
922 * ClientSession messages with metadata are v2
dbd0c8bf 923 */
7cfa0313
JS		 924 msg->hdr.version = cpu_to_le16(2);
925 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 926
927 /* The write pointer, following the session_head structure */
928 p = msg->front.iov_base + sizeof(*h);
929
930 /* Number of entries in the map */
931 ceph_encode_32(&p, metadata_key_count);
932
933 /* Two length-prefixed strings for each entry in the map */
d37b1d99 934 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 935 size_t const key_len = strlen(metadata[i][0]);
936 size_t const val_len = strlen(metadata[i][1]);
937
938 ceph_encode_32(&p, key_len);
939 memcpy(p, metadata[i][0], key_len);
940 p += key_len;
941 ceph_encode_32(&p, val_len);
942 memcpy(p, metadata[i][1], val_len);
943 p += val_len;
944 }
945
2f2dc053
SW		 946 return msg;
947}
948
949/*
950 * send session open request.
951 *
952 * called under mdsc->mutex
953 */
954static int __open_session(struct ceph_mds_client *mdsc,
955 struct ceph_mds_session *session)
956{
957 struct ceph_msg *msg;
958 int mstate;
959 int mds = session->s_mds;
2f2dc053
SW		 960
961 /* wait for mds to go active? */
962 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
963 dout("open_session to mds%d (%s)\n", mds,
964 ceph_mds_state_name(mstate));
965 session->s_state = CEPH_MDS_SESSION_OPENING;
966 session->s_renew_requested = jiffies;
967
968 /* send connect message */
dbd0c8bf 969 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 970 if (!msg)
971 return -ENOMEM;
2f2dc053 972 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 973 return 0;
974}
975
ed0552a1
SW		 976/*
977 * open sessions for any export targets for the given mds
978 *
979 * called under mdsc->mutex
980 */
5d72d13c
YZ		 981static struct ceph_mds_session *
982__open_export_target_session(struct ceph_mds_client *mdsc, int target)
983{
984 struct ceph_mds_session *session;
985
986 session = __ceph_lookup_mds_session(mdsc, target);
987 if (!session) {
988 session = register_session(mdsc, target);
989 if (IS_ERR(session))
990 return session;
991 }
992 if (session->s_state == CEPH_MDS_SESSION_NEW ||
993 session->s_state == CEPH_MDS_SESSION_CLOSING)
994 __open_session(mdsc, session);
995
996 return session;
997}
998
999struct ceph_mds_session *
1000ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1001{
1002 struct ceph_mds_session *session;
1003
1004 dout("open_export_target_session to mds%d\n", target);
1005
1006 mutex_lock(&mdsc->mutex);
1007 session = __open_export_target_session(mdsc, target);
1008 mutex_unlock(&mdsc->mutex);
1009
1010 return session;
1011}
1012
ed0552a1
SW		 1013static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1014 struct ceph_mds_session *session)
1015{
1016 struct ceph_mds_info *mi;
1017 struct ceph_mds_session *ts;
1018 int i, mds = session->s_mds;
ed0552a1 1019
76201b63 1020 if (mds >= mdsc->mdsmap->m_num_mds)
ed0552a1 1021 return;
5d72d13c 1022
ed0552a1
SW		 1023 mi = &mdsc->mdsmap->m_info[mds];
1024 dout("open_export_target_sessions for mds%d (%d targets)\n",
1025 session->s_mds, mi->num_export_targets);
1026
1027 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1028 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1029 if (!IS_ERR(ts))
1030 ceph_put_mds_session(ts);
ed0552a1
SW		 1031 }
1032}
1033
154f42c2
SW		 1034void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1035 struct ceph_mds_session *session)
1036{
1037 mutex_lock(&mdsc->mutex);
1038 __open_export_target_sessions(mdsc, session);
1039 mutex_unlock(&mdsc->mutex);
1040}
1041
2f2dc053
SW		 1042/*
1043 * session caps
1044 */
1045
c8a96a31
JL		 1046static void detach_cap_releases(struct ceph_mds_session *session,
1047 struct list_head *target)
2f2dc053 1048{
c8a96a31
JL		 1049 lockdep_assert_held(&session->s_cap_lock);
1050
1051 list_splice_init(&session->s_cap_releases, target);
745a8e3b 1052 session->s_num_cap_releases = 0;
c8a96a31
JL		 1053 dout("dispose_cap_releases mds%d\n", session->s_mds);
1054}
2f2dc053 1055
c8a96a31
JL		 1056static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1057 struct list_head *dispose)
1058{
1059 while (!list_empty(dispose)) {
745a8e3b
YZ		 1060 struct ceph_cap *cap;
1061 /* zero out the in-progress message */
c8a96a31 1062 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
745a8e3b
YZ		 1063 list_del(&cap->session_caps);
1064 ceph_put_cap(mdsc, cap);
2f2dc053 1065 }
2f2dc053
SW		 1066}
1067
1c841a96
YZ		 1068static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1069 struct ceph_mds_session *session)
1070{
1071 struct ceph_mds_request *req;
1072 struct rb_node *p;
1073
1074 dout("cleanup_session_requests mds%d\n", session->s_mds);
1075 mutex_lock(&mdsc->mutex);
1076 while (!list_empty(&session->s_unsafe)) {
1077 req = list_first_entry(&session->s_unsafe,
1078 struct ceph_mds_request, r_unsafe_item);
3e0708b9
YZ		 1079 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1080 req->r_tid);
1c841a96
YZ		 1081 __unregister_request(mdsc, req);
1082 }
1083 /* zero r_attempts, so kick_requests() will re-send requests */
1084 p = rb_first(&mdsc->request_tree);
1085 while (p) {
1086 req = rb_entry(p, struct ceph_mds_request, r_node);
1087 p = rb_next(p);
1088 if (req->r_session &&
1089 req->r_session->s_mds == session->s_mds)
1090 req->r_attempts = 0;
1091 }
1092 mutex_unlock(&mdsc->mutex);
1093}
1094
2f2dc053 1095/*
f818a736
SW		 1096 * Helper to safely iterate over all caps associated with a session, with
1097 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1098 *
f818a736 1099 * Caller must hold session s_mutex.
2f2dc053
SW		 1100 */
1101static int iterate_session_caps(struct ceph_mds_session *session,
1102 int (*cb)(struct inode *, struct ceph_cap *,
1103 void *), void *arg)
1104{
7c1332b8
SW		 1105 struct list_head *p;
1106 struct ceph_cap *cap;
1107 struct inode *inode, *last_inode = NULL;
1108 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1109 int ret;
1110
1111 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1112 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1113 p = session->s_caps.next;
1114 while (p != &session->s_caps) {
1115 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1116 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1117 if (!inode) {
1118 p = p->next;
2f2dc053 1119 continue;
7c1332b8
SW		 1120 }
1121 session->s_cap_iterator = cap;
2f2dc053 1122 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1123
1124 if (last_inode) {
1125 iput(last_inode);
1126 last_inode = NULL;
1127 }
1128 if (old_cap) {
37151668 1129 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1130 old_cap = NULL;
1131 }
1132
2f2dc053 1133 ret = cb(inode, cap, arg);
7c1332b8
SW		 1134 last_inode = inode;
1135
2f2dc053 1136 spin_lock(&session->s_cap_lock);
7c1332b8 1137 p = p->next;
d37b1d99 1138 if (!cap->ci) {
7c1332b8
SW		 1139 dout("iterate_session_caps finishing cap %p removal\n",
1140 cap);
1141 BUG_ON(cap->session != session);
745a8e3b 1142 cap->session = NULL;
7c1332b8
SW		 1143 list_del_init(&cap->session_caps);
1144 session->s_nr_caps--;
1de82206
YZ		 1145 if (cap->queue_release)
1146 __ceph_queue_cap_release(session, cap);
1147 else
745a8e3b 1148 old_cap = cap; /* put_cap it w/o locks held */
7c1332b8 1149 }
5dacf091
SW		 1150 if (ret < 0)
1151 goto out;
2f2dc053 1152 }
5dacf091
SW		 1153 ret = 0;
1154out:
7c1332b8 1155 session->s_cap_iterator = NULL;
2f2dc053 1156 spin_unlock(&session->s_cap_lock);
7c1332b8 1157
e96a650a 1158 iput(last_inode);
7c1332b8 1159 if (old_cap)
37151668 1160 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1161
5dacf091 1162 return ret;
2f2dc053
SW		 1163}
1164
1165static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1166 void *arg)
2f2dc053 1167{
6c93df5d 1168 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
2f2dc053 1169 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1170 LIST_HEAD(to_remove);
6c93df5d
YZ		 1171 bool drop = false;
1172 bool invalidate = false;
6c99f254 1173
2f2dc053
SW		 1174 dout("removing cap %p, ci is %p, inode is %p\n",
1175 cap, ci, &ci->vfs_inode);
be655596 1176 spin_lock(&ci->i_ceph_lock);
a096b09a 1177 __ceph_remove_cap(cap, false);
571ade33 1178 if (!ci->i_auth_cap) {
553adfd9 1179 struct ceph_cap_flush *cf;
6c93df5d 1180 struct ceph_mds_client *mdsc = fsc->mdsc;
6c99f254 1181
77310320
YZ		 1182 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1183
6c93df5d 1184 if (ci->i_wrbuffer_ref > 0 &&
52953d55 1185 READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
6c93df5d
YZ		 1186 invalidate = true;
1187
e4500b5e
YZ		 1188 while (!list_empty(&ci->i_cap_flush_list)) {
1189 cf = list_first_entry(&ci->i_cap_flush_list,
1190 struct ceph_cap_flush, i_list);
8cdcc07d 1191 list_move(&cf->i_list, &to_remove);
553adfd9
YZ		 1192 }
1193
6c99f254 1194 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1195
e4500b5e
YZ		 1196 list_for_each_entry(cf, &to_remove, i_list)
1197 list_del(&cf->g_list);
8310b089 1198
6c99f254 1199 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1200 pr_warn_ratelimited(
1201 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1202 ceph_cap_string(ci->i_dirty_caps),
1203 inode, ceph_ino(inode));
1204 ci->i_dirty_caps = 0;
1205 list_del_init(&ci->i_dirty_item);
6c93df5d 1206 drop = true;
6c99f254
SW		 1207 }
1208 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1209 pr_warn_ratelimited(
1210 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1211 ceph_cap_string(ci->i_flushing_caps),
1212 inode, ceph_ino(inode));
1213 ci->i_flushing_caps = 0;
1214 list_del_init(&ci->i_flushing_item);
1215 mdsc->num_cap_flushing--;
6c93df5d 1216 drop = true;
6c99f254 1217 }
6c99f254 1218 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1219
b3f8d68f
YZ		 1220 if (atomic_read(&ci->i_filelock_ref) > 0) {
1221 /* make further file lock syscall return -EIO */
1222 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1223 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1224 inode, ceph_ino(inode));
1225 }
1226
f66fd9f0 1227 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1228 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1229 ci->i_prealloc_cap_flush = NULL;
1230 }
6c99f254 1231 }
be655596 1232 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1233 while (!list_empty(&to_remove)) {
1234 struct ceph_cap_flush *cf;
1235 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1236 struct ceph_cap_flush, i_list);
1237 list_del(&cf->i_list);
f66fd9f0 1238 ceph_free_cap_flush(cf);
553adfd9 1239 }
77310320
YZ		 1240
1241 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1242 if (invalidate)
1243 ceph_queue_invalidate(inode);
77310320 1244 if (drop)
6c99f254 1245 iput(inode);
2f2dc053
SW		 1246 return 0;
1247}
1248
1249/*
1250 * caller must hold session s_mutex
1251 */
1252static void remove_session_caps(struct ceph_mds_session *session)
1253{
6c93df5d
YZ		 1254 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1255 struct super_block *sb = fsc->sb;
c8a96a31
JL		 1256 LIST_HEAD(dispose);
1257
2f2dc053 1258 dout("remove_session_caps on %p\n", session);
6c93df5d 1259 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1260
c8799fc4
YZ		 1261 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1262
6f60f889
YZ		 1263 spin_lock(&session->s_cap_lock);
1264 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1265 struct inode *inode;
1266 struct ceph_cap *cap, *prev = NULL;
1267 struct ceph_vino vino;
1268 /*
1269 * iterate_session_caps() skips inodes that are being
1270 * deleted, we need to wait until deletions are complete.
1271 * __wait_on_freeing_inode() is designed for the job,
1272 * but it is not exported, so use lookup inode function
1273 * to access it.
1274 */
1275 while (!list_empty(&session->s_caps)) {
1276 cap = list_entry(session->s_caps.next,
1277 struct ceph_cap, session_caps);
1278 if (cap == prev)
1279 break;
1280 prev = cap;
1281 vino = cap->ci->i_vino;
1282 spin_unlock(&session->s_cap_lock);
1283
ed284c49 1284 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1285 iput(inode);
1286
1287 spin_lock(&session->s_cap_lock);
1288 }
1289 }
745a8e3b
YZ		 1290
1291 // drop cap expires and unlock s_cap_lock
c8a96a31 1292 detach_cap_releases(session, &dispose);
6f60f889 1293
2f2dc053 1294 BUG_ON(session->s_nr_caps > 0);
6c99f254 1295 BUG_ON(!list_empty(&session->s_cap_flushing));
c8a96a31
JL		 1296 spin_unlock(&session->s_cap_lock);
1297 dispose_cap_releases(session->s_mdsc, &dispose);
2f2dc053
SW		 1298}
1299
1300/*
1301 * wake up any threads waiting on this session's caps. if the cap is
1302 * old (didn't get renewed on the client reconnect), remove it now.
1303 *
1304 * caller must hold s_mutex.
1305 */
1306static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1307 void *arg)
1308{
0dc2570f
SW		 1309 struct ceph_inode_info *ci = ceph_inode(inode);
1310
0dc2570f 1311 if (arg) {
be655596 1312 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1313 ci->i_wanted_max_size = 0;
1314 ci->i_requested_max_size = 0;
be655596 1315 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1316 }
e5360309 1317 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1318 return 0;
1319}
1320
0dc2570f
SW		 1321static void wake_up_session_caps(struct ceph_mds_session *session,
1322 int reconnect)
2f2dc053
SW		 1323{
1324 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1325 iterate_session_caps(session, wake_up_session_cb,
1326 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1327}
1328
1329/*
1330 * Send periodic message to MDS renewing all currently held caps. The
1331 * ack will reset the expiration for all caps from this session.
1332 *
1333 * caller holds s_mutex
1334 */
1335static int send_renew_caps(struct ceph_mds_client *mdsc,
1336 struct ceph_mds_session *session)
1337{
1338 struct ceph_msg *msg;
1339 int state;
1340
1341 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1342 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1343 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1344 session->s_renew_requested = jiffies;
2f2dc053
SW		 1345
1346 /* do not try to renew caps until a recovering mds has reconnected
1347 * with its clients. */
1348 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1349 if (state < CEPH_MDS_STATE_RECONNECT) {
1350 dout("send_renew_caps ignoring mds%d (%s)\n",
1351 session->s_mds, ceph_mds_state_name(state));
1352 return 0;
1353 }
1354
1355 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1356 ceph_mds_state_name(state));
2f2dc053
SW		 1357 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1358 ++session->s_renew_seq);
a79832f2
SW		 1359 if (!msg)
1360 return -ENOMEM;
2f2dc053
SW		 1361 ceph_con_send(&session->s_con, msg);
1362 return 0;
1363}
1364
186e4f7a
YZ		 1365static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1366 struct ceph_mds_session *session, u64 seq)
1367{
1368 struct ceph_msg *msg;
1369
1370 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1371 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1372 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1373 if (!msg)
1374 return -ENOMEM;
1375 ceph_con_send(&session->s_con, msg);
1376 return 0;
1377}
1378
1379
2f2dc053
SW		 1380/*
1381 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1382 *
1383 * Called under session->s_mutex
2f2dc053
SW		 1384 */
1385static void renewed_caps(struct ceph_mds_client *mdsc,
1386 struct ceph_mds_session *session, int is_renew)
1387{
1388 int was_stale;
1389 int wake = 0;
1390
1391 spin_lock(&session->s_cap_lock);
1ce208a6 1392 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1393
1394 session->s_cap_ttl = session->s_renew_requested +
1395 mdsc->mdsmap->m_session_timeout*HZ;
1396
1397 if (was_stale) {
1398 if (time_before(jiffies, session->s_cap_ttl)) {
1399 pr_info("mds%d caps renewed\n", session->s_mds);
1400 wake = 1;
1401 } else {
1402 pr_info("mds%d caps still stale\n", session->s_mds);
1403 }
1404 }
1405 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1406 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1407 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1408 spin_unlock(&session->s_cap_lock);
1409
1410 if (wake)
0dc2570f 1411 wake_up_session_caps(session, 0);
2f2dc053
SW		 1412}
1413
1414/*
1415 * send a session close request
1416 */
1417static int request_close_session(struct ceph_mds_client *mdsc,
1418 struct ceph_mds_session *session)
1419{
1420 struct ceph_msg *msg;
2f2dc053
SW		 1421
1422 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1423 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1424 session->s_seq);
1425 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1426 if (!msg)
1427 return -ENOMEM;
1428 ceph_con_send(&session->s_con, msg);
fcff415c 1429 return 1;
2f2dc053
SW		 1430}
1431
1432/*
1433 * Called with s_mutex held.
1434 */
1435static int __close_session(struct ceph_mds_client *mdsc,
1436 struct ceph_mds_session *session)
1437{
1438 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1439 return 0;
1440 session->s_state = CEPH_MDS_SESSION_CLOSING;
1441 return request_close_session(mdsc, session);
1442}
1443
040d7860
YZ		 1444static bool drop_negative_children(struct dentry *dentry)
1445{
1446 struct dentry *child;
1447 bool all_negative = true;
1448
1449 if (!d_is_dir(dentry))
1450 goto out;
1451
1452 spin_lock(&dentry->d_lock);
1453 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1454 if (d_really_is_positive(child)) {
1455 all_negative = false;
1456 break;
1457 }
1458 }
1459 spin_unlock(&dentry->d_lock);
1460
1461 if (all_negative)
1462 shrink_dcache_parent(dentry);
1463out:
1464 return all_negative;
1465}
1466
2f2dc053
SW		 1467/*
1468 * Trim old(er) caps.
1469 *
1470 * Because we can't cache an inode without one or more caps, we do
1471 * this indirectly: if a cap is unused, we prune its aliases, at which
1472 * point the inode will hopefully get dropped to.
1473 *
1474 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1475 * memory pressure from the MDS, though, so it needn't be perfect.
1476 */
1477static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1478{
1479 struct ceph_mds_session *session = arg;
1480 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1481 int used, wanted, oissued, mine;
2f2dc053
SW		 1482
1483 if (session->s_trim_caps <= 0)
1484 return -1;
1485
be655596 1486 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1487 mine = cap->issued | cap->implemented;
1488 used = __ceph_caps_used(ci);
979abfdd 1489 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1490 oissued = __ceph_caps_issued_other(ci, cap);
1491
979abfdd 1492 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1493 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1494 ceph_cap_string(used), ceph_cap_string(wanted));
1495 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1496 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1497 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1498 goto out;
1499 if ((used | wanted) & CEPH_CAP_ANY_WR)
1500 goto out;
89aa5930
YZ		 1501 /* Note: it's possible that i_filelock_ref becomes non-zero
1502 * after dropping auth caps. It doesn't hurt because reply
1503 * of lock mds request will re-add auth caps. */
1504 if (atomic_read(&ci->i_filelock_ref) > 0)
1505 goto out;
979abfdd 1506 }
5e804ac4
YZ		 1507 /* The inode has cached pages, but it's no longer used.
1508 * we can safely drop it */
1509 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1510 !(oissued & CEPH_CAP_FILE_CACHE)) {
1511 used = 0;
1512 oissued = 0;
1513 }
979abfdd 1514 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1515 goto out; /* we need these caps */
1516
2f2dc053
SW		 1517 if (oissued) {
1518 /* we aren't the only cap.. just remove us */
a096b09a 1519 __ceph_remove_cap(cap, true);
040d7860 1520 session->s_trim_caps--;
2f2dc053 1521 } else {
040d7860 1522 struct dentry *dentry;
5e804ac4 1523 /* try dropping referring dentries */
be655596 1524 spin_unlock(&ci->i_ceph_lock);
040d7860
YZ		 1525 dentry = d_find_any_alias(inode);
1526 if (dentry && drop_negative_children(dentry)) {
1527 int count;
1528 dput(dentry);
1529 d_prune_aliases(inode);
1530 count = atomic_read(&inode->i_count);
1531 if (count == 1)
1532 session->s_trim_caps--;
1533 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1534 inode, cap, count);
1535 } else {
1536 dput(dentry);
1537 }
2f2dc053
SW		 1538 return 0;
1539 }
1540
1541out:
be655596 1542 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1543 return 0;
1544}
1545
1546/*
1547 * Trim session cap count down to some max number.
1548 */
1549static int trim_caps(struct ceph_mds_client *mdsc,
1550 struct ceph_mds_session *session,
1551 int max_caps)
1552{
1553 int trim_caps = session->s_nr_caps - max_caps;
1554
1555 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1556 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1557 if (trim_caps > 0) {
1558 session->s_trim_caps = trim_caps;
1559 iterate_session_caps(session, trim_caps_cb, session);
1560 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1561 session->s_mds, session->s_nr_caps, max_caps,
1562 trim_caps - session->s_trim_caps);
5dacf091 1563 session->s_trim_caps = 0;
2f2dc053 1564 }
a56371d9 1565
1de82206 1566 ceph_flush_cap_releases(mdsc, session);
2f2dc053
SW		 1567 return 0;
1568}
1569
8310b089
YZ		 1570static int check_caps_flush(struct ceph_mds_client *mdsc,
1571 u64 want_flush_tid)
1572{
8310b089
YZ		 1573 int ret = 1;
1574
1575 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1576 if (!list_empty(&mdsc->cap_flush_list)) {
1577 struct ceph_cap_flush *cf =
1578 list_first_entry(&mdsc->cap_flush_list,
1579 struct ceph_cap_flush, g_list);
1580 if (cf->tid <= want_flush_tid) {
1581 dout("check_caps_flush still flushing tid "
1582 "%llu <= %llu\n", cf->tid, want_flush_tid);
1583 ret = 0;
1584 }
8310b089
YZ		 1585 }
1586 spin_unlock(&mdsc->cap_dirty_lock);
1587 return ret;
d3383a8e
YZ		 1588}
1589
2f2dc053
SW		 1590/*
1591 * flush all dirty inode data to disk.
1592 *
8310b089 1593 * returns true if we've flushed through want_flush_tid
2f2dc053 1594 */
affbc19a 1595static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1596 u64 want_flush_tid)
2f2dc053 1597{
0e294387 1598 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1599
1600 wait_event(mdsc->cap_flushing_wq,
1601 check_caps_flush(mdsc, want_flush_tid));
1602
1603 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1604}
1605
1606/*
1607 * called under s_mutex
1608 */
1de82206
YZ		 1609static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1610 struct ceph_mds_session *session)
2f2dc053 1611{
745a8e3b
YZ		 1612 struct ceph_msg *msg = NULL;
1613 struct ceph_mds_cap_release *head;
1614 struct ceph_mds_cap_item *item;
92475f05 1615 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
745a8e3b
YZ		 1616 struct ceph_cap *cap;
1617 LIST_HEAD(tmp_list);
1618 int num_cap_releases;
92475f05
JL		 1619 __le32 barrier, *cap_barrier;
1620
1621 down_read(&osdc->lock);
1622 barrier = cpu_to_le32(osdc->epoch_barrier);
1623 up_read(&osdc->lock);
2f2dc053 1624
0f8605f2 1625 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1626again:
1627 list_splice_init(&session->s_cap_releases, &tmp_list);
1628 num_cap_releases = session->s_num_cap_releases;
1629 session->s_num_cap_releases = 0;
2f2dc053 1630 spin_unlock(&session->s_cap_lock);
e01a5946 1631
745a8e3b
YZ		 1632 while (!list_empty(&tmp_list)) {
1633 if (!msg) {
1634 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1635 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1636 if (!msg)
1637 goto out_err;
1638 head = msg->front.iov_base;
1639 head->num = cpu_to_le32(0);
1640 msg->front.iov_len = sizeof(*head);
92475f05
JL		 1641
1642 msg->hdr.version = cpu_to_le16(2);
1643 msg->hdr.compat_version = cpu_to_le16(1);
745a8e3b 1644 }
92475f05 1645
745a8e3b
YZ		 1646 cap = list_first_entry(&tmp_list, struct ceph_cap,
1647 session_caps);
1648 list_del(&cap->session_caps);
1649 num_cap_releases--;
e01a5946 1650
00bd8edb 1651 head = msg->front.iov_base;
745a8e3b
YZ		 1652 le32_add_cpu(&head->num, 1);
1653 item = msg->front.iov_base + msg->front.iov_len;
1654 item->ino = cpu_to_le64(cap->cap_ino);
1655 item->cap_id = cpu_to_le64(cap->cap_id);
1656 item->migrate_seq = cpu_to_le32(cap->mseq);
1657 item->seq = cpu_to_le32(cap->issue_seq);
1658 msg->front.iov_len += sizeof(*item);
1659
1660 ceph_put_cap(mdsc, cap);
1661
1662 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
92475f05
JL		 1663 // Append cap_barrier field
1664 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1665 *cap_barrier = barrier;
1666 msg->front.iov_len += sizeof(*cap_barrier);
1667
745a8e3b
YZ		 1668 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1669 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1670 ceph_con_send(&session->s_con, msg);
1671 msg = NULL;
1672 }
00bd8edb 1673 }
e01a5946 1674
745a8e3b 1675 BUG_ON(num_cap_releases != 0);
e01a5946 1676
745a8e3b
YZ		 1677 spin_lock(&session->s_cap_lock);
1678 if (!list_empty(&session->s_cap_releases))
1679 goto again;
1680 spin_unlock(&session->s_cap_lock);
1681
1682 if (msg) {
92475f05
JL		 1683 // Append cap_barrier field
1684 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1685 *cap_barrier = barrier;
1686 msg->front.iov_len += sizeof(*cap_barrier);
1687
745a8e3b
YZ		 1688 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1689 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1690 ceph_con_send(&session->s_con, msg);
e01a5946 1691 }
745a8e3b
YZ		 1692 return;
1693out_err:
1694 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1695 session->s_mds);
1696 spin_lock(&session->s_cap_lock);
1697 list_splice(&tmp_list, &session->s_cap_releases);
1698 session->s_num_cap_releases += num_cap_releases;
1699 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1700}
1701
1de82206
YZ		 1702static void ceph_cap_release_work(struct work_struct *work)
1703{
1704 struct ceph_mds_session *session =
1705 container_of(work, struct ceph_mds_session, s_cap_release_work);
1706
1707 mutex_lock(&session->s_mutex);
1708 if (session->s_state == CEPH_MDS_SESSION_OPEN ||
1709 session->s_state == CEPH_MDS_SESSION_HUNG)
1710 ceph_send_cap_releases(session->s_mdsc, session);
1711 mutex_unlock(&session->s_mutex);
1712 ceph_put_mds_session(session);
1713}
1714
1715void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
1716 struct ceph_mds_session *session)
1717{
1718 if (mdsc->stopping)
1719 return;
1720
1721 get_session(session);
1722 if (queue_work(mdsc->fsc->cap_wq,
1723 &session->s_cap_release_work)) {
1724 dout("cap release work queued\n");
1725 } else {
1726 ceph_put_mds_session(session);
1727 dout("failed to queue cap release work\n");
1728 }
1729}
1730
1731/*
1732 * caller holds session->s_cap_lock
1733 */
1734void __ceph_queue_cap_release(struct ceph_mds_session *session,
1735 struct ceph_cap *cap)
1736{
1737 list_add_tail(&cap->session_caps, &session->s_cap_releases);
1738 session->s_num_cap_releases++;
1739
1740 if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
1741 ceph_flush_cap_releases(session->s_mdsc, session);
1742}
1743
2f2dc053
SW		 1744/*
1745 * requests
1746 */
1747
54008399
YZ		 1748int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1749 struct inode *dir)
1750{
1751 struct ceph_inode_info *ci = ceph_inode(dir);
1752 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1753 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1754 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1755 int order, num_entries;
1756
1757 spin_lock(&ci->i_ceph_lock);
1758 num_entries = ci->i_files + ci->i_subdirs;
1759 spin_unlock(&ci->i_ceph_lock);
1760 num_entries = max(num_entries, 1);
1761 num_entries = min(num_entries, opt->max_readdir);
1762
1763 order = get_order(size * num_entries);
1764 while (order >= 0) {
2a5beea3
YZ		 1765 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1766 __GFP_NOWARN,
1767 order);
1768 if (rinfo->dir_entries)
54008399
YZ		 1769 break;
1770 order--;
1771 }
2a5beea3 1772 if (!rinfo->dir_entries)
54008399
YZ		 1773 return -ENOMEM;
1774
1775 num_entries = (PAGE_SIZE << order) / size;
1776 num_entries = min(num_entries, opt->max_readdir);
1777
1778 rinfo->dir_buf_size = PAGE_SIZE << order;
1779 req->r_num_caps = num_entries + 1;
1780 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1781 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1782 return 0;
1783}
1784
2f2dc053
SW		 1785/*
1786 * Create an mds request.
1787 */
1788struct ceph_mds_request *
1789ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1790{
1791 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1792
1793 if (!req)
1794 return ERR_PTR(-ENOMEM);
1795
b4556396 1796 mutex_init(&req->r_fill_mutex);
37151668 1797 req->r_mdsc = mdsc;
2f2dc053
SW		 1798 req->r_started = jiffies;
1799 req->r_resend_mds = -1;
1800 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1801 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1802 req->r_fmode = -1;
153c8e6b 1803 kref_init(&req->r_kref);
fcd00b68 1804 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1805 INIT_LIST_HEAD(&req->r_wait);
1806 init_completion(&req->r_completion);
1807 init_completion(&req->r_safe_completion);
1808 INIT_LIST_HEAD(&req->r_unsafe_item);
1809
56199016 1810 req->r_stamp = timespec_trunc(current_kernel_time(), mdsc->fsc->sb->s_time_gran);
b8e69066 1811
2f2dc053
SW		 1812 req->r_op = op;
1813 req->r_direct_mode = mode;
1814 return req;
1815}
1816
1817/*
44ca18f2 1818 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1819 *
1820 * called under mdsc->mutex.
1821 */
44ca18f2
SW		 1822static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1823{
1824 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1825 return NULL;
1826 return rb_entry(rb_first(&mdsc->request_tree),
1827 struct ceph_mds_request, r_node);
1828}
1829
e8a7b8b1 1830static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1831{
e8a7b8b1 1832 return mdsc->oldest_tid;
2f2dc053
SW		 1833}
1834
1835/*
1836 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1837 * on build_path_from_dentry in fs/cifs/dir.c.
1838 *
1839 * If @stop_on_nosnap, generate path relative to the first non-snapped
1840 * inode.
1841 *
1842 * Encode hidden .snap dirs as a double /, i.e.
1843 * foo/.snap/bar -> foo//bar
1844 */
1845char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1846 int stop_on_nosnap)
1847{
1848 struct dentry *temp;
1849 char *path;
1850 int len, pos;
1b71fe2e 1851 unsigned seq;
2f2dc053 1852
d37b1d99 1853 if (!dentry)
2f2dc053
SW		 1854 return ERR_PTR(-EINVAL);
1855
1856retry:
1857 len = 0;
1b71fe2e
AV		 1858 seq = read_seqbegin(&rename_lock);
1859 rcu_read_lock();
2f2dc053 1860 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1861 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1862 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1863 len++; /* slash only */
1864 else if (stop_on_nosnap && inode &&
1865 ceph_snap(inode) == CEPH_NOSNAP)
1866 break;
1867 else
1868 len += 1 + temp->d_name.len;
1869 temp = temp->d_parent;
2f2dc053 1870 }
1b71fe2e 1871 rcu_read_unlock();
2f2dc053
SW		 1872 if (len)
1873 len--; /* no leading '/' */
1874
1875 path = kmalloc(len+1, GFP_NOFS);
d37b1d99 1876 if (!path)
2f2dc053
SW		 1877 return ERR_PTR(-ENOMEM);
1878 pos = len;
1879 path[pos] = 0; /* trailing null */
1b71fe2e 1880 rcu_read_lock();
2f2dc053 1881 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1882 struct inode *inode;
2f2dc053 1883
1b71fe2e 1884 spin_lock(&temp->d_lock);
2b0143b5 1885 inode = d_inode(temp);
2f2dc053 1886 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1887 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1888 pos, temp);
1889 } else if (stop_on_nosnap && inode &&
1890 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1891 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1892 break;
1893 } else {
1894 pos -= temp->d_name.len;
1b71fe2e
AV		 1895 if (pos < 0) {
1896 spin_unlock(&temp->d_lock);
2f2dc053 1897 break;
1b71fe2e 1898 }
2f2dc053
SW		 1899 strncpy(path + pos, temp->d_name.name,
1900 temp->d_name.len);
2f2dc053 1901 }
1b71fe2e 1902 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1903 if (pos)
1904 path[--pos] = '/';
1905 temp = temp->d_parent;
2f2dc053 1906 }
1b71fe2e
AV		 1907 rcu_read_unlock();
1908 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1909 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1910 "expected, namelen is %d, pos is %d\n", len, pos);
1911 /* presumably this is only possible if racing with a
1912 rename of one of the parent directories (we can not
1913 lock the dentries above us to prevent this, but
1914 retrying should be harmless) */
1915 kfree(path);
1916 goto retry;
1917 }
1918
2b0143b5 1919 *base = ceph_ino(d_inode(temp));
2f2dc053 1920 *plen = len;
104648ad 1921 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1922 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1923 return path;
1924}
1925
fd36a717 1926static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2f2dc053
SW		 1927 const char **ppath, int *ppathlen, u64 *pino,
1928 int *pfreepath)
1929{
1930 char *path;
1931
c6b0b656 1932 rcu_read_lock();
fd36a717
JL		 1933 if (!dir)
1934 dir = d_inode_rcu(dentry->d_parent);
c6b0b656
JL		 1935 if (dir && ceph_snap(dir) == CEPH_NOSNAP) {
1936 *pino = ceph_ino(dir);
1937 rcu_read_unlock();
2f2dc053
SW		 1938 *ppath = dentry->d_name.name;
1939 *ppathlen = dentry->d_name.len;
1940 return 0;
1941 }
c6b0b656 1942 rcu_read_unlock();
2f2dc053
SW		 1943 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1944 if (IS_ERR(path))
1945 return PTR_ERR(path);
1946 *ppath = path;
1947 *pfreepath = 1;
1948 return 0;
1949}
1950
1951static int build_inode_path(struct inode *inode,
1952 const char **ppath, int *ppathlen, u64 *pino,
1953 int *pfreepath)
1954{
1955 struct dentry *dentry;
1956 char *path;
1957
1958 if (ceph_snap(inode) == CEPH_NOSNAP) {
1959 *pino = ceph_ino(inode);
1960 *ppathlen = 0;
1961 return 0;
1962 }
1963 dentry = d_find_alias(inode);
1964 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1965 dput(dentry);
1966 if (IS_ERR(path))
1967 return PTR_ERR(path);
1968 *ppath = path;
1969 *pfreepath = 1;
1970 return 0;
1971}
1972
1973/*
1974 * request arguments may be specified via an inode *, a dentry *, or
1975 * an explicit ino+path.
1976 */
1977static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
fd36a717
JL		 1978 struct inode *rdiri, const char *rpath,
1979 u64 rino, const char **ppath, int *pathlen,
2f2dc053
SW		 1980 u64 *ino, int *freepath)
1981{
1982 int r = 0;
1983
1984 if (rinode) {
1985 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1986 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1987 ceph_snap(rinode));
1988 } else if (rdentry) {
fd36a717
JL		 1989 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
1990 freepath);
2f2dc053
SW		 1991 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1992 *ppath);
795858db 1993 } else if (rpath || rino) {
2f2dc053
SW		 1994 *ino = rino;
1995 *ppath = rpath;
b000056a 1996 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1997 dout(" path %.*s\n", *pathlen, rpath);
1998 }
1999
2000 return r;
2001}
2002
2003/*
2004 * called under mdsc->mutex
2005 */
2006static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
2007 struct ceph_mds_request *req,
6e6f0923 2008 int mds, bool drop_cap_releases)
2f2dc053
SW		 2009{
2010 struct ceph_msg *msg;
2011 struct ceph_mds_request_head *head;
2012 const char *path1 = NULL;
2013 const char *path2 = NULL;
2014 u64 ino1 = 0, ino2 = 0;
2015 int pathlen1 = 0, pathlen2 = 0;
2016 int freepath1 = 0, freepath2 = 0;
2017 int len;
2018 u16 releases;
2019 void *p, *end;
2020 int ret;
2021
2022 ret = set_request_path_attr(req->r_inode, req->r_dentry,
3dd69aab 2023 req->r_parent, req->r_path1, req->r_ino1.ino,
2f2dc053
SW		 2024 &path1, &pathlen1, &ino1, &freepath1);
2025 if (ret < 0) {
2026 msg = ERR_PTR(ret);
2027 goto out;
2028 }
2029
2030 ret = set_request_path_attr(NULL, req->r_old_dentry,
fd36a717 2031 req->r_old_dentry_dir,
2f2dc053
SW		 2032 req->r_path2, req->r_ino2.ino,
2033 &path2, &pathlen2, &ino2, &freepath2);
2034 if (ret < 0) {
2035 msg = ERR_PTR(ret);
2036 goto out_free1;
2037 }
2038
2039 len = sizeof(*head) +
b8e69066 2040 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 2041 sizeof(struct ceph_timespec);
2f2dc053
SW		 2042
2043 /* calculate (max) length for cap releases */
2044 len += sizeof(struct ceph_mds_request_release) *
2045 (!!req->r_inode_drop + !!req->r_dentry_drop +
2046 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2047 if (req->r_dentry_drop)
2048 len += req->r_dentry->d_name.len;
2049 if (req->r_old_dentry_drop)
2050 len += req->r_old_dentry->d_name.len;
2051
b61c2763 2052 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 2053 if (!msg) {
2054 msg = ERR_PTR(-ENOMEM);
2f2dc053 2055 goto out_free2;
a79832f2 2056 }
2f2dc053 2057
7cfa0313 2058 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 2059 msg->hdr.tid = cpu_to_le64(req->r_tid);
2060
2f2dc053
SW		 2061 head = msg->front.iov_base;
2062 p = msg->front.iov_base + sizeof(*head);
2063 end = msg->front.iov_base + msg->front.iov_len;
2064
2065 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2066 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 2067 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2068 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 2069 head->args = req->r_args;
2070
2071 ceph_encode_filepath(&p, end, ino1, path1);
2072 ceph_encode_filepath(&p, end, ino2, path2);
2073
e979cf50
SW		 2074 /* make note of release offset, in case we need to replay */
2075 req->r_request_release_offset = p - msg->front.iov_base;
2076
2f2dc053
SW		 2077 /* cap releases */
2078 releases = 0;
2079 if (req->r_inode_drop)
2080 releases += ceph_encode_inode_release(&p,
2b0143b5 2081 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 2082 mds, req->r_inode_drop, req->r_inode_unless, 0);
2083 if (req->r_dentry_drop)
2084 releases += ceph_encode_dentry_release(&p, req->r_dentry,
3dd69aab 2085 req->r_parent, mds, req->r_dentry_drop,
ca6c8ae0 2086 req->r_dentry_unless);
2f2dc053
SW		 2087 if (req->r_old_dentry_drop)
2088 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
ca6c8ae0
JL		 2089 req->r_old_dentry_dir, mds,
2090 req->r_old_dentry_drop,
2091 req->r_old_dentry_unless);
2f2dc053
SW		 2092 if (req->r_old_inode_drop)
2093 releases += ceph_encode_inode_release(&p,
2b0143b5 2094 d_inode(req->r_old_dentry),
2f2dc053 2095 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 2096
2097 if (drop_cap_releases) {
2098 releases = 0;
2099 p = msg->front.iov_base + req->r_request_release_offset;
2100 }
2101
2f2dc053
SW		 2102 head->num_releases = cpu_to_le16(releases);
2103
b8e69066 2104 /* time stamp */
1f041a89
YZ		 2105 {
2106 struct ceph_timespec ts;
2107 ceph_encode_timespec(&ts, &req->r_stamp);
2108 ceph_encode_copy(&p, &ts, sizeof(ts));
2109 }
b8e69066 2110
2f2dc053
SW		 2111 BUG_ON(p > end);
2112 msg->front.iov_len = p - msg->front.iov_base;
2113 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2114
25e6bae3
YZ		 2115 if (req->r_pagelist) {
2116 struct ceph_pagelist *pagelist = req->r_pagelist;
0e1a5ee6 2117 refcount_inc(&pagelist->refcnt);
25e6bae3
YZ		 2118 ceph_msg_data_add_pagelist(msg, pagelist);
2119 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2120 } else {
2121 msg->hdr.data_len = 0;
ebf18f47 2122 }
02afca6c 2123
2f2dc053
SW		 2124 msg->hdr.data_off = cpu_to_le16(0);
2125
2126out_free2:
2127 if (freepath2)
2128 kfree((char *)path2);
2129out_free1:
2130 if (freepath1)
2131 kfree((char *)path1);
2132out:
2133 return msg;
2134}
2135
2136/*
2137 * called under mdsc->mutex if error, under no mutex if
2138 * success.
2139 */
2140static void complete_request(struct ceph_mds_client *mdsc,
2141 struct ceph_mds_request *req)
2142{
2143 if (req->r_callback)
2144 req->r_callback(mdsc, req);
2145 else
03066f23 2146 complete_all(&req->r_completion);
2f2dc053
SW		 2147}
2148
2149/*
2150 * called under mdsc->mutex
2151 */
2152static int __prepare_send_request(struct ceph_mds_client *mdsc,
2153 struct ceph_mds_request *req,
6e6f0923 2154 int mds, bool drop_cap_releases)
2f2dc053
SW		 2155{
2156 struct ceph_mds_request_head *rhead;
2157 struct ceph_msg *msg;
2158 int flags = 0;
2159
2f2dc053 2160 req->r_attempts++;
e55b71f8
GF		 2161 if (req->r_inode) {
2162 struct ceph_cap *cap =
2163 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2164
2165 if (cap)
2166 req->r_sent_on_mseq = cap->mseq;
2167 else
2168 req->r_sent_on_mseq = -1;
2169 }
2f2dc053
SW		 2170 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2171 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2172
bc2de10d 2173 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
c5c9a0bf 2174 void *p;
01a92f17
SW		 2175 /*
2176 * Replay. Do not regenerate message (and rebuild
2177 * paths, etc.); just use the original message.
2178 * Rebuilding paths will break for renames because
2179 * d_move mangles the src name.
2180 */
2181 msg = req->r_request;
2182 rhead = msg->front.iov_base;
2183
2184 flags = le32_to_cpu(rhead->flags);
2185 flags |= CEPH_MDS_FLAG_REPLAY;
2186 rhead->flags = cpu_to_le32(flags);
2187
2188 if (req->r_target_inode)
2189 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2190
2191 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2192
2193 /* remove cap/dentry releases from message */
2194 rhead->num_releases = 0;
c5c9a0bf
YZ		 2195
2196 /* time stamp */
2197 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2198 {
2199 struct ceph_timespec ts;
2200 ceph_encode_timespec(&ts, &req->r_stamp);
2201 ceph_encode_copy(&p, &ts, sizeof(ts));
2202 }
c5c9a0bf
YZ		 2203
2204 msg->front.iov_len = p - msg->front.iov_base;
2205 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2206 return 0;
2207 }
2208
2f2dc053
SW		 2209 if (req->r_request) {
2210 ceph_msg_put(req->r_request);
2211 req->r_request = NULL;
2212 }
6e6f0923 2213 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2214 if (IS_ERR(msg)) {
e1518c7c 2215 req->r_err = PTR_ERR(msg);
a79832f2 2216 return PTR_ERR(msg);
2f2dc053
SW		 2217 }
2218 req->r_request = msg;
2219
2220 rhead = msg->front.iov_base;
2f2dc053 2221 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
bc2de10d 2222 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2f2dc053 2223 flags |= CEPH_MDS_FLAG_REPLAY;
3dd69aab 2224 if (req->r_parent)
2f2dc053
SW		 2225 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2226 rhead->flags = cpu_to_le32(flags);
2227 rhead->num_fwd = req->r_num_fwd;
2228 rhead->num_retry = req->r_attempts - 1;
01a92f17 2229 rhead->ino = 0;
2f2dc053 2230
3dd69aab 2231 dout(" r_parent = %p\n", req->r_parent);
2f2dc053
SW		 2232 return 0;
2233}
2234
2235/*
2236 * send request, or put it on the appropriate wait list.
2237 */
2238static int __do_request(struct ceph_mds_client *mdsc,
2239 struct ceph_mds_request *req)
2240{
2241 struct ceph_mds_session *session = NULL;
2242 int mds = -1;
48fec5d0 2243 int err = 0;
2f2dc053 2244
bc2de10d
JL		 2245 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2246 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
eb1b8af3 2247 __unregister_request(mdsc, req);
2f2dc053 2248 goto out;
eb1b8af3 2249 }
2f2dc053
SW		 2250
2251 if (req->r_timeout &&
2252 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2253 dout("do_request timed out\n");
2254 err = -EIO;
2255 goto finish;
2256 }
52953d55 2257 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
48fec5d0
YZ		 2258 dout("do_request forced umount\n");
2259 err = -EIO;
2260 goto finish;
2261 }
52953d55 2262 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
e9e427f0
YZ		 2263 if (mdsc->mdsmap_err) {
2264 err = mdsc->mdsmap_err;
2265 dout("do_request mdsmap err %d\n", err);
2266 goto finish;
2267 }
cc8e8342
YZ		 2268 if (mdsc->mdsmap->m_epoch == 0) {
2269 dout("do_request no mdsmap, waiting for map\n");
2270 list_add(&req->r_wait, &mdsc->waiting_for_map);
2271 goto finish;
2272 }
e9e427f0
YZ		 2273 if (!(mdsc->fsc->mount_options->flags &
2274 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2275 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2276 err = -ENOENT;
2277 pr_info("probably no mds server is up\n");
2278 goto finish;
2279 }
2280 }
2f2dc053 2281
dc69e2e9
SW		 2282 put_request_session(req);
2283
2f2dc053
SW		 2284 mds = __choose_mds(mdsc, req);
2285 if (mds < 0 ||
2286 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2287 dout("do_request no mds or not active, waiting for map\n");
2288 list_add(&req->r_wait, &mdsc->waiting_for_map);
2289 goto out;
2290 }
2291
2292 /* get, open session */
2293 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2294 if (!session) {
2f2dc053 2295 session = register_session(mdsc, mds);
9c423956
SW		 2296 if (IS_ERR(session)) {
2297 err = PTR_ERR(session);
2298 goto finish;
2299 }
2300 }
dc69e2e9
SW		 2301 req->r_session = get_session(session);
2302
2f2dc053 2303 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2304 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2305 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2306 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2307 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2308 err = -EACCES;
2309 goto out_session;
2310 }
2f2dc053
SW		 2311 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2312 session->s_state == CEPH_MDS_SESSION_CLOSING)
2313 __open_session(mdsc, session);
2314 list_add(&req->r_wait, &session->s_waiting);
2315 goto out_session;
2316 }
2317
2318 /* send request */
2f2dc053
SW		 2319 req->r_resend_mds = -1; /* forget any previous mds hint */
2320
2321 if (req->r_request_started == 0) /* note request start time */
2322 req->r_request_started = jiffies;
2323
6e6f0923 2324 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2325 if (!err) {
2326 ceph_msg_get(req->r_request);
2327 ceph_con_send(&session->s_con, req->r_request);
2328 }
2329
2330out_session:
2331 ceph_put_mds_session(session);
48fec5d0
YZ		 2332finish:
2333 if (err) {
2334 dout("__do_request early error %d\n", err);
2335 req->r_err = err;
2336 complete_request(mdsc, req);
2337 __unregister_request(mdsc, req);
2338 }
2f2dc053
SW		 2339out:
2340 return err;
2f2dc053
SW		 2341}
2342
2343/*
2344 * called under mdsc->mutex
2345 */
2346static void __wake_requests(struct ceph_mds_client *mdsc,
2347 struct list_head *head)
2348{
ed75ec2c
YZ		 2349 struct ceph_mds_request *req;
2350 LIST_HEAD(tmp_list);
2351
2352 list_splice_init(head, &tmp_list);
2f2dc053 2353
ed75ec2c
YZ		 2354 while (!list_empty(&tmp_list)) {
2355 req = list_entry(tmp_list.next,
2356 struct ceph_mds_request, r_wait);
2f2dc053 2357 list_del_init(&req->r_wait);
7971bd92 2358 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2359 __do_request(mdsc, req);
2360 }
2361}
2362
2363/*
2364 * Wake up threads with requests pending for @mds, so that they can
29790f26 2365 * resubmit their requests to a possibly different mds.
2f2dc053 2366 */
29790f26 2367static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2368{
44ca18f2 2369 struct ceph_mds_request *req;
282c1052 2370 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2371
2372 dout("kick_requests mds%d\n", mds);
282c1052 2373 while (p) {
44ca18f2 2374 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2375 p = rb_next(p);
bc2de10d 2376 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
44ca18f2 2377 continue;
3de22be6
YZ		 2378 if (req->r_attempts > 0)
2379 continue; /* only new requests */
44ca18f2
SW		 2380 if (req->r_session &&
2381 req->r_session->s_mds == mds) {
2382 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2383 list_del_init(&req->r_wait);
44ca18f2 2384 __do_request(mdsc, req);
2f2dc053
SW		 2385 }
2386 }
2387}
2388
2389void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2390 struct ceph_mds_request *req)
2391{
2392 dout("submit_request on %p\n", req);
2393 mutex_lock(&mdsc->mutex);
2394 __register_request(mdsc, req, NULL);
2395 __do_request(mdsc, req);
2396 mutex_unlock(&mdsc->mutex);
2397}
2398
2399/*
2400 * Synchrously perform an mds request. Take care of all of the
2401 * session setup, forwarding, retry details.
2402 */
2403int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2404 struct inode *dir,
2405 struct ceph_mds_request *req)
2406{
2407 int err;
2408
2409 dout("do_request on %p\n", req);
2410
3dd69aab 2411 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2f2dc053
SW		 2412 if (req->r_inode)
2413 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3dd69aab
JL		 2414 if (req->r_parent)
2415 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
844d87c3 2416 if (req->r_old_dentry_dir)
41b02e1f
SW		 2417 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2418 CEPH_CAP_PIN);
2f2dc053
SW		 2419
2420 /* issue */
2421 mutex_lock(&mdsc->mutex);
2422 __register_request(mdsc, req, dir);
2423 __do_request(mdsc, req);
2424
e1518c7c
SW		 2425 if (req->r_err) {
2426 err = req->r_err;
e1518c7c 2427 goto out;
2f2dc053
SW		 2428 }
2429
e1518c7c
SW		 2430 /* wait */
2431 mutex_unlock(&mdsc->mutex);
2432 dout("do_request waiting\n");
5be73034 2433 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2434 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2435 } else {
5be73034
ID		 2436 long timeleft = wait_for_completion_killable_timeout(
2437 &req->r_completion,
2438 ceph_timeout_jiffies(req->r_timeout));
2439 if (timeleft > 0)
2440 err = 0;
2441 else if (!timeleft)
2442 err = -EIO; /* timed out */
2443 else
2444 err = timeleft; /* killed */
e1518c7c
SW		 2445 }
2446 dout("do_request waited, got %d\n", err);
2447 mutex_lock(&mdsc->mutex);
5b1daecd 2448
e1518c7c 2449 /* only abort if we didn't race with a real reply */
bc2de10d 2450 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
e1518c7c
SW		 2451 err = le32_to_cpu(req->r_reply_info.head->result);
2452 } else if (err < 0) {
2453 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2454
2455 /*
2456 * ensure we aren't running concurrently with
2457 * ceph_fill_trace or ceph_readdir_prepopulate, which
2458 * rely on locks (dir mutex) held by our caller.
2459 */
2460 mutex_lock(&req->r_fill_mutex);
e1518c7c 2461 req->r_err = err;
bc2de10d 2462 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
b4556396 2463 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2464
3dd69aab 2465 if (req->r_parent &&
167c9e35
SW		 2466 (req->r_op & CEPH_MDS_OP_WRITE))
2467 ceph_invalidate_dir_request(req);
2f2dc053 2468 } else {
e1518c7c 2469 err = req->r_err;
2f2dc053 2470 }
2f2dc053 2471
e1518c7c
SW		 2472out:
2473 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2474 dout("do_request %p done, result %d\n", req, err);
2475 return err;
2476}
2477
167c9e35 2478/*
2f276c51 2479 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2480 * namespace request.
2481 */
2482void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2483{
3dd69aab 2484 struct inode *inode = req->r_parent;
167c9e35 2485
2f276c51 2486 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2487
2f276c51 2488 ceph_dir_clear_complete(inode);
167c9e35
SW		 2489 if (req->r_dentry)
2490 ceph_invalidate_dentry_lease(req->r_dentry);
2491 if (req->r_old_dentry)
2492 ceph_invalidate_dentry_lease(req->r_old_dentry);
2493}
2494
2f2dc053
SW		 2495/*
2496 * Handle mds reply.
2497 *
2498 * We take the session mutex and parse and process the reply immediately.
2499 * This preserves the logical ordering of replies, capabilities, etc., sent
2500 * by the MDS as they are applied to our local cache.
2501 */
2502static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2503{
2504 struct ceph_mds_client *mdsc = session->s_mdsc;
2505 struct ceph_mds_request *req;
2506 struct ceph_mds_reply_head *head = msg->front.iov_base;
2507 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2508 struct ceph_snap_realm *realm;
2f2dc053
SW		 2509 u64 tid;
2510 int err, result;
2600d2dd 2511 int mds = session->s_mds;
2f2dc053 2512
2f2dc053
SW		 2513 if (msg->front.iov_len < sizeof(*head)) {
2514 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2515 ceph_msg_dump(msg);
2f2dc053
SW		 2516 return;
2517 }
2518
2519 /* get request, session */
6df058c0 2520 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2521 mutex_lock(&mdsc->mutex);
fcd00b68 2522 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2523 if (!req) {
2524 dout("handle_reply on unknown tid %llu\n", tid);
2525 mutex_unlock(&mdsc->mutex);
2526 return;
2527 }
2528 dout("handle_reply %p\n", req);
2f2dc053
SW		 2529
2530 /* correct session? */
d96d6049 2531 if (req->r_session != session) {
2f2dc053
SW		 2532 pr_err("mdsc_handle_reply got %llu on session mds%d"
2533 " not mds%d\n", tid, session->s_mds,
2534 req->r_session ? req->r_session->s_mds : -1);
2535 mutex_unlock(&mdsc->mutex);
2536 goto out;
2537 }
2538
2539 /* dup? */
bc2de10d
JL		 2540 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2541 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
f3ae1b97 2542 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2543 head->safe ? "safe" : "unsafe", tid, mds);
2544 mutex_unlock(&mdsc->mutex);
2545 goto out;
2546 }
bc2de10d 2547 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
f3ae1b97 2548 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2549 tid, mds);
2550 mutex_unlock(&mdsc->mutex);
2551 goto out;
2552 }
2f2dc053
SW		 2553
2554 result = le32_to_cpu(head->result);
2555
2556 /*
e55b71f8
GF		 2557 * Handle an ESTALE
2558 * if we're not talking to the authority, send to them
2559 * if the authority has changed while we weren't looking,
2560 * send to new authority
2561 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2562 */
2563 if (result == -ESTALE) {
e55b71f8 2564 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2565 req->r_resend_mds = -1;
ca18bede 2566 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2567 dout("not using auth, setting for that now");
2568 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2569 __do_request(mdsc, req);
2570 mutex_unlock(&mdsc->mutex);
2571 goto out;
e55b71f8 2572 } else {
ca18bede
YZ		 2573 int mds = __choose_mds(mdsc, req);
2574 if (mds >= 0 && mds != req->r_session->s_mds) {
2575 dout("but auth changed, so resending");
e55b71f8
GF		 2576 __do_request(mdsc, req);
2577 mutex_unlock(&mdsc->mutex);
2578 goto out;
2579 }
2f2dc053 2580 }
e55b71f8 2581 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2582 }
2583
e55b71f8 2584
2f2dc053 2585 if (head->safe) {
bc2de10d 2586 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2f2dc053 2587 __unregister_request(mdsc, req);
2f2dc053 2588
bc2de10d 2589 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053
SW		 2590 /*
2591 * We already handled the unsafe response, now do the
2592 * cleanup. No need to examine the response; the MDS
2593 * doesn't include any result info in the safe
2594 * response. And even if it did, there is nothing
2595 * useful we could do with a revised return value.
2596 */
2597 dout("got safe reply %llu, mds%d\n", tid, mds);
2f2dc053
SW		 2598
2599 /* last unsafe request during umount? */
44ca18f2 2600 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2601 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2602 mutex_unlock(&mdsc->mutex);
2603 goto out;
2604 }
e1518c7c 2605 } else {
bc2de10d 2606 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2f2dc053 2607 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2608 if (req->r_unsafe_dir) {
2609 struct ceph_inode_info *ci =
2610 ceph_inode(req->r_unsafe_dir);
2611 spin_lock(&ci->i_unsafe_lock);
2612 list_add_tail(&req->r_unsafe_dir_item,
2613 &ci->i_unsafe_dirops);
2614 spin_unlock(&ci->i_unsafe_lock);
2615 }
2f2dc053
SW		 2616 }
2617
2618 dout("handle_reply tid %lld result %d\n", tid, result);
2619 rinfo = &req->r_reply_info;
14303d20 2620 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2621 mutex_unlock(&mdsc->mutex);
2622
2623 mutex_lock(&session->s_mutex);
2624 if (err < 0) {
25933abd 2625 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2626 ceph_msg_dump(msg);
2f2dc053
SW		 2627 goto out_err;
2628 }
2629
2630 /* snap trace */
982d6011 2631 realm = NULL;
2f2dc053
SW		 2632 if (rinfo->snapblob_len) {
2633 down_write(&mdsc->snap_rwsem);
2634 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2635 rinfo->snapblob + rinfo->snapblob_len,
2636 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2637 &realm);
2f2dc053
SW		 2638 downgrade_write(&mdsc->snap_rwsem);
2639 } else {
2640 down_read(&mdsc->snap_rwsem);
2641 }
2642
2643 /* insert trace into our cache */
b4556396 2644 mutex_lock(&req->r_fill_mutex);
315f2408 2645 current->journal_info = req;
f5a03b08 2646 err = ceph_fill_trace(mdsc->fsc->sb, req);
2f2dc053 2647 if (err == 0) {
6e8575fa 2648 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2649 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2650 ceph_readdir_prepopulate(req, req->r_session);
37151668 2651 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2652 }
315f2408 2653 current->journal_info = NULL;
b4556396 2654 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2655
2656 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2657 if (realm)
2658 ceph_put_snap_realm(mdsc, realm);
68cd5b4b 2659
bc2de10d
JL		 2660 if (err == 0 && req->r_target_inode &&
2661 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 2662 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2663 spin_lock(&ci->i_unsafe_lock);
2664 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2665 spin_unlock(&ci->i_unsafe_lock);
2666 }
2f2dc053 2667out_err:
e1518c7c 2668 mutex_lock(&mdsc->mutex);
bc2de10d 2669 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
e1518c7c
SW		 2670 if (err) {
2671 req->r_err = err;
2672 } else {
5fdb1389 2673 req->r_reply = ceph_msg_get(msg);
bc2de10d 2674 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
e1518c7c 2675 }
2f2dc053 2676 } else {
e1518c7c 2677 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2678 }
e1518c7c 2679 mutex_unlock(&mdsc->mutex);
2f2dc053 2680
2f2dc053
SW		 2681 mutex_unlock(&session->s_mutex);
2682
2683 /* kick calling process */
2684 complete_request(mdsc, req);
2685out:
2686 ceph_mdsc_put_request(req);
2687 return;
2688}
2689
2690
2691
2692/*
2693 * handle mds notification that our request has been forwarded.
2694 */
2600d2dd
SW		 2695static void handle_forward(struct ceph_mds_client *mdsc,
2696 struct ceph_mds_session *session,
2697 struct ceph_msg *msg)
2f2dc053
SW		 2698{
2699 struct ceph_mds_request *req;
a1ea787c 2700 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2701 u32 next_mds;
2702 u32 fwd_seq;
2f2dc053
SW		 2703 int err = -EINVAL;
2704 void *p = msg->front.iov_base;
2705 void *end = p + msg->front.iov_len;
2f2dc053 2706
a1ea787c 2707 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2708 next_mds = ceph_decode_32(&p);
2709 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2710
2711 mutex_lock(&mdsc->mutex);
fcd00b68 2712 req = lookup_get_request(mdsc, tid);
2f2dc053 2713 if (!req) {
2a8e5e36 2714 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2715 goto out; /* dup reply? */
2716 }
2717
bc2de10d 2718 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2a8e5e36
SW		 2719 dout("forward tid %llu aborted, unregistering\n", tid);
2720 __unregister_request(mdsc, req);
2721 } else if (fwd_seq <= req->r_num_fwd) {
2722 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2723 tid, next_mds, req->r_num_fwd, fwd_seq);
2724 } else {
2725 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2726 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2727 BUG_ON(req->r_err);
bc2de10d 2728 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3de22be6 2729 req->r_attempts = 0;
2f2dc053
SW		 2730 req->r_num_fwd = fwd_seq;
2731 req->r_resend_mds = next_mds;
2732 put_request_session(req);
2733 __do_request(mdsc, req);
2734 }
2735 ceph_mdsc_put_request(req);
2736out:
2737 mutex_unlock(&mdsc->mutex);
2738 return;
2739
2740bad:
2741 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2742}
2743
2744/*
2745 * handle a mds session control message
2746 */
2747static void handle_session(struct ceph_mds_session *session,
2748 struct ceph_msg *msg)
2749{
2750 struct ceph_mds_client *mdsc = session->s_mdsc;
2751 u32 op;
2752 u64 seq;
2600d2dd 2753 int mds = session->s_mds;
2f2dc053
SW		 2754 struct ceph_mds_session_head *h = msg->front.iov_base;
2755 int wake = 0;
2756
2f2dc053
SW		 2757 /* decode */
2758 if (msg->front.iov_len != sizeof(*h))
2759 goto bad;
2760 op = le32_to_cpu(h->op);
2761 seq = le64_to_cpu(h->seq);
2762
2763 mutex_lock(&mdsc->mutex);
0a07fc8c
YZ		 2764 if (op == CEPH_SESSION_CLOSE) {
2765 get_session(session);
2600d2dd 2766 __unregister_session(mdsc, session);
0a07fc8c 2767 }
2f2dc053
SW		 2768 /* FIXME: this ttl calculation is generous */
2769 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2770 mutex_unlock(&mdsc->mutex);
2771
2772 mutex_lock(&session->s_mutex);
2773
2774 dout("handle_session mds%d %s %p state %s seq %llu\n",
2775 mds, ceph_session_op_name(op), session,
a687ecaf 2776 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2777
2778 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2779 session->s_state = CEPH_MDS_SESSION_OPEN;
2780 pr_info("mds%d came back\n", session->s_mds);
2781 }
2782
2783 switch (op) {
2784 case CEPH_SESSION_OPEN:
29790f26
SW		 2785 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2786 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2787 session->s_state = CEPH_MDS_SESSION_OPEN;
2788 renewed_caps(mdsc, session, 0);
2789 wake = 1;
2790 if (mdsc->stopping)
2791 __close_session(mdsc, session);
2792 break;
2793
2794 case CEPH_SESSION_RENEWCAPS:
2795 if (session->s_renew_seq == seq)
2796 renewed_caps(mdsc, session, 1);
2797 break;
2798
2799 case CEPH_SESSION_CLOSE:
29790f26
SW		 2800 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2801 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2802 cleanup_session_requests(mdsc, session);
2f2dc053 2803 remove_session_caps(session);
656e4382 2804 wake = 2; /* for good measure */
f3c60c59 2805 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2806 break;
2807
2808 case CEPH_SESSION_STALE:
2809 pr_info("mds%d caps went stale, renewing\n",
2810 session->s_mds);
d8fb02ab 2811 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2812 session->s_cap_gen++;
1ce208a6 2813 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2814 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2815 send_renew_caps(mdsc, session);
2816 break;
2817
2818 case CEPH_SESSION_RECALL_STATE:
2819 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2820 break;
2821
186e4f7a
YZ		 2822 case CEPH_SESSION_FLUSHMSG:
2823 send_flushmsg_ack(mdsc, session, seq);
2824 break;
2825
03f4fcb0
YZ		 2826 case CEPH_SESSION_FORCE_RO:
2827 dout("force_session_readonly %p\n", session);
2828 spin_lock(&session->s_cap_lock);
2829 session->s_readonly = true;
2830 spin_unlock(&session->s_cap_lock);
2831 wake_up_session_caps(session, 0);
2832 break;
2833
fcff415c
YZ		 2834 case CEPH_SESSION_REJECT:
2835 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2836 pr_info("mds%d rejected session\n", session->s_mds);
2837 session->s_state = CEPH_MDS_SESSION_REJECTED;
2838 cleanup_session_requests(mdsc, session);
2839 remove_session_caps(session);
2840 wake = 2; /* for good measure */
2841 break;
2842
2f2dc053
SW		 2843 default:
2844 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2845 WARN_ON(1);
2846 }
2847
2848 mutex_unlock(&session->s_mutex);
2849 if (wake) {
2850 mutex_lock(&mdsc->mutex);
2851 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2852 if (wake == 2)
2853 kick_requests(mdsc, mds);
2f2dc053
SW		 2854 mutex_unlock(&mdsc->mutex);
2855 }
0a07fc8c
YZ		 2856 if (op == CEPH_SESSION_CLOSE)
2857 ceph_put_mds_session(session);
2f2dc053
SW		 2858 return;
2859
2860bad:
2861 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2862 (int)msg->front.iov_len);
9ec7cab1 2863 ceph_msg_dump(msg);
2f2dc053
SW		 2864 return;
2865}
2866
2867
2868/*
2869 * called under session->mutex.
2870 */
2871static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2872 struct ceph_mds_session *session)
2873{
2874 struct ceph_mds_request *req, *nreq;
3de22be6 2875 struct rb_node *p;
2f2dc053
SW		 2876 int err;
2877
2878 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2879
2880 mutex_lock(&mdsc->mutex);
2881 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2882 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2883 if (!err) {
2884 ceph_msg_get(req->r_request);
2885 ceph_con_send(&session->s_con, req->r_request);
2886 }
2887 }
3de22be6
YZ		 2888
2889 /*
2890 * also re-send old requests when MDS enters reconnect stage. So that MDS
2891 * can process completed request in clientreplay stage.
2892 */
2893 p = rb_first(&mdsc->request_tree);
2894 while (p) {
2895 req = rb_entry(p, struct ceph_mds_request, r_node);
2896 p = rb_next(p);
bc2de10d 2897 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3de22be6
YZ		 2898 continue;
2899 if (req->r_attempts == 0)
2900 continue; /* only old requests */
2901 if (req->r_session &&
2902 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2903 err = __prepare_send_request(mdsc, req,
2904 session->s_mds, true);
3de22be6
YZ		 2905 if (!err) {
2906 ceph_msg_get(req->r_request);
2907 ceph_con_send(&session->s_con, req->r_request);
2908 }
2909 }
2910 }
2f2dc053
SW		 2911 mutex_unlock(&mdsc->mutex);
2912}
2913
2914/*
2915 * Encode information about a cap for a reconnect with the MDS.
2916 */
2f2dc053
SW		 2917static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2918 void *arg)
2919{
20cb34ae
SW		 2920 union {
2921 struct ceph_mds_cap_reconnect v2;
2922 struct ceph_mds_cap_reconnect_v1 v1;
2923 } rec;
b3f8d68f 2924 struct ceph_inode_info *ci = cap->ci;
20cb34ae
SW		 2925 struct ceph_reconnect_state *recon_state = arg;
2926 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2927 char *path;
2928 int pathlen, err;
2929 u64 pathbase;
3469ed0d 2930 u64 snap_follows;
2f2dc053
SW		 2931 struct dentry *dentry;
2932
2f2dc053
SW		 2933 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2934 inode, ceph_vinop(inode), cap, cap->cap_id,
2935 ceph_cap_string(cap->issued));
93cea5be
SW		 2936 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2937 if (err)
2938 return err;
2f2dc053
SW		 2939
2940 dentry = d_find_alias(inode);
2941 if (dentry) {
2942 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2943 if (IS_ERR(path)) {
2944 err = PTR_ERR(path);
e072f8aa 2945 goto out_dput;
2f2dc053
SW		 2946 }
2947 } else {
2948 path = NULL;
2949 pathlen = 0;
4eacd4cb 2950 pathbase = 0;
2f2dc053 2951 }
2f2dc053 2952
be655596 2953 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2954 cap->seq = 0; /* reset cap seq */
2955 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2956 cap->mseq = 0; /* and migrate_seq */
99a9c273 2957 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2958
121f22a1 2959 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2960 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2961 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2962 rec.v2.issued = cpu_to_le32(cap->issued);
2963 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2964 rec.v2.pathbase = cpu_to_le64(pathbase);
ec1dff25
JL		 2965 rec.v2.flock_len = (__force __le32)
2966 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
20cb34ae
SW		 2967 } else {
2968 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2969 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2970 rec.v1.issued = cpu_to_le32(cap->issued);
2971 rec.v1.size = cpu_to_le64(inode->i_size);
2972 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2973 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2974 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2975 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2976 }
3469ed0d
YZ		 2977
2978 if (list_empty(&ci->i_cap_snaps)) {
92776fd2 2979 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3469ed0d
YZ		 2980 } else {
2981 struct ceph_cap_snap *capsnap =
2982 list_first_entry(&ci->i_cap_snaps,
2983 struct ceph_cap_snap, ci_item);
2984 snap_follows = capsnap->follows;
20cb34ae 2985 }
be655596 2986 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2987
121f22a1 2988 if (recon_state->msg_version >= 2) {
40819f6f 2989 int num_fcntl_locks, num_flock_locks;
4deb14a2 2990 struct ceph_filelock *flocks = NULL;
121f22a1
YZ		 2991 size_t struct_len, total_len = 0;
2992 u8 struct_v = 0;
39be95e9
JS		 2993
2994encode_again:
b3f8d68f
YZ		 2995 if (rec.v2.flock_len) {
2996 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2997 } else {
2998 num_fcntl_locks = 0;
2999 num_flock_locks = 0;
3000 }
4deb14a2
YZ		 3001 if (num_fcntl_locks + num_flock_locks > 0) {
3002 flocks = kmalloc((num_fcntl_locks + num_flock_locks) *
3003 sizeof(struct ceph_filelock), GFP_NOFS);
3004 if (!flocks) {
3005 err = -ENOMEM;
3006 goto out_free;
3007 }
3008 err = ceph_encode_locks_to_buffer(inode, flocks,
3009 num_fcntl_locks,
3010 num_flock_locks);
3011 if (err) {
3012 kfree(flocks);
3013 flocks = NULL;
3014 if (err == -ENOSPC)
3015 goto encode_again;
3016 goto out_free;
3017 }
3018 } else {
39be95e9 3019 kfree(flocks);
4deb14a2 3020 flocks = NULL;
39be95e9 3021 }
121f22a1
YZ		 3022
3023 if (recon_state->msg_version >= 3) {
3024 /* version, compat_version and struct_len */
3025 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 3026 struct_v = 2;
121f22a1 3027 }
39be95e9
JS		 3028 /*
3029 * number of encoded locks is stable, so copy to pagelist
3030 */
121f22a1
YZ		 3031 struct_len = 2 * sizeof(u32) +
3032 (num_fcntl_locks + num_flock_locks) *
3033 sizeof(struct ceph_filelock);
3034 rec.v2.flock_len = cpu_to_le32(struct_len);
3035
3036 struct_len += sizeof(rec.v2);
3037 struct_len += sizeof(u32) + pathlen;
3038
3469ed0d
YZ		 3039 if (struct_v >= 2)
3040 struct_len += sizeof(u64); /* snap_follows */
3041
121f22a1
YZ		 3042 total_len += struct_len;
3043 err = ceph_pagelist_reserve(pagelist, total_len);
3044
3045 if (!err) {
3046 if (recon_state->msg_version >= 3) {
3047 ceph_pagelist_encode_8(pagelist, struct_v);
3048 ceph_pagelist_encode_8(pagelist, 1);
3049 ceph_pagelist_encode_32(pagelist, struct_len);
3050 }
3051 ceph_pagelist_encode_string(pagelist, path, pathlen);
3052 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3053 ceph_locks_to_pagelist(flocks, pagelist,
3054 num_fcntl_locks,
3055 num_flock_locks);
3469ed0d
YZ		 3056 if (struct_v >= 2)
3057 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 3058 }
39be95e9 3059 kfree(flocks);
3612abbd 3060 } else {
121f22a1
YZ		 3061 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
3062 err = ceph_pagelist_reserve(pagelist, size);
3063 if (!err) {
3064 ceph_pagelist_encode_string(pagelist, path, pathlen);
3065 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3066 }
40819f6f 3067 }
44c99757
YZ		 3068
3069 recon_state->nr_caps++;
e072f8aa 3070out_free:
2f2dc053 3071 kfree(path);
e072f8aa 3072out_dput:
2f2dc053 3073 dput(dentry);
93cea5be 3074 return err;
2f2dc053
SW		 3075}
3076
3077
3078/*
3079 * If an MDS fails and recovers, clients need to reconnect in order to
3080 * reestablish shared state. This includes all caps issued through
3081 * this session _and_ the snap_realm hierarchy. Because it's not
3082 * clear which snap realms the mds cares about, we send everything we
3083 * know about.. that ensures we'll then get any new info the
3084 * recovering MDS might have.
3085 *
3086 * This is a relatively heavyweight operation, but it's rare.
3087 *
3088 * called with mdsc->mutex held.
3089 */
34b6c855
SW		 3090static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3091 struct ceph_mds_session *session)
2f2dc053 3092{
2f2dc053 3093 struct ceph_msg *reply;
a105f00c 3094 struct rb_node *p;
34b6c855 3095 int mds = session->s_mds;
9abf82b8 3096 int err = -ENOMEM;
44c99757 3097 int s_nr_caps;
93cea5be 3098 struct ceph_pagelist *pagelist;
20cb34ae 3099 struct ceph_reconnect_state recon_state;
c8a96a31 3100 LIST_HEAD(dispose);
2f2dc053 3101
34b6c855 3102 pr_info("mds%d reconnect start\n", mds);
2f2dc053 3103
93cea5be
SW		 3104 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
3105 if (!pagelist)
3106 goto fail_nopagelist;
3107 ceph_pagelist_init(pagelist);
3108
b61c2763 3109 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 3110 if (!reply)
93cea5be 3111 goto fail_nomsg;
93cea5be 3112
34b6c855
SW		 3113 mutex_lock(&session->s_mutex);
3114 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3115 session->s_seq = 0;
2f2dc053 3116
2f2dc053 3117 dout("session %p state %s\n", session,
a687ecaf 3118 ceph_session_state_name(session->s_state));
2f2dc053 3119
99a9c273
YZ		 3120 spin_lock(&session->s_gen_ttl_lock);
3121 session->s_cap_gen++;
3122 spin_unlock(&session->s_gen_ttl_lock);
3123
3124 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 3125 /* don't know if session is readonly */
3126 session->s_readonly = 0;
99a9c273
YZ		 3127 /*
3128 * notify __ceph_remove_cap() that we are composing cap reconnect.
3129 * If a cap get released before being added to the cap reconnect,
3130 * __ceph_remove_cap() should skip queuing cap release.
3131 */
3132 session->s_cap_reconnect = 1;
e01a5946 3133 /* drop old cap expires; we're about to reestablish that state */
c8a96a31
JL		 3134 detach_cap_releases(session, &dispose);
3135 spin_unlock(&session->s_cap_lock);
3136 dispose_cap_releases(mdsc, &dispose);
e01a5946 3137
5d23371f 3138 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 3139 if (mdsc->fsc->sb->s_root)
3140 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 3141
3142 ceph_con_close(&session->s_con);
3143 ceph_con_open(&session->s_con,
3144 CEPH_ENTITY_TYPE_MDS, mds,
3145 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3146
3147 /* replay unsafe requests */
3148 replay_unsafe_requests(mdsc, session);
3149
3150 down_read(&mdsc->snap_rwsem);
3151
2f2dc053 3152 /* traverse this session's caps */
44c99757
YZ		 3153 s_nr_caps = session->s_nr_caps;
3154 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 3155 if (err)
3156 goto fail;
20cb34ae 3157
44c99757 3158 recon_state.nr_caps = 0;
20cb34ae 3159 recon_state.pagelist = pagelist;
121f22a1
YZ		 3160 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
3161 recon_state.msg_version = 3;
3162 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
3163 recon_state.msg_version = 2;
3164 else
3165 recon_state.msg_version = 1;
20cb34ae 3166 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 3167 if (err < 0)
9abf82b8 3168 goto fail;
2f2dc053 3169
99a9c273
YZ		 3170 spin_lock(&session->s_cap_lock);
3171 session->s_cap_reconnect = 0;
3172 spin_unlock(&session->s_cap_lock);
3173
2f2dc053
SW		 3174 /*
3175 * snaprealms. we provide mds with the ino, seq (version), and
3176 * parent for all of our realms. If the mds has any newer info,
3177 * it will tell us.
3178 */
a105f00c
SW		 3179 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3180 struct ceph_snap_realm *realm =
3181 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3182 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3183
3184 dout(" adding snap realm %llx seq %lld parent %llx\n",
3185 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3186 sr_rec.ino = cpu_to_le64(realm->ino);
3187 sr_rec.seq = cpu_to_le64(realm->seq);
3188 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3189 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3190 if (err)
3191 goto fail;
2f2dc053 3192 }
2f2dc053 3193
121f22a1 3194 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3195
3196 /* raced with cap release? */
3197 if (s_nr_caps != recon_state.nr_caps) {
3198 struct page *page = list_first_entry(&pagelist->head,
3199 struct page, lru);
3200 __le32 *addr = kmap_atomic(page);
3201 *addr = cpu_to_le32(recon_state.nr_caps);
3202 kunmap_atomic(addr);
ebf18f47 3203 }
44c99757
YZ		 3204
3205 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3206 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3207
3208 ceph_early_kick_flushing_caps(mdsc, session);
3209
2f2dc053
SW		 3210 ceph_con_send(&session->s_con, reply);
3211
9abf82b8
SW		 3212 mutex_unlock(&session->s_mutex);
3213
3214 mutex_lock(&mdsc->mutex);
3215 __wake_requests(mdsc, &session->s_waiting);
3216 mutex_unlock(&mdsc->mutex);
3217
2f2dc053 3218 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3219 return;
3220
93cea5be 3221fail:
2f2dc053 3222 ceph_msg_put(reply);
9abf82b8
SW		 3223 up_read(&mdsc->snap_rwsem);
3224 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3225fail_nomsg:
3226 ceph_pagelist_release(pagelist);
93cea5be 3227fail_nopagelist:
9abf82b8 3228 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3229 return;
2f2dc053
SW		 3230}
3231
3232
3233/*
3234 * compare old and new mdsmaps, kicking requests
3235 * and closing out old connections as necessary
3236 *
3237 * called under mdsc->mutex.
3238 */
3239static void check_new_map(struct ceph_mds_client *mdsc,
3240 struct ceph_mdsmap *newmap,
3241 struct ceph_mdsmap *oldmap)
3242{
3243 int i;
3244 int oldstate, newstate;
3245 struct ceph_mds_session *s;
3246
3247 dout("check_new_map new %u old %u\n",
3248 newmap->m_epoch, oldmap->m_epoch);
3249
76201b63 3250 for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
d37b1d99 3251 if (!mdsc->sessions[i])
2f2dc053
SW		 3252 continue;
3253 s = mdsc->sessions[i];
3254 oldstate = ceph_mdsmap_get_state(oldmap, i);
3255 newstate = ceph_mdsmap_get_state(newmap, i);
3256
0deb01c9 3257 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3258 i, ceph_mds_state_name(oldstate),
0deb01c9 3259 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3260 ceph_mds_state_name(newstate),
0deb01c9 3261 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3262 ceph_session_state_name(s->s_state));
2f2dc053 3263
76201b63 3264 if (i >= newmap->m_num_mds ||
3e8f43a0 3265 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3266 ceph_mdsmap_get_addr(newmap, i),
3267 sizeof(struct ceph_entity_addr))) {
3268 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3269 /* the session never opened, just close it
3270 * out now */
0a07fc8c 3271 get_session(s);
2600d2dd 3272 __unregister_session(mdsc, s);
2f2dc053 3273 __wake_requests(mdsc, &s->s_waiting);
0a07fc8c 3274 ceph_put_mds_session(s);
2827528d
YZ		 3275 } else if (i >= newmap->m_num_mds) {
3276 /* force close session for stopped mds */
3277 get_session(s);
2600d2dd 3278 __unregister_session(mdsc, s);
2827528d
YZ		 3279 __wake_requests(mdsc, &s->s_waiting);
3280 kick_requests(mdsc, i);
3281 mutex_unlock(&mdsc->mutex);
3282
3283 mutex_lock(&s->s_mutex);
3284 cleanup_session_requests(mdsc, s);
3285 remove_session_caps(s);
3286 mutex_unlock(&s->s_mutex);
3287
3288 ceph_put_mds_session(s);
3289
3290 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3291 } else {
3292 /* just close it */
3293 mutex_unlock(&mdsc->mutex);
3294 mutex_lock(&s->s_mutex);
3295 mutex_lock(&mdsc->mutex);
3296 ceph_con_close(&s->s_con);
3297 mutex_unlock(&s->s_mutex);
3298 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3299 }
2f2dc053
SW		 3300 } else if (oldstate == newstate) {
3301 continue; /* nothing new with this mds */
3302 }
3303
3304 /*
3305 * send reconnect?
3306 */
3307 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3308 newstate >= CEPH_MDS_STATE_RECONNECT) {
3309 mutex_unlock(&mdsc->mutex);
3310 send_mds_reconnect(mdsc, s);
3311 mutex_lock(&mdsc->mutex);
3312 }
2f2dc053
SW		 3313
3314 /*
29790f26 3315 * kick request on any mds that has gone active.
2f2dc053
SW		 3316 */
3317 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3318 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3319 if (oldstate != CEPH_MDS_STATE_CREATING &&
3320 oldstate != CEPH_MDS_STATE_STARTING)
3321 pr_info("mds%d recovery completed\n", s->s_mds);
3322 kick_requests(mdsc, i);
2f2dc053 3323 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3324 wake_up_session_caps(s, 1);
2f2dc053
SW		 3325 }
3326 }
cb170a22 3327
76201b63 3328 for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
cb170a22
SW		 3329 s = mdsc->sessions[i];
3330 if (!s)
3331 continue;
3332 if (!ceph_mdsmap_is_laggy(newmap, i))
3333 continue;
3334 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3335 s->s_state == CEPH_MDS_SESSION_HUNG ||
3336 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3337 dout(" connecting to export targets of laggy mds%d\n",
3338 i);
3339 __open_export_target_sessions(mdsc, s);
3340 }
3341 }
2f2dc053
SW		 3342}
3343
3344
3345
3346/*
3347 * leases
3348 */
3349
3350/*
3351 * caller must hold session s_mutex, dentry->d_lock
3352 */
3353void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3354{
3355 struct ceph_dentry_info *di = ceph_dentry(dentry);
3356
3357 ceph_put_mds_session(di->lease_session);
3358 di->lease_session = NULL;
3359}
3360
2600d2dd
SW		 3361static void handle_lease(struct ceph_mds_client *mdsc,
3362 struct ceph_mds_session *session,
3363 struct ceph_msg *msg)
2f2dc053 3364{
3d14c5d2 3365 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3366 struct inode *inode;
2f2dc053
SW		 3367 struct dentry *parent, *dentry;
3368 struct ceph_dentry_info *di;
2600d2dd 3369 int mds = session->s_mds;
2f2dc053 3370 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3371 u32 seq;
2f2dc053 3372 struct ceph_vino vino;
2f2dc053
SW		 3373 struct qstr dname;
3374 int release = 0;
3375
2f2dc053
SW		 3376 dout("handle_lease from mds%d\n", mds);
3377
3378 /* decode */
3379 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3380 goto bad;
3381 vino.ino = le64_to_cpu(h->ino);
3382 vino.snap = CEPH_NOSNAP;
1e5ea23d 3383 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3384 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3385 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3386 if (dname.len != get_unaligned_le32(h+1))
3387 goto bad;
3388
2f2dc053
SW		 3389 /* lookup inode */
3390 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3391 dout("handle_lease %s, ino %llx %p %.*s\n",
3392 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3393 dname.len, dname.name);
6cd3bcad
YZ		 3394
3395 mutex_lock(&session->s_mutex);
3396 session->s_seq++;
3397
d37b1d99 3398 if (!inode) {
2f2dc053
SW		 3399 dout("handle_lease no inode %llx\n", vino.ino);
3400 goto release;
3401 }
2f2dc053
SW		 3402
3403 /* dentry */
3404 parent = d_find_alias(inode);
3405 if (!parent) {
3406 dout("no parent dentry on inode %p\n", inode);
3407 WARN_ON(1);
3408 goto release; /* hrm... */
3409 }
8387ff25 3410 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3411 dentry = d_lookup(parent, &dname);
3412 dput(parent);
3413 if (!dentry)
3414 goto release;
3415
3416 spin_lock(&dentry->d_lock);
3417 di = ceph_dentry(dentry);
3418 switch (h->action) {
3419 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3420 if (di->lease_session == session) {
1e5ea23d
SW		 3421 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3422 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3423 __ceph_mdsc_drop_dentry_lease(dentry);
3424 }
3425 release = 1;
3426 break;
3427
3428 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3429 if (di->lease_session == session &&
2f2dc053
SW		 3430 di->lease_gen == session->s_cap_gen &&
3431 di->lease_renew_from &&
3432 di->lease_renew_after == 0) {
3433 unsigned long duration =
3563dbdd 3434 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3435
1e5ea23d 3436 di->lease_seq = seq;
9b16f03c 3437 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3438 di->lease_renew_after = di->lease_renew_from +
3439 (duration >> 1);
3440 di->lease_renew_from = 0;
3441 }
3442 break;
3443 }
3444 spin_unlock(&dentry->d_lock);
3445 dput(dentry);
3446
3447 if (!release)
3448 goto out;
3449
3450release:
3451 /* let's just reuse the same message */
3452 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3453 ceph_msg_get(msg);
3454 ceph_con_send(&session->s_con, msg);
3455
3456out:
3457 iput(inode);
3458 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3459 return;
3460
3461bad:
3462 pr_err("corrupt lease message\n");
9ec7cab1 3463 ceph_msg_dump(msg);
2f2dc053
SW		 3464}
3465
3466void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3467 struct inode *inode,
3468 struct dentry *dentry, char action,
3469 u32 seq)
3470{
3471 struct ceph_msg *msg;
3472 struct ceph_mds_lease *lease;
3473 int len = sizeof(*lease) + sizeof(u32);
3474 int dnamelen = 0;
3475
3476 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3477 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3478 dnamelen = dentry->d_name.len;
3479 len += dnamelen;
3480
b61c2763 3481 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3482 if (!msg)
2f2dc053
SW		 3483 return;
3484 lease = msg->front.iov_base;
3485 lease->action = action;
2f2dc053
SW		 3486 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3487 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3488 lease->seq = cpu_to_le32(seq);
3489 put_unaligned_le32(dnamelen, lease + 1);
3490 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3491
3492 /*
3493 * if this is a preemptive lease RELEASE, no need to
3494 * flush request stream, since the actual request will
3495 * soon follow.
3496 */
3497 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3498
3499 ceph_con_send(&session->s_con, msg);
3500}
3501
2f2dc053
SW		 3502/*
3503 * drop all leases (and dentry refs) in preparation for umount
3504 */
3505static void drop_leases(struct ceph_mds_client *mdsc)
3506{
3507 int i;
3508
3509 dout("drop_leases\n");
3510 mutex_lock(&mdsc->mutex);
3511 for (i = 0; i < mdsc->max_sessions; i++) {
3512 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3513 if (!s)
3514 continue;
3515 mutex_unlock(&mdsc->mutex);
3516 mutex_lock(&s->s_mutex);
3517 mutex_unlock(&s->s_mutex);
3518 ceph_put_mds_session(s);
3519 mutex_lock(&mdsc->mutex);
3520 }
3521 mutex_unlock(&mdsc->mutex);
3522}
3523
3524
3525
3526/*
3527 * delayed work -- periodically trim expired leases, renew caps with mds
3528 */
3529static void schedule_delayed(struct ceph_mds_client *mdsc)
3530{
3531 int delay = 5;
3532 unsigned hz = round_jiffies_relative(HZ * delay);
3533 schedule_delayed_work(&mdsc->delayed_work, hz);
3534}
3535
3536static void delayed_work(struct work_struct *work)
3537{
3538 int i;
3539 struct ceph_mds_client *mdsc =
3540 container_of(work, struct ceph_mds_client, delayed_work.work);
3541 int renew_interval;
3542 int renew_caps;
3543
3544 dout("mdsc delayed_work\n");
afcdaea3 3545 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3546
3547 mutex_lock(&mdsc->mutex);
3548 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3549 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3550 mdsc->last_renew_caps);
3551 if (renew_caps)
3552 mdsc->last_renew_caps = jiffies;
3553
3554 for (i = 0; i < mdsc->max_sessions; i++) {
3555 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
d37b1d99 3556 if (!s)
2f2dc053
SW		 3557 continue;
3558 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3559 dout("resending session close request for mds%d\n",
3560 s->s_mds);
3561 request_close_session(mdsc, s);
3562 ceph_put_mds_session(s);
3563 continue;
3564 }
3565 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3566 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3567 s->s_state = CEPH_MDS_SESSION_HUNG;
3568 pr_info("mds%d hung\n", s->s_mds);
3569 }
3570 }
3571 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3572 /* this mds is failed or recovering, just wait */
3573 ceph_put_mds_session(s);
3574 continue;
3575 }
3576 mutex_unlock(&mdsc->mutex);
3577
3578 mutex_lock(&s->s_mutex);
3579 if (renew_caps)
3580 send_renew_caps(mdsc, s);
3581 else
3582 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3583 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3584 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3585 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3586 mutex_unlock(&s->s_mutex);
3587 ceph_put_mds_session(s);
3588
3589 mutex_lock(&mdsc->mutex);
3590 }
3591 mutex_unlock(&mdsc->mutex);
3592
3593 schedule_delayed(mdsc);
3594}
3595
3d14c5d2 3596int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3597
2f2dc053 3598{
3d14c5d2
YS		 3599 struct ceph_mds_client *mdsc;
3600
3601 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3602 if (!mdsc)
3603 return -ENOMEM;
3604 mdsc->fsc = fsc;
3605 fsc->mdsc = mdsc;
2f2dc053
SW		 3606 mutex_init(&mdsc->mutex);
3607 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
d37b1d99 3608 if (!mdsc->mdsmap) {
fb3101b6 3609 kfree(mdsc);
2d06eeb8 3610 return -ENOMEM;
fb3101b6 3611 }
2d06eeb8 3612
2f2dc053 3613 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3614 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3615 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3616 mdsc->sessions = NULL;
86d8f67b 3617 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3618 mdsc->max_sessions = 0;
3619 mdsc->stopping = 0;
affbc19a 3620 mdsc->last_snap_seq = 0;
2f2dc053 3621 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3622 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3623 INIT_LIST_HEAD(&mdsc->snap_empty);
3624 spin_lock_init(&mdsc->snap_empty_lock);
3625 mdsc->last_tid = 0;
e8a7b8b1 3626 mdsc->oldest_tid = 0;
44ca18f2 3627 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3628 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3629 mdsc->last_renew_caps = jiffies;
3630 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3631 spin_lock_init(&mdsc->cap_delay_lock);
3632 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3633 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3634 mdsc->last_cap_flush_tid = 1;
e4500b5e 3635 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3636 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3637 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3638 mdsc->num_cap_flushing = 0;
3639 spin_lock_init(&mdsc->cap_dirty_lock);
3640 init_waitqueue_head(&mdsc->cap_flushing_wq);
3641 spin_lock_init(&mdsc->dentry_lru_lock);
3642 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3643
37151668 3644 ceph_caps_init(mdsc);
3d14c5d2 3645 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3646
10183a69
YZ		 3647 init_rwsem(&mdsc->pool_perm_rwsem);
3648 mdsc->pool_perm_tree = RB_ROOT;
3649
717e6f28
YZ		 3650 strncpy(mdsc->nodename, utsname()->nodename,
3651 sizeof(mdsc->nodename) - 1);
5f44f142 3652 return 0;
2f2dc053
SW		 3653}
3654
3655/*
3656 * Wait for safe replies on open mds requests. If we time out, drop
3657 * all requests from the tree to avoid dangling dentry refs.
3658 */
3659static void wait_requests(struct ceph_mds_client *mdsc)
3660{
a319bf56 3661 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3662 struct ceph_mds_request *req;
2f2dc053
SW		 3663
3664 mutex_lock(&mdsc->mutex);
44ca18f2 3665 if (__get_oldest_req(mdsc)) {
2f2dc053 3666 mutex_unlock(&mdsc->mutex);
44ca18f2 3667
2f2dc053
SW		 3668 dout("wait_requests waiting for requests\n");
3669 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3670 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3671
3672 /* tear down remaining requests */
44ca18f2
SW		 3673 mutex_lock(&mdsc->mutex);
3674 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3675 dout("wait_requests timed out on tid %llu\n",
3676 req->r_tid);
44ca18f2 3677 __unregister_request(mdsc, req);
2f2dc053
SW		 3678 }
3679 }
3680 mutex_unlock(&mdsc->mutex);
3681 dout("wait_requests done\n");
3682}
3683
3684/*
3685 * called before mount is ro, and before dentries are torn down.
3686 * (hmm, does this still race with new lookups?)
3687 */
3688void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3689{
3690 dout("pre_umount\n");
3691 mdsc->stopping = 1;
3692
3693 drop_leases(mdsc);
afcdaea3 3694 ceph_flush_dirty_caps(mdsc);
2f2dc053 3695 wait_requests(mdsc);
17c688c3
SW		 3696
3697 /*
3698 * wait for reply handlers to drop their request refs and
3699 * their inode/dcache refs
3700 */
3701 ceph_msgr_flush();
2f2dc053
SW		 3702}
3703
3704/*
3705 * wait for all write mds requests to flush.
3706 */
3707static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3708{
80fc7314 3709 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3710 struct rb_node *n;
2f2dc053
SW		 3711
3712 mutex_lock(&mdsc->mutex);
3713 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3714restart:
44ca18f2
SW		 3715 req = __get_oldest_req(mdsc);
3716 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3717 /* find next request */
3718 n = rb_next(&req->r_node);
3719 if (n)
3720 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3721 else
3722 nextreq = NULL;
e8a7b8b1
YZ		 3723 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3724 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3725 /* write op */
3726 ceph_mdsc_get_request(req);
80fc7314
SW		 3727 if (nextreq)
3728 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3729 mutex_unlock(&mdsc->mutex);
3730 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3731 req->r_tid, want_tid);
3732 wait_for_completion(&req->r_safe_completion);
3733 mutex_lock(&mdsc->mutex);
44ca18f2 3734 ceph_mdsc_put_request(req);
80fc7314
SW		 3735 if (!nextreq)
3736 break; /* next dne before, so we're done! */
3737 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3738 /* next request was removed from tree */
3739 ceph_mdsc_put_request(nextreq);
3740 goto restart;
3741 }
3742 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3743 }
80fc7314 3744 req = nextreq;
2f2dc053
SW		 3745 }
3746 mutex_unlock(&mdsc->mutex);
3747 dout("wait_unsafe_requests done\n");
3748}
3749
3750void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3751{
0e294387 3752 u64 want_tid, want_flush;
2f2dc053 3753
52953d55 3754 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3755 return;
3756
2f2dc053
SW		 3757 dout("sync\n");
3758 mutex_lock(&mdsc->mutex);
3759 want_tid = mdsc->last_tid;
2f2dc053 3760 mutex_unlock(&mdsc->mutex);
2f2dc053 3761
afcdaea3 3762 ceph_flush_dirty_caps(mdsc);
d3383a8e 3763 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3764 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3765 if (!list_empty(&mdsc->cap_flush_list)) {
3766 struct ceph_cap_flush *cf =
3767 list_last_entry(&mdsc->cap_flush_list,
3768 struct ceph_cap_flush, g_list);
3769 cf->wake = true;
3770 }
d3383a8e
YZ		 3771 spin_unlock(&mdsc->cap_dirty_lock);
3772
0e294387
YZ		 3773 dout("sync want tid %lld flush_seq %lld\n",
3774 want_tid, want_flush);
2f2dc053
SW		 3775
3776 wait_unsafe_requests(mdsc, want_tid);
0e294387 3777 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3778}
3779
f3c60c59
SW		 3780/*
3781 * true if all sessions are closed, or we force unmount
3782 */
fcff415c 3783static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3784{
52953d55 3785 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3786 return true;
fcff415c 3787 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3788}
2f2dc053
SW		 3789
3790/*
3791 * called after sb is ro.
3792 */
3793void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3794{
a319bf56 3795 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3796 struct ceph_mds_session *session;
3797 int i;
fcff415c 3798 int skipped = 0;
2f2dc053
SW		 3799
3800 dout("close_sessions\n");
3801
2f2dc053 3802 /* close sessions */
f3c60c59
SW		 3803 mutex_lock(&mdsc->mutex);
3804 for (i = 0; i < mdsc->max_sessions; i++) {
3805 session = __ceph_lookup_mds_session(mdsc, i);
3806 if (!session)
3807 continue;
2f2dc053 3808 mutex_unlock(&mdsc->mutex);
f3c60c59 3809 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3810 if (__close_session(mdsc, session) <= 0)
3811 skipped++;
f3c60c59
SW		 3812 mutex_unlock(&session->s_mutex);
3813 ceph_put_mds_session(session);
2f2dc053
SW		 3814 mutex_lock(&mdsc->mutex);
3815 }
f3c60c59
SW		 3816 mutex_unlock(&mdsc->mutex);
3817
3818 dout("waiting for sessions to close\n");
fcff415c
YZ		 3819 wait_event_timeout(mdsc->session_close_wq,
3820 done_closing_sessions(mdsc, skipped),
a319bf56 3821 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3822
3823 /* tear down remaining sessions */
f3c60c59 3824 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3825 for (i = 0; i < mdsc->max_sessions; i++) {
3826 if (mdsc->sessions[i]) {
3827 session = get_session(mdsc->sessions[i]);
2600d2dd 3828 __unregister_session(mdsc, session);
2f2dc053
SW		 3829 mutex_unlock(&mdsc->mutex);
3830 mutex_lock(&session->s_mutex);
3831 remove_session_caps(session);
3832 mutex_unlock(&session->s_mutex);
3833 ceph_put_mds_session(session);
3834 mutex_lock(&mdsc->mutex);
3835 }
3836 }
2f2dc053 3837 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3838 mutex_unlock(&mdsc->mutex);
3839
3840 ceph_cleanup_empty_realms(mdsc);
3841
3842 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3843
3844 dout("stopped\n");
3845}
3846
48fec5d0
YZ		 3847void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3848{
3849 struct ceph_mds_session *session;
3850 int mds;
3851
3852 dout("force umount\n");
3853
3854 mutex_lock(&mdsc->mutex);
3855 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3856 session = __ceph_lookup_mds_session(mdsc, mds);
3857 if (!session)
3858 continue;
3859 mutex_unlock(&mdsc->mutex);
3860 mutex_lock(&session->s_mutex);
3861 __close_session(mdsc, session);
3862 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3863 cleanup_session_requests(mdsc, session);
3864 remove_session_caps(session);
3865 }
3866 mutex_unlock(&session->s_mutex);
3867 ceph_put_mds_session(session);
3868 mutex_lock(&mdsc->mutex);
3869 kick_requests(mdsc, mds);
3870 }
3871 __wake_requests(mdsc, &mdsc->waiting_for_map);
3872 mutex_unlock(&mdsc->mutex);
3873}
3874
3d14c5d2 3875static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3876{
3877 dout("stop\n");
3878 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3879 if (mdsc->mdsmap)
3880 ceph_mdsmap_destroy(mdsc->mdsmap);
3881 kfree(mdsc->sessions);
37151668 3882 ceph_caps_finalize(mdsc);
10183a69 3883 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3884}
3885
3d14c5d2
YS		 3886void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3887{
3888 struct ceph_mds_client *mdsc = fsc->mdsc;
ef550f6f 3889 dout("mdsc_destroy %p\n", mdsc);
ef550f6f
SW		 3890
3891 /* flush out any connection work with references to us */
3892 ceph_msgr_flush();
3893
62a65f36
YZ		 3894 ceph_mdsc_stop(mdsc);
3895
3d14c5d2
YS		 3896 fsc->mdsc = NULL;
3897 kfree(mdsc);
ef550f6f 3898 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3899}
3900
430afbad
YZ		 3901void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3902{
3903 struct ceph_fs_client *fsc = mdsc->fsc;
3904 const char *mds_namespace = fsc->mount_options->mds_namespace;
3905 void *p = msg->front.iov_base;
3906 void *end = p + msg->front.iov_len;
3907 u32 epoch;
3908 u32 map_len;
3909 u32 num_fs;
3910 u32 mount_fscid = (u32)-1;
3911 u8 struct_v, struct_cv;
3912 int err = -EINVAL;
3913
3914 ceph_decode_need(&p, end, sizeof(u32), bad);
3915 epoch = ceph_decode_32(&p);
3916
3917 dout("handle_fsmap epoch %u\n", epoch);
3918
3919 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3920 struct_v = ceph_decode_8(&p);
3921 struct_cv = ceph_decode_8(&p);
3922 map_len = ceph_decode_32(&p);
3923
3924 ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
3925 p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
3926
3927 num_fs = ceph_decode_32(&p);
3928 while (num_fs-- > 0) {
3929 void *info_p, *info_end;
3930 u32 info_len;
3931 u8 info_v, info_cv;
3932 u32 fscid, namelen;
3933
3934 ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
3935 info_v = ceph_decode_8(&p);
3936 info_cv = ceph_decode_8(&p);
3937 info_len = ceph_decode_32(&p);
3938 ceph_decode_need(&p, end, info_len, bad);
3939 info_p = p;
3940 info_end = p + info_len;
3941 p = info_end;
3942
3943 ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
3944 fscid = ceph_decode_32(&info_p);
3945 namelen = ceph_decode_32(&info_p);
3946 ceph_decode_need(&info_p, info_end, namelen, bad);
3947
3948 if (mds_namespace &&
3949 strlen(mds_namespace) == namelen &&
3950 !strncmp(mds_namespace, (char *)info_p, namelen)) {
3951 mount_fscid = fscid;
3952 break;
3953 }
3954 }
3955
3956 ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
3957 if (mount_fscid != (u32)-1) {
3958 fsc->client->monc.fs_cluster_id = mount_fscid;
3959 ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
3960 0, true);
3961 ceph_monc_renew_subs(&fsc->client->monc);
3962 } else {
3963 err = -ENOENT;
3964 goto err_out;
3965 }
3966 return;
76bd6ec4 3967
430afbad
YZ		 3968bad:
3969 pr_err("error decoding fsmap\n");
3970err_out:
3971 mutex_lock(&mdsc->mutex);
76bd6ec4 3972 mdsc->mdsmap_err = err;
430afbad
YZ		 3973 __wake_requests(mdsc, &mdsc->waiting_for_map);
3974 mutex_unlock(&mdsc->mutex);
430afbad 3975}
2f2dc053
SW		 3976
3977/*
3978 * handle mds map update.
3979 */
430afbad 3980void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
2f2dc053
SW		 3981{
3982 u32 epoch;
3983 u32 maplen;
3984 void *p = msg->front.iov_base;
3985 void *end = p + msg->front.iov_len;
3986 struct ceph_mdsmap *newmap, *oldmap;
3987 struct ceph_fsid fsid;
3988 int err = -EINVAL;
3989
3990 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3991 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3992 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3993 return;
c89136ea
SW		 3994 epoch = ceph_decode_32(&p);
3995 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3996 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3997
3998 /* do we need it? */
2f2dc053
SW		 3999 mutex_lock(&mdsc->mutex);
4000 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
4001 dout("handle_map epoch %u <= our %u\n",
4002 epoch, mdsc->mdsmap->m_epoch);
4003 mutex_unlock(&mdsc->mutex);
4004 return;
4005 }
4006
4007 newmap = ceph_mdsmap_decode(&p, end);
4008 if (IS_ERR(newmap)) {
4009 err = PTR_ERR(newmap);
4010 goto bad_unlock;
4011 }
4012
4013 /* swap into place */
4014 if (mdsc->mdsmap) {
4015 oldmap = mdsc->mdsmap;
4016 mdsc->mdsmap = newmap;
4017 check_new_map(mdsc, newmap, oldmap);
4018 ceph_mdsmap_destroy(oldmap);
4019 } else {
4020 mdsc->mdsmap = newmap; /* first mds map */
4021 }
3d14c5d2 4022 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 4023
4024 __wake_requests(mdsc, &mdsc->waiting_for_map);
82dcabad
ID		 4025 ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
4026 mdsc->mdsmap->m_epoch);
2f2dc053
SW		 4027
4028 mutex_unlock(&mdsc->mutex);
4029 schedule_delayed(mdsc);
4030 return;
4031
4032bad_unlock:
4033 mutex_unlock(&mdsc->mutex);
4034bad:
4035 pr_err("error decoding mdsmap %d\n", err);
4036 return;
4037}
4038
4039static struct ceph_connection *con_get(struct ceph_connection *con)
4040{
4041 struct ceph_mds_session *s = con->private;
4042
4043 if (get_session(s)) {
3997c01d 4044 dout("mdsc con_get %p ok (%d)\n", s, refcount_read(&s->s_ref));
2f2dc053
SW		 4045 return con;
4046 }
4047 dout("mdsc con_get %p FAIL\n", s);
4048 return NULL;
4049}
4050
4051static void con_put(struct ceph_connection *con)
4052{
4053 struct ceph_mds_session *s = con->private;
4054
3997c01d 4055 dout("mdsc con_put %p (%d)\n", s, refcount_read(&s->s_ref) - 1);
2f2dc053
SW		 4056 ceph_put_mds_session(s);
4057}
4058
4059/*
4060 * if the client is unresponsive for long enough, the mds will kill
4061 * the session entirely.
4062 */
4063static void peer_reset(struct ceph_connection *con)
4064{
4065 struct ceph_mds_session *s = con->private;
7e70f0ed 4066 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 4067
f3ae1b97 4068 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 4069 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 4070}
4071
4072static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4073{
4074 struct ceph_mds_session *s = con->private;
4075 struct ceph_mds_client *mdsc = s->s_mdsc;
4076 int type = le16_to_cpu(msg->hdr.type);
4077
2600d2dd
SW		 4078 mutex_lock(&mdsc->mutex);
4079 if (__verify_registered_session(mdsc, s) < 0) {
4080 mutex_unlock(&mdsc->mutex);
4081 goto out;
4082 }
4083 mutex_unlock(&mdsc->mutex);
4084
2f2dc053
SW		 4085 switch (type) {
4086 case CEPH_MSG_MDS_MAP:
430afbad
YZ		 4087 ceph_mdsc_handle_mdsmap(mdsc, msg);
4088 break;
4089 case CEPH_MSG_FS_MAP_USER:
4090 ceph_mdsc_handle_fsmap(mdsc, msg);
2f2dc053
SW		 4091 break;
4092 case CEPH_MSG_CLIENT_SESSION:
4093 handle_session(s, msg);
4094 break;
4095 case CEPH_MSG_CLIENT_REPLY:
4096 handle_reply(s, msg);
4097 break;
4098 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 4099 handle_forward(mdsc, s, msg);
2f2dc053
SW		 4100 break;
4101 case CEPH_MSG_CLIENT_CAPS:
4102 ceph_handle_caps(s, msg);
4103 break;
4104 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 4105 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 4106 break;
4107 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 4108 handle_lease(mdsc, s, msg);
2f2dc053
SW		 4109 break;
4110
4111 default:
4112 pr_err("received unknown message type %d %s\n", type,
4113 ceph_msg_type_name(type));
4114 }
2600d2dd 4115out:
2f2dc053
SW		 4116 ceph_msg_put(msg);
4117}
4118
4e7a5dcd
SW		 4119/*
4120 * authentication
4121 */
a3530df3
AE		 4122
4123/*
4124 * Note: returned pointer is the address of a structure that's
4125 * managed separately. Caller must *not* attempt to free it.
4126 */
4127static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 4128 int *proto, int force_new)
4e7a5dcd
SW		 4129{
4130 struct ceph_mds_session *s = con->private;
4131 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4132 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 4133 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 4134
74f1869f 4135 if (force_new && auth->authorizer) {
6c1ea260 4136 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 4137 auth->authorizer = NULL;
4e7a5dcd 4138 }
27859f97
SW		 4139 if (!auth->authorizer) {
4140 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4141 auth);
0bed9b5c
SW		 4142 if (ret)
4143 return ERR_PTR(ret);
27859f97
SW		 4144 } else {
4145 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4146 auth);
a255651d 4147 if (ret)
a3530df3 4148 return ERR_PTR(ret);
4e7a5dcd 4149 }
4e7a5dcd 4150 *proto = ac->protocol;
74f1869f 4151
a3530df3 4152 return auth;
4e7a5dcd
SW		 4153}
4154
4155
0dde5848 4156static int verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 4157{
4158 struct ceph_mds_session *s = con->private;
4159 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4160 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 4161
0dde5848 4162 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4e7a5dcd
SW		 4163}
4164
9bd2e6f8
SW		 4165static int invalidate_authorizer(struct ceph_connection *con)
4166{
4167 struct ceph_mds_session *s = con->private;
4168 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4169 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 4170
27859f97 4171 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 4172
3d14c5d2 4173 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 4174}
4175
53ded495
AE		 4176static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4177 struct ceph_msg_header *hdr, int *skip)
4178{
4179 struct ceph_msg *msg;
4180 int type = (int) le16_to_cpu(hdr->type);
4181 int front_len = (int) le32_to_cpu(hdr->front_len);
4182
4183 if (con->in_msg)
4184 return con->in_msg;
4185
4186 *skip = 0;
4187 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4188 if (!msg) {
4189 pr_err("unable to allocate msg type %d len %d\n",
4190 type, front_len);
4191 return NULL;
4192 }
53ded495
AE		 4193
4194 return msg;
4195}
4196
79dbd1ba 4197static int mds_sign_message(struct ceph_msg *msg)
33d07337 4198{
79dbd1ba 4199 struct ceph_mds_session *s = msg->con->private;
33d07337 4200 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4201
33d07337
YZ		 4202 return ceph_auth_sign_message(auth, msg);
4203}
4204
79dbd1ba 4205static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 4206{
79dbd1ba 4207 struct ceph_mds_session *s = msg->con->private;
33d07337 4208 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4209
33d07337
YZ		 4210 return ceph_auth_check_message_signature(auth, msg);
4211}
4212
9e32789f 4213static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4214 .get = con_get,
4215 .put = con_put,
4216 .dispatch = dispatch,
4e7a5dcd
SW		 4217 .get_authorizer = get_authorizer,
4218 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4219 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4220 .peer_reset = peer_reset,
53ded495 4221 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4222 .sign_message = mds_sign_message,
4223 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4224};
4225
2f2dc053 4226/* eof */
ubuntu-bionic-kernel mirror