projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: use timespec64 for r_mtime
[linux-block.git] / fs / ceph / mds_client.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
3d14c5d2 2#include <linux/ceph/ceph_debug.h>
2f2dc053 3
496e5955 4#include <linux/fs.h>
2f2dc053 5#include <linux/wait.h>
5a0e3ad6 6#include <linux/slab.h>
54008399 7#include <linux/gfp.h>
2f2dc053 8#include <linux/sched.h>
3d14c5d2
YS		 9#include <linux/debugfs.h>
10#include <linux/seq_file.h>
3e0708b9 11#include <linux/ratelimit.h>
2f2dc053 12
2f2dc053 13#include "super.h"
3d14c5d2
YS		 14#include "mds_client.h"
15
1fe60e51 16#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 17#include <linux/ceph/messenger.h>
18#include <linux/ceph/decode.h>
19#include <linux/ceph/pagelist.h>
20#include <linux/ceph/auth.h>
21#include <linux/ceph/debugfs.h>
2f2dc053
SW		 22
23/*
24 * A cluster of MDS (metadata server) daemons is responsible for
25 * managing the file system namespace (the directory hierarchy and
26 * inodes) and for coordinating shared access to storage. Metadata is
27 * partitioning hierarchically across a number of servers, and that
28 * partition varies over time as the cluster adjusts the distribution
29 * in order to balance load.
30 *
31 * The MDS client is primarily responsible to managing synchronous
32 * metadata requests for operations like open, unlink, and so forth.
33 * If there is a MDS failure, we find out about it when we (possibly
34 * request and) receive a new MDS map, and can resubmit affected
35 * requests.
36 *
37 * For the most part, though, we take advantage of a lossless
38 * communications channel to the MDS, and do not need to worry about
39 * timing out or resubmitting requests.
40 *
41 * We maintain a stateful "session" with each MDS we interact with.
42 * Within each session, we sent periodic heartbeat messages to ensure
43 * any capabilities or leases we have been issues remain valid. If
44 * the session times out and goes stale, our leases and capabilities
45 * are no longer valid.
46 */
47
20cb34ae 48struct ceph_reconnect_state {
44c99757 49 int nr_caps;
20cb34ae 50 struct ceph_pagelist *pagelist;
121f22a1 51 unsigned msg_version;
20cb34ae
SW		 52};
53
2f2dc053
SW		 54static void __wake_requests(struct ceph_mds_client *mdsc,
55 struct list_head *head);
56
9e32789f 57static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 58
59
60/*
61 * mds reply parsing
62 */
63
64/*
65 * parse individual inode info
66 */
67static int parse_reply_info_in(void **p, void *end,
14303d20 68 struct ceph_mds_reply_info_in *info,
12b4629a 69 u64 features)
2f2dc053
SW		 70{
71 int err = -EIO;
72
73 info->in = *p;
74 *p += sizeof(struct ceph_mds_reply_inode) +
75 sizeof(*info->in->fragtree.splits) *
76 le32_to_cpu(info->in->fragtree.nsplits);
77
78 ceph_decode_32_safe(p, end, info->symlink_len, bad);
79 ceph_decode_need(p, end, info->symlink_len, bad);
80 info->symlink = *p;
81 *p += info->symlink_len;
82
14303d20
SW		 83 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
84 ceph_decode_copy_safe(p, end, &info->dir_layout,
85 sizeof(info->dir_layout), bad);
86 else
87 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
88
2f2dc053
SW		 89 ceph_decode_32_safe(p, end, info->xattr_len, bad);
90 ceph_decode_need(p, end, info->xattr_len, bad);
91 info->xattr_data = *p;
92 *p += info->xattr_len;
fb01d1f8
YZ		 93
94 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
95 ceph_decode_64_safe(p, end, info->inline_version, bad);
96 ceph_decode_32_safe(p, end, info->inline_len, bad);
97 ceph_decode_need(p, end, info->inline_len, bad);
98 info->inline_data = *p;
99 *p += info->inline_len;
100 } else
101 info->inline_version = CEPH_INLINE_NONE;
102
fb18a575
LH		 103 if (features & CEPH_FEATURE_MDS_QUOTA) {
104 u8 struct_v, struct_compat;
105 u32 struct_len;
106
107 /*
108 * both struct_v and struct_compat are expected to be >= 1
109 */
110 ceph_decode_8_safe(p, end, struct_v, bad);
111 ceph_decode_8_safe(p, end, struct_compat, bad);
112 if (!struct_v || !struct_compat)
113 goto bad;
114 ceph_decode_32_safe(p, end, struct_len, bad);
115 ceph_decode_need(p, end, struct_len, bad);
116 ceph_decode_64_safe(p, end, info->max_bytes, bad);
117 ceph_decode_64_safe(p, end, info->max_files, bad);
118 } else {
119 info->max_bytes = 0;
120 info->max_files = 0;
121 }
122
779fe0fb
YZ		 123 info->pool_ns_len = 0;
124 info->pool_ns_data = NULL;
5ea5c5e0
YZ		 125 if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
126 ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
779fe0fb
YZ		 127 if (info->pool_ns_len > 0) {
128 ceph_decode_need(p, end, info->pool_ns_len, bad);
129 info->pool_ns_data = *p;
130 *p += info->pool_ns_len;
131 }
5ea5c5e0
YZ		 132 }
133
2f2dc053
SW		 134 return 0;
135bad:
136 return err;
137}
138
139/*
140 * parse a normal reply, which may contain a (dir+)dentry and/or a
141 * target inode.
142 */
143static int parse_reply_info_trace(void **p, void *end,
14303d20 144 struct ceph_mds_reply_info_parsed *info,
12b4629a 145 u64 features)
2f2dc053
SW		 146{
147 int err;
148
149 if (info->head->is_dentry) {
14303d20 150 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 151 if (err < 0)
152 goto out_bad;
153
154 if (unlikely(*p + sizeof(*info->dirfrag) > end))
155 goto bad;
156 info->dirfrag = *p;
157 *p += sizeof(*info->dirfrag) +
158 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
159 if (unlikely(*p > end))
160 goto bad;
161
162 ceph_decode_32_safe(p, end, info->dname_len, bad);
163 ceph_decode_need(p, end, info->dname_len, bad);
164 info->dname = *p;
165 *p += info->dname_len;
166 info->dlease = *p;
167 *p += sizeof(*info->dlease);
168 }
169
170 if (info->head->is_target) {
14303d20 171 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 172 if (err < 0)
173 goto out_bad;
174 }
175
176 if (unlikely(*p != end))
177 goto bad;
178 return 0;
179
180bad:
181 err = -EIO;
182out_bad:
183 pr_err("problem parsing mds trace %d\n", err);
184 return err;
185}
186
187/*
188 * parse readdir results
189 */
190static int parse_reply_info_dir(void **p, void *end,
14303d20 191 struct ceph_mds_reply_info_parsed *info,
12b4629a 192 u64 features)
2f2dc053
SW		 193{
194 u32 num, i = 0;
195 int err;
196
197 info->dir_dir = *p;
198 if (*p + sizeof(*info->dir_dir) > end)
199 goto bad;
200 *p += sizeof(*info->dir_dir) +
201 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
202 if (*p > end)
203 goto bad;
204
205 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea 206 num = ceph_decode_32(p);
956d39d6
YZ		 207 {
208 u16 flags = ceph_decode_16(p);
209 info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
210 info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
f3c4ebe6 211 info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
79162547 212 info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
956d39d6 213 }
2f2dc053
SW		 214 if (num == 0)
215 goto done;
216
2a5beea3
YZ		 217 BUG_ON(!info->dir_entries);
218 if ((unsigned long)(info->dir_entries + num) >
219 (unsigned long)info->dir_entries + info->dir_buf_size) {
54008399
YZ		 220 pr_err("dir contents are larger than expected\n");
221 WARN_ON(1);
222 goto bad;
223 }
2f2dc053 224
54008399 225 info->dir_nr = num;
2f2dc053 226 while (num) {
2a5beea3 227 struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
2f2dc053
SW		 228 /* dentry */
229 ceph_decode_need(p, end, sizeof(u32)*2, bad);
2a5beea3
YZ		 230 rde->name_len = ceph_decode_32(p);
231 ceph_decode_need(p, end, rde->name_len, bad);
232 rde->name = *p;
233 *p += rde->name_len;
234 dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
235 rde->lease = *p;
2f2dc053
SW		 236 *p += sizeof(struct ceph_mds_reply_lease);
237
238 /* inode */
2a5beea3 239 err = parse_reply_info_in(p, end, &rde->inode, features);
2f2dc053
SW		 240 if (err < 0)
241 goto out_bad;
8974eebd
YZ		 242 /* ceph_readdir_prepopulate() will update it */
243 rde->offset = 0;
2f2dc053
SW		 244 i++;
245 num--;
246 }
247
248done:
249 if (*p != end)
250 goto bad;
251 return 0;
252
253bad:
254 err = -EIO;
255out_bad:
256 pr_err("problem parsing dir contents %d\n", err);
257 return err;
258}
259
25933abd
HS		 260/*
261 * parse fcntl F_GETLK results
262 */
263static int parse_reply_info_filelock(void **p, void *end,
14303d20 264 struct ceph_mds_reply_info_parsed *info,
12b4629a 265 u64 features)
25933abd
HS		 266{
267 if (*p + sizeof(*info->filelock_reply) > end)
268 goto bad;
269
270 info->filelock_reply = *p;
271 *p += sizeof(*info->filelock_reply);
272
273 if (unlikely(*p != end))
274 goto bad;
275 return 0;
276
277bad:
278 return -EIO;
279}
280
6e8575fa
SL		 281/*
282 * parse create results
283 */
284static int parse_reply_info_create(void **p, void *end,
285 struct ceph_mds_reply_info_parsed *info,
12b4629a 286 u64 features)
6e8575fa
SL		 287{
288 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
289 if (*p == end) {
290 info->has_create_ino = false;
291 } else {
292 info->has_create_ino = true;
293 info->ino = ceph_decode_64(p);
294 }
295 }
296
297 if (unlikely(*p != end))
298 goto bad;
299 return 0;
300
301bad:
302 return -EIO;
303}
304
25933abd
HS		 305/*
306 * parse extra results
307 */
308static int parse_reply_info_extra(void **p, void *end,
14303d20 309 struct ceph_mds_reply_info_parsed *info,
12b4629a 310 u64 features)
25933abd 311{
6df8c9d8
JL		 312 u32 op = le32_to_cpu(info->head->op);
313
314 if (op == CEPH_MDS_OP_GETFILELOCK)
14303d20 315 return parse_reply_info_filelock(p, end, info, features);
6df8c9d8 316 else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
14303d20 317 return parse_reply_info_dir(p, end, info, features);
6df8c9d8 318 else if (op == CEPH_MDS_OP_CREATE)
6e8575fa
SL		 319 return parse_reply_info_create(p, end, info, features);
320 else
321 return -EIO;
25933abd
HS		 322}
323
2f2dc053
SW		 324/*
325 * parse entire mds reply
326 */
327static int parse_reply_info(struct ceph_msg *msg,
14303d20 328 struct ceph_mds_reply_info_parsed *info,
12b4629a 329 u64 features)
2f2dc053
SW		 330{
331 void *p, *end;
332 u32 len;
333 int err;
334
335 info->head = msg->front.iov_base;
336 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
337 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
338
339 /* trace */
340 ceph_decode_32_safe(&p, end, len, bad);
341 if (len > 0) {
32852a81 342 ceph_decode_need(&p, end, len, bad);
14303d20 343 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 344 if (err < 0)
345 goto out_bad;
346 }
347
25933abd 348 /* extra */
2f2dc053
SW		 349 ceph_decode_32_safe(&p, end, len, bad);
350 if (len > 0) {
32852a81 351 ceph_decode_need(&p, end, len, bad);
14303d20 352 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 353 if (err < 0)
354 goto out_bad;
355 }
356
357 /* snap blob */
358 ceph_decode_32_safe(&p, end, len, bad);
359 info->snapblob_len = len;
360 info->snapblob = p;
361 p += len;
362
363 if (p != end)
364 goto bad;
365 return 0;
366
367bad:
368 err = -EIO;
369out_bad:
370 pr_err("mds parse_reply err %d\n", err);
371 return err;
372}
373
374static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
375{
2a5beea3 376 if (!info->dir_entries)
54008399 377 return;
2a5beea3 378 free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
2f2dc053
SW		 379}
380
381
382/*
383 * sessions
384 */
a687ecaf 385const char *ceph_session_state_name(int s)
2f2dc053
SW		 386{
387 switch (s) {
388 case CEPH_MDS_SESSION_NEW: return "new";
389 case CEPH_MDS_SESSION_OPENING: return "opening";
390 case CEPH_MDS_SESSION_OPEN: return "open";
391 case CEPH_MDS_SESSION_HUNG: return "hung";
392 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 393 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053 394 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
fcff415c 395 case CEPH_MDS_SESSION_REJECTED: return "rejected";
2f2dc053
SW		 396 default: return "???";
397 }
398}
399
400static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
401{
3997c01d 402 if (refcount_inc_not_zero(&s->s_ref)) {
2f2dc053 403 dout("mdsc get_session %p %d -> %d\n", s,
3997c01d 404 refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
2f2dc053
SW		 405 return s;
406 } else {
4c069a58 407 dout("mdsc get_session %p 0 -- FAIL\n", s);
2f2dc053
SW		 408 return NULL;
409 }
410}
411
412void ceph_put_mds_session(struct ceph_mds_session *s)
413{
414 dout("mdsc put_session %p %d -> %d\n", s,
3997c01d
ER		 415 refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
416 if (refcount_dec_and_test(&s->s_ref)) {
6c4a1915 417 if (s->s_auth.authorizer)
6c1ea260 418 ceph_auth_destroy_authorizer(s->s_auth.authorizer);
2f2dc053 419 kfree(s);
4e7a5dcd 420 }
2f2dc053
SW		 421}
422
423/*
424 * called under mdsc->mutex
425 */
426struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
427 int mds)
428{
429 struct ceph_mds_session *session;
430
d37b1d99 431 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053
SW		 432 return NULL;
433 session = mdsc->sessions[mds];
434 dout("lookup_mds_session %p %d\n", session,
3997c01d 435 refcount_read(&session->s_ref));
2f2dc053
SW		 436 get_session(session);
437 return session;
438}
439
440static bool __have_session(struct ceph_mds_client *mdsc, int mds)
441{
98cfda81 442 if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
2f2dc053 443 return false;
98cfda81
CX		 444 else
445 return true;
2f2dc053
SW		 446}
447
2600d2dd
SW		 448static int __verify_registered_session(struct ceph_mds_client *mdsc,
449 struct ceph_mds_session *s)
450{
451 if (s->s_mds >= mdsc->max_sessions ||
452 mdsc->sessions[s->s_mds] != s)
453 return -ENOENT;
454 return 0;
455}
456
2f2dc053
SW		 457/*
458 * create+register a new session for given mds.
459 * called under mdsc->mutex.
460 */
461static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
462 int mds)
463{
464 struct ceph_mds_session *s;
465
76201b63 466 if (mds >= mdsc->mdsmap->m_num_mds)
c338c07c
NY		 467 return ERR_PTR(-EINVAL);
468
2f2dc053 469 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 470 if (!s)
471 return ERR_PTR(-ENOMEM);
47474d0b
CX		 472
473 if (mds >= mdsc->max_sessions) {
474 int newmax = 1 << get_count_order(mds + 1);
475 struct ceph_mds_session **sa;
476
477 dout("%s: realloc to %d\n", __func__, newmax);
478 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
479 if (!sa)
480 goto fail_realloc;
481 if (mdsc->sessions) {
482 memcpy(sa, mdsc->sessions,
483 mdsc->max_sessions * sizeof(void *));
484 kfree(mdsc->sessions);
485 }
486 mdsc->sessions = sa;
487 mdsc->max_sessions = newmax;
488 }
489
490 dout("%s: mds%d\n", __func__, mds);
2f2dc053
SW		 491 s->s_mdsc = mdsc;
492 s->s_mds = mds;
493 s->s_state = CEPH_MDS_SESSION_NEW;
494 s->s_ttl = 0;
495 s->s_seq = 0;
496 mutex_init(&s->s_mutex);
497
b7a9e5dd 498 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 499
d8fb02ab 500 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 501 s->s_cap_gen = 0;
1ce208a6 502 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 503
504 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 505 s->s_renew_requested = 0;
506 s->s_renew_seq = 0;
507 INIT_LIST_HEAD(&s->s_caps);
508 s->s_nr_caps = 0;
5dacf091 509 s->s_trim_caps = 0;
3997c01d 510 refcount_set(&s->s_ref, 1);
2f2dc053
SW		 511 INIT_LIST_HEAD(&s->s_waiting);
512 INIT_LIST_HEAD(&s->s_unsafe);
513 s->s_num_cap_releases = 0;
99a9c273 514 s->s_cap_reconnect = 0;
7c1332b8 515 s->s_cap_iterator = NULL;
2f2dc053 516 INIT_LIST_HEAD(&s->s_cap_releases);
2f2dc053 517 INIT_LIST_HEAD(&s->s_cap_flushing);
2f2dc053 518
2f2dc053 519 mdsc->sessions[mds] = s;
86d8f67b 520 atomic_inc(&mdsc->num_sessions);
3997c01d 521 refcount_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 522
b7a9e5dd
SW		 523 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
524 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 525
2f2dc053 526 return s;
42ce56e5
SW		 527
528fail_realloc:
529 kfree(s);
530 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 531}
532
533/*
534 * called under mdsc->mutex
535 */
2600d2dd 536static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 537 struct ceph_mds_session *s)
2f2dc053 538{
2600d2dd
SW		 539 dout("__unregister_session mds%d %p\n", s->s_mds, s);
540 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 541 mdsc->sessions[s->s_mds] = NULL;
542 ceph_con_close(&s->s_con);
543 ceph_put_mds_session(s);
86d8f67b 544 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 545}
546
547/*
548 * drop session refs in request.
549 *
550 * should be last request ref, or hold mdsc->mutex
551 */
552static void put_request_session(struct ceph_mds_request *req)
553{
554 if (req->r_session) {
555 ceph_put_mds_session(req->r_session);
556 req->r_session = NULL;
557 }
558}
559
153c8e6b 560void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 561{
153c8e6b
SW		 562 struct ceph_mds_request *req = container_of(kref,
563 struct ceph_mds_request,
564 r_kref);
54008399 565 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 566 if (req->r_request)
567 ceph_msg_put(req->r_request);
54008399 568 if (req->r_reply)
153c8e6b 569 ceph_msg_put(req->r_reply);
153c8e6b 570 if (req->r_inode) {
41b02e1f 571 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 572 iput(req->r_inode);
573 }
3dd69aab
JL		 574 if (req->r_parent)
575 ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
e96a650a 576 iput(req->r_target_inode);
153c8e6b
SW		 577 if (req->r_dentry)
578 dput(req->r_dentry);
844d87c3
SW		 579 if (req->r_old_dentry)
580 dput(req->r_old_dentry);
581 if (req->r_old_dentry_dir) {
41b02e1f
SW		 582 /*
583 * track (and drop pins for) r_old_dentry_dir
584 * separately, since r_old_dentry's d_parent may have
585 * changed between the dir mutex being dropped and
586 * this request being freed.
587 */
588 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
589 CEPH_CAP_PIN);
41b02e1f 590 iput(req->r_old_dentry_dir);
2f2dc053 591 }
153c8e6b
SW		 592 kfree(req->r_path1);
593 kfree(req->r_path2);
25e6bae3
YZ		 594 if (req->r_pagelist)
595 ceph_pagelist_release(req->r_pagelist);
153c8e6b 596 put_request_session(req);
37151668 597 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 598 kfree(req);
2f2dc053
SW		 599}
600
fcd00b68
ID		 601DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
602
2f2dc053
SW		 603/*
604 * lookup session, bump ref if found.
605 *
606 * called under mdsc->mutex.
607 */
fcd00b68
ID		 608static struct ceph_mds_request *
609lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
2f2dc053
SW		 610{
611 struct ceph_mds_request *req;
44ca18f2 612
fcd00b68
ID		 613 req = lookup_request(&mdsc->request_tree, tid);
614 if (req)
615 ceph_mdsc_get_request(req);
44ca18f2 616
fcd00b68 617 return req;
2f2dc053
SW		 618}
619
620/*
621 * Register an in-flight request, and assign a tid. Link to directory
622 * are modifying (if any).
623 *
624 * Called under mdsc->mutex.
625 */
626static void __register_request(struct ceph_mds_client *mdsc,
627 struct ceph_mds_request *req,
628 struct inode *dir)
629{
e30ee581
ZZ		 630 int ret = 0;
631
2f2dc053 632 req->r_tid = ++mdsc->last_tid;
e30ee581
ZZ		 633 if (req->r_num_caps) {
634 ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
635 req->r_num_caps);
636 if (ret < 0) {
637 pr_err("__register_request %p "
638 "failed to reserve caps: %d\n", req, ret);
639 /* set req->r_err to fail early from __do_request */
640 req->r_err = ret;
641 return;
642 }
643 }
2f2dc053
SW		 644 dout("__register_request %p tid %lld\n", req, req->r_tid);
645 ceph_mdsc_get_request(req);
fcd00b68 646 insert_request(&mdsc->request_tree, req);
2f2dc053 647
cb4276cc
SW		 648 req->r_uid = current_fsuid();
649 req->r_gid = current_fsgid();
650
e8a7b8b1
YZ		 651 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
652 mdsc->oldest_tid = req->r_tid;
653
2f2dc053 654 if (dir) {
3b663780 655 ihold(dir);
2f2dc053 656 req->r_unsafe_dir = dir;
2f2dc053
SW		 657 }
658}
659
660static void __unregister_request(struct ceph_mds_client *mdsc,
661 struct ceph_mds_request *req)
662{
663 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1 664
df963ea8
JL		 665 /* Never leave an unregistered request on an unsafe list! */
666 list_del_init(&req->r_unsafe_item);
667
e8a7b8b1
YZ		 668 if (req->r_tid == mdsc->oldest_tid) {
669 struct rb_node *p = rb_next(&req->r_node);
670 mdsc->oldest_tid = 0;
671 while (p) {
672 struct ceph_mds_request *next_req =
673 rb_entry(p, struct ceph_mds_request, r_node);
674 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
675 mdsc->oldest_tid = next_req->r_tid;
676 break;
677 }
678 p = rb_next(p);
679 }
680 }
681
fcd00b68 682 erase_request(&mdsc->request_tree, req);
2f2dc053 683
bc2de10d
JL		 684 if (req->r_unsafe_dir &&
685 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053 686 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
2f2dc053
SW		 687 spin_lock(&ci->i_unsafe_lock);
688 list_del_init(&req->r_unsafe_dir_item);
689 spin_unlock(&ci->i_unsafe_lock);
4c06ace8 690 }
bc2de10d
JL		 691 if (req->r_target_inode &&
692 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 693 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
694 spin_lock(&ci->i_unsafe_lock);
695 list_del_init(&req->r_unsafe_target_item);
696 spin_unlock(&ci->i_unsafe_lock);
697 }
3b663780 698
4c06ace8 699 if (req->r_unsafe_dir) {
3b663780
SW		 700 iput(req->r_unsafe_dir);
701 req->r_unsafe_dir = NULL;
2f2dc053 702 }
94aa8ae1 703
fc55d2c9
YZ		 704 complete_all(&req->r_safe_completion);
705
94aa8ae1 706 ceph_mdsc_put_request(req);
2f2dc053
SW		 707}
708
30c71233
JL		 709/*
710 * Walk back up the dentry tree until we hit a dentry representing a
711 * non-snapshot inode. We do this using the rcu_read_lock (which must be held
712 * when calling this) to ensure that the objects won't disappear while we're
713 * working with them. Once we hit a candidate dentry, we attempt to take a
714 * reference to it, and return that as the result.
715 */
f1075480
DC		 716static struct inode *get_nonsnap_parent(struct dentry *dentry)
717{
718 struct inode *inode = NULL;
30c71233
JL		 719
720 while (dentry && !IS_ROOT(dentry)) {
721 inode = d_inode_rcu(dentry);
722 if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
723 break;
724 dentry = dentry->d_parent;
725 }
726 if (inode)
727 inode = igrab(inode);
728 return inode;
729}
730
2f2dc053
SW		 731/*
732 * Choose mds to send request to next. If there is a hint set in the
733 * request (e.g., due to a prior forward hint from the mds), use that.
734 * Otherwise, consult frag tree and/or caps to identify the
735 * appropriate mds. If all else fails, choose randomly.
736 *
737 * Called under mdsc->mutex.
738 */
739static int __choose_mds(struct ceph_mds_client *mdsc,
740 struct ceph_mds_request *req)
741{
742 struct inode *inode;
743 struct ceph_inode_info *ci;
744 struct ceph_cap *cap;
745 int mode = req->r_direct_mode;
746 int mds = -1;
747 u32 hash = req->r_direct_hash;
bc2de10d 748 bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
2f2dc053
SW		 749
750 /*
751 * is there a specific mds we should try? ignore hint if we have
752 * no session and the mds is not up (active or recovering).
753 */
754 if (req->r_resend_mds >= 0 &&
755 (__have_session(mdsc, req->r_resend_mds) ||
756 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
757 dout("choose_mds using resend_mds mds%d\n",
758 req->r_resend_mds);
759 return req->r_resend_mds;
760 }
761
762 if (mode == USE_RANDOM_MDS)
763 goto random;
764
765 inode = NULL;
766 if (req->r_inode) {
5d37ca14
YZ		 767 if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
768 inode = req->r_inode;
769 ihold(inode);
770 } else {
38f340cc
YZ		 771 /* req->r_dentry is non-null for LSSNAP request */
772 rcu_read_lock();
773 inode = get_nonsnap_parent(req->r_dentry);
774 rcu_read_unlock();
775 dout("__choose_mds using snapdir's parent %p\n", inode);
5d37ca14 776 }
38f340cc 777 } else if (req->r_dentry) {
d79698da 778 /* ignore race with rename; old or new d_parent is okay */
30c71233
JL		 779 struct dentry *parent;
780 struct inode *dir;
781
782 rcu_read_lock();
783 parent = req->r_dentry->d_parent;
3dd69aab 784 dir = req->r_parent ? : d_inode_rcu(parent);
eb6bb1c5 785
30c71233
JL		 786 if (!dir || dir->i_sb != mdsc->fsc->sb) {
787 /* not this fs or parent went negative */
2b0143b5 788 inode = d_inode(req->r_dentry);
30c71233
JL		 789 if (inode)
790 ihold(inode);
eb6bb1c5
SW		 791 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
792 /* direct snapped/virtual snapdir requests
793 * based on parent dir inode */
30c71233 794 inode = get_nonsnap_parent(parent);
eb6bb1c5 795 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 796 } else {
eb6bb1c5 797 /* dentry target */
2b0143b5 798 inode = d_inode(req->r_dentry);
ca18bede
YZ		 799 if (!inode || mode == USE_AUTH_MDS) {
800 /* dir + name */
30c71233 801 inode = igrab(dir);
ca18bede
YZ		 802 hash = ceph_dentry_hash(dir, req->r_dentry);
803 is_hash = true;
30c71233
JL		 804 } else {
805 ihold(inode);
ca18bede 806 }
2f2dc053 807 }
30c71233 808 rcu_read_unlock();
2f2dc053 809 }
eb6bb1c5 810
2f2dc053
SW		 811 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
812 (int)hash, mode);
813 if (!inode)
814 goto random;
815 ci = ceph_inode(inode);
816
817 if (is_hash && S_ISDIR(inode->i_mode)) {
818 struct ceph_inode_frag frag;
819 int found;
820
821 ceph_choose_frag(ci, hash, &frag, &found);
822 if (found) {
823 if (mode == USE_ANY_MDS && frag.ndist > 0) {
824 u8 r;
825
826 /* choose a random replica */
827 get_random_bytes(&r, 1);
828 r %= frag.ndist;
829 mds = frag.dist[r];
830 dout("choose_mds %p %llx.%llx "
831 "frag %u mds%d (%d/%d)\n",
832 inode, ceph_vinop(inode),
d66bbd44 833 frag.frag, mds,
2f2dc053 834 (int)r, frag.ndist);
d66bbd44
SW		 835 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
836 CEPH_MDS_STATE_ACTIVE)
30c71233 837 goto out;
2f2dc053
SW		 838 }
839
840 /* since this file/dir wasn't known to be
841 * replicated, then we want to look for the
842 * authoritative mds. */
843 mode = USE_AUTH_MDS;
844 if (frag.mds >= 0) {
845 /* choose auth mds */
846 mds = frag.mds;
847 dout("choose_mds %p %llx.%llx "
848 "frag %u mds%d (auth)\n",
849 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 850 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
851 CEPH_MDS_STATE_ACTIVE)
30c71233 852 goto out;
2f2dc053
SW		 853 }
854 }
855 }
856
be655596 857 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 858 cap = NULL;
859 if (mode == USE_AUTH_MDS)
860 cap = ci->i_auth_cap;
861 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
862 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
863 if (!cap) {
be655596 864 spin_unlock(&ci->i_ceph_lock);
30c71233 865 iput(inode);
2f2dc053
SW		 866 goto random;
867 }
868 mds = cap->session->s_mds;
869 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
870 inode, ceph_vinop(inode), mds,
871 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 872 spin_unlock(&ci->i_ceph_lock);
30c71233
JL		 873out:
874 iput(inode);
2f2dc053
SW		 875 return mds;
876
877random:
878 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
879 dout("choose_mds chose random mds%d\n", mds);
880 return mds;
881}
882
883
884/*
885 * session messages
886 */
887static struct ceph_msg *create_session_msg(u32 op, u64 seq)
888{
889 struct ceph_msg *msg;
890 struct ceph_mds_session_head *h;
891
b61c2763
SW		 892 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
893 false);
a79832f2 894 if (!msg) {
2f2dc053 895 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 896 return NULL;
2f2dc053
SW		 897 }
898 h = msg->front.iov_base;
899 h->op = cpu_to_le32(op);
900 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 901
902 return msg;
903}
904
905/*
906 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
907 * to include additional client metadata fields.
908 */
909static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
910{
911 struct ceph_msg *msg;
912 struct ceph_mds_session_head *h;
913 int i = -1;
914 int metadata_bytes = 0;
915 int metadata_key_count = 0;
916 struct ceph_options *opt = mdsc->fsc->client->options;
3f384954 917 struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
dbd0c8bf
JS		 918 void *p;
919
a6a5ce4f 920 const char* metadata[][2] = {
717e6f28
YZ		 921 {"hostname", mdsc->nodename},
922 {"kernel_version", init_utsname()->release},
3f384954
YZ		 923 {"entity_id", opt->name ? : ""},
924 {"root", fsopt->server_path ? : "/"},
dbd0c8bf
JS		 925 {NULL, NULL}
926 };
927
928 /* Calculate serialized length of metadata */
929 metadata_bytes = 4; /* map length */
d37b1d99 930 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 931 metadata_bytes += 8 + strlen(metadata[i][0]) +
932 strlen(metadata[i][1]);
933 metadata_key_count++;
934 }
935
936 /* Allocate the message */
937 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
938 GFP_NOFS, false);
939 if (!msg) {
940 pr_err("create_session_msg ENOMEM creating msg\n");
941 return NULL;
942 }
943 h = msg->front.iov_base;
944 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
945 h->seq = cpu_to_le64(seq);
946
947 /*
948 * Serialize client metadata into waiting buffer space, using
949 * the format that userspace expects for map<string, string>
7cfa0313
JS		 950 *
951 * ClientSession messages with metadata are v2
dbd0c8bf 952 */
7cfa0313
JS		 953 msg->hdr.version = cpu_to_le16(2);
954 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 955
956 /* The write pointer, following the session_head structure */
957 p = msg->front.iov_base + sizeof(*h);
958
959 /* Number of entries in the map */
960 ceph_encode_32(&p, metadata_key_count);
961
962 /* Two length-prefixed strings for each entry in the map */
d37b1d99 963 for (i = 0; metadata[i][0]; ++i) {
dbd0c8bf
JS		 964 size_t const key_len = strlen(metadata[i][0]);
965 size_t const val_len = strlen(metadata[i][1]);
966
967 ceph_encode_32(&p, key_len);
968 memcpy(p, metadata[i][0], key_len);
969 p += key_len;
970 ceph_encode_32(&p, val_len);
971 memcpy(p, metadata[i][1], val_len);
972 p += val_len;
973 }
974
2f2dc053
SW		 975 return msg;
976}
977
978/*
979 * send session open request.
980 *
981 * called under mdsc->mutex
982 */
983static int __open_session(struct ceph_mds_client *mdsc,
984 struct ceph_mds_session *session)
985{
986 struct ceph_msg *msg;
987 int mstate;
988 int mds = session->s_mds;
2f2dc053
SW		 989
990 /* wait for mds to go active? */
991 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
992 dout("open_session to mds%d (%s)\n", mds,
993 ceph_mds_state_name(mstate));
994 session->s_state = CEPH_MDS_SESSION_OPENING;
995 session->s_renew_requested = jiffies;
996
997 /* send connect message */
dbd0c8bf 998 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 999 if (!msg)
1000 return -ENOMEM;
2f2dc053 1001 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 1002 return 0;
1003}
1004
ed0552a1
SW		 1005/*
1006 * open sessions for any export targets for the given mds
1007 *
1008 * called under mdsc->mutex
1009 */
5d72d13c
YZ		 1010static struct ceph_mds_session *
1011__open_export_target_session(struct ceph_mds_client *mdsc, int target)
1012{
1013 struct ceph_mds_session *session;
1014
1015 session = __ceph_lookup_mds_session(mdsc, target);
1016 if (!session) {
1017 session = register_session(mdsc, target);
1018 if (IS_ERR(session))
1019 return session;
1020 }
1021 if (session->s_state == CEPH_MDS_SESSION_NEW ||
1022 session->s_state == CEPH_MDS_SESSION_CLOSING)
1023 __open_session(mdsc, session);
1024
1025 return session;
1026}
1027
1028struct ceph_mds_session *
1029ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1030{
1031 struct ceph_mds_session *session;
1032
1033 dout("open_export_target_session to mds%d\n", target);
1034
1035 mutex_lock(&mdsc->mutex);
1036 session = __open_export_target_session(mdsc, target);
1037 mutex_unlock(&mdsc->mutex);
1038
1039 return session;
1040}
1041
ed0552a1
SW		 1042static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1043 struct ceph_mds_session *session)
1044{
1045 struct ceph_mds_info *mi;
1046 struct ceph_mds_session *ts;
1047 int i, mds = session->s_mds;
ed0552a1 1048
76201b63 1049 if (mds >= mdsc->mdsmap->m_num_mds)
ed0552a1 1050 return;
5d72d13c 1051
ed0552a1
SW		 1052 mi = &mdsc->mdsmap->m_info[mds];
1053 dout("open_export_target_sessions for mds%d (%d targets)\n",
1054 session->s_mds, mi->num_export_targets);
1055
1056 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1057 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1058 if (!IS_ERR(ts))
1059 ceph_put_mds_session(ts);
ed0552a1
SW		 1060 }
1061}
1062
154f42c2
SW		 1063void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1064 struct ceph_mds_session *session)
1065{
1066 mutex_lock(&mdsc->mutex);
1067 __open_export_target_sessions(mdsc, session);
1068 mutex_unlock(&mdsc->mutex);
1069}
1070
2f2dc053
SW		 1071/*
1072 * session caps
1073 */
1074
c8a96a31
JL		 1075static void detach_cap_releases(struct ceph_mds_session *session,
1076 struct list_head *target)
2f2dc053 1077{
c8a96a31
JL		 1078 lockdep_assert_held(&session->s_cap_lock);
1079
1080 list_splice_init(&session->s_cap_releases, target);
745a8e3b 1081 session->s_num_cap_releases = 0;
c8a96a31
JL		 1082 dout("dispose_cap_releases mds%d\n", session->s_mds);
1083}
2f2dc053 1084
c8a96a31
JL		 1085static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1086 struct list_head *dispose)
1087{
1088 while (!list_empty(dispose)) {
745a8e3b
YZ		 1089 struct ceph_cap *cap;
1090 /* zero out the in-progress message */
c8a96a31 1091 cap = list_first_entry(dispose, struct ceph_cap, session_caps);
745a8e3b
YZ		 1092 list_del(&cap->session_caps);
1093 ceph_put_cap(mdsc, cap);
2f2dc053 1094 }
2f2dc053
SW		 1095}
1096
1c841a96
YZ		 1097static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1098 struct ceph_mds_session *session)
1099{
1100 struct ceph_mds_request *req;
1101 struct rb_node *p;
1102
1103 dout("cleanup_session_requests mds%d\n", session->s_mds);
1104 mutex_lock(&mdsc->mutex);
1105 while (!list_empty(&session->s_unsafe)) {
1106 req = list_first_entry(&session->s_unsafe,
1107 struct ceph_mds_request, r_unsafe_item);
3e0708b9
YZ		 1108 pr_warn_ratelimited(" dropping unsafe request %llu\n",
1109 req->r_tid);
1c841a96
YZ		 1110 __unregister_request(mdsc, req);
1111 }
1112 /* zero r_attempts, so kick_requests() will re-send requests */
1113 p = rb_first(&mdsc->request_tree);
1114 while (p) {
1115 req = rb_entry(p, struct ceph_mds_request, r_node);
1116 p = rb_next(p);
1117 if (req->r_session &&
1118 req->r_session->s_mds == session->s_mds)
1119 req->r_attempts = 0;
1120 }
1121 mutex_unlock(&mdsc->mutex);
1122}
1123
2f2dc053 1124/*
f818a736
SW		 1125 * Helper to safely iterate over all caps associated with a session, with
1126 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1127 *
f818a736 1128 * Caller must hold session s_mutex.
2f2dc053
SW		 1129 */
1130static int iterate_session_caps(struct ceph_mds_session *session,
1131 int (*cb)(struct inode *, struct ceph_cap *,
1132 void *), void *arg)
1133{
7c1332b8
SW		 1134 struct list_head *p;
1135 struct ceph_cap *cap;
1136 struct inode *inode, *last_inode = NULL;
1137 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1138 int ret;
1139
1140 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1141 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1142 p = session->s_caps.next;
1143 while (p != &session->s_caps) {
1144 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1145 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1146 if (!inode) {
1147 p = p->next;
2f2dc053 1148 continue;
7c1332b8
SW		 1149 }
1150 session->s_cap_iterator = cap;
2f2dc053 1151 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1152
1153 if (last_inode) {
1154 iput(last_inode);
1155 last_inode = NULL;
1156 }
1157 if (old_cap) {
37151668 1158 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1159 old_cap = NULL;
1160 }
1161
2f2dc053 1162 ret = cb(inode, cap, arg);
7c1332b8
SW		 1163 last_inode = inode;
1164
2f2dc053 1165 spin_lock(&session->s_cap_lock);
7c1332b8 1166 p = p->next;
d37b1d99 1167 if (!cap->ci) {
7c1332b8
SW		 1168 dout("iterate_session_caps finishing cap %p removal\n",
1169 cap);
1170 BUG_ON(cap->session != session);
745a8e3b 1171 cap->session = NULL;
7c1332b8
SW		 1172 list_del_init(&cap->session_caps);
1173 session->s_nr_caps--;
745a8e3b
YZ		 1174 if (cap->queue_release) {
1175 list_add_tail(&cap->session_caps,
1176 &session->s_cap_releases);
1177 session->s_num_cap_releases++;
1178 } else {
1179 old_cap = cap; /* put_cap it w/o locks held */
1180 }
7c1332b8 1181 }
5dacf091
SW		 1182 if (ret < 0)
1183 goto out;
2f2dc053 1184 }
5dacf091
SW		 1185 ret = 0;
1186out:
7c1332b8 1187 session->s_cap_iterator = NULL;
2f2dc053 1188 spin_unlock(&session->s_cap_lock);
7c1332b8 1189
e96a650a 1190 iput(last_inode);
7c1332b8 1191 if (old_cap)
37151668 1192 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1193
5dacf091 1194 return ret;
2f2dc053
SW		 1195}
1196
1197static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1198 void *arg)
2f2dc053 1199{
6c93df5d 1200 struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
2f2dc053 1201 struct ceph_inode_info *ci = ceph_inode(inode);
553adfd9 1202 LIST_HEAD(to_remove);
6c93df5d
YZ		 1203 bool drop = false;
1204 bool invalidate = false;
6c99f254 1205
2f2dc053
SW		 1206 dout("removing cap %p, ci is %p, inode is %p\n",
1207 cap, ci, &ci->vfs_inode);
be655596 1208 spin_lock(&ci->i_ceph_lock);
a096b09a 1209 __ceph_remove_cap(cap, false);
571ade33 1210 if (!ci->i_auth_cap) {
553adfd9 1211 struct ceph_cap_flush *cf;
6c93df5d 1212 struct ceph_mds_client *mdsc = fsc->mdsc;
6c99f254 1213
77310320
YZ		 1214 ci->i_ceph_flags |= CEPH_I_CAP_DROPPED;
1215
6c93df5d 1216 if (ci->i_wrbuffer_ref > 0 &&
52953d55 1217 READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
6c93df5d
YZ		 1218 invalidate = true;
1219
e4500b5e
YZ		 1220 while (!list_empty(&ci->i_cap_flush_list)) {
1221 cf = list_first_entry(&ci->i_cap_flush_list,
1222 struct ceph_cap_flush, i_list);
8cdcc07d 1223 list_move(&cf->i_list, &to_remove);
553adfd9
YZ		 1224 }
1225
6c99f254 1226 spin_lock(&mdsc->cap_dirty_lock);
8310b089 1227
e4500b5e
YZ		 1228 list_for_each_entry(cf, &to_remove, i_list)
1229 list_del(&cf->g_list);
8310b089 1230
6c99f254 1231 if (!list_empty(&ci->i_dirty_item)) {
3e0708b9
YZ		 1232 pr_warn_ratelimited(
1233 " dropping dirty %s state for %p %lld\n",
6c99f254
SW		 1234 ceph_cap_string(ci->i_dirty_caps),
1235 inode, ceph_ino(inode));
1236 ci->i_dirty_caps = 0;
1237 list_del_init(&ci->i_dirty_item);
6c93df5d 1238 drop = true;
6c99f254
SW		 1239 }
1240 if (!list_empty(&ci->i_flushing_item)) {
3e0708b9
YZ		 1241 pr_warn_ratelimited(
1242 " dropping dirty+flushing %s state for %p %lld\n",
6c99f254
SW		 1243 ceph_cap_string(ci->i_flushing_caps),
1244 inode, ceph_ino(inode));
1245 ci->i_flushing_caps = 0;
1246 list_del_init(&ci->i_flushing_item);
1247 mdsc->num_cap_flushing--;
6c93df5d 1248 drop = true;
6c99f254 1249 }
6c99f254 1250 spin_unlock(&mdsc->cap_dirty_lock);
553adfd9 1251
b3f8d68f
YZ		 1252 if (atomic_read(&ci->i_filelock_ref) > 0) {
1253 /* make further file lock syscall return -EIO */
1254 ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1255 pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1256 inode, ceph_ino(inode));
1257 }
1258
f66fd9f0 1259 if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
e4500b5e 1260 list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
f66fd9f0
YZ		 1261 ci->i_prealloc_cap_flush = NULL;
1262 }
6c99f254 1263 }
be655596 1264 spin_unlock(&ci->i_ceph_lock);
553adfd9
YZ		 1265 while (!list_empty(&to_remove)) {
1266 struct ceph_cap_flush *cf;
1267 cf = list_first_entry(&to_remove,
e4500b5e
YZ		 1268 struct ceph_cap_flush, i_list);
1269 list_del(&cf->i_list);
f66fd9f0 1270 ceph_free_cap_flush(cf);
553adfd9 1271 }
77310320
YZ		 1272
1273 wake_up_all(&ci->i_cap_wq);
6c93df5d
YZ		 1274 if (invalidate)
1275 ceph_queue_invalidate(inode);
77310320 1276 if (drop)
6c99f254 1277 iput(inode);
2f2dc053
SW		 1278 return 0;
1279}
1280
1281/*
1282 * caller must hold session s_mutex
1283 */
1284static void remove_session_caps(struct ceph_mds_session *session)
1285{
6c93df5d
YZ		 1286 struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1287 struct super_block *sb = fsc->sb;
c8a96a31
JL		 1288 LIST_HEAD(dispose);
1289
2f2dc053 1290 dout("remove_session_caps on %p\n", session);
6c93df5d 1291 iterate_session_caps(session, remove_session_caps_cb, fsc);
6f60f889 1292
c8799fc4
YZ		 1293 wake_up_all(&fsc->mdsc->cap_flushing_wq);
1294
6f60f889
YZ		 1295 spin_lock(&session->s_cap_lock);
1296 if (session->s_nr_caps > 0) {
6f60f889
YZ		 1297 struct inode *inode;
1298 struct ceph_cap *cap, *prev = NULL;
1299 struct ceph_vino vino;
1300 /*
1301 * iterate_session_caps() skips inodes that are being
1302 * deleted, we need to wait until deletions are complete.
1303 * __wait_on_freeing_inode() is designed for the job,
1304 * but it is not exported, so use lookup inode function
1305 * to access it.
1306 */
1307 while (!list_empty(&session->s_caps)) {
1308 cap = list_entry(session->s_caps.next,
1309 struct ceph_cap, session_caps);
1310 if (cap == prev)
1311 break;
1312 prev = cap;
1313 vino = cap->ci->i_vino;
1314 spin_unlock(&session->s_cap_lock);
1315
ed284c49 1316 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1317 iput(inode);
1318
1319 spin_lock(&session->s_cap_lock);
1320 }
1321 }
745a8e3b
YZ		 1322
1323 // drop cap expires and unlock s_cap_lock
c8a96a31 1324 detach_cap_releases(session, &dispose);
6f60f889 1325
2f2dc053 1326 BUG_ON(session->s_nr_caps > 0);
6c99f254 1327 BUG_ON(!list_empty(&session->s_cap_flushing));
c8a96a31
JL		 1328 spin_unlock(&session->s_cap_lock);
1329 dispose_cap_releases(session->s_mdsc, &dispose);
2f2dc053
SW		 1330}
1331
1332/*
1333 * wake up any threads waiting on this session's caps. if the cap is
1334 * old (didn't get renewed on the client reconnect), remove it now.
1335 *
1336 * caller must hold s_mutex.
1337 */
1338static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1339 void *arg)
1340{
0dc2570f
SW		 1341 struct ceph_inode_info *ci = ceph_inode(inode);
1342
0dc2570f 1343 if (arg) {
be655596 1344 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1345 ci->i_wanted_max_size = 0;
1346 ci->i_requested_max_size = 0;
be655596 1347 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1348 }
e5360309 1349 wake_up_all(&ci->i_cap_wq);
2f2dc053
SW		 1350 return 0;
1351}
1352
0dc2570f
SW		 1353static void wake_up_session_caps(struct ceph_mds_session *session,
1354 int reconnect)
2f2dc053
SW		 1355{
1356 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1357 iterate_session_caps(session, wake_up_session_cb,
1358 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1359}
1360
1361/*
1362 * Send periodic message to MDS renewing all currently held caps. The
1363 * ack will reset the expiration for all caps from this session.
1364 *
1365 * caller holds s_mutex
1366 */
1367static int send_renew_caps(struct ceph_mds_client *mdsc,
1368 struct ceph_mds_session *session)
1369{
1370 struct ceph_msg *msg;
1371 int state;
1372
1373 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1374 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1375 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1376 session->s_renew_requested = jiffies;
2f2dc053
SW		 1377
1378 /* do not try to renew caps until a recovering mds has reconnected
1379 * with its clients. */
1380 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1381 if (state < CEPH_MDS_STATE_RECONNECT) {
1382 dout("send_renew_caps ignoring mds%d (%s)\n",
1383 session->s_mds, ceph_mds_state_name(state));
1384 return 0;
1385 }
1386
1387 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1388 ceph_mds_state_name(state));
2f2dc053
SW		 1389 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1390 ++session->s_renew_seq);
a79832f2
SW		 1391 if (!msg)
1392 return -ENOMEM;
2f2dc053
SW		 1393 ceph_con_send(&session->s_con, msg);
1394 return 0;
1395}
1396
186e4f7a
YZ		 1397static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1398 struct ceph_mds_session *session, u64 seq)
1399{
1400 struct ceph_msg *msg;
1401
1402 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1403 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1404 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1405 if (!msg)
1406 return -ENOMEM;
1407 ceph_con_send(&session->s_con, msg);
1408 return 0;
1409}
1410
1411
2f2dc053
SW		 1412/*
1413 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1414 *
1415 * Called under session->s_mutex
2f2dc053
SW		 1416 */
1417static void renewed_caps(struct ceph_mds_client *mdsc,
1418 struct ceph_mds_session *session, int is_renew)
1419{
1420 int was_stale;
1421 int wake = 0;
1422
1423 spin_lock(&session->s_cap_lock);
1ce208a6 1424 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1425
1426 session->s_cap_ttl = session->s_renew_requested +
1427 mdsc->mdsmap->m_session_timeout*HZ;
1428
1429 if (was_stale) {
1430 if (time_before(jiffies, session->s_cap_ttl)) {
1431 pr_info("mds%d caps renewed\n", session->s_mds);
1432 wake = 1;
1433 } else {
1434 pr_info("mds%d caps still stale\n", session->s_mds);
1435 }
1436 }
1437 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1438 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1439 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1440 spin_unlock(&session->s_cap_lock);
1441
1442 if (wake)
0dc2570f 1443 wake_up_session_caps(session, 0);
2f2dc053
SW		 1444}
1445
1446/*
1447 * send a session close request
1448 */
1449static int request_close_session(struct ceph_mds_client *mdsc,
1450 struct ceph_mds_session *session)
1451{
1452 struct ceph_msg *msg;
2f2dc053
SW		 1453
1454 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1455 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1456 session->s_seq);
1457 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1458 if (!msg)
1459 return -ENOMEM;
1460 ceph_con_send(&session->s_con, msg);
fcff415c 1461 return 1;
2f2dc053
SW		 1462}
1463
1464/*
1465 * Called with s_mutex held.
1466 */
1467static int __close_session(struct ceph_mds_client *mdsc,
1468 struct ceph_mds_session *session)
1469{
1470 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1471 return 0;
1472 session->s_state = CEPH_MDS_SESSION_CLOSING;
1473 return request_close_session(mdsc, session);
1474}
1475
040d7860
YZ		 1476static bool drop_negative_children(struct dentry *dentry)
1477{
1478 struct dentry *child;
1479 bool all_negative = true;
1480
1481 if (!d_is_dir(dentry))
1482 goto out;
1483
1484 spin_lock(&dentry->d_lock);
1485 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1486 if (d_really_is_positive(child)) {
1487 all_negative = false;
1488 break;
1489 }
1490 }
1491 spin_unlock(&dentry->d_lock);
1492
1493 if (all_negative)
1494 shrink_dcache_parent(dentry);
1495out:
1496 return all_negative;
1497}
1498
2f2dc053
SW		 1499/*
1500 * Trim old(er) caps.
1501 *
1502 * Because we can't cache an inode without one or more caps, we do
1503 * this indirectly: if a cap is unused, we prune its aliases, at which
1504 * point the inode will hopefully get dropped to.
1505 *
1506 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1507 * memory pressure from the MDS, though, so it needn't be perfect.
1508 */
1509static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1510{
1511 struct ceph_mds_session *session = arg;
1512 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1513 int used, wanted, oissued, mine;
2f2dc053
SW		 1514
1515 if (session->s_trim_caps <= 0)
1516 return -1;
1517
be655596 1518 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1519 mine = cap->issued | cap->implemented;
1520 used = __ceph_caps_used(ci);
979abfdd 1521 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1522 oissued = __ceph_caps_issued_other(ci, cap);
1523
979abfdd 1524 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1525 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1526 ceph_cap_string(used), ceph_cap_string(wanted));
1527 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1528 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1529 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1530 goto out;
1531 if ((used | wanted) & CEPH_CAP_ANY_WR)
1532 goto out;
89aa5930
YZ		 1533 /* Note: it's possible that i_filelock_ref becomes non-zero
1534 * after dropping auth caps. It doesn't hurt because reply
1535 * of lock mds request will re-add auth caps. */
1536 if (atomic_read(&ci->i_filelock_ref) > 0)
1537 goto out;
979abfdd 1538 }
5e804ac4
YZ		 1539 /* The inode has cached pages, but it's no longer used.
1540 * we can safely drop it */
1541 if (wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1542 !(oissued & CEPH_CAP_FILE_CACHE)) {
1543 used = 0;
1544 oissued = 0;
1545 }
979abfdd 1546 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1547 goto out; /* we need these caps */
1548
2f2dc053
SW		 1549 if (oissued) {
1550 /* we aren't the only cap.. just remove us */
a096b09a 1551 __ceph_remove_cap(cap, true);
040d7860 1552 session->s_trim_caps--;
2f2dc053 1553 } else {
040d7860 1554 struct dentry *dentry;
5e804ac4 1555 /* try dropping referring dentries */
be655596 1556 spin_unlock(&ci->i_ceph_lock);
040d7860
YZ		 1557 dentry = d_find_any_alias(inode);
1558 if (dentry && drop_negative_children(dentry)) {
1559 int count;
1560 dput(dentry);
1561 d_prune_aliases(inode);
1562 count = atomic_read(&inode->i_count);
1563 if (count == 1)
1564 session->s_trim_caps--;
1565 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1566 inode, cap, count);
1567 } else {
1568 dput(dentry);
1569 }
2f2dc053
SW		 1570 return 0;
1571 }
1572
1573out:
be655596 1574 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1575 return 0;
1576}
1577
1578/*
1579 * Trim session cap count down to some max number.
1580 */
e30ee581
ZZ		 1581int ceph_trim_caps(struct ceph_mds_client *mdsc,
1582 struct ceph_mds_session *session,
1583 int max_caps)
2f2dc053
SW		 1584{
1585 int trim_caps = session->s_nr_caps - max_caps;
1586
1587 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1588 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1589 if (trim_caps > 0) {
1590 session->s_trim_caps = trim_caps;
1591 iterate_session_caps(session, trim_caps_cb, session);
1592 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1593 session->s_mds, session->s_nr_caps, max_caps,
1594 trim_caps - session->s_trim_caps);
5dacf091 1595 session->s_trim_caps = 0;
2f2dc053 1596 }
a56371d9 1597
a56371d9 1598 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1599 return 0;
1600}
1601
8310b089
YZ		 1602static int check_caps_flush(struct ceph_mds_client *mdsc,
1603 u64 want_flush_tid)
1604{
8310b089
YZ		 1605 int ret = 1;
1606
1607 spin_lock(&mdsc->cap_dirty_lock);
e4500b5e
YZ		 1608 if (!list_empty(&mdsc->cap_flush_list)) {
1609 struct ceph_cap_flush *cf =
1610 list_first_entry(&mdsc->cap_flush_list,
1611 struct ceph_cap_flush, g_list);
1612 if (cf->tid <= want_flush_tid) {
1613 dout("check_caps_flush still flushing tid "
1614 "%llu <= %llu\n", cf->tid, want_flush_tid);
1615 ret = 0;
1616 }
8310b089
YZ		 1617 }
1618 spin_unlock(&mdsc->cap_dirty_lock);
1619 return ret;
d3383a8e
YZ		 1620}
1621
2f2dc053
SW		 1622/*
1623 * flush all dirty inode data to disk.
1624 *
8310b089 1625 * returns true if we've flushed through want_flush_tid
2f2dc053 1626 */
affbc19a 1627static void wait_caps_flush(struct ceph_mds_client *mdsc,
0e294387 1628 u64 want_flush_tid)
2f2dc053 1629{
0e294387 1630 dout("check_caps_flush want %llu\n", want_flush_tid);
8310b089
YZ		 1631
1632 wait_event(mdsc->cap_flushing_wq,
1633 check_caps_flush(mdsc, want_flush_tid));
1634
1635 dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
2f2dc053
SW		 1636}
1637
1638/*
1639 * called under s_mutex
1640 */
3d7ded4d
SW		 1641void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1642 struct ceph_mds_session *session)
2f2dc053 1643{
745a8e3b
YZ		 1644 struct ceph_msg *msg = NULL;
1645 struct ceph_mds_cap_release *head;
1646 struct ceph_mds_cap_item *item;
92475f05 1647 struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
745a8e3b
YZ		 1648 struct ceph_cap *cap;
1649 LIST_HEAD(tmp_list);
1650 int num_cap_releases;
92475f05
JL		 1651 __le32 barrier, *cap_barrier;
1652
1653 down_read(&osdc->lock);
1654 barrier = cpu_to_le32(osdc->epoch_barrier);
1655 up_read(&osdc->lock);
2f2dc053 1656
0f8605f2 1657 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1658again:
1659 list_splice_init(&session->s_cap_releases, &tmp_list);
1660 num_cap_releases = session->s_num_cap_releases;
1661 session->s_num_cap_releases = 0;
2f2dc053 1662 spin_unlock(&session->s_cap_lock);
e01a5946 1663
745a8e3b
YZ		 1664 while (!list_empty(&tmp_list)) {
1665 if (!msg) {
1666 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
09cbfeaf 1667 PAGE_SIZE, GFP_NOFS, false);
745a8e3b
YZ		 1668 if (!msg)
1669 goto out_err;
1670 head = msg->front.iov_base;
1671 head->num = cpu_to_le32(0);
1672 msg->front.iov_len = sizeof(*head);
92475f05
JL		 1673
1674 msg->hdr.version = cpu_to_le16(2);
1675 msg->hdr.compat_version = cpu_to_le16(1);
745a8e3b 1676 }
92475f05 1677
745a8e3b
YZ		 1678 cap = list_first_entry(&tmp_list, struct ceph_cap,
1679 session_caps);
1680 list_del(&cap->session_caps);
1681 num_cap_releases--;
e01a5946 1682
00bd8edb 1683 head = msg->front.iov_base;
745a8e3b
YZ		 1684 le32_add_cpu(&head->num, 1);
1685 item = msg->front.iov_base + msg->front.iov_len;
1686 item->ino = cpu_to_le64(cap->cap_ino);
1687 item->cap_id = cpu_to_le64(cap->cap_id);
1688 item->migrate_seq = cpu_to_le32(cap->mseq);
1689 item->seq = cpu_to_le32(cap->issue_seq);
1690 msg->front.iov_len += sizeof(*item);
1691
1692 ceph_put_cap(mdsc, cap);
1693
1694 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
92475f05
JL		 1695 // Append cap_barrier field
1696 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1697 *cap_barrier = barrier;
1698 msg->front.iov_len += sizeof(*cap_barrier);
1699
745a8e3b
YZ		 1700 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1701 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1702 ceph_con_send(&session->s_con, msg);
1703 msg = NULL;
1704 }
00bd8edb 1705 }
e01a5946 1706
745a8e3b 1707 BUG_ON(num_cap_releases != 0);
e01a5946 1708
745a8e3b
YZ		 1709 spin_lock(&session->s_cap_lock);
1710 if (!list_empty(&session->s_cap_releases))
1711 goto again;
1712 spin_unlock(&session->s_cap_lock);
1713
1714 if (msg) {
92475f05
JL		 1715 // Append cap_barrier field
1716 cap_barrier = msg->front.iov_base + msg->front.iov_len;
1717 *cap_barrier = barrier;
1718 msg->front.iov_len += sizeof(*cap_barrier);
1719
745a8e3b
YZ		 1720 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1721 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1722 ceph_con_send(&session->s_con, msg);
e01a5946 1723 }
745a8e3b
YZ		 1724 return;
1725out_err:
1726 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1727 session->s_mds);
1728 spin_lock(&session->s_cap_lock);
1729 list_splice(&tmp_list, &session->s_cap_releases);
1730 session->s_num_cap_releases += num_cap_releases;
1731 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1732}
1733
2f2dc053
SW		 1734/*
1735 * requests
1736 */
1737
54008399
YZ		 1738int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1739 struct inode *dir)
1740{
1741 struct ceph_inode_info *ci = ceph_inode(dir);
1742 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1743 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2a5beea3 1744 size_t size = sizeof(struct ceph_mds_reply_dir_entry);
54008399
YZ		 1745 int order, num_entries;
1746
1747 spin_lock(&ci->i_ceph_lock);
1748 num_entries = ci->i_files + ci->i_subdirs;
1749 spin_unlock(&ci->i_ceph_lock);
1750 num_entries = max(num_entries, 1);
1751 num_entries = min(num_entries, opt->max_readdir);
1752
1753 order = get_order(size * num_entries);
1754 while (order >= 0) {
2a5beea3
YZ		 1755 rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
1756 __GFP_NOWARN,
1757 order);
1758 if (rinfo->dir_entries)
54008399
YZ		 1759 break;
1760 order--;
1761 }
2a5beea3 1762 if (!rinfo->dir_entries)
54008399
YZ		 1763 return -ENOMEM;
1764
1765 num_entries = (PAGE_SIZE << order) / size;
1766 num_entries = min(num_entries, opt->max_readdir);
1767
1768 rinfo->dir_buf_size = PAGE_SIZE << order;
1769 req->r_num_caps = num_entries + 1;
1770 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1771 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1772 return 0;
1773}
1774
2f2dc053
SW		 1775/*
1776 * Create an mds request.
1777 */
1778struct ceph_mds_request *
1779ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1780{
1781 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
63ecae7e 1782 struct timespec64 ts;
2f2dc053
SW		 1783
1784 if (!req)
1785 return ERR_PTR(-ENOMEM);
1786
b4556396 1787 mutex_init(&req->r_fill_mutex);
37151668 1788 req->r_mdsc = mdsc;
2f2dc053
SW		 1789 req->r_started = jiffies;
1790 req->r_resend_mds = -1;
1791 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
68cd5b4b 1792 INIT_LIST_HEAD(&req->r_unsafe_target_item);
2f2dc053 1793 req->r_fmode = -1;
153c8e6b 1794 kref_init(&req->r_kref);
fcd00b68 1795 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 1796 INIT_LIST_HEAD(&req->r_wait);
1797 init_completion(&req->r_completion);
1798 init_completion(&req->r_safe_completion);
1799 INIT_LIST_HEAD(&req->r_unsafe_item);
1800
63ecae7e
AB		 1801 ktime_get_coarse_real_ts64(&ts);
1802 req->r_stamp = timespec64_to_timespec(timespec64_trunc(ts,
1803 mdsc->fsc->sb->s_time_gran));
b8e69066 1804
2f2dc053
SW		 1805 req->r_op = op;
1806 req->r_direct_mode = mode;
1807 return req;
1808}
1809
1810/*
44ca18f2 1811 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1812 *
1813 * called under mdsc->mutex.
1814 */
44ca18f2
SW		 1815static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1816{
1817 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1818 return NULL;
1819 return rb_entry(rb_first(&mdsc->request_tree),
1820 struct ceph_mds_request, r_node);
1821}
1822
e8a7b8b1 1823static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1824{
e8a7b8b1 1825 return mdsc->oldest_tid;
2f2dc053
SW		 1826}
1827
1828/*
1829 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1830 * on build_path_from_dentry in fs/cifs/dir.c.
1831 *
1832 * If @stop_on_nosnap, generate path relative to the first non-snapped
1833 * inode.
1834 *
1835 * Encode hidden .snap dirs as a double /, i.e.
1836 * foo/.snap/bar -> foo//bar
1837 */
1838char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1839 int stop_on_nosnap)
1840{
1841 struct dentry *temp;
1842 char *path;
1843 int len, pos;
1b71fe2e 1844 unsigned seq;
2f2dc053 1845
d37b1d99 1846 if (!dentry)
2f2dc053
SW		 1847 return ERR_PTR(-EINVAL);
1848
1849retry:
1850 len = 0;
1b71fe2e
AV		 1851 seq = read_seqbegin(&rename_lock);
1852 rcu_read_lock();
2f2dc053 1853 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1854 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1855 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1856 len++; /* slash only */
1857 else if (stop_on_nosnap && inode &&
1858 ceph_snap(inode) == CEPH_NOSNAP)
1859 break;
1860 else
1861 len += 1 + temp->d_name.len;
1862 temp = temp->d_parent;
2f2dc053 1863 }
1b71fe2e 1864 rcu_read_unlock();
2f2dc053
SW		 1865 if (len)
1866 len--; /* no leading '/' */
1867
1868 path = kmalloc(len+1, GFP_NOFS);
d37b1d99 1869 if (!path)
2f2dc053
SW		 1870 return ERR_PTR(-ENOMEM);
1871 pos = len;
1872 path[pos] = 0; /* trailing null */
1b71fe2e 1873 rcu_read_lock();
2f2dc053 1874 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1875 struct inode *inode;
2f2dc053 1876
1b71fe2e 1877 spin_lock(&temp->d_lock);
2b0143b5 1878 inode = d_inode(temp);
2f2dc053 1879 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1880 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1881 pos, temp);
1882 } else if (stop_on_nosnap && inode &&
1883 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1884 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1885 break;
1886 } else {
1887 pos -= temp->d_name.len;
1b71fe2e
AV		 1888 if (pos < 0) {
1889 spin_unlock(&temp->d_lock);
2f2dc053 1890 break;
1b71fe2e 1891 }
2f2dc053
SW		 1892 strncpy(path + pos, temp->d_name.name,
1893 temp->d_name.len);
2f2dc053 1894 }
1b71fe2e 1895 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1896 if (pos)
1897 path[--pos] = '/';
1898 temp = temp->d_parent;
2f2dc053 1899 }
1b71fe2e
AV		 1900 rcu_read_unlock();
1901 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1902 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1903 "expected, namelen is %d, pos is %d\n", len, pos);
1904 /* presumably this is only possible if racing with a
1905 rename of one of the parent directories (we can not
1906 lock the dentries above us to prevent this, but
1907 retrying should be harmless) */
1908 kfree(path);
1909 goto retry;
1910 }
1911
2b0143b5 1912 *base = ceph_ino(d_inode(temp));
2f2dc053 1913 *plen = len;
104648ad 1914 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1915 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1916 return path;
1917}
1918
fd36a717 1919static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2f2dc053
SW		 1920 const char **ppath, int *ppathlen, u64 *pino,
1921 int *pfreepath)
1922{
1923 char *path;
1924
c6b0b656 1925 rcu_read_lock();
fd36a717
JL		 1926 if (!dir)
1927 dir = d_inode_rcu(dentry->d_parent);
c6b0b656
JL		 1928 if (dir && ceph_snap(dir) == CEPH_NOSNAP) {
1929 *pino = ceph_ino(dir);
1930 rcu_read_unlock();
2f2dc053
SW		 1931 *ppath = dentry->d_name.name;
1932 *ppathlen = dentry->d_name.len;
1933 return 0;
1934 }
c6b0b656 1935 rcu_read_unlock();
2f2dc053
SW		 1936 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1937 if (IS_ERR(path))
1938 return PTR_ERR(path);
1939 *ppath = path;
1940 *pfreepath = 1;
1941 return 0;
1942}
1943
1944static int build_inode_path(struct inode *inode,
1945 const char **ppath, int *ppathlen, u64 *pino,
1946 int *pfreepath)
1947{
1948 struct dentry *dentry;
1949 char *path;
1950
1951 if (ceph_snap(inode) == CEPH_NOSNAP) {
1952 *pino = ceph_ino(inode);
1953 *ppathlen = 0;
1954 return 0;
1955 }
1956 dentry = d_find_alias(inode);
1957 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1958 dput(dentry);
1959 if (IS_ERR(path))
1960 return PTR_ERR(path);
1961 *ppath = path;
1962 *pfreepath = 1;
1963 return 0;
1964}
1965
1966/*
1967 * request arguments may be specified via an inode *, a dentry *, or
1968 * an explicit ino+path.
1969 */
1970static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
fd36a717
JL		 1971 struct inode *rdiri, const char *rpath,
1972 u64 rino, const char **ppath, int *pathlen,
2f2dc053
SW		 1973 u64 *ino, int *freepath)
1974{
1975 int r = 0;
1976
1977 if (rinode) {
1978 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1979 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1980 ceph_snap(rinode));
1981 } else if (rdentry) {
fd36a717
JL		 1982 r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
1983 freepath);
2f2dc053
SW		 1984 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1985 *ppath);
795858db 1986 } else if (rpath || rino) {
2f2dc053
SW		 1987 *ino = rino;
1988 *ppath = rpath;
b000056a 1989 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1990 dout(" path %.*s\n", *pathlen, rpath);
1991 }
1992
1993 return r;
1994}
1995
1996/*
1997 * called under mdsc->mutex
1998 */
1999static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
2000 struct ceph_mds_request *req,
6e6f0923 2001 int mds, bool drop_cap_releases)
2f2dc053
SW		 2002{
2003 struct ceph_msg *msg;
2004 struct ceph_mds_request_head *head;
2005 const char *path1 = NULL;
2006 const char *path2 = NULL;
2007 u64 ino1 = 0, ino2 = 0;
2008 int pathlen1 = 0, pathlen2 = 0;
2009 int freepath1 = 0, freepath2 = 0;
2010 int len;
2011 u16 releases;
2012 void *p, *end;
2013 int ret;
2014
2015 ret = set_request_path_attr(req->r_inode, req->r_dentry,
3dd69aab 2016 req->r_parent, req->r_path1, req->r_ino1.ino,
2f2dc053
SW		 2017 &path1, &pathlen1, &ino1, &freepath1);
2018 if (ret < 0) {
2019 msg = ERR_PTR(ret);
2020 goto out;
2021 }
2022
2023 ret = set_request_path_attr(NULL, req->r_old_dentry,
fd36a717 2024 req->r_old_dentry_dir,
2f2dc053
SW		 2025 req->r_path2, req->r_ino2.ino,
2026 &path2, &pathlen2, &ino2, &freepath2);
2027 if (ret < 0) {
2028 msg = ERR_PTR(ret);
2029 goto out_free1;
2030 }
2031
2032 len = sizeof(*head) +
b8e69066 2033 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
777d738a 2034 sizeof(struct ceph_timespec);
2f2dc053
SW		 2035
2036 /* calculate (max) length for cap releases */
2037 len += sizeof(struct ceph_mds_request_release) *
2038 (!!req->r_inode_drop + !!req->r_dentry_drop +
2039 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2040 if (req->r_dentry_drop)
2041 len += req->r_dentry->d_name.len;
2042 if (req->r_old_dentry_drop)
2043 len += req->r_old_dentry->d_name.len;
2044
b61c2763 2045 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 2046 if (!msg) {
2047 msg = ERR_PTR(-ENOMEM);
2f2dc053 2048 goto out_free2;
a79832f2 2049 }
2f2dc053 2050
7cfa0313 2051 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 2052 msg->hdr.tid = cpu_to_le64(req->r_tid);
2053
2f2dc053
SW		 2054 head = msg->front.iov_base;
2055 p = msg->front.iov_base + sizeof(*head);
2056 end = msg->front.iov_base + msg->front.iov_len;
2057
2058 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2059 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 2060 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2061 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 2062 head->args = req->r_args;
2063
2064 ceph_encode_filepath(&p, end, ino1, path1);
2065 ceph_encode_filepath(&p, end, ino2, path2);
2066
e979cf50
SW		 2067 /* make note of release offset, in case we need to replay */
2068 req->r_request_release_offset = p - msg->front.iov_base;
2069
2f2dc053
SW		 2070 /* cap releases */
2071 releases = 0;
2072 if (req->r_inode_drop)
2073 releases += ceph_encode_inode_release(&p,
2b0143b5 2074 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 2075 mds, req->r_inode_drop, req->r_inode_unless, 0);
2076 if (req->r_dentry_drop)
2077 releases += ceph_encode_dentry_release(&p, req->r_dentry,
3dd69aab 2078 req->r_parent, mds, req->r_dentry_drop,
ca6c8ae0 2079 req->r_dentry_unless);
2f2dc053
SW		 2080 if (req->r_old_dentry_drop)
2081 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
ca6c8ae0
JL		 2082 req->r_old_dentry_dir, mds,
2083 req->r_old_dentry_drop,
2084 req->r_old_dentry_unless);
2f2dc053
SW		 2085 if (req->r_old_inode_drop)
2086 releases += ceph_encode_inode_release(&p,
2b0143b5 2087 d_inode(req->r_old_dentry),
2f2dc053 2088 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 2089
2090 if (drop_cap_releases) {
2091 releases = 0;
2092 p = msg->front.iov_base + req->r_request_release_offset;
2093 }
2094
2f2dc053
SW		 2095 head->num_releases = cpu_to_le16(releases);
2096
b8e69066 2097 /* time stamp */
1f041a89
YZ		 2098 {
2099 struct ceph_timespec ts;
2100 ceph_encode_timespec(&ts, &req->r_stamp);
2101 ceph_encode_copy(&p, &ts, sizeof(ts));
2102 }
b8e69066 2103
2f2dc053
SW		 2104 BUG_ON(p > end);
2105 msg->front.iov_len = p - msg->front.iov_base;
2106 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2107
25e6bae3
YZ		 2108 if (req->r_pagelist) {
2109 struct ceph_pagelist *pagelist = req->r_pagelist;
0e1a5ee6 2110 refcount_inc(&pagelist->refcnt);
25e6bae3
YZ		 2111 ceph_msg_data_add_pagelist(msg, pagelist);
2112 msg->hdr.data_len = cpu_to_le32(pagelist->length);
2113 } else {
2114 msg->hdr.data_len = 0;
ebf18f47 2115 }
02afca6c 2116
2f2dc053
SW		 2117 msg->hdr.data_off = cpu_to_le16(0);
2118
2119out_free2:
2120 if (freepath2)
2121 kfree((char *)path2);
2122out_free1:
2123 if (freepath1)
2124 kfree((char *)path1);
2125out:
2126 return msg;
2127}
2128
2129/*
2130 * called under mdsc->mutex if error, under no mutex if
2131 * success.
2132 */
2133static void complete_request(struct ceph_mds_client *mdsc,
2134 struct ceph_mds_request *req)
2135{
2136 if (req->r_callback)
2137 req->r_callback(mdsc, req);
2138 else
03066f23 2139 complete_all(&req->r_completion);
2f2dc053
SW		 2140}
2141
2142/*
2143 * called under mdsc->mutex
2144 */
2145static int __prepare_send_request(struct ceph_mds_client *mdsc,
2146 struct ceph_mds_request *req,
6e6f0923 2147 int mds, bool drop_cap_releases)
2f2dc053
SW		 2148{
2149 struct ceph_mds_request_head *rhead;
2150 struct ceph_msg *msg;
2151 int flags = 0;
2152
2f2dc053 2153 req->r_attempts++;
e55b71f8
GF		 2154 if (req->r_inode) {
2155 struct ceph_cap *cap =
2156 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2157
2158 if (cap)
2159 req->r_sent_on_mseq = cap->mseq;
2160 else
2161 req->r_sent_on_mseq = -1;
2162 }
2f2dc053
SW		 2163 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2164 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2165
bc2de10d 2166 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
c5c9a0bf 2167 void *p;
01a92f17
SW		 2168 /*
2169 * Replay. Do not regenerate message (and rebuild
2170 * paths, etc.); just use the original message.
2171 * Rebuilding paths will break for renames because
2172 * d_move mangles the src name.
2173 */
2174 msg = req->r_request;
2175 rhead = msg->front.iov_base;
2176
2177 flags = le32_to_cpu(rhead->flags);
2178 flags |= CEPH_MDS_FLAG_REPLAY;
2179 rhead->flags = cpu_to_le32(flags);
2180
2181 if (req->r_target_inode)
2182 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2183
2184 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2185
2186 /* remove cap/dentry releases from message */
2187 rhead->num_releases = 0;
c5c9a0bf
YZ		 2188
2189 /* time stamp */
2190 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2191 {
2192 struct ceph_timespec ts;
2193 ceph_encode_timespec(&ts, &req->r_stamp);
2194 ceph_encode_copy(&p, &ts, sizeof(ts));
2195 }
c5c9a0bf
YZ		 2196
2197 msg->front.iov_len = p - msg->front.iov_base;
2198 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2199 return 0;
2200 }
2201
2f2dc053
SW		 2202 if (req->r_request) {
2203 ceph_msg_put(req->r_request);
2204 req->r_request = NULL;
2205 }
6e6f0923 2206 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2207 if (IS_ERR(msg)) {
e1518c7c 2208 req->r_err = PTR_ERR(msg);
a79832f2 2209 return PTR_ERR(msg);
2f2dc053
SW		 2210 }
2211 req->r_request = msg;
2212
2213 rhead = msg->front.iov_base;
2f2dc053 2214 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
bc2de10d 2215 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2f2dc053 2216 flags |= CEPH_MDS_FLAG_REPLAY;
3dd69aab 2217 if (req->r_parent)
2f2dc053
SW		 2218 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2219 rhead->flags = cpu_to_le32(flags);
2220 rhead->num_fwd = req->r_num_fwd;
2221 rhead->num_retry = req->r_attempts - 1;
01a92f17 2222 rhead->ino = 0;
2f2dc053 2223
3dd69aab 2224 dout(" r_parent = %p\n", req->r_parent);
2f2dc053
SW		 2225 return 0;
2226}
2227
2228/*
2229 * send request, or put it on the appropriate wait list.
2230 */
2231static int __do_request(struct ceph_mds_client *mdsc,
2232 struct ceph_mds_request *req)
2233{
2234 struct ceph_mds_session *session = NULL;
2235 int mds = -1;
48fec5d0 2236 int err = 0;
2f2dc053 2237
bc2de10d
JL		 2238 if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2239 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
eb1b8af3 2240 __unregister_request(mdsc, req);
2f2dc053 2241 goto out;
eb1b8af3 2242 }
2f2dc053
SW		 2243
2244 if (req->r_timeout &&
2245 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2246 dout("do_request timed out\n");
2247 err = -EIO;
2248 goto finish;
2249 }
52953d55 2250 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
48fec5d0
YZ		 2251 dout("do_request forced umount\n");
2252 err = -EIO;
2253 goto finish;
2254 }
52953d55 2255 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
e9e427f0
YZ		 2256 if (mdsc->mdsmap_err) {
2257 err = mdsc->mdsmap_err;
2258 dout("do_request mdsmap err %d\n", err);
2259 goto finish;
2260 }
cc8e8342
YZ		 2261 if (mdsc->mdsmap->m_epoch == 0) {
2262 dout("do_request no mdsmap, waiting for map\n");
2263 list_add(&req->r_wait, &mdsc->waiting_for_map);
2264 goto finish;
2265 }
e9e427f0
YZ		 2266 if (!(mdsc->fsc->mount_options->flags &
2267 CEPH_MOUNT_OPT_MOUNTWAIT) &&
2268 !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2269 err = -ENOENT;
2270 pr_info("probably no mds server is up\n");
2271 goto finish;
2272 }
2273 }
2f2dc053 2274
dc69e2e9
SW		 2275 put_request_session(req);
2276
2f2dc053
SW		 2277 mds = __choose_mds(mdsc, req);
2278 if (mds < 0 ||
2279 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2280 dout("do_request no mds or not active, waiting for map\n");
2281 list_add(&req->r_wait, &mdsc->waiting_for_map);
2282 goto out;
2283 }
2284
2285 /* get, open session */
2286 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2287 if (!session) {
2f2dc053 2288 session = register_session(mdsc, mds);
9c423956
SW		 2289 if (IS_ERR(session)) {
2290 err = PTR_ERR(session);
2291 goto finish;
2292 }
2293 }
dc69e2e9
SW		 2294 req->r_session = get_session(session);
2295
2f2dc053 2296 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2297 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2298 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2299 session->s_state != CEPH_MDS_SESSION_HUNG) {
fcff415c
YZ		 2300 if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2301 err = -EACCES;
2302 goto out_session;
2303 }
2f2dc053
SW		 2304 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2305 session->s_state == CEPH_MDS_SESSION_CLOSING)
2306 __open_session(mdsc, session);
2307 list_add(&req->r_wait, &session->s_waiting);
2308 goto out_session;
2309 }
2310
2311 /* send request */
2f2dc053
SW		 2312 req->r_resend_mds = -1; /* forget any previous mds hint */
2313
2314 if (req->r_request_started == 0) /* note request start time */
2315 req->r_request_started = jiffies;
2316
6e6f0923 2317 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2318 if (!err) {
2319 ceph_msg_get(req->r_request);
2320 ceph_con_send(&session->s_con, req->r_request);
2321 }
2322
2323out_session:
2324 ceph_put_mds_session(session);
48fec5d0
YZ		 2325finish:
2326 if (err) {
2327 dout("__do_request early error %d\n", err);
2328 req->r_err = err;
2329 complete_request(mdsc, req);
2330 __unregister_request(mdsc, req);
2331 }
2f2dc053
SW		 2332out:
2333 return err;
2f2dc053
SW		 2334}
2335
2336/*
2337 * called under mdsc->mutex
2338 */
2339static void __wake_requests(struct ceph_mds_client *mdsc,
2340 struct list_head *head)
2341{
ed75ec2c
YZ		 2342 struct ceph_mds_request *req;
2343 LIST_HEAD(tmp_list);
2344
2345 list_splice_init(head, &tmp_list);
2f2dc053 2346
ed75ec2c
YZ		 2347 while (!list_empty(&tmp_list)) {
2348 req = list_entry(tmp_list.next,
2349 struct ceph_mds_request, r_wait);
2f2dc053 2350 list_del_init(&req->r_wait);
7971bd92 2351 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2352 __do_request(mdsc, req);
2353 }
2354}
2355
2356/*
2357 * Wake up threads with requests pending for @mds, so that they can
29790f26 2358 * resubmit their requests to a possibly different mds.
2f2dc053 2359 */
29790f26 2360static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2361{
44ca18f2 2362 struct ceph_mds_request *req;
282c1052 2363 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2364
2365 dout("kick_requests mds%d\n", mds);
282c1052 2366 while (p) {
44ca18f2 2367 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2368 p = rb_next(p);
bc2de10d 2369 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
44ca18f2 2370 continue;
3de22be6
YZ		 2371 if (req->r_attempts > 0)
2372 continue; /* only new requests */
44ca18f2
SW		 2373 if (req->r_session &&
2374 req->r_session->s_mds == mds) {
2375 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2376 list_del_init(&req->r_wait);
44ca18f2 2377 __do_request(mdsc, req);
2f2dc053
SW		 2378 }
2379 }
2380}
2381
2382void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2383 struct ceph_mds_request *req)
2384{
2385 dout("submit_request on %p\n", req);
2386 mutex_lock(&mdsc->mutex);
2387 __register_request(mdsc, req, NULL);
2388 __do_request(mdsc, req);
2389 mutex_unlock(&mdsc->mutex);
2390}
2391
2392/*
2393 * Synchrously perform an mds request. Take care of all of the
2394 * session setup, forwarding, retry details.
2395 */
2396int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2397 struct inode *dir,
2398 struct ceph_mds_request *req)
2399{
2400 int err;
2401
2402 dout("do_request on %p\n", req);
2403
3dd69aab 2404 /* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2f2dc053
SW		 2405 if (req->r_inode)
2406 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
3dd69aab
JL		 2407 if (req->r_parent)
2408 ceph_get_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
844d87c3 2409 if (req->r_old_dentry_dir)
41b02e1f
SW		 2410 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2411 CEPH_CAP_PIN);
2f2dc053
SW		 2412
2413 /* issue */
2414 mutex_lock(&mdsc->mutex);
2415 __register_request(mdsc, req, dir);
2416 __do_request(mdsc, req);
2417
e1518c7c
SW		 2418 if (req->r_err) {
2419 err = req->r_err;
e1518c7c 2420 goto out;
2f2dc053
SW		 2421 }
2422
e1518c7c
SW		 2423 /* wait */
2424 mutex_unlock(&mdsc->mutex);
2425 dout("do_request waiting\n");
5be73034 2426 if (!req->r_timeout && req->r_wait_for_completion) {
9280be24 2427 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2428 } else {
5be73034
ID		 2429 long timeleft = wait_for_completion_killable_timeout(
2430 &req->r_completion,
2431 ceph_timeout_jiffies(req->r_timeout));
2432 if (timeleft > 0)
2433 err = 0;
2434 else if (!timeleft)
2435 err = -EIO; /* timed out */
2436 else
2437 err = timeleft; /* killed */
e1518c7c
SW		 2438 }
2439 dout("do_request waited, got %d\n", err);
2440 mutex_lock(&mdsc->mutex);
5b1daecd 2441
e1518c7c 2442 /* only abort if we didn't race with a real reply */
bc2de10d 2443 if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
e1518c7c
SW		 2444 err = le32_to_cpu(req->r_reply_info.head->result);
2445 } else if (err < 0) {
2446 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2447
2448 /*
2449 * ensure we aren't running concurrently with
2450 * ceph_fill_trace or ceph_readdir_prepopulate, which
2451 * rely on locks (dir mutex) held by our caller.
2452 */
2453 mutex_lock(&req->r_fill_mutex);
e1518c7c 2454 req->r_err = err;
bc2de10d 2455 set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
b4556396 2456 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2457
3dd69aab 2458 if (req->r_parent &&
167c9e35
SW		 2459 (req->r_op & CEPH_MDS_OP_WRITE))
2460 ceph_invalidate_dir_request(req);
2f2dc053 2461 } else {
e1518c7c 2462 err = req->r_err;
2f2dc053 2463 }
2f2dc053 2464
e1518c7c
SW		 2465out:
2466 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2467 dout("do_request %p done, result %d\n", req, err);
2468 return err;
2469}
2470
167c9e35 2471/*
2f276c51 2472 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2473 * namespace request.
2474 */
2475void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2476{
8d8f371c
YZ		 2477 struct inode *dir = req->r_parent;
2478 struct inode *old_dir = req->r_old_dentry_dir;
167c9e35 2479
8d8f371c 2480 dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
167c9e35 2481
8d8f371c
YZ		 2482 ceph_dir_clear_complete(dir);
2483 if (old_dir)
2484 ceph_dir_clear_complete(old_dir);
167c9e35
SW		 2485 if (req->r_dentry)
2486 ceph_invalidate_dentry_lease(req->r_dentry);
2487 if (req->r_old_dentry)
2488 ceph_invalidate_dentry_lease(req->r_old_dentry);
2489}
2490
2f2dc053
SW		 2491/*
2492 * Handle mds reply.
2493 *
2494 * We take the session mutex and parse and process the reply immediately.
2495 * This preserves the logical ordering of replies, capabilities, etc., sent
2496 * by the MDS as they are applied to our local cache.
2497 */
2498static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2499{
2500 struct ceph_mds_client *mdsc = session->s_mdsc;
2501 struct ceph_mds_request *req;
2502 struct ceph_mds_reply_head *head = msg->front.iov_base;
2503 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2504 struct ceph_snap_realm *realm;
2f2dc053
SW		 2505 u64 tid;
2506 int err, result;
2600d2dd 2507 int mds = session->s_mds;
2f2dc053 2508
2f2dc053
SW		 2509 if (msg->front.iov_len < sizeof(*head)) {
2510 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2511 ceph_msg_dump(msg);
2f2dc053
SW		 2512 return;
2513 }
2514
2515 /* get request, session */
6df058c0 2516 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053 2517 mutex_lock(&mdsc->mutex);
fcd00b68 2518 req = lookup_get_request(mdsc, tid);
2f2dc053
SW		 2519 if (!req) {
2520 dout("handle_reply on unknown tid %llu\n", tid);
2521 mutex_unlock(&mdsc->mutex);
2522 return;
2523 }
2524 dout("handle_reply %p\n", req);
2f2dc053
SW		 2525
2526 /* correct session? */
d96d6049 2527 if (req->r_session != session) {
2f2dc053
SW		 2528 pr_err("mdsc_handle_reply got %llu on session mds%d"
2529 " not mds%d\n", tid, session->s_mds,
2530 req->r_session ? req->r_session->s_mds : -1);
2531 mutex_unlock(&mdsc->mutex);
2532 goto out;
2533 }
2534
2535 /* dup? */
bc2de10d
JL		 2536 if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
2537 (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
f3ae1b97 2538 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2539 head->safe ? "safe" : "unsafe", tid, mds);
2540 mutex_unlock(&mdsc->mutex);
2541 goto out;
2542 }
bc2de10d 2543 if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
f3ae1b97 2544 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2545 tid, mds);
2546 mutex_unlock(&mdsc->mutex);
2547 goto out;
2548 }
2f2dc053
SW		 2549
2550 result = le32_to_cpu(head->result);
2551
2552 /*
e55b71f8
GF		 2553 * Handle an ESTALE
2554 * if we're not talking to the authority, send to them
2555 * if the authority has changed while we weren't looking,
2556 * send to new authority
2557 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2558 */
2559 if (result == -ESTALE) {
4c069a58 2560 dout("got ESTALE on request %llu\n", req->r_tid);
51da8e8c 2561 req->r_resend_mds = -1;
ca18bede 2562 if (req->r_direct_mode != USE_AUTH_MDS) {
4c069a58 2563 dout("not using auth, setting for that now\n");
e55b71f8 2564 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2565 __do_request(mdsc, req);
2566 mutex_unlock(&mdsc->mutex);
2567 goto out;
e55b71f8 2568 } else {
ca18bede
YZ		 2569 int mds = __choose_mds(mdsc, req);
2570 if (mds >= 0 && mds != req->r_session->s_mds) {
4c069a58 2571 dout("but auth changed, so resending\n");
e55b71f8
GF		 2572 __do_request(mdsc, req);
2573 mutex_unlock(&mdsc->mutex);
2574 goto out;
2575 }
2f2dc053 2576 }
4c069a58 2577 dout("have to return ESTALE on request %llu\n", req->r_tid);
2f2dc053
SW		 2578 }
2579
e55b71f8 2580
2f2dc053 2581 if (head->safe) {
bc2de10d 2582 set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
2f2dc053 2583 __unregister_request(mdsc, req);
2f2dc053 2584
bc2de10d 2585 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2f2dc053
SW		 2586 /*
2587 * We already handled the unsafe response, now do the
2588 * cleanup. No need to examine the response; the MDS
2589 * doesn't include any result info in the safe
2590 * response. And even if it did, there is nothing
2591 * useful we could do with a revised return value.
2592 */
2593 dout("got safe reply %llu, mds%d\n", tid, mds);
2f2dc053
SW		 2594
2595 /* last unsafe request during umount? */
44ca18f2 2596 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2597 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2598 mutex_unlock(&mdsc->mutex);
2599 goto out;
2600 }
e1518c7c 2601 } else {
bc2de10d 2602 set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
2f2dc053 2603 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
4c06ace8
YZ		 2604 if (req->r_unsafe_dir) {
2605 struct ceph_inode_info *ci =
2606 ceph_inode(req->r_unsafe_dir);
2607 spin_lock(&ci->i_unsafe_lock);
2608 list_add_tail(&req->r_unsafe_dir_item,
2609 &ci->i_unsafe_dirops);
2610 spin_unlock(&ci->i_unsafe_lock);
2611 }
2f2dc053
SW		 2612 }
2613
2614 dout("handle_reply tid %lld result %d\n", tid, result);
2615 rinfo = &req->r_reply_info;
14303d20 2616 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2617 mutex_unlock(&mdsc->mutex);
2618
2619 mutex_lock(&session->s_mutex);
2620 if (err < 0) {
25933abd 2621 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2622 ceph_msg_dump(msg);
2f2dc053
SW		 2623 goto out_err;
2624 }
2625
2626 /* snap trace */
982d6011 2627 realm = NULL;
2f2dc053
SW		 2628 if (rinfo->snapblob_len) {
2629 down_write(&mdsc->snap_rwsem);
2630 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2631 rinfo->snapblob + rinfo->snapblob_len,
2632 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2633 &realm);
2f2dc053
SW		 2634 downgrade_write(&mdsc->snap_rwsem);
2635 } else {
2636 down_read(&mdsc->snap_rwsem);
2637 }
2638
2639 /* insert trace into our cache */
b4556396 2640 mutex_lock(&req->r_fill_mutex);
315f2408 2641 current->journal_info = req;
f5a03b08 2642 err = ceph_fill_trace(mdsc->fsc->sb, req);
2f2dc053 2643 if (err == 0) {
6e8575fa 2644 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2645 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2646 ceph_readdir_prepopulate(req, req->r_session);
37151668 2647 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2648 }
315f2408 2649 current->journal_info = NULL;
b4556396 2650 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2651
2652 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2653 if (realm)
2654 ceph_put_snap_realm(mdsc, realm);
68cd5b4b 2655
bc2de10d
JL		 2656 if (err == 0 && req->r_target_inode &&
2657 test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
68cd5b4b
YZ		 2658 struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
2659 spin_lock(&ci->i_unsafe_lock);
2660 list_add_tail(&req->r_unsafe_target_item, &ci->i_unsafe_iops);
2661 spin_unlock(&ci->i_unsafe_lock);
2662 }
2f2dc053 2663out_err:
e1518c7c 2664 mutex_lock(&mdsc->mutex);
bc2de10d 2665 if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
e1518c7c
SW		 2666 if (err) {
2667 req->r_err = err;
2668 } else {
5fdb1389 2669 req->r_reply = ceph_msg_get(msg);
bc2de10d 2670 set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
e1518c7c 2671 }
2f2dc053 2672 } else {
e1518c7c 2673 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2674 }
e1518c7c 2675 mutex_unlock(&mdsc->mutex);
2f2dc053 2676
2f2dc053
SW		 2677 mutex_unlock(&session->s_mutex);
2678
2679 /* kick calling process */
2680 complete_request(mdsc, req);
2681out:
2682 ceph_mdsc_put_request(req);
2683 return;
2684}
2685
2686
2687
2688/*
2689 * handle mds notification that our request has been forwarded.
2690 */
2600d2dd
SW		 2691static void handle_forward(struct ceph_mds_client *mdsc,
2692 struct ceph_mds_session *session,
2693 struct ceph_msg *msg)
2f2dc053
SW		 2694{
2695 struct ceph_mds_request *req;
a1ea787c 2696 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2697 u32 next_mds;
2698 u32 fwd_seq;
2f2dc053
SW		 2699 int err = -EINVAL;
2700 void *p = msg->front.iov_base;
2701 void *end = p + msg->front.iov_len;
2f2dc053 2702
a1ea787c 2703 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2704 next_mds = ceph_decode_32(&p);
2705 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2706
2707 mutex_lock(&mdsc->mutex);
fcd00b68 2708 req = lookup_get_request(mdsc, tid);
2f2dc053 2709 if (!req) {
2a8e5e36 2710 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2711 goto out; /* dup reply? */
2712 }
2713
bc2de10d 2714 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
2a8e5e36
SW		 2715 dout("forward tid %llu aborted, unregistering\n", tid);
2716 __unregister_request(mdsc, req);
2717 } else if (fwd_seq <= req->r_num_fwd) {
2718 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2719 tid, next_mds, req->r_num_fwd, fwd_seq);
2720 } else {
2721 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2722 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2723 BUG_ON(req->r_err);
bc2de10d 2724 BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3de22be6 2725 req->r_attempts = 0;
2f2dc053
SW		 2726 req->r_num_fwd = fwd_seq;
2727 req->r_resend_mds = next_mds;
2728 put_request_session(req);
2729 __do_request(mdsc, req);
2730 }
2731 ceph_mdsc_put_request(req);
2732out:
2733 mutex_unlock(&mdsc->mutex);
2734 return;
2735
2736bad:
2737 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2738}
2739
2740/*
2741 * handle a mds session control message
2742 */
2743static void handle_session(struct ceph_mds_session *session,
2744 struct ceph_msg *msg)
2745{
2746 struct ceph_mds_client *mdsc = session->s_mdsc;
2747 u32 op;
2748 u64 seq;
2600d2dd 2749 int mds = session->s_mds;
2f2dc053
SW		 2750 struct ceph_mds_session_head *h = msg->front.iov_base;
2751 int wake = 0;
2752
2f2dc053
SW		 2753 /* decode */
2754 if (msg->front.iov_len != sizeof(*h))
2755 goto bad;
2756 op = le32_to_cpu(h->op);
2757 seq = le64_to_cpu(h->seq);
2758
2759 mutex_lock(&mdsc->mutex);
0a07fc8c
YZ		 2760 if (op == CEPH_SESSION_CLOSE) {
2761 get_session(session);
2600d2dd 2762 __unregister_session(mdsc, session);
0a07fc8c 2763 }
2f2dc053
SW		 2764 /* FIXME: this ttl calculation is generous */
2765 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2766 mutex_unlock(&mdsc->mutex);
2767
2768 mutex_lock(&session->s_mutex);
2769
2770 dout("handle_session mds%d %s %p state %s seq %llu\n",
2771 mds, ceph_session_op_name(op), session,
a687ecaf 2772 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2773
2774 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2775 session->s_state = CEPH_MDS_SESSION_OPEN;
2776 pr_info("mds%d came back\n", session->s_mds);
2777 }
2778
2779 switch (op) {
2780 case CEPH_SESSION_OPEN:
29790f26
SW		 2781 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2782 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2783 session->s_state = CEPH_MDS_SESSION_OPEN;
2784 renewed_caps(mdsc, session, 0);
2785 wake = 1;
2786 if (mdsc->stopping)
2787 __close_session(mdsc, session);
2788 break;
2789
2790 case CEPH_SESSION_RENEWCAPS:
2791 if (session->s_renew_seq == seq)
2792 renewed_caps(mdsc, session, 1);
2793 break;
2794
2795 case CEPH_SESSION_CLOSE:
29790f26
SW		 2796 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2797 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2798 cleanup_session_requests(mdsc, session);
2f2dc053 2799 remove_session_caps(session);
656e4382 2800 wake = 2; /* for good measure */
f3c60c59 2801 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2802 break;
2803
2804 case CEPH_SESSION_STALE:
2805 pr_info("mds%d caps went stale, renewing\n",
2806 session->s_mds);
d8fb02ab 2807 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2808 session->s_cap_gen++;
1ce208a6 2809 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2810 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2811 send_renew_caps(mdsc, session);
2812 break;
2813
2814 case CEPH_SESSION_RECALL_STATE:
e30ee581 2815 ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2f2dc053
SW		 2816 break;
2817
186e4f7a
YZ		 2818 case CEPH_SESSION_FLUSHMSG:
2819 send_flushmsg_ack(mdsc, session, seq);
2820 break;
2821
03f4fcb0
YZ		 2822 case CEPH_SESSION_FORCE_RO:
2823 dout("force_session_readonly %p\n", session);
2824 spin_lock(&session->s_cap_lock);
2825 session->s_readonly = true;
2826 spin_unlock(&session->s_cap_lock);
2827 wake_up_session_caps(session, 0);
2828 break;
2829
fcff415c
YZ		 2830 case CEPH_SESSION_REJECT:
2831 WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
2832 pr_info("mds%d rejected session\n", session->s_mds);
2833 session->s_state = CEPH_MDS_SESSION_REJECTED;
2834 cleanup_session_requests(mdsc, session);
2835 remove_session_caps(session);
2836 wake = 2; /* for good measure */
2837 break;
2838
2f2dc053
SW		 2839 default:
2840 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2841 WARN_ON(1);
2842 }
2843
2844 mutex_unlock(&session->s_mutex);
2845 if (wake) {
2846 mutex_lock(&mdsc->mutex);
2847 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2848 if (wake == 2)
2849 kick_requests(mdsc, mds);
2f2dc053
SW		 2850 mutex_unlock(&mdsc->mutex);
2851 }
0a07fc8c
YZ		 2852 if (op == CEPH_SESSION_CLOSE)
2853 ceph_put_mds_session(session);
2f2dc053
SW		 2854 return;
2855
2856bad:
2857 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2858 (int)msg->front.iov_len);
9ec7cab1 2859 ceph_msg_dump(msg);
2f2dc053
SW		 2860 return;
2861}
2862
2863
2864/*
2865 * called under session->mutex.
2866 */
2867static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2868 struct ceph_mds_session *session)
2869{
2870 struct ceph_mds_request *req, *nreq;
3de22be6 2871 struct rb_node *p;
2f2dc053
SW		 2872 int err;
2873
2874 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2875
2876 mutex_lock(&mdsc->mutex);
2877 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2878 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2879 if (!err) {
2880 ceph_msg_get(req->r_request);
2881 ceph_con_send(&session->s_con, req->r_request);
2882 }
2883 }
3de22be6
YZ		 2884
2885 /*
2886 * also re-send old requests when MDS enters reconnect stage. So that MDS
2887 * can process completed request in clientreplay stage.
2888 */
2889 p = rb_first(&mdsc->request_tree);
2890 while (p) {
2891 req = rb_entry(p, struct ceph_mds_request, r_node);
2892 p = rb_next(p);
bc2de10d 2893 if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3de22be6
YZ		 2894 continue;
2895 if (req->r_attempts == 0)
2896 continue; /* only old requests */
2897 if (req->r_session &&
2898 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2899 err = __prepare_send_request(mdsc, req,
2900 session->s_mds, true);
3de22be6
YZ		 2901 if (!err) {
2902 ceph_msg_get(req->r_request);
2903 ceph_con_send(&session->s_con, req->r_request);
2904 }
2905 }
2906 }
2f2dc053
SW		 2907 mutex_unlock(&mdsc->mutex);
2908}
2909
2910/*
2911 * Encode information about a cap for a reconnect with the MDS.
2912 */
2f2dc053
SW		 2913static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2914 void *arg)
2915{
20cb34ae
SW		 2916 union {
2917 struct ceph_mds_cap_reconnect v2;
2918 struct ceph_mds_cap_reconnect_v1 v1;
2919 } rec;
b3f8d68f 2920 struct ceph_inode_info *ci = cap->ci;
20cb34ae
SW		 2921 struct ceph_reconnect_state *recon_state = arg;
2922 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2923 char *path;
2924 int pathlen, err;
2925 u64 pathbase;
3469ed0d 2926 u64 snap_follows;
2f2dc053
SW		 2927 struct dentry *dentry;
2928
2f2dc053
SW		 2929 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2930 inode, ceph_vinop(inode), cap, cap->cap_id,
2931 ceph_cap_string(cap->issued));
93cea5be
SW		 2932 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2933 if (err)
2934 return err;
2f2dc053
SW		 2935
2936 dentry = d_find_alias(inode);
2937 if (dentry) {
2938 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2939 if (IS_ERR(path)) {
2940 err = PTR_ERR(path);
e072f8aa 2941 goto out_dput;
2f2dc053
SW		 2942 }
2943 } else {
2944 path = NULL;
2945 pathlen = 0;
4eacd4cb 2946 pathbase = 0;
2f2dc053 2947 }
2f2dc053 2948
be655596 2949 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2950 cap->seq = 0; /* reset cap seq */
2951 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2952 cap->mseq = 0; /* and migrate_seq */
99a9c273 2953 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae 2954
121f22a1 2955 if (recon_state->msg_version >= 2) {
20cb34ae
SW		 2956 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2957 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2958 rec.v2.issued = cpu_to_le32(cap->issued);
2959 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2960 rec.v2.pathbase = cpu_to_le64(pathbase);
ec1dff25
JL		 2961 rec.v2.flock_len = (__force __le32)
2962 ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
20cb34ae
SW		 2963 } else {
2964 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2965 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2966 rec.v1.issued = cpu_to_le32(cap->issued);
2967 rec.v1.size = cpu_to_le64(inode->i_size);
9bbeab41
AB		 2968 ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
2969 ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
20cb34ae
SW		 2970 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2971 rec.v1.pathbase = cpu_to_le64(pathbase);
20cb34ae 2972 }
3469ed0d
YZ		 2973
2974 if (list_empty(&ci->i_cap_snaps)) {
92776fd2 2975 snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3469ed0d
YZ		 2976 } else {
2977 struct ceph_cap_snap *capsnap =
2978 list_first_entry(&ci->i_cap_snaps,
2979 struct ceph_cap_snap, ci_item);
2980 snap_follows = capsnap->follows;
20cb34ae 2981 }
be655596 2982 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2983
121f22a1 2984 if (recon_state->msg_version >= 2) {
40819f6f 2985 int num_fcntl_locks, num_flock_locks;
4deb14a2 2986 struct ceph_filelock *flocks = NULL;
121f22a1
YZ		 2987 size_t struct_len, total_len = 0;
2988 u8 struct_v = 0;
39be95e9
JS		 2989
2990encode_again:
b3f8d68f
YZ		 2991 if (rec.v2.flock_len) {
2992 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2993 } else {
2994 num_fcntl_locks = 0;
2995 num_flock_locks = 0;
2996 }
4deb14a2 2997 if (num_fcntl_locks + num_flock_locks > 0) {
6da2ec56
KC		 2998 flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
2999 sizeof(struct ceph_filelock),
3000 GFP_NOFS);
4deb14a2
YZ		 3001 if (!flocks) {
3002 err = -ENOMEM;
3003 goto out_free;
3004 }
3005 err = ceph_encode_locks_to_buffer(inode, flocks,
3006 num_fcntl_locks,
3007 num_flock_locks);
3008 if (err) {
3009 kfree(flocks);
3010 flocks = NULL;
3011 if (err == -ENOSPC)
3012 goto encode_again;
3013 goto out_free;
3014 }
3015 } else {
39be95e9 3016 kfree(flocks);
4deb14a2 3017 flocks = NULL;
39be95e9 3018 }
121f22a1
YZ		 3019
3020 if (recon_state->msg_version >= 3) {
3021 /* version, compat_version and struct_len */
3022 total_len = 2 * sizeof(u8) + sizeof(u32);
3469ed0d 3023 struct_v = 2;
121f22a1 3024 }
39be95e9
JS		 3025 /*
3026 * number of encoded locks is stable, so copy to pagelist
3027 */
121f22a1
YZ		 3028 struct_len = 2 * sizeof(u32) +
3029 (num_fcntl_locks + num_flock_locks) *
3030 sizeof(struct ceph_filelock);
3031 rec.v2.flock_len = cpu_to_le32(struct_len);
3032
3033 struct_len += sizeof(rec.v2);
3034 struct_len += sizeof(u32) + pathlen;
3035
3469ed0d
YZ		 3036 if (struct_v >= 2)
3037 struct_len += sizeof(u64); /* snap_follows */
3038
121f22a1
YZ		 3039 total_len += struct_len;
3040 err = ceph_pagelist_reserve(pagelist, total_len);
3041
3042 if (!err) {
3043 if (recon_state->msg_version >= 3) {
3044 ceph_pagelist_encode_8(pagelist, struct_v);
3045 ceph_pagelist_encode_8(pagelist, 1);
3046 ceph_pagelist_encode_32(pagelist, struct_len);
3047 }
3048 ceph_pagelist_encode_string(pagelist, path, pathlen);
3049 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3050 ceph_locks_to_pagelist(flocks, pagelist,
3051 num_fcntl_locks,
3052 num_flock_locks);
3469ed0d
YZ		 3053 if (struct_v >= 2)
3054 ceph_pagelist_encode_64(pagelist, snap_follows);
121f22a1 3055 }
39be95e9 3056 kfree(flocks);
3612abbd 3057 } else {
121f22a1
YZ		 3058 size_t size = sizeof(u32) + pathlen + sizeof(rec.v1);
3059 err = ceph_pagelist_reserve(pagelist, size);
3060 if (!err) {
3061 ceph_pagelist_encode_string(pagelist, path, pathlen);
3062 ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3063 }
40819f6f 3064 }
44c99757
YZ		 3065
3066 recon_state->nr_caps++;
e072f8aa 3067out_free:
2f2dc053 3068 kfree(path);
e072f8aa 3069out_dput:
2f2dc053 3070 dput(dentry);
93cea5be 3071 return err;
2f2dc053
SW		 3072}
3073
3074
3075/*
3076 * If an MDS fails and recovers, clients need to reconnect in order to
3077 * reestablish shared state. This includes all caps issued through
3078 * this session _and_ the snap_realm hierarchy. Because it's not
3079 * clear which snap realms the mds cares about, we send everything we
3080 * know about.. that ensures we'll then get any new info the
3081 * recovering MDS might have.
3082 *
3083 * This is a relatively heavyweight operation, but it's rare.
3084 *
3085 * called with mdsc->mutex held.
3086 */
34b6c855
SW		 3087static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3088 struct ceph_mds_session *session)
2f2dc053 3089{
2f2dc053 3090 struct ceph_msg *reply;
a105f00c 3091 struct rb_node *p;
34b6c855 3092 int mds = session->s_mds;
9abf82b8 3093 int err = -ENOMEM;
44c99757 3094 int s_nr_caps;
93cea5be 3095 struct ceph_pagelist *pagelist;
20cb34ae 3096 struct ceph_reconnect_state recon_state;
c8a96a31 3097 LIST_HEAD(dispose);
2f2dc053 3098
34b6c855 3099 pr_info("mds%d reconnect start\n", mds);
2f2dc053 3100
93cea5be
SW		 3101 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
3102 if (!pagelist)
3103 goto fail_nopagelist;
3104 ceph_pagelist_init(pagelist);
3105
b61c2763 3106 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 3107 if (!reply)
93cea5be 3108 goto fail_nomsg;
93cea5be 3109
34b6c855
SW		 3110 mutex_lock(&session->s_mutex);
3111 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3112 session->s_seq = 0;
2f2dc053 3113
2f2dc053 3114 dout("session %p state %s\n", session,
a687ecaf 3115 ceph_session_state_name(session->s_state));
2f2dc053 3116
99a9c273
YZ		 3117 spin_lock(&session->s_gen_ttl_lock);
3118 session->s_cap_gen++;
3119 spin_unlock(&session->s_gen_ttl_lock);
3120
3121 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 3122 /* don't know if session is readonly */
3123 session->s_readonly = 0;
99a9c273
YZ		 3124 /*
3125 * notify __ceph_remove_cap() that we are composing cap reconnect.
3126 * If a cap get released before being added to the cap reconnect,
3127 * __ceph_remove_cap() should skip queuing cap release.
3128 */
3129 session->s_cap_reconnect = 1;
e01a5946 3130 /* drop old cap expires; we're about to reestablish that state */
c8a96a31
JL		 3131 detach_cap_releases(session, &dispose);
3132 spin_unlock(&session->s_cap_lock);
3133 dispose_cap_releases(mdsc, &dispose);
e01a5946 3134
5d23371f 3135 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 3136 if (mdsc->fsc->sb->s_root)
3137 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 3138
3139 ceph_con_close(&session->s_con);
3140 ceph_con_open(&session->s_con,
3141 CEPH_ENTITY_TYPE_MDS, mds,
3142 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3143
3144 /* replay unsafe requests */
3145 replay_unsafe_requests(mdsc, session);
3146
3147 down_read(&mdsc->snap_rwsem);
3148
2f2dc053 3149 /* traverse this session's caps */
44c99757
YZ		 3150 s_nr_caps = session->s_nr_caps;
3151 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 3152 if (err)
3153 goto fail;
20cb34ae 3154
44c99757 3155 recon_state.nr_caps = 0;
20cb34ae 3156 recon_state.pagelist = pagelist;
121f22a1
YZ		 3157 if (session->s_con.peer_features & CEPH_FEATURE_MDSENC)
3158 recon_state.msg_version = 3;
3159 else if (session->s_con.peer_features & CEPH_FEATURE_FLOCK)
3160 recon_state.msg_version = 2;
3161 else
3162 recon_state.msg_version = 1;
20cb34ae 3163 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 3164 if (err < 0)
9abf82b8 3165 goto fail;
2f2dc053 3166
99a9c273
YZ		 3167 spin_lock(&session->s_cap_lock);
3168 session->s_cap_reconnect = 0;
3169 spin_unlock(&session->s_cap_lock);
3170
2f2dc053
SW		 3171 /*
3172 * snaprealms. we provide mds with the ino, seq (version), and
3173 * parent for all of our realms. If the mds has any newer info,
3174 * it will tell us.
3175 */
a105f00c
SW		 3176 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3177 struct ceph_snap_realm *realm =
3178 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 3179 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 3180
3181 dout(" adding snap realm %llx seq %lld parent %llx\n",
3182 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 3183 sr_rec.ino = cpu_to_le64(realm->ino);
3184 sr_rec.seq = cpu_to_le64(realm->seq);
3185 sr_rec.parent = cpu_to_le64(realm->parent_ino);
3186 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3187 if (err)
3188 goto fail;
2f2dc053 3189 }
2f2dc053 3190
121f22a1 3191 reply->hdr.version = cpu_to_le16(recon_state.msg_version);
44c99757
YZ		 3192
3193 /* raced with cap release? */
3194 if (s_nr_caps != recon_state.nr_caps) {
3195 struct page *page = list_first_entry(&pagelist->head,
3196 struct page, lru);
3197 __le32 *addr = kmap_atomic(page);
3198 *addr = cpu_to_le32(recon_state.nr_caps);
3199 kunmap_atomic(addr);
ebf18f47 3200 }
44c99757
YZ		 3201
3202 reply->hdr.data_len = cpu_to_le32(pagelist->length);
3203 ceph_msg_data_add_pagelist(reply, pagelist);
e548e9b9
YZ		 3204
3205 ceph_early_kick_flushing_caps(mdsc, session);
3206
2f2dc053
SW		 3207 ceph_con_send(&session->s_con, reply);
3208
9abf82b8
SW		 3209 mutex_unlock(&session->s_mutex);
3210
3211 mutex_lock(&mdsc->mutex);
3212 __wake_requests(mdsc, &session->s_waiting);
3213 mutex_unlock(&mdsc->mutex);
3214
2f2dc053 3215 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 3216 return;
3217
93cea5be 3218fail:
2f2dc053 3219 ceph_msg_put(reply);
9abf82b8
SW		 3220 up_read(&mdsc->snap_rwsem);
3221 mutex_unlock(&session->s_mutex);
93cea5be
SW		 3222fail_nomsg:
3223 ceph_pagelist_release(pagelist);
93cea5be 3224fail_nopagelist:
9abf82b8 3225 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 3226 return;
2f2dc053
SW		 3227}
3228
3229
3230/*
3231 * compare old and new mdsmaps, kicking requests
3232 * and closing out old connections as necessary
3233 *
3234 * called under mdsc->mutex.
3235 */
3236static void check_new_map(struct ceph_mds_client *mdsc,
3237 struct ceph_mdsmap *newmap,
3238 struct ceph_mdsmap *oldmap)
3239{
3240 int i;
3241 int oldstate, newstate;
3242 struct ceph_mds_session *s;
3243
3244 dout("check_new_map new %u old %u\n",
3245 newmap->m_epoch, oldmap->m_epoch);
3246
76201b63 3247 for (i = 0; i < oldmap->m_num_mds && i < mdsc->max_sessions; i++) {
d37b1d99 3248 if (!mdsc->sessions[i])
2f2dc053
SW		 3249 continue;
3250 s = mdsc->sessions[i];
3251 oldstate = ceph_mdsmap_get_state(oldmap, i);
3252 newstate = ceph_mdsmap_get_state(newmap, i);
3253
0deb01c9 3254 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3255 i, ceph_mds_state_name(oldstate),
0deb01c9 3256 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3257 ceph_mds_state_name(newstate),
0deb01c9 3258 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3259 ceph_session_state_name(s->s_state));
2f2dc053 3260
76201b63 3261 if (i >= newmap->m_num_mds ||
3e8f43a0 3262 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3263 ceph_mdsmap_get_addr(newmap, i),
3264 sizeof(struct ceph_entity_addr))) {
3265 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3266 /* the session never opened, just close it
3267 * out now */
0a07fc8c 3268 get_session(s);
2600d2dd 3269 __unregister_session(mdsc, s);
2f2dc053 3270 __wake_requests(mdsc, &s->s_waiting);
0a07fc8c 3271 ceph_put_mds_session(s);
2827528d
YZ		 3272 } else if (i >= newmap->m_num_mds) {
3273 /* force close session for stopped mds */
3274 get_session(s);
2600d2dd 3275 __unregister_session(mdsc, s);
2827528d
YZ		 3276 __wake_requests(mdsc, &s->s_waiting);
3277 kick_requests(mdsc, i);
3278 mutex_unlock(&mdsc->mutex);
3279
3280 mutex_lock(&s->s_mutex);
3281 cleanup_session_requests(mdsc, s);
3282 remove_session_caps(s);
3283 mutex_unlock(&s->s_mutex);
3284
3285 ceph_put_mds_session(s);
3286
3287 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3288 } else {
3289 /* just close it */
3290 mutex_unlock(&mdsc->mutex);
3291 mutex_lock(&s->s_mutex);
3292 mutex_lock(&mdsc->mutex);
3293 ceph_con_close(&s->s_con);
3294 mutex_unlock(&s->s_mutex);
3295 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3296 }
2f2dc053
SW		 3297 } else if (oldstate == newstate) {
3298 continue; /* nothing new with this mds */
3299 }
3300
3301 /*
3302 * send reconnect?
3303 */
3304 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3305 newstate >= CEPH_MDS_STATE_RECONNECT) {
3306 mutex_unlock(&mdsc->mutex);
3307 send_mds_reconnect(mdsc, s);
3308 mutex_lock(&mdsc->mutex);
3309 }
2f2dc053
SW		 3310
3311 /*
29790f26 3312 * kick request on any mds that has gone active.
2f2dc053
SW		 3313 */
3314 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3315 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3316 if (oldstate != CEPH_MDS_STATE_CREATING &&
3317 oldstate != CEPH_MDS_STATE_STARTING)
3318 pr_info("mds%d recovery completed\n", s->s_mds);
3319 kick_requests(mdsc, i);
2f2dc053 3320 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3321 wake_up_session_caps(s, 1);
2f2dc053
SW		 3322 }
3323 }
cb170a22 3324
76201b63 3325 for (i = 0; i < newmap->m_num_mds && i < mdsc->max_sessions; i++) {
cb170a22
SW		 3326 s = mdsc->sessions[i];
3327 if (!s)
3328 continue;
3329 if (!ceph_mdsmap_is_laggy(newmap, i))
3330 continue;
3331 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3332 s->s_state == CEPH_MDS_SESSION_HUNG ||
3333 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3334 dout(" connecting to export targets of laggy mds%d\n",
3335 i);
3336 __open_export_target_sessions(mdsc, s);
3337 }
3338 }
2f2dc053
SW		 3339}
3340
3341
3342
3343/*
3344 * leases
3345 */
3346
3347/*
3348 * caller must hold session s_mutex, dentry->d_lock
3349 */
3350void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3351{
3352 struct ceph_dentry_info *di = ceph_dentry(dentry);
3353
3354 ceph_put_mds_session(di->lease_session);
3355 di->lease_session = NULL;
3356}
3357
2600d2dd
SW		 3358static void handle_lease(struct ceph_mds_client *mdsc,
3359 struct ceph_mds_session *session,
3360 struct ceph_msg *msg)
2f2dc053 3361{
3d14c5d2 3362 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3363 struct inode *inode;
2f2dc053
SW		 3364 struct dentry *parent, *dentry;
3365 struct ceph_dentry_info *di;
2600d2dd 3366 int mds = session->s_mds;
2f2dc053 3367 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3368 u32 seq;
2f2dc053 3369 struct ceph_vino vino;
2f2dc053
SW		 3370 struct qstr dname;
3371 int release = 0;
3372
2f2dc053
SW		 3373 dout("handle_lease from mds%d\n", mds);
3374
3375 /* decode */
3376 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3377 goto bad;
3378 vino.ino = le64_to_cpu(h->ino);
3379 vino.snap = CEPH_NOSNAP;
1e5ea23d 3380 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3381 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3382 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3383 if (dname.len != get_unaligned_le32(h+1))
3384 goto bad;
3385
2f2dc053
SW		 3386 /* lookup inode */
3387 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3388 dout("handle_lease %s, ino %llx %p %.*s\n",
3389 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3390 dname.len, dname.name);
6cd3bcad
YZ		 3391
3392 mutex_lock(&session->s_mutex);
3393 session->s_seq++;
3394
d37b1d99 3395 if (!inode) {
2f2dc053
SW		 3396 dout("handle_lease no inode %llx\n", vino.ino);
3397 goto release;
3398 }
2f2dc053
SW		 3399
3400 /* dentry */
3401 parent = d_find_alias(inode);
3402 if (!parent) {
3403 dout("no parent dentry on inode %p\n", inode);
3404 WARN_ON(1);
3405 goto release; /* hrm... */
3406 }
8387ff25 3407 dname.hash = full_name_hash(parent, dname.name, dname.len);
2f2dc053
SW		 3408 dentry = d_lookup(parent, &dname);
3409 dput(parent);
3410 if (!dentry)
3411 goto release;
3412
3413 spin_lock(&dentry->d_lock);
3414 di = ceph_dentry(dentry);
3415 switch (h->action) {
3416 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3417 if (di->lease_session == session) {
1e5ea23d
SW		 3418 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3419 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3420 __ceph_mdsc_drop_dentry_lease(dentry);
3421 }
3422 release = 1;
3423 break;
3424
3425 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3426 if (di->lease_session == session &&
2f2dc053
SW		 3427 di->lease_gen == session->s_cap_gen &&
3428 di->lease_renew_from &&
3429 di->lease_renew_after == 0) {
3430 unsigned long duration =
3563dbdd 3431 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3432
1e5ea23d 3433 di->lease_seq = seq;
9b16f03c 3434 di->time = di->lease_renew_from + duration;
2f2dc053
SW		 3435 di->lease_renew_after = di->lease_renew_from +
3436 (duration >> 1);
3437 di->lease_renew_from = 0;
3438 }
3439 break;
3440 }
3441 spin_unlock(&dentry->d_lock);
3442 dput(dentry);
3443
3444 if (!release)
3445 goto out;
3446
3447release:
3448 /* let's just reuse the same message */
3449 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3450 ceph_msg_get(msg);
3451 ceph_con_send(&session->s_con, msg);
3452
3453out:
3454 iput(inode);
3455 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3456 return;
3457
3458bad:
3459 pr_err("corrupt lease message\n");
9ec7cab1 3460 ceph_msg_dump(msg);
2f2dc053
SW		 3461}
3462
3463void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3464 struct inode *inode,
3465 struct dentry *dentry, char action,
3466 u32 seq)
3467{
3468 struct ceph_msg *msg;
3469 struct ceph_mds_lease *lease;
3470 int len = sizeof(*lease) + sizeof(u32);
3471 int dnamelen = 0;
3472
3473 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3474 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3475 dnamelen = dentry->d_name.len;
3476 len += dnamelen;
3477
b61c2763 3478 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3479 if (!msg)
2f2dc053
SW		 3480 return;
3481 lease = msg->front.iov_base;
3482 lease->action = action;
2f2dc053
SW		 3483 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3484 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3485 lease->seq = cpu_to_le32(seq);
3486 put_unaligned_le32(dnamelen, lease + 1);
3487 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3488
3489 /*
3490 * if this is a preemptive lease RELEASE, no need to
3491 * flush request stream, since the actual request will
3492 * soon follow.
3493 */
3494 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3495
3496 ceph_con_send(&session->s_con, msg);
3497}
3498
2f2dc053 3499/*
7aac453a 3500 * lock unlock sessions, to wait ongoing session activities
2f2dc053 3501 */
7aac453a 3502static void lock_unlock_sessions(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3503{
3504 int i;
3505
2f2dc053
SW		 3506 mutex_lock(&mdsc->mutex);
3507 for (i = 0; i < mdsc->max_sessions; i++) {
3508 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3509 if (!s)
3510 continue;
3511 mutex_unlock(&mdsc->mutex);
3512 mutex_lock(&s->s_mutex);
3513 mutex_unlock(&s->s_mutex);
3514 ceph_put_mds_session(s);
3515 mutex_lock(&mdsc->mutex);
3516 }
3517 mutex_unlock(&mdsc->mutex);
3518}
3519
3520
3521
3522/*
3523 * delayed work -- periodically trim expired leases, renew caps with mds
3524 */
3525static void schedule_delayed(struct ceph_mds_client *mdsc)
3526{
3527 int delay = 5;
3528 unsigned hz = round_jiffies_relative(HZ * delay);
3529 schedule_delayed_work(&mdsc->delayed_work, hz);
3530}
3531
3532static void delayed_work(struct work_struct *work)
3533{
3534 int i;
3535 struct ceph_mds_client *mdsc =
3536 container_of(work, struct ceph_mds_client, delayed_work.work);
3537 int renew_interval;
3538 int renew_caps;
3539
3540 dout("mdsc delayed_work\n");
afcdaea3 3541 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3542
3543 mutex_lock(&mdsc->mutex);
3544 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3545 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3546 mdsc->last_renew_caps);
3547 if (renew_caps)
3548 mdsc->last_renew_caps = jiffies;
3549
3550 for (i = 0; i < mdsc->max_sessions; i++) {
3551 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
d37b1d99 3552 if (!s)
2f2dc053
SW		 3553 continue;
3554 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3555 dout("resending session close request for mds%d\n",
3556 s->s_mds);
3557 request_close_session(mdsc, s);
3558 ceph_put_mds_session(s);
3559 continue;
3560 }
3561 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3562 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3563 s->s_state = CEPH_MDS_SESSION_HUNG;
3564 pr_info("mds%d hung\n", s->s_mds);
3565 }
3566 }
3567 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3568 /* this mds is failed or recovering, just wait */
3569 ceph_put_mds_session(s);
3570 continue;
3571 }
3572 mutex_unlock(&mdsc->mutex);
3573
3574 mutex_lock(&s->s_mutex);
3575 if (renew_caps)
3576 send_renew_caps(mdsc, s);
3577 else
3578 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3579 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3580 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3581 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3582 mutex_unlock(&s->s_mutex);
3583 ceph_put_mds_session(s);
3584
3585 mutex_lock(&mdsc->mutex);
3586 }
3587 mutex_unlock(&mdsc->mutex);
3588
3589 schedule_delayed(mdsc);
3590}
3591
3d14c5d2 3592int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3593
2f2dc053 3594{
3d14c5d2
YS		 3595 struct ceph_mds_client *mdsc;
3596
3597 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3598 if (!mdsc)
3599 return -ENOMEM;
3600 mdsc->fsc = fsc;
2f2dc053
SW		 3601 mutex_init(&mdsc->mutex);
3602 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
d37b1d99 3603 if (!mdsc->mdsmap) {
fb3101b6 3604 kfree(mdsc);
2d06eeb8 3605 return -ENOMEM;
fb3101b6 3606 }
2d06eeb8 3607
50c55aec 3608 fsc->mdsc = mdsc;
2f2dc053 3609 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3610 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3611 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3612 mdsc->sessions = NULL;
86d8f67b 3613 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3614 mdsc->max_sessions = 0;
3615 mdsc->stopping = 0;
d557c48d 3616 atomic64_set(&mdsc->quotarealms_count, 0);
affbc19a 3617 mdsc->last_snap_seq = 0;
2f2dc053 3618 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3619 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3620 INIT_LIST_HEAD(&mdsc->snap_empty);
3621 spin_lock_init(&mdsc->snap_empty_lock);
3622 mdsc->last_tid = 0;
e8a7b8b1 3623 mdsc->oldest_tid = 0;
44ca18f2 3624 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3625 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3626 mdsc->last_renew_caps = jiffies;
3627 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3628 spin_lock_init(&mdsc->cap_delay_lock);
3629 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3630 spin_lock_init(&mdsc->snap_flush_lock);
553adfd9 3631 mdsc->last_cap_flush_tid = 1;
e4500b5e 3632 INIT_LIST_HEAD(&mdsc->cap_flush_list);
2f2dc053 3633 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3634 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3635 mdsc->num_cap_flushing = 0;
3636 spin_lock_init(&mdsc->cap_dirty_lock);
3637 init_waitqueue_head(&mdsc->cap_flushing_wq);
3638 spin_lock_init(&mdsc->dentry_lru_lock);
3639 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3640
37151668 3641 ceph_caps_init(mdsc);
3d14c5d2 3642 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3643
10183a69
YZ		 3644 init_rwsem(&mdsc->pool_perm_rwsem);
3645 mdsc->pool_perm_tree = RB_ROOT;
3646
dfeb84d4
YZ		 3647 strscpy(mdsc->nodename, utsname()->nodename,
3648 sizeof(mdsc->nodename));
5f44f142 3649 return 0;
2f2dc053
SW		 3650}
3651
3652/*
3653 * Wait for safe replies on open mds requests. If we time out, drop
3654 * all requests from the tree to avoid dangling dentry refs.
3655 */
3656static void wait_requests(struct ceph_mds_client *mdsc)
3657{
a319bf56 3658 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3659 struct ceph_mds_request *req;
2f2dc053
SW		 3660
3661 mutex_lock(&mdsc->mutex);
44ca18f2 3662 if (__get_oldest_req(mdsc)) {
2f2dc053 3663 mutex_unlock(&mdsc->mutex);
44ca18f2 3664
2f2dc053
SW		 3665 dout("wait_requests waiting for requests\n");
3666 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3667 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3668
3669 /* tear down remaining requests */
44ca18f2
SW		 3670 mutex_lock(&mdsc->mutex);
3671 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3672 dout("wait_requests timed out on tid %llu\n",
3673 req->r_tid);
44ca18f2 3674 __unregister_request(mdsc, req);
2f2dc053
SW		 3675 }
3676 }
3677 mutex_unlock(&mdsc->mutex);
3678 dout("wait_requests done\n");
3679}
3680
3681/*
3682 * called before mount is ro, and before dentries are torn down.
3683 * (hmm, does this still race with new lookups?)
3684 */
3685void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3686{
3687 dout("pre_umount\n");
3688 mdsc->stopping = 1;
3689
7aac453a 3690 lock_unlock_sessions(mdsc);
afcdaea3 3691 ceph_flush_dirty_caps(mdsc);
2f2dc053 3692 wait_requests(mdsc);
17c688c3
SW		 3693
3694 /*
3695 * wait for reply handlers to drop their request refs and
3696 * their inode/dcache refs
3697 */
3698 ceph_msgr_flush();
2f2dc053
SW		 3699}
3700
3701/*
3702 * wait for all write mds requests to flush.
3703 */
3704static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3705{
80fc7314 3706 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3707 struct rb_node *n;
2f2dc053
SW		 3708
3709 mutex_lock(&mdsc->mutex);
3710 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3711restart:
44ca18f2
SW		 3712 req = __get_oldest_req(mdsc);
3713 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3714 /* find next request */
3715 n = rb_next(&req->r_node);
3716 if (n)
3717 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3718 else
3719 nextreq = NULL;
e8a7b8b1
YZ		 3720 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3721 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3722 /* write op */
3723 ceph_mdsc_get_request(req);
80fc7314
SW		 3724 if (nextreq)
3725 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3726 mutex_unlock(&mdsc->mutex);
3727 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3728 req->r_tid, want_tid);
3729 wait_for_completion(&req->r_safe_completion);
3730 mutex_lock(&mdsc->mutex);
44ca18f2 3731 ceph_mdsc_put_request(req);
80fc7314
SW		 3732 if (!nextreq)
3733 break; /* next dne before, so we're done! */
3734 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3735 /* next request was removed from tree */
3736 ceph_mdsc_put_request(nextreq);
3737 goto restart;
3738 }
3739 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3740 }
80fc7314 3741 req = nextreq;
2f2dc053
SW		 3742 }
3743 mutex_unlock(&mdsc->mutex);
3744 dout("wait_unsafe_requests done\n");
3745}
3746
3747void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3748{
0e294387 3749 u64 want_tid, want_flush;
2f2dc053 3750
52953d55 3751 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3752 return;
3753
2f2dc053
SW		 3754 dout("sync\n");
3755 mutex_lock(&mdsc->mutex);
3756 want_tid = mdsc->last_tid;
2f2dc053 3757 mutex_unlock(&mdsc->mutex);
2f2dc053 3758
afcdaea3 3759 ceph_flush_dirty_caps(mdsc);
d3383a8e 3760 spin_lock(&mdsc->cap_dirty_lock);
8310b089 3761 want_flush = mdsc->last_cap_flush_tid;
c8799fc4
YZ		 3762 if (!list_empty(&mdsc->cap_flush_list)) {
3763 struct ceph_cap_flush *cf =
3764 list_last_entry(&mdsc->cap_flush_list,
3765 struct ceph_cap_flush, g_list);
3766 cf->wake = true;
3767 }
d3383a8e
YZ		 3768 spin_unlock(&mdsc->cap_dirty_lock);
3769
0e294387
YZ		 3770 dout("sync want tid %lld flush_seq %lld\n",
3771 want_tid, want_flush);
2f2dc053
SW		 3772
3773 wait_unsafe_requests(mdsc, want_tid);
0e294387 3774 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3775}
3776
f3c60c59
SW		 3777/*
3778 * true if all sessions are closed, or we force unmount
3779 */
fcff415c 3780static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
f3c60c59 3781{
52953d55 3782 if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3783 return true;
fcff415c 3784 return atomic_read(&mdsc->num_sessions) <= skipped;
f3c60c59 3785}
2f2dc053
SW		 3786
3787/*
3788 * called after sb is ro.
3789 */
3790void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3791{
a319bf56 3792 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3793 struct ceph_mds_session *session;
3794 int i;
fcff415c 3795 int skipped = 0;
2f2dc053
SW		 3796
3797 dout("close_sessions\n");
3798
2f2dc053 3799 /* close sessions */
f3c60c59
SW		 3800 mutex_lock(&mdsc->mutex);
3801 for (i = 0; i < mdsc->max_sessions; i++) {
3802 session = __ceph_lookup_mds_session(mdsc, i);
3803 if (!session)
3804 continue;
2f2dc053 3805 mutex_unlock(&mdsc->mutex);
f3c60c59 3806 mutex_lock(&session->s_mutex);
fcff415c
YZ		 3807 if (__close_session(mdsc, session) <= 0)
3808 skipped++;
f3c60c59
SW		 3809 mutex_unlock(&session->s_mutex);
3810 ceph_put_mds_session(session);
2f2dc053
SW		 3811 mutex_lock(&mdsc->mutex);
3812 }
f3c60c59
SW		 3813 mutex_unlock(&mdsc->mutex);
3814
3815 dout("waiting for sessions to close\n");
fcff415c
YZ		 3816 wait_event_timeout(mdsc->session_close_wq,
3817 done_closing_sessions(mdsc, skipped),
a319bf56 3818 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3819
3820 /* tear down remaining sessions */
f3c60c59 3821 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3822 for (i = 0; i < mdsc->max_sessions; i++) {
3823 if (mdsc->sessions[i]) {
3824 session = get_session(mdsc->sessions[i]);
2600d2dd 3825 __unregister_session(mdsc, session);
2f2dc053
SW		 3826 mutex_unlock(&mdsc->mutex);
3827 mutex_lock(&session->s_mutex);
3828 remove_session_caps(session);
3829 mutex_unlock(&session->s_mutex);
3830 ceph_put_mds_session(session);
3831 mutex_lock(&mdsc->mutex);
3832 }
3833 }
2f2dc053 3834 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3835 mutex_unlock(&mdsc->mutex);
3836
3837 ceph_cleanup_empty_realms(mdsc);
3838
3839 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3840
3841 dout("stopped\n");
3842}
3843
48fec5d0
YZ		 3844void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
3845{
3846 struct ceph_mds_session *session;
3847 int mds;
3848
3849 dout("force umount\n");
3850
3851 mutex_lock(&mdsc->mutex);
3852 for (mds = 0; mds < mdsc->max_sessions; mds++) {
3853 session = __ceph_lookup_mds_session(mdsc, mds);
3854 if (!session)
3855 continue;
3856 mutex_unlock(&mdsc->mutex);
3857 mutex_lock(&session->s_mutex);
3858 __close_session(mdsc, session);
3859 if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
3860 cleanup_session_requests(mdsc, session);
3861 remove_session_caps(session);
3862 }
3863 mutex_unlock(&session->s_mutex);
3864 ceph_put_mds_session(session);
3865 mutex_lock(&mdsc->mutex);
3866 kick_requests(mdsc, mds);
3867 }
3868 __wake_requests(mdsc, &mdsc->waiting_for_map);
3869 mutex_unlock(&mdsc->mutex);
3870}
3871
3d14c5d2 3872static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3873{
3874 dout("stop\n");
3875 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3876 if (mdsc->mdsmap)
3877 ceph_mdsmap_destroy(mdsc->mdsmap);
3878 kfree(mdsc->sessions);
37151668 3879 ceph_caps_finalize(mdsc);
10183a69 3880 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3881}
3882
3d14c5d2
YS		 3883void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3884{
3885 struct ceph_mds_client *mdsc = fsc->mdsc;
ef550f6f 3886 dout("mdsc_destroy %p\n", mdsc);
ef550f6f 3887
50c55aec
CX		 3888 if (!mdsc)
3889 return;
3890
ef550f6f
SW		 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 4109 break;
fb18a575
LH		 4110 case CEPH_MSG_CLIENT_QUOTA:
4111 ceph_handle_quota(mdsc, s, msg);
4112 break;
2f2dc053
SW		 4113
4114 default:
4115 pr_err("received unknown message type %d %s\n", type,
4116 ceph_msg_type_name(type));
4117 }
2600d2dd 4118out:
2f2dc053
SW		 4119 ceph_msg_put(msg);
4120}
4121
4e7a5dcd
SW		 4122/*
4123 * authentication
4124 */
a3530df3
AE		 4125
4126/*
4127 * Note: returned pointer is the address of a structure that's
4128 * managed separately. Caller must *not* attempt to free it.
4129 */
4130static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 4131 int *proto, int force_new)
4e7a5dcd
SW		 4132{
4133 struct ceph_mds_session *s = con->private;
4134 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4135 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 4136 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 4137
74f1869f 4138 if (force_new && auth->authorizer) {
6c1ea260 4139 ceph_auth_destroy_authorizer(auth->authorizer);
74f1869f 4140 auth->authorizer = NULL;
4e7a5dcd 4141 }
27859f97
SW		 4142 if (!auth->authorizer) {
4143 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4144 auth);
0bed9b5c
SW		 4145 if (ret)
4146 return ERR_PTR(ret);
27859f97
SW		 4147 } else {
4148 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4149 auth);
a255651d 4150 if (ret)
a3530df3 4151 return ERR_PTR(ret);
4e7a5dcd 4152 }
4e7a5dcd 4153 *proto = ac->protocol;
74f1869f 4154
a3530df3 4155 return auth;
4e7a5dcd
SW		 4156}
4157
4158
0dde5848 4159static int verify_authorizer_reply(struct ceph_connection *con)
4e7a5dcd
SW		 4160{
4161 struct ceph_mds_session *s = con->private;
4162 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4163 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 4164
0dde5848 4165 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4e7a5dcd
SW		 4166}
4167
9bd2e6f8
SW		 4168static int invalidate_authorizer(struct ceph_connection *con)
4169{
4170 struct ceph_mds_session *s = con->private;
4171 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 4172 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 4173
27859f97 4174 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 4175
3d14c5d2 4176 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 4177}
4178
53ded495
AE		 4179static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4180 struct ceph_msg_header *hdr, int *skip)
4181{
4182 struct ceph_msg *msg;
4183 int type = (int) le16_to_cpu(hdr->type);
4184 int front_len = (int) le32_to_cpu(hdr->front_len);
4185
4186 if (con->in_msg)
4187 return con->in_msg;
4188
4189 *skip = 0;
4190 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4191 if (!msg) {
4192 pr_err("unable to allocate msg type %d len %d\n",
4193 type, front_len);
4194 return NULL;
4195 }
53ded495
AE		 4196
4197 return msg;
4198}
4199
79dbd1ba 4200static int mds_sign_message(struct ceph_msg *msg)
33d07337 4201{
79dbd1ba 4202 struct ceph_mds_session *s = msg->con->private;
33d07337 4203 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4204
33d07337
YZ		 4205 return ceph_auth_sign_message(auth, msg);
4206}
4207
79dbd1ba 4208static int mds_check_message_signature(struct ceph_msg *msg)
33d07337 4209{
79dbd1ba 4210 struct ceph_mds_session *s = msg->con->private;
33d07337 4211 struct ceph_auth_handshake *auth = &s->s_auth;
79dbd1ba 4212
33d07337
YZ		 4213 return ceph_auth_check_message_signature(auth, msg);
4214}
4215
9e32789f 4216static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 4217 .get = con_get,
4218 .put = con_put,
4219 .dispatch = dispatch,
4e7a5dcd
SW		 4220 .get_authorizer = get_authorizer,
4221 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 4222 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 4223 .peer_reset = peer_reset,
53ded495 4224 .alloc_msg = mds_alloc_msg,
79dbd1ba
ID		 4225 .sign_message = mds_sign_message,
4226 .check_message_signature = mds_check_message_signature,
2f2dc053
SW		 4227};
4228
2f2dc053 4229/* eof */
Linux 6.x block layer and io_uring tree(s)