]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/ceph/mds_client.c
projects / mirror_ubuntu-eoan-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: avoid block operation when !TASK_RUNNING (ceph_get_caps)
[mirror_ubuntu-eoan-kernel.git] / fs / ceph / mds_client.c
CommitLineData
3d14c5d2 1#include <linux/ceph/ceph_debug.h>
2f2dc053 2
496e5955 3#include <linux/fs.h>
2f2dc053 4#include <linux/wait.h>
5a0e3ad6 5#include <linux/slab.h>
54008399 6#include <linux/gfp.h>
2f2dc053 7#include <linux/sched.h>
3d14c5d2
YS		 8#include <linux/debugfs.h>
9#include <linux/seq_file.h>
dbd0c8bf 10#include <linux/utsname.h>
2f2dc053 11
2f2dc053 12#include "super.h"
3d14c5d2
YS		 13#include "mds_client.h"
14
1fe60e51 15#include <linux/ceph/ceph_features.h>
3d14c5d2
YS		 16#include <linux/ceph/messenger.h>
17#include <linux/ceph/decode.h>
18#include <linux/ceph/pagelist.h>
19#include <linux/ceph/auth.h>
20#include <linux/ceph/debugfs.h>
2f2dc053
SW		 21
22/*
23 * A cluster of MDS (metadata server) daemons is responsible for
24 * managing the file system namespace (the directory hierarchy and
25 * inodes) and for coordinating shared access to storage. Metadata is
26 * partitioning hierarchically across a number of servers, and that
27 * partition varies over time as the cluster adjusts the distribution
28 * in order to balance load.
29 *
30 * The MDS client is primarily responsible to managing synchronous
31 * metadata requests for operations like open, unlink, and so forth.
32 * If there is a MDS failure, we find out about it when we (possibly
33 * request and) receive a new MDS map, and can resubmit affected
34 * requests.
35 *
36 * For the most part, though, we take advantage of a lossless
37 * communications channel to the MDS, and do not need to worry about
38 * timing out or resubmitting requests.
39 *
40 * We maintain a stateful "session" with each MDS we interact with.
41 * Within each session, we sent periodic heartbeat messages to ensure
42 * any capabilities or leases we have been issues remain valid. If
43 * the session times out and goes stale, our leases and capabilities
44 * are no longer valid.
45 */
46
20cb34ae 47struct ceph_reconnect_state {
44c99757 48 int nr_caps;
20cb34ae
SW		 49 struct ceph_pagelist *pagelist;
50 bool flock;
51};
52
2f2dc053
SW		 53static void __wake_requests(struct ceph_mds_client *mdsc,
54 struct list_head *head);
55
9e32789f 56static const struct ceph_connection_operations mds_con_ops;
2f2dc053
SW		 57
58
59/*
60 * mds reply parsing
61 */
62
63/*
64 * parse individual inode info
65 */
66static int parse_reply_info_in(void **p, void *end,
14303d20 67 struct ceph_mds_reply_info_in *info,
12b4629a 68 u64 features)
2f2dc053
SW		 69{
70 int err = -EIO;
71
72 info->in = *p;
73 *p += sizeof(struct ceph_mds_reply_inode) +
74 sizeof(*info->in->fragtree.splits) *
75 le32_to_cpu(info->in->fragtree.nsplits);
76
77 ceph_decode_32_safe(p, end, info->symlink_len, bad);
78 ceph_decode_need(p, end, info->symlink_len, bad);
79 info->symlink = *p;
80 *p += info->symlink_len;
81
14303d20
SW		 82 if (features & CEPH_FEATURE_DIRLAYOUTHASH)
83 ceph_decode_copy_safe(p, end, &info->dir_layout,
84 sizeof(info->dir_layout), bad);
85 else
86 memset(&info->dir_layout, 0, sizeof(info->dir_layout));
87
2f2dc053
SW		 88 ceph_decode_32_safe(p, end, info->xattr_len, bad);
89 ceph_decode_need(p, end, info->xattr_len, bad);
90 info->xattr_data = *p;
91 *p += info->xattr_len;
fb01d1f8
YZ		 92
93 if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
94 ceph_decode_64_safe(p, end, info->inline_version, bad);
95 ceph_decode_32_safe(p, end, info->inline_len, bad);
96 ceph_decode_need(p, end, info->inline_len, bad);
97 info->inline_data = *p;
98 *p += info->inline_len;
99 } else
100 info->inline_version = CEPH_INLINE_NONE;
101
2f2dc053
SW		 102 return 0;
103bad:
104 return err;
105}
106
107/*
108 * parse a normal reply, which may contain a (dir+)dentry and/or a
109 * target inode.
110 */
111static int parse_reply_info_trace(void **p, void *end,
14303d20 112 struct ceph_mds_reply_info_parsed *info,
12b4629a 113 u64 features)
2f2dc053
SW		 114{
115 int err;
116
117 if (info->head->is_dentry) {
14303d20 118 err = parse_reply_info_in(p, end, &info->diri, features);
2f2dc053
SW		 119 if (err < 0)
120 goto out_bad;
121
122 if (unlikely(*p + sizeof(*info->dirfrag) > end))
123 goto bad;
124 info->dirfrag = *p;
125 *p += sizeof(*info->dirfrag) +
126 sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
127 if (unlikely(*p > end))
128 goto bad;
129
130 ceph_decode_32_safe(p, end, info->dname_len, bad);
131 ceph_decode_need(p, end, info->dname_len, bad);
132 info->dname = *p;
133 *p += info->dname_len;
134 info->dlease = *p;
135 *p += sizeof(*info->dlease);
136 }
137
138 if (info->head->is_target) {
14303d20 139 err = parse_reply_info_in(p, end, &info->targeti, features);
2f2dc053
SW		 140 if (err < 0)
141 goto out_bad;
142 }
143
144 if (unlikely(*p != end))
145 goto bad;
146 return 0;
147
148bad:
149 err = -EIO;
150out_bad:
151 pr_err("problem parsing mds trace %d\n", err);
152 return err;
153}
154
155/*
156 * parse readdir results
157 */
158static int parse_reply_info_dir(void **p, void *end,
14303d20 159 struct ceph_mds_reply_info_parsed *info,
12b4629a 160 u64 features)
2f2dc053
SW		 161{
162 u32 num, i = 0;
163 int err;
164
165 info->dir_dir = *p;
166 if (*p + sizeof(*info->dir_dir) > end)
167 goto bad;
168 *p += sizeof(*info->dir_dir) +
169 sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
170 if (*p > end)
171 goto bad;
172
173 ceph_decode_need(p, end, sizeof(num) + 2, bad);
c89136ea
SW		 174 num = ceph_decode_32(p);
175 info->dir_end = ceph_decode_8(p);
176 info->dir_complete = ceph_decode_8(p);
2f2dc053
SW		 177 if (num == 0)
178 goto done;
179
54008399 180 BUG_ON(!info->dir_in);
2f2dc053
SW		 181 info->dir_dname = (void *)(info->dir_in + num);
182 info->dir_dname_len = (void *)(info->dir_dname + num);
183 info->dir_dlease = (void *)(info->dir_dname_len + num);
54008399
YZ		 184 if ((unsigned long)(info->dir_dlease + num) >
185 (unsigned long)info->dir_in + info->dir_buf_size) {
186 pr_err("dir contents are larger than expected\n");
187 WARN_ON(1);
188 goto bad;
189 }
2f2dc053 190
54008399 191 info->dir_nr = num;
2f2dc053
SW		 192 while (num) {
193 /* dentry */
194 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea 195 info->dir_dname_len[i] = ceph_decode_32(p);
2f2dc053
SW		 196 ceph_decode_need(p, end, info->dir_dname_len[i], bad);
197 info->dir_dname[i] = *p;
198 *p += info->dir_dname_len[i];
199 dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
200 info->dir_dname[i]);
201 info->dir_dlease[i] = *p;
202 *p += sizeof(struct ceph_mds_reply_lease);
203
204 /* inode */
14303d20 205 err = parse_reply_info_in(p, end, &info->dir_in[i], features);
2f2dc053
SW		 206 if (err < 0)
207 goto out_bad;
208 i++;
209 num--;
210 }
211
212done:
213 if (*p != end)
214 goto bad;
215 return 0;
216
217bad:
218 err = -EIO;
219out_bad:
220 pr_err("problem parsing dir contents %d\n", err);
221 return err;
222}
223
25933abd
HS		 224/*
225 * parse fcntl F_GETLK results
226 */
227static int parse_reply_info_filelock(void **p, void *end,
14303d20 228 struct ceph_mds_reply_info_parsed *info,
12b4629a 229 u64 features)
25933abd
HS		 230{
231 if (*p + sizeof(*info->filelock_reply) > end)
232 goto bad;
233
234 info->filelock_reply = *p;
235 *p += sizeof(*info->filelock_reply);
236
237 if (unlikely(*p != end))
238 goto bad;
239 return 0;
240
241bad:
242 return -EIO;
243}
244
6e8575fa
SL		 245/*
246 * parse create results
247 */
248static int parse_reply_info_create(void **p, void *end,
249 struct ceph_mds_reply_info_parsed *info,
12b4629a 250 u64 features)
6e8575fa
SL		 251{
252 if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
253 if (*p == end) {
254 info->has_create_ino = false;
255 } else {
256 info->has_create_ino = true;
257 info->ino = ceph_decode_64(p);
258 }
259 }
260
261 if (unlikely(*p != end))
262 goto bad;
263 return 0;
264
265bad:
266 return -EIO;
267}
268
25933abd
HS		 269/*
270 * parse extra results
271 */
272static int parse_reply_info_extra(void **p, void *end,
14303d20 273 struct ceph_mds_reply_info_parsed *info,
12b4629a 274 u64 features)
25933abd
HS		 275{
276 if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
14303d20 277 return parse_reply_info_filelock(p, end, info, features);
8a034497
YZ		 278 else if (info->head->op == CEPH_MDS_OP_READDIR ||
279 info->head->op == CEPH_MDS_OP_LSSNAP)
14303d20 280 return parse_reply_info_dir(p, end, info, features);
6e8575fa
SL		 281 else if (info->head->op == CEPH_MDS_OP_CREATE)
282 return parse_reply_info_create(p, end, info, features);
283 else
284 return -EIO;
25933abd
HS		 285}
286
2f2dc053
SW		 287/*
288 * parse entire mds reply
289 */
290static int parse_reply_info(struct ceph_msg *msg,
14303d20 291 struct ceph_mds_reply_info_parsed *info,
12b4629a 292 u64 features)
2f2dc053
SW		 293{
294 void *p, *end;
295 u32 len;
296 int err;
297
298 info->head = msg->front.iov_base;
299 p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
300 end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
301
302 /* trace */
303 ceph_decode_32_safe(&p, end, len, bad);
304 if (len > 0) {
32852a81 305 ceph_decode_need(&p, end, len, bad);
14303d20 306 err = parse_reply_info_trace(&p, p+len, info, features);
2f2dc053
SW		 307 if (err < 0)
308 goto out_bad;
309 }
310
25933abd 311 /* extra */
2f2dc053
SW		 312 ceph_decode_32_safe(&p, end, len, bad);
313 if (len > 0) {
32852a81 314 ceph_decode_need(&p, end, len, bad);
14303d20 315 err = parse_reply_info_extra(&p, p+len, info, features);
2f2dc053
SW		 316 if (err < 0)
317 goto out_bad;
318 }
319
320 /* snap blob */
321 ceph_decode_32_safe(&p, end, len, bad);
322 info->snapblob_len = len;
323 info->snapblob = p;
324 p += len;
325
326 if (p != end)
327 goto bad;
328 return 0;
329
330bad:
331 err = -EIO;
332out_bad:
333 pr_err("mds parse_reply err %d\n", err);
334 return err;
335}
336
337static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
338{
54008399
YZ		 339 if (!info->dir_in)
340 return;
341 free_pages((unsigned long)info->dir_in, get_order(info->dir_buf_size));
2f2dc053
SW		 342}
343
344
345/*
346 * sessions
347 */
a687ecaf 348const char *ceph_session_state_name(int s)
2f2dc053
SW		 349{
350 switch (s) {
351 case CEPH_MDS_SESSION_NEW: return "new";
352 case CEPH_MDS_SESSION_OPENING: return "opening";
353 case CEPH_MDS_SESSION_OPEN: return "open";
354 case CEPH_MDS_SESSION_HUNG: return "hung";
355 case CEPH_MDS_SESSION_CLOSING: return "closing";
44ca18f2 356 case CEPH_MDS_SESSION_RESTARTING: return "restarting";
2f2dc053
SW		 357 case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
358 default: return "???";
359 }
360}
361
362static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
363{
364 if (atomic_inc_not_zero(&s->s_ref)) {
365 dout("mdsc get_session %p %d -> %d\n", s,
366 atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
367 return s;
368 } else {
369 dout("mdsc get_session %p 0 -- FAIL", s);
370 return NULL;
371 }
372}
373
374void ceph_put_mds_session(struct ceph_mds_session *s)
375{
376 dout("mdsc put_session %p %d -> %d\n", s,
377 atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
4e7a5dcd 378 if (atomic_dec_and_test(&s->s_ref)) {
6c4a1915 379 if (s->s_auth.authorizer)
27859f97
SW		 380 ceph_auth_destroy_authorizer(
381 s->s_mdsc->fsc->client->monc.auth,
382 s->s_auth.authorizer);
2f2dc053 383 kfree(s);
4e7a5dcd 384 }
2f2dc053
SW		 385}
386
387/*
388 * called under mdsc->mutex
389 */
390struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
391 int mds)
392{
393 struct ceph_mds_session *session;
394
395 if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
396 return NULL;
397 session = mdsc->sessions[mds];
398 dout("lookup_mds_session %p %d\n", session,
399 atomic_read(&session->s_ref));
400 get_session(session);
401 return session;
402}
403
404static bool __have_session(struct ceph_mds_client *mdsc, int mds)
405{
406 if (mds >= mdsc->max_sessions)
407 return false;
408 return mdsc->sessions[mds];
409}
410
2600d2dd
SW		 411static int __verify_registered_session(struct ceph_mds_client *mdsc,
412 struct ceph_mds_session *s)
413{
414 if (s->s_mds >= mdsc->max_sessions ||
415 mdsc->sessions[s->s_mds] != s)
416 return -ENOENT;
417 return 0;
418}
419
2f2dc053
SW		 420/*
421 * create+register a new session for given mds.
422 * called under mdsc->mutex.
423 */
424static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
425 int mds)
426{
427 struct ceph_mds_session *s;
428
c338c07c
NY		 429 if (mds >= mdsc->mdsmap->m_max_mds)
430 return ERR_PTR(-EINVAL);
431
2f2dc053 432 s = kzalloc(sizeof(*s), GFP_NOFS);
4736b009
DC		 433 if (!s)
434 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 435 s->s_mdsc = mdsc;
436 s->s_mds = mds;
437 s->s_state = CEPH_MDS_SESSION_NEW;
438 s->s_ttl = 0;
439 s->s_seq = 0;
440 mutex_init(&s->s_mutex);
441
b7a9e5dd 442 ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
2f2dc053 443
d8fb02ab 444 spin_lock_init(&s->s_gen_ttl_lock);
2f2dc053 445 s->s_cap_gen = 0;
1ce208a6 446 s->s_cap_ttl = jiffies - 1;
d8fb02ab
AE		 447
448 spin_lock_init(&s->s_cap_lock);
2f2dc053
SW		 449 s->s_renew_requested = 0;
450 s->s_renew_seq = 0;
451 INIT_LIST_HEAD(&s->s_caps);
452 s->s_nr_caps = 0;
5dacf091 453 s->s_trim_caps = 0;
2f2dc053
SW		 454 atomic_set(&s->s_ref, 1);
455 INIT_LIST_HEAD(&s->s_waiting);
456 INIT_LIST_HEAD(&s->s_unsafe);
457 s->s_num_cap_releases = 0;
99a9c273 458 s->s_cap_reconnect = 0;
7c1332b8 459 s->s_cap_iterator = NULL;
2f2dc053
SW		 460 INIT_LIST_HEAD(&s->s_cap_releases);
461 INIT_LIST_HEAD(&s->s_cap_releases_done);
462 INIT_LIST_HEAD(&s->s_cap_flushing);
463 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
464
465 dout("register_session mds%d\n", mds);
466 if (mds >= mdsc->max_sessions) {
467 int newmax = 1 << get_count_order(mds+1);
468 struct ceph_mds_session **sa;
469
470 dout("register_session realloc to %d\n", newmax);
471 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
472 if (sa == NULL)
42ce56e5 473 goto fail_realloc;
2f2dc053
SW		 474 if (mdsc->sessions) {
475 memcpy(sa, mdsc->sessions,
476 mdsc->max_sessions * sizeof(void *));
477 kfree(mdsc->sessions);
478 }
479 mdsc->sessions = sa;
480 mdsc->max_sessions = newmax;
481 }
482 mdsc->sessions[mds] = s;
483 atomic_inc(&s->s_ref); /* one ref to sessions[], one to caller */
42ce56e5 484
b7a9e5dd
SW		 485 ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
486 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
42ce56e5 487
2f2dc053 488 return s;
42ce56e5
SW		 489
490fail_realloc:
491 kfree(s);
492 return ERR_PTR(-ENOMEM);
2f2dc053
SW		 493}
494
495/*
496 * called under mdsc->mutex
497 */
2600d2dd 498static void __unregister_session(struct ceph_mds_client *mdsc,
42ce56e5 499 struct ceph_mds_session *s)
2f2dc053 500{
2600d2dd
SW		 501 dout("__unregister_session mds%d %p\n", s->s_mds, s);
502 BUG_ON(mdsc->sessions[s->s_mds] != s);
42ce56e5
SW		 503 mdsc->sessions[s->s_mds] = NULL;
504 ceph_con_close(&s->s_con);
505 ceph_put_mds_session(s);
2f2dc053
SW		 506}
507
508/*
509 * drop session refs in request.
510 *
511 * should be last request ref, or hold mdsc->mutex
512 */
513static void put_request_session(struct ceph_mds_request *req)
514{
515 if (req->r_session) {
516 ceph_put_mds_session(req->r_session);
517 req->r_session = NULL;
518 }
519}
520
153c8e6b 521void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 522{
153c8e6b
SW		 523 struct ceph_mds_request *req = container_of(kref,
524 struct ceph_mds_request,
525 r_kref);
54008399 526 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 527 if (req->r_request)
528 ceph_msg_put(req->r_request);
54008399 529 if (req->r_reply)
153c8e6b 530 ceph_msg_put(req->r_reply);
153c8e6b 531 if (req->r_inode) {
41b02e1f 532 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 533 iput(req->r_inode);
534 }
535 if (req->r_locked_dir)
41b02e1f 536 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
e96a650a 537 iput(req->r_target_inode);
153c8e6b
SW		 538 if (req->r_dentry)
539 dput(req->r_dentry);
844d87c3
SW		 540 if (req->r_old_dentry)
541 dput(req->r_old_dentry);
542 if (req->r_old_dentry_dir) {
41b02e1f
SW		 543 /*
544 * track (and drop pins for) r_old_dentry_dir
545 * separately, since r_old_dentry's d_parent may have
546 * changed between the dir mutex being dropped and
547 * this request being freed.
548 */
549 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
550 CEPH_CAP_PIN);
41b02e1f 551 iput(req->r_old_dentry_dir);
2f2dc053 552 }
153c8e6b
SW		 553 kfree(req->r_path1);
554 kfree(req->r_path2);
25e6bae3
YZ		 555 if (req->r_pagelist)
556 ceph_pagelist_release(req->r_pagelist);
153c8e6b 557 put_request_session(req);
37151668 558 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 559 kfree(req);
2f2dc053
SW		 560}
561
562/*
563 * lookup session, bump ref if found.
564 *
565 * called under mdsc->mutex.
566 */
567static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
568 u64 tid)
569{
570 struct ceph_mds_request *req;
44ca18f2
SW		 571 struct rb_node *n = mdsc->request_tree.rb_node;
572
573 while (n) {
574 req = rb_entry(n, struct ceph_mds_request, r_node);
575 if (tid < req->r_tid)
576 n = n->rb_left;
577 else if (tid > req->r_tid)
578 n = n->rb_right;
579 else {
580 ceph_mdsc_get_request(req);
581 return req;
582 }
583 }
584 return NULL;
585}
586
587static void __insert_request(struct ceph_mds_client *mdsc,
588 struct ceph_mds_request *new)
589{
590 struct rb_node **p = &mdsc->request_tree.rb_node;
591 struct rb_node *parent = NULL;
592 struct ceph_mds_request *req = NULL;
593
594 while (*p) {
595 parent = *p;
596 req = rb_entry(parent, struct ceph_mds_request, r_node);
597 if (new->r_tid < req->r_tid)
598 p = &(*p)->rb_left;
599 else if (new->r_tid > req->r_tid)
600 p = &(*p)->rb_right;
601 else
602 BUG();
603 }
604
605 rb_link_node(&new->r_node, parent, p);
606 rb_insert_color(&new->r_node, &mdsc->request_tree);
2f2dc053
SW		 607}
608
609/*
610 * Register an in-flight request, and assign a tid. Link to directory
611 * are modifying (if any).
612 *
613 * Called under mdsc->mutex.
614 */
615static void __register_request(struct ceph_mds_client *mdsc,
616 struct ceph_mds_request *req,
617 struct inode *dir)
618{
619 req->r_tid = ++mdsc->last_tid;
620 if (req->r_num_caps)
37151668
YS		 621 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
622 req->r_num_caps);
2f2dc053
SW		 623 dout("__register_request %p tid %lld\n", req, req->r_tid);
624 ceph_mdsc_get_request(req);
44ca18f2 625 __insert_request(mdsc, req);
2f2dc053 626
cb4276cc
SW		 627 req->r_uid = current_fsuid();
628 req->r_gid = current_fsgid();
629
2f2dc053
SW		 630 if (dir) {
631 struct ceph_inode_info *ci = ceph_inode(dir);
632
3b663780 633 ihold(dir);
2f2dc053
SW		 634 spin_lock(&ci->i_unsafe_lock);
635 req->r_unsafe_dir = dir;
636 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
637 spin_unlock(&ci->i_unsafe_lock);
638 }
639}
640
641static void __unregister_request(struct ceph_mds_client *mdsc,
642 struct ceph_mds_request *req)
643{
644 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
44ca18f2 645 rb_erase(&req->r_node, &mdsc->request_tree);
80fc7314 646 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 647
648 if (req->r_unsafe_dir) {
649 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
650
651 spin_lock(&ci->i_unsafe_lock);
652 list_del_init(&req->r_unsafe_dir_item);
653 spin_unlock(&ci->i_unsafe_lock);
3b663780
SW		 654
655 iput(req->r_unsafe_dir);
656 req->r_unsafe_dir = NULL;
2f2dc053 657 }
94aa8ae1 658
fc55d2c9
YZ		 659 complete_all(&req->r_safe_completion);
660
94aa8ae1 661 ceph_mdsc_put_request(req);
2f2dc053
SW		 662}
663
664/*
665 * Choose mds to send request to next. If there is a hint set in the
666 * request (e.g., due to a prior forward hint from the mds), use that.
667 * Otherwise, consult frag tree and/or caps to identify the
668 * appropriate mds. If all else fails, choose randomly.
669 *
670 * Called under mdsc->mutex.
671 */
7fd7d101 672static struct dentry *get_nonsnap_parent(struct dentry *dentry)
eb6bb1c5 673{
d79698da
SW		 674 /*
675 * we don't need to worry about protecting the d_parent access
676 * here because we never renaming inside the snapped namespace
677 * except to resplice to another snapdir, and either the old or new
678 * result is a valid result.
679 */
eb6bb1c5
SW		 680 while (!IS_ROOT(dentry) && ceph_snap(dentry->d_inode) != CEPH_NOSNAP)
681 dentry = dentry->d_parent;
682 return dentry;
683}
684
2f2dc053
SW		 685static int __choose_mds(struct ceph_mds_client *mdsc,
686 struct ceph_mds_request *req)
687{
688 struct inode *inode;
689 struct ceph_inode_info *ci;
690 struct ceph_cap *cap;
691 int mode = req->r_direct_mode;
692 int mds = -1;
693 u32 hash = req->r_direct_hash;
694 bool is_hash = req->r_direct_is_hash;
695
696 /*
697 * is there a specific mds we should try? ignore hint if we have
698 * no session and the mds is not up (active or recovering).
699 */
700 if (req->r_resend_mds >= 0 &&
701 (__have_session(mdsc, req->r_resend_mds) ||
702 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
703 dout("choose_mds using resend_mds mds%d\n",
704 req->r_resend_mds);
705 return req->r_resend_mds;
706 }
707
708 if (mode == USE_RANDOM_MDS)
709 goto random;
710
711 inode = NULL;
712 if (req->r_inode) {
713 inode = req->r_inode;
714 } else if (req->r_dentry) {
d79698da
SW		 715 /* ignore race with rename; old or new d_parent is okay */
716 struct dentry *parent = req->r_dentry->d_parent;
717 struct inode *dir = parent->d_inode;
eb6bb1c5 718
3d14c5d2 719 if (dir->i_sb != mdsc->fsc->sb) {
eb6bb1c5
SW		 720 /* not this fs! */
721 inode = req->r_dentry->d_inode;
722 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
723 /* direct snapped/virtual snapdir requests
724 * based on parent dir inode */
d79698da 725 struct dentry *dn = get_nonsnap_parent(parent);
eb6bb1c5
SW		 726 inode = dn->d_inode;
727 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 728 } else {
eb6bb1c5 729 /* dentry target */
2f2dc053 730 inode = req->r_dentry->d_inode;
ca18bede
YZ		 731 if (!inode || mode == USE_AUTH_MDS) {
732 /* dir + name */
733 inode = dir;
734 hash = ceph_dentry_hash(dir, req->r_dentry);
735 is_hash = true;
736 }
2f2dc053
SW		 737 }
738 }
eb6bb1c5 739
2f2dc053
SW		 740 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
741 (int)hash, mode);
742 if (!inode)
743 goto random;
744 ci = ceph_inode(inode);
745
746 if (is_hash && S_ISDIR(inode->i_mode)) {
747 struct ceph_inode_frag frag;
748 int found;
749
750 ceph_choose_frag(ci, hash, &frag, &found);
751 if (found) {
752 if (mode == USE_ANY_MDS && frag.ndist > 0) {
753 u8 r;
754
755 /* choose a random replica */
756 get_random_bytes(&r, 1);
757 r %= frag.ndist;
758 mds = frag.dist[r];
759 dout("choose_mds %p %llx.%llx "
760 "frag %u mds%d (%d/%d)\n",
761 inode, ceph_vinop(inode),
d66bbd44 762 frag.frag, mds,
2f2dc053 763 (int)r, frag.ndist);
d66bbd44
SW		 764 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
765 CEPH_MDS_STATE_ACTIVE)
766 return mds;
2f2dc053
SW		 767 }
768
769 /* since this file/dir wasn't known to be
770 * replicated, then we want to look for the
771 * authoritative mds. */
772 mode = USE_AUTH_MDS;
773 if (frag.mds >= 0) {
774 /* choose auth mds */
775 mds = frag.mds;
776 dout("choose_mds %p %llx.%llx "
777 "frag %u mds%d (auth)\n",
778 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 779 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
780 CEPH_MDS_STATE_ACTIVE)
781 return mds;
2f2dc053
SW		 782 }
783 }
784 }
785
be655596 786 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 787 cap = NULL;
788 if (mode == USE_AUTH_MDS)
789 cap = ci->i_auth_cap;
790 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
791 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
792 if (!cap) {
be655596 793 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 794 goto random;
795 }
796 mds = cap->session->s_mds;
797 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
798 inode, ceph_vinop(inode), mds,
799 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 800 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 801 return mds;
802
803random:
804 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
805 dout("choose_mds chose random mds%d\n", mds);
806 return mds;
807}
808
809
810/*
811 * session messages
812 */
813static struct ceph_msg *create_session_msg(u32 op, u64 seq)
814{
815 struct ceph_msg *msg;
816 struct ceph_mds_session_head *h;
817
b61c2763
SW		 818 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
819 false);
a79832f2 820 if (!msg) {
2f2dc053 821 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 822 return NULL;
2f2dc053
SW		 823 }
824 h = msg->front.iov_base;
825 h->op = cpu_to_le32(op);
826 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 827
828 return msg;
829}
830
831/*
832 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
833 * to include additional client metadata fields.
834 */
835static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
836{
837 struct ceph_msg *msg;
838 struct ceph_mds_session_head *h;
839 int i = -1;
840 int metadata_bytes = 0;
841 int metadata_key_count = 0;
842 struct ceph_options *opt = mdsc->fsc->client->options;
843 void *p;
844
845 const char* metadata[3][2] = {
846 {"hostname", utsname()->nodename},
847 {"entity_id", opt->name ? opt->name : ""},
848 {NULL, NULL}
849 };
850
851 /* Calculate serialized length of metadata */
852 metadata_bytes = 4; /* map length */
853 for (i = 0; metadata[i][0] != NULL; ++i) {
854 metadata_bytes += 8 + strlen(metadata[i][0]) +
855 strlen(metadata[i][1]);
856 metadata_key_count++;
857 }
858
859 /* Allocate the message */
860 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
861 GFP_NOFS, false);
862 if (!msg) {
863 pr_err("create_session_msg ENOMEM creating msg\n");
864 return NULL;
865 }
866 h = msg->front.iov_base;
867 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
868 h->seq = cpu_to_le64(seq);
869
870 /*
871 * Serialize client metadata into waiting buffer space, using
872 * the format that userspace expects for map<string, string>
7cfa0313
JS		 873 *
874 * ClientSession messages with metadata are v2
dbd0c8bf 875 */
7cfa0313
JS		 876 msg->hdr.version = cpu_to_le16(2);
877 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 878
879 /* The write pointer, following the session_head structure */
880 p = msg->front.iov_base + sizeof(*h);
881
882 /* Number of entries in the map */
883 ceph_encode_32(&p, metadata_key_count);
884
885 /* Two length-prefixed strings for each entry in the map */
886 for (i = 0; metadata[i][0] != NULL; ++i) {
887 size_t const key_len = strlen(metadata[i][0]);
888 size_t const val_len = strlen(metadata[i][1]);
889
890 ceph_encode_32(&p, key_len);
891 memcpy(p, metadata[i][0], key_len);
892 p += key_len;
893 ceph_encode_32(&p, val_len);
894 memcpy(p, metadata[i][1], val_len);
895 p += val_len;
896 }
897
2f2dc053
SW		 898 return msg;
899}
900
901/*
902 * send session open request.
903 *
904 * called under mdsc->mutex
905 */
906static int __open_session(struct ceph_mds_client *mdsc,
907 struct ceph_mds_session *session)
908{
909 struct ceph_msg *msg;
910 int mstate;
911 int mds = session->s_mds;
2f2dc053
SW		 912
913 /* wait for mds to go active? */
914 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
915 dout("open_session to mds%d (%s)\n", mds,
916 ceph_mds_state_name(mstate));
917 session->s_state = CEPH_MDS_SESSION_OPENING;
918 session->s_renew_requested = jiffies;
919
920 /* send connect message */
dbd0c8bf 921 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 922 if (!msg)
923 return -ENOMEM;
2f2dc053 924 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 925 return 0;
926}
927
ed0552a1
SW		 928/*
929 * open sessions for any export targets for the given mds
930 *
931 * called under mdsc->mutex
932 */
5d72d13c
YZ		 933static struct ceph_mds_session *
934__open_export_target_session(struct ceph_mds_client *mdsc, int target)
935{
936 struct ceph_mds_session *session;
937
938 session = __ceph_lookup_mds_session(mdsc, target);
939 if (!session) {
940 session = register_session(mdsc, target);
941 if (IS_ERR(session))
942 return session;
943 }
944 if (session->s_state == CEPH_MDS_SESSION_NEW ||
945 session->s_state == CEPH_MDS_SESSION_CLOSING)
946 __open_session(mdsc, session);
947
948 return session;
949}
950
951struct ceph_mds_session *
952ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
953{
954 struct ceph_mds_session *session;
955
956 dout("open_export_target_session to mds%d\n", target);
957
958 mutex_lock(&mdsc->mutex);
959 session = __open_export_target_session(mdsc, target);
960 mutex_unlock(&mdsc->mutex);
961
962 return session;
963}
964
ed0552a1
SW		 965static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
966 struct ceph_mds_session *session)
967{
968 struct ceph_mds_info *mi;
969 struct ceph_mds_session *ts;
970 int i, mds = session->s_mds;
ed0552a1
SW		 971
972 if (mds >= mdsc->mdsmap->m_max_mds)
973 return;
5d72d13c 974
ed0552a1
SW		 975 mi = &mdsc->mdsmap->m_info[mds];
976 dout("open_export_target_sessions for mds%d (%d targets)\n",
977 session->s_mds, mi->num_export_targets);
978
979 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 980 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
981 if (!IS_ERR(ts))
982 ceph_put_mds_session(ts);
ed0552a1
SW		 983 }
984}
985
154f42c2
SW		 986void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
987 struct ceph_mds_session *session)
988{
989 mutex_lock(&mdsc->mutex);
990 __open_export_target_sessions(mdsc, session);
991 mutex_unlock(&mdsc->mutex);
992}
993
2f2dc053
SW		 994/*
995 * session caps
996 */
997
998/*
999 * Free preallocated cap messages assigned to this session
1000 */
1001static void cleanup_cap_releases(struct ceph_mds_session *session)
1002{
1003 struct ceph_msg *msg;
1004
1005 spin_lock(&session->s_cap_lock);
1006 while (!list_empty(&session->s_cap_releases)) {
1007 msg = list_first_entry(&session->s_cap_releases,
1008 struct ceph_msg, list_head);
1009 list_del_init(&msg->list_head);
1010 ceph_msg_put(msg);
1011 }
1012 while (!list_empty(&session->s_cap_releases_done)) {
1013 msg = list_first_entry(&session->s_cap_releases_done,
1014 struct ceph_msg, list_head);
1015 list_del_init(&msg->list_head);
1016 ceph_msg_put(msg);
1017 }
1018 spin_unlock(&session->s_cap_lock);
1019}
1020
1021/*
f818a736
SW		 1022 * Helper to safely iterate over all caps associated with a session, with
1023 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1024 *
f818a736 1025 * Caller must hold session s_mutex.
2f2dc053
SW		 1026 */
1027static int iterate_session_caps(struct ceph_mds_session *session,
1028 int (*cb)(struct inode *, struct ceph_cap *,
1029 void *), void *arg)
1030{
7c1332b8
SW		 1031 struct list_head *p;
1032 struct ceph_cap *cap;
1033 struct inode *inode, *last_inode = NULL;
1034 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1035 int ret;
1036
1037 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1038 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1039 p = session->s_caps.next;
1040 while (p != &session->s_caps) {
1041 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1042 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1043 if (!inode) {
1044 p = p->next;
2f2dc053 1045 continue;
7c1332b8
SW		 1046 }
1047 session->s_cap_iterator = cap;
2f2dc053 1048 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1049
1050 if (last_inode) {
1051 iput(last_inode);
1052 last_inode = NULL;
1053 }
1054 if (old_cap) {
37151668 1055 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1056 old_cap = NULL;
1057 }
1058
2f2dc053 1059 ret = cb(inode, cap, arg);
7c1332b8
SW		 1060 last_inode = inode;
1061
2f2dc053 1062 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1063 p = p->next;
1064 if (cap->ci == NULL) {
1065 dout("iterate_session_caps finishing cap %p removal\n",
1066 cap);
1067 BUG_ON(cap->session != session);
1068 list_del_init(&cap->session_caps);
1069 session->s_nr_caps--;
1070 cap->session = NULL;
1071 old_cap = cap; /* put_cap it w/o locks held */
1072 }
5dacf091
SW		 1073 if (ret < 0)
1074 goto out;
2f2dc053 1075 }
5dacf091
SW		 1076 ret = 0;
1077out:
7c1332b8 1078 session->s_cap_iterator = NULL;
2f2dc053 1079 spin_unlock(&session->s_cap_lock);
7c1332b8 1080
e96a650a 1081 iput(last_inode);
7c1332b8 1082 if (old_cap)
37151668 1083 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1084
5dacf091 1085 return ret;
2f2dc053
SW		 1086}
1087
1088static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1089 void *arg)
2f2dc053
SW		 1090{
1091 struct ceph_inode_info *ci = ceph_inode(inode);
6c99f254
SW		 1092 int drop = 0;
1093
2f2dc053
SW		 1094 dout("removing cap %p, ci is %p, inode is %p\n",
1095 cap, ci, &ci->vfs_inode);
be655596 1096 spin_lock(&ci->i_ceph_lock);
a096b09a 1097 __ceph_remove_cap(cap, false);
6c99f254
SW		 1098 if (!__ceph_is_any_real_caps(ci)) {
1099 struct ceph_mds_client *mdsc =
3d14c5d2 1100 ceph_sb_to_client(inode->i_sb)->mdsc;
6c99f254
SW		 1101
1102 spin_lock(&mdsc->cap_dirty_lock);
1103 if (!list_empty(&ci->i_dirty_item)) {
1104 pr_info(" dropping dirty %s state for %p %lld\n",
1105 ceph_cap_string(ci->i_dirty_caps),
1106 inode, ceph_ino(inode));
1107 ci->i_dirty_caps = 0;
1108 list_del_init(&ci->i_dirty_item);
1109 drop = 1;
1110 }
1111 if (!list_empty(&ci->i_flushing_item)) {
1112 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1113 ceph_cap_string(ci->i_flushing_caps),
1114 inode, ceph_ino(inode));
1115 ci->i_flushing_caps = 0;
1116 list_del_init(&ci->i_flushing_item);
1117 mdsc->num_cap_flushing--;
1118 drop = 1;
1119 }
1120 if (drop && ci->i_wrbuffer_ref) {
1121 pr_info(" dropping dirty data for %p %lld\n",
1122 inode, ceph_ino(inode));
1123 ci->i_wrbuffer_ref = 0;
1124 ci->i_wrbuffer_ref_head = 0;
1125 drop++;
1126 }
1127 spin_unlock(&mdsc->cap_dirty_lock);
1128 }
be655596 1129 spin_unlock(&ci->i_ceph_lock);
6c99f254
SW		 1130 while (drop--)
1131 iput(inode);
2f2dc053
SW		 1132 return 0;
1133}
1134
1135/*
1136 * caller must hold session s_mutex
1137 */
1138static void remove_session_caps(struct ceph_mds_session *session)
1139{
1140 dout("remove_session_caps on %p\n", session);
1141 iterate_session_caps(session, remove_session_caps_cb, NULL);
6f60f889
YZ		 1142
1143 spin_lock(&session->s_cap_lock);
1144 if (session->s_nr_caps > 0) {
1145 struct super_block *sb = session->s_mdsc->fsc->sb;
1146 struct inode *inode;
1147 struct ceph_cap *cap, *prev = NULL;
1148 struct ceph_vino vino;
1149 /*
1150 * iterate_session_caps() skips inodes that are being
1151 * deleted, we need to wait until deletions are complete.
1152 * __wait_on_freeing_inode() is designed for the job,
1153 * but it is not exported, so use lookup inode function
1154 * to access it.
1155 */
1156 while (!list_empty(&session->s_caps)) {
1157 cap = list_entry(session->s_caps.next,
1158 struct ceph_cap, session_caps);
1159 if (cap == prev)
1160 break;
1161 prev = cap;
1162 vino = cap->ci->i_vino;
1163 spin_unlock(&session->s_cap_lock);
1164
ed284c49 1165 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1166 iput(inode);
1167
1168 spin_lock(&session->s_cap_lock);
1169 }
1170 }
1171 spin_unlock(&session->s_cap_lock);
1172
2f2dc053 1173 BUG_ON(session->s_nr_caps > 0);
6c99f254 1174 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1175 cleanup_cap_releases(session);
1176}
1177
1178/*
1179 * wake up any threads waiting on this session's caps. if the cap is
1180 * old (didn't get renewed on the client reconnect), remove it now.
1181 *
1182 * caller must hold s_mutex.
1183 */
1184static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1185 void *arg)
1186{
0dc2570f
SW		 1187 struct ceph_inode_info *ci = ceph_inode(inode);
1188
03066f23 1189 wake_up_all(&ci->i_cap_wq);
0dc2570f 1190 if (arg) {
be655596 1191 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1192 ci->i_wanted_max_size = 0;
1193 ci->i_requested_max_size = 0;
be655596 1194 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1195 }
2f2dc053
SW		 1196 return 0;
1197}
1198
0dc2570f
SW		 1199static void wake_up_session_caps(struct ceph_mds_session *session,
1200 int reconnect)
2f2dc053
SW		 1201{
1202 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1203 iterate_session_caps(session, wake_up_session_cb,
1204 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1205}
1206
1207/*
1208 * Send periodic message to MDS renewing all currently held caps. The
1209 * ack will reset the expiration for all caps from this session.
1210 *
1211 * caller holds s_mutex
1212 */
1213static int send_renew_caps(struct ceph_mds_client *mdsc,
1214 struct ceph_mds_session *session)
1215{
1216 struct ceph_msg *msg;
1217 int state;
1218
1219 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1220 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1221 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1222 session->s_renew_requested = jiffies;
2f2dc053
SW		 1223
1224 /* do not try to renew caps until a recovering mds has reconnected
1225 * with its clients. */
1226 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1227 if (state < CEPH_MDS_STATE_RECONNECT) {
1228 dout("send_renew_caps ignoring mds%d (%s)\n",
1229 session->s_mds, ceph_mds_state_name(state));
1230 return 0;
1231 }
1232
1233 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1234 ceph_mds_state_name(state));
2f2dc053
SW		 1235 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1236 ++session->s_renew_seq);
a79832f2
SW		 1237 if (!msg)
1238 return -ENOMEM;
2f2dc053
SW		 1239 ceph_con_send(&session->s_con, msg);
1240 return 0;
1241}
1242
186e4f7a
YZ		 1243static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1244 struct ceph_mds_session *session, u64 seq)
1245{
1246 struct ceph_msg *msg;
1247
1248 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1249 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1250 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1251 if (!msg)
1252 return -ENOMEM;
1253 ceph_con_send(&session->s_con, msg);
1254 return 0;
1255}
1256
1257
2f2dc053
SW		 1258/*
1259 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1260 *
1261 * Called under session->s_mutex
2f2dc053
SW		 1262 */
1263static void renewed_caps(struct ceph_mds_client *mdsc,
1264 struct ceph_mds_session *session, int is_renew)
1265{
1266 int was_stale;
1267 int wake = 0;
1268
1269 spin_lock(&session->s_cap_lock);
1ce208a6 1270 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1271
1272 session->s_cap_ttl = session->s_renew_requested +
1273 mdsc->mdsmap->m_session_timeout*HZ;
1274
1275 if (was_stale) {
1276 if (time_before(jiffies, session->s_cap_ttl)) {
1277 pr_info("mds%d caps renewed\n", session->s_mds);
1278 wake = 1;
1279 } else {
1280 pr_info("mds%d caps still stale\n", session->s_mds);
1281 }
1282 }
1283 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1284 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1285 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1286 spin_unlock(&session->s_cap_lock);
1287
1288 if (wake)
0dc2570f 1289 wake_up_session_caps(session, 0);
2f2dc053
SW		 1290}
1291
1292/*
1293 * send a session close request
1294 */
1295static int request_close_session(struct ceph_mds_client *mdsc,
1296 struct ceph_mds_session *session)
1297{
1298 struct ceph_msg *msg;
2f2dc053
SW		 1299
1300 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1301 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1302 session->s_seq);
1303 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1304 if (!msg)
1305 return -ENOMEM;
1306 ceph_con_send(&session->s_con, msg);
1307 return 0;
2f2dc053
SW		 1308}
1309
1310/*
1311 * Called with s_mutex held.
1312 */
1313static int __close_session(struct ceph_mds_client *mdsc,
1314 struct ceph_mds_session *session)
1315{
1316 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1317 return 0;
1318 session->s_state = CEPH_MDS_SESSION_CLOSING;
1319 return request_close_session(mdsc, session);
1320}
1321
1322/*
1323 * Trim old(er) caps.
1324 *
1325 * Because we can't cache an inode without one or more caps, we do
1326 * this indirectly: if a cap is unused, we prune its aliases, at which
1327 * point the inode will hopefully get dropped to.
1328 *
1329 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1330 * memory pressure from the MDS, though, so it needn't be perfect.
1331 */
1332static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1333{
1334 struct ceph_mds_session *session = arg;
1335 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1336 int used, wanted, oissued, mine;
2f2dc053
SW		 1337
1338 if (session->s_trim_caps <= 0)
1339 return -1;
1340
be655596 1341 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1342 mine = cap->issued | cap->implemented;
1343 used = __ceph_caps_used(ci);
979abfdd 1344 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1345 oissued = __ceph_caps_issued_other(ci, cap);
1346
979abfdd 1347 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1348 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1349 ceph_cap_string(used), ceph_cap_string(wanted));
1350 if (cap == ci->i_auth_cap) {
1351 if (ci->i_dirty_caps | ci->i_flushing_caps)
1352 goto out;
1353 if ((used | wanted) & CEPH_CAP_ANY_WR)
1354 goto out;
1355 }
1356 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1357 goto out; /* we need these caps */
1358
1359 session->s_trim_caps--;
1360 if (oissued) {
1361 /* we aren't the only cap.. just remove us */
a096b09a 1362 __ceph_remove_cap(cap, true);
2f2dc053
SW		 1363 } else {
1364 /* try to drop referring dentries */
be655596 1365 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1366 d_prune_aliases(inode);
1367 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1368 inode, cap, atomic_read(&inode->i_count));
1369 return 0;
1370 }
1371
1372out:
be655596 1373 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1374 return 0;
1375}
1376
1377/*
1378 * Trim session cap count down to some max number.
1379 */
1380static int trim_caps(struct ceph_mds_client *mdsc,
1381 struct ceph_mds_session *session,
1382 int max_caps)
1383{
1384 int trim_caps = session->s_nr_caps - max_caps;
1385
1386 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1387 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1388 if (trim_caps > 0) {
1389 session->s_trim_caps = trim_caps;
1390 iterate_session_caps(session, trim_caps_cb, session);
1391 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1392 session->s_mds, session->s_nr_caps, max_caps,
1393 trim_caps - session->s_trim_caps);
5dacf091 1394 session->s_trim_caps = 0;
2f2dc053 1395 }
a56371d9
YZ		 1396
1397 ceph_add_cap_releases(mdsc, session);
1398 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1399 return 0;
1400}
1401
1402/*
1403 * Allocate cap_release messages. If there is a partially full message
1404 * in the queue, try to allocate enough to cover it's remainder, so that
1405 * we can send it immediately.
1406 *
1407 * Called under s_mutex.
1408 */
2b2300d6 1409int ceph_add_cap_releases(struct ceph_mds_client *mdsc,
ee6b272b 1410 struct ceph_mds_session *session)
2f2dc053 1411{
38e8883e 1412 struct ceph_msg *msg, *partial = NULL;
2f2dc053
SW		 1413 struct ceph_mds_cap_release *head;
1414 int err = -ENOMEM;
3d14c5d2 1415 int extra = mdsc->fsc->mount_options->cap_release_safety;
38e8883e 1416 int num;
2f2dc053 1417
38e8883e
SW		 1418 dout("add_cap_releases %p mds%d extra %d\n", session, session->s_mds,
1419 extra);
2f2dc053
SW		 1420
1421 spin_lock(&session->s_cap_lock);
1422
1423 if (!list_empty(&session->s_cap_releases)) {
1424 msg = list_first_entry(&session->s_cap_releases,
1425 struct ceph_msg,
1426 list_head);
1427 head = msg->front.iov_base;
38e8883e
SW		 1428 num = le32_to_cpu(head->num);
1429 if (num) {
1430 dout(" partial %p with (%d/%d)\n", msg, num,
1431 (int)CEPH_CAPS_PER_RELEASE);
1432 extra += CEPH_CAPS_PER_RELEASE - num;
1433 partial = msg;
1434 }
2f2dc053 1435 }
2f2dc053
SW		 1436 while (session->s_num_cap_releases < session->s_nr_caps + extra) {
1437 spin_unlock(&session->s_cap_lock);
34d23762 1438 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE, PAGE_CACHE_SIZE,
b61c2763 1439 GFP_NOFS, false);
2f2dc053
SW		 1440 if (!msg)
1441 goto out_unlocked;
1442 dout("add_cap_releases %p msg %p now %d\n", session, msg,
1443 (int)msg->front.iov_len);
1444 head = msg->front.iov_base;
1445 head->num = cpu_to_le32(0);
1446 msg->front.iov_len = sizeof(*head);
1447 spin_lock(&session->s_cap_lock);
1448 list_add(&msg->list_head, &session->s_cap_releases);
1449 session->s_num_cap_releases += CEPH_CAPS_PER_RELEASE;
1450 }
1451
38e8883e
SW		 1452 if (partial) {
1453 head = partial->front.iov_base;
1454 num = le32_to_cpu(head->num);
1455 dout(" queueing partial %p with %d/%d\n", partial, num,
1456 (int)CEPH_CAPS_PER_RELEASE);
1457 list_move_tail(&partial->list_head,
1458 &session->s_cap_releases_done);
1459 session->s_num_cap_releases -= CEPH_CAPS_PER_RELEASE - num;
2f2dc053
SW		 1460 }
1461 err = 0;
1462 spin_unlock(&session->s_cap_lock);
1463out_unlocked:
1464 return err;
1465}
1466
d3383a8e
YZ		 1467static int check_cap_flush(struct inode *inode, u64 want_flush_seq)
1468{
1469 struct ceph_inode_info *ci = ceph_inode(inode);
1470 int ret;
1471 spin_lock(&ci->i_ceph_lock);
1472 if (ci->i_flushing_caps)
1473 ret = ci->i_cap_flush_seq >= want_flush_seq;
1474 else
1475 ret = 1;
1476 spin_unlock(&ci->i_ceph_lock);
1477 return ret;
1478}
1479
2f2dc053
SW		 1480/*
1481 * flush all dirty inode data to disk.
1482 *
1483 * returns true if we've flushed through want_flush_seq
1484 */
d3383a8e 1485static void wait_caps_flush(struct ceph_mds_client *mdsc, u64 want_flush_seq)
2f2dc053 1486{
d3383a8e 1487 int mds;
2f2dc053
SW		 1488
1489 dout("check_cap_flush want %lld\n", want_flush_seq);
1490 mutex_lock(&mdsc->mutex);
d3383a8e 1491 for (mds = 0; mds < mdsc->max_sessions; mds++) {
2f2dc053 1492 struct ceph_mds_session *session = mdsc->sessions[mds];
d3383a8e 1493 struct inode *inode = NULL;
2f2dc053
SW		 1494
1495 if (!session)
1496 continue;
1497 get_session(session);
1498 mutex_unlock(&mdsc->mutex);
1499
1500 mutex_lock(&session->s_mutex);
1501 if (!list_empty(&session->s_cap_flushing)) {
1502 struct ceph_inode_info *ci =
1503 list_entry(session->s_cap_flushing.next,
1504 struct ceph_inode_info,
1505 i_flushing_item);
2f2dc053 1506
d3383a8e 1507 if (!check_cap_flush(&ci->vfs_inode, want_flush_seq)) {
2f2dc053 1508 dout("check_cap_flush still flushing %p "
d3383a8e
YZ		 1509 "seq %lld <= %lld to mds%d\n",
1510 &ci->vfs_inode, ci->i_cap_flush_seq,
1511 want_flush_seq, session->s_mds);
1512 inode = igrab(&ci->vfs_inode);
2f2dc053 1513 }
2f2dc053
SW		 1514 }
1515 mutex_unlock(&session->s_mutex);
1516 ceph_put_mds_session(session);
1517
d3383a8e
YZ		 1518 if (inode) {
1519 wait_event(mdsc->cap_flushing_wq,
1520 check_cap_flush(inode, want_flush_seq));
1521 iput(inode);
1522 }
1523
2f2dc053
SW		 1524 mutex_lock(&mdsc->mutex);
1525 }
1526
1527 mutex_unlock(&mdsc->mutex);
1528 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
2f2dc053
SW		 1529}
1530
1531/*
1532 * called under s_mutex
1533 */
3d7ded4d
SW		 1534void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1535 struct ceph_mds_session *session)
2f2dc053
SW		 1536{
1537 struct ceph_msg *msg;
1538
1539 dout("send_cap_releases mds%d\n", session->s_mds);
0f8605f2
SW		 1540 spin_lock(&session->s_cap_lock);
1541 while (!list_empty(&session->s_cap_releases_done)) {
2f2dc053
SW		 1542 msg = list_first_entry(&session->s_cap_releases_done,
1543 struct ceph_msg, list_head);
1544 list_del_init(&msg->list_head);
1545 spin_unlock(&session->s_cap_lock);
1546 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1547 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1548 ceph_con_send(&session->s_con, msg);
0f8605f2 1549 spin_lock(&session->s_cap_lock);
2f2dc053
SW		 1550 }
1551 spin_unlock(&session->s_cap_lock);
1552}
1553
e01a5946
SW		 1554static void discard_cap_releases(struct ceph_mds_client *mdsc,
1555 struct ceph_mds_session *session)
1556{
1557 struct ceph_msg *msg;
1558 struct ceph_mds_cap_release *head;
1559 unsigned num;
1560
1561 dout("discard_cap_releases mds%d\n", session->s_mds);
e01a5946 1562
00bd8edb
YZ		 1563 if (!list_empty(&session->s_cap_releases)) {
1564 /* zero out the in-progress message */
1565 msg = list_first_entry(&session->s_cap_releases,
1566 struct ceph_msg, list_head);
1567 head = msg->front.iov_base;
1568 num = le32_to_cpu(head->num);
1569 dout("discard_cap_releases mds%d %p %u\n",
1570 session->s_mds, msg, num);
1571 head->num = cpu_to_le32(0);
1572 msg->front.iov_len = sizeof(*head);
1573 session->s_num_cap_releases += num;
1574 }
e01a5946
SW		 1575
1576 /* requeue completed messages */
1577 while (!list_empty(&session->s_cap_releases_done)) {
1578 msg = list_first_entry(&session->s_cap_releases_done,
1579 struct ceph_msg, list_head);
1580 list_del_init(&msg->list_head);
1581
1582 head = msg->front.iov_base;
1583 num = le32_to_cpu(head->num);
1584 dout("discard_cap_releases mds%d %p %u\n", session->s_mds, msg,
1585 num);
1586 session->s_num_cap_releases += num;
1587 head->num = cpu_to_le32(0);
1588 msg->front.iov_len = sizeof(*head);
1589 list_add(&msg->list_head, &session->s_cap_releases);
1590 }
e01a5946
SW		 1591}
1592
2f2dc053
SW		 1593/*
1594 * requests
1595 */
1596
54008399
YZ		 1597int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1598 struct inode *dir)
1599{
1600 struct ceph_inode_info *ci = ceph_inode(dir);
1601 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1602 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1603 size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1604 sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1605 int order, num_entries;
1606
1607 spin_lock(&ci->i_ceph_lock);
1608 num_entries = ci->i_files + ci->i_subdirs;
1609 spin_unlock(&ci->i_ceph_lock);
1610 num_entries = max(num_entries, 1);
1611 num_entries = min(num_entries, opt->max_readdir);
1612
1613 order = get_order(size * num_entries);
1614 while (order >= 0) {
1615 rinfo->dir_in = (void*)__get_free_pages(GFP_NOFS | __GFP_NOWARN,
1616 order);
1617 if (rinfo->dir_in)
1618 break;
1619 order--;
1620 }
1621 if (!rinfo->dir_in)
1622 return -ENOMEM;
1623
1624 num_entries = (PAGE_SIZE << order) / size;
1625 num_entries = min(num_entries, opt->max_readdir);
1626
1627 rinfo->dir_buf_size = PAGE_SIZE << order;
1628 req->r_num_caps = num_entries + 1;
1629 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1630 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1631 return 0;
1632}
1633
2f2dc053
SW		 1634/*
1635 * Create an mds request.
1636 */
1637struct ceph_mds_request *
1638ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1639{
1640 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1641
1642 if (!req)
1643 return ERR_PTR(-ENOMEM);
1644
b4556396 1645 mutex_init(&req->r_fill_mutex);
37151668 1646 req->r_mdsc = mdsc;
2f2dc053
SW		 1647 req->r_started = jiffies;
1648 req->r_resend_mds = -1;
1649 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1650 req->r_fmode = -1;
153c8e6b 1651 kref_init(&req->r_kref);
2f2dc053
SW		 1652 INIT_LIST_HEAD(&req->r_wait);
1653 init_completion(&req->r_completion);
1654 init_completion(&req->r_safe_completion);
1655 INIT_LIST_HEAD(&req->r_unsafe_item);
1656
b8e69066
SW		 1657 req->r_stamp = CURRENT_TIME;
1658
2f2dc053
SW		 1659 req->r_op = op;
1660 req->r_direct_mode = mode;
1661 return req;
1662}
1663
1664/*
44ca18f2 1665 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1666 *
1667 * called under mdsc->mutex.
1668 */
44ca18f2
SW		 1669static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1670{
1671 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1672 return NULL;
1673 return rb_entry(rb_first(&mdsc->request_tree),
1674 struct ceph_mds_request, r_node);
1675}
1676
2f2dc053
SW		 1677static u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1678{
44ca18f2
SW		 1679 struct ceph_mds_request *req = __get_oldest_req(mdsc);
1680
1681 if (req)
1682 return req->r_tid;
1683 return 0;
2f2dc053
SW		 1684}
1685
1686/*
1687 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1688 * on build_path_from_dentry in fs/cifs/dir.c.
1689 *
1690 * If @stop_on_nosnap, generate path relative to the first non-snapped
1691 * inode.
1692 *
1693 * Encode hidden .snap dirs as a double /, i.e.
1694 * foo/.snap/bar -> foo//bar
1695 */
1696char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1697 int stop_on_nosnap)
1698{
1699 struct dentry *temp;
1700 char *path;
1701 int len, pos;
1b71fe2e 1702 unsigned seq;
2f2dc053
SW		 1703
1704 if (dentry == NULL)
1705 return ERR_PTR(-EINVAL);
1706
1707retry:
1708 len = 0;
1b71fe2e
AV		 1709 seq = read_seqbegin(&rename_lock);
1710 rcu_read_lock();
2f2dc053
SW		 1711 for (temp = dentry; !IS_ROOT(temp);) {
1712 struct inode *inode = temp->d_inode;
1713 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1714 len++; /* slash only */
1715 else if (stop_on_nosnap && inode &&
1716 ceph_snap(inode) == CEPH_NOSNAP)
1717 break;
1718 else
1719 len += 1 + temp->d_name.len;
1720 temp = temp->d_parent;
2f2dc053 1721 }
1b71fe2e 1722 rcu_read_unlock();
2f2dc053
SW		 1723 if (len)
1724 len--; /* no leading '/' */
1725
1726 path = kmalloc(len+1, GFP_NOFS);
1727 if (path == NULL)
1728 return ERR_PTR(-ENOMEM);
1729 pos = len;
1730 path[pos] = 0; /* trailing null */
1b71fe2e 1731 rcu_read_lock();
2f2dc053 1732 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1733 struct inode *inode;
2f2dc053 1734
1b71fe2e
AV		 1735 spin_lock(&temp->d_lock);
1736 inode = temp->d_inode;
2f2dc053 1737 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1738 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1739 pos, temp);
1740 } else if (stop_on_nosnap && inode &&
1741 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1742 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1743 break;
1744 } else {
1745 pos -= temp->d_name.len;
1b71fe2e
AV		 1746 if (pos < 0) {
1747 spin_unlock(&temp->d_lock);
2f2dc053 1748 break;
1b71fe2e 1749 }
2f2dc053
SW		 1750 strncpy(path + pos, temp->d_name.name,
1751 temp->d_name.len);
2f2dc053 1752 }
1b71fe2e 1753 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1754 if (pos)
1755 path[--pos] = '/';
1756 temp = temp->d_parent;
2f2dc053 1757 }
1b71fe2e
AV		 1758 rcu_read_unlock();
1759 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1760 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1761 "expected, namelen is %d, pos is %d\n", len, pos);
1762 /* presumably this is only possible if racing with a
1763 rename of one of the parent directories (we can not
1764 lock the dentries above us to prevent this, but
1765 retrying should be harmless) */
1766 kfree(path);
1767 goto retry;
1768 }
1769
1770 *base = ceph_ino(temp->d_inode);
1771 *plen = len;
104648ad 1772 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1773 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1774 return path;
1775}
1776
1777static int build_dentry_path(struct dentry *dentry,
1778 const char **ppath, int *ppathlen, u64 *pino,
1779 int *pfreepath)
1780{
1781 char *path;
1782
1783 if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) {
1784 *pino = ceph_ino(dentry->d_parent->d_inode);
1785 *ppath = dentry->d_name.name;
1786 *ppathlen = dentry->d_name.len;
1787 return 0;
1788 }
1789 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1790 if (IS_ERR(path))
1791 return PTR_ERR(path);
1792 *ppath = path;
1793 *pfreepath = 1;
1794 return 0;
1795}
1796
1797static int build_inode_path(struct inode *inode,
1798 const char **ppath, int *ppathlen, u64 *pino,
1799 int *pfreepath)
1800{
1801 struct dentry *dentry;
1802 char *path;
1803
1804 if (ceph_snap(inode) == CEPH_NOSNAP) {
1805 *pino = ceph_ino(inode);
1806 *ppathlen = 0;
1807 return 0;
1808 }
1809 dentry = d_find_alias(inode);
1810 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1811 dput(dentry);
1812 if (IS_ERR(path))
1813 return PTR_ERR(path);
1814 *ppath = path;
1815 *pfreepath = 1;
1816 return 0;
1817}
1818
1819/*
1820 * request arguments may be specified via an inode *, a dentry *, or
1821 * an explicit ino+path.
1822 */
1823static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1824 const char *rpath, u64 rino,
1825 const char **ppath, int *pathlen,
1826 u64 *ino, int *freepath)
1827{
1828 int r = 0;
1829
1830 if (rinode) {
1831 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1832 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1833 ceph_snap(rinode));
1834 } else if (rdentry) {
1835 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1836 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1837 *ppath);
795858db 1838 } else if (rpath || rino) {
2f2dc053
SW		 1839 *ino = rino;
1840 *ppath = rpath;
b000056a 1841 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1842 dout(" path %.*s\n", *pathlen, rpath);
1843 }
1844
1845 return r;
1846}
1847
1848/*
1849 * called under mdsc->mutex
1850 */
1851static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1852 struct ceph_mds_request *req,
1853 int mds)
1854{
1855 struct ceph_msg *msg;
1856 struct ceph_mds_request_head *head;
1857 const char *path1 = NULL;
1858 const char *path2 = NULL;
1859 u64 ino1 = 0, ino2 = 0;
1860 int pathlen1 = 0, pathlen2 = 0;
1861 int freepath1 = 0, freepath2 = 0;
1862 int len;
1863 u16 releases;
1864 void *p, *end;
1865 int ret;
1866
1867 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1868 req->r_path1, req->r_ino1.ino,
1869 &path1, &pathlen1, &ino1, &freepath1);
1870 if (ret < 0) {
1871 msg = ERR_PTR(ret);
1872 goto out;
1873 }
1874
1875 ret = set_request_path_attr(NULL, req->r_old_dentry,
1876 req->r_path2, req->r_ino2.ino,
1877 &path2, &pathlen2, &ino2, &freepath2);
1878 if (ret < 0) {
1879 msg = ERR_PTR(ret);
1880 goto out_free1;
1881 }
1882
1883 len = sizeof(*head) +
b8e69066
SW		 1884 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
1885 sizeof(struct timespec);
2f2dc053
SW		 1886
1887 /* calculate (max) length for cap releases */
1888 len += sizeof(struct ceph_mds_request_release) *
1889 (!!req->r_inode_drop + !!req->r_dentry_drop +
1890 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1891 if (req->r_dentry_drop)
1892 len += req->r_dentry->d_name.len;
1893 if (req->r_old_dentry_drop)
1894 len += req->r_old_dentry->d_name.len;
1895
b61c2763 1896 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1897 if (!msg) {
1898 msg = ERR_PTR(-ENOMEM);
2f2dc053 1899 goto out_free2;
a79832f2 1900 }
2f2dc053 1901
7cfa0313 1902 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1903 msg->hdr.tid = cpu_to_le64(req->r_tid);
1904
2f2dc053
SW		 1905 head = msg->front.iov_base;
1906 p = msg->front.iov_base + sizeof(*head);
1907 end = msg->front.iov_base + msg->front.iov_len;
1908
1909 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1910 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1911 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1912 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1913 head->args = req->r_args;
1914
1915 ceph_encode_filepath(&p, end, ino1, path1);
1916 ceph_encode_filepath(&p, end, ino2, path2);
1917
e979cf50
SW		 1918 /* make note of release offset, in case we need to replay */
1919 req->r_request_release_offset = p - msg->front.iov_base;
1920
2f2dc053
SW		 1921 /* cap releases */
1922 releases = 0;
1923 if (req->r_inode_drop)
1924 releases += ceph_encode_inode_release(&p,
1925 req->r_inode ? req->r_inode : req->r_dentry->d_inode,
1926 mds, req->r_inode_drop, req->r_inode_unless, 0);
1927 if (req->r_dentry_drop)
1928 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1929 mds, req->r_dentry_drop, req->r_dentry_unless);
1930 if (req->r_old_dentry_drop)
1931 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1932 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1933 if (req->r_old_inode_drop)
1934 releases += ceph_encode_inode_release(&p,
1935 req->r_old_dentry->d_inode,
1936 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1937 head->num_releases = cpu_to_le16(releases);
1938
b8e69066
SW		 1939 /* time stamp */
1940 ceph_encode_copy(&p, &req->r_stamp, sizeof(req->r_stamp));
1941
2f2dc053
SW		 1942 BUG_ON(p > end);
1943 msg->front.iov_len = p - msg->front.iov_base;
1944 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1945
25e6bae3
YZ		 1946 if (req->r_pagelist) {
1947 struct ceph_pagelist *pagelist = req->r_pagelist;
1948 atomic_inc(&pagelist->refcnt);
1949 ceph_msg_data_add_pagelist(msg, pagelist);
1950 msg->hdr.data_len = cpu_to_le32(pagelist->length);
1951 } else {
1952 msg->hdr.data_len = 0;
ebf18f47 1953 }
02afca6c 1954
2f2dc053
SW		 1955 msg->hdr.data_off = cpu_to_le16(0);
1956
1957out_free2:
1958 if (freepath2)
1959 kfree((char *)path2);
1960out_free1:
1961 if (freepath1)
1962 kfree((char *)path1);
1963out:
1964 return msg;
1965}
1966
1967/*
1968 * called under mdsc->mutex if error, under no mutex if
1969 * success.
1970 */
1971static void complete_request(struct ceph_mds_client *mdsc,
1972 struct ceph_mds_request *req)
1973{
1974 if (req->r_callback)
1975 req->r_callback(mdsc, req);
1976 else
03066f23 1977 complete_all(&req->r_completion);
2f2dc053
SW		 1978}
1979
1980/*
1981 * called under mdsc->mutex
1982 */
1983static int __prepare_send_request(struct ceph_mds_client *mdsc,
1984 struct ceph_mds_request *req,
1985 int mds)
1986{
1987 struct ceph_mds_request_head *rhead;
1988 struct ceph_msg *msg;
1989 int flags = 0;
1990
2f2dc053 1991 req->r_attempts++;
e55b71f8
GF		 1992 if (req->r_inode) {
1993 struct ceph_cap *cap =
1994 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
1995
1996 if (cap)
1997 req->r_sent_on_mseq = cap->mseq;
1998 else
1999 req->r_sent_on_mseq = -1;
2000 }
2f2dc053
SW		 2001 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2002 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2003
01a92f17 2004 if (req->r_got_unsafe) {
c5c9a0bf 2005 void *p;
01a92f17
SW		 2006 /*
2007 * Replay. Do not regenerate message (and rebuild
2008 * paths, etc.); just use the original message.
2009 * Rebuilding paths will break for renames because
2010 * d_move mangles the src name.
2011 */
2012 msg = req->r_request;
2013 rhead = msg->front.iov_base;
2014
2015 flags = le32_to_cpu(rhead->flags);
2016 flags |= CEPH_MDS_FLAG_REPLAY;
2017 rhead->flags = cpu_to_le32(flags);
2018
2019 if (req->r_target_inode)
2020 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2021
2022 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2023
2024 /* remove cap/dentry releases from message */
2025 rhead->num_releases = 0;
c5c9a0bf
YZ		 2026
2027 /* time stamp */
2028 p = msg->front.iov_base + req->r_request_release_offset;
2029 ceph_encode_copy(&p, &req->r_stamp, sizeof(req->r_stamp));
2030
2031 msg->front.iov_len = p - msg->front.iov_base;
2032 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2033 return 0;
2034 }
2035
2f2dc053
SW		 2036 if (req->r_request) {
2037 ceph_msg_put(req->r_request);
2038 req->r_request = NULL;
2039 }
2040 msg = create_request_message(mdsc, req, mds);
2041 if (IS_ERR(msg)) {
e1518c7c 2042 req->r_err = PTR_ERR(msg);
2f2dc053 2043 complete_request(mdsc, req);
a79832f2 2044 return PTR_ERR(msg);
2f2dc053
SW		 2045 }
2046 req->r_request = msg;
2047
2048 rhead = msg->front.iov_base;
2f2dc053
SW		 2049 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2050 if (req->r_got_unsafe)
2051 flags |= CEPH_MDS_FLAG_REPLAY;
2052 if (req->r_locked_dir)
2053 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2054 rhead->flags = cpu_to_le32(flags);
2055 rhead->num_fwd = req->r_num_fwd;
2056 rhead->num_retry = req->r_attempts - 1;
01a92f17 2057 rhead->ino = 0;
2f2dc053
SW		 2058
2059 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 2060 return 0;
2061}
2062
2063/*
2064 * send request, or put it on the appropriate wait list.
2065 */
2066static int __do_request(struct ceph_mds_client *mdsc,
2067 struct ceph_mds_request *req)
2068{
2069 struct ceph_mds_session *session = NULL;
2070 int mds = -1;
2071 int err = -EAGAIN;
2072
eb1b8af3
YZ		 2073 if (req->r_err || req->r_got_result) {
2074 if (req->r_aborted)
2075 __unregister_request(mdsc, req);
2f2dc053 2076 goto out;
eb1b8af3 2077 }
2f2dc053
SW		 2078
2079 if (req->r_timeout &&
2080 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2081 dout("do_request timed out\n");
2082 err = -EIO;
2083 goto finish;
2084 }
2085
dc69e2e9
SW		 2086 put_request_session(req);
2087
2f2dc053
SW		 2088 mds = __choose_mds(mdsc, req);
2089 if (mds < 0 ||
2090 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2091 dout("do_request no mds or not active, waiting for map\n");
2092 list_add(&req->r_wait, &mdsc->waiting_for_map);
2093 goto out;
2094 }
2095
2096 /* get, open session */
2097 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2098 if (!session) {
2f2dc053 2099 session = register_session(mdsc, mds);
9c423956
SW		 2100 if (IS_ERR(session)) {
2101 err = PTR_ERR(session);
2102 goto finish;
2103 }
2104 }
dc69e2e9
SW		 2105 req->r_session = get_session(session);
2106
2f2dc053 2107 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2108 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2109 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2110 session->s_state != CEPH_MDS_SESSION_HUNG) {
2111 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2112 session->s_state == CEPH_MDS_SESSION_CLOSING)
2113 __open_session(mdsc, session);
2114 list_add(&req->r_wait, &session->s_waiting);
2115 goto out_session;
2116 }
2117
2118 /* send request */
2f2dc053
SW		 2119 req->r_resend_mds = -1; /* forget any previous mds hint */
2120
2121 if (req->r_request_started == 0) /* note request start time */
2122 req->r_request_started = jiffies;
2123
2124 err = __prepare_send_request(mdsc, req, mds);
2125 if (!err) {
2126 ceph_msg_get(req->r_request);
2127 ceph_con_send(&session->s_con, req->r_request);
2128 }
2129
2130out_session:
2131 ceph_put_mds_session(session);
2132out:
2133 return err;
2134
2135finish:
e1518c7c 2136 req->r_err = err;
2f2dc053
SW		 2137 complete_request(mdsc, req);
2138 goto out;
2139}
2140
2141/*
2142 * called under mdsc->mutex
2143 */
2144static void __wake_requests(struct ceph_mds_client *mdsc,
2145 struct list_head *head)
2146{
ed75ec2c
YZ		 2147 struct ceph_mds_request *req;
2148 LIST_HEAD(tmp_list);
2149
2150 list_splice_init(head, &tmp_list);
2f2dc053 2151
ed75ec2c
YZ		 2152 while (!list_empty(&tmp_list)) {
2153 req = list_entry(tmp_list.next,
2154 struct ceph_mds_request, r_wait);
2f2dc053 2155 list_del_init(&req->r_wait);
7971bd92 2156 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2157 __do_request(mdsc, req);
2158 }
2159}
2160
2161/*
2162 * Wake up threads with requests pending for @mds, so that they can
29790f26 2163 * resubmit their requests to a possibly different mds.
2f2dc053 2164 */
29790f26 2165static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2166{
44ca18f2 2167 struct ceph_mds_request *req;
282c1052 2168 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2169
2170 dout("kick_requests mds%d\n", mds);
282c1052 2171 while (p) {
44ca18f2 2172 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2173 p = rb_next(p);
44ca18f2
SW		 2174 if (req->r_got_unsafe)
2175 continue;
2176 if (req->r_session &&
2177 req->r_session->s_mds == mds) {
2178 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2179 list_del_init(&req->r_wait);
44ca18f2 2180 __do_request(mdsc, req);
2f2dc053
SW		 2181 }
2182 }
2183}
2184
2185void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2186 struct ceph_mds_request *req)
2187{
2188 dout("submit_request on %p\n", req);
2189 mutex_lock(&mdsc->mutex);
2190 __register_request(mdsc, req, NULL);
2191 __do_request(mdsc, req);
2192 mutex_unlock(&mdsc->mutex);
2193}
2194
2195/*
2196 * Synchrously perform an mds request. Take care of all of the
2197 * session setup, forwarding, retry details.
2198 */
2199int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2200 struct inode *dir,
2201 struct ceph_mds_request *req)
2202{
2203 int err;
2204
2205 dout("do_request on %p\n", req);
2206
2207 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2208 if (req->r_inode)
2209 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2210 if (req->r_locked_dir)
2211 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
844d87c3 2212 if (req->r_old_dentry_dir)
41b02e1f
SW		 2213 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2214 CEPH_CAP_PIN);
2f2dc053
SW		 2215
2216 /* issue */
2217 mutex_lock(&mdsc->mutex);
2218 __register_request(mdsc, req, dir);
2219 __do_request(mdsc, req);
2220
e1518c7c
SW		 2221 if (req->r_err) {
2222 err = req->r_err;
2223 __unregister_request(mdsc, req);
2224 dout("do_request early error %d\n", err);
2225 goto out;
2f2dc053
SW		 2226 }
2227
e1518c7c
SW		 2228 /* wait */
2229 mutex_unlock(&mdsc->mutex);
2230 dout("do_request waiting\n");
2231 if (req->r_timeout) {
aa91647c 2232 err = (long)wait_for_completion_killable_timeout(
e1518c7c
SW		 2233 &req->r_completion, req->r_timeout);
2234 if (err == 0)
2235 err = -EIO;
9280be24
YZ		 2236 } else if (req->r_wait_for_completion) {
2237 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2238 } else {
aa91647c 2239 err = wait_for_completion_killable(&req->r_completion);
e1518c7c
SW		 2240 }
2241 dout("do_request waited, got %d\n", err);
2242 mutex_lock(&mdsc->mutex);
5b1daecd 2243
e1518c7c
SW		 2244 /* only abort if we didn't race with a real reply */
2245 if (req->r_got_result) {
2246 err = le32_to_cpu(req->r_reply_info.head->result);
2247 } else if (err < 0) {
2248 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2249
2250 /*
2251 * ensure we aren't running concurrently with
2252 * ceph_fill_trace or ceph_readdir_prepopulate, which
2253 * rely on locks (dir mutex) held by our caller.
2254 */
2255 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 2256 req->r_err = err;
2257 req->r_aborted = true;
b4556396 2258 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2259
e1518c7c 2260 if (req->r_locked_dir &&
167c9e35
SW		 2261 (req->r_op & CEPH_MDS_OP_WRITE))
2262 ceph_invalidate_dir_request(req);
2f2dc053 2263 } else {
e1518c7c 2264 err = req->r_err;
2f2dc053 2265 }
2f2dc053 2266
e1518c7c
SW		 2267out:
2268 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2269 dout("do_request %p done, result %d\n", req, err);
2270 return err;
2271}
2272
167c9e35 2273/*
2f276c51 2274 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2275 * namespace request.
2276 */
2277void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2278{
2279 struct inode *inode = req->r_locked_dir;
167c9e35 2280
2f276c51 2281 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2282
2f276c51 2283 ceph_dir_clear_complete(inode);
167c9e35
SW		 2284 if (req->r_dentry)
2285 ceph_invalidate_dentry_lease(req->r_dentry);
2286 if (req->r_old_dentry)
2287 ceph_invalidate_dentry_lease(req->r_old_dentry);
2288}
2289
2f2dc053
SW		 2290/*
2291 * Handle mds reply.
2292 *
2293 * We take the session mutex and parse and process the reply immediately.
2294 * This preserves the logical ordering of replies, capabilities, etc., sent
2295 * by the MDS as they are applied to our local cache.
2296 */
2297static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2298{
2299 struct ceph_mds_client *mdsc = session->s_mdsc;
2300 struct ceph_mds_request *req;
2301 struct ceph_mds_reply_head *head = msg->front.iov_base;
2302 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2303 struct ceph_snap_realm *realm;
2f2dc053
SW		 2304 u64 tid;
2305 int err, result;
2600d2dd 2306 int mds = session->s_mds;
2f2dc053 2307
2f2dc053
SW		 2308 if (msg->front.iov_len < sizeof(*head)) {
2309 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2310 ceph_msg_dump(msg);
2f2dc053
SW		 2311 return;
2312 }
2313
2314 /* get request, session */
6df058c0 2315 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2316 mutex_lock(&mdsc->mutex);
2317 req = __lookup_request(mdsc, tid);
2318 if (!req) {
2319 dout("handle_reply on unknown tid %llu\n", tid);
2320 mutex_unlock(&mdsc->mutex);
2321 return;
2322 }
2323 dout("handle_reply %p\n", req);
2f2dc053
SW		 2324
2325 /* correct session? */
d96d6049 2326 if (req->r_session != session) {
2f2dc053
SW		 2327 pr_err("mdsc_handle_reply got %llu on session mds%d"
2328 " not mds%d\n", tid, session->s_mds,
2329 req->r_session ? req->r_session->s_mds : -1);
2330 mutex_unlock(&mdsc->mutex);
2331 goto out;
2332 }
2333
2334 /* dup? */
2335 if ((req->r_got_unsafe && !head->safe) ||
2336 (req->r_got_safe && head->safe)) {
f3ae1b97 2337 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2338 head->safe ? "safe" : "unsafe", tid, mds);
2339 mutex_unlock(&mdsc->mutex);
2340 goto out;
2341 }
85792d0d 2342 if (req->r_got_safe && !head->safe) {
f3ae1b97 2343 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2344 tid, mds);
2345 mutex_unlock(&mdsc->mutex);
2346 goto out;
2347 }
2f2dc053
SW		 2348
2349 result = le32_to_cpu(head->result);
2350
2351 /*
e55b71f8
GF		 2352 * Handle an ESTALE
2353 * if we're not talking to the authority, send to them
2354 * if the authority has changed while we weren't looking,
2355 * send to new authority
2356 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2357 */
2358 if (result == -ESTALE) {
e55b71f8 2359 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2360 req->r_resend_mds = -1;
ca18bede 2361 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2362 dout("not using auth, setting for that now");
2363 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2364 __do_request(mdsc, req);
2365 mutex_unlock(&mdsc->mutex);
2366 goto out;
e55b71f8 2367 } else {
ca18bede
YZ		 2368 int mds = __choose_mds(mdsc, req);
2369 if (mds >= 0 && mds != req->r_session->s_mds) {
2370 dout("but auth changed, so resending");
e55b71f8
GF		 2371 __do_request(mdsc, req);
2372 mutex_unlock(&mdsc->mutex);
2373 goto out;
2374 }
2f2dc053 2375 }
e55b71f8 2376 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2377 }
2378
e55b71f8 2379
2f2dc053
SW		 2380 if (head->safe) {
2381 req->r_got_safe = true;
2382 __unregister_request(mdsc, req);
2f2dc053
SW		 2383
2384 if (req->r_got_unsafe) {
2385 /*
2386 * We already handled the unsafe response, now do the
2387 * cleanup. No need to examine the response; the MDS
2388 * doesn't include any result info in the safe
2389 * response. And even if it did, there is nothing
2390 * useful we could do with a revised return value.
2391 */
2392 dout("got safe reply %llu, mds%d\n", tid, mds);
2393 list_del_init(&req->r_unsafe_item);
2394
2395 /* last unsafe request during umount? */
44ca18f2 2396 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2397 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2398 mutex_unlock(&mdsc->mutex);
2399 goto out;
2400 }
e1518c7c 2401 } else {
2f2dc053
SW		 2402 req->r_got_unsafe = true;
2403 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2404 }
2405
2406 dout("handle_reply tid %lld result %d\n", tid, result);
2407 rinfo = &req->r_reply_info;
14303d20 2408 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2409 mutex_unlock(&mdsc->mutex);
2410
2411 mutex_lock(&session->s_mutex);
2412 if (err < 0) {
25933abd 2413 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2414 ceph_msg_dump(msg);
2f2dc053
SW		 2415 goto out_err;
2416 }
2417
2418 /* snap trace */
982d6011 2419 realm = NULL;
2f2dc053
SW		 2420 if (rinfo->snapblob_len) {
2421 down_write(&mdsc->snap_rwsem);
2422 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2423 rinfo->snapblob + rinfo->snapblob_len,
2424 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2425 &realm);
2f2dc053
SW		 2426 downgrade_write(&mdsc->snap_rwsem);
2427 } else {
2428 down_read(&mdsc->snap_rwsem);
2429 }
2430
2431 /* insert trace into our cache */
b4556396 2432 mutex_lock(&req->r_fill_mutex);
3d14c5d2 2433 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053 2434 if (err == 0) {
6e8575fa 2435 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2436 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2437 ceph_readdir_prepopulate(req, req->r_session);
37151668 2438 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2439 }
b4556396 2440 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2441
2442 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2443 if (realm)
2444 ceph_put_snap_realm(mdsc, realm);
2f2dc053 2445out_err:
e1518c7c
SW		 2446 mutex_lock(&mdsc->mutex);
2447 if (!req->r_aborted) {
2448 if (err) {
2449 req->r_err = err;
2450 } else {
2451 req->r_reply = msg;
2452 ceph_msg_get(msg);
2453 req->r_got_result = true;
2454 }
2f2dc053 2455 } else {
e1518c7c 2456 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2457 }
e1518c7c 2458 mutex_unlock(&mdsc->mutex);
2f2dc053 2459
ee6b272b 2460 ceph_add_cap_releases(mdsc, req->r_session);
2f2dc053
SW		 2461 mutex_unlock(&session->s_mutex);
2462
2463 /* kick calling process */
2464 complete_request(mdsc, req);
2465out:
2466 ceph_mdsc_put_request(req);
2467 return;
2468}
2469
2470
2471
2472/*
2473 * handle mds notification that our request has been forwarded.
2474 */
2600d2dd
SW		 2475static void handle_forward(struct ceph_mds_client *mdsc,
2476 struct ceph_mds_session *session,
2477 struct ceph_msg *msg)
2f2dc053
SW		 2478{
2479 struct ceph_mds_request *req;
a1ea787c 2480 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2481 u32 next_mds;
2482 u32 fwd_seq;
2f2dc053
SW		 2483 int err = -EINVAL;
2484 void *p = msg->front.iov_base;
2485 void *end = p + msg->front.iov_len;
2f2dc053 2486
a1ea787c 2487 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2488 next_mds = ceph_decode_32(&p);
2489 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2490
2491 mutex_lock(&mdsc->mutex);
2492 req = __lookup_request(mdsc, tid);
2493 if (!req) {
2a8e5e36 2494 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2495 goto out; /* dup reply? */
2496 }
2497
2a8e5e36
SW		 2498 if (req->r_aborted) {
2499 dout("forward tid %llu aborted, unregistering\n", tid);
2500 __unregister_request(mdsc, req);
2501 } else if (fwd_seq <= req->r_num_fwd) {
2502 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2503 tid, next_mds, req->r_num_fwd, fwd_seq);
2504 } else {
2505 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2506 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2507 BUG_ON(req->r_err);
2508 BUG_ON(req->r_got_result);
2f2dc053
SW		 2509 req->r_num_fwd = fwd_seq;
2510 req->r_resend_mds = next_mds;
2511 put_request_session(req);
2512 __do_request(mdsc, req);
2513 }
2514 ceph_mdsc_put_request(req);
2515out:
2516 mutex_unlock(&mdsc->mutex);
2517 return;
2518
2519bad:
2520 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2521}
2522
2523/*
2524 * handle a mds session control message
2525 */
2526static void handle_session(struct ceph_mds_session *session,
2527 struct ceph_msg *msg)
2528{
2529 struct ceph_mds_client *mdsc = session->s_mdsc;
2530 u32 op;
2531 u64 seq;
2600d2dd 2532 int mds = session->s_mds;
2f2dc053
SW		 2533 struct ceph_mds_session_head *h = msg->front.iov_base;
2534 int wake = 0;
2535
2f2dc053
SW		 2536 /* decode */
2537 if (msg->front.iov_len != sizeof(*h))
2538 goto bad;
2539 op = le32_to_cpu(h->op);
2540 seq = le64_to_cpu(h->seq);
2541
2542 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2543 if (op == CEPH_SESSION_CLOSE)
2544 __unregister_session(mdsc, session);
2f2dc053
SW		 2545 /* FIXME: this ttl calculation is generous */
2546 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2547 mutex_unlock(&mdsc->mutex);
2548
2549 mutex_lock(&session->s_mutex);
2550
2551 dout("handle_session mds%d %s %p state %s seq %llu\n",
2552 mds, ceph_session_op_name(op), session,
a687ecaf 2553 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2554
2555 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2556 session->s_state = CEPH_MDS_SESSION_OPEN;
2557 pr_info("mds%d came back\n", session->s_mds);
2558 }
2559
2560 switch (op) {
2561 case CEPH_SESSION_OPEN:
29790f26
SW		 2562 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2563 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2564 session->s_state = CEPH_MDS_SESSION_OPEN;
2565 renewed_caps(mdsc, session, 0);
2566 wake = 1;
2567 if (mdsc->stopping)
2568 __close_session(mdsc, session);
2569 break;
2570
2571 case CEPH_SESSION_RENEWCAPS:
2572 if (session->s_renew_seq == seq)
2573 renewed_caps(mdsc, session, 1);
2574 break;
2575
2576 case CEPH_SESSION_CLOSE:
29790f26
SW		 2577 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2578 pr_info("mds%d reconnect denied\n", session->s_mds);
2f2dc053 2579 remove_session_caps(session);
656e4382 2580 wake = 2; /* for good measure */
f3c60c59 2581 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2582 break;
2583
2584 case CEPH_SESSION_STALE:
2585 pr_info("mds%d caps went stale, renewing\n",
2586 session->s_mds);
d8fb02ab 2587 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2588 session->s_cap_gen++;
1ce208a6 2589 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2590 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2591 send_renew_caps(mdsc, session);
2592 break;
2593
2594 case CEPH_SESSION_RECALL_STATE:
2595 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2596 break;
2597
186e4f7a
YZ		 2598 case CEPH_SESSION_FLUSHMSG:
2599 send_flushmsg_ack(mdsc, session, seq);
2600 break;
2601
03f4fcb0
YZ		 2602 case CEPH_SESSION_FORCE_RO:
2603 dout("force_session_readonly %p\n", session);
2604 spin_lock(&session->s_cap_lock);
2605 session->s_readonly = true;
2606 spin_unlock(&session->s_cap_lock);
2607 wake_up_session_caps(session, 0);
2608 break;
2609
2f2dc053
SW		 2610 default:
2611 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2612 WARN_ON(1);
2613 }
2614
2615 mutex_unlock(&session->s_mutex);
2616 if (wake) {
2617 mutex_lock(&mdsc->mutex);
2618 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2619 if (wake == 2)
2620 kick_requests(mdsc, mds);
2f2dc053
SW		 2621 mutex_unlock(&mdsc->mutex);
2622 }
2623 return;
2624
2625bad:
2626 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2627 (int)msg->front.iov_len);
9ec7cab1 2628 ceph_msg_dump(msg);
2f2dc053
SW		 2629 return;
2630}
2631
2632
2633/*
2634 * called under session->mutex.
2635 */
2636static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2637 struct ceph_mds_session *session)
2638{
2639 struct ceph_mds_request *req, *nreq;
2640 int err;
2641
2642 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2643
2644 mutex_lock(&mdsc->mutex);
2645 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2646 err = __prepare_send_request(mdsc, req, session->s_mds);
2647 if (!err) {
2648 ceph_msg_get(req->r_request);
2649 ceph_con_send(&session->s_con, req->r_request);
2650 }
2651 }
2652 mutex_unlock(&mdsc->mutex);
2653}
2654
2655/*
2656 * Encode information about a cap for a reconnect with the MDS.
2657 */
2f2dc053
SW		 2658static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2659 void *arg)
2660{
20cb34ae
SW		 2661 union {
2662 struct ceph_mds_cap_reconnect v2;
2663 struct ceph_mds_cap_reconnect_v1 v1;
2664 } rec;
2665 size_t reclen;
2f2dc053 2666 struct ceph_inode_info *ci;
20cb34ae
SW		 2667 struct ceph_reconnect_state *recon_state = arg;
2668 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2669 char *path;
2670 int pathlen, err;
2671 u64 pathbase;
2672 struct dentry *dentry;
2673
2674 ci = cap->ci;
2675
2676 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2677 inode, ceph_vinop(inode), cap, cap->cap_id,
2678 ceph_cap_string(cap->issued));
93cea5be
SW		 2679 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2680 if (err)
2681 return err;
2f2dc053
SW		 2682
2683 dentry = d_find_alias(inode);
2684 if (dentry) {
2685 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2686 if (IS_ERR(path)) {
2687 err = PTR_ERR(path);
e072f8aa 2688 goto out_dput;
2f2dc053
SW		 2689 }
2690 } else {
2691 path = NULL;
2692 pathlen = 0;
2693 }
93cea5be
SW		 2694 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2695 if (err)
e072f8aa 2696 goto out_free;
2f2dc053 2697
be655596 2698 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2699 cap->seq = 0; /* reset cap seq */
2700 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2701 cap->mseq = 0; /* and migrate_seq */
99a9c273 2702 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae
SW		 2703
2704 if (recon_state->flock) {
2705 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2706 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2707 rec.v2.issued = cpu_to_le32(cap->issued);
2708 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2709 rec.v2.pathbase = cpu_to_le64(pathbase);
2710 rec.v2.flock_len = 0;
2711 reclen = sizeof(rec.v2);
2712 } else {
2713 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2714 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2715 rec.v1.issued = cpu_to_le32(cap->issued);
2716 rec.v1.size = cpu_to_le64(inode->i_size);
2717 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2718 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2719 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2720 rec.v1.pathbase = cpu_to_le64(pathbase);
2721 reclen = sizeof(rec.v1);
2722 }
be655596 2723 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2724
40819f6f
GF		 2725 if (recon_state->flock) {
2726 int num_fcntl_locks, num_flock_locks;
39be95e9
JS		 2727 struct ceph_filelock *flocks;
2728
2729encode_again:
1c8c601a 2730 spin_lock(&inode->i_lock);
39be95e9 2731 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
1c8c601a 2732 spin_unlock(&inode->i_lock);
39be95e9
JS		 2733 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2734 sizeof(struct ceph_filelock), GFP_NOFS);
2735 if (!flocks) {
2736 err = -ENOMEM;
2737 goto out_free;
2738 }
1c8c601a 2739 spin_lock(&inode->i_lock);
39be95e9
JS		 2740 err = ceph_encode_locks_to_buffer(inode, flocks,
2741 num_fcntl_locks,
2742 num_flock_locks);
1c8c601a 2743 spin_unlock(&inode->i_lock);
39be95e9
JS		 2744 if (err) {
2745 kfree(flocks);
2746 if (err == -ENOSPC)
2747 goto encode_again;
2748 goto out_free;
2749 }
2750 /*
2751 * number of encoded locks is stable, so copy to pagelist
2752 */
2753 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2754 (num_fcntl_locks+num_flock_locks) *
2755 sizeof(struct ceph_filelock));
2756 err = ceph_pagelist_append(pagelist, &rec, reclen);
2757 if (!err)
2758 err = ceph_locks_to_pagelist(flocks, pagelist,
2759 num_fcntl_locks,
2760 num_flock_locks);
2761 kfree(flocks);
3612abbd
SW		 2762 } else {
2763 err = ceph_pagelist_append(pagelist, &rec, reclen);
40819f6f 2764 }
44c99757
YZ		 2765
2766 recon_state->nr_caps++;
e072f8aa 2767out_free:
2f2dc053 2768 kfree(path);
e072f8aa 2769out_dput:
2f2dc053 2770 dput(dentry);
93cea5be 2771 return err;
2f2dc053
SW		 2772}
2773
2774
2775/*
2776 * If an MDS fails and recovers, clients need to reconnect in order to
2777 * reestablish shared state. This includes all caps issued through
2778 * this session _and_ the snap_realm hierarchy. Because it's not
2779 * clear which snap realms the mds cares about, we send everything we
2780 * know about.. that ensures we'll then get any new info the
2781 * recovering MDS might have.
2782 *
2783 * This is a relatively heavyweight operation, but it's rare.
2784 *
2785 * called with mdsc->mutex held.
2786 */
34b6c855
SW		 2787static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2788 struct ceph_mds_session *session)
2f2dc053 2789{
2f2dc053 2790 struct ceph_msg *reply;
a105f00c 2791 struct rb_node *p;
34b6c855 2792 int mds = session->s_mds;
9abf82b8 2793 int err = -ENOMEM;
44c99757 2794 int s_nr_caps;
93cea5be 2795 struct ceph_pagelist *pagelist;
20cb34ae 2796 struct ceph_reconnect_state recon_state;
2f2dc053 2797
34b6c855 2798 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2799
93cea5be
SW		 2800 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2801 if (!pagelist)
2802 goto fail_nopagelist;
2803 ceph_pagelist_init(pagelist);
2804
b61c2763 2805 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 2806 if (!reply)
93cea5be 2807 goto fail_nomsg;
93cea5be 2808
34b6c855
SW		 2809 mutex_lock(&session->s_mutex);
2810 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2811 session->s_seq = 0;
2f2dc053 2812
2f2dc053 2813 dout("session %p state %s\n", session,
a687ecaf 2814 ceph_session_state_name(session->s_state));
2f2dc053 2815
99a9c273
YZ		 2816 spin_lock(&session->s_gen_ttl_lock);
2817 session->s_cap_gen++;
2818 spin_unlock(&session->s_gen_ttl_lock);
2819
2820 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 2821 /* don't know if session is readonly */
2822 session->s_readonly = 0;
99a9c273
YZ		 2823 /*
2824 * notify __ceph_remove_cap() that we are composing cap reconnect.
2825 * If a cap get released before being added to the cap reconnect,
2826 * __ceph_remove_cap() should skip queuing cap release.
2827 */
2828 session->s_cap_reconnect = 1;
e01a5946
SW		 2829 /* drop old cap expires; we're about to reestablish that state */
2830 discard_cap_releases(mdsc, session);
99a9c273 2831 spin_unlock(&session->s_cap_lock);
e01a5946 2832
5d23371f
YZ		 2833 /* trim unused caps to reduce MDS's cache rejoin time */
2834 shrink_dcache_parent(mdsc->fsc->sb->s_root);
2835
2836 ceph_con_close(&session->s_con);
2837 ceph_con_open(&session->s_con,
2838 CEPH_ENTITY_TYPE_MDS, mds,
2839 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2840
2841 /* replay unsafe requests */
2842 replay_unsafe_requests(mdsc, session);
2843
2844 down_read(&mdsc->snap_rwsem);
2845
2f2dc053 2846 /* traverse this session's caps */
44c99757
YZ		 2847 s_nr_caps = session->s_nr_caps;
2848 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 2849 if (err)
2850 goto fail;
20cb34ae 2851
44c99757 2852 recon_state.nr_caps = 0;
20cb34ae
SW		 2853 recon_state.pagelist = pagelist;
2854 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2855 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2856 if (err < 0)
9abf82b8 2857 goto fail;
2f2dc053 2858
99a9c273
YZ		 2859 spin_lock(&session->s_cap_lock);
2860 session->s_cap_reconnect = 0;
2861 spin_unlock(&session->s_cap_lock);
2862
2f2dc053
SW		 2863 /*
2864 * snaprealms. we provide mds with the ino, seq (version), and
2865 * parent for all of our realms. If the mds has any newer info,
2866 * it will tell us.
2867 */
a105f00c
SW		 2868 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2869 struct ceph_snap_realm *realm =
2870 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 2871 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 2872
2873 dout(" adding snap realm %llx seq %lld parent %llx\n",
2874 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 2875 sr_rec.ino = cpu_to_le64(realm->ino);
2876 sr_rec.seq = cpu_to_le64(realm->seq);
2877 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2878 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2879 if (err)
2880 goto fail;
2f2dc053 2881 }
2f2dc053 2882
20cb34ae
SW		 2883 if (recon_state.flock)
2884 reply->hdr.version = cpu_to_le16(2);
44c99757
YZ		 2885
2886 /* raced with cap release? */
2887 if (s_nr_caps != recon_state.nr_caps) {
2888 struct page *page = list_first_entry(&pagelist->head,
2889 struct page, lru);
2890 __le32 *addr = kmap_atomic(page);
2891 *addr = cpu_to_le32(recon_state.nr_caps);
2892 kunmap_atomic(addr);
ebf18f47 2893 }
44c99757
YZ		 2894
2895 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2896 ceph_msg_data_add_pagelist(reply, pagelist);
2f2dc053
SW		 2897 ceph_con_send(&session->s_con, reply);
2898
9abf82b8
SW		 2899 mutex_unlock(&session->s_mutex);
2900
2901 mutex_lock(&mdsc->mutex);
2902 __wake_requests(mdsc, &session->s_waiting);
2903 mutex_unlock(&mdsc->mutex);
2904
2f2dc053 2905 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 2906 return;
2907
93cea5be 2908fail:
2f2dc053 2909 ceph_msg_put(reply);
9abf82b8
SW		 2910 up_read(&mdsc->snap_rwsem);
2911 mutex_unlock(&session->s_mutex);
93cea5be
SW		 2912fail_nomsg:
2913 ceph_pagelist_release(pagelist);
93cea5be 2914fail_nopagelist:
9abf82b8 2915 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 2916 return;
2f2dc053
SW		 2917}
2918
2919
2920/*
2921 * compare old and new mdsmaps, kicking requests
2922 * and closing out old connections as necessary
2923 *
2924 * called under mdsc->mutex.
2925 */
2926static void check_new_map(struct ceph_mds_client *mdsc,
2927 struct ceph_mdsmap *newmap,
2928 struct ceph_mdsmap *oldmap)
2929{
2930 int i;
2931 int oldstate, newstate;
2932 struct ceph_mds_session *s;
2933
2934 dout("check_new_map new %u old %u\n",
2935 newmap->m_epoch, oldmap->m_epoch);
2936
2937 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2938 if (mdsc->sessions[i] == NULL)
2939 continue;
2940 s = mdsc->sessions[i];
2941 oldstate = ceph_mdsmap_get_state(oldmap, i);
2942 newstate = ceph_mdsmap_get_state(newmap, i);
2943
0deb01c9 2944 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 2945 i, ceph_mds_state_name(oldstate),
0deb01c9 2946 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 2947 ceph_mds_state_name(newstate),
0deb01c9 2948 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 2949 ceph_session_state_name(s->s_state));
2f2dc053 2950
3e8f43a0
YZ		 2951 if (i >= newmap->m_max_mds ||
2952 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 2953 ceph_mdsmap_get_addr(newmap, i),
2954 sizeof(struct ceph_entity_addr))) {
2955 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
2956 /* the session never opened, just close it
2957 * out now */
2958 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 2959 __unregister_session(mdsc, s);
2f2dc053
SW		 2960 } else {
2961 /* just close it */
2962 mutex_unlock(&mdsc->mutex);
2963 mutex_lock(&s->s_mutex);
2964 mutex_lock(&mdsc->mutex);
2965 ceph_con_close(&s->s_con);
2966 mutex_unlock(&s->s_mutex);
2967 s->s_state = CEPH_MDS_SESSION_RESTARTING;
2968 }
2969
2970 /* kick any requests waiting on the recovering mds */
29790f26 2971 kick_requests(mdsc, i);
2f2dc053
SW		 2972 } else if (oldstate == newstate) {
2973 continue; /* nothing new with this mds */
2974 }
2975
2976 /*
2977 * send reconnect?
2978 */
2979 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 2980 newstate >= CEPH_MDS_STATE_RECONNECT) {
2981 mutex_unlock(&mdsc->mutex);
2982 send_mds_reconnect(mdsc, s);
2983 mutex_lock(&mdsc->mutex);
2984 }
2f2dc053
SW		 2985
2986 /*
29790f26 2987 * kick request on any mds that has gone active.
2f2dc053
SW		 2988 */
2989 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
2990 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 2991 if (oldstate != CEPH_MDS_STATE_CREATING &&
2992 oldstate != CEPH_MDS_STATE_STARTING)
2993 pr_info("mds%d recovery completed\n", s->s_mds);
2994 kick_requests(mdsc, i);
2f2dc053 2995 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 2996 wake_up_session_caps(s, 1);
2f2dc053
SW		 2997 }
2998 }
cb170a22
SW		 2999
3000 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3001 s = mdsc->sessions[i];
3002 if (!s)
3003 continue;
3004 if (!ceph_mdsmap_is_laggy(newmap, i))
3005 continue;
3006 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3007 s->s_state == CEPH_MDS_SESSION_HUNG ||
3008 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3009 dout(" connecting to export targets of laggy mds%d\n",
3010 i);
3011 __open_export_target_sessions(mdsc, s);
3012 }
3013 }
2f2dc053
SW		 3014}
3015
3016
3017
3018/*
3019 * leases
3020 */
3021
3022/*
3023 * caller must hold session s_mutex, dentry->d_lock
3024 */
3025void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3026{
3027 struct ceph_dentry_info *di = ceph_dentry(dentry);
3028
3029 ceph_put_mds_session(di->lease_session);
3030 di->lease_session = NULL;
3031}
3032
2600d2dd
SW		 3033static void handle_lease(struct ceph_mds_client *mdsc,
3034 struct ceph_mds_session *session,
3035 struct ceph_msg *msg)
2f2dc053 3036{
3d14c5d2 3037 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3038 struct inode *inode;
2f2dc053
SW		 3039 struct dentry *parent, *dentry;
3040 struct ceph_dentry_info *di;
2600d2dd 3041 int mds = session->s_mds;
2f2dc053 3042 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3043 u32 seq;
2f2dc053 3044 struct ceph_vino vino;
2f2dc053
SW		 3045 struct qstr dname;
3046 int release = 0;
3047
2f2dc053
SW		 3048 dout("handle_lease from mds%d\n", mds);
3049
3050 /* decode */
3051 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3052 goto bad;
3053 vino.ino = le64_to_cpu(h->ino);
3054 vino.snap = CEPH_NOSNAP;
1e5ea23d 3055 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3056 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3057 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3058 if (dname.len != get_unaligned_le32(h+1))
3059 goto bad;
3060
2f2dc053
SW		 3061 /* lookup inode */
3062 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3063 dout("handle_lease %s, ino %llx %p %.*s\n",
3064 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3065 dname.len, dname.name);
6cd3bcad
YZ		 3066
3067 mutex_lock(&session->s_mutex);
3068 session->s_seq++;
3069
2f2dc053
SW		 3070 if (inode == NULL) {
3071 dout("handle_lease no inode %llx\n", vino.ino);
3072 goto release;
3073 }
2f2dc053
SW		 3074
3075 /* dentry */
3076 parent = d_find_alias(inode);
3077 if (!parent) {
3078 dout("no parent dentry on inode %p\n", inode);
3079 WARN_ON(1);
3080 goto release; /* hrm... */
3081 }
3082 dname.hash = full_name_hash(dname.name, dname.len);
3083 dentry = d_lookup(parent, &dname);
3084 dput(parent);
3085 if (!dentry)
3086 goto release;
3087
3088 spin_lock(&dentry->d_lock);
3089 di = ceph_dentry(dentry);
3090 switch (h->action) {
3091 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3092 if (di->lease_session == session) {
1e5ea23d
SW		 3093 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3094 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3095 __ceph_mdsc_drop_dentry_lease(dentry);
3096 }
3097 release = 1;
3098 break;
3099
3100 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3101 if (di->lease_session == session &&
2f2dc053
SW		 3102 di->lease_gen == session->s_cap_gen &&
3103 di->lease_renew_from &&
3104 di->lease_renew_after == 0) {
3105 unsigned long duration =
3106 le32_to_cpu(h->duration_ms) * HZ / 1000;
3107
1e5ea23d 3108 di->lease_seq = seq;
2f2dc053
SW		 3109 dentry->d_time = di->lease_renew_from + duration;
3110 di->lease_renew_after = di->lease_renew_from +
3111 (duration >> 1);
3112 di->lease_renew_from = 0;
3113 }
3114 break;
3115 }
3116 spin_unlock(&dentry->d_lock);
3117 dput(dentry);
3118
3119 if (!release)
3120 goto out;
3121
3122release:
3123 /* let's just reuse the same message */
3124 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3125 ceph_msg_get(msg);
3126 ceph_con_send(&session->s_con, msg);
3127
3128out:
3129 iput(inode);
3130 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3131 return;
3132
3133bad:
3134 pr_err("corrupt lease message\n");
9ec7cab1 3135 ceph_msg_dump(msg);
2f2dc053
SW		 3136}
3137
3138void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3139 struct inode *inode,
3140 struct dentry *dentry, char action,
3141 u32 seq)
3142{
3143 struct ceph_msg *msg;
3144 struct ceph_mds_lease *lease;
3145 int len = sizeof(*lease) + sizeof(u32);
3146 int dnamelen = 0;
3147
3148 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3149 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3150 dnamelen = dentry->d_name.len;
3151 len += dnamelen;
3152
b61c2763 3153 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3154 if (!msg)
2f2dc053
SW		 3155 return;
3156 lease = msg->front.iov_base;
3157 lease->action = action;
2f2dc053
SW		 3158 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3159 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3160 lease->seq = cpu_to_le32(seq);
3161 put_unaligned_le32(dnamelen, lease + 1);
3162 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3163
3164 /*
3165 * if this is a preemptive lease RELEASE, no need to
3166 * flush request stream, since the actual request will
3167 * soon follow.
3168 */
3169 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3170
3171 ceph_con_send(&session->s_con, msg);
3172}
3173
3174/*
3175 * Preemptively release a lease we expect to invalidate anyway.
3176 * Pass @inode always, @dentry is optional.
3177 */
3178void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2f90b852 3179 struct dentry *dentry)
2f2dc053
SW		 3180{
3181 struct ceph_dentry_info *di;
3182 struct ceph_mds_session *session;
3183 u32 seq;
3184
3185 BUG_ON(inode == NULL);
3186 BUG_ON(dentry == NULL);
2f2dc053
SW		 3187
3188 /* is dentry lease valid? */
3189 spin_lock(&dentry->d_lock);
3190 di = ceph_dentry(dentry);
3191 if (!di || !di->lease_session ||
3192 di->lease_session->s_mds < 0 ||
3193 di->lease_gen != di->lease_session->s_cap_gen ||
3194 !time_before(jiffies, dentry->d_time)) {
3195 dout("lease_release inode %p dentry %p -- "
2f90b852
SW		 3196 "no lease\n",
3197 inode, dentry);
2f2dc053
SW		 3198 spin_unlock(&dentry->d_lock);
3199 return;
3200 }
3201
3202 /* we do have a lease on this dentry; note mds and seq */
3203 session = ceph_get_mds_session(di->lease_session);
3204 seq = di->lease_seq;
3205 __ceph_mdsc_drop_dentry_lease(dentry);
3206 spin_unlock(&dentry->d_lock);
3207
2f90b852
SW		 3208 dout("lease_release inode %p dentry %p to mds%d\n",
3209 inode, dentry, session->s_mds);
2f2dc053
SW		 3210 ceph_mdsc_lease_send_msg(session, inode, dentry,
3211 CEPH_MDS_LEASE_RELEASE, seq);
3212 ceph_put_mds_session(session);
3213}
3214
3215/*
3216 * drop all leases (and dentry refs) in preparation for umount
3217 */
3218static void drop_leases(struct ceph_mds_client *mdsc)
3219{
3220 int i;
3221
3222 dout("drop_leases\n");
3223 mutex_lock(&mdsc->mutex);
3224 for (i = 0; i < mdsc->max_sessions; i++) {
3225 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3226 if (!s)
3227 continue;
3228 mutex_unlock(&mdsc->mutex);
3229 mutex_lock(&s->s_mutex);
3230 mutex_unlock(&s->s_mutex);
3231 ceph_put_mds_session(s);
3232 mutex_lock(&mdsc->mutex);
3233 }
3234 mutex_unlock(&mdsc->mutex);
3235}
3236
3237
3238
3239/*
3240 * delayed work -- periodically trim expired leases, renew caps with mds
3241 */
3242static void schedule_delayed(struct ceph_mds_client *mdsc)
3243{
3244 int delay = 5;
3245 unsigned hz = round_jiffies_relative(HZ * delay);
3246 schedule_delayed_work(&mdsc->delayed_work, hz);
3247}
3248
3249static void delayed_work(struct work_struct *work)
3250{
3251 int i;
3252 struct ceph_mds_client *mdsc =
3253 container_of(work, struct ceph_mds_client, delayed_work.work);
3254 int renew_interval;
3255 int renew_caps;
3256
3257 dout("mdsc delayed_work\n");
afcdaea3 3258 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3259
3260 mutex_lock(&mdsc->mutex);
3261 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3262 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3263 mdsc->last_renew_caps);
3264 if (renew_caps)
3265 mdsc->last_renew_caps = jiffies;
3266
3267 for (i = 0; i < mdsc->max_sessions; i++) {
3268 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3269 if (s == NULL)
3270 continue;
3271 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3272 dout("resending session close request for mds%d\n",
3273 s->s_mds);
3274 request_close_session(mdsc, s);
3275 ceph_put_mds_session(s);
3276 continue;
3277 }
3278 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3279 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3280 s->s_state = CEPH_MDS_SESSION_HUNG;
3281 pr_info("mds%d hung\n", s->s_mds);
3282 }
3283 }
3284 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3285 /* this mds is failed or recovering, just wait */
3286 ceph_put_mds_session(s);
3287 continue;
3288 }
3289 mutex_unlock(&mdsc->mutex);
3290
3291 mutex_lock(&s->s_mutex);
3292 if (renew_caps)
3293 send_renew_caps(mdsc, s);
3294 else
3295 ceph_con_keepalive(&s->s_con);
ee6b272b 3296 ceph_add_cap_releases(mdsc, s);
aab53dd9
SW		 3297 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3298 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3299 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3300 mutex_unlock(&s->s_mutex);
3301 ceph_put_mds_session(s);
3302
3303 mutex_lock(&mdsc->mutex);
3304 }
3305 mutex_unlock(&mdsc->mutex);
3306
3307 schedule_delayed(mdsc);
3308}
3309
3d14c5d2 3310int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3311
2f2dc053 3312{
3d14c5d2
YS		 3313 struct ceph_mds_client *mdsc;
3314
3315 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3316 if (!mdsc)
3317 return -ENOMEM;
3318 mdsc->fsc = fsc;
3319 fsc->mdsc = mdsc;
2f2dc053
SW		 3320 mutex_init(&mdsc->mutex);
3321 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3322 if (mdsc->mdsmap == NULL) {
3323 kfree(mdsc);
2d06eeb8 3324 return -ENOMEM;
fb3101b6 3325 }
2d06eeb8 3326
2f2dc053 3327 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3328 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3329 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3330 mdsc->sessions = NULL;
3331 mdsc->max_sessions = 0;
3332 mdsc->stopping = 0;
3333 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3334 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3335 INIT_LIST_HEAD(&mdsc->snap_empty);
3336 spin_lock_init(&mdsc->snap_empty_lock);
3337 mdsc->last_tid = 0;
44ca18f2 3338 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3339 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3340 mdsc->last_renew_caps = jiffies;
3341 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3342 spin_lock_init(&mdsc->cap_delay_lock);
3343 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3344 spin_lock_init(&mdsc->snap_flush_lock);
3345 mdsc->cap_flush_seq = 0;
3346 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3347 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3348 mdsc->num_cap_flushing = 0;
3349 spin_lock_init(&mdsc->cap_dirty_lock);
3350 init_waitqueue_head(&mdsc->cap_flushing_wq);
3351 spin_lock_init(&mdsc->dentry_lru_lock);
3352 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3353
37151668 3354 ceph_caps_init(mdsc);
3d14c5d2 3355 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3356
5f44f142 3357 return 0;
2f2dc053
SW		 3358}
3359
3360/*
3361 * Wait for safe replies on open mds requests. If we time out, drop
3362 * all requests from the tree to avoid dangling dentry refs.
3363 */
3364static void wait_requests(struct ceph_mds_client *mdsc)
3365{
3366 struct ceph_mds_request *req;
3d14c5d2 3367 struct ceph_fs_client *fsc = mdsc->fsc;
2f2dc053
SW		 3368
3369 mutex_lock(&mdsc->mutex);
44ca18f2 3370 if (__get_oldest_req(mdsc)) {
2f2dc053 3371 mutex_unlock(&mdsc->mutex);
44ca18f2 3372
2f2dc053
SW		 3373 dout("wait_requests waiting for requests\n");
3374 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3d14c5d2 3375 fsc->client->options->mount_timeout * HZ);
2f2dc053
SW		 3376
3377 /* tear down remaining requests */
44ca18f2
SW		 3378 mutex_lock(&mdsc->mutex);
3379 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3380 dout("wait_requests timed out on tid %llu\n",
3381 req->r_tid);
44ca18f2 3382 __unregister_request(mdsc, req);
2f2dc053
SW		 3383 }
3384 }
3385 mutex_unlock(&mdsc->mutex);
3386 dout("wait_requests done\n");
3387}
3388
3389/*
3390 * called before mount is ro, and before dentries are torn down.
3391 * (hmm, does this still race with new lookups?)
3392 */
3393void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3394{
3395 dout("pre_umount\n");
3396 mdsc->stopping = 1;
3397
3398 drop_leases(mdsc);
afcdaea3 3399 ceph_flush_dirty_caps(mdsc);
2f2dc053 3400 wait_requests(mdsc);
17c688c3
SW		 3401
3402 /*
3403 * wait for reply handlers to drop their request refs and
3404 * their inode/dcache refs
3405 */
3406 ceph_msgr_flush();
2f2dc053
SW		 3407}
3408
3409/*
3410 * wait for all write mds requests to flush.
3411 */
3412static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3413{
80fc7314 3414 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3415 struct rb_node *n;
2f2dc053
SW		 3416
3417 mutex_lock(&mdsc->mutex);
3418 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3419restart:
44ca18f2
SW		 3420 req = __get_oldest_req(mdsc);
3421 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3422 /* find next request */
3423 n = rb_next(&req->r_node);
3424 if (n)
3425 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3426 else
3427 nextreq = NULL;
44ca18f2
SW		 3428 if ((req->r_op & CEPH_MDS_OP_WRITE)) {
3429 /* write op */
3430 ceph_mdsc_get_request(req);
80fc7314
SW		 3431 if (nextreq)
3432 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3433 mutex_unlock(&mdsc->mutex);
3434 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3435 req->r_tid, want_tid);
3436 wait_for_completion(&req->r_safe_completion);
3437 mutex_lock(&mdsc->mutex);
44ca18f2 3438 ceph_mdsc_put_request(req);
80fc7314
SW		 3439 if (!nextreq)
3440 break; /* next dne before, so we're done! */
3441 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3442 /* next request was removed from tree */
3443 ceph_mdsc_put_request(nextreq);
3444 goto restart;
3445 }
3446 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3447 }
80fc7314 3448 req = nextreq;
2f2dc053
SW		 3449 }
3450 mutex_unlock(&mdsc->mutex);
3451 dout("wait_unsafe_requests done\n");
3452}
3453
3454void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3455{
3456 u64 want_tid, want_flush;
3457
3d14c5d2 3458 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3459 return;
3460
2f2dc053
SW		 3461 dout("sync\n");
3462 mutex_lock(&mdsc->mutex);
3463 want_tid = mdsc->last_tid;
2f2dc053 3464 mutex_unlock(&mdsc->mutex);
2f2dc053 3465
afcdaea3 3466 ceph_flush_dirty_caps(mdsc);
d3383a8e
YZ		 3467 spin_lock(&mdsc->cap_dirty_lock);
3468 want_flush = mdsc->cap_flush_seq;
3469 spin_unlock(&mdsc->cap_dirty_lock);
3470
3471 dout("sync want tid %lld flush_seq %lld\n", want_tid, want_flush);
2f2dc053
SW		 3472
3473 wait_unsafe_requests(mdsc, want_tid);
d3383a8e 3474 wait_caps_flush(mdsc, want_flush);
2f2dc053
SW		 3475}
3476
f3c60c59
SW		 3477/*
3478 * true if all sessions are closed, or we force unmount
3479 */
7fd7d101 3480static bool done_closing_sessions(struct ceph_mds_client *mdsc)
f3c60c59
SW		 3481{
3482 int i, n = 0;
3483
3d14c5d2 3484 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
f3c60c59
SW		 3485 return true;
3486
3487 mutex_lock(&mdsc->mutex);
3488 for (i = 0; i < mdsc->max_sessions; i++)
3489 if (mdsc->sessions[i])
3490 n++;
3491 mutex_unlock(&mdsc->mutex);
3492 return n == 0;
3493}
2f2dc053
SW		 3494
3495/*
3496 * called after sb is ro.
3497 */
3498void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3499{
3500 struct ceph_mds_session *session;
3501 int i;
3d14c5d2
YS		 3502 struct ceph_fs_client *fsc = mdsc->fsc;
3503 unsigned long timeout = fsc->client->options->mount_timeout * HZ;
2f2dc053
SW		 3504
3505 dout("close_sessions\n");
3506
2f2dc053 3507 /* close sessions */
f3c60c59
SW		 3508 mutex_lock(&mdsc->mutex);
3509 for (i = 0; i < mdsc->max_sessions; i++) {
3510 session = __ceph_lookup_mds_session(mdsc, i);
3511 if (!session)
3512 continue;
2f2dc053 3513 mutex_unlock(&mdsc->mutex);
f3c60c59
SW		 3514 mutex_lock(&session->s_mutex);
3515 __close_session(mdsc, session);
3516 mutex_unlock(&session->s_mutex);
3517 ceph_put_mds_session(session);
2f2dc053
SW		 3518 mutex_lock(&mdsc->mutex);
3519 }
f3c60c59
SW		 3520 mutex_unlock(&mdsc->mutex);
3521
3522 dout("waiting for sessions to close\n");
3523 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3524 timeout);
2f2dc053
SW		 3525
3526 /* tear down remaining sessions */
f3c60c59 3527 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3528 for (i = 0; i < mdsc->max_sessions; i++) {
3529 if (mdsc->sessions[i]) {
3530 session = get_session(mdsc->sessions[i]);
2600d2dd 3531 __unregister_session(mdsc, session);
2f2dc053
SW		 3532 mutex_unlock(&mdsc->mutex);
3533 mutex_lock(&session->s_mutex);
3534 remove_session_caps(session);
3535 mutex_unlock(&session->s_mutex);
3536 ceph_put_mds_session(session);
3537 mutex_lock(&mdsc->mutex);
3538 }
3539 }
2f2dc053 3540 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3541 mutex_unlock(&mdsc->mutex);
3542
3543 ceph_cleanup_empty_realms(mdsc);
3544
3545 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3546
3547 dout("stopped\n");
3548}
3549
3d14c5d2 3550static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3551{
3552 dout("stop\n");
3553 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3554 if (mdsc->mdsmap)
3555 ceph_mdsmap_destroy(mdsc->mdsmap);
3556 kfree(mdsc->sessions);
37151668 3557 ceph_caps_finalize(mdsc);
2f2dc053
SW		 3558}
3559
3d14c5d2
YS		 3560void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3561{
3562 struct ceph_mds_client *mdsc = fsc->mdsc;
3563
ef550f6f 3564 dout("mdsc_destroy %p\n", mdsc);
3d14c5d2 3565 ceph_mdsc_stop(mdsc);
ef550f6f
SW		 3566
3567 /* flush out any connection work with references to us */
3568 ceph_msgr_flush();
3569
3d14c5d2
YS		 3570 fsc->mdsc = NULL;
3571 kfree(mdsc);
ef550f6f 3572 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3573}
3574
2f2dc053
SW		 3575
3576/*
3577 * handle mds map update.
3578 */
3579void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3580{
3581 u32 epoch;
3582 u32 maplen;
3583 void *p = msg->front.iov_base;
3584 void *end = p + msg->front.iov_len;
3585 struct ceph_mdsmap *newmap, *oldmap;
3586 struct ceph_fsid fsid;
3587 int err = -EINVAL;
3588
3589 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3590 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3591 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3592 return;
c89136ea
SW		 3593 epoch = ceph_decode_32(&p);
3594 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3595 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3596
3597 /* do we need it? */
3d14c5d2 3598 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
2f2dc053
SW		 3599 mutex_lock(&mdsc->mutex);
3600 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3601 dout("handle_map epoch %u <= our %u\n",
3602 epoch, mdsc->mdsmap->m_epoch);
3603 mutex_unlock(&mdsc->mutex);
3604 return;
3605 }
3606
3607 newmap = ceph_mdsmap_decode(&p, end);
3608 if (IS_ERR(newmap)) {
3609 err = PTR_ERR(newmap);
3610 goto bad_unlock;
3611 }
3612
3613 /* swap into place */
3614 if (mdsc->mdsmap) {
3615 oldmap = mdsc->mdsmap;
3616 mdsc->mdsmap = newmap;
3617 check_new_map(mdsc, newmap, oldmap);
3618 ceph_mdsmap_destroy(oldmap);
3619 } else {
3620 mdsc->mdsmap = newmap; /* first mds map */
3621 }
3d14c5d2 3622 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3623
3624 __wake_requests(mdsc, &mdsc->waiting_for_map);
3625
3626 mutex_unlock(&mdsc->mutex);
3627 schedule_delayed(mdsc);
3628 return;
3629
3630bad_unlock:
3631 mutex_unlock(&mdsc->mutex);
3632bad:
3633 pr_err("error decoding mdsmap %d\n", err);
3634 return;
3635}
3636
3637static struct ceph_connection *con_get(struct ceph_connection *con)
3638{
3639 struct ceph_mds_session *s = con->private;
3640
3641 if (get_session(s)) {
2600d2dd 3642 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3643 return con;
3644 }
3645 dout("mdsc con_get %p FAIL\n", s);
3646 return NULL;
3647}
3648
3649static void con_put(struct ceph_connection *con)
3650{
3651 struct ceph_mds_session *s = con->private;
3652
7d8e18a6 3653 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
2f2dc053
SW		 3654 ceph_put_mds_session(s);
3655}
3656
3657/*
3658 * if the client is unresponsive for long enough, the mds will kill
3659 * the session entirely.
3660 */
3661static void peer_reset(struct ceph_connection *con)
3662{
3663 struct ceph_mds_session *s = con->private;
7e70f0ed 3664 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3665
f3ae1b97 3666 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3667 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3668}
3669
3670static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3671{
3672 struct ceph_mds_session *s = con->private;
3673 struct ceph_mds_client *mdsc = s->s_mdsc;
3674 int type = le16_to_cpu(msg->hdr.type);
3675
2600d2dd
SW		 3676 mutex_lock(&mdsc->mutex);
3677 if (__verify_registered_session(mdsc, s) < 0) {
3678 mutex_unlock(&mdsc->mutex);
3679 goto out;
3680 }
3681 mutex_unlock(&mdsc->mutex);
3682
2f2dc053
SW		 3683 switch (type) {
3684 case CEPH_MSG_MDS_MAP:
3685 ceph_mdsc_handle_map(mdsc, msg);
3686 break;
3687 case CEPH_MSG_CLIENT_SESSION:
3688 handle_session(s, msg);
3689 break;
3690 case CEPH_MSG_CLIENT_REPLY:
3691 handle_reply(s, msg);
3692 break;
3693 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3694 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3695 break;
3696 case CEPH_MSG_CLIENT_CAPS:
3697 ceph_handle_caps(s, msg);
3698 break;
3699 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3700 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3701 break;
3702 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3703 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3704 break;
3705
3706 default:
3707 pr_err("received unknown message type %d %s\n", type,
3708 ceph_msg_type_name(type));
3709 }
2600d2dd 3710out:
2f2dc053
SW		 3711 ceph_msg_put(msg);
3712}
3713
4e7a5dcd
SW		 3714/*
3715 * authentication
3716 */
a3530df3
AE		 3717
3718/*
3719 * Note: returned pointer is the address of a structure that's
3720 * managed separately. Caller must *not* attempt to free it.
3721 */
3722static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3723 int *proto, int force_new)
4e7a5dcd
SW		 3724{
3725 struct ceph_mds_session *s = con->private;
3726 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3727 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 3728 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 3729
74f1869f 3730 if (force_new && auth->authorizer) {
27859f97 3731 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f 3732 auth->authorizer = NULL;
4e7a5dcd 3733 }
27859f97
SW		 3734 if (!auth->authorizer) {
3735 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3736 auth);
0bed9b5c
SW		 3737 if (ret)
3738 return ERR_PTR(ret);
27859f97
SW		 3739 } else {
3740 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3741 auth);
a255651d 3742 if (ret)
a3530df3 3743 return ERR_PTR(ret);
4e7a5dcd 3744 }
4e7a5dcd 3745 *proto = ac->protocol;
74f1869f 3746
a3530df3 3747 return auth;
4e7a5dcd
SW		 3748}
3749
3750
3751static int verify_authorizer_reply(struct ceph_connection *con, int len)
3752{
3753 struct ceph_mds_session *s = con->private;
3754 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3755 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 3756
27859f97 3757 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
4e7a5dcd
SW		 3758}
3759
9bd2e6f8
SW		 3760static int invalidate_authorizer(struct ceph_connection *con)
3761{
3762 struct ceph_mds_session *s = con->private;
3763 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3764 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 3765
27859f97 3766 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 3767
3d14c5d2 3768 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3769}
3770
53ded495
AE		 3771static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3772 struct ceph_msg_header *hdr, int *skip)
3773{
3774 struct ceph_msg *msg;
3775 int type = (int) le16_to_cpu(hdr->type);
3776 int front_len = (int) le32_to_cpu(hdr->front_len);
3777
3778 if (con->in_msg)
3779 return con->in_msg;
3780
3781 *skip = 0;
3782 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3783 if (!msg) {
3784 pr_err("unable to allocate msg type %d len %d\n",
3785 type, front_len);
3786 return NULL;
3787 }
53ded495
AE		 3788
3789 return msg;
3790}
3791
33d07337
YZ		 3792static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
3793{
3794 struct ceph_mds_session *s = con->private;
3795 struct ceph_auth_handshake *auth = &s->s_auth;
3796 return ceph_auth_sign_message(auth, msg);
3797}
3798
3799static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
3800{
3801 struct ceph_mds_session *s = con->private;
3802 struct ceph_auth_handshake *auth = &s->s_auth;
3803 return ceph_auth_check_message_signature(auth, msg);
3804}
3805
9e32789f 3806static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 3807 .get = con_get,
3808 .put = con_put,
3809 .dispatch = dispatch,
4e7a5dcd
SW		 3810 .get_authorizer = get_authorizer,
3811 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3812 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 3813 .peer_reset = peer_reset,
53ded495 3814 .alloc_msg = mds_alloc_msg,
33d07337
YZ		 3815 .sign_message = sign_message,
3816 .check_message_signature = check_message_signature,
2f2dc053
SW		 3817};
3818
2f2dc053 3819/* eof */
Ubuntu Eoan Linux kernel mirror