projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
libceph: a couple tweaks for wait loops
[linux-2.6-block.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44 * are no longer valid.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 460 INIT_LIST_HEAD(&s->s_cap_releases);
2f2dc053
SW		 461 INIT_LIST_HEAD(&s->s_cap_flushing);
462 INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
463
464 dout("register_session mds%d\n", mds);
465 if (mds >= mdsc->max_sessions) {
466 int newmax = 1 << get_count_order(mds+1);
467 struct ceph_mds_session **sa;
468
469 dout("register_session realloc to %d\n", newmax);
470 sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
471 if (sa == NULL)
42ce56e5 472 goto fail_realloc;
2f2dc053
SW		 473 if (mdsc->sessions) {
474 memcpy(sa, mdsc->sessions,
475 mdsc->max_sessions * sizeof(void *));
476 kfree(mdsc->sessions);
477 }
478 mdsc->sessions = sa;
479 mdsc->max_sessions = newmax;
480 }
481 mdsc->sessions[mds] = s;
86d8f67b 482 atomic_inc(&mdsc->num_sessions);
2f2dc053 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);
86d8f67b 506 atomic_dec(&mdsc->num_sessions);
2f2dc053
SW		 507}
508
509/*
510 * drop session refs in request.
511 *
512 * should be last request ref, or hold mdsc->mutex
513 */
514static void put_request_session(struct ceph_mds_request *req)
515{
516 if (req->r_session) {
517 ceph_put_mds_session(req->r_session);
518 req->r_session = NULL;
519 }
520}
521
153c8e6b 522void ceph_mdsc_release_request(struct kref *kref)
2f2dc053 523{
153c8e6b
SW		 524 struct ceph_mds_request *req = container_of(kref,
525 struct ceph_mds_request,
526 r_kref);
54008399 527 destroy_reply_info(&req->r_reply_info);
153c8e6b
SW		 528 if (req->r_request)
529 ceph_msg_put(req->r_request);
54008399 530 if (req->r_reply)
153c8e6b 531 ceph_msg_put(req->r_reply);
153c8e6b 532 if (req->r_inode) {
41b02e1f 533 ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
153c8e6b
SW		 534 iput(req->r_inode);
535 }
536 if (req->r_locked_dir)
41b02e1f 537 ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
e96a650a 538 iput(req->r_target_inode);
153c8e6b
SW		 539 if (req->r_dentry)
540 dput(req->r_dentry);
844d87c3
SW		 541 if (req->r_old_dentry)
542 dput(req->r_old_dentry);
543 if (req->r_old_dentry_dir) {
41b02e1f
SW		 544 /*
545 * track (and drop pins for) r_old_dentry_dir
546 * separately, since r_old_dentry's d_parent may have
547 * changed between the dir mutex being dropped and
548 * this request being freed.
549 */
550 ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
551 CEPH_CAP_PIN);
41b02e1f 552 iput(req->r_old_dentry_dir);
2f2dc053 553 }
153c8e6b
SW		 554 kfree(req->r_path1);
555 kfree(req->r_path2);
25e6bae3
YZ		 556 if (req->r_pagelist)
557 ceph_pagelist_release(req->r_pagelist);
153c8e6b 558 put_request_session(req);
37151668 559 ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
153c8e6b 560 kfree(req);
2f2dc053
SW		 561}
562
563/*
564 * lookup session, bump ref if found.
565 *
566 * called under mdsc->mutex.
567 */
568static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
569 u64 tid)
570{
571 struct ceph_mds_request *req;
44ca18f2
SW		 572 struct rb_node *n = mdsc->request_tree.rb_node;
573
574 while (n) {
575 req = rb_entry(n, struct ceph_mds_request, r_node);
576 if (tid < req->r_tid)
577 n = n->rb_left;
578 else if (tid > req->r_tid)
579 n = n->rb_right;
580 else {
581 ceph_mdsc_get_request(req);
582 return req;
583 }
584 }
585 return NULL;
586}
587
588static void __insert_request(struct ceph_mds_client *mdsc,
589 struct ceph_mds_request *new)
590{
591 struct rb_node **p = &mdsc->request_tree.rb_node;
592 struct rb_node *parent = NULL;
593 struct ceph_mds_request *req = NULL;
594
595 while (*p) {
596 parent = *p;
597 req = rb_entry(parent, struct ceph_mds_request, r_node);
598 if (new->r_tid < req->r_tid)
599 p = &(*p)->rb_left;
600 else if (new->r_tid > req->r_tid)
601 p = &(*p)->rb_right;
602 else
603 BUG();
604 }
605
606 rb_link_node(&new->r_node, parent, p);
607 rb_insert_color(&new->r_node, &mdsc->request_tree);
2f2dc053
SW		 608}
609
610/*
611 * Register an in-flight request, and assign a tid. Link to directory
612 * are modifying (if any).
613 *
614 * Called under mdsc->mutex.
615 */
616static void __register_request(struct ceph_mds_client *mdsc,
617 struct ceph_mds_request *req,
618 struct inode *dir)
619{
620 req->r_tid = ++mdsc->last_tid;
621 if (req->r_num_caps)
37151668
YS		 622 ceph_reserve_caps(mdsc, &req->r_caps_reservation,
623 req->r_num_caps);
2f2dc053
SW		 624 dout("__register_request %p tid %lld\n", req, req->r_tid);
625 ceph_mdsc_get_request(req);
44ca18f2 626 __insert_request(mdsc, req);
2f2dc053 627
cb4276cc
SW		 628 req->r_uid = current_fsuid();
629 req->r_gid = current_fsgid();
630
e8a7b8b1
YZ		 631 if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
632 mdsc->oldest_tid = req->r_tid;
633
2f2dc053
SW		 634 if (dir) {
635 struct ceph_inode_info *ci = ceph_inode(dir);
636
3b663780 637 ihold(dir);
2f2dc053
SW		 638 spin_lock(&ci->i_unsafe_lock);
639 req->r_unsafe_dir = dir;
640 list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
641 spin_unlock(&ci->i_unsafe_lock);
642 }
643}
644
645static void __unregister_request(struct ceph_mds_client *mdsc,
646 struct ceph_mds_request *req)
647{
648 dout("__unregister_request %p tid %lld\n", req, req->r_tid);
e8a7b8b1
YZ		 649
650 if (req->r_tid == mdsc->oldest_tid) {
651 struct rb_node *p = rb_next(&req->r_node);
652 mdsc->oldest_tid = 0;
653 while (p) {
654 struct ceph_mds_request *next_req =
655 rb_entry(p, struct ceph_mds_request, r_node);
656 if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
657 mdsc->oldest_tid = next_req->r_tid;
658 break;
659 }
660 p = rb_next(p);
661 }
662 }
663
44ca18f2 664 rb_erase(&req->r_node, &mdsc->request_tree);
80fc7314 665 RB_CLEAR_NODE(&req->r_node);
2f2dc053
SW		 666
667 if (req->r_unsafe_dir) {
668 struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
669
670 spin_lock(&ci->i_unsafe_lock);
671 list_del_init(&req->r_unsafe_dir_item);
672 spin_unlock(&ci->i_unsafe_lock);
3b663780
SW		 673
674 iput(req->r_unsafe_dir);
675 req->r_unsafe_dir = NULL;
2f2dc053 676 }
94aa8ae1 677
fc55d2c9
YZ		 678 complete_all(&req->r_safe_completion);
679
94aa8ae1 680 ceph_mdsc_put_request(req);
2f2dc053
SW		 681}
682
683/*
684 * Choose mds to send request to next. If there is a hint set in the
685 * request (e.g., due to a prior forward hint from the mds), use that.
686 * Otherwise, consult frag tree and/or caps to identify the
687 * appropriate mds. If all else fails, choose randomly.
688 *
689 * Called under mdsc->mutex.
690 */
7fd7d101 691static struct dentry *get_nonsnap_parent(struct dentry *dentry)
eb6bb1c5 692{
d79698da
SW		 693 /*
694 * we don't need to worry about protecting the d_parent access
695 * here because we never renaming inside the snapped namespace
696 * except to resplice to another snapdir, and either the old or new
697 * result is a valid result.
698 */
2b0143b5 699 while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
eb6bb1c5
SW		 700 dentry = dentry->d_parent;
701 return dentry;
702}
703
2f2dc053
SW		 704static int __choose_mds(struct ceph_mds_client *mdsc,
705 struct ceph_mds_request *req)
706{
707 struct inode *inode;
708 struct ceph_inode_info *ci;
709 struct ceph_cap *cap;
710 int mode = req->r_direct_mode;
711 int mds = -1;
712 u32 hash = req->r_direct_hash;
713 bool is_hash = req->r_direct_is_hash;
714
715 /*
716 * is there a specific mds we should try? ignore hint if we have
717 * no session and the mds is not up (active or recovering).
718 */
719 if (req->r_resend_mds >= 0 &&
720 (__have_session(mdsc, req->r_resend_mds) ||
721 ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
722 dout("choose_mds using resend_mds mds%d\n",
723 req->r_resend_mds);
724 return req->r_resend_mds;
725 }
726
727 if (mode == USE_RANDOM_MDS)
728 goto random;
729
730 inode = NULL;
731 if (req->r_inode) {
732 inode = req->r_inode;
733 } else if (req->r_dentry) {
d79698da
SW		 734 /* ignore race with rename; old or new d_parent is okay */
735 struct dentry *parent = req->r_dentry->d_parent;
2b0143b5 736 struct inode *dir = d_inode(parent);
eb6bb1c5 737
3d14c5d2 738 if (dir->i_sb != mdsc->fsc->sb) {
eb6bb1c5 739 /* not this fs! */
2b0143b5 740 inode = d_inode(req->r_dentry);
eb6bb1c5
SW		 741 } else if (ceph_snap(dir) != CEPH_NOSNAP) {
742 /* direct snapped/virtual snapdir requests
743 * based on parent dir inode */
d79698da 744 struct dentry *dn = get_nonsnap_parent(parent);
2b0143b5 745 inode = d_inode(dn);
eb6bb1c5 746 dout("__choose_mds using nonsnap parent %p\n", inode);
ca18bede 747 } else {
eb6bb1c5 748 /* dentry target */
2b0143b5 749 inode = d_inode(req->r_dentry);
ca18bede
YZ		 750 if (!inode || mode == USE_AUTH_MDS) {
751 /* dir + name */
752 inode = dir;
753 hash = ceph_dentry_hash(dir, req->r_dentry);
754 is_hash = true;
755 }
2f2dc053
SW		 756 }
757 }
eb6bb1c5 758
2f2dc053
SW		 759 dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
760 (int)hash, mode);
761 if (!inode)
762 goto random;
763 ci = ceph_inode(inode);
764
765 if (is_hash && S_ISDIR(inode->i_mode)) {
766 struct ceph_inode_frag frag;
767 int found;
768
769 ceph_choose_frag(ci, hash, &frag, &found);
770 if (found) {
771 if (mode == USE_ANY_MDS && frag.ndist > 0) {
772 u8 r;
773
774 /* choose a random replica */
775 get_random_bytes(&r, 1);
776 r %= frag.ndist;
777 mds = frag.dist[r];
778 dout("choose_mds %p %llx.%llx "
779 "frag %u mds%d (%d/%d)\n",
780 inode, ceph_vinop(inode),
d66bbd44 781 frag.frag, mds,
2f2dc053 782 (int)r, frag.ndist);
d66bbd44
SW		 783 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
784 CEPH_MDS_STATE_ACTIVE)
785 return mds;
2f2dc053
SW		 786 }
787
788 /* since this file/dir wasn't known to be
789 * replicated, then we want to look for the
790 * authoritative mds. */
791 mode = USE_AUTH_MDS;
792 if (frag.mds >= 0) {
793 /* choose auth mds */
794 mds = frag.mds;
795 dout("choose_mds %p %llx.%llx "
796 "frag %u mds%d (auth)\n",
797 inode, ceph_vinop(inode), frag.frag, mds);
d66bbd44
SW		 798 if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
799 CEPH_MDS_STATE_ACTIVE)
800 return mds;
2f2dc053
SW		 801 }
802 }
803 }
804
be655596 805 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 806 cap = NULL;
807 if (mode == USE_AUTH_MDS)
808 cap = ci->i_auth_cap;
809 if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
810 cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
811 if (!cap) {
be655596 812 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 813 goto random;
814 }
815 mds = cap->session->s_mds;
816 dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
817 inode, ceph_vinop(inode), mds,
818 cap == ci->i_auth_cap ? "auth " : "", cap);
be655596 819 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 820 return mds;
821
822random:
823 mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
824 dout("choose_mds chose random mds%d\n", mds);
825 return mds;
826}
827
828
829/*
830 * session messages
831 */
832static struct ceph_msg *create_session_msg(u32 op, u64 seq)
833{
834 struct ceph_msg *msg;
835 struct ceph_mds_session_head *h;
836
b61c2763
SW		 837 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
838 false);
a79832f2 839 if (!msg) {
2f2dc053 840 pr_err("create_session_msg ENOMEM creating msg\n");
a79832f2 841 return NULL;
2f2dc053
SW		 842 }
843 h = msg->front.iov_base;
844 h->op = cpu_to_le32(op);
845 h->seq = cpu_to_le64(seq);
dbd0c8bf
JS		 846
847 return msg;
848}
849
850/*
851 * session message, specialization for CEPH_SESSION_REQUEST_OPEN
852 * to include additional client metadata fields.
853 */
854static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
855{
856 struct ceph_msg *msg;
857 struct ceph_mds_session_head *h;
858 int i = -1;
859 int metadata_bytes = 0;
860 int metadata_key_count = 0;
861 struct ceph_options *opt = mdsc->fsc->client->options;
862 void *p;
863
a6a5ce4f 864 const char* metadata[][2] = {
dbd0c8bf 865 {"hostname", utsname()->nodename},
a6a5ce4f 866 {"kernel_version", utsname()->release},
dbd0c8bf
JS		 867 {"entity_id", opt->name ? opt->name : ""},
868 {NULL, NULL}
869 };
870
871 /* Calculate serialized length of metadata */
872 metadata_bytes = 4; /* map length */
873 for (i = 0; metadata[i][0] != NULL; ++i) {
874 metadata_bytes += 8 + strlen(metadata[i][0]) +
875 strlen(metadata[i][1]);
876 metadata_key_count++;
877 }
878
879 /* Allocate the message */
880 msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
881 GFP_NOFS, false);
882 if (!msg) {
883 pr_err("create_session_msg ENOMEM creating msg\n");
884 return NULL;
885 }
886 h = msg->front.iov_base;
887 h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
888 h->seq = cpu_to_le64(seq);
889
890 /*
891 * Serialize client metadata into waiting buffer space, using
892 * the format that userspace expects for map<string, string>
7cfa0313
JS		 893 *
894 * ClientSession messages with metadata are v2
dbd0c8bf 895 */
7cfa0313
JS		 896 msg->hdr.version = cpu_to_le16(2);
897 msg->hdr.compat_version = cpu_to_le16(1);
dbd0c8bf
JS		 898
899 /* The write pointer, following the session_head structure */
900 p = msg->front.iov_base + sizeof(*h);
901
902 /* Number of entries in the map */
903 ceph_encode_32(&p, metadata_key_count);
904
905 /* Two length-prefixed strings for each entry in the map */
906 for (i = 0; metadata[i][0] != NULL; ++i) {
907 size_t const key_len = strlen(metadata[i][0]);
908 size_t const val_len = strlen(metadata[i][1]);
909
910 ceph_encode_32(&p, key_len);
911 memcpy(p, metadata[i][0], key_len);
912 p += key_len;
913 ceph_encode_32(&p, val_len);
914 memcpy(p, metadata[i][1], val_len);
915 p += val_len;
916 }
917
2f2dc053
SW		 918 return msg;
919}
920
921/*
922 * send session open request.
923 *
924 * called under mdsc->mutex
925 */
926static int __open_session(struct ceph_mds_client *mdsc,
927 struct ceph_mds_session *session)
928{
929 struct ceph_msg *msg;
930 int mstate;
931 int mds = session->s_mds;
2f2dc053
SW		 932
933 /* wait for mds to go active? */
934 mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
935 dout("open_session to mds%d (%s)\n", mds,
936 ceph_mds_state_name(mstate));
937 session->s_state = CEPH_MDS_SESSION_OPENING;
938 session->s_renew_requested = jiffies;
939
940 /* send connect message */
dbd0c8bf 941 msg = create_session_open_msg(mdsc, session->s_seq);
a79832f2
SW		 942 if (!msg)
943 return -ENOMEM;
2f2dc053 944 ceph_con_send(&session->s_con, msg);
2f2dc053
SW		 945 return 0;
946}
947
ed0552a1
SW		 948/*
949 * open sessions for any export targets for the given mds
950 *
951 * called under mdsc->mutex
952 */
5d72d13c
YZ		 953static struct ceph_mds_session *
954__open_export_target_session(struct ceph_mds_client *mdsc, int target)
955{
956 struct ceph_mds_session *session;
957
958 session = __ceph_lookup_mds_session(mdsc, target);
959 if (!session) {
960 session = register_session(mdsc, target);
961 if (IS_ERR(session))
962 return session;
963 }
964 if (session->s_state == CEPH_MDS_SESSION_NEW ||
965 session->s_state == CEPH_MDS_SESSION_CLOSING)
966 __open_session(mdsc, session);
967
968 return session;
969}
970
971struct ceph_mds_session *
972ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
973{
974 struct ceph_mds_session *session;
975
976 dout("open_export_target_session to mds%d\n", target);
977
978 mutex_lock(&mdsc->mutex);
979 session = __open_export_target_session(mdsc, target);
980 mutex_unlock(&mdsc->mutex);
981
982 return session;
983}
984
ed0552a1
SW		 985static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
986 struct ceph_mds_session *session)
987{
988 struct ceph_mds_info *mi;
989 struct ceph_mds_session *ts;
990 int i, mds = session->s_mds;
ed0552a1
SW		 991
992 if (mds >= mdsc->mdsmap->m_max_mds)
993 return;
5d72d13c 994
ed0552a1
SW		 995 mi = &mdsc->mdsmap->m_info[mds];
996 dout("open_export_target_sessions for mds%d (%d targets)\n",
997 session->s_mds, mi->num_export_targets);
998
999 for (i = 0; i < mi->num_export_targets; i++) {
5d72d13c
YZ		 1000 ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1001 if (!IS_ERR(ts))
1002 ceph_put_mds_session(ts);
ed0552a1
SW		 1003 }
1004}
1005
154f42c2
SW		 1006void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1007 struct ceph_mds_session *session)
1008{
1009 mutex_lock(&mdsc->mutex);
1010 __open_export_target_sessions(mdsc, session);
1011 mutex_unlock(&mdsc->mutex);
1012}
1013
2f2dc053
SW		 1014/*
1015 * session caps
1016 */
1017
745a8e3b
YZ		 1018/* caller holds s_cap_lock, we drop it */
1019static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1020 struct ceph_mds_session *session)
1021 __releases(session->s_cap_lock)
2f2dc053 1022{
745a8e3b
YZ		 1023 LIST_HEAD(tmp_list);
1024 list_splice_init(&session->s_cap_releases, &tmp_list);
1025 session->s_num_cap_releases = 0;
1026 spin_unlock(&session->s_cap_lock);
2f2dc053 1027
745a8e3b
YZ		 1028 dout("cleanup_cap_releases mds%d\n", session->s_mds);
1029 while (!list_empty(&tmp_list)) {
1030 struct ceph_cap *cap;
1031 /* zero out the in-progress message */
1032 cap = list_first_entry(&tmp_list,
1033 struct ceph_cap, session_caps);
1034 list_del(&cap->session_caps);
1035 ceph_put_cap(mdsc, cap);
2f2dc053 1036 }
2f2dc053
SW		 1037}
1038
1c841a96
YZ		 1039static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1040 struct ceph_mds_session *session)
1041{
1042 struct ceph_mds_request *req;
1043 struct rb_node *p;
1044
1045 dout("cleanup_session_requests mds%d\n", session->s_mds);
1046 mutex_lock(&mdsc->mutex);
1047 while (!list_empty(&session->s_unsafe)) {
1048 req = list_first_entry(&session->s_unsafe,
1049 struct ceph_mds_request, r_unsafe_item);
1050 list_del_init(&req->r_unsafe_item);
1051 pr_info(" dropping unsafe request %llu\n", req->r_tid);
1052 __unregister_request(mdsc, req);
1053 }
1054 /* zero r_attempts, so kick_requests() will re-send requests */
1055 p = rb_first(&mdsc->request_tree);
1056 while (p) {
1057 req = rb_entry(p, struct ceph_mds_request, r_node);
1058 p = rb_next(p);
1059 if (req->r_session &&
1060 req->r_session->s_mds == session->s_mds)
1061 req->r_attempts = 0;
1062 }
1063 mutex_unlock(&mdsc->mutex);
1064}
1065
2f2dc053 1066/*
f818a736
SW		 1067 * Helper to safely iterate over all caps associated with a session, with
1068 * special care taken to handle a racing __ceph_remove_cap().
2f2dc053 1069 *
f818a736 1070 * Caller must hold session s_mutex.
2f2dc053
SW		 1071 */
1072static int iterate_session_caps(struct ceph_mds_session *session,
1073 int (*cb)(struct inode *, struct ceph_cap *,
1074 void *), void *arg)
1075{
7c1332b8
SW		 1076 struct list_head *p;
1077 struct ceph_cap *cap;
1078 struct inode *inode, *last_inode = NULL;
1079 struct ceph_cap *old_cap = NULL;
2f2dc053
SW		 1080 int ret;
1081
1082 dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1083 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1084 p = session->s_caps.next;
1085 while (p != &session->s_caps) {
1086 cap = list_entry(p, struct ceph_cap, session_caps);
2f2dc053 1087 inode = igrab(&cap->ci->vfs_inode);
7c1332b8
SW		 1088 if (!inode) {
1089 p = p->next;
2f2dc053 1090 continue;
7c1332b8
SW		 1091 }
1092 session->s_cap_iterator = cap;
2f2dc053 1093 spin_unlock(&session->s_cap_lock);
7c1332b8
SW		 1094
1095 if (last_inode) {
1096 iput(last_inode);
1097 last_inode = NULL;
1098 }
1099 if (old_cap) {
37151668 1100 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8
SW		 1101 old_cap = NULL;
1102 }
1103
2f2dc053 1104 ret = cb(inode, cap, arg);
7c1332b8
SW		 1105 last_inode = inode;
1106
2f2dc053 1107 spin_lock(&session->s_cap_lock);
7c1332b8
SW		 1108 p = p->next;
1109 if (cap->ci == NULL) {
1110 dout("iterate_session_caps finishing cap %p removal\n",
1111 cap);
1112 BUG_ON(cap->session != session);
745a8e3b 1113 cap->session = NULL;
7c1332b8
SW		 1114 list_del_init(&cap->session_caps);
1115 session->s_nr_caps--;
745a8e3b
YZ		 1116 if (cap->queue_release) {
1117 list_add_tail(&cap->session_caps,
1118 &session->s_cap_releases);
1119 session->s_num_cap_releases++;
1120 } else {
1121 old_cap = cap; /* put_cap it w/o locks held */
1122 }
7c1332b8 1123 }
5dacf091
SW		 1124 if (ret < 0)
1125 goto out;
2f2dc053 1126 }
5dacf091
SW		 1127 ret = 0;
1128out:
7c1332b8 1129 session->s_cap_iterator = NULL;
2f2dc053 1130 spin_unlock(&session->s_cap_lock);
7c1332b8 1131
e96a650a 1132 iput(last_inode);
7c1332b8 1133 if (old_cap)
37151668 1134 ceph_put_cap(session->s_mdsc, old_cap);
7c1332b8 1135
5dacf091 1136 return ret;
2f2dc053
SW		 1137}
1138
1139static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
6c99f254 1140 void *arg)
2f2dc053
SW		 1141{
1142 struct ceph_inode_info *ci = ceph_inode(inode);
6c99f254
SW		 1143 int drop = 0;
1144
2f2dc053
SW		 1145 dout("removing cap %p, ci is %p, inode is %p\n",
1146 cap, ci, &ci->vfs_inode);
be655596 1147 spin_lock(&ci->i_ceph_lock);
a096b09a 1148 __ceph_remove_cap(cap, false);
571ade33 1149 if (!ci->i_auth_cap) {
6c99f254 1150 struct ceph_mds_client *mdsc =
3d14c5d2 1151 ceph_sb_to_client(inode->i_sb)->mdsc;
6c99f254
SW		 1152
1153 spin_lock(&mdsc->cap_dirty_lock);
1154 if (!list_empty(&ci->i_dirty_item)) {
1155 pr_info(" dropping dirty %s state for %p %lld\n",
1156 ceph_cap_string(ci->i_dirty_caps),
1157 inode, ceph_ino(inode));
1158 ci->i_dirty_caps = 0;
1159 list_del_init(&ci->i_dirty_item);
1160 drop = 1;
1161 }
1162 if (!list_empty(&ci->i_flushing_item)) {
1163 pr_info(" dropping dirty+flushing %s state for %p %lld\n",
1164 ceph_cap_string(ci->i_flushing_caps),
1165 inode, ceph_ino(inode));
1166 ci->i_flushing_caps = 0;
1167 list_del_init(&ci->i_flushing_item);
1168 mdsc->num_cap_flushing--;
1169 drop = 1;
1170 }
6c99f254
SW		 1171 spin_unlock(&mdsc->cap_dirty_lock);
1172 }
be655596 1173 spin_unlock(&ci->i_ceph_lock);
6c99f254
SW		 1174 while (drop--)
1175 iput(inode);
2f2dc053
SW		 1176 return 0;
1177}
1178
1179/*
1180 * caller must hold session s_mutex
1181 */
1182static void remove_session_caps(struct ceph_mds_session *session)
1183{
1184 dout("remove_session_caps on %p\n", session);
1185 iterate_session_caps(session, remove_session_caps_cb, NULL);
6f60f889
YZ		 1186
1187 spin_lock(&session->s_cap_lock);
1188 if (session->s_nr_caps > 0) {
1189 struct super_block *sb = session->s_mdsc->fsc->sb;
1190 struct inode *inode;
1191 struct ceph_cap *cap, *prev = NULL;
1192 struct ceph_vino vino;
1193 /*
1194 * iterate_session_caps() skips inodes that are being
1195 * deleted, we need to wait until deletions are complete.
1196 * __wait_on_freeing_inode() is designed for the job,
1197 * but it is not exported, so use lookup inode function
1198 * to access it.
1199 */
1200 while (!list_empty(&session->s_caps)) {
1201 cap = list_entry(session->s_caps.next,
1202 struct ceph_cap, session_caps);
1203 if (cap == prev)
1204 break;
1205 prev = cap;
1206 vino = cap->ci->i_vino;
1207 spin_unlock(&session->s_cap_lock);
1208
ed284c49 1209 inode = ceph_find_inode(sb, vino);
6f60f889
YZ		 1210 iput(inode);
1211
1212 spin_lock(&session->s_cap_lock);
1213 }
1214 }
745a8e3b
YZ		 1215
1216 // drop cap expires and unlock s_cap_lock
1217 cleanup_cap_releases(session->s_mdsc, session);
6f60f889 1218
2f2dc053 1219 BUG_ON(session->s_nr_caps > 0);
6c99f254 1220 BUG_ON(!list_empty(&session->s_cap_flushing));
2f2dc053
SW		 1221}
1222
1223/*
1224 * wake up any threads waiting on this session's caps. if the cap is
1225 * old (didn't get renewed on the client reconnect), remove it now.
1226 *
1227 * caller must hold s_mutex.
1228 */
1229static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1230 void *arg)
1231{
0dc2570f
SW		 1232 struct ceph_inode_info *ci = ceph_inode(inode);
1233
03066f23 1234 wake_up_all(&ci->i_cap_wq);
0dc2570f 1235 if (arg) {
be655596 1236 spin_lock(&ci->i_ceph_lock);
0dc2570f
SW		 1237 ci->i_wanted_max_size = 0;
1238 ci->i_requested_max_size = 0;
be655596 1239 spin_unlock(&ci->i_ceph_lock);
0dc2570f 1240 }
2f2dc053
SW		 1241 return 0;
1242}
1243
0dc2570f
SW		 1244static void wake_up_session_caps(struct ceph_mds_session *session,
1245 int reconnect)
2f2dc053
SW		 1246{
1247 dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
0dc2570f
SW		 1248 iterate_session_caps(session, wake_up_session_cb,
1249 (void *)(unsigned long)reconnect);
2f2dc053
SW		 1250}
1251
1252/*
1253 * Send periodic message to MDS renewing all currently held caps. The
1254 * ack will reset the expiration for all caps from this session.
1255 *
1256 * caller holds s_mutex
1257 */
1258static int send_renew_caps(struct ceph_mds_client *mdsc,
1259 struct ceph_mds_session *session)
1260{
1261 struct ceph_msg *msg;
1262 int state;
1263
1264 if (time_after_eq(jiffies, session->s_cap_ttl) &&
1265 time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1266 pr_info("mds%d caps stale\n", session->s_mds);
e4cb4cb8 1267 session->s_renew_requested = jiffies;
2f2dc053
SW		 1268
1269 /* do not try to renew caps until a recovering mds has reconnected
1270 * with its clients. */
1271 state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1272 if (state < CEPH_MDS_STATE_RECONNECT) {
1273 dout("send_renew_caps ignoring mds%d (%s)\n",
1274 session->s_mds, ceph_mds_state_name(state));
1275 return 0;
1276 }
1277
1278 dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1279 ceph_mds_state_name(state));
2f2dc053
SW		 1280 msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1281 ++session->s_renew_seq);
a79832f2
SW		 1282 if (!msg)
1283 return -ENOMEM;
2f2dc053
SW		 1284 ceph_con_send(&session->s_con, msg);
1285 return 0;
1286}
1287
186e4f7a
YZ		 1288static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1289 struct ceph_mds_session *session, u64 seq)
1290{
1291 struct ceph_msg *msg;
1292
1293 dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
a687ecaf 1294 session->s_mds, ceph_session_state_name(session->s_state), seq);
186e4f7a
YZ		 1295 msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1296 if (!msg)
1297 return -ENOMEM;
1298 ceph_con_send(&session->s_con, msg);
1299 return 0;
1300}
1301
1302
2f2dc053
SW		 1303/*
1304 * Note new cap ttl, and any transition from stale -> not stale (fresh?).
0dc2570f
SW		 1305 *
1306 * Called under session->s_mutex
2f2dc053
SW		 1307 */
1308static void renewed_caps(struct ceph_mds_client *mdsc,
1309 struct ceph_mds_session *session, int is_renew)
1310{
1311 int was_stale;
1312 int wake = 0;
1313
1314 spin_lock(&session->s_cap_lock);
1ce208a6 1315 was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
2f2dc053
SW		 1316
1317 session->s_cap_ttl = session->s_renew_requested +
1318 mdsc->mdsmap->m_session_timeout*HZ;
1319
1320 if (was_stale) {
1321 if (time_before(jiffies, session->s_cap_ttl)) {
1322 pr_info("mds%d caps renewed\n", session->s_mds);
1323 wake = 1;
1324 } else {
1325 pr_info("mds%d caps still stale\n", session->s_mds);
1326 }
1327 }
1328 dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1329 session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1330 time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1331 spin_unlock(&session->s_cap_lock);
1332
1333 if (wake)
0dc2570f 1334 wake_up_session_caps(session, 0);
2f2dc053
SW		 1335}
1336
1337/*
1338 * send a session close request
1339 */
1340static int request_close_session(struct ceph_mds_client *mdsc,
1341 struct ceph_mds_session *session)
1342{
1343 struct ceph_msg *msg;
2f2dc053
SW		 1344
1345 dout("request_close_session mds%d state %s seq %lld\n",
a687ecaf 1346 session->s_mds, ceph_session_state_name(session->s_state),
2f2dc053
SW		 1347 session->s_seq);
1348 msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
a79832f2
SW		 1349 if (!msg)
1350 return -ENOMEM;
1351 ceph_con_send(&session->s_con, msg);
1352 return 0;
2f2dc053
SW		 1353}
1354
1355/*
1356 * Called with s_mutex held.
1357 */
1358static int __close_session(struct ceph_mds_client *mdsc,
1359 struct ceph_mds_session *session)
1360{
1361 if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1362 return 0;
1363 session->s_state = CEPH_MDS_SESSION_CLOSING;
1364 return request_close_session(mdsc, session);
1365}
1366
1367/*
1368 * Trim old(er) caps.
1369 *
1370 * Because we can't cache an inode without one or more caps, we do
1371 * this indirectly: if a cap is unused, we prune its aliases, at which
1372 * point the inode will hopefully get dropped to.
1373 *
1374 * Yes, this is a bit sloppy. Our only real goal here is to respond to
1375 * memory pressure from the MDS, though, so it needn't be perfect.
1376 */
1377static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1378{
1379 struct ceph_mds_session *session = arg;
1380 struct ceph_inode_info *ci = ceph_inode(inode);
979abfdd 1381 int used, wanted, oissued, mine;
2f2dc053
SW		 1382
1383 if (session->s_trim_caps <= 0)
1384 return -1;
1385
be655596 1386 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 1387 mine = cap->issued | cap->implemented;
1388 used = __ceph_caps_used(ci);
979abfdd 1389 wanted = __ceph_caps_file_wanted(ci);
2f2dc053
SW		 1390 oissued = __ceph_caps_issued_other(ci, cap);
1391
979abfdd 1392 dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
2f2dc053 1393 inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
979abfdd
YZ		 1394 ceph_cap_string(used), ceph_cap_string(wanted));
1395 if (cap == ci->i_auth_cap) {
622f3e25
YZ		 1396 if (ci->i_dirty_caps || ci->i_flushing_caps ||
1397 !list_empty(&ci->i_cap_snaps))
979abfdd
YZ		 1398 goto out;
1399 if ((used | wanted) & CEPH_CAP_ANY_WR)
1400 goto out;
1401 }
1402 if ((used | wanted) & ~oissued & mine)
2f2dc053
SW		 1403 goto out; /* we need these caps */
1404
1405 session->s_trim_caps--;
1406 if (oissued) {
1407 /* we aren't the only cap.. just remove us */
a096b09a 1408 __ceph_remove_cap(cap, true);
2f2dc053
SW		 1409 } else {
1410 /* try to drop referring dentries */
be655596 1411 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1412 d_prune_aliases(inode);
1413 dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1414 inode, cap, atomic_read(&inode->i_count));
1415 return 0;
1416 }
1417
1418out:
be655596 1419 spin_unlock(&ci->i_ceph_lock);
2f2dc053
SW		 1420 return 0;
1421}
1422
1423/*
1424 * Trim session cap count down to some max number.
1425 */
1426static int trim_caps(struct ceph_mds_client *mdsc,
1427 struct ceph_mds_session *session,
1428 int max_caps)
1429{
1430 int trim_caps = session->s_nr_caps - max_caps;
1431
1432 dout("trim_caps mds%d start: %d / %d, trim %d\n",
1433 session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1434 if (trim_caps > 0) {
1435 session->s_trim_caps = trim_caps;
1436 iterate_session_caps(session, trim_caps_cb, session);
1437 dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1438 session->s_mds, session->s_nr_caps, max_caps,
1439 trim_caps - session->s_trim_caps);
5dacf091 1440 session->s_trim_caps = 0;
2f2dc053 1441 }
a56371d9 1442
a56371d9 1443 ceph_send_cap_releases(mdsc, session);
2f2dc053
SW		 1444 return 0;
1445}
1446
affbc19a
YZ		 1447static int check_cap_flush(struct ceph_inode_info *ci,
1448 u64 want_flush_seq, u64 want_snap_seq)
d3383a8e 1449{
affbc19a 1450 int ret1 = 1, ret2 = 1;
d3383a8e 1451 spin_lock(&ci->i_ceph_lock);
affbc19a
YZ		 1452 if (want_flush_seq > 0 && ci->i_flushing_caps)
1453 ret1 = ci->i_cap_flush_seq >= want_flush_seq;
1454
1455 if (want_snap_seq > 0 && !list_empty(&ci->i_cap_snaps)) {
1456 struct ceph_cap_snap *capsnap =
1457 list_first_entry(&ci->i_cap_snaps,
1458 struct ceph_cap_snap, ci_item);
1459 ret2 = capsnap->follows >= want_snap_seq;
1460 }
d3383a8e 1461 spin_unlock(&ci->i_ceph_lock);
affbc19a 1462 return ret1 && ret2;
d3383a8e
YZ		 1463}
1464
2f2dc053
SW		 1465/*
1466 * flush all dirty inode data to disk.
1467 *
1468 * returns true if we've flushed through want_flush_seq
1469 */
affbc19a
YZ		 1470static void wait_caps_flush(struct ceph_mds_client *mdsc,
1471 u64 want_flush_seq, u64 want_snap_seq)
2f2dc053 1472{
d3383a8e 1473 int mds;
2f2dc053
SW		 1474
1475 dout("check_cap_flush want %lld\n", want_flush_seq);
1476 mutex_lock(&mdsc->mutex);
affbc19a 1477 for (mds = 0; mds < mdsc->max_sessions; ) {
2f2dc053 1478 struct ceph_mds_session *session = mdsc->sessions[mds];
affbc19a 1479 struct inode *inode1 = NULL, *inode2 = NULL;
2f2dc053 1480
affbc19a
YZ		 1481 if (!session) {
1482 mds++;
2f2dc053 1483 continue;
affbc19a 1484 }
2f2dc053
SW		 1485 get_session(session);
1486 mutex_unlock(&mdsc->mutex);
1487
1488 mutex_lock(&session->s_mutex);
1489 if (!list_empty(&session->s_cap_flushing)) {
1490 struct ceph_inode_info *ci =
affbc19a
YZ		 1491 list_first_entry(&session->s_cap_flushing,
1492 struct ceph_inode_info,
1493 i_flushing_item);
2f2dc053 1494
affbc19a 1495 if (!check_cap_flush(ci, want_flush_seq, 0)) {
2f2dc053 1496 dout("check_cap_flush still flushing %p "
d3383a8e
YZ		 1497 "seq %lld <= %lld to mds%d\n",
1498 &ci->vfs_inode, ci->i_cap_flush_seq,
affbc19a
YZ		 1499 want_flush_seq, mds);
1500 inode1 = igrab(&ci->vfs_inode);
1501 }
1502 }
1503 if (!list_empty(&session->s_cap_snaps_flushing)) {
1504 struct ceph_cap_snap *capsnap =
1505 list_first_entry(&session->s_cap_snaps_flushing,
1506 struct ceph_cap_snap,
1507 flushing_item);
1508 struct ceph_inode_info *ci = capsnap->ci;
1509 if (!check_cap_flush(ci, 0, want_snap_seq)) {
1510 dout("check_cap_flush still flushing snap %p "
1511 "follows %lld <= %lld to mds%d\n",
1512 &ci->vfs_inode, capsnap->follows,
1513 want_snap_seq, mds);
1514 inode2 = igrab(&ci->vfs_inode);
2f2dc053 1515 }
2f2dc053
SW		 1516 }
1517 mutex_unlock(&session->s_mutex);
1518 ceph_put_mds_session(session);
1519
affbc19a 1520 if (inode1) {
d3383a8e 1521 wait_event(mdsc->cap_flushing_wq,
affbc19a
YZ		 1522 check_cap_flush(ceph_inode(inode1),
1523 want_flush_seq, 0));
1524 iput(inode1);
1525 }
1526 if (inode2) {
1527 wait_event(mdsc->cap_flushing_wq,
1528 check_cap_flush(ceph_inode(inode2),
1529 0, want_snap_seq));
1530 iput(inode2);
d3383a8e
YZ		 1531 }
1532
affbc19a
YZ		 1533 if (!inode1 && !inode2)
1534 mds++;
1535
2f2dc053
SW		 1536 mutex_lock(&mdsc->mutex);
1537 }
1538
1539 mutex_unlock(&mdsc->mutex);
1540 dout("check_cap_flush ok, flushed thru %lld\n", want_flush_seq);
2f2dc053
SW		 1541}
1542
1543/*
1544 * called under s_mutex
1545 */
3d7ded4d
SW		 1546void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1547 struct ceph_mds_session *session)
2f2dc053 1548{
745a8e3b
YZ		 1549 struct ceph_msg *msg = NULL;
1550 struct ceph_mds_cap_release *head;
1551 struct ceph_mds_cap_item *item;
1552 struct ceph_cap *cap;
1553 LIST_HEAD(tmp_list);
1554 int num_cap_releases;
2f2dc053 1555
0f8605f2 1556 spin_lock(&session->s_cap_lock);
745a8e3b
YZ		 1557again:
1558 list_splice_init(&session->s_cap_releases, &tmp_list);
1559 num_cap_releases = session->s_num_cap_releases;
1560 session->s_num_cap_releases = 0;
2f2dc053 1561 spin_unlock(&session->s_cap_lock);
e01a5946 1562
745a8e3b
YZ		 1563 while (!list_empty(&tmp_list)) {
1564 if (!msg) {
1565 msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1566 PAGE_CACHE_SIZE, GFP_NOFS, false);
1567 if (!msg)
1568 goto out_err;
1569 head = msg->front.iov_base;
1570 head->num = cpu_to_le32(0);
1571 msg->front.iov_len = sizeof(*head);
1572 }
1573 cap = list_first_entry(&tmp_list, struct ceph_cap,
1574 session_caps);
1575 list_del(&cap->session_caps);
1576 num_cap_releases--;
e01a5946 1577
00bd8edb 1578 head = msg->front.iov_base;
745a8e3b
YZ		 1579 le32_add_cpu(&head->num, 1);
1580 item = msg->front.iov_base + msg->front.iov_len;
1581 item->ino = cpu_to_le64(cap->cap_ino);
1582 item->cap_id = cpu_to_le64(cap->cap_id);
1583 item->migrate_seq = cpu_to_le32(cap->mseq);
1584 item->seq = cpu_to_le32(cap->issue_seq);
1585 msg->front.iov_len += sizeof(*item);
1586
1587 ceph_put_cap(mdsc, cap);
1588
1589 if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1590 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1591 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1592 ceph_con_send(&session->s_con, msg);
1593 msg = NULL;
1594 }
00bd8edb 1595 }
e01a5946 1596
745a8e3b 1597 BUG_ON(num_cap_releases != 0);
e01a5946 1598
745a8e3b
YZ		 1599 spin_lock(&session->s_cap_lock);
1600 if (!list_empty(&session->s_cap_releases))
1601 goto again;
1602 spin_unlock(&session->s_cap_lock);
1603
1604 if (msg) {
1605 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1606 dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1607 ceph_con_send(&session->s_con, msg);
e01a5946 1608 }
745a8e3b
YZ		 1609 return;
1610out_err:
1611 pr_err("send_cap_releases mds%d, failed to allocate message\n",
1612 session->s_mds);
1613 spin_lock(&session->s_cap_lock);
1614 list_splice(&tmp_list, &session->s_cap_releases);
1615 session->s_num_cap_releases += num_cap_releases;
1616 spin_unlock(&session->s_cap_lock);
e01a5946
SW		 1617}
1618
2f2dc053
SW		 1619/*
1620 * requests
1621 */
1622
54008399
YZ		 1623int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1624 struct inode *dir)
1625{
1626 struct ceph_inode_info *ci = ceph_inode(dir);
1627 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1628 struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1629 size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1630 sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1631 int order, num_entries;
1632
1633 spin_lock(&ci->i_ceph_lock);
1634 num_entries = ci->i_files + ci->i_subdirs;
1635 spin_unlock(&ci->i_ceph_lock);
1636 num_entries = max(num_entries, 1);
1637 num_entries = min(num_entries, opt->max_readdir);
1638
1639 order = get_order(size * num_entries);
1640 while (order >= 0) {
1641 rinfo->dir_in = (void*)__get_free_pages(GFP_NOFS | __GFP_NOWARN,
1642 order);
1643 if (rinfo->dir_in)
1644 break;
1645 order--;
1646 }
1647 if (!rinfo->dir_in)
1648 return -ENOMEM;
1649
1650 num_entries = (PAGE_SIZE << order) / size;
1651 num_entries = min(num_entries, opt->max_readdir);
1652
1653 rinfo->dir_buf_size = PAGE_SIZE << order;
1654 req->r_num_caps = num_entries + 1;
1655 req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1656 req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1657 return 0;
1658}
1659
2f2dc053
SW		 1660/*
1661 * Create an mds request.
1662 */
1663struct ceph_mds_request *
1664ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1665{
1666 struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1667
1668 if (!req)
1669 return ERR_PTR(-ENOMEM);
1670
b4556396 1671 mutex_init(&req->r_fill_mutex);
37151668 1672 req->r_mdsc = mdsc;
2f2dc053
SW		 1673 req->r_started = jiffies;
1674 req->r_resend_mds = -1;
1675 INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1676 req->r_fmode = -1;
153c8e6b 1677 kref_init(&req->r_kref);
2f2dc053
SW		 1678 INIT_LIST_HEAD(&req->r_wait);
1679 init_completion(&req->r_completion);
1680 init_completion(&req->r_safe_completion);
1681 INIT_LIST_HEAD(&req->r_unsafe_item);
1682
b8e69066
SW		 1683 req->r_stamp = CURRENT_TIME;
1684
2f2dc053
SW		 1685 req->r_op = op;
1686 req->r_direct_mode = mode;
1687 return req;
1688}
1689
1690/*
44ca18f2 1691 * return oldest (lowest) request, tid in request tree, 0 if none.
2f2dc053
SW		 1692 *
1693 * called under mdsc->mutex.
1694 */
44ca18f2
SW		 1695static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1696{
1697 if (RB_EMPTY_ROOT(&mdsc->request_tree))
1698 return NULL;
1699 return rb_entry(rb_first(&mdsc->request_tree),
1700 struct ceph_mds_request, r_node);
1701}
1702
e8a7b8b1 1703static inline u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2f2dc053 1704{
e8a7b8b1 1705 return mdsc->oldest_tid;
2f2dc053
SW		 1706}
1707
1708/*
1709 * Build a dentry's path. Allocate on heap; caller must kfree. Based
1710 * on build_path_from_dentry in fs/cifs/dir.c.
1711 *
1712 * If @stop_on_nosnap, generate path relative to the first non-snapped
1713 * inode.
1714 *
1715 * Encode hidden .snap dirs as a double /, i.e.
1716 * foo/.snap/bar -> foo//bar
1717 */
1718char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1719 int stop_on_nosnap)
1720{
1721 struct dentry *temp;
1722 char *path;
1723 int len, pos;
1b71fe2e 1724 unsigned seq;
2f2dc053
SW		 1725
1726 if (dentry == NULL)
1727 return ERR_PTR(-EINVAL);
1728
1729retry:
1730 len = 0;
1b71fe2e
AV		 1731 seq = read_seqbegin(&rename_lock);
1732 rcu_read_lock();
2f2dc053 1733 for (temp = dentry; !IS_ROOT(temp);) {
2b0143b5 1734 struct inode *inode = d_inode(temp);
2f2dc053
SW		 1735 if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1736 len++; /* slash only */
1737 else if (stop_on_nosnap && inode &&
1738 ceph_snap(inode) == CEPH_NOSNAP)
1739 break;
1740 else
1741 len += 1 + temp->d_name.len;
1742 temp = temp->d_parent;
2f2dc053 1743 }
1b71fe2e 1744 rcu_read_unlock();
2f2dc053
SW		 1745 if (len)
1746 len--; /* no leading '/' */
1747
1748 path = kmalloc(len+1, GFP_NOFS);
1749 if (path == NULL)
1750 return ERR_PTR(-ENOMEM);
1751 pos = len;
1752 path[pos] = 0; /* trailing null */
1b71fe2e 1753 rcu_read_lock();
2f2dc053 1754 for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1b71fe2e 1755 struct inode *inode;
2f2dc053 1756
1b71fe2e 1757 spin_lock(&temp->d_lock);
2b0143b5 1758 inode = d_inode(temp);
2f2dc053 1759 if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
104648ad 1760 dout("build_path path+%d: %p SNAPDIR\n",
2f2dc053
SW		 1761 pos, temp);
1762 } else if (stop_on_nosnap && inode &&
1763 ceph_snap(inode) == CEPH_NOSNAP) {
9d5a09e6 1764 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1765 break;
1766 } else {
1767 pos -= temp->d_name.len;
1b71fe2e
AV		 1768 if (pos < 0) {
1769 spin_unlock(&temp->d_lock);
2f2dc053 1770 break;
1b71fe2e 1771 }
2f2dc053
SW		 1772 strncpy(path + pos, temp->d_name.name,
1773 temp->d_name.len);
2f2dc053 1774 }
1b71fe2e 1775 spin_unlock(&temp->d_lock);
2f2dc053
SW		 1776 if (pos)
1777 path[--pos] = '/';
1778 temp = temp->d_parent;
2f2dc053 1779 }
1b71fe2e
AV		 1780 rcu_read_unlock();
1781 if (pos != 0 || read_seqretry(&rename_lock, seq)) {
104648ad 1782 pr_err("build_path did not end path lookup where "
2f2dc053
SW		 1783 "expected, namelen is %d, pos is %d\n", len, pos);
1784 /* presumably this is only possible if racing with a
1785 rename of one of the parent directories (we can not
1786 lock the dentries above us to prevent this, but
1787 retrying should be harmless) */
1788 kfree(path);
1789 goto retry;
1790 }
1791
2b0143b5 1792 *base = ceph_ino(d_inode(temp));
2f2dc053 1793 *plen = len;
104648ad 1794 dout("build_path on %p %d built %llx '%.*s'\n",
84d08fa8 1795 dentry, d_count(dentry), *base, len, path);
2f2dc053
SW		 1796 return path;
1797}
1798
1799static int build_dentry_path(struct dentry *dentry,
1800 const char **ppath, int *ppathlen, u64 *pino,
1801 int *pfreepath)
1802{
1803 char *path;
1804
2b0143b5
DH		 1805 if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1806 *pino = ceph_ino(d_inode(dentry->d_parent));
2f2dc053
SW		 1807 *ppath = dentry->d_name.name;
1808 *ppathlen = dentry->d_name.len;
1809 return 0;
1810 }
1811 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1812 if (IS_ERR(path))
1813 return PTR_ERR(path);
1814 *ppath = path;
1815 *pfreepath = 1;
1816 return 0;
1817}
1818
1819static int build_inode_path(struct inode *inode,
1820 const char **ppath, int *ppathlen, u64 *pino,
1821 int *pfreepath)
1822{
1823 struct dentry *dentry;
1824 char *path;
1825
1826 if (ceph_snap(inode) == CEPH_NOSNAP) {
1827 *pino = ceph_ino(inode);
1828 *ppathlen = 0;
1829 return 0;
1830 }
1831 dentry = d_find_alias(inode);
1832 path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1833 dput(dentry);
1834 if (IS_ERR(path))
1835 return PTR_ERR(path);
1836 *ppath = path;
1837 *pfreepath = 1;
1838 return 0;
1839}
1840
1841/*
1842 * request arguments may be specified via an inode *, a dentry *, or
1843 * an explicit ino+path.
1844 */
1845static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1846 const char *rpath, u64 rino,
1847 const char **ppath, int *pathlen,
1848 u64 *ino, int *freepath)
1849{
1850 int r = 0;
1851
1852 if (rinode) {
1853 r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1854 dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1855 ceph_snap(rinode));
1856 } else if (rdentry) {
1857 r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1858 dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1859 *ppath);
795858db 1860 } else if (rpath || rino) {
2f2dc053
SW		 1861 *ino = rino;
1862 *ppath = rpath;
b000056a 1863 *pathlen = rpath ? strlen(rpath) : 0;
2f2dc053
SW		 1864 dout(" path %.*s\n", *pathlen, rpath);
1865 }
1866
1867 return r;
1868}
1869
1870/*
1871 * called under mdsc->mutex
1872 */
1873static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1874 struct ceph_mds_request *req,
6e6f0923 1875 int mds, bool drop_cap_releases)
2f2dc053
SW		 1876{
1877 struct ceph_msg *msg;
1878 struct ceph_mds_request_head *head;
1879 const char *path1 = NULL;
1880 const char *path2 = NULL;
1881 u64 ino1 = 0, ino2 = 0;
1882 int pathlen1 = 0, pathlen2 = 0;
1883 int freepath1 = 0, freepath2 = 0;
1884 int len;
1885 u16 releases;
1886 void *p, *end;
1887 int ret;
1888
1889 ret = set_request_path_attr(req->r_inode, req->r_dentry,
1890 req->r_path1, req->r_ino1.ino,
1891 &path1, &pathlen1, &ino1, &freepath1);
1892 if (ret < 0) {
1893 msg = ERR_PTR(ret);
1894 goto out;
1895 }
1896
1897 ret = set_request_path_attr(NULL, req->r_old_dentry,
1898 req->r_path2, req->r_ino2.ino,
1899 &path2, &pathlen2, &ino2, &freepath2);
1900 if (ret < 0) {
1901 msg = ERR_PTR(ret);
1902 goto out_free1;
1903 }
1904
1905 len = sizeof(*head) +
b8e69066
SW		 1906 pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
1907 sizeof(struct timespec);
2f2dc053
SW		 1908
1909 /* calculate (max) length for cap releases */
1910 len += sizeof(struct ceph_mds_request_release) *
1911 (!!req->r_inode_drop + !!req->r_dentry_drop +
1912 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1913 if (req->r_dentry_drop)
1914 len += req->r_dentry->d_name.len;
1915 if (req->r_old_dentry_drop)
1916 len += req->r_old_dentry->d_name.len;
1917
b61c2763 1918 msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
a79832f2
SW		 1919 if (!msg) {
1920 msg = ERR_PTR(-ENOMEM);
2f2dc053 1921 goto out_free2;
a79832f2 1922 }
2f2dc053 1923
7cfa0313 1924 msg->hdr.version = cpu_to_le16(2);
6df058c0
SW		 1925 msg->hdr.tid = cpu_to_le64(req->r_tid);
1926
2f2dc053
SW		 1927 head = msg->front.iov_base;
1928 p = msg->front.iov_base + sizeof(*head);
1929 end = msg->front.iov_base + msg->front.iov_len;
1930
1931 head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1932 head->op = cpu_to_le32(req->r_op);
ff3d0046
EB		 1933 head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1934 head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2f2dc053
SW		 1935 head->args = req->r_args;
1936
1937 ceph_encode_filepath(&p, end, ino1, path1);
1938 ceph_encode_filepath(&p, end, ino2, path2);
1939
e979cf50
SW		 1940 /* make note of release offset, in case we need to replay */
1941 req->r_request_release_offset = p - msg->front.iov_base;
1942
2f2dc053
SW		 1943 /* cap releases */
1944 releases = 0;
1945 if (req->r_inode_drop)
1946 releases += ceph_encode_inode_release(&p,
2b0143b5 1947 req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2f2dc053
SW		 1948 mds, req->r_inode_drop, req->r_inode_unless, 0);
1949 if (req->r_dentry_drop)
1950 releases += ceph_encode_dentry_release(&p, req->r_dentry,
1951 mds, req->r_dentry_drop, req->r_dentry_unless);
1952 if (req->r_old_dentry_drop)
1953 releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1954 mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1955 if (req->r_old_inode_drop)
1956 releases += ceph_encode_inode_release(&p,
2b0143b5 1957 d_inode(req->r_old_dentry),
2f2dc053 1958 mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
6e6f0923
YZ		 1959
1960 if (drop_cap_releases) {
1961 releases = 0;
1962 p = msg->front.iov_base + req->r_request_release_offset;
1963 }
1964
2f2dc053
SW		 1965 head->num_releases = cpu_to_le16(releases);
1966
b8e69066 1967 /* time stamp */
1f041a89
YZ		 1968 {
1969 struct ceph_timespec ts;
1970 ceph_encode_timespec(&ts, &req->r_stamp);
1971 ceph_encode_copy(&p, &ts, sizeof(ts));
1972 }
b8e69066 1973
2f2dc053
SW		 1974 BUG_ON(p > end);
1975 msg->front.iov_len = p - msg->front.iov_base;
1976 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1977
25e6bae3
YZ		 1978 if (req->r_pagelist) {
1979 struct ceph_pagelist *pagelist = req->r_pagelist;
1980 atomic_inc(&pagelist->refcnt);
1981 ceph_msg_data_add_pagelist(msg, pagelist);
1982 msg->hdr.data_len = cpu_to_le32(pagelist->length);
1983 } else {
1984 msg->hdr.data_len = 0;
ebf18f47 1985 }
02afca6c 1986
2f2dc053
SW		 1987 msg->hdr.data_off = cpu_to_le16(0);
1988
1989out_free2:
1990 if (freepath2)
1991 kfree((char *)path2);
1992out_free1:
1993 if (freepath1)
1994 kfree((char *)path1);
1995out:
1996 return msg;
1997}
1998
1999/*
2000 * called under mdsc->mutex if error, under no mutex if
2001 * success.
2002 */
2003static void complete_request(struct ceph_mds_client *mdsc,
2004 struct ceph_mds_request *req)
2005{
2006 if (req->r_callback)
2007 req->r_callback(mdsc, req);
2008 else
03066f23 2009 complete_all(&req->r_completion);
2f2dc053
SW		 2010}
2011
2012/*
2013 * called under mdsc->mutex
2014 */
2015static int __prepare_send_request(struct ceph_mds_client *mdsc,
2016 struct ceph_mds_request *req,
6e6f0923 2017 int mds, bool drop_cap_releases)
2f2dc053
SW		 2018{
2019 struct ceph_mds_request_head *rhead;
2020 struct ceph_msg *msg;
2021 int flags = 0;
2022
2f2dc053 2023 req->r_attempts++;
e55b71f8
GF		 2024 if (req->r_inode) {
2025 struct ceph_cap *cap =
2026 ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2027
2028 if (cap)
2029 req->r_sent_on_mseq = cap->mseq;
2030 else
2031 req->r_sent_on_mseq = -1;
2032 }
2f2dc053
SW		 2033 dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2034 req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2035
01a92f17 2036 if (req->r_got_unsafe) {
c5c9a0bf 2037 void *p;
01a92f17
SW		 2038 /*
2039 * Replay. Do not regenerate message (and rebuild
2040 * paths, etc.); just use the original message.
2041 * Rebuilding paths will break for renames because
2042 * d_move mangles the src name.
2043 */
2044 msg = req->r_request;
2045 rhead = msg->front.iov_base;
2046
2047 flags = le32_to_cpu(rhead->flags);
2048 flags |= CEPH_MDS_FLAG_REPLAY;
2049 rhead->flags = cpu_to_le32(flags);
2050
2051 if (req->r_target_inode)
2052 rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2053
2054 rhead->num_retry = req->r_attempts - 1;
e979cf50
SW		 2055
2056 /* remove cap/dentry releases from message */
2057 rhead->num_releases = 0;
c5c9a0bf
YZ		 2058
2059 /* time stamp */
2060 p = msg->front.iov_base + req->r_request_release_offset;
1f041a89
YZ		 2061 {
2062 struct ceph_timespec ts;
2063 ceph_encode_timespec(&ts, &req->r_stamp);
2064 ceph_encode_copy(&p, &ts, sizeof(ts));
2065 }
c5c9a0bf
YZ		 2066
2067 msg->front.iov_len = p - msg->front.iov_base;
2068 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
01a92f17
SW		 2069 return 0;
2070 }
2071
2f2dc053
SW		 2072 if (req->r_request) {
2073 ceph_msg_put(req->r_request);
2074 req->r_request = NULL;
2075 }
6e6f0923 2076 msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2f2dc053 2077 if (IS_ERR(msg)) {
e1518c7c 2078 req->r_err = PTR_ERR(msg);
2f2dc053 2079 complete_request(mdsc, req);
a79832f2 2080 return PTR_ERR(msg);
2f2dc053
SW		 2081 }
2082 req->r_request = msg;
2083
2084 rhead = msg->front.iov_base;
2f2dc053
SW		 2085 rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2086 if (req->r_got_unsafe)
2087 flags |= CEPH_MDS_FLAG_REPLAY;
2088 if (req->r_locked_dir)
2089 flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2090 rhead->flags = cpu_to_le32(flags);
2091 rhead->num_fwd = req->r_num_fwd;
2092 rhead->num_retry = req->r_attempts - 1;
01a92f17 2093 rhead->ino = 0;
2f2dc053
SW		 2094
2095 dout(" r_locked_dir = %p\n", req->r_locked_dir);
2f2dc053
SW		 2096 return 0;
2097}
2098
2099/*
2100 * send request, or put it on the appropriate wait list.
2101 */
2102static int __do_request(struct ceph_mds_client *mdsc,
2103 struct ceph_mds_request *req)
2104{
2105 struct ceph_mds_session *session = NULL;
2106 int mds = -1;
2107 int err = -EAGAIN;
2108
eb1b8af3
YZ		 2109 if (req->r_err || req->r_got_result) {
2110 if (req->r_aborted)
2111 __unregister_request(mdsc, req);
2f2dc053 2112 goto out;
eb1b8af3 2113 }
2f2dc053
SW		 2114
2115 if (req->r_timeout &&
2116 time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2117 dout("do_request timed out\n");
2118 err = -EIO;
2119 goto finish;
2120 }
2121
dc69e2e9
SW		 2122 put_request_session(req);
2123
2f2dc053
SW		 2124 mds = __choose_mds(mdsc, req);
2125 if (mds < 0 ||
2126 ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2127 dout("do_request no mds or not active, waiting for map\n");
2128 list_add(&req->r_wait, &mdsc->waiting_for_map);
2129 goto out;
2130 }
2131
2132 /* get, open session */
2133 session = __ceph_lookup_mds_session(mdsc, mds);
9c423956 2134 if (!session) {
2f2dc053 2135 session = register_session(mdsc, mds);
9c423956
SW		 2136 if (IS_ERR(session)) {
2137 err = PTR_ERR(session);
2138 goto finish;
2139 }
2140 }
dc69e2e9
SW		 2141 req->r_session = get_session(session);
2142
2f2dc053 2143 dout("do_request mds%d session %p state %s\n", mds, session,
a687ecaf 2144 ceph_session_state_name(session->s_state));
2f2dc053
SW		 2145 if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2146 session->s_state != CEPH_MDS_SESSION_HUNG) {
2147 if (session->s_state == CEPH_MDS_SESSION_NEW ||
2148 session->s_state == CEPH_MDS_SESSION_CLOSING)
2149 __open_session(mdsc, session);
2150 list_add(&req->r_wait, &session->s_waiting);
2151 goto out_session;
2152 }
2153
2154 /* send request */
2f2dc053
SW		 2155 req->r_resend_mds = -1; /* forget any previous mds hint */
2156
2157 if (req->r_request_started == 0) /* note request start time */
2158 req->r_request_started = jiffies;
2159
6e6f0923 2160 err = __prepare_send_request(mdsc, req, mds, false);
2f2dc053
SW		 2161 if (!err) {
2162 ceph_msg_get(req->r_request);
2163 ceph_con_send(&session->s_con, req->r_request);
2164 }
2165
2166out_session:
2167 ceph_put_mds_session(session);
2168out:
2169 return err;
2170
2171finish:
e1518c7c 2172 req->r_err = err;
2f2dc053
SW		 2173 complete_request(mdsc, req);
2174 goto out;
2175}
2176
2177/*
2178 * called under mdsc->mutex
2179 */
2180static void __wake_requests(struct ceph_mds_client *mdsc,
2181 struct list_head *head)
2182{
ed75ec2c
YZ		 2183 struct ceph_mds_request *req;
2184 LIST_HEAD(tmp_list);
2185
2186 list_splice_init(head, &tmp_list);
2f2dc053 2187
ed75ec2c
YZ		 2188 while (!list_empty(&tmp_list)) {
2189 req = list_entry(tmp_list.next,
2190 struct ceph_mds_request, r_wait);
2f2dc053 2191 list_del_init(&req->r_wait);
7971bd92 2192 dout(" wake request %p tid %llu\n", req, req->r_tid);
2f2dc053
SW		 2193 __do_request(mdsc, req);
2194 }
2195}
2196
2197/*
2198 * Wake up threads with requests pending for @mds, so that they can
29790f26 2199 * resubmit their requests to a possibly different mds.
2f2dc053 2200 */
29790f26 2201static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2f2dc053 2202{
44ca18f2 2203 struct ceph_mds_request *req;
282c1052 2204 struct rb_node *p = rb_first(&mdsc->request_tree);
2f2dc053
SW		 2205
2206 dout("kick_requests mds%d\n", mds);
282c1052 2207 while (p) {
44ca18f2 2208 req = rb_entry(p, struct ceph_mds_request, r_node);
282c1052 2209 p = rb_next(p);
44ca18f2
SW		 2210 if (req->r_got_unsafe)
2211 continue;
3de22be6
YZ		 2212 if (req->r_attempts > 0)
2213 continue; /* only new requests */
44ca18f2
SW		 2214 if (req->r_session &&
2215 req->r_session->s_mds == mds) {
2216 dout(" kicking tid %llu\n", req->r_tid);
03974e81 2217 list_del_init(&req->r_wait);
44ca18f2 2218 __do_request(mdsc, req);
2f2dc053
SW		 2219 }
2220 }
2221}
2222
2223void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2224 struct ceph_mds_request *req)
2225{
2226 dout("submit_request on %p\n", req);
2227 mutex_lock(&mdsc->mutex);
2228 __register_request(mdsc, req, NULL);
2229 __do_request(mdsc, req);
2230 mutex_unlock(&mdsc->mutex);
2231}
2232
2233/*
2234 * Synchrously perform an mds request. Take care of all of the
2235 * session setup, forwarding, retry details.
2236 */
2237int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2238 struct inode *dir,
2239 struct ceph_mds_request *req)
2240{
2241 int err;
2242
2243 dout("do_request on %p\n", req);
2244
2245 /* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2246 if (req->r_inode)
2247 ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2248 if (req->r_locked_dir)
2249 ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
844d87c3 2250 if (req->r_old_dentry_dir)
41b02e1f
SW		 2251 ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2252 CEPH_CAP_PIN);
2f2dc053
SW		 2253
2254 /* issue */
2255 mutex_lock(&mdsc->mutex);
2256 __register_request(mdsc, req, dir);
2257 __do_request(mdsc, req);
2258
e1518c7c
SW		 2259 if (req->r_err) {
2260 err = req->r_err;
2261 __unregister_request(mdsc, req);
2262 dout("do_request early error %d\n", err);
2263 goto out;
2f2dc053
SW		 2264 }
2265
e1518c7c
SW		 2266 /* wait */
2267 mutex_unlock(&mdsc->mutex);
2268 dout("do_request waiting\n");
2269 if (req->r_timeout) {
aa91647c 2270 err = (long)wait_for_completion_killable_timeout(
a319bf56
ID		 2271 &req->r_completion,
2272 ceph_timeout_jiffies(req->r_timeout));
e1518c7c
SW		 2273 if (err == 0)
2274 err = -EIO;
9280be24
YZ		 2275 } else if (req->r_wait_for_completion) {
2276 err = req->r_wait_for_completion(mdsc, req);
e1518c7c 2277 } else {
aa91647c 2278 err = wait_for_completion_killable(&req->r_completion);
e1518c7c
SW		 2279 }
2280 dout("do_request waited, got %d\n", err);
2281 mutex_lock(&mdsc->mutex);
5b1daecd 2282
e1518c7c
SW		 2283 /* only abort if we didn't race with a real reply */
2284 if (req->r_got_result) {
2285 err = le32_to_cpu(req->r_reply_info.head->result);
2286 } else if (err < 0) {
2287 dout("aborted request %lld with %d\n", req->r_tid, err);
b4556396
SW		 2288
2289 /*
2290 * ensure we aren't running concurrently with
2291 * ceph_fill_trace or ceph_readdir_prepopulate, which
2292 * rely on locks (dir mutex) held by our caller.
2293 */
2294 mutex_lock(&req->r_fill_mutex);
e1518c7c
SW		 2295 req->r_err = err;
2296 req->r_aborted = true;
b4556396 2297 mutex_unlock(&req->r_fill_mutex);
5b1daecd 2298
e1518c7c 2299 if (req->r_locked_dir &&
167c9e35
SW		 2300 (req->r_op & CEPH_MDS_OP_WRITE))
2301 ceph_invalidate_dir_request(req);
2f2dc053 2302 } else {
e1518c7c 2303 err = req->r_err;
2f2dc053 2304 }
2f2dc053 2305
e1518c7c
SW		 2306out:
2307 mutex_unlock(&mdsc->mutex);
2f2dc053
SW		 2308 dout("do_request %p done, result %d\n", req, err);
2309 return err;
2310}
2311
167c9e35 2312/*
2f276c51 2313 * Invalidate dir's completeness, dentry lease state on an aborted MDS
167c9e35
SW		 2314 * namespace request.
2315 */
2316void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2317{
2318 struct inode *inode = req->r_locked_dir;
167c9e35 2319
2f276c51 2320 dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
167c9e35 2321
2f276c51 2322 ceph_dir_clear_complete(inode);
167c9e35
SW		 2323 if (req->r_dentry)
2324 ceph_invalidate_dentry_lease(req->r_dentry);
2325 if (req->r_old_dentry)
2326 ceph_invalidate_dentry_lease(req->r_old_dentry);
2327}
2328
2f2dc053
SW		 2329/*
2330 * Handle mds reply.
2331 *
2332 * We take the session mutex and parse and process the reply immediately.
2333 * This preserves the logical ordering of replies, capabilities, etc., sent
2334 * by the MDS as they are applied to our local cache.
2335 */
2336static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2337{
2338 struct ceph_mds_client *mdsc = session->s_mdsc;
2339 struct ceph_mds_request *req;
2340 struct ceph_mds_reply_head *head = msg->front.iov_base;
2341 struct ceph_mds_reply_info_parsed *rinfo; /* parsed reply info */
982d6011 2342 struct ceph_snap_realm *realm;
2f2dc053
SW		 2343 u64 tid;
2344 int err, result;
2600d2dd 2345 int mds = session->s_mds;
2f2dc053 2346
2f2dc053
SW		 2347 if (msg->front.iov_len < sizeof(*head)) {
2348 pr_err("mdsc_handle_reply got corrupt (short) reply\n");
9ec7cab1 2349 ceph_msg_dump(msg);
2f2dc053
SW		 2350 return;
2351 }
2352
2353 /* get request, session */
6df058c0 2354 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2355 mutex_lock(&mdsc->mutex);
2356 req = __lookup_request(mdsc, tid);
2357 if (!req) {
2358 dout("handle_reply on unknown tid %llu\n", tid);
2359 mutex_unlock(&mdsc->mutex);
2360 return;
2361 }
2362 dout("handle_reply %p\n", req);
2f2dc053
SW		 2363
2364 /* correct session? */
d96d6049 2365 if (req->r_session != session) {
2f2dc053
SW		 2366 pr_err("mdsc_handle_reply got %llu on session mds%d"
2367 " not mds%d\n", tid, session->s_mds,
2368 req->r_session ? req->r_session->s_mds : -1);
2369 mutex_unlock(&mdsc->mutex);
2370 goto out;
2371 }
2372
2373 /* dup? */
2374 if ((req->r_got_unsafe && !head->safe) ||
2375 (req->r_got_safe && head->safe)) {
f3ae1b97 2376 pr_warn("got a dup %s reply on %llu from mds%d\n",
2f2dc053
SW		 2377 head->safe ? "safe" : "unsafe", tid, mds);
2378 mutex_unlock(&mdsc->mutex);
2379 goto out;
2380 }
85792d0d 2381 if (req->r_got_safe && !head->safe) {
f3ae1b97 2382 pr_warn("got unsafe after safe on %llu from mds%d\n",
85792d0d
SW		 2383 tid, mds);
2384 mutex_unlock(&mdsc->mutex);
2385 goto out;
2386 }
2f2dc053
SW		 2387
2388 result = le32_to_cpu(head->result);
2389
2390 /*
e55b71f8
GF		 2391 * Handle an ESTALE
2392 * if we're not talking to the authority, send to them
2393 * if the authority has changed while we weren't looking,
2394 * send to new authority
2395 * Otherwise we just have to return an ESTALE
2f2dc053
SW		 2396 */
2397 if (result == -ESTALE) {
e55b71f8 2398 dout("got ESTALE on request %llu", req->r_tid);
51da8e8c 2399 req->r_resend_mds = -1;
ca18bede 2400 if (req->r_direct_mode != USE_AUTH_MDS) {
e55b71f8
GF		 2401 dout("not using auth, setting for that now");
2402 req->r_direct_mode = USE_AUTH_MDS;
2f2dc053
SW		 2403 __do_request(mdsc, req);
2404 mutex_unlock(&mdsc->mutex);
2405 goto out;
e55b71f8 2406 } else {
ca18bede
YZ		 2407 int mds = __choose_mds(mdsc, req);
2408 if (mds >= 0 && mds != req->r_session->s_mds) {
2409 dout("but auth changed, so resending");
e55b71f8
GF		 2410 __do_request(mdsc, req);
2411 mutex_unlock(&mdsc->mutex);
2412 goto out;
2413 }
2f2dc053 2414 }
e55b71f8 2415 dout("have to return ESTALE on request %llu", req->r_tid);
2f2dc053
SW		 2416 }
2417
e55b71f8 2418
2f2dc053
SW		 2419 if (head->safe) {
2420 req->r_got_safe = true;
2421 __unregister_request(mdsc, req);
2f2dc053
SW		 2422
2423 if (req->r_got_unsafe) {
2424 /*
2425 * We already handled the unsafe response, now do the
2426 * cleanup. No need to examine the response; the MDS
2427 * doesn't include any result info in the safe
2428 * response. And even if it did, there is nothing
2429 * useful we could do with a revised return value.
2430 */
2431 dout("got safe reply %llu, mds%d\n", tid, mds);
2432 list_del_init(&req->r_unsafe_item);
2433
2434 /* last unsafe request during umount? */
44ca18f2 2435 if (mdsc->stopping && !__get_oldest_req(mdsc))
03066f23 2436 complete_all(&mdsc->safe_umount_waiters);
2f2dc053
SW		 2437 mutex_unlock(&mdsc->mutex);
2438 goto out;
2439 }
e1518c7c 2440 } else {
2f2dc053
SW		 2441 req->r_got_unsafe = true;
2442 list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2443 }
2444
2445 dout("handle_reply tid %lld result %d\n", tid, result);
2446 rinfo = &req->r_reply_info;
14303d20 2447 err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2f2dc053
SW		 2448 mutex_unlock(&mdsc->mutex);
2449
2450 mutex_lock(&session->s_mutex);
2451 if (err < 0) {
25933abd 2452 pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
9ec7cab1 2453 ceph_msg_dump(msg);
2f2dc053
SW		 2454 goto out_err;
2455 }
2456
2457 /* snap trace */
982d6011 2458 realm = NULL;
2f2dc053
SW		 2459 if (rinfo->snapblob_len) {
2460 down_write(&mdsc->snap_rwsem);
2461 ceph_update_snap_trace(mdsc, rinfo->snapblob,
982d6011
YZ		 2462 rinfo->snapblob + rinfo->snapblob_len,
2463 le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2464 &realm);
2f2dc053
SW		 2465 downgrade_write(&mdsc->snap_rwsem);
2466 } else {
2467 down_read(&mdsc->snap_rwsem);
2468 }
2469
2470 /* insert trace into our cache */
b4556396 2471 mutex_lock(&req->r_fill_mutex);
3d14c5d2 2472 err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2f2dc053 2473 if (err == 0) {
6e8575fa 2474 if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
81c6aea5 2475 req->r_op == CEPH_MDS_OP_LSSNAP))
2f2dc053 2476 ceph_readdir_prepopulate(req, req->r_session);
37151668 2477 ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2f2dc053 2478 }
b4556396 2479 mutex_unlock(&req->r_fill_mutex);
2f2dc053
SW		 2480
2481 up_read(&mdsc->snap_rwsem);
982d6011
YZ		 2482 if (realm)
2483 ceph_put_snap_realm(mdsc, realm);
2f2dc053 2484out_err:
e1518c7c
SW		 2485 mutex_lock(&mdsc->mutex);
2486 if (!req->r_aborted) {
2487 if (err) {
2488 req->r_err = err;
2489 } else {
2490 req->r_reply = msg;
2491 ceph_msg_get(msg);
2492 req->r_got_result = true;
2493 }
2f2dc053 2494 } else {
e1518c7c 2495 dout("reply arrived after request %lld was aborted\n", tid);
2f2dc053 2496 }
e1518c7c 2497 mutex_unlock(&mdsc->mutex);
2f2dc053 2498
2f2dc053
SW		 2499 mutex_unlock(&session->s_mutex);
2500
2501 /* kick calling process */
2502 complete_request(mdsc, req);
2503out:
2504 ceph_mdsc_put_request(req);
2505 return;
2506}
2507
2508
2509
2510/*
2511 * handle mds notification that our request has been forwarded.
2512 */
2600d2dd
SW		 2513static void handle_forward(struct ceph_mds_client *mdsc,
2514 struct ceph_mds_session *session,
2515 struct ceph_msg *msg)
2f2dc053
SW		 2516{
2517 struct ceph_mds_request *req;
a1ea787c 2518 u64 tid = le64_to_cpu(msg->hdr.tid);
2f2dc053
SW		 2519 u32 next_mds;
2520 u32 fwd_seq;
2f2dc053
SW		 2521 int err = -EINVAL;
2522 void *p = msg->front.iov_base;
2523 void *end = p + msg->front.iov_len;
2f2dc053 2524
a1ea787c 2525 ceph_decode_need(&p, end, 2*sizeof(u32), bad);
c89136ea
SW		 2526 next_mds = ceph_decode_32(&p);
2527 fwd_seq = ceph_decode_32(&p);
2f2dc053
SW		 2528
2529 mutex_lock(&mdsc->mutex);
2530 req = __lookup_request(mdsc, tid);
2531 if (!req) {
2a8e5e36 2532 dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2f2dc053
SW		 2533 goto out; /* dup reply? */
2534 }
2535
2a8e5e36
SW		 2536 if (req->r_aborted) {
2537 dout("forward tid %llu aborted, unregistering\n", tid);
2538 __unregister_request(mdsc, req);
2539 } else if (fwd_seq <= req->r_num_fwd) {
2540 dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2f2dc053
SW		 2541 tid, next_mds, req->r_num_fwd, fwd_seq);
2542 } else {
2543 /* resend. forward race not possible; mds would drop */
2a8e5e36
SW		 2544 dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2545 BUG_ON(req->r_err);
2546 BUG_ON(req->r_got_result);
3de22be6 2547 req->r_attempts = 0;
2f2dc053
SW		 2548 req->r_num_fwd = fwd_seq;
2549 req->r_resend_mds = next_mds;
2550 put_request_session(req);
2551 __do_request(mdsc, req);
2552 }
2553 ceph_mdsc_put_request(req);
2554out:
2555 mutex_unlock(&mdsc->mutex);
2556 return;
2557
2558bad:
2559 pr_err("mdsc_handle_forward decode error err=%d\n", err);
2560}
2561
2562/*
2563 * handle a mds session control message
2564 */
2565static void handle_session(struct ceph_mds_session *session,
2566 struct ceph_msg *msg)
2567{
2568 struct ceph_mds_client *mdsc = session->s_mdsc;
2569 u32 op;
2570 u64 seq;
2600d2dd 2571 int mds = session->s_mds;
2f2dc053
SW		 2572 struct ceph_mds_session_head *h = msg->front.iov_base;
2573 int wake = 0;
2574
2f2dc053
SW		 2575 /* decode */
2576 if (msg->front.iov_len != sizeof(*h))
2577 goto bad;
2578 op = le32_to_cpu(h->op);
2579 seq = le64_to_cpu(h->seq);
2580
2581 mutex_lock(&mdsc->mutex);
2600d2dd
SW		 2582 if (op == CEPH_SESSION_CLOSE)
2583 __unregister_session(mdsc, session);
2f2dc053
SW		 2584 /* FIXME: this ttl calculation is generous */
2585 session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2586 mutex_unlock(&mdsc->mutex);
2587
2588 mutex_lock(&session->s_mutex);
2589
2590 dout("handle_session mds%d %s %p state %s seq %llu\n",
2591 mds, ceph_session_op_name(op), session,
a687ecaf 2592 ceph_session_state_name(session->s_state), seq);
2f2dc053
SW		 2593
2594 if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2595 session->s_state = CEPH_MDS_SESSION_OPEN;
2596 pr_info("mds%d came back\n", session->s_mds);
2597 }
2598
2599 switch (op) {
2600 case CEPH_SESSION_OPEN:
29790f26
SW		 2601 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2602 pr_info("mds%d reconnect success\n", session->s_mds);
2f2dc053
SW		 2603 session->s_state = CEPH_MDS_SESSION_OPEN;
2604 renewed_caps(mdsc, session, 0);
2605 wake = 1;
2606 if (mdsc->stopping)
2607 __close_session(mdsc, session);
2608 break;
2609
2610 case CEPH_SESSION_RENEWCAPS:
2611 if (session->s_renew_seq == seq)
2612 renewed_caps(mdsc, session, 1);
2613 break;
2614
2615 case CEPH_SESSION_CLOSE:
29790f26
SW		 2616 if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2617 pr_info("mds%d reconnect denied\n", session->s_mds);
1c841a96 2618 cleanup_session_requests(mdsc, session);
2f2dc053 2619 remove_session_caps(session);
656e4382 2620 wake = 2; /* for good measure */
f3c60c59 2621 wake_up_all(&mdsc->session_close_wq);
2f2dc053
SW		 2622 break;
2623
2624 case CEPH_SESSION_STALE:
2625 pr_info("mds%d caps went stale, renewing\n",
2626 session->s_mds);
d8fb02ab 2627 spin_lock(&session->s_gen_ttl_lock);
2f2dc053 2628 session->s_cap_gen++;
1ce208a6 2629 session->s_cap_ttl = jiffies - 1;
d8fb02ab 2630 spin_unlock(&session->s_gen_ttl_lock);
2f2dc053
SW		 2631 send_renew_caps(mdsc, session);
2632 break;
2633
2634 case CEPH_SESSION_RECALL_STATE:
2635 trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2636 break;
2637
186e4f7a
YZ		 2638 case CEPH_SESSION_FLUSHMSG:
2639 send_flushmsg_ack(mdsc, session, seq);
2640 break;
2641
03f4fcb0
YZ		 2642 case CEPH_SESSION_FORCE_RO:
2643 dout("force_session_readonly %p\n", session);
2644 spin_lock(&session->s_cap_lock);
2645 session->s_readonly = true;
2646 spin_unlock(&session->s_cap_lock);
2647 wake_up_session_caps(session, 0);
2648 break;
2649
2f2dc053
SW		 2650 default:
2651 pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2652 WARN_ON(1);
2653 }
2654
2655 mutex_unlock(&session->s_mutex);
2656 if (wake) {
2657 mutex_lock(&mdsc->mutex);
2658 __wake_requests(mdsc, &session->s_waiting);
656e4382
YZ		 2659 if (wake == 2)
2660 kick_requests(mdsc, mds);
2f2dc053
SW		 2661 mutex_unlock(&mdsc->mutex);
2662 }
2663 return;
2664
2665bad:
2666 pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2667 (int)msg->front.iov_len);
9ec7cab1 2668 ceph_msg_dump(msg);
2f2dc053
SW		 2669 return;
2670}
2671
2672
2673/*
2674 * called under session->mutex.
2675 */
2676static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2677 struct ceph_mds_session *session)
2678{
2679 struct ceph_mds_request *req, *nreq;
3de22be6 2680 struct rb_node *p;
2f2dc053
SW		 2681 int err;
2682
2683 dout("replay_unsafe_requests mds%d\n", session->s_mds);
2684
2685 mutex_lock(&mdsc->mutex);
2686 list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
6e6f0923 2687 err = __prepare_send_request(mdsc, req, session->s_mds, true);
2f2dc053
SW		 2688 if (!err) {
2689 ceph_msg_get(req->r_request);
2690 ceph_con_send(&session->s_con, req->r_request);
2691 }
2692 }
3de22be6
YZ		 2693
2694 /*
2695 * also re-send old requests when MDS enters reconnect stage. So that MDS
2696 * can process completed request in clientreplay stage.
2697 */
2698 p = rb_first(&mdsc->request_tree);
2699 while (p) {
2700 req = rb_entry(p, struct ceph_mds_request, r_node);
2701 p = rb_next(p);
2702 if (req->r_got_unsafe)
2703 continue;
2704 if (req->r_attempts == 0)
2705 continue; /* only old requests */
2706 if (req->r_session &&
2707 req->r_session->s_mds == session->s_mds) {
6e6f0923
YZ		 2708 err = __prepare_send_request(mdsc, req,
2709 session->s_mds, true);
3de22be6
YZ		 2710 if (!err) {
2711 ceph_msg_get(req->r_request);
2712 ceph_con_send(&session->s_con, req->r_request);
2713 }
2714 }
2715 }
2f2dc053
SW		 2716 mutex_unlock(&mdsc->mutex);
2717}
2718
2719/*
2720 * Encode information about a cap for a reconnect with the MDS.
2721 */
2f2dc053
SW		 2722static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2723 void *arg)
2724{
20cb34ae
SW		 2725 union {
2726 struct ceph_mds_cap_reconnect v2;
2727 struct ceph_mds_cap_reconnect_v1 v1;
2728 } rec;
2729 size_t reclen;
2f2dc053 2730 struct ceph_inode_info *ci;
20cb34ae
SW		 2731 struct ceph_reconnect_state *recon_state = arg;
2732 struct ceph_pagelist *pagelist = recon_state->pagelist;
2f2dc053
SW		 2733 char *path;
2734 int pathlen, err;
2735 u64 pathbase;
2736 struct dentry *dentry;
2737
2738 ci = cap->ci;
2739
2740 dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2741 inode, ceph_vinop(inode), cap, cap->cap_id,
2742 ceph_cap_string(cap->issued));
93cea5be
SW		 2743 err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2744 if (err)
2745 return err;
2f2dc053
SW		 2746
2747 dentry = d_find_alias(inode);
2748 if (dentry) {
2749 path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2750 if (IS_ERR(path)) {
2751 err = PTR_ERR(path);
e072f8aa 2752 goto out_dput;
2f2dc053
SW		 2753 }
2754 } else {
2755 path = NULL;
2756 pathlen = 0;
2757 }
93cea5be
SW		 2758 err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2759 if (err)
e072f8aa 2760 goto out_free;
2f2dc053 2761
be655596 2762 spin_lock(&ci->i_ceph_lock);
2f2dc053
SW		 2763 cap->seq = 0; /* reset cap seq */
2764 cap->issue_seq = 0; /* and issue_seq */
667ca05c 2765 cap->mseq = 0; /* and migrate_seq */
99a9c273 2766 cap->cap_gen = cap->session->s_cap_gen;
20cb34ae
SW		 2767
2768 if (recon_state->flock) {
2769 rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2770 rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2771 rec.v2.issued = cpu_to_le32(cap->issued);
2772 rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2773 rec.v2.pathbase = cpu_to_le64(pathbase);
2774 rec.v2.flock_len = 0;
2775 reclen = sizeof(rec.v2);
2776 } else {
2777 rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2778 rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2779 rec.v1.issued = cpu_to_le32(cap->issued);
2780 rec.v1.size = cpu_to_le64(inode->i_size);
2781 ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2782 ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2783 rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2784 rec.v1.pathbase = cpu_to_le64(pathbase);
2785 reclen = sizeof(rec.v1);
2786 }
be655596 2787 spin_unlock(&ci->i_ceph_lock);
2f2dc053 2788
40819f6f
GF		 2789 if (recon_state->flock) {
2790 int num_fcntl_locks, num_flock_locks;
39be95e9
JS		 2791 struct ceph_filelock *flocks;
2792
2793encode_again:
39be95e9 2794 ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
39be95e9
JS		 2795 flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2796 sizeof(struct ceph_filelock), GFP_NOFS);
2797 if (!flocks) {
2798 err = -ENOMEM;
2799 goto out_free;
2800 }
39be95e9
JS		 2801 err = ceph_encode_locks_to_buffer(inode, flocks,
2802 num_fcntl_locks,
2803 num_flock_locks);
39be95e9
JS		 2804 if (err) {
2805 kfree(flocks);
2806 if (err == -ENOSPC)
2807 goto encode_again;
2808 goto out_free;
2809 }
2810 /*
2811 * number of encoded locks is stable, so copy to pagelist
2812 */
2813 rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2814 (num_fcntl_locks+num_flock_locks) *
2815 sizeof(struct ceph_filelock));
2816 err = ceph_pagelist_append(pagelist, &rec, reclen);
2817 if (!err)
2818 err = ceph_locks_to_pagelist(flocks, pagelist,
2819 num_fcntl_locks,
2820 num_flock_locks);
2821 kfree(flocks);
3612abbd
SW		 2822 } else {
2823 err = ceph_pagelist_append(pagelist, &rec, reclen);
40819f6f 2824 }
44c99757
YZ		 2825
2826 recon_state->nr_caps++;
e072f8aa 2827out_free:
2f2dc053 2828 kfree(path);
e072f8aa 2829out_dput:
2f2dc053 2830 dput(dentry);
93cea5be 2831 return err;
2f2dc053
SW		 2832}
2833
2834
2835/*
2836 * If an MDS fails and recovers, clients need to reconnect in order to
2837 * reestablish shared state. This includes all caps issued through
2838 * this session _and_ the snap_realm hierarchy. Because it's not
2839 * clear which snap realms the mds cares about, we send everything we
2840 * know about.. that ensures we'll then get any new info the
2841 * recovering MDS might have.
2842 *
2843 * This is a relatively heavyweight operation, but it's rare.
2844 *
2845 * called with mdsc->mutex held.
2846 */
34b6c855
SW		 2847static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2848 struct ceph_mds_session *session)
2f2dc053 2849{
2f2dc053 2850 struct ceph_msg *reply;
a105f00c 2851 struct rb_node *p;
34b6c855 2852 int mds = session->s_mds;
9abf82b8 2853 int err = -ENOMEM;
44c99757 2854 int s_nr_caps;
93cea5be 2855 struct ceph_pagelist *pagelist;
20cb34ae 2856 struct ceph_reconnect_state recon_state;
2f2dc053 2857
34b6c855 2858 pr_info("mds%d reconnect start\n", mds);
2f2dc053 2859
93cea5be
SW		 2860 pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2861 if (!pagelist)
2862 goto fail_nopagelist;
2863 ceph_pagelist_init(pagelist);
2864
b61c2763 2865 reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
a79832f2 2866 if (!reply)
93cea5be 2867 goto fail_nomsg;
93cea5be 2868
34b6c855
SW		 2869 mutex_lock(&session->s_mutex);
2870 session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2871 session->s_seq = 0;
2f2dc053 2872
2f2dc053 2873 dout("session %p state %s\n", session,
a687ecaf 2874 ceph_session_state_name(session->s_state));
2f2dc053 2875
99a9c273
YZ		 2876 spin_lock(&session->s_gen_ttl_lock);
2877 session->s_cap_gen++;
2878 spin_unlock(&session->s_gen_ttl_lock);
2879
2880 spin_lock(&session->s_cap_lock);
03f4fcb0
YZ		 2881 /* don't know if session is readonly */
2882 session->s_readonly = 0;
99a9c273
YZ		 2883 /*
2884 * notify __ceph_remove_cap() that we are composing cap reconnect.
2885 * If a cap get released before being added to the cap reconnect,
2886 * __ceph_remove_cap() should skip queuing cap release.
2887 */
2888 session->s_cap_reconnect = 1;
e01a5946 2889 /* drop old cap expires; we're about to reestablish that state */
745a8e3b 2890 cleanup_cap_releases(mdsc, session);
e01a5946 2891
5d23371f 2892 /* trim unused caps to reduce MDS's cache rejoin time */
c0bd50e2
YZ		 2893 if (mdsc->fsc->sb->s_root)
2894 shrink_dcache_parent(mdsc->fsc->sb->s_root);
5d23371f
YZ		 2895
2896 ceph_con_close(&session->s_con);
2897 ceph_con_open(&session->s_con,
2898 CEPH_ENTITY_TYPE_MDS, mds,
2899 ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2900
2901 /* replay unsafe requests */
2902 replay_unsafe_requests(mdsc, session);
2903
2904 down_read(&mdsc->snap_rwsem);
2905
2f2dc053 2906 /* traverse this session's caps */
44c99757
YZ		 2907 s_nr_caps = session->s_nr_caps;
2908 err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
93cea5be
SW		 2909 if (err)
2910 goto fail;
20cb34ae 2911
44c99757 2912 recon_state.nr_caps = 0;
20cb34ae
SW		 2913 recon_state.pagelist = pagelist;
2914 recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2915 err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2f2dc053 2916 if (err < 0)
9abf82b8 2917 goto fail;
2f2dc053 2918
99a9c273
YZ		 2919 spin_lock(&session->s_cap_lock);
2920 session->s_cap_reconnect = 0;
2921 spin_unlock(&session->s_cap_lock);
2922
2f2dc053
SW		 2923 /*
2924 * snaprealms. we provide mds with the ino, seq (version), and
2925 * parent for all of our realms. If the mds has any newer info,
2926 * it will tell us.
2927 */
a105f00c
SW		 2928 for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2929 struct ceph_snap_realm *realm =
2930 rb_entry(p, struct ceph_snap_realm, node);
93cea5be 2931 struct ceph_mds_snaprealm_reconnect sr_rec;
2f2dc053
SW		 2932
2933 dout(" adding snap realm %llx seq %lld parent %llx\n",
2934 realm->ino, realm->seq, realm->parent_ino);
93cea5be
SW		 2935 sr_rec.ino = cpu_to_le64(realm->ino);
2936 sr_rec.seq = cpu_to_le64(realm->seq);
2937 sr_rec.parent = cpu_to_le64(realm->parent_ino);
2938 err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2939 if (err)
2940 goto fail;
2f2dc053 2941 }
2f2dc053 2942
20cb34ae
SW		 2943 if (recon_state.flock)
2944 reply->hdr.version = cpu_to_le16(2);
44c99757
YZ		 2945
2946 /* raced with cap release? */
2947 if (s_nr_caps != recon_state.nr_caps) {
2948 struct page *page = list_first_entry(&pagelist->head,
2949 struct page, lru);
2950 __le32 *addr = kmap_atomic(page);
2951 *addr = cpu_to_le32(recon_state.nr_caps);
2952 kunmap_atomic(addr);
ebf18f47 2953 }
44c99757
YZ		 2954
2955 reply->hdr.data_len = cpu_to_le32(pagelist->length);
2956 ceph_msg_data_add_pagelist(reply, pagelist);
2f2dc053
SW		 2957 ceph_con_send(&session->s_con, reply);
2958
9abf82b8
SW		 2959 mutex_unlock(&session->s_mutex);
2960
2961 mutex_lock(&mdsc->mutex);
2962 __wake_requests(mdsc, &session->s_waiting);
2963 mutex_unlock(&mdsc->mutex);
2964
2f2dc053 2965 up_read(&mdsc->snap_rwsem);
2f2dc053
SW		 2966 return;
2967
93cea5be 2968fail:
2f2dc053 2969 ceph_msg_put(reply);
9abf82b8
SW		 2970 up_read(&mdsc->snap_rwsem);
2971 mutex_unlock(&session->s_mutex);
93cea5be
SW		 2972fail_nomsg:
2973 ceph_pagelist_release(pagelist);
93cea5be 2974fail_nopagelist:
9abf82b8 2975 pr_err("error %d preparing reconnect for mds%d\n", err, mds);
9abf82b8 2976 return;
2f2dc053
SW		 2977}
2978
2979
2980/*
2981 * compare old and new mdsmaps, kicking requests
2982 * and closing out old connections as necessary
2983 *
2984 * called under mdsc->mutex.
2985 */
2986static void check_new_map(struct ceph_mds_client *mdsc,
2987 struct ceph_mdsmap *newmap,
2988 struct ceph_mdsmap *oldmap)
2989{
2990 int i;
2991 int oldstate, newstate;
2992 struct ceph_mds_session *s;
2993
2994 dout("check_new_map new %u old %u\n",
2995 newmap->m_epoch, oldmap->m_epoch);
2996
2997 for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
2998 if (mdsc->sessions[i] == NULL)
2999 continue;
3000 s = mdsc->sessions[i];
3001 oldstate = ceph_mdsmap_get_state(oldmap, i);
3002 newstate = ceph_mdsmap_get_state(newmap, i);
3003
0deb01c9 3004 dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
2f2dc053 3005 i, ceph_mds_state_name(oldstate),
0deb01c9 3006 ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
2f2dc053 3007 ceph_mds_state_name(newstate),
0deb01c9 3008 ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
a687ecaf 3009 ceph_session_state_name(s->s_state));
2f2dc053 3010
3e8f43a0
YZ		 3011 if (i >= newmap->m_max_mds ||
3012 memcmp(ceph_mdsmap_get_addr(oldmap, i),
2f2dc053
SW		 3013 ceph_mdsmap_get_addr(newmap, i),
3014 sizeof(struct ceph_entity_addr))) {
3015 if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3016 /* the session never opened, just close it
3017 * out now */
3018 __wake_requests(mdsc, &s->s_waiting);
2600d2dd 3019 __unregister_session(mdsc, s);
2f2dc053
SW		 3020 } else {
3021 /* just close it */
3022 mutex_unlock(&mdsc->mutex);
3023 mutex_lock(&s->s_mutex);
3024 mutex_lock(&mdsc->mutex);
3025 ceph_con_close(&s->s_con);
3026 mutex_unlock(&s->s_mutex);
3027 s->s_state = CEPH_MDS_SESSION_RESTARTING;
3028 }
2f2dc053
SW		 3029 } else if (oldstate == newstate) {
3030 continue; /* nothing new with this mds */
3031 }
3032
3033 /*
3034 * send reconnect?
3035 */
3036 if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
34b6c855
SW		 3037 newstate >= CEPH_MDS_STATE_RECONNECT) {
3038 mutex_unlock(&mdsc->mutex);
3039 send_mds_reconnect(mdsc, s);
3040 mutex_lock(&mdsc->mutex);
3041 }
2f2dc053
SW		 3042
3043 /*
29790f26 3044 * kick request on any mds that has gone active.
2f2dc053
SW		 3045 */
3046 if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3047 newstate >= CEPH_MDS_STATE_ACTIVE) {
29790f26
SW		 3048 if (oldstate != CEPH_MDS_STATE_CREATING &&
3049 oldstate != CEPH_MDS_STATE_STARTING)
3050 pr_info("mds%d recovery completed\n", s->s_mds);
3051 kick_requests(mdsc, i);
2f2dc053 3052 ceph_kick_flushing_caps(mdsc, s);
0dc2570f 3053 wake_up_session_caps(s, 1);
2f2dc053
SW		 3054 }
3055 }
cb170a22
SW		 3056
3057 for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3058 s = mdsc->sessions[i];
3059 if (!s)
3060 continue;
3061 if (!ceph_mdsmap_is_laggy(newmap, i))
3062 continue;
3063 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3064 s->s_state == CEPH_MDS_SESSION_HUNG ||
3065 s->s_state == CEPH_MDS_SESSION_CLOSING) {
3066 dout(" connecting to export targets of laggy mds%d\n",
3067 i);
3068 __open_export_target_sessions(mdsc, s);
3069 }
3070 }
2f2dc053
SW		 3071}
3072
3073
3074
3075/*
3076 * leases
3077 */
3078
3079/*
3080 * caller must hold session s_mutex, dentry->d_lock
3081 */
3082void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3083{
3084 struct ceph_dentry_info *di = ceph_dentry(dentry);
3085
3086 ceph_put_mds_session(di->lease_session);
3087 di->lease_session = NULL;
3088}
3089
2600d2dd
SW		 3090static void handle_lease(struct ceph_mds_client *mdsc,
3091 struct ceph_mds_session *session,
3092 struct ceph_msg *msg)
2f2dc053 3093{
3d14c5d2 3094 struct super_block *sb = mdsc->fsc->sb;
2f2dc053 3095 struct inode *inode;
2f2dc053
SW		 3096 struct dentry *parent, *dentry;
3097 struct ceph_dentry_info *di;
2600d2dd 3098 int mds = session->s_mds;
2f2dc053 3099 struct ceph_mds_lease *h = msg->front.iov_base;
1e5ea23d 3100 u32 seq;
2f2dc053 3101 struct ceph_vino vino;
2f2dc053
SW		 3102 struct qstr dname;
3103 int release = 0;
3104
2f2dc053
SW		 3105 dout("handle_lease from mds%d\n", mds);
3106
3107 /* decode */
3108 if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3109 goto bad;
3110 vino.ino = le64_to_cpu(h->ino);
3111 vino.snap = CEPH_NOSNAP;
1e5ea23d 3112 seq = le32_to_cpu(h->seq);
2f2dc053
SW		 3113 dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3114 dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3115 if (dname.len != get_unaligned_le32(h+1))
3116 goto bad;
3117
2f2dc053
SW		 3118 /* lookup inode */
3119 inode = ceph_find_inode(sb, vino);
2f90b852
SW		 3120 dout("handle_lease %s, ino %llx %p %.*s\n",
3121 ceph_lease_op_name(h->action), vino.ino, inode,
1e5ea23d 3122 dname.len, dname.name);
6cd3bcad
YZ		 3123
3124 mutex_lock(&session->s_mutex);
3125 session->s_seq++;
3126
2f2dc053
SW		 3127 if (inode == NULL) {
3128 dout("handle_lease no inode %llx\n", vino.ino);
3129 goto release;
3130 }
2f2dc053
SW		 3131
3132 /* dentry */
3133 parent = d_find_alias(inode);
3134 if (!parent) {
3135 dout("no parent dentry on inode %p\n", inode);
3136 WARN_ON(1);
3137 goto release; /* hrm... */
3138 }
3139 dname.hash = full_name_hash(dname.name, dname.len);
3140 dentry = d_lookup(parent, &dname);
3141 dput(parent);
3142 if (!dentry)
3143 goto release;
3144
3145 spin_lock(&dentry->d_lock);
3146 di = ceph_dentry(dentry);
3147 switch (h->action) {
3148 case CEPH_MDS_LEASE_REVOKE:
3d8eb7a9 3149 if (di->lease_session == session) {
1e5ea23d
SW		 3150 if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3151 h->seq = cpu_to_le32(di->lease_seq);
2f2dc053
SW		 3152 __ceph_mdsc_drop_dentry_lease(dentry);
3153 }
3154 release = 1;
3155 break;
3156
3157 case CEPH_MDS_LEASE_RENEW:
3d8eb7a9 3158 if (di->lease_session == session &&
2f2dc053
SW		 3159 di->lease_gen == session->s_cap_gen &&
3160 di->lease_renew_from &&
3161 di->lease_renew_after == 0) {
3162 unsigned long duration =
3563dbdd 3163 msecs_to_jiffies(le32_to_cpu(h->duration_ms));
2f2dc053 3164
1e5ea23d 3165 di->lease_seq = seq;
2f2dc053
SW		 3166 dentry->d_time = di->lease_renew_from + duration;
3167 di->lease_renew_after = di->lease_renew_from +
3168 (duration >> 1);
3169 di->lease_renew_from = 0;
3170 }
3171 break;
3172 }
3173 spin_unlock(&dentry->d_lock);
3174 dput(dentry);
3175
3176 if (!release)
3177 goto out;
3178
3179release:
3180 /* let's just reuse the same message */
3181 h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3182 ceph_msg_get(msg);
3183 ceph_con_send(&session->s_con, msg);
3184
3185out:
3186 iput(inode);
3187 mutex_unlock(&session->s_mutex);
2f2dc053
SW		 3188 return;
3189
3190bad:
3191 pr_err("corrupt lease message\n");
9ec7cab1 3192 ceph_msg_dump(msg);
2f2dc053
SW		 3193}
3194
3195void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3196 struct inode *inode,
3197 struct dentry *dentry, char action,
3198 u32 seq)
3199{
3200 struct ceph_msg *msg;
3201 struct ceph_mds_lease *lease;
3202 int len = sizeof(*lease) + sizeof(u32);
3203 int dnamelen = 0;
3204
3205 dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3206 inode, dentry, ceph_lease_op_name(action), session->s_mds);
3207 dnamelen = dentry->d_name.len;
3208 len += dnamelen;
3209
b61c2763 3210 msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
a79832f2 3211 if (!msg)
2f2dc053
SW		 3212 return;
3213 lease = msg->front.iov_base;
3214 lease->action = action;
2f2dc053
SW		 3215 lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3216 lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3217 lease->seq = cpu_to_le32(seq);
3218 put_unaligned_le32(dnamelen, lease + 1);
3219 memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3220
3221 /*
3222 * if this is a preemptive lease RELEASE, no need to
3223 * flush request stream, since the actual request will
3224 * soon follow.
3225 */
3226 msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3227
3228 ceph_con_send(&session->s_con, msg);
3229}
3230
3231/*
3232 * Preemptively release a lease we expect to invalidate anyway.
3233 * Pass @inode always, @dentry is optional.
3234 */
3235void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
2f90b852 3236 struct dentry *dentry)
2f2dc053
SW		 3237{
3238 struct ceph_dentry_info *di;
3239 struct ceph_mds_session *session;
3240 u32 seq;
3241
3242 BUG_ON(inode == NULL);
3243 BUG_ON(dentry == NULL);
2f2dc053
SW		 3244
3245 /* is dentry lease valid? */
3246 spin_lock(&dentry->d_lock);
3247 di = ceph_dentry(dentry);
3248 if (!di || !di->lease_session ||
3249 di->lease_session->s_mds < 0 ||
3250 di->lease_gen != di->lease_session->s_cap_gen ||
3251 !time_before(jiffies, dentry->d_time)) {
3252 dout("lease_release inode %p dentry %p -- "
2f90b852
SW		 3253 "no lease\n",
3254 inode, dentry);
2f2dc053
SW		 3255 spin_unlock(&dentry->d_lock);
3256 return;
3257 }
3258
3259 /* we do have a lease on this dentry; note mds and seq */
3260 session = ceph_get_mds_session(di->lease_session);
3261 seq = di->lease_seq;
3262 __ceph_mdsc_drop_dentry_lease(dentry);
3263 spin_unlock(&dentry->d_lock);
3264
2f90b852
SW		 3265 dout("lease_release inode %p dentry %p to mds%d\n",
3266 inode, dentry, session->s_mds);
2f2dc053
SW		 3267 ceph_mdsc_lease_send_msg(session, inode, dentry,
3268 CEPH_MDS_LEASE_RELEASE, seq);
3269 ceph_put_mds_session(session);
3270}
3271
3272/*
3273 * drop all leases (and dentry refs) in preparation for umount
3274 */
3275static void drop_leases(struct ceph_mds_client *mdsc)
3276{
3277 int i;
3278
3279 dout("drop_leases\n");
3280 mutex_lock(&mdsc->mutex);
3281 for (i = 0; i < mdsc->max_sessions; i++) {
3282 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3283 if (!s)
3284 continue;
3285 mutex_unlock(&mdsc->mutex);
3286 mutex_lock(&s->s_mutex);
3287 mutex_unlock(&s->s_mutex);
3288 ceph_put_mds_session(s);
3289 mutex_lock(&mdsc->mutex);
3290 }
3291 mutex_unlock(&mdsc->mutex);
3292}
3293
3294
3295
3296/*
3297 * delayed work -- periodically trim expired leases, renew caps with mds
3298 */
3299static void schedule_delayed(struct ceph_mds_client *mdsc)
3300{
3301 int delay = 5;
3302 unsigned hz = round_jiffies_relative(HZ * delay);
3303 schedule_delayed_work(&mdsc->delayed_work, hz);
3304}
3305
3306static void delayed_work(struct work_struct *work)
3307{
3308 int i;
3309 struct ceph_mds_client *mdsc =
3310 container_of(work, struct ceph_mds_client, delayed_work.work);
3311 int renew_interval;
3312 int renew_caps;
3313
3314 dout("mdsc delayed_work\n");
afcdaea3 3315 ceph_check_delayed_caps(mdsc);
2f2dc053
SW		 3316
3317 mutex_lock(&mdsc->mutex);
3318 renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3319 renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3320 mdsc->last_renew_caps);
3321 if (renew_caps)
3322 mdsc->last_renew_caps = jiffies;
3323
3324 for (i = 0; i < mdsc->max_sessions; i++) {
3325 struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3326 if (s == NULL)
3327 continue;
3328 if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3329 dout("resending session close request for mds%d\n",
3330 s->s_mds);
3331 request_close_session(mdsc, s);
3332 ceph_put_mds_session(s);
3333 continue;
3334 }
3335 if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3336 if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3337 s->s_state = CEPH_MDS_SESSION_HUNG;
3338 pr_info("mds%d hung\n", s->s_mds);
3339 }
3340 }
3341 if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3342 /* this mds is failed or recovering, just wait */
3343 ceph_put_mds_session(s);
3344 continue;
3345 }
3346 mutex_unlock(&mdsc->mutex);
3347
3348 mutex_lock(&s->s_mutex);
3349 if (renew_caps)
3350 send_renew_caps(mdsc, s);
3351 else
3352 ceph_con_keepalive(&s->s_con);
aab53dd9
SW		 3353 if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3354 s->s_state == CEPH_MDS_SESSION_HUNG)
3d7ded4d 3355 ceph_send_cap_releases(mdsc, s);
2f2dc053
SW		 3356 mutex_unlock(&s->s_mutex);
3357 ceph_put_mds_session(s);
3358
3359 mutex_lock(&mdsc->mutex);
3360 }
3361 mutex_unlock(&mdsc->mutex);
3362
3363 schedule_delayed(mdsc);
3364}
3365
3d14c5d2 3366int ceph_mdsc_init(struct ceph_fs_client *fsc)
2f2dc053 3367
2f2dc053 3368{
3d14c5d2
YS		 3369 struct ceph_mds_client *mdsc;
3370
3371 mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3372 if (!mdsc)
3373 return -ENOMEM;
3374 mdsc->fsc = fsc;
3375 fsc->mdsc = mdsc;
2f2dc053
SW		 3376 mutex_init(&mdsc->mutex);
3377 mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
fb3101b6 3378 if (mdsc->mdsmap == NULL) {
3379 kfree(mdsc);
2d06eeb8 3380 return -ENOMEM;
fb3101b6 3381 }
2d06eeb8 3382
2f2dc053 3383 init_completion(&mdsc->safe_umount_waiters);
f3c60c59 3384 init_waitqueue_head(&mdsc->session_close_wq);
2f2dc053
SW		 3385 INIT_LIST_HEAD(&mdsc->waiting_for_map);
3386 mdsc->sessions = NULL;
86d8f67b 3387 atomic_set(&mdsc->num_sessions, 0);
2f2dc053
SW		 3388 mdsc->max_sessions = 0;
3389 mdsc->stopping = 0;
affbc19a 3390 mdsc->last_snap_seq = 0;
2f2dc053 3391 init_rwsem(&mdsc->snap_rwsem);
a105f00c 3392 mdsc->snap_realms = RB_ROOT;
2f2dc053
SW		 3393 INIT_LIST_HEAD(&mdsc->snap_empty);
3394 spin_lock_init(&mdsc->snap_empty_lock);
3395 mdsc->last_tid = 0;
e8a7b8b1 3396 mdsc->oldest_tid = 0;
44ca18f2 3397 mdsc->request_tree = RB_ROOT;
2f2dc053
SW		 3398 INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3399 mdsc->last_renew_caps = jiffies;
3400 INIT_LIST_HEAD(&mdsc->cap_delay_list);
3401 spin_lock_init(&mdsc->cap_delay_lock);
3402 INIT_LIST_HEAD(&mdsc->snap_flush_list);
3403 spin_lock_init(&mdsc->snap_flush_lock);
3404 mdsc->cap_flush_seq = 0;
3405 INIT_LIST_HEAD(&mdsc->cap_dirty);
db354052 3406 INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
2f2dc053
SW		 3407 mdsc->num_cap_flushing = 0;
3408 spin_lock_init(&mdsc->cap_dirty_lock);
3409 init_waitqueue_head(&mdsc->cap_flushing_wq);
3410 spin_lock_init(&mdsc->dentry_lru_lock);
3411 INIT_LIST_HEAD(&mdsc->dentry_lru);
2d06eeb8 3412
37151668 3413 ceph_caps_init(mdsc);
3d14c5d2 3414 ceph_adjust_min_caps(mdsc, fsc->min_caps);
37151668 3415
10183a69
YZ		 3416 init_rwsem(&mdsc->pool_perm_rwsem);
3417 mdsc->pool_perm_tree = RB_ROOT;
3418
5f44f142 3419 return 0;
2f2dc053
SW		 3420}
3421
3422/*
3423 * Wait for safe replies on open mds requests. If we time out, drop
3424 * all requests from the tree to avoid dangling dentry refs.
3425 */
3426static void wait_requests(struct ceph_mds_client *mdsc)
3427{
a319bf56 3428 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053 3429 struct ceph_mds_request *req;
2f2dc053
SW		 3430
3431 mutex_lock(&mdsc->mutex);
44ca18f2 3432 if (__get_oldest_req(mdsc)) {
2f2dc053 3433 mutex_unlock(&mdsc->mutex);
44ca18f2 3434
2f2dc053
SW		 3435 dout("wait_requests waiting for requests\n");
3436 wait_for_completion_timeout(&mdsc->safe_umount_waiters,
a319bf56 3437 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3438
3439 /* tear down remaining requests */
44ca18f2
SW		 3440 mutex_lock(&mdsc->mutex);
3441 while ((req = __get_oldest_req(mdsc))) {
2f2dc053
SW		 3442 dout("wait_requests timed out on tid %llu\n",
3443 req->r_tid);
44ca18f2 3444 __unregister_request(mdsc, req);
2f2dc053
SW		 3445 }
3446 }
3447 mutex_unlock(&mdsc->mutex);
3448 dout("wait_requests done\n");
3449}
3450
3451/*
3452 * called before mount is ro, and before dentries are torn down.
3453 * (hmm, does this still race with new lookups?)
3454 */
3455void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3456{
3457 dout("pre_umount\n");
3458 mdsc->stopping = 1;
3459
3460 drop_leases(mdsc);
afcdaea3 3461 ceph_flush_dirty_caps(mdsc);
2f2dc053 3462 wait_requests(mdsc);
17c688c3
SW		 3463
3464 /*
3465 * wait for reply handlers to drop their request refs and
3466 * their inode/dcache refs
3467 */
3468 ceph_msgr_flush();
2f2dc053
SW		 3469}
3470
3471/*
3472 * wait for all write mds requests to flush.
3473 */
3474static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3475{
80fc7314 3476 struct ceph_mds_request *req = NULL, *nextreq;
44ca18f2 3477 struct rb_node *n;
2f2dc053
SW		 3478
3479 mutex_lock(&mdsc->mutex);
3480 dout("wait_unsafe_requests want %lld\n", want_tid);
80fc7314 3481restart:
44ca18f2
SW		 3482 req = __get_oldest_req(mdsc);
3483 while (req && req->r_tid <= want_tid) {
80fc7314
SW		 3484 /* find next request */
3485 n = rb_next(&req->r_node);
3486 if (n)
3487 nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3488 else
3489 nextreq = NULL;
e8a7b8b1
YZ		 3490 if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3491 (req->r_op & CEPH_MDS_OP_WRITE)) {
44ca18f2
SW		 3492 /* write op */
3493 ceph_mdsc_get_request(req);
80fc7314
SW		 3494 if (nextreq)
3495 ceph_mdsc_get_request(nextreq);
44ca18f2
SW		 3496 mutex_unlock(&mdsc->mutex);
3497 dout("wait_unsafe_requests wait on %llu (want %llu)\n",
3498 req->r_tid, want_tid);
3499 wait_for_completion(&req->r_safe_completion);
3500 mutex_lock(&mdsc->mutex);
44ca18f2 3501 ceph_mdsc_put_request(req);
80fc7314
SW		 3502 if (!nextreq)
3503 break; /* next dne before, so we're done! */
3504 if (RB_EMPTY_NODE(&nextreq->r_node)) {
3505 /* next request was removed from tree */
3506 ceph_mdsc_put_request(nextreq);
3507 goto restart;
3508 }
3509 ceph_mdsc_put_request(nextreq); /* won't go away */
44ca18f2 3510 }
80fc7314 3511 req = nextreq;
2f2dc053
SW		 3512 }
3513 mutex_unlock(&mdsc->mutex);
3514 dout("wait_unsafe_requests done\n");
3515}
3516
3517void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3518{
affbc19a 3519 u64 want_tid, want_flush, want_snap;
2f2dc053 3520
3d14c5d2 3521 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
56b7cf95
SW		 3522 return;
3523
2f2dc053
SW		 3524 dout("sync\n");
3525 mutex_lock(&mdsc->mutex);
3526 want_tid = mdsc->last_tid;
2f2dc053 3527 mutex_unlock(&mdsc->mutex);
2f2dc053 3528
afcdaea3 3529 ceph_flush_dirty_caps(mdsc);
d3383a8e
YZ		 3530 spin_lock(&mdsc->cap_dirty_lock);
3531 want_flush = mdsc->cap_flush_seq;
3532 spin_unlock(&mdsc->cap_dirty_lock);
3533
affbc19a
YZ		 3534 down_read(&mdsc->snap_rwsem);
3535 want_snap = mdsc->last_snap_seq;
3536 up_read(&mdsc->snap_rwsem);
3537
3538 dout("sync want tid %lld flush_seq %lld snap_seq %lld\n",
3539 want_tid, want_flush, want_snap);
2f2dc053
SW		 3540
3541 wait_unsafe_requests(mdsc, want_tid);
affbc19a 3542 wait_caps_flush(mdsc, want_flush, want_snap);
2f2dc053
SW		 3543}
3544
f3c60c59
SW		 3545/*
3546 * true if all sessions are closed, or we force unmount
3547 */
7fd7d101 3548static bool done_closing_sessions(struct ceph_mds_client *mdsc)
f3c60c59 3549{
3d14c5d2 3550 if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
f3c60c59 3551 return true;
86d8f67b 3552 return atomic_read(&mdsc->num_sessions) == 0;
f3c60c59 3553}
2f2dc053
SW		 3554
3555/*
3556 * called after sb is ro.
3557 */
3558void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3559{
a319bf56 3560 struct ceph_options *opts = mdsc->fsc->client->options;
2f2dc053
SW		 3561 struct ceph_mds_session *session;
3562 int i;
2f2dc053
SW		 3563
3564 dout("close_sessions\n");
3565
2f2dc053 3566 /* close sessions */
f3c60c59
SW		 3567 mutex_lock(&mdsc->mutex);
3568 for (i = 0; i < mdsc->max_sessions; i++) {
3569 session = __ceph_lookup_mds_session(mdsc, i);
3570 if (!session)
3571 continue;
2f2dc053 3572 mutex_unlock(&mdsc->mutex);
f3c60c59
SW		 3573 mutex_lock(&session->s_mutex);
3574 __close_session(mdsc, session);
3575 mutex_unlock(&session->s_mutex);
3576 ceph_put_mds_session(session);
2f2dc053
SW		 3577 mutex_lock(&mdsc->mutex);
3578 }
f3c60c59
SW		 3579 mutex_unlock(&mdsc->mutex);
3580
3581 dout("waiting for sessions to close\n");
3582 wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
a319bf56 3583 ceph_timeout_jiffies(opts->mount_timeout));
2f2dc053
SW		 3584
3585 /* tear down remaining sessions */
f3c60c59 3586 mutex_lock(&mdsc->mutex);
2f2dc053
SW		 3587 for (i = 0; i < mdsc->max_sessions; i++) {
3588 if (mdsc->sessions[i]) {
3589 session = get_session(mdsc->sessions[i]);
2600d2dd 3590 __unregister_session(mdsc, session);
2f2dc053
SW		 3591 mutex_unlock(&mdsc->mutex);
3592 mutex_lock(&session->s_mutex);
3593 remove_session_caps(session);
3594 mutex_unlock(&session->s_mutex);
3595 ceph_put_mds_session(session);
3596 mutex_lock(&mdsc->mutex);
3597 }
3598 }
2f2dc053 3599 WARN_ON(!list_empty(&mdsc->cap_delay_list));
2f2dc053
SW		 3600 mutex_unlock(&mdsc->mutex);
3601
3602 ceph_cleanup_empty_realms(mdsc);
3603
3604 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3605
3606 dout("stopped\n");
3607}
3608
3d14c5d2 3609static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
2f2dc053
SW		 3610{
3611 dout("stop\n");
3612 cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3613 if (mdsc->mdsmap)
3614 ceph_mdsmap_destroy(mdsc->mdsmap);
3615 kfree(mdsc->sessions);
37151668 3616 ceph_caps_finalize(mdsc);
10183a69 3617 ceph_pool_perm_destroy(mdsc);
2f2dc053
SW		 3618}
3619
3d14c5d2
YS		 3620void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3621{
3622 struct ceph_mds_client *mdsc = fsc->mdsc;
3623
ef550f6f 3624 dout("mdsc_destroy %p\n", mdsc);
3d14c5d2 3625 ceph_mdsc_stop(mdsc);
ef550f6f
SW		 3626
3627 /* flush out any connection work with references to us */
3628 ceph_msgr_flush();
3629
3d14c5d2
YS		 3630 fsc->mdsc = NULL;
3631 kfree(mdsc);
ef550f6f 3632 dout("mdsc_destroy %p done\n", mdsc);
3d14c5d2
YS		 3633}
3634
2f2dc053
SW		 3635
3636/*
3637 * handle mds map update.
3638 */
3639void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3640{
3641 u32 epoch;
3642 u32 maplen;
3643 void *p = msg->front.iov_base;
3644 void *end = p + msg->front.iov_len;
3645 struct ceph_mdsmap *newmap, *oldmap;
3646 struct ceph_fsid fsid;
3647 int err = -EINVAL;
3648
3649 ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3650 ceph_decode_copy(&p, &fsid, sizeof(fsid));
3d14c5d2 3651 if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
0743304d 3652 return;
c89136ea
SW		 3653 epoch = ceph_decode_32(&p);
3654 maplen = ceph_decode_32(&p);
2f2dc053
SW		 3655 dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3656
3657 /* do we need it? */
3d14c5d2 3658 ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
2f2dc053
SW		 3659 mutex_lock(&mdsc->mutex);
3660 if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3661 dout("handle_map epoch %u <= our %u\n",
3662 epoch, mdsc->mdsmap->m_epoch);
3663 mutex_unlock(&mdsc->mutex);
3664 return;
3665 }
3666
3667 newmap = ceph_mdsmap_decode(&p, end);
3668 if (IS_ERR(newmap)) {
3669 err = PTR_ERR(newmap);
3670 goto bad_unlock;
3671 }
3672
3673 /* swap into place */
3674 if (mdsc->mdsmap) {
3675 oldmap = mdsc->mdsmap;
3676 mdsc->mdsmap = newmap;
3677 check_new_map(mdsc, newmap, oldmap);
3678 ceph_mdsmap_destroy(oldmap);
3679 } else {
3680 mdsc->mdsmap = newmap; /* first mds map */
3681 }
3d14c5d2 3682 mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
2f2dc053
SW		 3683
3684 __wake_requests(mdsc, &mdsc->waiting_for_map);
3685
3686 mutex_unlock(&mdsc->mutex);
3687 schedule_delayed(mdsc);
3688 return;
3689
3690bad_unlock:
3691 mutex_unlock(&mdsc->mutex);
3692bad:
3693 pr_err("error decoding mdsmap %d\n", err);
3694 return;
3695}
3696
3697static struct ceph_connection *con_get(struct ceph_connection *con)
3698{
3699 struct ceph_mds_session *s = con->private;
3700
3701 if (get_session(s)) {
2600d2dd 3702 dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
2f2dc053
SW		 3703 return con;
3704 }
3705 dout("mdsc con_get %p FAIL\n", s);
3706 return NULL;
3707}
3708
3709static void con_put(struct ceph_connection *con)
3710{
3711 struct ceph_mds_session *s = con->private;
3712
7d8e18a6 3713 dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
2f2dc053
SW		 3714 ceph_put_mds_session(s);
3715}
3716
3717/*
3718 * if the client is unresponsive for long enough, the mds will kill
3719 * the session entirely.
3720 */
3721static void peer_reset(struct ceph_connection *con)
3722{
3723 struct ceph_mds_session *s = con->private;
7e70f0ed 3724 struct ceph_mds_client *mdsc = s->s_mdsc;
2f2dc053 3725
f3ae1b97 3726 pr_warn("mds%d closed our session\n", s->s_mds);
7e70f0ed 3727 send_mds_reconnect(mdsc, s);
2f2dc053
SW		 3728}
3729
3730static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3731{
3732 struct ceph_mds_session *s = con->private;
3733 struct ceph_mds_client *mdsc = s->s_mdsc;
3734 int type = le16_to_cpu(msg->hdr.type);
3735
2600d2dd
SW		 3736 mutex_lock(&mdsc->mutex);
3737 if (__verify_registered_session(mdsc, s) < 0) {
3738 mutex_unlock(&mdsc->mutex);
3739 goto out;
3740 }
3741 mutex_unlock(&mdsc->mutex);
3742
2f2dc053
SW		 3743 switch (type) {
3744 case CEPH_MSG_MDS_MAP:
3745 ceph_mdsc_handle_map(mdsc, msg);
3746 break;
3747 case CEPH_MSG_CLIENT_SESSION:
3748 handle_session(s, msg);
3749 break;
3750 case CEPH_MSG_CLIENT_REPLY:
3751 handle_reply(s, msg);
3752 break;
3753 case CEPH_MSG_CLIENT_REQUEST_FORWARD:
2600d2dd 3754 handle_forward(mdsc, s, msg);
2f2dc053
SW		 3755 break;
3756 case CEPH_MSG_CLIENT_CAPS:
3757 ceph_handle_caps(s, msg);
3758 break;
3759 case CEPH_MSG_CLIENT_SNAP:
2600d2dd 3760 ceph_handle_snap(mdsc, s, msg);
2f2dc053
SW		 3761 break;
3762 case CEPH_MSG_CLIENT_LEASE:
2600d2dd 3763 handle_lease(mdsc, s, msg);
2f2dc053
SW		 3764 break;
3765
3766 default:
3767 pr_err("received unknown message type %d %s\n", type,
3768 ceph_msg_type_name(type));
3769 }
2600d2dd 3770out:
2f2dc053
SW		 3771 ceph_msg_put(msg);
3772}
3773
4e7a5dcd
SW		 3774/*
3775 * authentication
3776 */
a3530df3
AE		 3777
3778/*
3779 * Note: returned pointer is the address of a structure that's
3780 * managed separately. Caller must *not* attempt to free it.
3781 */
3782static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
8f43fb53 3783 int *proto, int force_new)
4e7a5dcd
SW		 3784{
3785 struct ceph_mds_session *s = con->private;
3786 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3787 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
74f1869f 3788 struct ceph_auth_handshake *auth = &s->s_auth;
4e7a5dcd 3789
74f1869f 3790 if (force_new && auth->authorizer) {
27859f97 3791 ceph_auth_destroy_authorizer(ac, auth->authorizer);
74f1869f 3792 auth->authorizer = NULL;
4e7a5dcd 3793 }
27859f97
SW		 3794 if (!auth->authorizer) {
3795 int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3796 auth);
0bed9b5c
SW		 3797 if (ret)
3798 return ERR_PTR(ret);
27859f97
SW		 3799 } else {
3800 int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3801 auth);
a255651d 3802 if (ret)
a3530df3 3803 return ERR_PTR(ret);
4e7a5dcd 3804 }
4e7a5dcd 3805 *proto = ac->protocol;
74f1869f 3806
a3530df3 3807 return auth;
4e7a5dcd
SW		 3808}
3809
3810
3811static int verify_authorizer_reply(struct ceph_connection *con, int len)
3812{
3813 struct ceph_mds_session *s = con->private;
3814 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3815 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4e7a5dcd 3816
27859f97 3817 return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
4e7a5dcd
SW		 3818}
3819
9bd2e6f8
SW		 3820static int invalidate_authorizer(struct ceph_connection *con)
3821{
3822 struct ceph_mds_session *s = con->private;
3823 struct ceph_mds_client *mdsc = s->s_mdsc;
3d14c5d2 3824 struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
9bd2e6f8 3825
27859f97 3826 ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
9bd2e6f8 3827
3d14c5d2 3828 return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
9bd2e6f8
SW		 3829}
3830
53ded495
AE		 3831static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3832 struct ceph_msg_header *hdr, int *skip)
3833{
3834 struct ceph_msg *msg;
3835 int type = (int) le16_to_cpu(hdr->type);
3836 int front_len = (int) le32_to_cpu(hdr->front_len);
3837
3838 if (con->in_msg)
3839 return con->in_msg;
3840
3841 *skip = 0;
3842 msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3843 if (!msg) {
3844 pr_err("unable to allocate msg type %d len %d\n",
3845 type, front_len);
3846 return NULL;
3847 }
53ded495
AE		 3848
3849 return msg;
3850}
3851
33d07337
YZ		 3852static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
3853{
3854 struct ceph_mds_session *s = con->private;
3855 struct ceph_auth_handshake *auth = &s->s_auth;
3856 return ceph_auth_sign_message(auth, msg);
3857}
3858
3859static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
3860{
3861 struct ceph_mds_session *s = con->private;
3862 struct ceph_auth_handshake *auth = &s->s_auth;
3863 return ceph_auth_check_message_signature(auth, msg);
3864}
3865
9e32789f 3866static const struct ceph_connection_operations mds_con_ops = {
2f2dc053
SW		 3867 .get = con_get,
3868 .put = con_put,
3869 .dispatch = dispatch,
4e7a5dcd
SW		 3870 .get_authorizer = get_authorizer,
3871 .verify_authorizer_reply = verify_authorizer_reply,
9bd2e6f8 3872 .invalidate_authorizer = invalidate_authorizer,
2f2dc053 3873 .peer_reset = peer_reset,
53ded495 3874 .alloc_msg = mds_alloc_msg,
33d07337
YZ		 3875 .sign_message = sign_message,
3876 .check_message_signature = check_message_signature,
2f2dc053
SW		 3877};
3878
2f2dc053 3879/* eof */
Linux 6.x block layer and io_uring tree(s)