xref: /linux/fs/ceph/mds_client.c (revision 93df8a1ed6231727c5db94a80b1a6bd5ee67cec3)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/fs.h>
4 #include <linux/wait.h>
5 #include <linux/slab.h>
6 #include <linux/gfp.h>
7 #include <linux/sched.h>
8 #include <linux/debugfs.h>
9 #include <linux/seq_file.h>
10 #include <linux/utsname.h>
11 #include <linux/ratelimit.h>
12 
13 #include "super.h"
14 #include "mds_client.h"
15 
16 #include <linux/ceph/ceph_features.h>
17 #include <linux/ceph/messenger.h>
18 #include <linux/ceph/decode.h>
19 #include <linux/ceph/pagelist.h>
20 #include <linux/ceph/auth.h>
21 #include <linux/ceph/debugfs.h>
22 
23 /*
24  * A cluster of MDS (metadata server) daemons is responsible for
25  * managing the file system namespace (the directory hierarchy and
26  * inodes) and for coordinating shared access to storage.  Metadata is
27  * partitioning hierarchically across a number of servers, and that
28  * partition varies over time as the cluster adjusts the distribution
29  * in order to balance load.
30  *
31  * The MDS client is primarily responsible to managing synchronous
32  * metadata requests for operations like open, unlink, and so forth.
33  * If there is a MDS failure, we find out about it when we (possibly
34  * request and) receive a new MDS map, and can resubmit affected
35  * requests.
36  *
37  * For the most part, though, we take advantage of a lossless
38  * communications channel to the MDS, and do not need to worry about
39  * timing out or resubmitting requests.
40  *
41  * We maintain a stateful "session" with each MDS we interact with.
42  * Within each session, we sent periodic heartbeat messages to ensure
43  * any capabilities or leases we have been issues remain valid.  If
44  * the session times out and goes stale, our leases and capabilities
45  * are no longer valid.
46  */
47 
48 struct ceph_reconnect_state {
49 	int nr_caps;
50 	struct ceph_pagelist *pagelist;
51 	bool flock;
52 };
53 
54 static void __wake_requests(struct ceph_mds_client *mdsc,
55 			    struct list_head *head);
56 
57 static const struct ceph_connection_operations mds_con_ops;
58 
59 
60 /*
61  * mds reply parsing
62  */
63 
64 /*
65  * parse individual inode info
66  */
67 static int parse_reply_info_in(void **p, void *end,
68 			       struct ceph_mds_reply_info_in *info,
69 			       u64 features)
70 {
71 	int err = -EIO;
72 
73 	info->in = *p;
74 	*p += sizeof(struct ceph_mds_reply_inode) +
75 		sizeof(*info->in->fragtree.splits) *
76 		le32_to_cpu(info->in->fragtree.nsplits);
77 
78 	ceph_decode_32_safe(p, end, info->symlink_len, bad);
79 	ceph_decode_need(p, end, info->symlink_len, bad);
80 	info->symlink = *p;
81 	*p += info->symlink_len;
82 
83 	if (features & CEPH_FEATURE_DIRLAYOUTHASH)
84 		ceph_decode_copy_safe(p, end, &info->dir_layout,
85 				      sizeof(info->dir_layout), bad);
86 	else
87 		memset(&info->dir_layout, 0, sizeof(info->dir_layout));
88 
89 	ceph_decode_32_safe(p, end, info->xattr_len, bad);
90 	ceph_decode_need(p, end, info->xattr_len, bad);
91 	info->xattr_data = *p;
92 	*p += info->xattr_len;
93 
94 	if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
95 		ceph_decode_64_safe(p, end, info->inline_version, bad);
96 		ceph_decode_32_safe(p, end, info->inline_len, bad);
97 		ceph_decode_need(p, end, info->inline_len, bad);
98 		info->inline_data = *p;
99 		*p += info->inline_len;
100 	} else
101 		info->inline_version = CEPH_INLINE_NONE;
102 
103 	return 0;
104 bad:
105 	return err;
106 }
107 
108 /*
109  * parse a normal reply, which may contain a (dir+)dentry and/or a
110  * target inode.
111  */
112 static int parse_reply_info_trace(void **p, void *end,
113 				  struct ceph_mds_reply_info_parsed *info,
114 				  u64 features)
115 {
116 	int err;
117 
118 	if (info->head->is_dentry) {
119 		err = parse_reply_info_in(p, end, &info->diri, features);
120 		if (err < 0)
121 			goto out_bad;
122 
123 		if (unlikely(*p + sizeof(*info->dirfrag) > end))
124 			goto bad;
125 		info->dirfrag = *p;
126 		*p += sizeof(*info->dirfrag) +
127 			sizeof(u32)*le32_to_cpu(info->dirfrag->ndist);
128 		if (unlikely(*p > end))
129 			goto bad;
130 
131 		ceph_decode_32_safe(p, end, info->dname_len, bad);
132 		ceph_decode_need(p, end, info->dname_len, bad);
133 		info->dname = *p;
134 		*p += info->dname_len;
135 		info->dlease = *p;
136 		*p += sizeof(*info->dlease);
137 	}
138 
139 	if (info->head->is_target) {
140 		err = parse_reply_info_in(p, end, &info->targeti, features);
141 		if (err < 0)
142 			goto out_bad;
143 	}
144 
145 	if (unlikely(*p != end))
146 		goto bad;
147 	return 0;
148 
149 bad:
150 	err = -EIO;
151 out_bad:
152 	pr_err("problem parsing mds trace %d\n", err);
153 	return err;
154 }
155 
156 /*
157  * parse readdir results
158  */
159 static int parse_reply_info_dir(void **p, void *end,
160 				struct ceph_mds_reply_info_parsed *info,
161 				u64 features)
162 {
163 	u32 num, i = 0;
164 	int err;
165 
166 	info->dir_dir = *p;
167 	if (*p + sizeof(*info->dir_dir) > end)
168 		goto bad;
169 	*p += sizeof(*info->dir_dir) +
170 		sizeof(u32)*le32_to_cpu(info->dir_dir->ndist);
171 	if (*p > end)
172 		goto bad;
173 
174 	ceph_decode_need(p, end, sizeof(num) + 2, bad);
175 	num = ceph_decode_32(p);
176 	info->dir_end = ceph_decode_8(p);
177 	info->dir_complete = ceph_decode_8(p);
178 	if (num == 0)
179 		goto done;
180 
181 	BUG_ON(!info->dir_in);
182 	info->dir_dname = (void *)(info->dir_in + num);
183 	info->dir_dname_len = (void *)(info->dir_dname + num);
184 	info->dir_dlease = (void *)(info->dir_dname_len + num);
185 	if ((unsigned long)(info->dir_dlease + num) >
186 	    (unsigned long)info->dir_in + info->dir_buf_size) {
187 		pr_err("dir contents are larger than expected\n");
188 		WARN_ON(1);
189 		goto bad;
190 	}
191 
192 	info->dir_nr = num;
193 	while (num) {
194 		/* dentry */
195 		ceph_decode_need(p, end, sizeof(u32)*2, bad);
196 		info->dir_dname_len[i] = ceph_decode_32(p);
197 		ceph_decode_need(p, end, info->dir_dname_len[i], bad);
198 		info->dir_dname[i] = *p;
199 		*p += info->dir_dname_len[i];
200 		dout("parsed dir dname '%.*s'\n", info->dir_dname_len[i],
201 		     info->dir_dname[i]);
202 		info->dir_dlease[i] = *p;
203 		*p += sizeof(struct ceph_mds_reply_lease);
204 
205 		/* inode */
206 		err = parse_reply_info_in(p, end, &info->dir_in[i], features);
207 		if (err < 0)
208 			goto out_bad;
209 		i++;
210 		num--;
211 	}
212 
213 done:
214 	if (*p != end)
215 		goto bad;
216 	return 0;
217 
218 bad:
219 	err = -EIO;
220 out_bad:
221 	pr_err("problem parsing dir contents %d\n", err);
222 	return err;
223 }
224 
225 /*
226  * parse fcntl F_GETLK results
227  */
228 static int parse_reply_info_filelock(void **p, void *end,
229 				     struct ceph_mds_reply_info_parsed *info,
230 				     u64 features)
231 {
232 	if (*p + sizeof(*info->filelock_reply) > end)
233 		goto bad;
234 
235 	info->filelock_reply = *p;
236 	*p += sizeof(*info->filelock_reply);
237 
238 	if (unlikely(*p != end))
239 		goto bad;
240 	return 0;
241 
242 bad:
243 	return -EIO;
244 }
245 
246 /*
247  * parse create results
248  */
249 static int parse_reply_info_create(void **p, void *end,
250 				  struct ceph_mds_reply_info_parsed *info,
251 				  u64 features)
252 {
253 	if (features & CEPH_FEATURE_REPLY_CREATE_INODE) {
254 		if (*p == end) {
255 			info->has_create_ino = false;
256 		} else {
257 			info->has_create_ino = true;
258 			info->ino = ceph_decode_64(p);
259 		}
260 	}
261 
262 	if (unlikely(*p != end))
263 		goto bad;
264 	return 0;
265 
266 bad:
267 	return -EIO;
268 }
269 
270 /*
271  * parse extra results
272  */
273 static int parse_reply_info_extra(void **p, void *end,
274 				  struct ceph_mds_reply_info_parsed *info,
275 				  u64 features)
276 {
277 	if (info->head->op == CEPH_MDS_OP_GETFILELOCK)
278 		return parse_reply_info_filelock(p, end, info, features);
279 	else if (info->head->op == CEPH_MDS_OP_READDIR ||
280 		 info->head->op == CEPH_MDS_OP_LSSNAP)
281 		return parse_reply_info_dir(p, end, info, features);
282 	else if (info->head->op == CEPH_MDS_OP_CREATE)
283 		return parse_reply_info_create(p, end, info, features);
284 	else
285 		return -EIO;
286 }
287 
288 /*
289  * parse entire mds reply
290  */
291 static int parse_reply_info(struct ceph_msg *msg,
292 			    struct ceph_mds_reply_info_parsed *info,
293 			    u64 features)
294 {
295 	void *p, *end;
296 	u32 len;
297 	int err;
298 
299 	info->head = msg->front.iov_base;
300 	p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
301 	end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
302 
303 	/* trace */
304 	ceph_decode_32_safe(&p, end, len, bad);
305 	if (len > 0) {
306 		ceph_decode_need(&p, end, len, bad);
307 		err = parse_reply_info_trace(&p, p+len, info, features);
308 		if (err < 0)
309 			goto out_bad;
310 	}
311 
312 	/* extra */
313 	ceph_decode_32_safe(&p, end, len, bad);
314 	if (len > 0) {
315 		ceph_decode_need(&p, end, len, bad);
316 		err = parse_reply_info_extra(&p, p+len, info, features);
317 		if (err < 0)
318 			goto out_bad;
319 	}
320 
321 	/* snap blob */
322 	ceph_decode_32_safe(&p, end, len, bad);
323 	info->snapblob_len = len;
324 	info->snapblob = p;
325 	p += len;
326 
327 	if (p != end)
328 		goto bad;
329 	return 0;
330 
331 bad:
332 	err = -EIO;
333 out_bad:
334 	pr_err("mds parse_reply err %d\n", err);
335 	return err;
336 }
337 
338 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
339 {
340 	if (!info->dir_in)
341 		return;
342 	free_pages((unsigned long)info->dir_in, get_order(info->dir_buf_size));
343 }
344 
345 
346 /*
347  * sessions
348  */
349 const char *ceph_session_state_name(int s)
350 {
351 	switch (s) {
352 	case CEPH_MDS_SESSION_NEW: return "new";
353 	case CEPH_MDS_SESSION_OPENING: return "opening";
354 	case CEPH_MDS_SESSION_OPEN: return "open";
355 	case CEPH_MDS_SESSION_HUNG: return "hung";
356 	case CEPH_MDS_SESSION_CLOSING: return "closing";
357 	case CEPH_MDS_SESSION_RESTARTING: return "restarting";
358 	case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
359 	default: return "???";
360 	}
361 }
362 
363 static struct ceph_mds_session *get_session(struct ceph_mds_session *s)
364 {
365 	if (atomic_inc_not_zero(&s->s_ref)) {
366 		dout("mdsc get_session %p %d -> %d\n", s,
367 		     atomic_read(&s->s_ref)-1, atomic_read(&s->s_ref));
368 		return s;
369 	} else {
370 		dout("mdsc get_session %p 0 -- FAIL", s);
371 		return NULL;
372 	}
373 }
374 
375 void ceph_put_mds_session(struct ceph_mds_session *s)
376 {
377 	dout("mdsc put_session %p %d -> %d\n", s,
378 	     atomic_read(&s->s_ref), atomic_read(&s->s_ref)-1);
379 	if (atomic_dec_and_test(&s->s_ref)) {
380 		if (s->s_auth.authorizer)
381 			ceph_auth_destroy_authorizer(
382 				s->s_mdsc->fsc->client->monc.auth,
383 				s->s_auth.authorizer);
384 		kfree(s);
385 	}
386 }
387 
388 /*
389  * called under mdsc->mutex
390  */
391 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
392 						   int mds)
393 {
394 	struct ceph_mds_session *session;
395 
396 	if (mds >= mdsc->max_sessions || mdsc->sessions[mds] == NULL)
397 		return NULL;
398 	session = mdsc->sessions[mds];
399 	dout("lookup_mds_session %p %d\n", session,
400 	     atomic_read(&session->s_ref));
401 	get_session(session);
402 	return session;
403 }
404 
405 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
406 {
407 	if (mds >= mdsc->max_sessions)
408 		return false;
409 	return mdsc->sessions[mds];
410 }
411 
412 static int __verify_registered_session(struct ceph_mds_client *mdsc,
413 				       struct ceph_mds_session *s)
414 {
415 	if (s->s_mds >= mdsc->max_sessions ||
416 	    mdsc->sessions[s->s_mds] != s)
417 		return -ENOENT;
418 	return 0;
419 }
420 
421 /*
422  * create+register a new session for given mds.
423  * called under mdsc->mutex.
424  */
425 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
426 						 int mds)
427 {
428 	struct ceph_mds_session *s;
429 
430 	if (mds >= mdsc->mdsmap->m_max_mds)
431 		return ERR_PTR(-EINVAL);
432 
433 	s = kzalloc(sizeof(*s), GFP_NOFS);
434 	if (!s)
435 		return ERR_PTR(-ENOMEM);
436 	s->s_mdsc = mdsc;
437 	s->s_mds = mds;
438 	s->s_state = CEPH_MDS_SESSION_NEW;
439 	s->s_ttl = 0;
440 	s->s_seq = 0;
441 	mutex_init(&s->s_mutex);
442 
443 	ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
444 
445 	spin_lock_init(&s->s_gen_ttl_lock);
446 	s->s_cap_gen = 0;
447 	s->s_cap_ttl = jiffies - 1;
448 
449 	spin_lock_init(&s->s_cap_lock);
450 	s->s_renew_requested = 0;
451 	s->s_renew_seq = 0;
452 	INIT_LIST_HEAD(&s->s_caps);
453 	s->s_nr_caps = 0;
454 	s->s_trim_caps = 0;
455 	atomic_set(&s->s_ref, 1);
456 	INIT_LIST_HEAD(&s->s_waiting);
457 	INIT_LIST_HEAD(&s->s_unsafe);
458 	s->s_num_cap_releases = 0;
459 	s->s_cap_reconnect = 0;
460 	s->s_cap_iterator = NULL;
461 	INIT_LIST_HEAD(&s->s_cap_releases);
462 	INIT_LIST_HEAD(&s->s_cap_flushing);
463 	INIT_LIST_HEAD(&s->s_cap_snaps_flushing);
464 
465 	dout("register_session mds%d\n", mds);
466 	if (mds >= mdsc->max_sessions) {
467 		int newmax = 1 << get_count_order(mds+1);
468 		struct ceph_mds_session **sa;
469 
470 		dout("register_session realloc to %d\n", newmax);
471 		sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
472 		if (sa == NULL)
473 			goto fail_realloc;
474 		if (mdsc->sessions) {
475 			memcpy(sa, mdsc->sessions,
476 			       mdsc->max_sessions * sizeof(void *));
477 			kfree(mdsc->sessions);
478 		}
479 		mdsc->sessions = sa;
480 		mdsc->max_sessions = newmax;
481 	}
482 	mdsc->sessions[mds] = s;
483 	atomic_inc(&mdsc->num_sessions);
484 	atomic_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
485 
486 	ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
487 		      ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
488 
489 	return s;
490 
491 fail_realloc:
492 	kfree(s);
493 	return ERR_PTR(-ENOMEM);
494 }
495 
496 /*
497  * called under mdsc->mutex
498  */
499 static void __unregister_session(struct ceph_mds_client *mdsc,
500 			       struct ceph_mds_session *s)
501 {
502 	dout("__unregister_session mds%d %p\n", s->s_mds, s);
503 	BUG_ON(mdsc->sessions[s->s_mds] != s);
504 	mdsc->sessions[s->s_mds] = NULL;
505 	ceph_con_close(&s->s_con);
506 	ceph_put_mds_session(s);
507 	atomic_dec(&mdsc->num_sessions);
508 }
509 
510 /*
511  * drop session refs in request.
512  *
513  * should be last request ref, or hold mdsc->mutex
514  */
515 static void put_request_session(struct ceph_mds_request *req)
516 {
517 	if (req->r_session) {
518 		ceph_put_mds_session(req->r_session);
519 		req->r_session = NULL;
520 	}
521 }
522 
523 void ceph_mdsc_release_request(struct kref *kref)
524 {
525 	struct ceph_mds_request *req = container_of(kref,
526 						    struct ceph_mds_request,
527 						    r_kref);
528 	destroy_reply_info(&req->r_reply_info);
529 	if (req->r_request)
530 		ceph_msg_put(req->r_request);
531 	if (req->r_reply)
532 		ceph_msg_put(req->r_reply);
533 	if (req->r_inode) {
534 		ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
535 		iput(req->r_inode);
536 	}
537 	if (req->r_locked_dir)
538 		ceph_put_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
539 	iput(req->r_target_inode);
540 	if (req->r_dentry)
541 		dput(req->r_dentry);
542 	if (req->r_old_dentry)
543 		dput(req->r_old_dentry);
544 	if (req->r_old_dentry_dir) {
545 		/*
546 		 * track (and drop pins for) r_old_dentry_dir
547 		 * separately, since r_old_dentry's d_parent may have
548 		 * changed between the dir mutex being dropped and
549 		 * this request being freed.
550 		 */
551 		ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
552 				  CEPH_CAP_PIN);
553 		iput(req->r_old_dentry_dir);
554 	}
555 	kfree(req->r_path1);
556 	kfree(req->r_path2);
557 	if (req->r_pagelist)
558 		ceph_pagelist_release(req->r_pagelist);
559 	put_request_session(req);
560 	ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
561 	kfree(req);
562 }
563 
564 /*
565  * lookup session, bump ref if found.
566  *
567  * called under mdsc->mutex.
568  */
569 static struct ceph_mds_request *__lookup_request(struct ceph_mds_client *mdsc,
570 					     u64 tid)
571 {
572 	struct ceph_mds_request *req;
573 	struct rb_node *n = mdsc->request_tree.rb_node;
574 
575 	while (n) {
576 		req = rb_entry(n, struct ceph_mds_request, r_node);
577 		if (tid < req->r_tid)
578 			n = n->rb_left;
579 		else if (tid > req->r_tid)
580 			n = n->rb_right;
581 		else {
582 			ceph_mdsc_get_request(req);
583 			return req;
584 		}
585 	}
586 	return NULL;
587 }
588 
589 static void __insert_request(struct ceph_mds_client *mdsc,
590 			     struct ceph_mds_request *new)
591 {
592 	struct rb_node **p = &mdsc->request_tree.rb_node;
593 	struct rb_node *parent = NULL;
594 	struct ceph_mds_request *req = NULL;
595 
596 	while (*p) {
597 		parent = *p;
598 		req = rb_entry(parent, struct ceph_mds_request, r_node);
599 		if (new->r_tid < req->r_tid)
600 			p = &(*p)->rb_left;
601 		else if (new->r_tid > req->r_tid)
602 			p = &(*p)->rb_right;
603 		else
604 			BUG();
605 	}
606 
607 	rb_link_node(&new->r_node, parent, p);
608 	rb_insert_color(&new->r_node, &mdsc->request_tree);
609 }
610 
611 /*
612  * Register an in-flight request, and assign a tid.  Link to directory
613  * are modifying (if any).
614  *
615  * Called under mdsc->mutex.
616  */
617 static void __register_request(struct ceph_mds_client *mdsc,
618 			       struct ceph_mds_request *req,
619 			       struct inode *dir)
620 {
621 	req->r_tid = ++mdsc->last_tid;
622 	if (req->r_num_caps)
623 		ceph_reserve_caps(mdsc, &req->r_caps_reservation,
624 				  req->r_num_caps);
625 	dout("__register_request %p tid %lld\n", req, req->r_tid);
626 	ceph_mdsc_get_request(req);
627 	__insert_request(mdsc, req);
628 
629 	req->r_uid = current_fsuid();
630 	req->r_gid = current_fsgid();
631 
632 	if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
633 		mdsc->oldest_tid = req->r_tid;
634 
635 	if (dir) {
636 		struct ceph_inode_info *ci = ceph_inode(dir);
637 
638 		ihold(dir);
639 		spin_lock(&ci->i_unsafe_lock);
640 		req->r_unsafe_dir = dir;
641 		list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
642 		spin_unlock(&ci->i_unsafe_lock);
643 	}
644 }
645 
646 static void __unregister_request(struct ceph_mds_client *mdsc,
647 				 struct ceph_mds_request *req)
648 {
649 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
650 
651 	if (req->r_tid == mdsc->oldest_tid) {
652 		struct rb_node *p = rb_next(&req->r_node);
653 		mdsc->oldest_tid = 0;
654 		while (p) {
655 			struct ceph_mds_request *next_req =
656 				rb_entry(p, struct ceph_mds_request, r_node);
657 			if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
658 				mdsc->oldest_tid = next_req->r_tid;
659 				break;
660 			}
661 			p = rb_next(p);
662 		}
663 	}
664 
665 	rb_erase(&req->r_node, &mdsc->request_tree);
666 	RB_CLEAR_NODE(&req->r_node);
667 
668 	if (req->r_unsafe_dir) {
669 		struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
670 
671 		spin_lock(&ci->i_unsafe_lock);
672 		list_del_init(&req->r_unsafe_dir_item);
673 		spin_unlock(&ci->i_unsafe_lock);
674 
675 		iput(req->r_unsafe_dir);
676 		req->r_unsafe_dir = NULL;
677 	}
678 
679 	complete_all(&req->r_safe_completion);
680 
681 	ceph_mdsc_put_request(req);
682 }
683 
684 /*
685  * Choose mds to send request to next.  If there is a hint set in the
686  * request (e.g., due to a prior forward hint from the mds), use that.
687  * Otherwise, consult frag tree and/or caps to identify the
688  * appropriate mds.  If all else fails, choose randomly.
689  *
690  * Called under mdsc->mutex.
691  */
692 static struct dentry *get_nonsnap_parent(struct dentry *dentry)
693 {
694 	/*
695 	 * we don't need to worry about protecting the d_parent access
696 	 * here because we never renaming inside the snapped namespace
697 	 * except to resplice to another snapdir, and either the old or new
698 	 * result is a valid result.
699 	 */
700 	while (!IS_ROOT(dentry) && ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
701 		dentry = dentry->d_parent;
702 	return dentry;
703 }
704 
705 static int __choose_mds(struct ceph_mds_client *mdsc,
706 			struct ceph_mds_request *req)
707 {
708 	struct inode *inode;
709 	struct ceph_inode_info *ci;
710 	struct ceph_cap *cap;
711 	int mode = req->r_direct_mode;
712 	int mds = -1;
713 	u32 hash = req->r_direct_hash;
714 	bool is_hash = req->r_direct_is_hash;
715 
716 	/*
717 	 * is there a specific mds we should try?  ignore hint if we have
718 	 * no session and the mds is not up (active or recovering).
719 	 */
720 	if (req->r_resend_mds >= 0 &&
721 	    (__have_session(mdsc, req->r_resend_mds) ||
722 	     ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
723 		dout("choose_mds using resend_mds mds%d\n",
724 		     req->r_resend_mds);
725 		return req->r_resend_mds;
726 	}
727 
728 	if (mode == USE_RANDOM_MDS)
729 		goto random;
730 
731 	inode = NULL;
732 	if (req->r_inode) {
733 		inode = req->r_inode;
734 	} else if (req->r_dentry) {
735 		/* ignore race with rename; old or new d_parent is okay */
736 		struct dentry *parent = req->r_dentry->d_parent;
737 		struct inode *dir = d_inode(parent);
738 
739 		if (dir->i_sb != mdsc->fsc->sb) {
740 			/* not this fs! */
741 			inode = d_inode(req->r_dentry);
742 		} else if (ceph_snap(dir) != CEPH_NOSNAP) {
743 			/* direct snapped/virtual snapdir requests
744 			 * based on parent dir inode */
745 			struct dentry *dn = get_nonsnap_parent(parent);
746 			inode = d_inode(dn);
747 			dout("__choose_mds using nonsnap parent %p\n", inode);
748 		} else {
749 			/* dentry target */
750 			inode = d_inode(req->r_dentry);
751 			if (!inode || mode == USE_AUTH_MDS) {
752 				/* dir + name */
753 				inode = dir;
754 				hash = ceph_dentry_hash(dir, req->r_dentry);
755 				is_hash = true;
756 			}
757 		}
758 	}
759 
760 	dout("__choose_mds %p is_hash=%d (%d) mode %d\n", inode, (int)is_hash,
761 	     (int)hash, mode);
762 	if (!inode)
763 		goto random;
764 	ci = ceph_inode(inode);
765 
766 	if (is_hash && S_ISDIR(inode->i_mode)) {
767 		struct ceph_inode_frag frag;
768 		int found;
769 
770 		ceph_choose_frag(ci, hash, &frag, &found);
771 		if (found) {
772 			if (mode == USE_ANY_MDS && frag.ndist > 0) {
773 				u8 r;
774 
775 				/* choose a random replica */
776 				get_random_bytes(&r, 1);
777 				r %= frag.ndist;
778 				mds = frag.dist[r];
779 				dout("choose_mds %p %llx.%llx "
780 				     "frag %u mds%d (%d/%d)\n",
781 				     inode, ceph_vinop(inode),
782 				     frag.frag, mds,
783 				     (int)r, frag.ndist);
784 				if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
785 				    CEPH_MDS_STATE_ACTIVE)
786 					return mds;
787 			}
788 
789 			/* since this file/dir wasn't known to be
790 			 * replicated, then we want to look for the
791 			 * authoritative mds. */
792 			mode = USE_AUTH_MDS;
793 			if (frag.mds >= 0) {
794 				/* choose auth mds */
795 				mds = frag.mds;
796 				dout("choose_mds %p %llx.%llx "
797 				     "frag %u mds%d (auth)\n",
798 				     inode, ceph_vinop(inode), frag.frag, mds);
799 				if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
800 				    CEPH_MDS_STATE_ACTIVE)
801 					return mds;
802 			}
803 		}
804 	}
805 
806 	spin_lock(&ci->i_ceph_lock);
807 	cap = NULL;
808 	if (mode == USE_AUTH_MDS)
809 		cap = ci->i_auth_cap;
810 	if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
811 		cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
812 	if (!cap) {
813 		spin_unlock(&ci->i_ceph_lock);
814 		goto random;
815 	}
816 	mds = cap->session->s_mds;
817 	dout("choose_mds %p %llx.%llx mds%d (%scap %p)\n",
818 	     inode, ceph_vinop(inode), mds,
819 	     cap == ci->i_auth_cap ? "auth " : "", cap);
820 	spin_unlock(&ci->i_ceph_lock);
821 	return mds;
822 
823 random:
824 	mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
825 	dout("choose_mds chose random mds%d\n", mds);
826 	return mds;
827 }
828 
829 
830 /*
831  * session messages
832  */
833 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
834 {
835 	struct ceph_msg *msg;
836 	struct ceph_mds_session_head *h;
837 
838 	msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
839 			   false);
840 	if (!msg) {
841 		pr_err("create_session_msg ENOMEM creating msg\n");
842 		return NULL;
843 	}
844 	h = msg->front.iov_base;
845 	h->op = cpu_to_le32(op);
846 	h->seq = cpu_to_le64(seq);
847 
848 	return msg;
849 }
850 
851 /*
852  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
853  * to include additional client metadata fields.
854  */
855 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
856 {
857 	struct ceph_msg *msg;
858 	struct ceph_mds_session_head *h;
859 	int i = -1;
860 	int metadata_bytes = 0;
861 	int metadata_key_count = 0;
862 	struct ceph_options *opt = mdsc->fsc->client->options;
863 	void *p;
864 
865 	const char* metadata[][2] = {
866 		{"hostname", utsname()->nodename},
867 		{"kernel_version", utsname()->release},
868 		{"entity_id", opt->name ? opt->name : ""},
869 		{NULL, NULL}
870 	};
871 
872 	/* Calculate serialized length of metadata */
873 	metadata_bytes = 4;  /* map length */
874 	for (i = 0; metadata[i][0] != NULL; ++i) {
875 		metadata_bytes += 8 + strlen(metadata[i][0]) +
876 			strlen(metadata[i][1]);
877 		metadata_key_count++;
878 	}
879 
880 	/* Allocate the message */
881 	msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + metadata_bytes,
882 			   GFP_NOFS, false);
883 	if (!msg) {
884 		pr_err("create_session_msg ENOMEM creating msg\n");
885 		return NULL;
886 	}
887 	h = msg->front.iov_base;
888 	h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
889 	h->seq = cpu_to_le64(seq);
890 
891 	/*
892 	 * Serialize client metadata into waiting buffer space, using
893 	 * the format that userspace expects for map<string, string>
894 	 *
895 	 * ClientSession messages with metadata are v2
896 	 */
897 	msg->hdr.version = cpu_to_le16(2);
898 	msg->hdr.compat_version = cpu_to_le16(1);
899 
900 	/* The write pointer, following the session_head structure */
901 	p = msg->front.iov_base + sizeof(*h);
902 
903 	/* Number of entries in the map */
904 	ceph_encode_32(&p, metadata_key_count);
905 
906 	/* Two length-prefixed strings for each entry in the map */
907 	for (i = 0; metadata[i][0] != NULL; ++i) {
908 		size_t const key_len = strlen(metadata[i][0]);
909 		size_t const val_len = strlen(metadata[i][1]);
910 
911 		ceph_encode_32(&p, key_len);
912 		memcpy(p, metadata[i][0], key_len);
913 		p += key_len;
914 		ceph_encode_32(&p, val_len);
915 		memcpy(p, metadata[i][1], val_len);
916 		p += val_len;
917 	}
918 
919 	return msg;
920 }
921 
922 /*
923  * send session open request.
924  *
925  * called under mdsc->mutex
926  */
927 static int __open_session(struct ceph_mds_client *mdsc,
928 			  struct ceph_mds_session *session)
929 {
930 	struct ceph_msg *msg;
931 	int mstate;
932 	int mds = session->s_mds;
933 
934 	/* wait for mds to go active? */
935 	mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
936 	dout("open_session to mds%d (%s)\n", mds,
937 	     ceph_mds_state_name(mstate));
938 	session->s_state = CEPH_MDS_SESSION_OPENING;
939 	session->s_renew_requested = jiffies;
940 
941 	/* send connect message */
942 	msg = create_session_open_msg(mdsc, session->s_seq);
943 	if (!msg)
944 		return -ENOMEM;
945 	ceph_con_send(&session->s_con, msg);
946 	return 0;
947 }
948 
949 /*
950  * open sessions for any export targets for the given mds
951  *
952  * called under mdsc->mutex
953  */
954 static struct ceph_mds_session *
955 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
956 {
957 	struct ceph_mds_session *session;
958 
959 	session = __ceph_lookup_mds_session(mdsc, target);
960 	if (!session) {
961 		session = register_session(mdsc, target);
962 		if (IS_ERR(session))
963 			return session;
964 	}
965 	if (session->s_state == CEPH_MDS_SESSION_NEW ||
966 	    session->s_state == CEPH_MDS_SESSION_CLOSING)
967 		__open_session(mdsc, session);
968 
969 	return session;
970 }
971 
972 struct ceph_mds_session *
973 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
974 {
975 	struct ceph_mds_session *session;
976 
977 	dout("open_export_target_session to mds%d\n", target);
978 
979 	mutex_lock(&mdsc->mutex);
980 	session = __open_export_target_session(mdsc, target);
981 	mutex_unlock(&mdsc->mutex);
982 
983 	return session;
984 }
985 
986 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
987 					  struct ceph_mds_session *session)
988 {
989 	struct ceph_mds_info *mi;
990 	struct ceph_mds_session *ts;
991 	int i, mds = session->s_mds;
992 
993 	if (mds >= mdsc->mdsmap->m_max_mds)
994 		return;
995 
996 	mi = &mdsc->mdsmap->m_info[mds];
997 	dout("open_export_target_sessions for mds%d (%d targets)\n",
998 	     session->s_mds, mi->num_export_targets);
999 
1000 	for (i = 0; i < mi->num_export_targets; i++) {
1001 		ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1002 		if (!IS_ERR(ts))
1003 			ceph_put_mds_session(ts);
1004 	}
1005 }
1006 
1007 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1008 					   struct ceph_mds_session *session)
1009 {
1010 	mutex_lock(&mdsc->mutex);
1011 	__open_export_target_sessions(mdsc, session);
1012 	mutex_unlock(&mdsc->mutex);
1013 }
1014 
1015 /*
1016  * session caps
1017  */
1018 
1019 /* caller holds s_cap_lock, we drop it */
1020 static void cleanup_cap_releases(struct ceph_mds_client *mdsc,
1021 				 struct ceph_mds_session *session)
1022 	__releases(session->s_cap_lock)
1023 {
1024 	LIST_HEAD(tmp_list);
1025 	list_splice_init(&session->s_cap_releases, &tmp_list);
1026 	session->s_num_cap_releases = 0;
1027 	spin_unlock(&session->s_cap_lock);
1028 
1029 	dout("cleanup_cap_releases mds%d\n", session->s_mds);
1030 	while (!list_empty(&tmp_list)) {
1031 		struct ceph_cap *cap;
1032 		/* zero out the in-progress message */
1033 		cap = list_first_entry(&tmp_list,
1034 					struct ceph_cap, session_caps);
1035 		list_del(&cap->session_caps);
1036 		ceph_put_cap(mdsc, cap);
1037 	}
1038 }
1039 
1040 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1041 				     struct ceph_mds_session *session)
1042 {
1043 	struct ceph_mds_request *req;
1044 	struct rb_node *p;
1045 
1046 	dout("cleanup_session_requests mds%d\n", session->s_mds);
1047 	mutex_lock(&mdsc->mutex);
1048 	while (!list_empty(&session->s_unsafe)) {
1049 		req = list_first_entry(&session->s_unsafe,
1050 				       struct ceph_mds_request, r_unsafe_item);
1051 		list_del_init(&req->r_unsafe_item);
1052 		pr_warn_ratelimited(" dropping unsafe request %llu\n",
1053 				    req->r_tid);
1054 		__unregister_request(mdsc, req);
1055 	}
1056 	/* zero r_attempts, so kick_requests() will re-send requests */
1057 	p = rb_first(&mdsc->request_tree);
1058 	while (p) {
1059 		req = rb_entry(p, struct ceph_mds_request, r_node);
1060 		p = rb_next(p);
1061 		if (req->r_session &&
1062 		    req->r_session->s_mds == session->s_mds)
1063 			req->r_attempts = 0;
1064 	}
1065 	mutex_unlock(&mdsc->mutex);
1066 }
1067 
1068 /*
1069  * Helper to safely iterate over all caps associated with a session, with
1070  * special care taken to handle a racing __ceph_remove_cap().
1071  *
1072  * Caller must hold session s_mutex.
1073  */
1074 static int iterate_session_caps(struct ceph_mds_session *session,
1075 				 int (*cb)(struct inode *, struct ceph_cap *,
1076 					    void *), void *arg)
1077 {
1078 	struct list_head *p;
1079 	struct ceph_cap *cap;
1080 	struct inode *inode, *last_inode = NULL;
1081 	struct ceph_cap *old_cap = NULL;
1082 	int ret;
1083 
1084 	dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1085 	spin_lock(&session->s_cap_lock);
1086 	p = session->s_caps.next;
1087 	while (p != &session->s_caps) {
1088 		cap = list_entry(p, struct ceph_cap, session_caps);
1089 		inode = igrab(&cap->ci->vfs_inode);
1090 		if (!inode) {
1091 			p = p->next;
1092 			continue;
1093 		}
1094 		session->s_cap_iterator = cap;
1095 		spin_unlock(&session->s_cap_lock);
1096 
1097 		if (last_inode) {
1098 			iput(last_inode);
1099 			last_inode = NULL;
1100 		}
1101 		if (old_cap) {
1102 			ceph_put_cap(session->s_mdsc, old_cap);
1103 			old_cap = NULL;
1104 		}
1105 
1106 		ret = cb(inode, cap, arg);
1107 		last_inode = inode;
1108 
1109 		spin_lock(&session->s_cap_lock);
1110 		p = p->next;
1111 		if (cap->ci == NULL) {
1112 			dout("iterate_session_caps  finishing cap %p removal\n",
1113 			     cap);
1114 			BUG_ON(cap->session != session);
1115 			cap->session = NULL;
1116 			list_del_init(&cap->session_caps);
1117 			session->s_nr_caps--;
1118 			if (cap->queue_release) {
1119 				list_add_tail(&cap->session_caps,
1120 					      &session->s_cap_releases);
1121 				session->s_num_cap_releases++;
1122 			} else {
1123 				old_cap = cap;  /* put_cap it w/o locks held */
1124 			}
1125 		}
1126 		if (ret < 0)
1127 			goto out;
1128 	}
1129 	ret = 0;
1130 out:
1131 	session->s_cap_iterator = NULL;
1132 	spin_unlock(&session->s_cap_lock);
1133 
1134 	iput(last_inode);
1135 	if (old_cap)
1136 		ceph_put_cap(session->s_mdsc, old_cap);
1137 
1138 	return ret;
1139 }
1140 
1141 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1142 				  void *arg)
1143 {
1144 	struct ceph_inode_info *ci = ceph_inode(inode);
1145 	LIST_HEAD(to_remove);
1146 	int drop = 0;
1147 
1148 	dout("removing cap %p, ci is %p, inode is %p\n",
1149 	     cap, ci, &ci->vfs_inode);
1150 	spin_lock(&ci->i_ceph_lock);
1151 	__ceph_remove_cap(cap, false);
1152 	if (!ci->i_auth_cap) {
1153 		struct ceph_cap_flush *cf;
1154 		struct ceph_mds_client *mdsc =
1155 			ceph_sb_to_client(inode->i_sb)->mdsc;
1156 
1157 		while (true) {
1158 			struct rb_node *n = rb_first(&ci->i_cap_flush_tree);
1159 			if (!n)
1160 				break;
1161 			cf = rb_entry(n, struct ceph_cap_flush, i_node);
1162 			rb_erase(&cf->i_node, &ci->i_cap_flush_tree);
1163 			list_add(&cf->list, &to_remove);
1164 		}
1165 
1166 		spin_lock(&mdsc->cap_dirty_lock);
1167 
1168 		list_for_each_entry(cf, &to_remove, list)
1169 			rb_erase(&cf->g_node, &mdsc->cap_flush_tree);
1170 
1171 		if (!list_empty(&ci->i_dirty_item)) {
1172 			pr_warn_ratelimited(
1173 				" dropping dirty %s state for %p %lld\n",
1174 				ceph_cap_string(ci->i_dirty_caps),
1175 				inode, ceph_ino(inode));
1176 			ci->i_dirty_caps = 0;
1177 			list_del_init(&ci->i_dirty_item);
1178 			drop = 1;
1179 		}
1180 		if (!list_empty(&ci->i_flushing_item)) {
1181 			pr_warn_ratelimited(
1182 				" dropping dirty+flushing %s state for %p %lld\n",
1183 				ceph_cap_string(ci->i_flushing_caps),
1184 				inode, ceph_ino(inode));
1185 			ci->i_flushing_caps = 0;
1186 			list_del_init(&ci->i_flushing_item);
1187 			mdsc->num_cap_flushing--;
1188 			drop = 1;
1189 		}
1190 		spin_unlock(&mdsc->cap_dirty_lock);
1191 
1192 		if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1193 			list_add(&ci->i_prealloc_cap_flush->list, &to_remove);
1194 			ci->i_prealloc_cap_flush = NULL;
1195 		}
1196 	}
1197 	spin_unlock(&ci->i_ceph_lock);
1198 	while (!list_empty(&to_remove)) {
1199 		struct ceph_cap_flush *cf;
1200 		cf = list_first_entry(&to_remove,
1201 				      struct ceph_cap_flush, list);
1202 		list_del(&cf->list);
1203 		ceph_free_cap_flush(cf);
1204 	}
1205 	while (drop--)
1206 		iput(inode);
1207 	return 0;
1208 }
1209 
1210 /*
1211  * caller must hold session s_mutex
1212  */
1213 static void remove_session_caps(struct ceph_mds_session *session)
1214 {
1215 	dout("remove_session_caps on %p\n", session);
1216 	iterate_session_caps(session, remove_session_caps_cb, NULL);
1217 
1218 	spin_lock(&session->s_cap_lock);
1219 	if (session->s_nr_caps > 0) {
1220 		struct super_block *sb = session->s_mdsc->fsc->sb;
1221 		struct inode *inode;
1222 		struct ceph_cap *cap, *prev = NULL;
1223 		struct ceph_vino vino;
1224 		/*
1225 		 * iterate_session_caps() skips inodes that are being
1226 		 * deleted, we need to wait until deletions are complete.
1227 		 * __wait_on_freeing_inode() is designed for the job,
1228 		 * but it is not exported, so use lookup inode function
1229 		 * to access it.
1230 		 */
1231 		while (!list_empty(&session->s_caps)) {
1232 			cap = list_entry(session->s_caps.next,
1233 					 struct ceph_cap, session_caps);
1234 			if (cap == prev)
1235 				break;
1236 			prev = cap;
1237 			vino = cap->ci->i_vino;
1238 			spin_unlock(&session->s_cap_lock);
1239 
1240 			inode = ceph_find_inode(sb, vino);
1241 			iput(inode);
1242 
1243 			spin_lock(&session->s_cap_lock);
1244 		}
1245 	}
1246 
1247 	// drop cap expires and unlock s_cap_lock
1248 	cleanup_cap_releases(session->s_mdsc, session);
1249 
1250 	BUG_ON(session->s_nr_caps > 0);
1251 	BUG_ON(!list_empty(&session->s_cap_flushing));
1252 }
1253 
1254 /*
1255  * wake up any threads waiting on this session's caps.  if the cap is
1256  * old (didn't get renewed on the client reconnect), remove it now.
1257  *
1258  * caller must hold s_mutex.
1259  */
1260 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1261 			      void *arg)
1262 {
1263 	struct ceph_inode_info *ci = ceph_inode(inode);
1264 
1265 	wake_up_all(&ci->i_cap_wq);
1266 	if (arg) {
1267 		spin_lock(&ci->i_ceph_lock);
1268 		ci->i_wanted_max_size = 0;
1269 		ci->i_requested_max_size = 0;
1270 		spin_unlock(&ci->i_ceph_lock);
1271 	}
1272 	return 0;
1273 }
1274 
1275 static void wake_up_session_caps(struct ceph_mds_session *session,
1276 				 int reconnect)
1277 {
1278 	dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1279 	iterate_session_caps(session, wake_up_session_cb,
1280 			     (void *)(unsigned long)reconnect);
1281 }
1282 
1283 /*
1284  * Send periodic message to MDS renewing all currently held caps.  The
1285  * ack will reset the expiration for all caps from this session.
1286  *
1287  * caller holds s_mutex
1288  */
1289 static int send_renew_caps(struct ceph_mds_client *mdsc,
1290 			   struct ceph_mds_session *session)
1291 {
1292 	struct ceph_msg *msg;
1293 	int state;
1294 
1295 	if (time_after_eq(jiffies, session->s_cap_ttl) &&
1296 	    time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1297 		pr_info("mds%d caps stale\n", session->s_mds);
1298 	session->s_renew_requested = jiffies;
1299 
1300 	/* do not try to renew caps until a recovering mds has reconnected
1301 	 * with its clients. */
1302 	state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1303 	if (state < CEPH_MDS_STATE_RECONNECT) {
1304 		dout("send_renew_caps ignoring mds%d (%s)\n",
1305 		     session->s_mds, ceph_mds_state_name(state));
1306 		return 0;
1307 	}
1308 
1309 	dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1310 		ceph_mds_state_name(state));
1311 	msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1312 				 ++session->s_renew_seq);
1313 	if (!msg)
1314 		return -ENOMEM;
1315 	ceph_con_send(&session->s_con, msg);
1316 	return 0;
1317 }
1318 
1319 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1320 			     struct ceph_mds_session *session, u64 seq)
1321 {
1322 	struct ceph_msg *msg;
1323 
1324 	dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1325 	     session->s_mds, ceph_session_state_name(session->s_state), seq);
1326 	msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1327 	if (!msg)
1328 		return -ENOMEM;
1329 	ceph_con_send(&session->s_con, msg);
1330 	return 0;
1331 }
1332 
1333 
1334 /*
1335  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1336  *
1337  * Called under session->s_mutex
1338  */
1339 static void renewed_caps(struct ceph_mds_client *mdsc,
1340 			 struct ceph_mds_session *session, int is_renew)
1341 {
1342 	int was_stale;
1343 	int wake = 0;
1344 
1345 	spin_lock(&session->s_cap_lock);
1346 	was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1347 
1348 	session->s_cap_ttl = session->s_renew_requested +
1349 		mdsc->mdsmap->m_session_timeout*HZ;
1350 
1351 	if (was_stale) {
1352 		if (time_before(jiffies, session->s_cap_ttl)) {
1353 			pr_info("mds%d caps renewed\n", session->s_mds);
1354 			wake = 1;
1355 		} else {
1356 			pr_info("mds%d caps still stale\n", session->s_mds);
1357 		}
1358 	}
1359 	dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1360 	     session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1361 	     time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1362 	spin_unlock(&session->s_cap_lock);
1363 
1364 	if (wake)
1365 		wake_up_session_caps(session, 0);
1366 }
1367 
1368 /*
1369  * send a session close request
1370  */
1371 static int request_close_session(struct ceph_mds_client *mdsc,
1372 				 struct ceph_mds_session *session)
1373 {
1374 	struct ceph_msg *msg;
1375 
1376 	dout("request_close_session mds%d state %s seq %lld\n",
1377 	     session->s_mds, ceph_session_state_name(session->s_state),
1378 	     session->s_seq);
1379 	msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1380 	if (!msg)
1381 		return -ENOMEM;
1382 	ceph_con_send(&session->s_con, msg);
1383 	return 0;
1384 }
1385 
1386 /*
1387  * Called with s_mutex held.
1388  */
1389 static int __close_session(struct ceph_mds_client *mdsc,
1390 			 struct ceph_mds_session *session)
1391 {
1392 	if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1393 		return 0;
1394 	session->s_state = CEPH_MDS_SESSION_CLOSING;
1395 	return request_close_session(mdsc, session);
1396 }
1397 
1398 /*
1399  * Trim old(er) caps.
1400  *
1401  * Because we can't cache an inode without one or more caps, we do
1402  * this indirectly: if a cap is unused, we prune its aliases, at which
1403  * point the inode will hopefully get dropped to.
1404  *
1405  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1406  * memory pressure from the MDS, though, so it needn't be perfect.
1407  */
1408 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1409 {
1410 	struct ceph_mds_session *session = arg;
1411 	struct ceph_inode_info *ci = ceph_inode(inode);
1412 	int used, wanted, oissued, mine;
1413 
1414 	if (session->s_trim_caps <= 0)
1415 		return -1;
1416 
1417 	spin_lock(&ci->i_ceph_lock);
1418 	mine = cap->issued | cap->implemented;
1419 	used = __ceph_caps_used(ci);
1420 	wanted = __ceph_caps_file_wanted(ci);
1421 	oissued = __ceph_caps_issued_other(ci, cap);
1422 
1423 	dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1424 	     inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1425 	     ceph_cap_string(used), ceph_cap_string(wanted));
1426 	if (cap == ci->i_auth_cap) {
1427 		if (ci->i_dirty_caps || ci->i_flushing_caps ||
1428 		    !list_empty(&ci->i_cap_snaps))
1429 			goto out;
1430 		if ((used | wanted) & CEPH_CAP_ANY_WR)
1431 			goto out;
1432 	}
1433 	if ((used | wanted) & ~oissued & mine)
1434 		goto out;   /* we need these caps */
1435 
1436 	session->s_trim_caps--;
1437 	if (oissued) {
1438 		/* we aren't the only cap.. just remove us */
1439 		__ceph_remove_cap(cap, true);
1440 	} else {
1441 		/* try to drop referring dentries */
1442 		spin_unlock(&ci->i_ceph_lock);
1443 		d_prune_aliases(inode);
1444 		dout("trim_caps_cb %p cap %p  pruned, count now %d\n",
1445 		     inode, cap, atomic_read(&inode->i_count));
1446 		return 0;
1447 	}
1448 
1449 out:
1450 	spin_unlock(&ci->i_ceph_lock);
1451 	return 0;
1452 }
1453 
1454 /*
1455  * Trim session cap count down to some max number.
1456  */
1457 static int trim_caps(struct ceph_mds_client *mdsc,
1458 		     struct ceph_mds_session *session,
1459 		     int max_caps)
1460 {
1461 	int trim_caps = session->s_nr_caps - max_caps;
1462 
1463 	dout("trim_caps mds%d start: %d / %d, trim %d\n",
1464 	     session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1465 	if (trim_caps > 0) {
1466 		session->s_trim_caps = trim_caps;
1467 		iterate_session_caps(session, trim_caps_cb, session);
1468 		dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1469 		     session->s_mds, session->s_nr_caps, max_caps,
1470 			trim_caps - session->s_trim_caps);
1471 		session->s_trim_caps = 0;
1472 	}
1473 
1474 	ceph_send_cap_releases(mdsc, session);
1475 	return 0;
1476 }
1477 
1478 static int check_capsnap_flush(struct ceph_inode_info *ci,
1479 			       u64 want_snap_seq)
1480 {
1481 	int ret = 1;
1482 	spin_lock(&ci->i_ceph_lock);
1483 	if (want_snap_seq > 0 && !list_empty(&ci->i_cap_snaps)) {
1484 		struct ceph_cap_snap *capsnap =
1485 			list_first_entry(&ci->i_cap_snaps,
1486 					 struct ceph_cap_snap, ci_item);
1487 		ret = capsnap->follows >= want_snap_seq;
1488 	}
1489 	spin_unlock(&ci->i_ceph_lock);
1490 	return ret;
1491 }
1492 
1493 static int check_caps_flush(struct ceph_mds_client *mdsc,
1494 			    u64 want_flush_tid)
1495 {
1496 	struct rb_node *n;
1497 	struct ceph_cap_flush *cf;
1498 	int ret = 1;
1499 
1500 	spin_lock(&mdsc->cap_dirty_lock);
1501 	n = rb_first(&mdsc->cap_flush_tree);
1502 	cf = n ? rb_entry(n, struct ceph_cap_flush, g_node) : NULL;
1503 	if (cf && cf->tid <= want_flush_tid) {
1504 		dout("check_caps_flush still flushing tid %llu <= %llu\n",
1505 		     cf->tid, want_flush_tid);
1506 		ret = 0;
1507 	}
1508 	spin_unlock(&mdsc->cap_dirty_lock);
1509 	return ret;
1510 }
1511 
1512 /*
1513  * flush all dirty inode data to disk.
1514  *
1515  * returns true if we've flushed through want_flush_tid
1516  */
1517 static void wait_caps_flush(struct ceph_mds_client *mdsc,
1518 			    u64 want_flush_tid, u64 want_snap_seq)
1519 {
1520 	int mds;
1521 
1522 	dout("check_caps_flush want %llu snap want %llu\n",
1523 	     want_flush_tid, want_snap_seq);
1524 	mutex_lock(&mdsc->mutex);
1525 	for (mds = 0; mds < mdsc->max_sessions; ) {
1526 		struct ceph_mds_session *session = mdsc->sessions[mds];
1527 		struct inode *inode = NULL;
1528 
1529 		if (!session) {
1530 			mds++;
1531 			continue;
1532 		}
1533 		get_session(session);
1534 		mutex_unlock(&mdsc->mutex);
1535 
1536 		mutex_lock(&session->s_mutex);
1537 		if (!list_empty(&session->s_cap_snaps_flushing)) {
1538 			struct ceph_cap_snap *capsnap =
1539 				list_first_entry(&session->s_cap_snaps_flushing,
1540 						 struct ceph_cap_snap,
1541 						 flushing_item);
1542 			struct ceph_inode_info *ci = capsnap->ci;
1543 			if (!check_capsnap_flush(ci, want_snap_seq)) {
1544 				dout("check_cap_flush still flushing snap %p "
1545 				     "follows %lld <= %lld to mds%d\n",
1546 				     &ci->vfs_inode, capsnap->follows,
1547 				     want_snap_seq, mds);
1548 				inode = igrab(&ci->vfs_inode);
1549 			}
1550 		}
1551 		mutex_unlock(&session->s_mutex);
1552 		ceph_put_mds_session(session);
1553 
1554 		if (inode) {
1555 			wait_event(mdsc->cap_flushing_wq,
1556 				   check_capsnap_flush(ceph_inode(inode),
1557 						       want_snap_seq));
1558 			iput(inode);
1559 		} else {
1560 			mds++;
1561 		}
1562 
1563 		mutex_lock(&mdsc->mutex);
1564 	}
1565 	mutex_unlock(&mdsc->mutex);
1566 
1567 	wait_event(mdsc->cap_flushing_wq,
1568 		   check_caps_flush(mdsc, want_flush_tid));
1569 
1570 	dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
1571 }
1572 
1573 /*
1574  * called under s_mutex
1575  */
1576 void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1577 			    struct ceph_mds_session *session)
1578 {
1579 	struct ceph_msg *msg = NULL;
1580 	struct ceph_mds_cap_release *head;
1581 	struct ceph_mds_cap_item *item;
1582 	struct ceph_cap *cap;
1583 	LIST_HEAD(tmp_list);
1584 	int num_cap_releases;
1585 
1586 	spin_lock(&session->s_cap_lock);
1587 again:
1588 	list_splice_init(&session->s_cap_releases, &tmp_list);
1589 	num_cap_releases = session->s_num_cap_releases;
1590 	session->s_num_cap_releases = 0;
1591 	spin_unlock(&session->s_cap_lock);
1592 
1593 	while (!list_empty(&tmp_list)) {
1594 		if (!msg) {
1595 			msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
1596 					PAGE_CACHE_SIZE, GFP_NOFS, false);
1597 			if (!msg)
1598 				goto out_err;
1599 			head = msg->front.iov_base;
1600 			head->num = cpu_to_le32(0);
1601 			msg->front.iov_len = sizeof(*head);
1602 		}
1603 		cap = list_first_entry(&tmp_list, struct ceph_cap,
1604 					session_caps);
1605 		list_del(&cap->session_caps);
1606 		num_cap_releases--;
1607 
1608 		head = msg->front.iov_base;
1609 		le32_add_cpu(&head->num, 1);
1610 		item = msg->front.iov_base + msg->front.iov_len;
1611 		item->ino = cpu_to_le64(cap->cap_ino);
1612 		item->cap_id = cpu_to_le64(cap->cap_id);
1613 		item->migrate_seq = cpu_to_le32(cap->mseq);
1614 		item->seq = cpu_to_le32(cap->issue_seq);
1615 		msg->front.iov_len += sizeof(*item);
1616 
1617 		ceph_put_cap(mdsc, cap);
1618 
1619 		if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
1620 			msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1621 			dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1622 			ceph_con_send(&session->s_con, msg);
1623 			msg = NULL;
1624 		}
1625 	}
1626 
1627 	BUG_ON(num_cap_releases != 0);
1628 
1629 	spin_lock(&session->s_cap_lock);
1630 	if (!list_empty(&session->s_cap_releases))
1631 		goto again;
1632 	spin_unlock(&session->s_cap_lock);
1633 
1634 	if (msg) {
1635 		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1636 		dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
1637 		ceph_con_send(&session->s_con, msg);
1638 	}
1639 	return;
1640 out_err:
1641 	pr_err("send_cap_releases mds%d, failed to allocate message\n",
1642 		session->s_mds);
1643 	spin_lock(&session->s_cap_lock);
1644 	list_splice(&tmp_list, &session->s_cap_releases);
1645 	session->s_num_cap_releases += num_cap_releases;
1646 	spin_unlock(&session->s_cap_lock);
1647 }
1648 
1649 /*
1650  * requests
1651  */
1652 
1653 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
1654 				    struct inode *dir)
1655 {
1656 	struct ceph_inode_info *ci = ceph_inode(dir);
1657 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1658 	struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
1659 	size_t size = sizeof(*rinfo->dir_in) + sizeof(*rinfo->dir_dname_len) +
1660 		      sizeof(*rinfo->dir_dname) + sizeof(*rinfo->dir_dlease);
1661 	int order, num_entries;
1662 
1663 	spin_lock(&ci->i_ceph_lock);
1664 	num_entries = ci->i_files + ci->i_subdirs;
1665 	spin_unlock(&ci->i_ceph_lock);
1666 	num_entries = max(num_entries, 1);
1667 	num_entries = min(num_entries, opt->max_readdir);
1668 
1669 	order = get_order(size * num_entries);
1670 	while (order >= 0) {
1671 		rinfo->dir_in = (void*)__get_free_pages(GFP_KERNEL |
1672 							__GFP_NOWARN,
1673 							order);
1674 		if (rinfo->dir_in)
1675 			break;
1676 		order--;
1677 	}
1678 	if (!rinfo->dir_in)
1679 		return -ENOMEM;
1680 
1681 	num_entries = (PAGE_SIZE << order) / size;
1682 	num_entries = min(num_entries, opt->max_readdir);
1683 
1684 	rinfo->dir_buf_size = PAGE_SIZE << order;
1685 	req->r_num_caps = num_entries + 1;
1686 	req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
1687 	req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
1688 	return 0;
1689 }
1690 
1691 /*
1692  * Create an mds request.
1693  */
1694 struct ceph_mds_request *
1695 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
1696 {
1697 	struct ceph_mds_request *req = kzalloc(sizeof(*req), GFP_NOFS);
1698 
1699 	if (!req)
1700 		return ERR_PTR(-ENOMEM);
1701 
1702 	mutex_init(&req->r_fill_mutex);
1703 	req->r_mdsc = mdsc;
1704 	req->r_started = jiffies;
1705 	req->r_resend_mds = -1;
1706 	INIT_LIST_HEAD(&req->r_unsafe_dir_item);
1707 	req->r_fmode = -1;
1708 	kref_init(&req->r_kref);
1709 	INIT_LIST_HEAD(&req->r_wait);
1710 	init_completion(&req->r_completion);
1711 	init_completion(&req->r_safe_completion);
1712 	INIT_LIST_HEAD(&req->r_unsafe_item);
1713 
1714 	req->r_stamp = CURRENT_TIME;
1715 
1716 	req->r_op = op;
1717 	req->r_direct_mode = mode;
1718 	return req;
1719 }
1720 
1721 /*
1722  * return oldest (lowest) request, tid in request tree, 0 if none.
1723  *
1724  * called under mdsc->mutex.
1725  */
1726 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
1727 {
1728 	if (RB_EMPTY_ROOT(&mdsc->request_tree))
1729 		return NULL;
1730 	return rb_entry(rb_first(&mdsc->request_tree),
1731 			struct ceph_mds_request, r_node);
1732 }
1733 
1734 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
1735 {
1736 	return mdsc->oldest_tid;
1737 }
1738 
1739 /*
1740  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
1741  * on build_path_from_dentry in fs/cifs/dir.c.
1742  *
1743  * If @stop_on_nosnap, generate path relative to the first non-snapped
1744  * inode.
1745  *
1746  * Encode hidden .snap dirs as a double /, i.e.
1747  *   foo/.snap/bar -> foo//bar
1748  */
1749 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *base,
1750 			   int stop_on_nosnap)
1751 {
1752 	struct dentry *temp;
1753 	char *path;
1754 	int len, pos;
1755 	unsigned seq;
1756 
1757 	if (dentry == NULL)
1758 		return ERR_PTR(-EINVAL);
1759 
1760 retry:
1761 	len = 0;
1762 	seq = read_seqbegin(&rename_lock);
1763 	rcu_read_lock();
1764 	for (temp = dentry; !IS_ROOT(temp);) {
1765 		struct inode *inode = d_inode(temp);
1766 		if (inode && ceph_snap(inode) == CEPH_SNAPDIR)
1767 			len++;  /* slash only */
1768 		else if (stop_on_nosnap && inode &&
1769 			 ceph_snap(inode) == CEPH_NOSNAP)
1770 			break;
1771 		else
1772 			len += 1 + temp->d_name.len;
1773 		temp = temp->d_parent;
1774 	}
1775 	rcu_read_unlock();
1776 	if (len)
1777 		len--;  /* no leading '/' */
1778 
1779 	path = kmalloc(len+1, GFP_NOFS);
1780 	if (path == NULL)
1781 		return ERR_PTR(-ENOMEM);
1782 	pos = len;
1783 	path[pos] = 0;	/* trailing null */
1784 	rcu_read_lock();
1785 	for (temp = dentry; !IS_ROOT(temp) && pos != 0; ) {
1786 		struct inode *inode;
1787 
1788 		spin_lock(&temp->d_lock);
1789 		inode = d_inode(temp);
1790 		if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1791 			dout("build_path path+%d: %p SNAPDIR\n",
1792 			     pos, temp);
1793 		} else if (stop_on_nosnap && inode &&
1794 			   ceph_snap(inode) == CEPH_NOSNAP) {
1795 			spin_unlock(&temp->d_lock);
1796 			break;
1797 		} else {
1798 			pos -= temp->d_name.len;
1799 			if (pos < 0) {
1800 				spin_unlock(&temp->d_lock);
1801 				break;
1802 			}
1803 			strncpy(path + pos, temp->d_name.name,
1804 				temp->d_name.len);
1805 		}
1806 		spin_unlock(&temp->d_lock);
1807 		if (pos)
1808 			path[--pos] = '/';
1809 		temp = temp->d_parent;
1810 	}
1811 	rcu_read_unlock();
1812 	if (pos != 0 || read_seqretry(&rename_lock, seq)) {
1813 		pr_err("build_path did not end path lookup where "
1814 		       "expected, namelen is %d, pos is %d\n", len, pos);
1815 		/* presumably this is only possible if racing with a
1816 		   rename of one of the parent directories (we can not
1817 		   lock the dentries above us to prevent this, but
1818 		   retrying should be harmless) */
1819 		kfree(path);
1820 		goto retry;
1821 	}
1822 
1823 	*base = ceph_ino(d_inode(temp));
1824 	*plen = len;
1825 	dout("build_path on %p %d built %llx '%.*s'\n",
1826 	     dentry, d_count(dentry), *base, len, path);
1827 	return path;
1828 }
1829 
1830 static int build_dentry_path(struct dentry *dentry,
1831 			     const char **ppath, int *ppathlen, u64 *pino,
1832 			     int *pfreepath)
1833 {
1834 	char *path;
1835 
1836 	if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP) {
1837 		*pino = ceph_ino(d_inode(dentry->d_parent));
1838 		*ppath = dentry->d_name.name;
1839 		*ppathlen = dentry->d_name.len;
1840 		return 0;
1841 	}
1842 	path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1843 	if (IS_ERR(path))
1844 		return PTR_ERR(path);
1845 	*ppath = path;
1846 	*pfreepath = 1;
1847 	return 0;
1848 }
1849 
1850 static int build_inode_path(struct inode *inode,
1851 			    const char **ppath, int *ppathlen, u64 *pino,
1852 			    int *pfreepath)
1853 {
1854 	struct dentry *dentry;
1855 	char *path;
1856 
1857 	if (ceph_snap(inode) == CEPH_NOSNAP) {
1858 		*pino = ceph_ino(inode);
1859 		*ppathlen = 0;
1860 		return 0;
1861 	}
1862 	dentry = d_find_alias(inode);
1863 	path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
1864 	dput(dentry);
1865 	if (IS_ERR(path))
1866 		return PTR_ERR(path);
1867 	*ppath = path;
1868 	*pfreepath = 1;
1869 	return 0;
1870 }
1871 
1872 /*
1873  * request arguments may be specified via an inode *, a dentry *, or
1874  * an explicit ino+path.
1875  */
1876 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
1877 				  const char *rpath, u64 rino,
1878 				  const char **ppath, int *pathlen,
1879 				  u64 *ino, int *freepath)
1880 {
1881 	int r = 0;
1882 
1883 	if (rinode) {
1884 		r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
1885 		dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
1886 		     ceph_snap(rinode));
1887 	} else if (rdentry) {
1888 		r = build_dentry_path(rdentry, ppath, pathlen, ino, freepath);
1889 		dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
1890 		     *ppath);
1891 	} else if (rpath || rino) {
1892 		*ino = rino;
1893 		*ppath = rpath;
1894 		*pathlen = rpath ? strlen(rpath) : 0;
1895 		dout(" path %.*s\n", *pathlen, rpath);
1896 	}
1897 
1898 	return r;
1899 }
1900 
1901 /*
1902  * called under mdsc->mutex
1903  */
1904 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
1905 					       struct ceph_mds_request *req,
1906 					       int mds, bool drop_cap_releases)
1907 {
1908 	struct ceph_msg *msg;
1909 	struct ceph_mds_request_head *head;
1910 	const char *path1 = NULL;
1911 	const char *path2 = NULL;
1912 	u64 ino1 = 0, ino2 = 0;
1913 	int pathlen1 = 0, pathlen2 = 0;
1914 	int freepath1 = 0, freepath2 = 0;
1915 	int len;
1916 	u16 releases;
1917 	void *p, *end;
1918 	int ret;
1919 
1920 	ret = set_request_path_attr(req->r_inode, req->r_dentry,
1921 			      req->r_path1, req->r_ino1.ino,
1922 			      &path1, &pathlen1, &ino1, &freepath1);
1923 	if (ret < 0) {
1924 		msg = ERR_PTR(ret);
1925 		goto out;
1926 	}
1927 
1928 	ret = set_request_path_attr(NULL, req->r_old_dentry,
1929 			      req->r_path2, req->r_ino2.ino,
1930 			      &path2, &pathlen2, &ino2, &freepath2);
1931 	if (ret < 0) {
1932 		msg = ERR_PTR(ret);
1933 		goto out_free1;
1934 	}
1935 
1936 	len = sizeof(*head) +
1937 		pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
1938 		sizeof(struct timespec);
1939 
1940 	/* calculate (max) length for cap releases */
1941 	len += sizeof(struct ceph_mds_request_release) *
1942 		(!!req->r_inode_drop + !!req->r_dentry_drop +
1943 		 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
1944 	if (req->r_dentry_drop)
1945 		len += req->r_dentry->d_name.len;
1946 	if (req->r_old_dentry_drop)
1947 		len += req->r_old_dentry->d_name.len;
1948 
1949 	msg = ceph_msg_new(CEPH_MSG_CLIENT_REQUEST, len, GFP_NOFS, false);
1950 	if (!msg) {
1951 		msg = ERR_PTR(-ENOMEM);
1952 		goto out_free2;
1953 	}
1954 
1955 	msg->hdr.version = cpu_to_le16(2);
1956 	msg->hdr.tid = cpu_to_le64(req->r_tid);
1957 
1958 	head = msg->front.iov_base;
1959 	p = msg->front.iov_base + sizeof(*head);
1960 	end = msg->front.iov_base + msg->front.iov_len;
1961 
1962 	head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
1963 	head->op = cpu_to_le32(req->r_op);
1964 	head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
1965 	head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
1966 	head->args = req->r_args;
1967 
1968 	ceph_encode_filepath(&p, end, ino1, path1);
1969 	ceph_encode_filepath(&p, end, ino2, path2);
1970 
1971 	/* make note of release offset, in case we need to replay */
1972 	req->r_request_release_offset = p - msg->front.iov_base;
1973 
1974 	/* cap releases */
1975 	releases = 0;
1976 	if (req->r_inode_drop)
1977 		releases += ceph_encode_inode_release(&p,
1978 		      req->r_inode ? req->r_inode : d_inode(req->r_dentry),
1979 		      mds, req->r_inode_drop, req->r_inode_unless, 0);
1980 	if (req->r_dentry_drop)
1981 		releases += ceph_encode_dentry_release(&p, req->r_dentry,
1982 		       mds, req->r_dentry_drop, req->r_dentry_unless);
1983 	if (req->r_old_dentry_drop)
1984 		releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
1985 		       mds, req->r_old_dentry_drop, req->r_old_dentry_unless);
1986 	if (req->r_old_inode_drop)
1987 		releases += ceph_encode_inode_release(&p,
1988 		      d_inode(req->r_old_dentry),
1989 		      mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
1990 
1991 	if (drop_cap_releases) {
1992 		releases = 0;
1993 		p = msg->front.iov_base + req->r_request_release_offset;
1994 	}
1995 
1996 	head->num_releases = cpu_to_le16(releases);
1997 
1998 	/* time stamp */
1999 	{
2000 		struct ceph_timespec ts;
2001 		ceph_encode_timespec(&ts, &req->r_stamp);
2002 		ceph_encode_copy(&p, &ts, sizeof(ts));
2003 	}
2004 
2005 	BUG_ON(p > end);
2006 	msg->front.iov_len = p - msg->front.iov_base;
2007 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2008 
2009 	if (req->r_pagelist) {
2010 		struct ceph_pagelist *pagelist = req->r_pagelist;
2011 		atomic_inc(&pagelist->refcnt);
2012 		ceph_msg_data_add_pagelist(msg, pagelist);
2013 		msg->hdr.data_len = cpu_to_le32(pagelist->length);
2014 	} else {
2015 		msg->hdr.data_len = 0;
2016 	}
2017 
2018 	msg->hdr.data_off = cpu_to_le16(0);
2019 
2020 out_free2:
2021 	if (freepath2)
2022 		kfree((char *)path2);
2023 out_free1:
2024 	if (freepath1)
2025 		kfree((char *)path1);
2026 out:
2027 	return msg;
2028 }
2029 
2030 /*
2031  * called under mdsc->mutex if error, under no mutex if
2032  * success.
2033  */
2034 static void complete_request(struct ceph_mds_client *mdsc,
2035 			     struct ceph_mds_request *req)
2036 {
2037 	if (req->r_callback)
2038 		req->r_callback(mdsc, req);
2039 	else
2040 		complete_all(&req->r_completion);
2041 }
2042 
2043 /*
2044  * called under mdsc->mutex
2045  */
2046 static int __prepare_send_request(struct ceph_mds_client *mdsc,
2047 				  struct ceph_mds_request *req,
2048 				  int mds, bool drop_cap_releases)
2049 {
2050 	struct ceph_mds_request_head *rhead;
2051 	struct ceph_msg *msg;
2052 	int flags = 0;
2053 
2054 	req->r_attempts++;
2055 	if (req->r_inode) {
2056 		struct ceph_cap *cap =
2057 			ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2058 
2059 		if (cap)
2060 			req->r_sent_on_mseq = cap->mseq;
2061 		else
2062 			req->r_sent_on_mseq = -1;
2063 	}
2064 	dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2065 	     req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2066 
2067 	if (req->r_got_unsafe) {
2068 		void *p;
2069 		/*
2070 		 * Replay.  Do not regenerate message (and rebuild
2071 		 * paths, etc.); just use the original message.
2072 		 * Rebuilding paths will break for renames because
2073 		 * d_move mangles the src name.
2074 		 */
2075 		msg = req->r_request;
2076 		rhead = msg->front.iov_base;
2077 
2078 		flags = le32_to_cpu(rhead->flags);
2079 		flags |= CEPH_MDS_FLAG_REPLAY;
2080 		rhead->flags = cpu_to_le32(flags);
2081 
2082 		if (req->r_target_inode)
2083 			rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2084 
2085 		rhead->num_retry = req->r_attempts - 1;
2086 
2087 		/* remove cap/dentry releases from message */
2088 		rhead->num_releases = 0;
2089 
2090 		/* time stamp */
2091 		p = msg->front.iov_base + req->r_request_release_offset;
2092 		{
2093 			struct ceph_timespec ts;
2094 			ceph_encode_timespec(&ts, &req->r_stamp);
2095 			ceph_encode_copy(&p, &ts, sizeof(ts));
2096 		}
2097 
2098 		msg->front.iov_len = p - msg->front.iov_base;
2099 		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2100 		return 0;
2101 	}
2102 
2103 	if (req->r_request) {
2104 		ceph_msg_put(req->r_request);
2105 		req->r_request = NULL;
2106 	}
2107 	msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2108 	if (IS_ERR(msg)) {
2109 		req->r_err = PTR_ERR(msg);
2110 		complete_request(mdsc, req);
2111 		return PTR_ERR(msg);
2112 	}
2113 	req->r_request = msg;
2114 
2115 	rhead = msg->front.iov_base;
2116 	rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2117 	if (req->r_got_unsafe)
2118 		flags |= CEPH_MDS_FLAG_REPLAY;
2119 	if (req->r_locked_dir)
2120 		flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2121 	rhead->flags = cpu_to_le32(flags);
2122 	rhead->num_fwd = req->r_num_fwd;
2123 	rhead->num_retry = req->r_attempts - 1;
2124 	rhead->ino = 0;
2125 
2126 	dout(" r_locked_dir = %p\n", req->r_locked_dir);
2127 	return 0;
2128 }
2129 
2130 /*
2131  * send request, or put it on the appropriate wait list.
2132  */
2133 static int __do_request(struct ceph_mds_client *mdsc,
2134 			struct ceph_mds_request *req)
2135 {
2136 	struct ceph_mds_session *session = NULL;
2137 	int mds = -1;
2138 	int err = -EAGAIN;
2139 
2140 	if (req->r_err || req->r_got_result) {
2141 		if (req->r_aborted)
2142 			__unregister_request(mdsc, req);
2143 		goto out;
2144 	}
2145 
2146 	if (req->r_timeout &&
2147 	    time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2148 		dout("do_request timed out\n");
2149 		err = -EIO;
2150 		goto finish;
2151 	}
2152 
2153 	put_request_session(req);
2154 
2155 	mds = __choose_mds(mdsc, req);
2156 	if (mds < 0 ||
2157 	    ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2158 		dout("do_request no mds or not active, waiting for map\n");
2159 		list_add(&req->r_wait, &mdsc->waiting_for_map);
2160 		goto out;
2161 	}
2162 
2163 	/* get, open session */
2164 	session = __ceph_lookup_mds_session(mdsc, mds);
2165 	if (!session) {
2166 		session = register_session(mdsc, mds);
2167 		if (IS_ERR(session)) {
2168 			err = PTR_ERR(session);
2169 			goto finish;
2170 		}
2171 	}
2172 	req->r_session = get_session(session);
2173 
2174 	dout("do_request mds%d session %p state %s\n", mds, session,
2175 	     ceph_session_state_name(session->s_state));
2176 	if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2177 	    session->s_state != CEPH_MDS_SESSION_HUNG) {
2178 		if (session->s_state == CEPH_MDS_SESSION_NEW ||
2179 		    session->s_state == CEPH_MDS_SESSION_CLOSING)
2180 			__open_session(mdsc, session);
2181 		list_add(&req->r_wait, &session->s_waiting);
2182 		goto out_session;
2183 	}
2184 
2185 	/* send request */
2186 	req->r_resend_mds = -1;   /* forget any previous mds hint */
2187 
2188 	if (req->r_request_started == 0)   /* note request start time */
2189 		req->r_request_started = jiffies;
2190 
2191 	err = __prepare_send_request(mdsc, req, mds, false);
2192 	if (!err) {
2193 		ceph_msg_get(req->r_request);
2194 		ceph_con_send(&session->s_con, req->r_request);
2195 	}
2196 
2197 out_session:
2198 	ceph_put_mds_session(session);
2199 out:
2200 	return err;
2201 
2202 finish:
2203 	req->r_err = err;
2204 	complete_request(mdsc, req);
2205 	goto out;
2206 }
2207 
2208 /*
2209  * called under mdsc->mutex
2210  */
2211 static void __wake_requests(struct ceph_mds_client *mdsc,
2212 			    struct list_head *head)
2213 {
2214 	struct ceph_mds_request *req;
2215 	LIST_HEAD(tmp_list);
2216 
2217 	list_splice_init(head, &tmp_list);
2218 
2219 	while (!list_empty(&tmp_list)) {
2220 		req = list_entry(tmp_list.next,
2221 				 struct ceph_mds_request, r_wait);
2222 		list_del_init(&req->r_wait);
2223 		dout(" wake request %p tid %llu\n", req, req->r_tid);
2224 		__do_request(mdsc, req);
2225 	}
2226 }
2227 
2228 /*
2229  * Wake up threads with requests pending for @mds, so that they can
2230  * resubmit their requests to a possibly different mds.
2231  */
2232 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2233 {
2234 	struct ceph_mds_request *req;
2235 	struct rb_node *p = rb_first(&mdsc->request_tree);
2236 
2237 	dout("kick_requests mds%d\n", mds);
2238 	while (p) {
2239 		req = rb_entry(p, struct ceph_mds_request, r_node);
2240 		p = rb_next(p);
2241 		if (req->r_got_unsafe)
2242 			continue;
2243 		if (req->r_attempts > 0)
2244 			continue; /* only new requests */
2245 		if (req->r_session &&
2246 		    req->r_session->s_mds == mds) {
2247 			dout(" kicking tid %llu\n", req->r_tid);
2248 			list_del_init(&req->r_wait);
2249 			__do_request(mdsc, req);
2250 		}
2251 	}
2252 }
2253 
2254 void ceph_mdsc_submit_request(struct ceph_mds_client *mdsc,
2255 			      struct ceph_mds_request *req)
2256 {
2257 	dout("submit_request on %p\n", req);
2258 	mutex_lock(&mdsc->mutex);
2259 	__register_request(mdsc, req, NULL);
2260 	__do_request(mdsc, req);
2261 	mutex_unlock(&mdsc->mutex);
2262 }
2263 
2264 /*
2265  * Synchrously perform an mds request.  Take care of all of the
2266  * session setup, forwarding, retry details.
2267  */
2268 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2269 			 struct inode *dir,
2270 			 struct ceph_mds_request *req)
2271 {
2272 	int err;
2273 
2274 	dout("do_request on %p\n", req);
2275 
2276 	/* take CAP_PIN refs for r_inode, r_locked_dir, r_old_dentry */
2277 	if (req->r_inode)
2278 		ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2279 	if (req->r_locked_dir)
2280 		ceph_get_cap_refs(ceph_inode(req->r_locked_dir), CEPH_CAP_PIN);
2281 	if (req->r_old_dentry_dir)
2282 		ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2283 				  CEPH_CAP_PIN);
2284 
2285 	/* issue */
2286 	mutex_lock(&mdsc->mutex);
2287 	__register_request(mdsc, req, dir);
2288 	__do_request(mdsc, req);
2289 
2290 	if (req->r_err) {
2291 		err = req->r_err;
2292 		__unregister_request(mdsc, req);
2293 		dout("do_request early error %d\n", err);
2294 		goto out;
2295 	}
2296 
2297 	/* wait */
2298 	mutex_unlock(&mdsc->mutex);
2299 	dout("do_request waiting\n");
2300 	if (!req->r_timeout && req->r_wait_for_completion) {
2301 		err = req->r_wait_for_completion(mdsc, req);
2302 	} else {
2303 		long timeleft = wait_for_completion_killable_timeout(
2304 					&req->r_completion,
2305 					ceph_timeout_jiffies(req->r_timeout));
2306 		if (timeleft > 0)
2307 			err = 0;
2308 		else if (!timeleft)
2309 			err = -EIO;  /* timed out */
2310 		else
2311 			err = timeleft;  /* killed */
2312 	}
2313 	dout("do_request waited, got %d\n", err);
2314 	mutex_lock(&mdsc->mutex);
2315 
2316 	/* only abort if we didn't race with a real reply */
2317 	if (req->r_got_result) {
2318 		err = le32_to_cpu(req->r_reply_info.head->result);
2319 	} else if (err < 0) {
2320 		dout("aborted request %lld with %d\n", req->r_tid, err);
2321 
2322 		/*
2323 		 * ensure we aren't running concurrently with
2324 		 * ceph_fill_trace or ceph_readdir_prepopulate, which
2325 		 * rely on locks (dir mutex) held by our caller.
2326 		 */
2327 		mutex_lock(&req->r_fill_mutex);
2328 		req->r_err = err;
2329 		req->r_aborted = true;
2330 		mutex_unlock(&req->r_fill_mutex);
2331 
2332 		if (req->r_locked_dir &&
2333 		    (req->r_op & CEPH_MDS_OP_WRITE))
2334 			ceph_invalidate_dir_request(req);
2335 	} else {
2336 		err = req->r_err;
2337 	}
2338 
2339 out:
2340 	mutex_unlock(&mdsc->mutex);
2341 	dout("do_request %p done, result %d\n", req, err);
2342 	return err;
2343 }
2344 
2345 /*
2346  * Invalidate dir's completeness, dentry lease state on an aborted MDS
2347  * namespace request.
2348  */
2349 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2350 {
2351 	struct inode *inode = req->r_locked_dir;
2352 
2353 	dout("invalidate_dir_request %p (complete, lease(s))\n", inode);
2354 
2355 	ceph_dir_clear_complete(inode);
2356 	if (req->r_dentry)
2357 		ceph_invalidate_dentry_lease(req->r_dentry);
2358 	if (req->r_old_dentry)
2359 		ceph_invalidate_dentry_lease(req->r_old_dentry);
2360 }
2361 
2362 /*
2363  * Handle mds reply.
2364  *
2365  * We take the session mutex and parse and process the reply immediately.
2366  * This preserves the logical ordering of replies, capabilities, etc., sent
2367  * by the MDS as they are applied to our local cache.
2368  */
2369 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2370 {
2371 	struct ceph_mds_client *mdsc = session->s_mdsc;
2372 	struct ceph_mds_request *req;
2373 	struct ceph_mds_reply_head *head = msg->front.iov_base;
2374 	struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2375 	struct ceph_snap_realm *realm;
2376 	u64 tid;
2377 	int err, result;
2378 	int mds = session->s_mds;
2379 
2380 	if (msg->front.iov_len < sizeof(*head)) {
2381 		pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2382 		ceph_msg_dump(msg);
2383 		return;
2384 	}
2385 
2386 	/* get request, session */
2387 	tid = le64_to_cpu(msg->hdr.tid);
2388 	mutex_lock(&mdsc->mutex);
2389 	req = __lookup_request(mdsc, tid);
2390 	if (!req) {
2391 		dout("handle_reply on unknown tid %llu\n", tid);
2392 		mutex_unlock(&mdsc->mutex);
2393 		return;
2394 	}
2395 	dout("handle_reply %p\n", req);
2396 
2397 	/* correct session? */
2398 	if (req->r_session != session) {
2399 		pr_err("mdsc_handle_reply got %llu on session mds%d"
2400 		       " not mds%d\n", tid, session->s_mds,
2401 		       req->r_session ? req->r_session->s_mds : -1);
2402 		mutex_unlock(&mdsc->mutex);
2403 		goto out;
2404 	}
2405 
2406 	/* dup? */
2407 	if ((req->r_got_unsafe && !head->safe) ||
2408 	    (req->r_got_safe && head->safe)) {
2409 		pr_warn("got a dup %s reply on %llu from mds%d\n",
2410 			   head->safe ? "safe" : "unsafe", tid, mds);
2411 		mutex_unlock(&mdsc->mutex);
2412 		goto out;
2413 	}
2414 	if (req->r_got_safe && !head->safe) {
2415 		pr_warn("got unsafe after safe on %llu from mds%d\n",
2416 			   tid, mds);
2417 		mutex_unlock(&mdsc->mutex);
2418 		goto out;
2419 	}
2420 
2421 	result = le32_to_cpu(head->result);
2422 
2423 	/*
2424 	 * Handle an ESTALE
2425 	 * if we're not talking to the authority, send to them
2426 	 * if the authority has changed while we weren't looking,
2427 	 * send to new authority
2428 	 * Otherwise we just have to return an ESTALE
2429 	 */
2430 	if (result == -ESTALE) {
2431 		dout("got ESTALE on request %llu", req->r_tid);
2432 		req->r_resend_mds = -1;
2433 		if (req->r_direct_mode != USE_AUTH_MDS) {
2434 			dout("not using auth, setting for that now");
2435 			req->r_direct_mode = USE_AUTH_MDS;
2436 			__do_request(mdsc, req);
2437 			mutex_unlock(&mdsc->mutex);
2438 			goto out;
2439 		} else  {
2440 			int mds = __choose_mds(mdsc, req);
2441 			if (mds >= 0 && mds != req->r_session->s_mds) {
2442 				dout("but auth changed, so resending");
2443 				__do_request(mdsc, req);
2444 				mutex_unlock(&mdsc->mutex);
2445 				goto out;
2446 			}
2447 		}
2448 		dout("have to return ESTALE on request %llu", req->r_tid);
2449 	}
2450 
2451 
2452 	if (head->safe) {
2453 		req->r_got_safe = true;
2454 		__unregister_request(mdsc, req);
2455 
2456 		if (req->r_got_unsafe) {
2457 			/*
2458 			 * We already handled the unsafe response, now do the
2459 			 * cleanup.  No need to examine the response; the MDS
2460 			 * doesn't include any result info in the safe
2461 			 * response.  And even if it did, there is nothing
2462 			 * useful we could do with a revised return value.
2463 			 */
2464 			dout("got safe reply %llu, mds%d\n", tid, mds);
2465 			list_del_init(&req->r_unsafe_item);
2466 
2467 			/* last unsafe request during umount? */
2468 			if (mdsc->stopping && !__get_oldest_req(mdsc))
2469 				complete_all(&mdsc->safe_umount_waiters);
2470 			mutex_unlock(&mdsc->mutex);
2471 			goto out;
2472 		}
2473 	} else {
2474 		req->r_got_unsafe = true;
2475 		list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
2476 	}
2477 
2478 	dout("handle_reply tid %lld result %d\n", tid, result);
2479 	rinfo = &req->r_reply_info;
2480 	err = parse_reply_info(msg, rinfo, session->s_con.peer_features);
2481 	mutex_unlock(&mdsc->mutex);
2482 
2483 	mutex_lock(&session->s_mutex);
2484 	if (err < 0) {
2485 		pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
2486 		ceph_msg_dump(msg);
2487 		goto out_err;
2488 	}
2489 
2490 	/* snap trace */
2491 	realm = NULL;
2492 	if (rinfo->snapblob_len) {
2493 		down_write(&mdsc->snap_rwsem);
2494 		ceph_update_snap_trace(mdsc, rinfo->snapblob,
2495 				rinfo->snapblob + rinfo->snapblob_len,
2496 				le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
2497 				&realm);
2498 		downgrade_write(&mdsc->snap_rwsem);
2499 	} else {
2500 		down_read(&mdsc->snap_rwsem);
2501 	}
2502 
2503 	/* insert trace into our cache */
2504 	mutex_lock(&req->r_fill_mutex);
2505 	err = ceph_fill_trace(mdsc->fsc->sb, req, req->r_session);
2506 	if (err == 0) {
2507 		if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
2508 				    req->r_op == CEPH_MDS_OP_LSSNAP))
2509 			ceph_readdir_prepopulate(req, req->r_session);
2510 		ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
2511 	}
2512 	mutex_unlock(&req->r_fill_mutex);
2513 
2514 	up_read(&mdsc->snap_rwsem);
2515 	if (realm)
2516 		ceph_put_snap_realm(mdsc, realm);
2517 out_err:
2518 	mutex_lock(&mdsc->mutex);
2519 	if (!req->r_aborted) {
2520 		if (err) {
2521 			req->r_err = err;
2522 		} else {
2523 			req->r_reply = msg;
2524 			ceph_msg_get(msg);
2525 			req->r_got_result = true;
2526 		}
2527 	} else {
2528 		dout("reply arrived after request %lld was aborted\n", tid);
2529 	}
2530 	mutex_unlock(&mdsc->mutex);
2531 
2532 	mutex_unlock(&session->s_mutex);
2533 
2534 	/* kick calling process */
2535 	complete_request(mdsc, req);
2536 out:
2537 	ceph_mdsc_put_request(req);
2538 	return;
2539 }
2540 
2541 
2542 
2543 /*
2544  * handle mds notification that our request has been forwarded.
2545  */
2546 static void handle_forward(struct ceph_mds_client *mdsc,
2547 			   struct ceph_mds_session *session,
2548 			   struct ceph_msg *msg)
2549 {
2550 	struct ceph_mds_request *req;
2551 	u64 tid = le64_to_cpu(msg->hdr.tid);
2552 	u32 next_mds;
2553 	u32 fwd_seq;
2554 	int err = -EINVAL;
2555 	void *p = msg->front.iov_base;
2556 	void *end = p + msg->front.iov_len;
2557 
2558 	ceph_decode_need(&p, end, 2*sizeof(u32), bad);
2559 	next_mds = ceph_decode_32(&p);
2560 	fwd_seq = ceph_decode_32(&p);
2561 
2562 	mutex_lock(&mdsc->mutex);
2563 	req = __lookup_request(mdsc, tid);
2564 	if (!req) {
2565 		dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
2566 		goto out;  /* dup reply? */
2567 	}
2568 
2569 	if (req->r_aborted) {
2570 		dout("forward tid %llu aborted, unregistering\n", tid);
2571 		__unregister_request(mdsc, req);
2572 	} else if (fwd_seq <= req->r_num_fwd) {
2573 		dout("forward tid %llu to mds%d - old seq %d <= %d\n",
2574 		     tid, next_mds, req->r_num_fwd, fwd_seq);
2575 	} else {
2576 		/* resend. forward race not possible; mds would drop */
2577 		dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
2578 		BUG_ON(req->r_err);
2579 		BUG_ON(req->r_got_result);
2580 		req->r_attempts = 0;
2581 		req->r_num_fwd = fwd_seq;
2582 		req->r_resend_mds = next_mds;
2583 		put_request_session(req);
2584 		__do_request(mdsc, req);
2585 	}
2586 	ceph_mdsc_put_request(req);
2587 out:
2588 	mutex_unlock(&mdsc->mutex);
2589 	return;
2590 
2591 bad:
2592 	pr_err("mdsc_handle_forward decode error err=%d\n", err);
2593 }
2594 
2595 /*
2596  * handle a mds session control message
2597  */
2598 static void handle_session(struct ceph_mds_session *session,
2599 			   struct ceph_msg *msg)
2600 {
2601 	struct ceph_mds_client *mdsc = session->s_mdsc;
2602 	u32 op;
2603 	u64 seq;
2604 	int mds = session->s_mds;
2605 	struct ceph_mds_session_head *h = msg->front.iov_base;
2606 	int wake = 0;
2607 
2608 	/* decode */
2609 	if (msg->front.iov_len != sizeof(*h))
2610 		goto bad;
2611 	op = le32_to_cpu(h->op);
2612 	seq = le64_to_cpu(h->seq);
2613 
2614 	mutex_lock(&mdsc->mutex);
2615 	if (op == CEPH_SESSION_CLOSE)
2616 		__unregister_session(mdsc, session);
2617 	/* FIXME: this ttl calculation is generous */
2618 	session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
2619 	mutex_unlock(&mdsc->mutex);
2620 
2621 	mutex_lock(&session->s_mutex);
2622 
2623 	dout("handle_session mds%d %s %p state %s seq %llu\n",
2624 	     mds, ceph_session_op_name(op), session,
2625 	     ceph_session_state_name(session->s_state), seq);
2626 
2627 	if (session->s_state == CEPH_MDS_SESSION_HUNG) {
2628 		session->s_state = CEPH_MDS_SESSION_OPEN;
2629 		pr_info("mds%d came back\n", session->s_mds);
2630 	}
2631 
2632 	switch (op) {
2633 	case CEPH_SESSION_OPEN:
2634 		if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2635 			pr_info("mds%d reconnect success\n", session->s_mds);
2636 		session->s_state = CEPH_MDS_SESSION_OPEN;
2637 		renewed_caps(mdsc, session, 0);
2638 		wake = 1;
2639 		if (mdsc->stopping)
2640 			__close_session(mdsc, session);
2641 		break;
2642 
2643 	case CEPH_SESSION_RENEWCAPS:
2644 		if (session->s_renew_seq == seq)
2645 			renewed_caps(mdsc, session, 1);
2646 		break;
2647 
2648 	case CEPH_SESSION_CLOSE:
2649 		if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
2650 			pr_info("mds%d reconnect denied\n", session->s_mds);
2651 		cleanup_session_requests(mdsc, session);
2652 		remove_session_caps(session);
2653 		wake = 2; /* for good measure */
2654 		wake_up_all(&mdsc->session_close_wq);
2655 		break;
2656 
2657 	case CEPH_SESSION_STALE:
2658 		pr_info("mds%d caps went stale, renewing\n",
2659 			session->s_mds);
2660 		spin_lock(&session->s_gen_ttl_lock);
2661 		session->s_cap_gen++;
2662 		session->s_cap_ttl = jiffies - 1;
2663 		spin_unlock(&session->s_gen_ttl_lock);
2664 		send_renew_caps(mdsc, session);
2665 		break;
2666 
2667 	case CEPH_SESSION_RECALL_STATE:
2668 		trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
2669 		break;
2670 
2671 	case CEPH_SESSION_FLUSHMSG:
2672 		send_flushmsg_ack(mdsc, session, seq);
2673 		break;
2674 
2675 	case CEPH_SESSION_FORCE_RO:
2676 		dout("force_session_readonly %p\n", session);
2677 		spin_lock(&session->s_cap_lock);
2678 		session->s_readonly = true;
2679 		spin_unlock(&session->s_cap_lock);
2680 		wake_up_session_caps(session, 0);
2681 		break;
2682 
2683 	default:
2684 		pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
2685 		WARN_ON(1);
2686 	}
2687 
2688 	mutex_unlock(&session->s_mutex);
2689 	if (wake) {
2690 		mutex_lock(&mdsc->mutex);
2691 		__wake_requests(mdsc, &session->s_waiting);
2692 		if (wake == 2)
2693 			kick_requests(mdsc, mds);
2694 		mutex_unlock(&mdsc->mutex);
2695 	}
2696 	return;
2697 
2698 bad:
2699 	pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
2700 	       (int)msg->front.iov_len);
2701 	ceph_msg_dump(msg);
2702 	return;
2703 }
2704 
2705 
2706 /*
2707  * called under session->mutex.
2708  */
2709 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
2710 				   struct ceph_mds_session *session)
2711 {
2712 	struct ceph_mds_request *req, *nreq;
2713 	struct rb_node *p;
2714 	int err;
2715 
2716 	dout("replay_unsafe_requests mds%d\n", session->s_mds);
2717 
2718 	mutex_lock(&mdsc->mutex);
2719 	list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item) {
2720 		err = __prepare_send_request(mdsc, req, session->s_mds, true);
2721 		if (!err) {
2722 			ceph_msg_get(req->r_request);
2723 			ceph_con_send(&session->s_con, req->r_request);
2724 		}
2725 	}
2726 
2727 	/*
2728 	 * also re-send old requests when MDS enters reconnect stage. So that MDS
2729 	 * can process completed request in clientreplay stage.
2730 	 */
2731 	p = rb_first(&mdsc->request_tree);
2732 	while (p) {
2733 		req = rb_entry(p, struct ceph_mds_request, r_node);
2734 		p = rb_next(p);
2735 		if (req->r_got_unsafe)
2736 			continue;
2737 		if (req->r_attempts == 0)
2738 			continue; /* only old requests */
2739 		if (req->r_session &&
2740 		    req->r_session->s_mds == session->s_mds) {
2741 			err = __prepare_send_request(mdsc, req,
2742 						     session->s_mds, true);
2743 			if (!err) {
2744 				ceph_msg_get(req->r_request);
2745 				ceph_con_send(&session->s_con, req->r_request);
2746 			}
2747 		}
2748 	}
2749 	mutex_unlock(&mdsc->mutex);
2750 }
2751 
2752 /*
2753  * Encode information about a cap for a reconnect with the MDS.
2754  */
2755 static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
2756 			  void *arg)
2757 {
2758 	union {
2759 		struct ceph_mds_cap_reconnect v2;
2760 		struct ceph_mds_cap_reconnect_v1 v1;
2761 	} rec;
2762 	size_t reclen;
2763 	struct ceph_inode_info *ci;
2764 	struct ceph_reconnect_state *recon_state = arg;
2765 	struct ceph_pagelist *pagelist = recon_state->pagelist;
2766 	char *path;
2767 	int pathlen, err;
2768 	u64 pathbase;
2769 	struct dentry *dentry;
2770 
2771 	ci = cap->ci;
2772 
2773 	dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
2774 	     inode, ceph_vinop(inode), cap, cap->cap_id,
2775 	     ceph_cap_string(cap->issued));
2776 	err = ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
2777 	if (err)
2778 		return err;
2779 
2780 	dentry = d_find_alias(inode);
2781 	if (dentry) {
2782 		path = ceph_mdsc_build_path(dentry, &pathlen, &pathbase, 0);
2783 		if (IS_ERR(path)) {
2784 			err = PTR_ERR(path);
2785 			goto out_dput;
2786 		}
2787 	} else {
2788 		path = NULL;
2789 		pathlen = 0;
2790 	}
2791 	err = ceph_pagelist_encode_string(pagelist, path, pathlen);
2792 	if (err)
2793 		goto out_free;
2794 
2795 	spin_lock(&ci->i_ceph_lock);
2796 	cap->seq = 0;        /* reset cap seq */
2797 	cap->issue_seq = 0;  /* and issue_seq */
2798 	cap->mseq = 0;       /* and migrate_seq */
2799 	cap->cap_gen = cap->session->s_cap_gen;
2800 
2801 	if (recon_state->flock) {
2802 		rec.v2.cap_id = cpu_to_le64(cap->cap_id);
2803 		rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2804 		rec.v2.issued = cpu_to_le32(cap->issued);
2805 		rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2806 		rec.v2.pathbase = cpu_to_le64(pathbase);
2807 		rec.v2.flock_len = 0;
2808 		reclen = sizeof(rec.v2);
2809 	} else {
2810 		rec.v1.cap_id = cpu_to_le64(cap->cap_id);
2811 		rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
2812 		rec.v1.issued = cpu_to_le32(cap->issued);
2813 		rec.v1.size = cpu_to_le64(inode->i_size);
2814 		ceph_encode_timespec(&rec.v1.mtime, &inode->i_mtime);
2815 		ceph_encode_timespec(&rec.v1.atime, &inode->i_atime);
2816 		rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
2817 		rec.v1.pathbase = cpu_to_le64(pathbase);
2818 		reclen = sizeof(rec.v1);
2819 	}
2820 	spin_unlock(&ci->i_ceph_lock);
2821 
2822 	if (recon_state->flock) {
2823 		int num_fcntl_locks, num_flock_locks;
2824 		struct ceph_filelock *flocks;
2825 
2826 encode_again:
2827 		ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
2828 		flocks = kmalloc((num_fcntl_locks+num_flock_locks) *
2829 				 sizeof(struct ceph_filelock), GFP_NOFS);
2830 		if (!flocks) {
2831 			err = -ENOMEM;
2832 			goto out_free;
2833 		}
2834 		err = ceph_encode_locks_to_buffer(inode, flocks,
2835 						  num_fcntl_locks,
2836 						  num_flock_locks);
2837 		if (err) {
2838 			kfree(flocks);
2839 			if (err == -ENOSPC)
2840 				goto encode_again;
2841 			goto out_free;
2842 		}
2843 		/*
2844 		 * number of encoded locks is stable, so copy to pagelist
2845 		 */
2846 		rec.v2.flock_len = cpu_to_le32(2*sizeof(u32) +
2847 				    (num_fcntl_locks+num_flock_locks) *
2848 				    sizeof(struct ceph_filelock));
2849 		err = ceph_pagelist_append(pagelist, &rec, reclen);
2850 		if (!err)
2851 			err = ceph_locks_to_pagelist(flocks, pagelist,
2852 						     num_fcntl_locks,
2853 						     num_flock_locks);
2854 		kfree(flocks);
2855 	} else {
2856 		err = ceph_pagelist_append(pagelist, &rec, reclen);
2857 	}
2858 
2859 	recon_state->nr_caps++;
2860 out_free:
2861 	kfree(path);
2862 out_dput:
2863 	dput(dentry);
2864 	return err;
2865 }
2866 
2867 
2868 /*
2869  * If an MDS fails and recovers, clients need to reconnect in order to
2870  * reestablish shared state.  This includes all caps issued through
2871  * this session _and_ the snap_realm hierarchy.  Because it's not
2872  * clear which snap realms the mds cares about, we send everything we
2873  * know about.. that ensures we'll then get any new info the
2874  * recovering MDS might have.
2875  *
2876  * This is a relatively heavyweight operation, but it's rare.
2877  *
2878  * called with mdsc->mutex held.
2879  */
2880 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
2881 			       struct ceph_mds_session *session)
2882 {
2883 	struct ceph_msg *reply;
2884 	struct rb_node *p;
2885 	int mds = session->s_mds;
2886 	int err = -ENOMEM;
2887 	int s_nr_caps;
2888 	struct ceph_pagelist *pagelist;
2889 	struct ceph_reconnect_state recon_state;
2890 
2891 	pr_info("mds%d reconnect start\n", mds);
2892 
2893 	pagelist = kmalloc(sizeof(*pagelist), GFP_NOFS);
2894 	if (!pagelist)
2895 		goto fail_nopagelist;
2896 	ceph_pagelist_init(pagelist);
2897 
2898 	reply = ceph_msg_new(CEPH_MSG_CLIENT_RECONNECT, 0, GFP_NOFS, false);
2899 	if (!reply)
2900 		goto fail_nomsg;
2901 
2902 	mutex_lock(&session->s_mutex);
2903 	session->s_state = CEPH_MDS_SESSION_RECONNECTING;
2904 	session->s_seq = 0;
2905 
2906 	dout("session %p state %s\n", session,
2907 	     ceph_session_state_name(session->s_state));
2908 
2909 	spin_lock(&session->s_gen_ttl_lock);
2910 	session->s_cap_gen++;
2911 	spin_unlock(&session->s_gen_ttl_lock);
2912 
2913 	spin_lock(&session->s_cap_lock);
2914 	/* don't know if session is readonly */
2915 	session->s_readonly = 0;
2916 	/*
2917 	 * notify __ceph_remove_cap() that we are composing cap reconnect.
2918 	 * If a cap get released before being added to the cap reconnect,
2919 	 * __ceph_remove_cap() should skip queuing cap release.
2920 	 */
2921 	session->s_cap_reconnect = 1;
2922 	/* drop old cap expires; we're about to reestablish that state */
2923 	cleanup_cap_releases(mdsc, session);
2924 
2925 	/* trim unused caps to reduce MDS's cache rejoin time */
2926 	if (mdsc->fsc->sb->s_root)
2927 		shrink_dcache_parent(mdsc->fsc->sb->s_root);
2928 
2929 	ceph_con_close(&session->s_con);
2930 	ceph_con_open(&session->s_con,
2931 		      CEPH_ENTITY_TYPE_MDS, mds,
2932 		      ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
2933 
2934 	/* replay unsafe requests */
2935 	replay_unsafe_requests(mdsc, session);
2936 
2937 	down_read(&mdsc->snap_rwsem);
2938 
2939 	/* traverse this session's caps */
2940 	s_nr_caps = session->s_nr_caps;
2941 	err = ceph_pagelist_encode_32(pagelist, s_nr_caps);
2942 	if (err)
2943 		goto fail;
2944 
2945 	recon_state.nr_caps = 0;
2946 	recon_state.pagelist = pagelist;
2947 	recon_state.flock = session->s_con.peer_features & CEPH_FEATURE_FLOCK;
2948 	err = iterate_session_caps(session, encode_caps_cb, &recon_state);
2949 	if (err < 0)
2950 		goto fail;
2951 
2952 	spin_lock(&session->s_cap_lock);
2953 	session->s_cap_reconnect = 0;
2954 	spin_unlock(&session->s_cap_lock);
2955 
2956 	/*
2957 	 * snaprealms.  we provide mds with the ino, seq (version), and
2958 	 * parent for all of our realms.  If the mds has any newer info,
2959 	 * it will tell us.
2960 	 */
2961 	for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
2962 		struct ceph_snap_realm *realm =
2963 			rb_entry(p, struct ceph_snap_realm, node);
2964 		struct ceph_mds_snaprealm_reconnect sr_rec;
2965 
2966 		dout(" adding snap realm %llx seq %lld parent %llx\n",
2967 		     realm->ino, realm->seq, realm->parent_ino);
2968 		sr_rec.ino = cpu_to_le64(realm->ino);
2969 		sr_rec.seq = cpu_to_le64(realm->seq);
2970 		sr_rec.parent = cpu_to_le64(realm->parent_ino);
2971 		err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
2972 		if (err)
2973 			goto fail;
2974 	}
2975 
2976 	if (recon_state.flock)
2977 		reply->hdr.version = cpu_to_le16(2);
2978 
2979 	/* raced with cap release? */
2980 	if (s_nr_caps != recon_state.nr_caps) {
2981 		struct page *page = list_first_entry(&pagelist->head,
2982 						     struct page, lru);
2983 		__le32 *addr = kmap_atomic(page);
2984 		*addr = cpu_to_le32(recon_state.nr_caps);
2985 		kunmap_atomic(addr);
2986 	}
2987 
2988 	reply->hdr.data_len = cpu_to_le32(pagelist->length);
2989 	ceph_msg_data_add_pagelist(reply, pagelist);
2990 
2991 	ceph_early_kick_flushing_caps(mdsc, session);
2992 
2993 	ceph_con_send(&session->s_con, reply);
2994 
2995 	mutex_unlock(&session->s_mutex);
2996 
2997 	mutex_lock(&mdsc->mutex);
2998 	__wake_requests(mdsc, &session->s_waiting);
2999 	mutex_unlock(&mdsc->mutex);
3000 
3001 	up_read(&mdsc->snap_rwsem);
3002 	return;
3003 
3004 fail:
3005 	ceph_msg_put(reply);
3006 	up_read(&mdsc->snap_rwsem);
3007 	mutex_unlock(&session->s_mutex);
3008 fail_nomsg:
3009 	ceph_pagelist_release(pagelist);
3010 fail_nopagelist:
3011 	pr_err("error %d preparing reconnect for mds%d\n", err, mds);
3012 	return;
3013 }
3014 
3015 
3016 /*
3017  * compare old and new mdsmaps, kicking requests
3018  * and closing out old connections as necessary
3019  *
3020  * called under mdsc->mutex.
3021  */
3022 static void check_new_map(struct ceph_mds_client *mdsc,
3023 			  struct ceph_mdsmap *newmap,
3024 			  struct ceph_mdsmap *oldmap)
3025 {
3026 	int i;
3027 	int oldstate, newstate;
3028 	struct ceph_mds_session *s;
3029 
3030 	dout("check_new_map new %u old %u\n",
3031 	     newmap->m_epoch, oldmap->m_epoch);
3032 
3033 	for (i = 0; i < oldmap->m_max_mds && i < mdsc->max_sessions; i++) {
3034 		if (mdsc->sessions[i] == NULL)
3035 			continue;
3036 		s = mdsc->sessions[i];
3037 		oldstate = ceph_mdsmap_get_state(oldmap, i);
3038 		newstate = ceph_mdsmap_get_state(newmap, i);
3039 
3040 		dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
3041 		     i, ceph_mds_state_name(oldstate),
3042 		     ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
3043 		     ceph_mds_state_name(newstate),
3044 		     ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
3045 		     ceph_session_state_name(s->s_state));
3046 
3047 		if (i >= newmap->m_max_mds ||
3048 		    memcmp(ceph_mdsmap_get_addr(oldmap, i),
3049 			   ceph_mdsmap_get_addr(newmap, i),
3050 			   sizeof(struct ceph_entity_addr))) {
3051 			if (s->s_state == CEPH_MDS_SESSION_OPENING) {
3052 				/* the session never opened, just close it
3053 				 * out now */
3054 				__wake_requests(mdsc, &s->s_waiting);
3055 				__unregister_session(mdsc, s);
3056 			} else {
3057 				/* just close it */
3058 				mutex_unlock(&mdsc->mutex);
3059 				mutex_lock(&s->s_mutex);
3060 				mutex_lock(&mdsc->mutex);
3061 				ceph_con_close(&s->s_con);
3062 				mutex_unlock(&s->s_mutex);
3063 				s->s_state = CEPH_MDS_SESSION_RESTARTING;
3064 			}
3065 		} else if (oldstate == newstate) {
3066 			continue;  /* nothing new with this mds */
3067 		}
3068 
3069 		/*
3070 		 * send reconnect?
3071 		 */
3072 		if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
3073 		    newstate >= CEPH_MDS_STATE_RECONNECT) {
3074 			mutex_unlock(&mdsc->mutex);
3075 			send_mds_reconnect(mdsc, s);
3076 			mutex_lock(&mdsc->mutex);
3077 		}
3078 
3079 		/*
3080 		 * kick request on any mds that has gone active.
3081 		 */
3082 		if (oldstate < CEPH_MDS_STATE_ACTIVE &&
3083 		    newstate >= CEPH_MDS_STATE_ACTIVE) {
3084 			if (oldstate != CEPH_MDS_STATE_CREATING &&
3085 			    oldstate != CEPH_MDS_STATE_STARTING)
3086 				pr_info("mds%d recovery completed\n", s->s_mds);
3087 			kick_requests(mdsc, i);
3088 			ceph_kick_flushing_caps(mdsc, s);
3089 			wake_up_session_caps(s, 1);
3090 		}
3091 	}
3092 
3093 	for (i = 0; i < newmap->m_max_mds && i < mdsc->max_sessions; i++) {
3094 		s = mdsc->sessions[i];
3095 		if (!s)
3096 			continue;
3097 		if (!ceph_mdsmap_is_laggy(newmap, i))
3098 			continue;
3099 		if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3100 		    s->s_state == CEPH_MDS_SESSION_HUNG ||
3101 		    s->s_state == CEPH_MDS_SESSION_CLOSING) {
3102 			dout(" connecting to export targets of laggy mds%d\n",
3103 			     i);
3104 			__open_export_target_sessions(mdsc, s);
3105 		}
3106 	}
3107 }
3108 
3109 
3110 
3111 /*
3112  * leases
3113  */
3114 
3115 /*
3116  * caller must hold session s_mutex, dentry->d_lock
3117  */
3118 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
3119 {
3120 	struct ceph_dentry_info *di = ceph_dentry(dentry);
3121 
3122 	ceph_put_mds_session(di->lease_session);
3123 	di->lease_session = NULL;
3124 }
3125 
3126 static void handle_lease(struct ceph_mds_client *mdsc,
3127 			 struct ceph_mds_session *session,
3128 			 struct ceph_msg *msg)
3129 {
3130 	struct super_block *sb = mdsc->fsc->sb;
3131 	struct inode *inode;
3132 	struct dentry *parent, *dentry;
3133 	struct ceph_dentry_info *di;
3134 	int mds = session->s_mds;
3135 	struct ceph_mds_lease *h = msg->front.iov_base;
3136 	u32 seq;
3137 	struct ceph_vino vino;
3138 	struct qstr dname;
3139 	int release = 0;
3140 
3141 	dout("handle_lease from mds%d\n", mds);
3142 
3143 	/* decode */
3144 	if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
3145 		goto bad;
3146 	vino.ino = le64_to_cpu(h->ino);
3147 	vino.snap = CEPH_NOSNAP;
3148 	seq = le32_to_cpu(h->seq);
3149 	dname.name = (void *)h + sizeof(*h) + sizeof(u32);
3150 	dname.len = msg->front.iov_len - sizeof(*h) - sizeof(u32);
3151 	if (dname.len != get_unaligned_le32(h+1))
3152 		goto bad;
3153 
3154 	/* lookup inode */
3155 	inode = ceph_find_inode(sb, vino);
3156 	dout("handle_lease %s, ino %llx %p %.*s\n",
3157 	     ceph_lease_op_name(h->action), vino.ino, inode,
3158 	     dname.len, dname.name);
3159 
3160 	mutex_lock(&session->s_mutex);
3161 	session->s_seq++;
3162 
3163 	if (inode == NULL) {
3164 		dout("handle_lease no inode %llx\n", vino.ino);
3165 		goto release;
3166 	}
3167 
3168 	/* dentry */
3169 	parent = d_find_alias(inode);
3170 	if (!parent) {
3171 		dout("no parent dentry on inode %p\n", inode);
3172 		WARN_ON(1);
3173 		goto release;  /* hrm... */
3174 	}
3175 	dname.hash = full_name_hash(dname.name, dname.len);
3176 	dentry = d_lookup(parent, &dname);
3177 	dput(parent);
3178 	if (!dentry)
3179 		goto release;
3180 
3181 	spin_lock(&dentry->d_lock);
3182 	di = ceph_dentry(dentry);
3183 	switch (h->action) {
3184 	case CEPH_MDS_LEASE_REVOKE:
3185 		if (di->lease_session == session) {
3186 			if (ceph_seq_cmp(di->lease_seq, seq) > 0)
3187 				h->seq = cpu_to_le32(di->lease_seq);
3188 			__ceph_mdsc_drop_dentry_lease(dentry);
3189 		}
3190 		release = 1;
3191 		break;
3192 
3193 	case CEPH_MDS_LEASE_RENEW:
3194 		if (di->lease_session == session &&
3195 		    di->lease_gen == session->s_cap_gen &&
3196 		    di->lease_renew_from &&
3197 		    di->lease_renew_after == 0) {
3198 			unsigned long duration =
3199 				msecs_to_jiffies(le32_to_cpu(h->duration_ms));
3200 
3201 			di->lease_seq = seq;
3202 			dentry->d_time = di->lease_renew_from + duration;
3203 			di->lease_renew_after = di->lease_renew_from +
3204 				(duration >> 1);
3205 			di->lease_renew_from = 0;
3206 		}
3207 		break;
3208 	}
3209 	spin_unlock(&dentry->d_lock);
3210 	dput(dentry);
3211 
3212 	if (!release)
3213 		goto out;
3214 
3215 release:
3216 	/* let's just reuse the same message */
3217 	h->action = CEPH_MDS_LEASE_REVOKE_ACK;
3218 	ceph_msg_get(msg);
3219 	ceph_con_send(&session->s_con, msg);
3220 
3221 out:
3222 	iput(inode);
3223 	mutex_unlock(&session->s_mutex);
3224 	return;
3225 
3226 bad:
3227 	pr_err("corrupt lease message\n");
3228 	ceph_msg_dump(msg);
3229 }
3230 
3231 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
3232 			      struct inode *inode,
3233 			      struct dentry *dentry, char action,
3234 			      u32 seq)
3235 {
3236 	struct ceph_msg *msg;
3237 	struct ceph_mds_lease *lease;
3238 	int len = sizeof(*lease) + sizeof(u32);
3239 	int dnamelen = 0;
3240 
3241 	dout("lease_send_msg inode %p dentry %p %s to mds%d\n",
3242 	     inode, dentry, ceph_lease_op_name(action), session->s_mds);
3243 	dnamelen = dentry->d_name.len;
3244 	len += dnamelen;
3245 
3246 	msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
3247 	if (!msg)
3248 		return;
3249 	lease = msg->front.iov_base;
3250 	lease->action = action;
3251 	lease->ino = cpu_to_le64(ceph_vino(inode).ino);
3252 	lease->first = lease->last = cpu_to_le64(ceph_vino(inode).snap);
3253 	lease->seq = cpu_to_le32(seq);
3254 	put_unaligned_le32(dnamelen, lease + 1);
3255 	memcpy((void *)(lease + 1) + 4, dentry->d_name.name, dnamelen);
3256 
3257 	/*
3258 	 * if this is a preemptive lease RELEASE, no need to
3259 	 * flush request stream, since the actual request will
3260 	 * soon follow.
3261 	 */
3262 	msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
3263 
3264 	ceph_con_send(&session->s_con, msg);
3265 }
3266 
3267 /*
3268  * Preemptively release a lease we expect to invalidate anyway.
3269  * Pass @inode always, @dentry is optional.
3270  */
3271 void ceph_mdsc_lease_release(struct ceph_mds_client *mdsc, struct inode *inode,
3272 			     struct dentry *dentry)
3273 {
3274 	struct ceph_dentry_info *di;
3275 	struct ceph_mds_session *session;
3276 	u32 seq;
3277 
3278 	BUG_ON(inode == NULL);
3279 	BUG_ON(dentry == NULL);
3280 
3281 	/* is dentry lease valid? */
3282 	spin_lock(&dentry->d_lock);
3283 	di = ceph_dentry(dentry);
3284 	if (!di || !di->lease_session ||
3285 	    di->lease_session->s_mds < 0 ||
3286 	    di->lease_gen != di->lease_session->s_cap_gen ||
3287 	    !time_before(jiffies, dentry->d_time)) {
3288 		dout("lease_release inode %p dentry %p -- "
3289 		     "no lease\n",
3290 		     inode, dentry);
3291 		spin_unlock(&dentry->d_lock);
3292 		return;
3293 	}
3294 
3295 	/* we do have a lease on this dentry; note mds and seq */
3296 	session = ceph_get_mds_session(di->lease_session);
3297 	seq = di->lease_seq;
3298 	__ceph_mdsc_drop_dentry_lease(dentry);
3299 	spin_unlock(&dentry->d_lock);
3300 
3301 	dout("lease_release inode %p dentry %p to mds%d\n",
3302 	     inode, dentry, session->s_mds);
3303 	ceph_mdsc_lease_send_msg(session, inode, dentry,
3304 				 CEPH_MDS_LEASE_RELEASE, seq);
3305 	ceph_put_mds_session(session);
3306 }
3307 
3308 /*
3309  * drop all leases (and dentry refs) in preparation for umount
3310  */
3311 static void drop_leases(struct ceph_mds_client *mdsc)
3312 {
3313 	int i;
3314 
3315 	dout("drop_leases\n");
3316 	mutex_lock(&mdsc->mutex);
3317 	for (i = 0; i < mdsc->max_sessions; i++) {
3318 		struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3319 		if (!s)
3320 			continue;
3321 		mutex_unlock(&mdsc->mutex);
3322 		mutex_lock(&s->s_mutex);
3323 		mutex_unlock(&s->s_mutex);
3324 		ceph_put_mds_session(s);
3325 		mutex_lock(&mdsc->mutex);
3326 	}
3327 	mutex_unlock(&mdsc->mutex);
3328 }
3329 
3330 
3331 
3332 /*
3333  * delayed work -- periodically trim expired leases, renew caps with mds
3334  */
3335 static void schedule_delayed(struct ceph_mds_client *mdsc)
3336 {
3337 	int delay = 5;
3338 	unsigned hz = round_jiffies_relative(HZ * delay);
3339 	schedule_delayed_work(&mdsc->delayed_work, hz);
3340 }
3341 
3342 static void delayed_work(struct work_struct *work)
3343 {
3344 	int i;
3345 	struct ceph_mds_client *mdsc =
3346 		container_of(work, struct ceph_mds_client, delayed_work.work);
3347 	int renew_interval;
3348 	int renew_caps;
3349 
3350 	dout("mdsc delayed_work\n");
3351 	ceph_check_delayed_caps(mdsc);
3352 
3353 	mutex_lock(&mdsc->mutex);
3354 	renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
3355 	renew_caps = time_after_eq(jiffies, HZ*renew_interval +
3356 				   mdsc->last_renew_caps);
3357 	if (renew_caps)
3358 		mdsc->last_renew_caps = jiffies;
3359 
3360 	for (i = 0; i < mdsc->max_sessions; i++) {
3361 		struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
3362 		if (s == NULL)
3363 			continue;
3364 		if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
3365 			dout("resending session close request for mds%d\n",
3366 			     s->s_mds);
3367 			request_close_session(mdsc, s);
3368 			ceph_put_mds_session(s);
3369 			continue;
3370 		}
3371 		if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
3372 			if (s->s_state == CEPH_MDS_SESSION_OPEN) {
3373 				s->s_state = CEPH_MDS_SESSION_HUNG;
3374 				pr_info("mds%d hung\n", s->s_mds);
3375 			}
3376 		}
3377 		if (s->s_state < CEPH_MDS_SESSION_OPEN) {
3378 			/* this mds is failed or recovering, just wait */
3379 			ceph_put_mds_session(s);
3380 			continue;
3381 		}
3382 		mutex_unlock(&mdsc->mutex);
3383 
3384 		mutex_lock(&s->s_mutex);
3385 		if (renew_caps)
3386 			send_renew_caps(mdsc, s);
3387 		else
3388 			ceph_con_keepalive(&s->s_con);
3389 		if (s->s_state == CEPH_MDS_SESSION_OPEN ||
3390 		    s->s_state == CEPH_MDS_SESSION_HUNG)
3391 			ceph_send_cap_releases(mdsc, s);
3392 		mutex_unlock(&s->s_mutex);
3393 		ceph_put_mds_session(s);
3394 
3395 		mutex_lock(&mdsc->mutex);
3396 	}
3397 	mutex_unlock(&mdsc->mutex);
3398 
3399 	schedule_delayed(mdsc);
3400 }
3401 
3402 int ceph_mdsc_init(struct ceph_fs_client *fsc)
3403 
3404 {
3405 	struct ceph_mds_client *mdsc;
3406 
3407 	mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
3408 	if (!mdsc)
3409 		return -ENOMEM;
3410 	mdsc->fsc = fsc;
3411 	fsc->mdsc = mdsc;
3412 	mutex_init(&mdsc->mutex);
3413 	mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
3414 	if (mdsc->mdsmap == NULL) {
3415 		kfree(mdsc);
3416 		return -ENOMEM;
3417 	}
3418 
3419 	init_completion(&mdsc->safe_umount_waiters);
3420 	init_waitqueue_head(&mdsc->session_close_wq);
3421 	INIT_LIST_HEAD(&mdsc->waiting_for_map);
3422 	mdsc->sessions = NULL;
3423 	atomic_set(&mdsc->num_sessions, 0);
3424 	mdsc->max_sessions = 0;
3425 	mdsc->stopping = 0;
3426 	mdsc->last_snap_seq = 0;
3427 	init_rwsem(&mdsc->snap_rwsem);
3428 	mdsc->snap_realms = RB_ROOT;
3429 	INIT_LIST_HEAD(&mdsc->snap_empty);
3430 	spin_lock_init(&mdsc->snap_empty_lock);
3431 	mdsc->last_tid = 0;
3432 	mdsc->oldest_tid = 0;
3433 	mdsc->request_tree = RB_ROOT;
3434 	INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
3435 	mdsc->last_renew_caps = jiffies;
3436 	INIT_LIST_HEAD(&mdsc->cap_delay_list);
3437 	spin_lock_init(&mdsc->cap_delay_lock);
3438 	INIT_LIST_HEAD(&mdsc->snap_flush_list);
3439 	spin_lock_init(&mdsc->snap_flush_lock);
3440 	mdsc->last_cap_flush_tid = 1;
3441 	mdsc->cap_flush_tree = RB_ROOT;
3442 	INIT_LIST_HEAD(&mdsc->cap_dirty);
3443 	INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
3444 	mdsc->num_cap_flushing = 0;
3445 	spin_lock_init(&mdsc->cap_dirty_lock);
3446 	init_waitqueue_head(&mdsc->cap_flushing_wq);
3447 	spin_lock_init(&mdsc->dentry_lru_lock);
3448 	INIT_LIST_HEAD(&mdsc->dentry_lru);
3449 
3450 	ceph_caps_init(mdsc);
3451 	ceph_adjust_min_caps(mdsc, fsc->min_caps);
3452 
3453 	init_rwsem(&mdsc->pool_perm_rwsem);
3454 	mdsc->pool_perm_tree = RB_ROOT;
3455 
3456 	return 0;
3457 }
3458 
3459 /*
3460  * Wait for safe replies on open mds requests.  If we time out, drop
3461  * all requests from the tree to avoid dangling dentry refs.
3462  */
3463 static void wait_requests(struct ceph_mds_client *mdsc)
3464 {
3465 	struct ceph_options *opts = mdsc->fsc->client->options;
3466 	struct ceph_mds_request *req;
3467 
3468 	mutex_lock(&mdsc->mutex);
3469 	if (__get_oldest_req(mdsc)) {
3470 		mutex_unlock(&mdsc->mutex);
3471 
3472 		dout("wait_requests waiting for requests\n");
3473 		wait_for_completion_timeout(&mdsc->safe_umount_waiters,
3474 				    ceph_timeout_jiffies(opts->mount_timeout));
3475 
3476 		/* tear down remaining requests */
3477 		mutex_lock(&mdsc->mutex);
3478 		while ((req = __get_oldest_req(mdsc))) {
3479 			dout("wait_requests timed out on tid %llu\n",
3480 			     req->r_tid);
3481 			__unregister_request(mdsc, req);
3482 		}
3483 	}
3484 	mutex_unlock(&mdsc->mutex);
3485 	dout("wait_requests done\n");
3486 }
3487 
3488 /*
3489  * called before mount is ro, and before dentries are torn down.
3490  * (hmm, does this still race with new lookups?)
3491  */
3492 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
3493 {
3494 	dout("pre_umount\n");
3495 	mdsc->stopping = 1;
3496 
3497 	drop_leases(mdsc);
3498 	ceph_flush_dirty_caps(mdsc);
3499 	wait_requests(mdsc);
3500 
3501 	/*
3502 	 * wait for reply handlers to drop their request refs and
3503 	 * their inode/dcache refs
3504 	 */
3505 	ceph_msgr_flush();
3506 }
3507 
3508 /*
3509  * wait for all write mds requests to flush.
3510  */
3511 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
3512 {
3513 	struct ceph_mds_request *req = NULL, *nextreq;
3514 	struct rb_node *n;
3515 
3516 	mutex_lock(&mdsc->mutex);
3517 	dout("wait_unsafe_requests want %lld\n", want_tid);
3518 restart:
3519 	req = __get_oldest_req(mdsc);
3520 	while (req && req->r_tid <= want_tid) {
3521 		/* find next request */
3522 		n = rb_next(&req->r_node);
3523 		if (n)
3524 			nextreq = rb_entry(n, struct ceph_mds_request, r_node);
3525 		else
3526 			nextreq = NULL;
3527 		if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
3528 		    (req->r_op & CEPH_MDS_OP_WRITE)) {
3529 			/* write op */
3530 			ceph_mdsc_get_request(req);
3531 			if (nextreq)
3532 				ceph_mdsc_get_request(nextreq);
3533 			mutex_unlock(&mdsc->mutex);
3534 			dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
3535 			     req->r_tid, want_tid);
3536 			wait_for_completion(&req->r_safe_completion);
3537 			mutex_lock(&mdsc->mutex);
3538 			ceph_mdsc_put_request(req);
3539 			if (!nextreq)
3540 				break;  /* next dne before, so we're done! */
3541 			if (RB_EMPTY_NODE(&nextreq->r_node)) {
3542 				/* next request was removed from tree */
3543 				ceph_mdsc_put_request(nextreq);
3544 				goto restart;
3545 			}
3546 			ceph_mdsc_put_request(nextreq);  /* won't go away */
3547 		}
3548 		req = nextreq;
3549 	}
3550 	mutex_unlock(&mdsc->mutex);
3551 	dout("wait_unsafe_requests done\n");
3552 }
3553 
3554 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
3555 {
3556 	u64 want_tid, want_flush, want_snap;
3557 
3558 	if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3559 		return;
3560 
3561 	dout("sync\n");
3562 	mutex_lock(&mdsc->mutex);
3563 	want_tid = mdsc->last_tid;
3564 	mutex_unlock(&mdsc->mutex);
3565 
3566 	ceph_flush_dirty_caps(mdsc);
3567 	spin_lock(&mdsc->cap_dirty_lock);
3568 	want_flush = mdsc->last_cap_flush_tid;
3569 	spin_unlock(&mdsc->cap_dirty_lock);
3570 
3571 	down_read(&mdsc->snap_rwsem);
3572 	want_snap = mdsc->last_snap_seq;
3573 	up_read(&mdsc->snap_rwsem);
3574 
3575 	dout("sync want tid %lld flush_seq %lld snap_seq %lld\n",
3576 	     want_tid, want_flush, want_snap);
3577 
3578 	wait_unsafe_requests(mdsc, want_tid);
3579 	wait_caps_flush(mdsc, want_flush, want_snap);
3580 }
3581 
3582 /*
3583  * true if all sessions are closed, or we force unmount
3584  */
3585 static bool done_closing_sessions(struct ceph_mds_client *mdsc)
3586 {
3587 	if (mdsc->fsc->mount_state == CEPH_MOUNT_SHUTDOWN)
3588 		return true;
3589 	return atomic_read(&mdsc->num_sessions) == 0;
3590 }
3591 
3592 /*
3593  * called after sb is ro.
3594  */
3595 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
3596 {
3597 	struct ceph_options *opts = mdsc->fsc->client->options;
3598 	struct ceph_mds_session *session;
3599 	int i;
3600 
3601 	dout("close_sessions\n");
3602 
3603 	/* close sessions */
3604 	mutex_lock(&mdsc->mutex);
3605 	for (i = 0; i < mdsc->max_sessions; i++) {
3606 		session = __ceph_lookup_mds_session(mdsc, i);
3607 		if (!session)
3608 			continue;
3609 		mutex_unlock(&mdsc->mutex);
3610 		mutex_lock(&session->s_mutex);
3611 		__close_session(mdsc, session);
3612 		mutex_unlock(&session->s_mutex);
3613 		ceph_put_mds_session(session);
3614 		mutex_lock(&mdsc->mutex);
3615 	}
3616 	mutex_unlock(&mdsc->mutex);
3617 
3618 	dout("waiting for sessions to close\n");
3619 	wait_event_timeout(mdsc->session_close_wq, done_closing_sessions(mdsc),
3620 			   ceph_timeout_jiffies(opts->mount_timeout));
3621 
3622 	/* tear down remaining sessions */
3623 	mutex_lock(&mdsc->mutex);
3624 	for (i = 0; i < mdsc->max_sessions; i++) {
3625 		if (mdsc->sessions[i]) {
3626 			session = get_session(mdsc->sessions[i]);
3627 			__unregister_session(mdsc, session);
3628 			mutex_unlock(&mdsc->mutex);
3629 			mutex_lock(&session->s_mutex);
3630 			remove_session_caps(session);
3631 			mutex_unlock(&session->s_mutex);
3632 			ceph_put_mds_session(session);
3633 			mutex_lock(&mdsc->mutex);
3634 		}
3635 	}
3636 	WARN_ON(!list_empty(&mdsc->cap_delay_list));
3637 	mutex_unlock(&mdsc->mutex);
3638 
3639 	ceph_cleanup_empty_realms(mdsc);
3640 
3641 	cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3642 
3643 	dout("stopped\n");
3644 }
3645 
3646 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
3647 {
3648 	dout("stop\n");
3649 	cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
3650 	if (mdsc->mdsmap)
3651 		ceph_mdsmap_destroy(mdsc->mdsmap);
3652 	kfree(mdsc->sessions);
3653 	ceph_caps_finalize(mdsc);
3654 	ceph_pool_perm_destroy(mdsc);
3655 }
3656 
3657 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
3658 {
3659 	struct ceph_mds_client *mdsc = fsc->mdsc;
3660 
3661 	dout("mdsc_destroy %p\n", mdsc);
3662 	ceph_mdsc_stop(mdsc);
3663 
3664 	/* flush out any connection work with references to us */
3665 	ceph_msgr_flush();
3666 
3667 	fsc->mdsc = NULL;
3668 	kfree(mdsc);
3669 	dout("mdsc_destroy %p done\n", mdsc);
3670 }
3671 
3672 
3673 /*
3674  * handle mds map update.
3675  */
3676 void ceph_mdsc_handle_map(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
3677 {
3678 	u32 epoch;
3679 	u32 maplen;
3680 	void *p = msg->front.iov_base;
3681 	void *end = p + msg->front.iov_len;
3682 	struct ceph_mdsmap *newmap, *oldmap;
3683 	struct ceph_fsid fsid;
3684 	int err = -EINVAL;
3685 
3686 	ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
3687 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
3688 	if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
3689 		return;
3690 	epoch = ceph_decode_32(&p);
3691 	maplen = ceph_decode_32(&p);
3692 	dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
3693 
3694 	/* do we need it? */
3695 	ceph_monc_got_mdsmap(&mdsc->fsc->client->monc, epoch);
3696 	mutex_lock(&mdsc->mutex);
3697 	if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
3698 		dout("handle_map epoch %u <= our %u\n",
3699 		     epoch, mdsc->mdsmap->m_epoch);
3700 		mutex_unlock(&mdsc->mutex);
3701 		return;
3702 	}
3703 
3704 	newmap = ceph_mdsmap_decode(&p, end);
3705 	if (IS_ERR(newmap)) {
3706 		err = PTR_ERR(newmap);
3707 		goto bad_unlock;
3708 	}
3709 
3710 	/* swap into place */
3711 	if (mdsc->mdsmap) {
3712 		oldmap = mdsc->mdsmap;
3713 		mdsc->mdsmap = newmap;
3714 		check_new_map(mdsc, newmap, oldmap);
3715 		ceph_mdsmap_destroy(oldmap);
3716 	} else {
3717 		mdsc->mdsmap = newmap;  /* first mds map */
3718 	}
3719 	mdsc->fsc->sb->s_maxbytes = mdsc->mdsmap->m_max_file_size;
3720 
3721 	__wake_requests(mdsc, &mdsc->waiting_for_map);
3722 
3723 	mutex_unlock(&mdsc->mutex);
3724 	schedule_delayed(mdsc);
3725 	return;
3726 
3727 bad_unlock:
3728 	mutex_unlock(&mdsc->mutex);
3729 bad:
3730 	pr_err("error decoding mdsmap %d\n", err);
3731 	return;
3732 }
3733 
3734 static struct ceph_connection *con_get(struct ceph_connection *con)
3735 {
3736 	struct ceph_mds_session *s = con->private;
3737 
3738 	if (get_session(s)) {
3739 		dout("mdsc con_get %p ok (%d)\n", s, atomic_read(&s->s_ref));
3740 		return con;
3741 	}
3742 	dout("mdsc con_get %p FAIL\n", s);
3743 	return NULL;
3744 }
3745 
3746 static void con_put(struct ceph_connection *con)
3747 {
3748 	struct ceph_mds_session *s = con->private;
3749 
3750 	dout("mdsc con_put %p (%d)\n", s, atomic_read(&s->s_ref) - 1);
3751 	ceph_put_mds_session(s);
3752 }
3753 
3754 /*
3755  * if the client is unresponsive for long enough, the mds will kill
3756  * the session entirely.
3757  */
3758 static void peer_reset(struct ceph_connection *con)
3759 {
3760 	struct ceph_mds_session *s = con->private;
3761 	struct ceph_mds_client *mdsc = s->s_mdsc;
3762 
3763 	pr_warn("mds%d closed our session\n", s->s_mds);
3764 	send_mds_reconnect(mdsc, s);
3765 }
3766 
3767 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
3768 {
3769 	struct ceph_mds_session *s = con->private;
3770 	struct ceph_mds_client *mdsc = s->s_mdsc;
3771 	int type = le16_to_cpu(msg->hdr.type);
3772 
3773 	mutex_lock(&mdsc->mutex);
3774 	if (__verify_registered_session(mdsc, s) < 0) {
3775 		mutex_unlock(&mdsc->mutex);
3776 		goto out;
3777 	}
3778 	mutex_unlock(&mdsc->mutex);
3779 
3780 	switch (type) {
3781 	case CEPH_MSG_MDS_MAP:
3782 		ceph_mdsc_handle_map(mdsc, msg);
3783 		break;
3784 	case CEPH_MSG_CLIENT_SESSION:
3785 		handle_session(s, msg);
3786 		break;
3787 	case CEPH_MSG_CLIENT_REPLY:
3788 		handle_reply(s, msg);
3789 		break;
3790 	case CEPH_MSG_CLIENT_REQUEST_FORWARD:
3791 		handle_forward(mdsc, s, msg);
3792 		break;
3793 	case CEPH_MSG_CLIENT_CAPS:
3794 		ceph_handle_caps(s, msg);
3795 		break;
3796 	case CEPH_MSG_CLIENT_SNAP:
3797 		ceph_handle_snap(mdsc, s, msg);
3798 		break;
3799 	case CEPH_MSG_CLIENT_LEASE:
3800 		handle_lease(mdsc, s, msg);
3801 		break;
3802 
3803 	default:
3804 		pr_err("received unknown message type %d %s\n", type,
3805 		       ceph_msg_type_name(type));
3806 	}
3807 out:
3808 	ceph_msg_put(msg);
3809 }
3810 
3811 /*
3812  * authentication
3813  */
3814 
3815 /*
3816  * Note: returned pointer is the address of a structure that's
3817  * managed separately.  Caller must *not* attempt to free it.
3818  */
3819 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
3820 					int *proto, int force_new)
3821 {
3822 	struct ceph_mds_session *s = con->private;
3823 	struct ceph_mds_client *mdsc = s->s_mdsc;
3824 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3825 	struct ceph_auth_handshake *auth = &s->s_auth;
3826 
3827 	if (force_new && auth->authorizer) {
3828 		ceph_auth_destroy_authorizer(ac, auth->authorizer);
3829 		auth->authorizer = NULL;
3830 	}
3831 	if (!auth->authorizer) {
3832 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3833 						      auth);
3834 		if (ret)
3835 			return ERR_PTR(ret);
3836 	} else {
3837 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
3838 						      auth);
3839 		if (ret)
3840 			return ERR_PTR(ret);
3841 	}
3842 	*proto = ac->protocol;
3843 
3844 	return auth;
3845 }
3846 
3847 
3848 static int verify_authorizer_reply(struct ceph_connection *con, int len)
3849 {
3850 	struct ceph_mds_session *s = con->private;
3851 	struct ceph_mds_client *mdsc = s->s_mdsc;
3852 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3853 
3854 	return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer, len);
3855 }
3856 
3857 static int invalidate_authorizer(struct ceph_connection *con)
3858 {
3859 	struct ceph_mds_session *s = con->private;
3860 	struct ceph_mds_client *mdsc = s->s_mdsc;
3861 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
3862 
3863 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
3864 
3865 	return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
3866 }
3867 
3868 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
3869 				struct ceph_msg_header *hdr, int *skip)
3870 {
3871 	struct ceph_msg *msg;
3872 	int type = (int) le16_to_cpu(hdr->type);
3873 	int front_len = (int) le32_to_cpu(hdr->front_len);
3874 
3875 	if (con->in_msg)
3876 		return con->in_msg;
3877 
3878 	*skip = 0;
3879 	msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
3880 	if (!msg) {
3881 		pr_err("unable to allocate msg type %d len %d\n",
3882 		       type, front_len);
3883 		return NULL;
3884 	}
3885 
3886 	return msg;
3887 }
3888 
3889 static int sign_message(struct ceph_connection *con, struct ceph_msg *msg)
3890 {
3891        struct ceph_mds_session *s = con->private;
3892        struct ceph_auth_handshake *auth = &s->s_auth;
3893        return ceph_auth_sign_message(auth, msg);
3894 }
3895 
3896 static int check_message_signature(struct ceph_connection *con, struct ceph_msg *msg)
3897 {
3898        struct ceph_mds_session *s = con->private;
3899        struct ceph_auth_handshake *auth = &s->s_auth;
3900        return ceph_auth_check_message_signature(auth, msg);
3901 }
3902 
3903 static const struct ceph_connection_operations mds_con_ops = {
3904 	.get = con_get,
3905 	.put = con_put,
3906 	.dispatch = dispatch,
3907 	.get_authorizer = get_authorizer,
3908 	.verify_authorizer_reply = verify_authorizer_reply,
3909 	.invalidate_authorizer = invalidate_authorizer,
3910 	.peer_reset = peer_reset,
3911 	.alloc_msg = mds_alloc_msg,
3912 	.sign_message = sign_message,
3913 	.check_message_signature = check_message_signature,
3914 };
3915 
3916 /* eof */
3917