xref: /openbmc/linux/fs/ceph/mds_client.c (revision adb57164)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/fs.h>
5 #include <linux/wait.h>
6 #include <linux/slab.h>
7 #include <linux/gfp.h>
8 #include <linux/sched.h>
9 #include <linux/debugfs.h>
10 #include <linux/seq_file.h>
11 #include <linux/ratelimit.h>
12 #include <linux/bits.h>
13 
14 #include "super.h"
15 #include "mds_client.h"
16 
17 #include <linux/ceph/ceph_features.h>
18 #include <linux/ceph/messenger.h>
19 #include <linux/ceph/decode.h>
20 #include <linux/ceph/pagelist.h>
21 #include <linux/ceph/auth.h>
22 #include <linux/ceph/debugfs.h>
23 
24 #define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
25 
26 /*
27  * A cluster of MDS (metadata server) daemons is responsible for
28  * managing the file system namespace (the directory hierarchy and
29  * inodes) and for coordinating shared access to storage.  Metadata is
30  * partitioning hierarchically across a number of servers, and that
31  * partition varies over time as the cluster adjusts the distribution
32  * in order to balance load.
33  *
34  * The MDS client is primarily responsible to managing synchronous
35  * metadata requests for operations like open, unlink, and so forth.
36  * If there is a MDS failure, we find out about it when we (possibly
37  * request and) receive a new MDS map, and can resubmit affected
38  * requests.
39  *
40  * For the most part, though, we take advantage of a lossless
41  * communications channel to the MDS, and do not need to worry about
42  * timing out or resubmitting requests.
43  *
44  * We maintain a stateful "session" with each MDS we interact with.
45  * Within each session, we sent periodic heartbeat messages to ensure
46  * any capabilities or leases we have been issues remain valid.  If
47  * the session times out and goes stale, our leases and capabilities
48  * are no longer valid.
49  */
50 
51 struct ceph_reconnect_state {
52 	struct ceph_mds_session *session;
53 	int nr_caps, nr_realms;
54 	struct ceph_pagelist *pagelist;
55 	unsigned msg_version;
56 	bool allow_multi;
57 };
58 
59 static void __wake_requests(struct ceph_mds_client *mdsc,
60 			    struct list_head *head);
61 static void ceph_cap_release_work(struct work_struct *work);
62 static void ceph_cap_reclaim_work(struct work_struct *work);
63 
64 static const struct ceph_connection_operations mds_con_ops;
65 
66 
67 /*
68  * mds reply parsing
69  */
70 
71 static int parse_reply_info_quota(void **p, void *end,
72 				  struct ceph_mds_reply_info_in *info)
73 {
74 	u8 struct_v, struct_compat;
75 	u32 struct_len;
76 
77 	ceph_decode_8_safe(p, end, struct_v, bad);
78 	ceph_decode_8_safe(p, end, struct_compat, bad);
79 	/* struct_v is expected to be >= 1. we only
80 	 * understand encoding with struct_compat == 1. */
81 	if (!struct_v || struct_compat != 1)
82 		goto bad;
83 	ceph_decode_32_safe(p, end, struct_len, bad);
84 	ceph_decode_need(p, end, struct_len, bad);
85 	end = *p + struct_len;
86 	ceph_decode_64_safe(p, end, info->max_bytes, bad);
87 	ceph_decode_64_safe(p, end, info->max_files, bad);
88 	*p = end;
89 	return 0;
90 bad:
91 	return -EIO;
92 }
93 
94 /*
95  * parse individual inode info
96  */
97 static int parse_reply_info_in(void **p, void *end,
98 			       struct ceph_mds_reply_info_in *info,
99 			       u64 features)
100 {
101 	int err = 0;
102 	u8 struct_v = 0;
103 
104 	if (features == (u64)-1) {
105 		u32 struct_len;
106 		u8 struct_compat;
107 		ceph_decode_8_safe(p, end, struct_v, bad);
108 		ceph_decode_8_safe(p, end, struct_compat, bad);
109 		/* struct_v is expected to be >= 1. we only understand
110 		 * encoding with struct_compat == 1. */
111 		if (!struct_v || struct_compat != 1)
112 			goto bad;
113 		ceph_decode_32_safe(p, end, struct_len, bad);
114 		ceph_decode_need(p, end, struct_len, bad);
115 		end = *p + struct_len;
116 	}
117 
118 	ceph_decode_need(p, end, sizeof(struct ceph_mds_reply_inode), bad);
119 	info->in = *p;
120 	*p += sizeof(struct ceph_mds_reply_inode) +
121 		sizeof(*info->in->fragtree.splits) *
122 		le32_to_cpu(info->in->fragtree.nsplits);
123 
124 	ceph_decode_32_safe(p, end, info->symlink_len, bad);
125 	ceph_decode_need(p, end, info->symlink_len, bad);
126 	info->symlink = *p;
127 	*p += info->symlink_len;
128 
129 	ceph_decode_copy_safe(p, end, &info->dir_layout,
130 			      sizeof(info->dir_layout), bad);
131 	ceph_decode_32_safe(p, end, info->xattr_len, bad);
132 	ceph_decode_need(p, end, info->xattr_len, bad);
133 	info->xattr_data = *p;
134 	*p += info->xattr_len;
135 
136 	if (features == (u64)-1) {
137 		/* inline data */
138 		ceph_decode_64_safe(p, end, info->inline_version, bad);
139 		ceph_decode_32_safe(p, end, info->inline_len, bad);
140 		ceph_decode_need(p, end, info->inline_len, bad);
141 		info->inline_data = *p;
142 		*p += info->inline_len;
143 		/* quota */
144 		err = parse_reply_info_quota(p, end, info);
145 		if (err < 0)
146 			goto out_bad;
147 		/* pool namespace */
148 		ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
149 		if (info->pool_ns_len > 0) {
150 			ceph_decode_need(p, end, info->pool_ns_len, bad);
151 			info->pool_ns_data = *p;
152 			*p += info->pool_ns_len;
153 		}
154 
155 		/* btime */
156 		ceph_decode_need(p, end, sizeof(info->btime), bad);
157 		ceph_decode_copy(p, &info->btime, sizeof(info->btime));
158 
159 		/* change attribute */
160 		ceph_decode_64_safe(p, end, info->change_attr, bad);
161 
162 		/* dir pin */
163 		if (struct_v >= 2) {
164 			ceph_decode_32_safe(p, end, info->dir_pin, bad);
165 		} else {
166 			info->dir_pin = -ENODATA;
167 		}
168 
169 		/* snapshot birth time, remains zero for v<=2 */
170 		if (struct_v >= 3) {
171 			ceph_decode_need(p, end, sizeof(info->snap_btime), bad);
172 			ceph_decode_copy(p, &info->snap_btime,
173 					 sizeof(info->snap_btime));
174 		} else {
175 			memset(&info->snap_btime, 0, sizeof(info->snap_btime));
176 		}
177 
178 		*p = end;
179 	} else {
180 		if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
181 			ceph_decode_64_safe(p, end, info->inline_version, bad);
182 			ceph_decode_32_safe(p, end, info->inline_len, bad);
183 			ceph_decode_need(p, end, info->inline_len, bad);
184 			info->inline_data = *p;
185 			*p += info->inline_len;
186 		} else
187 			info->inline_version = CEPH_INLINE_NONE;
188 
189 		if (features & CEPH_FEATURE_MDS_QUOTA) {
190 			err = parse_reply_info_quota(p, end, info);
191 			if (err < 0)
192 				goto out_bad;
193 		} else {
194 			info->max_bytes = 0;
195 			info->max_files = 0;
196 		}
197 
198 		info->pool_ns_len = 0;
199 		info->pool_ns_data = NULL;
200 		if (features & CEPH_FEATURE_FS_FILE_LAYOUT_V2) {
201 			ceph_decode_32_safe(p, end, info->pool_ns_len, bad);
202 			if (info->pool_ns_len > 0) {
203 				ceph_decode_need(p, end, info->pool_ns_len, bad);
204 				info->pool_ns_data = *p;
205 				*p += info->pool_ns_len;
206 			}
207 		}
208 
209 		if (features & CEPH_FEATURE_FS_BTIME) {
210 			ceph_decode_need(p, end, sizeof(info->btime), bad);
211 			ceph_decode_copy(p, &info->btime, sizeof(info->btime));
212 			ceph_decode_64_safe(p, end, info->change_attr, bad);
213 		}
214 
215 		info->dir_pin = -ENODATA;
216 		/* info->snap_btime remains zero */
217 	}
218 	return 0;
219 bad:
220 	err = -EIO;
221 out_bad:
222 	return err;
223 }
224 
225 static int parse_reply_info_dir(void **p, void *end,
226 				struct ceph_mds_reply_dirfrag **dirfrag,
227 				u64 features)
228 {
229 	if (features == (u64)-1) {
230 		u8 struct_v, struct_compat;
231 		u32 struct_len;
232 		ceph_decode_8_safe(p, end, struct_v, bad);
233 		ceph_decode_8_safe(p, end, struct_compat, bad);
234 		/* struct_v is expected to be >= 1. we only understand
235 		 * encoding whose struct_compat == 1. */
236 		if (!struct_v || struct_compat != 1)
237 			goto bad;
238 		ceph_decode_32_safe(p, end, struct_len, bad);
239 		ceph_decode_need(p, end, struct_len, bad);
240 		end = *p + struct_len;
241 	}
242 
243 	ceph_decode_need(p, end, sizeof(**dirfrag), bad);
244 	*dirfrag = *p;
245 	*p += sizeof(**dirfrag) + sizeof(u32) * le32_to_cpu((*dirfrag)->ndist);
246 	if (unlikely(*p > end))
247 		goto bad;
248 	if (features == (u64)-1)
249 		*p = end;
250 	return 0;
251 bad:
252 	return -EIO;
253 }
254 
255 static int parse_reply_info_lease(void **p, void *end,
256 				  struct ceph_mds_reply_lease **lease,
257 				  u64 features)
258 {
259 	if (features == (u64)-1) {
260 		u8 struct_v, struct_compat;
261 		u32 struct_len;
262 		ceph_decode_8_safe(p, end, struct_v, bad);
263 		ceph_decode_8_safe(p, end, struct_compat, bad);
264 		/* struct_v is expected to be >= 1. we only understand
265 		 * encoding whose struct_compat == 1. */
266 		if (!struct_v || struct_compat != 1)
267 			goto bad;
268 		ceph_decode_32_safe(p, end, struct_len, bad);
269 		ceph_decode_need(p, end, struct_len, bad);
270 		end = *p + struct_len;
271 	}
272 
273 	ceph_decode_need(p, end, sizeof(**lease), bad);
274 	*lease = *p;
275 	*p += sizeof(**lease);
276 	if (features == (u64)-1)
277 		*p = end;
278 	return 0;
279 bad:
280 	return -EIO;
281 }
282 
283 /*
284  * parse a normal reply, which may contain a (dir+)dentry and/or a
285  * target inode.
286  */
287 static int parse_reply_info_trace(void **p, void *end,
288 				  struct ceph_mds_reply_info_parsed *info,
289 				  u64 features)
290 {
291 	int err;
292 
293 	if (info->head->is_dentry) {
294 		err = parse_reply_info_in(p, end, &info->diri, features);
295 		if (err < 0)
296 			goto out_bad;
297 
298 		err = parse_reply_info_dir(p, end, &info->dirfrag, features);
299 		if (err < 0)
300 			goto out_bad;
301 
302 		ceph_decode_32_safe(p, end, info->dname_len, bad);
303 		ceph_decode_need(p, end, info->dname_len, bad);
304 		info->dname = *p;
305 		*p += info->dname_len;
306 
307 		err = parse_reply_info_lease(p, end, &info->dlease, features);
308 		if (err < 0)
309 			goto out_bad;
310 	}
311 
312 	if (info->head->is_target) {
313 		err = parse_reply_info_in(p, end, &info->targeti, features);
314 		if (err < 0)
315 			goto out_bad;
316 	}
317 
318 	if (unlikely(*p != end))
319 		goto bad;
320 	return 0;
321 
322 bad:
323 	err = -EIO;
324 out_bad:
325 	pr_err("problem parsing mds trace %d\n", err);
326 	return err;
327 }
328 
329 /*
330  * parse readdir results
331  */
332 static int parse_reply_info_readdir(void **p, void *end,
333 				struct ceph_mds_reply_info_parsed *info,
334 				u64 features)
335 {
336 	u32 num, i = 0;
337 	int err;
338 
339 	err = parse_reply_info_dir(p, end, &info->dir_dir, features);
340 	if (err < 0)
341 		goto out_bad;
342 
343 	ceph_decode_need(p, end, sizeof(num) + 2, bad);
344 	num = ceph_decode_32(p);
345 	{
346 		u16 flags = ceph_decode_16(p);
347 		info->dir_end = !!(flags & CEPH_READDIR_FRAG_END);
348 		info->dir_complete = !!(flags & CEPH_READDIR_FRAG_COMPLETE);
349 		info->hash_order = !!(flags & CEPH_READDIR_HASH_ORDER);
350 		info->offset_hash = !!(flags & CEPH_READDIR_OFFSET_HASH);
351 	}
352 	if (num == 0)
353 		goto done;
354 
355 	BUG_ON(!info->dir_entries);
356 	if ((unsigned long)(info->dir_entries + num) >
357 	    (unsigned long)info->dir_entries + info->dir_buf_size) {
358 		pr_err("dir contents are larger than expected\n");
359 		WARN_ON(1);
360 		goto bad;
361 	}
362 
363 	info->dir_nr = num;
364 	while (num) {
365 		struct ceph_mds_reply_dir_entry *rde = info->dir_entries + i;
366 		/* dentry */
367 		ceph_decode_32_safe(p, end, rde->name_len, bad);
368 		ceph_decode_need(p, end, rde->name_len, bad);
369 		rde->name = *p;
370 		*p += rde->name_len;
371 		dout("parsed dir dname '%.*s'\n", rde->name_len, rde->name);
372 
373 		/* dentry lease */
374 		err = parse_reply_info_lease(p, end, &rde->lease, features);
375 		if (err)
376 			goto out_bad;
377 		/* inode */
378 		err = parse_reply_info_in(p, end, &rde->inode, features);
379 		if (err < 0)
380 			goto out_bad;
381 		/* ceph_readdir_prepopulate() will update it */
382 		rde->offset = 0;
383 		i++;
384 		num--;
385 	}
386 
387 done:
388 	/* Skip over any unrecognized fields */
389 	*p = end;
390 	return 0;
391 
392 bad:
393 	err = -EIO;
394 out_bad:
395 	pr_err("problem parsing dir contents %d\n", err);
396 	return err;
397 }
398 
399 /*
400  * parse fcntl F_GETLK results
401  */
402 static int parse_reply_info_filelock(void **p, void *end,
403 				     struct ceph_mds_reply_info_parsed *info,
404 				     u64 features)
405 {
406 	if (*p + sizeof(*info->filelock_reply) > end)
407 		goto bad;
408 
409 	info->filelock_reply = *p;
410 
411 	/* Skip over any unrecognized fields */
412 	*p = end;
413 	return 0;
414 bad:
415 	return -EIO;
416 }
417 
418 
419 #if BITS_PER_LONG == 64
420 
421 #define DELEGATED_INO_AVAILABLE		xa_mk_value(1)
422 
423 static int ceph_parse_deleg_inos(void **p, void *end,
424 				 struct ceph_mds_session *s)
425 {
426 	u32 sets;
427 
428 	ceph_decode_32_safe(p, end, sets, bad);
429 	dout("got %u sets of delegated inodes\n", sets);
430 	while (sets--) {
431 		u64 start, len, ino;
432 
433 		ceph_decode_64_safe(p, end, start, bad);
434 		ceph_decode_64_safe(p, end, len, bad);
435 		while (len--) {
436 			int err = xa_insert(&s->s_delegated_inos, ino = start++,
437 					    DELEGATED_INO_AVAILABLE,
438 					    GFP_KERNEL);
439 			if (!err) {
440 				dout("added delegated inode 0x%llx\n",
441 				     start - 1);
442 			} else if (err == -EBUSY) {
443 				pr_warn("ceph: MDS delegated inode 0x%llx more than once.\n",
444 					start - 1);
445 			} else {
446 				return err;
447 			}
448 		}
449 	}
450 	return 0;
451 bad:
452 	return -EIO;
453 }
454 
455 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
456 {
457 	unsigned long ino;
458 	void *val;
459 
460 	xa_for_each(&s->s_delegated_inos, ino, val) {
461 		val = xa_erase(&s->s_delegated_inos, ino);
462 		if (val == DELEGATED_INO_AVAILABLE)
463 			return ino;
464 	}
465 	return 0;
466 }
467 
468 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
469 {
470 	return xa_insert(&s->s_delegated_inos, ino, DELEGATED_INO_AVAILABLE,
471 			 GFP_KERNEL);
472 }
473 #else /* BITS_PER_LONG == 64 */
474 /*
475  * FIXME: xarrays can't handle 64-bit indexes on a 32-bit arch. For now, just
476  * ignore delegated_inos on 32 bit arch. Maybe eventually add xarrays for top
477  * and bottom words?
478  */
479 static int ceph_parse_deleg_inos(void **p, void *end,
480 				 struct ceph_mds_session *s)
481 {
482 	u32 sets;
483 
484 	ceph_decode_32_safe(p, end, sets, bad);
485 	if (sets)
486 		ceph_decode_skip_n(p, end, sets * 2 * sizeof(__le64), bad);
487 	return 0;
488 bad:
489 	return -EIO;
490 }
491 
492 u64 ceph_get_deleg_ino(struct ceph_mds_session *s)
493 {
494 	return 0;
495 }
496 
497 int ceph_restore_deleg_ino(struct ceph_mds_session *s, u64 ino)
498 {
499 	return 0;
500 }
501 #endif /* BITS_PER_LONG == 64 */
502 
503 /*
504  * parse create results
505  */
506 static int parse_reply_info_create(void **p, void *end,
507 				  struct ceph_mds_reply_info_parsed *info,
508 				  u64 features, struct ceph_mds_session *s)
509 {
510 	int ret;
511 
512 	if (features == (u64)-1 ||
513 	    (features & CEPH_FEATURE_REPLY_CREATE_INODE)) {
514 		if (*p == end) {
515 			/* Malformed reply? */
516 			info->has_create_ino = false;
517 		} else if (test_bit(CEPHFS_FEATURE_DELEG_INO, &s->s_features)) {
518 			u8 struct_v, struct_compat;
519 			u32 len;
520 
521 			info->has_create_ino = true;
522 			ceph_decode_8_safe(p, end, struct_v, bad);
523 			ceph_decode_8_safe(p, end, struct_compat, bad);
524 			ceph_decode_32_safe(p, end, len, bad);
525 			ceph_decode_64_safe(p, end, info->ino, bad);
526 			ret = ceph_parse_deleg_inos(p, end, s);
527 			if (ret)
528 				return ret;
529 		} else {
530 			/* legacy */
531 			ceph_decode_64_safe(p, end, info->ino, bad);
532 			info->has_create_ino = true;
533 		}
534 	} else {
535 		if (*p != end)
536 			goto bad;
537 	}
538 
539 	/* Skip over any unrecognized fields */
540 	*p = end;
541 	return 0;
542 bad:
543 	return -EIO;
544 }
545 
546 /*
547  * parse extra results
548  */
549 static int parse_reply_info_extra(void **p, void *end,
550 				  struct ceph_mds_reply_info_parsed *info,
551 				  u64 features, struct ceph_mds_session *s)
552 {
553 	u32 op = le32_to_cpu(info->head->op);
554 
555 	if (op == CEPH_MDS_OP_GETFILELOCK)
556 		return parse_reply_info_filelock(p, end, info, features);
557 	else if (op == CEPH_MDS_OP_READDIR || op == CEPH_MDS_OP_LSSNAP)
558 		return parse_reply_info_readdir(p, end, info, features);
559 	else if (op == CEPH_MDS_OP_CREATE)
560 		return parse_reply_info_create(p, end, info, features, s);
561 	else
562 		return -EIO;
563 }
564 
565 /*
566  * parse entire mds reply
567  */
568 static int parse_reply_info(struct ceph_mds_session *s, struct ceph_msg *msg,
569 			    struct ceph_mds_reply_info_parsed *info,
570 			    u64 features)
571 {
572 	void *p, *end;
573 	u32 len;
574 	int err;
575 
576 	info->head = msg->front.iov_base;
577 	p = msg->front.iov_base + sizeof(struct ceph_mds_reply_head);
578 	end = p + msg->front.iov_len - sizeof(struct ceph_mds_reply_head);
579 
580 	/* trace */
581 	ceph_decode_32_safe(&p, end, len, bad);
582 	if (len > 0) {
583 		ceph_decode_need(&p, end, len, bad);
584 		err = parse_reply_info_trace(&p, p+len, info, features);
585 		if (err < 0)
586 			goto out_bad;
587 	}
588 
589 	/* extra */
590 	ceph_decode_32_safe(&p, end, len, bad);
591 	if (len > 0) {
592 		ceph_decode_need(&p, end, len, bad);
593 		err = parse_reply_info_extra(&p, p+len, info, features, s);
594 		if (err < 0)
595 			goto out_bad;
596 	}
597 
598 	/* snap blob */
599 	ceph_decode_32_safe(&p, end, len, bad);
600 	info->snapblob_len = len;
601 	info->snapblob = p;
602 	p += len;
603 
604 	if (p != end)
605 		goto bad;
606 	return 0;
607 
608 bad:
609 	err = -EIO;
610 out_bad:
611 	pr_err("mds parse_reply err %d\n", err);
612 	return err;
613 }
614 
615 static void destroy_reply_info(struct ceph_mds_reply_info_parsed *info)
616 {
617 	if (!info->dir_entries)
618 		return;
619 	free_pages((unsigned long)info->dir_entries, get_order(info->dir_buf_size));
620 }
621 
622 
623 /*
624  * sessions
625  */
626 const char *ceph_session_state_name(int s)
627 {
628 	switch (s) {
629 	case CEPH_MDS_SESSION_NEW: return "new";
630 	case CEPH_MDS_SESSION_OPENING: return "opening";
631 	case CEPH_MDS_SESSION_OPEN: return "open";
632 	case CEPH_MDS_SESSION_HUNG: return "hung";
633 	case CEPH_MDS_SESSION_CLOSING: return "closing";
634 	case CEPH_MDS_SESSION_CLOSED: return "closed";
635 	case CEPH_MDS_SESSION_RESTARTING: return "restarting";
636 	case CEPH_MDS_SESSION_RECONNECTING: return "reconnecting";
637 	case CEPH_MDS_SESSION_REJECTED: return "rejected";
638 	default: return "???";
639 	}
640 }
641 
642 struct ceph_mds_session *ceph_get_mds_session(struct ceph_mds_session *s)
643 {
644 	if (refcount_inc_not_zero(&s->s_ref)) {
645 		dout("mdsc get_session %p %d -> %d\n", s,
646 		     refcount_read(&s->s_ref)-1, refcount_read(&s->s_ref));
647 		return s;
648 	} else {
649 		dout("mdsc get_session %p 0 -- FAIL\n", s);
650 		return NULL;
651 	}
652 }
653 
654 void ceph_put_mds_session(struct ceph_mds_session *s)
655 {
656 	dout("mdsc put_session %p %d -> %d\n", s,
657 	     refcount_read(&s->s_ref), refcount_read(&s->s_ref)-1);
658 	if (refcount_dec_and_test(&s->s_ref)) {
659 		if (s->s_auth.authorizer)
660 			ceph_auth_destroy_authorizer(s->s_auth.authorizer);
661 		xa_destroy(&s->s_delegated_inos);
662 		kfree(s);
663 	}
664 }
665 
666 /*
667  * called under mdsc->mutex
668  */
669 struct ceph_mds_session *__ceph_lookup_mds_session(struct ceph_mds_client *mdsc,
670 						   int mds)
671 {
672 	if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
673 		return NULL;
674 	return ceph_get_mds_session(mdsc->sessions[mds]);
675 }
676 
677 static bool __have_session(struct ceph_mds_client *mdsc, int mds)
678 {
679 	if (mds >= mdsc->max_sessions || !mdsc->sessions[mds])
680 		return false;
681 	else
682 		return true;
683 }
684 
685 static int __verify_registered_session(struct ceph_mds_client *mdsc,
686 				       struct ceph_mds_session *s)
687 {
688 	if (s->s_mds >= mdsc->max_sessions ||
689 	    mdsc->sessions[s->s_mds] != s)
690 		return -ENOENT;
691 	return 0;
692 }
693 
694 /*
695  * create+register a new session for given mds.
696  * called under mdsc->mutex.
697  */
698 static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc,
699 						 int mds)
700 {
701 	struct ceph_mds_session *s;
702 
703 	if (mds >= mdsc->mdsmap->possible_max_rank)
704 		return ERR_PTR(-EINVAL);
705 
706 	s = kzalloc(sizeof(*s), GFP_NOFS);
707 	if (!s)
708 		return ERR_PTR(-ENOMEM);
709 
710 	if (mds >= mdsc->max_sessions) {
711 		int newmax = 1 << get_count_order(mds + 1);
712 		struct ceph_mds_session **sa;
713 
714 		dout("%s: realloc to %d\n", __func__, newmax);
715 		sa = kcalloc(newmax, sizeof(void *), GFP_NOFS);
716 		if (!sa)
717 			goto fail_realloc;
718 		if (mdsc->sessions) {
719 			memcpy(sa, mdsc->sessions,
720 			       mdsc->max_sessions * sizeof(void *));
721 			kfree(mdsc->sessions);
722 		}
723 		mdsc->sessions = sa;
724 		mdsc->max_sessions = newmax;
725 	}
726 
727 	dout("%s: mds%d\n", __func__, mds);
728 	s->s_mdsc = mdsc;
729 	s->s_mds = mds;
730 	s->s_state = CEPH_MDS_SESSION_NEW;
731 	s->s_ttl = 0;
732 	s->s_seq = 0;
733 	mutex_init(&s->s_mutex);
734 
735 	ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr);
736 
737 	spin_lock_init(&s->s_gen_ttl_lock);
738 	s->s_cap_gen = 1;
739 	s->s_cap_ttl = jiffies - 1;
740 
741 	spin_lock_init(&s->s_cap_lock);
742 	s->s_renew_requested = 0;
743 	s->s_renew_seq = 0;
744 	INIT_LIST_HEAD(&s->s_caps);
745 	s->s_nr_caps = 0;
746 	refcount_set(&s->s_ref, 1);
747 	INIT_LIST_HEAD(&s->s_waiting);
748 	INIT_LIST_HEAD(&s->s_unsafe);
749 	xa_init(&s->s_delegated_inos);
750 	s->s_num_cap_releases = 0;
751 	s->s_cap_reconnect = 0;
752 	s->s_cap_iterator = NULL;
753 	INIT_LIST_HEAD(&s->s_cap_releases);
754 	INIT_WORK(&s->s_cap_release_work, ceph_cap_release_work);
755 
756 	INIT_LIST_HEAD(&s->s_cap_flushing);
757 
758 	mdsc->sessions[mds] = s;
759 	atomic_inc(&mdsc->num_sessions);
760 	refcount_inc(&s->s_ref);  /* one ref to sessions[], one to caller */
761 
762 	ceph_con_open(&s->s_con, CEPH_ENTITY_TYPE_MDS, mds,
763 		      ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
764 
765 	return s;
766 
767 fail_realloc:
768 	kfree(s);
769 	return ERR_PTR(-ENOMEM);
770 }
771 
772 /*
773  * called under mdsc->mutex
774  */
775 static void __unregister_session(struct ceph_mds_client *mdsc,
776 			       struct ceph_mds_session *s)
777 {
778 	dout("__unregister_session mds%d %p\n", s->s_mds, s);
779 	BUG_ON(mdsc->sessions[s->s_mds] != s);
780 	mdsc->sessions[s->s_mds] = NULL;
781 	ceph_con_close(&s->s_con);
782 	ceph_put_mds_session(s);
783 	atomic_dec(&mdsc->num_sessions);
784 }
785 
786 /*
787  * drop session refs in request.
788  *
789  * should be last request ref, or hold mdsc->mutex
790  */
791 static void put_request_session(struct ceph_mds_request *req)
792 {
793 	if (req->r_session) {
794 		ceph_put_mds_session(req->r_session);
795 		req->r_session = NULL;
796 	}
797 }
798 
799 void ceph_mdsc_release_request(struct kref *kref)
800 {
801 	struct ceph_mds_request *req = container_of(kref,
802 						    struct ceph_mds_request,
803 						    r_kref);
804 	ceph_mdsc_release_dir_caps(req);
805 	destroy_reply_info(&req->r_reply_info);
806 	if (req->r_request)
807 		ceph_msg_put(req->r_request);
808 	if (req->r_reply)
809 		ceph_msg_put(req->r_reply);
810 	if (req->r_inode) {
811 		ceph_put_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
812 		/* avoid calling iput_final() in mds dispatch threads */
813 		ceph_async_iput(req->r_inode);
814 	}
815 	if (req->r_parent) {
816 		ceph_put_cap_refs(ceph_inode(req->r_parent), CEPH_CAP_PIN);
817 		ceph_async_iput(req->r_parent);
818 	}
819 	ceph_async_iput(req->r_target_inode);
820 	if (req->r_dentry)
821 		dput(req->r_dentry);
822 	if (req->r_old_dentry)
823 		dput(req->r_old_dentry);
824 	if (req->r_old_dentry_dir) {
825 		/*
826 		 * track (and drop pins for) r_old_dentry_dir
827 		 * separately, since r_old_dentry's d_parent may have
828 		 * changed between the dir mutex being dropped and
829 		 * this request being freed.
830 		 */
831 		ceph_put_cap_refs(ceph_inode(req->r_old_dentry_dir),
832 				  CEPH_CAP_PIN);
833 		ceph_async_iput(req->r_old_dentry_dir);
834 	}
835 	kfree(req->r_path1);
836 	kfree(req->r_path2);
837 	if (req->r_pagelist)
838 		ceph_pagelist_release(req->r_pagelist);
839 	put_request_session(req);
840 	ceph_unreserve_caps(req->r_mdsc, &req->r_caps_reservation);
841 	WARN_ON_ONCE(!list_empty(&req->r_wait));
842 	kmem_cache_free(ceph_mds_request_cachep, req);
843 }
844 
845 DEFINE_RB_FUNCS(request, struct ceph_mds_request, r_tid, r_node)
846 
847 /*
848  * lookup session, bump ref if found.
849  *
850  * called under mdsc->mutex.
851  */
852 static struct ceph_mds_request *
853 lookup_get_request(struct ceph_mds_client *mdsc, u64 tid)
854 {
855 	struct ceph_mds_request *req;
856 
857 	req = lookup_request(&mdsc->request_tree, tid);
858 	if (req)
859 		ceph_mdsc_get_request(req);
860 
861 	return req;
862 }
863 
864 /*
865  * Register an in-flight request, and assign a tid.  Link to directory
866  * are modifying (if any).
867  *
868  * Called under mdsc->mutex.
869  */
870 static void __register_request(struct ceph_mds_client *mdsc,
871 			       struct ceph_mds_request *req,
872 			       struct inode *dir)
873 {
874 	int ret = 0;
875 
876 	req->r_tid = ++mdsc->last_tid;
877 	if (req->r_num_caps) {
878 		ret = ceph_reserve_caps(mdsc, &req->r_caps_reservation,
879 					req->r_num_caps);
880 		if (ret < 0) {
881 			pr_err("__register_request %p "
882 			       "failed to reserve caps: %d\n", req, ret);
883 			/* set req->r_err to fail early from __do_request */
884 			req->r_err = ret;
885 			return;
886 		}
887 	}
888 	dout("__register_request %p tid %lld\n", req, req->r_tid);
889 	ceph_mdsc_get_request(req);
890 	insert_request(&mdsc->request_tree, req);
891 
892 	req->r_uid = current_fsuid();
893 	req->r_gid = current_fsgid();
894 
895 	if (mdsc->oldest_tid == 0 && req->r_op != CEPH_MDS_OP_SETFILELOCK)
896 		mdsc->oldest_tid = req->r_tid;
897 
898 	if (dir) {
899 		struct ceph_inode_info *ci = ceph_inode(dir);
900 
901 		ihold(dir);
902 		req->r_unsafe_dir = dir;
903 		spin_lock(&ci->i_unsafe_lock);
904 		list_add_tail(&req->r_unsafe_dir_item, &ci->i_unsafe_dirops);
905 		spin_unlock(&ci->i_unsafe_lock);
906 	}
907 }
908 
909 static void __unregister_request(struct ceph_mds_client *mdsc,
910 				 struct ceph_mds_request *req)
911 {
912 	dout("__unregister_request %p tid %lld\n", req, req->r_tid);
913 
914 	/* Never leave an unregistered request on an unsafe list! */
915 	list_del_init(&req->r_unsafe_item);
916 
917 	if (req->r_tid == mdsc->oldest_tid) {
918 		struct rb_node *p = rb_next(&req->r_node);
919 		mdsc->oldest_tid = 0;
920 		while (p) {
921 			struct ceph_mds_request *next_req =
922 				rb_entry(p, struct ceph_mds_request, r_node);
923 			if (next_req->r_op != CEPH_MDS_OP_SETFILELOCK) {
924 				mdsc->oldest_tid = next_req->r_tid;
925 				break;
926 			}
927 			p = rb_next(p);
928 		}
929 	}
930 
931 	erase_request(&mdsc->request_tree, req);
932 
933 	if (req->r_unsafe_dir) {
934 		struct ceph_inode_info *ci = ceph_inode(req->r_unsafe_dir);
935 		spin_lock(&ci->i_unsafe_lock);
936 		list_del_init(&req->r_unsafe_dir_item);
937 		spin_unlock(&ci->i_unsafe_lock);
938 	}
939 	if (req->r_target_inode &&
940 	    test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
941 		struct ceph_inode_info *ci = ceph_inode(req->r_target_inode);
942 		spin_lock(&ci->i_unsafe_lock);
943 		list_del_init(&req->r_unsafe_target_item);
944 		spin_unlock(&ci->i_unsafe_lock);
945 	}
946 
947 	if (req->r_unsafe_dir) {
948 		/* avoid calling iput_final() in mds dispatch threads */
949 		ceph_async_iput(req->r_unsafe_dir);
950 		req->r_unsafe_dir = NULL;
951 	}
952 
953 	complete_all(&req->r_safe_completion);
954 
955 	ceph_mdsc_put_request(req);
956 }
957 
958 /*
959  * Walk back up the dentry tree until we hit a dentry representing a
960  * non-snapshot inode. We do this using the rcu_read_lock (which must be held
961  * when calling this) to ensure that the objects won't disappear while we're
962  * working with them. Once we hit a candidate dentry, we attempt to take a
963  * reference to it, and return that as the result.
964  */
965 static struct inode *get_nonsnap_parent(struct dentry *dentry)
966 {
967 	struct inode *inode = NULL;
968 
969 	while (dentry && !IS_ROOT(dentry)) {
970 		inode = d_inode_rcu(dentry);
971 		if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
972 			break;
973 		dentry = dentry->d_parent;
974 	}
975 	if (inode)
976 		inode = igrab(inode);
977 	return inode;
978 }
979 
980 /*
981  * Choose mds to send request to next.  If there is a hint set in the
982  * request (e.g., due to a prior forward hint from the mds), use that.
983  * Otherwise, consult frag tree and/or caps to identify the
984  * appropriate mds.  If all else fails, choose randomly.
985  *
986  * Called under mdsc->mutex.
987  */
988 static int __choose_mds(struct ceph_mds_client *mdsc,
989 			struct ceph_mds_request *req,
990 			bool *random)
991 {
992 	struct inode *inode;
993 	struct ceph_inode_info *ci;
994 	struct ceph_cap *cap;
995 	int mode = req->r_direct_mode;
996 	int mds = -1;
997 	u32 hash = req->r_direct_hash;
998 	bool is_hash = test_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
999 
1000 	if (random)
1001 		*random = false;
1002 
1003 	/*
1004 	 * is there a specific mds we should try?  ignore hint if we have
1005 	 * no session and the mds is not up (active or recovering).
1006 	 */
1007 	if (req->r_resend_mds >= 0 &&
1008 	    (__have_session(mdsc, req->r_resend_mds) ||
1009 	     ceph_mdsmap_get_state(mdsc->mdsmap, req->r_resend_mds) > 0)) {
1010 		dout("%s using resend_mds mds%d\n", __func__,
1011 		     req->r_resend_mds);
1012 		return req->r_resend_mds;
1013 	}
1014 
1015 	if (mode == USE_RANDOM_MDS)
1016 		goto random;
1017 
1018 	inode = NULL;
1019 	if (req->r_inode) {
1020 		if (ceph_snap(req->r_inode) != CEPH_SNAPDIR) {
1021 			inode = req->r_inode;
1022 			ihold(inode);
1023 		} else {
1024 			/* req->r_dentry is non-null for LSSNAP request */
1025 			rcu_read_lock();
1026 			inode = get_nonsnap_parent(req->r_dentry);
1027 			rcu_read_unlock();
1028 			dout("%s using snapdir's parent %p\n", __func__, inode);
1029 		}
1030 	} else if (req->r_dentry) {
1031 		/* ignore race with rename; old or new d_parent is okay */
1032 		struct dentry *parent;
1033 		struct inode *dir;
1034 
1035 		rcu_read_lock();
1036 		parent = READ_ONCE(req->r_dentry->d_parent);
1037 		dir = req->r_parent ? : d_inode_rcu(parent);
1038 
1039 		if (!dir || dir->i_sb != mdsc->fsc->sb) {
1040 			/*  not this fs or parent went negative */
1041 			inode = d_inode(req->r_dentry);
1042 			if (inode)
1043 				ihold(inode);
1044 		} else if (ceph_snap(dir) != CEPH_NOSNAP) {
1045 			/* direct snapped/virtual snapdir requests
1046 			 * based on parent dir inode */
1047 			inode = get_nonsnap_parent(parent);
1048 			dout("%s using nonsnap parent %p\n", __func__, inode);
1049 		} else {
1050 			/* dentry target */
1051 			inode = d_inode(req->r_dentry);
1052 			if (!inode || mode == USE_AUTH_MDS) {
1053 				/* dir + name */
1054 				inode = igrab(dir);
1055 				hash = ceph_dentry_hash(dir, req->r_dentry);
1056 				is_hash = true;
1057 			} else {
1058 				ihold(inode);
1059 			}
1060 		}
1061 		rcu_read_unlock();
1062 	}
1063 
1064 	dout("%s %p is_hash=%d (0x%x) mode %d\n", __func__, inode, (int)is_hash,
1065 	     hash, mode);
1066 	if (!inode)
1067 		goto random;
1068 	ci = ceph_inode(inode);
1069 
1070 	if (is_hash && S_ISDIR(inode->i_mode)) {
1071 		struct ceph_inode_frag frag;
1072 		int found;
1073 
1074 		ceph_choose_frag(ci, hash, &frag, &found);
1075 		if (found) {
1076 			if (mode == USE_ANY_MDS && frag.ndist > 0) {
1077 				u8 r;
1078 
1079 				/* choose a random replica */
1080 				get_random_bytes(&r, 1);
1081 				r %= frag.ndist;
1082 				mds = frag.dist[r];
1083 				dout("%s %p %llx.%llx frag %u mds%d (%d/%d)\n",
1084 				     __func__, inode, ceph_vinop(inode),
1085 				     frag.frag, mds, (int)r, frag.ndist);
1086 				if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1087 				    CEPH_MDS_STATE_ACTIVE &&
1088 				    !ceph_mdsmap_is_laggy(mdsc->mdsmap, mds))
1089 					goto out;
1090 			}
1091 
1092 			/* since this file/dir wasn't known to be
1093 			 * replicated, then we want to look for the
1094 			 * authoritative mds. */
1095 			if (frag.mds >= 0) {
1096 				/* choose auth mds */
1097 				mds = frag.mds;
1098 				dout("%s %p %llx.%llx frag %u mds%d (auth)\n",
1099 				     __func__, inode, ceph_vinop(inode),
1100 				     frag.frag, mds);
1101 				if (ceph_mdsmap_get_state(mdsc->mdsmap, mds) >=
1102 				    CEPH_MDS_STATE_ACTIVE) {
1103 					if (mode == USE_ANY_MDS &&
1104 					    !ceph_mdsmap_is_laggy(mdsc->mdsmap,
1105 								  mds))
1106 						goto out;
1107 				}
1108 			}
1109 			mode = USE_AUTH_MDS;
1110 		}
1111 	}
1112 
1113 	spin_lock(&ci->i_ceph_lock);
1114 	cap = NULL;
1115 	if (mode == USE_AUTH_MDS)
1116 		cap = ci->i_auth_cap;
1117 	if (!cap && !RB_EMPTY_ROOT(&ci->i_caps))
1118 		cap = rb_entry(rb_first(&ci->i_caps), struct ceph_cap, ci_node);
1119 	if (!cap) {
1120 		spin_unlock(&ci->i_ceph_lock);
1121 		ceph_async_iput(inode);
1122 		goto random;
1123 	}
1124 	mds = cap->session->s_mds;
1125 	dout("%s %p %llx.%llx mds%d (%scap %p)\n", __func__,
1126 	     inode, ceph_vinop(inode), mds,
1127 	     cap == ci->i_auth_cap ? "auth " : "", cap);
1128 	spin_unlock(&ci->i_ceph_lock);
1129 out:
1130 	/* avoid calling iput_final() while holding mdsc->mutex or
1131 	 * in mds dispatch threads */
1132 	ceph_async_iput(inode);
1133 	return mds;
1134 
1135 random:
1136 	if (random)
1137 		*random = true;
1138 
1139 	mds = ceph_mdsmap_get_random_mds(mdsc->mdsmap);
1140 	dout("%s chose random mds%d\n", __func__, mds);
1141 	return mds;
1142 }
1143 
1144 
1145 /*
1146  * session messages
1147  */
1148 static struct ceph_msg *create_session_msg(u32 op, u64 seq)
1149 {
1150 	struct ceph_msg *msg;
1151 	struct ceph_mds_session_head *h;
1152 
1153 	msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h), GFP_NOFS,
1154 			   false);
1155 	if (!msg) {
1156 		pr_err("create_session_msg ENOMEM creating msg\n");
1157 		return NULL;
1158 	}
1159 	h = msg->front.iov_base;
1160 	h->op = cpu_to_le32(op);
1161 	h->seq = cpu_to_le64(seq);
1162 
1163 	return msg;
1164 }
1165 
1166 static const unsigned char feature_bits[] = CEPHFS_FEATURES_CLIENT_SUPPORTED;
1167 #define FEATURE_BYTES(c) (DIV_ROUND_UP((size_t)feature_bits[c - 1] + 1, 64) * 8)
1168 static void encode_supported_features(void **p, void *end)
1169 {
1170 	static const size_t count = ARRAY_SIZE(feature_bits);
1171 
1172 	if (count > 0) {
1173 		size_t i;
1174 		size_t size = FEATURE_BYTES(count);
1175 
1176 		BUG_ON(*p + 4 + size > end);
1177 		ceph_encode_32(p, size);
1178 		memset(*p, 0, size);
1179 		for (i = 0; i < count; i++)
1180 			((unsigned char*)(*p))[i / 8] |= BIT(feature_bits[i] % 8);
1181 		*p += size;
1182 	} else {
1183 		BUG_ON(*p + 4 > end);
1184 		ceph_encode_32(p, 0);
1185 	}
1186 }
1187 
1188 /*
1189  * session message, specialization for CEPH_SESSION_REQUEST_OPEN
1190  * to include additional client metadata fields.
1191  */
1192 static struct ceph_msg *create_session_open_msg(struct ceph_mds_client *mdsc, u64 seq)
1193 {
1194 	struct ceph_msg *msg;
1195 	struct ceph_mds_session_head *h;
1196 	int i = -1;
1197 	int extra_bytes = 0;
1198 	int metadata_key_count = 0;
1199 	struct ceph_options *opt = mdsc->fsc->client->options;
1200 	struct ceph_mount_options *fsopt = mdsc->fsc->mount_options;
1201 	size_t size, count;
1202 	void *p, *end;
1203 
1204 	const char* metadata[][2] = {
1205 		{"hostname", mdsc->nodename},
1206 		{"kernel_version", init_utsname()->release},
1207 		{"entity_id", opt->name ? : ""},
1208 		{"root", fsopt->server_path ? : "/"},
1209 		{NULL, NULL}
1210 	};
1211 
1212 	/* Calculate serialized length of metadata */
1213 	extra_bytes = 4;  /* map length */
1214 	for (i = 0; metadata[i][0]; ++i) {
1215 		extra_bytes += 8 + strlen(metadata[i][0]) +
1216 			strlen(metadata[i][1]);
1217 		metadata_key_count++;
1218 	}
1219 
1220 	/* supported feature */
1221 	size = 0;
1222 	count = ARRAY_SIZE(feature_bits);
1223 	if (count > 0)
1224 		size = FEATURE_BYTES(count);
1225 	extra_bytes += 4 + size;
1226 
1227 	/* Allocate the message */
1228 	msg = ceph_msg_new(CEPH_MSG_CLIENT_SESSION, sizeof(*h) + extra_bytes,
1229 			   GFP_NOFS, false);
1230 	if (!msg) {
1231 		pr_err("create_session_msg ENOMEM creating msg\n");
1232 		return NULL;
1233 	}
1234 	p = msg->front.iov_base;
1235 	end = p + msg->front.iov_len;
1236 
1237 	h = p;
1238 	h->op = cpu_to_le32(CEPH_SESSION_REQUEST_OPEN);
1239 	h->seq = cpu_to_le64(seq);
1240 
1241 	/*
1242 	 * Serialize client metadata into waiting buffer space, using
1243 	 * the format that userspace expects for map<string, string>
1244 	 *
1245 	 * ClientSession messages with metadata are v3
1246 	 */
1247 	msg->hdr.version = cpu_to_le16(3);
1248 	msg->hdr.compat_version = cpu_to_le16(1);
1249 
1250 	/* The write pointer, following the session_head structure */
1251 	p += sizeof(*h);
1252 
1253 	/* Number of entries in the map */
1254 	ceph_encode_32(&p, metadata_key_count);
1255 
1256 	/* Two length-prefixed strings for each entry in the map */
1257 	for (i = 0; metadata[i][0]; ++i) {
1258 		size_t const key_len = strlen(metadata[i][0]);
1259 		size_t const val_len = strlen(metadata[i][1]);
1260 
1261 		ceph_encode_32(&p, key_len);
1262 		memcpy(p, metadata[i][0], key_len);
1263 		p += key_len;
1264 		ceph_encode_32(&p, val_len);
1265 		memcpy(p, metadata[i][1], val_len);
1266 		p += val_len;
1267 	}
1268 
1269 	encode_supported_features(&p, end);
1270 	msg->front.iov_len = p - msg->front.iov_base;
1271 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
1272 
1273 	return msg;
1274 }
1275 
1276 /*
1277  * send session open request.
1278  *
1279  * called under mdsc->mutex
1280  */
1281 static int __open_session(struct ceph_mds_client *mdsc,
1282 			  struct ceph_mds_session *session)
1283 {
1284 	struct ceph_msg *msg;
1285 	int mstate;
1286 	int mds = session->s_mds;
1287 
1288 	/* wait for mds to go active? */
1289 	mstate = ceph_mdsmap_get_state(mdsc->mdsmap, mds);
1290 	dout("open_session to mds%d (%s)\n", mds,
1291 	     ceph_mds_state_name(mstate));
1292 	session->s_state = CEPH_MDS_SESSION_OPENING;
1293 	session->s_renew_requested = jiffies;
1294 
1295 	/* send connect message */
1296 	msg = create_session_open_msg(mdsc, session->s_seq);
1297 	if (!msg)
1298 		return -ENOMEM;
1299 	ceph_con_send(&session->s_con, msg);
1300 	return 0;
1301 }
1302 
1303 /*
1304  * open sessions for any export targets for the given mds
1305  *
1306  * called under mdsc->mutex
1307  */
1308 static struct ceph_mds_session *
1309 __open_export_target_session(struct ceph_mds_client *mdsc, int target)
1310 {
1311 	struct ceph_mds_session *session;
1312 
1313 	session = __ceph_lookup_mds_session(mdsc, target);
1314 	if (!session) {
1315 		session = register_session(mdsc, target);
1316 		if (IS_ERR(session))
1317 			return session;
1318 	}
1319 	if (session->s_state == CEPH_MDS_SESSION_NEW ||
1320 	    session->s_state == CEPH_MDS_SESSION_CLOSING)
1321 		__open_session(mdsc, session);
1322 
1323 	return session;
1324 }
1325 
1326 struct ceph_mds_session *
1327 ceph_mdsc_open_export_target_session(struct ceph_mds_client *mdsc, int target)
1328 {
1329 	struct ceph_mds_session *session;
1330 
1331 	dout("open_export_target_session to mds%d\n", target);
1332 
1333 	mutex_lock(&mdsc->mutex);
1334 	session = __open_export_target_session(mdsc, target);
1335 	mutex_unlock(&mdsc->mutex);
1336 
1337 	return session;
1338 }
1339 
1340 static void __open_export_target_sessions(struct ceph_mds_client *mdsc,
1341 					  struct ceph_mds_session *session)
1342 {
1343 	struct ceph_mds_info *mi;
1344 	struct ceph_mds_session *ts;
1345 	int i, mds = session->s_mds;
1346 
1347 	if (mds >= mdsc->mdsmap->possible_max_rank)
1348 		return;
1349 
1350 	mi = &mdsc->mdsmap->m_info[mds];
1351 	dout("open_export_target_sessions for mds%d (%d targets)\n",
1352 	     session->s_mds, mi->num_export_targets);
1353 
1354 	for (i = 0; i < mi->num_export_targets; i++) {
1355 		ts = __open_export_target_session(mdsc, mi->export_targets[i]);
1356 		if (!IS_ERR(ts))
1357 			ceph_put_mds_session(ts);
1358 	}
1359 }
1360 
1361 void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc,
1362 					   struct ceph_mds_session *session)
1363 {
1364 	mutex_lock(&mdsc->mutex);
1365 	__open_export_target_sessions(mdsc, session);
1366 	mutex_unlock(&mdsc->mutex);
1367 }
1368 
1369 /*
1370  * session caps
1371  */
1372 
1373 static void detach_cap_releases(struct ceph_mds_session *session,
1374 				struct list_head *target)
1375 {
1376 	lockdep_assert_held(&session->s_cap_lock);
1377 
1378 	list_splice_init(&session->s_cap_releases, target);
1379 	session->s_num_cap_releases = 0;
1380 	dout("dispose_cap_releases mds%d\n", session->s_mds);
1381 }
1382 
1383 static void dispose_cap_releases(struct ceph_mds_client *mdsc,
1384 				 struct list_head *dispose)
1385 {
1386 	while (!list_empty(dispose)) {
1387 		struct ceph_cap *cap;
1388 		/* zero out the in-progress message */
1389 		cap = list_first_entry(dispose, struct ceph_cap, session_caps);
1390 		list_del(&cap->session_caps);
1391 		ceph_put_cap(mdsc, cap);
1392 	}
1393 }
1394 
1395 static void cleanup_session_requests(struct ceph_mds_client *mdsc,
1396 				     struct ceph_mds_session *session)
1397 {
1398 	struct ceph_mds_request *req;
1399 	struct rb_node *p;
1400 	struct ceph_inode_info *ci;
1401 
1402 	dout("cleanup_session_requests mds%d\n", session->s_mds);
1403 	mutex_lock(&mdsc->mutex);
1404 	while (!list_empty(&session->s_unsafe)) {
1405 		req = list_first_entry(&session->s_unsafe,
1406 				       struct ceph_mds_request, r_unsafe_item);
1407 		pr_warn_ratelimited(" dropping unsafe request %llu\n",
1408 				    req->r_tid);
1409 		if (req->r_target_inode) {
1410 			/* dropping unsafe change of inode's attributes */
1411 			ci = ceph_inode(req->r_target_inode);
1412 			errseq_set(&ci->i_meta_err, -EIO);
1413 		}
1414 		if (req->r_unsafe_dir) {
1415 			/* dropping unsafe directory operation */
1416 			ci = ceph_inode(req->r_unsafe_dir);
1417 			errseq_set(&ci->i_meta_err, -EIO);
1418 		}
1419 		__unregister_request(mdsc, req);
1420 	}
1421 	/* zero r_attempts, so kick_requests() will re-send requests */
1422 	p = rb_first(&mdsc->request_tree);
1423 	while (p) {
1424 		req = rb_entry(p, struct ceph_mds_request, r_node);
1425 		p = rb_next(p);
1426 		if (req->r_session &&
1427 		    req->r_session->s_mds == session->s_mds)
1428 			req->r_attempts = 0;
1429 	}
1430 	mutex_unlock(&mdsc->mutex);
1431 }
1432 
1433 /*
1434  * Helper to safely iterate over all caps associated with a session, with
1435  * special care taken to handle a racing __ceph_remove_cap().
1436  *
1437  * Caller must hold session s_mutex.
1438  */
1439 int ceph_iterate_session_caps(struct ceph_mds_session *session,
1440 			      int (*cb)(struct inode *, struct ceph_cap *,
1441 					void *), void *arg)
1442 {
1443 	struct list_head *p;
1444 	struct ceph_cap *cap;
1445 	struct inode *inode, *last_inode = NULL;
1446 	struct ceph_cap *old_cap = NULL;
1447 	int ret;
1448 
1449 	dout("iterate_session_caps %p mds%d\n", session, session->s_mds);
1450 	spin_lock(&session->s_cap_lock);
1451 	p = session->s_caps.next;
1452 	while (p != &session->s_caps) {
1453 		cap = list_entry(p, struct ceph_cap, session_caps);
1454 		inode = igrab(&cap->ci->vfs_inode);
1455 		if (!inode) {
1456 			p = p->next;
1457 			continue;
1458 		}
1459 		session->s_cap_iterator = cap;
1460 		spin_unlock(&session->s_cap_lock);
1461 
1462 		if (last_inode) {
1463 			/* avoid calling iput_final() while holding
1464 			 * s_mutex or in mds dispatch threads */
1465 			ceph_async_iput(last_inode);
1466 			last_inode = NULL;
1467 		}
1468 		if (old_cap) {
1469 			ceph_put_cap(session->s_mdsc, old_cap);
1470 			old_cap = NULL;
1471 		}
1472 
1473 		ret = cb(inode, cap, arg);
1474 		last_inode = inode;
1475 
1476 		spin_lock(&session->s_cap_lock);
1477 		p = p->next;
1478 		if (!cap->ci) {
1479 			dout("iterate_session_caps  finishing cap %p removal\n",
1480 			     cap);
1481 			BUG_ON(cap->session != session);
1482 			cap->session = NULL;
1483 			list_del_init(&cap->session_caps);
1484 			session->s_nr_caps--;
1485 			if (cap->queue_release)
1486 				__ceph_queue_cap_release(session, cap);
1487 			else
1488 				old_cap = cap;  /* put_cap it w/o locks held */
1489 		}
1490 		if (ret < 0)
1491 			goto out;
1492 	}
1493 	ret = 0;
1494 out:
1495 	session->s_cap_iterator = NULL;
1496 	spin_unlock(&session->s_cap_lock);
1497 
1498 	ceph_async_iput(last_inode);
1499 	if (old_cap)
1500 		ceph_put_cap(session->s_mdsc, old_cap);
1501 
1502 	return ret;
1503 }
1504 
1505 static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
1506 				  void *arg)
1507 {
1508 	struct ceph_fs_client *fsc = (struct ceph_fs_client *)arg;
1509 	struct ceph_inode_info *ci = ceph_inode(inode);
1510 	LIST_HEAD(to_remove);
1511 	bool dirty_dropped = false;
1512 	bool invalidate = false;
1513 
1514 	dout("removing cap %p, ci is %p, inode is %p\n",
1515 	     cap, ci, &ci->vfs_inode);
1516 	spin_lock(&ci->i_ceph_lock);
1517 	__ceph_remove_cap(cap, false);
1518 	if (!ci->i_auth_cap) {
1519 		struct ceph_cap_flush *cf;
1520 		struct ceph_mds_client *mdsc = fsc->mdsc;
1521 
1522 		if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1523 			if (inode->i_data.nrpages > 0)
1524 				invalidate = true;
1525 			if (ci->i_wrbuffer_ref > 0)
1526 				mapping_set_error(&inode->i_data, -EIO);
1527 		}
1528 
1529 		while (!list_empty(&ci->i_cap_flush_list)) {
1530 			cf = list_first_entry(&ci->i_cap_flush_list,
1531 					      struct ceph_cap_flush, i_list);
1532 			list_move(&cf->i_list, &to_remove);
1533 		}
1534 
1535 		spin_lock(&mdsc->cap_dirty_lock);
1536 
1537 		list_for_each_entry(cf, &to_remove, i_list)
1538 			list_del(&cf->g_list);
1539 
1540 		if (!list_empty(&ci->i_dirty_item)) {
1541 			pr_warn_ratelimited(
1542 				" dropping dirty %s state for %p %lld\n",
1543 				ceph_cap_string(ci->i_dirty_caps),
1544 				inode, ceph_ino(inode));
1545 			ci->i_dirty_caps = 0;
1546 			list_del_init(&ci->i_dirty_item);
1547 			dirty_dropped = true;
1548 		}
1549 		if (!list_empty(&ci->i_flushing_item)) {
1550 			pr_warn_ratelimited(
1551 				" dropping dirty+flushing %s state for %p %lld\n",
1552 				ceph_cap_string(ci->i_flushing_caps),
1553 				inode, ceph_ino(inode));
1554 			ci->i_flushing_caps = 0;
1555 			list_del_init(&ci->i_flushing_item);
1556 			mdsc->num_cap_flushing--;
1557 			dirty_dropped = true;
1558 		}
1559 		spin_unlock(&mdsc->cap_dirty_lock);
1560 
1561 		if (dirty_dropped) {
1562 			errseq_set(&ci->i_meta_err, -EIO);
1563 
1564 			if (ci->i_wrbuffer_ref_head == 0 &&
1565 			    ci->i_wr_ref == 0 &&
1566 			    ci->i_dirty_caps == 0 &&
1567 			    ci->i_flushing_caps == 0) {
1568 				ceph_put_snap_context(ci->i_head_snapc);
1569 				ci->i_head_snapc = NULL;
1570 			}
1571 		}
1572 
1573 		if (atomic_read(&ci->i_filelock_ref) > 0) {
1574 			/* make further file lock syscall return -EIO */
1575 			ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
1576 			pr_warn_ratelimited(" dropping file locks for %p %lld\n",
1577 					    inode, ceph_ino(inode));
1578 		}
1579 
1580 		if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
1581 			list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
1582 			ci->i_prealloc_cap_flush = NULL;
1583 		}
1584 	}
1585 	spin_unlock(&ci->i_ceph_lock);
1586 	while (!list_empty(&to_remove)) {
1587 		struct ceph_cap_flush *cf;
1588 		cf = list_first_entry(&to_remove,
1589 				      struct ceph_cap_flush, i_list);
1590 		list_del(&cf->i_list);
1591 		ceph_free_cap_flush(cf);
1592 	}
1593 
1594 	wake_up_all(&ci->i_cap_wq);
1595 	if (invalidate)
1596 		ceph_queue_invalidate(inode);
1597 	if (dirty_dropped)
1598 		iput(inode);
1599 	return 0;
1600 }
1601 
1602 /*
1603  * caller must hold session s_mutex
1604  */
1605 static void remove_session_caps(struct ceph_mds_session *session)
1606 {
1607 	struct ceph_fs_client *fsc = session->s_mdsc->fsc;
1608 	struct super_block *sb = fsc->sb;
1609 	LIST_HEAD(dispose);
1610 
1611 	dout("remove_session_caps on %p\n", session);
1612 	ceph_iterate_session_caps(session, remove_session_caps_cb, fsc);
1613 
1614 	wake_up_all(&fsc->mdsc->cap_flushing_wq);
1615 
1616 	spin_lock(&session->s_cap_lock);
1617 	if (session->s_nr_caps > 0) {
1618 		struct inode *inode;
1619 		struct ceph_cap *cap, *prev = NULL;
1620 		struct ceph_vino vino;
1621 		/*
1622 		 * iterate_session_caps() skips inodes that are being
1623 		 * deleted, we need to wait until deletions are complete.
1624 		 * __wait_on_freeing_inode() is designed for the job,
1625 		 * but it is not exported, so use lookup inode function
1626 		 * to access it.
1627 		 */
1628 		while (!list_empty(&session->s_caps)) {
1629 			cap = list_entry(session->s_caps.next,
1630 					 struct ceph_cap, session_caps);
1631 			if (cap == prev)
1632 				break;
1633 			prev = cap;
1634 			vino = cap->ci->i_vino;
1635 			spin_unlock(&session->s_cap_lock);
1636 
1637 			inode = ceph_find_inode(sb, vino);
1638 			 /* avoid calling iput_final() while holding s_mutex */
1639 			ceph_async_iput(inode);
1640 
1641 			spin_lock(&session->s_cap_lock);
1642 		}
1643 	}
1644 
1645 	// drop cap expires and unlock s_cap_lock
1646 	detach_cap_releases(session, &dispose);
1647 
1648 	BUG_ON(session->s_nr_caps > 0);
1649 	BUG_ON(!list_empty(&session->s_cap_flushing));
1650 	spin_unlock(&session->s_cap_lock);
1651 	dispose_cap_releases(session->s_mdsc, &dispose);
1652 }
1653 
1654 enum {
1655 	RECONNECT,
1656 	RENEWCAPS,
1657 	FORCE_RO,
1658 };
1659 
1660 /*
1661  * wake up any threads waiting on this session's caps.  if the cap is
1662  * old (didn't get renewed on the client reconnect), remove it now.
1663  *
1664  * caller must hold s_mutex.
1665  */
1666 static int wake_up_session_cb(struct inode *inode, struct ceph_cap *cap,
1667 			      void *arg)
1668 {
1669 	struct ceph_inode_info *ci = ceph_inode(inode);
1670 	unsigned long ev = (unsigned long)arg;
1671 
1672 	if (ev == RECONNECT) {
1673 		spin_lock(&ci->i_ceph_lock);
1674 		ci->i_wanted_max_size = 0;
1675 		ci->i_requested_max_size = 0;
1676 		spin_unlock(&ci->i_ceph_lock);
1677 	} else if (ev == RENEWCAPS) {
1678 		if (cap->cap_gen < cap->session->s_cap_gen) {
1679 			/* mds did not re-issue stale cap */
1680 			spin_lock(&ci->i_ceph_lock);
1681 			cap->issued = cap->implemented = CEPH_CAP_PIN;
1682 			spin_unlock(&ci->i_ceph_lock);
1683 		}
1684 	} else if (ev == FORCE_RO) {
1685 	}
1686 	wake_up_all(&ci->i_cap_wq);
1687 	return 0;
1688 }
1689 
1690 static void wake_up_session_caps(struct ceph_mds_session *session, int ev)
1691 {
1692 	dout("wake_up_session_caps %p mds%d\n", session, session->s_mds);
1693 	ceph_iterate_session_caps(session, wake_up_session_cb,
1694 				  (void *)(unsigned long)ev);
1695 }
1696 
1697 /*
1698  * Send periodic message to MDS renewing all currently held caps.  The
1699  * ack will reset the expiration for all caps from this session.
1700  *
1701  * caller holds s_mutex
1702  */
1703 static int send_renew_caps(struct ceph_mds_client *mdsc,
1704 			   struct ceph_mds_session *session)
1705 {
1706 	struct ceph_msg *msg;
1707 	int state;
1708 
1709 	if (time_after_eq(jiffies, session->s_cap_ttl) &&
1710 	    time_after_eq(session->s_cap_ttl, session->s_renew_requested))
1711 		pr_info("mds%d caps stale\n", session->s_mds);
1712 	session->s_renew_requested = jiffies;
1713 
1714 	/* do not try to renew caps until a recovering mds has reconnected
1715 	 * with its clients. */
1716 	state = ceph_mdsmap_get_state(mdsc->mdsmap, session->s_mds);
1717 	if (state < CEPH_MDS_STATE_RECONNECT) {
1718 		dout("send_renew_caps ignoring mds%d (%s)\n",
1719 		     session->s_mds, ceph_mds_state_name(state));
1720 		return 0;
1721 	}
1722 
1723 	dout("send_renew_caps to mds%d (%s)\n", session->s_mds,
1724 		ceph_mds_state_name(state));
1725 	msg = create_session_msg(CEPH_SESSION_REQUEST_RENEWCAPS,
1726 				 ++session->s_renew_seq);
1727 	if (!msg)
1728 		return -ENOMEM;
1729 	ceph_con_send(&session->s_con, msg);
1730 	return 0;
1731 }
1732 
1733 static int send_flushmsg_ack(struct ceph_mds_client *mdsc,
1734 			     struct ceph_mds_session *session, u64 seq)
1735 {
1736 	struct ceph_msg *msg;
1737 
1738 	dout("send_flushmsg_ack to mds%d (%s)s seq %lld\n",
1739 	     session->s_mds, ceph_session_state_name(session->s_state), seq);
1740 	msg = create_session_msg(CEPH_SESSION_FLUSHMSG_ACK, seq);
1741 	if (!msg)
1742 		return -ENOMEM;
1743 	ceph_con_send(&session->s_con, msg);
1744 	return 0;
1745 }
1746 
1747 
1748 /*
1749  * Note new cap ttl, and any transition from stale -> not stale (fresh?).
1750  *
1751  * Called under session->s_mutex
1752  */
1753 static void renewed_caps(struct ceph_mds_client *mdsc,
1754 			 struct ceph_mds_session *session, int is_renew)
1755 {
1756 	int was_stale;
1757 	int wake = 0;
1758 
1759 	spin_lock(&session->s_cap_lock);
1760 	was_stale = is_renew && time_after_eq(jiffies, session->s_cap_ttl);
1761 
1762 	session->s_cap_ttl = session->s_renew_requested +
1763 		mdsc->mdsmap->m_session_timeout*HZ;
1764 
1765 	if (was_stale) {
1766 		if (time_before(jiffies, session->s_cap_ttl)) {
1767 			pr_info("mds%d caps renewed\n", session->s_mds);
1768 			wake = 1;
1769 		} else {
1770 			pr_info("mds%d caps still stale\n", session->s_mds);
1771 		}
1772 	}
1773 	dout("renewed_caps mds%d ttl now %lu, was %s, now %s\n",
1774 	     session->s_mds, session->s_cap_ttl, was_stale ? "stale" : "fresh",
1775 	     time_before(jiffies, session->s_cap_ttl) ? "stale" : "fresh");
1776 	spin_unlock(&session->s_cap_lock);
1777 
1778 	if (wake)
1779 		wake_up_session_caps(session, RENEWCAPS);
1780 }
1781 
1782 /*
1783  * send a session close request
1784  */
1785 static int request_close_session(struct ceph_mds_client *mdsc,
1786 				 struct ceph_mds_session *session)
1787 {
1788 	struct ceph_msg *msg;
1789 
1790 	dout("request_close_session mds%d state %s seq %lld\n",
1791 	     session->s_mds, ceph_session_state_name(session->s_state),
1792 	     session->s_seq);
1793 	msg = create_session_msg(CEPH_SESSION_REQUEST_CLOSE, session->s_seq);
1794 	if (!msg)
1795 		return -ENOMEM;
1796 	ceph_con_send(&session->s_con, msg);
1797 	return 1;
1798 }
1799 
1800 /*
1801  * Called with s_mutex held.
1802  */
1803 static int __close_session(struct ceph_mds_client *mdsc,
1804 			 struct ceph_mds_session *session)
1805 {
1806 	if (session->s_state >= CEPH_MDS_SESSION_CLOSING)
1807 		return 0;
1808 	session->s_state = CEPH_MDS_SESSION_CLOSING;
1809 	return request_close_session(mdsc, session);
1810 }
1811 
1812 static bool drop_negative_children(struct dentry *dentry)
1813 {
1814 	struct dentry *child;
1815 	bool all_negative = true;
1816 
1817 	if (!d_is_dir(dentry))
1818 		goto out;
1819 
1820 	spin_lock(&dentry->d_lock);
1821 	list_for_each_entry(child, &dentry->d_subdirs, d_child) {
1822 		if (d_really_is_positive(child)) {
1823 			all_negative = false;
1824 			break;
1825 		}
1826 	}
1827 	spin_unlock(&dentry->d_lock);
1828 
1829 	if (all_negative)
1830 		shrink_dcache_parent(dentry);
1831 out:
1832 	return all_negative;
1833 }
1834 
1835 /*
1836  * Trim old(er) caps.
1837  *
1838  * Because we can't cache an inode without one or more caps, we do
1839  * this indirectly: if a cap is unused, we prune its aliases, at which
1840  * point the inode will hopefully get dropped to.
1841  *
1842  * Yes, this is a bit sloppy.  Our only real goal here is to respond to
1843  * memory pressure from the MDS, though, so it needn't be perfect.
1844  */
1845 static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg)
1846 {
1847 	int *remaining = arg;
1848 	struct ceph_inode_info *ci = ceph_inode(inode);
1849 	int used, wanted, oissued, mine;
1850 
1851 	if (*remaining <= 0)
1852 		return -1;
1853 
1854 	spin_lock(&ci->i_ceph_lock);
1855 	mine = cap->issued | cap->implemented;
1856 	used = __ceph_caps_used(ci);
1857 	wanted = __ceph_caps_file_wanted(ci);
1858 	oissued = __ceph_caps_issued_other(ci, cap);
1859 
1860 	dout("trim_caps_cb %p cap %p mine %s oissued %s used %s wanted %s\n",
1861 	     inode, cap, ceph_cap_string(mine), ceph_cap_string(oissued),
1862 	     ceph_cap_string(used), ceph_cap_string(wanted));
1863 	if (cap == ci->i_auth_cap) {
1864 		if (ci->i_dirty_caps || ci->i_flushing_caps ||
1865 		    !list_empty(&ci->i_cap_snaps))
1866 			goto out;
1867 		if ((used | wanted) & CEPH_CAP_ANY_WR)
1868 			goto out;
1869 		/* Note: it's possible that i_filelock_ref becomes non-zero
1870 		 * after dropping auth caps. It doesn't hurt because reply
1871 		 * of lock mds request will re-add auth caps. */
1872 		if (atomic_read(&ci->i_filelock_ref) > 0)
1873 			goto out;
1874 	}
1875 	/* The inode has cached pages, but it's no longer used.
1876 	 * we can safely drop it */
1877 	if (S_ISREG(inode->i_mode) &&
1878 	    wanted == 0 && used == CEPH_CAP_FILE_CACHE &&
1879 	    !(oissued & CEPH_CAP_FILE_CACHE)) {
1880 	  used = 0;
1881 	  oissued = 0;
1882 	}
1883 	if ((used | wanted) & ~oissued & mine)
1884 		goto out;   /* we need these caps */
1885 
1886 	if (oissued) {
1887 		/* we aren't the only cap.. just remove us */
1888 		__ceph_remove_cap(cap, true);
1889 		(*remaining)--;
1890 	} else {
1891 		struct dentry *dentry;
1892 		/* try dropping referring dentries */
1893 		spin_unlock(&ci->i_ceph_lock);
1894 		dentry = d_find_any_alias(inode);
1895 		if (dentry && drop_negative_children(dentry)) {
1896 			int count;
1897 			dput(dentry);
1898 			d_prune_aliases(inode);
1899 			count = atomic_read(&inode->i_count);
1900 			if (count == 1)
1901 				(*remaining)--;
1902 			dout("trim_caps_cb %p cap %p pruned, count now %d\n",
1903 			     inode, cap, count);
1904 		} else {
1905 			dput(dentry);
1906 		}
1907 		return 0;
1908 	}
1909 
1910 out:
1911 	spin_unlock(&ci->i_ceph_lock);
1912 	return 0;
1913 }
1914 
1915 /*
1916  * Trim session cap count down to some max number.
1917  */
1918 int ceph_trim_caps(struct ceph_mds_client *mdsc,
1919 		   struct ceph_mds_session *session,
1920 		   int max_caps)
1921 {
1922 	int trim_caps = session->s_nr_caps - max_caps;
1923 
1924 	dout("trim_caps mds%d start: %d / %d, trim %d\n",
1925 	     session->s_mds, session->s_nr_caps, max_caps, trim_caps);
1926 	if (trim_caps > 0) {
1927 		int remaining = trim_caps;
1928 
1929 		ceph_iterate_session_caps(session, trim_caps_cb, &remaining);
1930 		dout("trim_caps mds%d done: %d / %d, trimmed %d\n",
1931 		     session->s_mds, session->s_nr_caps, max_caps,
1932 			trim_caps - remaining);
1933 	}
1934 
1935 	ceph_flush_cap_releases(mdsc, session);
1936 	return 0;
1937 }
1938 
1939 static int check_caps_flush(struct ceph_mds_client *mdsc,
1940 			    u64 want_flush_tid)
1941 {
1942 	int ret = 1;
1943 
1944 	spin_lock(&mdsc->cap_dirty_lock);
1945 	if (!list_empty(&mdsc->cap_flush_list)) {
1946 		struct ceph_cap_flush *cf =
1947 			list_first_entry(&mdsc->cap_flush_list,
1948 					 struct ceph_cap_flush, g_list);
1949 		if (cf->tid <= want_flush_tid) {
1950 			dout("check_caps_flush still flushing tid "
1951 			     "%llu <= %llu\n", cf->tid, want_flush_tid);
1952 			ret = 0;
1953 		}
1954 	}
1955 	spin_unlock(&mdsc->cap_dirty_lock);
1956 	return ret;
1957 }
1958 
1959 /*
1960  * flush all dirty inode data to disk.
1961  *
1962  * returns true if we've flushed through want_flush_tid
1963  */
1964 static void wait_caps_flush(struct ceph_mds_client *mdsc,
1965 			    u64 want_flush_tid)
1966 {
1967 	dout("check_caps_flush want %llu\n", want_flush_tid);
1968 
1969 	wait_event(mdsc->cap_flushing_wq,
1970 		   check_caps_flush(mdsc, want_flush_tid));
1971 
1972 	dout("check_caps_flush ok, flushed thru %llu\n", want_flush_tid);
1973 }
1974 
1975 /*
1976  * called under s_mutex
1977  */
1978 static void ceph_send_cap_releases(struct ceph_mds_client *mdsc,
1979 				   struct ceph_mds_session *session)
1980 {
1981 	struct ceph_msg *msg = NULL;
1982 	struct ceph_mds_cap_release *head;
1983 	struct ceph_mds_cap_item *item;
1984 	struct ceph_osd_client *osdc = &mdsc->fsc->client->osdc;
1985 	struct ceph_cap *cap;
1986 	LIST_HEAD(tmp_list);
1987 	int num_cap_releases;
1988 	__le32	barrier, *cap_barrier;
1989 
1990 	down_read(&osdc->lock);
1991 	barrier = cpu_to_le32(osdc->epoch_barrier);
1992 	up_read(&osdc->lock);
1993 
1994 	spin_lock(&session->s_cap_lock);
1995 again:
1996 	list_splice_init(&session->s_cap_releases, &tmp_list);
1997 	num_cap_releases = session->s_num_cap_releases;
1998 	session->s_num_cap_releases = 0;
1999 	spin_unlock(&session->s_cap_lock);
2000 
2001 	while (!list_empty(&tmp_list)) {
2002 		if (!msg) {
2003 			msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPRELEASE,
2004 					PAGE_SIZE, GFP_NOFS, false);
2005 			if (!msg)
2006 				goto out_err;
2007 			head = msg->front.iov_base;
2008 			head->num = cpu_to_le32(0);
2009 			msg->front.iov_len = sizeof(*head);
2010 
2011 			msg->hdr.version = cpu_to_le16(2);
2012 			msg->hdr.compat_version = cpu_to_le16(1);
2013 		}
2014 
2015 		cap = list_first_entry(&tmp_list, struct ceph_cap,
2016 					session_caps);
2017 		list_del(&cap->session_caps);
2018 		num_cap_releases--;
2019 
2020 		head = msg->front.iov_base;
2021 		put_unaligned_le32(get_unaligned_le32(&head->num) + 1,
2022 				   &head->num);
2023 		item = msg->front.iov_base + msg->front.iov_len;
2024 		item->ino = cpu_to_le64(cap->cap_ino);
2025 		item->cap_id = cpu_to_le64(cap->cap_id);
2026 		item->migrate_seq = cpu_to_le32(cap->mseq);
2027 		item->seq = cpu_to_le32(cap->issue_seq);
2028 		msg->front.iov_len += sizeof(*item);
2029 
2030 		ceph_put_cap(mdsc, cap);
2031 
2032 		if (le32_to_cpu(head->num) == CEPH_CAPS_PER_RELEASE) {
2033 			// Append cap_barrier field
2034 			cap_barrier = msg->front.iov_base + msg->front.iov_len;
2035 			*cap_barrier = barrier;
2036 			msg->front.iov_len += sizeof(*cap_barrier);
2037 
2038 			msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2039 			dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2040 			ceph_con_send(&session->s_con, msg);
2041 			msg = NULL;
2042 		}
2043 	}
2044 
2045 	BUG_ON(num_cap_releases != 0);
2046 
2047 	spin_lock(&session->s_cap_lock);
2048 	if (!list_empty(&session->s_cap_releases))
2049 		goto again;
2050 	spin_unlock(&session->s_cap_lock);
2051 
2052 	if (msg) {
2053 		// Append cap_barrier field
2054 		cap_barrier = msg->front.iov_base + msg->front.iov_len;
2055 		*cap_barrier = barrier;
2056 		msg->front.iov_len += sizeof(*cap_barrier);
2057 
2058 		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2059 		dout("send_cap_releases mds%d %p\n", session->s_mds, msg);
2060 		ceph_con_send(&session->s_con, msg);
2061 	}
2062 	return;
2063 out_err:
2064 	pr_err("send_cap_releases mds%d, failed to allocate message\n",
2065 		session->s_mds);
2066 	spin_lock(&session->s_cap_lock);
2067 	list_splice(&tmp_list, &session->s_cap_releases);
2068 	session->s_num_cap_releases += num_cap_releases;
2069 	spin_unlock(&session->s_cap_lock);
2070 }
2071 
2072 static void ceph_cap_release_work(struct work_struct *work)
2073 {
2074 	struct ceph_mds_session *session =
2075 		container_of(work, struct ceph_mds_session, s_cap_release_work);
2076 
2077 	mutex_lock(&session->s_mutex);
2078 	if (session->s_state == CEPH_MDS_SESSION_OPEN ||
2079 	    session->s_state == CEPH_MDS_SESSION_HUNG)
2080 		ceph_send_cap_releases(session->s_mdsc, session);
2081 	mutex_unlock(&session->s_mutex);
2082 	ceph_put_mds_session(session);
2083 }
2084 
2085 void ceph_flush_cap_releases(struct ceph_mds_client *mdsc,
2086 		             struct ceph_mds_session *session)
2087 {
2088 	if (mdsc->stopping)
2089 		return;
2090 
2091 	ceph_get_mds_session(session);
2092 	if (queue_work(mdsc->fsc->cap_wq,
2093 		       &session->s_cap_release_work)) {
2094 		dout("cap release work queued\n");
2095 	} else {
2096 		ceph_put_mds_session(session);
2097 		dout("failed to queue cap release work\n");
2098 	}
2099 }
2100 
2101 /*
2102  * caller holds session->s_cap_lock
2103  */
2104 void __ceph_queue_cap_release(struct ceph_mds_session *session,
2105 			      struct ceph_cap *cap)
2106 {
2107 	list_add_tail(&cap->session_caps, &session->s_cap_releases);
2108 	session->s_num_cap_releases++;
2109 
2110 	if (!(session->s_num_cap_releases % CEPH_CAPS_PER_RELEASE))
2111 		ceph_flush_cap_releases(session->s_mdsc, session);
2112 }
2113 
2114 static void ceph_cap_reclaim_work(struct work_struct *work)
2115 {
2116 	struct ceph_mds_client *mdsc =
2117 		container_of(work, struct ceph_mds_client, cap_reclaim_work);
2118 	int ret = ceph_trim_dentries(mdsc);
2119 	if (ret == -EAGAIN)
2120 		ceph_queue_cap_reclaim_work(mdsc);
2121 }
2122 
2123 void ceph_queue_cap_reclaim_work(struct ceph_mds_client *mdsc)
2124 {
2125 	if (mdsc->stopping)
2126 		return;
2127 
2128         if (queue_work(mdsc->fsc->cap_wq, &mdsc->cap_reclaim_work)) {
2129                 dout("caps reclaim work queued\n");
2130         } else {
2131                 dout("failed to queue caps release work\n");
2132         }
2133 }
2134 
2135 void ceph_reclaim_caps_nr(struct ceph_mds_client *mdsc, int nr)
2136 {
2137 	int val;
2138 	if (!nr)
2139 		return;
2140 	val = atomic_add_return(nr, &mdsc->cap_reclaim_pending);
2141 	if ((val % CEPH_CAPS_PER_RELEASE) < nr) {
2142 		atomic_set(&mdsc->cap_reclaim_pending, 0);
2143 		ceph_queue_cap_reclaim_work(mdsc);
2144 	}
2145 }
2146 
2147 /*
2148  * requests
2149  */
2150 
2151 int ceph_alloc_readdir_reply_buffer(struct ceph_mds_request *req,
2152 				    struct inode *dir)
2153 {
2154 	struct ceph_inode_info *ci = ceph_inode(dir);
2155 	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
2156 	struct ceph_mount_options *opt = req->r_mdsc->fsc->mount_options;
2157 	size_t size = sizeof(struct ceph_mds_reply_dir_entry);
2158 	unsigned int num_entries;
2159 	int order;
2160 
2161 	spin_lock(&ci->i_ceph_lock);
2162 	num_entries = ci->i_files + ci->i_subdirs;
2163 	spin_unlock(&ci->i_ceph_lock);
2164 	num_entries = max(num_entries, 1U);
2165 	num_entries = min(num_entries, opt->max_readdir);
2166 
2167 	order = get_order(size * num_entries);
2168 	while (order >= 0) {
2169 		rinfo->dir_entries = (void*)__get_free_pages(GFP_KERNEL |
2170 							     __GFP_NOWARN,
2171 							     order);
2172 		if (rinfo->dir_entries)
2173 			break;
2174 		order--;
2175 	}
2176 	if (!rinfo->dir_entries)
2177 		return -ENOMEM;
2178 
2179 	num_entries = (PAGE_SIZE << order) / size;
2180 	num_entries = min(num_entries, opt->max_readdir);
2181 
2182 	rinfo->dir_buf_size = PAGE_SIZE << order;
2183 	req->r_num_caps = num_entries + 1;
2184 	req->r_args.readdir.max_entries = cpu_to_le32(num_entries);
2185 	req->r_args.readdir.max_bytes = cpu_to_le32(opt->max_readdir_bytes);
2186 	return 0;
2187 }
2188 
2189 /*
2190  * Create an mds request.
2191  */
2192 struct ceph_mds_request *
2193 ceph_mdsc_create_request(struct ceph_mds_client *mdsc, int op, int mode)
2194 {
2195 	struct ceph_mds_request *req;
2196 
2197 	req = kmem_cache_zalloc(ceph_mds_request_cachep, GFP_NOFS);
2198 	if (!req)
2199 		return ERR_PTR(-ENOMEM);
2200 
2201 	mutex_init(&req->r_fill_mutex);
2202 	req->r_mdsc = mdsc;
2203 	req->r_started = jiffies;
2204 	req->r_resend_mds = -1;
2205 	INIT_LIST_HEAD(&req->r_unsafe_dir_item);
2206 	INIT_LIST_HEAD(&req->r_unsafe_target_item);
2207 	req->r_fmode = -1;
2208 	kref_init(&req->r_kref);
2209 	RB_CLEAR_NODE(&req->r_node);
2210 	INIT_LIST_HEAD(&req->r_wait);
2211 	init_completion(&req->r_completion);
2212 	init_completion(&req->r_safe_completion);
2213 	INIT_LIST_HEAD(&req->r_unsafe_item);
2214 
2215 	ktime_get_coarse_real_ts64(&req->r_stamp);
2216 
2217 	req->r_op = op;
2218 	req->r_direct_mode = mode;
2219 	return req;
2220 }
2221 
2222 /*
2223  * return oldest (lowest) request, tid in request tree, 0 if none.
2224  *
2225  * called under mdsc->mutex.
2226  */
2227 static struct ceph_mds_request *__get_oldest_req(struct ceph_mds_client *mdsc)
2228 {
2229 	if (RB_EMPTY_ROOT(&mdsc->request_tree))
2230 		return NULL;
2231 	return rb_entry(rb_first(&mdsc->request_tree),
2232 			struct ceph_mds_request, r_node);
2233 }
2234 
2235 static inline  u64 __get_oldest_tid(struct ceph_mds_client *mdsc)
2236 {
2237 	return mdsc->oldest_tid;
2238 }
2239 
2240 /*
2241  * Build a dentry's path.  Allocate on heap; caller must kfree.  Based
2242  * on build_path_from_dentry in fs/cifs/dir.c.
2243  *
2244  * If @stop_on_nosnap, generate path relative to the first non-snapped
2245  * inode.
2246  *
2247  * Encode hidden .snap dirs as a double /, i.e.
2248  *   foo/.snap/bar -> foo//bar
2249  */
2250 char *ceph_mdsc_build_path(struct dentry *dentry, int *plen, u64 *pbase,
2251 			   int stop_on_nosnap)
2252 {
2253 	struct dentry *temp;
2254 	char *path;
2255 	int pos;
2256 	unsigned seq;
2257 	u64 base;
2258 
2259 	if (!dentry)
2260 		return ERR_PTR(-EINVAL);
2261 
2262 	path = __getname();
2263 	if (!path)
2264 		return ERR_PTR(-ENOMEM);
2265 retry:
2266 	pos = PATH_MAX - 1;
2267 	path[pos] = '\0';
2268 
2269 	seq = read_seqbegin(&rename_lock);
2270 	rcu_read_lock();
2271 	temp = dentry;
2272 	for (;;) {
2273 		struct inode *inode;
2274 
2275 		spin_lock(&temp->d_lock);
2276 		inode = d_inode(temp);
2277 		if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
2278 			dout("build_path path+%d: %p SNAPDIR\n",
2279 			     pos, temp);
2280 		} else if (stop_on_nosnap && inode && dentry != temp &&
2281 			   ceph_snap(inode) == CEPH_NOSNAP) {
2282 			spin_unlock(&temp->d_lock);
2283 			pos++; /* get rid of any prepended '/' */
2284 			break;
2285 		} else {
2286 			pos -= temp->d_name.len;
2287 			if (pos < 0) {
2288 				spin_unlock(&temp->d_lock);
2289 				break;
2290 			}
2291 			memcpy(path + pos, temp->d_name.name, temp->d_name.len);
2292 		}
2293 		spin_unlock(&temp->d_lock);
2294 		temp = READ_ONCE(temp->d_parent);
2295 
2296 		/* Are we at the root? */
2297 		if (IS_ROOT(temp))
2298 			break;
2299 
2300 		/* Are we out of buffer? */
2301 		if (--pos < 0)
2302 			break;
2303 
2304 		path[pos] = '/';
2305 	}
2306 	base = ceph_ino(d_inode(temp));
2307 	rcu_read_unlock();
2308 
2309 	if (read_seqretry(&rename_lock, seq))
2310 		goto retry;
2311 
2312 	if (pos < 0) {
2313 		/*
2314 		 * A rename didn't occur, but somehow we didn't end up where
2315 		 * we thought we would. Throw a warning and try again.
2316 		 */
2317 		pr_warn("build_path did not end path lookup where "
2318 			"expected, pos is %d\n", pos);
2319 		goto retry;
2320 	}
2321 
2322 	*pbase = base;
2323 	*plen = PATH_MAX - 1 - pos;
2324 	dout("build_path on %p %d built %llx '%.*s'\n",
2325 	     dentry, d_count(dentry), base, *plen, path + pos);
2326 	return path + pos;
2327 }
2328 
2329 static int build_dentry_path(struct dentry *dentry, struct inode *dir,
2330 			     const char **ppath, int *ppathlen, u64 *pino,
2331 			     bool *pfreepath, bool parent_locked)
2332 {
2333 	char *path;
2334 
2335 	rcu_read_lock();
2336 	if (!dir)
2337 		dir = d_inode_rcu(dentry->d_parent);
2338 	if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP) {
2339 		*pino = ceph_ino(dir);
2340 		rcu_read_unlock();
2341 		*ppath = dentry->d_name.name;
2342 		*ppathlen = dentry->d_name.len;
2343 		return 0;
2344 	}
2345 	rcu_read_unlock();
2346 	path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2347 	if (IS_ERR(path))
2348 		return PTR_ERR(path);
2349 	*ppath = path;
2350 	*pfreepath = true;
2351 	return 0;
2352 }
2353 
2354 static int build_inode_path(struct inode *inode,
2355 			    const char **ppath, int *ppathlen, u64 *pino,
2356 			    bool *pfreepath)
2357 {
2358 	struct dentry *dentry;
2359 	char *path;
2360 
2361 	if (ceph_snap(inode) == CEPH_NOSNAP) {
2362 		*pino = ceph_ino(inode);
2363 		*ppathlen = 0;
2364 		return 0;
2365 	}
2366 	dentry = d_find_alias(inode);
2367 	path = ceph_mdsc_build_path(dentry, ppathlen, pino, 1);
2368 	dput(dentry);
2369 	if (IS_ERR(path))
2370 		return PTR_ERR(path);
2371 	*ppath = path;
2372 	*pfreepath = true;
2373 	return 0;
2374 }
2375 
2376 /*
2377  * request arguments may be specified via an inode *, a dentry *, or
2378  * an explicit ino+path.
2379  */
2380 static int set_request_path_attr(struct inode *rinode, struct dentry *rdentry,
2381 				  struct inode *rdiri, const char *rpath,
2382 				  u64 rino, const char **ppath, int *pathlen,
2383 				  u64 *ino, bool *freepath, bool parent_locked)
2384 {
2385 	int r = 0;
2386 
2387 	if (rinode) {
2388 		r = build_inode_path(rinode, ppath, pathlen, ino, freepath);
2389 		dout(" inode %p %llx.%llx\n", rinode, ceph_ino(rinode),
2390 		     ceph_snap(rinode));
2391 	} else if (rdentry) {
2392 		r = build_dentry_path(rdentry, rdiri, ppath, pathlen, ino,
2393 					freepath, parent_locked);
2394 		dout(" dentry %p %llx/%.*s\n", rdentry, *ino, *pathlen,
2395 		     *ppath);
2396 	} else if (rpath || rino) {
2397 		*ino = rino;
2398 		*ppath = rpath;
2399 		*pathlen = rpath ? strlen(rpath) : 0;
2400 		dout(" path %.*s\n", *pathlen, rpath);
2401 	}
2402 
2403 	return r;
2404 }
2405 
2406 /*
2407  * called under mdsc->mutex
2408  */
2409 static struct ceph_msg *create_request_message(struct ceph_mds_client *mdsc,
2410 					       struct ceph_mds_request *req,
2411 					       int mds, bool drop_cap_releases)
2412 {
2413 	struct ceph_msg *msg;
2414 	struct ceph_mds_request_head *head;
2415 	const char *path1 = NULL;
2416 	const char *path2 = NULL;
2417 	u64 ino1 = 0, ino2 = 0;
2418 	int pathlen1 = 0, pathlen2 = 0;
2419 	bool freepath1 = false, freepath2 = false;
2420 	int len;
2421 	u16 releases;
2422 	void *p, *end;
2423 	int ret;
2424 
2425 	ret = set_request_path_attr(req->r_inode, req->r_dentry,
2426 			      req->r_parent, req->r_path1, req->r_ino1.ino,
2427 			      &path1, &pathlen1, &ino1, &freepath1,
2428 			      test_bit(CEPH_MDS_R_PARENT_LOCKED,
2429 					&req->r_req_flags));
2430 	if (ret < 0) {
2431 		msg = ERR_PTR(ret);
2432 		goto out;
2433 	}
2434 
2435 	/* If r_old_dentry is set, then assume that its parent is locked */
2436 	ret = set_request_path_attr(NULL, req->r_old_dentry,
2437 			      req->r_old_dentry_dir,
2438 			      req->r_path2, req->r_ino2.ino,
2439 			      &path2, &pathlen2, &ino2, &freepath2, true);
2440 	if (ret < 0) {
2441 		msg = ERR_PTR(ret);
2442 		goto out_free1;
2443 	}
2444 
2445 	len = sizeof(*head) +
2446 		pathlen1 + pathlen2 + 2*(1 + sizeof(u32) + sizeof(u64)) +
2447 		sizeof(struct ceph_timespec);
2448 
2449 	/* calculate (max) length for cap releases */
2450 	len += sizeof(struct ceph_mds_request_release) *
2451 		(!!req->r_inode_drop + !!req->r_dentry_drop +
2452 		 !!req->r_old_inode_drop + !!req->r_old_dentry_drop);
2453 	if (req->r_dentry_drop)
2454 		len += pathlen1;
2455 	if (req->r_old_dentry_drop)
2456 		len += pathlen2;
2457 
2458 	msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
2459 	if (!msg) {
2460 		msg = ERR_PTR(-ENOMEM);
2461 		goto out_free2;
2462 	}
2463 
2464 	msg->hdr.version = cpu_to_le16(2);
2465 	msg->hdr.tid = cpu_to_le64(req->r_tid);
2466 
2467 	head = msg->front.iov_base;
2468 	p = msg->front.iov_base + sizeof(*head);
2469 	end = msg->front.iov_base + msg->front.iov_len;
2470 
2471 	head->mdsmap_epoch = cpu_to_le32(mdsc->mdsmap->m_epoch);
2472 	head->op = cpu_to_le32(req->r_op);
2473 	head->caller_uid = cpu_to_le32(from_kuid(&init_user_ns, req->r_uid));
2474 	head->caller_gid = cpu_to_le32(from_kgid(&init_user_ns, req->r_gid));
2475 	head->ino = cpu_to_le64(req->r_deleg_ino);
2476 	head->args = req->r_args;
2477 
2478 	ceph_encode_filepath(&p, end, ino1, path1);
2479 	ceph_encode_filepath(&p, end, ino2, path2);
2480 
2481 	/* make note of release offset, in case we need to replay */
2482 	req->r_request_release_offset = p - msg->front.iov_base;
2483 
2484 	/* cap releases */
2485 	releases = 0;
2486 	if (req->r_inode_drop)
2487 		releases += ceph_encode_inode_release(&p,
2488 		      req->r_inode ? req->r_inode : d_inode(req->r_dentry),
2489 		      mds, req->r_inode_drop, req->r_inode_unless,
2490 		      req->r_op == CEPH_MDS_OP_READDIR);
2491 	if (req->r_dentry_drop)
2492 		releases += ceph_encode_dentry_release(&p, req->r_dentry,
2493 				req->r_parent, mds, req->r_dentry_drop,
2494 				req->r_dentry_unless);
2495 	if (req->r_old_dentry_drop)
2496 		releases += ceph_encode_dentry_release(&p, req->r_old_dentry,
2497 				req->r_old_dentry_dir, mds,
2498 				req->r_old_dentry_drop,
2499 				req->r_old_dentry_unless);
2500 	if (req->r_old_inode_drop)
2501 		releases += ceph_encode_inode_release(&p,
2502 		      d_inode(req->r_old_dentry),
2503 		      mds, req->r_old_inode_drop, req->r_old_inode_unless, 0);
2504 
2505 	if (drop_cap_releases) {
2506 		releases = 0;
2507 		p = msg->front.iov_base + req->r_request_release_offset;
2508 	}
2509 
2510 	head->num_releases = cpu_to_le16(releases);
2511 
2512 	/* time stamp */
2513 	{
2514 		struct ceph_timespec ts;
2515 		ceph_encode_timespec64(&ts, &req->r_stamp);
2516 		ceph_encode_copy(&p, &ts, sizeof(ts));
2517 	}
2518 
2519 	BUG_ON(p > end);
2520 	msg->front.iov_len = p - msg->front.iov_base;
2521 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2522 
2523 	if (req->r_pagelist) {
2524 		struct ceph_pagelist *pagelist = req->r_pagelist;
2525 		ceph_msg_data_add_pagelist(msg, pagelist);
2526 		msg->hdr.data_len = cpu_to_le32(pagelist->length);
2527 	} else {
2528 		msg->hdr.data_len = 0;
2529 	}
2530 
2531 	msg->hdr.data_off = cpu_to_le16(0);
2532 
2533 out_free2:
2534 	if (freepath2)
2535 		ceph_mdsc_free_path((char *)path2, pathlen2);
2536 out_free1:
2537 	if (freepath1)
2538 		ceph_mdsc_free_path((char *)path1, pathlen1);
2539 out:
2540 	return msg;
2541 }
2542 
2543 /*
2544  * called under mdsc->mutex if error, under no mutex if
2545  * success.
2546  */
2547 static void complete_request(struct ceph_mds_client *mdsc,
2548 			     struct ceph_mds_request *req)
2549 {
2550 	if (req->r_callback)
2551 		req->r_callback(mdsc, req);
2552 	complete_all(&req->r_completion);
2553 }
2554 
2555 /*
2556  * called under mdsc->mutex
2557  */
2558 static int __prepare_send_request(struct ceph_mds_client *mdsc,
2559 				  struct ceph_mds_request *req,
2560 				  int mds, bool drop_cap_releases)
2561 {
2562 	struct ceph_mds_request_head *rhead;
2563 	struct ceph_msg *msg;
2564 	int flags = 0;
2565 
2566 	req->r_attempts++;
2567 	if (req->r_inode) {
2568 		struct ceph_cap *cap =
2569 			ceph_get_cap_for_mds(ceph_inode(req->r_inode), mds);
2570 
2571 		if (cap)
2572 			req->r_sent_on_mseq = cap->mseq;
2573 		else
2574 			req->r_sent_on_mseq = -1;
2575 	}
2576 	dout("prepare_send_request %p tid %lld %s (attempt %d)\n", req,
2577 	     req->r_tid, ceph_mds_op_name(req->r_op), req->r_attempts);
2578 
2579 	if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
2580 		void *p;
2581 		/*
2582 		 * Replay.  Do not regenerate message (and rebuild
2583 		 * paths, etc.); just use the original message.
2584 		 * Rebuilding paths will break for renames because
2585 		 * d_move mangles the src name.
2586 		 */
2587 		msg = req->r_request;
2588 		rhead = msg->front.iov_base;
2589 
2590 		flags = le32_to_cpu(rhead->flags);
2591 		flags |= CEPH_MDS_FLAG_REPLAY;
2592 		rhead->flags = cpu_to_le32(flags);
2593 
2594 		if (req->r_target_inode)
2595 			rhead->ino = cpu_to_le64(ceph_ino(req->r_target_inode));
2596 
2597 		rhead->num_retry = req->r_attempts - 1;
2598 
2599 		/* remove cap/dentry releases from message */
2600 		rhead->num_releases = 0;
2601 
2602 		/* time stamp */
2603 		p = msg->front.iov_base + req->r_request_release_offset;
2604 		{
2605 			struct ceph_timespec ts;
2606 			ceph_encode_timespec64(&ts, &req->r_stamp);
2607 			ceph_encode_copy(&p, &ts, sizeof(ts));
2608 		}
2609 
2610 		msg->front.iov_len = p - msg->front.iov_base;
2611 		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
2612 		return 0;
2613 	}
2614 
2615 	if (req->r_request) {
2616 		ceph_msg_put(req->r_request);
2617 		req->r_request = NULL;
2618 	}
2619 	msg = create_request_message(mdsc, req, mds, drop_cap_releases);
2620 	if (IS_ERR(msg)) {
2621 		req->r_err = PTR_ERR(msg);
2622 		return PTR_ERR(msg);
2623 	}
2624 	req->r_request = msg;
2625 
2626 	rhead = msg->front.iov_base;
2627 	rhead->oldest_client_tid = cpu_to_le64(__get_oldest_tid(mdsc));
2628 	if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2629 		flags |= CEPH_MDS_FLAG_REPLAY;
2630 	if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags))
2631 		flags |= CEPH_MDS_FLAG_ASYNC;
2632 	if (req->r_parent)
2633 		flags |= CEPH_MDS_FLAG_WANT_DENTRY;
2634 	rhead->flags = cpu_to_le32(flags);
2635 	rhead->num_fwd = req->r_num_fwd;
2636 	rhead->num_retry = req->r_attempts - 1;
2637 
2638 	dout(" r_parent = %p\n", req->r_parent);
2639 	return 0;
2640 }
2641 
2642 /*
2643  * called under mdsc->mutex
2644  */
2645 static int __send_request(struct ceph_mds_client *mdsc,
2646 			  struct ceph_mds_session *session,
2647 			  struct ceph_mds_request *req,
2648 			  bool drop_cap_releases)
2649 {
2650 	int err;
2651 
2652 	err = __prepare_send_request(mdsc, req, session->s_mds,
2653 				     drop_cap_releases);
2654 	if (!err) {
2655 		ceph_msg_get(req->r_request);
2656 		ceph_con_send(&session->s_con, req->r_request);
2657 	}
2658 
2659 	return err;
2660 }
2661 
2662 /*
2663  * send request, or put it on the appropriate wait list.
2664  */
2665 static void __do_request(struct ceph_mds_client *mdsc,
2666 			struct ceph_mds_request *req)
2667 {
2668 	struct ceph_mds_session *session = NULL;
2669 	int mds = -1;
2670 	int err = 0;
2671 	bool random;
2672 
2673 	if (req->r_err || test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2674 		if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
2675 			__unregister_request(mdsc, req);
2676 		return;
2677 	}
2678 
2679 	if (req->r_timeout &&
2680 	    time_after_eq(jiffies, req->r_started + req->r_timeout)) {
2681 		dout("do_request timed out\n");
2682 		err = -ETIMEDOUT;
2683 		goto finish;
2684 	}
2685 	if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
2686 		dout("do_request forced umount\n");
2687 		err = -EIO;
2688 		goto finish;
2689 	}
2690 	if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_MOUNTING) {
2691 		if (mdsc->mdsmap_err) {
2692 			err = mdsc->mdsmap_err;
2693 			dout("do_request mdsmap err %d\n", err);
2694 			goto finish;
2695 		}
2696 		if (mdsc->mdsmap->m_epoch == 0) {
2697 			dout("do_request no mdsmap, waiting for map\n");
2698 			list_add(&req->r_wait, &mdsc->waiting_for_map);
2699 			return;
2700 		}
2701 		if (!(mdsc->fsc->mount_options->flags &
2702 		      CEPH_MOUNT_OPT_MOUNTWAIT) &&
2703 		    !ceph_mdsmap_is_cluster_available(mdsc->mdsmap)) {
2704 			err = -EHOSTUNREACH;
2705 			goto finish;
2706 		}
2707 	}
2708 
2709 	put_request_session(req);
2710 
2711 	mds = __choose_mds(mdsc, req, &random);
2712 	if (mds < 0 ||
2713 	    ceph_mdsmap_get_state(mdsc->mdsmap, mds) < CEPH_MDS_STATE_ACTIVE) {
2714 		if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2715 			err = -EJUKEBOX;
2716 			goto finish;
2717 		}
2718 		dout("do_request no mds or not active, waiting for map\n");
2719 		list_add(&req->r_wait, &mdsc->waiting_for_map);
2720 		return;
2721 	}
2722 
2723 	/* get, open session */
2724 	session = __ceph_lookup_mds_session(mdsc, mds);
2725 	if (!session) {
2726 		session = register_session(mdsc, mds);
2727 		if (IS_ERR(session)) {
2728 			err = PTR_ERR(session);
2729 			goto finish;
2730 		}
2731 	}
2732 	req->r_session = ceph_get_mds_session(session);
2733 
2734 	dout("do_request mds%d session %p state %s\n", mds, session,
2735 	     ceph_session_state_name(session->s_state));
2736 	if (session->s_state != CEPH_MDS_SESSION_OPEN &&
2737 	    session->s_state != CEPH_MDS_SESSION_HUNG) {
2738 		if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
2739 			err = -EACCES;
2740 			goto out_session;
2741 		}
2742 		/*
2743 		 * We cannot queue async requests since the caps and delegated
2744 		 * inodes are bound to the session. Just return -EJUKEBOX and
2745 		 * let the caller retry a sync request in that case.
2746 		 */
2747 		if (test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags)) {
2748 			err = -EJUKEBOX;
2749 			goto out_session;
2750 		}
2751 		if (session->s_state == CEPH_MDS_SESSION_NEW ||
2752 		    session->s_state == CEPH_MDS_SESSION_CLOSING) {
2753 			__open_session(mdsc, session);
2754 			/* retry the same mds later */
2755 			if (random)
2756 				req->r_resend_mds = mds;
2757 		}
2758 		list_add(&req->r_wait, &session->s_waiting);
2759 		goto out_session;
2760 	}
2761 
2762 	/* send request */
2763 	req->r_resend_mds = -1;   /* forget any previous mds hint */
2764 
2765 	if (req->r_request_started == 0)   /* note request start time */
2766 		req->r_request_started = jiffies;
2767 
2768 	err = __send_request(mdsc, session, req, false);
2769 
2770 out_session:
2771 	ceph_put_mds_session(session);
2772 finish:
2773 	if (err) {
2774 		dout("__do_request early error %d\n", err);
2775 		req->r_err = err;
2776 		complete_request(mdsc, req);
2777 		__unregister_request(mdsc, req);
2778 	}
2779 	return;
2780 }
2781 
2782 /*
2783  * called under mdsc->mutex
2784  */
2785 static void __wake_requests(struct ceph_mds_client *mdsc,
2786 			    struct list_head *head)
2787 {
2788 	struct ceph_mds_request *req;
2789 	LIST_HEAD(tmp_list);
2790 
2791 	list_splice_init(head, &tmp_list);
2792 
2793 	while (!list_empty(&tmp_list)) {
2794 		req = list_entry(tmp_list.next,
2795 				 struct ceph_mds_request, r_wait);
2796 		list_del_init(&req->r_wait);
2797 		dout(" wake request %p tid %llu\n", req, req->r_tid);
2798 		__do_request(mdsc, req);
2799 	}
2800 }
2801 
2802 /*
2803  * Wake up threads with requests pending for @mds, so that they can
2804  * resubmit their requests to a possibly different mds.
2805  */
2806 static void kick_requests(struct ceph_mds_client *mdsc, int mds)
2807 {
2808 	struct ceph_mds_request *req;
2809 	struct rb_node *p = rb_first(&mdsc->request_tree);
2810 
2811 	dout("kick_requests mds%d\n", mds);
2812 	while (p) {
2813 		req = rb_entry(p, struct ceph_mds_request, r_node);
2814 		p = rb_next(p);
2815 		if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
2816 			continue;
2817 		if (req->r_attempts > 0)
2818 			continue; /* only new requests */
2819 		if (req->r_session &&
2820 		    req->r_session->s_mds == mds) {
2821 			dout(" kicking tid %llu\n", req->r_tid);
2822 			list_del_init(&req->r_wait);
2823 			__do_request(mdsc, req);
2824 		}
2825 	}
2826 }
2827 
2828 int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
2829 			      struct ceph_mds_request *req)
2830 {
2831 	int err = 0;
2832 
2833 	/* take CAP_PIN refs for r_inode, r_parent, r_old_dentry */
2834 	if (req->r_inode)
2835 		ceph_get_cap_refs(ceph_inode(req->r_inode), CEPH_CAP_PIN);
2836 	if (req->r_parent) {
2837 		struct ceph_inode_info *ci = ceph_inode(req->r_parent);
2838 		int fmode = (req->r_op & CEPH_MDS_OP_WRITE) ?
2839 			    CEPH_FILE_MODE_WR : CEPH_FILE_MODE_RD;
2840 		spin_lock(&ci->i_ceph_lock);
2841 		ceph_take_cap_refs(ci, CEPH_CAP_PIN, false);
2842 		__ceph_touch_fmode(ci, mdsc, fmode);
2843 		spin_unlock(&ci->i_ceph_lock);
2844 		ihold(req->r_parent);
2845 	}
2846 	if (req->r_old_dentry_dir)
2847 		ceph_get_cap_refs(ceph_inode(req->r_old_dentry_dir),
2848 				  CEPH_CAP_PIN);
2849 
2850 	if (req->r_inode) {
2851 		err = ceph_wait_on_async_create(req->r_inode);
2852 		if (err) {
2853 			dout("%s: wait for async create returned: %d\n",
2854 			     __func__, err);
2855 			return err;
2856 		}
2857 	}
2858 
2859 	if (!err && req->r_old_inode) {
2860 		err = ceph_wait_on_async_create(req->r_old_inode);
2861 		if (err) {
2862 			dout("%s: wait for async create returned: %d\n",
2863 			     __func__, err);
2864 			return err;
2865 		}
2866 	}
2867 
2868 	dout("submit_request on %p for inode %p\n", req, dir);
2869 	mutex_lock(&mdsc->mutex);
2870 	__register_request(mdsc, req, dir);
2871 	__do_request(mdsc, req);
2872 	err = req->r_err;
2873 	mutex_unlock(&mdsc->mutex);
2874 	return err;
2875 }
2876 
2877 static int ceph_mdsc_wait_request(struct ceph_mds_client *mdsc,
2878 				  struct ceph_mds_request *req)
2879 {
2880 	int err;
2881 
2882 	/* wait */
2883 	dout("do_request waiting\n");
2884 	if (!req->r_timeout && req->r_wait_for_completion) {
2885 		err = req->r_wait_for_completion(mdsc, req);
2886 	} else {
2887 		long timeleft = wait_for_completion_killable_timeout(
2888 					&req->r_completion,
2889 					ceph_timeout_jiffies(req->r_timeout));
2890 		if (timeleft > 0)
2891 			err = 0;
2892 		else if (!timeleft)
2893 			err = -ETIMEDOUT;  /* timed out */
2894 		else
2895 			err = timeleft;  /* killed */
2896 	}
2897 	dout("do_request waited, got %d\n", err);
2898 	mutex_lock(&mdsc->mutex);
2899 
2900 	/* only abort if we didn't race with a real reply */
2901 	if (test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags)) {
2902 		err = le32_to_cpu(req->r_reply_info.head->result);
2903 	} else if (err < 0) {
2904 		dout("aborted request %lld with %d\n", req->r_tid, err);
2905 
2906 		/*
2907 		 * ensure we aren't running concurrently with
2908 		 * ceph_fill_trace or ceph_readdir_prepopulate, which
2909 		 * rely on locks (dir mutex) held by our caller.
2910 		 */
2911 		mutex_lock(&req->r_fill_mutex);
2912 		req->r_err = err;
2913 		set_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags);
2914 		mutex_unlock(&req->r_fill_mutex);
2915 
2916 		if (req->r_parent &&
2917 		    (req->r_op & CEPH_MDS_OP_WRITE))
2918 			ceph_invalidate_dir_request(req);
2919 	} else {
2920 		err = req->r_err;
2921 	}
2922 
2923 	mutex_unlock(&mdsc->mutex);
2924 	return err;
2925 }
2926 
2927 /*
2928  * Synchrously perform an mds request.  Take care of all of the
2929  * session setup, forwarding, retry details.
2930  */
2931 int ceph_mdsc_do_request(struct ceph_mds_client *mdsc,
2932 			 struct inode *dir,
2933 			 struct ceph_mds_request *req)
2934 {
2935 	int err;
2936 
2937 	dout("do_request on %p\n", req);
2938 
2939 	/* issue */
2940 	err = ceph_mdsc_submit_request(mdsc, dir, req);
2941 	if (!err)
2942 		err = ceph_mdsc_wait_request(mdsc, req);
2943 	dout("do_request %p done, result %d\n", req, err);
2944 	return err;
2945 }
2946 
2947 /*
2948  * Invalidate dir's completeness, dentry lease state on an aborted MDS
2949  * namespace request.
2950  */
2951 void ceph_invalidate_dir_request(struct ceph_mds_request *req)
2952 {
2953 	struct inode *dir = req->r_parent;
2954 	struct inode *old_dir = req->r_old_dentry_dir;
2955 
2956 	dout("invalidate_dir_request %p %p (complete, lease(s))\n", dir, old_dir);
2957 
2958 	ceph_dir_clear_complete(dir);
2959 	if (old_dir)
2960 		ceph_dir_clear_complete(old_dir);
2961 	if (req->r_dentry)
2962 		ceph_invalidate_dentry_lease(req->r_dentry);
2963 	if (req->r_old_dentry)
2964 		ceph_invalidate_dentry_lease(req->r_old_dentry);
2965 }
2966 
2967 /*
2968  * Handle mds reply.
2969  *
2970  * We take the session mutex and parse and process the reply immediately.
2971  * This preserves the logical ordering of replies, capabilities, etc., sent
2972  * by the MDS as they are applied to our local cache.
2973  */
2974 static void handle_reply(struct ceph_mds_session *session, struct ceph_msg *msg)
2975 {
2976 	struct ceph_mds_client *mdsc = session->s_mdsc;
2977 	struct ceph_mds_request *req;
2978 	struct ceph_mds_reply_head *head = msg->front.iov_base;
2979 	struct ceph_mds_reply_info_parsed *rinfo;  /* parsed reply info */
2980 	struct ceph_snap_realm *realm;
2981 	u64 tid;
2982 	int err, result;
2983 	int mds = session->s_mds;
2984 
2985 	if (msg->front.iov_len < sizeof(*head)) {
2986 		pr_err("mdsc_handle_reply got corrupt (short) reply\n");
2987 		ceph_msg_dump(msg);
2988 		return;
2989 	}
2990 
2991 	/* get request, session */
2992 	tid = le64_to_cpu(msg->hdr.tid);
2993 	mutex_lock(&mdsc->mutex);
2994 	req = lookup_get_request(mdsc, tid);
2995 	if (!req) {
2996 		dout("handle_reply on unknown tid %llu\n", tid);
2997 		mutex_unlock(&mdsc->mutex);
2998 		return;
2999 	}
3000 	dout("handle_reply %p\n", req);
3001 
3002 	/* correct session? */
3003 	if (req->r_session != session) {
3004 		pr_err("mdsc_handle_reply got %llu on session mds%d"
3005 		       " not mds%d\n", tid, session->s_mds,
3006 		       req->r_session ? req->r_session->s_mds : -1);
3007 		mutex_unlock(&mdsc->mutex);
3008 		goto out;
3009 	}
3010 
3011 	/* dup? */
3012 	if ((test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags) && !head->safe) ||
3013 	    (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags) && head->safe)) {
3014 		pr_warn("got a dup %s reply on %llu from mds%d\n",
3015 			   head->safe ? "safe" : "unsafe", tid, mds);
3016 		mutex_unlock(&mdsc->mutex);
3017 		goto out;
3018 	}
3019 	if (test_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags)) {
3020 		pr_warn("got unsafe after safe on %llu from mds%d\n",
3021 			   tid, mds);
3022 		mutex_unlock(&mdsc->mutex);
3023 		goto out;
3024 	}
3025 
3026 	result = le32_to_cpu(head->result);
3027 
3028 	/*
3029 	 * Handle an ESTALE
3030 	 * if we're not talking to the authority, send to them
3031 	 * if the authority has changed while we weren't looking,
3032 	 * send to new authority
3033 	 * Otherwise we just have to return an ESTALE
3034 	 */
3035 	if (result == -ESTALE) {
3036 		dout("got ESTALE on request %llu\n", req->r_tid);
3037 		req->r_resend_mds = -1;
3038 		if (req->r_direct_mode != USE_AUTH_MDS) {
3039 			dout("not using auth, setting for that now\n");
3040 			req->r_direct_mode = USE_AUTH_MDS;
3041 			__do_request(mdsc, req);
3042 			mutex_unlock(&mdsc->mutex);
3043 			goto out;
3044 		} else  {
3045 			int mds = __choose_mds(mdsc, req, NULL);
3046 			if (mds >= 0 && mds != req->r_session->s_mds) {
3047 				dout("but auth changed, so resending\n");
3048 				__do_request(mdsc, req);
3049 				mutex_unlock(&mdsc->mutex);
3050 				goto out;
3051 			}
3052 		}
3053 		dout("have to return ESTALE on request %llu\n", req->r_tid);
3054 	}
3055 
3056 
3057 	if (head->safe) {
3058 		set_bit(CEPH_MDS_R_GOT_SAFE, &req->r_req_flags);
3059 		__unregister_request(mdsc, req);
3060 
3061 		/* last request during umount? */
3062 		if (mdsc->stopping && !__get_oldest_req(mdsc))
3063 			complete_all(&mdsc->safe_umount_waiters);
3064 
3065 		if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3066 			/*
3067 			 * We already handled the unsafe response, now do the
3068 			 * cleanup.  No need to examine the response; the MDS
3069 			 * doesn't include any result info in the safe
3070 			 * response.  And even if it did, there is nothing
3071 			 * useful we could do with a revised return value.
3072 			 */
3073 			dout("got safe reply %llu, mds%d\n", tid, mds);
3074 
3075 			mutex_unlock(&mdsc->mutex);
3076 			goto out;
3077 		}
3078 	} else {
3079 		set_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags);
3080 		list_add_tail(&req->r_unsafe_item, &req->r_session->s_unsafe);
3081 	}
3082 
3083 	dout("handle_reply tid %lld result %d\n", tid, result);
3084 	rinfo = &req->r_reply_info;
3085 	if (test_bit(CEPHFS_FEATURE_REPLY_ENCODING, &session->s_features))
3086 		err = parse_reply_info(session, msg, rinfo, (u64)-1);
3087 	else
3088 		err = parse_reply_info(session, msg, rinfo, session->s_con.peer_features);
3089 	mutex_unlock(&mdsc->mutex);
3090 
3091 	mutex_lock(&session->s_mutex);
3092 	if (err < 0) {
3093 		pr_err("mdsc_handle_reply got corrupt reply mds%d(tid:%lld)\n", mds, tid);
3094 		ceph_msg_dump(msg);
3095 		goto out_err;
3096 	}
3097 
3098 	/* snap trace */
3099 	realm = NULL;
3100 	if (rinfo->snapblob_len) {
3101 		down_write(&mdsc->snap_rwsem);
3102 		ceph_update_snap_trace(mdsc, rinfo->snapblob,
3103 				rinfo->snapblob + rinfo->snapblob_len,
3104 				le32_to_cpu(head->op) == CEPH_MDS_OP_RMSNAP,
3105 				&realm);
3106 		downgrade_write(&mdsc->snap_rwsem);
3107 	} else {
3108 		down_read(&mdsc->snap_rwsem);
3109 	}
3110 
3111 	/* insert trace into our cache */
3112 	mutex_lock(&req->r_fill_mutex);
3113 	current->journal_info = req;
3114 	err = ceph_fill_trace(mdsc->fsc->sb, req);
3115 	if (err == 0) {
3116 		if (result == 0 && (req->r_op == CEPH_MDS_OP_READDIR ||
3117 				    req->r_op == CEPH_MDS_OP_LSSNAP))
3118 			ceph_readdir_prepopulate(req, req->r_session);
3119 	}
3120 	current->journal_info = NULL;
3121 	mutex_unlock(&req->r_fill_mutex);
3122 
3123 	up_read(&mdsc->snap_rwsem);
3124 	if (realm)
3125 		ceph_put_snap_realm(mdsc, realm);
3126 
3127 	if (err == 0) {
3128 		if (req->r_target_inode &&
3129 		    test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags)) {
3130 			struct ceph_inode_info *ci =
3131 				ceph_inode(req->r_target_inode);
3132 			spin_lock(&ci->i_unsafe_lock);
3133 			list_add_tail(&req->r_unsafe_target_item,
3134 				      &ci->i_unsafe_iops);
3135 			spin_unlock(&ci->i_unsafe_lock);
3136 		}
3137 
3138 		ceph_unreserve_caps(mdsc, &req->r_caps_reservation);
3139 	}
3140 out_err:
3141 	mutex_lock(&mdsc->mutex);
3142 	if (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3143 		if (err) {
3144 			req->r_err = err;
3145 		} else {
3146 			req->r_reply =  ceph_msg_get(msg);
3147 			set_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags);
3148 		}
3149 	} else {
3150 		dout("reply arrived after request %lld was aborted\n", tid);
3151 	}
3152 	mutex_unlock(&mdsc->mutex);
3153 
3154 	mutex_unlock(&session->s_mutex);
3155 
3156 	/* kick calling process */
3157 	complete_request(mdsc, req);
3158 out:
3159 	ceph_mdsc_put_request(req);
3160 	return;
3161 }
3162 
3163 
3164 
3165 /*
3166  * handle mds notification that our request has been forwarded.
3167  */
3168 static void handle_forward(struct ceph_mds_client *mdsc,
3169 			   struct ceph_mds_session *session,
3170 			   struct ceph_msg *msg)
3171 {
3172 	struct ceph_mds_request *req;
3173 	u64 tid = le64_to_cpu(msg->hdr.tid);
3174 	u32 next_mds;
3175 	u32 fwd_seq;
3176 	int err = -EINVAL;
3177 	void *p = msg->front.iov_base;
3178 	void *end = p + msg->front.iov_len;
3179 
3180 	ceph_decode_need(&p, end, 2*sizeof(u32), bad);
3181 	next_mds = ceph_decode_32(&p);
3182 	fwd_seq = ceph_decode_32(&p);
3183 
3184 	mutex_lock(&mdsc->mutex);
3185 	req = lookup_get_request(mdsc, tid);
3186 	if (!req) {
3187 		dout("forward tid %llu to mds%d - req dne\n", tid, next_mds);
3188 		goto out;  /* dup reply? */
3189 	}
3190 
3191 	if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
3192 		dout("forward tid %llu aborted, unregistering\n", tid);
3193 		__unregister_request(mdsc, req);
3194 	} else if (fwd_seq <= req->r_num_fwd) {
3195 		dout("forward tid %llu to mds%d - old seq %d <= %d\n",
3196 		     tid, next_mds, req->r_num_fwd, fwd_seq);
3197 	} else {
3198 		/* resend. forward race not possible; mds would drop */
3199 		dout("forward tid %llu to mds%d (we resend)\n", tid, next_mds);
3200 		BUG_ON(req->r_err);
3201 		BUG_ON(test_bit(CEPH_MDS_R_GOT_RESULT, &req->r_req_flags));
3202 		req->r_attempts = 0;
3203 		req->r_num_fwd = fwd_seq;
3204 		req->r_resend_mds = next_mds;
3205 		put_request_session(req);
3206 		__do_request(mdsc, req);
3207 	}
3208 	ceph_mdsc_put_request(req);
3209 out:
3210 	mutex_unlock(&mdsc->mutex);
3211 	return;
3212 
3213 bad:
3214 	pr_err("mdsc_handle_forward decode error err=%d\n", err);
3215 }
3216 
3217 static int __decode_session_metadata(void **p, void *end,
3218 				     bool *blacklisted)
3219 {
3220 	/* map<string,string> */
3221 	u32 n;
3222 	bool err_str;
3223 	ceph_decode_32_safe(p, end, n, bad);
3224 	while (n-- > 0) {
3225 		u32 len;
3226 		ceph_decode_32_safe(p, end, len, bad);
3227 		ceph_decode_need(p, end, len, bad);
3228 		err_str = !strncmp(*p, "error_string", len);
3229 		*p += len;
3230 		ceph_decode_32_safe(p, end, len, bad);
3231 		ceph_decode_need(p, end, len, bad);
3232 		if (err_str && strnstr(*p, "blacklisted", len))
3233 			*blacklisted = true;
3234 		*p += len;
3235 	}
3236 	return 0;
3237 bad:
3238 	return -1;
3239 }
3240 
3241 /*
3242  * handle a mds session control message
3243  */
3244 static void handle_session(struct ceph_mds_session *session,
3245 			   struct ceph_msg *msg)
3246 {
3247 	struct ceph_mds_client *mdsc = session->s_mdsc;
3248 	int mds = session->s_mds;
3249 	int msg_version = le16_to_cpu(msg->hdr.version);
3250 	void *p = msg->front.iov_base;
3251 	void *end = p + msg->front.iov_len;
3252 	struct ceph_mds_session_head *h;
3253 	u32 op;
3254 	u64 seq;
3255 	unsigned long features = 0;
3256 	int wake = 0;
3257 	bool blacklisted = false;
3258 
3259 	/* decode */
3260 	ceph_decode_need(&p, end, sizeof(*h), bad);
3261 	h = p;
3262 	p += sizeof(*h);
3263 
3264 	op = le32_to_cpu(h->op);
3265 	seq = le64_to_cpu(h->seq);
3266 
3267 	if (msg_version >= 3) {
3268 		u32 len;
3269 		/* version >= 2, metadata */
3270 		if (__decode_session_metadata(&p, end, &blacklisted) < 0)
3271 			goto bad;
3272 		/* version >= 3, feature bits */
3273 		ceph_decode_32_safe(&p, end, len, bad);
3274 		ceph_decode_need(&p, end, len, bad);
3275 		memcpy(&features, p, min_t(size_t, len, sizeof(features)));
3276 		p += len;
3277 	}
3278 
3279 	mutex_lock(&mdsc->mutex);
3280 	if (op == CEPH_SESSION_CLOSE) {
3281 		ceph_get_mds_session(session);
3282 		__unregister_session(mdsc, session);
3283 	}
3284 	/* FIXME: this ttl calculation is generous */
3285 	session->s_ttl = jiffies + HZ*mdsc->mdsmap->m_session_autoclose;
3286 	mutex_unlock(&mdsc->mutex);
3287 
3288 	mutex_lock(&session->s_mutex);
3289 
3290 	dout("handle_session mds%d %s %p state %s seq %llu\n",
3291 	     mds, ceph_session_op_name(op), session,
3292 	     ceph_session_state_name(session->s_state), seq);
3293 
3294 	if (session->s_state == CEPH_MDS_SESSION_HUNG) {
3295 		session->s_state = CEPH_MDS_SESSION_OPEN;
3296 		pr_info("mds%d came back\n", session->s_mds);
3297 	}
3298 
3299 	switch (op) {
3300 	case CEPH_SESSION_OPEN:
3301 		if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3302 			pr_info("mds%d reconnect success\n", session->s_mds);
3303 		session->s_state = CEPH_MDS_SESSION_OPEN;
3304 		session->s_features = features;
3305 		renewed_caps(mdsc, session, 0);
3306 		wake = 1;
3307 		if (mdsc->stopping)
3308 			__close_session(mdsc, session);
3309 		break;
3310 
3311 	case CEPH_SESSION_RENEWCAPS:
3312 		if (session->s_renew_seq == seq)
3313 			renewed_caps(mdsc, session, 1);
3314 		break;
3315 
3316 	case CEPH_SESSION_CLOSE:
3317 		if (session->s_state == CEPH_MDS_SESSION_RECONNECTING)
3318 			pr_info("mds%d reconnect denied\n", session->s_mds);
3319 		session->s_state = CEPH_MDS_SESSION_CLOSED;
3320 		cleanup_session_requests(mdsc, session);
3321 		remove_session_caps(session);
3322 		wake = 2; /* for good measure */
3323 		wake_up_all(&mdsc->session_close_wq);
3324 		break;
3325 
3326 	case CEPH_SESSION_STALE:
3327 		pr_info("mds%d caps went stale, renewing\n",
3328 			session->s_mds);
3329 		spin_lock(&session->s_gen_ttl_lock);
3330 		session->s_cap_gen++;
3331 		session->s_cap_ttl = jiffies - 1;
3332 		spin_unlock(&session->s_gen_ttl_lock);
3333 		send_renew_caps(mdsc, session);
3334 		break;
3335 
3336 	case CEPH_SESSION_RECALL_STATE:
3337 		ceph_trim_caps(mdsc, session, le32_to_cpu(h->max_caps));
3338 		break;
3339 
3340 	case CEPH_SESSION_FLUSHMSG:
3341 		send_flushmsg_ack(mdsc, session, seq);
3342 		break;
3343 
3344 	case CEPH_SESSION_FORCE_RO:
3345 		dout("force_session_readonly %p\n", session);
3346 		spin_lock(&session->s_cap_lock);
3347 		session->s_readonly = true;
3348 		spin_unlock(&session->s_cap_lock);
3349 		wake_up_session_caps(session, FORCE_RO);
3350 		break;
3351 
3352 	case CEPH_SESSION_REJECT:
3353 		WARN_ON(session->s_state != CEPH_MDS_SESSION_OPENING);
3354 		pr_info("mds%d rejected session\n", session->s_mds);
3355 		session->s_state = CEPH_MDS_SESSION_REJECTED;
3356 		cleanup_session_requests(mdsc, session);
3357 		remove_session_caps(session);
3358 		if (blacklisted)
3359 			mdsc->fsc->blacklisted = true;
3360 		wake = 2; /* for good measure */
3361 		break;
3362 
3363 	default:
3364 		pr_err("mdsc_handle_session bad op %d mds%d\n", op, mds);
3365 		WARN_ON(1);
3366 	}
3367 
3368 	mutex_unlock(&session->s_mutex);
3369 	if (wake) {
3370 		mutex_lock(&mdsc->mutex);
3371 		__wake_requests(mdsc, &session->s_waiting);
3372 		if (wake == 2)
3373 			kick_requests(mdsc, mds);
3374 		mutex_unlock(&mdsc->mutex);
3375 	}
3376 	if (op == CEPH_SESSION_CLOSE)
3377 		ceph_put_mds_session(session);
3378 	return;
3379 
3380 bad:
3381 	pr_err("mdsc_handle_session corrupt message mds%d len %d\n", mds,
3382 	       (int)msg->front.iov_len);
3383 	ceph_msg_dump(msg);
3384 	return;
3385 }
3386 
3387 void ceph_mdsc_release_dir_caps(struct ceph_mds_request *req)
3388 {
3389 	int dcaps;
3390 
3391 	dcaps = xchg(&req->r_dir_caps, 0);
3392 	if (dcaps) {
3393 		dout("releasing r_dir_caps=%s\n", ceph_cap_string(dcaps));
3394 		ceph_put_cap_refs(ceph_inode(req->r_parent), dcaps);
3395 	}
3396 }
3397 
3398 /*
3399  * called under session->mutex.
3400  */
3401 static void replay_unsafe_requests(struct ceph_mds_client *mdsc,
3402 				   struct ceph_mds_session *session)
3403 {
3404 	struct ceph_mds_request *req, *nreq;
3405 	struct rb_node *p;
3406 
3407 	dout("replay_unsafe_requests mds%d\n", session->s_mds);
3408 
3409 	mutex_lock(&mdsc->mutex);
3410 	list_for_each_entry_safe(req, nreq, &session->s_unsafe, r_unsafe_item)
3411 		__send_request(mdsc, session, req, true);
3412 
3413 	/*
3414 	 * also re-send old requests when MDS enters reconnect stage. So that MDS
3415 	 * can process completed request in clientreplay stage.
3416 	 */
3417 	p = rb_first(&mdsc->request_tree);
3418 	while (p) {
3419 		req = rb_entry(p, struct ceph_mds_request, r_node);
3420 		p = rb_next(p);
3421 		if (test_bit(CEPH_MDS_R_GOT_UNSAFE, &req->r_req_flags))
3422 			continue;
3423 		if (req->r_attempts == 0)
3424 			continue; /* only old requests */
3425 		if (!req->r_session)
3426 			continue;
3427 		if (req->r_session->s_mds != session->s_mds)
3428 			continue;
3429 
3430 		ceph_mdsc_release_dir_caps(req);
3431 
3432 		__send_request(mdsc, session, req, true);
3433 	}
3434 	mutex_unlock(&mdsc->mutex);
3435 }
3436 
3437 static int send_reconnect_partial(struct ceph_reconnect_state *recon_state)
3438 {
3439 	struct ceph_msg *reply;
3440 	struct ceph_pagelist *_pagelist;
3441 	struct page *page;
3442 	__le32 *addr;
3443 	int err = -ENOMEM;
3444 
3445 	if (!recon_state->allow_multi)
3446 		return -ENOSPC;
3447 
3448 	/* can't handle message that contains both caps and realm */
3449 	BUG_ON(!recon_state->nr_caps == !recon_state->nr_realms);
3450 
3451 	/* pre-allocate new pagelist */
3452 	_pagelist = ceph_pagelist_alloc(GFP_NOFS);
3453 	if (!_pagelist)
3454 		return -ENOMEM;
3455 
3456 	reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3457 	if (!reply)
3458 		goto fail_msg;
3459 
3460 	/* placeholder for nr_caps */
3461 	err = ceph_pagelist_encode_32(_pagelist, 0);
3462 	if (err < 0)
3463 		goto fail;
3464 
3465 	if (recon_state->nr_caps) {
3466 		/* currently encoding caps */
3467 		err = ceph_pagelist_encode_32(recon_state->pagelist, 0);
3468 		if (err)
3469 			goto fail;
3470 	} else {
3471 		/* placeholder for nr_realms (currently encoding relams) */
3472 		err = ceph_pagelist_encode_32(_pagelist, 0);
3473 		if (err < 0)
3474 			goto fail;
3475 	}
3476 
3477 	err = ceph_pagelist_encode_8(recon_state->pagelist, 1);
3478 	if (err)
3479 		goto fail;
3480 
3481 	page = list_first_entry(&recon_state->pagelist->head, struct page, lru);
3482 	addr = kmap_atomic(page);
3483 	if (recon_state->nr_caps) {
3484 		/* currently encoding caps */
3485 		*addr = cpu_to_le32(recon_state->nr_caps);
3486 	} else {
3487 		/* currently encoding relams */
3488 		*(addr + 1) = cpu_to_le32(recon_state->nr_realms);
3489 	}
3490 	kunmap_atomic(addr);
3491 
3492 	reply->hdr.version = cpu_to_le16(5);
3493 	reply->hdr.compat_version = cpu_to_le16(4);
3494 
3495 	reply->hdr.data_len = cpu_to_le32(recon_state->pagelist->length);
3496 	ceph_msg_data_add_pagelist(reply, recon_state->pagelist);
3497 
3498 	ceph_con_send(&recon_state->session->s_con, reply);
3499 	ceph_pagelist_release(recon_state->pagelist);
3500 
3501 	recon_state->pagelist = _pagelist;
3502 	recon_state->nr_caps = 0;
3503 	recon_state->nr_realms = 0;
3504 	recon_state->msg_version = 5;
3505 	return 0;
3506 fail:
3507 	ceph_msg_put(reply);
3508 fail_msg:
3509 	ceph_pagelist_release(_pagelist);
3510 	return err;
3511 }
3512 
3513 /*
3514  * Encode information about a cap for a reconnect with the MDS.
3515  */
3516 static int reconnect_caps_cb(struct inode *inode, struct ceph_cap *cap,
3517 			  void *arg)
3518 {
3519 	union {
3520 		struct ceph_mds_cap_reconnect v2;
3521 		struct ceph_mds_cap_reconnect_v1 v1;
3522 	} rec;
3523 	struct ceph_inode_info *ci = cap->ci;
3524 	struct ceph_reconnect_state *recon_state = arg;
3525 	struct ceph_pagelist *pagelist = recon_state->pagelist;
3526 	int err;
3527 	u64 snap_follows;
3528 
3529 	dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
3530 	     inode, ceph_vinop(inode), cap, cap->cap_id,
3531 	     ceph_cap_string(cap->issued));
3532 
3533 	spin_lock(&ci->i_ceph_lock);
3534 	cap->seq = 0;        /* reset cap seq */
3535 	cap->issue_seq = 0;  /* and issue_seq */
3536 	cap->mseq = 0;       /* and migrate_seq */
3537 	cap->cap_gen = cap->session->s_cap_gen;
3538 
3539 	/* These are lost when the session goes away */
3540 	if (S_ISDIR(inode->i_mode)) {
3541 		if (cap->issued & CEPH_CAP_DIR_CREATE) {
3542 			ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
3543 			memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
3544 		}
3545 		cap->issued &= ~CEPH_CAP_ANY_DIR_OPS;
3546 	}
3547 
3548 	if (recon_state->msg_version >= 2) {
3549 		rec.v2.cap_id = cpu_to_le64(cap->cap_id);
3550 		rec.v2.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3551 		rec.v2.issued = cpu_to_le32(cap->issued);
3552 		rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3553 		rec.v2.pathbase = 0;
3554 		rec.v2.flock_len = (__force __le32)
3555 			((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1);
3556 	} else {
3557 		rec.v1.cap_id = cpu_to_le64(cap->cap_id);
3558 		rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
3559 		rec.v1.issued = cpu_to_le32(cap->issued);
3560 		rec.v1.size = cpu_to_le64(inode->i_size);
3561 		ceph_encode_timespec64(&rec.v1.mtime, &inode->i_mtime);
3562 		ceph_encode_timespec64(&rec.v1.atime, &inode->i_atime);
3563 		rec.v1.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
3564 		rec.v1.pathbase = 0;
3565 	}
3566 
3567 	if (list_empty(&ci->i_cap_snaps)) {
3568 		snap_follows = ci->i_head_snapc ? ci->i_head_snapc->seq : 0;
3569 	} else {
3570 		struct ceph_cap_snap *capsnap =
3571 			list_first_entry(&ci->i_cap_snaps,
3572 					 struct ceph_cap_snap, ci_item);
3573 		snap_follows = capsnap->follows;
3574 	}
3575 	spin_unlock(&ci->i_ceph_lock);
3576 
3577 	if (recon_state->msg_version >= 2) {
3578 		int num_fcntl_locks, num_flock_locks;
3579 		struct ceph_filelock *flocks = NULL;
3580 		size_t struct_len, total_len = sizeof(u64);
3581 		u8 struct_v = 0;
3582 
3583 encode_again:
3584 		if (rec.v2.flock_len) {
3585 			ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
3586 		} else {
3587 			num_fcntl_locks = 0;
3588 			num_flock_locks = 0;
3589 		}
3590 		if (num_fcntl_locks + num_flock_locks > 0) {
3591 			flocks = kmalloc_array(num_fcntl_locks + num_flock_locks,
3592 					       sizeof(struct ceph_filelock),
3593 					       GFP_NOFS);
3594 			if (!flocks) {
3595 				err = -ENOMEM;
3596 				goto out_err;
3597 			}
3598 			err = ceph_encode_locks_to_buffer(inode, flocks,
3599 							  num_fcntl_locks,
3600 							  num_flock_locks);
3601 			if (err) {
3602 				kfree(flocks);
3603 				flocks = NULL;
3604 				if (err == -ENOSPC)
3605 					goto encode_again;
3606 				goto out_err;
3607 			}
3608 		} else {
3609 			kfree(flocks);
3610 			flocks = NULL;
3611 		}
3612 
3613 		if (recon_state->msg_version >= 3) {
3614 			/* version, compat_version and struct_len */
3615 			total_len += 2 * sizeof(u8) + sizeof(u32);
3616 			struct_v = 2;
3617 		}
3618 		/*
3619 		 * number of encoded locks is stable, so copy to pagelist
3620 		 */
3621 		struct_len = 2 * sizeof(u32) +
3622 			    (num_fcntl_locks + num_flock_locks) *
3623 			    sizeof(struct ceph_filelock);
3624 		rec.v2.flock_len = cpu_to_le32(struct_len);
3625 
3626 		struct_len += sizeof(u32) + sizeof(rec.v2);
3627 
3628 		if (struct_v >= 2)
3629 			struct_len += sizeof(u64); /* snap_follows */
3630 
3631 		total_len += struct_len;
3632 
3633 		if (pagelist->length + total_len > RECONNECT_MAX_SIZE) {
3634 			err = send_reconnect_partial(recon_state);
3635 			if (err)
3636 				goto out_freeflocks;
3637 			pagelist = recon_state->pagelist;
3638 		}
3639 
3640 		err = ceph_pagelist_reserve(pagelist, total_len);
3641 		if (err)
3642 			goto out_freeflocks;
3643 
3644 		ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3645 		if (recon_state->msg_version >= 3) {
3646 			ceph_pagelist_encode_8(pagelist, struct_v);
3647 			ceph_pagelist_encode_8(pagelist, 1);
3648 			ceph_pagelist_encode_32(pagelist, struct_len);
3649 		}
3650 		ceph_pagelist_encode_string(pagelist, NULL, 0);
3651 		ceph_pagelist_append(pagelist, &rec, sizeof(rec.v2));
3652 		ceph_locks_to_pagelist(flocks, pagelist,
3653 				       num_fcntl_locks, num_flock_locks);
3654 		if (struct_v >= 2)
3655 			ceph_pagelist_encode_64(pagelist, snap_follows);
3656 out_freeflocks:
3657 		kfree(flocks);
3658 	} else {
3659 		u64 pathbase = 0;
3660 		int pathlen = 0;
3661 		char *path = NULL;
3662 		struct dentry *dentry;
3663 
3664 		dentry = d_find_alias(inode);
3665 		if (dentry) {
3666 			path = ceph_mdsc_build_path(dentry,
3667 						&pathlen, &pathbase, 0);
3668 			dput(dentry);
3669 			if (IS_ERR(path)) {
3670 				err = PTR_ERR(path);
3671 				goto out_err;
3672 			}
3673 			rec.v1.pathbase = cpu_to_le64(pathbase);
3674 		}
3675 
3676 		err = ceph_pagelist_reserve(pagelist,
3677 					    sizeof(u64) + sizeof(u32) +
3678 					    pathlen + sizeof(rec.v1));
3679 		if (err) {
3680 			goto out_freepath;
3681 		}
3682 
3683 		ceph_pagelist_encode_64(pagelist, ceph_ino(inode));
3684 		ceph_pagelist_encode_string(pagelist, path, pathlen);
3685 		ceph_pagelist_append(pagelist, &rec, sizeof(rec.v1));
3686 out_freepath:
3687 		ceph_mdsc_free_path(path, pathlen);
3688 	}
3689 
3690 out_err:
3691 	if (err >= 0)
3692 		recon_state->nr_caps++;
3693 	return err;
3694 }
3695 
3696 static int encode_snap_realms(struct ceph_mds_client *mdsc,
3697 			      struct ceph_reconnect_state *recon_state)
3698 {
3699 	struct rb_node *p;
3700 	struct ceph_pagelist *pagelist = recon_state->pagelist;
3701 	int err = 0;
3702 
3703 	if (recon_state->msg_version >= 4) {
3704 		err = ceph_pagelist_encode_32(pagelist, mdsc->num_snap_realms);
3705 		if (err < 0)
3706 			goto fail;
3707 	}
3708 
3709 	/*
3710 	 * snaprealms.  we provide mds with the ino, seq (version), and
3711 	 * parent for all of our realms.  If the mds has any newer info,
3712 	 * it will tell us.
3713 	 */
3714 	for (p = rb_first(&mdsc->snap_realms); p; p = rb_next(p)) {
3715 		struct ceph_snap_realm *realm =
3716 		       rb_entry(p, struct ceph_snap_realm, node);
3717 		struct ceph_mds_snaprealm_reconnect sr_rec;
3718 
3719 		if (recon_state->msg_version >= 4) {
3720 			size_t need = sizeof(u8) * 2 + sizeof(u32) +
3721 				      sizeof(sr_rec);
3722 
3723 			if (pagelist->length + need > RECONNECT_MAX_SIZE) {
3724 				err = send_reconnect_partial(recon_state);
3725 				if (err)
3726 					goto fail;
3727 				pagelist = recon_state->pagelist;
3728 			}
3729 
3730 			err = ceph_pagelist_reserve(pagelist, need);
3731 			if (err)
3732 				goto fail;
3733 
3734 			ceph_pagelist_encode_8(pagelist, 1);
3735 			ceph_pagelist_encode_8(pagelist, 1);
3736 			ceph_pagelist_encode_32(pagelist, sizeof(sr_rec));
3737 		}
3738 
3739 		dout(" adding snap realm %llx seq %lld parent %llx\n",
3740 		     realm->ino, realm->seq, realm->parent_ino);
3741 		sr_rec.ino = cpu_to_le64(realm->ino);
3742 		sr_rec.seq = cpu_to_le64(realm->seq);
3743 		sr_rec.parent = cpu_to_le64(realm->parent_ino);
3744 
3745 		err = ceph_pagelist_append(pagelist, &sr_rec, sizeof(sr_rec));
3746 		if (err)
3747 			goto fail;
3748 
3749 		recon_state->nr_realms++;
3750 	}
3751 fail:
3752 	return err;
3753 }
3754 
3755 
3756 /*
3757  * If an MDS fails and recovers, clients need to reconnect in order to
3758  * reestablish shared state.  This includes all caps issued through
3759  * this session _and_ the snap_realm hierarchy.  Because it's not
3760  * clear which snap realms the mds cares about, we send everything we
3761  * know about.. that ensures we'll then get any new info the
3762  * recovering MDS might have.
3763  *
3764  * This is a relatively heavyweight operation, but it's rare.
3765  *
3766  * called with mdsc->mutex held.
3767  */
3768 static void send_mds_reconnect(struct ceph_mds_client *mdsc,
3769 			       struct ceph_mds_session *session)
3770 {
3771 	struct ceph_msg *reply;
3772 	int mds = session->s_mds;
3773 	int err = -ENOMEM;
3774 	struct ceph_reconnect_state recon_state = {
3775 		.session = session,
3776 	};
3777 	LIST_HEAD(dispose);
3778 
3779 	pr_info("mds%d reconnect start\n", mds);
3780 
3781 	recon_state.pagelist = ceph_pagelist_alloc(GFP_NOFS);
3782 	if (!recon_state.pagelist)
3783 		goto fail_nopagelist;
3784 
3785 	reply = ceph_msg_new2(CEPH_MSG_CLIENT_RECONNECT, 0, 1, GFP_NOFS, false);
3786 	if (!reply)
3787 		goto fail_nomsg;
3788 
3789 	xa_destroy(&session->s_delegated_inos);
3790 
3791 	mutex_lock(&session->s_mutex);
3792 	session->s_state = CEPH_MDS_SESSION_RECONNECTING;
3793 	session->s_seq = 0;
3794 
3795 	dout("session %p state %s\n", session,
3796 	     ceph_session_state_name(session->s_state));
3797 
3798 	spin_lock(&session->s_gen_ttl_lock);
3799 	session->s_cap_gen++;
3800 	spin_unlock(&session->s_gen_ttl_lock);
3801 
3802 	spin_lock(&session->s_cap_lock);
3803 	/* don't know if session is readonly */
3804 	session->s_readonly = 0;
3805 	/*
3806 	 * notify __ceph_remove_cap() that we are composing cap reconnect.
3807 	 * If a cap get released before being added to the cap reconnect,
3808 	 * __ceph_remove_cap() should skip queuing cap release.
3809 	 */
3810 	session->s_cap_reconnect = 1;
3811 	/* drop old cap expires; we're about to reestablish that state */
3812 	detach_cap_releases(session, &dispose);
3813 	spin_unlock(&session->s_cap_lock);
3814 	dispose_cap_releases(mdsc, &dispose);
3815 
3816 	/* trim unused caps to reduce MDS's cache rejoin time */
3817 	if (mdsc->fsc->sb->s_root)
3818 		shrink_dcache_parent(mdsc->fsc->sb->s_root);
3819 
3820 	ceph_con_close(&session->s_con);
3821 	ceph_con_open(&session->s_con,
3822 		      CEPH_ENTITY_TYPE_MDS, mds,
3823 		      ceph_mdsmap_get_addr(mdsc->mdsmap, mds));
3824 
3825 	/* replay unsafe requests */
3826 	replay_unsafe_requests(mdsc, session);
3827 
3828 	ceph_early_kick_flushing_caps(mdsc, session);
3829 
3830 	down_read(&mdsc->snap_rwsem);
3831 
3832 	/* placeholder for nr_caps */
3833 	err = ceph_pagelist_encode_32(recon_state.pagelist, 0);
3834 	if (err)
3835 		goto fail;
3836 
3837 	if (test_bit(CEPHFS_FEATURE_MULTI_RECONNECT, &session->s_features)) {
3838 		recon_state.msg_version = 3;
3839 		recon_state.allow_multi = true;
3840 	} else if (session->s_con.peer_features & CEPH_FEATURE_MDSENC) {
3841 		recon_state.msg_version = 3;
3842 	} else {
3843 		recon_state.msg_version = 2;
3844 	}
3845 	/* trsaverse this session's caps */
3846 	err = ceph_iterate_session_caps(session, reconnect_caps_cb, &recon_state);
3847 
3848 	spin_lock(&session->s_cap_lock);
3849 	session->s_cap_reconnect = 0;
3850 	spin_unlock(&session->s_cap_lock);
3851 
3852 	if (err < 0)
3853 		goto fail;
3854 
3855 	/* check if all realms can be encoded into current message */
3856 	if (mdsc->num_snap_realms) {
3857 		size_t total_len =
3858 			recon_state.pagelist->length +
3859 			mdsc->num_snap_realms *
3860 			sizeof(struct ceph_mds_snaprealm_reconnect);
3861 		if (recon_state.msg_version >= 4) {
3862 			/* number of realms */
3863 			total_len += sizeof(u32);
3864 			/* version, compat_version and struct_len */
3865 			total_len += mdsc->num_snap_realms *
3866 				     (2 * sizeof(u8) + sizeof(u32));
3867 		}
3868 		if (total_len > RECONNECT_MAX_SIZE) {
3869 			if (!recon_state.allow_multi) {
3870 				err = -ENOSPC;
3871 				goto fail;
3872 			}
3873 			if (recon_state.nr_caps) {
3874 				err = send_reconnect_partial(&recon_state);
3875 				if (err)
3876 					goto fail;
3877 			}
3878 			recon_state.msg_version = 5;
3879 		}
3880 	}
3881 
3882 	err = encode_snap_realms(mdsc, &recon_state);
3883 	if (err < 0)
3884 		goto fail;
3885 
3886 	if (recon_state.msg_version >= 5) {
3887 		err = ceph_pagelist_encode_8(recon_state.pagelist, 0);
3888 		if (err < 0)
3889 			goto fail;
3890 	}
3891 
3892 	if (recon_state.nr_caps || recon_state.nr_realms) {
3893 		struct page *page =
3894 			list_first_entry(&recon_state.pagelist->head,
3895 					struct page, lru);
3896 		__le32 *addr = kmap_atomic(page);
3897 		if (recon_state.nr_caps) {
3898 			WARN_ON(recon_state.nr_realms != mdsc->num_snap_realms);
3899 			*addr = cpu_to_le32(recon_state.nr_caps);
3900 		} else if (recon_state.msg_version >= 4) {
3901 			*(addr + 1) = cpu_to_le32(recon_state.nr_realms);
3902 		}
3903 		kunmap_atomic(addr);
3904 	}
3905 
3906 	reply->hdr.version = cpu_to_le16(recon_state.msg_version);
3907 	if (recon_state.msg_version >= 4)
3908 		reply->hdr.compat_version = cpu_to_le16(4);
3909 
3910 	reply->hdr.data_len = cpu_to_le32(recon_state.pagelist->length);
3911 	ceph_msg_data_add_pagelist(reply, recon_state.pagelist);
3912 
3913 	ceph_con_send(&session->s_con, reply);
3914 
3915 	mutex_unlock(&session->s_mutex);
3916 
3917 	mutex_lock(&mdsc->mutex);
3918 	__wake_requests(mdsc, &session->s_waiting);
3919 	mutex_unlock(&mdsc->mutex);
3920 
3921 	up_read(&mdsc->snap_rwsem);
3922 	ceph_pagelist_release(recon_state.pagelist);
3923 	return;
3924 
3925 fail:
3926 	ceph_msg_put(reply);
3927 	up_read(&mdsc->snap_rwsem);
3928 	mutex_unlock(&session->s_mutex);
3929 fail_nomsg:
3930 	ceph_pagelist_release(recon_state.pagelist);
3931 fail_nopagelist:
3932 	pr_err("error %d preparing reconnect for mds%d\n", err, mds);
3933 	return;
3934 }
3935 
3936 
3937 /*
3938  * compare old and new mdsmaps, kicking requests
3939  * and closing out old connections as necessary
3940  *
3941  * called under mdsc->mutex.
3942  */
3943 static void check_new_map(struct ceph_mds_client *mdsc,
3944 			  struct ceph_mdsmap *newmap,
3945 			  struct ceph_mdsmap *oldmap)
3946 {
3947 	int i;
3948 	int oldstate, newstate;
3949 	struct ceph_mds_session *s;
3950 
3951 	dout("check_new_map new %u old %u\n",
3952 	     newmap->m_epoch, oldmap->m_epoch);
3953 
3954 	for (i = 0; i < oldmap->possible_max_rank && i < mdsc->max_sessions; i++) {
3955 		if (!mdsc->sessions[i])
3956 			continue;
3957 		s = mdsc->sessions[i];
3958 		oldstate = ceph_mdsmap_get_state(oldmap, i);
3959 		newstate = ceph_mdsmap_get_state(newmap, i);
3960 
3961 		dout("check_new_map mds%d state %s%s -> %s%s (session %s)\n",
3962 		     i, ceph_mds_state_name(oldstate),
3963 		     ceph_mdsmap_is_laggy(oldmap, i) ? " (laggy)" : "",
3964 		     ceph_mds_state_name(newstate),
3965 		     ceph_mdsmap_is_laggy(newmap, i) ? " (laggy)" : "",
3966 		     ceph_session_state_name(s->s_state));
3967 
3968 		if (i >= newmap->possible_max_rank) {
3969 			/* force close session for stopped mds */
3970 			ceph_get_mds_session(s);
3971 			__unregister_session(mdsc, s);
3972 			__wake_requests(mdsc, &s->s_waiting);
3973 			mutex_unlock(&mdsc->mutex);
3974 
3975 			mutex_lock(&s->s_mutex);
3976 			cleanup_session_requests(mdsc, s);
3977 			remove_session_caps(s);
3978 			mutex_unlock(&s->s_mutex);
3979 
3980 			ceph_put_mds_session(s);
3981 
3982 			mutex_lock(&mdsc->mutex);
3983 			kick_requests(mdsc, i);
3984 			continue;
3985 		}
3986 
3987 		if (memcmp(ceph_mdsmap_get_addr(oldmap, i),
3988 			   ceph_mdsmap_get_addr(newmap, i),
3989 			   sizeof(struct ceph_entity_addr))) {
3990 			/* just close it */
3991 			mutex_unlock(&mdsc->mutex);
3992 			mutex_lock(&s->s_mutex);
3993 			mutex_lock(&mdsc->mutex);
3994 			ceph_con_close(&s->s_con);
3995 			mutex_unlock(&s->s_mutex);
3996 			s->s_state = CEPH_MDS_SESSION_RESTARTING;
3997 		} else if (oldstate == newstate) {
3998 			continue;  /* nothing new with this mds */
3999 		}
4000 
4001 		/*
4002 		 * send reconnect?
4003 		 */
4004 		if (s->s_state == CEPH_MDS_SESSION_RESTARTING &&
4005 		    newstate >= CEPH_MDS_STATE_RECONNECT) {
4006 			mutex_unlock(&mdsc->mutex);
4007 			send_mds_reconnect(mdsc, s);
4008 			mutex_lock(&mdsc->mutex);
4009 		}
4010 
4011 		/*
4012 		 * kick request on any mds that has gone active.
4013 		 */
4014 		if (oldstate < CEPH_MDS_STATE_ACTIVE &&
4015 		    newstate >= CEPH_MDS_STATE_ACTIVE) {
4016 			if (oldstate != CEPH_MDS_STATE_CREATING &&
4017 			    oldstate != CEPH_MDS_STATE_STARTING)
4018 				pr_info("mds%d recovery completed\n", s->s_mds);
4019 			kick_requests(mdsc, i);
4020 			ceph_kick_flushing_caps(mdsc, s);
4021 			wake_up_session_caps(s, RECONNECT);
4022 		}
4023 	}
4024 
4025 	for (i = 0; i < newmap->possible_max_rank && i < mdsc->max_sessions; i++) {
4026 		s = mdsc->sessions[i];
4027 		if (!s)
4028 			continue;
4029 		if (!ceph_mdsmap_is_laggy(newmap, i))
4030 			continue;
4031 		if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4032 		    s->s_state == CEPH_MDS_SESSION_HUNG ||
4033 		    s->s_state == CEPH_MDS_SESSION_CLOSING) {
4034 			dout(" connecting to export targets of laggy mds%d\n",
4035 			     i);
4036 			__open_export_target_sessions(mdsc, s);
4037 		}
4038 	}
4039 }
4040 
4041 
4042 
4043 /*
4044  * leases
4045  */
4046 
4047 /*
4048  * caller must hold session s_mutex, dentry->d_lock
4049  */
4050 void __ceph_mdsc_drop_dentry_lease(struct dentry *dentry)
4051 {
4052 	struct ceph_dentry_info *di = ceph_dentry(dentry);
4053 
4054 	ceph_put_mds_session(di->lease_session);
4055 	di->lease_session = NULL;
4056 }
4057 
4058 static void handle_lease(struct ceph_mds_client *mdsc,
4059 			 struct ceph_mds_session *session,
4060 			 struct ceph_msg *msg)
4061 {
4062 	struct super_block *sb = mdsc->fsc->sb;
4063 	struct inode *inode;
4064 	struct dentry *parent, *dentry;
4065 	struct ceph_dentry_info *di;
4066 	int mds = session->s_mds;
4067 	struct ceph_mds_lease *h = msg->front.iov_base;
4068 	u32 seq;
4069 	struct ceph_vino vino;
4070 	struct qstr dname;
4071 	int release = 0;
4072 
4073 	dout("handle_lease from mds%d\n", mds);
4074 
4075 	/* decode */
4076 	if (msg->front.iov_len < sizeof(*h) + sizeof(u32))
4077 		goto bad;
4078 	vino.ino = le64_to_cpu(h->ino);
4079 	vino.snap = CEPH_NOSNAP;
4080 	seq = le32_to_cpu(h->seq);
4081 	dname.len = get_unaligned_le32(h + 1);
4082 	if (msg->front.iov_len < sizeof(*h) + sizeof(u32) + dname.len)
4083 		goto bad;
4084 	dname.name = (void *)(h + 1) + sizeof(u32);
4085 
4086 	/* lookup inode */
4087 	inode = ceph_find_inode(sb, vino);
4088 	dout("handle_lease %s, ino %llx %p %.*s\n",
4089 	     ceph_lease_op_name(h->action), vino.ino, inode,
4090 	     dname.len, dname.name);
4091 
4092 	mutex_lock(&session->s_mutex);
4093 	session->s_seq++;
4094 
4095 	if (!inode) {
4096 		dout("handle_lease no inode %llx\n", vino.ino);
4097 		goto release;
4098 	}
4099 
4100 	/* dentry */
4101 	parent = d_find_alias(inode);
4102 	if (!parent) {
4103 		dout("no parent dentry on inode %p\n", inode);
4104 		WARN_ON(1);
4105 		goto release;  /* hrm... */
4106 	}
4107 	dname.hash = full_name_hash(parent, dname.name, dname.len);
4108 	dentry = d_lookup(parent, &dname);
4109 	dput(parent);
4110 	if (!dentry)
4111 		goto release;
4112 
4113 	spin_lock(&dentry->d_lock);
4114 	di = ceph_dentry(dentry);
4115 	switch (h->action) {
4116 	case CEPH_MDS_LEASE_REVOKE:
4117 		if (di->lease_session == session) {
4118 			if (ceph_seq_cmp(di->lease_seq, seq) > 0)
4119 				h->seq = cpu_to_le32(di->lease_seq);
4120 			__ceph_mdsc_drop_dentry_lease(dentry);
4121 		}
4122 		release = 1;
4123 		break;
4124 
4125 	case CEPH_MDS_LEASE_RENEW:
4126 		if (di->lease_session == session &&
4127 		    di->lease_gen == session->s_cap_gen &&
4128 		    di->lease_renew_from &&
4129 		    di->lease_renew_after == 0) {
4130 			unsigned long duration =
4131 				msecs_to_jiffies(le32_to_cpu(h->duration_ms));
4132 
4133 			di->lease_seq = seq;
4134 			di->time = di->lease_renew_from + duration;
4135 			di->lease_renew_after = di->lease_renew_from +
4136 				(duration >> 1);
4137 			di->lease_renew_from = 0;
4138 		}
4139 		break;
4140 	}
4141 	spin_unlock(&dentry->d_lock);
4142 	dput(dentry);
4143 
4144 	if (!release)
4145 		goto out;
4146 
4147 release:
4148 	/* let's just reuse the same message */
4149 	h->action = CEPH_MDS_LEASE_REVOKE_ACK;
4150 	ceph_msg_get(msg);
4151 	ceph_con_send(&session->s_con, msg);
4152 
4153 out:
4154 	mutex_unlock(&session->s_mutex);
4155 	/* avoid calling iput_final() in mds dispatch threads */
4156 	ceph_async_iput(inode);
4157 	return;
4158 
4159 bad:
4160 	pr_err("corrupt lease message\n");
4161 	ceph_msg_dump(msg);
4162 }
4163 
4164 void ceph_mdsc_lease_send_msg(struct ceph_mds_session *session,
4165 			      struct dentry *dentry, char action,
4166 			      u32 seq)
4167 {
4168 	struct ceph_msg *msg;
4169 	struct ceph_mds_lease *lease;
4170 	struct inode *dir;
4171 	int len = sizeof(*lease) + sizeof(u32) + NAME_MAX;
4172 
4173 	dout("lease_send_msg identry %p %s to mds%d\n",
4174 	     dentry, ceph_lease_op_name(action), session->s_mds);
4175 
4176 	msg = ceph_msg_new(CEPH_MSG_CLIENT_LEASE, len, GFP_NOFS, false);
4177 	if (!msg)
4178 		return;
4179 	lease = msg->front.iov_base;
4180 	lease->action = action;
4181 	lease->seq = cpu_to_le32(seq);
4182 
4183 	spin_lock(&dentry->d_lock);
4184 	dir = d_inode(dentry->d_parent);
4185 	lease->ino = cpu_to_le64(ceph_ino(dir));
4186 	lease->first = lease->last = cpu_to_le64(ceph_snap(dir));
4187 
4188 	put_unaligned_le32(dentry->d_name.len, lease + 1);
4189 	memcpy((void *)(lease + 1) + 4,
4190 	       dentry->d_name.name, dentry->d_name.len);
4191 	spin_unlock(&dentry->d_lock);
4192 	/*
4193 	 * if this is a preemptive lease RELEASE, no need to
4194 	 * flush request stream, since the actual request will
4195 	 * soon follow.
4196 	 */
4197 	msg->more_to_follow = (action == CEPH_MDS_LEASE_RELEASE);
4198 
4199 	ceph_con_send(&session->s_con, msg);
4200 }
4201 
4202 /*
4203  * lock unlock sessions, to wait ongoing session activities
4204  */
4205 static void lock_unlock_sessions(struct ceph_mds_client *mdsc)
4206 {
4207 	int i;
4208 
4209 	mutex_lock(&mdsc->mutex);
4210 	for (i = 0; i < mdsc->max_sessions; i++) {
4211 		struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4212 		if (!s)
4213 			continue;
4214 		mutex_unlock(&mdsc->mutex);
4215 		mutex_lock(&s->s_mutex);
4216 		mutex_unlock(&s->s_mutex);
4217 		ceph_put_mds_session(s);
4218 		mutex_lock(&mdsc->mutex);
4219 	}
4220 	mutex_unlock(&mdsc->mutex);
4221 }
4222 
4223 static void maybe_recover_session(struct ceph_mds_client *mdsc)
4224 {
4225 	struct ceph_fs_client *fsc = mdsc->fsc;
4226 
4227 	if (!ceph_test_mount_opt(fsc, CLEANRECOVER))
4228 		return;
4229 
4230 	if (READ_ONCE(fsc->mount_state) != CEPH_MOUNT_MOUNTED)
4231 		return;
4232 
4233 	if (!READ_ONCE(fsc->blacklisted))
4234 		return;
4235 
4236 	if (fsc->last_auto_reconnect &&
4237 	    time_before(jiffies, fsc->last_auto_reconnect + HZ * 60 * 30))
4238 		return;
4239 
4240 	pr_info("auto reconnect after blacklisted\n");
4241 	fsc->last_auto_reconnect = jiffies;
4242 	ceph_force_reconnect(fsc->sb);
4243 }
4244 
4245 /*
4246  * delayed work -- periodically trim expired leases, renew caps with mds
4247  */
4248 static void schedule_delayed(struct ceph_mds_client *mdsc)
4249 {
4250 	int delay = 5;
4251 	unsigned hz = round_jiffies_relative(HZ * delay);
4252 	schedule_delayed_work(&mdsc->delayed_work, hz);
4253 }
4254 
4255 static void delayed_work(struct work_struct *work)
4256 {
4257 	int i;
4258 	struct ceph_mds_client *mdsc =
4259 		container_of(work, struct ceph_mds_client, delayed_work.work);
4260 	int renew_interval;
4261 	int renew_caps;
4262 
4263 	dout("mdsc delayed_work\n");
4264 
4265 	mutex_lock(&mdsc->mutex);
4266 	renew_interval = mdsc->mdsmap->m_session_timeout >> 2;
4267 	renew_caps = time_after_eq(jiffies, HZ*renew_interval +
4268 				   mdsc->last_renew_caps);
4269 	if (renew_caps)
4270 		mdsc->last_renew_caps = jiffies;
4271 
4272 	for (i = 0; i < mdsc->max_sessions; i++) {
4273 		struct ceph_mds_session *s = __ceph_lookup_mds_session(mdsc, i);
4274 		if (!s)
4275 			continue;
4276 		if (s->s_state == CEPH_MDS_SESSION_CLOSING) {
4277 			dout("resending session close request for mds%d\n",
4278 			     s->s_mds);
4279 			request_close_session(mdsc, s);
4280 			ceph_put_mds_session(s);
4281 			continue;
4282 		}
4283 		if (s->s_ttl && time_after(jiffies, s->s_ttl)) {
4284 			if (s->s_state == CEPH_MDS_SESSION_OPEN) {
4285 				s->s_state = CEPH_MDS_SESSION_HUNG;
4286 				pr_info("mds%d hung\n", s->s_mds);
4287 			}
4288 		}
4289 		if (s->s_state == CEPH_MDS_SESSION_NEW ||
4290 		    s->s_state == CEPH_MDS_SESSION_RESTARTING ||
4291 		    s->s_state == CEPH_MDS_SESSION_REJECTED) {
4292 			/* this mds is failed or recovering, just wait */
4293 			ceph_put_mds_session(s);
4294 			continue;
4295 		}
4296 		mutex_unlock(&mdsc->mutex);
4297 
4298 		mutex_lock(&s->s_mutex);
4299 		if (renew_caps)
4300 			send_renew_caps(mdsc, s);
4301 		else
4302 			ceph_con_keepalive(&s->s_con);
4303 		if (s->s_state == CEPH_MDS_SESSION_OPEN ||
4304 		    s->s_state == CEPH_MDS_SESSION_HUNG)
4305 			ceph_send_cap_releases(mdsc, s);
4306 		mutex_unlock(&s->s_mutex);
4307 		ceph_put_mds_session(s);
4308 
4309 		mutex_lock(&mdsc->mutex);
4310 	}
4311 	mutex_unlock(&mdsc->mutex);
4312 
4313 	ceph_check_delayed_caps(mdsc);
4314 
4315 	ceph_queue_cap_reclaim_work(mdsc);
4316 
4317 	ceph_trim_snapid_map(mdsc);
4318 
4319 	maybe_recover_session(mdsc);
4320 
4321 	schedule_delayed(mdsc);
4322 }
4323 
4324 int ceph_mdsc_init(struct ceph_fs_client *fsc)
4325 
4326 {
4327 	struct ceph_mds_client *mdsc;
4328 
4329 	mdsc = kzalloc(sizeof(struct ceph_mds_client), GFP_NOFS);
4330 	if (!mdsc)
4331 		return -ENOMEM;
4332 	mdsc->fsc = fsc;
4333 	mutex_init(&mdsc->mutex);
4334 	mdsc->mdsmap = kzalloc(sizeof(*mdsc->mdsmap), GFP_NOFS);
4335 	if (!mdsc->mdsmap) {
4336 		kfree(mdsc);
4337 		return -ENOMEM;
4338 	}
4339 
4340 	fsc->mdsc = mdsc;
4341 	init_completion(&mdsc->safe_umount_waiters);
4342 	init_waitqueue_head(&mdsc->session_close_wq);
4343 	INIT_LIST_HEAD(&mdsc->waiting_for_map);
4344 	mdsc->sessions = NULL;
4345 	atomic_set(&mdsc->num_sessions, 0);
4346 	mdsc->max_sessions = 0;
4347 	mdsc->stopping = 0;
4348 	atomic64_set(&mdsc->quotarealms_count, 0);
4349 	mdsc->quotarealms_inodes = RB_ROOT;
4350 	mutex_init(&mdsc->quotarealms_inodes_mutex);
4351 	mdsc->last_snap_seq = 0;
4352 	init_rwsem(&mdsc->snap_rwsem);
4353 	mdsc->snap_realms = RB_ROOT;
4354 	INIT_LIST_HEAD(&mdsc->snap_empty);
4355 	mdsc->num_snap_realms = 0;
4356 	spin_lock_init(&mdsc->snap_empty_lock);
4357 	mdsc->last_tid = 0;
4358 	mdsc->oldest_tid = 0;
4359 	mdsc->request_tree = RB_ROOT;
4360 	INIT_DELAYED_WORK(&mdsc->delayed_work, delayed_work);
4361 	mdsc->last_renew_caps = jiffies;
4362 	INIT_LIST_HEAD(&mdsc->cap_delay_list);
4363 	INIT_LIST_HEAD(&mdsc->cap_wait_list);
4364 	spin_lock_init(&mdsc->cap_delay_lock);
4365 	INIT_LIST_HEAD(&mdsc->snap_flush_list);
4366 	spin_lock_init(&mdsc->snap_flush_lock);
4367 	mdsc->last_cap_flush_tid = 1;
4368 	INIT_LIST_HEAD(&mdsc->cap_flush_list);
4369 	INIT_LIST_HEAD(&mdsc->cap_dirty);
4370 	INIT_LIST_HEAD(&mdsc->cap_dirty_migrating);
4371 	mdsc->num_cap_flushing = 0;
4372 	spin_lock_init(&mdsc->cap_dirty_lock);
4373 	init_waitqueue_head(&mdsc->cap_flushing_wq);
4374 	INIT_WORK(&mdsc->cap_reclaim_work, ceph_cap_reclaim_work);
4375 	atomic_set(&mdsc->cap_reclaim_pending, 0);
4376 
4377 	spin_lock_init(&mdsc->dentry_list_lock);
4378 	INIT_LIST_HEAD(&mdsc->dentry_leases);
4379 	INIT_LIST_HEAD(&mdsc->dentry_dir_leases);
4380 
4381 	ceph_caps_init(mdsc);
4382 	ceph_adjust_caps_max_min(mdsc, fsc->mount_options);
4383 
4384 	spin_lock_init(&mdsc->snapid_map_lock);
4385 	mdsc->snapid_map_tree = RB_ROOT;
4386 	INIT_LIST_HEAD(&mdsc->snapid_map_lru);
4387 
4388 	init_rwsem(&mdsc->pool_perm_rwsem);
4389 	mdsc->pool_perm_tree = RB_ROOT;
4390 
4391 	strscpy(mdsc->nodename, utsname()->nodename,
4392 		sizeof(mdsc->nodename));
4393 	return 0;
4394 }
4395 
4396 /*
4397  * Wait for safe replies on open mds requests.  If we time out, drop
4398  * all requests from the tree to avoid dangling dentry refs.
4399  */
4400 static void wait_requests(struct ceph_mds_client *mdsc)
4401 {
4402 	struct ceph_options *opts = mdsc->fsc->client->options;
4403 	struct ceph_mds_request *req;
4404 
4405 	mutex_lock(&mdsc->mutex);
4406 	if (__get_oldest_req(mdsc)) {
4407 		mutex_unlock(&mdsc->mutex);
4408 
4409 		dout("wait_requests waiting for requests\n");
4410 		wait_for_completion_timeout(&mdsc->safe_umount_waiters,
4411 				    ceph_timeout_jiffies(opts->mount_timeout));
4412 
4413 		/* tear down remaining requests */
4414 		mutex_lock(&mdsc->mutex);
4415 		while ((req = __get_oldest_req(mdsc))) {
4416 			dout("wait_requests timed out on tid %llu\n",
4417 			     req->r_tid);
4418 			list_del_init(&req->r_wait);
4419 			__unregister_request(mdsc, req);
4420 		}
4421 	}
4422 	mutex_unlock(&mdsc->mutex);
4423 	dout("wait_requests done\n");
4424 }
4425 
4426 /*
4427  * called before mount is ro, and before dentries are torn down.
4428  * (hmm, does this still race with new lookups?)
4429  */
4430 void ceph_mdsc_pre_umount(struct ceph_mds_client *mdsc)
4431 {
4432 	dout("pre_umount\n");
4433 	mdsc->stopping = 1;
4434 
4435 	lock_unlock_sessions(mdsc);
4436 	ceph_flush_dirty_caps(mdsc);
4437 	wait_requests(mdsc);
4438 
4439 	/*
4440 	 * wait for reply handlers to drop their request refs and
4441 	 * their inode/dcache refs
4442 	 */
4443 	ceph_msgr_flush();
4444 
4445 	ceph_cleanup_quotarealms_inodes(mdsc);
4446 }
4447 
4448 /*
4449  * wait for all write mds requests to flush.
4450  */
4451 static void wait_unsafe_requests(struct ceph_mds_client *mdsc, u64 want_tid)
4452 {
4453 	struct ceph_mds_request *req = NULL, *nextreq;
4454 	struct rb_node *n;
4455 
4456 	mutex_lock(&mdsc->mutex);
4457 	dout("wait_unsafe_requests want %lld\n", want_tid);
4458 restart:
4459 	req = __get_oldest_req(mdsc);
4460 	while (req && req->r_tid <= want_tid) {
4461 		/* find next request */
4462 		n = rb_next(&req->r_node);
4463 		if (n)
4464 			nextreq = rb_entry(n, struct ceph_mds_request, r_node);
4465 		else
4466 			nextreq = NULL;
4467 		if (req->r_op != CEPH_MDS_OP_SETFILELOCK &&
4468 		    (req->r_op & CEPH_MDS_OP_WRITE)) {
4469 			/* write op */
4470 			ceph_mdsc_get_request(req);
4471 			if (nextreq)
4472 				ceph_mdsc_get_request(nextreq);
4473 			mutex_unlock(&mdsc->mutex);
4474 			dout("wait_unsafe_requests  wait on %llu (want %llu)\n",
4475 			     req->r_tid, want_tid);
4476 			wait_for_completion(&req->r_safe_completion);
4477 			mutex_lock(&mdsc->mutex);
4478 			ceph_mdsc_put_request(req);
4479 			if (!nextreq)
4480 				break;  /* next dne before, so we're done! */
4481 			if (RB_EMPTY_NODE(&nextreq->r_node)) {
4482 				/* next request was removed from tree */
4483 				ceph_mdsc_put_request(nextreq);
4484 				goto restart;
4485 			}
4486 			ceph_mdsc_put_request(nextreq);  /* won't go away */
4487 		}
4488 		req = nextreq;
4489 	}
4490 	mutex_unlock(&mdsc->mutex);
4491 	dout("wait_unsafe_requests done\n");
4492 }
4493 
4494 void ceph_mdsc_sync(struct ceph_mds_client *mdsc)
4495 {
4496 	u64 want_tid, want_flush;
4497 
4498 	if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4499 		return;
4500 
4501 	dout("sync\n");
4502 	mutex_lock(&mdsc->mutex);
4503 	want_tid = mdsc->last_tid;
4504 	mutex_unlock(&mdsc->mutex);
4505 
4506 	ceph_flush_dirty_caps(mdsc);
4507 	spin_lock(&mdsc->cap_dirty_lock);
4508 	want_flush = mdsc->last_cap_flush_tid;
4509 	if (!list_empty(&mdsc->cap_flush_list)) {
4510 		struct ceph_cap_flush *cf =
4511 			list_last_entry(&mdsc->cap_flush_list,
4512 					struct ceph_cap_flush, g_list);
4513 		cf->wake = true;
4514 	}
4515 	spin_unlock(&mdsc->cap_dirty_lock);
4516 
4517 	dout("sync want tid %lld flush_seq %lld\n",
4518 	     want_tid, want_flush);
4519 
4520 	wait_unsafe_requests(mdsc, want_tid);
4521 	wait_caps_flush(mdsc, want_flush);
4522 }
4523 
4524 /*
4525  * true if all sessions are closed, or we force unmount
4526  */
4527 static bool done_closing_sessions(struct ceph_mds_client *mdsc, int skipped)
4528 {
4529 	if (READ_ONCE(mdsc->fsc->mount_state) == CEPH_MOUNT_SHUTDOWN)
4530 		return true;
4531 	return atomic_read(&mdsc->num_sessions) <= skipped;
4532 }
4533 
4534 /*
4535  * called after sb is ro.
4536  */
4537 void ceph_mdsc_close_sessions(struct ceph_mds_client *mdsc)
4538 {
4539 	struct ceph_options *opts = mdsc->fsc->client->options;
4540 	struct ceph_mds_session *session;
4541 	int i;
4542 	int skipped = 0;
4543 
4544 	dout("close_sessions\n");
4545 
4546 	/* close sessions */
4547 	mutex_lock(&mdsc->mutex);
4548 	for (i = 0; i < mdsc->max_sessions; i++) {
4549 		session = __ceph_lookup_mds_session(mdsc, i);
4550 		if (!session)
4551 			continue;
4552 		mutex_unlock(&mdsc->mutex);
4553 		mutex_lock(&session->s_mutex);
4554 		if (__close_session(mdsc, session) <= 0)
4555 			skipped++;
4556 		mutex_unlock(&session->s_mutex);
4557 		ceph_put_mds_session(session);
4558 		mutex_lock(&mdsc->mutex);
4559 	}
4560 	mutex_unlock(&mdsc->mutex);
4561 
4562 	dout("waiting for sessions to close\n");
4563 	wait_event_timeout(mdsc->session_close_wq,
4564 			   done_closing_sessions(mdsc, skipped),
4565 			   ceph_timeout_jiffies(opts->mount_timeout));
4566 
4567 	/* tear down remaining sessions */
4568 	mutex_lock(&mdsc->mutex);
4569 	for (i = 0; i < mdsc->max_sessions; i++) {
4570 		if (mdsc->sessions[i]) {
4571 			session = ceph_get_mds_session(mdsc->sessions[i]);
4572 			__unregister_session(mdsc, session);
4573 			mutex_unlock(&mdsc->mutex);
4574 			mutex_lock(&session->s_mutex);
4575 			remove_session_caps(session);
4576 			mutex_unlock(&session->s_mutex);
4577 			ceph_put_mds_session(session);
4578 			mutex_lock(&mdsc->mutex);
4579 		}
4580 	}
4581 	WARN_ON(!list_empty(&mdsc->cap_delay_list));
4582 	mutex_unlock(&mdsc->mutex);
4583 
4584 	ceph_cleanup_snapid_map(mdsc);
4585 	ceph_cleanup_empty_realms(mdsc);
4586 
4587 	cancel_work_sync(&mdsc->cap_reclaim_work);
4588 	cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4589 
4590 	dout("stopped\n");
4591 }
4592 
4593 void ceph_mdsc_force_umount(struct ceph_mds_client *mdsc)
4594 {
4595 	struct ceph_mds_session *session;
4596 	int mds;
4597 
4598 	dout("force umount\n");
4599 
4600 	mutex_lock(&mdsc->mutex);
4601 	for (mds = 0; mds < mdsc->max_sessions; mds++) {
4602 		session = __ceph_lookup_mds_session(mdsc, mds);
4603 		if (!session)
4604 			continue;
4605 
4606 		if (session->s_state == CEPH_MDS_SESSION_REJECTED)
4607 			__unregister_session(mdsc, session);
4608 		__wake_requests(mdsc, &session->s_waiting);
4609 		mutex_unlock(&mdsc->mutex);
4610 
4611 		mutex_lock(&session->s_mutex);
4612 		__close_session(mdsc, session);
4613 		if (session->s_state == CEPH_MDS_SESSION_CLOSING) {
4614 			cleanup_session_requests(mdsc, session);
4615 			remove_session_caps(session);
4616 		}
4617 		mutex_unlock(&session->s_mutex);
4618 		ceph_put_mds_session(session);
4619 
4620 		mutex_lock(&mdsc->mutex);
4621 		kick_requests(mdsc, mds);
4622 	}
4623 	__wake_requests(mdsc, &mdsc->waiting_for_map);
4624 	mutex_unlock(&mdsc->mutex);
4625 }
4626 
4627 static void ceph_mdsc_stop(struct ceph_mds_client *mdsc)
4628 {
4629 	dout("stop\n");
4630 	cancel_delayed_work_sync(&mdsc->delayed_work); /* cancel timer */
4631 	if (mdsc->mdsmap)
4632 		ceph_mdsmap_destroy(mdsc->mdsmap);
4633 	kfree(mdsc->sessions);
4634 	ceph_caps_finalize(mdsc);
4635 	ceph_pool_perm_destroy(mdsc);
4636 }
4637 
4638 void ceph_mdsc_destroy(struct ceph_fs_client *fsc)
4639 {
4640 	struct ceph_mds_client *mdsc = fsc->mdsc;
4641 	dout("mdsc_destroy %p\n", mdsc);
4642 
4643 	if (!mdsc)
4644 		return;
4645 
4646 	/* flush out any connection work with references to us */
4647 	ceph_msgr_flush();
4648 
4649 	ceph_mdsc_stop(mdsc);
4650 
4651 	fsc->mdsc = NULL;
4652 	kfree(mdsc);
4653 	dout("mdsc_destroy %p done\n", mdsc);
4654 }
4655 
4656 void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4657 {
4658 	struct ceph_fs_client *fsc = mdsc->fsc;
4659 	const char *mds_namespace = fsc->mount_options->mds_namespace;
4660 	void *p = msg->front.iov_base;
4661 	void *end = p + msg->front.iov_len;
4662 	u32 epoch;
4663 	u32 map_len;
4664 	u32 num_fs;
4665 	u32 mount_fscid = (u32)-1;
4666 	u8 struct_v, struct_cv;
4667 	int err = -EINVAL;
4668 
4669 	ceph_decode_need(&p, end, sizeof(u32), bad);
4670 	epoch = ceph_decode_32(&p);
4671 
4672 	dout("handle_fsmap epoch %u\n", epoch);
4673 
4674 	ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
4675 	struct_v = ceph_decode_8(&p);
4676 	struct_cv = ceph_decode_8(&p);
4677 	map_len = ceph_decode_32(&p);
4678 
4679 	ceph_decode_need(&p, end, sizeof(u32) * 3, bad);
4680 	p += sizeof(u32) * 2; /* skip epoch and legacy_client_fscid */
4681 
4682 	num_fs = ceph_decode_32(&p);
4683 	while (num_fs-- > 0) {
4684 		void *info_p, *info_end;
4685 		u32 info_len;
4686 		u8 info_v, info_cv;
4687 		u32 fscid, namelen;
4688 
4689 		ceph_decode_need(&p, end, 2 + sizeof(u32), bad);
4690 		info_v = ceph_decode_8(&p);
4691 		info_cv = ceph_decode_8(&p);
4692 		info_len = ceph_decode_32(&p);
4693 		ceph_decode_need(&p, end, info_len, bad);
4694 		info_p = p;
4695 		info_end = p + info_len;
4696 		p = info_end;
4697 
4698 		ceph_decode_need(&info_p, info_end, sizeof(u32) * 2, bad);
4699 		fscid = ceph_decode_32(&info_p);
4700 		namelen = ceph_decode_32(&info_p);
4701 		ceph_decode_need(&info_p, info_end, namelen, bad);
4702 
4703 		if (mds_namespace &&
4704 		    strlen(mds_namespace) == namelen &&
4705 		    !strncmp(mds_namespace, (char *)info_p, namelen)) {
4706 			mount_fscid = fscid;
4707 			break;
4708 		}
4709 	}
4710 
4711 	ceph_monc_got_map(&fsc->client->monc, CEPH_SUB_FSMAP, epoch);
4712 	if (mount_fscid != (u32)-1) {
4713 		fsc->client->monc.fs_cluster_id = mount_fscid;
4714 		ceph_monc_want_map(&fsc->client->monc, CEPH_SUB_MDSMAP,
4715 				   0, true);
4716 		ceph_monc_renew_subs(&fsc->client->monc);
4717 	} else {
4718 		err = -ENOENT;
4719 		goto err_out;
4720 	}
4721 	return;
4722 
4723 bad:
4724 	pr_err("error decoding fsmap\n");
4725 err_out:
4726 	mutex_lock(&mdsc->mutex);
4727 	mdsc->mdsmap_err = err;
4728 	__wake_requests(mdsc, &mdsc->waiting_for_map);
4729 	mutex_unlock(&mdsc->mutex);
4730 }
4731 
4732 /*
4733  * handle mds map update.
4734  */
4735 void ceph_mdsc_handle_mdsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg)
4736 {
4737 	u32 epoch;
4738 	u32 maplen;
4739 	void *p = msg->front.iov_base;
4740 	void *end = p + msg->front.iov_len;
4741 	struct ceph_mdsmap *newmap, *oldmap;
4742 	struct ceph_fsid fsid;
4743 	int err = -EINVAL;
4744 
4745 	ceph_decode_need(&p, end, sizeof(fsid)+2*sizeof(u32), bad);
4746 	ceph_decode_copy(&p, &fsid, sizeof(fsid));
4747 	if (ceph_check_fsid(mdsc->fsc->client, &fsid) < 0)
4748 		return;
4749 	epoch = ceph_decode_32(&p);
4750 	maplen = ceph_decode_32(&p);
4751 	dout("handle_map epoch %u len %d\n", epoch, (int)maplen);
4752 
4753 	/* do we need it? */
4754 	mutex_lock(&mdsc->mutex);
4755 	if (mdsc->mdsmap && epoch <= mdsc->mdsmap->m_epoch) {
4756 		dout("handle_map epoch %u <= our %u\n",
4757 		     epoch, mdsc->mdsmap->m_epoch);
4758 		mutex_unlock(&mdsc->mutex);
4759 		return;
4760 	}
4761 
4762 	newmap = ceph_mdsmap_decode(&p, end);
4763 	if (IS_ERR(newmap)) {
4764 		err = PTR_ERR(newmap);
4765 		goto bad_unlock;
4766 	}
4767 
4768 	/* swap into place */
4769 	if (mdsc->mdsmap) {
4770 		oldmap = mdsc->mdsmap;
4771 		mdsc->mdsmap = newmap;
4772 		check_new_map(mdsc, newmap, oldmap);
4773 		ceph_mdsmap_destroy(oldmap);
4774 	} else {
4775 		mdsc->mdsmap = newmap;  /* first mds map */
4776 	}
4777 	mdsc->fsc->max_file_size = min((loff_t)mdsc->mdsmap->m_max_file_size,
4778 					MAX_LFS_FILESIZE);
4779 
4780 	__wake_requests(mdsc, &mdsc->waiting_for_map);
4781 	ceph_monc_got_map(&mdsc->fsc->client->monc, CEPH_SUB_MDSMAP,
4782 			  mdsc->mdsmap->m_epoch);
4783 
4784 	mutex_unlock(&mdsc->mutex);
4785 	schedule_delayed(mdsc);
4786 	return;
4787 
4788 bad_unlock:
4789 	mutex_unlock(&mdsc->mutex);
4790 bad:
4791 	pr_err("error decoding mdsmap %d\n", err);
4792 	return;
4793 }
4794 
4795 static struct ceph_connection *con_get(struct ceph_connection *con)
4796 {
4797 	struct ceph_mds_session *s = con->private;
4798 
4799 	if (ceph_get_mds_session(s))
4800 		return con;
4801 	return NULL;
4802 }
4803 
4804 static void con_put(struct ceph_connection *con)
4805 {
4806 	struct ceph_mds_session *s = con->private;
4807 
4808 	ceph_put_mds_session(s);
4809 }
4810 
4811 /*
4812  * if the client is unresponsive for long enough, the mds will kill
4813  * the session entirely.
4814  */
4815 static void peer_reset(struct ceph_connection *con)
4816 {
4817 	struct ceph_mds_session *s = con->private;
4818 	struct ceph_mds_client *mdsc = s->s_mdsc;
4819 
4820 	pr_warn("mds%d closed our session\n", s->s_mds);
4821 	send_mds_reconnect(mdsc, s);
4822 }
4823 
4824 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
4825 {
4826 	struct ceph_mds_session *s = con->private;
4827 	struct ceph_mds_client *mdsc = s->s_mdsc;
4828 	int type = le16_to_cpu(msg->hdr.type);
4829 
4830 	mutex_lock(&mdsc->mutex);
4831 	if (__verify_registered_session(mdsc, s) < 0) {
4832 		mutex_unlock(&mdsc->mutex);
4833 		goto out;
4834 	}
4835 	mutex_unlock(&mdsc->mutex);
4836 
4837 	switch (type) {
4838 	case CEPH_MSG_MDS_MAP:
4839 		ceph_mdsc_handle_mdsmap(mdsc, msg);
4840 		break;
4841 	case CEPH_MSG_FS_MAP_USER:
4842 		ceph_mdsc_handle_fsmap(mdsc, msg);
4843 		break;
4844 	case CEPH_MSG_CLIENT_SESSION:
4845 		handle_session(s, msg);
4846 		break;
4847 	case CEPH_MSG_CLIENT_REPLY:
4848 		handle_reply(s, msg);
4849 		break;
4850 	case CEPH_MSG_CLIENT_REQUEST_FORWARD:
4851 		handle_forward(mdsc, s, msg);
4852 		break;
4853 	case CEPH_MSG_CLIENT_CAPS:
4854 		ceph_handle_caps(s, msg);
4855 		break;
4856 	case CEPH_MSG_CLIENT_SNAP:
4857 		ceph_handle_snap(mdsc, s, msg);
4858 		break;
4859 	case CEPH_MSG_CLIENT_LEASE:
4860 		handle_lease(mdsc, s, msg);
4861 		break;
4862 	case CEPH_MSG_CLIENT_QUOTA:
4863 		ceph_handle_quota(mdsc, s, msg);
4864 		break;
4865 
4866 	default:
4867 		pr_err("received unknown message type %d %s\n", type,
4868 		       ceph_msg_type_name(type));
4869 	}
4870 out:
4871 	ceph_msg_put(msg);
4872 }
4873 
4874 /*
4875  * authentication
4876  */
4877 
4878 /*
4879  * Note: returned pointer is the address of a structure that's
4880  * managed separately.  Caller must *not* attempt to free it.
4881  */
4882 static struct ceph_auth_handshake *get_authorizer(struct ceph_connection *con,
4883 					int *proto, int force_new)
4884 {
4885 	struct ceph_mds_session *s = con->private;
4886 	struct ceph_mds_client *mdsc = s->s_mdsc;
4887 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4888 	struct ceph_auth_handshake *auth = &s->s_auth;
4889 
4890 	if (force_new && auth->authorizer) {
4891 		ceph_auth_destroy_authorizer(auth->authorizer);
4892 		auth->authorizer = NULL;
4893 	}
4894 	if (!auth->authorizer) {
4895 		int ret = ceph_auth_create_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4896 						      auth);
4897 		if (ret)
4898 			return ERR_PTR(ret);
4899 	} else {
4900 		int ret = ceph_auth_update_authorizer(ac, CEPH_ENTITY_TYPE_MDS,
4901 						      auth);
4902 		if (ret)
4903 			return ERR_PTR(ret);
4904 	}
4905 	*proto = ac->protocol;
4906 
4907 	return auth;
4908 }
4909 
4910 static int add_authorizer_challenge(struct ceph_connection *con,
4911 				    void *challenge_buf, int challenge_buf_len)
4912 {
4913 	struct ceph_mds_session *s = con->private;
4914 	struct ceph_mds_client *mdsc = s->s_mdsc;
4915 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4916 
4917 	return ceph_auth_add_authorizer_challenge(ac, s->s_auth.authorizer,
4918 					    challenge_buf, challenge_buf_len);
4919 }
4920 
4921 static int verify_authorizer_reply(struct ceph_connection *con)
4922 {
4923 	struct ceph_mds_session *s = con->private;
4924 	struct ceph_mds_client *mdsc = s->s_mdsc;
4925 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4926 
4927 	return ceph_auth_verify_authorizer_reply(ac, s->s_auth.authorizer);
4928 }
4929 
4930 static int invalidate_authorizer(struct ceph_connection *con)
4931 {
4932 	struct ceph_mds_session *s = con->private;
4933 	struct ceph_mds_client *mdsc = s->s_mdsc;
4934 	struct ceph_auth_client *ac = mdsc->fsc->client->monc.auth;
4935 
4936 	ceph_auth_invalidate_authorizer(ac, CEPH_ENTITY_TYPE_MDS);
4937 
4938 	return ceph_monc_validate_auth(&mdsc->fsc->client->monc);
4939 }
4940 
4941 static struct ceph_msg *mds_alloc_msg(struct ceph_connection *con,
4942 				struct ceph_msg_header *hdr, int *skip)
4943 {
4944 	struct ceph_msg *msg;
4945 	int type = (int) le16_to_cpu(hdr->type);
4946 	int front_len = (int) le32_to_cpu(hdr->front_len);
4947 
4948 	if (con->in_msg)
4949 		return con->in_msg;
4950 
4951 	*skip = 0;
4952 	msg = ceph_msg_new(type, front_len, GFP_NOFS, false);
4953 	if (!msg) {
4954 		pr_err("unable to allocate msg type %d len %d\n",
4955 		       type, front_len);
4956 		return NULL;
4957 	}
4958 
4959 	return msg;
4960 }
4961 
4962 static int mds_sign_message(struct ceph_msg *msg)
4963 {
4964        struct ceph_mds_session *s = msg->con->private;
4965        struct ceph_auth_handshake *auth = &s->s_auth;
4966 
4967        return ceph_auth_sign_message(auth, msg);
4968 }
4969 
4970 static int mds_check_message_signature(struct ceph_msg *msg)
4971 {
4972        struct ceph_mds_session *s = msg->con->private;
4973        struct ceph_auth_handshake *auth = &s->s_auth;
4974 
4975        return ceph_auth_check_message_signature(auth, msg);
4976 }
4977 
4978 static const struct ceph_connection_operations mds_con_ops = {
4979 	.get = con_get,
4980 	.put = con_put,
4981 	.dispatch = dispatch,
4982 	.get_authorizer = get_authorizer,
4983 	.add_authorizer_challenge = add_authorizer_challenge,
4984 	.verify_authorizer_reply = verify_authorizer_reply,
4985 	.invalidate_authorizer = invalidate_authorizer,
4986 	.peer_reset = peer_reset,
4987 	.alloc_msg = mds_alloc_msg,
4988 	.sign_message = mds_sign_message,
4989 	.check_message_signature = mds_check_message_signature,
4990 };
4991 
4992 /* eof */
4993